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.
24 #include "fail-open.h"
27 #include "ofproto-provider.h"
28 #include "openvswitch/dynamic-string.h"
29 #include "openvswitch/ofp-actions.h"
30 #include "openvswitch/ofp-msgs.h"
31 #include "openvswitch/ofp-util.h"
32 #include "openvswitch/ofpbuf.h"
33 #include "openvswitch/vconn.h"
34 #include "openvswitch/vlog.h"
36 #include "poll-loop.h"
39 #include "openvswitch/shash.h"
45 VLOG_DEFINE_THIS_MODULE(connmgr);
46 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
48 /* An OpenFlow connection.
54 * 'ofproto_mutex' must be held whenever an ofconn is created or destroyed or,
55 * more or less equivalently, whenever an ofconn is added to or removed from a
56 * connmgr. 'ofproto_mutex' doesn't protect the data inside the ofconn, except
57 * as specifically noted below. */
59 /* Configuration that persists from one connection to the next. */
61 struct ovs_list node; /* In struct connmgr's "all_conns" list. */
62 struct hmap_node hmap_node; /* In struct connmgr's "controllers" map. */
64 struct connmgr *connmgr; /* Connection's manager. */
65 struct rconn *rconn; /* OpenFlow connection. */
66 enum ofconn_type type; /* Type. */
67 enum ofproto_band band; /* In-band or out-of-band? */
68 bool enable_async_msgs; /* Initially enable async messages? */
70 /* State that should be cleared from one connection to the next. */
73 enum ofp12_controller_role role; /* Role. */
74 enum ofputil_protocol protocol; /* Current protocol variant. */
75 enum nx_packet_in_format packet_in_format; /* OFPT_PACKET_IN format. */
77 /* OFPT_PACKET_IN related data. */
78 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
79 #define N_SCHEDULERS 2
80 struct pinsched *schedulers[N_SCHEDULERS];
81 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
82 int miss_send_len; /* Bytes to send of buffered packets. */
83 uint16_t controller_id; /* Connection controller ID. */
85 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
86 * requests, and the maximum number before we stop reading OpenFlow
88 #define OFCONN_REPLY_MAX 100
89 struct rconn_packet_counter *reply_counter;
91 /* Asynchronous message configuration in each possible role.
93 * A 1-bit enables sending an asynchronous message for one possible reason
94 * that the message might be generated, a 0-bit disables it. */
95 struct ofputil_async_cfg *async_cfg;
97 /* Flow table operation logging. */
98 int n_add, n_delete, n_modify; /* Number of unreported ops of each kind. */
99 long long int first_op, last_op; /* Range of times for unreported ops. */
100 long long int next_op_report; /* Time to report ops, or LLONG_MAX. */
101 long long int op_backoff; /* Earliest time to report ops again. */
103 /* Flow monitors (e.g. NXST_FLOW_MONITOR). */
105 /* Configuration. Contains "struct ofmonitor"s. */
106 struct hmap monitors OVS_GUARDED_BY(ofproto_mutex);
110 * When too many flow monitor notifications back up in the transmit buffer,
111 * we pause the transmission of further notifications. These members track
112 * the flow control state.
114 * When notifications are flowing, 'monitor_paused' is 0. When
115 * notifications are paused, 'monitor_paused' is the value of
116 * 'monitor_seqno' at the point we paused.
118 * 'monitor_counter' counts the OpenFlow messages and bytes currently in
119 * flight. This value growing too large triggers pausing. */
120 uint64_t monitor_paused OVS_GUARDED_BY(ofproto_mutex);
121 struct rconn_packet_counter *monitor_counter OVS_GUARDED_BY(ofproto_mutex);
123 /* State of monitors for a single ongoing flow_mod.
125 * 'updates' is a list of "struct ofpbuf"s that contain
126 * NXST_FLOW_MONITOR_REPLY messages representing the changes made by the
129 * When 'updates' is nonempty, 'sent_abbrev_update' is true if 'updates'
130 * contains an update event of type NXFME_ABBREV and false otherwise.. */
131 struct ovs_list updates OVS_GUARDED_BY(ofproto_mutex);
132 bool sent_abbrev_update OVS_GUARDED_BY(ofproto_mutex);
134 /* Active bundles. Contains "struct ofp_bundle"s. */
138 static struct ofconn *ofconn_create(struct connmgr *, struct rconn *,
139 enum ofconn_type, bool enable_async_msgs)
140 OVS_REQUIRES(ofproto_mutex);
141 static void ofconn_destroy(struct ofconn *) OVS_REQUIRES(ofproto_mutex);
142 static void ofconn_flush(struct ofconn *) OVS_REQUIRES(ofproto_mutex);
144 static void ofconn_reconfigure(struct ofconn *,
145 const struct ofproto_controller *);
147 static void ofconn_run(struct ofconn *,
148 void (*handle_openflow)(struct ofconn *,
149 const struct ofpbuf *ofp_msg));
150 static void ofconn_wait(struct ofconn *);
152 static void ofconn_log_flow_mods(struct ofconn *);
154 static const char *ofconn_get_target(const struct ofconn *);
155 static char *ofconn_make_name(const struct connmgr *, const char *target);
157 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
159 static void ofconn_send(const struct ofconn *, struct ofpbuf *,
160 struct rconn_packet_counter *);
162 static void do_send_packet_ins(struct ofconn *, struct ovs_list *txq);
164 /* A listener for incoming OpenFlow "service" connections. */
166 struct hmap_node node; /* In struct connmgr's "services" hmap. */
167 struct pvconn *pvconn; /* OpenFlow connection listener. */
169 /* These are not used by ofservice directly. They are settings for
170 * accepted "struct ofconn"s from the pvconn. */
171 int probe_interval; /* Max idle time before probing, in seconds. */
172 int rate_limit; /* Max packet-in rate in packets per second. */
173 int burst_limit; /* Limit on accumulating packet credits. */
174 bool enable_async_msgs; /* Initially enable async messages? */
175 uint8_t dscp; /* DSCP Value for controller connection */
176 uint32_t allowed_versions; /* OpenFlow protocol versions that may
177 * be negotiated for a session. */
180 static void ofservice_reconfigure(struct ofservice *,
181 const struct ofproto_controller *);
182 static int ofservice_create(struct connmgr *mgr, const char *target,
183 uint32_t allowed_versions, uint8_t dscp);
184 static void ofservice_destroy(struct connmgr *, struct ofservice *);
185 static struct ofservice *ofservice_lookup(struct connmgr *,
188 /* Connection manager for an OpenFlow switch. */
190 struct ofproto *ofproto;
192 char *local_port_name;
194 /* OpenFlow connections. */
195 struct hmap controllers; /* All OFCONN_PRIMARY controllers. */
196 struct ovs_list all_conns; /* All controllers. */
197 uint64_t master_election_id; /* monotonically increasing sequence number
198 * for master election */
199 bool master_election_id_defined;
201 /* OpenFlow listeners. */
202 struct hmap services; /* Contains "struct ofservice"s. */
203 struct pvconn **snoops;
207 struct fail_open *fail_open;
208 enum ofproto_fail_mode fail_mode;
210 /* In-band control. */
211 struct in_band *in_band;
212 struct sockaddr_in *extra_in_band_remotes;
213 size_t n_extra_remotes;
217 static void update_in_band_remotes(struct connmgr *);
218 static void add_snooper(struct connmgr *, struct vconn *);
219 static void ofmonitor_run(struct connmgr *);
220 static void ofmonitor_wait(struct connmgr *);
222 /* Creates and returns a new connection manager owned by 'ofproto'. 'name' is
223 * a name for the ofproto suitable for using in log messages.
224 * 'local_port_name' is the name of the local port (OFPP_LOCAL) within
227 connmgr_create(struct ofproto *ofproto,
228 const char *name, const char *local_port_name)
232 mgr = xmalloc(sizeof *mgr);
233 mgr->ofproto = ofproto;
234 mgr->name = xstrdup(name);
235 mgr->local_port_name = xstrdup(local_port_name);
237 hmap_init(&mgr->controllers);
238 ovs_list_init(&mgr->all_conns);
239 mgr->master_election_id = 0;
240 mgr->master_election_id_defined = false;
242 hmap_init(&mgr->services);
246 mgr->fail_open = NULL;
247 mgr->fail_mode = OFPROTO_FAIL_SECURE;
250 mgr->extra_in_band_remotes = NULL;
251 mgr->n_extra_remotes = 0;
252 mgr->in_band_queue = -1;
257 /* Frees 'mgr' and all of its resources. */
259 connmgr_destroy(struct connmgr *mgr)
261 struct ofservice *ofservice, *next_ofservice;
262 struct ofconn *ofconn, *next_ofconn;
269 ovs_mutex_lock(&ofproto_mutex);
270 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &mgr->all_conns) {
271 ofconn_destroy(ofconn);
273 ovs_mutex_unlock(&ofproto_mutex);
275 hmap_destroy(&mgr->controllers);
277 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &mgr->services) {
278 ofservice_destroy(mgr, ofservice);
280 hmap_destroy(&mgr->services);
282 for (i = 0; i < mgr->n_snoops; i++) {
283 pvconn_close(mgr->snoops[i]);
287 fail_open_destroy(mgr->fail_open);
288 mgr->fail_open = NULL;
290 in_band_destroy(mgr->in_band);
292 free(mgr->extra_in_band_remotes);
294 free(mgr->local_port_name);
299 /* Does all of the periodic maintenance required by 'mgr'. Calls
300 * 'handle_openflow' for each message received on an OpenFlow connection,
301 * passing along the OpenFlow connection itself and the message that was sent.
