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 *);
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 update_in_band_remotes(mgr);
641 update_fail_open(mgr);
642 if (had_controllers != connmgr_has_controllers(mgr)) {
643 ofproto_flush_flows(mgr->ofproto);
645 ovs_mutex_unlock(&ofproto_mutex);
648 /* Drops the connections between 'mgr' and all of its primary and secondary
649 * controllers, forcing them to reconnect. */
651 connmgr_reconnect(const struct connmgr *mgr)
653 struct ofconn *ofconn;
655 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
656 rconn_reconnect(ofconn->rconn);
660 /* Sets the "snoops" for 'mgr' to the pvconn targets listed in 'snoops'.
662 * A "snoop" is a pvconn to which every OpenFlow message to or from the most
663 * important controller on 'mgr' is mirrored. */
665 connmgr_set_snoops(struct connmgr *mgr, const struct sset *snoops)
667 return set_pvconns(&mgr->snoops, &mgr->n_snoops, snoops);
670 /* Adds each of the snoops currently configured on 'mgr' to 'snoops'. */
672 connmgr_get_snoops(const struct connmgr *mgr, struct sset *snoops)
676 for (i = 0; i < mgr->n_snoops; i++) {
677 sset_add(snoops, pvconn_get_name(mgr->snoops[i]));
681 /* Returns true if 'mgr' has at least one snoop, false if it has none. */
683 connmgr_has_snoops(const struct connmgr *mgr)
685 return mgr->n_snoops > 0;
688 /* Creates a new controller for 'target' in 'mgr'. update_controller() needs
689 * to be called later to finish the new ofconn's configuration. */
691 add_controller(struct connmgr *mgr, const char *target, uint8_t dscp,
692 uint32_t allowed_versions)
693 OVS_REQUIRES(ofproto_mutex)
695 char *name = ofconn_make_name(mgr, target);
696 struct ofconn *ofconn;
698 ofconn = ofconn_create(mgr, rconn_create(5, 8, dscp, allowed_versions),
699 OFCONN_PRIMARY, true);
700 ofconn->pktbuf = pktbuf_create();
701 rconn_connect(ofconn->rconn, target, name);
702 hmap_insert(&mgr->controllers, &ofconn->hmap_node, hash_string(target, 0));
707 static struct ofconn *
708 find_controller_by_target(struct connmgr *mgr, const char *target)
710 struct ofconn *ofconn;
712 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
713 hash_string(target, 0), &mgr->controllers) {
714 if (!strcmp(ofconn_get_target(ofconn), target)) {
722 update_in_band_remotes(struct connmgr *mgr)
724 struct sockaddr_in *addrs;
725 size_t max_addrs, n_addrs;
726 struct ofconn *ofconn;
729 /* Allocate enough memory for as many remotes as we could possibly have. */
730 max_addrs = mgr->n_extra_remotes + hmap_count(&mgr->controllers);
731 addrs = xmalloc(max_addrs * sizeof *addrs);
734 /* Add all the remotes. */
735 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
736 struct sockaddr_in *sin = &addrs[n_addrs];
737 const char *target = rconn_get_target(ofconn->rconn);
739 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
743 if (stream_parse_target_with_default_port(target,
749 for (i = 0; i < mgr->n_extra_remotes; i++) {
750 addrs[n_addrs++] = mgr->extra_in_band_remotes[i];
753 /* Create or update or destroy in-band. */
756 in_band_create(mgr->ofproto, mgr->local_port_name, &mgr->in_band);
758 in_band_set_queue(mgr->in_band, mgr->in_band_queue);
760 /* in_band_run() needs a chance to delete any existing in-band flows.
761 * We will destroy mgr->in_band after it's done with that. */
764 in_band_set_remotes(mgr->in_band, addrs, n_addrs);
772 update_fail_open(struct connmgr *mgr)
774 if (connmgr_has_controllers(mgr)
775 && mgr->fail_mode == OFPROTO_FAIL_STANDALONE) {
776 if (!mgr->fail_open) {
777 mgr->fail_open = fail_open_create(mgr->ofproto, mgr);
780 fail_open_destroy(mgr->fail_open);
781 mgr->fail_open = NULL;
786 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
787 const struct sset *sset)
789 struct pvconn **pvconns = *pvconnsp;
790 size_t n_pvconns = *n_pvconnsp;
795 for (i = 0; i < n_pvconns; i++) {
796 pvconn_close(pvconns[i]);
800 pvconns = xmalloc(sset_count(sset) * sizeof *pvconns);
802 SSET_FOR_EACH (name, sset) {
803 struct pvconn *pvconn;
805 error = pvconn_open(name, 0, 0, &pvconn);
807 pvconns[n_pvconns++] = pvconn;
809 VLOG_ERR("failed to listen on %s: %s", name, ovs_strerror(error));
817 *n_pvconnsp = n_pvconns;
822 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
823 * means that 'ofconn' is more interesting for monitoring than a lower return
826 snoop_preference(const struct ofconn *ofconn)
828 switch (ofconn->role) {
829 case OFPCR12_ROLE_MASTER:
831 case OFPCR12_ROLE_EQUAL:
833 case OFPCR12_ROLE_SLAVE:
835 case OFPCR12_ROLE_NOCHANGE:
837 /* Shouldn't happen. */
842 /* One of 'mgr''s "snoop" pvconns has accepted a new connection on 'vconn'.
843 * Connects this vconn to a controller. */
845 add_snooper(struct connmgr *mgr, struct vconn *vconn)
847 struct ofconn *ofconn, *best;
849 /* Pick a controller for monitoring. */
851 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
852 if (ofconn->type == OFCONN_PRIMARY
853 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
859 rconn_add_monitor(best->rconn, vconn);
861 VLOG_INFO_RL(&rl, "no controller connection to snoop");
866 /* Public ofconn functions. */
868 /* Returns the connection type, either OFCONN_PRIMARY or OFCONN_SERVICE. */
870 ofconn_get_type(const struct ofconn *ofconn)
875 /* If a master election id is defined, stores it into '*idp' and returns
876 * true. Otherwise, stores UINT64_MAX into '*idp' and returns false. */
878 ofconn_get_master_election_id(const struct ofconn *ofconn, uint64_t *idp)
880 *idp = (ofconn->connmgr->master_election_id_defined
881 ? ofconn->connmgr->master_election_id
883 return ofconn->connmgr->master_election_id_defined;
886 /* Sets the master election id.
