2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 Nicira, Inc.
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"
26 #include "fail-open.h"
29 #include "ofp-actions.h"
33 #include "ofproto-provider.h"
35 #include "poll-loop.h"
42 #include "openvswitch/vconn.h"
43 #include "openvswitch/vlog.h"
47 VLOG_DEFINE_THIS_MODULE(connmgr);
48 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
50 /* An OpenFlow connection.
56 * 'ofproto_mutex' must be held whenever an ofconn is created or destroyed or,
57 * more or less equivalently, whenever an ofconn is added to or removed from a
58 * connmgr. 'ofproto_mutex' doesn't protect the data inside the ofconn, except
59 * as specifically noted below. */
61 /* Configuration that persists from one connection to the next. */
63 struct ovs_list node; /* In struct connmgr's "all_conns" list. */
64 struct hmap_node hmap_node; /* In struct connmgr's "controllers" map. */
66 struct connmgr *connmgr; /* Connection's manager. */
67 struct rconn *rconn; /* OpenFlow connection. */
68 enum ofconn_type type; /* Type. */
69 enum ofproto_band band; /* In-band or out-of-band? */
70 bool enable_async_msgs; /* Initially enable async messages? */
72 /* State that should be cleared from one connection to the next. */
75 enum ofp12_controller_role role; /* Role. */
76 enum ofputil_protocol protocol; /* Current protocol variant. */
77 enum nx_packet_in_format packet_in_format; /* OFPT_PACKET_IN format. */
79 /* OFPT_PACKET_IN related data. */
80 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
81 #define N_SCHEDULERS 2
82 struct pinsched *schedulers[N_SCHEDULERS];
83 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
84 int miss_send_len; /* Bytes to send of buffered packets. */
85 uint16_t controller_id; /* Connection controller ID. */
87 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
88 * requests, and the maximum number before we stop reading OpenFlow
90 #define OFCONN_REPLY_MAX 100
91 struct rconn_packet_counter *reply_counter;
93 /* Asynchronous message configuration in each possible role.
95 * A 1-bit enables sending an asynchronous message for one possible reason
96 * that the message might be generated, a 0-bit disables it. */
97 struct ofputil_async_cfg *async_cfg;
99 /* Flow table operation logging. */
100 int n_add, n_delete, n_modify; /* Number of unreported ops of each kind. */
101 long long int first_op, last_op; /* Range of times for unreported ops. */
102 long long int next_op_report; /* Time to report ops, or LLONG_MAX. */
103 long long int op_backoff; /* Earliest time to report ops again. */
105 /* Flow monitors (e.g. NXST_FLOW_MONITOR). */
107 /* Configuration. Contains "struct ofmonitor"s. */
108 struct hmap monitors OVS_GUARDED_BY(ofproto_mutex);
112 * When too many flow monitor notifications back up in the transmit buffer,
113 * we pause the transmission of further notifications. These members track
114 * the flow control state.
116 * When notifications are flowing, 'monitor_paused' is 0. When
117 * notifications are paused, 'monitor_paused' is the value of
118 * 'monitor_seqno' at the point we paused.
120 * 'monitor_counter' counts the OpenFlow messages and bytes currently in
121 * flight. This value growing too large triggers pausing. */
122 uint64_t monitor_paused OVS_GUARDED_BY(ofproto_mutex);
123 struct rconn_packet_counter *monitor_counter OVS_GUARDED_BY(ofproto_mutex);
125 /* State of monitors for a single ongoing flow_mod.
127 * 'updates' is a list of "struct ofpbuf"s that contain
128 * NXST_FLOW_MONITOR_REPLY messages representing the changes made by the
131 * When 'updates' is nonempty, 'sent_abbrev_update' is true if 'updates'
132 * contains an update event of type NXFME_ABBREV and false otherwise.. */
133 struct ovs_list updates OVS_GUARDED_BY(ofproto_mutex);
134 bool sent_abbrev_update OVS_GUARDED_BY(ofproto_mutex);
136 /* Active bundles. Contains "struct ofp_bundle"s. */
140 static struct ofconn *ofconn_create(struct connmgr *, struct rconn *,
141 enum ofconn_type, bool enable_async_msgs)
142 OVS_REQUIRES(ofproto_mutex);
143 static void ofconn_destroy(struct ofconn *) OVS_REQUIRES(ofproto_mutex);
144 static void ofconn_flush(struct ofconn *) OVS_REQUIRES(ofproto_mutex);
146 static void ofconn_reconfigure(struct ofconn *,
147 const struct ofproto_controller *);
149 static void ofconn_run(struct ofconn *,
150 void (*handle_openflow)(struct ofconn *,
151 const struct ofpbuf *ofp_msg));
152 static void ofconn_wait(struct ofconn *);
154 static void ofconn_log_flow_mods(struct ofconn *);
156 static const char *ofconn_get_target(const struct ofconn *);
157 static char *ofconn_make_name(const struct connmgr *, const char *target);
159 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
161 static void ofconn_send(const struct ofconn *, struct ofpbuf *,
162 struct rconn_packet_counter *);
164 static void do_send_packet_ins(struct ofconn *, struct ovs_list *txq);
166 /* A listener for incoming OpenFlow "service" connections. */
168 struct hmap_node node; /* In struct connmgr's "services" hmap. */
169 struct pvconn *pvconn; /* OpenFlow connection listener. */
171 /* These are not used by ofservice directly. They are settings for
172 * accepted "struct ofconn"s from the pvconn. */
173 int probe_interval; /* Max idle time before probing, in seconds. */
174 int rate_limit; /* Max packet-in rate in packets per second. */
175 int burst_limit; /* Limit on accumulating packet credits. */
176 bool enable_async_msgs; /* Initially enable async messages? */
177 uint8_t dscp; /* DSCP Value for controller connection */
178 uint32_t allowed_versions; /* OpenFlow protocol versions that may
179 * be negotiated for a session. */
182 static void ofservice_reconfigure(struct ofservice *,
183 const struct ofproto_controller *);
184 static int ofservice_create(struct connmgr *mgr, const char *target,
185 uint32_t allowed_versions, uint8_t dscp);
186 static void ofservice_destroy(struct connmgr *, struct ofservice *);
187 static struct ofservice *ofservice_lookup(struct connmgr *,
190 /* Connection manager for an OpenFlow switch. */
192 struct ofproto *ofproto;
194 char *local_port_name;
196 /* OpenFlow connections. */
197 struct hmap controllers; /* All OFCONN_PRIMARY controllers. */
198 struct ovs_list all_conns; /* All controllers. */
199 uint64_t master_election_id; /* monotonically increasing sequence number
200 * for master election */
201 bool master_election_id_defined;
203 /* OpenFlow listeners. */
204 struct hmap services; /* Contains "struct ofservice"s. */
205 struct pvconn **snoops;
209 struct fail_open *fail_open;
210 enum ofproto_fail_mode fail_mode;
212 /* In-band control. */
213 struct in_band *in_band;
214 struct sockaddr_in *extra_in_band_remotes;
215 size_t n_extra_remotes;
219 static void update_in_band_remotes(struct connmgr *);
220 static void add_snooper(struct connmgr *, struct vconn *);
221 static void ofmonitor_run(struct connmgr *);
222 static void ofmonitor_wait(struct connmgr *);
224 /* Creates and returns a new connection manager owned by 'ofproto'. 'name' is
225 * a name for the ofproto suitable for using in log messages.
226 * 'local_port_name' is the name of the local port (OFPP_LOCAL) within
229 connmgr_create(struct ofproto *ofproto,
230 const char *name, const char *local_port_name)
234 mgr = xmalloc(sizeof *mgr);
235 mgr->ofproto = ofproto;
236 mgr->name = xstrdup(name);
237 mgr->local_port_name = xstrdup(local_port_name);
239 hmap_init(&mgr->controllers);
240 list_init(&mgr->all_conns);
241 mgr->master_election_id = 0;
242 mgr->master_election_id_defined = false;
244 hmap_init(&mgr->services);
248 mgr->fail_open = NULL;
249 mgr->fail_mode = OFPROTO_FAIL_SECURE;
252 mgr->extra_in_band_remotes = NULL;
253 mgr->n_extra_remotes = 0;
254 mgr->in_band_queue = -1;
259 /* Frees 'mgr' and all of its resources. */
261 connmgr_destroy(struct connmgr *mgr)
263 struct ofservice *ofservice, *next_ofservice;
264 struct ofconn *ofconn, *next_ofconn;
271 ovs_mutex_lock(&ofproto_mutex);
272 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &mgr->all_conns) {
273 ofconn_destroy(ofconn);
275 ovs_mutex_unlock(&ofproto_mutex);
277 hmap_destroy(&mgr->controllers);
279 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &mgr->services) {
280 ofservice_destroy(mgr, ofservice);
282 hmap_destroy(&mgr->services);
284 for (i = 0; i < mgr->n_snoops; i++) {
285 pvconn_close(mgr->snoops[i]);
289 fail_open_destroy(mgr->fail_open);
290 mgr->fail_open = NULL;
292 in_band_destroy(mgr->in_band);
294 free(mgr->extra_in_band_remotes);
296 free(mgr->local_port_name);
301 /* Does all of the periodic maintenance required by 'mgr'. Calls
302 * 'handle_openflow' for each message received on an OpenFlow connection,
303 * passing along the OpenFlow connection itself and the message that was sent.
