2 * Copyright (c) 2008, 2009 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
18 #include "reconnect.h"
23 #include "poll-loop.h"
25 #define THIS_MODULE VLM_reconnect
30 STATE(BACKOFF, 1 << 1) \
31 STATE(CONNECTING, 1 << 2) \
32 STATE(ACTIVE, 1 << 3) \
34 STATE(RECONNECT, 1 << 5)
36 #define STATE(NAME, VALUE) S_##NAME = VALUE,
42 is_connected_state(enum state state)
44 return (state & (S_ACTIVE | S_IDLE)) != 0;
56 long long int state_entered;
58 long long int last_received;
59 long long int last_connected;
60 unsigned int max_tries;
62 /* These values are simply for statistics reporting, not otherwise used
63 * directly by anything internal. */
64 long long int creation_time;
65 unsigned int n_attempted_connections, n_successful_connections;
66 unsigned int total_connected_duration;
70 static void reconnect_transition__(struct reconnect *, long long int now,
72 static long long int reconnect_deadline__(const struct reconnect *);
73 static bool reconnect_may_retry(struct reconnect *);
76 reconnect_state_name__(enum state state)
79 #define STATE(NAME, VALUE) case S_##NAME: return #NAME;
86 /* Creates and returns a new reconnect FSM with default settings. The FSM is
87 * initially disabled. The caller will likely want to call reconnect_enable()
88 * and reconnect_set_name() on the returned object. */
90 reconnect_create(long long int now)
92 struct reconnect *fsm = xzalloc(sizeof *fsm);
94 fsm->name = xstrdup("void");
95 fsm->min_backoff = 1000;
96 fsm->max_backoff = 8000;
97 fsm->probe_interval = 5000;
100 fsm->state_entered = now;
102 fsm->last_received = now;
103 fsm->last_connected = now;
104 fsm->max_tries = UINT_MAX;
105 fsm->creation_time = now;
112 reconnect_destroy(struct reconnect *fsm)
120 /* Returns 'fsm''s name. */
122 reconnect_get_name(const struct reconnect *fsm)
127 /* Sets 'fsm''s name to 'name'. If 'name' is null, then "void" is used
130 * The name set for 'fsm' is used in log messages. */
132 reconnect_set_name(struct reconnect *fsm, const char *name)
135 fsm->name = xstrdup(name ? name : "void");
138 /* Return the minimum number of milliseconds to back off between consecutive
139 * connection attempts. The default is 1000 ms. */
141 reconnect_get_min_backoff(const struct reconnect *fsm)
143 return fsm->min_backoff;
146 /* Return the maximum number of milliseconds to back off between consecutive
147 * connection attempts. The default is 8000 ms. */
149 reconnect_get_max_backoff(const struct reconnect *fsm)
151 return fsm->max_backoff;
154 /* Returns the "probe interval" for 'fsm' in milliseconds. If this is zero, it
155 * disables the connection keepalive feature. If it is nonzero, then if the
156 * interval passes while 'fsm' is connected and without reconnect_received()
157 * being called for 'fsm', reconnect_run() returns RECONNECT_PROBE. If the
158 * interval passes again without reconnect_received() being called,
159 * reconnect_run() returns RECONNECT_DISCONNECT for 'fsm'. */
161 reconnect_get_probe_interval(const struct reconnect *fsm)
163 return fsm->probe_interval;
166 /* Limits the maximum number of times that 'fsm' will ask the client to try to
167 * reconnect to 'max_tries'. UINT_MAX (the default) means an unlimited number
170 * After the number of tries has expired, the 'fsm' will disable itself
171 * instead of backing off and retrying. */
173 reconnect_set_max_tries(struct reconnect *fsm, unsigned int max_tries)
175 fsm->max_tries = max_tries;
178 /* Returns the current remaining number of connection attempts, UINT_MAX if
179 * the number is unlimited. */
181 reconnect_get_max_tries(struct reconnect *fsm)
183 return fsm->max_tries;
186 /* Configures the backoff parameters for 'fsm'. 'min_backoff' is the minimum
187 * number of milliseconds, and 'max_backoff' is the maximum, between connection
190 * 'min_backoff' must be at least 1000, and 'max_backoff' must be greater than
191 * or equal to 'min_backoff'. */
193 reconnect_set_backoff(struct reconnect *fsm, int min_backoff, int max_backoff)
195 fsm->min_backoff = MAX(min_backoff, 1000);
196 fsm->max_backoff = max_backoff ? MAX(max_backoff, 1000) : 8000;
197 if (fsm->min_backoff > fsm->max_backoff) {
198 fsm->max_backoff = fsm->min_backoff;
201 if (fsm->state == S_BACKOFF && fsm->backoff > max_backoff) {
202 fsm->backoff = max_backoff;
206 /* Sets the "probe interval" for 'fsm' to 'probe_interval', in milliseconds.
