2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
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"
31 #include "poll-loop.h"
40 VLOG_DEFINE_THIS_MODULE(connmgr);
41 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
43 /* An OpenFlow connection. */
45 struct connmgr *connmgr; /* Connection's manager. */
46 struct list node; /* In struct connmgr's "all_conns" list. */
47 struct rconn *rconn; /* OpenFlow connection. */
48 enum ofconn_type type; /* Type. */
49 enum nx_flow_format flow_format; /* Currently selected flow format. */
50 bool flow_mod_table_id; /* NXT_FLOW_MOD_TABLE_ID enabled? */
52 /* Asynchronous flow table operation support. */
53 struct list opgroups; /* Contains pending "ofopgroups", if any. */
54 struct ofpbuf *blocked; /* Postponed OpenFlow message, if any. */
55 bool retry; /* True if 'blocked' is ready to try again. */
57 /* OFPT_PACKET_IN related data. */
58 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
59 #define N_SCHEDULERS 2
60 struct pinsched *schedulers[N_SCHEDULERS];
61 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
62 int miss_send_len; /* Bytes to send of buffered packets. */
64 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
65 * requests, and the maximum number before we stop reading OpenFlow
67 #define OFCONN_REPLY_MAX 100
68 struct rconn_packet_counter *reply_counter;
70 /* type == OFCONN_PRIMARY only. */
71 enum nx_role role; /* Role. */
72 struct hmap_node hmap_node; /* In struct connmgr's "controllers" map. */
73 enum ofproto_band band; /* In-band or out-of-band? */
76 static struct ofconn *ofconn_create(struct connmgr *, struct rconn *,
78 static void ofconn_destroy(struct ofconn *);
80 static void ofconn_reconfigure(struct ofconn *,
81 const struct ofproto_controller *);
83 static void ofconn_run(struct ofconn *,
84 bool (*handle_openflow)(struct ofconn *,
85 struct ofpbuf *ofp_msg));
86 static void ofconn_wait(struct ofconn *, bool handling_openflow);
88 static const char *ofconn_get_target(const struct ofconn *);
89 static char *ofconn_make_name(const struct connmgr *, const char *target);
91 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
93 static bool ofconn_receives_async_msgs(const struct ofconn *);
95 static void ofconn_send(const struct ofconn *, struct ofpbuf *,
96 struct rconn_packet_counter *);
98 static void do_send_packet_in(struct ofpbuf *, void *ofconn_);
100 /* A listener for incoming OpenFlow "service" connections. */
102 struct hmap_node node; /* In struct connmgr's "services" hmap. */
103 struct pvconn *pvconn; /* OpenFlow connection listener. */
105 /* These are not used by ofservice directly. They are settings for
106 * accepted "struct ofconn"s from the pvconn. */
107 int probe_interval; /* Max idle time before probing, in seconds. */
108 int rate_limit; /* Max packet-in rate in packets per second. */
109 int burst_limit; /* Limit on accumulating packet credits. */
112 static void ofservice_reconfigure(struct ofservice *,
113 const struct ofproto_controller *);
114 static int ofservice_create(struct connmgr *, const char *target);
115 static void ofservice_destroy(struct connmgr *, struct ofservice *);
116 static struct ofservice *ofservice_lookup(struct connmgr *,
119 /* Connection manager for an OpenFlow switch. */
121 struct ofproto *ofproto;
123 char *local_port_name;
125 /* OpenFlow connections. */
126 struct hmap controllers; /* Controller "struct ofconn"s. */
127 struct list all_conns; /* Contains "struct ofconn"s. */
129 /* OpenFlow listeners. */
130 struct hmap services; /* Contains "struct ofservice"s. */
131 struct pvconn **snoops;
135 struct fail_open *fail_open;
136 enum ofproto_fail_mode fail_mode;
138 /* In-band control. */
139 struct in_band *in_band;
140 long long int next_in_band_update;
141 struct sockaddr_in *extra_in_band_remotes;
142 size_t n_extra_remotes;
146 static void update_in_band_remotes(struct connmgr *);
147 static void add_snooper(struct connmgr *, struct vconn *);
149 /* Creates and returns a new connection manager owned by 'ofproto'. 'name' is
150 * a name for the ofproto suitable for using in log messages.
151 * 'local_port_name' is the name of the local port (OFPP_LOCAL) within
154 connmgr_create(struct ofproto *ofproto,
155 const char *name, const char *local_port_name)
159 mgr = xmalloc(sizeof *mgr);
160 mgr->ofproto = ofproto;
161 mgr->name = xstrdup(name);
162 mgr->local_port_name = xstrdup(local_port_name);
164 hmap_init(&mgr->controllers);
165 list_init(&mgr->all_conns);
167 hmap_init(&mgr->services);
171 mgr->fail_open = NULL;
172 mgr->fail_mode = OFPROTO_FAIL_SECURE;
175 mgr->next_in_band_update = LLONG_MAX;
176 mgr->extra_in_band_remotes = NULL;
177 mgr->n_extra_remotes = 0;
178 mgr->in_band_queue = -1;
183 /* Frees 'mgr' and all of its resources. */
185 connmgr_destroy(struct connmgr *mgr)
187 struct ofservice *ofservice, *next_ofservice;
188 struct ofconn *ofconn, *next_ofconn;
195 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &mgr->all_conns) {
196 ofconn_destroy(ofconn);
198 hmap_destroy(&mgr->controllers);
200 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &mgr->services) {
201 ofservice_destroy(mgr, ofservice);
203 hmap_destroy(&mgr->services);
205 for (i = 0; i < mgr->n_snoops; i++) {
206 pvconn_close(mgr->snoops[i]);
210 fail_open_destroy(mgr->fail_open);
211 mgr->fail_open = NULL;
213 in_band_destroy(mgr->in_band);
215 free(mgr->extra_in_band_remotes);
217 free(mgr->local_port_name);
222 /* Does all of the periodic maintenance required by 'mgr'.