304 * 'handle_openflow' must not modify or free the message. */
306 connmgr_run(struct connmgr *mgr,
307 void (*handle_openflow)(struct ofconn *,
308 const struct ofpbuf *ofp_msg))
309 OVS_EXCLUDED(ofproto_mutex)
311 struct ofconn *ofconn, *next_ofconn;
312 struct ofservice *ofservice;
316 if (!in_band_run(mgr->in_band)) {
317 in_band_destroy(mgr->in_band);
322 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &mgr->all_conns) {
323 ofconn_run(ofconn, handle_openflow);
327 /* Fail-open maintenance. Do this after processing the ofconns since
328 * fail-open checks the status of the controller rconn. */
329 if (mgr->fail_open) {
330 fail_open_run(mgr->fail_open);
333 HMAP_FOR_EACH (ofservice, node, &mgr->services) {
337 retval = pvconn_accept(ofservice->pvconn, &vconn);
342 /* Passing default value for creation of the rconn */
343 rconn = rconn_create(ofservice->probe_interval, 0, ofservice->dscp,
344 vconn_get_allowed_versions(vconn));
345 name = ofconn_make_name(mgr, vconn_get_name(vconn));
346 rconn_connect_unreliably(rconn, vconn, name);
349 ovs_mutex_lock(&ofproto_mutex);
350 ofconn = ofconn_create(mgr, rconn, OFCONN_SERVICE,
351 ofservice->enable_async_msgs);
352 ovs_mutex_unlock(&ofproto_mutex);
354 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
355 ofservice->burst_limit);
356 } else if (retval != EAGAIN) {
357 VLOG_WARN_RL(&rl, "accept failed (%s)", ovs_strerror(retval));
361 for (i = 0; i < mgr->n_snoops; i++) {
365 retval = pvconn_accept(mgr->snoops[i], &vconn);
367 add_snooper(mgr, vconn);
368 } else if (retval != EAGAIN) {
369 VLOG_WARN_RL(&rl, "accept failed (%s)", ovs_strerror(retval));
374 /* Causes the poll loop to wake up when connmgr_run() needs to run. */
376 connmgr_wait(struct connmgr *mgr)
378 struct ofservice *ofservice;
379 struct ofconn *ofconn;
382 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
387 in_band_wait(mgr->in_band);
389 if (mgr->fail_open) {
390 fail_open_wait(mgr->fail_open);
392 HMAP_FOR_EACH (ofservice, node, &mgr->services) {
393 pvconn_wait(ofservice->pvconn);
395 for (i = 0; i < mgr->n_snoops; i++) {
396 pvconn_wait(mgr->snoops[i]);
400 /* Adds some memory usage statistics for 'mgr' into 'usage', for use with
401 * memory_report(). */
403 connmgr_get_memory_usage(const struct connmgr *mgr, struct simap *usage)
405 const struct ofconn *ofconn;
406 unsigned int packets = 0;
407 unsigned int ofconns = 0;
409 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
414 packets += rconn_count_txqlen(ofconn->rconn);
415 for (i = 0; i < N_SCHEDULERS; i++) {
416 struct pinsched_stats stats;
418 pinsched_get_stats(ofconn->schedulers[i], &stats);
419 packets += stats.n_queued;
421 packets += pktbuf_count_packets(ofconn->pktbuf);
423 simap_increase(usage, "ofconns", ofconns);
424 simap_increase(usage, "packets", packets);
427 /* Returns the ofproto that owns 'ofconn''s connmgr. */
429 ofconn_get_ofproto(const struct ofconn *ofconn)
431 return ofconn->connmgr->ofproto;
434 /* OpenFlow configuration. */
436 static void add_controller(struct connmgr *, const char *target, uint8_t dscp,
437 uint32_t allowed_versions)
438 OVS_REQUIRES(ofproto_mutex);
439 static struct ofconn *find_controller_by_target(struct connmgr *,
441 static void update_fail_open(struct connmgr *) OVS_EXCLUDED(ofproto_mutex);
442 static int set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
443 const struct sset *);
445 /* Returns true if 'mgr' has any configured primary controllers.
447 * Service controllers do not count, but configured primary controllers do
448 * count whether or not they are currently connected. */
450 connmgr_has_controllers(const struct connmgr *mgr)
452 return !hmap_is_empty(&mgr->controllers);
455 /* Initializes 'info' and populates it with information about each configured
456 * primary controller. The keys in 'info' are the controllers' targets; the
457 * data values are corresponding "struct ofproto_controller_info".
459 * The caller owns 'info' and everything in it and should free it when it is no
462 connmgr_get_controller_info(struct connmgr *mgr, struct shash *info)
464 const struct ofconn *ofconn;
466 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
467 const struct rconn *rconn = ofconn->rconn;
468 const char *target = rconn_get_target(rconn);
470 if (!shash_find(info, target)) {
471 struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo);
472 time_t now = time_now();
473 time_t last_connection = rconn_get_last_connection(rconn);
474 time_t last_disconnect = rconn_get_last_disconnect(rconn);
475 int last_error = rconn_get_last_error(rconn);
478 shash_add(info, target, cinfo);
480 cinfo->is_connected = rconn_is_connected(rconn);
481 cinfo->role = ofconn->role;
483 smap_init(&cinfo->pairs);
485 smap_add(&cinfo->pairs, "last_error",
486 ovs_retval_to_string(last_error));
489 smap_add(&cinfo->pairs, "state", rconn_get_state(rconn));
491 if (last_connection != TIME_MIN) {
492 smap_add_format(&cinfo->pairs, "sec_since_connect",
493 "%ld", (long int) (now - last_connection));
496 if (last_disconnect != TIME_MIN) {
497 smap_add_format(&cinfo->pairs, "sec_since_disconnect",
498 "%ld", (long int) (now - last_disconnect));
501 for (i = 0; i < N_SCHEDULERS; i++) {
502 if (ofconn->schedulers[i]) {
503 const char *name = i ? "miss" : "action";
504 struct pinsched_stats stats;
506 pinsched_get_stats(ofconn->schedulers[i], &stats);
507 smap_add_nocopy(&cinfo->pairs,
508 xasprintf("packet-in-%s-backlog", name),
509 xasprintf("%u", stats.n_queued));
510 smap_add_nocopy(&cinfo->pairs,
511 xasprintf("packet-in-%s-bypassed", name),
512 xasprintf("%llu", stats.n_normal));
513 smap_add_nocopy(&cinfo->pairs,
514 xasprintf("packet-in-%s-queued", name),
515 xasprintf("%llu", stats.n_limited));
516 smap_add_nocopy(&cinfo->pairs,
517 xasprintf("packet-in-%s-dropped", name),
518 xasprintf("%llu", stats.n_queue_dropped));
526 connmgr_free_controller_info(struct shash *info)
528 struct shash_node *node;
530 SHASH_FOR_EACH (node, info) {
531 struct ofproto_controller_info *cinfo = node->data;
532 smap_destroy(&cinfo->pairs);
538 /* Changes 'mgr''s set of controllers to the 'n_controllers' controllers in
541 connmgr_set_controllers(struct connmgr *mgr,
542 const struct ofproto_controller *controllers,
543 size_t n_controllers, uint32_t allowed_versions)
544 OVS_EXCLUDED(ofproto_mutex)
546 bool had_controllers = connmgr_has_controllers(mgr);
547 struct shash new_controllers;
548 struct ofconn *ofconn, *next_ofconn;
549 struct ofservice *ofservice, *next_ofservice;
552 /* Required to add and remove ofconns. This could probably be narrowed to
553 * cover a smaller amount of code, if that yielded some benefit. */
554 ovs_mutex_lock(&ofproto_mutex);
556 /* Create newly configured controllers and services.
