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.
41 #include "poll-loop.h"
42 #include "ofp-print.h"
47 #define THIS_MODULE VLM_rconn
52 STATE(BACKOFF, 1 << 1) \
53 STATE(CONNECTING, 1 << 2) \
54 STATE(ACTIVE, 1 << 3) \
57 #define STATE(NAME, VALUE) S_##NAME = VALUE,
63 state_name(enum state state)
66 #define STATE(NAME, VALUE) case S_##NAME: return #NAME;
73 /* A reliable connection to an OpenFlow switch or controller.
75 * See the large comment in rconn.h for more information. */
79 unsigned int min_timeout;
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 bool timeout(struct rconn *, unsigned int secs);
117 static void state_transition(struct rconn *, enum state);
118 static int try_send(struct rconn *);
119 static void reconnect(struct rconn *);
120 static void disconnect(struct rconn *, int error);
121 static void question_connectivity(struct rconn *);
123 /* Creates a new rconn, connects it (reliably) to 'name', and returns it. */
125 rconn_new(const char *name, int txq_limit, int inactivity_probe_interval,
128 struct rconn *rc = rconn_create(txq_limit, inactivity_probe_interval,
130 rconn_connect(rc, name);
134 /* Creates a new rconn, connects it (unreliably) to 'vconn', and returns it. */
136 rconn_new_from_vconn(const char *name, int txq_limit, struct vconn *vconn)
138 struct rconn *rc = rconn_create(txq_limit, 60, 0);
139 rconn_connect_unreliably(rc, name, vconn);
143 /* Creates and returns a new rconn.
145 * 'txq_limit' is the maximum length of the send queue, in packets.
147 * 'probe_interval' is a number of seconds. If the interval passes once
148 * without an OpenFlow message being received from the peer, the rconn sends
149 * out an "echo request" message. If the interval passes again without a
150 * message being received, the rconn disconnects and re-connects to the peer.
151 * Setting 'probe_interval' to 0 disables this behavior.
153 * 'max_backoff' is the maximum number of seconds between attempts to connect
154 * to the peer. The actual interval starts at 1 second and doubles on each
155 * failure until it reaches 'max_backoff'. If 0 is specified, the default of
156 * 60 seconds is used. */
158 rconn_create(int txq_limit, int probe_interval, int max_backoff)
160 struct rconn *rc = xcalloc(1, sizeof *rc);
163 rc->state_entered = time(0);
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(0);
178 rc->last_connected = time(0);
180 rc->packets_sent = 0;
182 rc->questionable_connectivity = false;
183 rc->last_questioned = time(0);
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);
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(0);
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 run_VOID(struct rconn *rc)
250 reconnect(struct rconn *rc)
254 VLOG_WARN("%s: connecting...", rc->name);
255 retval = vconn_open(rc->name, &rc->vconn);
257 rc->backoff_deadline = time(0) + rc->backoff;
258 state_transition(rc, S_CONNECTING);
260 VLOG_WARN("%s: connection failed (%s)", rc->name, strerror(retval));
266 run_BACKOFF(struct rconn *rc)
268 if (timeout(rc, rc->backoff)) {
274 run_CONNECTING(struct rconn *rc)
276 int error = vconn_connect(rc->vconn);
278 VLOG_WARN("%s: connected", rc->name);
279 if (vconn_is_passive(rc->vconn)) {
280 fatal(0, "%s: passive vconn not supported in switch",
283 state_transition(rc, S_ACTIVE);
284 rc->last_connected = rc->state_entered;
285 } else if (error != EAGAIN) {
286 VLOG_WARN("%s: connection failed (%s)", rc->name, strerror(error));
287 disconnect(rc, error);