207 * If this is zero, it disables the connection keepalive feature. If it is
208 * nonzero, then if the interval passes while 'fsm' is connected and without
209 * reconnect_received() being called for 'fsm', reconnect_run() returns
210 * RECONNECT_PROBE. If the interval passes again without reconnect_received()
211 * being called, reconnect_run() returns RECONNECT_DISCONNECT for 'fsm'.
213 * If 'probe_interval' is nonzero, then it will be forced to a value of at
216 reconnect_set_probe_interval(struct reconnect *fsm, int probe_interval)
218 fsm->probe_interval = probe_interval ? MAX(1000, probe_interval) : 0;
221 /* Returns true if 'fsm' has been enabled with reconnect_enable(). Calling
222 * another function that indicates a change in connection state, such as
223 * reconnect_disconnected() or reconnect_force_reconnect(), will also enable
224 * a reconnect FSM. */
226 reconnect_is_enabled(const struct reconnect *fsm)
228 return fsm->state != S_VOID;
231 /* If 'fsm' is disabled (the default for newly created FSMs), enables it, so
232 * that the next call to reconnect_run() for 'fsm' will return
235 * If 'fsm' is not disabled, this function has no effect. */
237 reconnect_enable(struct reconnect *fsm, long long int now)
239 if (fsm->state == S_VOID && reconnect_may_retry(fsm)) {
240 reconnect_transition__(fsm, now, S_BACKOFF);
245 /* Disables 'fsm'. Until 'fsm' is enabled again, reconnect_run() will always
248 reconnect_disable(struct reconnect *fsm, long long int now)
250 if (fsm->state != S_VOID) {
251 reconnect_transition__(fsm, now, S_VOID);
255 /* If 'fsm' is enabled and currently connected (or attempting to connect),
256 * forces reconnect_run() for 'fsm' to return RECONNECT_DISCONNECT the next
257 * time it is called, which should cause the client to drop the connection (or
258 * attempt), back off, and then reconnect. */
260 reconnect_force_reconnect(struct reconnect *fsm, long long int now)
262 if (fsm->state & (S_CONNECTING | S_ACTIVE | S_IDLE)) {
263 reconnect_transition__(fsm, now, S_RECONNECT);
267 /* Tell 'fsm' that the connection dropped or that a connection attempt failed.
268 * 'error' specifies the reason: a positive value represents an errno value,
269 * EOF indicates that the connection was closed by the peer (e.g. read()
270 * returned 0), and 0 indicates no specific error.
272 * The FSM will back off, then reconnect. */
274 reconnect_disconnected(struct reconnect *fsm, long long int now, int error)
276 if (!(fsm->state & (S_BACKOFF | S_VOID))) {
277 /* Report what happened. */
278 if (fsm->state & (S_ACTIVE | S_IDLE)) {
280 VLOG_WARN("%s: connection dropped (%s)",
281 fsm->name, strerror(error));
282 } else if (error == EOF) {
283 VLOG_INFO("%s: connection closed by peer", fsm->name);
285 VLOG_INFO("%s: connection dropped", fsm->name);
289 VLOG_WARN("%s: connection attempt failed (%s)",
290 fsm->name, strerror(error));
292 VLOG_INFO("%s: connection attempt timed out", fsm->name);
297 if (fsm->state & (S_ACTIVE | S_IDLE)
298 && fsm->last_received - fsm->last_connected >= fsm->backoff) {
299 fsm->backoff = fsm->min_backoff;
301 if (fsm->backoff < fsm->min_backoff) {
302 fsm->backoff = fsm->min_backoff;
303 } else if (fsm->backoff >= fsm->max_backoff / 2) {
304 fsm->backoff = fsm->max_backoff;
308 VLOG_INFO("%s: waiting %.3g seconds before reconnect\n",
309 fsm->name, fsm->backoff / 1000.0);
312 reconnect_transition__(fsm, now,
313 reconnect_may_retry(fsm) ? S_BACKOFF : S_VOID);
317 /* Tell 'fsm' that a connection attempt is in progress.
