2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
25 #include "dynamic-string.h"
26 #include "fail-open.h"
29 #include "ofp-actions.h"
33 #include "ofproto-provider.h"
35 #include "poll-loop.h"
47 VLOG_DEFINE_THIS_MODULE(connmgr);
48 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
50 /* An OpenFlow connection.
56 * 'ofproto_mutex' must be held whenever an ofconn is created or destroyed or,
57 * more or less equivalently, whenever an ofconn is added to or removed from a
58 * connmgr. 'ofproto_mutex' doesn't protect the data inside the ofconn, except
59 * as specifically noted below. */
61 /* Configuration that persists from one connection to the next. */
63 struct list node; /* In struct connmgr's "all_conns" list. */
64 struct hmap_node hmap_node; /* In struct connmgr's "controllers" map. */
66 struct connmgr *connmgr; /* Connection's manager. */
67 struct rconn *rconn; /* OpenFlow connection. */
68 enum ofconn_type type; /* Type. */
69 enum ofproto_band band; /* In-band or out-of-band? */
70 bool enable_async_msgs; /* Initially enable async messages? */
72 /* State that should be cleared from one connection to the next. */
75 enum ofp12_controller_role role; /* Role. */
76 enum ofputil_protocol protocol; /* Current protocol variant. */
77 enum nx_packet_in_format packet_in_format; /* OFPT_PACKET_IN format. */
79 /* OFPT_PACKET_IN related data. */
80 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
81 #define N_SCHEDULERS 2
82 struct pinsched *schedulers[N_SCHEDULERS];
83 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
84 int miss_send_len; /* Bytes to send of buffered packets. */
85 uint16_t controller_id; /* Connection controller ID. */
87 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
88 * requests, and the maximum number before we stop reading OpenFlow
90 #define OFCONN_REPLY_MAX 100
91 struct rconn_packet_counter *reply_counter;
93 /* Asynchronous message configuration in each possible roles.
95 * A 1-bit enables sending an asynchronous message for one possible reason
96 * that the message might be generated, a 0-bit disables it. */
97 uint32_t master_async_config[OAM_N_TYPES]; /* master, other */
98 uint32_t slave_async_config[OAM_N_TYPES]; /* slave */
100 /* Flow table operation logging. */
101 int n_add, n_delete, n_modify; /* Number of unreported ops of each kind. */
102 long long int first_op, last_op; /* Range of times for unreported ops. */
103 long long int next_op_report; /* Time to report ops, or LLONG_MAX. */
104 long long int op_backoff; /* Earliest time to report ops again. */
106 /* Flow monitors (e.g. NXST_FLOW_MONITOR). */
108 /* Configuration. Contains "struct ofmonitor"s. */
109 struct hmap monitors OVS_GUARDED_BY(ofproto_mutex);
113 * When too many flow monitor notifications back up in the transmit buffer,
114 * we pause the transmission of further notifications. These members track
115 * the flow control state.
117 * When notifications are flowing, 'monitor_paused' is 0. When
118 * notifications are paused, 'monitor_paused' is the value of
119 * 'monitor_seqno' at the point we paused.
121 * 'monitor_counter' counts the OpenFlow messages and bytes currently in
122 * flight. This value growing too large triggers pausing. */
123 uint64_t monitor_paused OVS_GUARDED_BY(ofproto_mutex);
124 struct rconn_packet_counter *monitor_counter OVS_GUARDED_BY(ofproto_mutex);
126 /* State of monitors for a single ongoing flow_mod.
128 * 'updates' is a list of "struct ofpbuf"s that contain
129 * NXST_FLOW_MONITOR_REPLY messages representing the changes made by the
132 * When 'updates' is nonempty, 'sent_abbrev_update' is true if 'updates'
133 * contains an update event of type NXFME_ABBREV and false otherwise.. */
134 struct list updates OVS_GUARDED_BY(ofproto_mutex);
135 bool sent_abbrev_update OVS_GUARDED_BY(ofproto_mutex);
137 /* Active bundles. Contains "struct ofp_bundle"s. */
141 static struct ofconn *ofconn_create(struct connmgr *, struct rconn *,
142 enum ofconn_type, bool enable_async_msgs)
143 OVS_REQUIRES(ofproto_mutex);
144 static void ofconn_destroy(struct ofconn *) OVS_REQUIRES(ofproto_mutex);
145 static void ofconn_flush(struct ofconn *) OVS_REQUIRES(ofproto_mutex);
147 static void ofconn_reconfigure(struct ofconn *,
148 const struct ofproto_controller *);
150 static void ofconn_run(struct ofconn *,
151 void (*handle_openflow)(struct ofconn *,
152 const struct ofpbuf *ofp_msg));
153 static void ofconn_wait(struct ofconn *);
155 static void ofconn_log_flow_mods(struct ofconn *);
157 static const char *ofconn_get_target(const struct ofconn *);
158 static char *ofconn_make_name(const struct connmgr *, const char *target);
160 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
162 static void ofconn_send(const struct ofconn *, struct ofpbuf *,
163 struct rconn_packet_counter *);
165 static void do_send_packet_ins(struct ofconn *, struct list *txq);
167 /* A listener for incoming OpenFlow "service" connections. */
169 struct hmap_node node; /* In struct connmgr's "services" hmap. */
170 struct pvconn *pvconn; /* OpenFlow connection listener. */
172 /* These are not used by ofservice directly. They are settings for
173 * accepted "struct ofconn"s from the pvconn. */
174 int probe_interval; /* Max idle time before probing, in seconds. */
175 int rate_limit; /* Max packet-in rate in packets per second. */
176 int burst_limit; /* Limit on accumulating packet credits. */
177 bool enable_async_msgs; /* Initially enable async messages? */
178 uint8_t dscp; /* DSCP Value for controller connection */
179 uint32_t allowed_versions; /* OpenFlow protocol versions that may
180 * be negotiated for a session. */
183 static void ofservice_reconfigure(struct ofservice *,
184 const struct ofproto_controller *);
185 static int ofservice_create(struct connmgr *mgr, const char *target,
186 uint32_t allowed_versions, uint8_t dscp);
187 static void ofservice_destroy(struct connmgr *, struct ofservice *);
188 static struct ofservice *ofservice_lookup(struct connmgr *,
191 /* Connection manager for an OpenFlow switch. */
193 struct ofproto *ofproto;
195 char *local_port_name;
197 /* OpenFlow connections. */
198 struct hmap controllers; /* Controller "struct ofconn"s. */
199 struct list all_conns; /* Contains "struct ofconn"s. */
200 uint64_t master_election_id; /* monotonically increasing sequence number
201 * for master election */
202 bool master_election_id_defined;
204 /* OpenFlow listeners. */
205 struct hmap services; /* Contains "struct ofservice"s. */
206 struct pvconn **snoops;
210 struct fail_open *fail_open;
211 enum ofproto_fail_mode fail_mode;
213 /* In-band control. */
214 struct in_band *in_band;
215 struct sockaddr_in *extra_in_band_remotes;
216 size_t n_extra_remotes;
220 static void update_in_band_remotes(struct connmgr *);
221 static void add_snooper(struct connmgr *, struct vconn *);
222 static void ofmonitor_run(struct connmgr *);
223 static void ofmonitor_wait(struct connmgr *);
225 /* Creates and returns a new connection manager owned by 'ofproto'. 'name' is
226 * a name for the ofproto suitable for using in log messages.
227 * 'local_port_name' is the name of the local port (OFPP_LOCAL) within
230 connmgr_create(struct ofproto *ofproto,
231 const char *name, const char *local_port_name)
235 mgr = xmalloc(sizeof *mgr);
236 mgr->ofproto = ofproto;
237 mgr->name = xstrdup(name);
238 mgr->local_port_name = xstrdup(local_port_name);
240 hmap_init(&mgr->controllers);
241 list_init(&mgr->all_conns);
242 mgr->master_election_id = 0;
243 mgr->master_election_id_defined = false;
245 hmap_init(&mgr->services);
249 mgr->fail_open = NULL;
250 mgr->fail_mode = OFPROTO_FAIL_SECURE;
253 mgr->extra_in_band_remotes = NULL;
254 mgr->n_extra_remotes = 0;
255 mgr->in_band_queue = -1;
260 /* Frees 'mgr' and all of its resources. */
262 connmgr_destroy(struct connmgr *mgr)
264 struct ofservice *ofservice, *next_ofservice;
265 struct ofconn *ofconn, *next_ofconn;
272 ovs_mutex_lock(&ofproto_mutex);
273 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &mgr->all_conns) {
274 ofconn_destroy(ofconn);
276 ovs_mutex_unlock(&ofproto_mutex);
278 hmap_destroy(&mgr->controllers);
280 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &mgr->services) {
281 ofservice_destroy(mgr, ofservice);
283 hmap_destroy(&mgr->services);
285 for (i = 0; i < mgr->n_snoops; i++) {
286 pvconn_close(mgr->snoops[i]);
290 fail_open_destroy(mgr->fail_open);
291 mgr->fail_open = NULL;
293 in_band_destroy(mgr->in_band);
295 free(mgr->extra_in_band_remotes);
297 free(mgr->local_port_name);
302 /* Does all of the periodic maintenance required by 'mgr'. Calls
303 * 'handle_openflow' for each message received on an OpenFlow connection,
304 * passing along the OpenFlow connection itself and the message that was sent.