888 * Returns true if successful, false if the id is stale
891 ofconn_set_master_election_id(struct ofconn *ofconn, uint64_t id)
893 if (ofconn->connmgr->master_election_id_defined
895 /* Unsigned difference interpreted as a two's complement signed
897 (int64_t)(id - ofconn->connmgr->master_election_id) < 0) {
900 ofconn->connmgr->master_election_id = id;
901 ofconn->connmgr->master_election_id_defined = true;
906 /* Returns the role configured for 'ofconn'.
908 * The default role, if no other role has been set, is OFPCR12_ROLE_EQUAL. */
909 enum ofp12_controller_role
910 ofconn_get_role(const struct ofconn *ofconn)
916 ofconn_send_role_status(struct ofconn *ofconn, uint32_t role, uint8_t reason)
918 struct ofputil_role_status status;
921 status.reason = reason;
923 ofconn_get_master_election_id(ofconn, &status.generation_id);
925 buf = ofputil_encode_role_status(&status, ofconn_get_protocol(ofconn));
927 ofconn_send(ofconn, buf, NULL);
930 /* Changes 'ofconn''s role to 'role'. If 'role' is OFPCR12_ROLE_MASTER then
931 * any existing master is demoted to a slave. */
933 ofconn_set_role(struct ofconn *ofconn, enum ofp12_controller_role role)
935 if (role != ofconn->role && role == OFPCR12_ROLE_MASTER) {
936 struct ofconn *other;
938 HMAP_FOR_EACH (other, hmap_node, &ofconn->connmgr->controllers) {
939 if (other->role == OFPCR12_ROLE_MASTER) {
940 other->role = OFPCR12_ROLE_SLAVE;
941 ofconn_send_role_status(other, OFPCR12_ROLE_SLAVE, OFPCRR_MASTER_REQUEST);
949 ofconn_set_invalid_ttl_to_controller(struct ofconn *ofconn, bool enable)
951 uint32_t bit = 1u << OFPR_INVALID_TTL;
953 ofconn->master_async_config[OAM_PACKET_IN] |= bit;
955 ofconn->master_async_config[OAM_PACKET_IN] &= ~bit;
960 ofconn_get_invalid_ttl_to_controller(struct ofconn *ofconn)
962 uint32_t bit = 1u << OFPR_INVALID_TTL;
963 return (ofconn->master_async_config[OAM_PACKET_IN] & bit) != 0;
966 /* Returns the currently configured protocol for 'ofconn', one of OFPUTIL_P_*.
968 * Returns OFPUTIL_P_NONE, which is not a valid protocol, if 'ofconn' hasn't
969 * completed version negotiation. This can't happen if at least one OpenFlow
970 * message, other than OFPT_HELLO, has been received on the connection (such as
971 * in ofproto.c's message handling code), since version negotiation is a
972 * prerequisite for starting to receive messages. This means that
973 * OFPUTIL_P_NONE is a special case that most callers need not worry about. */
974 enum ofputil_protocol
975 ofconn_get_protocol(const struct ofconn *ofconn)
977 if (ofconn->protocol == OFPUTIL_P_NONE &&
978 rconn_is_connected(ofconn->rconn)) {
979 int version = rconn_get_version(ofconn->rconn);
981 ofconn_set_protocol(CONST_CAST(struct ofconn *, ofconn),
982 ofputil_protocol_from_ofp_version(version));
986 return ofconn->protocol;
989 /* Sets the protocol for 'ofconn' to 'protocol' (one of OFPUTIL_P_*).
991 * (This doesn't actually send anything to accomplish this. Presumably the
992 * caller already did that.) */
994 ofconn_set_protocol(struct ofconn *ofconn, enum ofputil_protocol protocol)
996 ofconn->protocol = protocol;
999 /* Returns the currently configured packet in format for 'ofconn', one of
1002 * The default, if no other format has been set, is NXPIF_OPENFLOW10. */
1003 enum nx_packet_in_format
1004 ofconn_get_packet_in_format(struct ofconn *ofconn)
1006 return ofconn->packet_in_format;
1009 /* Sets the packet in format for 'ofconn' to 'packet_in_format' (one of
1012 ofconn_set_packet_in_format(struct ofconn *ofconn,
1013 enum nx_packet_in_format packet_in_format)
1015 ofconn->packet_in_format = packet_in_format;
1018 /* Sets the controller connection ID for 'ofconn' to 'controller_id'.
