]> git.proxmox.com Git - mirror_ovs.git/blob - ofproto/connmgr.c
ofproto: Honour NXFMF_OWN flag of flow monitors
[mirror_ovs.git] / ofproto / connmgr.c
1 /*
2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
3 *
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:
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
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.
15 */
16
17 #include <config.h>
18
19 #include "connmgr.h"
20
21 #include <errno.h>
22 #include <stdlib.h>
23
24 #include "coverage.h"
25 #include "dynamic-string.h"
26 #include "fail-open.h"
27 #include "in-band.h"
28 #include "odp-util.h"
29 #include "ofp-actions.h"
30 #include "ofp-msgs.h"
31 #include "ofp-util.h"
32 #include "ofpbuf.h"
33 #include "ofproto-provider.h"
34 #include "pinsched.h"
35 #include "poll-loop.h"
36 #include "pktbuf.h"
37 #include "rconn.h"
38 #include "shash.h"
39 #include "simap.h"
40 #include "stream.h"
41 #include "timeval.h"
42 #include "vconn.h"
43 #include "vlog.h"
44
45 #include "bundles.h"
46
47 VLOG_DEFINE_THIS_MODULE(connmgr);
48 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
49
50 /* An OpenFlow connection.
51 *
52 *
53 * Thread-safety
54 * =============
55 *
56 * 'ofproto_mutex' must be held whenever an ofconn is created or destroyed or,
57 * more or less equivalently, whenever an ofconn is added to or removed from a
58 * connmgr. 'ofproto_mutex' doesn't protect the data inside the ofconn, except
59 * as specifically noted below. */
60 struct ofconn {
61 /* Configuration that persists from one connection to the next. */
62
63 struct list node; /* In struct connmgr's "all_conns" list. */
64 struct hmap_node hmap_node; /* In struct connmgr's "controllers" map. */
65
66 struct connmgr *connmgr; /* Connection's manager. */
67 struct rconn *rconn; /* OpenFlow connection. */
68 enum ofconn_type type; /* Type. */
69 enum ofproto_band band; /* In-band or out-of-band? */
70 bool enable_async_msgs; /* Initially enable async messages? */
71
72 /* State that should be cleared from one connection to the next. */
73
74 /* OpenFlow state. */
75 enum ofp12_controller_role role; /* Role. */
76 enum ofputil_protocol protocol; /* Current protocol variant. */
77 enum nx_packet_in_format packet_in_format; /* OFPT_PACKET_IN format. */
78
79 /* OFPT_PACKET_IN related data. */
80 struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */
81 #define N_SCHEDULERS 2
82 struct pinsched *schedulers[N_SCHEDULERS];
83 struct pktbuf *pktbuf; /* OpenFlow packet buffers. */
84 int miss_send_len; /* Bytes to send of buffered packets. */
85 uint16_t controller_id; /* Connection controller ID. */
86
87 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
88 * requests, and the maximum number before we stop reading OpenFlow
89 * requests. */
90 #define OFCONN_REPLY_MAX 100
91 struct rconn_packet_counter *reply_counter;
92
93 /* Asynchronous message configuration in each possible roles.
94 *
95 * A 1-bit enables sending an asynchronous message for one possible reason
96 * that the message might be generated, a 0-bit disables it. */
97 uint32_t master_async_config[OAM_N_TYPES]; /* master, other */
98 uint32_t slave_async_config[OAM_N_TYPES]; /* slave */
99
100 /* Flow table operation logging. */
101 int n_add, n_delete, n_modify; /* Number of unreported ops of each kind. */
102 long long int first_op, last_op; /* Range of times for unreported ops. */
103 long long int next_op_report; /* Time to report ops, or LLONG_MAX. */
104 long long int op_backoff; /* Earliest time to report ops again. */
105
106 /* Flow monitors (e.g. NXST_FLOW_MONITOR). */
107
108 /* Configuration. Contains "struct ofmonitor"s. */
109 struct hmap monitors OVS_GUARDED_BY(ofproto_mutex);
110
111 /* Flow control.
112 *
113 * When too many flow monitor notifications back up in the transmit buffer,
114 * we pause the transmission of further notifications. These members track
115 * the flow control state.
116 *
117 * When notifications are flowing, 'monitor_paused' is 0. When
118 * notifications are paused, 'monitor_paused' is the value of
119 * 'monitor_seqno' at the point we paused.
120 *
121 * 'monitor_counter' counts the OpenFlow messages and bytes currently in
122 * flight. This value growing too large triggers pausing. */
123 uint64_t monitor_paused OVS_GUARDED_BY(ofproto_mutex);
124 struct rconn_packet_counter *monitor_counter OVS_GUARDED_BY(ofproto_mutex);
125
126 /* State of monitors for a single ongoing flow_mod.
127 *
128 * 'updates' is a list of "struct ofpbuf"s that contain
129 * NXST_FLOW_MONITOR_REPLY messages representing the changes made by the
130 * current flow_mod.
131 *
132 * When 'updates' is nonempty, 'sent_abbrev_update' is true if 'updates'
133 * contains an update event of type NXFME_ABBREV and false otherwise.. */
134 struct list updates OVS_GUARDED_BY(ofproto_mutex);
135 bool sent_abbrev_update OVS_GUARDED_BY(ofproto_mutex);
136
137 /* Active bundles. Contains "struct ofp_bundle"s. */
138 struct hmap bundles;
139 };
140
141 static struct ofconn *ofconn_create(struct connmgr *, struct rconn *,
142 enum ofconn_type, bool enable_async_msgs)
143 OVS_REQUIRES(ofproto_mutex);
144 static void ofconn_destroy(struct ofconn *) OVS_REQUIRES(ofproto_mutex);
145 static void ofconn_flush(struct ofconn *) OVS_REQUIRES(ofproto_mutex);
146
147 static void ofconn_reconfigure(struct ofconn *,
148 const struct ofproto_controller *);
149
150 static void ofconn_run(struct ofconn *,
151 void (*handle_openflow)(struct ofconn *,
152 const struct ofpbuf *ofp_msg));
153 static void ofconn_wait(struct ofconn *);
154
155 static void ofconn_log_flow_mods(struct ofconn *);
156
157 static const char *ofconn_get_target(const struct ofconn *);
158 static char *ofconn_make_name(const struct connmgr *, const char *target);
159
160 static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst);
161
162 static void ofconn_send(const struct ofconn *, struct ofpbuf *,
163 struct rconn_packet_counter *);
164
165 static void do_send_packet_ins(struct ofconn *, struct list *txq);
166
167 /* A listener for incoming OpenFlow "service" connections. */
168 struct ofservice {
169 struct hmap_node node; /* In struct connmgr's "services" hmap. */
170 struct pvconn *pvconn; /* OpenFlow connection listener. */
171
172 /* These are not used by ofservice directly. They are settings for
173 * accepted "struct ofconn"s from the pvconn. */
174 int probe_interval; /* Max idle time before probing, in seconds. */
175 int rate_limit; /* Max packet-in rate in packets per second. */
176 int burst_limit; /* Limit on accumulating packet credits. */
177 bool enable_async_msgs; /* Initially enable async messages? */
178 uint8_t dscp; /* DSCP Value for controller connection */
179 uint32_t allowed_versions; /* OpenFlow protocol versions that may
180 * be negotiated for a session. */
181 };
182
183 static void ofservice_reconfigure(struct ofservice *,
184 const struct ofproto_controller *);
185 static int ofservice_create(struct connmgr *mgr, const char *target,
186 uint32_t allowed_versions, uint8_t dscp);
187 static void ofservice_destroy(struct connmgr *, struct ofservice *);
188 static struct ofservice *ofservice_lookup(struct connmgr *,
189 const char *target);
190
191 /* Connection manager for an OpenFlow switch. */
192 struct connmgr {
193 struct ofproto *ofproto;
194 char *name;
195 char *local_port_name;
196
197 /* OpenFlow connections. */
198 struct hmap controllers; /* Controller "struct ofconn"s. */
199 struct list all_conns; /* Contains "struct ofconn"s. */
200 uint64_t master_election_id; /* monotonically increasing sequence number
201 * for master election */
202 bool master_election_id_defined;
203
204 /* OpenFlow listeners. */
205 struct hmap services; /* Contains "struct ofservice"s. */
206 struct pvconn **snoops;
207 size_t n_snoops;
208
209 /* Fail open. */
210 struct fail_open *fail_open;
211 enum ofproto_fail_mode fail_mode;
212
213 /* In-band control. */
214 struct in_band *in_band;
215 struct sockaddr_in *extra_in_band_remotes;
216 size_t n_extra_remotes;
217 int in_band_queue;
218 };
219
220 static void update_in_band_remotes(struct connmgr *);
221 static void add_snooper(struct connmgr *, struct vconn *);
222 static void ofmonitor_run(struct connmgr *);
223 static void ofmonitor_wait(struct connmgr *);
224
225 /* Creates and returns a new connection manager owned by 'ofproto'. 'name' is
226 * a name for the ofproto suitable for using in log messages.
227 * 'local_port_name' is the name of the local port (OFPP_LOCAL) within
228 * 'ofproto'. */
229 struct connmgr *
230 connmgr_create(struct ofproto *ofproto,
231 const char *name, const char *local_port_name)
232 {
233 struct connmgr *mgr;
234
235 mgr = xmalloc(sizeof *mgr);
236 mgr->ofproto = ofproto;
237 mgr->name = xstrdup(name);
238 mgr->local_port_name = xstrdup(local_port_name);
239
240 hmap_init(&mgr->controllers);
241 list_init(&mgr->all_conns);
242 mgr->master_election_id = 0;
243 mgr->master_election_id_defined = false;
244
245 hmap_init(&mgr->services);
246 mgr->snoops = NULL;
247 mgr->n_snoops = 0;
248
249 mgr->fail_open = NULL;
250 mgr->fail_mode = OFPROTO_FAIL_SECURE;
251
252 mgr->in_band = NULL;
253 mgr->extra_in_band_remotes = NULL;
254 mgr->n_extra_remotes = 0;
255 mgr->in_band_queue = -1;
256
257 return mgr;
258 }
259
260 /* Frees 'mgr' and all of its resources. */
261 void
262 connmgr_destroy(struct connmgr *mgr)
263 {
264 struct ofservice *ofservice, *next_ofservice;
265 struct ofconn *ofconn, *next_ofconn;
266 size_t i;
267
268 if (!mgr) {
269 return;
270 }
271
272 ovs_mutex_lock(&ofproto_mutex);
273 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &mgr->all_conns) {
274 ofconn_destroy(ofconn);
275 }
276 ovs_mutex_unlock(&ofproto_mutex);
277
278 hmap_destroy(&mgr->controllers);
279
280 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &mgr->services) {
281 ofservice_destroy(mgr, ofservice);
282 }
283 hmap_destroy(&mgr->services);
284
285 for (i = 0; i < mgr->n_snoops; i++) {
286 pvconn_close(mgr->snoops[i]);
287 }
288 free(mgr->snoops);
289
290 fail_open_destroy(mgr->fail_open);
291 mgr->fail_open = NULL;
292
293 in_band_destroy(mgr->in_band);
294 mgr->in_band = NULL;
295 free(mgr->extra_in_band_remotes);
296 free(mgr->name);
297 free(mgr->local_port_name);
298
299 free(mgr);
300 }
301
302 /* Does all of the periodic maintenance required by 'mgr'. Calls
303 * 'handle_openflow' for each message received on an OpenFlow connection,
304 * passing along the OpenFlow connection itself and the message that was sent.