305 * 'handle_openflow' must not modify or free the message. */
307 connmgr_run(struct connmgr *mgr,
308 void (*handle_openflow)(struct ofconn *,
309 const struct ofpbuf *ofp_msg))
310 OVS_EXCLUDED(ofproto_mutex)
312 struct ofconn *ofconn, *next_ofconn;
313 struct ofservice *ofservice;
317 if (!in_band_run(mgr->in_band)) {
318 in_band_destroy(mgr->in_band);
323 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &mgr->all_conns) {
324 ofconn_run(ofconn, handle_openflow);
328 /* Fail-open maintenance. Do this after processing the ofconns since
329 * fail-open checks the status of the controller rconn. */
330 if (mgr->fail_open) {
331 fail_open_run(mgr->fail_open);
334 HMAP_FOR_EACH (ofservice, node, &mgr->services) {
338 retval = pvconn_accept(ofservice->pvconn, &vconn);
343 /* Passing default value for creation of the rconn */
344 rconn = rconn_create(ofservice->probe_interval, 0, ofservice->dscp,
345 vconn_get_allowed_versions(vconn));
346 name = ofconn_make_name(mgr, vconn_get_name(vconn));
347 rconn_connect_unreliably(rconn, vconn, name);
350 ovs_mutex_lock(&ofproto_mutex);
351 ofconn = ofconn_create(mgr, rconn, OFCONN_SERVICE,
352 ofservice->enable_async_msgs);
353 ovs_mutex_unlock(&ofproto_mutex);
355 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
356 ofservice->burst_limit);
357 } else if (retval != EAGAIN) {
358 VLOG_WARN_RL(&rl, "accept failed (%s)", ovs_strerror(retval));
362 for (i = 0; i < mgr->n_snoops; i++) {
366 retval = pvconn_accept(mgr->snoops[i], &vconn);
368 add_snooper(mgr, vconn);
369 } else if (retval != EAGAIN) {
370 VLOG_WARN_RL(&rl, "accept failed (%s)", ovs_strerror(retval));
375 /* Causes the poll loop to wake up when connmgr_run() needs to run. */
377 connmgr_wait(struct connmgr *mgr)
379 struct ofservice *ofservice;
380 struct ofconn *ofconn;
383 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
388 in_band_wait(mgr->in_band);
390 if (mgr->fail_open) {
391 fail_open_wait(mgr->fail_open);
393 HMAP_FOR_EACH (ofservice, node, &mgr->services) {
394 pvconn_wait(ofservice->pvconn);
396 for (i = 0; i < mgr->n_snoops; i++) {
397 pvconn_wait(mgr->snoops[i]);
401 /* Adds some memory usage statistics for 'mgr' into 'usage', for use with
402 * memory_report(). */
404 connmgr_get_memory_usage(const struct connmgr *mgr, struct simap *usage)
406 const struct ofconn *ofconn;
407 unsigned int packets = 0;
408 unsigned int ofconns = 0;
410 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
415 packets += rconn_count_txqlen(ofconn->rconn);
416 for (i = 0; i < N_SCHEDULERS; i++) {
417 packets += pinsched_count_txqlen(ofconn->schedulers[i]);
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);
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));
502 connmgr_free_controller_info(struct shash *info)
504 struct shash_node *node;
506 SHASH_FOR_EACH (node, info) {
507 struct ofproto_controller_info *cinfo = node->data;
508 smap_destroy(&cinfo->pairs);
514 /* Changes 'mgr''s set of controllers to the 'n_controllers' controllers in
517 connmgr_set_controllers(struct connmgr *mgr,
518 const struct ofproto_controller *controllers,
519 size_t n_controllers, uint32_t allowed_versions)
520 OVS_EXCLUDED(ofproto_mutex)
522 bool had_controllers = connmgr_has_controllers(mgr);
523 struct shash new_controllers;
524 struct ofconn *ofconn, *next_ofconn;
525 struct ofservice *ofservice, *next_ofservice;
528 /* Required to add and remove ofconns. This could probably be narrowed to
529 * cover a smaller amount of code, if that yielded some benefit. */
530 ovs_mutex_lock(&ofproto_mutex);
532 /* Create newly configured controllers and services.
533 * Create a name to ofproto_controller mapping in 'new_controllers'. */
534 shash_init(&new_controllers);
535 for (i = 0; i < n_controllers; i++) {
536 const struct ofproto_controller *c = &controllers[i];
538 if (!vconn_verify_name(c->target)) {
540 ofconn = find_controller_by_target(mgr, c->target);
542 VLOG_INFO("%s: added primary controller \"%s\"",
543 mgr->name, c->target);
545 } else if (rconn_get_allowed_versions(ofconn->rconn) !=
547 VLOG_INFO("%s: re-added primary controller \"%s\"",
548 mgr->name, c->target);
550 ofconn_destroy(ofconn);
553 add_controller(mgr, c->target, c->dscp, allowed_versions);
555 } else if (!pvconn_verify_name(c->target)) {
557 ofservice = ofservice_lookup(mgr, c->target);
559 VLOG_INFO("%s: added service controller \"%s\"",
560 mgr->name, c->target);
562 } else if (ofservice->allowed_versions != allowed_versions) {
563 VLOG_INFO("%s: re-added service controller \"%s\"",
564 mgr->name, c->target);
565 ofservice_destroy(mgr, ofservice);
569 ofservice_create(mgr, c->target, allowed_versions, c->dscp);
572 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
573 mgr->name, c->target);
577 shash_add_once(&new_controllers, c->target, &controllers[i]);
580 /* Delete controllers that are no longer configured.
581 * Update configuration of all now-existing controllers. */
582 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &mgr->controllers) {
583 const char *target = ofconn_get_target(ofconn);
584 struct ofproto_controller *c;
586 c = shash_find_data(&new_controllers, target);
588 VLOG_INFO("%s: removed primary controller \"%s\"",
590 ofconn_destroy(ofconn);
592 ofconn_reconfigure(ofconn, c);
596 /* Delete services that are no longer configured.
597 * Update configuration of all now-existing services. */
598 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &mgr->services) {
599 const char *target = pvconn_get_name(ofservice->pvconn);
600 struct ofproto_controller *c;
602 c = shash_find_data(&new_controllers, target);
604 VLOG_INFO("%s: removed service controller \"%s\"",
606 ofservice_destroy(mgr, ofservice);
608 ofservice_reconfigure(ofservice, c);
612 shash_destroy(&new_controllers);
614 ovs_mutex_unlock(&ofproto_mutex);
616 update_in_band_remotes(mgr);
617 update_fail_open(mgr);
618 if (had_controllers != connmgr_has_controllers(mgr)) {
619 ofproto_flush_flows(mgr->ofproto);
623 /* Drops the connections between 'mgr' and all of its primary and secondary
624 * controllers, forcing them to reconnect. */
626 connmgr_reconnect(const struct connmgr *mgr)
628 struct ofconn *ofconn;
630 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
631 rconn_reconnect(ofconn->rconn);
635 /* Sets the "snoops" for 'mgr' to the pvconn targets listed in 'snoops'.