288 } else if (timeout(rc, MAX(1, rc->backoff))) {
289 VLOG_WARN("%s: connection timed out", rc->name);
290 rc->backoff_deadline = TIME_MAX; /* Prevent resetting backoff. */
296 do_tx_work(struct rconn *rc)
298 while (rc->txq.n > 0) {
299 int error = try_send(rc);
307 run_ACTIVE(struct rconn *rc)
309 if (rc->probe_interval) {
310 unsigned int base = MAX(rc->last_received, rc->state_entered);
311 unsigned int arg = base + rc->probe_interval - rc->state_entered;
312 if (timeout(rc, arg)) {
313 queue_push_tail(&rc->txq, make_echo_request());
314 VLOG_DBG("%s: idle %u seconds, sending inactivity probe",
315 rc->name, (unsigned int) (time(0) - base));
316 state_transition(rc, S_IDLE);
325 run_IDLE(struct rconn *rc)
327 if (timeout(rc, rc->probe_interval)) {
328 question_connectivity(rc);
329 VLOG_ERR("%s: no response to inactivity probe after %u "
330 "seconds, disconnecting",
331 rc->name, elapsed_in_this_state(rc));
338 /* Performs whatever activities are necessary to maintain 'rc': if 'rc' is
339 * disconnected, attempts to (re)connect, backing off as necessary; if 'rc' is
340 * connected, attempts to send packets in the send queue, if any. */
342 rconn_run(struct rconn *rc)
346 old_state = rc->state;
347 rc->min_timeout = UINT_MAX;
349 #define STATE(NAME, VALUE) case S_##NAME: run_##NAME(rc); break;
355 } while (rc->state != old_state);
358 /* Causes the next call to poll_block() to wake up when rconn_run() should be
361 rconn_run_wait(struct rconn *rc)
363 if (rc->min_timeout != UINT_MAX) {
364 poll_timer_wait(sat_mul(rc->min_timeout, 1000));
366 /* Reset timeout to 1 second. This will have no effect ordinarily, because
367 * rconn_run() will typically set it back to a higher value. If, however,
368 * the caller fails to call rconn_run() before its next call to
369 * rconn_wait() we won't potentially block forever. */
372 if ((rc->state & (S_ACTIVE | S_IDLE)) && rc->txq.n) {
373 vconn_wait(rc->vconn, WAIT_SEND);
377 /* Attempts to receive a packet from 'rc'. If successful, returns the packet;
378 * otherwise, returns a null pointer. The caller is responsible for freeing
379 * the packet (with buffer_delete()). */
381 rconn_recv(struct rconn *rc)
383 if (rc->state & (S_ACTIVE | S_IDLE)) {
384 struct buffer *buffer;
385 int error = vconn_recv(rc->vconn, &buffer);
387 rc->last_received = time(0);
388 if (rc->state == S_IDLE) {
389 state_transition(rc, S_ACTIVE);
392 } else if (error != EAGAIN) {
393 disconnect(rc, error);
399 /* Causes the next call to poll_block() to wake up when a packet may be ready
400 * to be received by vconn_recv() on 'rc'. */
402 rconn_recv_wait(struct rconn *rc)
405 vconn_wait(rc->vconn, WAIT_RECV);
409 /* Sends 'b' on 'rc'. Returns 0 if successful, EAGAIN if at least 'txq_limit'
410 * packets are already queued, otherwise a positive errno value. */
412 do_send(struct rconn *rc, struct buffer *b, int txq_limit)
415 if (rc->txq.n < txq_limit) {
416 queue_push_tail(&rc->txq, b);
417 if (rc->txq.n == 1) {
429 /* Sends 'b' on 'rc'. Returns 0 if successful, EAGAIN if the send queue is
430 * full, or ENOTCONN if 'rc' is not currently connected.
432 * There is no rconn_send_wait() function: an rconn has a send queue that it
433 * takes care of sending if you call rconn_run(), which will have the side
434 * effect of waking up poll_block(). */
436 rconn_send(struct rconn *rc, struct buffer *b)
438 return do_send(rc, b, rc->txq_limit);
441 /* Sends 'b' on 'rc'. Returns 0 if successful, EAGAIN if the send queue is
442 * full, otherwise a positive errno value.
