2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 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 "fail-open.h"
28 #include "ofp-actions.h"
32 #include "ofproto-provider.h"
34 #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 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 /* Asynchronous flow table operation support. */
77 struct list opgroups; /* Contains pending "ofopgroups", if any. */
78 struct ofpbuf *blocked; /* Postponed OpenFlow message, if any. */
79 bool retry; /* True if 'blocked' is ready to try again. */
81 /* OFPT_PACKET_IN related data. */
82 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
83 #define N_SCHEDULERS 2
84 struct pinsched *schedulers[N_SCHEDULERS];
85 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
86 int miss_send_len; /* Bytes to send of buffered packets. */
87 uint16_t controller_id; /* Connection controller ID. */
89 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
90 * requests, and the maximum number before we stop reading OpenFlow
92 #define OFCONN_REPLY_MAX 100
93 struct rconn_packet_counter *reply_counter;
95 /* Asynchronous message configuration in each possible roles.
97 * A 1-bit enables sending an asynchronous message for one possible reason
98 * that the message might be generated, a 0-bit disables it. */
99 uint32_t master_async_config[OAM_N_TYPES]; /* master, other */
100 uint32_t slave_async_config[OAM_N_TYPES]; /* slave */
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 list updates OVS_GUARDED_BY(ofproto_mutex);
131 bool sent_abbrev_update OVS_GUARDED_BY(ofproto_mutex);
134 static struct ofconn *ofconn_create(struct connmgr *, struct rconn *,
135 enum ofconn_type, bool enable_async_msgs)
136 OVS_REQUIRES(ofproto_mutex);
137 static void ofconn_destroy(struct ofconn *) OVS_REQUIRES(ofproto_mutex);
138 static void ofconn_flush(struct ofconn *) OVS_REQUIRES(ofproto_mutex);
140 static void ofconn_reconfigure(struct ofconn *,
141 const struct ofproto_controller *);
143 static void ofconn_run(struct ofconn *,
144 bool (*handle_openflow)(struct ofconn *,
145 const struct ofpbuf *ofp_msg));
146 static void ofconn_wait(struct ofconn *, bool handling_openflow);
148 static const char *ofconn_get_target(const struct ofconn *);
149 static char *ofconn_make_name(const struct connmgr *, const char *target);
151 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
153 static void ofconn_send(const struct ofconn *, struct ofpbuf *,
154 struct rconn_packet_counter *);
156 static void do_send_packet_ins(struct ofconn *, struct list *txq);
158 /* A listener for incoming OpenFlow "service" connections. */
160 struct hmap_node node; /* In struct connmgr's "services" hmap. */
161 struct pvconn *pvconn; /* OpenFlow connection listener. */
163 /* These are not used by ofservice directly. They are settings for
164 * accepted "struct ofconn"s from the pvconn. */
165 int probe_interval; /* Max idle time before probing, in seconds. */
166 int rate_limit; /* Max packet-in rate in packets per second. */
167 int burst_limit; /* Limit on accumulating packet credits. */
168 bool enable_async_msgs; /* Initially enable async messages? */
169 uint8_t dscp; /* DSCP Value for controller connection */
170 uint32_t allowed_versions; /* OpenFlow protocol versions that may
171 * be negotiated for a session. */
174 static void ofservice_reconfigure(struct ofservice *,
175 const struct ofproto_controller *);
176 static int ofservice_create(struct connmgr *mgr, const char *target,
177 uint32_t allowed_versions, uint8_t dscp);
178 static void ofservice_destroy(struct connmgr *, struct ofservice *);
179 static struct ofservice *ofservice_lookup(struct connmgr *,
182 /* Connection manager for an OpenFlow switch. */
184 struct ofproto *ofproto;
186 char *local_port_name;
188 /* OpenFlow connections. */
189 struct hmap controllers; /* Controller "struct ofconn"s. */
190 struct list all_conns; /* Contains "struct ofconn"s. */
191 uint64_t master_election_id; /* monotonically increasing sequence number
192 * for master election */
193 bool master_election_id_defined;
195 /* OpenFlow listeners. */
196 struct hmap services; /* Contains "struct ofservice"s. */
197 struct pvconn **snoops;
201 struct fail_open *fail_open;
202 enum ofproto_fail_mode fail_mode;
204 /* In-band control. */
205 struct in_band *in_band;
206 struct sockaddr_in *extra_in_band_remotes;
207 size_t n_extra_remotes;
211 static void update_in_band_remotes(struct connmgr *);
212 static void add_snooper(struct connmgr *, struct vconn *);
213 static void ofmonitor_run(struct connmgr *);
214 static void ofmonitor_wait(struct connmgr *);
216 /* Creates and returns a new connection manager owned by 'ofproto'. 'name' is
217 * a name for the ofproto suitable for using in log messages.
218 * 'local_port_name' is the name of the local port (OFPP_LOCAL) within
221 connmgr_create(struct ofproto *ofproto,
222 const char *name, const char *local_port_name)
226 mgr = xmalloc(sizeof *mgr);
227 mgr->ofproto = ofproto;
228 mgr->name = xstrdup(name);
229 mgr->local_port_name = xstrdup(local_port_name);
231 hmap_init(&mgr->controllers);
232 list_init(&mgr->all_conns);
233 mgr->master_election_id = 0;
234 mgr->master_election_id_defined = false;
236 hmap_init(&mgr->services);
240 mgr->fail_open = NULL;
241 mgr->fail_mode = OFPROTO_FAIL_SECURE;
244 mgr->extra_in_band_remotes = NULL;
245 mgr->n_extra_remotes = 0;
246 mgr->in_band_queue = -1;
251 /* Frees 'mgr' and all of its resources. */
253 connmgr_destroy(struct connmgr *mgr)
255 struct ofservice *ofservice, *next_ofservice;
256 struct ofconn *ofconn, *next_ofconn;
263 ovs_mutex_lock(&ofproto_mutex);
264 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &mgr->all_conns) {
265 ofconn_destroy(ofconn);
267 ovs_mutex_unlock(&ofproto_mutex);
269 hmap_destroy(&mgr->controllers);
271 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &mgr->services) {
272 ofservice_destroy(mgr, ofservice);
274 hmap_destroy(&mgr->services);
276 for (i = 0; i < mgr->n_snoops; i++) {
277 pvconn_close(mgr->snoops[i]);
281 fail_open_destroy(mgr->fail_open);
282 mgr->fail_open = NULL;
284 in_band_destroy(mgr->in_band);
286 free(mgr->extra_in_band_remotes);
288 free(mgr->local_port_name);
293 /* Does all of the periodic maintenance required by 'mgr'.
295 * If 'handle_openflow' is nonnull, calls 'handle_openflow' for each message
296 * received on an OpenFlow connection, passing along the OpenFlow connection
297 * itself and the message that was sent. If 'handle_openflow' returns true,
298 * the message is considered to be fully processed. If 'handle_openflow'
299 * returns false, the message is considered not to have been processed at all;
300 * it will be stored and re-presented to 'handle_openflow' following the next
301 * call to connmgr_retry(). 'handle_openflow' must not modify or free the
304 * If 'handle_openflow' is NULL, no OpenFlow messages will be processed and
305 * other activities that could affect the flow table (in-band processing,
306 * fail-open processing) are suppressed too. */
308 connmgr_run(struct connmgr *mgr,
309 bool (*handle_openflow)(struct ofconn *,
310 const struct ofpbuf *ofp_msg))
311 OVS_EXCLUDED(ofproto_mutex)
313 struct ofconn *ofconn, *next_ofconn;
314 struct ofservice *ofservice;
317 if (handle_openflow && mgr->in_band) {
318 if (!in_band_run(mgr->in_band)) {
319 in_band_destroy(mgr->in_band);
324 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &mgr->all_conns) {
325 ofconn_run(ofconn, handle_openflow);
329 /* Fail-open maintenance. Do this after processing the ofconns since
330 * fail-open checks the status of the controller rconn. */
331 if (handle_openflow && mgr->fail_open) {
332 fail_open_run(mgr->fail_open);
335 HMAP_FOR_EACH (ofservice, node, &mgr->services) {
339 retval = pvconn_accept(ofservice->pvconn, &vconn);
344 /* Passing default value for creation of the rconn */
345 rconn = rconn_create(ofservice->probe_interval, 0, ofservice->dscp,
346 vconn_get_allowed_versions(vconn));
347 name = ofconn_make_name(mgr, vconn_get_name(vconn));
348 rconn_connect_unreliably(rconn, vconn, name);
351 ovs_mutex_lock(&ofproto_mutex);
352 ofconn = ofconn_create(mgr, rconn, OFCONN_SERVICE,
353 ofservice->enable_async_msgs);
354 ovs_mutex_unlock(&ofproto_mutex);
356 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
357 ofservice->burst_limit);
358 } else if (retval != EAGAIN) {
359 VLOG_WARN_RL(&rl, "accept failed (%s)", ovs_strerror(retval));
363 for (i = 0; i < mgr->n_snoops; i++) {
367 retval = pvconn_accept(mgr->snoops[i], &vconn);
369 add_snooper(mgr, vconn);
370 } else if (retval != EAGAIN) {
371 VLOG_WARN_RL(&rl, "accept failed (%s)", ovs_strerror(retval));
376 /* Causes the poll loop to wake up when connmgr_run() needs to run.