637 * A "snoop" is a pvconn to which every OpenFlow message to or from the most
638 * important controller on 'mgr' is mirrored. */
640 connmgr_set_snoops(struct connmgr *mgr, const struct sset *snoops)
642 return set_pvconns(&mgr->snoops, &mgr->n_snoops, snoops);
645 /* Adds each of the snoops currently configured on 'mgr' to 'snoops'. */
647 connmgr_get_snoops(const struct connmgr *mgr, struct sset *snoops)
651 for (i = 0; i < mgr->n_snoops; i++) {
652 sset_add(snoops, pvconn_get_name(mgr->snoops[i]));
656 /* Returns true if 'mgr' has at least one snoop, false if it has none. */
658 connmgr_has_snoops(const struct connmgr *mgr)
660 return mgr->n_snoops > 0;
663 /* Creates a new controller for 'target' in 'mgr'. update_controller() needs
664 * to be called later to finish the new ofconn's configuration. */
666 add_controller(struct connmgr *mgr, const char *target, uint8_t dscp,
667 uint32_t allowed_versions)
668 OVS_REQUIRES(ofproto_mutex)
670 char *name = ofconn_make_name(mgr, target);
671 struct ofconn *ofconn;
673 ofconn = ofconn_create(mgr, rconn_create(5, 8, dscp, allowed_versions),
674 OFCONN_PRIMARY, true);
675 ofconn->pktbuf = pktbuf_create();
676 rconn_connect(ofconn->rconn, target, name);
677 hmap_insert(&mgr->controllers, &ofconn->hmap_node, hash_string(target, 0));
682 static struct ofconn *
683 find_controller_by_target(struct connmgr *mgr, const char *target)
685 struct ofconn *ofconn;
687 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
688 hash_string(target, 0), &mgr->controllers) {
689 if (!strcmp(ofconn_get_target(ofconn), target)) {
697 update_in_band_remotes(struct connmgr *mgr)
699 struct sockaddr_in *addrs;
700 size_t max_addrs, n_addrs;
701 struct ofconn *ofconn;
704 /* Allocate enough memory for as many remotes as we could possibly have. */
705 max_addrs = mgr->n_extra_remotes + hmap_count(&mgr->controllers);
706 addrs = xmalloc(max_addrs * sizeof *addrs);
709 /* Add all the remotes. */
710 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
711 const char *target = rconn_get_target(ofconn->rconn);
712 struct sockaddr_storage ss;
714 if (ofconn->band == OFPROTO_IN_BAND
715 && stream_parse_target_with_default_port(target, OFP_OLD_PORT, &ss)
716 && ss.ss_family == AF_INET) {
717 addrs[n_addrs++] = *(struct sockaddr_in *) &ss;
720 for (i = 0; i < mgr->n_extra_remotes; i++) {
721 addrs[n_addrs++] = mgr->extra_in_band_remotes[i];
724 /* Create or update or destroy in-band. */
727 in_band_create(mgr->ofproto, mgr->local_port_name, &mgr->in_band);
729 in_band_set_queue(mgr->in_band, mgr->in_band_queue);
731 /* in_band_run() needs a chance to delete any existing in-band flows.
732 * We will destroy mgr->in_band after it's done with that. */
735 in_band_set_remotes(mgr->in_band, addrs, n_addrs);
743 update_fail_open(struct connmgr *mgr)
744 OVS_EXCLUDED(ofproto_mutex)
746 if (connmgr_has_controllers(mgr)
747 && mgr->fail_mode == OFPROTO_FAIL_STANDALONE) {
748 if (!mgr->fail_open) {
749 mgr->fail_open = fail_open_create(mgr->ofproto, mgr);
752 fail_open_destroy(mgr->fail_open);
753 mgr->fail_open = NULL;
758 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
759 const struct sset *sset)
761 struct pvconn **pvconns = *pvconnsp;
762 size_t n_pvconns = *n_pvconnsp;
767 for (i = 0; i < n_pvconns; i++) {
768 pvconn_close(pvconns[i]);
772 pvconns = xmalloc(sset_count(sset) * sizeof *pvconns);
774 SSET_FOR_EACH (name, sset) {
775 struct pvconn *pvconn;
777 error = pvconn_open(name, 0, 0, &pvconn);
779 pvconns[n_pvconns++] = pvconn;
781 VLOG_ERR("failed to listen on %s: %s", name, ovs_strerror(error));
789 *n_pvconnsp = n_pvconns;
794 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
795 * means that 'ofconn' is more interesting for monitoring than a lower return
798 snoop_preference(const struct ofconn *ofconn)
800 switch (ofconn->role) {
801 case OFPCR12_ROLE_MASTER:
803 case OFPCR12_ROLE_EQUAL:
805 case OFPCR12_ROLE_SLAVE:
807 case OFPCR12_ROLE_NOCHANGE:
809 /* Shouldn't happen. */
814 /* One of 'mgr''s "snoop" pvconns has accepted a new connection on 'vconn'.
815 * Connects this vconn to a controller. */
817 add_snooper(struct connmgr *mgr, struct vconn *vconn)
819 struct ofconn *ofconn, *best;
821 /* Pick a controller for monitoring. */
823 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
824 if (ofconn->type == OFCONN_PRIMARY
825 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
831 rconn_add_monitor(best->rconn, vconn);
833 VLOG_INFO_RL(&rl, "no controller connection to snoop");
838 /* Public ofconn functions. */
840 /* Returns the connection type, either OFCONN_PRIMARY or OFCONN_SERVICE. */
842 ofconn_get_type(const struct ofconn *ofconn)
847 /* If a master election id is defined, stores it into '*idp' and returns
848 * true. Otherwise, stores UINT64_MAX into '*idp' and returns false. */
850 ofconn_get_master_election_id(const struct ofconn *ofconn, uint64_t *idp)
852 *idp = (ofconn->connmgr->master_election_id_defined
853 ? ofconn->connmgr->master_election_id
855 return ofconn->connmgr->master_election_id_defined;
858 /* Sets the master election id.
860 * Returns true if successful, false if the id is stale
863 ofconn_set_master_election_id(struct ofconn *ofconn, uint64_t id)
865 if (ofconn->connmgr->master_election_id_defined
867 /* Unsigned difference interpreted as a two's complement signed
869 (int64_t)(id - ofconn->connmgr->master_election_id) < 0) {
872 ofconn->connmgr->master_election_id = id;
873 ofconn->connmgr->master_election_id_defined = true;
878 /* Returns the role configured for 'ofconn'.
880 * The default role, if no other role has been set, is OFPCR12_ROLE_EQUAL. */
881 enum ofp12_controller_role
882 ofconn_get_role(const struct ofconn *ofconn)
888 ofconn_send_role_status(struct ofconn *ofconn, uint32_t role, uint8_t reason)
890 struct ofputil_role_status status;
893 status.reason = reason;
895 ofconn_get_master_election_id(ofconn, &status.generation_id);
897 buf = ofputil_encode_role_status(&status, ofconn_get_protocol(ofconn));
899 ofconn_send(ofconn, buf, NULL);
902 /* Changes 'ofconn''s role to 'role'. If 'role' is OFPCR12_ROLE_MASTER then
903 * any existing master is demoted to a slave. */
905 ofconn_set_role(struct ofconn *ofconn, enum ofp12_controller_role role)
907 if (role != ofconn->role && role == OFPCR12_ROLE_MASTER) {
908 struct ofconn *other;
910 HMAP_FOR_EACH (other, hmap_node, &ofconn->connmgr->controllers) {
911 if (other->role == OFPCR12_ROLE_MASTER) {
912 other->role = OFPCR12_ROLE_SLAVE;
913 ofconn_send_role_status(other, OFPCR12_ROLE_SLAVE, OFPCRR_MASTER_REQUEST);
921 ofconn_set_invalid_ttl_to_controller(struct ofconn *ofconn, bool enable)
923 uint32_t bit = 1u << OFPR_INVALID_TTL;
925 ofconn->master_async_config[OAM_PACKET_IN] |= bit;
927 ofconn->master_async_config[OAM_PACKET_IN] &= ~bit;
932 ofconn_get_invalid_ttl_to_controller(struct ofconn *ofconn)
934 uint32_t bit = 1u << OFPR_INVALID_TTL;
935 return (ofconn->master_async_config[OAM_PACKET_IN] & bit) != 0;
938 /* Returns the currently configured protocol for 'ofconn', one of OFPUTIL_P_*.
940 * Returns OFPUTIL_P_NONE, which is not a valid protocol, if 'ofconn' hasn't
941 * completed version negotiation. This can't happen if at least one OpenFlow
942 * message, other than OFPT_HELLO, has been received on the connection (such as
943 * in ofproto.c's message handling code), since version negotiation is a
944 * prerequisite for starting to receive messages. This means that
945 * OFPUTIL_P_NONE is a special case that most callers need not worry about. */
946 enum ofputil_protocol
947 ofconn_get_protocol(const struct ofconn *ofconn)
949 if (ofconn->protocol == OFPUTIL_P_NONE &&
950 rconn_is_connected(ofconn->rconn)) {
951 int version = rconn_get_version(ofconn->rconn);
953 ofconn_set_protocol(CONST_CAST(struct ofconn *, ofconn),
954 ofputil_protocol_from_ofp_version(version));
958 return ofconn->protocol;
961 /* Sets the protocol for 'ofconn' to 'protocol' (one of OFPUTIL_P_*).
