2 * Copyright (c) 2008, 2009, 2010, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
18 #include "reconnect.h"
22 #include "poll-loop.h"
24 #include "openvswitch/vlog.h"
26 VLOG_DEFINE_THIS_MODULE(reconnect);
30 STATE(BACKOFF, 1 << 1) \
31 STATE(CONNECTING, 1 << 3) \
32 STATE(ACTIVE, 1 << 4) \
34 STATE(RECONNECT, 1 << 6) \
35 STATE(LISTENING, 1 << 7)
37 #define STATE(NAME, VALUE) S_##NAME = VALUE,
43 is_connected_state(enum state state)
45 return (state & (S_ACTIVE | S_IDLE)) != 0;
55 enum vlog_level info; /* Used for informational messages. */
59 long long int state_entered;
61 long long int last_activity;
62 long long int last_connected;
63 long long int last_disconnected;
64 unsigned int max_tries;
66 /* These values are simply for statistics reporting, not otherwise used
67 * directly by anything internal. */
68 long long int creation_time;
69 unsigned int n_attempted_connections, n_successful_connections;
70 unsigned int total_connected_duration;
74 static void reconnect_transition__(struct reconnect *, long long int now,
76 static long long int reconnect_deadline__(const struct reconnect *);
77 static bool reconnect_may_retry(struct reconnect *);
80 reconnect_state_name__(enum state state)
83 #define STATE(NAME, VALUE) case S_##NAME: return #NAME;
90 /* Creates and returns a new reconnect FSM with default settings. The FSM is
91 * initially disabled. The caller will likely want to call reconnect_enable()
92 * and reconnect_set_name() on the returned object. */
94 reconnect_create(long long int now)
96 struct reconnect *fsm = xzalloc(sizeof *fsm);
98 fsm->name = xstrdup("void");
99 fsm->min_backoff = RECONNECT_DEFAULT_MIN_BACKOFF;
100 fsm->max_backoff = RECONNECT_DEFAULT_MAX_BACKOFF;
101 fsm->probe_interval = RECONNECT_DEFAULT_PROBE_INTERVAL;
102 fsm->passive = false;
103 fsm->info = VLL_INFO;
106 fsm->state_entered = now;
108 fsm->last_activity = now;
109 fsm->last_connected = LLONG_MAX;
110 fsm->last_disconnected = LLONG_MAX;
111 fsm->max_tries = UINT_MAX;
112 fsm->creation_time = now;
119 reconnect_destroy(struct reconnect *fsm)
127 /* If 'quiet' is true, 'fsm' will log informational messages at level VLL_DBG,
128 * by default keeping them out of log files. This is appropriate if the
129 * connection is one that is expected to be short-lived, so that the log
130 * messages are merely distracting.
132 * If 'quiet' is false, 'fsm' logs informational messages at level VLL_INFO.
133 * This is the default.
135 * This setting has no effect on the log level of debugging, warning, or error
138 reconnect_set_quiet(struct reconnect *fsm, bool quiet)
140 fsm->info = quiet ? VLL_DBG : VLL_INFO;
143 /* Returns 'fsm''s name. */
145 reconnect_get_name(const struct reconnect *fsm)
150 /* Sets 'fsm''s name to 'name'. If 'name' is null, then "void" is used
153 * The name set for 'fsm' is used in log messages. */
155 reconnect_set_name(struct reconnect *fsm, const char *name)
158 fsm->name = xstrdup(name ? name : "void");
161 /* Return the minimum number of milliseconds to back off between consecutive
162 * connection attempts. The default is RECONNECT_DEFAULT_MIN_BACKOFF. */
164 reconnect_get_min_backoff(const struct reconnect *fsm)
166 return fsm->min_backoff;
169 /* Return the maximum number of milliseconds to back off between consecutive
170 * connection attempts. The default is RECONNECT_DEFAULT_MAX_BACKOFF. */
172 reconnect_get_max_backoff(const struct reconnect *fsm)
174 return fsm->max_backoff;
177 /* Returns the "probe interval" for 'fsm' in milliseconds. If this is zero, it
178 * disables the connection keepalive feature. If it is nonzero, then if the
179 * interval passes while 'fsm' is connected and without reconnect_activity()
180 * being called for 'fsm', reconnect_run() returns RECONNECT_PROBE. If the
181 * interval passes again without reconnect_activity() being called,
182 * reconnect_run() returns RECONNECT_DISCONNECT for 'fsm'. */
184 reconnect_get_probe_interval(const struct reconnect *fsm)
186 return fsm->probe_interval;
189 /* Limits the maximum number of times that 'fsm' will ask the client to try to
190 * reconnect to 'max_tries'. UINT_MAX (the default) means an unlimited number
193 * After the number of tries has expired, the 'fsm' will disable itself
194 * instead of backing off and retrying. */
196 reconnect_set_max_tries(struct reconnect *fsm, unsigned int max_tries)
198 fsm->max_tries = max_tries;
201 /* Returns the current remaining number of connection attempts, UINT_MAX if
202 * the number is unlimited. */
204 reconnect_get_max_tries(struct reconnect *fsm)
206 return fsm->max_tries;
209 /* Configures the backoff parameters for 'fsm'. 'min_backoff' is the minimum
210 * number of milliseconds, and 'max_backoff' is the maximum, between connection
211 * attempts. The current backoff is also the duration that 'fsm' is willing to
212 * wait for a given connection to succeed or fail.