378 * If 'handling_openflow' is true, arriving OpenFlow messages and other
379 * activities that affect the flow table will wake up the poll loop. If
380 * 'handling_openflow' is false, they will not. */
382 connmgr_wait(struct connmgr *mgr, bool handling_openflow)
384 struct ofservice *ofservice;
385 struct ofconn *ofconn;
388 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
389 ofconn_wait(ofconn, handling_openflow);
392 if (handling_openflow && mgr->in_band) {
393 in_band_wait(mgr->in_band);
395 if (handling_openflow && mgr->fail_open) {
396 fail_open_wait(mgr->fail_open);
398 HMAP_FOR_EACH (ofservice, node, &mgr->services) {
399 pvconn_wait(ofservice->pvconn);
401 for (i = 0; i < mgr->n_snoops; i++) {
402 pvconn_wait(mgr->snoops[i]);
406 /* Adds some memory usage statistics for 'mgr' into 'usage', for use with
407 * memory_report(). */
409 connmgr_get_memory_usage(const struct connmgr *mgr, struct simap *usage)
411 const struct ofconn *ofconn;
412 unsigned int packets = 0;
413 unsigned int ofconns = 0;
415 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
420 packets += rconn_count_txqlen(ofconn->rconn);
421 for (i = 0; i < N_SCHEDULERS; i++) {
422 packets += pinsched_count_txqlen(ofconn->schedulers[i]);
424 packets += pktbuf_count_packets(ofconn->pktbuf);
426 simap_increase(usage, "ofconns", ofconns);
427 simap_increase(usage, "packets", packets);
430 /* Returns the ofproto that owns 'ofconn''s connmgr. */
432 ofconn_get_ofproto(const struct ofconn *ofconn)
434 return ofconn->connmgr->ofproto;
437 /* If processing of OpenFlow messages was blocked on any 'mgr' ofconns by
438 * returning false to the 'handle_openflow' callback to connmgr_run(), this
439 * re-enables them. */
441 connmgr_retry(struct connmgr *mgr)
443 struct ofconn *ofconn;
445 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
446 ofconn->retry = true;
450 /* OpenFlow configuration. */
452 static void add_controller(struct connmgr *, const char *target, uint8_t dscp,
453 uint32_t allowed_versions)
454 OVS_REQUIRES(ofproto_mutex);
455 static struct ofconn *find_controller_by_target(struct connmgr *,
457 static void update_fail_open(struct connmgr *) OVS_EXCLUDED(ofproto_mutex);
458 static int set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
459 const struct sset *);
461 /* Returns true if 'mgr' has any configured primary controllers.
463 * Service controllers do not count, but configured primary controllers do
464 * count whether or not they are currently connected. */
466 connmgr_has_controllers(const struct connmgr *mgr)
468 return !hmap_is_empty(&mgr->controllers);
471 /* Initializes 'info' and populates it with information about each configured
472 * primary controller. The keys in 'info' are the controllers' targets; the
473 * data values are corresponding "struct ofproto_controller_info".
475 * The caller owns 'info' and everything in it and should free it when it is no
478 connmgr_get_controller_info(struct connmgr *mgr, struct shash *info)
480 const struct ofconn *ofconn;
482 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
483 const struct rconn *rconn = ofconn->rconn;
484 const char *target = rconn_get_target(rconn);
486 if (!shash_find(info, target)) {
487 struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo);
488 time_t now = time_now();
489 time_t last_connection = rconn_get_last_connection(rconn);
490 time_t last_disconnect = rconn_get_last_disconnect(rconn);
491 int last_error = rconn_get_last_error(rconn);
493 shash_add(info, target, cinfo);
495 cinfo->is_connected = rconn_is_connected(rconn);
496 cinfo->role = ofconn->role;
501 cinfo->pairs.keys[cinfo->pairs.n] = "last_error";
502 cinfo->pairs.values[cinfo->pairs.n++]
503 = xstrdup(ovs_retval_to_string(last_error));
506 cinfo->pairs.keys[cinfo->pairs.n] = "state";
507 cinfo->pairs.values[cinfo->pairs.n++]
508 = xstrdup(rconn_get_state(rconn));
510 if (last_connection != TIME_MIN) {
511 cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_connect";
512 cinfo->pairs.values[cinfo->pairs.n++]
513 = xasprintf("%ld", (long int) (now - last_connection));
516 if (last_disconnect != TIME_MIN) {
517 cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_disconnect";
518 cinfo->pairs.values[cinfo->pairs.n++]
519 = xasprintf("%ld", (long int) (now - last_disconnect));
526 connmgr_free_controller_info(struct shash *info)
528 struct shash_node *node;
530 SHASH_FOR_EACH (node, info) {
531 struct ofproto_controller_info *cinfo = node->data;
532 while (cinfo->pairs.n) {
533 free(CONST_CAST(char *, cinfo->pairs.values[--cinfo->pairs.n]));
540 /* Changes 'mgr''s set of controllers to the 'n_controllers' controllers in
543 connmgr_set_controllers(struct connmgr *mgr,
544 const struct ofproto_controller *controllers,
545 size_t n_controllers, uint32_t allowed_versions)
546 OVS_EXCLUDED(ofproto_mutex)
548 bool had_controllers = connmgr_has_controllers(mgr);
549 struct shash new_controllers;
550 struct ofconn *ofconn, *next_ofconn;
551 struct ofservice *ofservice, *next_ofservice;
554 /* Required to add and remove ofconns. This could probably be narrowed to
555 * cover a smaller amount of code, if that yielded some benefit. */
556 ovs_mutex_lock(&ofproto_mutex);
558 /* Create newly configured controllers and services.
559 * Create a name to ofproto_controller mapping in 'new_controllers'. */
560 shash_init(&new_controllers);
561 for (i = 0; i < n_controllers; i++) {
562 const struct ofproto_controller *c = &controllers[i];
564 if (!vconn_verify_name(c->target)) {
566 ofconn = find_controller_by_target(mgr, c->target);
568 VLOG_INFO("%s: added primary controller \"%s\"",
569 mgr->name, c->target);
571 } else if (rconn_get_allowed_versions(ofconn->rconn) !=
573 VLOG_INFO("%s: re-added primary controller \"%s\"",
574 mgr->name, c->target);
576 ofconn_destroy(ofconn);
579 add_controller(mgr, c->target, c->dscp, allowed_versions);
581 } else if (!pvconn_verify_name(c->target)) {
583 ofservice = ofservice_lookup(mgr, c->target);
585 VLOG_INFO("%s: added service controller \"%s\"",
586 mgr->name, c->target);
588 } else if (ofservice->allowed_versions != allowed_versions) {
589 VLOG_INFO("%s: re-added service controller \"%s\"",
590 mgr->name, c->target);
591 ofservice_destroy(mgr, ofservice);
595 ofservice_create(mgr, c->target, allowed_versions, c->dscp);
598 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
599 mgr->name, c->target);
603 shash_add_once(&new_controllers, c->target, &controllers[i]);
606 /* Delete controllers that are no longer configured.
