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"
44 VLOG_DEFINE_THIS_MODULE(connmgr);
45 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
47 /* An OpenFlow connection.
53 * 'ofproto_mutex' must be held whenever an ofconn is created or destroyed or,
54 * more or less equivalently, whenever an ofconn is added to or removed from a
55 * connmgr. 'ofproto_mutex' doesn't protect the data inside the ofconn, except
56 * as specifically noted below. */
58 /* Configuration that persists from one connection to the next. */
60 struct ovs_list node; /* In struct connmgr's "all_conns" list. */
61 struct hmap_node hmap_node; /* In struct connmgr's "controllers" map. */
63 struct connmgr *connmgr; /* Connection's manager. */
64 struct rconn *rconn; /* OpenFlow connection. */
65 enum ofconn_type type; /* Type. */
66 enum ofproto_band band; /* In-band or out-of-band? */
67 bool enable_async_msgs; /* Initially enable async messages? */
69 /* State that should be cleared from one connection to the next. */
72 enum ofp12_controller_role role; /* Role. */
73 enum ofputil_protocol protocol; /* Current protocol variant. */
74 enum nx_packet_in_format packet_in_format; /* OFPT_PACKET_IN format. */
76 /* OFPT_PACKET_IN related data. */
77 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
78 #define N_SCHEDULERS 2
79 struct pinsched *schedulers[N_SCHEDULERS];
80 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
81 int miss_send_len; /* Bytes to send of buffered packets. */
82 uint16_t controller_id; /* Connection controller ID. */
84 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
85 * requests, and the maximum number before we stop reading OpenFlow
87 #define OFCONN_REPLY_MAX 100
88 struct rconn_packet_counter *reply_counter;
90 /* Asynchronous message configuration in each possible role.
92 * A 1-bit enables sending an asynchronous message for one possible reason
93 * that the message might be generated, a 0-bit disables it. */
94 struct ofputil_async_cfg *async_cfg;
96 /* Flow table operation logging. */
97 int n_add, n_delete, n_modify; /* Number of unreported ops of each kind. */
98 long long int first_op, last_op; /* Range of times for unreported ops. */
99 long long int next_op_report; /* Time to report ops, or LLONG_MAX. */
100 long long int op_backoff; /* Earliest time to report ops again. */
102 /* Flow monitors (e.g. NXST_FLOW_MONITOR). */
104 /* Configuration. Contains "struct ofmonitor"s. */
105 struct hmap monitors OVS_GUARDED_BY(ofproto_mutex);
109 * When too many flow monitor notifications back up in the transmit buffer,
110 * we pause the transmission of further notifications. These members track
111 * the flow control state.
113 * When notifications are flowing, 'monitor_paused' is 0. When
114 * notifications are paused, 'monitor_paused' is the value of
115 * 'monitor_seqno' at the point we paused.
117 * 'monitor_counter' counts the OpenFlow messages and bytes currently in
118 * flight. This value growing too large triggers pausing. */
119 uint64_t monitor_paused OVS_GUARDED_BY(ofproto_mutex);
120 struct rconn_packet_counter *monitor_counter OVS_GUARDED_BY(ofproto_mutex);
122 /* State of monitors for a single ongoing flow_mod.
124 * 'updates' is a list of "struct ofpbuf"s that contain
125 * NXST_FLOW_MONITOR_REPLY messages representing the changes made by the
128 * When 'updates' is nonempty, 'sent_abbrev_update' is true if 'updates'
129 * contains an update event of type NXFME_ABBREV and false otherwise.. */
130 struct ovs_list updates OVS_GUARDED_BY(ofproto_mutex);
131 bool sent_abbrev_update OVS_GUARDED_BY(ofproto_mutex);
133 /* Active bundles. Contains "struct ofp_bundle"s. */
137 static struct ofconn *ofconn_create(struct connmgr *, struct rconn *,
138 enum ofconn_type, bool enable_async_msgs)
139 OVS_REQUIRES(ofproto_mutex);
140 static void ofconn_destroy(struct ofconn *) OVS_REQUIRES(ofproto_mutex);
141 static void ofconn_flush(struct ofconn *) OVS_REQUIRES(ofproto_mutex);
143 static void ofconn_reconfigure(struct ofconn *,
144 const struct ofproto_controller *);
146 static void ofconn_run(struct ofconn *,
147 void (*handle_openflow)(struct ofconn *,
148 const struct ofpbuf *ofp_msg));
149 static void ofconn_wait(struct ofconn *);
151 static void ofconn_log_flow_mods(struct ofconn *);
153 static const char *ofconn_get_target(const struct ofconn *);
154 static char *ofconn_make_name(const struct connmgr *, const char *target);
156 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
158 static void ofconn_send(const struct ofconn *, struct ofpbuf *,
159 struct rconn_packet_counter *);
161 static void do_send_packet_ins(struct ofconn *, struct ovs_list *txq);
163 /* A listener for incoming OpenFlow "service" connections. */
165 struct hmap_node node; /* In struct connmgr's "services" hmap. */
166 struct pvconn *pvconn; /* OpenFlow connection listener. */
168 /* These are not used by ofservice directly. They are settings for
169 * accepted "struct ofconn"s from the pvconn. */
170 int probe_interval; /* Max idle time before probing, in seconds. */
171 int rate_limit; /* Max packet-in rate in packets per second. */
172 int burst_limit; /* Limit on accumulating packet credits. */
173 bool enable_async_msgs; /* Initially enable async messages? */
174 uint8_t dscp; /* DSCP Value for controller connection */
175 uint32_t allowed_versions; /* OpenFlow protocol versions that may
176 * be negotiated for a session. */
179 static void ofservice_reconfigure(struct ofservice *,
180 const struct ofproto_controller *);
181 static int ofservice_create(struct connmgr *mgr, const char *target,
182 uint32_t allowed_versions, uint8_t dscp);
183 static void ofservice_destroy(struct connmgr *, struct ofservice *);
184 static struct ofservice *ofservice_lookup(struct connmgr *,
187 /* Connection manager for an OpenFlow switch. */
189 struct ofproto *ofproto;
191 char *local_port_name;
193 /* OpenFlow connections. */
194 struct hmap controllers; /* All OFCONN_PRIMARY controllers. */
195 struct ovs_list all_conns; /* All controllers. */
196 uint64_t master_election_id; /* monotonically increasing sequence number
197 * for master election */
198 bool master_election_id_defined;
200 /* OpenFlow listeners. */
201 struct hmap services; /* Contains "struct ofservice"s. */
202 struct pvconn **snoops;
206 struct fail_open *fail_open;
207 enum ofproto_fail_mode fail_mode;
209 /* In-band control. */
210 struct in_band *in_band;
211 struct sockaddr_in *extra_in_band_remotes;
212 size_t n_extra_remotes;
216 static void update_in_band_remotes(struct connmgr *);
217 static void add_snooper(struct connmgr *, struct vconn *);
218 static void ofmonitor_run(struct connmgr *);
219 static void ofmonitor_wait(struct connmgr *);
221 /* Creates and returns a new connection manager owned by 'ofproto'. 'name' is
222 * a name for the ofproto suitable for using in log messages.
223 * 'local_port_name' is the name of the local port (OFPP_LOCAL) within
226 connmgr_create(struct ofproto *ofproto,
227 const char *name, const char *local_port_name)
231 mgr = xmalloc(sizeof *mgr);
232 mgr->ofproto = ofproto;
233 mgr->name = xstrdup(name);
234 mgr->local_port_name = xstrdup(local_port_name);
236 hmap_init(&mgr->controllers);
237 ovs_list_init(&mgr->all_conns);
238 mgr->master_election_id = 0;
239 mgr->master_election_id_defined = false;
241 hmap_init(&mgr->services);
245 mgr->fail_open = NULL;
246 mgr->fail_mode = OFPROTO_FAIL_SECURE;
249 mgr->extra_in_band_remotes = NULL;
250 mgr->n_extra_remotes = 0;
251 mgr->in_band_queue = -1;
256 /* Frees 'mgr' and all of its resources. */
258 connmgr_destroy(struct connmgr *mgr)
260 struct ofservice *ofservice, *next_ofservice;
261 struct ofconn *ofconn, *next_ofconn;
268 ovs_mutex_lock(&ofproto_mutex);
269 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &mgr->all_conns) {
270 ofconn_destroy(ofconn);
272 ovs_mutex_unlock(&ofproto_mutex);
274 hmap_destroy(&mgr->controllers);
276 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &mgr->services) {
277 ofservice_destroy(mgr, ofservice);
279 hmap_destroy(&mgr->services);
281 for (i = 0; i < mgr->n_snoops; i++) {
282 pvconn_close(mgr->snoops[i]);
286 fail_open_destroy(mgr->fail_open);
287 mgr->fail_open = NULL;
289 in_band_destroy(mgr->in_band);
291 free(mgr->extra_in_band_remotes);
293 free(mgr->local_port_name);
298 /* Does all of the periodic maintenance required by 'mgr'. Calls
299 * 'handle_openflow' for each message received on an OpenFlow connection,
300 * passing along the OpenFlow connection itself and the message that was sent.
