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"
48 #define THIS_MODULE VLM_rconn
53 STATE(BACKOFF, 1 << 1) \
54 STATE(CONNECTING, 1 << 2) \
55 STATE(ACTIVE, 1 << 3) \
58 #define STATE(NAME, VALUE) S_##NAME = VALUE,
64 state_name(enum state state)
67 #define STATE(NAME, VALUE) case S_##NAME: return #NAME;
74 /* A reliable connection to an OpenFlow switch or controller.
76 * See the large comment in rconn.h for more information. */
90 time_t backoff_deadline;
92 time_t last_connected;
94 unsigned int packets_sent;
96 /* If we can't connect to the peer, it could be for any number of reasons.
97 * Usually, one would assume it is because the peer is not running or
98 * because the network is partitioned. But it could also be because the
99 * network topology has changed, in which case the upper layer will need to
100 * reassess it (in particular, obtain a new IP address via DHCP and find
101 * the new location of the controller). We set this flag when we suspect
102 * that this could be the case. */
103 bool questionable_connectivity;
104 time_t last_questioned;
106 /* Throughout this file, "probe" is shorthand for "inactivity probe".
107 * When nothing has been received from the peer for a while, we send out
108 * an echo request as an inactivity probe packet. We should receive back
110 int probe_interval; /* Secs of inactivity before sending probe. */
113 static unsigned int sat_add(unsigned int x, unsigned int y);
114 static unsigned int sat_mul(unsigned int x, unsigned int y);
115 static unsigned int elapsed_in_this_state(const struct rconn *);
116 static unsigned int timeout(const struct rconn *);
117 static bool timed_out(const struct rconn *);
118 static void state_transition(struct rconn *, enum state);
119 static int try_send(struct rconn *);
120 static int reconnect(struct rconn *);
121 static void disconnect(struct rconn *, int error);
122 static void question_connectivity(struct rconn *);
124 /* Creates a new rconn, connects it (reliably) to 'name', and returns it. */
126 rconn_new(const char *name, int txq_limit, int inactivity_probe_interval,
129 struct rconn *rc = rconn_create(txq_limit, inactivity_probe_interval,
131 rconn_connect(rc, name);
135 /* Creates a new rconn, connects it (unreliably) to 'vconn', and returns it. */
137 rconn_new_from_vconn(const char *name, int txq_limit, struct vconn *vconn)
139 struct rconn *rc = rconn_create(txq_limit, 60, 0);
140 rconn_connect_unreliably(rc, name, vconn);
144 /* Creates and returns a new rconn.
146 * 'txq_limit' is the maximum length of the send queue, in packets.
148 * 'probe_interval' is a number of seconds. If the interval passes once
149 * without an OpenFlow message being received from the peer, the rconn sends
150 * out an "echo request" message. If the interval passes again without a
151 * message being received, the rconn disconnects and re-connects to the peer.
152 * Setting 'probe_interval' to 0 disables this behavior.
154 * 'max_backoff' is the maximum number of seconds between attempts to connect
155 * to the peer. The actual interval starts at 1 second and doubles on each
156 * failure until it reaches 'max_backoff'. If 0 is specified, the default of
157 * 60 seconds is used. */
159 rconn_create(int txq_limit, int probe_interval, int max_backoff)
161 struct rconn *rc = xcalloc(1, sizeof *rc);
164 rc->state_entered = time_now();
167 rc->name = xstrdup("void");
168 rc->reliable = false;
170 queue_init(&rc->txq);
171 assert(txq_limit > 0);
172 rc->txq_limit = txq_limit;
175 rc->max_backoff = max_backoff ? max_backoff : 60;
176 rc->backoff_deadline = TIME_MIN;
177 rc->last_received = time_now();
178 rc->last_connected = time_now();
180 rc->packets_sent = 0;
182 rc->questionable_connectivity = false;
183 rc->last_questioned = time_now();
185 rc->probe_interval = probe_interval ? MAX(5, probe_interval) : 0;
191 rconn_connect(struct rconn *rc, const char *name)
193 rconn_disconnect(rc);
195 rc->name = xstrdup(name);
197 return reconnect(rc);
201 rconn_connect_unreliably(struct rconn *rc,
202 const char *name, struct vconn *vconn)
204 assert(vconn != NULL);
205 rconn_disconnect(rc);
207 rc->name = xstrdup(name);
208 rc->reliable = false;
210 rc->last_connected = time_now();
211 state_transition(rc, S_ACTIVE);
215 rconn_disconnect(struct rconn *rc)
218 vconn_close(rc->vconn);
222 rc->name = xstrdup("void");
223 rc->reliable = false;
226 rc->backoff_deadline = TIME_MIN;
228 state_transition(rc, S_VOID);