305 * 'handle_openflow' must not modify or free the message. */
306 void
307 connmgr_run(struct connmgr *mgr,
308 void (*handle_openflow)(struct ofconn *,
309 const struct ofpbuf *ofp_msg))
310 OVS_EXCLUDED(ofproto_mutex)
311 {
312 struct ofconn *ofconn, *next_ofconn;
313 struct ofservice *ofservice;
314 size_t i;
315
316 if (mgr->in_band) {
317 if (!in_band_run(mgr->in_band)) {
318 in_band_destroy(mgr->in_band);
319 mgr->in_band = NULL;
320 }
321 }
322
323 LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &mgr->all_conns) {
324 ofconn_run(ofconn, handle_openflow);
325 }
326 ofmonitor_run(mgr);
327
328 /* Fail-open maintenance. Do this after processing the ofconns since
329 * fail-open checks the status of the controller rconn. */
330 if (mgr->fail_open) {
331 fail_open_run(mgr->fail_open);
332 }
333
334 HMAP_FOR_EACH (ofservice, node, &mgr->services) {
335 struct vconn *vconn;
336 int retval;
337
338 retval = pvconn_accept(ofservice->pvconn, &vconn);
339 if (!retval) {
340 struct rconn *rconn;
341 char *name;
342
343 /* Passing default value for creation of the rconn */
344 rconn = rconn_create(ofservice->probe_interval, 0, ofservice->dscp,
345 vconn_get_allowed_versions(vconn));
346 name = ofconn_make_name(mgr, vconn_get_name(vconn));
347 rconn_connect_unreliably(rconn, vconn, name);
348 free(name);
349
350 ovs_mutex_lock(&ofproto_mutex);
351 ofconn = ofconn_create(mgr, rconn, OFCONN_SERVICE,
352 ofservice->enable_async_msgs);
353 ovs_mutex_unlock(&ofproto_mutex);
354
355 ofconn_set_rate_limit(ofconn, ofservice->rate_limit,
356 ofservice->burst_limit);
357 } else if (retval != EAGAIN) {
358 VLOG_WARN_RL(&rl, "accept failed (%s)", ovs_strerror(retval));
359 }
360 }
361
362 for (i = 0; i < mgr->n_snoops; i++) {
363 struct vconn *vconn;
364 int retval;
365
366 retval = pvconn_accept(mgr->snoops[i], &vconn);
367 if (!retval) {
368 add_snooper(mgr, vconn);
369 } else if (retval != EAGAIN) {
370 VLOG_WARN_RL(&rl, "accept failed (%s)", ovs_strerror(retval));
371 }
372 }
373 }
374
375 /* Causes the poll loop to wake up when connmgr_run() needs to run. */
376 void
377 connmgr_wait(struct connmgr *mgr)
378 {
379 struct ofservice *ofservice;
380 struct ofconn *ofconn;
381 size_t i;
382
383 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
384 ofconn_wait(ofconn);
385 }
386 ofmonitor_wait(mgr);
387 if (mgr->in_band) {
388 in_band_wait(mgr->in_band);
389 }
390 if (mgr->fail_open) {
391 fail_open_wait(mgr->fail_open);
392 }
393 HMAP_FOR_EACH (ofservice, node, &mgr->services) {
394 pvconn_wait(ofservice->pvconn);
395 }
396 for (i = 0; i < mgr->n_snoops; i++) {
397 pvconn_wait(mgr->snoops[i]);
398 }
399 }
400
401 /* Adds some memory usage statistics for 'mgr' into 'usage', for use with
402 * memory_report(). */
403 void
404 connmgr_get_memory_usage(const struct connmgr *mgr, struct simap *usage)
405 {
406 const struct ofconn *ofconn;
407 unsigned int packets = 0;
408 unsigned int ofconns = 0;
409
410 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
411 int i;
412
413 ofconns++;
414
415 packets += rconn_count_txqlen(ofconn->rconn);
416 for (i = 0; i < N_SCHEDULERS; i++) {
417 packets += pinsched_count_txqlen(ofconn->schedulers[i]);
418 }
419 packets += pktbuf_count_packets(ofconn->pktbuf);
420 }
421 simap_increase(usage, "ofconns", ofconns);
422 simap_increase(usage, "packets", packets);
423 }
424
425 /* Returns the ofproto that owns 'ofconn''s connmgr. */
426 struct ofproto *
427 ofconn_get_ofproto(const struct ofconn *ofconn)
428 {
429 return ofconn->connmgr->ofproto;
430 }
431 \f
432 /* OpenFlow configuration. */
433
434 static void add_controller(struct connmgr *, const char *target, uint8_t dscp,
435 uint32_t allowed_versions)
436 OVS_REQUIRES(ofproto_mutex);
437 static struct ofconn *find_controller_by_target(struct connmgr *,
438 const char *target);
439 static void update_fail_open(struct connmgr *) OVS_EXCLUDED(ofproto_mutex);
440 static int set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
441 const struct sset *);
442
443 /* Returns true if 'mgr' has any configured primary controllers.
444 *
445 * Service controllers do not count, but configured primary controllers do
446 * count whether or not they are currently connected. */
447 bool
448 connmgr_has_controllers(const struct connmgr *mgr)
449 {
450 return !hmap_is_empty(&mgr->controllers);
451 }
452
453 /* Initializes 'info' and populates it with information about each configured
454 * primary controller. The keys in 'info' are the controllers' targets; the
455 * data values are corresponding "struct ofproto_controller_info".
456 *
457 * The caller owns 'info' and everything in it and should free it when it is no
458 * longer needed. */
459 void
460 connmgr_get_controller_info(struct connmgr *mgr, struct shash *info)
461 {
462 const struct ofconn *ofconn;
463
464 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
465 const struct rconn *rconn = ofconn->rconn;
466 const char *target = rconn_get_target(rconn);
467
468 if (!shash_find(info, target)) {
469 struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo);
470 time_t now = time_now();
471 time_t last_connection = rconn_get_last_connection(rconn);
472 time_t last_disconnect = rconn_get_last_disconnect(rconn);
473 int last_error = rconn_get_last_error(rconn);
474
475 shash_add(info, target, cinfo);
476
477 cinfo->is_connected = rconn_is_connected(rconn);
478 cinfo->role = ofconn->role;
479
480 cinfo->pairs.n = 0;
481
482 if (last_error) {
483 cinfo->pairs.keys[cinfo->pairs.n] = "last_error";
484 cinfo->pairs.values[cinfo->pairs.n++]
485 = xstrdup(ovs_retval_to_string(last_error));
486 }
487
488 cinfo->pairs.keys[cinfo->pairs.n] = "state";
489 cinfo->pairs.values[cinfo->pairs.n++]
490 = xstrdup(rconn_get_state(rconn));
491
492 if (last_connection != TIME_MIN) {
493 cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_connect";
494 cinfo->pairs.values[cinfo->pairs.n++]
495 = xasprintf("%ld", (long int) (now - last_connection));
496 }
497
498 if (last_disconnect != TIME_MIN) {
499 cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_disconnect";
500 cinfo->pairs.values[cinfo->pairs.n++]
501 = xasprintf("%ld", (long int) (now - last_disconnect));
502 }
503 }
504 }
505 }
506
507 void
508 connmgr_free_controller_info(struct shash *info)
509 {
510 struct shash_node *node;
511
512 SHASH_FOR_EACH (node, info) {
513 struct ofproto_controller_info *cinfo = node->data;
514 while (cinfo->pairs.n) {
515 free(CONST_CAST(char *, cinfo->pairs.values[--cinfo->pairs.n]));
516 }
517 free(cinfo);
518 }
519 shash_destroy(info);
520 }
521
522 /* Changes 'mgr''s set of controllers to the 'n_controllers' controllers in
523 * 'controllers'. */
524 void
525 connmgr_set_controllers(struct connmgr *mgr,
526 const struct ofproto_controller *controllers,
527 size_t n_controllers, uint32_t allowed_versions)
528 OVS_EXCLUDED(ofproto_mutex)
529 {
530 bool had_controllers = connmgr_has_controllers(mgr);
531 struct shash new_controllers;
532 struct ofconn *ofconn, *next_ofconn;
533 struct ofservice *ofservice, *next_ofservice;
534 size_t i;
535
536 /* Required to add and remove ofconns. This could probably be narrowed to
537 * cover a smaller amount of code, if that yielded some benefit. */
538 ovs_mutex_lock(&ofproto_mutex);
539
540 /* Create newly configured controllers and services.
541 * Create a name to ofproto_controller mapping in 'new_controllers'. */
542 shash_init(&new_controllers);
543 for (i = 0; i < n_controllers; i++) {
544 const struct ofproto_controller *c = &controllers[i];
545
546 if (!vconn_verify_name(c->target)) {
547 bool add = false;
548 ofconn = find_controller_by_target(mgr, c->target);
549 if (!ofconn) {
550 VLOG_INFO("%s: added primary controller \"%s\"",
551 mgr->name, c->target);
552 add = true;
553 } else if (rconn_get_allowed_versions(ofconn->rconn) !=
554 allowed_versions) {
555 VLOG_INFO("%s: re-added primary controller \"%s\"",
556 mgr->name, c->target);
557 add = true;
558 ofconn_destroy(ofconn);
559 }
560 if (add) {
561 add_controller(mgr, c->target, c->dscp, allowed_versions);
562 }
563 } else if (!pvconn_verify_name(c->target)) {
564 bool add = false;
565 ofservice = ofservice_lookup(mgr, c->target);
566 if (!ofservice) {
567 VLOG_INFO("%s: added service controller \"%s\"",
568 mgr->name, c->target);
569 add = true;
570 } else if (ofservice->allowed_versions != allowed_versions) {
571 VLOG_INFO("%s: re-added service controller \"%s\"",
572 mgr->name, c->target);
573 ofservice_destroy(mgr, ofservice);
574 add = true;
575 }
576 if (add) {
577 ofservice_create(mgr, c->target, allowed_versions, c->dscp);
578 }
579 } else {
580 VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"",
581 mgr->name, c->target);
582 continue;
583 }
584
585 shash_add_once(&new_controllers, c->target, &controllers[i]);
586 }
587
588 /* Delete controllers that are no longer configured.