963 * (This doesn't actually send anything to accomplish this. Presumably the
964 * caller already did that.) */
966 ofconn_set_protocol(struct ofconn *ofconn, enum ofputil_protocol protocol)
968 ofconn->protocol = protocol;
971 /* Returns the currently configured packet in format for 'ofconn', one of
974 * The default, if no other format has been set, is NXPIF_OPENFLOW10. */
975 enum nx_packet_in_format
976 ofconn_get_packet_in_format(struct ofconn *ofconn)
978 return ofconn->packet_in_format;
981 /* Sets the packet in format for 'ofconn' to 'packet_in_format' (one of
984 ofconn_set_packet_in_format(struct ofconn *ofconn,
985 enum nx_packet_in_format packet_in_format)
987 ofconn->packet_in_format = packet_in_format;
990 /* Sets the controller connection ID for 'ofconn' to 'controller_id'.
992 * The connection controller ID is used for OFPP_CONTROLLER and
993 * NXAST_CONTROLLER actions. See "struct nx_action_controller" for details. */
995 ofconn_set_controller_id(struct ofconn *ofconn, uint16_t controller_id)
997 ofconn->controller_id = controller_id;
1000 /* Returns the default miss send length for 'ofconn'. */
1002 ofconn_get_miss_send_len(const struct ofconn *ofconn)
1004 return ofconn->miss_send_len;
1007 /* Sets the default miss send length for 'ofconn' to 'miss_send_len'. */
1009 ofconn_set_miss_send_len(struct ofconn *ofconn, int miss_send_len)
1011 ofconn->miss_send_len = miss_send_len;
1015 ofconn_set_async_config(struct ofconn *ofconn,
1016 const uint32_t master_masks[OAM_N_TYPES],
1017 const uint32_t slave_masks[OAM_N_TYPES])
1019 size_t size = sizeof ofconn->master_async_config;
1020 memcpy(ofconn->master_async_config, master_masks, size);
1021 memcpy(ofconn->slave_async_config, slave_masks, size);
1025 ofconn_get_async_config(struct ofconn *ofconn,
1026 uint32_t *master_masks, uint32_t *slave_masks)
1028 size_t size = sizeof ofconn->master_async_config;
1029 memcpy(master_masks, ofconn->master_async_config, size);
1030 memcpy(slave_masks, ofconn->slave_async_config, size);
1033 /* Sends 'msg' on 'ofconn', accounting it as a reply. (If there is a
1034 * sufficient number of OpenFlow replies in-flight on a single ofconn, then the
1035 * connmgr will stop accepting new OpenFlow requests on that ofconn until the
1036 * controller has accepted some of the replies.) */
1038 ofconn_send_reply(const struct ofconn *ofconn, struct ofpbuf *msg)
1040 ofconn_send(ofconn, msg, ofconn->reply_counter);
1043 /* Sends each of the messages in list 'replies' on 'ofconn' in order,
1044 * accounting them as replies. */
1046 ofconn_send_replies(const struct ofconn *ofconn, struct list *replies)
1048 struct ofpbuf *reply, *next;
1050 LIST_FOR_EACH_SAFE (reply, next, list_node, replies) {
1051 list_remove(&reply->list_node);
1052 ofconn_send_reply(ofconn, reply);
1056 /* Sends 'error' on 'ofconn', as a reply to 'request'. Only at most the
1057 * first 64 bytes of 'request' are used. */
1059 ofconn_send_error(const struct ofconn *ofconn,
1060 const struct ofp_header *request, enum ofperr error)
1062 static struct vlog_rate_limit err_rl = VLOG_RATE_LIMIT_INIT(10, 10);
1063 struct ofpbuf *reply;
1065 reply = ofperr_encode_reply(error, request);
1066 if (!VLOG_DROP_INFO(&err_rl)) {
1067 const char *type_name;
1071 request_len = ntohs(request->length);
1072 type_name = (!ofpraw_decode_partial(&raw, request,
1073 MIN(64, request_len))
1074 ? ofpraw_get_name(raw)
1077 VLOG_INFO("%s: sending %s error reply to %s message",
1078 rconn_get_name(ofconn->rconn), ofperr_to_string(error),
1081 ofconn_send_reply(ofconn, reply);
1084 /* Same as pktbuf_retrieve(), using the pktbuf owned by 'ofconn'. */
1086 ofconn_pktbuf_retrieve(struct ofconn *ofconn, uint32_t id,
1087 struct ofpbuf **bufferp, ofp_port_t *in_port)
1089 return pktbuf_retrieve(ofconn->pktbuf, id, bufferp, in_port);
1092 /* Reports that a flow_mod operation of the type specified by 'command' was
1093 * successfully executed by 'ofconn', so that the connmgr can log it. */
1095 ofconn_report_flow_mod(struct ofconn *ofconn,
1096 enum ofp_flow_mod_command command)
1106 case OFPFC_MODIFY_STRICT:
1111 case OFPFC_DELETE_STRICT:
1117 if (ofconn->next_op_report == LLONG_MAX) {
1118 ofconn->first_op = now;
1119 ofconn->next_op_report = MAX(now + 10 * 1000, ofconn->op_backoff);
1120 ofconn->op_backoff = ofconn->next_op_report + 60 * 1000;
1122 ofconn->last_op = now;
1126 ofconn_get_bundles(struct ofconn *ofconn)
1128 return &ofconn->bundles;
1132 /* Private ofconn functions. */
1135 ofconn_get_target(const struct ofconn *ofconn)
1137 return rconn_get_target(ofconn->rconn);
1140 static struct ofconn *
1141 ofconn_create(struct connmgr *mgr, struct rconn *rconn, enum ofconn_type type,
1142 bool enable_async_msgs)
1144 struct ofconn *ofconn;
1146 ofconn = xzalloc(sizeof *ofconn);
1147 ofconn->connmgr = mgr;
1148 list_push_back(&mgr->all_conns, &ofconn->node);
1149 ofconn->rconn = rconn;
1150 ofconn->type = type;
1151 ofconn->enable_async_msgs = enable_async_msgs;
1153 hmap_init(&ofconn->monitors);
1154 list_init(&ofconn->updates);
1156 hmap_init(&ofconn->bundles);
1158 ofconn_flush(ofconn);
1163 /* Clears all of the state in 'ofconn' that should not persist from one
1164 * connection to the next. */
1166 ofconn_flush(struct ofconn *ofconn)
1167 OVS_REQUIRES(ofproto_mutex)
1169 struct ofmonitor *monitor, *next_monitor;
1172 ofconn_log_flow_mods(ofconn);
1174 ofconn->role = OFPCR12_ROLE_EQUAL;
1175 ofconn_set_protocol(ofconn, OFPUTIL_P_NONE);
1176 ofconn->packet_in_format = NXPIF_OPENFLOW10;
1178 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1179 ofconn->packet_in_counter = rconn_packet_counter_create();
1180 for (i = 0; i < N_SCHEDULERS; i++) {
1181 if (ofconn->schedulers[i]) {
1184 pinsched_get_limits(ofconn->schedulers[i], &rate, &burst);
1185 pinsched_destroy(ofconn->schedulers[i]);
1186 ofconn->schedulers[i] = pinsched_create(rate, burst);
1189 if (ofconn->pktbuf) {
1190 pktbuf_destroy(ofconn->pktbuf);
1191 ofconn->pktbuf = pktbuf_create();
1193 ofconn->miss_send_len = (ofconn->type == OFCONN_PRIMARY
1194 ? OFP_DEFAULT_MISS_SEND_LEN
1196 ofconn->controller_id = 0;
1198 rconn_packet_counter_destroy(ofconn->reply_counter);
1199 ofconn->reply_counter = rconn_packet_counter_create();
1201 if (ofconn->enable_async_msgs) {
1202 uint32_t *master = ofconn->master_async_config;
1203 uint32_t *slave = ofconn->slave_async_config;
1205 /* "master" and "other" roles get all asynchronous messages by default,
1206 * except that the controller needs to enable nonstandard "packet-in"
