1 /* Copyright (c) 2008 The Board of Trustees of The Leland Stanford
4 * We are making the OpenFlow specification and associated documentation
5 * (Software) available for public use and benefit with the expectation
6 * that others will use, modify and enhance the Software and contribute
7 * those enhancements back to the community. However, since we would
8 * like to make the Software available for broadest use, with as few
9 * restrictions as possible permission is hereby granted, free of
10 * charge, to any person obtaining a copy of this Software to deal in
11 * the Software under the copyrights without restriction, including
12 * without limitation the rights to use, copy, modify, merge, publish,
13 * distribute, sublicense, and/or sell copies of the Software, and to
14 * permit persons to whom the Software is furnished to do so, subject to
15 * the following conditions:
17 * The above copyright notice and this permission notice shall be
18 * included in all copies or substantial portions of the Software.
20 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
21 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
22 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
23 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
24 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
25 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
26 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
29 * The name and trademarks of copyright holder(s) may NOT be used in
30 * advertising or publicity pertaining to the Software or any
31 * derivatives without specific, written prior permission.
42 #include "poll-loop.h"
43 #include "ofp-print.h"
49 #define THIS_MODULE VLM_rconn
54 STATE(BACKOFF, 1 << 1) \
55 STATE(CONNECTING, 1 << 2) \
56 STATE(ACTIVE, 1 << 3) \
59 #define STATE(NAME, VALUE) S_##NAME = VALUE,
65 state_name(enum state state)
68 #define STATE(NAME, VALUE) case S_##NAME: return #NAME;
75 /* A reliable connection to an OpenFlow switch or controller.
77 * See the large comment in rconn.h for more information. */
90 time_t backoff_deadline;
92 time_t last_connected;
93 unsigned int packets_sent;
95 /* These values are simply for statistics reporting, not used directly by
96 * anything internal to the rconn (or the secchan for that matter). */
97 unsigned int packets_received;
98 unsigned int n_attempted_connections, n_successful_connections;
100 unsigned long int total_time_connected;
102 /* If we can't connect to the peer, it could be for any number of reasons.
103 * Usually, one would assume it is because the peer is not running or
104 * because the network is partitioned. But it could also be because the
105 * network topology has changed, in which case the upper layer will need to
106 * reassess it (in particular, obtain a new IP address via DHCP and find
107 * the new location of the controller). We set this flag when we suspect
108 * that this could be the case. */
109 bool questionable_connectivity;
110 time_t last_questioned;
112 /* Throughout this file, "probe" is shorthand for "inactivity probe".
113 * When nothing has been received from the peer for a while, we send out
114 * an echo request as an inactivity probe packet. We should receive back
116 int probe_interval; /* Secs of inactivity before sending probe. */
119 static unsigned int elapsed_in_this_state(const struct rconn *);
120 static unsigned int timeout(const struct rconn *);
121 static bool timed_out(const struct rconn *);
122 static void state_transition(struct rconn *, enum state);
123 static int try_send(struct rconn *);
124 static int reconnect(struct rconn *);
125 static void disconnect(struct rconn *, int error);
126 static void flush_queue(struct rconn *);
127 static void question_connectivity(struct rconn *);
129 /* Creates a new rconn, connects it (reliably) to 'name', and returns it. */
131 rconn_new(const char *name, int inactivity_probe_interval, int max_backoff)
133 struct rconn *rc = rconn_create(inactivity_probe_interval, max_backoff);
134 rconn_connect(rc, name);
138 /* Creates a new rconn, connects it (unreliably) to 'vconn', and returns it. */
140 rconn_new_from_vconn(const char *name, struct vconn *vconn)
142 struct rconn *rc = rconn_create(60, 0);
143 rconn_connect_unreliably(rc, name, vconn);
147 /* Creates and returns a new rconn.
149 * 'probe_interval' is a number of seconds. If the interval passes once
150 * without an OpenFlow message being received from the peer, the rconn sends
151 * out an "echo request" message. If the interval passes again without a
152 * message being received, the rconn disconnects and re-connects to the peer.
153 * Setting 'probe_interval' to 0 disables this behavior.