302 * 'handle_openflow' must not modify or free the message. */
304 connmgr_run(struct connmgr *mgr,
305 void (*handle_openflow)(struct ofconn *,
306 const struct ofpbuf *ofp_msg))
307 OVS_EXCLUDED(ofproto_mutex)
309 struct ofconn *ofconn, *next_ofconn;
310 struct ofservice *ofservice;
314 if (!in_band_run(mgr->in_band)) {
315 in_band_destroy(mgr->in_band);
320 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &mgr->all_conns) {
321 ofconn_run(ofconn, handle_openflow);
325 /* Fail-open maintenance. Do this after processing the ofconns since
326 * fail-open checks the status of the controller rconn. */
327 if (mgr->fail_open) {
328 fail_open_run(mgr->fail_open);
331 HMAP_FOR_EACH (ofservice, node, &mgr->services) {
335 retval = pvconn_accept(ofservice->pvconn, &vconn);
340 /* Passing default value for creation of the rconn */
341 rconn = rconn_create(ofservice->probe_interval, 0, ofservice->dscp,
342 vconn_get_allowed_versions(vconn));
343 name = ofconn_make_name(mgr, vconn_get_name(vconn));
344 rconn_connect_unreliably(rconn, vconn, name);
347 ovs_mutex_lock(&ofproto_mutex);
348 ofconn = ofconn_create(mgr, rconn, OFCONN_SERVICE,
349 ofservice->enable_async_msgs);
350 ovs_mutex_unlock(&ofproto_mutex);
352 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
353 ofservice->burst_limit);
354 } else if (retval != EAGAIN) {
355 VLOG_WARN_RL(&rl, "accept failed (%s)", ovs_strerror(retval));
359 for (i = 0; i < mgr->n_snoops; i++) {
363 retval = pvconn_accept(mgr->snoops[i], &vconn);
365 add_snooper(mgr, vconn);
366 } else if (retval != EAGAIN) {
367 VLOG_WARN_RL(&rl, "accept failed (%s)", ovs_strerror(retval));
372 /* Causes the poll loop to wake up when connmgr_run() needs to run. */
374 connmgr_wait(struct connmgr *mgr)
376 struct ofservice *ofservice;
377 struct ofconn *ofconn;
380 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
385 in_band_wait(mgr->in_band);
387 if (mgr->fail_open) {
388 fail_open_wait(mgr->fail_open);
390 HMAP_FOR_EACH (ofservice, node, &mgr->services) {
391 pvconn_wait(ofservice->pvconn);
393 for (i = 0; i < mgr->n_snoops; i++) {
394 pvconn_wait(mgr->snoops[i]);
398 /* Adds some memory usage statistics for 'mgr' into 'usage', for use with
399 * memory_report(). */
401 connmgr_get_memory_usage(const struct connmgr *mgr, struct simap *usage)
403 const struct ofconn *ofconn;
404 unsigned int packets = 0;
405 unsigned int ofconns = 0;
407 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
412 packets += rconn_count_txqlen(ofconn->rconn);
413 for (i = 0; i < N_SCHEDULERS; i++) {
414 struct pinsched_stats stats;
416 pinsched_get_stats(ofconn->schedulers[i], &stats);
417 packets += stats.n_queued;
419 packets += pktbuf_count_packets(ofconn->pktbuf);
421 simap_increase(usage, "ofconns", ofconns);
422 simap_increase(usage, "packets", packets);
425 /* Returns the ofproto that owns 'ofconn''s connmgr. */
427 ofconn_get_ofproto(const struct ofconn *ofconn)
429 return ofconn->connmgr->ofproto;
432 /* OpenFlow configuration. */
434 static void add_controller(struct connmgr *, const char *target, uint8_t dscp,
435 uint32_t allowed_versions)
436 OVS_REQUIRES(ofproto_mutex);
437 static struct ofconn *find_controller_by_target(struct connmgr *,
439 static void update_fail_open(struct connmgr *) OVS_EXCLUDED(ofproto_mutex);
440 static int set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
441 const struct sset *);
443 /* Returns true if 'mgr' has any configured primary controllers.
445 * Service controllers do not count, but configured primary controllers do
446 * count whether or not they are currently connected. */
448 connmgr_has_controllers(const struct connmgr *mgr)
450 return !hmap_is_empty(&mgr->controllers);
453 /* Initializes 'info' and populates it with information about each configured
454 * primary controller. The keys in 'info' are the controllers' targets; the
455 * data values are corresponding "struct ofproto_controller_info".
457 * The caller owns 'info' and everything in it and should free it when it is no
460 connmgr_get_controller_info(struct connmgr *mgr, struct shash *info)
462 const struct ofconn *ofconn;
464 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
465 const struct rconn *rconn = ofconn->rconn;
466 const char *target = rconn_get_target(rconn);
468 if (!shash_find(info, target)) {
469 struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo);
470 time_t now = time_now();
471 time_t last_connection = rconn_get_last_connection(rconn);
472 time_t last_disconnect = rconn_get_last_disconnect(rconn);
473 int last_error = rconn_get_last_error(rconn);
476 shash_add(info, target, cinfo);
478 cinfo->is_connected = rconn_is_connected(rconn);
479 cinfo->role = ofconn->role;
481 smap_init(&cinfo->pairs);
483 smap_add(&cinfo->pairs, "last_error",
484 ovs_retval_to_string(last_error));
487 smap_add(&cinfo->pairs, "state", rconn_get_state(rconn));
489 if (last_connection != TIME_MIN) {
490 smap_add_format(&cinfo->pairs, "sec_since_connect",
491 "%ld", (long int) (now - last_connection));
494 if (last_disconnect != TIME_MIN) {
495 smap_add_format(&cinfo->pairs, "sec_since_disconnect",
496 "%ld", (long int) (now - last_disconnect));
499 for (i = 0; i < N_SCHEDULERS; i++) {
500 if (ofconn->schedulers[i]) {
501 const char *name = i ? "miss" : "action";
502 struct pinsched_stats stats;
504 pinsched_get_stats(ofconn->schedulers[i], &stats);
505 smap_add_nocopy(&cinfo->pairs,
506 xasprintf("packet-in-%s-backlog", name),
507 xasprintf("%u", stats.n_queued));
508 smap_add_nocopy(&cinfo->pairs,
509 xasprintf("packet-in-%s-bypassed", name),
510 xasprintf("%llu", stats.n_normal));
511 smap_add_nocopy(&cinfo->pairs,
512 xasprintf("packet-in-%s-queued", name),
513 xasprintf("%llu", stats.n_limited));
514 smap_add_nocopy(&cinfo->pairs,
515 xasprintf("packet-in-%s-dropped", name),
516 xasprintf("%llu", stats.n_queue_dropped));
524 connmgr_free_controller_info(struct shash *info)
526 struct shash_node *node;
528 SHASH_FOR_EACH (node, info) {
529 struct ofproto_controller_info *cinfo = node->data;
530 smap_destroy(&cinfo->pairs);
536 /* Changes 'mgr''s set of controllers to the 'n_controllers' controllers in
539 connmgr_set_controllers(struct connmgr *mgr,
540 const struct ofproto_controller *controllers,
541 size_t n_controllers, uint32_t allowed_versions)
542 OVS_EXCLUDED(ofproto_mutex)
544 bool had_controllers = connmgr_has_controllers(mgr);
545 struct shash new_controllers;
546 struct ofconn *ofconn, *next_ofconn;
547 struct ofservice *ofservice, *next_ofservice;
550 /* Required to add and remove ofconns. This could probably be narrowed to
551 * cover a smaller amount of code, if that yielded some benefit. */
552 ovs_mutex_lock(&ofproto_mutex);
554 /* Create newly configured controllers and services.