224 * If 'handle_openflow' is nonnull, calls 'handle_openflow' for each message
225 * received on an OpenFlow connection, passing along the OpenFlow connection
226 * itself and the message that was sent. If 'handle_openflow' returns true,
227 * the message is considered to be fully processed. If 'handle_openflow'
228 * returns false, the message is considered not to have been processed at all;
229 * it will be stored and re-presented to 'handle_openflow' following the next
230 * call to connmgr_retry(). 'handle_openflow' must not modify or free the
233 * If 'handle_openflow' is NULL, no OpenFlow messages will be processed and
234 * other activities that could affect the flow table (in-band processing,
235 * fail-open processing) are suppressed too. */
237 connmgr_run(struct connmgr *mgr,
238 bool (*handle_openflow)(struct ofconn *, struct ofpbuf *ofp_msg))
240 struct ofconn *ofconn, *next_ofconn;
241 struct ofservice *ofservice;
244 if (handle_openflow && mgr->in_band) {
245 if (time_msec() >= mgr->next_in_band_update) {
246 update_in_band_remotes(mgr);
248 in_band_run(mgr->in_band);
251 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &mgr->all_conns) {
252 ofconn_run(ofconn, handle_openflow);
255 /* Fail-open maintenance. Do this after processing the ofconns since
256 * fail-open checks the status of the controller rconn. */
257 if (handle_openflow && mgr->fail_open) {
258 fail_open_run(mgr->fail_open);
261 HMAP_FOR_EACH (ofservice, node, &mgr->services) {
265 retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn);
270 rconn = rconn_create(ofservice->probe_interval, 0);
271 name = ofconn_make_name(mgr, vconn_get_name(vconn));
272 rconn_connect_unreliably(rconn, vconn, name);
275 ofconn = ofconn_create(mgr, rconn, OFCONN_SERVICE);
276 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
277 ofservice->burst_limit);
278 } else if (retval != EAGAIN) {
279 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
283 for (i = 0; i < mgr->n_snoops; i++) {
287 retval = pvconn_accept(mgr->snoops[i], OFP_VERSION, &vconn);
289 add_snooper(mgr, vconn);
290 } else if (retval != EAGAIN) {
291 VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
296 /* Causes the poll loop to wake up when connmgr_run() needs to run.
298 * If 'handling_openflow' is true, arriving OpenFlow messages and other
299 * activities that affect the flow table will wake up the poll loop. If
300 * 'handling_openflow' is false, they will not. */
302 connmgr_wait(struct connmgr *mgr, bool handling_openflow)
304 struct ofservice *ofservice;
305 struct ofconn *ofconn;
308 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
309 ofconn_wait(ofconn, handling_openflow);
311 if (handling_openflow && mgr->in_band) {
312 poll_timer_wait_until(mgr->next_in_band_update);
313 in_band_wait(mgr->in_band);
315 if (handling_openflow && mgr->fail_open) {
316 fail_open_wait(mgr->fail_open);
318 HMAP_FOR_EACH (ofservice, node, &mgr->services) {
319 pvconn_wait(ofservice->pvconn);
321 for (i = 0; i < mgr->n_snoops; i++) {
322 pvconn_wait(mgr->snoops[i]);
326 /* Returns the ofproto that owns 'ofconn''s connmgr. */
328 ofconn_get_ofproto(const struct ofconn *ofconn)
330 return ofconn->connmgr->ofproto;
333 /* If processing of OpenFlow messages was blocked on any 'mgr' ofconns by
334 * returning false to the 'handle_openflow' callback to connmgr_run(), this
335 * re-enables them. */
337 connmgr_retry(struct connmgr *mgr)
339 struct ofconn *ofconn;
341 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
342 ofconn->retry = true;
346 /* OpenFlow configuration. */
348 static void add_controller(struct connmgr *, const char *target);
349 static struct ofconn *find_controller_by_target(struct connmgr *,
351 static void update_fail_open(struct connmgr *);
352 static int set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
353 const struct sset *);
355 /* Returns true if 'mgr' has any configured primary controllers.
357 * Service controllers do not count, but configured primary controllers do
358 * count whether or not they are currently connected. */
360 connmgr_has_controllers(const struct connmgr *mgr)
362 return !hmap_is_empty(&mgr->controllers);
365 /* Initializes 'info' and populates it with information about each configured
366 * primary controller. The keys in 'info' are the controllers' targets; the
367 * data values are corresponding "struct ofproto_controller_info".