1020 * The connection controller ID is used for OFPP_CONTROLLER and
1021 * NXAST_CONTROLLER actions. See "struct nx_action_controller" for details. */
1023 ofconn_set_controller_id(struct ofconn *ofconn, uint16_t controller_id)
1025 ofconn->controller_id = controller_id;
1028 /* Returns the default miss send length for 'ofconn'. */
1030 ofconn_get_miss_send_len(const struct ofconn *ofconn)
1032 return ofconn->miss_send_len;
1035 /* Sets the default miss send length for 'ofconn' to 'miss_send_len'. */
1037 ofconn_set_miss_send_len(struct ofconn *ofconn, int miss_send_len)
1039 ofconn->miss_send_len = miss_send_len;
1043 ofconn_set_async_config(struct ofconn *ofconn,
1044 const uint32_t master_masks[OAM_N_TYPES],
1045 const uint32_t slave_masks[OAM_N_TYPES])
1047 size_t size = sizeof ofconn->master_async_config;
1048 memcpy(ofconn->master_async_config, master_masks, size);
1049 memcpy(ofconn->slave_async_config, slave_masks, size);
1053 ofconn_get_async_config(struct ofconn *ofconn,
1054 uint32_t *master_masks, uint32_t *slave_masks)
1056 size_t size = sizeof ofconn->master_async_config;
1057 memcpy(master_masks, ofconn->master_async_config, size);
1058 memcpy(slave_masks, ofconn->slave_async_config, size);
1061 /* Sends 'msg' on 'ofconn', accounting it as a reply. (If there is a
1062 * sufficient number of OpenFlow replies in-flight on a single ofconn, then the
1063 * connmgr will stop accepting new OpenFlow requests on that ofconn until the
1064 * controller has accepted some of the replies.) */
1066 ofconn_send_reply(const struct ofconn *ofconn, struct ofpbuf *msg)
1068 ofconn_send(ofconn, msg, ofconn->reply_counter);
1071 /* Sends each of the messages in list 'replies' on 'ofconn' in order,
1072 * accounting them as replies. */
1074 ofconn_send_replies(const struct ofconn *ofconn, struct list *replies)
1076 struct ofpbuf *reply, *next;
1078 LIST_FOR_EACH_SAFE (reply, next, list_node, replies) {
1079 list_remove(&reply->list_node);
1080 ofconn_send_reply(ofconn, reply);
1084 /* Sends 'error' on 'ofconn', as a reply to 'request'. Only at most the
1085 * first 64 bytes of 'request' are used. */
1087 ofconn_send_error(const struct ofconn *ofconn,
1088 const struct ofp_header *request, enum ofperr error)
1090 static struct vlog_rate_limit err_rl = VLOG_RATE_LIMIT_INIT(10, 10);
1091 struct ofpbuf *reply;
1093 reply = ofperr_encode_reply(error, request);
1094 if (!VLOG_DROP_INFO(&err_rl)) {
1095 const char *type_name;
1099 request_len = ntohs(request->length);
1100 type_name = (!ofpraw_decode_partial(&raw, request,
1101 MIN(64, request_len))
1102 ? ofpraw_get_name(raw)
1105 VLOG_INFO("%s: sending %s error reply to %s message",
1106 rconn_get_name(ofconn->rconn), ofperr_to_string(error),
1109 ofconn_send_reply(ofconn, reply);
1112 /* Same as pktbuf_retrieve(), using the pktbuf owned by 'ofconn'. */
1114 ofconn_pktbuf_retrieve(struct ofconn *ofconn, uint32_t id,
1115 struct ofpbuf **bufferp, ofp_port_t *in_port)
1117 return pktbuf_retrieve(ofconn->pktbuf, id, bufferp, in_port);
1120 /* Returns true if 'ofconn' has any pending opgroups. */
1122 ofconn_has_pending_opgroups(const struct ofconn *ofconn)
1124 return !list_is_empty(&ofconn->opgroups);
1127 /* Adds 'ofconn_node' to 'ofconn''s list of pending opgroups.
1129 * If 'ofconn' is destroyed or its connection drops, then 'ofconn' will remove
1130 * 'ofconn_node' from the list and re-initialize it with list_init(). The
1131 * client may, therefore, use list_is_empty(ofconn_node) to determine whether
1132 * 'ofconn_node' is still associated with an active ofconn.
1134 * The client may also remove ofconn_node from the list itself, with
1137 ofconn_add_opgroup(struct ofconn *ofconn, struct list *ofconn_node)
1139 list_push_back(&ofconn->opgroups, ofconn_node);
1142 /* Private ofconn functions. */
1145 ofconn_get_target(const struct ofconn *ofconn)
1147 return rconn_get_target(ofconn->rconn);
1150 static struct ofconn *
1151 ofconn_create(struct connmgr *mgr, struct rconn *rconn, enum ofconn_type type,
1152 bool enable_async_msgs)
1154 struct ofconn *ofconn;
1156 ofconn = xzalloc(sizeof *ofconn);
1157 ofconn->connmgr = mgr;
1158 list_push_back(&mgr->all_conns, &ofconn->node);
1159 ofconn->rconn = rconn;
1160 ofconn->type = type;
1161 ofconn->enable_async_msgs = enable_async_msgs;
1163 list_init(&ofconn->opgroups);
1165 hmap_init(&ofconn->monitors);
1166 list_init(&ofconn->updates);
1168 ofconn_flush(ofconn);
1173 /* Clears all of the state in 'ofconn' that should not persist from one
1174 * connection to the next. */
1176 ofconn_flush(struct ofconn *ofconn)
1177 OVS_REQUIRES(ofproto_mutex)
1179 struct ofmonitor *monitor, *next_monitor;
1182 ofconn->role = OFPCR12_ROLE_EQUAL;
1183 ofconn_set_protocol(ofconn, OFPUTIL_P_NONE);
1184 ofconn->packet_in_format = NXPIF_OPENFLOW10;
1186 /* Disassociate 'ofconn' from all of the ofopgroups that it initiated that
1187 * have not yet completed. (Those ofopgroups will still run to completion
1188 * in the usual way, but any errors that they run into will not be reported
1189 * on any OpenFlow channel.)