555 * Create a name to ofproto_controller mapping in 'new_controllers'. */
556 shash_init(&new_controllers);
557 for (i = 0; i < n_controllers; i++) {
558 const struct ofproto_controller *c = &controllers[i];
560 if (!vconn_verify_name(c->target)) {
562 ofconn = find_controller_by_target(mgr, c->target);
564 VLOG_INFO("%s: added primary controller \"%s\"",
565 mgr->name, c->target);
567 } else if (rconn_get_allowed_versions(ofconn->rconn) !=
569 VLOG_INFO("%s: re-added primary controller \"%s\"",
570 mgr->name, c->target);
572 ofconn_destroy(ofconn);
575 add_controller(mgr, c->target, c->dscp, allowed_versions);
577 } else if (!pvconn_verify_name(c->target)) {
579 ofservice = ofservice_lookup(mgr, c->target);
581 VLOG_INFO("%s: added service controller \"%s\"",
582 mgr->name, c->target);
584 } else if (ofservice->allowed_versions != allowed_versions) {
585 VLOG_INFO("%s: re-added service controller \"%s\"",
586 mgr->name, c->target);
587 ofservice_destroy(mgr, ofservice);
591 ofservice_create(mgr, c->target, allowed_versions, c->dscp);
594 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
595 mgr->name, c->target);
599 shash_add_once(&new_controllers, c->target, &controllers[i]);
602 /* Delete controllers that are no longer configured.
603 * Update configuration of all now-existing controllers. */
604 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &mgr->controllers) {
605 const char *target = ofconn_get_target(ofconn);
606 struct ofproto_controller *c;
608 c = shash_find_data(&new_controllers, target);
610 VLOG_INFO("%s: removed primary controller \"%s\"",
612 ofconn_destroy(ofconn);
614 ofconn_reconfigure(ofconn, c);
618 /* Delete services that are no longer configured.
619 * Update configuration of all now-existing services. */
620 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &mgr->services) {
621 const char *target = pvconn_get_name(ofservice->pvconn);
622 struct ofproto_controller *c;
624 c = shash_find_data(&new_controllers, target);
626 VLOG_INFO("%s: removed service controller \"%s\"",
628 ofservice_destroy(mgr, ofservice);
630 ofservice_reconfigure(ofservice, c);
634 shash_destroy(&new_controllers);
636 ovs_mutex_unlock(&ofproto_mutex);
638 update_in_band_remotes(mgr);
639 update_fail_open(mgr);
640 if (had_controllers != connmgr_has_controllers(mgr)) {
641 ofproto_flush_flows(mgr->ofproto);
645 /* Drops the connections between 'mgr' and all of its primary and secondary
646 * controllers, forcing them to reconnect. */
648 connmgr_reconnect(const struct connmgr *mgr)
650 struct ofconn *ofconn;
652 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
653 rconn_reconnect(ofconn->rconn);
657 /* Sets the "snoops" for 'mgr' to the pvconn targets listed in 'snoops'.
659 * A "snoop" is a pvconn to which every OpenFlow message to or from the most
660 * important controller on 'mgr' is mirrored. */
662 connmgr_set_snoops(struct connmgr *mgr, const struct sset *snoops)
664 return set_pvconns(&mgr->snoops, &mgr->n_snoops, snoops);
667 /* Adds each of the snoops currently configured on 'mgr' to 'snoops'. */
669 connmgr_get_snoops(const struct connmgr *mgr, struct sset *snoops)
673 for (i = 0; i < mgr->n_snoops; i++) {
674 sset_add(snoops, pvconn_get_name(mgr->snoops[i]));
678 /* Returns true if 'mgr' has at least one snoop, false if it has none. */
680 connmgr_has_snoops(const struct connmgr *mgr)
682 return mgr->n_snoops > 0;
685 /* Creates a new controller for 'target' in 'mgr'. update_controller() needs
686 * to be called later to finish the new ofconn's configuration. */
688 add_controller(struct connmgr *mgr, const char *target, uint8_t dscp,
689 uint32_t allowed_versions)
690 OVS_REQUIRES(ofproto_mutex)
692 char *name = ofconn_make_name(mgr, target);
693 struct ofconn *ofconn;
695 ofconn = ofconn_create(mgr, rconn_create(5, 8, dscp, allowed_versions),
696 OFCONN_PRIMARY, true);
697 ofconn->pktbuf = pktbuf_create();
698 rconn_connect(ofconn->rconn, target, name);
699 hmap_insert(&mgr->controllers, &ofconn->hmap_node, hash_string(target, 0));
704 static struct ofconn *
705 find_controller_by_target(struct connmgr *mgr, const char *target)
707 struct ofconn *ofconn;
709 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
710 hash_string(target, 0), &mgr->controllers) {
711 if (!strcmp(ofconn_get_target(ofconn), target)) {
719 update_in_band_remotes(struct connmgr *mgr)
721 struct sockaddr_in *addrs;
722 size_t max_addrs, n_addrs;
723 struct ofconn *ofconn;
726 /* Allocate enough memory for as many remotes as we could possibly have. */
727 max_addrs = mgr->n_extra_remotes + hmap_count(&mgr->controllers);
728 addrs = xmalloc(max_addrs * sizeof *addrs);
731 /* Add all the remotes. */
732 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
733 const char *target = rconn_get_target(ofconn->rconn);
735 struct sockaddr_storage ss;
736 struct sockaddr_in in;
739 if (ofconn->band == OFPROTO_IN_BAND
740 && stream_parse_target_with_default_port(target, OFP_PORT, &sa.ss)
741 && sa.ss.ss_family == AF_INET) {
742 addrs[n_addrs++] = sa.in;
745 for (i = 0; i < mgr->n_extra_remotes; i++) {
746 addrs[n_addrs++] = mgr->extra_in_band_remotes[i];
749 /* Create or update or destroy in-band. */
752 in_band_create(mgr->ofproto, mgr->local_port_name, &mgr->in_band);
754 in_band_set_queue(mgr->in_band, mgr->in_band_queue);
756 /* in_band_run() needs a chance to delete any existing in-band flows.
757 * We will destroy mgr->in_band after it's done with that. */
760 in_band_set_remotes(mgr->in_band, addrs, n_addrs);
768 update_fail_open(struct connmgr *mgr)
769 OVS_EXCLUDED(ofproto_mutex)
771 if (connmgr_has_controllers(mgr)
772 && mgr->fail_mode == OFPROTO_FAIL_STANDALONE) {
773 if (!mgr->fail_open) {
774 mgr->fail_open = fail_open_create(mgr->ofproto, mgr);
777 fail_open_destroy(mgr->fail_open);
778 mgr->fail_open = NULL;
783 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
784 const struct sset *sset)
786 struct pvconn **pvconns = *pvconnsp;
787 size_t n_pvconns = *n_pvconnsp;
792 for (i = 0; i < n_pvconns; i++) {
793 pvconn_close(pvconns[i]);
797 pvconns = xmalloc(sset_count(sset) * sizeof *pvconns);
799 SSET_FOR_EACH (name, sset) {
800 struct pvconn *pvconn;
802 error = pvconn_open(name, 0, 0, &pvconn);
804 pvconns[n_pvconns++] = pvconn;
806 VLOG_ERR("failed to listen on %s: %s", name, ovs_strerror(error));
814 *n_pvconnsp = n_pvconns;
819 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
820 * means that 'ofconn' is more interesting for monitoring than a lower return
823 snoop_preference(const struct ofconn *ofconn)
825 switch (ofconn->role) {
826 case OFPCR12_ROLE_MASTER:
828 case OFPCR12_ROLE_EQUAL:
830 case OFPCR12_ROLE_SLAVE:
832 case OFPCR12_ROLE_NOCHANGE:
834 /* Shouldn't happen. */
839 /* One of 'mgr''s "snoop" pvconns has accepted a new connection on 'vconn'.
840 * Connects this vconn to a controller. */
842 add_snooper(struct connmgr *mgr, struct vconn *vconn)
844 struct ofconn *ofconn, *best;
846 /* Pick a controller for monitoring. */
848 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
849 if (ofconn->type == OFCONN_PRIMARY
850 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
856 rconn_add_monitor(best->rconn, vconn);
858 VLOG_INFO_RL(&rl, "no controller connection to snoop");
863 /* Public ofconn functions. */
865 /* Returns the connection type, either OFCONN_PRIMARY or OFCONN_SERVICE. */
867 ofconn_get_type(const struct ofconn *ofconn)
872 /* If a master election id is defined, stores it into '*idp' and returns
873 * true. Otherwise, stores UINT64_MAX into '*idp' and returns false. */
875 ofconn_get_master_election_id(const struct ofconn *ofconn, uint64_t *idp)
877 *idp = (ofconn->connmgr->master_election_id_defined
878 ? ofconn->connmgr->master_election_id
880 return ofconn->connmgr->master_election_id_defined;
883 /* Sets the master election id.
885 * Returns true if successful, false if the id is stale
888 ofconn_set_master_election_id(struct ofconn *ofconn, uint64_t id)
890 if (ofconn->connmgr->master_election_id_defined
892 /* Unsigned difference interpreted as a two's complement signed
894 (int64_t)(id - ofconn->connmgr->master_election_id) < 0) {
897 ofconn->connmgr->master_election_id = id;
898 ofconn->connmgr->master_election_id_defined = true;
903 /* Returns the role configured for 'ofconn'.