369 * The caller owns 'info' and everything in it and should free it when it is no
372 connmgr_get_controller_info(struct connmgr *mgr, struct shash *info)
374 const struct ofconn *ofconn;
376 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
377 const struct rconn *rconn = ofconn->rconn;
378 const char *target = rconn_get_target(rconn);
380 if (!shash_find(info, target)) {
381 struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo);
382 time_t now = time_now();
383 time_t last_connection = rconn_get_last_connection(rconn);
384 time_t last_disconnect = rconn_get_last_disconnect(rconn);
385 int last_error = rconn_get_last_error(rconn);
387 shash_add(info, target, cinfo);
389 cinfo->is_connected = rconn_is_connected(rconn);
390 cinfo->role = ofconn->role;
395 cinfo->pairs.keys[cinfo->pairs.n] = "last_error";
396 cinfo->pairs.values[cinfo->pairs.n++]
397 = xstrdup(ovs_retval_to_string(last_error));
400 cinfo->pairs.keys[cinfo->pairs.n] = "state";
401 cinfo->pairs.values[cinfo->pairs.n++]
402 = xstrdup(rconn_get_state(rconn));
404 if (last_connection != TIME_MIN) {
405 cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_connect";
406 cinfo->pairs.values[cinfo->pairs.n++]
407 = xasprintf("%ld", (long int) (now - last_connection));
410 if (last_disconnect != TIME_MIN) {
411 cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_disconnect";
412 cinfo->pairs.values[cinfo->pairs.n++]
413 = xasprintf("%ld", (long int) (now - last_disconnect));
419 /* Changes 'mgr''s set of controllers to the 'n_controllers' controllers in
422 connmgr_set_controllers(struct connmgr *mgr,
423 const struct ofproto_controller *controllers,
424 size_t n_controllers)
426 bool had_controllers = connmgr_has_controllers(mgr);
427 struct shash new_controllers;
428 struct ofconn *ofconn, *next_ofconn;
429 struct ofservice *ofservice, *next_ofservice;
432 /* Create newly configured controllers and services.
433 * Create a name to ofproto_controller mapping in 'new_controllers'. */
434 shash_init(&new_controllers);
435 for (i = 0; i < n_controllers; i++) {
436 const struct ofproto_controller *c = &controllers[i];
438 if (!vconn_verify_name(c->target)) {
439 if (!find_controller_by_target(mgr, c->target)) {
440 add_controller(mgr, c->target);
442 } else if (!pvconn_verify_name(c->target)) {
443 if (!ofservice_lookup(mgr, c->target)) {
444 ofservice_create(mgr, c->target);
447 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
448 mgr->name, c->target);
452 shash_add_once(&new_controllers, c->target, &controllers[i]);
455 /* Delete controllers that are no longer configured.
456 * Update configuration of all now-existing controllers. */
457 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &mgr->controllers) {
458 struct ofproto_controller *c;
460 c = shash_find_data(&new_controllers, ofconn_get_target(ofconn));
462 ofconn_destroy(ofconn);
464 ofconn_reconfigure(ofconn, c);
468 /* Delete services that are no longer configured.
469 * Update configuration of all now-existing services. */
470 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &mgr->services) {
471 struct ofproto_controller *c;
473 c = shash_find_data(&new_controllers,
474 pvconn_get_name(ofservice->pvconn));
476 ofservice_destroy(mgr, ofservice);
478 ofservice_reconfigure(ofservice, c);
482 shash_destroy(&new_controllers);
484 update_in_band_remotes(mgr);
485 update_fail_open(mgr);
486 if (had_controllers != connmgr_has_controllers(mgr)) {
487 ofproto_flush_flows(mgr->ofproto);
491 /* Drops the connections between 'mgr' and all of its primary and secondary
492 * controllers, forcing them to reconnect. */
494 connmgr_reconnect(const struct connmgr *mgr)
496 struct ofconn *ofconn;
498 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
499 rconn_reconnect(ofconn->rconn);
503 /* Sets the "snoops" for 'mgr' to the pvconn targets listed in 'snoops'.
505 * A "snoop" is a pvconn to which every OpenFlow message to or from the most
506 * important controller on 'mgr' is mirrored. */
508 connmgr_set_snoops(struct connmgr *mgr, const struct sset *snoops)
510 return set_pvconns(&mgr->snoops, &mgr->n_snoops, snoops);
513 /* Adds each of the snoops currently configured on 'mgr' to 'snoops'. */
515 connmgr_get_snoops(const struct connmgr *mgr, struct sset *snoops)
519 for (i = 0; i < mgr->n_snoops; i++) {
520 sset_add(snoops, pvconn_get_name(mgr->snoops[i]));
524 /* Returns true if 'mgr' has at least one snoop, false if it has none. */