214 * 'min_backoff' must be at least 1000, and 'max_backoff' must be greater than
215 * or equal to 'min_backoff'.
217 * Pass 0 for 'min_backoff' or 'max_backoff' or both to use the defaults. */
219 reconnect_set_backoff(struct reconnect *fsm, int min_backoff, int max_backoff)
221 fsm->min_backoff = MAX(min_backoff, 1000);
222 fsm->max_backoff = (max_backoff
223 ? MAX(max_backoff, 1000)
224 : RECONNECT_DEFAULT_MAX_BACKOFF);
225 if (fsm->min_backoff > fsm->max_backoff) {
226 fsm->max_backoff = fsm->min_backoff;
229 if (fsm->state == S_BACKOFF && fsm->backoff > max_backoff) {
230 fsm->backoff = max_backoff;
234 /* Sets the "probe interval" for 'fsm' to 'probe_interval', in milliseconds.
235 * If this is zero, it disables the connection keepalive feature. If it is
236 * nonzero, then if the interval passes while 'fsm' is connected and without
237 * reconnect_activity() being called for 'fsm', reconnect_run() returns
238 * RECONNECT_PROBE. If the interval passes again without reconnect_activity()
239 * being called, reconnect_run() returns RECONNECT_DISCONNECT for 'fsm'.
241 * If 'probe_interval' is nonzero, then it will be forced to a value of at
244 reconnect_set_probe_interval(struct reconnect *fsm, int probe_interval)
246 fsm->probe_interval = probe_interval ? MAX(1000, probe_interval) : 0;
249 /* Returns true if 'fsm' is in passive mode, false if 'fsm' is in active mode
252 reconnect_is_passive(const struct reconnect *fsm)
257 /* Configures 'fsm' for active or passive mode. In active mode (the default),
258 * the FSM is attempting to connect to a remote host. In passive mode, the FSM
259 * is listening for connections from a remote host. */
261 reconnect_set_passive(struct reconnect *fsm, bool passive, long long int now)
263 if (fsm->passive != passive) {
264 fsm->passive = passive;
267 ? fsm->state & (S_CONNECTING | S_RECONNECT)
268 : fsm->state == S_LISTENING && reconnect_may_retry(fsm)) {
269 reconnect_transition__(fsm, now, S_BACKOFF);
275 /* Returns true if 'fsm' has been enabled with reconnect_enable(). Calling
276 * another function that indicates a change in connection state, such as
277 * reconnect_disconnected() or reconnect_force_reconnect(), will also enable
278 * a reconnect FSM. */
280 reconnect_is_enabled(const struct reconnect *fsm)
282 return fsm->state != S_VOID;
285 /* If 'fsm' is disabled (the default for newly created FSMs), enables it, so
286 * that the next call to reconnect_run() for 'fsm' will return
289 * If 'fsm' is not disabled, this function has no effect. */
291 reconnect_enable(struct reconnect *fsm, long long int now)
293 if (fsm->state == S_VOID && reconnect_may_retry(fsm)) {
294 reconnect_transition__(fsm, now, S_BACKOFF);
299 /* Disables 'fsm'. Until 'fsm' is enabled again, reconnect_run() will always
302 reconnect_disable(struct reconnect *fsm, long long int now)
304 if (fsm->state != S_VOID) {
305 reconnect_transition__(fsm, now, S_VOID);
309 /* If 'fsm' is enabled and currently connected (or attempting to connect),
310 * forces reconnect_run() for 'fsm' to return RECONNECT_DISCONNECT the next
311 * time it is called, which should cause the client to drop the connection (or
312 * attempt), back off, and then reconnect. */
314 reconnect_force_reconnect(struct reconnect *fsm, long long int now)
316 if (fsm->state & (S_CONNECTING | S_ACTIVE | S_IDLE)) {
317 reconnect_transition__(fsm, now, S_RECONNECT);
321 /* Tell 'fsm' that the connection dropped or that a connection attempt failed.