589 * Update configuration of all now-existing controllers. */
590 HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &mgr->controllers) {
591 const char *target = ofconn_get_target(ofconn);
592 struct ofproto_controller *c;
593
594 c = shash_find_data(&new_controllers, target);
595 if (!c) {
596 VLOG_INFO("%s: removed primary controller \"%s\"",
597 mgr->name, target);
598 ofconn_destroy(ofconn);
599 } else {
600 ofconn_reconfigure(ofconn, c);
601 }
602 }
603
604 /* Delete services that are no longer configured.
605 * Update configuration of all now-existing services. */
606 HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &mgr->services) {
607 const char *target = pvconn_get_name(ofservice->pvconn);
608 struct ofproto_controller *c;
609
610 c = shash_find_data(&new_controllers, target);
611 if (!c) {
612 VLOG_INFO("%s: removed service controller \"%s\"",
613 mgr->name, target);
614 ofservice_destroy(mgr, ofservice);
615 } else {
616 ofservice_reconfigure(ofservice, c);
617 }
618 }
619
620 shash_destroy(&new_controllers);
621
622 ovs_mutex_unlock(&ofproto_mutex);
623
624 update_in_band_remotes(mgr);
625 update_fail_open(mgr);
626 if (had_controllers != connmgr_has_controllers(mgr)) {
627 ofproto_flush_flows(mgr->ofproto);
628 }
629 }
630
631 /* Drops the connections between 'mgr' and all of its primary and secondary
632 * controllers, forcing them to reconnect. */
633 void
634 connmgr_reconnect(const struct connmgr *mgr)
635 {
636 struct ofconn *ofconn;
637
638 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
639 rconn_reconnect(ofconn->rconn);
640 }
641 }
642
643 /* Sets the "snoops" for 'mgr' to the pvconn targets listed in 'snoops'.
644 *
645 * A "snoop" is a pvconn to which every OpenFlow message to or from the most
646 * important controller on 'mgr' is mirrored. */
647 int
648 connmgr_set_snoops(struct connmgr *mgr, const struct sset *snoops)
649 {
650 return set_pvconns(&mgr->snoops, &mgr->n_snoops, snoops);
651 }
652
653 /* Adds each of the snoops currently configured on 'mgr' to 'snoops'. */
654 void
655 connmgr_get_snoops(const struct connmgr *mgr, struct sset *snoops)
656 {
657 size_t i;
658
659 for (i = 0; i < mgr->n_snoops; i++) {
660 sset_add(snoops, pvconn_get_name(mgr->snoops[i]));
661 }
662 }
663
664 /* Returns true if 'mgr' has at least one snoop, false if it has none. */
665 bool
666 connmgr_has_snoops(const struct connmgr *mgr)
667 {
668 return mgr->n_snoops > 0;
669 }
670
671 /* Creates a new controller for 'target' in 'mgr'. update_controller() needs
672 * to be called later to finish the new ofconn's configuration. */
673 static void
674 add_controller(struct connmgr *mgr, const char *target, uint8_t dscp,
675 uint32_t allowed_versions)
676 OVS_REQUIRES(ofproto_mutex)
677 {
678 char *name = ofconn_make_name(mgr, target);
679 struct ofconn *ofconn;
680
681 ofconn = ofconn_create(mgr, rconn_create(5, 8, dscp, allowed_versions),
682 OFCONN_PRIMARY, true);
683 ofconn->pktbuf = pktbuf_create();
684 rconn_connect(ofconn->rconn, target, name);
685 hmap_insert(&mgr->controllers, &ofconn->hmap_node, hash_string(target, 0));
686
687 free(name);
688 }
689
690 static struct ofconn *
691 find_controller_by_target(struct connmgr *mgr, const char *target)
692 {
693 struct ofconn *ofconn;
694
695 HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node,
696 hash_string(target, 0), &mgr->controllers) {
697 if (!strcmp(ofconn_get_target(ofconn), target)) {
698 return ofconn;
699 }
700 }
701 return NULL;
702 }
703
704 static void
705 update_in_band_remotes(struct connmgr *mgr)
706 {
707 struct sockaddr_in *addrs;
708 size_t max_addrs, n_addrs;
709 struct ofconn *ofconn;
710 size_t i;
711
712 /* Allocate enough memory for as many remotes as we could possibly have. */
713 max_addrs = mgr->n_extra_remotes + hmap_count(&mgr->controllers);
714 addrs = xmalloc(max_addrs * sizeof *addrs);
715 n_addrs = 0;
716
717 /* Add all the remotes. */
718 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
719 const char *target = rconn_get_target(ofconn->rconn);
720 struct sockaddr_storage ss;
721
722 if (ofconn->band == OFPROTO_IN_BAND
723 && stream_parse_target_with_default_port(target, OFP_OLD_PORT, &ss)
724 && ss.ss_family == AF_INET) {
725 addrs[n_addrs++] = *(struct sockaddr_in *) &ss;
726 }
727 }
728 for (i = 0; i < mgr->n_extra_remotes; i++) {
729 addrs[n_addrs++] = mgr->extra_in_band_remotes[i];
730 }
731
732 /* Create or update or destroy in-band. */
733 if (n_addrs) {
734 if (!mgr->in_band) {
735 in_band_create(mgr->ofproto, mgr->local_port_name, &mgr->in_band);
736 }
737 in_band_set_queue(mgr->in_band, mgr->in_band_queue);
738 } else {
739 /* in_band_run() needs a chance to delete any existing in-band flows.
740 * We will destroy mgr->in_band after it's done with that. */
741 }
742 if (mgr->in_band) {
743 in_band_set_remotes(mgr->in_band, addrs, n_addrs);
744 }
745
746 /* Clean up. */
747 free(addrs);
748 }
749
750 static void
751 update_fail_open(struct connmgr *mgr)
752 OVS_EXCLUDED(ofproto_mutex)
753 {
754 if (connmgr_has_controllers(mgr)
755 && mgr->fail_mode == OFPROTO_FAIL_STANDALONE) {
756 if (!mgr->fail_open) {
757 mgr->fail_open = fail_open_create(mgr->ofproto, mgr);
758 }
759 } else {
760 fail_open_destroy(mgr->fail_open);
761 mgr->fail_open = NULL;
762 }
763 }
764
765 static int
766 set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp,
767 const struct sset *sset)
768 {
769 struct pvconn **pvconns = *pvconnsp;
770 size_t n_pvconns = *n_pvconnsp;
771 const char *name;
772 int retval = 0;
773 size_t i;
774
775 for (i = 0; i < n_pvconns; i++) {
776 pvconn_close(pvconns[i]);
777 }
778 free(pvconns);
779
780 pvconns = xmalloc(sset_count(sset) * sizeof *pvconns);
781 n_pvconns = 0;
782 SSET_FOR_EACH (name, sset) {
783 struct pvconn *pvconn;
784 int error;
785 error = pvconn_open(name, 0, 0, &pvconn);
786 if (!error) {
787 pvconns[n_pvconns++] = pvconn;
788 } else {
789 VLOG_ERR("failed to listen on %s: %s", name, ovs_strerror(error));
790 if (!retval) {
791 retval = error;
792 }
793 }
794 }
795
796 *pvconnsp = pvconns;
797 *n_pvconnsp = n_pvconns;
798
799 return retval;
800 }
801
802 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
803 * means that 'ofconn' is more interesting for monitoring than a lower return
804 * value. */
805 static int
806 snoop_preference(const struct ofconn *ofconn)
807 {
808 switch (ofconn->role) {
809 case OFPCR12_ROLE_MASTER:
810 return 3;
811 case OFPCR12_ROLE_EQUAL:
812 return 2;
813 case OFPCR12_ROLE_SLAVE:
814 return 1;
815 case OFPCR12_ROLE_NOCHANGE:
816 default:
817 /* Shouldn't happen. */
818 return 0;
819 }
820 }
821
822 /* One of 'mgr''s "snoop" pvconns has accepted a new connection on 'vconn'.
823 * Connects this vconn to a controller. */
824 static void
825 add_snooper(struct connmgr *mgr, struct vconn *vconn)
826 {
827 struct ofconn *ofconn, *best;
828
829 /* Pick a controller for monitoring. */
830 best = NULL;
831 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
832 if (ofconn->type == OFCONN_PRIMARY
833 && (!best || snoop_preference(ofconn) > snoop_preference(best))) {
834 best = ofconn;
835 }
836 }
837
838 if (best) {
839 rconn_add_monitor(best->rconn, vconn);
840 } else {
841 VLOG_INFO_RL(&rl, "no controller connection to snoop");
842 vconn_close(vconn);
843 }
844 }
845 \f
846 /* Public ofconn functions. */
847
848 /* Returns the connection type, either OFCONN_PRIMARY or OFCONN_SERVICE. */
849 enum ofconn_type
850 ofconn_get_type(const struct ofconn *ofconn)
851 {
852 return ofconn->type;
853 }
854
855 /* If a master election id is defined, stores it into '*idp' and returns
856 * true. Otherwise, stores UINT64_MAX into '*idp' and returns false. */
857 bool
858 ofconn_get_master_election_id(const struct ofconn *ofconn, uint64_t *idp)
859 {
860 *idp = (ofconn->connmgr->master_election_id_defined
861 ? ofconn->connmgr->master_election_id
862 : UINT64_MAX);
863 return ofconn->connmgr->master_election_id_defined;
864 }
865
866 /* Sets the master election id.
867 *
868 * Returns true if successful, false if the id is stale
869 */
870 bool
871 ofconn_set_master_election_id(struct ofconn *ofconn, uint64_t id)
872 {
873 if (ofconn->connmgr->master_election_id_defined
874 &&
875 /* Unsigned difference interpreted as a two's complement signed
876 * value */
877 (int64_t)(id - ofconn->connmgr->master_election_id) < 0) {
878 return false;
879 }
880 ofconn->connmgr->master_election_id = id;
881 ofconn->connmgr->master_election_id_defined = true;
882
883 return true;
884 }
885
886 /* Returns the role configured for 'ofconn'.