526 connmgr_has_snoops(const struct connmgr *mgr)
528 return mgr->n_snoops > 0;
531 /* Creates a new controller for 'target' in 'mgr'. update_controller() needs
532 * to be called later to finish the new ofconn's configuration. */
534 add_controller(struct connmgr *mgr, const char *target)
536 char *name = ofconn_make_name(mgr, target);
537 struct ofconn *ofconn;
539 ofconn = ofconn_create(mgr, rconn_create(5, 8), OFCONN_PRIMARY);
540 ofconn->pktbuf = pktbuf_create();
541 ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
542 rconn_connect(ofconn->rconn, target, name);
543 hmap_insert(&mgr->controllers, &ofconn->hmap_node, hash_string(target, 0));
548 static struct ofconn *
549 find_controller_by_target(struct connmgr *mgr, const char *target)
551 struct ofconn *ofconn;
553 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
554 hash_string(target, 0), &mgr->controllers) {
555 if (!strcmp(ofconn_get_target(ofconn), target)) {
563 update_in_band_remotes(struct connmgr *mgr)
565 struct sockaddr_in *addrs;
566 size_t max_addrs, n_addrs;
567 struct ofconn *ofconn;
570 /* Allocate enough memory for as many remotes as we could possibly have. */
571 max_addrs = mgr->n_extra_remotes + hmap_count(&mgr->controllers);
572 addrs = xmalloc(max_addrs * sizeof *addrs);
575 /* Add all the remotes. */
576 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
577 struct sockaddr_in *sin = &addrs[n_addrs];
579 if (ofconn->band == OFPROTO_OUT_OF_BAND) {
583 sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn);
584 if (sin->sin_addr.s_addr) {
585 sin->sin_port = rconn_get_remote_port(ofconn->rconn);
589 for (i = 0; i < mgr->n_extra_remotes; i++) {
590 addrs[n_addrs++] = mgr->extra_in_band_remotes[i];
593 /* Create or update or destroy in-band. */
596 in_band_create(mgr->ofproto, mgr->local_port_name, &mgr->in_band);
599 in_band_set_remotes(mgr->in_band, addrs, n_addrs);
601 in_band_set_queue(mgr->in_band, mgr->in_band_queue);
602 mgr->next_in_band_update = time_msec() + 1000;
604 in_band_destroy(mgr->in_band);
613 update_fail_open(struct connmgr *mgr)
615 if (connmgr_has_controllers(mgr)
616 && mgr->fail_mode == OFPROTO_FAIL_STANDALONE) {
617 if (!mgr->fail_open) {
618 mgr->fail_open = fail_open_create(mgr->ofproto, mgr);
621 fail_open_destroy(mgr->fail_open);
622 mgr->fail_open = NULL;
627 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
628 const struct sset *sset)
630 struct pvconn **pvconns = *pvconnsp;
631 size_t n_pvconns = *n_pvconnsp;
636 for (i = 0; i < n_pvconns; i++) {
637 pvconn_close(pvconns[i]);
641 pvconns = xmalloc(sset_count(sset) * sizeof *pvconns);
643 SSET_FOR_EACH (name, sset) {
644 struct pvconn *pvconn;
647 error = pvconn_open(name, &pvconn);
649 pvconns[n_pvconns++] = pvconn;
651 VLOG_ERR("failed to listen on %s: %s", name, strerror(error));
659 *n_pvconnsp = n_pvconns;
664 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
665 * means that 'ofconn' is more interesting for monitoring than a lower return
668 snoop_preference(const struct ofconn *ofconn)
670 switch (ofconn->role) {
678 /* Shouldn't happen. */
683 /* One of 'mgr''s "snoop" pvconns has accepted a new connection on 'vconn'.
684 * Connects this vconn to a controller. */
686 add_snooper(struct connmgr *mgr, struct vconn *vconn)
688 struct ofconn *ofconn, *best;
690 /* Pick a controller for monitoring. */
692 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
693 if (ofconn->type == OFCONN_PRIMARY
694 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
700 rconn_add_monitor(best->rconn, vconn);
702 VLOG_INFO_RL(&rl, "no controller connection to snoop");
707 /* Public ofconn functions. */
709 /* Returns the connection type, either OFCONN_PRIMARY or OFCONN_SERVICE. */
711 ofconn_get_type(const struct ofconn *ofconn)
716 /* Returns the role configured for 'ofconn'.
718 * The default role, if no other role has been set, is NX_ROLE_OTHER. */
720 ofconn_get_role(const struct ofconn *ofconn)
725 /* Changes 'ofconn''s role to 'role'. If 'role' is NX_ROLE_MASTER then any
726 * existing master is demoted to a slave. */
728 ofconn_set_role(struct ofconn *ofconn, enum nx_role role)
730 if (role == NX_ROLE_MASTER) {
731 struct ofconn *other;
733 HMAP_FOR_EACH (other, hmap_node, &ofconn->connmgr->controllers) {
734 if (other->role == NX_ROLE_MASTER) {
735 other->role = NX_ROLE_SLAVE;
742 /* Returns the currently configured flow format for 'ofconn', one of NXFF_*.