1207 * reasons itself. */
1208 master[OAM_PACKET_IN] = (1u << OFPR_NO_MATCH) | (1u << OFPR_ACTION);
1209 master[OAM_PORT_STATUS] = ((1u << OFPPR_ADD)
1210 | (1u << OFPPR_DELETE)
1211 | (1u << OFPPR_MODIFY));
1212 master[OAM_FLOW_REMOVED] = ((1u << OFPRR_IDLE_TIMEOUT)
1213 | (1u << OFPRR_HARD_TIMEOUT)
1214 | (1u << OFPRR_DELETE));
1216 /* "slave" role gets port status updates by default. */
1217 slave[OAM_PACKET_IN] = 0;
1218 slave[OAM_PORT_STATUS] = ((1u << OFPPR_ADD)
1219 | (1u << OFPPR_DELETE)
1220 | (1u << OFPPR_MODIFY));
1221 slave[OAM_FLOW_REMOVED] = 0;
1223 memset(ofconn->master_async_config, 0,
1224 sizeof ofconn->master_async_config);
1225 memset(ofconn->slave_async_config, 0,
1226 sizeof ofconn->slave_async_config);
1229 ofconn->n_add = ofconn->n_delete = ofconn->n_modify = 0;
1230 ofconn->first_op = ofconn->last_op = LLONG_MIN;
1231 ofconn->next_op_report = LLONG_MAX;
1232 ofconn->op_backoff = LLONG_MIN;
1234 HMAP_FOR_EACH_SAFE (monitor, next_monitor, ofconn_node,
1235 &ofconn->monitors) {
1236 ofmonitor_destroy(monitor);
1238 rconn_packet_counter_destroy(ofconn->monitor_counter);
1239 ofconn->monitor_counter = rconn_packet_counter_create();
1240 ofpbuf_list_delete(&ofconn->updates); /* ...but it should be empty. */
1244 ofconn_destroy(struct ofconn *ofconn)
1245 OVS_REQUIRES(ofproto_mutex)
1247 ofconn_flush(ofconn);
1249 if (ofconn->type == OFCONN_PRIMARY) {
1250 hmap_remove(&ofconn->connmgr->controllers, &ofconn->hmap_node);
1253 ofp_bundle_remove_all(ofconn);
1255 hmap_destroy(&ofconn->monitors);
1256 list_remove(&ofconn->node);
1257 rconn_destroy(ofconn->rconn);
1258 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1259 rconn_packet_counter_destroy(ofconn->reply_counter);
1260 pktbuf_destroy(ofconn->pktbuf);
1261 rconn_packet_counter_destroy(ofconn->monitor_counter);
1265 /* Reconfigures 'ofconn' to match 'c'. 'ofconn' and 'c' must have the same
1268 ofconn_reconfigure(struct ofconn *ofconn, const struct ofproto_controller *c)
1272 ofconn->band = c->band;
1273 ofconn->enable_async_msgs = c->enable_async_msgs;
1275 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
1277 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
1278 rconn_set_probe_interval(ofconn->rconn, probe_interval);
1280 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
1282 /* If dscp value changed reconnect. */
1283 if (c->dscp != rconn_get_dscp(ofconn->rconn)) {
1284 rconn_set_dscp(ofconn->rconn, c->dscp);
1285 rconn_reconnect(ofconn->rconn);
1289 /* Returns true if it makes sense for 'ofconn' to receive and process OpenFlow
1292 ofconn_may_recv(const struct ofconn *ofconn)
1294 int count = rconn_packet_counter_n_packets(ofconn->reply_counter);
1295 return count < OFCONN_REPLY_MAX;
1299 ofconn_run(struct ofconn *ofconn,
1300 void (*handle_openflow)(struct ofconn *,
1301 const struct ofpbuf *ofp_msg))
1303 struct connmgr *mgr = ofconn->connmgr;
1306 for (i = 0; i < N_SCHEDULERS; i++) {
1309 pinsched_run(ofconn->schedulers[i], &txq);
1310 do_send_packet_ins(ofconn, &txq);
1313 rconn_run(ofconn->rconn);
1315 /* Limit the number of iterations to avoid starving other tasks. */
1316 for (i = 0; i < 50 && ofconn_may_recv(ofconn); i++) {
1317 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1322 if (mgr->fail_open) {
1323 fail_open_maybe_recover(mgr->fail_open);
1326 handle_openflow(ofconn, of_msg);
1327 ofpbuf_delete(of_msg);
1330 if (time_msec() >= ofconn->next_op_report) {
1331 ofconn_log_flow_mods(ofconn);
1334 ovs_mutex_lock(&ofproto_mutex);
1335 if (!rconn_is_alive(ofconn->rconn)) {
1336 ofconn_destroy(ofconn);
1337 } else if (!rconn_is_connected(ofconn->rconn)) {
1338 ofconn_flush(ofconn);
1340 ovs_mutex_unlock(&ofproto_mutex);
1344 ofconn_wait(struct ofconn *ofconn)
1348 for (i = 0; i < N_SCHEDULERS; i++) {
1349 pinsched_wait(ofconn->schedulers[i]);
1351 rconn_run_wait(ofconn->rconn);
1352 if (ofconn_may_recv(ofconn)) {
1353 rconn_recv_wait(ofconn->rconn);
1355 if (ofconn->next_op_report != LLONG_MAX) {
1356 poll_timer_wait_until(ofconn->next_op_report);
1361 ofconn_log_flow_mods(struct ofconn *ofconn)
1363 int n_flow_mods = ofconn->n_add + ofconn->n_delete + ofconn->n_modify;
1365 long long int ago = (time_msec() - ofconn->first_op) / 1000;
1366 long long int interval = (ofconn->last_op - ofconn->first_op) / 1000;
1370 ds_put_format(&s, "%d flow_mods ", n_flow_mods);
1371 if (interval == ago) {
1372 ds_put_format(&s, "in the last %lld s", ago);
1373 } else if (interval) {
1374 ds_put_format(&s, "in the %lld s starting %lld s ago",
1377 ds_put_format(&s, "%lld s ago", ago);
1380 ds_put_cstr(&s, " (");
1381 if (ofconn->n_add) {
1382 ds_put_format(&s, "%d adds, ", ofconn->n_add);
1384 if (ofconn->n_delete) {
1385 ds_put_format(&s, "%d deletes, ", ofconn->n_delete);
1387 if (ofconn->n_modify) {
1388 ds_put_format(&s, "%d modifications, ", ofconn->n_modify);
1391 ds_put_char(&s, ')');
1393 VLOG_INFO("%s: %s", rconn_get_name(ofconn->rconn), ds_cstr(&s));
1396 ofconn->n_add = ofconn->n_delete = ofconn->n_modify = 0;
1398 ofconn->next_op_report = LLONG_MAX;
1401 /* Returns true if 'ofconn' should receive asynchronous messages of the given
1402 * OAM_* 'type' and 'reason', which should be a OFPR_* value for OAM_PACKET_IN,
1403 * a OFPPR_* value for OAM_PORT_STATUS, or an OFPRR_* value for
1404 * OAM_FLOW_REMOVED. Returns false if the message should not be sent on
1407 ofconn_receives_async_msg(const struct ofconn *ofconn,
1408 enum ofconn_async_msg_type type,
1409 unsigned int reason)
1411 const uint32_t *async_config;
1413 ovs_assert(reason < 32);
1414 ovs_assert((unsigned int) type < OAM_N_TYPES);
1416 if (ofconn_get_protocol(ofconn) == OFPUTIL_P_NONE
1417 || !rconn_is_connected(ofconn->rconn)) {
1421 /* Keep the following code in sync with the documentation in the
1422 * "Asynchronous Messages" section in DESIGN. */
1424 if (ofconn->type == OFCONN_SERVICE && !ofconn->miss_send_len) {
1425 /* Service connections don't get asynchronous messages unless they have
1426 * explicitly asked for them by setting a nonzero miss send length. */
1430 async_config = (ofconn->role == OFPCR12_ROLE_SLAVE
1431 ? ofconn->slave_async_config
1432 : ofconn->master_async_config);
1433 if (!(async_config[type] & (1u << reason))) {
1440 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
1441 * packet rather than to send the packet to the controller.
1443 * This function returns false to indicate the packet should be dropped if
1444 * the controller action was the result of the default table-miss behaviour
1445 * and the controller is using OpenFlow1.3+.