607 * Update configuration of all now-existing controllers. */
608 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &mgr->controllers) {
609 const char *target = ofconn_get_target(ofconn);
610 struct ofproto_controller *c;
612 c = shash_find_data(&new_controllers, target);
614 VLOG_INFO("%s: removed primary controller \"%s\"",
616 ofconn_destroy(ofconn);
618 ofconn_reconfigure(ofconn, c);
622 /* Delete services that are no longer configured.
623 * Update configuration of all now-existing services. */
624 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &mgr->services) {
625 const char *target = pvconn_get_name(ofservice->pvconn);
626 struct ofproto_controller *c;
628 c = shash_find_data(&new_controllers, target);
630 VLOG_INFO("%s: removed service controller \"%s\"",
632 ofservice_destroy(mgr, ofservice);
634 ofservice_reconfigure(ofservice, c);
638 shash_destroy(&new_controllers);
640 ovs_mutex_unlock(&ofproto_mutex);
642 update_in_band_remotes(mgr);
643 update_fail_open(mgr);
644 if (had_controllers != connmgr_has_controllers(mgr)) {
645 ofproto_flush_flows(mgr->ofproto);
649 /* Drops the connections between 'mgr' and all of its primary and secondary
650 * controllers, forcing them to reconnect. */
652 connmgr_reconnect(const struct connmgr *mgr)
654 struct ofconn *ofconn;
656 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
657 rconn_reconnect(ofconn->rconn);
661 /* Sets the "snoops" for 'mgr' to the pvconn targets listed in 'snoops'.
663 * A "snoop" is a pvconn to which every OpenFlow message to or from the most
664 * important controller on 'mgr' is mirrored. */
666 connmgr_set_snoops(struct connmgr *mgr, const struct sset *snoops)
668 return set_pvconns(&mgr->snoops, &mgr->n_snoops, snoops);
671 /* Adds each of the snoops currently configured on 'mgr' to 'snoops'. */
673 connmgr_get_snoops(const struct connmgr *mgr, struct sset *snoops)
677 for (i = 0; i < mgr->n_snoops; i++) {
678 sset_add(snoops, pvconn_get_name(mgr->snoops[i]));
682 /* Returns true if 'mgr' has at least one snoop, false if it has none. */
684 connmgr_has_snoops(const struct connmgr *mgr)
686 return mgr->n_snoops > 0;
689 /* Creates a new controller for 'target' in 'mgr'. update_controller() needs
690 * to be called later to finish the new ofconn's configuration. */
692 add_controller(struct connmgr *mgr, const char *target, uint8_t dscp,
693 uint32_t allowed_versions)
694 OVS_REQUIRES(ofproto_mutex)
696 char *name = ofconn_make_name(mgr, target);
697 struct ofconn *ofconn;
699 ofconn = ofconn_create(mgr, rconn_create(5, 8, dscp, allowed_versions),
700 OFCONN_PRIMARY, true);
701 ofconn->pktbuf = pktbuf_create();
702 rconn_connect(ofconn->rconn, target, name);
703 hmap_insert(&mgr->controllers, &ofconn->hmap_node, hash_string(target, 0));
708 static struct ofconn *
709 find_controller_by_target(struct connmgr *mgr, const char *target)
711 struct ofconn *ofconn;
713 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
714 hash_string(target, 0), &mgr->controllers) {
715 if (!strcmp(ofconn_get_target(ofconn), target)) {
723 update_in_band_remotes(struct connmgr *mgr)
725 struct sockaddr_in *addrs;
726 size_t max_addrs, n_addrs;
727 struct ofconn *ofconn;
730 /* Allocate enough memory for as many remotes as we could possibly have. */
731 max_addrs = mgr->n_extra_remotes + hmap_count(&mgr->controllers);
732 addrs = xmalloc(max_addrs * sizeof *addrs);
735 /* Add all the remotes. */
736 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
737 struct sockaddr_in *sin = &addrs[n_addrs];
738 const char *target = rconn_get_target(ofconn->rconn);
740 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
744 if (stream_parse_target_with_default_port(target,
750 for (i = 0; i < mgr->n_extra_remotes; i++) {
751 addrs[n_addrs++] = mgr->extra_in_band_remotes[i];
754 /* Create or update or destroy in-band. */
757 in_band_create(mgr->ofproto, mgr->local_port_name, &mgr->in_band);
759 in_band_set_queue(mgr->in_band, mgr->in_band_queue);
761 /* in_band_run() needs a chance to delete any existing in-band flows.
762 * We will destroy mgr->in_band after it's done with that. */
765 in_band_set_remotes(mgr->in_band, addrs, n_addrs);
773 update_fail_open(struct connmgr *mgr)
774 OVS_EXCLUDED(ofproto_mutex)
776 if (connmgr_has_controllers(mgr)
777 && mgr->fail_mode == OFPROTO_FAIL_STANDALONE) {
778 if (!mgr->fail_open) {
779 mgr->fail_open = fail_open_create(mgr->ofproto, mgr);
782 fail_open_destroy(mgr->fail_open);
783 mgr->fail_open = NULL;
788 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
789 const struct sset *sset)
791 struct pvconn **pvconns = *pvconnsp;
792 size_t n_pvconns = *n_pvconnsp;
797 for (i = 0; i < n_pvconns; i++) {
798 pvconn_close(pvconns[i]);
802 pvconns = xmalloc(sset_count(sset) * sizeof *pvconns);
804 SSET_FOR_EACH (name, sset) {
805 struct pvconn *pvconn;
807 error = pvconn_open(name, 0, 0, &pvconn);
809 pvconns[n_pvconns++] = pvconn;
811 VLOG_ERR("failed to listen on %s: %s", name, ovs_strerror(error));
819 *n_pvconnsp = n_pvconns;
824 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
825 * means that 'ofconn' is more interesting for monitoring than a lower return
828 snoop_preference(const struct ofconn *ofconn)
830 switch (ofconn->role) {
831 case OFPCR12_ROLE_MASTER:
833 case OFPCR12_ROLE_EQUAL:
835 case OFPCR12_ROLE_SLAVE:
837 case OFPCR12_ROLE_NOCHANGE:
839 /* Shouldn't happen. */
844 /* One of 'mgr''s "snoop" pvconns has accepted a new connection on 'vconn'.
845 * Connects this vconn to a controller. */
847 add_snooper(struct connmgr *mgr, struct vconn *vconn)
849 struct ofconn *ofconn, *best;
851 /* Pick a controller for monitoring. */
853 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
854 if (ofconn->type == OFCONN_PRIMARY
855 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
861 rconn_add_monitor(best->rconn, vconn);
863 VLOG_INFO_RL(&rl, "no controller connection to snoop");
868 /* Public ofconn functions. */
870 /* Returns the connection type, either OFCONN_PRIMARY or OFCONN_SERVICE. */
872 ofconn_get_type(const struct ofconn *ofconn)
877 /* If a master election id is defined, stores it into '*idp' and returns
878 * true. Otherwise, stores UINT64_MAX into '*idp' and returns false. */
880 ofconn_get_master_election_id(const struct ofconn *ofconn, uint64_t *idp)
882 *idp = (ofconn->connmgr->master_election_id_defined
883 ? ofconn->connmgr->master_election_id
885 return ofconn->connmgr->master_election_id_defined;
888 /* Sets the master election id.
890 * Returns true if successful, false if the id is stale
893 ofconn_set_master_election_id(struct ofconn *ofconn, uint64_t id)
895 if (ofconn->connmgr->master_election_id_defined
897 /* Unsigned difference interpreted as a two's complement signed
899 (int64_t)(id - ofconn->connmgr->master_election_id) < 0) {
902 ofconn->connmgr->master_election_id = id;
903 ofconn->connmgr->master_election_id_defined = true;
908 /* Returns the role configured for 'ofconn'.