301 * 'handle_openflow' must not modify or free the message. */
303 connmgr_run(struct connmgr *mgr,
304 void (*handle_openflow)(struct ofconn *,
305 const struct ofpbuf *ofp_msg))
306 OVS_EXCLUDED(ofproto_mutex)
308 struct ofconn *ofconn, *next_ofconn;
309 struct ofservice *ofservice;
313 if (!in_band_run(mgr->in_band)) {
314 in_band_destroy(mgr->in_band);
319 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &mgr->all_conns) {
320 ofconn_run(ofconn, handle_openflow);
324 /* Fail-open maintenance. Do this after processing the ofconns since
325 * fail-open checks the status of the controller rconn. */
326 if (mgr->fail_open) {
327 fail_open_run(mgr->fail_open);
330 HMAP_FOR_EACH (ofservice, node, &mgr->services) {
334 retval = pvconn_accept(ofservice->pvconn, &vconn);
339 /* Passing default value for creation of the rconn */
340 rconn = rconn_create(ofservice->probe_interval, 0, ofservice->dscp,
341 vconn_get_allowed_versions(vconn));
342 name = ofconn_make_name(mgr, vconn_get_name(vconn));
343 rconn_connect_unreliably(rconn, vconn, name);
346 ovs_mutex_lock(&ofproto_mutex);
347 ofconn = ofconn_create(mgr, rconn, OFCONN_SERVICE,
348 ofservice->enable_async_msgs);
349 ovs_mutex_unlock(&ofproto_mutex);
351 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
352 ofservice->burst_limit);
353 } else if (retval != EAGAIN) {
354 VLOG_WARN_RL(&rl, "accept failed (%s)", ovs_strerror(retval));
358 for (i = 0; i < mgr->n_snoops; i++) {
362 retval = pvconn_accept(mgr->snoops[i], &vconn);
364 add_snooper(mgr, vconn);
365 } else if (retval != EAGAIN) {
366 VLOG_WARN_RL(&rl, "accept failed (%s)", ovs_strerror(retval));
371 /* Causes the poll loop to wake up when connmgr_run() needs to run. */
373 connmgr_wait(struct connmgr *mgr)
375 struct ofservice *ofservice;
376 struct ofconn *ofconn;
379 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
384 in_band_wait(mgr->in_band);
386 if (mgr->fail_open) {
387 fail_open_wait(mgr->fail_open);
389 HMAP_FOR_EACH (ofservice, node, &mgr->services) {
390 pvconn_wait(ofservice->pvconn);
392 for (i = 0; i < mgr->n_snoops; i++) {
393 pvconn_wait(mgr->snoops[i]);
397 /* Adds some memory usage statistics for 'mgr' into 'usage', for use with
398 * memory_report(). */
400 connmgr_get_memory_usage(const struct connmgr *mgr, struct simap *usage)
402 const struct ofconn *ofconn;
403 unsigned int packets = 0;
404 unsigned int ofconns = 0;
406 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
411 packets += rconn_count_txqlen(ofconn->rconn);
412 for (i = 0; i < N_SCHEDULERS; i++) {
413 struct pinsched_stats stats;
415 pinsched_get_stats(ofconn->schedulers[i], &stats);
416 packets += stats.n_queued;
418 packets += pktbuf_count_packets(ofconn->pktbuf);
420 simap_increase(usage, "ofconns", ofconns);
421 simap_increase(usage, "packets", packets);
424 /* Returns the ofproto that owns 'ofconn''s connmgr. */
426 ofconn_get_ofproto(const struct ofconn *ofconn)
428 return ofconn->connmgr->ofproto;
431 /* OpenFlow configuration. */
433 static void add_controller(struct connmgr *, const char *target, uint8_t dscp,
434 uint32_t allowed_versions)
435 OVS_REQUIRES(ofproto_mutex);
436 static struct ofconn *find_controller_by_target(struct connmgr *,
438 static void update_fail_open(struct connmgr *) OVS_EXCLUDED(ofproto_mutex);
439 static int set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
440 const struct sset *);
442 /* Returns true if 'mgr' has any configured primary controllers.
444 * Service controllers do not count, but configured primary controllers do
445 * count whether or not they are currently connected. */
447 connmgr_has_controllers(const struct connmgr *mgr)
449 return !hmap_is_empty(&mgr->controllers);
452 /* Initializes 'info' and populates it with information about each configured
453 * primary controller. The keys in 'info' are the controllers' targets; the
454 * data values are corresponding "struct ofproto_controller_info".
456 * The caller owns 'info' and everything in it and should free it when it is no
459 connmgr_get_controller_info(struct connmgr *mgr, struct shash *info)
461 const struct ofconn *ofconn;
463 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
464 const struct rconn *rconn = ofconn->rconn;
465 const char *target = rconn_get_target(rconn);
467 if (!shash_find(info, target)) {
468 struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo);
469 time_t now = time_now();
470 time_t last_connection = rconn_get_last_connection(rconn);
471 time_t last_disconnect = rconn_get_last_disconnect(rconn);
472 int last_error = rconn_get_last_error(rconn);
475 shash_add(info, target, cinfo);
477 cinfo->is_connected = rconn_is_connected(rconn);
478 cinfo->role = ofconn->role;
480 smap_init(&cinfo->pairs);
482 smap_add(&cinfo->pairs, "last_error",
483 ovs_retval_to_string(last_error));
486 smap_add(&cinfo->pairs, "state", rconn_get_state(rconn));
488 if (last_connection != TIME_MIN) {
489 smap_add_format(&cinfo->pairs, "sec_since_connect",
490 "%ld", (long int) (now - last_connection));
493 if (last_disconnect != TIME_MIN) {
494 smap_add_format(&cinfo->pairs, "sec_since_disconnect",
495 "%ld", (long int) (now - last_disconnect));
498 for (i = 0; i < N_SCHEDULERS; i++) {
499 if (ofconn->schedulers[i]) {
500 const char *name = i ? "miss" : "action";
501 struct pinsched_stats stats;
503 pinsched_get_stats(ofconn->schedulers[i], &stats);
504 smap_add_nocopy(&cinfo->pairs,
505 xasprintf("packet-in-%s-backlog", name),
506 xasprintf("%u", stats.n_queued));
507 smap_add_nocopy(&cinfo->pairs,
508 xasprintf("packet-in-%s-bypassed", name),
509 xasprintf("%llu", stats.n_normal));
510 smap_add_nocopy(&cinfo->pairs,
511 xasprintf("packet-in-%s-queued", name),
512 xasprintf("%llu", stats.n_limited));
513 smap_add_nocopy(&cinfo->pairs,
514 xasprintf("packet-in-%s-dropped", name),
515 xasprintf("%llu", stats.n_queue_dropped));
523 connmgr_free_controller_info(struct shash *info)
525 struct shash_node *node;
527 SHASH_FOR_EACH (node, info) {
528 struct ofproto_controller_info *cinfo = node->data;
529 smap_destroy(&cinfo->pairs);
535 /* Changes 'mgr''s set of controllers to the 'n_controllers' controllers in
538 connmgr_set_controllers(struct connmgr *mgr,
539 const struct ofproto_controller *controllers,
540 size_t n_controllers, uint32_t allowed_versions)
541 OVS_EXCLUDED(ofproto_mutex)
543 bool had_controllers = connmgr_has_controllers(mgr);
544 struct shash new_controllers;
545 struct ofconn *ofconn, *next_ofconn;
546 struct ofservice *ofservice, *next_ofservice;
549 /* Required to add and remove ofconns. This could probably be narrowed to
550 * cover a smaller amount of code, if that yielded some benefit. */
551 ovs_mutex_lock(&ofproto_mutex);
553 /* Create newly configured controllers and services.
554 * Create a name to ofproto_controller mapping in 'new_controllers'. */
555 shash_init(&new_controllers);
556 for (i = 0; i < n_controllers; i++) {
557 const struct ofproto_controller *c = &controllers[i];
559 if (!vconn_verify_name(c->target)) {
561 ofconn = find_controller_by_target(mgr, c->target);
563 VLOG_INFO("%s: added primary controller \"%s\"",
564 mgr->name, c->target);
566 } else if (rconn_get_allowed_versions(ofconn->rconn) !=
568 VLOG_INFO("%s: re-added primary controller \"%s\"",
569 mgr->name, c->target);
571 ofconn_destroy(ofconn);
574 add_controller(mgr, c->target, c->dscp, allowed_versions);
576 } else if (!pvconn_verify_name(c->target)) {
578 ofservice = ofservice_lookup(mgr, c->target);
580 VLOG_INFO("%s: added service controller \"%s\"",
581 mgr->name, c->target);
583 } else if (ofservice->allowed_versions != allowed_versions) {
584 VLOG_INFO("%s: re-added service controller \"%s\"",
585 mgr->name, c->target);
586 ofservice_destroy(mgr, ofservice);
590 ofservice_create(mgr, c->target, allowed_versions, c->dscp);
593 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
594 mgr->name, c->target);
598 shash_add_once(&new_controllers, c->target, &controllers[i]);
601 /* Delete controllers that are no longer configured.