1447 * Otherwise true is returned to indicate the packet should be forwarded to
1450 ofconn_wants_packet_in_on_miss(struct ofconn *ofconn,
1451 const struct ofproto_packet_in *pin)
1453 if (pin->miss_type == OFPROTO_PACKET_IN_MISS_WITHOUT_FLOW) {
1454 enum ofputil_protocol protocol = ofconn_get_protocol(ofconn);
1456 if (protocol != OFPUTIL_P_NONE
1457 && ofputil_protocol_to_ofp_version(protocol) >= OFP13_VERSION) {
1458 enum ofproto_table_config config;
1460 config = ofproto_table_get_config(ofconn->connmgr->ofproto,
1462 if (config == OFPROTO_TABLE_MISS_DEFAULT) {
1470 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
1471 * packet rather than to send the packet to the controller.
1473 * This function returns false to indicate that a packet_in message
1474 * for a "table-miss" should be sent to at least one controller.
1475 * That is there is at least one controller with controller_id 0
1476 * which connected using an OpenFlow version earlier than OpenFlow1.3.
1480 * This logic assumes that "table-miss" packet_in messages
1481 * are always sent to controller_id 0. */
1483 connmgr_wants_packet_in_on_miss(struct connmgr *mgr)
1485 struct ofconn *ofconn;
1487 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1488 enum ofputil_protocol protocol = ofconn_get_protocol(ofconn);
1490 if (ofconn->controller_id == 0 &&
1491 (protocol == OFPUTIL_P_NONE ||
1492 ofputil_protocol_to_ofp_version(protocol) < OFP13_VERSION)) {
1499 /* Returns a human-readable name for an OpenFlow connection between 'mgr' and
1500 * 'target', suitable for use in log messages for identifying the connection.
1502 * The name is dynamically allocated. The caller should free it (with free())
1503 * when it is no longer needed. */
1505 ofconn_make_name(const struct connmgr *mgr, const char *target)
1507 return xasprintf("%s<->%s", mgr->name, target);
1511 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1515 for (i = 0; i < N_SCHEDULERS; i++) {
1516 struct pinsched **s = &ofconn->schedulers[i];
1520 *s = pinsched_create(rate, burst);
1522 pinsched_set_limits(*s, rate, burst);
1525 pinsched_destroy(*s);
1532 ofconn_send(const struct ofconn *ofconn, struct ofpbuf *msg,
1533 struct rconn_packet_counter *counter)
1535 ofpmsg_update_length(msg);
1536 rconn_send(ofconn->rconn, msg, counter);
1539 /* Sending asynchronous messages. */
1541 static void schedule_packet_in(struct ofconn *, struct ofproto_packet_in,
1542 enum ofp_packet_in_reason wire_reason);
1544 /* Sends an OFPT_PORT_STATUS message with 'opp' and 'reason' to appropriate
1545 * controllers managed by 'mgr'. For messages caused by a controller
1546 * OFPT_PORT_MOD, specify 'source' as the controller connection that sent the
1547 * request; otherwise, specify 'source' as NULL. */
1549 connmgr_send_port_status(struct connmgr *mgr, struct ofconn *source,
1550 const struct ofputil_phy_port *pp, uint8_t reason)
1552 /* XXX Should limit the number of queued port status change messages. */
1553 struct ofputil_port_status ps;
1554 struct ofconn *ofconn;
1558 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1559 if (ofconn_receives_async_msg(ofconn, OAM_PORT_STATUS, reason)) {
1562 /* Before 1.5, OpenFlow specified that OFPT_PORT_MOD should not
1563 * generate OFPT_PORT_STATUS messages. That requirement was a
1564 * relic of how OpenFlow originally supported a single controller,
1565 * so that one could expect the controller to already know the
1566 * changes it had made.
1568 * EXT-338 changes OpenFlow 1.5 OFPT_PORT_MOD to send
1569 * OFPT_PORT_STATUS messages to every controller. This is
1570 * obviously more useful in the multi-controller case. We could
1571 * always implement it that way in OVS, but that would risk
1572 * confusing controllers that are intended for single-controller
1573 * use only. (Imagine a controller that generates an OFPT_PORT_MOD
1574 * in response to any OFPT_PORT_STATUS!)
1576 * So this compromises: for OpenFlow 1.4 and earlier, it generates
1577 * OFPT_PORT_STATUS for OFPT_PORT_MOD, but not back to the
1578 * originating controller. In a single-controller environment, in
1579 * particular, this means that it will never generate
1580 * OFPT_PORT_STATUS for OFPT_PORT_MOD at all. */
1581 if (ofconn == source
1582 && rconn_get_version(ofconn->rconn) < OFP15_VERSION) {
1586 msg = ofputil_encode_port_status(&ps, ofconn_get_protocol(ofconn));
1587 ofconn_send(ofconn, msg, NULL);
1592 /* Sends an OFPT_FLOW_REMOVED or NXT_FLOW_REMOVED message based on 'fr' to
1593 * appropriate controllers managed by 'mgr'. */
1595 connmgr_send_flow_removed(struct connmgr *mgr,
1596 const struct ofputil_flow_removed *fr)
1598 struct ofconn *ofconn;
1600 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1601 if (ofconn_receives_async_msg(ofconn, OAM_FLOW_REMOVED, fr->reason)) {
1604 /* Account flow expirations as replies to OpenFlow requests. That
1605 * works because preventing OpenFlow requests from being processed
1606 * also prevents new flows from being added (and expiring). (It
1607 * also prevents processing OpenFlow requests that would not add
1608 * new flows, so it is imperfect.) */
1609 msg = ofputil_encode_flow_removed(fr, ofconn_get_protocol(ofconn));
1610 ofconn_send_reply(ofconn, msg);
1615 /* Normally a send-to-controller action uses reason OFPR_ACTION. However, in
1616 * OpenFlow 1.3 and later, packet_ins generated by a send-to-controller action
1617 * in a "table-miss" flow (one with priority 0 and completely wildcarded) are
1618 * sent as OFPR_NO_MATCH. This function returns the reason that should
1619 * actually be sent on 'ofconn' for 'pin'. */
1620 static enum ofp_packet_in_reason
1621 wire_reason(struct ofconn *ofconn, const struct ofproto_packet_in *pin)
1623 if (pin->miss_type == OFPROTO_PACKET_IN_MISS_FLOW
1624 && pin->up.reason == OFPR_ACTION) {
1625 enum ofputil_protocol protocol = ofconn_get_protocol(ofconn);
1627 if (protocol != OFPUTIL_P_NONE
1628 && ofputil_protocol_to_ofp_version(protocol) >= OFP13_VERSION) {
1629 return OFPR_NO_MATCH;
1632 return pin->up.reason;
1635 /* Given 'pin', sends an OFPT_PACKET_IN message to each OpenFlow controller as
1636 * necessary according to their individual configurations.
1638 * The caller doesn't need to fill in pin->buffer_id or pin->total_len. */
1640 connmgr_send_packet_in(struct connmgr *mgr,
1641 const struct ofproto_packet_in *pin)
1643 struct ofconn *ofconn;
1645 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1646 enum ofp_packet_in_reason reason = wire_reason(ofconn, pin);
1648 if (ofconn_wants_packet_in_on_miss(ofconn, pin)
1649 && ofconn_receives_async_msg(ofconn, OAM_PACKET_IN, pin->up.reason)
1650 && ofconn->controller_id == pin->controller_id) {
1651 schedule_packet_in(ofconn, *pin, reason);
1657 do_send_packet_ins(struct ofconn *ofconn, struct list *txq)
1659 struct ofpbuf *pin, *next_pin;
1661 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, txq) {
1662 list_remove(&pin->list_node);
1664 if (rconn_send_with_limit(ofconn->rconn, pin,
1665 ofconn->packet_in_counter, 100) == EAGAIN) {
1666 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5);
1668 VLOG_INFO_RL(&rl, "%s: dropping packet-in due to queue overflow",
1669 rconn_get_name(ofconn->rconn));
1674 /* Takes 'pin', composes an OpenFlow packet-in message from it, and passes it
1675 * to 'ofconn''s packet scheduler for sending. */
1677 schedule_packet_in(struct ofconn *ofconn, struct ofproto_packet_in pin,
1678 enum ofp_packet_in_reason wire_reason)
1680 struct connmgr *mgr = ofconn->connmgr;
1681 uint16_t controller_max_len;
1684 pin.up.total_len = pin.up.packet_len;
1686 pin.up.reason = wire_reason;
1687 if (pin.up.reason == OFPR_ACTION) {
1688 controller_max_len = pin.send_len; /* max_len */
1690 controller_max_len = ofconn->miss_send_len;
1693 /* Get OpenFlow buffer_id.