905 * The default role, if no other role has been set, is OFPCR12_ROLE_EQUAL. */
906 enum ofp12_controller_role
907 ofconn_get_role(const struct ofconn *ofconn)
913 ofconn_send_role_status(struct ofconn *ofconn, uint32_t role, uint8_t reason)
915 struct ofputil_role_status status;
918 status.reason = reason;
920 ofconn_get_master_election_id(ofconn, &status.generation_id);
922 buf = ofputil_encode_role_status(&status, ofconn_get_protocol(ofconn));
924 ofconn_send(ofconn, buf, NULL);
928 /* Changes 'ofconn''s role to 'role'. If 'role' is OFPCR12_ROLE_MASTER then
929 * any existing master is demoted to a slave. */
931 ofconn_set_role(struct ofconn *ofconn, enum ofp12_controller_role role)
933 if (role != ofconn->role && role == OFPCR12_ROLE_MASTER) {
934 struct ofconn *other;
936 LIST_FOR_EACH (other, node, &ofconn->connmgr->all_conns) {
937 if (other->role == OFPCR12_ROLE_MASTER) {
938 other->role = OFPCR12_ROLE_SLAVE;
939 ofconn_send_role_status(other, OFPCR12_ROLE_SLAVE, OFPCRR_MASTER_REQUEST);
947 ofconn_set_invalid_ttl_to_controller(struct ofconn *ofconn, bool enable)
949 struct ofputil_async_cfg ac = ofconn_get_async_config(ofconn);
950 uint32_t bit = 1u << OFPR_INVALID_TTL;
952 ac.master[OAM_PACKET_IN] |= bit;
954 ac.master[OAM_PACKET_IN] &= ~bit;
956 ofconn_set_async_config(ofconn, &ac);
960 ofconn_get_invalid_ttl_to_controller(struct ofconn *ofconn)
962 struct ofputil_async_cfg ac = ofconn_get_async_config(ofconn);
963 uint32_t bit = 1u << OFPR_INVALID_TTL;
964 return (ac.master[OAM_PACKET_IN] & bit) != 0;
967 /* Returns the currently configured protocol for 'ofconn', one of OFPUTIL_P_*.
969 * Returns OFPUTIL_P_NONE, which is not a valid protocol, if 'ofconn' hasn't
970 * completed version negotiation. This can't happen if at least one OpenFlow
971 * message, other than OFPT_HELLO, has been received on the connection (such as
972 * in ofproto.c's message handling code), since version negotiation is a
973 * prerequisite for starting to receive messages. This means that
974 * OFPUTIL_P_NONE is a special case that most callers need not worry about. */
975 enum ofputil_protocol
976 ofconn_get_protocol(const struct ofconn *ofconn)
978 if (ofconn->protocol == OFPUTIL_P_NONE &&
979 rconn_is_connected(ofconn->rconn)) {
980 int version = rconn_get_version(ofconn->rconn);
982 ofconn_set_protocol(CONST_CAST(struct ofconn *, ofconn),
983 ofputil_protocol_from_ofp_version(version));
987 return ofconn->protocol;
990 /* Sets the protocol for 'ofconn' to 'protocol' (one of OFPUTIL_P_*).
992 * (This doesn't actually send anything to accomplish this. Presumably the
993 * caller already did that.) */
995 ofconn_set_protocol(struct ofconn *ofconn, enum ofputil_protocol protocol)
997 ofconn->protocol = protocol;
1000 /* Returns the currently configured packet in format for 'ofconn', one of
1003 * The default, if no other format has been set, is NXPIF_STANDARD. */
1004 enum nx_packet_in_format
1005 ofconn_get_packet_in_format(struct ofconn *ofconn)
1007 return ofconn->packet_in_format;
1010 /* Sets the packet in format for 'ofconn' to 'packet_in_format' (one of
1013 ofconn_set_packet_in_format(struct ofconn *ofconn,
1014 enum nx_packet_in_format packet_in_format)
1016 ofconn->packet_in_format = packet_in_format;
1019 /* Sets the controller connection ID for 'ofconn' to 'controller_id'.
1021 * The connection controller ID is used for OFPP_CONTROLLER and
1022 * NXAST_CONTROLLER actions. See "struct nx_action_controller" for details. */
1024 ofconn_set_controller_id(struct ofconn *ofconn, uint16_t controller_id)
1026 ofconn->controller_id = controller_id;
1029 /* Returns the default miss send length for 'ofconn'. */
1031 ofconn_get_miss_send_len(const struct ofconn *ofconn)
1033 return ofconn->miss_send_len;
1036 /* Sets the default miss send length for 'ofconn' to 'miss_send_len'. */
1038 ofconn_set_miss_send_len(struct ofconn *ofconn, int miss_send_len)
1040 ofconn->miss_send_len = miss_send_len;
1044 ofconn_set_async_config(struct ofconn *ofconn,
1045 const struct ofputil_async_cfg *ac)
1047 if (!ofconn->async_cfg) {
1048 ofconn->async_cfg = xmalloc(sizeof *ofconn->async_cfg);
1050 *ofconn->async_cfg = *ac;
1053 struct ofputil_async_cfg
1054 ofconn_get_async_config(const struct ofconn *ofconn)
1056 if (ofconn->async_cfg) {
1057 return *ofconn->async_cfg;
1060 int version = rconn_get_version(ofconn->rconn);
1061 return (version < 0 || !ofconn->enable_async_msgs
1062 ? OFPUTIL_ASYNC_CFG_INIT
1063 : ofputil_async_cfg_default(version));
1066 /* Sends 'msg' on 'ofconn', accounting it as a reply. (If there is a
1067 * sufficient number of OpenFlow replies in-flight on a single ofconn, then the
1068 * connmgr will stop accepting new OpenFlow requests on that ofconn until the
1069 * controller has accepted some of the replies.) */
1071 ofconn_send_reply(const struct ofconn *ofconn, struct ofpbuf *msg)
1073 ofconn_send(ofconn, msg, ofconn->reply_counter);
1076 /* Sends each of the messages in list 'replies' on 'ofconn' in order,
1077 * accounting them as replies. */
1079 ofconn_send_replies(const struct ofconn *ofconn, struct ovs_list *replies)
1081 struct ofpbuf *reply;
1083 LIST_FOR_EACH_POP (reply, list_node, replies) {
1084 ofconn_send_reply(ofconn, reply);
1088 /* Sends 'error' on 'ofconn', as a reply to 'request'. Only at most the
1089 * first 64 bytes of 'request' are used. */
1091 ofconn_send_error(const struct ofconn *ofconn,
1092 const struct ofp_header *request, enum ofperr error)
1094 static struct vlog_rate_limit err_rl = VLOG_RATE_LIMIT_INIT(10, 10);
1095 struct ofpbuf *reply;
1097 reply = ofperr_encode_reply(error, request);
1098 if (!VLOG_DROP_INFO(&err_rl)) {
1099 const char *type_name;
1103 request_len = ntohs(request->length);
1104 type_name = (!ofpraw_decode_partial(&raw, request,
1105 MIN(64, request_len))
1106 ? ofpraw_get_name(raw)
1109 VLOG_INFO("%s: sending %s error reply to %s message",
1110 rconn_get_name(ofconn->rconn), ofperr_to_string(error),
1113 ofconn_send_reply(ofconn, reply);
1116 /* Same as pktbuf_retrieve(), using the pktbuf owned by 'ofconn'. */
1118 ofconn_pktbuf_retrieve(struct ofconn *ofconn, uint32_t id,
1119 struct dp_packet **bufferp, ofp_port_t *in_port)
1121 return pktbuf_retrieve(ofconn->pktbuf, id, bufferp, in_port);
1124 /* Reports that a flow_mod operation of the type specified by 'command' was
1125 * successfully executed by 'ofconn', so that the connmgr can log it. */
1127 ofconn_report_flow_mod(struct ofconn *ofconn,
1128 enum ofp_flow_mod_command command)
1138 case OFPFC_MODIFY_STRICT:
1143 case OFPFC_DELETE_STRICT:
1149 if (ofconn->next_op_report == LLONG_MAX) {
1150 ofconn->first_op = now;
1151 ofconn->next_op_report = MAX(now + 10 * 1000, ofconn->op_backoff);
1152 ofconn->op_backoff = ofconn->next_op_report + 60 * 1000;
1154 ofconn->last_op = now;
1157 /* OpenFlow 1.4 bundles. */
1159 static inline uint32_t
1160 bundle_hash(uint32_t id)
1162 return hash_int(id, 0);
1166 ofconn_get_bundle(struct ofconn *ofconn, uint32_t id)
1168 struct ofp_bundle *bundle;
1170 HMAP_FOR_EACH_IN_BUCKET(bundle, node, bundle_hash(id), &ofconn->bundles) {
1171 if (bundle->id == id) {
1180 ofconn_insert_bundle(struct ofconn *ofconn, struct ofp_bundle *bundle)
1182 /* XXX: Check the limit of open bundles */
1184 hmap_insert(&ofconn->bundles, &bundle->node, bundle_hash(bundle->id));
1190 ofconn_remove_bundle(struct ofconn *ofconn, struct ofp_bundle *bundle)
1192 hmap_remove(&ofconn->bundles, &bundle->node);
1198 bundle_remove_all(struct ofconn *ofconn)