231 /* Disconnects 'rc' and frees the underlying storage. */
233 rconn_destroy(struct rconn *rc)
237 vconn_close(rc->vconn);
238 queue_destroy(&rc->txq);
244 timeout_VOID(const struct rconn *rc)
250 run_VOID(struct rconn *rc)
256 reconnect(struct rconn *rc)
260 VLOG_WARN("%s: connecting...", rc->name);
261 retval = vconn_open(rc->name, &rc->vconn);
263 rc->backoff_deadline = time_now() + rc->backoff;
264 state_transition(rc, S_CONNECTING);
266 VLOG_WARN("%s: connection failed (%s)", rc->name, strerror(retval));
273 timeout_BACKOFF(const struct rconn *rc)
279 run_BACKOFF(struct rconn *rc)
287 timeout_CONNECTING(const struct rconn *rc)
289 return MAX(1, rc->backoff);
293 run_CONNECTING(struct rconn *rc)
295 int retval = vconn_connect(rc->vconn);
297 VLOG_WARN("%s: connected", rc->name);
298 if (vconn_is_passive(rc->vconn)) {
299 error(0, "%s: passive vconn not supported", rc->name);
300 state_transition(rc, S_VOID);
302 state_transition(rc, S_ACTIVE);
303 rc->last_connected = rc->state_entered;
305 } else if (retval != EAGAIN) {
306 VLOG_WARN("%s: connection failed (%s)", rc->name, strerror(retval));
307 disconnect(rc, retval);
308 } else if (timed_out(rc)) {
309 VLOG_WARN("%s: connection timed out", rc->name);
310 rc->backoff_deadline = TIME_MAX; /* Prevent resetting backoff. */
316 do_tx_work(struct rconn *rc)
318 while (rc->txq.n > 0) {
319 int error = try_send(rc);
327 timeout_ACTIVE(const struct rconn *rc)
329 if (rc->probe_interval) {
330 unsigned int base = MAX(rc->last_received, rc->state_entered);
331 unsigned int arg = base + rc->probe_interval - rc->state_entered;
338 run_ACTIVE(struct rconn *rc)
341 unsigned int base = MAX(rc->last_received, rc->state_entered);
342 queue_push_tail(&rc->txq, make_echo_request());
343 VLOG_DBG("%s: idle %u seconds, sending inactivity probe",
344 rc->name, (unsigned int) (time_now() - base));
345 state_transition(rc, S_IDLE);
353 timeout_IDLE(const struct rconn *rc)
355 return rc->probe_interval;
359 run_IDLE(struct rconn *rc)
362 question_connectivity(rc);
363 VLOG_ERR("%s: no response to inactivity probe after %u "
364 "seconds, disconnecting",
365 rc->name, elapsed_in_this_state(rc));
372 /* Performs whatever activities are necessary to maintain 'rc': if 'rc' is
373 * disconnected, attempts to (re)connect, backing off as necessary; if 'rc' is
374 * connected, attempts to send packets in the send queue, if any. */
376 rconn_run(struct rconn *rc)
380 old_state = rc->state;
382 #define STATE(NAME, VALUE) case S_##NAME: run_##NAME(rc); break;
388 } while (rc->state != old_state);
391 /* Causes the next call to poll_block() to wake up when rconn_run() should be
394 rconn_run_wait(struct rconn *rc)
396 unsigned int timeo = timeout(rc);
397 if (timeo != UINT_MAX) {
398 poll_timer_wait(sat_mul(timeo, 1000));
401 if ((rc->state & (S_ACTIVE | S_IDLE)) && rc->txq.n) {
402 vconn_wait(rc->vconn, WAIT_SEND);
406 /* Attempts to receive a packet from 'rc'. If successful, returns the packet;
407 * otherwise, returns a null pointer. The caller is responsible for freeing
408 * the packet (with buffer_delete()). */
410 rconn_recv(struct rconn *rc)
412 if (rc->state & (S_ACTIVE | S_IDLE)) {
413 struct buffer *buffer;
414 int error = vconn_recv(rc->vconn, &buffer);
416 rc->last_received = time_now();
417 if (rc->state == S_IDLE) {
418 state_transition(rc, S_ACTIVE);
421 } else if (error != EAGAIN) {
422 disconnect(rc, error);
428 /* Causes the next call to poll_block() to wake up when a packet may be ready
429 * to be received by vconn_recv() on 'rc'. */
431 rconn_recv_wait(struct rconn *rc)
434 vconn_wait(rc->vconn, WAIT_RECV);
438 /* Sends 'b' on 'rc'. Returns 0 if successful, EAGAIN if at least 'txq_limit'
439 * packets are already queued, otherwise a positive errno value. */
441 do_send(struct rconn *rc, struct buffer *b, int txq_limit)
444 if (rc->txq.n < txq_limit) {
445 queue_push_tail(&rc->txq, b);
446 if (rc->txq.n == 1) {
458 /* Sends 'b' on 'rc'. Returns 0 if successful, EAGAIN if the send queue is
459 * full, or ENOTCONN if 'rc' is not currently connected.
461 * There is no rconn_send_wait() function: an rconn has a send queue that it
462 * takes care of sending if you call rconn_run(), which will have the side
463 * effect of waking up poll_block(). */
465 rconn_send(struct rconn *rc, struct buffer *b)
467 return do_send(rc, b, rc->txq_limit);
470 /* Sends 'b' on 'rc'. Returns 0 if successful, EAGAIN if the send queue is
471 * full, otherwise a positive errno value.