155 * 'max_backoff' is the maximum number of seconds between attempts to connect
156 * to the peer. The actual interval starts at 1 second and doubles on each
157 * failure until it reaches 'max_backoff'. If 0 is specified, the default of
158 * 60 seconds is used. */
160 rconn_create(int probe_interval, int max_backoff)
162 struct rconn *rc = xcalloc(1, sizeof *rc);
165 rc->state_entered = time_now();
168 rc->name = xstrdup("void");
169 rc->reliable = false;
171 queue_init(&rc->txq);
174 rc->max_backoff = max_backoff ? max_backoff : 60;
175 rc->backoff_deadline = TIME_MIN;
176 rc->last_received = time_now();
177 rc->last_connected = time_now();
179 rc->packets_sent = 0;
181 rc->packets_received = 0;
182 rc->n_attempted_connections = 0;
183 rc->n_successful_connections = 0;
184 rc->creation_time = time_now();
185 rc->total_time_connected = 0;
187 rc->questionable_connectivity = false;
188 rc->last_questioned = time_now();
190 rc->probe_interval = probe_interval ? MAX(5, probe_interval) : 0;
196 rconn_connect(struct rconn *rc, const char *name)
198 rconn_disconnect(rc);
200 rc->name = xstrdup(name);
202 return reconnect(rc);
206 rconn_connect_unreliably(struct rconn *rc,
207 const char *name, struct vconn *vconn)
209 assert(vconn != NULL);
210 rconn_disconnect(rc);
212 rc->name = xstrdup(name);
213 rc->reliable = false;
215 rc->last_connected = time_now();
216 state_transition(rc, S_ACTIVE);
220 rconn_disconnect(struct rconn *rc)
223 vconn_close(rc->vconn);
227 rc->name = xstrdup("void");
228 rc->reliable = false;
231 rc->backoff_deadline = TIME_MIN;
233 state_transition(rc, S_VOID);
236 /* Disconnects 'rc' and frees the underlying storage. */
238 rconn_destroy(struct rconn *rc)
242 vconn_close(rc->vconn);
244 queue_destroy(&rc->txq);
250 timeout_VOID(const struct rconn *rc)
256 run_VOID(struct rconn *rc)
262 reconnect(struct rconn *rc)
266 VLOG_WARN("%s: connecting...", rc->name);
267 rc->n_attempted_connections++;
268 retval = vconn_open(rc->name, &rc->vconn);
270 rc->backoff_deadline = time_now() + rc->backoff;
271 state_transition(rc, S_CONNECTING);
273 VLOG_WARN("%s: connection failed (%s)", rc->name, strerror(retval));
280 timeout_BACKOFF(const struct rconn *rc)
286 run_BACKOFF(struct rconn *rc)
294 timeout_CONNECTING(const struct rconn *rc)
296 return MAX(1, rc->backoff);
300 run_CONNECTING(struct rconn *rc)
302 int retval = vconn_connect(rc->vconn);
304 VLOG_WARN("%s: connected", rc->name);
305 rc->n_successful_connections++;
306 if (vconn_is_passive(rc->vconn)) {
307 error(0, "%s: passive vconn not supported", rc->name);
308 state_transition(rc, S_VOID);
310 state_transition(rc, S_ACTIVE);
311 rc->last_connected = rc->state_entered;
313 } else if (retval != EAGAIN) {
314 VLOG_WARN("%s: connection failed (%s)", rc->name, strerror(retval));
315 disconnect(rc, retval);
316 } else if (timed_out(rc)) {
317 VLOG_WARN("%s: connection timed out", rc->name);
318 rc->backoff_deadline = TIME_MAX; /* Prevent resetting backoff. */
324 do_tx_work(struct rconn *rc)
329 while (rc->txq.n > 0) {
330 int error = try_send(rc);
336 poll_immediate_wake();
341 timeout_ACTIVE(const struct rconn *rc)
343 if (rc->probe_interval) {
344 unsigned int base = MAX(rc->last_received, rc->state_entered);
345 unsigned int arg = base + rc->probe_interval - rc->state_entered;
352 run_ACTIVE(struct rconn *rc)
355 unsigned int base = MAX(rc->last_received, rc->state_entered);
356 rconn_send(rc, make_echo_request(), NULL);
357 VLOG_DBG("%s: idle %u seconds, sending inactivity probe",
358 rc->name, (unsigned int) (time_now() - base));
359 state_transition(rc, S_IDLE);
367 timeout_IDLE(const struct rconn *rc)
369 return rc->probe_interval;
373 run_IDLE(struct rconn *rc)
376 question_connectivity(rc);
377 VLOG_ERR("%s: no response to inactivity probe after %u "
378 "seconds, disconnecting",
379 rc->name, elapsed_in_this_state(rc));
386 /* Performs whatever activities are necessary to maintain 'rc': if 'rc' is
387 * disconnected, attempts to (re)connect, backing off as necessary; if 'rc' is
388 * connected, attempts to send packets in the send queue, if any. */
390 rconn_run(struct rconn *rc)
394 old_state = rc->state;
396 #define STATE(NAME, VALUE) case S_##NAME: run_##NAME(rc); break;
402 } while (rc->state != old_state);
405 /* Causes the next call to poll_block() to wake up when rconn_run() should be