1191 * Also discard any blocked operation on 'ofconn'. */
1192 while (!list_is_empty(&ofconn->opgroups)) {
1193 list_init(list_pop_front(&ofconn->opgroups));
1195 ofpbuf_delete(ofconn->blocked);
1196 ofconn->blocked = NULL;
1198 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1199 ofconn->packet_in_counter = rconn_packet_counter_create();
1200 for (i = 0; i < N_SCHEDULERS; i++) {
1201 if (ofconn->schedulers[i]) {
1204 pinsched_get_limits(ofconn->schedulers[i], &rate, &burst);
1205 pinsched_destroy(ofconn->schedulers[i]);
1206 ofconn->schedulers[i] = pinsched_create(rate, burst);
1209 if (ofconn->pktbuf) {
1210 pktbuf_destroy(ofconn->pktbuf);
1211 ofconn->pktbuf = pktbuf_create();
1213 ofconn->miss_send_len = (ofconn->type == OFCONN_PRIMARY
1214 ? OFP_DEFAULT_MISS_SEND_LEN
1216 ofconn->controller_id = 0;
1218 rconn_packet_counter_destroy(ofconn->reply_counter);
1219 ofconn->reply_counter = rconn_packet_counter_create();
1221 if (ofconn->enable_async_msgs) {
1222 uint32_t *master = ofconn->master_async_config;
1223 uint32_t *slave = ofconn->slave_async_config;
1225 /* "master" and "other" roles get all asynchronous messages by default,
1226 * except that the controller needs to enable nonstandard "packet-in"
1227 * reasons itself. */
1228 master[OAM_PACKET_IN] = (1u << OFPR_NO_MATCH) | (1u << OFPR_ACTION);
1229 master[OAM_PORT_STATUS] = ((1u << OFPPR_ADD)
1230 | (1u << OFPPR_DELETE)
1231 | (1u << OFPPR_MODIFY));
1232 master[OAM_FLOW_REMOVED] = ((1u << OFPRR_IDLE_TIMEOUT)
1233 | (1u << OFPRR_HARD_TIMEOUT)
1234 | (1u << OFPRR_DELETE));
1236 /* "slave" role gets port status updates by default. */
1237 slave[OAM_PACKET_IN] = 0;
1238 slave[OAM_PORT_STATUS] = ((1u << OFPPR_ADD)
1239 | (1u << OFPPR_DELETE)
1240 | (1u << OFPPR_MODIFY));
1241 slave[OAM_FLOW_REMOVED] = 0;
1243 memset(ofconn->master_async_config, 0,
1244 sizeof ofconn->master_async_config);
1245 memset(ofconn->slave_async_config, 0,
1246 sizeof ofconn->slave_async_config);
1249 HMAP_FOR_EACH_SAFE (monitor, next_monitor, ofconn_node,
1250 &ofconn->monitors) {
1251 ofmonitor_destroy(monitor);
1253 rconn_packet_counter_destroy(ofconn->monitor_counter);
1254 ofconn->monitor_counter = rconn_packet_counter_create();
1255 ofpbuf_list_delete(&ofconn->updates); /* ...but it should be empty. */
1259 ofconn_destroy(struct ofconn *ofconn)
1260 OVS_REQUIRES(ofproto_mutex)
1262 ofconn_flush(ofconn);
1264 if (ofconn->type == OFCONN_PRIMARY) {
1265 hmap_remove(&ofconn->connmgr->controllers, &ofconn->hmap_node);
1268 hmap_destroy(&ofconn->monitors);
1269 list_remove(&ofconn->node);
1270 rconn_destroy(ofconn->rconn);
1271 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1272 rconn_packet_counter_destroy(ofconn->reply_counter);
1273 pktbuf_destroy(ofconn->pktbuf);
1274 rconn_packet_counter_destroy(ofconn->monitor_counter);
1278 /* Reconfigures 'ofconn' to match 'c'. 'ofconn' and 'c' must have the same
1281 ofconn_reconfigure(struct ofconn *ofconn, const struct ofproto_controller *c)
1285 ofconn->band = c->band;
1286 ofconn->enable_async_msgs = c->enable_async_msgs;
1288 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
1290 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
1291 rconn_set_probe_interval(ofconn->rconn, probe_interval);
1293 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
1295 /* If dscp value changed reconnect. */
1296 if (c->dscp != rconn_get_dscp(ofconn->rconn)) {
1297 rconn_set_dscp(ofconn->rconn, c->dscp);
1298 rconn_reconnect(ofconn->rconn);
1302 /* Returns true if it makes sense for 'ofconn' to receive and process OpenFlow
1305 ofconn_may_recv(const struct ofconn *ofconn)
1307 int count = rconn_packet_counter_n_packets(ofconn->reply_counter);
1308 return (!ofconn->blocked || ofconn->retry) && count < OFCONN_REPLY_MAX;
1312 ofconn_run(struct ofconn *ofconn,
1313 bool (*handle_openflow)(struct ofconn *,
1314 const struct ofpbuf *ofp_msg))
1316 struct connmgr *mgr = ofconn->connmgr;
1319 for (i = 0; i < N_SCHEDULERS; i++) {
1322 pinsched_run(ofconn->schedulers[i], &txq);
1323 do_send_packet_ins(ofconn, &txq);
1326 rconn_run(ofconn->rconn);
1328 if (handle_openflow) {
1329 /* Limit the number of iterations to avoid starving other tasks. */
1330 for (i = 0; i < 50 && ofconn_may_recv(ofconn); i++) {
1331 struct ofpbuf *of_msg;
1333 of_msg = (ofconn->blocked
1335 : rconn_recv(ofconn->rconn));
1339 if (mgr->fail_open) {
1340 fail_open_maybe_recover(mgr->fail_open);
1343 if (handle_openflow(ofconn, of_msg)) {
1344 ofpbuf_delete(of_msg);
1345 ofconn->blocked = NULL;
1347 ofconn->blocked = of_msg;
1348 ofconn->retry = false;
1353 ovs_mutex_lock(&ofproto_mutex);
1354 if (!rconn_is_alive(ofconn->rconn)) {
1355 ofconn_destroy(ofconn);
1356 } else if (!rconn_is_connected(ofconn->rconn)) {
1357 ofconn_flush(ofconn);
1359 ovs_mutex_unlock(&ofproto_mutex);
1363 ofconn_wait(struct ofconn *ofconn, bool handling_openflow)
1367 for (i = 0; i < N_SCHEDULERS; i++) {
1368 pinsched_wait(ofconn->schedulers[i]);
1370 rconn_run_wait(ofconn->rconn);
1371 if (handling_openflow && ofconn_may_recv(ofconn)) {
1372 rconn_recv_wait(ofconn->rconn);
1376 /* Returns true if 'ofconn' should receive asynchronous messages of the given
1377 * OAM_* 'type' and 'reason', which should be a OFPR_* value for OAM_PACKET_IN,
1378 * a OFPPR_* value for OAM_PORT_STATUS, or an OFPRR_* value for
1379 * OAM_FLOW_REMOVED. Returns false if the message should not be sent on
1382 ofconn_receives_async_msg(const struct ofconn *ofconn,
1383 enum ofconn_async_msg_type type,
1384 unsigned int reason)
1386 const uint32_t *async_config;
1388 ovs_assert(reason < 32);
1389 ovs_assert((unsigned int) type < OAM_N_TYPES);
1391 if (ofconn_get_protocol(ofconn) == OFPUTIL_P_NONE
1392 || !rconn_is_connected(ofconn->rconn)) {
1396 /* Keep the following code in sync with the documentation in the
1397 * "Asynchronous Messages" section in DESIGN. */
1399 if (ofconn->type == OFCONN_SERVICE && !ofconn->miss_send_len) {
1400 /* Service connections don't get asynchronous messages unless they have
1401 * explicitly asked for them by setting a nonzero miss send length. */
1405 async_config = (ofconn->role == OFPCR12_ROLE_SLAVE
1406 ? ofconn->slave_async_config
1407 : ofconn->master_async_config);
1408 if (!(async_config[type] & (1u << reason))) {
1415 /* Returns a human-readable name for an OpenFlow connection between 'mgr' and
1416 * 'target', suitable for use in log messages for identifying the connection.