473 * Compared to rconn_send(), this function relaxes the queue limit, allowing
474 * more packets than usual to be queued. */
476 rconn_force_send(struct rconn *rc, struct buffer *b)
478 return do_send(rc, b, 2 * rc->txq_limit);
481 /* Returns true if 'rc''s send buffer is full,
482 * false if it has room for at least one more packet. */
484 rconn_is_full(const struct rconn *rc)
486 return rc->txq.n >= rc->txq_limit;
489 /* Returns the total number of packets successfully sent on the underlying
490 * vconn. A packet is not counted as sent while it is still queued in the
491 * rconn, only when it has been successfuly passed to the vconn. */
493 rconn_packets_sent(const struct rconn *rc)
495 return rc->packets_sent;
498 /* Returns 'rc''s name (the 'name' argument passed to rconn_new()). */
500 rconn_get_name(const struct rconn *rc)
505 /* Returns true if 'rconn' is connected or in the process of reconnecting,
506 * false if 'rconn' is disconnected and will not reconnect on its own. */
508 rconn_is_alive(const struct rconn *rconn)
510 return rconn->state != S_VOID;
513 /* Returns true if 'rconn' is connected, false otherwise. */
515 rconn_is_connected(const struct rconn *rconn)
517 return rconn->state & (S_ACTIVE | S_IDLE);
520 /* Returns 0 if 'rconn' is connected, otherwise the number of seconds that it
521 * has been disconnected. */
523 rconn_disconnected_duration(const struct rconn *rconn)
525 return rconn_is_connected(rconn) ? 0 : time_now() - rconn->last_received;
528 /* Returns the IP address of the peer, or 0 if the peer is not connected over
529 * an IP-based protocol or if its IP address is not known. */
531 rconn_get_ip(const struct rconn *rconn)
533 return rconn->vconn ? vconn_get_ip(rconn->vconn) : 0;
536 /* If 'rconn' can't connect to the peer, it could be for any number of reasons.
537 * Usually, one would assume it is because the peer is not running or because
538 * the network is partitioned. But it could also be because the network
539 * topology has changed, in which case the upper layer will need to reassess it
540 * (in particular, obtain a new IP address via DHCP and find the new location
541 * of the controller). When this appears that this might be the case, this
542 * function returns true. It also clears the questionability flag and prevents
543 * it from being set again for some time. */
545 rconn_is_connectivity_questionable(struct rconn *rconn)
547 bool questionable = rconn->questionable_connectivity;
548 rconn->questionable_connectivity = false;
552 /* Tries to send a packet from 'rc''s send buffer. Returns 0 if successful,
553 * otherwise a positive errno value. */
555 try_send(struct rconn *rc)
558 struct buffer *next = rc->txq.head->next;
559 retval = vconn_send(rc->vconn, rc->txq.head);
561 if (retval != EAGAIN) {
562 disconnect(rc, retval);
567 queue_advance_head(&rc->txq, next);
571 /* Disconnects 'rc'. 'error' is used only for logging purposes. If it is
572 * nonzero, then it should be EOF to indicate the connection was closed by the
573 * peer in a normal fashion or a positive errno value. */
575 disconnect(struct rconn *rc, int error)
578 time_t now = time_now();
580 if (rc->state & (S_CONNECTING | S_ACTIVE | S_IDLE)) {
582 VLOG_WARN("%s: connection dropped (%s)",
583 rc->name, strerror(error));
584 } else if (error == EOF) {
586 VLOG_WARN("%s: connection closed", rc->name);
589 VLOG_WARN("%s: connection dropped", rc->name);
591 vconn_close(rc->vconn);
593 queue_clear(&rc->txq);
596 if (now >= rc->backoff_deadline) {
599 rc->backoff = MIN(rc->max_backoff, MAX(1, 2 * rc->backoff));
600 VLOG_WARN("%s: waiting %d seconds before reconnect\n",
601 rc->name, rc->backoff);
603 rc->backoff_deadline = now + rc->backoff;
604 state_transition(rc, S_BACKOFF);
605 if (now - rc->last_connected > 60) {
606 question_connectivity(rc);
609 rconn_disconnect(rc);
614 elapsed_in_this_state(const struct rconn *rc)
616 return time_now() - rc->state_entered;
620 timeout(const struct rconn *rc)
623 #define STATE(NAME, VALUE) case S_##NAME: return timeout_##NAME(rc);
632 timed_out(const struct rconn *rc)
634 return time_now() >= sat_add(rc->state_entered, timeout(rc));
638 state_transition(struct rconn *rc, enum state state)
640 VLOG_DBG("%s: entering %s", rc->name, state_name(state));
642 rc->state_entered = time_now();
646 sat_add(unsigned int x, unsigned int y)
648 return x + y >= x ? x + y : UINT_MAX;
652 sat_mul(unsigned int x, unsigned int y)
655 return x <= UINT_MAX / y ? x * y : UINT_MAX;
659 question_connectivity(struct rconn *rc)
661 time_t now = time_now();
662 if (now - rc->last_questioned > 60) {
663 rc->questionable_connectivity = true;
664 rc->last_questioned = now;