887 *
888 * The default role, if no other role has been set, is OFPCR12_ROLE_EQUAL. */
889 enum ofp12_controller_role
890 ofconn_get_role(const struct ofconn *ofconn)
891 {
892 return ofconn->role;
893 }
894
895 void
896 ofconn_send_role_status(struct ofconn *ofconn, uint32_t role, uint8_t reason)
897 {
898 struct ofputil_role_status status;
899 struct ofpbuf *buf;
900
901 status.reason = reason;
902 status.role = role;
903 ofconn_get_master_election_id(ofconn, &status.generation_id);
904
905 buf = ofputil_encode_role_status(&status, ofconn_get_protocol(ofconn));
906
907 ofconn_send(ofconn, buf, NULL);
908 }
909
910 /* Changes 'ofconn''s role to 'role'. If 'role' is OFPCR12_ROLE_MASTER then
911 * any existing master is demoted to a slave. */
912 void
913 ofconn_set_role(struct ofconn *ofconn, enum ofp12_controller_role role)
914 {
915 if (role != ofconn->role && role == OFPCR12_ROLE_MASTER) {
916 struct ofconn *other;
917
918 HMAP_FOR_EACH (other, hmap_node, &ofconn->connmgr->controllers) {
919 if (other->role == OFPCR12_ROLE_MASTER) {
920 other->role = OFPCR12_ROLE_SLAVE;
921 ofconn_send_role_status(other, OFPCR12_ROLE_SLAVE, OFPCRR_MASTER_REQUEST);
922 }
923 }
924 }
925 ofconn->role = role;
926 }
927
928 void
929 ofconn_set_invalid_ttl_to_controller(struct ofconn *ofconn, bool enable)
930 {
931 uint32_t bit = 1u << OFPR_INVALID_TTL;
932 if (enable) {
933 ofconn->master_async_config[OAM_PACKET_IN] |= bit;
934 } else {
935 ofconn->master_async_config[OAM_PACKET_IN] &= ~bit;
936 }
937 }
938
939 bool
940 ofconn_get_invalid_ttl_to_controller(struct ofconn *ofconn)
941 {
942 uint32_t bit = 1u << OFPR_INVALID_TTL;
943 return (ofconn->master_async_config[OAM_PACKET_IN] & bit) != 0;
944 }
945
946 /* Returns the currently configured protocol for 'ofconn', one of OFPUTIL_P_*.
947 *
948 * Returns OFPUTIL_P_NONE, which is not a valid protocol, if 'ofconn' hasn't
949 * completed version negotiation. This can't happen if at least one OpenFlow
950 * message, other than OFPT_HELLO, has been received on the connection (such as
951 * in ofproto.c's message handling code), since version negotiation is a
952 * prerequisite for starting to receive messages. This means that
953 * OFPUTIL_P_NONE is a special case that most callers need not worry about. */
954 enum ofputil_protocol
955 ofconn_get_protocol(const struct ofconn *ofconn)
956 {
957 if (ofconn->protocol == OFPUTIL_P_NONE &&
958 rconn_is_connected(ofconn->rconn)) {
959 int version = rconn_get_version(ofconn->rconn);
960 if (version > 0) {
961 ofconn_set_protocol(CONST_CAST(struct ofconn *, ofconn),
962 ofputil_protocol_from_ofp_version(version));
963 }
964 }
965
966 return ofconn->protocol;
967 }
968
969 /* Sets the protocol for 'ofconn' to 'protocol' (one of OFPUTIL_P_*).
970 *
971 * (This doesn't actually send anything to accomplish this. Presumably the
972 * caller already did that.) */
973 void
974 ofconn_set_protocol(struct ofconn *ofconn, enum ofputil_protocol protocol)
975 {
976 ofconn->protocol = protocol;
977 }
978
979 /* Returns the currently configured packet in format for 'ofconn', one of
980 * NXPIF_*.
981 *
982 * The default, if no other format has been set, is NXPIF_OPENFLOW10. */
983 enum nx_packet_in_format
984 ofconn_get_packet_in_format(struct ofconn *ofconn)
985 {
986 return ofconn->packet_in_format;
987 }
988
989 /* Sets the packet in format for 'ofconn' to 'packet_in_format' (one of
990 * NXPIF_*). */
991 void
992 ofconn_set_packet_in_format(struct ofconn *ofconn,
993 enum nx_packet_in_format packet_in_format)
994 {
995 ofconn->packet_in_format = packet_in_format;
996 }
997
998 /* Sets the controller connection ID for 'ofconn' to 'controller_id'.
999 *
1000 * The connection controller ID is used for OFPP_CONTROLLER and
1001 * NXAST_CONTROLLER actions. See "struct nx_action_controller" for details. */
1002 void
1003 ofconn_set_controller_id(struct ofconn *ofconn, uint16_t controller_id)
1004 {
1005 ofconn->controller_id = controller_id;
1006 }
1007
1008 /* Returns the default miss send length for 'ofconn'. */
1009 int
1010 ofconn_get_miss_send_len(const struct ofconn *ofconn)
1011 {
1012 return ofconn->miss_send_len;
1013 }
1014
1015 /* Sets the default miss send length for 'ofconn' to 'miss_send_len'. */
1016 void
1017 ofconn_set_miss_send_len(struct ofconn *ofconn, int miss_send_len)
1018 {
1019 ofconn->miss_send_len = miss_send_len;
1020 }
1021
1022 void
1023 ofconn_set_async_config(struct ofconn *ofconn,
1024 const uint32_t master_masks[OAM_N_TYPES],
1025 const uint32_t slave_masks[OAM_N_TYPES])
1026 {
1027 size_t size = sizeof ofconn->master_async_config;
1028 memcpy(ofconn->master_async_config, master_masks, size);
1029 memcpy(ofconn->slave_async_config, slave_masks, size);
1030 }
1031
1032 void
1033 ofconn_get_async_config(struct ofconn *ofconn,
1034 uint32_t *master_masks, uint32_t *slave_masks)
1035 {
1036 size_t size = sizeof ofconn->master_async_config;
1037 memcpy(master_masks, ofconn->master_async_config, size);
1038 memcpy(slave_masks, ofconn->slave_async_config, size);
1039 }
1040
1041 /* Sends 'msg' on 'ofconn', accounting it as a reply. (If there is a
1042 * sufficient number of OpenFlow replies in-flight on a single ofconn, then the
1043 * connmgr will stop accepting new OpenFlow requests on that ofconn until the
1044 * controller has accepted some of the replies.) */
1045 void
1046 ofconn_send_reply(const struct ofconn *ofconn, struct ofpbuf *msg)
1047 {
1048 ofconn_send(ofconn, msg, ofconn->reply_counter);
1049 }
1050
1051 /* Sends each of the messages in list 'replies' on 'ofconn' in order,
1052 * accounting them as replies. */
1053 void
1054 ofconn_send_replies(const struct ofconn *ofconn, struct list *replies)
1055 {
1056 struct ofpbuf *reply, *next;
1057
1058 LIST_FOR_EACH_SAFE (reply, next, list_node, replies) {
1059 list_remove(&reply->list_node);
1060 ofconn_send_reply(ofconn, reply);
1061 }
1062 }
1063
1064 /* Sends 'error' on 'ofconn', as a reply to 'request'. Only at most the
1065 * first 64 bytes of 'request' are used. */
1066 void
1067 ofconn_send_error(const struct ofconn *ofconn,
1068 const struct ofp_header *request, enum ofperr error)
1069 {
1070 static struct vlog_rate_limit err_rl = VLOG_RATE_LIMIT_INIT(10, 10);
1071 struct ofpbuf *reply;
1072
1073 reply = ofperr_encode_reply(error, request);
1074 if (!VLOG_DROP_INFO(&err_rl)) {
1075 const char *type_name;
1076 size_t request_len;
1077 enum ofpraw raw;
1078
1079 request_len = ntohs(request->length);
1080 type_name = (!ofpraw_decode_partial(&raw, request,
1081 MIN(64, request_len))
1082 ? ofpraw_get_name(raw)
1083 : "invalid");
1084
1085 VLOG_INFO("%s: sending %s error reply to %s message",
1086 rconn_get_name(ofconn->rconn), ofperr_to_string(error),
1087 type_name);
1088 }
1089 ofconn_send_reply(ofconn, reply);
1090 }
1091
1092 /* Same as pktbuf_retrieve(), using the pktbuf owned by 'ofconn'. */
1093 enum ofperr
1094 ofconn_pktbuf_retrieve(struct ofconn *ofconn, uint32_t id,
1095 struct ofpbuf **bufferp, ofp_port_t *in_port)
1096 {
1097 return pktbuf_retrieve(ofconn->pktbuf, id, bufferp, in_port);
1098 }
1099
1100 /* Reports that a flow_mod operation of the type specified by 'command' was
1101 * successfully executed by 'ofconn', so that the connmgr can log it. */
1102 void
1103 ofconn_report_flow_mod(struct ofconn *ofconn,
1104 enum ofp_flow_mod_command command)
1105 {
1106 long long int now;
1107
1108 switch (command) {
1109 case OFPFC_ADD:
1110 ofconn->n_add++;
1111 break;
1112
1113 case OFPFC_MODIFY:
1114 case OFPFC_MODIFY_STRICT:
1115 ofconn->n_modify++;
1116 break;
1117
1118 case OFPFC_DELETE:
1119 case OFPFC_DELETE_STRICT:
1120 ofconn->n_delete++;
1121 break;
1122 }
1123
1124 now = time_msec();
1125 if (ofconn->next_op_report == LLONG_MAX) {
1126 ofconn->first_op = now;
1127 ofconn->next_op_report = MAX(now + 10 * 1000, ofconn->op_backoff);
1128 ofconn->op_backoff = ofconn->next_op_report + 60 * 1000;
1129 }
1130 ofconn->last_op = now;
1131 }
1132
1133 struct hmap *
1134 ofconn_get_bundles(struct ofconn *ofconn)
1135 {
1136 return &ofconn->bundles;
1137 }
1138
1139 \f
1140 /* Private ofconn functions. */
1141
1142 static const char *
1143 ofconn_get_target(const struct ofconn *ofconn)
1144 {
1145 return rconn_get_target(ofconn->rconn);
1146 }
1147
1148 static struct ofconn *
1149 ofconn_create(struct connmgr *mgr, struct rconn *rconn, enum ofconn_type type,
1150 bool enable_async_msgs)
1151 {
1152 struct ofconn *ofconn;
1153
1154 ofconn = xzalloc(sizeof *ofconn);
1155 ofconn->connmgr = mgr;
1156 list_push_back(&mgr->all_conns, &ofconn->node);
1157 ofconn->rconn = rconn;
1158 ofconn->type = type;
1159 ofconn->enable_async_msgs = enable_async_msgs;
1160
1161 hmap_init(&ofconn->monitors);
1162 list_init(&ofconn->updates);
1163
1164 hmap_init(&ofconn->bundles);
1165
1166 ofconn_flush(ofconn);
1167
1168 return ofconn;
1169 }
1170