910 * The default role, if no other role has been set, is OFPCR12_ROLE_EQUAL. */
911 enum ofp12_controller_role
912 ofconn_get_role(const struct ofconn *ofconn)
918 ofconn_send_role_status(struct ofconn *ofconn, uint32_t role, uint8_t reason)
920 struct ofputil_role_status status;
923 status.reason = reason;
925 ofconn_get_master_election_id(ofconn, &status.generation_id);
927 buf = ofputil_encode_role_status(&status, ofconn_get_protocol(ofconn));
929 ofconn_send(ofconn, buf, NULL);
932 /* Changes 'ofconn''s role to 'role'. If 'role' is OFPCR12_ROLE_MASTER then
933 * any existing master is demoted to a slave. */
935 ofconn_set_role(struct ofconn *ofconn, enum ofp12_controller_role role)
937 if (role != ofconn->role && role == OFPCR12_ROLE_MASTER) {
938 struct ofconn *other;
940 HMAP_FOR_EACH (other, hmap_node, &ofconn->connmgr->controllers) {
941 if (other->role == OFPCR12_ROLE_MASTER) {
942 other->role = OFPCR12_ROLE_SLAVE;
943 ofconn_send_role_status(other, OFPCR12_ROLE_SLAVE, OFPCRR_MASTER_REQUEST);
951 ofconn_set_invalid_ttl_to_controller(struct ofconn *ofconn, bool enable)
953 uint32_t bit = 1u << OFPR_INVALID_TTL;
955 ofconn->master_async_config[OAM_PACKET_IN] |= bit;
957 ofconn->master_async_config[OAM_PACKET_IN] &= ~bit;
962 ofconn_get_invalid_ttl_to_controller(struct ofconn *ofconn)
964 uint32_t bit = 1u << OFPR_INVALID_TTL;
965 return (ofconn->master_async_config[OAM_PACKET_IN] & bit) != 0;
968 /* Returns the currently configured protocol for 'ofconn', one of OFPUTIL_P_*.
970 * Returns OFPUTIL_P_NONE, which is not a valid protocol, if 'ofconn' hasn't
971 * completed version negotiation. This can't happen if at least one OpenFlow
972 * message, other than OFPT_HELLO, has been received on the connection (such as
973 * in ofproto.c's message handling code), since version negotiation is a
974 * prerequisite for starting to receive messages. This means that
975 * OFPUTIL_P_NONE is a special case that most callers need not worry about. */
976 enum ofputil_protocol
977 ofconn_get_protocol(const struct ofconn *ofconn)
979 if (ofconn->protocol == OFPUTIL_P_NONE &&
980 rconn_is_connected(ofconn->rconn)) {
981 int version = rconn_get_version(ofconn->rconn);
983 ofconn_set_protocol(CONST_CAST(struct ofconn *, ofconn),
984 ofputil_protocol_from_ofp_version(version));
988 return ofconn->protocol;
991 /* Sets the protocol for 'ofconn' to 'protocol' (one of OFPUTIL_P_*).
993 * (This doesn't actually send anything to accomplish this. Presumably the
994 * caller already did that.) */
996 ofconn_set_protocol(struct ofconn *ofconn, enum ofputil_protocol protocol)
998 ofconn->protocol = protocol;
1001 /* Returns the currently configured packet in format for 'ofconn', one of
1004 * The default, if no other format has been set, is NXPIF_OPENFLOW10. */
1005 enum nx_packet_in_format
1006 ofconn_get_packet_in_format(struct ofconn *ofconn)
1008 return ofconn->packet_in_format;
1011 /* Sets the packet in format for 'ofconn' to 'packet_in_format' (one of
1014 ofconn_set_packet_in_format(struct ofconn *ofconn,
1015 enum nx_packet_in_format packet_in_format)
1017 ofconn->packet_in_format = packet_in_format;
1020 /* Sets the controller connection ID for 'ofconn' to 'controller_id'.
1022 * The connection controller ID is used for OFPP_CONTROLLER and
1023 * NXAST_CONTROLLER actions. See "struct nx_action_controller" for details. */
1025 ofconn_set_controller_id(struct ofconn *ofconn, uint16_t controller_id)
1027 ofconn->controller_id = controller_id;
1030 /* Returns the default miss send length for 'ofconn'. */
1032 ofconn_get_miss_send_len(const struct ofconn *ofconn)
1034 return ofconn->miss_send_len;
1037 /* Sets the default miss send length for 'ofconn' to 'miss_send_len'. */
1039 ofconn_set_miss_send_len(struct ofconn *ofconn, int miss_send_len)
1041 ofconn->miss_send_len = miss_send_len;
1045 ofconn_set_async_config(struct ofconn *ofconn,
1046 const uint32_t master_masks[OAM_N_TYPES],
1047 const uint32_t slave_masks[OAM_N_TYPES])
1049 size_t size = sizeof ofconn->master_async_config;
1050 memcpy(ofconn->master_async_config, master_masks, size);
1051 memcpy(ofconn->slave_async_config, slave_masks, size);
1055 ofconn_get_async_config(struct ofconn *ofconn,
1056 uint32_t *master_masks, uint32_t *slave_masks)
1058 size_t size = sizeof ofconn->master_async_config;
1059 memcpy(master_masks, ofconn->master_async_config, size);
1060 memcpy(slave_masks, ofconn->slave_async_config, size);
1063 /* Sends 'msg' on 'ofconn', accounting it as a reply. (If there is a
1064 * sufficient number of OpenFlow replies in-flight on a single ofconn, then the
1065 * connmgr will stop accepting new OpenFlow requests on that ofconn until the
1066 * controller has accepted some of the replies.) */
1068 ofconn_send_reply(const struct ofconn *ofconn, struct ofpbuf *msg)
1070 ofconn_send(ofconn, msg, ofconn->reply_counter);
1073 /* Sends each of the messages in list 'replies' on 'ofconn' in order,
1074 * accounting them as replies. */
1076 ofconn_send_replies(const struct ofconn *ofconn, struct list *replies)
1078 struct ofpbuf *reply, *next;
1080 LIST_FOR_EACH_SAFE (reply, next, list_node, replies) {
1081 list_remove(&reply->list_node);
1082 ofconn_send_reply(ofconn, reply);
1086 /* Sends 'error' on 'ofconn', as a reply to 'request'. Only at most the
1087 * first 64 bytes of 'request' are used. */
1089 ofconn_send_error(const struct ofconn *ofconn,
1090 const struct ofp_header *request, enum ofperr error)
1092 static struct vlog_rate_limit err_rl = VLOG_RATE_LIMIT_INIT(10, 10);
1093 struct ofpbuf *reply;
1095 reply = ofperr_encode_reply(error, request);
1096 if (!VLOG_DROP_INFO(&err_rl)) {
1097 const char *type_name;
1101 request_len = ntohs(request->length);
1102 type_name = (!ofpraw_decode_partial(&raw, request,
1103 MIN(64, request_len))
1104 ? ofpraw_get_name(raw)
1107 VLOG_INFO("%s: sending %s error reply to %s message",
1108 rconn_get_name(ofconn->rconn), ofperr_to_string(error),
1111 ofconn_send_reply(ofconn, reply);
1114 /* Same as pktbuf_retrieve(), using the pktbuf owned by 'ofconn'. */
1116 ofconn_pktbuf_retrieve(struct ofconn *ofconn, uint32_t id,
1117 struct ofpbuf **bufferp, ofp_port_t *in_port)
1119 return pktbuf_retrieve(ofconn->pktbuf, id, bufferp, in_port);
1122 /* Returns true if 'ofconn' has any pending opgroups. */
1124 ofconn_has_pending_opgroups(const struct ofconn *ofconn)
1126 return !list_is_empty(&ofconn->opgroups);
1129 /* Adds 'ofconn_node' to 'ofconn''s list of pending opgroups.
1131 * If 'ofconn' is destroyed or its connection drops, then 'ofconn' will remove
1132 * 'ofconn_node' from the list and re-initialize it with list_init(). The
1133 * client may, therefore, use list_is_empty(ofconn_node) to determine whether
1134 * 'ofconn_node' is still associated with an active ofconn.