1200 struct ofp_bundle *b, *next;
1202 HMAP_FOR_EACH_SAFE (b, next, node, &ofconn->bundles) {
1203 ofp_bundle_remove__(ofconn, b, false);
1207 /* Private ofconn functions. */
1210 ofconn_get_target(const struct ofconn *ofconn)
1212 return rconn_get_target(ofconn->rconn);
1215 static struct ofconn *
1216 ofconn_create(struct connmgr *mgr, struct rconn *rconn, enum ofconn_type type,
1217 bool enable_async_msgs)
1219 struct ofconn *ofconn;
1221 ofconn = xzalloc(sizeof *ofconn);
1222 ofconn->connmgr = mgr;
1223 ovs_list_push_back(&mgr->all_conns, &ofconn->node);
1224 ofconn->rconn = rconn;
1225 ofconn->type = type;
1226 ofconn->enable_async_msgs = enable_async_msgs;
1228 hmap_init(&ofconn->monitors);
1229 ovs_list_init(&ofconn->updates);
1231 hmap_init(&ofconn->bundles);
1233 ofconn_flush(ofconn);
1238 /* Clears all of the state in 'ofconn' that should not persist from one
1239 * connection to the next. */
1241 ofconn_flush(struct ofconn *ofconn)
1242 OVS_REQUIRES(ofproto_mutex)
1244 struct ofmonitor *monitor, *next_monitor;
1247 ofconn_log_flow_mods(ofconn);
1249 ofconn->role = OFPCR12_ROLE_EQUAL;
1250 ofconn_set_protocol(ofconn, OFPUTIL_P_NONE);
1251 ofconn->packet_in_format = NXPIF_STANDARD;
1253 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1254 ofconn->packet_in_counter = rconn_packet_counter_create();
1255 for (i = 0; i < N_SCHEDULERS; i++) {
1256 if (ofconn->schedulers[i]) {
1259 pinsched_get_limits(ofconn->schedulers[i], &rate, &burst);
1260 pinsched_destroy(ofconn->schedulers[i]);
1261 ofconn->schedulers[i] = pinsched_create(rate, burst);
1264 if (ofconn->pktbuf) {
1265 pktbuf_destroy(ofconn->pktbuf);
1266 ofconn->pktbuf = pktbuf_create();
1268 ofconn->miss_send_len = (ofconn->type == OFCONN_PRIMARY
1269 ? OFP_DEFAULT_MISS_SEND_LEN
1271 ofconn->controller_id = 0;
1273 rconn_packet_counter_destroy(ofconn->reply_counter);
1274 ofconn->reply_counter = rconn_packet_counter_create();
1276 free(ofconn->async_cfg);
1277 ofconn->async_cfg = NULL;
1279 ofconn->n_add = ofconn->n_delete = ofconn->n_modify = 0;
1280 ofconn->first_op = ofconn->last_op = LLONG_MIN;
1281 ofconn->next_op_report = LLONG_MAX;
1282 ofconn->op_backoff = LLONG_MIN;
1284 HMAP_FOR_EACH_SAFE (monitor, next_monitor, ofconn_node,
1285 &ofconn->monitors) {
1286 ofmonitor_destroy(monitor);
1288 rconn_packet_counter_destroy(ofconn->monitor_counter);
1289 ofconn->monitor_counter = rconn_packet_counter_create();
1290 ofpbuf_list_delete(&ofconn->updates); /* ...but it should be empty. */
1294 ofconn_destroy(struct ofconn *ofconn)
1295 OVS_REQUIRES(ofproto_mutex)
1297 ofconn_flush(ofconn);
1299 if (ofconn->type == OFCONN_PRIMARY) {
1300 hmap_remove(&ofconn->connmgr->controllers, &ofconn->hmap_node);
1303 bundle_remove_all(ofconn);
1304 hmap_destroy(&ofconn->bundles);
1306 hmap_destroy(&ofconn->monitors);
1307 ovs_list_remove(&ofconn->node);
1308 rconn_destroy(ofconn->rconn);
1309 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1310 rconn_packet_counter_destroy(ofconn->reply_counter);
1311 pktbuf_destroy(ofconn->pktbuf);
1312 rconn_packet_counter_destroy(ofconn->monitor_counter);
1316 /* Reconfigures 'ofconn' to match 'c'. 'ofconn' and 'c' must have the same
1319 ofconn_reconfigure(struct ofconn *ofconn, const struct ofproto_controller *c)
1323 ofconn->band = c->band;
1324 ofconn->enable_async_msgs = c->enable_async_msgs;
1326 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
1328 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
1329 rconn_set_probe_interval(ofconn->rconn, probe_interval);
1331 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
1333 /* If dscp value changed reconnect. */
1334 if (c->dscp != rconn_get_dscp(ofconn->rconn)) {
1335 rconn_set_dscp(ofconn->rconn, c->dscp);
1336 rconn_reconnect(ofconn->rconn);
1340 /* Returns true if it makes sense for 'ofconn' to receive and process OpenFlow
1343 ofconn_may_recv(const struct ofconn *ofconn)
1345 int count = rconn_packet_counter_n_packets(ofconn->reply_counter);
1346 return count < OFCONN_REPLY_MAX;
1350 ofconn_run(struct ofconn *ofconn,
1351 void (*handle_openflow)(struct ofconn *,
1352 const struct ofpbuf *ofp_msg))
1354 struct connmgr *mgr = ofconn->connmgr;
1357 for (i = 0; i < N_SCHEDULERS; i++) {
1358 struct ovs_list txq;
1360 pinsched_run(ofconn->schedulers[i], &txq);
1361 do_send_packet_ins(ofconn, &txq);
1364 rconn_run(ofconn->rconn);
1366 /* Limit the number of iterations to avoid starving other tasks. */
1367 for (i = 0; i < 50 && ofconn_may_recv(ofconn); i++) {
1368 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1373 if (mgr->fail_open) {
1374 fail_open_maybe_recover(mgr->fail_open);
1377 handle_openflow(ofconn, of_msg);
1378 ofpbuf_delete(of_msg);
1381 if (time_msec() >= ofconn->next_op_report) {
1382 ofconn_log_flow_mods(ofconn);
1385 ovs_mutex_lock(&ofproto_mutex);
1386 if (!rconn_is_alive(ofconn->rconn)) {
1387 ofconn_destroy(ofconn);
1388 } else if (!rconn_is_connected(ofconn->rconn)) {
1389 ofconn_flush(ofconn);
1391 ovs_mutex_unlock(&ofproto_mutex);
1395 ofconn_wait(struct ofconn *ofconn)
1399 for (i = 0; i < N_SCHEDULERS; i++) {
1400 pinsched_wait(ofconn->schedulers[i]);
1402 rconn_run_wait(ofconn->rconn);
1403 if (ofconn_may_recv(ofconn)) {
1404 rconn_recv_wait(ofconn->rconn);
1406 if (ofconn->next_op_report != LLONG_MAX) {
1407 poll_timer_wait_until(ofconn->next_op_report);
1412 ofconn_log_flow_mods(struct ofconn *ofconn)
1414 int n_flow_mods = ofconn->n_add + ofconn->n_delete + ofconn->n_modify;
1416 long long int ago = (time_msec() - ofconn->first_op) / 1000;
1417 long long int interval = (ofconn->last_op - ofconn->first_op) / 1000;
1421 ds_put_format(&s, "%d flow_mods ", n_flow_mods);
1422 if (interval == ago) {
1423 ds_put_format(&s, "in the last %lld s", ago);
1424 } else if (interval) {
1425 ds_put_format(&s, "in the %lld s starting %lld s ago",
1428 ds_put_format(&s, "%lld s ago", ago);
1431 ds_put_cstr(&s, " (");
1432 if (ofconn->n_add) {
1433 ds_put_format(&s, "%d adds, ", ofconn->n_add);
1435 if (ofconn->n_delete) {
1436 ds_put_format(&s, "%d deletes, ", ofconn->n_delete);
1438 if (ofconn->n_modify) {
1439 ds_put_format(&s, "%d modifications, ", ofconn->n_modify);
1442 ds_put_char(&s, ')');
1444 VLOG_INFO("%s: %s", rconn_get_name(ofconn->rconn), ds_cstr(&s));
1447 ofconn->n_add = ofconn->n_delete = ofconn->n_modify = 0;
1449 ofconn->next_op_report = LLONG_MAX;
1452 /* Returns true if 'ofconn' should receive asynchronous messages of the given
1453 * OAM_* 'type' and 'reason', which should be a OFPR_* value for OAM_PACKET_IN,
1454 * a OFPPR_* value for OAM_PORT_STATUS, or an OFPRR_* value for
1455 * OAM_FLOW_REMOVED. Returns false if the message should not be sent on
1458 ofconn_receives_async_msg(const struct ofconn *ofconn,
1459 enum ofputil_async_msg_type type,
1460 unsigned int reason)
1462 ovs_assert(reason < 32);
1463 ovs_assert((unsigned int) type < OAM_N_TYPES);
1465 /* Keep the following code in sync with the documentation in the
1466 * "Asynchronous Messages" section in DESIGN. */
1468 if (ofconn->type == OFCONN_SERVICE && !ofconn->miss_send_len) {
1469 /* Service connections don't get asynchronous messages unless they have
1470 * explicitly asked for them by setting a nonzero miss send length. */
1474 struct ofputil_async_cfg ac = ofconn_get_async_config(ofconn);
1475 uint32_t *masks = (ofconn->role == OFPCR12_ROLE_SLAVE
1478 return (masks[type] & (1u << reason)) != 0;
1481 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
1482 * packet rather than to send the packet to the controller.