602 * Update configuration of all now-existing controllers. */
603 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &mgr->controllers) {
604 const char *target = ofconn_get_target(ofconn);
605 struct ofproto_controller *c;
607 c = shash_find_data(&new_controllers, target);
609 VLOG_INFO("%s: removed primary controller \"%s\"",
611 ofconn_destroy(ofconn);
613 ofconn_reconfigure(ofconn, c);
617 /* Delete services that are no longer configured.
618 * Update configuration of all now-existing services. */
619 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &mgr->services) {
620 const char *target = pvconn_get_name(ofservice->pvconn);
621 struct ofproto_controller *c;
623 c = shash_find_data(&new_controllers, target);
625 VLOG_INFO("%s: removed service controller \"%s\"",
627 ofservice_destroy(mgr, ofservice);
629 ofservice_reconfigure(ofservice, c);
633 shash_destroy(&new_controllers);
635 ovs_mutex_unlock(&ofproto_mutex);
637 update_in_band_remotes(mgr);
638 update_fail_open(mgr);
639 if (had_controllers != connmgr_has_controllers(mgr)) {
640 ofproto_flush_flows(mgr->ofproto);
644 /* Drops the connections between 'mgr' and all of its primary and secondary
645 * controllers, forcing them to reconnect. */
647 connmgr_reconnect(const struct connmgr *mgr)
649 struct ofconn *ofconn;
651 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
652 rconn_reconnect(ofconn->rconn);
656 /* Sets the "snoops" for 'mgr' to the pvconn targets listed in 'snoops'.
658 * A "snoop" is a pvconn to which every OpenFlow message to or from the most
659 * important controller on 'mgr' is mirrored. */
661 connmgr_set_snoops(struct connmgr *mgr, const struct sset *snoops)
663 return set_pvconns(&mgr->snoops, &mgr->n_snoops, snoops);
666 /* Adds each of the snoops currently configured on 'mgr' to 'snoops'. */
668 connmgr_get_snoops(const struct connmgr *mgr, struct sset *snoops)
672 for (i = 0; i < mgr->n_snoops; i++) {
673 sset_add(snoops, pvconn_get_name(mgr->snoops[i]));
677 /* Returns true if 'mgr' has at least one snoop, false if it has none. */
679 connmgr_has_snoops(const struct connmgr *mgr)
681 return mgr->n_snoops > 0;
684 /* Creates a new controller for 'target' in 'mgr'. update_controller() needs
685 * to be called later to finish the new ofconn's configuration. */
687 add_controller(struct connmgr *mgr, const char *target, uint8_t dscp,
688 uint32_t allowed_versions)
689 OVS_REQUIRES(ofproto_mutex)
691 char *name = ofconn_make_name(mgr, target);
692 struct ofconn *ofconn;
694 ofconn = ofconn_create(mgr, rconn_create(5, 8, dscp, allowed_versions),
695 OFCONN_PRIMARY, true);
696 ofconn->pktbuf = pktbuf_create();
697 rconn_connect(ofconn->rconn, target, name);
698 hmap_insert(&mgr->controllers, &ofconn->hmap_node, hash_string(target, 0));
703 static struct ofconn *
704 find_controller_by_target(struct connmgr *mgr, const char *target)
706 struct ofconn *ofconn;
708 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
709 hash_string(target, 0), &mgr->controllers) {
710 if (!strcmp(ofconn_get_target(ofconn), target)) {
718 update_in_band_remotes(struct connmgr *mgr)
720 struct sockaddr_in *addrs;
721 size_t max_addrs, n_addrs;
722 struct ofconn *ofconn;
725 /* Allocate enough memory for as many remotes as we could possibly have. */
726 max_addrs = mgr->n_extra_remotes + hmap_count(&mgr->controllers);
727 addrs = xmalloc(max_addrs * sizeof *addrs);
730 /* Add all the remotes. */
731 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
732 const char *target = rconn_get_target(ofconn->rconn);
734 struct sockaddr_storage ss;
735 struct sockaddr_in in;
738 if (ofconn->band == OFPROTO_IN_BAND
739 && stream_parse_target_with_default_port(target, OFP_PORT, &sa.ss)
740 && sa.ss.ss_family == AF_INET) {
741 addrs[n_addrs++] = sa.in;
744 for (i = 0; i < mgr->n_extra_remotes; i++) {
745 addrs[n_addrs++] = mgr->extra_in_band_remotes[i];
748 /* Create or update or destroy in-band. */
751 in_band_create(mgr->ofproto, mgr->local_port_name, &mgr->in_band);
753 in_band_set_queue(mgr->in_band, mgr->in_band_queue);
755 /* in_band_run() needs a chance to delete any existing in-band flows.
756 * We will destroy mgr->in_band after it's done with that. */
759 in_band_set_remotes(mgr->in_band, addrs, n_addrs);
767 update_fail_open(struct connmgr *mgr)
768 OVS_EXCLUDED(ofproto_mutex)
770 if (connmgr_has_controllers(mgr)
771 && mgr->fail_mode == OFPROTO_FAIL_STANDALONE) {
772 if (!mgr->fail_open) {
773 mgr->fail_open = fail_open_create(mgr->ofproto, mgr);
776 fail_open_destroy(mgr->fail_open);
777 mgr->fail_open = NULL;
782 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
783 const struct sset *sset)
785 struct pvconn **pvconns = *pvconnsp;
786 size_t n_pvconns = *n_pvconnsp;
791 for (i = 0; i < n_pvconns; i++) {
792 pvconn_close(pvconns[i]);
796 pvconns = xmalloc(sset_count(sset) * sizeof *pvconns);
798 SSET_FOR_EACH (name, sset) {
799 struct pvconn *pvconn;
801 error = pvconn_open(name, 0, 0, &pvconn);
803 pvconns[n_pvconns++] = pvconn;
805 VLOG_ERR("failed to listen on %s: %s", name, ovs_strerror(error));
813 *n_pvconnsp = n_pvconns;
818 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
819 * means that 'ofconn' is more interesting for monitoring than a lower return
822 snoop_preference(const struct ofconn *ofconn)
824 switch (ofconn->role) {
825 case OFPCR12_ROLE_MASTER:
827 case OFPCR12_ROLE_EQUAL:
829 case OFPCR12_ROLE_SLAVE:
831 case OFPCR12_ROLE_NOCHANGE:
833 /* Shouldn't happen. */
838 /* One of 'mgr''s "snoop" pvconns has accepted a new connection on 'vconn'.
839 * Connects this vconn to a controller. */
841 add_snooper(struct connmgr *mgr, struct vconn *vconn)
843 struct ofconn *ofconn, *best;
845 /* Pick a controller for monitoring. */
847 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
848 if (ofconn->type == OFCONN_PRIMARY
849 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
855 rconn_add_monitor(best->rconn, vconn);
857 VLOG_INFO_RL(&rl, "no controller connection to snoop");
862 /* Public ofconn functions. */
864 /* Returns the connection type, either OFCONN_PRIMARY or OFCONN_SERVICE. */
866 ofconn_get_type(const struct ofconn *ofconn)
871 /* If a master election id is defined, stores it into '*idp' and returns
872 * true. Otherwise, stores UINT64_MAX into '*idp' and returns false. */
874 ofconn_get_master_election_id(const struct ofconn *ofconn, uint64_t *idp)
876 *idp = (ofconn->connmgr->master_election_id_defined
877 ? ofconn->connmgr->master_election_id
879 return ofconn->connmgr->master_election_id_defined;
882 /* Sets the master election id.
884 * Returns true if successful, false if the id is stale
887 ofconn_set_master_election_id(struct ofconn *ofconn, uint64_t id)
889 if (ofconn->connmgr->master_election_id_defined
891 /* Unsigned difference interpreted as a two's complement signed
893 (int64_t)(id - ofconn->connmgr->master_election_id) < 0) {
896 ofconn->connmgr->master_election_id = id;
897 ofconn->connmgr->master_election_id_defined = true;
902 /* Returns the role configured for 'ofconn'.