744 * The default, if no other format has been set, is NXFF_OPENFLOW10. */
746 ofconn_get_flow_format(struct ofconn *ofconn)
748 return ofconn->flow_format;
751 /* Sets the flow format for 'ofconn' to 'flow_format' (one of NXFF_*). */
753 ofconn_set_flow_format(struct ofconn *ofconn, enum nx_flow_format flow_format)
755 ofconn->flow_format = flow_format;
758 /* Returns true if the NXT_FLOW_MOD_TABLE_ID extension is enabled, false
761 * By default the extension is not enabled. */
763 ofconn_get_flow_mod_table_id(const struct ofconn *ofconn)
765 return ofconn->flow_mod_table_id;
768 /* Enables or disables (according to 'enable') the NXT_FLOW_MOD_TABLE_ID
769 * extension on 'ofconn'. */
771 ofconn_set_flow_mod_table_id(struct ofconn *ofconn, bool enable)
773 ofconn->flow_mod_table_id = enable;
776 /* Returns the default miss send length for 'ofconn'. */
778 ofconn_get_miss_send_len(const struct ofconn *ofconn)
780 return ofconn->miss_send_len;
783 /* Sets the default miss send length for 'ofconn' to 'miss_send_len'. */
785 ofconn_set_miss_send_len(struct ofconn *ofconn, int miss_send_len)
787 ofconn->miss_send_len = miss_send_len;
790 /* Sends 'msg' on 'ofconn', accounting it as a reply. (If there is a
791 * sufficient number of OpenFlow replies in-flight on a single ofconn, then the
792 * connmgr will stop accepting new OpenFlow requests on that ofconn until the
793 * controller has accepted some of the replies.) */
795 ofconn_send_reply(const struct ofconn *ofconn, struct ofpbuf *msg)
797 ofconn_send(ofconn, msg, ofconn->reply_counter);
800 /* Sends each of the messages in list 'replies' on 'ofconn' in order,
801 * accounting them as replies. */
803 ofconn_send_replies(const struct ofconn *ofconn, struct list *replies)
805 struct ofpbuf *reply, *next;
807 LIST_FOR_EACH_SAFE (reply, next, list_node, replies) {
808 list_remove(&reply->list_node);
809 ofconn_send_reply(ofconn, reply);
813 /* Sends 'error', which should be an OpenFlow error created with
814 * e.g. ofp_mkerr(), on 'ofconn', as a reply to 'request'. Only at most the
815 * first 64 bytes of 'request' are used. */
817 ofconn_send_error(const struct ofconn *ofconn,
818 const struct ofp_header *request, int error)
820 struct ofpbuf *msg = ofputil_encode_error_msg(error, request);
822 ofconn_send_reply(ofconn, msg);
826 /* Same as pktbuf_retrieve(), using the pktbuf owned by 'ofconn'. */
828 ofconn_pktbuf_retrieve(struct ofconn *ofconn, uint32_t id,
829 struct ofpbuf **bufferp, uint16_t *in_port)
831 return pktbuf_retrieve(ofconn->pktbuf, id, bufferp, in_port);
834 /* Returns true if 'ofconn' has any pending opgroups. */
836 ofconn_has_pending_opgroups(const struct ofconn *ofconn)
838 return !list_is_empty(&ofconn->opgroups);
841 /* Returns the number of pending opgroups on 'ofconn'. */
843 ofconn_n_pending_opgroups(const struct ofconn *ofconn)
845 return list_size(&ofconn->opgroups);
848 /* Adds 'ofconn_node' to 'ofconn''s list of pending opgroups.
850 * If 'ofconn' is destroyed or its connection drops, then 'ofconn' will remove
851 * 'ofconn_node' from the list and re-initialize it with list_init(). The
852 * client may, therefore, use list_is_empty(ofconn_node) to determine whether
853 * 'ofconn_node' is still associated with an active ofconn.
855 * The client may also remove ofconn_node from the list itself, with
858 ofconn_add_opgroup(struct ofconn *ofconn, struct list *ofconn_node)
860 list_push_back(&ofconn->opgroups, ofconn_node);
863 /* Private ofconn functions. */
866 ofconn_get_target(const struct ofconn *ofconn)
868 return rconn_get_target(ofconn->rconn);
871 static struct ofconn *
872 ofconn_create(struct connmgr *mgr, struct rconn *rconn, enum ofconn_type type)
874 struct ofconn *ofconn = xzalloc(sizeof *ofconn);
875 ofconn->connmgr = mgr;
876 list_push_back(&mgr->all_conns, &ofconn->node);
877 ofconn->rconn = rconn;
879 ofconn->flow_format = NXFF_OPENFLOW10;
880 ofconn->flow_mod_table_id = false;
881 list_init(&ofconn->opgroups);
882 ofconn->role = NX_ROLE_OTHER;
883 ofconn->packet_in_counter = rconn_packet_counter_create ();
884 ofconn->pktbuf = NULL;
885 ofconn->miss_send_len = 0;
886 ofconn->reply_counter = rconn_packet_counter_create ();
890 /* Disassociates 'ofconn' from all of the ofopgroups that it initiated that
891 * have not yet completed. (Those ofopgroups will still run to completion in
892 * the usual way, but any errors that they run into will not be reported on any
895 * Also discards any blocked operation on 'ofconn'. */
897 ofconn_flush(struct ofconn *ofconn)
899 while (!list_is_empty(&ofconn->opgroups)) {
900 list_init(list_pop_front(&ofconn->opgroups));
902 ofpbuf_delete(ofconn->blocked);
903 ofconn->blocked = NULL;
907 ofconn_destroy(struct ofconn *ofconn)
909 ofconn_flush(ofconn);
911 if (ofconn->type == OFCONN_PRIMARY) {
912 hmap_remove(&ofconn->connmgr->controllers, &ofconn->hmap_node);
915 list_remove(&ofconn->node);
916 rconn_destroy(ofconn->rconn);
917 rconn_packet_counter_destroy(ofconn->packet_in_counter);
918 rconn_packet_counter_destroy(ofconn->reply_counter);
919 pktbuf_destroy(ofconn->pktbuf);
923 /* Reconfigures 'ofconn' to match 'c'. 'ofconn' and 'c' must have the same