322 * 'error' specifies the reason: a positive value represents an errno value,
323 * EOF indicates that the connection was closed by the peer (e.g. read()
324 * returned 0), and 0 indicates no specific error.
326 * The FSM will back off, then reconnect. */
328 reconnect_disconnected(struct reconnect *fsm, long long int now, int error)
330 if (!(fsm->state & (S_BACKOFF | S_VOID))) {
331 /* Report what happened. */
332 if (fsm->state & (S_ACTIVE | S_IDLE)) {
334 VLOG_WARN("%s: connection dropped (%s)",
335 fsm->name, ovs_strerror(error));
336 } else if (error == EOF) {
337 VLOG(fsm->info, "%s: connection closed by peer", fsm->name);
339 VLOG(fsm->info, "%s: connection dropped", fsm->name);
341 } else if (fsm->state == S_LISTENING) {
343 VLOG_WARN("%s: error listening for connections (%s)",
344 fsm->name, ovs_strerror(error));
346 VLOG(fsm->info, "%s: error listening for connections",
350 const char *type = fsm->passive ? "listen" : "connection";
352 VLOG_INFO("%s: %s attempt failed (%s)",
353 fsm->name, type, ovs_strerror(error));
355 VLOG(fsm->info, "%s: %s attempt timed out", fsm->name, type);
359 if (fsm->state & (S_ACTIVE | S_IDLE)) {
360 fsm->last_disconnected = now;
363 if (fsm->state & (S_ACTIVE | S_IDLE)
364 && (fsm->last_activity - fsm->last_connected >= fsm->backoff
366 fsm->backoff = fsm->passive ? 0 : fsm->min_backoff;
368 if (fsm->backoff < fsm->min_backoff) {
369 fsm->backoff = fsm->min_backoff;
370 } else if (fsm->backoff >= fsm->max_backoff / 2) {
371 fsm->backoff = fsm->max_backoff;
376 VLOG(fsm->info, "%s: waiting %.3g seconds before trying to "
377 "listen again", fsm->name, fsm->backoff / 1000.0);
379 VLOG(fsm->info, "%s: waiting %.3g seconds before reconnect",
380 fsm->name, fsm->backoff / 1000.0);
384 reconnect_transition__(fsm, now,
385 reconnect_may_retry(fsm) ? S_BACKOFF : S_VOID);
389 /* Tell 'fsm' that a connection or listening attempt is in progress.
391 * The FSM will start a timer, after which the connection or listening attempt
392 * will be aborted (by returning RECONNECT_DISCONNECT from
393 * reconnect_run()). */
395 reconnect_connecting(struct reconnect *fsm, long long int now)
397 if (fsm->state != S_CONNECTING) {
399 VLOG(fsm->info, "%s: listening...", fsm->name);
401 VLOG(fsm->info, "%s: connecting...", fsm->name);
403 reconnect_transition__(fsm, now, S_CONNECTING);
407 /* Tell 'fsm' that the client is listening for connection attempts. This state
408 * last indefinitely until the client reports some change.
410 * The natural progression from this state is for the client to report that a
411 * connection has been accepted or is in progress of being accepted, by calling
412 * reconnect_connecting() or reconnect_connected().
414 * The client may also report that listening failed (e.g. accept() returned an
415 * unexpected error such as ENOMEM) by calling reconnect_listen_error(), in
416 * which case the FSM will back off and eventually return RECONNECT_CONNECT
417 * from reconnect_run() to tell the client to try listening again. */
419 reconnect_listening(struct reconnect *fsm, long long int now)
421 if (fsm->state != S_LISTENING) {
422 VLOG(fsm->info, "%s: listening...", fsm->name);
423 reconnect_transition__(fsm, now, S_LISTENING);
427 /* Tell 'fsm' that the client's attempt to accept a connection failed
428 * (e.g. accept() returned an unexpected error such as ENOMEM).