904 * The default role, if no other role has been set, is OFPCR12_ROLE_EQUAL. */
905 enum ofp12_controller_role
906 ofconn_get_role(const struct ofconn *ofconn)
912 ofconn_send_role_status(struct ofconn *ofconn, uint32_t role, uint8_t reason)
914 struct ofputil_role_status status;
917 status.reason = reason;
919 ofconn_get_master_election_id(ofconn, &status.generation_id);
921 buf = ofputil_encode_role_status(&status, ofconn_get_protocol(ofconn));
923 ofconn_send(ofconn, buf, NULL);
927 /* Changes 'ofconn''s role to 'role'. If 'role' is OFPCR12_ROLE_MASTER then
928 * any existing master is demoted to a slave. */
930 ofconn_set_role(struct ofconn *ofconn, enum ofp12_controller_role role)
932 if (role != ofconn->role && role == OFPCR12_ROLE_MASTER) {
933 struct ofconn *other;
935 LIST_FOR_EACH (other, node, &ofconn->connmgr->all_conns) {
936 if (other->role == OFPCR12_ROLE_MASTER) {
937 other->role = OFPCR12_ROLE_SLAVE;
938 ofconn_send_role_status(other, OFPCR12_ROLE_SLAVE, OFPCRR_MASTER_REQUEST);
946 ofconn_set_invalid_ttl_to_controller(struct ofconn *ofconn, bool enable)
948 struct ofputil_async_cfg ac = ofconn_get_async_config(ofconn);
949 uint32_t bit = 1u << OFPR_INVALID_TTL;
951 ac.master[OAM_PACKET_IN] |= bit;
953 ac.master[OAM_PACKET_IN] &= ~bit;
955 ofconn_set_async_config(ofconn, &ac);
959 ofconn_get_invalid_ttl_to_controller(struct ofconn *ofconn)
961 struct ofputil_async_cfg ac = ofconn_get_async_config(ofconn);
962 uint32_t bit = 1u << OFPR_INVALID_TTL;
963 return (ac.master[OAM_PACKET_IN] & bit) != 0;
966 /* Returns the currently configured protocol for 'ofconn', one of OFPUTIL_P_*.
968 * Returns OFPUTIL_P_NONE, which is not a valid protocol, if 'ofconn' hasn't
969 * completed version negotiation. This can't happen if at least one OpenFlow
970 * message, other than OFPT_HELLO, has been received on the connection (such as
971 * in ofproto.c's message handling code), since version negotiation is a
972 * prerequisite for starting to receive messages. This means that
973 * OFPUTIL_P_NONE is a special case that most callers need not worry about. */
974 enum ofputil_protocol
975 ofconn_get_protocol(const struct ofconn *ofconn)
977 if (ofconn->protocol == OFPUTIL_P_NONE &&
978 rconn_is_connected(ofconn->rconn)) {
979 int version = rconn_get_version(ofconn->rconn);
981 ofconn_set_protocol(CONST_CAST(struct ofconn *, ofconn),
982 ofputil_protocol_from_ofp_version(version));
986 return ofconn->protocol;
989 /* Sets the protocol for 'ofconn' to 'protocol' (one of OFPUTIL_P_*).
991 * (This doesn't actually send anything to accomplish this. Presumably the
992 * caller already did that.) */
994 ofconn_set_protocol(struct ofconn *ofconn, enum ofputil_protocol protocol)
996 ofconn->protocol = protocol;
999 /* Returns the currently configured packet in format for 'ofconn', one of
1002 * The default, if no other format has been set, is NXPIF_STANDARD. */
1003 enum nx_packet_in_format
1004 ofconn_get_packet_in_format(struct ofconn *ofconn)
1006 return ofconn->packet_in_format;
1009 /* Sets the packet in format for 'ofconn' to 'packet_in_format' (one of
1012 ofconn_set_packet_in_format(struct ofconn *ofconn,
1013 enum nx_packet_in_format packet_in_format)
1015 ofconn->packet_in_format = packet_in_format;
1018 /* Sets the controller connection ID for 'ofconn' to 'controller_id'.
1020 * The connection controller ID is used for OFPP_CONTROLLER and
1021 * NXAST_CONTROLLER actions. See "struct nx_action_controller" for details. */
1023 ofconn_set_controller_id(struct ofconn *ofconn, uint16_t controller_id)
1025 ofconn->controller_id = controller_id;
1028 /* Returns the default miss send length for 'ofconn'. */
1030 ofconn_get_miss_send_len(const struct ofconn *ofconn)
1032 return ofconn->miss_send_len;
1035 /* Sets the default miss send length for 'ofconn' to 'miss_send_len'. */
1037 ofconn_set_miss_send_len(struct ofconn *ofconn, int miss_send_len)
1039 ofconn->miss_send_len = miss_send_len;
1043 ofconn_set_async_config(struct ofconn *ofconn,
1044 const struct ofputil_async_cfg *ac)
1046 if (!ofconn->async_cfg) {
1047 ofconn->async_cfg = xmalloc(sizeof *ofconn->async_cfg);
1049 *ofconn->async_cfg = *ac;
1052 struct ofputil_async_cfg
1053 ofconn_get_async_config(const struct ofconn *ofconn)
1055 if (ofconn->async_cfg) {
1056 return *ofconn->async_cfg;
1059 int version = rconn_get_version(ofconn->rconn);
1060 return (version < 0 || !ofconn->enable_async_msgs
1061 ? OFPUTIL_ASYNC_CFG_INIT
1062 : ofputil_async_cfg_default(version));
1065 /* Sends 'msg' on 'ofconn', accounting it as a reply. (If there is a
1066 * sufficient number of OpenFlow replies in-flight on a single ofconn, then the
1067 * connmgr will stop accepting new OpenFlow requests on that ofconn until the
1068 * controller has accepted some of the replies.) */
1070 ofconn_send_reply(const struct ofconn *ofconn, struct ofpbuf *msg)
1072 ofconn_send(ofconn, msg, ofconn->reply_counter);
1075 /* Sends each of the messages in list 'replies' on 'ofconn' in order,
1076 * accounting them as replies. */
1078 ofconn_send_replies(const struct ofconn *ofconn, struct ovs_list *replies)
1080 struct ofpbuf *reply;
1082 LIST_FOR_EACH_POP (reply, list_node, replies) {
1083 ofconn_send_reply(ofconn, reply);
1087 /* Sends 'error' on 'ofconn', as a reply to 'request'. Only at most the
1088 * first 64 bytes of 'request' are used. */
1090 ofconn_send_error(const struct ofconn *ofconn,
1091 const struct ofp_header *request, enum ofperr error)
1093 static struct vlog_rate_limit err_rl = VLOG_RATE_LIMIT_INIT(10, 10);
1094 struct ofpbuf *reply;
1096 reply = ofperr_encode_reply(error, request);
1097 if (!VLOG_DROP_INFO(&err_rl)) {
1098 const char *type_name;
1102 request_len = ntohs(request->length);
1103 type_name = (!ofpraw_decode_partial(&raw, request,
1104 MIN(64, request_len))
1105 ? ofpraw_get_name(raw)
1108 VLOG_INFO("%s: sending %s error reply to %s message",
1109 rconn_get_name(ofconn->rconn), ofperr_to_string(error),
1112 ofconn_send_reply(ofconn, reply);
1115 /* Same as pktbuf_retrieve(), using the pktbuf owned by 'ofconn'. */
1117 ofconn_pktbuf_retrieve(struct ofconn *ofconn, uint32_t id,
1118 struct dp_packet **bufferp, ofp_port_t *in_port)
1120 return pktbuf_retrieve(ofconn->pktbuf, id, bufferp, in_port);
1123 /* Reports that a flow_mod operation of the type specified by 'command' was
1124 * successfully executed by 'ofconn', so that the connmgr can log it. */
1126 ofconn_report_flow_mod(struct ofconn *ofconn,
1127 enum ofp_flow_mod_command command)
1137 case OFPFC_MODIFY_STRICT:
1142 case OFPFC_DELETE_STRICT:
1148 if (ofconn->next_op_report == LLONG_MAX) {
1149 ofconn->first_op = now;
1150 ofconn->next_op_report = MAX(now + 10 * 1000, ofconn->op_backoff);
1151 ofconn->op_backoff = ofconn->next_op_report + 60 * 1000;
1153 ofconn->last_op = now;
1156 /* OpenFlow 1.4 bundles. */
1158 static inline uint32_t
1159 bundle_hash(uint32_t id)
1161 return hash_int(id, 0);
1165 ofconn_get_bundle(struct ofconn *ofconn, uint32_t id)
1167 struct ofp_bundle *bundle;
1169 HMAP_FOR_EACH_IN_BUCKET(bundle, node, bundle_hash(id), &ofconn->bundles) {
1170 if (bundle->id == id) {
1179 ofconn_insert_bundle(struct ofconn *ofconn, struct ofp_bundle *bundle)
1181 /* XXX: Check the limit of open bundles */
1183 hmap_insert(&ofconn->bundles, &bundle->node, bundle_hash(bundle->id));
1189 ofconn_remove_bundle(struct ofconn *ofconn, struct ofp_bundle *bundle)
1191 hmap_remove(&ofconn->bundles, &bundle->node);
1197 bundle_remove_all(struct ofconn *ofconn)
