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. */
80 unsigned int min_timeout;
91 time_t backoff_deadline;
93 time_t last_connected;
95 unsigned int packets_sent;
97 /* If we can't connect to the peer, it could be for any number of reasons.
98 * Usually, one would assume it is because the peer is not running or
99 * because the network is partitioned. But it could also be because the
100 * network topology has changed, in which case the upper layer will need to
101 * reassess it (in particular, obtain a new IP address via DHCP and find
102 * the new location of the controller). We set this flag when we suspect
103 * that this could be the case. */
104 bool questionable_connectivity;
105 time_t last_questioned;
107 /* Throughout this file, "probe" is shorthand for "inactivity probe".
108 * When nothing has been received from the peer for a while, we send out
109 * an echo request as an inactivity probe packet. We should receive back
111 int probe_interval; /* Secs of inactivity before sending probe. */
114 static unsigned int sat_add(unsigned int x, unsigned int y);
115 static unsigned int sat_mul(unsigned int x, unsigned int y);
116 static unsigned int elapsed_in_this_state(const struct rconn *);
117 static bool timeout(struct rconn *, unsigned int secs);
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(0);
168 rc->name = xstrdup("void");
169 rc->reliable = false;
171 queue_init(&rc->txq);
172 assert(txq_limit > 0);
173 rc->txq_limit = txq_limit;
176 rc->max_backoff = max_backoff ? max_backoff : 60;
177 rc->backoff_deadline = TIME_MIN;
178 rc->last_received = time(0);
179 rc->last_connected = time(0);
181 rc->packets_sent = 0;
183 rc->questionable_connectivity = false;
184 rc->last_questioned = time(0);
186 rc->probe_interval = probe_interval ? MAX(5, probe_interval) : 0;
192 rconn_connect(struct rconn *rc, const char *name)
194 rconn_disconnect(rc);
196 rc->name = xstrdup(name);
198 return reconnect(rc);
202 rconn_connect_unreliably(struct rconn *rc,
203 const char *name, struct vconn *vconn)
205 assert(vconn != NULL);
206 rconn_disconnect(rc);
208 rc->name = xstrdup(name);
209 rc->reliable = false;
211 rc->last_connected = time(0);
212 state_transition(rc, S_ACTIVE);
216 rconn_disconnect(struct rconn *rc)
219 vconn_close(rc->vconn);
223 rc->name = xstrdup("void");
224 rc->reliable = false;
227 rc->backoff_deadline = TIME_MIN;
229 state_transition(rc, S_VOID);
232 /* Disconnects 'rc' and frees the underlying storage. */
234 rconn_destroy(struct rconn *rc)
238 vconn_close(rc->vconn);
239 queue_destroy(&rc->txq);
245 run_VOID(struct rconn *rc)
251 reconnect(struct rconn *rc)
255 VLOG_WARN("%s: connecting...", rc->name);
256 retval = vconn_open(rc->name, &rc->vconn);
258 rc->backoff_deadline = time(0) + rc->backoff;
259 state_transition(rc, S_CONNECTING);
261 VLOG_WARN("%s: connection failed (%s)", rc->name, strerror(retval));
268 run_BACKOFF(struct rconn *rc)
270 if (timeout(rc, rc->backoff)) {
276 run_CONNECTING(struct rconn *rc)
278 int retval = vconn_connect(rc->vconn);
280 VLOG_WARN("%s: connected", rc->name);
281 if (vconn_is_passive(rc->vconn)) {
282 error(0, "%s: passive vconn not supported", rc->name);
283 state_transition(rc, S_VOID);
285 state_transition(rc, S_ACTIVE);
286 rc->last_connected = rc->state_entered;
288 } else if (retval != EAGAIN) {
289 VLOG_WARN("%s: connection failed (%s)", rc->name, strerror(retval));
290 disconnect(rc, retval);
291 } else if (timeout(rc, MAX(1, rc->backoff))) {