1136 * The client may also remove ofconn_node from the list itself, with
1139 ofconn_add_opgroup(struct ofconn *ofconn, struct list *ofconn_node)
1141 list_push_back(&ofconn->opgroups, ofconn_node);
1144 /* Private ofconn functions. */
1147 ofconn_get_target(const struct ofconn *ofconn)
1149 return rconn_get_target(ofconn->rconn);
1152 static struct ofconn *
1153 ofconn_create(struct connmgr *mgr, struct rconn *rconn, enum ofconn_type type,
1154 bool enable_async_msgs)
1156 struct ofconn *ofconn;
1158 ofconn = xzalloc(sizeof *ofconn);
1159 ofconn->connmgr = mgr;
1160 list_push_back(&mgr->all_conns, &ofconn->node);
1161 ofconn->rconn = rconn;
1162 ofconn->type = type;
1163 ofconn->enable_async_msgs = enable_async_msgs;
1165 list_init(&ofconn->opgroups);
1167 hmap_init(&ofconn->monitors);
1168 list_init(&ofconn->updates);
1170 ofconn_flush(ofconn);
1175 /* Clears all of the state in 'ofconn' that should not persist from one
1176 * connection to the next. */
1178 ofconn_flush(struct ofconn *ofconn)
1179 OVS_REQUIRES(ofproto_mutex)
1181 struct ofmonitor *monitor, *next_monitor;
1184 ofconn->role = OFPCR12_ROLE_EQUAL;
1185 ofconn_set_protocol(ofconn, OFPUTIL_P_NONE);
1186 ofconn->packet_in_format = NXPIF_OPENFLOW10;
1188 /* Disassociate 'ofconn' from all of the ofopgroups that it initiated that
1189 * have not yet completed. (Those ofopgroups will still run to completion
1190 * in the usual way, but any errors that they run into will not be reported
1191 * on any OpenFlow channel.)
1193 * Also discard any blocked operation on 'ofconn'. */
1194 while (!list_is_empty(&ofconn->opgroups)) {
1195 list_init(list_pop_front(&ofconn->opgroups));
1197 ofpbuf_delete(ofconn->blocked);
1198 ofconn->blocked = NULL;
1200 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1201 ofconn->packet_in_counter = rconn_packet_counter_create();
1202 for (i = 0; i < N_SCHEDULERS; i++) {
1203 if (ofconn->schedulers[i]) {
1206 pinsched_get_limits(ofconn->schedulers[i], &rate, &burst);
1207 pinsched_destroy(ofconn->schedulers[i]);
1208 ofconn->schedulers[i] = pinsched_create(rate, burst);
1211 if (ofconn->pktbuf) {
1212 pktbuf_destroy(ofconn->pktbuf);
1213 ofconn->pktbuf = pktbuf_create();
1215 ofconn->miss_send_len = (ofconn->type == OFCONN_PRIMARY
1216 ? OFP_DEFAULT_MISS_SEND_LEN
1218 ofconn->controller_id = 0;
1220 rconn_packet_counter_destroy(ofconn->reply_counter);
1221 ofconn->reply_counter = rconn_packet_counter_create();
1223 if (ofconn->enable_async_msgs) {
1224 uint32_t *master = ofconn->master_async_config;
1225 uint32_t *slave = ofconn->slave_async_config;
1227 /* "master" and "other" roles get all asynchronous messages by default,
1228 * except that the controller needs to enable nonstandard "packet-in"
1229 * reasons itself. */
1230 master[OAM_PACKET_IN] = (1u << OFPR_NO_MATCH) | (1u << OFPR_ACTION);
1231 master[OAM_PORT_STATUS] = ((1u << OFPPR_ADD)
1232 | (1u << OFPPR_DELETE)
1233 | (1u << OFPPR_MODIFY));
1234 master[OAM_FLOW_REMOVED] = ((1u << OFPRR_IDLE_TIMEOUT)
1235 | (1u << OFPRR_HARD_TIMEOUT)
1236 | (1u << OFPRR_DELETE));
1238 /* "slave" role gets port status updates by default. */
1239 slave[OAM_PACKET_IN] = 0;
1240 slave[OAM_PORT_STATUS] = ((1u << OFPPR_ADD)
1241 | (1u << OFPPR_DELETE)
1242 | (1u << OFPPR_MODIFY));
1243 slave[OAM_FLOW_REMOVED] = 0;
1245 memset(ofconn->master_async_config, 0,
1246 sizeof ofconn->master_async_config);
1247 memset(ofconn->slave_async_config, 0,
1248 sizeof ofconn->slave_async_config);
1251 HMAP_FOR_EACH_SAFE (monitor, next_monitor, ofconn_node,
1252 &ofconn->monitors) {
1253 ofmonitor_destroy(monitor);
1255 rconn_packet_counter_destroy(ofconn->monitor_counter);
1256 ofconn->monitor_counter = rconn_packet_counter_create();
1257 ofpbuf_list_delete(&ofconn->updates); /* ...but it should be empty. */
1261 ofconn_destroy(struct ofconn *ofconn)
1262 OVS_REQUIRES(ofproto_mutex)
1264 ofconn_flush(ofconn);
1266 if (ofconn->type == OFCONN_PRIMARY) {
1267 hmap_remove(&ofconn->connmgr->controllers, &ofconn->hmap_node);
1270 hmap_destroy(&ofconn->monitors);
1271 list_remove(&ofconn->node);
1272 rconn_destroy(ofconn->rconn);
1273 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1274 rconn_packet_counter_destroy(ofconn->reply_counter);
1275 pktbuf_destroy(ofconn->pktbuf);
1276 rconn_packet_counter_destroy(ofconn->monitor_counter);
1280 /* Reconfigures 'ofconn' to match 'c'. 'ofconn' and 'c' must have the same
1283 ofconn_reconfigure(struct ofconn *ofconn, const struct ofproto_controller *c)
1287 ofconn->band = c->band;
1288 ofconn->enable_async_msgs = c->enable_async_msgs;
1290 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
1292 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
1293 rconn_set_probe_interval(ofconn->rconn, probe_interval);
1295 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
1297 /* If dscp value changed reconnect. */
1298 if (c->dscp != rconn_get_dscp(ofconn->rconn)) {
1299 rconn_set_dscp(ofconn->rconn, c->dscp);
1300 rconn_reconnect(ofconn->rconn);
1304 /* Returns true if it makes sense for 'ofconn' to receive and process OpenFlow
1307 ofconn_may_recv(const struct ofconn *ofconn)
1309 int count = rconn_packet_counter_n_packets(ofconn->reply_counter);
1310 return (!ofconn->blocked || ofconn->retry) && count < OFCONN_REPLY_MAX;
1314 ofconn_run(struct ofconn *ofconn,
1315 bool (*handle_openflow)(struct ofconn *,
1316 const struct ofpbuf *ofp_msg))
1318 struct connmgr *mgr = ofconn->connmgr;
1321 for (i = 0; i < N_SCHEDULERS; i++) {
1324 pinsched_run(ofconn->schedulers[i], &txq);
1325 do_send_packet_ins(ofconn, &txq);
1328 rconn_run(ofconn->rconn);
1330 if (handle_openflow) {
1331 /* Limit the number of iterations to avoid starving other tasks. */
1332 for (i = 0; i < 50 && ofconn_may_recv(ofconn); i++) {
1333 struct ofpbuf *of_msg;
1335 of_msg = (ofconn->blocked
1337 : rconn_recv(ofconn->rconn));
1341 if (mgr->fail_open) {
1342 fail_open_maybe_recover(mgr->fail_open);
1345 if (handle_openflow(ofconn, of_msg)) {
1346 ofpbuf_delete(of_msg);
1347 ofconn->blocked = NULL;
1349 ofconn->blocked = of_msg;
1350 ofconn->retry = false;
1355 ovs_mutex_lock(&ofproto_mutex);
1356 if (!rconn_is_alive(ofconn->rconn)) {
1357 ofconn_destroy(ofconn);
1358 } else if (!rconn_is_connected(ofconn->rconn)) {
1359 ofconn_flush(ofconn);
1361 ovs_mutex_unlock(&ofproto_mutex);
1365 ofconn_wait(struct ofconn *ofconn, bool handling_openflow)
1369 for (i = 0; i < N_SCHEDULERS; i++) {
1370 pinsched_wait(ofconn->schedulers[i]);
1372 rconn_run_wait(ofconn->rconn);
1373 if (handling_openflow && ofconn_may_recv(ofconn)) {
1374 rconn_recv_wait(ofconn->rconn);
1378 /* Returns true if 'ofconn' should receive asynchronous messages of the given
1379 * OAM_* 'type' and 'reason', which should be a OFPR_* value for OAM_PACKET_IN,
1380 * a OFPPR_* value for OAM_PORT_STATUS, or an OFPRR_* value for
1381 * OAM_FLOW_REMOVED. Returns false if the message should not be sent on
1384 ofconn_receives_async_msg(const struct ofconn *ofconn,
1385 enum ofconn_async_msg_type type,
1386 unsigned int reason)
1388 const uint32_t *async_config;
1390 ovs_assert(reason < 32);
1391 ovs_assert((unsigned int) type < OAM_N_TYPES);
1393 if (ofconn_get_protocol(ofconn) == OFPUTIL_P_NONE
1394 || !rconn_is_connected(ofconn->rconn)) {
1398 /* Keep the following code in sync with the documentation in the
1399 * "Asynchronous Messages" section in DESIGN. */
1401 if (ofconn->type == OFCONN_SERVICE && !ofconn->miss_send_len) {
1402 /* Service connections don't get asynchronous messages unless they have
1403 * explicitly asked for them by setting a nonzero miss send length. */
1407 async_config = (ofconn->role == OFPCR12_ROLE_SLAVE
1408 ? ofconn->slave_async_config
1409 : ofconn->master_async_config);
1410 if (!(async_config[type] & (1u << reason))) {
1417 /* Returns a human-readable name for an OpenFlow connection between 'mgr' and
1418 * 'target', suitable for use in log messages for identifying the connection.