1199 struct ofp_bundle *b, *next;
1201 HMAP_FOR_EACH_SAFE (b, next, node, &ofconn->bundles) {
1202 ofp_bundle_remove__(ofconn, b, false);
1206 /* Private ofconn functions. */
1209 ofconn_get_target(const struct ofconn *ofconn)
1211 return rconn_get_target(ofconn->rconn);
1214 static struct ofconn *
1215 ofconn_create(struct connmgr *mgr, struct rconn *rconn, enum ofconn_type type,
1216 bool enable_async_msgs)
1218 struct ofconn *ofconn;
1220 ofconn = xzalloc(sizeof *ofconn);
1221 ofconn->connmgr = mgr;
1222 ovs_list_push_back(&mgr->all_conns, &ofconn->node);
1223 ofconn->rconn = rconn;
1224 ofconn->type = type;
1225 ofconn->enable_async_msgs = enable_async_msgs;
1227 hmap_init(&ofconn->monitors);
1228 ovs_list_init(&ofconn->updates);
1230 hmap_init(&ofconn->bundles);
1232 ofconn_flush(ofconn);
1237 /* Clears all of the state in 'ofconn' that should not persist from one
1238 * connection to the next. */
1240 ofconn_flush(struct ofconn *ofconn)
1241 OVS_REQUIRES(ofproto_mutex)
1243 struct ofmonitor *monitor, *next_monitor;
1246 ofconn_log_flow_mods(ofconn);
1248 ofconn->role = OFPCR12_ROLE_EQUAL;
1249 ofconn_set_protocol(ofconn, OFPUTIL_P_NONE);
1250 ofconn->packet_in_format = NXPIF_STANDARD;
1252 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1253 ofconn->packet_in_counter = rconn_packet_counter_create();
1254 for (i = 0; i < N_SCHEDULERS; i++) {
1255 if (ofconn->schedulers[i]) {
1258 pinsched_get_limits(ofconn->schedulers[i], &rate, &burst);
1259 pinsched_destroy(ofconn->schedulers[i]);
1260 ofconn->schedulers[i] = pinsched_create(rate, burst);
1263 if (ofconn->pktbuf) {
1264 pktbuf_destroy(ofconn->pktbuf);
1265 ofconn->pktbuf = pktbuf_create();
1267 ofconn->miss_send_len = (ofconn->type == OFCONN_PRIMARY
1268 ? OFP_DEFAULT_MISS_SEND_LEN
1270 ofconn->controller_id = 0;
1272 rconn_packet_counter_destroy(ofconn->reply_counter);
1273 ofconn->reply_counter = rconn_packet_counter_create();
1275 free(ofconn->async_cfg);
1276 ofconn->async_cfg = NULL;
1278 ofconn->n_add = ofconn->n_delete = ofconn->n_modify = 0;
1279 ofconn->first_op = ofconn->last_op = LLONG_MIN;
1280 ofconn->next_op_report = LLONG_MAX;
1281 ofconn->op_backoff = LLONG_MIN;
1283 HMAP_FOR_EACH_SAFE (monitor, next_monitor, ofconn_node,
1284 &ofconn->monitors) {
1285 ofmonitor_destroy(monitor);
1287 rconn_packet_counter_destroy(ofconn->monitor_counter);
1288 ofconn->monitor_counter = rconn_packet_counter_create();
1289 ofpbuf_list_delete(&ofconn->updates); /* ...but it should be empty. */
1293 ofconn_destroy(struct ofconn *ofconn)
1294 OVS_REQUIRES(ofproto_mutex)
1296 ofconn_flush(ofconn);
1298 if (ofconn->type == OFCONN_PRIMARY) {
1299 hmap_remove(&ofconn->connmgr->controllers, &ofconn->hmap_node);
1302 bundle_remove_all(ofconn);
1303 hmap_destroy(&ofconn->bundles);
1305 hmap_destroy(&ofconn->monitors);
1306 ovs_list_remove(&ofconn->node);
1307 rconn_destroy(ofconn->rconn);
1308 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1309 rconn_packet_counter_destroy(ofconn->reply_counter);
1310 pktbuf_destroy(ofconn->pktbuf);
1311 rconn_packet_counter_destroy(ofconn->monitor_counter);
1315 /* Reconfigures 'ofconn' to match 'c'. 'ofconn' and 'c' must have the same
1318 ofconn_reconfigure(struct ofconn *ofconn, const struct ofproto_controller *c)
1322 ofconn->band = c->band;
1323 ofconn->enable_async_msgs = c->enable_async_msgs;
1325 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
1327 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
1328 rconn_set_probe_interval(ofconn->rconn, probe_interval);
1330 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
1332 /* If dscp value changed reconnect. */
1333 if (c->dscp != rconn_get_dscp(ofconn->rconn)) {
1334 rconn_set_dscp(ofconn->rconn, c->dscp);
1335 rconn_reconnect(ofconn->rconn);
1339 /* Returns true if it makes sense for 'ofconn' to receive and process OpenFlow
1342 ofconn_may_recv(const struct ofconn *ofconn)
1344 int count = rconn_packet_counter_n_packets(ofconn->reply_counter);
1345 return count < OFCONN_REPLY_MAX;
1349 ofconn_run(struct ofconn *ofconn,
1350 void (*handle_openflow)(struct ofconn *,
1351 const struct ofpbuf *ofp_msg))
1353 struct connmgr *mgr = ofconn->connmgr;
1356 for (i = 0; i < N_SCHEDULERS; i++) {
1357 struct ovs_list txq;
1359 pinsched_run(ofconn->schedulers[i], &txq);
1360 do_send_packet_ins(ofconn, &txq);
1363 rconn_run(ofconn->rconn);
1365 /* Limit the number of iterations to avoid starving other tasks. */
1366 for (i = 0; i < 50 && ofconn_may_recv(ofconn); i++) {
1367 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1372 if (mgr->fail_open) {
1373 fail_open_maybe_recover(mgr->fail_open);
1376 handle_openflow(ofconn, of_msg);
1377 ofpbuf_delete(of_msg);
1380 if (time_msec() >= ofconn->next_op_report) {
1381 ofconn_log_flow_mods(ofconn);
1384 ovs_mutex_lock(&ofproto_mutex);
1385 if (!rconn_is_alive(ofconn->rconn)) {
1386 ofconn_destroy(ofconn);
1387 } else if (!rconn_is_connected(ofconn->rconn)) {
1388 ofconn_flush(ofconn);
1390 ovs_mutex_unlock(&ofproto_mutex);
1394 ofconn_wait(struct ofconn *ofconn)
1398 for (i = 0; i < N_SCHEDULERS; i++) {
1399 pinsched_wait(ofconn->schedulers[i]);
1401 rconn_run_wait(ofconn->rconn);
1402 if (ofconn_may_recv(ofconn)) {
1403 rconn_recv_wait(ofconn->rconn);
1405 if (ofconn->next_op_report != LLONG_MAX) {
1406 poll_timer_wait_until(ofconn->next_op_report);
1411 ofconn_log_flow_mods(struct ofconn *ofconn)
1413 int n_flow_mods = ofconn->n_add + ofconn->n_delete + ofconn->n_modify;
1415 long long int ago = (time_msec() - ofconn->first_op) / 1000;
1416 long long int interval = (ofconn->last_op - ofconn->first_op) / 1000;
1420 ds_put_format(&s, "%d flow_mods ", n_flow_mods);
1421 if (interval == ago) {
1422 ds_put_format(&s, "in the last %lld s", ago);
1423 } else if (interval) {
1424 ds_put_format(&s, "in the %lld s starting %lld s ago",
1427 ds_put_format(&s, "%lld s ago", ago);
1430 ds_put_cstr(&s, " (");
1431 if (ofconn->n_add) {
1432 ds_put_format(&s, "%d adds, ", ofconn->n_add);
1434 if (ofconn->n_delete) {
1435 ds_put_format(&s, "%d deletes, ", ofconn->n_delete);
1437 if (ofconn->n_modify) {
1438 ds_put_format(&s, "%d modifications, ", ofconn->n_modify);
1441 ds_put_char(&s, ')');
1443 VLOG_INFO("%s: %s", rconn_get_name(ofconn->rconn), ds_cstr(&s));
1446 ofconn->n_add = ofconn->n_delete = ofconn->n_modify = 0;
1448 ofconn->next_op_report = LLONG_MAX;
1451 /* Returns true if 'ofconn' should receive asynchronous messages of the given
1452 * OAM_* 'type' and 'reason', which should be a OFPR_* value for OAM_PACKET_IN,
1453 * a OFPPR_* value for OAM_PORT_STATUS, or an OFPRR_* value for
1454 * OAM_FLOW_REMOVED. Returns false if the message should not be sent on
1457 ofconn_receives_async_msg(const struct ofconn *ofconn,
1458 enum ofputil_async_msg_type type,
1459 unsigned int reason)
1461 ovs_assert(reason < 32);
1462 ovs_assert((unsigned int) type < OAM_N_TYPES);
1464 /* Keep the following code in sync with the documentation in the
1465 * "Asynchronous Messages" section in DESIGN. */
1467 if (ofconn->type == OFCONN_SERVICE && !ofconn->miss_send_len) {
1468 /* Service connections don't get asynchronous messages unless they have
1469 * explicitly asked for them by setting a nonzero miss send length. */
1473 struct ofputil_async_cfg ac = ofconn_get_async_config(ofconn);
1474 uint32_t *masks = (ofconn->role == OFPCR12_ROLE_SLAVE
1477 return (masks[type] & (1u << reason)) != 0;
1480 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
1481 * packet rather than to send the packet to the controller.