292 VLOG_WARN("%s: connection timed out", rc->name);
293 rc->backoff_deadline = TIME_MAX; /* Prevent resetting backoff. */
299 do_tx_work(struct rconn *rc)
301 while (rc->txq.n > 0) {
302 int error = try_send(rc);
310 run_ACTIVE(struct rconn *rc)
312 if (rc->probe_interval) {
313 unsigned int base = MAX(rc->last_received, rc->state_entered);
314 unsigned int arg = base + rc->probe_interval - rc->state_entered;
315 if (timeout(rc, arg)) {
316 queue_push_tail(&rc->txq, make_echo_request());
317 VLOG_DBG("%s: idle %u seconds, sending inactivity probe",
318 rc->name, (unsigned int) (time(0) - base));
319 state_transition(rc, S_IDLE);
328 run_IDLE(struct rconn *rc)
330 if (timeout(rc, rc->probe_interval)) {
331 question_connectivity(rc);
332 VLOG_ERR("%s: no response to inactivity probe after %u "
333 "seconds, disconnecting",
334 rc->name, elapsed_in_this_state(rc));
341 /* Performs whatever activities are necessary to maintain 'rc': if 'rc' is
342 * disconnected, attempts to (re)connect, backing off as necessary; if 'rc' is
343 * connected, attempts to send packets in the send queue, if any. */
345 rconn_run(struct rconn *rc)
349 old_state = rc->state;
350 rc->min_timeout = UINT_MAX;
352 #define STATE(NAME, VALUE) case S_##NAME: run_##NAME(rc); break;
358 } while (rc->state != old_state);
361 /* Causes the next call to poll_block() to wake up when rconn_run() should be
364 rconn_run_wait(struct rconn *rc)
366 if (rc->min_timeout != UINT_MAX) {
367 poll_timer_wait(sat_mul(rc->min_timeout, 1000));
369 /* Reset timeout to 1 second. This will have no effect ordinarily, because
370 * rconn_run() will typically set it back to a higher value. If, however,
371 * the caller fails to call rconn_run() before its next call to
372 * rconn_wait() we won't potentially block forever. */
375 if ((rc->state & (S_ACTIVE | S_IDLE)) && rc->txq.n) {
376 vconn_wait(rc->vconn, WAIT_SEND);
380 /* Attempts to receive a packet from 'rc'. If successful, returns the packet;
381 * otherwise, returns a null pointer. The caller is responsible for freeing
382 * the packet (with buffer_delete()). */
384 rconn_recv(struct rconn *rc)
386 if (rc->state & (S_ACTIVE | S_IDLE)) {
387 struct buffer *buffer;
388 int error = vconn_recv(rc->vconn, &buffer);
390 rc->last_received = time(0);
391 if (rc->state == S_IDLE) {
392 state_transition(rc, S_ACTIVE);
395 } else if (error != EAGAIN) {
396 disconnect(rc, error);
402 /* Causes the next call to poll_block() to wake up when a packet may be ready
403 * to be received by vconn_recv() on 'rc'. */
405 rconn_recv_wait(struct rconn *rc)
408 vconn_wait(rc->vconn, WAIT_RECV);
412 /* Sends 'b' on 'rc'. Returns 0 if successful, EAGAIN if at least 'txq_limit'
413 * packets are already queued, otherwise a positive errno value. */
415 do_send(struct rconn *rc, struct buffer *b, int txq_limit)
418 if (rc->txq.n < txq_limit) {
419 queue_push_tail(&rc->txq, b);
420 if (rc->txq.n == 1) {
432 /* Sends 'b' on 'rc'. Returns 0 if successful, EAGAIN if the send queue is
433 * full, or ENOTCONN if 'rc' is not currently connected.
435 * There is no rconn_send_wait() function: an rconn has a send queue that it
436 * takes care of sending if you call rconn_run(), which will have the side
437 * effect of waking up poll_block(). */
439 rconn_send(struct rconn *rc, struct buffer *b)
441 return do_send(rc, b, rc->txq_limit);
444 /* Sends 'b' on 'rc'. Returns 0 if successful, EAGAIN if the send queue is
445 * full, otherwise a positive errno value.