1418 * The name is dynamically allocated. The caller should free it (with free())
1419 * when it is no longer needed. */
1421 ofconn_make_name(const struct connmgr *mgr, const char *target)
1423 return xasprintf("%s<->%s", mgr->name, target);
1427 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1431 for (i = 0; i < N_SCHEDULERS; i++) {
1432 struct pinsched **s = &ofconn->schedulers[i];
1436 *s = pinsched_create(rate, burst);
1438 pinsched_set_limits(*s, rate, burst);
1441 pinsched_destroy(*s);
1448 ofconn_send(const struct ofconn *ofconn, struct ofpbuf *msg,
1449 struct rconn_packet_counter *counter)
1451 ofpmsg_update_length(msg);
1452 rconn_send(ofconn->rconn, msg, counter);
1455 /* Sending asynchronous messages. */
1457 static void schedule_packet_in(struct ofconn *, struct ofproto_packet_in,
1458 enum ofp_packet_in_reason wire_reason);
1460 /* Sends an OFPT_PORT_STATUS message with 'opp' and 'reason' to appropriate
1461 * controllers managed by 'mgr'. */
1463 connmgr_send_port_status(struct connmgr *mgr,
1464 const struct ofputil_phy_port *pp, uint8_t reason)
1466 /* XXX Should limit the number of queued port status change messages. */
1467 struct ofputil_port_status ps;
1468 struct ofconn *ofconn;
1472 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1473 if (ofconn_receives_async_msg(ofconn, OAM_PORT_STATUS, reason)) {
1476 msg = ofputil_encode_port_status(&ps, ofconn_get_protocol(ofconn));
1477 ofconn_send(ofconn, msg, NULL);
1482 /* Sends an OFPT_FLOW_REMOVED or NXT_FLOW_REMOVED message based on 'fr' to
1483 * appropriate controllers managed by 'mgr'. */
1485 connmgr_send_flow_removed(struct connmgr *mgr,
1486 const struct ofputil_flow_removed *fr)
1488 struct ofconn *ofconn;
1490 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1491 if (ofconn_receives_async_msg(ofconn, OAM_FLOW_REMOVED, fr->reason)) {
1494 /* Account flow expirations as replies to OpenFlow requests. That
1495 * works because preventing OpenFlow requests from being processed
1496 * also prevents new flows from being added (and expiring). (It
1497 * also prevents processing OpenFlow requests that would not add
1498 * new flows, so it is imperfect.) */
1499 msg = ofputil_encode_flow_removed(fr, ofconn_get_protocol(ofconn));
1500 ofconn_send_reply(ofconn, msg);
1505 /* Normally a send-to-controller action uses reason OFPR_ACTION. However, in
1506 * OpenFlow 1.3 and later, packet_ins generated by a send-to-controller action
1507 * in a "table-miss" flow (one with priority 0 and completely wildcarded) are
1508 * sent as OFPR_NO_MATCH. This function returns the reason that should
1509 * actually be sent on 'ofconn' for 'pin'. */
1510 static enum ofp_packet_in_reason
1511 wire_reason(struct ofconn *ofconn, const struct ofproto_packet_in *pin)
1513 if (pin->generated_by_table_miss && pin->up.reason == OFPR_ACTION) {
1514 enum ofputil_protocol protocol = ofconn_get_protocol(ofconn);
1516 if (protocol != OFPUTIL_P_NONE
1517 && ofputil_protocol_to_ofp_version(protocol) >= OFP13_VERSION) {
1518 return OFPR_NO_MATCH;
1521 return pin->up.reason;
1524 /* Given 'pin', sends an OFPT_PACKET_IN message to each OpenFlow controller as
1525 * necessary according to their individual configurations.
1527 * The caller doesn't need to fill in pin->buffer_id or pin->total_len. */
1529 connmgr_send_packet_in(struct connmgr *mgr,
1530 const struct ofproto_packet_in *pin)
1532 struct ofconn *ofconn;
1534 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1535 enum ofp_packet_in_reason reason = wire_reason(ofconn, pin);
1537 if (ofconn_receives_async_msg(ofconn, OAM_PACKET_IN, reason)
1538 && ofconn->controller_id == pin->controller_id) {
1539 schedule_packet_in(ofconn, *pin, reason);
1545 do_send_packet_ins(struct ofconn *ofconn, struct list *txq)
1547 struct ofpbuf *pin, *next_pin;
1549 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, txq) {
1550 list_remove(&pin->list_node);
1552 rconn_send_with_limit(ofconn->rconn, pin,
1553 ofconn->packet_in_counter, 100);
1557 /* Takes 'pin', composes an OpenFlow packet-in message from it, and passes it
1558 * to 'ofconn''s packet scheduler for sending. */
1560 schedule_packet_in(struct ofconn *ofconn, struct ofproto_packet_in pin,
1561 enum ofp_packet_in_reason wire_reason)
1563 struct connmgr *mgr = ofconn->connmgr;
1564 uint16_t controller_max_len;
1567 pin.up.total_len = pin.up.packet_len;
1569 pin.up.reason = wire_reason;
1570 if (pin.up.reason == OFPR_ACTION) {
1571 controller_max_len = pin.send_len; /* max_len */
1573 controller_max_len = ofconn->miss_send_len;
1576 /* Get OpenFlow buffer_id.