319 * The FSM will start a timer, after which the connection attempt will be
320 * aborted (by returning RECONNECT_DISCONNECT from reconect_run()). */
322 reconnect_connecting(struct reconnect *fsm, long long int now)
324 if (fsm->state != S_CONNECTING) {
325 VLOG_INFO("%s: connecting...", fsm->name);
326 reconnect_transition__(fsm, now, S_CONNECTING);
330 /* Tell 'fsm' that the connection was successful.
332 * The FSM will start the probe interval timer, which is reset by
333 * reconnect_received(). If the timer expires, a probe will be sent (by
334 * returning RECONNECT_PROBE from reconnect_run()). If the timer expires
335 * again without being reset, the connection will be aborted (by returning
336 * RECONNECT_DISCONNECT from reconnect_run()). */
338 reconnect_connected(struct reconnect *fsm, long long int now)
340 if (!is_connected_state(fsm->state)) {
341 reconnect_connecting(fsm, now);
343 VLOG_INFO("%s: connected", fsm->name);
344 reconnect_transition__(fsm, now, S_ACTIVE);
345 fsm->last_connected = now;
349 /* Tell 'fsm' that the connection attempt failed.
351 * The FSM will back off and attempt to reconnect. */
353 reconnect_connect_failed(struct reconnect *fsm, long long int now, int error)
355 reconnect_connecting(fsm, now);
356 reconnect_disconnected(fsm, now, error);
359 /* Tell 'fsm' that some data was received. This resets the probe interval
360 * timer, so that the connection is known not to be idle. */
362 reconnect_received(struct reconnect *fsm, long long int now)
364 if (fsm->state != S_ACTIVE) {
365 reconnect_transition__(fsm, now, S_ACTIVE);
367 fsm->last_received = now;
371 reconnect_transition__(struct reconnect *fsm, long long int now,
374 if (fsm->state == S_CONNECTING) {
375 fsm->n_attempted_connections++;
376 if (state == S_ACTIVE) {
377 fsm->n_successful_connections++;
380 if (is_connected_state(fsm->state) != is_connected_state(state)) {
381 if (is_connected_state(fsm->state)) {
382 fsm->total_connected_duration += now - fsm->last_connected;
387 VLOG_DBG("%s: entering %s", fsm->name, reconnect_state_name__(state));
389 fsm->state_entered = now;
393 reconnect_deadline__(const struct reconnect *fsm)
395 assert(fsm->state_entered != LLONG_MIN);
396 switch (fsm->state) {
401 return fsm->state_entered + fsm->backoff;
404 return fsm->state_entered + MAX(1000, fsm->backoff);
407 if (fsm->probe_interval) {
408 long long int base = MAX(fsm->last_received, fsm->state_entered);
409 return base + fsm->probe_interval;
414 return fsm->state_entered + fsm->probe_interval;
417 return fsm->state_entered;
423 /* Assesses whether any action should be taken on 'fsm'. The return value is
426 * - 0: The client need not take any action.
428 * - RECONNECT_CONNECT: The client should start a connection attempt and
429 * indicate this by calling reconnect_connecting(). If the connection
430 * attempt has definitely succeeded, it should call
431 * reconnect_connected(). If the connection attempt has definitely
432 * failed, it should call reconnect_connect_failed().