1420 * The name is dynamically allocated. The caller should free it (with free())
1421 * when it is no longer needed. */
1423 ofconn_make_name(const struct connmgr *mgr, const char *target)
1425 return xasprintf("%s<->%s", mgr->name, target);
1429 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1433 for (i = 0; i < N_SCHEDULERS; i++) {
1434 struct pinsched **s = &ofconn->schedulers[i];
1438 *s = pinsched_create(rate, burst);
1440 pinsched_set_limits(*s, rate, burst);
1443 pinsched_destroy(*s);
1450 ofconn_send(const struct ofconn *ofconn, struct ofpbuf *msg,
1451 struct rconn_packet_counter *counter)
1453 ofpmsg_update_length(msg);
1454 rconn_send(ofconn->rconn, msg, counter);
1457 /* Sending asynchronous messages. */
1459 static void schedule_packet_in(struct ofconn *, struct ofproto_packet_in,
1460 enum ofp_packet_in_reason wire_reason);
1462 /* Sends an OFPT_PORT_STATUS message with 'opp' and 'reason' to appropriate
1463 * controllers managed by 'mgr'. */
1465 connmgr_send_port_status(struct connmgr *mgr,
1466 const struct ofputil_phy_port *pp, uint8_t reason)
1468 /* XXX Should limit the number of queued port status change messages. */
1469 struct ofputil_port_status ps;
1470 struct ofconn *ofconn;
1474 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1475 if (ofconn_receives_async_msg(ofconn, OAM_PORT_STATUS, reason)) {
1478 msg = ofputil_encode_port_status(&ps, ofconn_get_protocol(ofconn));
1479 ofconn_send(ofconn, msg, NULL);
1484 /* Sends an OFPT_FLOW_REMOVED or NXT_FLOW_REMOVED message based on 'fr' to
1485 * appropriate controllers managed by 'mgr'. */
1487 connmgr_send_flow_removed(struct connmgr *mgr,
1488 const struct ofputil_flow_removed *fr)
1490 struct ofconn *ofconn;
1492 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1493 if (ofconn_receives_async_msg(ofconn, OAM_FLOW_REMOVED, fr->reason)) {
1496 /* Account flow expirations as replies to OpenFlow requests. That
1497 * works because preventing OpenFlow requests from being processed
1498 * also prevents new flows from being added (and expiring). (It
1499 * also prevents processing OpenFlow requests that would not add
1500 * new flows, so it is imperfect.) */
1501 msg = ofputil_encode_flow_removed(fr, ofconn_get_protocol(ofconn));
1502 ofconn_send_reply(ofconn, msg);
1507 /* Normally a send-to-controller action uses reason OFPR_ACTION. However, in
1508 * OpenFlow 1.3 and later, packet_ins generated by a send-to-controller action
1509 * in a "table-miss" flow (one with priority 0 and completely wildcarded) are
1510 * sent as OFPR_NO_MATCH. This function returns the reason that should
1511 * actually be sent on 'ofconn' for 'pin'. */
1512 static enum ofp_packet_in_reason
1513 wire_reason(struct ofconn *ofconn, const struct ofproto_packet_in *pin)
1515 if (pin->generated_by_table_miss && pin->up.reason == OFPR_ACTION) {
1516 enum ofputil_protocol protocol = ofconn_get_protocol(ofconn);
1518 if (protocol != OFPUTIL_P_NONE
1519 && ofputil_protocol_to_ofp_version(protocol) >= OFP13_VERSION) {
1520 return OFPR_NO_MATCH;
1523 return pin->up.reason;
1526 /* Given 'pin', sends an OFPT_PACKET_IN message to each OpenFlow controller as
1527 * necessary according to their individual configurations.
1529 * The caller doesn't need to fill in pin->buffer_id or pin->total_len. */
1531 connmgr_send_packet_in(struct connmgr *mgr,
1532 const struct ofproto_packet_in *pin)
1534 struct ofconn *ofconn;
1536 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1537 enum ofp_packet_in_reason reason = wire_reason(ofconn, pin);
1539 if (ofconn_receives_async_msg(ofconn, OAM_PACKET_IN, reason)
1540 && ofconn->controller_id == pin->controller_id) {
1541 schedule_packet_in(ofconn, *pin, reason);
1547 do_send_packet_ins(struct ofconn *ofconn, struct list *txq)
1549 struct ofpbuf *pin, *next_pin;
1551 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, txq) {
1552 list_remove(&pin->list_node);
1554 if (rconn_send_with_limit(ofconn->rconn, pin,
1555 ofconn->packet_in_counter, 100) == EAGAIN) {
1556 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5);
1558 VLOG_INFO_RL(&rl, "%s: dropping packet-in due to queue overflow",
1559 rconn_get_name(ofconn->rconn));
1564 /* Takes 'pin', composes an OpenFlow packet-in message from it, and passes it
1565 * to 'ofconn''s packet scheduler for sending. */
1567 schedule_packet_in(struct ofconn *ofconn, struct ofproto_packet_in pin,
1568 enum ofp_packet_in_reason wire_reason)
1570 struct connmgr *mgr = ofconn->connmgr;
1571 uint16_t controller_max_len;
1574 pin.up.total_len = pin.up.packet_len;
1576 pin.up.reason = wire_reason;
1577 if (pin.up.reason == OFPR_ACTION) {
1578 controller_max_len = pin.send_len; /* max_len */
1580 controller_max_len = ofconn->miss_send_len;
1583 /* Get OpenFlow buffer_id.
1584 * For OpenFlow 1.2+, OFPCML_NO_BUFFER (== UINT16_MAX) specifies
1585 * unbuffered. This behaviour doesn't violate prior versions, too. */
1586 if (controller_max_len == UINT16_MAX) {
1587 pin.up.buffer_id = UINT32_MAX;
1588 } else if (mgr->fail_open && fail_open_is_active(mgr->fail_open)) {
1589 pin.up.buffer_id = pktbuf_get_null();
1590 } else if (!ofconn->pktbuf) {
1591 pin.up.buffer_id = UINT32_MAX;
1593 pin.up.buffer_id = pktbuf_save(ofconn->pktbuf,
1594 pin.up.packet, pin.up.packet_len,
1595 pin.up.fmd.in_port);
1598 /* Figure out how much of the packet to send.