1171 /* Clears all of the state in 'ofconn' that should not persist from one
1172 * connection to the next. */
1173 static void
1174 ofconn_flush(struct ofconn *ofconn)
1175 OVS_REQUIRES(ofproto_mutex)
1176 {
1177 struct ofmonitor *monitor, *next_monitor;
1178 int i;
1179
1180 ofconn_log_flow_mods(ofconn);
1181
1182 ofconn->role = OFPCR12_ROLE_EQUAL;
1183 ofconn_set_protocol(ofconn, OFPUTIL_P_NONE);
1184 ofconn->packet_in_format = NXPIF_OPENFLOW10;
1185
1186 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1187 ofconn->packet_in_counter = rconn_packet_counter_create();
1188 for (i = 0; i < N_SCHEDULERS; i++) {
1189 if (ofconn->schedulers[i]) {
1190 int rate, burst;
1191
1192 pinsched_get_limits(ofconn->schedulers[i], &rate, &burst);
1193 pinsched_destroy(ofconn->schedulers[i]);
1194 ofconn->schedulers[i] = pinsched_create(rate, burst);
1195 }
1196 }
1197 if (ofconn->pktbuf) {
1198 pktbuf_destroy(ofconn->pktbuf);
1199 ofconn->pktbuf = pktbuf_create();
1200 }
1201 ofconn->miss_send_len = (ofconn->type == OFCONN_PRIMARY
1202 ? OFP_DEFAULT_MISS_SEND_LEN
1203 : 0);
1204 ofconn->controller_id = 0;
1205
1206 rconn_packet_counter_destroy(ofconn->reply_counter);
1207 ofconn->reply_counter = rconn_packet_counter_create();
1208
1209 if (ofconn->enable_async_msgs) {
1210 uint32_t *master = ofconn->master_async_config;
1211 uint32_t *slave = ofconn->slave_async_config;
1212
1213 /* "master" and "other" roles get all asynchronous messages by default,
1214 * except that the controller needs to enable nonstandard "packet-in"
1215 * reasons itself. */
1216 master[OAM_PACKET_IN] = (1u << OFPR_NO_MATCH) | (1u << OFPR_ACTION);
1217 master[OAM_PORT_STATUS] = ((1u << OFPPR_ADD)
1218 | (1u << OFPPR_DELETE)
1219 | (1u << OFPPR_MODIFY));
1220 master[OAM_FLOW_REMOVED] = ((1u << OFPRR_IDLE_TIMEOUT)
1221 | (1u << OFPRR_HARD_TIMEOUT)
1222 | (1u << OFPRR_DELETE));
1223
1224 /* "slave" role gets port status updates by default. */
1225 slave[OAM_PACKET_IN] = 0;
1226 slave[OAM_PORT_STATUS] = ((1u << OFPPR_ADD)
1227 | (1u << OFPPR_DELETE)
1228 | (1u << OFPPR_MODIFY));
1229 slave[OAM_FLOW_REMOVED] = 0;
1230 } else {
1231 memset(ofconn->master_async_config, 0,
1232 sizeof ofconn->master_async_config);
1233 memset(ofconn->slave_async_config, 0,
1234 sizeof ofconn->slave_async_config);
1235 }
1236
1237 ofconn->n_add = ofconn->n_delete = ofconn->n_modify = 0;
1238 ofconn->first_op = ofconn->last_op = LLONG_MIN;
1239 ofconn->next_op_report = LLONG_MAX;
1240 ofconn->op_backoff = LLONG_MIN;
1241
1242 HMAP_FOR_EACH_SAFE (monitor, next_monitor, ofconn_node,
1243 &ofconn->monitors) {
1244 ofmonitor_destroy(monitor);
1245 }
1246 rconn_packet_counter_destroy(ofconn->monitor_counter);
1247 ofconn->monitor_counter = rconn_packet_counter_create();
1248 ofpbuf_list_delete(&ofconn->updates); /* ...but it should be empty. */
1249 }
1250
1251 static void
1252 ofconn_destroy(struct ofconn *ofconn)
1253 OVS_REQUIRES(ofproto_mutex)
1254 {
1255 ofconn_flush(ofconn);
1256
1257 if (ofconn->type == OFCONN_PRIMARY) {
1258 hmap_remove(&ofconn->connmgr->controllers, &ofconn->hmap_node);
1259 }
1260
1261 ofp_bundle_remove_all(ofconn);
1262
1263 hmap_destroy(&ofconn->monitors);
1264 list_remove(&ofconn->node);
1265 rconn_destroy(ofconn->rconn);
1266 rconn_packet_counter_destroy(ofconn->packet_in_counter);
1267 rconn_packet_counter_destroy(ofconn->reply_counter);
1268 pktbuf_destroy(ofconn->pktbuf);
1269 rconn_packet_counter_destroy(ofconn->monitor_counter);
1270 free(ofconn);
1271 }
1272
1273 /* Reconfigures 'ofconn' to match 'c'. 'ofconn' and 'c' must have the same
1274 * target. */
1275 static void
1276 ofconn_reconfigure(struct ofconn *ofconn, const struct ofproto_controller *c)
1277 {
1278 int probe_interval;
1279
1280 ofconn->band = c->band;
1281 ofconn->enable_async_msgs = c->enable_async_msgs;
1282
1283 rconn_set_max_backoff(ofconn->rconn, c->max_backoff);
1284
1285 probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0;
1286 rconn_set_probe_interval(ofconn->rconn, probe_interval);
1287
1288 ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit);
1289
1290 /* If dscp value changed reconnect. */
1291 if (c->dscp != rconn_get_dscp(ofconn->rconn)) {
1292 rconn_set_dscp(ofconn->rconn, c->dscp);
1293 rconn_reconnect(ofconn->rconn);
1294 }
1295 }
1296
1297 /* Returns true if it makes sense for 'ofconn' to receive and process OpenFlow
1298 * messages. */
1299 static bool
1300 ofconn_may_recv(const struct ofconn *ofconn)
1301 {
1302 int count = rconn_packet_counter_n_packets(ofconn->reply_counter);
1303 return count < OFCONN_REPLY_MAX;
1304 }
1305
1306 static void
1307 ofconn_run(struct ofconn *ofconn,
1308 void (*handle_openflow)(struct ofconn *,
1309 const struct ofpbuf *ofp_msg))
1310 {
1311 struct connmgr *mgr = ofconn->connmgr;
1312 size_t i;
1313
1314 for (i = 0; i < N_SCHEDULERS; i++) {
1315 struct list txq;
1316
1317 pinsched_run(ofconn->schedulers[i], &txq);
1318 do_send_packet_ins(ofconn, &txq);
1319 }
1320
1321 rconn_run(ofconn->rconn);
1322
1323 /* Limit the number of iterations to avoid starving other tasks. */
1324 for (i = 0; i < 50 && ofconn_may_recv(ofconn); i++) {
1325 struct ofpbuf *of_msg = rconn_recv(ofconn->rconn);
1326 if (!of_msg) {
1327 break;
1328 }
1329
1330 if (mgr->fail_open) {
1331 fail_open_maybe_recover(mgr->fail_open);
1332 }
1333
1334 handle_openflow(ofconn, of_msg);
1335 ofpbuf_delete(of_msg);
1336 }
1337
1338 if (time_msec() >= ofconn->next_op_report) {
1339 ofconn_log_flow_mods(ofconn);
1340 }
1341
1342 ovs_mutex_lock(&ofproto_mutex);
1343 if (!rconn_is_alive(ofconn->rconn)) {
1344 ofconn_destroy(ofconn);
1345 } else if (!rconn_is_connected(ofconn->rconn)) {
1346 ofconn_flush(ofconn);
1347 }
1348 ovs_mutex_unlock(&ofproto_mutex);
1349 }
1350
1351 static void
1352 ofconn_wait(struct ofconn *ofconn)
1353 {
1354 int i;
1355
1356 for (i = 0; i < N_SCHEDULERS; i++) {
1357 pinsched_wait(ofconn->schedulers[i]);
1358 }
1359 rconn_run_wait(ofconn->rconn);
1360 if (ofconn_may_recv(ofconn)) {
1361 rconn_recv_wait(ofconn->rconn);
1362 }
1363 if (ofconn->next_op_report != LLONG_MAX) {
1364 poll_timer_wait_until(ofconn->next_op_report);
1365 }
1366 }
1367
1368 static void
1369 ofconn_log_flow_mods(struct ofconn *ofconn)
1370 {
1371 int n_flow_mods = ofconn->n_add + ofconn->n_delete + ofconn->n_modify;
1372 if (n_flow_mods) {
1373 long long int ago = (time_msec() - ofconn->first_op) / 1000;
1374 long long int interval = (ofconn->last_op - ofconn->first_op) / 1000;
1375 struct ds s;
1376
1377 ds_init(&s);
1378 ds_put_format(&s, "%d flow_mods ", n_flow_mods);
1379 if (interval == ago) {
1380 ds_put_format(&s, "in the last %lld s", ago);
1381 } else if (interval) {
1382 ds_put_format(&s, "in the %lld s starting %lld s ago",
1383 interval, ago);
1384 } else {
1385 ds_put_format(&s, "%lld s ago", ago);
1386 }
1387
1388 ds_put_cstr(&s, " (");
1389 if (ofconn->n_add) {
1390 ds_put_format(&s, "%d adds, ", ofconn->n_add);
1391 }
1392 if (ofconn->n_delete) {
1393 ds_put_format(&s, "%d deletes, ", ofconn->n_delete);
1394 }
1395 if (ofconn->n_modify) {
1396 ds_put_format(&s, "%d modifications, ", ofconn->n_modify);
1397 }
1398 s.length -= 2;
1399 ds_put_char(&s, ')');
1400
1401 VLOG_INFO("%s: %s", rconn_get_name(ofconn->rconn), ds_cstr(&s));
1402 ds_destroy(&s);
1403
1404 ofconn->n_add = ofconn->n_delete = ofconn->n_modify = 0;
1405 }
1406 ofconn->next_op_report = LLONG_MAX;
1407 }
1408
1409 /* Returns true if 'ofconn' should receive asynchronous messages of the given
1410 * OAM_* 'type' and 'reason', which should be a OFPR_* value for OAM_PACKET_IN,
1411 * a OFPPR_* value for OAM_PORT_STATUS, or an OFPRR_* value for
1412 * OAM_FLOW_REMOVED. Returns false if the message should not be sent on
1413 * 'ofconn'. */
1414 static bool
1415 ofconn_receives_async_msg(const struct ofconn *ofconn,
1416 enum ofconn_async_msg_type type,
1417 unsigned int reason)
1418 {
1419 const uint32_t *async_config;
1420
1421 ovs_assert(reason < 32);
1422 ovs_assert((unsigned int) type < OAM_N_TYPES);
1423
1424 if (ofconn_get_protocol(ofconn) == OFPUTIL_P_NONE
1425 || !rconn_is_connected(ofconn->rconn)) {
1426 return false;
1427 }
1428
1429 /* Keep the following code in sync with the documentation in the
1430 * "Asynchronous Messages" section in DESIGN. */
1431
1432 if (ofconn->type == OFCONN_SERVICE && !ofconn->miss_send_len) {
1433 /* Service connections don't get asynchronous messages unless they have
1434 * explicitly asked for them by setting a nonzero miss send length. */
1435 return false;
1436 }
1437
1438 async_config = (ofconn->role == OFPCR12_ROLE_SLAVE
1439 ? ofconn->slave_async_config
1440 : ofconn->master_async_config);
1441 if (!(async_config[type] & (1u << reason))) {
1442 return false;
1443 }
1444
1445 return true;
1446 }
1447
1448 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
1449 * packet rather than to send the packet to the controller.