447 * Compared to rconn_send(), this function relaxes the queue limit, allowing
448 * more packets than usual to be queued. */
450 rconn_force_send(struct rconn *rc, struct buffer *b)
452 return do_send(rc, b, 2 * rc->txq_limit);
455 /* Returns true if 'rc''s send buffer is full,
456 * false if it has room for at least one more packet. */
458 rconn_is_full(const struct rconn *rc)
460 return rc->txq.n >= rc->txq_limit;
463 /* Returns the total number of packets successfully sent on the underlying
464 * vconn. A packet is not counted as sent while it is still queued in the
465 * rconn, only when it has been successfuly passed to the vconn. */
467 rconn_packets_sent(const struct rconn *rc)
469 return rc->packets_sent;
472 /* Returns 'rc''s name (the 'name' argument passed to rconn_new()). */
474 rconn_get_name(const struct rconn *rc)
479 /* Returns true if 'rconn' is connected or in the process of reconnecting,
480 * false if 'rconn' is disconnected and will not reconnect on its own. */
482 rconn_is_alive(const struct rconn *rconn)
484 return rconn->state != S_VOID;
487 /* Returns true if 'rconn' is connected, false otherwise. */
489 rconn_is_connected(const struct rconn *rconn)
491 return rconn->state & (S_ACTIVE | S_IDLE);
494 /* Returns 0 if 'rconn' is connected, otherwise the number of seconds that it
495 * has been disconnected. */
497 rconn_disconnected_duration(const struct rconn *rconn)
499 return rconn_is_connected(rconn) ? 0 : time(0) - rconn->last_connected;
502 /* Returns the IP address of the peer, or 0 if the peer is not connected over
503 * an IP-based protocol or if its IP address is not known. */
505 rconn_get_ip(const struct rconn *rconn)
507 return rconn->vconn ? vconn_get_ip(rconn->vconn) : 0;
510 /* If 'rconn' can't connect to the peer, it could be for any number of reasons.
511 * Usually, one would assume it is because the peer is not running or because
512 * the network is partitioned. But it could also be because the network
513 * topology has changed, in which case the upper layer will need to reassess it
514 * (in particular, obtain a new IP address via DHCP and find the new location
515 * of the controller). When this appears that this might be the case, this
516 * function returns true. It also clears the questionability flag and prevents
517 * it from being set again for some time. */
519 rconn_is_connectivity_questionable(struct rconn *rconn)
521 bool questionable = rconn->questionable_connectivity;
522 rconn->questionable_connectivity = false;
526 /* Tries to send a packet from 'rc''s send buffer. Returns 0 if successful,
527 * otherwise a positive errno value. */
529 try_send(struct rconn *rc)
532 struct buffer *next = rc->txq.head->next;
533 retval = vconn_send(rc->vconn, rc->txq.head);
535 if (retval != EAGAIN) {
536 disconnect(rc, retval);
541 queue_advance_head(&rc->txq, next);
545 /* Disconnects 'rc'. 'error' is used only for logging purposes. If it is
546 * nonzero, then it should be EOF to indicate the connection was closed by the
547 * peer in a normal fashion or a positive errno value. */
549 disconnect(struct rconn *rc, int error)
552 time_t now = time(0);
554 if (rc->state & (S_CONNECTING | S_ACTIVE | S_IDLE)) {
556 VLOG_WARN("%s: connection dropped (%s)",
557 rc->name, strerror(error));
558 } else if (error == EOF) {
560 VLOG_WARN("%s: connection closed", rc->name);
563 VLOG_WARN("%s: connection dropped", rc->name);
565 vconn_close(rc->vconn);
567 queue_clear(&rc->txq);
570 if (now >= rc->backoff_deadline) {
573 rc->backoff = MIN(rc->max_backoff, MAX(1, 2 * rc->backoff));
574 VLOG_WARN("%s: waiting %d seconds before reconnect\n",
575 rc->name, rc->backoff);
577 rc->backoff_deadline = now + rc->backoff;
578 state_transition(rc, S_BACKOFF);
579 if (now - rc->last_connected > 60) {
580 question_connectivity(rc);
583 rconn_disconnect(rc);
588 elapsed_in_this_state(const struct rconn *rc)
590 return time(0) - rc->state_entered;
594 timeout(struct rconn *rc, unsigned int secs)
596 rc->min_timeout = MIN(rc->min_timeout, secs);
597 return time(0) >= sat_add(rc->state_entered, secs);
601 state_transition(struct rconn *rc, enum state state)
603 VLOG_DBG("%s: entering %s", rc->name, state_name(state));
605 rc->state_entered = time(0);
609 sat_add(unsigned int x, unsigned int y)
611 return x + y >= x ? x + y : UINT_MAX;
615 sat_mul(unsigned int x, unsigned int y)
618 return x <= UINT_MAX / y ? x * y : UINT_MAX;
622 question_connectivity(struct rconn *rc)
624 time_t now = time(0);
625 if (now - rc->last_questioned > 60) {
626 rc->questionable_connectivity = true;
627 rc->last_questioned = now;