1577 * For OpenFlow 1.2+, OFPCML_NO_BUFFER (== UINT16_MAX) specifies
1578 * unbuffered. This behaviour doesn't violate prior versions, too. */
1579 if (controller_max_len == UINT16_MAX) {
1580 pin.up.buffer_id = UINT32_MAX;
1581 } else if (mgr->fail_open && fail_open_is_active(mgr->fail_open)) {
1582 pin.up.buffer_id = pktbuf_get_null();
1583 } else if (!ofconn->pktbuf) {
1584 pin.up.buffer_id = UINT32_MAX;
1586 pin.up.buffer_id = pktbuf_save(ofconn->pktbuf,
1587 pin.up.packet, pin.up.packet_len,
1588 pin.up.fmd.in_port);
1591 /* Figure out how much of the packet to send.
1592 * If not buffered, send the entire packet. Otherwise, depending on
1593 * the reason of packet-in, send what requested by the controller. */
1594 if (pin.up.buffer_id != UINT32_MAX
1595 && controller_max_len < pin.up.packet_len) {
1596 pin.up.packet_len = controller_max_len;
1599 /* Make OFPT_PACKET_IN and hand over to packet scheduler. */
1600 pinsched_send(ofconn->schedulers[pin.up.reason == OFPR_NO_MATCH ? 0 : 1],
1602 ofputil_encode_packet_in(&pin.up,
1603 ofconn_get_protocol(ofconn),
1604 ofconn->packet_in_format),
1606 do_send_packet_ins(ofconn, &txq);
1609 /* Fail-open settings. */
1611 /* Returns the failure handling mode (OFPROTO_FAIL_SECURE or
1612 * OFPROTO_FAIL_STANDALONE) for 'mgr'. */
1613 enum ofproto_fail_mode
1614 connmgr_get_fail_mode(const struct connmgr *mgr)
1616 return mgr->fail_mode;
1619 /* Sets the failure handling mode for 'mgr' to 'fail_mode' (either
1620 * OFPROTO_FAIL_SECURE or OFPROTO_FAIL_STANDALONE). */
1622 connmgr_set_fail_mode(struct connmgr *mgr, enum ofproto_fail_mode fail_mode)
1624 if (mgr->fail_mode != fail_mode) {
1625 mgr->fail_mode = fail_mode;
1626 update_fail_open(mgr);
1627 if (!connmgr_has_controllers(mgr)) {
1628 ofproto_flush_flows(mgr->ofproto);
1633 /* Fail-open implementation. */
1635 /* Returns the longest probe interval among the primary controllers configured
1636 * on 'mgr'. Returns 0 if there are no primary controllers. */
1638 connmgr_get_max_probe_interval(const struct connmgr *mgr)
1640 const struct ofconn *ofconn;
1641 int max_probe_interval;
1643 max_probe_interval = 0;
1644 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1645 int probe_interval = rconn_get_probe_interval(ofconn->rconn);
1646 max_probe_interval = MAX(max_probe_interval, probe_interval);
1648 return max_probe_interval;
1651 /* Returns the number of seconds for which all of 'mgr's primary controllers
1652 * have been disconnected. Returns 0 if 'mgr' has no primary controllers. */
1654 connmgr_failure_duration(const struct connmgr *mgr)
1656 const struct ofconn *ofconn;
1657 int min_failure_duration;
1659 if (!connmgr_has_controllers(mgr)) {
1663 min_failure_duration = INT_MAX;
1664 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1665 int failure_duration = rconn_failure_duration(ofconn->rconn);
1666 min_failure_duration = MIN(min_failure_duration, failure_duration);
1668 return min_failure_duration;
1671 /* Returns true if at least one primary controller is connected (regardless of
1672 * whether those controllers are believed to have authenticated and accepted
1673 * this switch), false if none of them are connected. */
1675 connmgr_is_any_controller_connected(const struct connmgr *mgr)
1677 const struct ofconn *ofconn;
1679 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1680 if (rconn_is_connected(ofconn->rconn)) {
1687 /* Returns true if at least one primary controller is believed to have
1688 * authenticated and accepted this switch, false otherwise. */
1690 connmgr_is_any_controller_admitted(const struct connmgr *mgr)
1692 const struct ofconn *ofconn;
1694 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1695 if (rconn_is_admitted(ofconn->rconn)) {
1702 /* In-band configuration. */
1704 static bool any_extras_changed(const struct connmgr *,
1705 const struct sockaddr_in *extras, size_t n);
1707 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'mgr''s
1708 * in-band control should guarantee access, in the same way that in-band
1709 * control guarantees access to OpenFlow controllers. */
1711 connmgr_set_extra_in_band_remotes(struct connmgr *mgr,
1712 const struct sockaddr_in *extras, size_t n)
1714 if (!any_extras_changed(mgr, extras, n)) {
1718 free(mgr->extra_in_band_remotes);
1719 mgr->n_extra_remotes = n;
1720 mgr->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
1722 update_in_band_remotes(mgr);
1725 /* Sets the OpenFlow queue used by flows set up by in-band control on
1726 * 'mgr' to 'queue_id'. If 'queue_id' is negative, then in-band control
1727 * flows will use the default queue. */
1729 connmgr_set_in_band_queue(struct connmgr *mgr, int queue_id)
1731 if (queue_id != mgr->in_band_queue) {
1732 mgr->in_band_queue = queue_id;
1733 update_in_band_remotes(mgr);
1738 any_extras_changed(const struct connmgr *mgr,
1739 const struct sockaddr_in *extras, size_t n)
1743 if (n != mgr->n_extra_remotes) {
1747 for (i = 0; i < n; i++) {
1748 const struct sockaddr_in *old = &mgr->extra_in_band_remotes[i];
1749 const struct sockaddr_in *new = &extras[i];
1751 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
1752 old->sin_port != new->sin_port) {
1760 /* In-band implementation. */
1763 connmgr_has_in_band(struct connmgr *mgr)
1765 return mgr->in_band != NULL;
1768 /* Fail-open and in-band implementation. */
1770 /* Called by 'ofproto' after all flows have been flushed, to allow fail-open
1771 * and standalone mode to re-create their flows.