1694 * For OpenFlow 1.2+, OFPCML_NO_BUFFER (== UINT16_MAX) specifies
1695 * unbuffered. This behaviour doesn't violate prior versions, too. */
1696 if (controller_max_len == UINT16_MAX) {
1697 pin.up.buffer_id = UINT32_MAX;
1698 } else if (mgr->fail_open && fail_open_is_active(mgr->fail_open)) {
1699 pin.up.buffer_id = pktbuf_get_null();
1700 } else if (!ofconn->pktbuf) {
1701 pin.up.buffer_id = UINT32_MAX;
1703 pin.up.buffer_id = pktbuf_save(ofconn->pktbuf,
1704 pin.up.packet, pin.up.packet_len,
1705 pin.up.fmd.in_port);
1708 /* Figure out how much of the packet to send.
1709 * If not buffered, send the entire packet. Otherwise, depending on
1710 * the reason of packet-in, send what requested by the controller. */
1711 if (pin.up.buffer_id != UINT32_MAX
1712 && controller_max_len < pin.up.packet_len) {
1713 pin.up.packet_len = controller_max_len;
1716 /* Make OFPT_PACKET_IN and hand over to packet scheduler. */
1717 pinsched_send(ofconn->schedulers[pin.up.reason == OFPR_NO_MATCH ? 0 : 1],
1719 ofputil_encode_packet_in(&pin.up,
1720 ofconn_get_protocol(ofconn),
1721 ofconn->packet_in_format),
1723 do_send_packet_ins(ofconn, &txq);
1726 /* Fail-open settings. */
1728 /* Returns the failure handling mode (OFPROTO_FAIL_SECURE or
1729 * OFPROTO_FAIL_STANDALONE) for 'mgr'. */
1730 enum ofproto_fail_mode
1731 connmgr_get_fail_mode(const struct connmgr *mgr)
1733 return mgr->fail_mode;
1736 /* Sets the failure handling mode for 'mgr' to 'fail_mode' (either
1737 * OFPROTO_FAIL_SECURE or OFPROTO_FAIL_STANDALONE). */
1739 connmgr_set_fail_mode(struct connmgr *mgr, enum ofproto_fail_mode fail_mode)
1741 if (mgr->fail_mode != fail_mode) {
1742 mgr->fail_mode = fail_mode;
1743 update_fail_open(mgr);
1744 if (!connmgr_has_controllers(mgr)) {
1745 ofproto_flush_flows(mgr->ofproto);
1750 /* Fail-open implementation. */
1752 /* Returns the longest probe interval among the primary controllers configured
1753 * on 'mgr'. Returns 0 if there are no primary controllers. */
1755 connmgr_get_max_probe_interval(const struct connmgr *mgr)
1757 const struct ofconn *ofconn;
1758 int max_probe_interval;
1760 max_probe_interval = 0;
1761 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1762 int probe_interval = rconn_get_probe_interval(ofconn->rconn);
1763 max_probe_interval = MAX(max_probe_interval, probe_interval);
1765 return max_probe_interval;
1768 /* Returns the number of seconds for which all of 'mgr's primary controllers
1769 * have been disconnected. Returns 0 if 'mgr' has no primary controllers. */
1771 connmgr_failure_duration(const struct connmgr *mgr)
1773 const struct ofconn *ofconn;
1774 int min_failure_duration;
1776 if (!connmgr_has_controllers(mgr)) {
1780 min_failure_duration = INT_MAX;
1781 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1782 int failure_duration = rconn_failure_duration(ofconn->rconn);
1783 min_failure_duration = MIN(min_failure_duration, failure_duration);
1785 return min_failure_duration;
1788 /* Returns true if at least one primary controller is connected (regardless of
1789 * whether those controllers are believed to have authenticated and accepted
1790 * this switch), false if none of them are connected. */
1792 connmgr_is_any_controller_connected(const struct connmgr *mgr)
1794 const struct ofconn *ofconn;
1796 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1797 if (rconn_is_connected(ofconn->rconn)) {
1804 /* Returns true if at least one primary controller is believed to have
1805 * authenticated and accepted this switch, false otherwise. */
1807 connmgr_is_any_controller_admitted(const struct connmgr *mgr)
1809 const struct ofconn *ofconn;
1811 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1812 if (rconn_is_admitted(ofconn->rconn)) {
1819 /* In-band configuration. */
1821 static bool any_extras_changed(const struct connmgr *,
1822 const struct sockaddr_in *extras, size_t n);
1824 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'mgr''s
1825 * in-band control should guarantee access, in the same way that in-band
1826 * control guarantees access to OpenFlow controllers. */
1828 connmgr_set_extra_in_band_remotes(struct connmgr *mgr,
1829 const struct sockaddr_in *extras, size_t n)
1831 if (!any_extras_changed(mgr, extras, n)) {
1835 free(mgr->extra_in_band_remotes);
1836 mgr->n_extra_remotes = n;
1837 mgr->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
1839 update_in_band_remotes(mgr);
1842 /* Sets the OpenFlow queue used by flows set up by in-band control on
1843 * 'mgr' to 'queue_id'. If 'queue_id' is negative, then in-band control
1844 * flows will use the default queue. */
1846 connmgr_set_in_band_queue(struct connmgr *mgr, int queue_id)
1848 if (queue_id != mgr->in_band_queue) {
1849 mgr->in_band_queue = queue_id;
1850 update_in_band_remotes(mgr);
1855 any_extras_changed(const struct connmgr *mgr,
1856 const struct sockaddr_in *extras, size_t n)
1860 if (n != mgr->n_extra_remotes) {
1864 for (i = 0; i < n; i++) {
1865 const struct sockaddr_in *old = &mgr->extra_in_band_remotes[i];
1866 const struct sockaddr_in *new = &extras[i];
1868 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
1869 old->sin_port != new->sin_port) {
1877 /* In-band implementation. */
1880 connmgr_has_in_band(struct connmgr *mgr)
1882 return mgr->in_band != NULL;
1885 /* Fail-open and in-band implementation. */
1887 /* Called by 'ofproto' after all flows have been flushed, to allow fail-open
1888 * and standalone mode to re-create their flows.
1890 * In-band control has more sophisticated code that manages flows itself. */
1892 connmgr_flushed(struct connmgr *mgr)
1893 OVS_EXCLUDED(ofproto_mutex)
1895 if (mgr->fail_open) {
1896 fail_open_flushed(mgr->fail_open);
1899 /* If there are no controllers and we're in standalone mode, set up a flow
1900 * that matches every packet and directs them to OFPP_NORMAL (which goes to
1901 * us). Otherwise, the switch is in secure mode and we won't pass any
1902 * traffic until a controller has been defined and it tells us to do so. */
1903 if (!connmgr_has_controllers(mgr)
1904 && mgr->fail_mode == OFPROTO_FAIL_STANDALONE) {
1905 struct ofpbuf ofpacts;
1908 ofpbuf_init(&ofpacts, OFPACT_OUTPUT_SIZE);
1909 ofpact_put_OUTPUT(&ofpacts)->port = OFPP_NORMAL;
1910 ofpact_pad(&ofpacts);
1912 match_init_catchall(&match);
1913 ofproto_add_flow(mgr->ofproto, &match, 0, ofpbuf_data(&ofpacts),
1914 ofpbuf_size(&ofpacts));
1916 ofpbuf_uninit(&ofpacts);
1920 /* Creates a new ofservice for 'target' in 'mgr'. Returns 0 if successful,
1921 * otherwise a positive errno value.
1923 * ofservice_reconfigure() must be called to fully configure the new
1926 ofservice_create(struct connmgr *mgr, const char *target,
1927 uint32_t allowed_versions, uint8_t dscp)
1929 struct ofservice *ofservice;
1930 struct pvconn *pvconn;
1933 error = pvconn_open(target, allowed_versions, dscp, &pvconn);
1938 ofservice = xzalloc(sizeof *ofservice);
1939 hmap_insert(&mgr->services, &ofservice->node, hash_string(target, 0));
1940 ofservice->pvconn = pvconn;
1941 ofservice->allowed_versions = allowed_versions;
1947 ofservice_destroy(struct connmgr *mgr, struct ofservice *ofservice)
1949 hmap_remove(&mgr->services, &ofservice->node);
1950 pvconn_close(ofservice->pvconn);
1955 ofservice_reconfigure(struct ofservice *ofservice,
1956 const struct ofproto_controller *c)
1958 ofservice->probe_interval = c->probe_interval;
1959 ofservice->rate_limit = c->rate_limit;
1960 ofservice->burst_limit = c->burst_limit;
1961 ofservice->enable_async_msgs = c->enable_async_msgs;
1962 ofservice->dscp = c->dscp;
1965 /* Finds and returns the ofservice within 'mgr' that has the given
1966 * 'target', or a null pointer if none exists. */
1967 static struct ofservice *
1968 ofservice_lookup(struct connmgr *mgr, const char *target)
1970 struct ofservice *ofservice;
1972 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1974 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1981 /* Flow monitors (NXST_FLOW_MONITOR). */
1983 /* A counter incremented when something significant happens to an OpenFlow
1986 * - When a rule is added, its 'add_seqno' and 'modify_seqno' are set to
1987 * the current value (which is then incremented).