1484 * This function returns true to indicate that a packet_in message
1485 * for a "table-miss" should be sent to at least one controller.
1486 * That is there is at least one controller with controller_id 0
1487 * which connected using an OpenFlow version earlier than OpenFlow1.3.
1491 * This logic assumes that "table-miss" packet_in messages
1492 * are always sent to controller_id 0. */
1494 connmgr_wants_packet_in_on_miss(struct connmgr *mgr) OVS_EXCLUDED(ofproto_mutex)
1496 struct ofconn *ofconn;
1498 ovs_mutex_lock(&ofproto_mutex);
1499 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1500 enum ofputil_protocol protocol = ofconn_get_protocol(ofconn);
1502 if (ofconn->controller_id == 0 &&
1503 (protocol == OFPUTIL_P_NONE ||
1504 ofputil_protocol_to_ofp_version(protocol) < OFP13_VERSION)) {
1505 ovs_mutex_unlock(&ofproto_mutex);
1509 ovs_mutex_unlock(&ofproto_mutex);
1514 /* Returns a human-readable name for an OpenFlow connection between 'mgr' and
1515 * 'target', suitable for use in log messages for identifying the connection.
1517 * The name is dynamically allocated. The caller should free it (with free())
1518 * when it is no longer needed. */
1520 ofconn_make_name(const struct connmgr *mgr, const char *target)
1522 return xasprintf("%s<->%s", mgr->name, target);
1526 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1530 for (i = 0; i < N_SCHEDULERS; i++) {
1531 struct pinsched **s = &ofconn->schedulers[i];
1535 *s = pinsched_create(rate, burst);
1537 pinsched_set_limits(*s, rate, burst);
1540 pinsched_destroy(*s);
1547 ofconn_send(const struct ofconn *ofconn, struct ofpbuf *msg,
1548 struct rconn_packet_counter *counter)
1550 ofpmsg_update_length(msg);
1551 rconn_send(ofconn->rconn, msg, counter);
1554 /* Sending asynchronous messages. */
1556 /* Sends an OFPT_PORT_STATUS message with 'opp' and 'reason' to appropriate
1557 * controllers managed by 'mgr'. For messages caused by a controller
1558 * OFPT_PORT_MOD, specify 'source' as the controller connection that sent the
1559 * request; otherwise, specify 'source' as NULL. */
1561 connmgr_send_port_status(struct connmgr *mgr, struct ofconn *source,
1562 const struct ofputil_phy_port *pp, uint8_t reason)
1564 /* XXX Should limit the number of queued port status change messages. */
1565 struct ofputil_port_status ps;
1566 struct ofconn *ofconn;
1570 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1571 if (ofconn_receives_async_msg(ofconn, OAM_PORT_STATUS, reason)) {
1574 /* Before 1.5, OpenFlow specified that OFPT_PORT_MOD should not
1575 * generate OFPT_PORT_STATUS messages. That requirement was a
1576 * relic of how OpenFlow originally supported a single controller,
1577 * so that one could expect the controller to already know the
1578 * changes it had made.
1580 * EXT-338 changes OpenFlow 1.5 OFPT_PORT_MOD to send
1581 * OFPT_PORT_STATUS messages to every controller. This is
1582 * obviously more useful in the multi-controller case. We could
1583 * always implement it that way in OVS, but that would risk
1584 * confusing controllers that are intended for single-controller
1585 * use only. (Imagine a controller that generates an OFPT_PORT_MOD
1586 * in response to any OFPT_PORT_STATUS!)
1588 * So this compromises: for OpenFlow 1.4 and earlier, it generates
1589 * OFPT_PORT_STATUS for OFPT_PORT_MOD, but not back to the
1590 * originating controller. In a single-controller environment, in
1591 * particular, this means that it will never generate
1592 * OFPT_PORT_STATUS for OFPT_PORT_MOD at all. */
1593 if (ofconn == source
1594 && rconn_get_version(ofconn->rconn) < OFP15_VERSION) {
1598 msg = ofputil_encode_port_status(&ps, ofconn_get_protocol(ofconn));
1599 ofconn_send(ofconn, msg, NULL);
1604 /* Sends an OFPT_REQUESTFORWARD message with 'request' and 'reason' to
1605 * appropriate controllers managed by 'mgr'. For messages caused by a
1606 * controller OFPT_GROUP_MOD and OFPT_METER_MOD, specify 'source' as the
1607 * controller connection that sent the request; otherwise, specify 'source'
1610 connmgr_send_requestforward(struct connmgr *mgr, const struct ofconn *source,
1611 const struct ofputil_requestforward *rf)
1613 struct ofconn *ofconn;
1615 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1616 if (ofconn_receives_async_msg(ofconn, OAM_REQUESTFORWARD, rf->reason)
1617 && rconn_get_version(ofconn->rconn) >= OFP14_VERSION
1618 && ofconn != source) {
1619 enum ofputil_protocol protocol = ofconn_get_protocol(ofconn);
1620 ofconn_send(ofconn, ofputil_encode_requestforward(rf, protocol),
1626 /* Sends an OFPT_FLOW_REMOVED or NXT_FLOW_REMOVED message based on 'fr' to
1627 * appropriate controllers managed by 'mgr'. */
1629 connmgr_send_flow_removed(struct connmgr *mgr,
1630 const struct ofputil_flow_removed *fr)
1632 struct ofconn *ofconn;
1634 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1635 if (ofconn_receives_async_msg(ofconn, OAM_FLOW_REMOVED, fr->reason)) {
1638 /* Account flow expirations as replies to OpenFlow requests. That
1639 * works because preventing OpenFlow requests from being processed
1640 * also prevents new flows from being added (and expiring). (It
1641 * also prevents processing OpenFlow requests that would not add
1642 * new flows, so it is imperfect.) */
1643 msg = ofputil_encode_flow_removed(fr, ofconn_get_protocol(ofconn));
1644 ofconn_send_reply(ofconn, msg);
1649 /* Sends an OFPT_TABLE_STATUS message with 'reason' to appropriate controllers
1650 * managed by 'mgr'. When the table state changes, the controller needs to be
1651 * informed with the OFPT_TABLE_STATUS message. The reason values
1652 * OFPTR_VACANCY_DOWN and OFPTR_VACANCY_UP identify a vacancy message. The
1653 * vacancy events are generated when the remaining space in the flow table
1654 * changes and crosses one of the vacancy thereshold specified by
1655 * OFPT_TABLE_MOD. */
1657 connmgr_send_table_status(struct connmgr *mgr,
1658 const struct ofputil_table_desc *td,
1661 struct ofputil_table_status ts;
1662 struct ofconn *ofconn;
1667 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1668 if (ofconn_receives_async_msg(ofconn, OAM_TABLE_STATUS, reason)) {
1671 msg = ofputil_encode_table_status(&ts,
1672 ofconn_get_protocol(ofconn));
1674 ofconn_send(ofconn, msg, NULL);
1680 /* Given 'pin', sends an OFPT_PACKET_IN message to each OpenFlow controller as
1681 * necessary according to their individual configurations. */
1683 connmgr_send_async_msg(struct connmgr *mgr,
1684 const struct ofproto_async_msg *am)
1686 struct ofconn *ofconn;
1688 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1689 enum ofputil_protocol protocol = ofconn_get_protocol(ofconn);
1690 if (protocol == OFPUTIL_P_NONE || !rconn_is_connected(ofconn->rconn)
1691 || ofconn->controller_id != am->controller_id
1692 || !ofconn_receives_async_msg(ofconn, am->oam,
1693 am->pin.up.public.reason)) {
1697 struct ofpbuf *msg = ofputil_encode_packet_in_private(
1698 &am->pin.up, protocol, ofconn->packet_in_format,
1699 am->pin.max_len >= 0 ? am->pin.max_len : ofconn->miss_send_len,
1702 struct ovs_list txq;
1703 bool is_miss = (am->pin.up.public.reason == OFPR_NO_MATCH ||
1704 am->pin.up.public.reason == OFPR_EXPLICIT_MISS ||
1705 am->pin.up.public.reason == OFPR_IMPLICIT_MISS);
1706 pinsched_send(ofconn->schedulers[is_miss],
1707 am->pin.up.public.flow_metadata.flow.in_port.ofp_port,
1709 do_send_packet_ins(ofconn, &txq);
1714 do_send_packet_ins(struct ofconn *ofconn, struct ovs_list *txq)
1718 LIST_FOR_EACH_POP (pin, list_node, txq) {
1719 if (rconn_send_with_limit(ofconn->rconn, pin,
1720 ofconn->packet_in_counter, 100) == EAGAIN) {