557 * Create a name to ofproto_controller mapping in 'new_controllers'. */
558 shash_init(&new_controllers);
559 for (i = 0; i < n_controllers; i++) {
560 const struct ofproto_controller *c = &controllers[i];
562 if (!vconn_verify_name(c->target)) {
564 ofconn = find_controller_by_target(mgr, c->target);
566 VLOG_INFO("%s: added primary controller \"%s\"",
567 mgr->name, c->target);
569 } else if (rconn_get_allowed_versions(ofconn->rconn) !=
571 VLOG_INFO("%s: re-added primary controller \"%s\"",
572 mgr->name, c->target);
574 ofconn_destroy(ofconn);
577 add_controller(mgr, c->target, c->dscp, allowed_versions);
579 } else if (!pvconn_verify_name(c->target)) {
581 ofservice = ofservice_lookup(mgr, c->target);
583 VLOG_INFO("%s: added service controller \"%s\"",
584 mgr->name, c->target);
586 } else if (ofservice->allowed_versions != allowed_versions) {
587 VLOG_INFO("%s: re-added service controller \"%s\"",
588 mgr->name, c->target);
589 ofservice_destroy(mgr, ofservice);
593 ofservice_create(mgr, c->target, allowed_versions, c->dscp);
596 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
597 mgr->name, c->target);
601 shash_add_once(&new_controllers, c->target, &controllers[i]);
604 /* Delete controllers that are no longer configured.
605 * Update configuration of all now-existing controllers. */
606 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &mgr->controllers) {
607 const char *target = ofconn_get_target(ofconn);
608 struct ofproto_controller *c;
610 c = shash_find_data(&new_controllers, target);
612 VLOG_INFO("%s: removed primary controller \"%s\"",
614 ofconn_destroy(ofconn);
616 ofconn_reconfigure(ofconn, c);
620 /* Delete services that are no longer configured.
621 * Update configuration of all now-existing services. */
622 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &mgr->services) {
623 const char *target = pvconn_get_name(ofservice->pvconn);
624 struct ofproto_controller *c;
626 c = shash_find_data(&new_controllers, target);
628 VLOG_INFO("%s: removed service controller \"%s\"",
630 ofservice_destroy(mgr, ofservice);
632 ofservice_reconfigure(ofservice, c);
636 shash_destroy(&new_controllers);
638 ovs_mutex_unlock(&ofproto_mutex);
640 update_in_band_remotes(mgr);
641 update_fail_open(mgr);
642 if (had_controllers != connmgr_has_controllers(mgr)) {
643 ofproto_flush_flows(mgr->ofproto);
647 /* Drops the connections between 'mgr' and all of its primary and secondary
648 * controllers, forcing them to reconnect. */
650 connmgr_reconnect(const struct connmgr *mgr)
652 struct ofconn *ofconn;
654 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
655 rconn_reconnect(ofconn->rconn);
659 /* Sets the "snoops" for 'mgr' to the pvconn targets listed in 'snoops'.
661 * A "snoop" is a pvconn to which every OpenFlow message to or from the most
662 * important controller on 'mgr' is mirrored. */
664 connmgr_set_snoops(struct connmgr *mgr, const struct sset *snoops)
666 return set_pvconns(&mgr->snoops, &mgr->n_snoops, snoops);
669 /* Adds each of the snoops currently configured on 'mgr' to 'snoops'. */
671 connmgr_get_snoops(const struct connmgr *mgr, struct sset *snoops)
675 for (i = 0; i < mgr->n_snoops; i++) {
676 sset_add(snoops, pvconn_get_name(mgr->snoops[i]));
680 /* Returns true if 'mgr' has at least one snoop, false if it has none. */
682 connmgr_has_snoops(const struct connmgr *mgr)
684 return mgr->n_snoops > 0;
687 /* Creates a new controller for 'target' in 'mgr'. update_controller() needs
688 * to be called later to finish the new ofconn's configuration. */
690 add_controller(struct connmgr *mgr, const char *target, uint8_t dscp,
691 uint32_t allowed_versions)
692 OVS_REQUIRES(ofproto_mutex)
694 char *name = ofconn_make_name(mgr, target);
695 struct ofconn *ofconn;
697 ofconn = ofconn_create(mgr, rconn_create(5, 8, dscp, allowed_versions),
698 OFCONN_PRIMARY, true);
699 ofconn->pktbuf = pktbuf_create();
700 rconn_connect(ofconn->rconn, target, name);
701 hmap_insert(&mgr->controllers, &ofconn->hmap_node, hash_string(target, 0));
706 static struct ofconn *
707 find_controller_by_target(struct connmgr *mgr, const char *target)
709 struct ofconn *ofconn;
711 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
712 hash_string(target, 0), &mgr->controllers) {
713 if (!strcmp(ofconn_get_target(ofconn), target)) {
721 update_in_band_remotes(struct connmgr *mgr)
723 struct sockaddr_in *addrs;
724 size_t max_addrs, n_addrs;
725 struct ofconn *ofconn;
728 /* Allocate enough memory for as many remotes as we could possibly have. */
729 max_addrs = mgr->n_extra_remotes + hmap_count(&mgr->controllers);
730 addrs = xmalloc(max_addrs * sizeof *addrs);
733 /* Add all the remotes. */
734 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
735 const char *target = rconn_get_target(ofconn->rconn);
737 struct sockaddr_storage ss;
738 struct sockaddr_in in;
741 if (ofconn->band == OFPROTO_IN_BAND
742 && stream_parse_target_with_default_port(target, OFP_PORT, &sa.ss)
743 && sa.ss.ss_family == AF_INET) {
744 addrs[n_addrs++] = sa.in;
747 for (i = 0; i < mgr->n_extra_remotes; i++) {
748 addrs[n_addrs++] = mgr->extra_in_band_remotes[i];
751 /* Create or update or destroy in-band. */
754 in_band_create(mgr->ofproto, mgr->local_port_name, &mgr->in_band);
756 in_band_set_queue(mgr->in_band, mgr->in_band_queue);
758 /* in_band_run() needs a chance to delete any existing in-band flows.
759 * We will destroy mgr->in_band after it's done with that. */
762 in_band_set_remotes(mgr->in_band, addrs, n_addrs);
770 update_fail_open(struct connmgr *mgr)
771 OVS_EXCLUDED(ofproto_mutex)
773 if (connmgr_has_controllers(mgr)
774 && mgr->fail_mode == OFPROTO_FAIL_STANDALONE) {
775 if (!mgr->fail_open) {
776 mgr->fail_open = fail_open_create(mgr->ofproto, mgr);
779 fail_open_destroy(mgr->fail_open);
780 mgr->fail_open = NULL;
785 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
786 const struct sset *sset)
788 struct pvconn **pvconns = *pvconnsp;
789 size_t n_pvconns = *n_pvconnsp;
794 for (i = 0; i < n_pvconns; i++) {
795 pvconn_close(pvconns[i]);
799 pvconns = xmalloc(sset_count(sset) * sizeof *pvconns);
801 SSET_FOR_EACH (name, sset) {
802 struct pvconn *pvconn;
804 error = pvconn_open(name, 0, 0, &pvconn);
806 pvconns[n_pvconns++] = pvconn;
808 VLOG_ERR("failed to listen on %s: %s", name, ovs_strerror(error));
816 *n_pvconnsp = n_pvconns;
821 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
822 * means that 'ofconn' is more interesting for monitoring than a lower return
825 snoop_preference(const struct ofconn *ofconn)
827 switch (ofconn->role) {
828 case OFPCR12_ROLE_MASTER:
830 case OFPCR12_ROLE_EQUAL:
832 case OFPCR12_ROLE_SLAVE:
834 case OFPCR12_ROLE_NOCHANGE:
836 /* Shouldn't happen. */
841 /* One of 'mgr''s "snoop" pvconns has accepted a new connection on 'vconn'.
842 * Connects this vconn to a controller. */
844 add_snooper(struct connmgr *mgr, struct vconn *vconn)
846 struct ofconn *ofconn, *best;
848 /* Pick a controller for monitoring. */
850 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
851 if (ofconn->type == OFCONN_PRIMARY
852 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
858 rconn_add_monitor(best->rconn, vconn);
860 VLOG_INFO_RL(&rl, "no controller connection to snoop");
865 /* Public ofconn functions. */
867 /* Returns the connection type, either OFCONN_PRIMARY or OFCONN_SERVICE. */
869 ofconn_get_type(const struct ofconn *ofconn)
874 /* If a master election id is defined, stores it into '*idp' and returns
875 * true. Otherwise, stores UINT64_MAX into '*idp' and returns false. */
877 ofconn_get_master_election_id(const struct ofconn *ofconn, uint64_t *idp)
879 *idp = (ofconn->connmgr->master_election_id_defined
880 ? ofconn->connmgr->master_election_id
882 return ofconn->connmgr->master_election_id_defined;
885 /* Sets the master election id.