1989 * - When a rule is modified, its 'modify_seqno' is set to the current
1990 * value (which is then incremented).
1992 * Thus, by comparing an old value of monitor_seqno against a rule's
1993 * 'add_seqno', one can tell whether the rule was added before or after the old
1994 * value was read, and similarly for 'modify_seqno'.
1996 * 32 bits should normally be sufficient (and would be nice, to save space in
1997 * each rule) but then we'd have to have some special cases for wraparound.
1999 * We initialize monitor_seqno to 1 to allow 0 to be used as an invalid
2001 static uint64_t monitor_seqno = 1;
2003 COVERAGE_DEFINE(ofmonitor_pause);
2004 COVERAGE_DEFINE(ofmonitor_resume);
2007 ofmonitor_create(const struct ofputil_flow_monitor_request *request,
2008 struct ofconn *ofconn, struct ofmonitor **monitorp)
2009 OVS_REQUIRES(ofproto_mutex)
2011 struct ofmonitor *m;
2015 m = ofmonitor_lookup(ofconn, request->id);
2017 return OFPERR_OFPMOFC_MONITOR_EXISTS;
2020 m = xmalloc(sizeof *m);
2022 hmap_insert(&ofconn->monitors, &m->ofconn_node, hash_int(request->id, 0));
2023 m->id = request->id;
2024 m->flags = request->flags;
2025 m->out_port = request->out_port;
2026 m->table_id = request->table_id;
2027 minimatch_init(&m->match, &request->match);
2034 ofmonitor_lookup(struct ofconn *ofconn, uint32_t id)
2035 OVS_REQUIRES(ofproto_mutex)
2037 struct ofmonitor *m;
2039 HMAP_FOR_EACH_IN_BUCKET (m, ofconn_node, hash_int(id, 0),
2040 &ofconn->monitors) {
2049 ofmonitor_destroy(struct ofmonitor *m)
2050 OVS_REQUIRES(ofproto_mutex)
2053 minimatch_destroy(&m->match);
2054 hmap_remove(&m->ofconn->monitors, &m->ofconn_node);
2060 ofmonitor_report(struct connmgr *mgr, struct rule *rule,
2061 enum nx_flow_update_event event,
2062 enum ofp_flow_removed_reason reason,
2063 const struct ofconn *abbrev_ofconn, ovs_be32 abbrev_xid,
2064 const struct rule_actions *old_actions)
2065 OVS_REQUIRES(ofproto_mutex)
2067 enum nx_flow_monitor_flags update;
2068 struct ofconn *ofconn;
2070 if (rule_is_hidden(rule)) {
2077 rule->add_seqno = rule->modify_seqno = monitor_seqno++;
2081 update = NXFMF_DELETE;
2084 case NXFME_MODIFIED:
2085 update = NXFMF_MODIFY;
2086 rule->modify_seqno = monitor_seqno++;
2094 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2095 enum nx_flow_monitor_flags flags = 0;
2096 struct ofmonitor *m;
2098 if (ofconn->monitor_paused) {
2099 /* Only send NXFME_DELETED notifications for flows that were added
2100 * before we paused. */
2101 if (event != NXFME_DELETED
2102 || rule->add_seqno > ofconn->monitor_paused) {
2107 HMAP_FOR_EACH (m, ofconn_node, &ofconn->monitors) {
2108 if (m->flags & update
2109 && (m->table_id == 0xff || m->table_id == rule->table_id)
2110 && (ofproto_rule_has_out_port(rule, m->out_port)
2112 && ofpacts_output_to_port(old_actions->ofpacts,
2113 old_actions->ofpacts_len,
2115 && cls_rule_is_loose_match(&rule->cr, &m->match)) {
2121 if (list_is_empty(&ofconn->updates)) {
2122 ofputil_start_flow_update(&ofconn->updates);
2123 ofconn->sent_abbrev_update = false;
2126 if (flags & NXFMF_OWN || ofconn != abbrev_ofconn
2127 || ofconn->monitor_paused) {
2128 struct ofputil_flow_update fu;
2132 fu.reason = event == NXFME_DELETED ? reason : 0;
2133 fu.table_id = rule->table_id;
2134 fu.cookie = rule->flow_cookie;
2135 minimatch_expand(&rule->cr.match, &match);
2137 fu.priority = rule->cr.priority;
2139 ovs_mutex_lock(&rule->mutex);
2140 fu.idle_timeout = rule->idle_timeout;
2141 fu.hard_timeout = rule->hard_timeout;
2142 ovs_mutex_unlock(&rule->mutex);
2144 if (flags & NXFMF_ACTIONS) {
2145 const struct rule_actions *actions = rule_get_actions(rule);
2146 fu.ofpacts = actions->ofpacts;
2147 fu.ofpacts_len = actions->ofpacts_len;
2152 ofputil_append_flow_update(&fu, &ofconn->updates);
2153 } else if (!ofconn->sent_abbrev_update) {
2154 struct ofputil_flow_update fu;
2156 fu.event = NXFME_ABBREV;
2157 fu.xid = abbrev_xid;
2158 ofputil_append_flow_update(&fu, &ofconn->updates);
2160 ofconn->sent_abbrev_update = true;
2167 ofmonitor_flush(struct connmgr *mgr)
2168 OVS_REQUIRES(ofproto_mutex)
2170 struct ofconn *ofconn;
2172 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2173 struct ofpbuf *msg, *next;
2175 LIST_FOR_EACH_SAFE (msg, next, list_node, &ofconn->updates) {
2176 unsigned int n_bytes;
2178 list_remove(&msg->list_node);
2179 ofconn_send(ofconn, msg, ofconn->monitor_counter);
2180 n_bytes = rconn_packet_counter_n_bytes(ofconn->monitor_counter);
2181 if (!ofconn->monitor_paused && n_bytes > 128 * 1024) {
2182 struct ofpbuf *pause;
2184 COVERAGE_INC(ofmonitor_pause);
2185 ofconn->monitor_paused = monitor_seqno++;
2186 pause = ofpraw_alloc_xid(OFPRAW_NXT_FLOW_MONITOR_PAUSED,
2187 OFP10_VERSION, htonl(0), 0);
2188 ofconn_send(ofconn, pause, ofconn->monitor_counter);
2195 ofmonitor_resume(struct ofconn *ofconn)
2196 OVS_REQUIRES(ofproto_mutex)
2198 struct rule_collection rules;
2199 struct ofpbuf *resumed;
2200 struct ofmonitor *m;
2203 rule_collection_init(&rules);
2204 HMAP_FOR_EACH (m, ofconn_node, &ofconn->monitors) {
2205 ofmonitor_collect_resume_rules(m, ofconn->monitor_paused, &rules);
2209 ofmonitor_compose_refresh_updates(&rules, &msgs);
2211 resumed = ofpraw_alloc_xid(OFPRAW_NXT_FLOW_MONITOR_RESUMED, OFP10_VERSION,
2213 list_push_back(&msgs, &resumed->list_node);
2214 ofconn_send_replies(ofconn, &msgs);
2216 ofconn->monitor_paused = 0;
2220 ofmonitor_may_resume(const struct ofconn *ofconn)
2221 OVS_REQUIRES(ofproto_mutex)
2223 return (ofconn->monitor_paused != 0
2224 && !rconn_packet_counter_n_packets(ofconn->monitor_counter));
2228 ofmonitor_run(struct connmgr *mgr)
2230 struct ofconn *ofconn;
2232 ovs_mutex_lock(&ofproto_mutex);
2233 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2234 if (ofmonitor_may_resume(ofconn)) {
2235 COVERAGE_INC(ofmonitor_resume);
2236 ofmonitor_resume(ofconn);
2239 ovs_mutex_unlock(&ofproto_mutex);
2243 ofmonitor_wait(struct connmgr *mgr)
2245 struct ofconn *ofconn;
2247 ovs_mutex_lock(&ofproto_mutex);
2248 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2249 if (ofmonitor_may_resume(ofconn)) {
2250 poll_immediate_wake();
2253 ovs_mutex_unlock(&ofproto_mutex);