1721 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5);
1723 VLOG_INFO_RL(&rl, "%s: dropping packet-in due to queue overflow",
1724 rconn_get_name(ofconn->rconn));
1729 /* Fail-open settings. */
1731 /* Returns the failure handling mode (OFPROTO_FAIL_SECURE or
1732 * OFPROTO_FAIL_STANDALONE) for 'mgr'. */
1733 enum ofproto_fail_mode
1734 connmgr_get_fail_mode(const struct connmgr *mgr)
1736 return mgr->fail_mode;
1739 /* Sets the failure handling mode for 'mgr' to 'fail_mode' (either
1740 * OFPROTO_FAIL_SECURE or OFPROTO_FAIL_STANDALONE). */
1742 connmgr_set_fail_mode(struct connmgr *mgr, enum ofproto_fail_mode fail_mode)
1744 if (mgr->fail_mode != fail_mode) {
1745 mgr->fail_mode = fail_mode;
1746 update_fail_open(mgr);
1747 if (!connmgr_has_controllers(mgr)) {
1748 ofproto_flush_flows(mgr->ofproto);
1753 /* Fail-open implementation. */
1755 /* Returns the longest probe interval among the primary controllers configured
1756 * on 'mgr'. Returns 0 if there are no primary controllers. */
1758 connmgr_get_max_probe_interval(const struct connmgr *mgr)
1760 const struct ofconn *ofconn;
1761 int max_probe_interval;
1763 max_probe_interval = 0;
1764 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1765 int probe_interval = rconn_get_probe_interval(ofconn->rconn);
1766 max_probe_interval = MAX(max_probe_interval, probe_interval);
1768 return max_probe_interval;
1771 /* Returns the number of seconds for which all of 'mgr's primary controllers
1772 * have been disconnected. Returns 0 if 'mgr' has no primary controllers. */
1774 connmgr_failure_duration(const struct connmgr *mgr)
1776 const struct ofconn *ofconn;
1777 int min_failure_duration;
1779 if (!connmgr_has_controllers(mgr)) {
1783 min_failure_duration = INT_MAX;
1784 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1785 int failure_duration = rconn_failure_duration(ofconn->rconn);
1786 min_failure_duration = MIN(min_failure_duration, failure_duration);
1788 return min_failure_duration;
1791 /* Returns true if at least one primary controller is connected (regardless of
1792 * whether those controllers are believed to have authenticated and accepted
1793 * this switch), false if none of them are connected. */
1795 connmgr_is_any_controller_connected(const struct connmgr *mgr)
1797 const struct ofconn *ofconn;
1799 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1800 if (rconn_is_connected(ofconn->rconn)) {
1807 /* Returns true if at least one primary controller is believed to have
1808 * authenticated and accepted this switch, false otherwise. */
1810 connmgr_is_any_controller_admitted(const struct connmgr *mgr)
1812 const struct ofconn *ofconn;
1814 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1815 if (rconn_is_admitted(ofconn->rconn)) {
1822 /* In-band configuration. */
1824 static bool any_extras_changed(const struct connmgr *,
1825 const struct sockaddr_in *extras, size_t n);
1827 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'mgr''s
1828 * in-band control should guarantee access, in the same way that in-band
1829 * control guarantees access to OpenFlow controllers. */
1831 connmgr_set_extra_in_band_remotes(struct connmgr *mgr,
1832 const struct sockaddr_in *extras, size_t n)
1834 if (!any_extras_changed(mgr, extras, n)) {
1838 free(mgr->extra_in_band_remotes);
1839 mgr->n_extra_remotes = n;
1840 mgr->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
1842 update_in_band_remotes(mgr);
1845 /* Sets the OpenFlow queue used by flows set up by in-band control on
1846 * 'mgr' to 'queue_id'. If 'queue_id' is negative, then in-band control
1847 * flows will use the default queue. */
1849 connmgr_set_in_band_queue(struct connmgr *mgr, int queue_id)
1851 if (queue_id != mgr->in_band_queue) {
1852 mgr->in_band_queue = queue_id;
1853 update_in_band_remotes(mgr);
1858 any_extras_changed(const struct connmgr *mgr,
1859 const struct sockaddr_in *extras, size_t n)
1863 if (n != mgr->n_extra_remotes) {
1867 for (i = 0; i < n; i++) {
1868 const struct sockaddr_in *old = &mgr->extra_in_band_remotes[i];
1869 const struct sockaddr_in *new = &extras[i];
1871 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
1872 old->sin_port != new->sin_port) {
1880 /* In-band implementation. */
1883 connmgr_has_in_band(struct connmgr *mgr)
1885 return mgr->in_band != NULL;
1888 /* Fail-open and in-band implementation. */
1890 /* Called by 'ofproto' after all flows have been flushed, to allow fail-open
1891 * and standalone mode to re-create their flows.
1893 * In-band control has more sophisticated code that manages flows itself. */
1895 connmgr_flushed(struct connmgr *mgr)
1896 OVS_EXCLUDED(ofproto_mutex)
1898 if (mgr->fail_open) {
1899 fail_open_flushed(mgr->fail_open);
1902 /* If there are no controllers and we're in standalone mode, set up a flow
1903 * that matches every packet and directs them to OFPP_NORMAL (which goes to
1904 * us). Otherwise, the switch is in secure mode and we won't pass any
1905 * traffic until a controller has been defined and it tells us to do so. */
1906 if (!connmgr_has_controllers(mgr)
1907 && mgr->fail_mode == OFPROTO_FAIL_STANDALONE) {
1908 struct ofpbuf ofpacts;
1911 ofpbuf_init(&ofpacts, OFPACT_OUTPUT_SIZE);
1912 ofpact_put_OUTPUT(&ofpacts)->port = OFPP_NORMAL;
1914 match_init_catchall(&match);
1915 ofproto_add_flow(mgr->ofproto, &match, 0, ofpacts.data,
1918 ofpbuf_uninit(&ofpacts);
1922 /* Returns the number of hidden rules created by the in-band and fail-open
1923 * implementations in table 0. (Subtracting this count from the number of
1924 * rules in the table 0 classifier, as maintained in struct oftable, yields
1925 * the number of flows that OVS should report via OpenFlow for table 0.) */
1927 connmgr_count_hidden_rules(const struct connmgr *mgr)
1931 n_hidden += in_band_count_rules(mgr->in_band);
1933 if (mgr->fail_open) {
1934 n_hidden += fail_open_count_rules(mgr->fail_open);
1939 /* Creates a new ofservice for 'target' in 'mgr'. Returns 0 if successful,
1940 * otherwise a positive errno value.
1942 * ofservice_reconfigure() must be called to fully configure the new
1945 ofservice_create(struct connmgr *mgr, const char *target,
1946 uint32_t allowed_versions, uint8_t dscp)
1948 struct ofservice *ofservice;
1949 struct pvconn *pvconn;
1952 error = pvconn_open(target, allowed_versions, dscp, &pvconn);
1957 ofservice = xzalloc(sizeof *ofservice);
1958 hmap_insert(&mgr->services, &ofservice->node, hash_string(target, 0));
1959 ofservice->pvconn = pvconn;
1960 ofservice->allowed_versions = allowed_versions;
1966 ofservice_destroy(struct connmgr *mgr, struct ofservice *ofservice)
1968 hmap_remove(&mgr->services, &ofservice->node);
1969 pvconn_close(ofservice->pvconn);
1974 ofservice_reconfigure(struct ofservice *ofservice,
1975 const struct ofproto_controller *c)
1977 ofservice->probe_interval = c->probe_interval;
1978 ofservice->rate_limit = c->rate_limit;
1979 ofservice->burst_limit = c->burst_limit;
1980 ofservice->enable_async_msgs = c->enable_async_msgs;
1981 ofservice->dscp = c->dscp;
1984 /* Finds and returns the ofservice within 'mgr' that has the given
1985 * 'target', or a null pointer if none exists. */
1986 static struct ofservice *
1987 ofservice_lookup(struct connmgr *mgr, const char *target)
1989 struct ofservice *ofservice;
1991 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1993 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
2000 /* Flow monitors (NXST_FLOW_MONITOR). */
2002 /* A counter incremented when something significant happens to an OpenFlow
2005 * - When a rule is added, its 'add_seqno' and 'modify_seqno' are set to
2006 * the current value (which is then incremented).