926 ofconn_reconfigure(struct ofconn *ofconn, const struct ofproto_controller *c)
930 ofconn->band = c->band;
932 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
934 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
935 rconn_set_probe_interval(ofconn->rconn, probe_interval);
937 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
940 /* Returns true if it makes sense for 'ofconn' to receive and process OpenFlow
943 ofconn_may_recv(const struct ofconn *ofconn)
945 int count = rconn_packet_counter_read (ofconn->reply_counter);
946 return (!ofconn->blocked || ofconn->retry) && count < OFCONN_REPLY_MAX;
950 ofconn_run(struct ofconn *ofconn,
951 bool (*handle_openflow)(struct ofconn *, struct ofpbuf *ofp_msg))
953 struct connmgr *mgr = ofconn->connmgr;
956 for (i = 0; i < N_SCHEDULERS; i++) {
957 pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn);
960 rconn_run(ofconn->rconn);
962 if (handle_openflow) {
963 /* Limit the number of iterations to avoid starving other tasks. */
964 for (i = 0; i < 50 && ofconn_may_recv(ofconn); i++) {
965 struct ofpbuf *of_msg;
967 of_msg = (ofconn->blocked
969 : rconn_recv(ofconn->rconn));
973 if (mgr->fail_open) {
974 fail_open_maybe_recover(mgr->fail_open);
977 if (handle_openflow(ofconn, of_msg)) {
978 ofpbuf_delete(of_msg);
979 ofconn->blocked = NULL;
981 ofconn->blocked = of_msg;
982 ofconn->retry = false;
987 if (!rconn_is_alive(ofconn->rconn)) {
988 ofconn_destroy(ofconn);
989 } else if (!rconn_is_connected(ofconn->rconn)) {
990 ofconn_flush(ofconn);
995 ofconn_wait(struct ofconn *ofconn, bool handling_openflow)
999 for (i = 0; i < N_SCHEDULERS; i++) {
1000 pinsched_wait(ofconn->schedulers[i]);
1002 rconn_run_wait(ofconn->rconn);
1003 if (handling_openflow && ofconn_may_recv(ofconn)) {
1004 rconn_recv_wait(ofconn->rconn);
1008 /* Returns true if 'ofconn' should receive asynchronous messages. */
1010 ofconn_receives_async_msgs(const struct ofconn *ofconn)
1012 if (!rconn_is_connected(ofconn->rconn)) {
1014 } else if (ofconn->type == OFCONN_PRIMARY) {
1015 /* Primary controllers always get asynchronous messages unless they
1016 * have configured themselves as "slaves". */
1017 return ofconn->role != NX_ROLE_SLAVE;
1019 /* Service connections don't get asynchronous messages unless they have
1020 * explicitly asked for them by setting a nonzero miss send length. */
1021 return ofconn->miss_send_len > 0;
1025 /* Returns a human-readable name for an OpenFlow connection between 'mgr' and
1026 * 'target', suitable for use in log messages for identifying the connection.
1028 * The name is dynamically allocated. The caller should free it (with free())
1029 * when it is no longer needed. */
1031 ofconn_make_name(const struct connmgr *mgr, const char *target)
1033 return xasprintf("%s<->%s", mgr->name, target);
1037 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1041 for (i = 0; i < N_SCHEDULERS; i++) {
1042 struct pinsched **s = &ofconn->schedulers[i];
1046 *s = pinsched_create(rate, burst);
1048 pinsched_set_limits(*s, rate, burst);
1051 pinsched_destroy(*s);
1058 ofconn_send(const struct ofconn *ofconn, struct ofpbuf *msg,
1059 struct rconn_packet_counter *counter)
1061 update_openflow_length(msg);
1062 if (rconn_send(ofconn->rconn, msg, counter)) {
1067 /* Sending asynchronous messages. */
1069 static void schedule_packet_in(struct ofconn *, struct ofputil_packet_in,
1070 const struct flow *, struct ofpbuf *rw_packet);
1072 /* Sends an OFPT_PORT_STATUS message with 'opp' and 'reason' to appropriate
1073 * controllers managed by 'mgr'. */
1075 connmgr_send_port_status(struct connmgr *mgr, const struct ofp_phy_port *opp,
1078 /* XXX Should limit the number of queued port status change messages. */
1079 struct ofconn *ofconn;
1081 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1082 struct ofp_port_status *ops;
1085 /* Primary controllers, even slaves, should always get port status
1086 updates. Otherwise obey ofconn_receives_async_msgs(). */
1087 if (ofconn->type != OFCONN_PRIMARY
1088 && !ofconn_receives_async_msgs(ofconn)) {
1092 ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b);
1093 ops->reason = reason;
1095 ofconn_send(ofconn, b, NULL);
1099 /* Sends an OFPT_FLOW_REMOVED or NXT_FLOW_REMOVED message based on 'fr' to
1100 * appropriate controllers managed by 'mgr'. */
1102 connmgr_send_flow_removed(struct connmgr *mgr,
1103 const struct ofputil_flow_removed *fr)
1105 struct ofconn *ofconn;
1107 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1110 if (!ofconn_receives_async_msgs(ofconn)) {
1114 /* Account flow expirations as replies to OpenFlow requests. That
1115 * works because preventing OpenFlow requests from being processed also
1116 * prevents new flows from being added (and expiring). (It also
1117 * prevents processing OpenFlow requests that would not add new flows,
1118 * so it is imperfect.) */
1119 msg = ofputil_encode_flow_removed(fr, ofconn->flow_format);
1120 ofconn_send_reply(ofconn, msg);
1124 /* Given 'pin', sends an OFPT_PACKET_IN message to each OpenFlow controller as
1125 * necessary according to their individual configurations.