1450 *
1451 * This function returns false to indicate the packet should be dropped if
1452 * the controller action was the result of the default table-miss behaviour
1453 * and the controller is using OpenFlow1.3+.
1454 *
1455 * Otherwise true is returned to indicate the packet should be forwarded to
1456 * the controller */
1457 static bool
1458 ofconn_wants_packet_in_on_miss(struct ofconn *ofconn,
1459 const struct ofproto_packet_in *pin)
1460 {
1461 if (pin->miss_type == OFPROTO_PACKET_IN_MISS_WITHOUT_FLOW) {
1462 enum ofputil_protocol protocol = ofconn_get_protocol(ofconn);
1463
1464 if (protocol != OFPUTIL_P_NONE
1465 && ofputil_protocol_to_ofp_version(protocol) >= OFP13_VERSION) {
1466 enum ofproto_table_config config;
1467
1468 config = ofproto_table_get_config(ofconn->connmgr->ofproto,
1469 pin->up.table_id);
1470 if (config == OFPROTO_TABLE_MISS_DEFAULT) {
1471 return false;
1472 }
1473 }
1474 }
1475 return true;
1476 }
1477
1478 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
1479 * packet rather than to send the packet to the controller.
1480 *
1481 * This function returns false to indicate that a packet_in message
1482 * for a "table-miss" should be sent to at least one controller.
1483 * That is there is at least one controller with controller_id 0
1484 * which connected using an OpenFlow version earlier than OpenFlow1.3.
1485 *
1486 * False otherwise.
1487 *
1488 * This logic assumes that "table-miss" packet_in messages
1489 * are always sent to controller_id 0. */
1490 bool
1491 connmgr_wants_packet_in_on_miss(struct connmgr *mgr)
1492 {
1493 struct ofconn *ofconn;
1494
1495 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1496 enum ofputil_protocol protocol = ofconn_get_protocol(ofconn);
1497
1498 if (ofconn->controller_id == 0 &&
1499 (protocol == OFPUTIL_P_NONE ||
1500 ofputil_protocol_to_ofp_version(protocol) < OFP13_VERSION)) {
1501 return true;
1502 }
1503 }
1504 return false;
1505 }
1506
1507 /* Returns a human-readable name for an OpenFlow connection between 'mgr' and
1508 * 'target', suitable for use in log messages for identifying the connection.
1509 *
1510 * The name is dynamically allocated. The caller should free it (with free())
1511 * when it is no longer needed. */
1512 static char *
1513 ofconn_make_name(const struct connmgr *mgr, const char *target)
1514 {
1515 return xasprintf("%s<->%s", mgr->name, target);
1516 }
1517
1518 static void
1519 ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst)
1520 {
1521 int i;
1522
1523 for (i = 0; i < N_SCHEDULERS; i++) {
1524 struct pinsched **s = &ofconn->schedulers[i];
1525
1526 if (rate > 0) {
1527 if (!*s) {
1528 *s = pinsched_create(rate, burst);
1529 } else {
1530 pinsched_set_limits(*s, rate, burst);
1531 }
1532 } else {
1533 pinsched_destroy(*s);
1534 *s = NULL;
1535 }
1536 }
1537 }
1538
1539 static void
1540 ofconn_send(const struct ofconn *ofconn, struct ofpbuf *msg,
1541 struct rconn_packet_counter *counter)
1542 {
1543 ofpmsg_update_length(msg);
1544 rconn_send(ofconn->rconn, msg, counter);
1545 }
1546 \f
1547 /* Sending asynchronous messages. */
1548
1549 static void schedule_packet_in(struct ofconn *, struct ofproto_packet_in,
1550 enum ofp_packet_in_reason wire_reason);
1551
1552 /* Sends an OFPT_PORT_STATUS message with 'opp' and 'reason' to appropriate
1553 * controllers managed by 'mgr'. For messages caused by a controller
1554 * OFPT_PORT_MOD, specify 'source' as the controller connection that sent the
1555 * request; otherwise, specify 'source' as NULL. */
1556 void
1557 connmgr_send_port_status(struct connmgr *mgr, struct ofconn *source,
1558 const struct ofputil_phy_port *pp, uint8_t reason)
1559 {
1560 /* XXX Should limit the number of queued port status change messages. */
1561 struct ofputil_port_status ps;
1562 struct ofconn *ofconn;
1563
1564 ps.reason = reason;
1565 ps.desc = *pp;
1566 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1567 if (ofconn_receives_async_msg(ofconn, OAM_PORT_STATUS, reason)) {
1568 struct ofpbuf *msg;
1569
1570 /* Before 1.5, OpenFlow specified that OFPT_PORT_MOD should not
1571 * generate OFPT_PORT_STATUS messages. That requirement was a
1572 * relic of how OpenFlow originally supported a single controller,
1573 * so that one could expect the controller to already know the
1574 * changes it had made.
1575 *
1576 * EXT-338 changes OpenFlow 1.5 OFPT_PORT_MOD to send
1577 * OFPT_PORT_STATUS messages to every controller. This is
1578 * obviously more useful in the multi-controller case. We could
1579 * always implement it that way in OVS, but that would risk
1580 * confusing controllers that are intended for single-controller
1581 * use only. (Imagine a controller that generates an OFPT_PORT_MOD
1582 * in response to any OFPT_PORT_STATUS!)
1583 *
1584 * So this compromises: for OpenFlow 1.4 and earlier, it generates
1585 * OFPT_PORT_STATUS for OFPT_PORT_MOD, but not back to the
1586 * originating controller. In a single-controller environment, in
1587 * particular, this means that it will never generate
1588 * OFPT_PORT_STATUS for OFPT_PORT_MOD at all. */
1589 if (ofconn == source
1590 && rconn_get_version(ofconn->rconn) < OFP15_VERSION) {
1591 continue;
1592 }
1593
1594 msg = ofputil_encode_port_status(&ps, ofconn_get_protocol(ofconn));
1595 ofconn_send(ofconn, msg, NULL);
1596 }
1597 }
1598 }
1599
1600 /* Sends an OFPT_FLOW_REMOVED or NXT_FLOW_REMOVED message based on 'fr' to
1601 * appropriate controllers managed by 'mgr'. */
1602 void
1603 connmgr_send_flow_removed(struct connmgr *mgr,
1604 const struct ofputil_flow_removed *fr)
1605 {
1606 struct ofconn *ofconn;
1607
1608 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1609 if (ofconn_receives_async_msg(ofconn, OAM_FLOW_REMOVED, fr->reason)) {
1610 struct ofpbuf *msg;
1611
1612 /* Account flow expirations as replies to OpenFlow requests. That
1613 * works because preventing OpenFlow requests from being processed
1614 * also prevents new flows from being added (and expiring). (It
1615 * also prevents processing OpenFlow requests that would not add
1616 * new flows, so it is imperfect.) */
1617 msg = ofputil_encode_flow_removed(fr, ofconn_get_protocol(ofconn));
1618 ofconn_send_reply(ofconn, msg);
1619 }
1620 }
1621 }
1622
1623 /* Normally a send-to-controller action uses reason OFPR_ACTION. However, in
1624 * OpenFlow 1.3 and later, packet_ins generated by a send-to-controller action
1625 * in a "table-miss" flow (one with priority 0 and completely wildcarded) are
1626 * sent as OFPR_NO_MATCH. This function returns the reason that should
1627 * actually be sent on 'ofconn' for 'pin'. */
1628 static enum ofp_packet_in_reason
1629 wire_reason(struct ofconn *ofconn, const struct ofproto_packet_in *pin)
1630 {
1631 if (pin->miss_type == OFPROTO_PACKET_IN_MISS_FLOW
1632 && pin->up.reason == OFPR_ACTION) {
1633 enum ofputil_protocol protocol = ofconn_get_protocol(ofconn);
1634
1635 if (protocol != OFPUTIL_P_NONE
1636 && ofputil_protocol_to_ofp_version(protocol) >= OFP13_VERSION) {
1637 return OFPR_NO_MATCH;
1638 }
1639 }
1640 return pin->up.reason;
1641 }
1642
1643 /* Given 'pin', sends an OFPT_PACKET_IN message to each OpenFlow controller as
1644 * necessary according to their individual configurations.
1645 *
1646 * The caller doesn't need to fill in pin->buffer_id or pin->total_len. */
1647 void
1648 connmgr_send_packet_in(struct connmgr *mgr,
1649 const struct ofproto_packet_in *pin)
1650 {
1651 struct ofconn *ofconn;
1652
1653 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
1654 enum ofp_packet_in_reason reason = wire_reason(ofconn, pin);
1655
1656 if (ofconn_wants_packet_in_on_miss(ofconn, pin)
1657 && ofconn_receives_async_msg(ofconn, OAM_PACKET_IN, pin->up.reason)
1658 && ofconn->controller_id == pin->controller_id) {
1659 schedule_packet_in(ofconn, *pin, reason);
1660 }
1661 }
1662 }
1663
1664 static void
1665 do_send_packet_ins(struct ofconn *ofconn, struct list *txq)
1666 {
1667 struct ofpbuf *pin, *next_pin;
1668
1669 LIST_FOR_EACH_SAFE (pin, next_pin, list_node, txq) {
1670 list_remove(&pin->list_node);
1671
1672 if (rconn_send_with_limit(ofconn->rconn, pin,
1673 ofconn->packet_in_counter, 100) == EAGAIN) {
1674 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5);
1675
1676 VLOG_INFO_RL(&rl, "%s: dropping packet-in due to queue overflow",
1677 rconn_get_name(ofconn->rconn));
1678 }
1679 }
1680 }
1681
1682 /* Takes 'pin', composes an OpenFlow packet-in message from it, and passes it
1683 * to 'ofconn''s packet scheduler for sending. */
1684 static void
1685 schedule_packet_in(struct ofconn *ofconn, struct ofproto_packet_in pin,
1686 enum ofp_packet_in_reason wire_reason)
1687 {
1688 struct connmgr *mgr = ofconn->connmgr;
1689 uint16_t controller_max_len;
1690 struct list txq;
1691
1692 pin.up.total_len = pin.up.packet_len;
1693
1694 pin.up.reason = wire_reason;
1695 if (pin.up.reason == OFPR_ACTION) {
1696 controller_max_len = pin.send_len; /* max_len */
1697 } else {
1698 controller_max_len = ofconn->miss_send_len;
1699 }
1700
1701 /* Get OpenFlow buffer_id.