2008 * - When a rule is modified, its 'modify_seqno' is set to the current
2009 * value (which is then incremented).
2011 * Thus, by comparing an old value of monitor_seqno against a rule's
2012 * 'add_seqno', one can tell whether the rule was added before or after the old
2013 * value was read, and similarly for 'modify_seqno'.
2015 * 32 bits should normally be sufficient (and would be nice, to save space in
2016 * each rule) but then we'd have to have some special cases for wraparound.
2018 * We initialize monitor_seqno to 1 to allow 0 to be used as an invalid
2020 static uint64_t monitor_seqno = 1;
2022 COVERAGE_DEFINE(ofmonitor_pause);
2023 COVERAGE_DEFINE(ofmonitor_resume);
2026 ofmonitor_create(const struct ofputil_flow_monitor_request *request,
2027 struct ofconn *ofconn, struct ofmonitor **monitorp)
2028 OVS_REQUIRES(ofproto_mutex)
2030 struct ofmonitor *m;
2034 m = ofmonitor_lookup(ofconn, request->id);
2036 return OFPERR_OFPMOFC_MONITOR_EXISTS;
2039 m = xmalloc(sizeof *m);
2041 hmap_insert(&ofconn->monitors, &m->ofconn_node, hash_int(request->id, 0));
2042 m->id = request->id;
2043 m->flags = request->flags;
2044 m->out_port = request->out_port;
2045 m->table_id = request->table_id;
2046 minimatch_init(&m->match, &request->match);
2053 ofmonitor_lookup(struct ofconn *ofconn, uint32_t id)
2054 OVS_REQUIRES(ofproto_mutex)
2056 struct ofmonitor *m;
2058 HMAP_FOR_EACH_IN_BUCKET (m, ofconn_node, hash_int(id, 0),
2059 &ofconn->monitors) {
2068 ofmonitor_destroy(struct ofmonitor *m)
2069 OVS_REQUIRES(ofproto_mutex)
2072 minimatch_destroy(&m->match);
2073 hmap_remove(&m->ofconn->monitors, &m->ofconn_node);
2079 ofmonitor_report(struct connmgr *mgr, struct rule *rule,
2080 enum nx_flow_update_event event,
2081 enum ofp_flow_removed_reason reason,
2082 const struct ofconn *abbrev_ofconn, ovs_be32 abbrev_xid,
2083 const struct rule_actions *old_actions)
2084 OVS_REQUIRES(ofproto_mutex)
2086 enum nx_flow_monitor_flags update;
2087 struct ofconn *ofconn;
2089 if (rule_is_hidden(rule)) {
2096 rule->add_seqno = rule->modify_seqno = monitor_seqno++;
2100 update = NXFMF_DELETE;
2103 case NXFME_MODIFIED:
2104 update = NXFMF_MODIFY;
2105 rule->modify_seqno = monitor_seqno++;
2113 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2114 enum nx_flow_monitor_flags flags = 0;
2115 struct ofmonitor *m;
2117 if (ofconn->monitor_paused) {
2118 /* Only send NXFME_DELETED notifications for flows that were added
2119 * before we paused. */
2120 if (event != NXFME_DELETED
2121 || rule->add_seqno > ofconn->monitor_paused) {
2126 HMAP_FOR_EACH (m, ofconn_node, &ofconn->monitors) {
2127 if (m->flags & update
2128 && (m->table_id == 0xff || m->table_id == rule->table_id)
2129 && (ofproto_rule_has_out_port(rule, m->out_port)
2131 && ofpacts_output_to_port(old_actions->ofpacts,
2132 old_actions->ofpacts_len,
2134 && cls_rule_is_loose_match(&rule->cr, &m->match)) {
2140 if (ovs_list_is_empty(&ofconn->updates)) {
2141 ofputil_start_flow_update(&ofconn->updates);
2142 ofconn->sent_abbrev_update = false;
2145 if (flags & NXFMF_OWN || ofconn != abbrev_ofconn
2146 || ofconn->monitor_paused) {
2147 struct ofputil_flow_update fu;
2151 fu.reason = event == NXFME_DELETED ? reason : 0;
2152 fu.table_id = rule->table_id;
2153 fu.cookie = rule->flow_cookie;
2154 minimatch_expand(&rule->cr.match, &match);
2156 fu.priority = rule->cr.priority;
2158 ovs_mutex_lock(&rule->mutex);
2159 fu.idle_timeout = rule->idle_timeout;
2160 fu.hard_timeout = rule->hard_timeout;
2161 ovs_mutex_unlock(&rule->mutex);
2163 if (flags & NXFMF_ACTIONS) {
2164 const struct rule_actions *actions = rule_get_actions(rule);
2165 fu.ofpacts = actions->ofpacts;
2166 fu.ofpacts_len = actions->ofpacts_len;
2171 ofputil_append_flow_update(&fu, &ofconn->updates);
2172 } else if (!ofconn->sent_abbrev_update) {
2173 struct ofputil_flow_update fu;
2175 fu.event = NXFME_ABBREV;
2176 fu.xid = abbrev_xid;
2177 ofputil_append_flow_update(&fu, &ofconn->updates);
2179 ofconn->sent_abbrev_update = true;
2186 ofmonitor_flush(struct connmgr *mgr)
2187 OVS_REQUIRES(ofproto_mutex)
2189 struct ofconn *ofconn;
2191 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2194 LIST_FOR_EACH_POP (msg, list_node, &ofconn->updates) {
2195 unsigned int n_bytes;
2197 ofconn_send(ofconn, msg, ofconn->monitor_counter);
2198 n_bytes = rconn_packet_counter_n_bytes(ofconn->monitor_counter);
2199 if (!ofconn->monitor_paused && n_bytes > 128 * 1024) {
2200 struct ofpbuf *pause;
2202 COVERAGE_INC(ofmonitor_pause);
2203 ofconn->monitor_paused = monitor_seqno++;
2204 pause = ofpraw_alloc_xid(OFPRAW_NXT_FLOW_MONITOR_PAUSED,
2205 OFP10_VERSION, htonl(0), 0);
2206 ofconn_send(ofconn, pause, ofconn->monitor_counter);
2213 ofmonitor_resume(struct ofconn *ofconn)
2214 OVS_REQUIRES(ofproto_mutex)
2216 struct rule_collection rules;
2217 struct ofpbuf *resumed;
2218 struct ofmonitor *m;
2219 struct ovs_list msgs;
2221 rule_collection_init(&rules);
2222 HMAP_FOR_EACH (m, ofconn_node, &ofconn->monitors) {
2223 ofmonitor_collect_resume_rules(m, ofconn->monitor_paused, &rules);
2226 ovs_list_init(&msgs);
2227 ofmonitor_compose_refresh_updates(&rules, &msgs);
2229 resumed = ofpraw_alloc_xid(OFPRAW_NXT_FLOW_MONITOR_RESUMED, OFP10_VERSION,
2231 ovs_list_push_back(&msgs, &resumed->list_node);
2232 ofconn_send_replies(ofconn, &msgs);
2234 ofconn->monitor_paused = 0;
2238 ofmonitor_may_resume(const struct ofconn *ofconn)
2239 OVS_REQUIRES(ofproto_mutex)
2241 return (ofconn->monitor_paused != 0
2242 && !rconn_packet_counter_n_packets(ofconn->monitor_counter));
2246 ofmonitor_run(struct connmgr *mgr)
2248 struct ofconn *ofconn;
2250 ovs_mutex_lock(&ofproto_mutex);
2251 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2252 if (ofmonitor_may_resume(ofconn)) {
2253 COVERAGE_INC(ofmonitor_resume);
2254 ofmonitor_resume(ofconn);
2257 ovs_mutex_unlock(&ofproto_mutex);
2261 ofmonitor_wait(struct connmgr *mgr)
2263 struct ofconn *ofconn;
2265 ovs_mutex_lock(&ofproto_mutex);
2266 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2267 if (ofmonitor_may_resume(ofconn)) {
2268 poll_immediate_wake();
2271 ovs_mutex_unlock(&ofproto_mutex);
2275 ofproto_async_msg_free(struct ofproto_async_msg *am)
2277 free(am->pin.up.public.packet);
2278 free(am->pin.up.public.userdata);
2279 free(am->pin.up.stack);
2280 free(am->pin.up.actions);
2281 free(am->pin.up.action_set);