1127 * 'rw_packet' may be NULL. Otherwise, 'rw_packet' must contain the same data
1128 * as pin->packet. (rw_packet == pin->packet is also valid.) Ownership of
1129 * 'rw_packet' is transferred to this function. */
1131 connmgr_send_packet_in(struct connmgr *mgr,
1132 const struct ofputil_packet_in *pin,
1133 const struct flow *flow, struct ofpbuf *rw_packet)
1135 struct ofconn *ofconn, *prev;
1138 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1139 if (ofconn_receives_async_msgs(ofconn)) {
1141 schedule_packet_in(prev, *pin, flow, NULL);
1147 schedule_packet_in(prev, *pin, flow, rw_packet);
1149 ofpbuf_delete(rw_packet);
1153 /* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */
1155 do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn_)
1157 struct ofconn *ofconn = ofconn_;
1159 rconn_send_with_limit(ofconn->rconn, ofp_packet_in,
1160 ofconn->packet_in_counter, 100);
1163 /* Takes 'pin', whose packet has the flow specified by 'flow', composes an
1164 * OpenFlow packet-in message from it, and passes it to 'ofconn''s packet
1165 * scheduler for sending.
1167 * 'rw_packet' may be NULL. Otherwise, 'rw_packet' must contain the same data
1168 * as pin->packet. (rw_packet == pin->packet is also valid.) Ownership of
1169 * 'rw_packet' is transferred to this function. */
1171 schedule_packet_in(struct ofconn *ofconn, struct ofputil_packet_in pin,
1172 const struct flow *flow, struct ofpbuf *rw_packet)
1174 struct connmgr *mgr = ofconn->connmgr;
1176 /* Get OpenFlow buffer_id. */
1177 if (pin.reason == OFPR_ACTION) {
1178 pin.buffer_id = UINT32_MAX;
1179 } else if (mgr->fail_open && fail_open_is_active(mgr->fail_open)) {
1180 pin.buffer_id = pktbuf_get_null();
1181 } else if (!ofconn->pktbuf) {
1182 pin.buffer_id = UINT32_MAX;
1184 pin.buffer_id = pktbuf_save(ofconn->pktbuf, pin.packet, flow->in_port);
1187 /* Figure out how much of the packet to send. */
1188 if (pin.reason == OFPR_NO_MATCH) {
1189 pin.send_len = pin.packet->size;
1191 /* Caller should have initialized 'send_len' to 'max_len' specified in
1192 * struct ofp_action_output. */
1194 if (pin.buffer_id != UINT32_MAX) {
1195 pin.send_len = MIN(pin.send_len, ofconn->miss_send_len);
1198 /* Make OFPT_PACKET_IN and hand over to packet scheduler. It might
1199 * immediately call into do_send_packet_in() or it might buffer it for a
1200 * while (until a later call to pinsched_run()). */
1201 pinsched_send(ofconn->schedulers[pin.reason == OFPR_NO_MATCH ? 0 : 1],
1202 flow->in_port, ofputil_encode_packet_in(&pin, rw_packet),
1203 do_send_packet_in, ofconn);
1206 /* Fail-open settings. */
1208 /* Returns the failure handling mode (OFPROTO_FAIL_SECURE or
1209 * OFPROTO_FAIL_STANDALONE) for 'mgr'. */
1210 enum ofproto_fail_mode
1211 connmgr_get_fail_mode(const struct connmgr *mgr)
1213 return mgr->fail_mode;
1216 /* Sets the failure handling mode for 'mgr' to 'fail_mode' (either
1217 * OFPROTO_FAIL_SECURE or OFPROTO_FAIL_STANDALONE). */
1219 connmgr_set_fail_mode(struct connmgr *mgr, enum ofproto_fail_mode fail_mode)
1221 if (mgr->fail_mode != fail_mode) {
1222 mgr->fail_mode = fail_mode;
1223 update_fail_open(mgr);
1224 if (!connmgr_has_controllers(mgr)) {
1225 ofproto_flush_flows(mgr->ofproto);
1230 /* Fail-open implementation. */
1232 /* Returns the longest probe interval among the primary controllers configured
1233 * on 'mgr'. Returns 0 if there are no primary controllers. */
1235 connmgr_get_max_probe_interval(const struct connmgr *mgr)
1237 const struct ofconn *ofconn;
1238 int max_probe_interval;
1240 max_probe_interval = 0;
1241 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1242 int probe_interval = rconn_get_probe_interval(ofconn->rconn);
1243 max_probe_interval = MAX(max_probe_interval, probe_interval);
1245 return max_probe_interval;
1248 /* Returns the number of seconds for which all of 'mgr's primary controllers
1249 * have been disconnected. Returns 0 if 'mgr' has no primary controllers. */
1251 connmgr_failure_duration(const struct connmgr *mgr)
1253 const struct ofconn *ofconn;
1254 int min_failure_duration;
1256 if (!connmgr_has_controllers(mgr)) {
1260 min_failure_duration = INT_MAX;
1261 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1262 int failure_duration = rconn_failure_duration(ofconn->rconn);
1263 min_failure_duration = MIN(min_failure_duration, failure_duration);
1265 return min_failure_duration;
1268 /* Returns true if at least one primary controller is connected (regardless of
1269 * whether those controllers are believed to have authenticated and accepted
1270 * this switch), false if none of them are connected. */
1272 connmgr_is_any_controller_connected(const struct connmgr *mgr)
1274 const struct ofconn *ofconn;
1276 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1277 if (rconn_is_connected(ofconn->rconn)) {
1284 /* Returns true if at least one primary controller is believed to have