1483 * This function returns true to indicate that a packet_in message
1484 * for a "table-miss" should be sent to at least one controller.
1485 * That is there is at least one controller with controller_id 0
1486 * which connected using an OpenFlow version earlier than OpenFlow1.3.
1490 * This logic assumes that "table-miss" packet_in messages
1491 * are always sent to controller_id 0. */
1493 connmgr_wants_packet_in_on_miss(struct connmgr *mgr) OVS_EXCLUDED(ofproto_mutex)
1495 struct ofconn *ofconn;
1497 ovs_mutex_lock(&ofproto_mutex);
1498 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1499 enum ofputil_protocol protocol = ofconn_get_protocol(ofconn);
1501 if (ofconn->controller_id == 0 &&
1502 (protocol == OFPUTIL_P_NONE ||
1503 ofputil_protocol_to_ofp_version(protocol) < OFP13_VERSION)) {
1504 ovs_mutex_unlock(&ofproto_mutex);
1508 ovs_mutex_unlock(&ofproto_mutex);
1513 /* Returns a human-readable name for an OpenFlow connection between 'mgr' and
1514 * 'target', suitable for use in log messages for identifying the connection.
1516 * The name is dynamically allocated. The caller should free it (with free())
1517 * when it is no longer needed. */
1519 ofconn_make_name(const struct connmgr *mgr, const char *target)
1521 return xasprintf("%s<->%s", mgr->name, target);
1525 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1529 for (i = 0; i < N_SCHEDULERS; i++) {
1530 struct pinsched **s = &ofconn->schedulers[i];
1534 *s = pinsched_create(rate, burst);
1536 pinsched_set_limits(*s, rate, burst);
1539 pinsched_destroy(*s);
1546 ofconn_send(const struct ofconn *ofconn, struct ofpbuf *msg,
1547 struct rconn_packet_counter *counter)
1549 ofpmsg_update_length(msg);
1550 rconn_send(ofconn->rconn, msg, counter);
1553 /* Sending asynchronous messages. */
1555 /* Sends an OFPT_PORT_STATUS message with 'opp' and 'reason' to appropriate
1556 * controllers managed by 'mgr'. For messages caused by a controller
1557 * OFPT_PORT_MOD, specify 'source' as the controller connection that sent the
1558 * request; otherwise, specify 'source' as NULL. */
1560 connmgr_send_port_status(struct connmgr *mgr, struct ofconn *source,
1561 const struct ofputil_phy_port *pp, uint8_t reason)
1563 /* XXX Should limit the number of queued port status change messages. */
1564 struct ofputil_port_status ps;
1565 struct ofconn *ofconn;
1569 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1570 if (ofconn_receives_async_msg(ofconn, OAM_PORT_STATUS, reason)) {
1573 /* Before 1.5, OpenFlow specified that OFPT_PORT_MOD should not
1574 * generate OFPT_PORT_STATUS messages. That requirement was a
1575 * relic of how OpenFlow originally supported a single controller,
1576 * so that one could expect the controller to already know the
1577 * changes it had made.
1579 * EXT-338 changes OpenFlow 1.5 OFPT_PORT_MOD to send
1580 * OFPT_PORT_STATUS messages to every controller. This is
1581 * obviously more useful in the multi-controller case. We could
1582 * always implement it that way in OVS, but that would risk
1583 * confusing controllers that are intended for single-controller
1584 * use only. (Imagine a controller that generates an OFPT_PORT_MOD
1585 * in response to any OFPT_PORT_STATUS!)
1587 * So this compromises: for OpenFlow 1.4 and earlier, it generates
1588 * OFPT_PORT_STATUS for OFPT_PORT_MOD, but not back to the
1589 * originating controller. In a single-controller environment, in
1590 * particular, this means that it will never generate
1591 * OFPT_PORT_STATUS for OFPT_PORT_MOD at all. */
1592 if (ofconn == source
1593 && rconn_get_version(ofconn->rconn) < OFP15_VERSION) {
1597 msg = ofputil_encode_port_status(&ps, ofconn_get_protocol(ofconn));
1598 ofconn_send(ofconn, msg, NULL);
1603 /* Sends an OFPT_REQUESTFORWARD message with 'request' and 'reason' to
1604 * appropriate controllers managed by 'mgr'. For messages caused by a
1605 * controller OFPT_GROUP_MOD and OFPT_METER_MOD, specify 'source' as the
1606 * controller connection that sent the request; otherwise, specify 'source'
1609 connmgr_send_requestforward(struct connmgr *mgr, const struct ofconn *source,
1610 const struct ofputil_requestforward *rf)
1612 struct ofconn *ofconn;
1614 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1615 if (ofconn_receives_async_msg(ofconn, OAM_REQUESTFORWARD, rf->reason)
1616 && rconn_get_version(ofconn->rconn) >= OFP14_VERSION
1617 && ofconn != source) {
1618 enum ofputil_protocol protocol = ofconn_get_protocol(ofconn);
1619 ofconn_send(ofconn, ofputil_encode_requestforward(rf, protocol),
1625 /* Sends an OFPT_FLOW_REMOVED or NXT_FLOW_REMOVED message based on 'fr' to
1626 * appropriate controllers managed by 'mgr'. */
1628 connmgr_send_flow_removed(struct connmgr *mgr,
1629 const struct ofputil_flow_removed *fr)
1631 struct ofconn *ofconn;
1633 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1634 if (ofconn_receives_async_msg(ofconn, OAM_FLOW_REMOVED, fr->reason)) {
1637 /* Account flow expirations as replies to OpenFlow requests. That
1638 * works because preventing OpenFlow requests from being processed
1639 * also prevents new flows from being added (and expiring). (It
1640 * also prevents processing OpenFlow requests that would not add
1641 * new flows, so it is imperfect.) */
1642 msg = ofputil_encode_flow_removed(fr, ofconn_get_protocol(ofconn));
1643 ofconn_send_reply(ofconn, msg);
1648 /* Sends an OFPT_TABLE_STATUS message with 'reason' to appropriate controllers
1649 * managed by 'mgr'. When the table state changes, the controller needs to be
1650 * informed with the OFPT_TABLE_STATUS message. The reason values
1651 * OFPTR_VACANCY_DOWN and OFPTR_VACANCY_UP identify a vacancy message. The
1652 * vacancy events are generated when the remaining space in the flow table
1653 * changes and crosses one of the vacancy thereshold specified by
1654 * OFPT_TABLE_MOD. */
1656 connmgr_send_table_status(struct connmgr *mgr,
1657 const struct ofputil_table_desc *td,
1660 struct ofputil_table_status ts;
1661 struct ofconn *ofconn;
1666 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1667 if (ofconn_receives_async_msg(ofconn, OAM_TABLE_STATUS, reason)) {
1670 msg = ofputil_encode_table_status(&ts,
1671 ofconn_get_protocol(ofconn));
1673 ofconn_send(ofconn, msg, NULL);
1679 /* Given 'pin', sends an OFPT_PACKET_IN message to each OpenFlow controller as
1680 * necessary according to their individual configurations. */
1682 connmgr_send_async_msg(struct connmgr *mgr,
1683 const struct ofproto_async_msg *am)
1685 struct ofconn *ofconn;
1687 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1688 enum ofputil_protocol protocol = ofconn_get_protocol(ofconn);
1689 if (protocol == OFPUTIL_P_NONE || !rconn_is_connected(ofconn->rconn)
1690 || ofconn->controller_id != am->controller_id
1691 || !ofconn_receives_async_msg(ofconn, am->oam,
1692 am->pin.up.public.reason)) {
1696 struct ofpbuf *msg = ofputil_encode_packet_in_private(
1697 &am->pin.up, protocol, ofconn->packet_in_format,
1698 am->pin.max_len >= 0 ? am->pin.max_len : ofconn->miss_send_len,
1701 struct ovs_list txq;
1702 bool is_miss = (am->pin.up.public.reason == OFPR_NO_MATCH ||
1703 am->pin.up.public.reason == OFPR_EXPLICIT_MISS ||
1704 am->pin.up.public.reason == OFPR_IMPLICIT_MISS);
1705 pinsched_send(ofconn->schedulers[is_miss],
1706 am->pin.up.public.flow_metadata.flow.in_port.ofp_port,
1708 do_send_packet_ins(ofconn, &txq);
1713 do_send_packet_ins(struct ofconn *ofconn, struct ovs_list *txq)
1717 LIST_FOR_EACH_POP (pin, list_node, txq) {
1718 if (rconn_send_with_limit(ofconn->rconn, pin,
1719 ofconn->packet_in_counter, 100) == EAGAIN) {