1702 * For OpenFlow 1.2+, OFPCML_NO_BUFFER (== UINT16_MAX) specifies
1703 * unbuffered. This behaviour doesn't violate prior versions, too. */
1704 if (controller_max_len == UINT16_MAX) {
1705 pin.up.buffer_id = UINT32_MAX;
1706 } else if (mgr->fail_open && fail_open_is_active(mgr->fail_open)) {
1707 pin.up.buffer_id = pktbuf_get_null();
1708 } else if (!ofconn->pktbuf) {
1709 pin.up.buffer_id = UINT32_MAX;
1710 } else {
1711 pin.up.buffer_id = pktbuf_save(ofconn->pktbuf,
1712 pin.up.packet, pin.up.packet_len,
1713 pin.up.fmd.in_port);
1714 }
1715
1716 /* Figure out how much of the packet to send.
1717 * If not buffered, send the entire packet. Otherwise, depending on
1718 * the reason of packet-in, send what requested by the controller. */
1719 if (pin.up.buffer_id != UINT32_MAX
1720 && controller_max_len < pin.up.packet_len) {
1721 pin.up.packet_len = controller_max_len;
1722 }
1723
1724 /* Make OFPT_PACKET_IN and hand over to packet scheduler. */
1725 pinsched_send(ofconn->schedulers[pin.up.reason == OFPR_NO_MATCH ? 0 : 1],
1726 pin.up.fmd.in_port,
1727 ofputil_encode_packet_in(&pin.up,
1728 ofconn_get_protocol(ofconn),
1729 ofconn->packet_in_format),
1730 &txq);
1731 do_send_packet_ins(ofconn, &txq);
1732 }
1733 \f
1734 /* Fail-open settings. */
1735
1736 /* Returns the failure handling mode (OFPROTO_FAIL_SECURE or
1737 * OFPROTO_FAIL_STANDALONE) for 'mgr'. */
1738 enum ofproto_fail_mode
1739 connmgr_get_fail_mode(const struct connmgr *mgr)
1740 {
1741 return mgr->fail_mode;
1742 }
1743
1744 /* Sets the failure handling mode for 'mgr' to 'fail_mode' (either
1745 * OFPROTO_FAIL_SECURE or OFPROTO_FAIL_STANDALONE). */
1746 void
1747 connmgr_set_fail_mode(struct connmgr *mgr, enum ofproto_fail_mode fail_mode)
1748 {
1749 if (mgr->fail_mode != fail_mode) {
1750 mgr->fail_mode = fail_mode;
1751 update_fail_open(mgr);
1752 if (!connmgr_has_controllers(mgr)) {
1753 ofproto_flush_flows(mgr->ofproto);
1754 }
1755 }
1756 }
1757 \f
1758 /* Fail-open implementation. */
1759
1760 /* Returns the longest probe interval among the primary controllers configured
1761 * on 'mgr'. Returns 0 if there are no primary controllers. */
1762 int
1763 connmgr_get_max_probe_interval(const struct connmgr *mgr)
1764 {
1765 const struct ofconn *ofconn;
1766 int max_probe_interval;
1767
1768 max_probe_interval = 0;
1769 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1770 int probe_interval = rconn_get_probe_interval(ofconn->rconn);
1771 max_probe_interval = MAX(max_probe_interval, probe_interval);
1772 }
1773 return max_probe_interval;
1774 }
1775
1776 /* Returns the number of seconds for which all of 'mgr's primary controllers
1777 * have been disconnected. Returns 0 if 'mgr' has no primary controllers. */
1778 int
1779 connmgr_failure_duration(const struct connmgr *mgr)
1780 {
1781 const struct ofconn *ofconn;
1782 int min_failure_duration;
1783
1784 if (!connmgr_has_controllers(mgr)) {
1785 return 0;
1786 }
1787
1788 min_failure_duration = INT_MAX;
1789 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1790 int failure_duration = rconn_failure_duration(ofconn->rconn);
1791 min_failure_duration = MIN(min_failure_duration, failure_duration);
1792 }
1793 return min_failure_duration;
1794 }
1795
1796 /* Returns true if at least one primary controller is connected (regardless of
1797 * whether those controllers are believed to have authenticated and accepted
1798 * this switch), false if none of them are connected. */
1799 bool
1800 connmgr_is_any_controller_connected(const struct connmgr *mgr)
1801 {
1802 const struct ofconn *ofconn;
1803
1804 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1805 if (rconn_is_connected(ofconn->rconn)) {
1806 return true;
1807 }
1808 }
1809 return false;
1810 }
1811
1812 /* Returns true if at least one primary controller is believed to have
1813 * authenticated and accepted this switch, false otherwise. */
1814 bool
1815 connmgr_is_any_controller_admitted(const struct connmgr *mgr)
1816 {
1817 const struct ofconn *ofconn;
1818
1819 HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) {
1820 if (rconn_is_admitted(ofconn->rconn)) {
1821 return true;
1822 }
1823 }
1824 return false;
1825 }
1826 \f
1827 /* In-band configuration. */
1828
1829 static bool any_extras_changed(const struct connmgr *,
1830 const struct sockaddr_in *extras, size_t n);
1831
1832 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'mgr''s
1833 * in-band control should guarantee access, in the same way that in-band
1834 * control guarantees access to OpenFlow controllers. */
1835 void
1836 connmgr_set_extra_in_band_remotes(struct connmgr *mgr,
1837 const struct sockaddr_in *extras, size_t n)
1838 {
1839 if (!any_extras_changed(mgr, extras, n)) {
1840 return;
1841 }
1842
1843 free(mgr->extra_in_band_remotes);
1844 mgr->n_extra_remotes = n;
1845 mgr->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras);
1846
1847 update_in_band_remotes(mgr);
1848 }
1849
1850 /* Sets the OpenFlow queue used by flows set up by in-band control on
1851 * 'mgr' to 'queue_id'. If 'queue_id' is negative, then in-band control
1852 * flows will use the default queue. */
1853 void
1854 connmgr_set_in_band_queue(struct connmgr *mgr, int queue_id)
1855 {
1856 if (queue_id != mgr->in_band_queue) {
1857 mgr->in_band_queue = queue_id;
1858 update_in_band_remotes(mgr);
1859 }
1860 }
1861
1862 static bool
1863 any_extras_changed(const struct connmgr *mgr,
1864 const struct sockaddr_in *extras, size_t n)
1865 {
1866 size_t i;
1867
1868 if (n != mgr->n_extra_remotes) {
1869 return true;
1870 }
1871
1872 for (i = 0; i < n; i++) {
1873 const struct sockaddr_in *old = &mgr->extra_in_band_remotes[i];
1874 const struct sockaddr_in *new = &extras[i];
1875
1876 if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
1877 old->sin_port != new->sin_port) {
1878 return true;
1879 }
1880 }
1881
1882 return false;
1883 }
1884 \f
1885 /* In-band implementation. */
1886
1887 bool
1888 connmgr_has_in_band(struct connmgr *mgr)
1889 {
1890 return mgr->in_band != NULL;
1891 }
1892 \f
1893 /* Fail-open and in-band implementation. */
1894
1895 /* Called by 'ofproto' after all flows have been flushed, to allow fail-open
1896 * and standalone mode to re-create their flows.
1897 *
1898 * In-band control has more sophisticated code that manages flows itself. */
1899 void
1900 connmgr_flushed(struct connmgr *mgr)
1901 OVS_EXCLUDED(ofproto_mutex)
1902 {
1903 if (mgr->fail_open) {
1904 fail_open_flushed(mgr->fail_open);
1905 }
1906
1907 /* If there are no controllers and we're in standalone mode, set up a flow
1908 * that matches every packet and directs them to OFPP_NORMAL (which goes to
1909 * us). Otherwise, the switch is in secure mode and we won't pass any
1910 * traffic until a controller has been defined and it tells us to do so. */
1911 if (!connmgr_has_controllers(mgr)
1912 && mgr->fail_mode == OFPROTO_FAIL_STANDALONE) {
1913 struct ofpbuf ofpacts;
1914 struct match match;
1915
1916 ofpbuf_init(&ofpacts, OFPACT_OUTPUT_SIZE);
1917 ofpact_put_OUTPUT(&ofpacts)->port = OFPP_NORMAL;
1918 ofpact_pad(&ofpacts);
1919
1920 match_init_catchall(&match);
1921 ofproto_add_flow(mgr->ofproto, &match, 0, ofpbuf_data(&ofpacts),
1922 ofpbuf_size(&ofpacts));
1923
1924 ofpbuf_uninit(&ofpacts);
1925 }
1926 }
1927 \f
1928 /* Creates a new ofservice for 'target' in 'mgr'. Returns 0 if successful,
1929 * otherwise a positive errno value.
1930 *
1931 * ofservice_reconfigure() must be called to fully configure the new
1932 * ofservice. */
1933 static int
1934 ofservice_create(struct connmgr *mgr, const char *target,
1935 uint32_t allowed_versions, uint8_t dscp)
1936 {
1937 struct ofservice *ofservice;
1938 struct pvconn *pvconn;
1939 int error;
1940
1941 error = pvconn_open(target, allowed_versions, dscp, &pvconn);
1942 if (error) {
1943 return error;
1944 }
1945
1946 ofservice = xzalloc(sizeof *ofservice);
1947 hmap_insert(&mgr->services, &ofservice->node, hash_string(target, 0));
1948 ofservice->pvconn = pvconn;
1949 ofservice->allowed_versions = allowed_versions;
1950
1951 return 0;
1952 }
1953
1954 static void
1955 ofservice_destroy(struct connmgr *mgr, struct ofservice *ofservice)
1956 {
1957 hmap_remove(&mgr->services, &ofservice->node);
1958 pvconn_close(ofservice->pvconn);
1959 free(ofservice);
1960 }
1961
1962 static void
1963 ofservice_reconfigure(struct ofservice *ofservice,
1964 const struct ofproto_controller *c)
1965 {
1966 ofservice->probe_interval = c->probe_interval;
1967 ofservice->rate_limit = c->rate_limit;
1968 ofservice->burst_limit = c->burst_limit;
1969 ofservice->enable_async_msgs = c->enable_async_msgs;
1970 ofservice->dscp = c->dscp;
1971 }
1972
1973 /* Finds and returns the ofservice within 'mgr' that has the given
1974 * 'target', or a null pointer if none exists. */
1975 static struct ofservice *
1976 ofservice_lookup(struct connmgr *mgr, const char *target)
1977 {
1978 struct ofservice *ofservice;
1979
1980 HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0),
1981 &mgr->services) {
1982 if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) {
1983 return ofservice;
1984 }
1985 }
1986 return NULL;
1987 }
1988 \f
1989 /* Flow monitors (NXST_FLOW_MONITOR). */
1990
1991 /* A counter incremented when something significant happens to an OpenFlow
1992 * rule.