1599 * If not buffered, send the entire packet. Otherwise, depending on
1600 * the reason of packet-in, send what requested by the controller. */
1601 if (pin.up.buffer_id != UINT32_MAX
1602 && controller_max_len < pin.up.packet_len) {
1603 pin.up.packet_len = controller_max_len;
1606 /* Make OFPT_PACKET_IN and hand over to packet scheduler. */
1607 pinsched_send(ofconn->schedulers[pin.up.reason == OFPR_NO_MATCH ? 0 : 1],
1609 ofputil_encode_packet_in(&pin.up,
1610 ofconn_get_protocol(ofconn),
1611 ofconn->packet_in_format),
1613 do_send_packet_ins(ofconn, &txq);
1616 /* Fail-open settings. */
1618 /* Returns the failure handling mode (OFPROTO_FAIL_SECURE or
1619 * OFPROTO_FAIL_STANDALONE) for 'mgr'. */
1620 enum ofproto_fail_mode
1621 connmgr_get_fail_mode(const struct connmgr *mgr)
1623 return mgr->fail_mode;
1626 /* Sets the failure handling mode for 'mgr' to 'fail_mode' (either
1627 * OFPROTO_FAIL_SECURE or OFPROTO_FAIL_STANDALONE). */
1629 connmgr_set_fail_mode(struct connmgr *mgr, enum ofproto_fail_mode fail_mode)
1631 if (mgr->fail_mode != fail_mode) {
1632 mgr->fail_mode = fail_mode;
1633 update_fail_open(mgr);
1634 if (!connmgr_has_controllers(mgr)) {
1635 ofproto_flush_flows(mgr->ofproto);
1640 /* Fail-open implementation. */
1642 /* Returns the longest probe interval among the primary controllers configured
1643 * on 'mgr'. Returns 0 if there are no primary controllers. */
1645 connmgr_get_max_probe_interval(const struct connmgr *mgr)
1647 const struct ofconn *ofconn;
1648 int max_probe_interval;
1650 max_probe_interval = 0;
1651 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1652 int probe_interval = rconn_get_probe_interval(ofconn->rconn);
1653 max_probe_interval = MAX(max_probe_interval, probe_interval);
1655 return max_probe_interval;
1658 /* Returns the number of seconds for which all of 'mgr's primary controllers
1659 * have been disconnected. Returns 0 if 'mgr' has no primary controllers. */
1661 connmgr_failure_duration(const struct connmgr *mgr)
1663 const struct ofconn *ofconn;
1664 int min_failure_duration;
1666 if (!connmgr_has_controllers(mgr)) {
1670 min_failure_duration = INT_MAX;
1671 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1672 int failure_duration = rconn_failure_duration(ofconn->rconn);
1673 min_failure_duration = MIN(min_failure_duration, failure_duration);
1675 return min_failure_duration;
1678 /* Returns true if at least one primary controller is connected (regardless of
1679 * whether those controllers are believed to have authenticated and accepted
1680 * this switch), false if none of them are connected. */
1682 connmgr_is_any_controller_connected(const struct connmgr *mgr)
1684 const struct ofconn *ofconn;
1686 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1687 if (rconn_is_connected(ofconn->rconn)) {
1694 /* Returns true if at least one primary controller is believed to have
1695 * authenticated and accepted this switch, false otherwise. */
1697 connmgr_is_any_controller_admitted(const struct connmgr *mgr)
1699 const struct ofconn *ofconn;
1701 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1702 if (rconn_is_admitted(ofconn->rconn)) {
1709 /* In-band configuration. */
1711 static bool any_extras_changed(const struct connmgr *,
1712 const struct sockaddr_in *extras, size_t n);
1714 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'mgr''s
1715 * in-band control should guarantee access, in the same way that in-band
1716 * control guarantees access to OpenFlow controllers. */
1718 connmgr_set_extra_in_band_remotes(struct connmgr *mgr,
1719 const struct sockaddr_in *extras, size_t n)
1721 if (!any_extras_changed(mgr, extras, n)) {
1725 free(mgr->extra_in_band_remotes);
1726 mgr->n_extra_remotes = n;
1727 mgr->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
1729 update_in_band_remotes(mgr);
1732 /* Sets the OpenFlow queue used by flows set up by in-band control on
1733 * 'mgr' to 'queue_id'. If 'queue_id' is negative, then in-band control
1734 * flows will use the default queue. */
1736 connmgr_set_in_band_queue(struct connmgr *mgr, int queue_id)
1738 if (queue_id != mgr->in_band_queue) {
1739 mgr->in_band_queue = queue_id;
1740 update_in_band_remotes(mgr);
1745 any_extras_changed(const struct connmgr *mgr,
1746 const struct sockaddr_in *extras, size_t n)
1750 if (n != mgr->n_extra_remotes) {
1754 for (i = 0; i < n; i++) {
1755 const struct sockaddr_in *old = &mgr->extra_in_band_remotes[i];
1756 const struct sockaddr_in *new = &extras[i];
1758 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
1759 old->sin_port != new->sin_port) {
1767 /* In-band implementation. */
1770 connmgr_has_in_band(struct connmgr *mgr)
1772 return mgr->in_band != NULL;
1775 /* Fail-open and in-band implementation. */
1777 /* Called by 'ofproto' after all flows have been flushed, to allow fail-open
1778 * and standalone mode to re-create their flows.
1780 * In-band control has more sophisticated code that manages flows itself. */
1782 connmgr_flushed(struct connmgr *mgr)
1783 OVS_EXCLUDED(ofproto_mutex)
1785 if (mgr->fail_open) {
1786 fail_open_flushed(mgr->fail_open);
1789 /* If there are no controllers and we're in standalone mode, set up a flow
1790 * that matches every packet and directs them to OFPP_NORMAL (which goes to
1791 * us). Otherwise, the switch is in secure mode and we won't pass any
1792 * traffic until a controller has been defined and it tells us to do so. */
1793 if (!connmgr_has_controllers(mgr)
1794 && mgr->fail_mode == OFPROTO_FAIL_STANDALONE) {
1795 struct ofpbuf ofpacts;
1798 ofpbuf_init(&ofpacts, OFPACT_OUTPUT_SIZE);
1799 ofpact_put_OUTPUT(&ofpacts)->port = OFPP_NORMAL;
1800 ofpact_pad(&ofpacts);
1802 match_init_catchall(&match);
1803 ofproto_add_flow(mgr->ofproto, &match, 0, ofpacts.data, ofpacts.size);
1805 ofpbuf_uninit(&ofpacts);
1809 /* Creates a new ofservice for 'target' in 'mgr'. Returns 0 if successful,
1810 * otherwise a positive errno value.
1812 * ofservice_reconfigure() must be called to fully configure the new
1815 ofservice_create(struct connmgr *mgr, const char *target,
1816 uint32_t allowed_versions, uint8_t dscp)
1818 struct ofservice *ofservice;
1819 struct pvconn *pvconn;
1822 error = pvconn_open(target, allowed_versions, dscp, &pvconn);
1827 ofservice = xzalloc(sizeof *ofservice);
1828 hmap_insert(&mgr->services, &ofservice->node, hash_string(target, 0));
1829 ofservice->pvconn = pvconn;
1830 ofservice->allowed_versions = allowed_versions;
1836 ofservice_destroy(struct connmgr *mgr, struct ofservice *ofservice)
1838 hmap_remove(&mgr->services, &ofservice->node);
1839 pvconn_close(ofservice->pvconn);
1844 ofservice_reconfigure(struct ofservice *ofservice,
1845 const struct ofproto_controller *c)
1847 ofservice->probe_interval = c->probe_interval;
1848 ofservice->rate_limit = c->rate_limit;
1849 ofservice->burst_limit = c->burst_limit;
1850 ofservice->enable_async_msgs = c->enable_async_msgs;
1851 ofservice->dscp = c->dscp;
1854 /* Finds and returns the ofservice within 'mgr' that has the given
1855 * 'target', or a null pointer if none exists. */
1856 static struct ofservice *
1857 ofservice_lookup(struct connmgr *mgr, const char *target)
1859 struct ofservice *ofservice;
1861 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1863 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1870 /* Flow monitors (NXST_FLOW_MONITOR). */
1872 /* A counter incremented when something significant happens to an OpenFlow
1875 * - When a rule is added, its 'add_seqno' and 'modify_seqno' are set to
1876 * the current value (which is then incremented).