444 * Compared to rconn_send(), this function relaxes the queue limit, allowing
445 * more packets than usual to be queued. */
447 rconn_force_send(struct rconn *rc, struct buffer *b)
449 return do_send(rc, b, 2 * rc->txq_limit);
452 /* Returns true if 'rc''s send buffer is full,
453 * false if it has room for at least one more packet. */
455 rconn_is_full(const struct rconn *rc)
457 return rc->txq.n >= rc->txq_limit;
460 /* Returns the total number of packets successfully sent on the underlying
461 * vconn. A packet is not counted as sent while it is still queued in the
462 * rconn, only when it has been successfuly passed to the vconn. */
464 rconn_packets_sent(const struct rconn *rc)
466 return rc->packets_sent;
469 /* Returns 'rc''s name (the 'name' argument passed to rconn_new()). */
471 rconn_get_name(const struct rconn *rc)
476 /* Returns true if 'rconn' is connected or in the process of reconnecting,
477 * false if 'rconn' is disconnected and will not reconnect on its own. */
479 rconn_is_alive(const struct rconn *rconn)
481 return rconn->state != S_VOID;
484 /* Returns true if 'rconn' is connected, false otherwise. */
486 rconn_is_connected(const struct rconn *rconn)
488 return rconn->state & (S_ACTIVE | S_IDLE);
491 /* Returns 0 if 'rconn' is connected, otherwise the number of seconds that it
492 * has been disconnected. */
494 rconn_disconnected_duration(const struct rconn *rconn)
496 return rconn_is_connected(rconn) ? 0 : time(0) - rconn->last_connected;
499 /* Returns the IP address of the peer, or 0 if the peer is not connected over
500 * an IP-based protocol or if its IP address is not known. */
502 rconn_get_ip(const struct rconn *rconn)
504 return rconn->vconn ? vconn_get_ip(rconn->vconn) : 0;
507 /* If 'rconn' can't connect to the peer, it could be for any number of reasons.
508 * Usually, one would assume it is because the peer is not running or because
509 * the network is partitioned. But it could also be because the network
510 * topology has changed, in which case the upper layer will need to reassess it
511 * (in particular, obtain a new IP address via DHCP and find the new location
512 * of the controller). When this appears that this might be the case, this
513 * function returns true. It also clears the questionability flag and prevents
514 * it from being set again for some time. */
516 rconn_is_connectivity_questionable(struct rconn *rconn)
518 bool questionable = rconn->questionable_connectivity;
519 rconn->questionable_connectivity = false;
523 /* Tries to send a packet from 'rc''s send buffer. Returns 0 if successful,
524 * otherwise a positive errno value. */
526 try_send(struct rconn *rc)
529 struct buffer *next = rc->txq.head->next;
530 retval = vconn_send(rc->vconn, rc->txq.head);
532 if (retval != EAGAIN) {
533 disconnect(rc, retval);
538 queue_advance_head(&rc->txq, next);
542 /* Disconnects 'rc'. 'error' is used only for logging purposes. If it is
543 * nonzero, then it should be EOF to indicate the connection was closed by the
544 * peer in a normal fashion or a positive errno value. */
546 disconnect(struct rconn *rc, int error)
549 time_t now = time(0);
551 if (rc->state & (S_CONNECTING | S_ACTIVE | S_IDLE)) {
553 VLOG_WARN("%s: connection dropped (%s)",
554 rc->name, strerror(error));
555 } else if (error == EOF) {
557 VLOG_WARN("%s: connection closed", rc->name);
560 VLOG_WARN("%s: connection dropped", rc->name);
562 vconn_close(rc->vconn);
564 queue_clear(&rc->txq);
567 if (now >= rc->backoff_deadline) {
570 rc->backoff = MIN(rc->max_backoff, MAX(1, 2 * rc->backoff));
571 VLOG_WARN("%s: waiting %d seconds before reconnect\n",
572 rc->name, rc->backoff);
574 rc->backoff_deadline = now + rc->backoff;
575 state_transition(rc, S_BACKOFF);
576 if (now - rc->last_connected > 60) {
577 question_connectivity(rc);
580 rconn_disconnect(rc);
585 elapsed_in_this_state(const struct rconn *rc)
587 return time(0) - rc->state_entered;
591 timeout(struct rconn *rc, unsigned int secs)
593 rc->min_timeout = MIN(rc->min_timeout, secs);
594 return time(0) >= sat_add(rc->state_entered, secs);
598 state_transition(struct rconn *rc, enum state state)
600 VLOG_DBG("%s: entering %s", rc->name, state_name(state));
602 rc->state_entered = time(0);
606 sat_add(unsigned int x, unsigned int y)
608 return x + y >= x ? x + y : UINT_MAX;
612 sat_mul(unsigned int x, unsigned int y)
615 return x <= UINT_MAX / y ? x * y : UINT_MAX;
619 question_connectivity(struct rconn *rc)
621 time_t now = time(0);
622 if (now - rc->last_questioned > 60) {
623 rc->questionable_connectivity = true;
624 rc->last_questioned = now;