1720 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5);
1722 VLOG_INFO_RL(&rl, "%s: dropping packet-in due to queue overflow",
1723 rconn_get_name(ofconn->rconn));
1728 /* Fail-open settings. */
1730 /* Returns the failure handling mode (OFPROTO_FAIL_SECURE or
1731 * OFPROTO_FAIL_STANDALONE) for 'mgr'. */
1732 enum ofproto_fail_mode
1733 connmgr_get_fail_mode(const struct connmgr *mgr)
1735 return mgr->fail_mode;
1738 /* Sets the failure handling mode for 'mgr' to 'fail_mode' (either
1739 * OFPROTO_FAIL_SECURE or OFPROTO_FAIL_STANDALONE). */
1741 connmgr_set_fail_mode(struct connmgr *mgr, enum ofproto_fail_mode fail_mode)
1743 if (mgr->fail_mode != fail_mode) {
1744 mgr->fail_mode = fail_mode;
1745 update_fail_open(mgr);
1746 if (!connmgr_has_controllers(mgr)) {
1747 ofproto_flush_flows(mgr->ofproto);
1752 /* Fail-open implementation. */
1754 /* Returns the longest probe interval among the primary controllers configured
1755 * on 'mgr'. Returns 0 if there are no primary controllers. */
1757 connmgr_get_max_probe_interval(const struct connmgr *mgr)
1759 const struct ofconn *ofconn;
1760 int max_probe_interval;
1762 max_probe_interval = 0;
1763 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1764 int probe_interval = rconn_get_probe_interval(ofconn->rconn);
1765 max_probe_interval = MAX(max_probe_interval, probe_interval);
1767 return max_probe_interval;
1770 /* Returns the number of seconds for which all of 'mgr's primary controllers
1771 * have been disconnected. Returns 0 if 'mgr' has no primary controllers. */
1773 connmgr_failure_duration(const struct connmgr *mgr)
1775 const struct ofconn *ofconn;
1776 int min_failure_duration;
1778 if (!connmgr_has_controllers(mgr)) {
1782 min_failure_duration = INT_MAX;
1783 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1784 int failure_duration = rconn_failure_duration(ofconn->rconn);
1785 min_failure_duration = MIN(min_failure_duration, failure_duration);
1787 return min_failure_duration;
1790 /* Returns true if at least one primary controller is connected (regardless of
1791 * whether those controllers are believed to have authenticated and accepted
1792 * this switch), false if none of them are connected. */
1794 connmgr_is_any_controller_connected(const struct connmgr *mgr)
1796 const struct ofconn *ofconn;
1798 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1799 if (rconn_is_connected(ofconn->rconn)) {
1806 /* Returns true if at least one primary controller is believed to have
1807 * authenticated and accepted this switch, false otherwise. */
1809 connmgr_is_any_controller_admitted(const struct connmgr *mgr)
1811 const struct ofconn *ofconn;
1813 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1814 if (rconn_is_admitted(ofconn->rconn)) {
1821 /* In-band configuration. */
1823 static bool any_extras_changed(const struct connmgr *,
1824 const struct sockaddr_in *extras, size_t n);
1826 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'mgr''s
1827 * in-band control should guarantee access, in the same way that in-band
1828 * control guarantees access to OpenFlow controllers. */
1830 connmgr_set_extra_in_band_remotes(struct connmgr *mgr,
1831 const struct sockaddr_in *extras, size_t n)
1833 if (!any_extras_changed(mgr, extras, n)) {
1837 free(mgr->extra_in_band_remotes);
1838 mgr->n_extra_remotes = n;
1839 mgr->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
1841 update_in_band_remotes(mgr);
1844 /* Sets the OpenFlow queue used by flows set up by in-band control on
1845 * 'mgr' to 'queue_id'. If 'queue_id' is negative, then in-band control
1846 * flows will use the default queue. */
1848 connmgr_set_in_band_queue(struct connmgr *mgr, int queue_id)
1850 if (queue_id != mgr->in_band_queue) {
1851 mgr->in_band_queue = queue_id;
1852 update_in_band_remotes(mgr);
1857 any_extras_changed(const struct connmgr *mgr,
1858 const struct sockaddr_in *extras, size_t n)
1862 if (n != mgr->n_extra_remotes) {
1866 for (i = 0; i < n; i++) {
1867 const struct sockaddr_in *old = &mgr->extra_in_band_remotes[i];
1868 const struct sockaddr_in *new = &extras[i];
1870 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
1871 old->sin_port != new->sin_port) {
1879 /* In-band implementation. */
1882 connmgr_has_in_band(struct connmgr *mgr)
1884 return mgr->in_band != NULL;
1887 /* Fail-open and in-band implementation. */
1889 /* Called by 'ofproto' after all flows have been flushed, to allow fail-open
1890 * and standalone mode to re-create their flows.
1892 * In-band control has more sophisticated code that manages flows itself. */
1894 connmgr_flushed(struct connmgr *mgr)
1895 OVS_EXCLUDED(ofproto_mutex)
1897 if (mgr->fail_open) {
1898 fail_open_flushed(mgr->fail_open);
1901 /* If there are no controllers and we're in standalone mode, set up a flow
1902 * that matches every packet and directs them to OFPP_NORMAL (which goes to
1903 * us). Otherwise, the switch is in secure mode and we won't pass any
1904 * traffic until a controller has been defined and it tells us to do so. */
1905 if (!connmgr_has_controllers(mgr)
1906 && mgr->fail_mode == OFPROTO_FAIL_STANDALONE) {
1907 struct ofpbuf ofpacts;
1910 ofpbuf_init(&ofpacts, OFPACT_OUTPUT_SIZE);
1911 ofpact_put_OUTPUT(&ofpacts)->port = OFPP_NORMAL;
1913 match_init_catchall(&match);
1914 ofproto_add_flow(mgr->ofproto, &match, 0, ofpacts.data,
1917 ofpbuf_uninit(&ofpacts);
1921 /* Returns the number of hidden rules created by the in-band and fail-open
1922 * implementations in table 0. (Subtracting this count from the number of
1923 * rules in the table 0 classifier, as maintained in struct oftable, yields
1924 * the number of flows that OVS should report via OpenFlow for table 0.) */
1926 connmgr_count_hidden_rules(const struct connmgr *mgr)
1930 n_hidden += in_band_count_rules(mgr->in_band);
1932 if (mgr->fail_open) {
1933 n_hidden += fail_open_count_rules(mgr->fail_open);
1938 /* Creates a new ofservice for 'target' in 'mgr'. Returns 0 if successful,
1939 * otherwise a positive errno value.
1941 * ofservice_reconfigure() must be called to fully configure the new
1944 ofservice_create(struct connmgr *mgr, const char *target,
1945 uint32_t allowed_versions, uint8_t dscp)
1947 struct ofservice *ofservice;
1948 struct pvconn *pvconn;
1951 error = pvconn_open(target, allowed_versions, dscp, &pvconn);
1956 ofservice = xzalloc(sizeof *ofservice);
1957 hmap_insert(&mgr->services, &ofservice->node, hash_string(target, 0));
1958 ofservice->pvconn = pvconn;
1959 ofservice->allowed_versions = allowed_versions;
1965 ofservice_destroy(struct connmgr *mgr, struct ofservice *ofservice)
1967 hmap_remove(&mgr->services, &ofservice->node);
1968 pvconn_close(ofservice->pvconn);
1973 ofservice_reconfigure(struct ofservice *ofservice,
1974 const struct ofproto_controller *c)
1976 ofservice->probe_interval = c->probe_interval;
1977 ofservice->rate_limit = c->rate_limit;
1978 ofservice->burst_limit = c->burst_limit;
1979 ofservice->enable_async_msgs = c->enable_async_msgs;
1980 ofservice->dscp = c->dscp;
1983 /* Finds and returns the ofservice within 'mgr' that has the given
1984 * 'target', or a null pointer if none exists. */
1985 static struct ofservice *
1986 ofservice_lookup(struct connmgr *mgr, const char *target)
1988 struct ofservice *ofservice;
1990 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1992 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1999 /* Flow monitors (NXST_FLOW_MONITOR). */
2001 /* A counter incremented when something significant happens to an OpenFlow
2004 * - When a rule is added, its 'add_seqno' and 'modify_seqno' are set to
2005 * the current value (which is then incremented).