887 * Returns true if successful, false if the id is stale
890 ofconn_set_master_election_id(struct ofconn *ofconn, uint64_t id)
892 if (ofconn->connmgr->master_election_id_defined
894 /* Unsigned difference interpreted as a two's complement signed
896 (int64_t)(id - ofconn->connmgr->master_election_id) < 0) {
899 ofconn->connmgr->master_election_id = id;
900 ofconn->connmgr->master_election_id_defined = true;
905 /* Returns the role configured for 'ofconn'.
907 * The default role, if no other role has been set, is OFPCR12_ROLE_EQUAL. */
908 enum ofp12_controller_role
909 ofconn_get_role(const struct ofconn *ofconn)
915 ofconn_send_role_status(struct ofconn *ofconn, uint32_t role, uint8_t reason)
917 struct ofputil_role_status status;
920 status.reason = reason;
922 ofconn_get_master_election_id(ofconn, &status.generation_id);
924 buf = ofputil_encode_role_status(&status, ofconn_get_protocol(ofconn));
926 ofconn_send(ofconn, buf, NULL);
930 /* Changes 'ofconn''s role to 'role'. If 'role' is OFPCR12_ROLE_MASTER then
931 * any existing master is demoted to a slave. */
933 ofconn_set_role(struct ofconn *ofconn, enum ofp12_controller_role role)
935 if (role != ofconn->role && role == OFPCR12_ROLE_MASTER) {
936 struct ofconn *other;
938 LIST_FOR_EACH (other, node, &ofconn->connmgr->all_conns) {
939 if (other->role == OFPCR12_ROLE_MASTER) {
940 other->role = OFPCR12_ROLE_SLAVE;
941 ofconn_send_role_status(other, OFPCR12_ROLE_SLAVE, OFPCRR_MASTER_REQUEST);
949 ofconn_set_invalid_ttl_to_controller(struct ofconn *ofconn, bool enable)
951 struct ofputil_async_cfg ac = ofconn_get_async_config(ofconn);
952 uint32_t bit = 1u << OFPR_INVALID_TTL;
954 ac.master[OAM_PACKET_IN] |= bit;
956 ac.master[OAM_PACKET_IN] &= ~bit;
958 ofconn_set_async_config(ofconn, &ac);
962 ofconn_get_invalid_ttl_to_controller(struct ofconn *ofconn)
964 struct ofputil_async_cfg ac = ofconn_get_async_config(ofconn);
965 uint32_t bit = 1u << OFPR_INVALID_TTL;
966 return (ac.master[OAM_PACKET_IN] & bit) != 0;
969 /* Returns the currently configured protocol for 'ofconn', one of OFPUTIL_P_*.
971 * Returns OFPUTIL_P_NONE, which is not a valid protocol, if 'ofconn' hasn't
972 * completed version negotiation. This can't happen if at least one OpenFlow
973 * message, other than OFPT_HELLO, has been received on the connection (such as
974 * in ofproto.c's message handling code), since version negotiation is a
975 * prerequisite for starting to receive messages. This means that
976 * OFPUTIL_P_NONE is a special case that most callers need not worry about. */
977 enum ofputil_protocol
978 ofconn_get_protocol(const struct ofconn *ofconn)
980 if (ofconn->protocol == OFPUTIL_P_NONE &&
981 rconn_is_connected(ofconn->rconn)) {
982 int version = rconn_get_version(ofconn->rconn);
984 ofconn_set_protocol(CONST_CAST(struct ofconn *, ofconn),
985 ofputil_protocol_from_ofp_version(version));
989 return ofconn->protocol;
992 /* Sets the protocol for 'ofconn' to 'protocol' (one of OFPUTIL_P_*).
994 * (This doesn't actually send anything to accomplish this. Presumably the
995 * caller already did that.) */
997 ofconn_set_protocol(struct ofconn *ofconn, enum ofputil_protocol protocol)
999 ofconn->protocol = protocol;
1002 /* Returns the currently configured packet in format for 'ofconn', one of
1005 * The default, if no other format has been set, is NXPIF_STANDARD. */
1006 enum nx_packet_in_format
1007 ofconn_get_packet_in_format(struct ofconn *ofconn)
1009 return ofconn->packet_in_format;
1012 /* Sets the packet in format for 'ofconn' to 'packet_in_format' (one of
1015 ofconn_set_packet_in_format(struct ofconn *ofconn,
1016 enum nx_packet_in_format packet_in_format)
1018 ofconn->packet_in_format = packet_in_format;
1021 /* Sets the controller connection ID for 'ofconn' to 'controller_id'.
1023 * The connection controller ID is used for OFPP_CONTROLLER and
1024 * NXAST_CONTROLLER actions. See "struct nx_action_controller" for details. */
1026 ofconn_set_controller_id(struct ofconn *ofconn, uint16_t controller_id)
1028 ofconn->controller_id = controller_id;
1031 /* Returns the default miss send length for 'ofconn'. */
1033 ofconn_get_miss_send_len(const struct ofconn *ofconn)
1035 return ofconn->miss_send_len;
1038 /* Sets the default miss send length for 'ofconn' to 'miss_send_len'. */
1040 ofconn_set_miss_send_len(struct ofconn *ofconn, int miss_send_len)
1042 ofconn->miss_send_len = miss_send_len;
1046 ofconn_set_async_config(struct ofconn *ofconn,
1047 const struct ofputil_async_cfg *ac)
1049 if (!ofconn->async_cfg) {
1050 ofconn->async_cfg = xmalloc(sizeof *ofconn->async_cfg);
1052 *ofconn->async_cfg = *ac;
1055 struct ofputil_async_cfg
1056 ofconn_get_async_config(const struct ofconn *ofconn)
1058 if (ofconn->async_cfg) {
1059 return *ofconn->async_cfg;
1062 int version = rconn_get_version(ofconn->rconn);
1063 return (version < 0 || !ofconn->enable_async_msgs
1064 ? OFPUTIL_ASYNC_CFG_INIT
1065 : ofputil_async_cfg_default(version));
1068 /* Sends 'msg' on 'ofconn', accounting it as a reply. (If there is a
1069 * sufficient number of OpenFlow replies in-flight on a single ofconn, then the
1070 * connmgr will stop accepting new OpenFlow requests on that ofconn until the
1071 * controller has accepted some of the replies.) */
1073 ofconn_send_reply(const struct ofconn *ofconn, struct ofpbuf *msg)
1075 ofconn_send(ofconn, msg, ofconn->reply_counter);
1078 /* Sends each of the messages in list 'replies' on 'ofconn' in order,
1079 * accounting them as replies. */
1081 ofconn_send_replies(const struct ofconn *ofconn, struct ovs_list *replies)
1083 struct ofpbuf *reply;
1085 LIST_FOR_EACH_POP (reply, list_node, replies) {
1086 ofconn_send_reply(ofconn, reply);
1090 /* Sends 'error' on 'ofconn', as a reply to 'request'. Only at most the
1091 * first 64 bytes of 'request' are used. */
1093 ofconn_send_error(const struct ofconn *ofconn,
1094 const struct ofp_header *request, enum ofperr error)
1096 static struct vlog_rate_limit err_rl = VLOG_RATE_LIMIT_INIT(10, 10);
1097 struct ofpbuf *reply;
1099 reply = ofperr_encode_reply(error, request);
1100 if (!VLOG_DROP_INFO(&err_rl)) {
1101 const char *type_name;
1105 request_len = ntohs(request->length);
1106 type_name = (!ofpraw_decode_partial(&raw, request,
1107 MIN(64, request_len))
1108 ? ofpraw_get_name(raw)
1111 VLOG_INFO("%s: sending %s error reply to %s message",
1112 rconn_get_name(ofconn->rconn), ofperr_to_string(error),
1115 ofconn_send_reply(ofconn, reply);
1118 /* Same as pktbuf_retrieve(), using the pktbuf owned by 'ofconn'. */
1120 ofconn_pktbuf_retrieve(struct ofconn *ofconn, uint32_t id,
1121 struct dp_packet **bufferp, ofp_port_t *in_port)
1123 return pktbuf_retrieve(ofconn->pktbuf, id, bufferp, in_port);
1126 /* Reports that a flow_mod operation of the type specified by 'command' was
1127 * successfully executed by 'ofconn', so that the connmgr can log it. */
1129 ofconn_report_flow_mod(struct ofconn *ofconn,
1130 enum ofp_flow_mod_command command)
1140 case OFPFC_MODIFY_STRICT:
1145 case OFPFC_DELETE_STRICT:
1151 if (ofconn->next_op_report == LLONG_MAX) {
1152 ofconn->first_op = now;
1153 ofconn->next_op_report = MAX(now + 10 * 1000, ofconn->op_backoff);
1154 ofconn->op_backoff = ofconn->next_op_report + 60 * 1000;
1156 ofconn->last_op = now;
1159 /* OpenFlow 1.4 bundles. */
1161 static inline uint32_t
1162 bundle_hash(uint32_t id)
1164 return hash_int(id, 0);
1168 ofconn_get_bundle(struct ofconn *ofconn, uint32_t id)
1170 struct ofp_bundle *bundle;
1172 HMAP_FOR_EACH_IN_BUCKET(bundle, node, bundle_hash(id), &ofconn->bundles) {
1173 if (bundle->id == id) {
1182 ofconn_insert_bundle(struct ofconn *ofconn, struct ofp_bundle *bundle)
1184 /* XXX: Check the limit of open bundles */