1773 * In-band control has more sophisticated code that manages flows itself. */
1775 connmgr_flushed(struct connmgr *mgr)
1776 OVS_EXCLUDED(ofproto_mutex)
1778 if (mgr->fail_open) {
1779 fail_open_flushed(mgr->fail_open);
1782 /* If there are no controllers and we're in standalone mode, set up a flow
1783 * that matches every packet and directs them to OFPP_NORMAL (which goes to
1784 * us). Otherwise, the switch is in secure mode and we won't pass any
1785 * traffic until a controller has been defined and it tells us to do so. */
1786 if (!connmgr_has_controllers(mgr)
1787 && mgr->fail_mode == OFPROTO_FAIL_STANDALONE) {
1788 struct ofpbuf ofpacts;
1791 ofpbuf_init(&ofpacts, OFPACT_OUTPUT_SIZE);
1792 ofpact_put_OUTPUT(&ofpacts)->port = OFPP_NORMAL;
1793 ofpact_pad(&ofpacts);
1795 match_init_catchall(&match);
1796 ofproto_add_flow(mgr->ofproto, &match, 0, ofpacts.data, ofpacts.size);
1798 ofpbuf_uninit(&ofpacts);
1802 /* Creates a new ofservice for 'target' in 'mgr'. Returns 0 if successful,
1803 * otherwise a positive errno value.
1805 * ofservice_reconfigure() must be called to fully configure the new
1808 ofservice_create(struct connmgr *mgr, const char *target,
1809 uint32_t allowed_versions, uint8_t dscp)
1811 struct ofservice *ofservice;
1812 struct pvconn *pvconn;
1815 error = pvconn_open(target, allowed_versions, dscp, &pvconn);
1820 ofservice = xzalloc(sizeof *ofservice);
1821 hmap_insert(&mgr->services, &ofservice->node, hash_string(target, 0));
1822 ofservice->pvconn = pvconn;
1823 ofservice->allowed_versions = allowed_versions;
1829 ofservice_destroy(struct connmgr *mgr, struct ofservice *ofservice)
1831 hmap_remove(&mgr->services, &ofservice->node);
1832 pvconn_close(ofservice->pvconn);
1837 ofservice_reconfigure(struct ofservice *ofservice,
1838 const struct ofproto_controller *c)
1840 ofservice->probe_interval = c->probe_interval;
1841 ofservice->rate_limit = c->rate_limit;
1842 ofservice->burst_limit = c->burst_limit;
1843 ofservice->enable_async_msgs = c->enable_async_msgs;
1844 ofservice->dscp = c->dscp;
1847 /* Finds and returns the ofservice within 'mgr' that has the given
1848 * 'target', or a null pointer if none exists. */
1849 static struct ofservice *
1850 ofservice_lookup(struct connmgr *mgr, const char *target)
1852 struct ofservice *ofservice;
1854 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1856 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1863 /* Flow monitors (NXST_FLOW_MONITOR). */
1865 /* A counter incremented when something significant happens to an OpenFlow
1868 * - When a rule is added, its 'add_seqno' and 'modify_seqno' are set to
1869 * the current value (which is then incremented).
1871 * - When a rule is modified, its 'modify_seqno' is set to the current
1872 * value (which is then incremented).
1874 * Thus, by comparing an old value of monitor_seqno against a rule's
1875 * 'add_seqno', one can tell whether the rule was added before or after the old
1876 * value was read, and similarly for 'modify_seqno'.
1878 * 32 bits should normally be sufficient (and would be nice, to save space in
1879 * each rule) but then we'd have to have some special cases for wraparound.