408 rconn_run_wait(struct rconn *rc)
410 unsigned int timeo = timeout(rc);
411 if (timeo != UINT_MAX) {
412 unsigned int expires = sat_add(rc->state_entered, timeo);
413 unsigned int remaining = sat_sub(expires, time_now());
414 poll_timer_wait(sat_mul(remaining, 1000));
417 if ((rc->state & (S_ACTIVE | S_IDLE)) && rc->txq.n) {
418 vconn_wait(rc->vconn, WAIT_SEND);
422 /* Attempts to receive a packet from 'rc'. If successful, returns the packet;
423 * otherwise, returns a null pointer. The caller is responsible for freeing
424 * the packet (with buffer_delete()). */
426 rconn_recv(struct rconn *rc)
428 if (rc->state & (S_ACTIVE | S_IDLE)) {
429 struct buffer *buffer;
430 int error = vconn_recv(rc->vconn, &buffer);
432 rc->last_received = time_now();
433 rc->packets_received++;
434 if (rc->state == S_IDLE) {
435 state_transition(rc, S_ACTIVE);
438 } else if (error != EAGAIN) {
439 disconnect(rc, error);
445 /* Causes the next call to poll_block() to wake up when a packet may be ready
446 * to be received by vconn_recv() on 'rc'. */
448 rconn_recv_wait(struct rconn *rc)
451 vconn_wait(rc->vconn, WAIT_RECV);
455 /* Sends 'b' on 'rc'. Returns 0 if successful (in which case 'b' is
456 * destroyed), or ENOTCONN if 'rc' is not currently connected (in which case
457 * the caller retains ownership of 'b').
459 * If 'n_queued' is non-null, then '*n_queued' will be incremented while the
460 * packet is in flight, then decremented when it has been sent (or discarded
461 * due to disconnection). Because 'b' may be sent (or discarded) before this
462 * function returns, the caller may not be able to observe any change in
465 * There is no rconn_send_wait() function: an rconn has a send queue that it
466 * takes care of sending if you call rconn_run(), which will have the side
467 * effect of waking up poll_block(). */
469 rconn_send(struct rconn *rc, struct buffer *b, int *n_queued)
471 if (rconn_is_connected(rc)) {
472 b->private = n_queued;
476 queue_push_tail(&rc->txq, b);
477 if (rc->txq.n == 1) {
486 /* Sends 'b' on 'rc'. Increments '*n_queued' while the packet is in flight; it
487 * will be decremented when it has been sent (or discarded due to
488 * disconnection). Returns 0 if successful, EAGAIN if '*n_queued' is already
489 * at least as large of 'queue_limit', or ENOTCONN if 'rc' is not currently
490 * connected. Regardless of return value, 'b' is destroyed.
492 * Because 'b' may be sent (or discarded) before this function returns, the
493 * caller may not be able to observe any change in '*n_queued'.
495 * There is no rconn_send_wait() function: an rconn has a send queue that it
496 * takes care of sending if you call rconn_run(), which will have the side
497 * effect of waking up poll_block(). */
499 rconn_send_with_limit(struct rconn *rc, struct buffer *b,
500 int *n_queued, int queue_limit)
503 retval = *n_queued >= queue_limit ? EAGAIN : rconn_send(rc, b, n_queued);
510 /* Returns the total number of packets successfully sent on the underlying
511 * vconn. A packet is not counted as sent while it is still queued in the
512 * rconn, only when it has been successfuly passed to the vconn. */
514 rconn_packets_sent(const struct rconn *rc)
516 return rc->packets_sent;
519 /* Returns 'rc''s name (the 'name' argument passed to rconn_new()). */
521 rconn_get_name(const struct rconn *rc)
526 /* Returns true if 'rconn' is connected or in the process of reconnecting,
527 * false if 'rconn' is disconnected and will not reconnect on its own. */
529 rconn_is_alive(const struct rconn *rconn)
531 return rconn->state != S_VOID;
534 /* Returns true if 'rconn' is connected, false otherwise. */
536 rconn_is_connected(const struct rconn *rconn)
538 return rconn->state & (S_ACTIVE | S_IDLE);
541 /* Returns 0 if 'rconn' is connected, otherwise the number of seconds that it
542 * has been disconnected. */
544 rconn_disconnected_duration(const struct rconn *rconn)
546 return rconn_is_connected(rconn) ? 0 : time_now() - rconn->last_received;
549 /* Returns the IP address of the peer, or 0 if the peer is not connected over
550 * an IP-based protocol or if its IP address is not known. */
552 rconn_get_ip(const struct rconn *rconn)
554 return rconn->vconn ? vconn_get_ip(rconn->vconn) : 0;
557 /* If 'rconn' can't connect to the peer, it could be for any number of reasons.