2007 * - When a rule is modified, its 'modify_seqno' is set to the current
2008 * value (which is then incremented).
2010 * Thus, by comparing an old value of monitor_seqno against a rule's
2011 * 'add_seqno', one can tell whether the rule was added before or after the old
2012 * value was read, and similarly for 'modify_seqno'.
2014 * 32 bits should normally be sufficient (and would be nice, to save space in
2015 * each rule) but then we'd have to have some special cases for wraparound.
2017 * We initialize monitor_seqno to 1 to allow 0 to be used as an invalid
2019 static uint64_t monitor_seqno = 1;
2021 COVERAGE_DEFINE(ofmonitor_pause);
2022 COVERAGE_DEFINE(ofmonitor_resume);
2025 ofmonitor_create(const struct ofputil_flow_monitor_request *request,
2026 struct ofconn *ofconn, struct ofmonitor **monitorp)
2027 OVS_REQUIRES(ofproto_mutex)
2029 struct ofmonitor *m;
2033 m = ofmonitor_lookup(ofconn, request->id);
2035 return OFPERR_OFPMOFC_MONITOR_EXISTS;
2038 m = xmalloc(sizeof *m);
2040 hmap_insert(&ofconn->monitors, &m->ofconn_node, hash_int(request->id, 0));
2041 m->id = request->id;
2042 m->flags = request->flags;
2043 m->out_port = request->out_port;
2044 m->table_id = request->table_id;
2045 minimatch_init(&m->match, &request->match);
2052 ofmonitor_lookup(struct ofconn *ofconn, uint32_t id)
2053 OVS_REQUIRES(ofproto_mutex)
2055 struct ofmonitor *m;
2057 HMAP_FOR_EACH_IN_BUCKET (m, ofconn_node, hash_int(id, 0),
2058 &ofconn->monitors) {
2067 ofmonitor_destroy(struct ofmonitor *m)
2068 OVS_REQUIRES(ofproto_mutex)
2071 minimatch_destroy(&m->match);
2072 hmap_remove(&m->ofconn->monitors, &m->ofconn_node);
2078 ofmonitor_report(struct connmgr *mgr, struct rule *rule,
2079 enum nx_flow_update_event event,
2080 enum ofp_flow_removed_reason reason,
2081 const struct ofconn *abbrev_ofconn, ovs_be32 abbrev_xid,
2082 const struct rule_actions *old_actions)
2083 OVS_REQUIRES(ofproto_mutex)
2085 enum nx_flow_monitor_flags update;
2086 struct ofconn *ofconn;
2088 if (rule_is_hidden(rule)) {
2095 rule->add_seqno = rule->modify_seqno = monitor_seqno++;
2099 update = NXFMF_DELETE;
2102 case NXFME_MODIFIED:
2103 update = NXFMF_MODIFY;
2104 rule->modify_seqno = monitor_seqno++;
2112 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2113 enum nx_flow_monitor_flags flags = 0;
2114 struct ofmonitor *m;
2116 if (ofconn->monitor_paused) {
2117 /* Only send NXFME_DELETED notifications for flows that were added
2118 * before we paused. */
2119 if (event != NXFME_DELETED
2120 || rule->add_seqno > ofconn->monitor_paused) {
2125 HMAP_FOR_EACH (m, ofconn_node, &ofconn->monitors) {
2126 if (m->flags & update
2127 && (m->table_id == 0xff || m->table_id == rule->table_id)
2128 && (ofproto_rule_has_out_port(rule, m->out_port)
2130 && ofpacts_output_to_port(old_actions->ofpacts,
2131 old_actions->ofpacts_len,
2133 && cls_rule_is_loose_match(&rule->cr, &m->match)) {
2139 if (ovs_list_is_empty(&ofconn->updates)) {
2140 ofputil_start_flow_update(&ofconn->updates);
2141 ofconn->sent_abbrev_update = false;
2144 if (flags & NXFMF_OWN || ofconn != abbrev_ofconn
2145 || ofconn->monitor_paused) {
2146 struct ofputil_flow_update fu;
2150 fu.reason = event == NXFME_DELETED ? reason : 0;
2151 fu.table_id = rule->table_id;
2152 fu.cookie = rule->flow_cookie;
2153 minimatch_expand(&rule->cr.match, &match);
2155 fu.priority = rule->cr.priority;
2157 ovs_mutex_lock(&rule->mutex);
2158 fu.idle_timeout = rule->idle_timeout;
2159 fu.hard_timeout = rule->hard_timeout;
2160 ovs_mutex_unlock(&rule->mutex);
2162 if (flags & NXFMF_ACTIONS) {
2163 const struct rule_actions *actions = rule_get_actions(rule);
2164 fu.ofpacts = actions->ofpacts;
2165 fu.ofpacts_len = actions->ofpacts_len;
2170 ofputil_append_flow_update(&fu, &ofconn->updates);
2171 } else if (!ofconn->sent_abbrev_update) {
2172 struct ofputil_flow_update fu;
2174 fu.event = NXFME_ABBREV;
2175 fu.xid = abbrev_xid;
2176 ofputil_append_flow_update(&fu, &ofconn->updates);
2178 ofconn->sent_abbrev_update = true;
2185 ofmonitor_flush(struct connmgr *mgr)
2186 OVS_REQUIRES(ofproto_mutex)
2188 struct ofconn *ofconn;
2190 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2193 LIST_FOR_EACH_POP (msg, list_node, &ofconn->updates) {
2194 unsigned int n_bytes;
2196 ofconn_send(ofconn, msg, ofconn->monitor_counter);
2197 n_bytes = rconn_packet_counter_n_bytes(ofconn->monitor_counter);
2198 if (!ofconn->monitor_paused && n_bytes > 128 * 1024) {
2199 struct ofpbuf *pause;
2201 COVERAGE_INC(ofmonitor_pause);
2202 ofconn->monitor_paused = monitor_seqno++;
2203 pause = ofpraw_alloc_xid(OFPRAW_NXT_FLOW_MONITOR_PAUSED,
2204 OFP10_VERSION, htonl(0), 0);
2205 ofconn_send(ofconn, pause, ofconn->monitor_counter);
2212 ofmonitor_resume(struct ofconn *ofconn)
2213 OVS_REQUIRES(ofproto_mutex)
2215 struct rule_collection rules;
2216 struct ofpbuf *resumed;
2217 struct ofmonitor *m;
2218 struct ovs_list msgs;
2220 rule_collection_init(&rules);
2221 HMAP_FOR_EACH (m, ofconn_node, &ofconn->monitors) {
2222 ofmonitor_collect_resume_rules(m, ofconn->monitor_paused, &rules);
2225 ovs_list_init(&msgs);
2226 ofmonitor_compose_refresh_updates(&rules, &msgs);
2228 resumed = ofpraw_alloc_xid(OFPRAW_NXT_FLOW_MONITOR_RESUMED, OFP10_VERSION,
2230 ovs_list_push_back(&msgs, &resumed->list_node);
2231 ofconn_send_replies(ofconn, &msgs);
2233 ofconn->monitor_paused = 0;
2237 ofmonitor_may_resume(const struct ofconn *ofconn)
2238 OVS_REQUIRES(ofproto_mutex)
2240 return (ofconn->monitor_paused != 0
2241 && !rconn_packet_counter_n_packets(ofconn->monitor_counter));
2245 ofmonitor_run(struct connmgr *mgr)
2247 struct ofconn *ofconn;
2249 ovs_mutex_lock(&ofproto_mutex);
2250 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2251 if (ofmonitor_may_resume(ofconn)) {
2252 COVERAGE_INC(ofmonitor_resume);
2253 ofmonitor_resume(ofconn);
2256 ovs_mutex_unlock(&ofproto_mutex);
2260 ofmonitor_wait(struct connmgr *mgr)
2262 struct ofconn *ofconn;
2264 ovs_mutex_lock(&ofproto_mutex);
2265 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2266 if (ofmonitor_may_resume(ofconn)) {
2267 poll_immediate_wake();
2270 ovs_mutex_unlock(&ofproto_mutex);
2274 ofproto_async_msg_free(struct ofproto_async_msg *am)
2276 free(am->pin.up.public.packet);
2277 free(am->pin.up.public.userdata);
2278 free(am->pin.up.stack);
2279 free(am->pin.up.actions);
2280 free(am->pin.up.action_set);