1881 * We initialize monitor_seqno to 1 to allow 0 to be used as an invalid
1883 static uint64_t monitor_seqno = 1;
1885 COVERAGE_DEFINE(ofmonitor_pause);
1886 COVERAGE_DEFINE(ofmonitor_resume);
1889 ofmonitor_create(const struct ofputil_flow_monitor_request *request,
1890 struct ofconn *ofconn, struct ofmonitor **monitorp)
1891 OVS_REQUIRES(ofproto_mutex)
1893 struct ofmonitor *m;
1897 m = ofmonitor_lookup(ofconn, request->id);
1899 return OFPERR_NXBRC_FM_DUPLICATE_ID;
1902 m = xmalloc(sizeof *m);
1904 hmap_insert(&ofconn->monitors, &m->ofconn_node, hash_int(request->id, 0));
1905 m->id = request->id;
1906 m->flags = request->flags;
1907 m->out_port = request->out_port;
1908 m->table_id = request->table_id;
1909 minimatch_init(&m->match, &request->match);
1916 ofmonitor_lookup(struct ofconn *ofconn, uint32_t id)
1917 OVS_REQUIRES(ofproto_mutex)
1919 struct ofmonitor *m;
1921 HMAP_FOR_EACH_IN_BUCKET (m, ofconn_node, hash_int(id, 0),
1922 &ofconn->monitors) {
1931 ofmonitor_destroy(struct ofmonitor *m)
1932 OVS_REQUIRES(ofproto_mutex)
1935 minimatch_destroy(&m->match);
1936 hmap_remove(&m->ofconn->monitors, &m->ofconn_node);
1942 ofmonitor_report(struct connmgr *mgr, struct rule *rule,
1943 enum nx_flow_update_event event,
1944 enum ofp_flow_removed_reason reason,
1945 const struct ofconn *abbrev_ofconn, ovs_be32 abbrev_xid)
1946 OVS_REQUIRES(ofproto_mutex)
1948 enum nx_flow_monitor_flags update;
1949 struct ofconn *ofconn;
1954 rule->add_seqno = rule->modify_seqno = monitor_seqno++;
1958 update = NXFMF_DELETE;
1961 case NXFME_MODIFIED:
1962 update = NXFMF_MODIFY;
1963 rule->modify_seqno = monitor_seqno++;
1971 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1972 enum nx_flow_monitor_flags flags = 0;
1973 struct ofmonitor *m;
1975 if (ofconn->monitor_paused) {
1976 /* Only send NXFME_DELETED notifications for flows that were added
1977 * before we paused. */
1978 if (event != NXFME_DELETED
1979 || rule->add_seqno > ofconn->monitor_paused) {
1984 HMAP_FOR_EACH (m, ofconn_node, &ofconn->monitors) {
1985 if (m->flags & update
1986 && (m->table_id == 0xff || m->table_id == rule->table_id)
1987 && ofoperation_has_out_port(rule->pending, m->out_port)
1988 && cls_rule_is_loose_match(&rule->cr, &m->match)) {
1994 if (list_is_empty(&ofconn->updates)) {
1995 ofputil_start_flow_update(&ofconn->updates);
1996 ofconn->sent_abbrev_update = false;
1999 if (ofconn != abbrev_ofconn || ofconn->monitor_paused) {
2000 struct ofputil_flow_update fu;
2004 fu.reason = event == NXFME_DELETED ? reason : 0;
2005 fu.table_id = rule->table_id;
2006 fu.cookie = rule->flow_cookie;
2007 minimatch_expand(&rule->cr.match, &match);
2009 fu.priority = rule->cr.priority;
2011 ovs_mutex_lock(&rule->mutex);
2012 fu.idle_timeout = rule->idle_timeout;
2013 fu.hard_timeout = rule->hard_timeout;
2014 ovs_mutex_unlock(&rule->mutex);
2016 if (flags & NXFMF_ACTIONS) {
2017 fu.ofpacts = rule->actions->ofpacts;
2018 fu.ofpacts_len = rule->actions->ofpacts_len;
2023 ofputil_append_flow_update(&fu, &ofconn->updates);
2024 } else if (!ofconn->sent_abbrev_update) {
2025 struct ofputil_flow_update fu;
2027 fu.event = NXFME_ABBREV;
2028 fu.xid = abbrev_xid;
2029 ofputil_append_flow_update(&fu, &ofconn->updates);
2031 ofconn->sent_abbrev_update = true;
2038 ofmonitor_flush(struct connmgr *mgr)
2039 OVS_REQUIRES(ofproto_mutex)
2041 struct ofconn *ofconn;
2043 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2044 struct ofpbuf *msg, *next;
2046 LIST_FOR_EACH_SAFE (msg, next, list_node, &ofconn->updates) {
2047 unsigned int n_bytes;
2049 list_remove(&msg->list_node);
2050 ofconn_send(ofconn, msg, ofconn->monitor_counter);
2051 n_bytes = rconn_packet_counter_n_bytes(ofconn->monitor_counter);
2052 if (!ofconn->monitor_paused && n_bytes > 128 * 1024) {
2053 struct ofpbuf *pause;
2055 COVERAGE_INC(ofmonitor_pause);
2056 ofconn->monitor_paused = monitor_seqno++;
2057 pause = ofpraw_alloc_xid(OFPRAW_NXT_FLOW_MONITOR_PAUSED,
2058 OFP10_VERSION, htonl(0), 0);
2059 ofconn_send(ofconn, pause, ofconn->monitor_counter);
2066 ofmonitor_resume(struct ofconn *ofconn)
2067 OVS_REQUIRES(ofproto_mutex)
2069 struct rule_collection rules;
2070 struct ofpbuf *resumed;
2071 struct ofmonitor *m;
2074 rule_collection_init(&rules);
2075 HMAP_FOR_EACH (m, ofconn_node, &ofconn->monitors) {
2076 ofmonitor_collect_resume_rules(m, ofconn->monitor_paused, &rules);
2080 ofmonitor_compose_refresh_updates(&rules, &msgs);
2082 resumed = ofpraw_alloc_xid(OFPRAW_NXT_FLOW_MONITOR_RESUMED, OFP10_VERSION,
2084 list_push_back(&msgs, &resumed->list_node);
2085 ofconn_send_replies(ofconn, &msgs);
2087 ofconn->monitor_paused = 0;
2091 ofmonitor_may_resume(const struct ofconn *ofconn)
2092 OVS_REQUIRES(ofproto_mutex)
2094 return (ofconn->monitor_paused != 0
2095 && !rconn_packet_counter_n_packets(ofconn->monitor_counter));
2099 ofmonitor_run(struct connmgr *mgr)
2101 struct ofconn *ofconn;
2103 ovs_mutex_lock(&ofproto_mutex);
2104 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2105 if (ofmonitor_may_resume(ofconn)) {
2106 COVERAGE_INC(ofmonitor_resume);
2107 ofmonitor_resume(ofconn);
2110 ovs_mutex_unlock(&ofproto_mutex);
2114 ofmonitor_wait(struct connmgr *mgr)
2116 struct ofconn *ofconn;
2118 ovs_mutex_lock(&ofproto_mutex);
2119 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2120 if (ofmonitor_may_resume(ofconn)) {
2121 poll_immediate_wake();
2124 ovs_mutex_unlock(&ofproto_mutex);