430 * If the FSM is currently listening (reconnect_listening() was called), it
431 * will back off and eventually return RECONNECT_CONNECT from reconnect_run()
432 * to tell the client to try listening again. If there is an active
433 * connection, this will be delayed until that connection drops. */
435 reconnect_listen_error(struct reconnect *fsm, long long int now, int error)
437 if (fsm->state == S_LISTENING) {
438 reconnect_disconnected(fsm, now, error);
442 /* Tell 'fsm' that the connection was successful.
444 * The FSM will start the probe interval timer, which is reset by
445 * reconnect_activity(). If the timer expires, a probe will be sent (by
446 * returning RECONNECT_PROBE from reconnect_run()). If the timer expires
447 * again without being reset, the connection will be aborted (by returning
448 * RECONNECT_DISCONNECT from reconnect_run()). */
450 reconnect_connected(struct reconnect *fsm, long long int now)
452 if (!is_connected_state(fsm->state)) {
453 reconnect_connecting(fsm, now);
455 VLOG(fsm->info, "%s: connected", fsm->name);
456 reconnect_transition__(fsm, now, S_ACTIVE);
457 fsm->last_connected = now;
461 /* Tell 'fsm' that the connection attempt failed.
463 * The FSM will back off and attempt to reconnect. */
465 reconnect_connect_failed(struct reconnect *fsm, long long int now, int error)
467 reconnect_connecting(fsm, now);
468 reconnect_disconnected(fsm, now, error);
471 /* Tell 'fsm' that some activity has occurred on the connection. This resets
472 * the probe interval timer, so that the connection is known not to be idle. */
474 reconnect_activity(struct reconnect *fsm, long long int now)
476 if (fsm->state != S_ACTIVE) {
477 reconnect_transition__(fsm, now, S_ACTIVE);
479 fsm->last_activity = now;
483 reconnect_transition__(struct reconnect *fsm, long long int now,
486 if (fsm->state == S_CONNECTING) {
487 fsm->n_attempted_connections++;
488 if (state == S_ACTIVE) {
489 fsm->n_successful_connections++;
492 if (is_connected_state(fsm->state) != is_connected_state(state)) {
493 if (is_connected_state(fsm->state)) {
494 fsm->total_connected_duration += now - fsm->last_connected;
499 VLOG_DBG("%s: entering %s", fsm->name, reconnect_state_name__(state));
501 fsm->state_entered = now;
505 reconnect_deadline__(const struct reconnect *fsm)
507 ovs_assert(fsm->state_entered != LLONG_MIN);
508 switch (fsm->state) {
514 return fsm->state_entered + fsm->backoff;
517 return fsm->state_entered + MAX(1000, fsm->backoff);
520 if (fsm->probe_interval) {
521 long long int base = MAX(fsm->last_activity, fsm->state_entered);
522 return base + fsm->probe_interval;
527 if (fsm->probe_interval) {
528 return fsm->state_entered + fsm->probe_interval;
533 return fsm->state_entered;
539 /* Assesses whether any action should be taken on 'fsm'. The return value is
542 * - 0: The client need not take any action.
544 * - Active client, RECONNECT_CONNECT: The client should start a connection
545 * attempt and indicate this by calling reconnect_connecting(). If the
546 * connection attempt has definitely succeeded, it should call
547 * reconnect_connected(). If the connection attempt has definitely
548 * failed, it should call reconnect_connect_failed().
550 * The FSM is smart enough to back off correctly after successful
551 * connections that quickly abort, so it is OK to call
552 * reconnect_connected() after a low-level successful connection
553 * (e.g. connect()) even if the connection might soon abort due to a
554 * failure at a high-level (e.g. SSL negotiation failure).
556 * - Passive client, RECONNECT_CONNECT: The client should try to listen for
557 * a connection, if it is not already listening. It should call
558 * reconnect_listening() if successful, otherwise reconnect_connecting()
559 * or reconnected_connect_failed() if the attempt is in progress or
560 * definitely failed, respectively.
562 * A listening passive client should constantly attempt to accept a new
563 * connection and report an accepted connection with
564 * reconnect_connected().
566 * - RECONNECT_DISCONNECT: The client should abort the current connection
567 * or connection attempt or listen attempt and call
568 * reconnect_disconnected() or reconnect_connect_failed() to indicate it.