1878 * - When a rule is modified, its 'modify_seqno' is set to the current
1879 * value (which is then incremented).
1881 * Thus, by comparing an old value of monitor_seqno against a rule's
1882 * 'add_seqno', one can tell whether the rule was added before or after the old
1883 * value was read, and similarly for 'modify_seqno'.
1885 * 32 bits should normally be sufficient (and would be nice, to save space in
1886 * each rule) but then we'd have to have some special cases for wraparound.
1888 * We initialize monitor_seqno to 1 to allow 0 to be used as an invalid
1890 static uint64_t monitor_seqno = 1;
1892 COVERAGE_DEFINE(ofmonitor_pause);
1893 COVERAGE_DEFINE(ofmonitor_resume);
1896 ofmonitor_create(const struct ofputil_flow_monitor_request *request,
1897 struct ofconn *ofconn, struct ofmonitor **monitorp)
1898 OVS_REQUIRES(ofproto_mutex)
1900 struct ofmonitor *m;
1904 m = ofmonitor_lookup(ofconn, request->id);
1906 return OFPERR_NXBRC_FM_DUPLICATE_ID;
1909 m = xmalloc(sizeof *m);
1911 hmap_insert(&ofconn->monitors, &m->ofconn_node, hash_int(request->id, 0));
1912 m->id = request->id;
1913 m->flags = request->flags;
1914 m->out_port = request->out_port;
1915 m->table_id = request->table_id;
1916 minimatch_init(&m->match, &request->match);
1923 ofmonitor_lookup(struct ofconn *ofconn, uint32_t id)
1924 OVS_REQUIRES(ofproto_mutex)
1926 struct ofmonitor *m;
1928 HMAP_FOR_EACH_IN_BUCKET (m, ofconn_node, hash_int(id, 0),
1929 &ofconn->monitors) {
1938 ofmonitor_destroy(struct ofmonitor *m)
1939 OVS_REQUIRES(ofproto_mutex)
1942 minimatch_destroy(&m->match);
1943 hmap_remove(&m->ofconn->monitors, &m->ofconn_node);
1949 ofmonitor_report(struct connmgr *mgr, struct rule *rule,
1950 enum nx_flow_update_event event,
1951 enum ofp_flow_removed_reason reason,
1952 const struct ofconn *abbrev_ofconn, ovs_be32 abbrev_xid)
1953 OVS_REQUIRES(ofproto_mutex)
1955 enum nx_flow_monitor_flags update;
1956 struct ofconn *ofconn;
1961 rule->add_seqno = rule->modify_seqno = monitor_seqno++;
1965 update = NXFMF_DELETE;
1968 case NXFME_MODIFIED:
1969 update = NXFMF_MODIFY;
1970 rule->modify_seqno = monitor_seqno++;
1978 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1979 enum nx_flow_monitor_flags flags = 0;
1980 struct ofmonitor *m;
1982 if (ofconn->monitor_paused) {
1983 /* Only send NXFME_DELETED notifications for flows that were added
1984 * before we paused. */
1985 if (event != NXFME_DELETED
1986 || rule->add_seqno > ofconn->monitor_paused) {
1991 HMAP_FOR_EACH (m, ofconn_node, &ofconn->monitors) {
1992 if (m->flags & update
1993 && (m->table_id == 0xff || m->table_id == rule->table_id)
1994 && ofoperation_has_out_port(rule->pending, m->out_port)
1995 && cls_rule_is_loose_match(&rule->cr, &m->match)) {
2001 if (list_is_empty(&ofconn->updates)) {
2002 ofputil_start_flow_update(&ofconn->updates);
2003 ofconn->sent_abbrev_update = false;
2006 if (ofconn != abbrev_ofconn || ofconn->monitor_paused) {
2007 struct ofputil_flow_update fu;
2011 fu.reason = event == NXFME_DELETED ? reason : 0;
2012 fu.table_id = rule->table_id;
2013 fu.cookie = rule->flow_cookie;
2014 minimatch_expand(&rule->cr.match, &match);
2016 fu.priority = rule->cr.priority;
2018 ovs_mutex_lock(&rule->mutex);
2019 fu.idle_timeout = rule->idle_timeout;
2020 fu.hard_timeout = rule->hard_timeout;
2021 ovs_mutex_unlock(&rule->mutex);
2023 if (flags & NXFMF_ACTIONS) {
2024 fu.ofpacts = rule->actions->ofpacts;
2025 fu.ofpacts_len = rule->actions->ofpacts_len;
2030 ofputil_append_flow_update(&fu, &ofconn->updates);
2031 } else if (!ofconn->sent_abbrev_update) {
2032 struct ofputil_flow_update fu;
2034 fu.event = NXFME_ABBREV;
2035 fu.xid = abbrev_xid;
2036 ofputil_append_flow_update(&fu, &ofconn->updates);
2038 ofconn->sent_abbrev_update = true;
2045 ofmonitor_flush(struct connmgr *mgr)
2046 OVS_REQUIRES(ofproto_mutex)
2048 struct ofconn *ofconn;
2050 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2051 struct ofpbuf *msg, *next;
2053 LIST_FOR_EACH_SAFE (msg, next, list_node, &ofconn->updates) {
2054 unsigned int n_bytes;
2056 list_remove(&msg->list_node);
2057 ofconn_send(ofconn, msg, ofconn->monitor_counter);
2058 n_bytes = rconn_packet_counter_n_bytes(ofconn->monitor_counter);
2059 if (!ofconn->monitor_paused && n_bytes > 128 * 1024) {
2060 struct ofpbuf *pause;
2062 COVERAGE_INC(ofmonitor_pause);
2063 ofconn->monitor_paused = monitor_seqno++;
2064 pause = ofpraw_alloc_xid(OFPRAW_NXT_FLOW_MONITOR_PAUSED,
2065 OFP10_VERSION, htonl(0), 0);
2066 ofconn_send(ofconn, pause, ofconn->monitor_counter);
2073 ofmonitor_resume(struct ofconn *ofconn)
2074 OVS_REQUIRES(ofproto_mutex)
2076 struct rule_collection rules;
2077 struct ofpbuf *resumed;
2078 struct ofmonitor *m;
2081 rule_collection_init(&rules);
2082 HMAP_FOR_EACH (m, ofconn_node, &ofconn->monitors) {
2083 ofmonitor_collect_resume_rules(m, ofconn->monitor_paused, &rules);
2087 ofmonitor_compose_refresh_updates(&rules, &msgs);
2089 resumed = ofpraw_alloc_xid(OFPRAW_NXT_FLOW_MONITOR_RESUMED, OFP10_VERSION,
2091 list_push_back(&msgs, &resumed->list_node);
2092 ofconn_send_replies(ofconn, &msgs);
2094 ofconn->monitor_paused = 0;
2098 ofmonitor_may_resume(const struct ofconn *ofconn)
2099 OVS_REQUIRES(ofproto_mutex)
2101 return (ofconn->monitor_paused != 0
2102 && !rconn_packet_counter_n_packets(ofconn->monitor_counter));
2106 ofmonitor_run(struct connmgr *mgr)
2108 struct ofconn *ofconn;
2110 ovs_mutex_lock(&ofproto_mutex);
2111 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2112 if (ofmonitor_may_resume(ofconn)) {
2113 COVERAGE_INC(ofmonitor_resume);
2114 ofmonitor_resume(ofconn);
2117 ovs_mutex_unlock(&ofproto_mutex);
2121 ofmonitor_wait(struct connmgr *mgr)
2123 struct ofconn *ofconn;
2125 ovs_mutex_lock(&ofproto_mutex);
2126 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2127 if (ofmonitor_may_resume(ofconn)) {
2128 poll_immediate_wake();
2131 ovs_mutex_unlock(&ofproto_mutex);