1186 hmap_insert(&ofconn->bundles, &bundle->node, bundle_hash(bundle->id));
1192 ofconn_remove_bundle(struct ofconn *ofconn, struct ofp_bundle *bundle)
1194 hmap_remove(&ofconn->bundles, &bundle->node);
1200 bundle_remove_all(struct ofconn *ofconn)
1202 struct ofp_bundle *b, *next;
1204 HMAP_FOR_EACH_SAFE (b, next, node, &ofconn->bundles) {
1205 ofp_bundle_remove__(ofconn, b, false);
1209 /* Private ofconn functions. */
1212 ofconn_get_target(const struct ofconn *ofconn)
1214 return rconn_get_target(ofconn->rconn);
1217 static struct ofconn *
1218 ofconn_create(struct connmgr *mgr, struct rconn *rconn, enum ofconn_type type,
1219 bool enable_async_msgs)
1221 struct ofconn *ofconn;
1223 ofconn = xzalloc(sizeof *ofconn);
1224 ofconn->connmgr = mgr;
1225 list_push_back(&mgr->all_conns, &ofconn->node);
1226 ofconn->rconn = rconn;
1227 ofconn->type = type;
1228 ofconn->enable_async_msgs = enable_async_msgs;
1230 hmap_init(&ofconn->monitors);
1231 list_init(&ofconn->updates);
1233 hmap_init(&ofconn->bundles);
1235 ofconn_flush(ofconn);
1240 /* Clears all of the state in 'ofconn' that should not persist from one
1241 * connection to the next. */
1243 ofconn_flush(struct ofconn *ofconn)
1244 OVS_REQUIRES(ofproto_mutex)
1246 struct ofmonitor *monitor, *next_monitor;
1249 ofconn_log_flow_mods(ofconn);
1251 ofconn->role = OFPCR12_ROLE_EQUAL;
1252 ofconn_set_protocol(ofconn, OFPUTIL_P_NONE);
1253 ofconn->packet_in_format = NXPIF_STANDARD;
1255 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1256 ofconn->packet_in_counter = rconn_packet_counter_create();
1257 for (i = 0; i < N_SCHEDULERS; i++) {
1258 if (ofconn->schedulers[i]) {
1261 pinsched_get_limits(ofconn->schedulers[i], &rate, &burst);
1262 pinsched_destroy(ofconn->schedulers[i]);
1263 ofconn->schedulers[i] = pinsched_create(rate, burst);
1266 if (ofconn->pktbuf) {
1267 pktbuf_destroy(ofconn->pktbuf);
1268 ofconn->pktbuf = pktbuf_create();
1270 ofconn->miss_send_len = (ofconn->type == OFCONN_PRIMARY
1271 ? OFP_DEFAULT_MISS_SEND_LEN
1273 ofconn->controller_id = 0;
1275 rconn_packet_counter_destroy(ofconn->reply_counter);
1276 ofconn->reply_counter = rconn_packet_counter_create();
1278 free(ofconn->async_cfg);
1279 ofconn->async_cfg = NULL;
1281 ofconn->n_add = ofconn->n_delete = ofconn->n_modify = 0;
1282 ofconn->first_op = ofconn->last_op = LLONG_MIN;
1283 ofconn->next_op_report = LLONG_MAX;
1284 ofconn->op_backoff = LLONG_MIN;
1286 HMAP_FOR_EACH_SAFE (monitor, next_monitor, ofconn_node,
1287 &ofconn->monitors) {
1288 ofmonitor_destroy(monitor);
1290 rconn_packet_counter_destroy(ofconn->monitor_counter);
1291 ofconn->monitor_counter = rconn_packet_counter_create();
1292 ofpbuf_list_delete(&ofconn->updates); /* ...but it should be empty. */
1296 ofconn_destroy(struct ofconn *ofconn)
1297 OVS_REQUIRES(ofproto_mutex)
1299 ofconn_flush(ofconn);
1301 if (ofconn->type == OFCONN_PRIMARY) {
1302 hmap_remove(&ofconn->connmgr->controllers, &ofconn->hmap_node);
1305 bundle_remove_all(ofconn);
1306 hmap_destroy(&ofconn->bundles);
1308 hmap_destroy(&ofconn->monitors);
1309 list_remove(&ofconn->node);
1310 rconn_destroy(ofconn->rconn);
1311 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1312 rconn_packet_counter_destroy(ofconn->reply_counter);
1313 pktbuf_destroy(ofconn->pktbuf);
1314 rconn_packet_counter_destroy(ofconn->monitor_counter);
1318 /* Reconfigures 'ofconn' to match 'c'. 'ofconn' and 'c' must have the same
1321 ofconn_reconfigure(struct ofconn *ofconn, const struct ofproto_controller *c)
1325 ofconn->band = c->band;
1326 ofconn->enable_async_msgs = c->enable_async_msgs;
1328 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
1330 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
1331 rconn_set_probe_interval(ofconn->rconn, probe_interval);
1333 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
1335 /* If dscp value changed reconnect. */
1336 if (c->dscp != rconn_get_dscp(ofconn->rconn)) {
1337 rconn_set_dscp(ofconn->rconn, c->dscp);
1338 rconn_reconnect(ofconn->rconn);
1342 /* Returns true if it makes sense for 'ofconn' to receive and process OpenFlow
1345 ofconn_may_recv(const struct ofconn *ofconn)
1347 int count = rconn_packet_counter_n_packets(ofconn->reply_counter);
1348 return count < OFCONN_REPLY_MAX;
1352 ofconn_run(struct ofconn *ofconn,
1353 void (*handle_openflow)(struct ofconn *,
1354 const struct ofpbuf *ofp_msg))
1356 struct connmgr *mgr = ofconn->connmgr;
1359 for (i = 0; i < N_SCHEDULERS; i++) {
1360 struct ovs_list txq;
1362 pinsched_run(ofconn->schedulers[i], &txq);
1363 do_send_packet_ins(ofconn, &txq);
1366 rconn_run(ofconn->rconn);
1368 /* Limit the number of iterations to avoid starving other tasks. */
1369 for (i = 0; i < 50 && ofconn_may_recv(ofconn); i++) {
1370 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1375 if (mgr->fail_open) {
1376 fail_open_maybe_recover(mgr->fail_open);
1379 handle_openflow(ofconn, of_msg);
1380 ofpbuf_delete(of_msg);
1383 if (time_msec() >= ofconn->next_op_report) {
1384 ofconn_log_flow_mods(ofconn);
1387 ovs_mutex_lock(&ofproto_mutex);
1388 if (!rconn_is_alive(ofconn->rconn)) {
1389 ofconn_destroy(ofconn);
1390 } else if (!rconn_is_connected(ofconn->rconn)) {
1391 ofconn_flush(ofconn);
1393 ovs_mutex_unlock(&ofproto_mutex);
1397 ofconn_wait(struct ofconn *ofconn)
1401 for (i = 0; i < N_SCHEDULERS; i++) {
1402 pinsched_wait(ofconn->schedulers[i]);
1404 rconn_run_wait(ofconn->rconn);
1405 if (ofconn_may_recv(ofconn)) {
1406 rconn_recv_wait(ofconn->rconn);
1408 if (ofconn->next_op_report != LLONG_MAX) {
1409 poll_timer_wait_until(ofconn->next_op_report);
1414 ofconn_log_flow_mods(struct ofconn *ofconn)
1416 int n_flow_mods = ofconn->n_add + ofconn->n_delete + ofconn->n_modify;
1418 long long int ago = (time_msec() - ofconn->first_op) / 1000;
1419 long long int interval = (ofconn->last_op - ofconn->first_op) / 1000;
1423 ds_put_format(&s, "%d flow_mods ", n_flow_mods);
1424 if (interval == ago) {
1425 ds_put_format(&s, "in the last %lld s", ago);
1426 } else if (interval) {
1427 ds_put_format(&s, "in the %lld s starting %lld s ago",
1430 ds_put_format(&s, "%lld s ago", ago);
1433 ds_put_cstr(&s, " (");
1434 if (ofconn->n_add) {
1435 ds_put_format(&s, "%d adds, ", ofconn->n_add);
1437 if (ofconn->n_delete) {
1438 ds_put_format(&s, "%d deletes, ", ofconn->n_delete);
1440 if (ofconn->n_modify) {
1441 ds_put_format(&s, "%d modifications, ", ofconn->n_modify);
1444 ds_put_char(&s, ')');
1446 VLOG_INFO("%s: %s", rconn_get_name(ofconn->rconn), ds_cstr(&s));
1449 ofconn->n_add = ofconn->n_delete = ofconn->n_modify = 0;
1451 ofconn->next_op_report = LLONG_MAX;
1454 /* Returns true if 'ofconn' should receive asynchronous messages of the given
1455 * OAM_* 'type' and 'reason', which should be a OFPR_* value for OAM_PACKET_IN,
1456 * a OFPPR_* value for OAM_PORT_STATUS, or an OFPRR_* value for
1457 * OAM_FLOW_REMOVED. Returns false if the message should not be sent on
1460 ofconn_receives_async_msg(const struct ofconn *ofconn,
1461 enum ofputil_async_msg_type type,
1462 unsigned int reason)
1464 ovs_assert(reason < 32);
1465 ovs_assert((unsigned int) type < OAM_N_TYPES);
1467 /* Keep the following code in sync with the documentation in the
1468 * "Asynchronous Messages" section in DESIGN. */
1470 if (ofconn->type == OFCONN_SERVICE && !ofconn->miss_send_len) {
1471 /* Service connections don't get asynchronous messages unless they have
1472 * explicitly asked for them by setting a nonzero miss send length. */
1476 struct ofputil_async_cfg ac = ofconn_get_async_config(ofconn);
1477 uint32_t *masks = (ofconn->role == OFPCR12_ROLE_SLAVE
1480 return (masks[type] & (1u << reason)) != 0;
1483 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
1484 * packet rather than to send the packet to the controller.