434 * The FSM is smart enough to back off correctly after successful
435 * connections that quickly abort, so it is OK to call
436 * reconnect_connected() after a low-level successful connection
437 * (e.g. connect()) even if the connection might soon abort due to a
438 * failure at a high-level (e.g. SSL negotiation failure).
440 * - RECONNECT_DISCONNECT: The client should abort the current connection
441 * or connection attempt and call reconnect_disconnected() or
442 * reconnect_connect_failed() to indicate it.
444 * - RECONNECT_PROBE: The client should send some kind of request to the
445 * peer that will elicit a response, to ensure that the connection is
446 * indeed in working order. (This will only be returned if the "probe
447 * interval" is nonzero--see reconnect_set_probe_interval()).
449 enum reconnect_action
450 reconnect_run(struct reconnect *fsm, long long int now)
452 if (now >= reconnect_deadline__(fsm)) {
453 switch (fsm->state) {
458 return RECONNECT_CONNECT;
461 return RECONNECT_DISCONNECT;
464 VLOG_DBG("%s: idle %lld ms, sending inactivity probe", fsm->name,
465 now - MAX(fsm->last_received, fsm->state_entered));
466 reconnect_transition__(fsm, now, S_IDLE);
467 return RECONNECT_PROBE;
470 VLOG_ERR("%s: no response to inactivity probe after %.3g "
471 "seconds, disconnecting",
472 fsm->name, (now - fsm->state_entered) / 1000.0);
473 return RECONNECT_DISCONNECT;
476 return RECONNECT_DISCONNECT;
481 return fsm->state == S_CONNECTING ? RECONNECT_CONNECT : 0;
485 /* Causes the next call to poll_block() to wake up when reconnect_run() should
486 * be called on 'fsm'. */
488 reconnect_wait(struct reconnect *fsm, long long int now)
490 int timeout = reconnect_timeout(fsm, now);
492 poll_timer_wait(timeout);
496 /* Returns the number of milliseconds after which reconnect_run() should be
497 * called on 'fsm' if nothing else notable happens in the meantime, or a
498 * negative number if this is currently unnecessary. */
500 reconnect_timeout(struct reconnect *fsm, long long int now)
502 long long int deadline = reconnect_deadline__(fsm);
503 if (deadline != LLONG_MAX) {
504 long long int remaining = deadline - now;
505 return MAX(0, MIN(INT_MAX, remaining));
510 /* Returns true if 'fsm' is currently believed to be connected, that is, if
511 * reconnect_connected() was called more recently than any call to
512 * reconnect_connect_failed() or reconnect_disconnected() or
513 * reconnect_disable(), and false otherwise. */
515 reconnect_is_connected(const struct reconnect *fsm)
517 return is_connected_state(fsm->state);
520 /* Returns the number of milliseconds for which 'fsm' has been continuously
521 * connected to its peer. (If 'fsm' is not currently connected, this is 0.) */
523 reconnect_get_connection_duration(const struct reconnect *fsm,
526 return reconnect_is_connected(fsm) ? now - fsm->last_connected : 0;
529 /* Copies various statistics for 'fsm' into '*stats'. */
531 reconnect_get_stats(const struct reconnect *fsm, long long int now,
532 struct reconnect_stats *stats)
534 stats->creation_time = fsm->creation_time;
535 stats->last_received = fsm->last_received;
536 stats->last_connected = fsm->last_connected;
537 stats->backoff = fsm->backoff;
538 stats->seqno = fsm->seqno;
539 stats->is_connected = reconnect_is_connected(fsm);
540 stats->current_connection_duration
541 = reconnect_get_connection_duration(fsm, now);
542 stats->total_connected_duration = (stats->current_connection_duration
543 + fsm->total_connected_duration);
544 stats->n_attempted_connections = fsm->n_attempted_connections;
545 stats->n_successful_connections = fsm->n_successful_connections;
546 stats->state = reconnect_state_name__(fsm->state);
547 stats->state_elapsed = now - fsm->state_entered;
551 reconnect_may_retry(struct reconnect *fsm)
553 bool may_retry = fsm->max_tries > 0;
554 if (may_retry && fsm->max_tries != UINT_MAX) {