558 * Usually, one would assume it is because the peer is not running or because
559 * the network is partitioned. But it could also be because the network
560 * topology has changed, in which case the upper layer will need to reassess it
561 * (in particular, obtain a new IP address via DHCP and find the new location
562 * of the controller). When this appears that this might be the case, this
563 * function returns true. It also clears the questionability flag and prevents
564 * it from being set again for some time. */
566 rconn_is_connectivity_questionable(struct rconn *rconn)
568 bool questionable = rconn->questionable_connectivity;
569 rconn->questionable_connectivity = false;
573 /* Returns the total number of packets successfully received by the underlying
576 rconn_packets_received(const struct rconn *rc)
578 return rc->packets_received;
581 /* Returns a string representing the internal state of 'rc'. The caller must
582 * not modify or free the string. */
584 rconn_get_state(const struct rconn *rc)
586 return state_name(rc->state);
589 /* Returns the number of connection attempts made by 'rc', including any
590 * ongoing attempt that has not yet succeeded or failed. */
592 rconn_get_attempted_connections(const struct rconn *rc)
594 return rc->n_attempted_connections;
597 /* Returns the number of successful connection attempts made by 'rc'. */
599 rconn_get_successful_connections(const struct rconn *rc)
601 return rc->n_successful_connections;
604 /* Returns the time at which the last successful connection was made by
607 rconn_get_last_connection(const struct rconn *rc)
609 return rc->last_connected;
612 /* Returns the time at which 'rc' was created. */
614 rconn_get_creation_time(const struct rconn *rc)
616 return rc->creation_time;
619 /* Returns the approximate number of seconds that 'rc' has been connected. */
621 rconn_get_total_time_connected(const struct rconn *rc)
623 return (rc->total_time_connected
624 + (rconn_is_connected(rc) ? elapsed_in_this_state(rc) : 0));
627 /* Returns the current amount of backoff, in seconds. This is the amount of
628 * time after which the rconn will transition from BACKOFF to CONNECTING. */
630 rconn_get_backoff(const struct rconn *rc)
635 /* Returns the number of seconds spent in this state so far. */
637 rconn_get_state_elapsed(const struct rconn *rc)
639 return elapsed_in_this_state(rc);
642 /* Tries to send a packet from 'rc''s send buffer. Returns 0 if successful,
643 * otherwise a positive errno value. */
645 try_send(struct rconn *rc)
648 struct buffer *next = rc->txq.head->next;
649 int *n_queued = rc->txq.head->private;
650 retval = vconn_send(rc->vconn, rc->txq.head);
652 if (retval != EAGAIN) {
653 disconnect(rc, retval);
661 queue_advance_head(&rc->txq, next);
665 /* Disconnects 'rc'. 'error' is used only for logging purposes. If it is
666 * nonzero, then it should be EOF to indicate the connection was closed by the
667 * peer in a normal fashion or a positive errno value. */
669 disconnect(struct rconn *rc, int error)
672 time_t now = time_now();
674 if (rc->state & (S_CONNECTING | S_ACTIVE | S_IDLE)) {
676 VLOG_WARN("%s: connection dropped (%s)",
677 rc->name, strerror(error));
678 } else if (error == EOF) {
680 VLOG_WARN("%s: connection closed", rc->name);
683 VLOG_WARN("%s: connection dropped", rc->name);
685 vconn_close(rc->vconn);
690 if (now >= rc->backoff_deadline) {
693 rc->backoff = MIN(rc->max_backoff, MAX(1, 2 * rc->backoff));
694 VLOG_WARN("%s: waiting %d seconds before reconnect\n",
695 rc->name, rc->backoff);
697 rc->backoff_deadline = now + rc->backoff;
698 state_transition(rc, S_BACKOFF);
699 if (now - rc->last_connected > 60) {
700 question_connectivity(rc);
703 rconn_disconnect(rc);
707 /* Drops all the packets from 'rc''s send queue and decrements their queue
710 flush_queue(struct rconn *rc)
715 while (rc->txq.n > 0) {
716 struct buffer *b = queue_pop_head(&rc->txq);
717 int *n_queued = b->private;
723 poll_immediate_wake();
727 elapsed_in_this_state(const struct rconn *rc)
729 return time_now() - rc->state_entered;
733 timeout(const struct rconn *rc)
736 #define STATE(NAME, VALUE) case S_##NAME: return timeout_##NAME(rc);
745 timed_out(const struct rconn *rc)
747 return time_now() >= sat_add(rc->state_entered, timeout(rc));
751 state_transition(struct rconn *rc, enum state state)
753 if (rconn_is_connected(rc)) {
754 rc->total_time_connected += elapsed_in_this_state(rc);
756 VLOG_DBG("%s: entering %s", rc->name, state_name(state));
758 rc->state_entered = time_now();
762 question_connectivity(struct rconn *rc)
764 time_t now = time_now();
765 if (now - rc->last_questioned > 60) {
766 rc->questionable_connectivity = true;
767 rc->last_questioned = now;