1486 * This function returns true to indicate that a packet_in message
1487 * for a "table-miss" should be sent to at least one controller.
1488 * That is there is at least one controller with controller_id 0
1489 * which connected using an OpenFlow version earlier than OpenFlow1.3.
1493 * This logic assumes that "table-miss" packet_in messages
1494 * are always sent to controller_id 0. */
1496 connmgr_wants_packet_in_on_miss(struct connmgr *mgr) OVS_EXCLUDED(ofproto_mutex)
1498 struct ofconn *ofconn;
1500 ovs_mutex_lock(&ofproto_mutex);
1501 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1502 enum ofputil_protocol protocol = ofconn_get_protocol(ofconn);
1504 if (ofconn->controller_id == 0 &&
1505 (protocol == OFPUTIL_P_NONE ||
1506 ofputil_protocol_to_ofp_version(protocol) < OFP13_VERSION)) {
1507 ovs_mutex_unlock(&ofproto_mutex);
1511 ovs_mutex_unlock(&ofproto_mutex);
1516 /* Returns a human-readable name for an OpenFlow connection between 'mgr' and
1517 * 'target', suitable for use in log messages for identifying the connection.
1519 * The name is dynamically allocated. The caller should free it (with free())
1520 * when it is no longer needed. */
1522 ofconn_make_name(const struct connmgr *mgr, const char *target)
1524 return xasprintf("%s<->%s", mgr->name, target);
1528 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1532 for (i = 0; i < N_SCHEDULERS; i++) {
1533 struct pinsched **s = &ofconn->schedulers[i];
1537 *s = pinsched_create(rate, burst);
1539 pinsched_set_limits(*s, rate, burst);
1542 pinsched_destroy(*s);
1549 ofconn_send(const struct ofconn *ofconn, struct ofpbuf *msg,
1550 struct rconn_packet_counter *counter)
1552 ofpmsg_update_length(msg);
1553 rconn_send(ofconn->rconn, msg, counter);
1556 /* Sending asynchronous messages. */
1558 /* Sends an OFPT_PORT_STATUS message with 'opp' and 'reason' to appropriate
1559 * controllers managed by 'mgr'. For messages caused by a controller
1560 * OFPT_PORT_MOD, specify 'source' as the controller connection that sent the
1561 * request; otherwise, specify 'source' as NULL. */
1563 connmgr_send_port_status(struct connmgr *mgr, struct ofconn *source,
1564 const struct ofputil_phy_port *pp, uint8_t reason)
1566 /* XXX Should limit the number of queued port status change messages. */
1567 struct ofputil_port_status ps;
1568 struct ofconn *ofconn;
1572 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1573 if (ofconn_receives_async_msg(ofconn, OAM_PORT_STATUS, reason)) {
1576 /* Before 1.5, OpenFlow specified that OFPT_PORT_MOD should not
1577 * generate OFPT_PORT_STATUS messages. That requirement was a
1578 * relic of how OpenFlow originally supported a single controller,
1579 * so that one could expect the controller to already know the
1580 * changes it had made.
1582 * EXT-338 changes OpenFlow 1.5 OFPT_PORT_MOD to send
1583 * OFPT_PORT_STATUS messages to every controller. This is
1584 * obviously more useful in the multi-controller case. We could
1585 * always implement it that way in OVS, but that would risk
1586 * confusing controllers that are intended for single-controller
1587 * use only. (Imagine a controller that generates an OFPT_PORT_MOD
1588 * in response to any OFPT_PORT_STATUS!)
1590 * So this compromises: for OpenFlow 1.4 and earlier, it generates
1591 * OFPT_PORT_STATUS for OFPT_PORT_MOD, but not back to the
1592 * originating controller. In a single-controller environment, in
1593 * particular, this means that it will never generate
1594 * OFPT_PORT_STATUS for OFPT_PORT_MOD at all. */
1595 if (ofconn == source
1596 && rconn_get_version(ofconn->rconn) < OFP15_VERSION) {
1600 msg = ofputil_encode_port_status(&ps, ofconn_get_protocol(ofconn));
1601 ofconn_send(ofconn, msg, NULL);
1606 /* Sends an OFPT_REQUESTFORWARD message with 'request' and 'reason' to
1607 * appropriate controllers managed by 'mgr'. For messages caused by a
1608 * controller OFPT_GROUP_MOD and OFPT_METER_MOD, specify 'source' as the
1609 * controller connection that sent the request; otherwise, specify 'source'
1612 connmgr_send_requestforward(struct connmgr *mgr, const struct ofconn *source,
1613 const struct ofputil_requestforward *rf)
1615 struct ofconn *ofconn;
1617 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1618 if (ofconn_receives_async_msg(ofconn, OAM_REQUESTFORWARD, rf->reason)
1619 && rconn_get_version(ofconn->rconn) >= OFP14_VERSION
1620 && ofconn != source) {
1621 enum ofputil_protocol protocol = ofconn_get_protocol(ofconn);
1622 ofconn_send(ofconn, ofputil_encode_requestforward(rf, protocol),
1628 /* Sends an OFPT_FLOW_REMOVED or NXT_FLOW_REMOVED message based on 'fr' to
1629 * appropriate controllers managed by 'mgr'. */
1631 connmgr_send_flow_removed(struct connmgr *mgr,
1632 const struct ofputil_flow_removed *fr)
1634 struct ofconn *ofconn;
1636 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1637 if (ofconn_receives_async_msg(ofconn, OAM_FLOW_REMOVED, fr->reason)) {
1640 /* Account flow expirations as replies to OpenFlow requests. That
1641 * works because preventing OpenFlow requests from being processed
1642 * also prevents new flows from being added (and expiring). (It
1643 * also prevents processing OpenFlow requests that would not add
1644 * new flows, so it is imperfect.) */
1645 msg = ofputil_encode_flow_removed(fr, ofconn_get_protocol(ofconn));
1646 ofconn_send_reply(ofconn, msg);
1651 /* Given 'pin', sends an OFPT_PACKET_IN message to each OpenFlow controller as
1652 * necessary according to their individual configurations. */
1654 connmgr_send_async_msg(struct connmgr *mgr,
1655 const struct ofproto_async_msg *am)
1657 struct ofconn *ofconn;
1659 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1660 enum ofputil_protocol protocol = ofconn_get_protocol(ofconn);
1661 if (protocol == OFPUTIL_P_NONE || !rconn_is_connected(ofconn->rconn)
1662 || ofconn->controller_id != am->controller_id
1663 || !ofconn_receives_async_msg(ofconn, am->oam,
1664 am->pin.up.public.reason)) {
1668 struct ofpbuf *msg = ofputil_encode_packet_in_private(
1669 &am->pin.up, protocol, ofconn->packet_in_format,
1670 am->pin.max_len >= 0 ? am->pin.max_len : ofconn->miss_send_len,
1673 struct ovs_list txq;
1674 bool is_miss = (am->pin.up.public.reason == OFPR_NO_MATCH ||
1675 am->pin.up.public.reason == OFPR_EXPLICIT_MISS ||
1676 am->pin.up.public.reason == OFPR_IMPLICIT_MISS);
1677 pinsched_send(ofconn->schedulers[is_miss],
1678 am->pin.up.public.flow_metadata.flow.in_port.ofp_port,
1680 do_send_packet_ins(ofconn, &txq);
1685 do_send_packet_ins(struct ofconn *ofconn, struct ovs_list *txq)
1689 LIST_FOR_EACH_POP (pin, list_node, txq) {
1690 if (rconn_send_with_limit(ofconn->rconn, pin,
1691 ofconn->packet_in_counter, 100) == EAGAIN) {
1692 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5);
1694 VLOG_INFO_RL(&rl, "%s: dropping packet-in due to queue overflow",
1695 rconn_get_name(ofconn->rconn));
1700 /* Fail-open settings. */
1702 /* Returns the failure handling mode (OFPROTO_FAIL_SECURE or
1703 * OFPROTO_FAIL_STANDALONE) for 'mgr'. */
1704 enum ofproto_fail_mode
1705 connmgr_get_fail_mode(const struct connmgr *mgr)
1707 return mgr->fail_mode;
1710 /* Sets the failure handling mode for 'mgr' to 'fail_mode' (either