1993 *
1994 * - When a rule is added, its 'add_seqno' and 'modify_seqno' are set to
1995 * the current value (which is then incremented).
1996 *
1997 * - When a rule is modified, its 'modify_seqno' is set to the current
1998 * value (which is then incremented).
1999 *
2000 * Thus, by comparing an old value of monitor_seqno against a rule's
2001 * 'add_seqno', one can tell whether the rule was added before or after the old
2002 * value was read, and similarly for 'modify_seqno'.
2003 *
2004 * 32 bits should normally be sufficient (and would be nice, to save space in
2005 * each rule) but then we'd have to have some special cases for wraparound.
2006 *
2007 * We initialize monitor_seqno to 1 to allow 0 to be used as an invalid
2008 * value. */
2009 static uint64_t monitor_seqno = 1;
2010
2011 COVERAGE_DEFINE(ofmonitor_pause);
2012 COVERAGE_DEFINE(ofmonitor_resume);
2013
2014 enum ofperr
2015 ofmonitor_create(const struct ofputil_flow_monitor_request *request,
2016 struct ofconn *ofconn, struct ofmonitor **monitorp)
2017 OVS_REQUIRES(ofproto_mutex)
2018 {
2019 struct ofmonitor *m;
2020
2021 *monitorp = NULL;
2022
2023 m = ofmonitor_lookup(ofconn, request->id);
2024 if (m) {
2025 return OFPERR_OFPMOFC_MONITOR_EXISTS;
2026 }
2027
2028 m = xmalloc(sizeof *m);
2029 m->ofconn = ofconn;
2030 hmap_insert(&ofconn->monitors, &m->ofconn_node, hash_int(request->id, 0));
2031 m->id = request->id;
2032 m->flags = request->flags;
2033 m->out_port = request->out_port;
2034 m->table_id = request->table_id;
2035 minimatch_init(&m->match, &request->match);
2036
2037 *monitorp = m;
2038 return 0;
2039 }
2040
2041 struct ofmonitor *
2042 ofmonitor_lookup(struct ofconn *ofconn, uint32_t id)
2043 OVS_REQUIRES(ofproto_mutex)
2044 {
2045 struct ofmonitor *m;
2046
2047 HMAP_FOR_EACH_IN_BUCKET (m, ofconn_node, hash_int(id, 0),
2048 &ofconn->monitors) {
2049 if (m->id == id) {
2050 return m;
2051 }
2052 }
2053 return NULL;
2054 }
2055
2056 void
2057 ofmonitor_destroy(struct ofmonitor *m)
2058 OVS_REQUIRES(ofproto_mutex)
2059 {
2060 if (m) {
2061 minimatch_destroy(&m->match);
2062 hmap_remove(&m->ofconn->monitors, &m->ofconn_node);
2063 free(m);
2064 }
2065 }
2066
2067 void
2068 ofmonitor_report(struct connmgr *mgr, struct rule *rule,
2069 enum nx_flow_update_event event,
2070 enum ofp_flow_removed_reason reason,
2071 const struct ofconn *abbrev_ofconn, ovs_be32 abbrev_xid)
2072 OVS_REQUIRES(ofproto_mutex)
2073 {
2074 enum nx_flow_monitor_flags update;
2075 struct ofconn *ofconn;
2076
2077 if (rule_is_hidden(rule)) {
2078 return;
2079 }
2080
2081 switch (event) {
2082 case NXFME_ADDED:
2083 update = NXFMF_ADD;
2084 rule->add_seqno = rule->modify_seqno = monitor_seqno++;
2085 break;
2086
2087 case NXFME_DELETED:
2088 update = NXFMF_DELETE;
2089 break;
2090
2091 case NXFME_MODIFIED:
2092 update = NXFMF_MODIFY;
2093 rule->modify_seqno = monitor_seqno++;
2094 break;
2095
2096 default:
2097 case NXFME_ABBREV:
2098 OVS_NOT_REACHED();
2099 }
2100
2101 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2102 enum nx_flow_monitor_flags flags = 0;
2103 struct ofmonitor *m;
2104
2105 if (ofconn->monitor_paused) {
2106 /* Only send NXFME_DELETED notifications for flows that were added
2107 * before we paused. */
2108 if (event != NXFME_DELETED
2109 || rule->add_seqno > ofconn->monitor_paused) {
2110 continue;
2111 }
2112 }
2113
2114 HMAP_FOR_EACH (m, ofconn_node, &ofconn->monitors) {
2115 if (m->flags & update
2116 && (m->table_id == 0xff || m->table_id == rule->table_id)
2117 && cls_rule_is_loose_match(&rule->cr, &m->match)) {
2118 flags |= m->flags;
2119 }
2120 }
2121
2122 if (flags) {
2123 if (list_is_empty(&ofconn->updates)) {
2124 ofputil_start_flow_update(&ofconn->updates);
2125 ofconn->sent_abbrev_update = false;
2126 }
2127
2128 if (flags & NXFMF_OWN || ofconn != abbrev_ofconn
2129 || ofconn->monitor_paused) {
2130 struct ofputil_flow_update fu;
2131 struct match match;
2132
2133 fu.event = event;
2134 fu.reason = event == NXFME_DELETED ? reason : 0;
2135 fu.table_id = rule->table_id;
2136 fu.cookie = rule->flow_cookie;
2137 minimatch_expand(&rule->cr.match, &match);
2138 fu.match = &match;
2139 fu.priority = rule->cr.priority;
2140
2141 ovs_mutex_lock(&rule->mutex);
2142 fu.idle_timeout = rule->idle_timeout;
2143 fu.hard_timeout = rule->hard_timeout;
2144 ovs_mutex_unlock(&rule->mutex);
2145
2146 if (flags & NXFMF_ACTIONS) {
2147 const struct rule_actions *actions = rule_get_actions(rule);
2148 fu.ofpacts = actions->ofpacts;
2149 fu.ofpacts_len = actions->ofpacts_len;
2150 } else {
2151 fu.ofpacts = NULL;
2152 fu.ofpacts_len = 0;
2153 }
2154 ofputil_append_flow_update(&fu, &ofconn->updates);
2155 } else if (!ofconn->sent_abbrev_update) {
2156 struct ofputil_flow_update fu;
2157
2158 fu.event = NXFME_ABBREV;
2159 fu.xid = abbrev_xid;
2160 ofputil_append_flow_update(&fu, &ofconn->updates);
2161
2162 ofconn->sent_abbrev_update = true;
2163 }
2164 }
2165 }
2166 }
2167
2168 void
2169 ofmonitor_flush(struct connmgr *mgr)
2170 OVS_REQUIRES(ofproto_mutex)
2171 {
2172 struct ofconn *ofconn;
2173
2174 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2175 struct ofpbuf *msg, *next;
2176
2177 LIST_FOR_EACH_SAFE (msg, next, list_node, &ofconn->updates) {
2178 unsigned int n_bytes;
2179
2180 list_remove(&msg->list_node);
2181 ofconn_send(ofconn, msg, ofconn->monitor_counter);
2182 n_bytes = rconn_packet_counter_n_bytes(ofconn->monitor_counter);
2183 if (!ofconn->monitor_paused && n_bytes > 128 * 1024) {
2184 struct ofpbuf *pause;
2185
2186 COVERAGE_INC(ofmonitor_pause);
2187 ofconn->monitor_paused = monitor_seqno++;
2188 pause = ofpraw_alloc_xid(OFPRAW_NXT_FLOW_MONITOR_PAUSED,
2189 OFP10_VERSION, htonl(0), 0);
2190 ofconn_send(ofconn, pause, ofconn->monitor_counter);
2191 }
2192 }
2193 }
2194 }
2195
2196 static void
2197 ofmonitor_resume(struct ofconn *ofconn)
2198 OVS_REQUIRES(ofproto_mutex)
2199 {
2200 struct rule_collection rules;
2201 struct ofpbuf *resumed;
2202 struct ofmonitor *m;
2203 struct list msgs;
2204
2205 rule_collection_init(&rules);
2206 HMAP_FOR_EACH (m, ofconn_node, &ofconn->monitors) {
2207 ofmonitor_collect_resume_rules(m, ofconn->monitor_paused, &rules);
2208 }
2209
2210 list_init(&msgs);
2211 ofmonitor_compose_refresh_updates(&rules, &msgs);
2212
2213 resumed = ofpraw_alloc_xid(OFPRAW_NXT_FLOW_MONITOR_RESUMED, OFP10_VERSION,
2214 htonl(0), 0);
2215 list_push_back(&msgs, &resumed->list_node);
2216 ofconn_send_replies(ofconn, &msgs);
2217
2218 ofconn->monitor_paused = 0;
2219 }
2220
2221 static bool
2222 ofmonitor_may_resume(const struct ofconn *ofconn)
2223 OVS_REQUIRES(ofproto_mutex)
2224 {
2225 return (ofconn->monitor_paused != 0
2226 && !rconn_packet_counter_n_packets(ofconn->monitor_counter));
2227 }
2228
2229 static void
2230 ofmonitor_run(struct connmgr *mgr)
2231 {
2232 struct ofconn *ofconn;
2233
2234 ovs_mutex_lock(&ofproto_mutex);
2235 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2236 if (ofmonitor_may_resume(ofconn)) {
2237 COVERAGE_INC(ofmonitor_resume);
2238 ofmonitor_resume(ofconn);
2239 }
2240 }
2241 ovs_mutex_unlock(&ofproto_mutex);
2242 }
2243
2244 static void
2245 ofmonitor_wait(struct connmgr *mgr)
2246 {
2247 struct ofconn *ofconn;
2248
2249 ovs_mutex_lock(&ofproto_mutex);
2250 LIST_FOR_EACH (ofconn, node, &mgr->all_conns) {
2251 if (ofmonitor_may_resume(ofconn)) {
2252 poll_immediate_wake();
2253 }
2254 }
2255 ovs_mutex_unlock(&ofproto_mutex);
2256 }