570 * - RECONNECT_PROBE: The client should send some kind of request to the
571 * peer that will elicit a response, to ensure that the connection is
572 * indeed in working order. (This will only be returned if the "probe
573 * interval" is nonzero--see reconnect_set_probe_interval()).
575 enum reconnect_action
576 reconnect_run(struct reconnect *fsm, long long int now)
578 if (now >= reconnect_deadline__(fsm)) {
579 switch (fsm->state) {
584 return RECONNECT_CONNECT;
587 return RECONNECT_DISCONNECT;
590 VLOG_DBG("%s: idle %lld ms, sending inactivity probe", fsm->name,
591 now - MAX(fsm->last_activity, fsm->state_entered));
592 reconnect_transition__(fsm, now, S_IDLE);
593 return RECONNECT_PROBE;
596 VLOG_ERR("%s: no response to inactivity probe after %.3g "
597 "seconds, disconnecting",
598 fsm->name, (now - fsm->state_entered) / 1000.0);
599 return RECONNECT_DISCONNECT;
602 return RECONNECT_DISCONNECT;
614 /* Causes the next call to poll_block() to wake up when reconnect_run() should
615 * be called on 'fsm'. */
617 reconnect_wait(struct reconnect *fsm, long long int now)
619 int timeout = reconnect_timeout(fsm, now);
621 poll_timer_wait(timeout);
625 /* Returns the number of milliseconds after which reconnect_run() should be
626 * called on 'fsm' if nothing else notable happens in the meantime, or a
627 * negative number if this is currently unnecessary. */
629 reconnect_timeout(struct reconnect *fsm, long long int now)
631 long long int deadline = reconnect_deadline__(fsm);
632 if (deadline != LLONG_MAX) {
633 long long int remaining = deadline - now;
634 return MAX(0, MIN(INT_MAX, remaining));
639 /* Returns true if 'fsm' is currently believed to be connected, that is, if
640 * reconnect_connected() was called more recently than any call to
641 * reconnect_connect_failed() or reconnect_disconnected() or
642 * reconnect_disable(), and false otherwise. */
644 reconnect_is_connected(const struct reconnect *fsm)
646 return is_connected_state(fsm->state);
649 /* Returns the number of milliseconds since 'fsm' last successfully connected
650 * to its peer (even if it has since disconnected). Returns UINT_MAX if never
653 reconnect_get_last_connect_elapsed(const struct reconnect *fsm,
656 return fsm->last_connected == LLONG_MAX ? UINT_MAX
657 : now - fsm->last_connected;
660 /* Returns the number of milliseconds since 'fsm' last disconnected
661 * from its peer (even if it has since reconnected). Returns UINT_MAX if never
664 reconnect_get_last_disconnect_elapsed(const struct reconnect *fsm,
667 return fsm->last_disconnected == LLONG_MAX ? UINT_MAX
668 : now - fsm->last_disconnected;
671 /* Copies various statistics for 'fsm' into '*stats'. */
673 reconnect_get_stats(const struct reconnect *fsm, long long int now,
674 struct reconnect_stats *stats)
676 stats->creation_time = fsm->creation_time;
677 stats->last_activity = fsm->last_activity;
678 stats->last_connected = fsm->last_connected;
679 stats->last_disconnected = fsm->last_disconnected;
680 stats->backoff = fsm->backoff;
681 stats->seqno = fsm->seqno;
682 stats->is_connected = reconnect_is_connected(fsm);
683 stats->msec_since_connect
684 = reconnect_get_last_connect_elapsed(fsm, now);
685 stats->msec_since_disconnect
686 = reconnect_get_last_disconnect_elapsed(fsm, now);
687 stats->total_connected_duration = fsm->total_connected_duration
688 + (is_connected_state(fsm->state)
689 ? reconnect_get_last_connect_elapsed(fsm, now) : 0);
690 stats->n_attempted_connections = fsm->n_attempted_connections;
691 stats->n_successful_connections = fsm->n_successful_connections;
692 stats->state = reconnect_state_name__(fsm->state);
693 stats->state_elapsed = now - fsm->state_entered;
697 reconnect_may_retry(struct reconnect *fsm)
699 bool may_retry = fsm->max_tries > 0;
700 if (may_retry && fsm->max_tries != UINT_MAX) {