1711 * OFPROTO_FAIL_SECURE or OFPROTO_FAIL_STANDALONE). */
1713 connmgr_set_fail_mode(struct connmgr *mgr, enum ofproto_fail_mode fail_mode)
1715 if (mgr->fail_mode != fail_mode) {
1716 mgr->fail_mode = fail_mode;
1717 update_fail_open(mgr);
1718 if (!connmgr_has_controllers(mgr)) {
1719 ofproto_flush_flows(mgr->ofproto);
1724 /* Fail-open implementation. */
1726 /* Returns the longest probe interval among the primary controllers configured
1727 * on 'mgr'. Returns 0 if there are no primary controllers. */
1729 connmgr_get_max_probe_interval(const struct connmgr *mgr)
1731 const struct ofconn *ofconn;
1732 int max_probe_interval;
1734 max_probe_interval = 0;
1735 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1736 int probe_interval = rconn_get_probe_interval(ofconn->rconn);
1737 max_probe_interval = MAX(max_probe_interval, probe_interval);
1739 return max_probe_interval;
1742 /* Returns the number of seconds for which all of 'mgr's primary controllers
1743 * have been disconnected. Returns 0 if 'mgr' has no primary controllers. */
1745 connmgr_failure_duration(const struct connmgr *mgr)
1747 const struct ofconn *ofconn;
1748 int min_failure_duration;
1750 if (!connmgr_has_controllers(mgr)) {
1754 min_failure_duration = INT_MAX;
1755 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1756 int failure_duration = rconn_failure_duration(ofconn->rconn);
1757 min_failure_duration = MIN(min_failure_duration, failure_duration);
1759 return min_failure_duration;
1762 /* Returns true if at least one primary controller is connected (regardless of
1763 * whether those controllers are believed to have authenticated and accepted
1764 * this switch), false if none of them are connected. */
1766 connmgr_is_any_controller_connected(const struct connmgr *mgr)
1768 const struct ofconn *ofconn;
1770 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1771 if (rconn_is_connected(ofconn->rconn)) {
1778 /* Returns true if at least one primary controller is believed to have
1779 * authenticated and accepted this switch, false otherwise. */
1781 connmgr_is_any_controller_admitted(const struct connmgr *mgr)
1783 const struct ofconn *ofconn;
1785 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1786 if (rconn_is_admitted(ofconn->rconn)) {
1793 /* In-band configuration. */
1795 static bool any_extras_changed(const struct connmgr *,
1796 const struct sockaddr_in *extras, size_t n);
1798 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'mgr''s
1799 * in-band control should guarantee access, in the same way that in-band
1800 * control guarantees access to OpenFlow controllers. */
1802 connmgr_set_extra_in_band_remotes(struct connmgr *mgr,
1803 const struct sockaddr_in *extras, size_t n)
1805 if (!any_extras_changed(mgr, extras, n)) {
1809 free(mgr->extra_in_band_remotes);
1810 mgr->n_extra_remotes = n;
1811 mgr->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
1813 update_in_band_remotes(mgr);
1816 /* Sets the OpenFlow queue used by flows set up by in-band control on
1817 * 'mgr' to 'queue_id'. If 'queue_id' is negative, then in-band control
1818 * flows will use the default queue. */
1820 connmgr_set_in_band_queue(struct connmgr *mgr, int queue_id)
1822 if (queue_id != mgr->in_band_queue) {
1823 mgr->in_band_queue = queue_id;
1824 update_in_band_remotes(mgr);
1829 any_extras_changed(const struct connmgr *mgr,
1830 const struct sockaddr_in *extras, size_t n)
1834 if (n != mgr->n_extra_remotes) {
1838 for (i = 0; i < n; i++) {
1839 const struct sockaddr_in *old = &mgr->extra_in_band_remotes[i];
1840 const struct sockaddr_in *new = &extras[i];
1842 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
1843 old->sin_port != new->sin_port) {
1851 /* In-band implementation. */
1854 connmgr_has_in_band(struct connmgr *mgr)
1856 return mgr->in_band != NULL;
1859 /* Fail-open and in-band implementation. */
1861 /* Called by 'ofproto' after all flows have been flushed, to allow fail-open
1862 * and standalone mode to re-create their flows.
1864 * In-band control has more sophisticated code that manages flows itself. */
1866 connmgr_flushed(struct connmgr *mgr)
1867 OVS_EXCLUDED(ofproto_mutex)
1869 if (mgr->fail_open) {
1870 fail_open_flushed(mgr->fail_open);
1873 /* If there are no controllers and we're in standalone mode, set up a flow
1874 * that matches every packet and directs them to OFPP_NORMAL (which goes to
1875 * us). Otherwise, the switch is in secure mode and we won't pass any
1876 * traffic until a controller has been defined and it tells us to do so. */
1877 if (!connmgr_has_controllers(mgr)
1878 && mgr->fail_mode == OFPROTO_FAIL_STANDALONE) {
1879 struct ofpbuf ofpacts;
1882 ofpbuf_init(&ofpacts, OFPACT_OUTPUT_SIZE);
1883 ofpact_put_OUTPUT(&ofpacts)->port = OFPP_NORMAL;
1885 match_init_catchall(&match);
1886 ofproto_add_flow(mgr->ofproto, &match, 0, ofpacts.data,
1889 ofpbuf_uninit(&ofpacts);
1893 /* Returns the number of hidden rules created by the in-band and fail-open
1894 * implementations in table 0. (Subtracting this count from the number of
1895 * rules in the table 0 classifier, as maintained in struct oftable, yields
1896 * the number of flows that OVS should report via OpenFlow for table 0.) */
1898 connmgr_count_hidden_rules(const struct connmgr *mgr)
1902 n_hidden += in_band_count_rules(mgr->in_band);
1904 if (mgr->fail_open) {
1905 n_hidden += fail_open_count_rules(mgr->fail_open);
1910 /* Creates a new ofservice for 'target' in 'mgr'. Returns 0 if successful,
1911 * otherwise a positive errno value.
1913 * ofservice_reconfigure() must be called to fully configure the new
1916 ofservice_create(struct connmgr *mgr, const char *target,
1917 uint32_t allowed_versions, uint8_t dscp)
1919 struct ofservice *ofservice;
1920 struct pvconn *pvconn;
1923 error = pvconn_open(target, allowed_versions, dscp, &pvconn);
1928 ofservice = xzalloc(sizeof *ofservice);
1929 hmap_insert(&mgr->services, &ofservice->node, hash_string(target, 0));
1930 ofservice->pvconn = pvconn;
1931 ofservice->allowed_versions = allowed_versions;
1937 ofservice_destroy(struct connmgr *mgr, struct ofservice *ofservice)
1939 hmap_remove(&mgr->services, &ofservice->node);
1940 pvconn_close(ofservice->pvconn);
1945 ofservice_reconfigure(struct ofservice *ofservice,
1946 const struct ofproto_controller *c)
1948 ofservice->probe_interval = c->probe_interval;
1949 ofservice->rate_limit = c->rate_limit;
1950 ofservice->burst_limit = c->burst_limit;
1951 ofservice->enable_async_msgs = c->enable_async_msgs;
1952 ofservice->dscp = c->dscp;
1955 /* Finds and returns the ofservice within 'mgr' that has the given
1956 * 'target', or a null pointer if none exists. */
1957 static struct ofservice *
1958 ofservice_lookup(struct connmgr *mgr, const char *target)
1960 struct ofservice *ofservice;
1962 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1964 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1971 /* Flow monitors (NXST_FLOW_MONITOR). */
1973 /* A counter incremented when something significant happens to an OpenFlow
1976 * - When a rule is added, its 'add_seqno' and 'modify_seqno' are set to
1977 * the current value (which is then incremented).
1979 * - When a rule is modified, its 'modify_seqno' is set to the current
1980 * value (which is then incremented).
1982 * Thus, by comparing an old value of monitor_seqno against a rule's
1983 * 'add_seqno', one can tell whether the rule was added before or after the old
1984 * value was read, and similarly for 'modify_seqno'.
1986 * 32 bits should normally be sufficient (and would be nice, to save space in
1987 * each rule) but then we'd have to have some special cases for wraparound.
1989 * We initialize monitor_seqno to 1 to allow 0 to be used as an invalid
1991 static uint64_t monitor_seqno = 1;
1993 COVERAGE_DEFINE(ofmonitor_pause);
1994 COVERAGE_DEFINE(ofmonitor_resume);
1997 ofmonitor_create(const struct ofputil_flow_monitor_request *request,
1998 struct ofconn *ofconn, struct ofmonitor **monitorp)
1999 OVS_REQUIRES(ofproto_mutex)
2001 struct ofmonitor *m;
2005 m = ofmonitor_lookup(ofconn, request->id);
2007 return OFPERR_OFPMOFC_MONITOR_EXISTS;
2010 m = xmalloc(sizeof *m);
2012 hmap_insert(&ofconn->monitors, &m->ofconn_node, hash_int(request->id, 0));
2013 m->id = request->id;
2014 m->flags = request->flags;
2015 m->out_port = request->out_port;
2016 m->table_id = request->table_id;
2017 minimatch_init(&m->match, &request->match);
2024 ofmonitor_lookup(struct ofconn *ofconn, uint32_t id)
2025 OVS_REQUIRES(ofproto_mutex)
2027 struct ofmonitor *m;
2029 HMAP_FOR_EACH_IN_BUCKET (m, ofconn_node, hash_int(id, 0),
2030 &ofconn->monitors) {
2039 ofmonitor_destroy(struct ofmonitor *m)
2040 OVS_REQUIRES(ofproto_mutex)
2043 minimatch_destroy(&m->match);
2044 hmap_remove(&m->ofconn->monitors, &m->ofconn_node);
2050 ofmonitor_report(struct connmgr *mgr, struct rule *rule,
2051 enum nx_flow_update_event event,
2052 enum ofp_flow_removed_reason reason,
2053 const struct ofconn *abbrev_ofconn, ovs_be32 abbrev_xid,
2054 const struct rule_actions *old_actions)
2055 OVS_REQUIRES(ofproto_mutex)
2057 enum nx_flow_monitor_flags update;
2058 struct ofconn *ofconn;
2060 if (rule_is_hidden(rule)) {
2067 rule->add_seqno = rule->modify_seqno = monitor_seqno++;
2071 update = NXFMF_DELETE;
2074 case NXFME_MODIFIED:
2075 update = NXFMF_MODIFY;
2076 rule->modify_seqno = monitor_seqno++;
2084 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2085 enum nx_flow_monitor_flags flags = 0;
2086 struct ofmonitor *m;
2088 if (ofconn->monitor_paused) {
2089 /* Only send NXFME_DELETED notifications for flows that were added
2090 * before we paused. */
2091 if (event != NXFME_DELETED
2092 || rule->add_seqno > ofconn->monitor_paused) {
2097 HMAP_FOR_EACH (m, ofconn_node, &ofconn->monitors) {
2098 if (m->flags & update
2099 && (m->table_id == 0xff || m->table_id == rule->table_id)
2100 && (ofproto_rule_has_out_port(rule, m->out_port)
2102 && ofpacts_output_to_port(old_actions->ofpacts,
2103 old_actions->ofpacts_len,
2105 && cls_rule_is_loose_match(&rule->cr, &m->match)) {
2111 if (list_is_empty(&ofconn->updates)) {
2112 ofputil_start_flow_update(&ofconn->updates);
2113 ofconn->sent_abbrev_update = false;
2116 if (flags & NXFMF_OWN || ofconn != abbrev_ofconn
2117 || ofconn->monitor_paused) {
2118 struct ofputil_flow_update fu;
2122 fu.reason = event == NXFME_DELETED ? reason : 0;
2123 fu.table_id = rule->table_id;
2124 fu.cookie = rule->flow_cookie;
2125 minimatch_expand(&rule->cr.match, &match);
2127 fu.priority = rule->cr.priority;
2129 ovs_mutex_lock(&rule->mutex);
2130 fu.idle_timeout = rule->idle_timeout;
2131 fu.hard_timeout = rule->hard_timeout;
2132 ovs_mutex_unlock(&rule->mutex);
2134 if (flags & NXFMF_ACTIONS) {
2135 const struct rule_actions *actions = rule_get_actions(rule);
2136 fu.ofpacts = actions->ofpacts;
2137 fu.ofpacts_len = actions->ofpacts_len;
2142 ofputil_append_flow_update(&fu, &ofconn->updates);
2143 } else if (!ofconn->sent_abbrev_update) {
2144 struct ofputil_flow_update fu;
2146 fu.event = NXFME_ABBREV;
2147 fu.xid = abbrev_xid;
2148 ofputil_append_flow_update(&fu, &ofconn->updates);
2150 ofconn->sent_abbrev_update = true;
2157 ofmonitor_flush(struct connmgr *mgr)
2158 OVS_REQUIRES(ofproto_mutex)
2160 struct ofconn *ofconn;
2162 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2165 LIST_FOR_EACH_POP (msg, list_node, &ofconn->updates) {
2166 unsigned int n_bytes;
2168 ofconn_send(ofconn, msg, ofconn->monitor_counter);
2169 n_bytes = rconn_packet_counter_n_bytes(ofconn->monitor_counter);
2170 if (!ofconn->monitor_paused && n_bytes > 128 * 1024) {
2171 struct ofpbuf *pause;
2173 COVERAGE_INC(ofmonitor_pause);
2174 ofconn->monitor_paused = monitor_seqno++;
2175 pause = ofpraw_alloc_xid(OFPRAW_NXT_FLOW_MONITOR_PAUSED,
2176 OFP10_VERSION, htonl(0), 0);
2177 ofconn_send(ofconn, pause, ofconn->monitor_counter);
2184 ofmonitor_resume(struct ofconn *ofconn)
2185 OVS_REQUIRES(ofproto_mutex)
2187 struct rule_collection rules;
2188 struct ofpbuf *resumed;
2189 struct ofmonitor *m;
2190 struct ovs_list msgs;
2192 rule_collection_init(&rules);
2193 HMAP_FOR_EACH (m, ofconn_node, &ofconn->monitors) {
2194 ofmonitor_collect_resume_rules(m, ofconn->monitor_paused, &rules);
2198 ofmonitor_compose_refresh_updates(&rules, &msgs);
2200 resumed = ofpraw_alloc_xid(OFPRAW_NXT_FLOW_MONITOR_RESUMED, OFP10_VERSION,
2202 list_push_back(&msgs, &resumed->list_node);
2203 ofconn_send_replies(ofconn, &msgs);
2205 ofconn->monitor_paused = 0;
2209 ofmonitor_may_resume(const struct ofconn *ofconn)
2210 OVS_REQUIRES(ofproto_mutex)
2212 return (ofconn->monitor_paused != 0
2213 && !rconn_packet_counter_n_packets(ofconn->monitor_counter));
2217 ofmonitor_run(struct connmgr *mgr)
2219 struct ofconn *ofconn;
2221 ovs_mutex_lock(&ofproto_mutex);
2222 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2223 if (ofmonitor_may_resume(ofconn)) {
2224 COVERAGE_INC(ofmonitor_resume);
2225 ofmonitor_resume(ofconn);
2228 ovs_mutex_unlock(&ofproto_mutex);
2232 ofmonitor_wait(struct connmgr *mgr)
2234 struct ofconn *ofconn;
2236 ovs_mutex_lock(&ofproto_mutex);
2237 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2238 if (ofmonitor_may_resume(ofconn)) {
2239 poll_immediate_wake();
2242 ovs_mutex_unlock(&ofproto_mutex);
2246 ofproto_async_msg_free(struct ofproto_async_msg *am)
2248 free(am->pin.up.public.packet);
2249 free(am->pin.up.public.userdata);
2250 free(am->pin.up.stack);
2251 free(am->pin.up.actions);
2252 free(am->pin.up.action_set);