1285 * authenticated and accepted this switch, false otherwise. */
1287 connmgr_is_any_controller_admitted(const struct connmgr *mgr)
1289 const struct ofconn *ofconn;
1291 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1292 if (rconn_is_admitted(ofconn->rconn)) {
1299 /* Sends 'packet' to each controller connected to 'mgr'. Takes ownership of
1302 connmgr_broadcast(struct connmgr *mgr, struct ofpbuf *packet)
1304 struct ofconn *ofconn, *prev;
1307 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1309 ofconn_send_reply(ofconn, ofpbuf_clone(packet));
1311 if (rconn_is_connected(ofconn->rconn)) {
1316 ofconn_send_reply(prev, packet);
1318 ofpbuf_delete(packet);
1322 /* In-band configuration. */
1324 static bool any_extras_changed(const struct connmgr *,
1325 const struct sockaddr_in *extras, size_t n);
1327 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'mgr''s
1328 * in-band control should guarantee access, in the same way that in-band
1329 * control guarantees access to OpenFlow controllers. */
1331 connmgr_set_extra_in_band_remotes(struct connmgr *mgr,
1332 const struct sockaddr_in *extras, size_t n)
1334 if (!any_extras_changed(mgr, extras, n)) {
1338 free(mgr->extra_in_band_remotes);
1339 mgr->n_extra_remotes = n;
1340 mgr->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
1342 update_in_band_remotes(mgr);
1345 /* Sets the OpenFlow queue used by flows set up by in-band control on
1346 * 'mgr' to 'queue_id'. If 'queue_id' is negative, then in-band control
1347 * flows will use the default queue. */
1349 connmgr_set_in_band_queue(struct connmgr *mgr, int queue_id)
1351 if (queue_id != mgr->in_band_queue) {
1352 mgr->in_band_queue = queue_id;
1353 update_in_band_remotes(mgr);
1358 any_extras_changed(const struct connmgr *mgr,
1359 const struct sockaddr_in *extras, size_t n)
1363 if (n != mgr->n_extra_remotes) {
1367 for (i = 0; i < n; i++) {
1368 const struct sockaddr_in *old = &mgr->extra_in_band_remotes[i];
1369 const struct sockaddr_in *new = &extras[i];
1371 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
1372 old->sin_port != new->sin_port) {
1380 /* In-band implementation. */
1383 connmgr_msg_in_hook(struct connmgr *mgr, const struct flow *flow,
1384 const struct ofpbuf *packet)
1386 return mgr->in_band && in_band_msg_in_hook(mgr->in_band, flow, packet);
1390 connmgr_may_set_up_flow(struct connmgr *mgr, const struct flow *flow,
1391 const struct nlattr *odp_actions,
1394 return !mgr->in_band || in_band_rule_check(flow, odp_actions, actions_len);
1397 /* Fail-open and in-band implementation. */
1399 /* Called by 'ofproto' after all flows have been flushed, to allow fail-open
1400 * and standalone mode to re-create their flows.
1402 * In-band control has more sophisticated code that manages flows itself. */
1404 connmgr_flushed(struct connmgr *mgr)
1406 if (mgr->fail_open) {
1407 fail_open_flushed(mgr->fail_open);
1410 /* If there are no controllers and we're in standalone mode, set up a flow
1411 * that matches every packet and directs them to OFPP_NORMAL (which goes to
1412 * us). Otherwise, the switch is in secure mode and we won't pass any
1413 * traffic until a controller has been defined and it tells us to do so. */
1414 if (!connmgr_has_controllers(mgr)
1415 && mgr->fail_mode == OFPROTO_FAIL_STANDALONE) {
1416 union ofp_action action;
1417 struct cls_rule rule;
1419 memset(&action, 0, sizeof action);
1420 action.type = htons(OFPAT_OUTPUT);
1421 action.output.len = htons(sizeof action);
1422 action.output.port = htons(OFPP_NORMAL);
1423 cls_rule_init_catchall(&rule, 0);
1424 ofproto_add_flow(mgr->ofproto, &rule, &action, 1);
1428 /* Creates a new ofservice for 'target' in 'mgr'. Returns 0 if successful,
1429 * otherwise a positive errno value.
1431 * ofservice_reconfigure() must be called to fully configure the new
1434 ofservice_create(struct connmgr *mgr, const char *target)
1436 struct ofservice *ofservice;
1437 struct pvconn *pvconn;
1440 error = pvconn_open(target, &pvconn);
1445 ofservice = xzalloc(sizeof *ofservice);
1446 hmap_insert(&mgr->services, &ofservice->node, hash_string(target, 0));
1447 ofservice->pvconn = pvconn;
1453 ofservice_destroy(struct connmgr *mgr, struct ofservice *ofservice)
1455 hmap_remove(&mgr->services, &ofservice->node);
1456 pvconn_close(ofservice->pvconn);
1461 ofservice_reconfigure(struct ofservice *ofservice,
1462 const struct ofproto_controller *c)
1464 ofservice->probe_interval = c->probe_interval;
1465 ofservice->rate_limit = c->rate_limit;
1466 ofservice->burst_limit = c->burst_limit;
1469 /* Finds and returns the ofservice within 'mgr' that has the given
1470 * 'target', or a null pointer if none exists. */
1471 static struct ofservice *
1472 ofservice_lookup(struct connmgr *mgr, const char *target)
1474 struct ofservice *ofservice;
1476 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1478 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {