]>
Commit | Line | Data |
---|---|---|
064af421 BP |
1 | /* |
2 | * Copyright (c) 2008, 2009 Nicira Networks. | |
3 | * | |
a14bc59f BP |
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: | |
064af421 | 7 | * |
a14bc59f BP |
8 | * http://www.apache.org/licenses/LICENSE-2.0 |
9 | * | |
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. | |
064af421 BP |
15 | */ |
16 | ||
17 | #include <config.h> | |
18 | #include "rconn.h" | |
19 | #include <assert.h> | |
20 | #include <errno.h> | |
21 | #include <limits.h> | |
22 | #include <stdlib.h> | |
23 | #include <string.h> | |
24 | #include "coverage.h" | |
25 | #include "ofpbuf.h" | |
26 | #include "openflow/openflow.h" | |
27 | #include "poll-loop.h" | |
28 | #include "sat-math.h" | |
29 | #include "timeval.h" | |
30 | #include "util.h" | |
31 | #include "vconn.h" | |
32 | ||
33 | #define THIS_MODULE VLM_rconn | |
34 | #include "vlog.h" | |
35 | ||
36 | #define STATES \ | |
37 | STATE(VOID, 1 << 0) \ | |
38 | STATE(BACKOFF, 1 << 1) \ | |
39 | STATE(CONNECTING, 1 << 2) \ | |
40 | STATE(ACTIVE, 1 << 3) \ | |
41 | STATE(IDLE, 1 << 4) | |
42 | enum state { | |
43 | #define STATE(NAME, VALUE) S_##NAME = VALUE, | |
44 | STATES | |
45 | #undef STATE | |
46 | }; | |
47 | ||
48 | static const char * | |
49 | state_name(enum state state) | |
50 | { | |
51 | switch (state) { | |
52 | #define STATE(NAME, VALUE) case S_##NAME: return #NAME; | |
53 | STATES | |
54 | #undef STATE | |
55 | } | |
56 | return "***ERROR***"; | |
57 | } | |
58 | ||
59 | /* A reliable connection to an OpenFlow switch or controller. | |
60 | * | |
61 | * See the large comment in rconn.h for more information. */ | |
62 | struct rconn { | |
63 | enum state state; | |
64 | time_t state_entered; | |
65 | ||
66 | struct vconn *vconn; | |
67 | char *name; | |
68 | bool reliable; | |
69 | ||
70 | struct ovs_queue txq; | |
71 | ||
72 | int backoff; | |
73 | int max_backoff; | |
74 | time_t backoff_deadline; | |
75 | time_t last_received; | |
76 | time_t last_connected; | |
77 | unsigned int packets_sent; | |
78 | unsigned int seqno; | |
79 | ||
80 | /* In S_ACTIVE and S_IDLE, probably_admitted reports whether we believe | |
81 | * that the peer has made a (positive) admission control decision on our | |
82 | * connection. If we have not yet been (probably) admitted, then the | |
83 | * connection does not reset the timer used for deciding whether the switch | |
84 | * should go into fail-open mode. | |
85 | * | |
86 | * last_admitted reports the last time we believe such a positive admission | |
87 | * control decision was made. */ | |
88 | bool probably_admitted; | |
89 | time_t last_admitted; | |
90 | ||
91 | /* These values are simply for statistics reporting, not used directly by | |
92 | * anything internal to the rconn (or the secchan for that matter). */ | |
93 | unsigned int packets_received; | |
94 | unsigned int n_attempted_connections, n_successful_connections; | |
95 | time_t creation_time; | |
96 | unsigned long int total_time_connected; | |
97 | ||
98 | /* If we can't connect to the peer, it could be for any number of reasons. | |
99 | * Usually, one would assume it is because the peer is not running or | |
100 | * because the network is partitioned. But it could also be because the | |
101 | * network topology has changed, in which case the upper layer will need to | |
102 | * reassess it (in particular, obtain a new IP address via DHCP and find | |
103 | * the new location of the controller). We set this flag when we suspect | |
104 | * that this could be the case. */ | |
105 | bool questionable_connectivity; | |
106 | time_t last_questioned; | |
107 | ||
108 | /* Throughout this file, "probe" is shorthand for "inactivity probe". | |
109 | * When nothing has been received from the peer for a while, we send out | |
110 | * an echo request as an inactivity probe packet. We should receive back | |
111 | * a response. */ | |
112 | int probe_interval; /* Secs of inactivity before sending probe. */ | |
113 | ||
114 | /* Messages sent or received are copied to the monitor connections. */ | |
115 | #define MAX_MONITORS 8 | |
116 | struct vconn *monitors[8]; | |
117 | size_t n_monitors; | |
118 | }; | |
119 | ||
120 | static unsigned int elapsed_in_this_state(const struct rconn *); | |
121 | static unsigned int timeout(const struct rconn *); | |
122 | static bool timed_out(const struct rconn *); | |
123 | static void state_transition(struct rconn *, enum state); | |
124 | static int try_send(struct rconn *); | |
125 | static int reconnect(struct rconn *); | |
126 | static void disconnect(struct rconn *, int error); | |
127 | static void flush_queue(struct rconn *); | |
128 | static void question_connectivity(struct rconn *); | |
129 | static void copy_to_monitor(struct rconn *, const struct ofpbuf *); | |
130 | static bool is_connected_state(enum state); | |
131 | static bool is_admitted_msg(const struct ofpbuf *); | |
132 | ||
133 | /* Creates a new rconn, connects it (reliably) to 'name', and returns it. */ | |
134 | struct rconn * | |
135 | rconn_new(const char *name, int inactivity_probe_interval, int max_backoff) | |
136 | { | |
137 | struct rconn *rc = rconn_create(inactivity_probe_interval, max_backoff); | |
138 | rconn_connect(rc, name); | |
139 | return rc; | |
140 | } | |
141 | ||
142 | /* Creates a new rconn, connects it (unreliably) to 'vconn', and returns it. */ | |
143 | struct rconn * | |
144 | rconn_new_from_vconn(const char *name, struct vconn *vconn) | |
145 | { | |
146 | struct rconn *rc = rconn_create(60, 0); | |
147 | rconn_connect_unreliably(rc, name, vconn); | |
148 | return rc; | |
149 | } | |
150 | ||
151 | /* Creates and returns a new rconn. | |
152 | * | |
153 | * 'probe_interval' is a number of seconds. If the interval passes once | |
154 | * without an OpenFlow message being received from the peer, the rconn sends | |
155 | * out an "echo request" message. If the interval passes again without a | |
156 | * message being received, the rconn disconnects and re-connects to the peer. | |
157 | * Setting 'probe_interval' to 0 disables this behavior. | |
158 | * | |
159 | * 'max_backoff' is the maximum number of seconds between attempts to connect | |
160 | * to the peer. The actual interval starts at 1 second and doubles on each | |
161 | * failure until it reaches 'max_backoff'. If 0 is specified, the default of | |
c9aaa877 | 162 | * 8 seconds is used. */ |
064af421 BP |
163 | struct rconn * |
164 | rconn_create(int probe_interval, int max_backoff) | |
165 | { | |
166 | struct rconn *rc = xcalloc(1, sizeof *rc); | |
167 | ||
168 | rc->state = S_VOID; | |
169 | rc->state_entered = time_now(); | |
170 | ||
171 | rc->vconn = NULL; | |
172 | rc->name = xstrdup("void"); | |
173 | rc->reliable = false; | |
174 | ||
175 | queue_init(&rc->txq); | |
176 | ||
177 | rc->backoff = 0; | |
c9aaa877 | 178 | rc->max_backoff = max_backoff ? max_backoff : 8; |
064af421 BP |
179 | rc->backoff_deadline = TIME_MIN; |
180 | rc->last_received = time_now(); | |
181 | rc->last_connected = time_now(); | |
182 | rc->seqno = 0; | |
183 | ||
184 | rc->packets_sent = 0; | |
185 | ||
186 | rc->probably_admitted = false; | |
187 | rc->last_admitted = time_now(); | |
188 | ||
189 | rc->packets_received = 0; | |
190 | rc->n_attempted_connections = 0; | |
191 | rc->n_successful_connections = 0; | |
192 | rc->creation_time = time_now(); | |
193 | rc->total_time_connected = 0; | |
194 | ||
195 | rc->questionable_connectivity = false; | |
196 | rc->last_questioned = time_now(); | |
197 | ||
198 | rconn_set_probe_interval(rc, probe_interval); | |
199 | ||
200 | rc->n_monitors = 0; | |
201 | ||
202 | return rc; | |
203 | } | |
204 | ||
205 | void | |
206 | rconn_set_max_backoff(struct rconn *rc, int max_backoff) | |
207 | { | |
208 | rc->max_backoff = MAX(1, max_backoff); | |
209 | if (rc->state == S_BACKOFF && rc->backoff > max_backoff) { | |
210 | rc->backoff = max_backoff; | |
211 | if (rc->backoff_deadline > time_now() + max_backoff) { | |
212 | rc->backoff_deadline = time_now() + max_backoff; | |
213 | } | |
214 | } | |
215 | } | |
216 | ||
217 | int | |
218 | rconn_get_max_backoff(const struct rconn *rc) | |
219 | { | |
220 | return rc->max_backoff; | |
221 | } | |
222 | ||
223 | void | |
224 | rconn_set_probe_interval(struct rconn *rc, int probe_interval) | |
225 | { | |
226 | rc->probe_interval = probe_interval ? MAX(5, probe_interval) : 0; | |
227 | } | |
228 | ||
229 | int | |
230 | rconn_get_probe_interval(const struct rconn *rc) | |
231 | { | |
232 | return rc->probe_interval; | |
233 | } | |
234 | ||
235 | int | |
236 | rconn_connect(struct rconn *rc, const char *name) | |
237 | { | |
238 | rconn_disconnect(rc); | |
239 | free(rc->name); | |
240 | rc->name = xstrdup(name); | |
241 | rc->reliable = true; | |
242 | return reconnect(rc); | |
243 | } | |
244 | ||
245 | void | |
246 | rconn_connect_unreliably(struct rconn *rc, | |
247 | const char *name, struct vconn *vconn) | |
248 | { | |
249 | assert(vconn != NULL); | |
250 | rconn_disconnect(rc); | |
251 | free(rc->name); | |
252 | rc->name = xstrdup(name); | |
253 | rc->reliable = false; | |
254 | rc->vconn = vconn; | |
255 | rc->last_connected = time_now(); | |
256 | state_transition(rc, S_ACTIVE); | |
257 | } | |
258 | ||
259 | /* If 'rc' is connected, forces it to drop the connection and reconnect. */ | |
260 | void | |
261 | rconn_reconnect(struct rconn *rc) | |
262 | { | |
263 | if (rc->state & (S_ACTIVE | S_IDLE)) { | |
264 | disconnect(rc, 0); | |
265 | } | |
266 | } | |
267 | ||
268 | void | |
269 | rconn_disconnect(struct rconn *rc) | |
270 | { | |
271 | if (rc->state != S_VOID) { | |
272 | if (rc->vconn) { | |
273 | vconn_close(rc->vconn); | |
274 | rc->vconn = NULL; | |
275 | } | |
276 | free(rc->name); | |
277 | rc->name = xstrdup("void"); | |
278 | rc->reliable = false; | |
279 | ||
280 | rc->backoff = 0; | |
281 | rc->backoff_deadline = TIME_MIN; | |
282 | ||
283 | state_transition(rc, S_VOID); | |
284 | } | |
285 | } | |
286 | ||
287 | /* Disconnects 'rc' and frees the underlying storage. */ | |
288 | void | |
289 | rconn_destroy(struct rconn *rc) | |
290 | { | |
291 | if (rc) { | |
292 | size_t i; | |
293 | ||
294 | free(rc->name); | |
295 | vconn_close(rc->vconn); | |
296 | flush_queue(rc); | |
297 | queue_destroy(&rc->txq); | |
298 | for (i = 0; i < rc->n_monitors; i++) { | |
299 | vconn_close(rc->monitors[i]); | |
300 | } | |
301 | free(rc); | |
302 | } | |
303 | } | |
304 | ||
305 | static unsigned int | |
306 | timeout_VOID(const struct rconn *rc UNUSED) | |
307 | { | |
308 | return UINT_MAX; | |
309 | } | |
310 | ||
311 | static void | |
312 | run_VOID(struct rconn *rc UNUSED) | |
313 | { | |
314 | /* Nothing to do. */ | |
315 | } | |
316 | ||
317 | static int | |
318 | reconnect(struct rconn *rc) | |
319 | { | |
320 | int retval; | |
321 | ||
322 | VLOG_INFO("%s: connecting...", rc->name); | |
323 | rc->n_attempted_connections++; | |
324 | retval = vconn_open(rc->name, OFP_VERSION, &rc->vconn); | |
325 | if (!retval) { | |
326 | rc->backoff_deadline = time_now() + rc->backoff; | |
327 | state_transition(rc, S_CONNECTING); | |
328 | } else { | |
329 | VLOG_WARN("%s: connection failed (%s)", rc->name, strerror(retval)); | |
330 | rc->backoff_deadline = TIME_MAX; /* Prevent resetting backoff. */ | |
331 | disconnect(rc, 0); | |
332 | } | |
333 | return retval; | |
334 | } | |
335 | ||
336 | static unsigned int | |
337 | timeout_BACKOFF(const struct rconn *rc) | |
338 | { | |
339 | return rc->backoff; | |
340 | } | |
341 | ||
342 | static void | |
343 | run_BACKOFF(struct rconn *rc) | |
344 | { | |
345 | if (timed_out(rc)) { | |
346 | reconnect(rc); | |
347 | } | |
348 | } | |
349 | ||
350 | static unsigned int | |
351 | timeout_CONNECTING(const struct rconn *rc) | |
352 | { | |
353 | return MAX(1, rc->backoff); | |
354 | } | |
355 | ||
356 | static void | |
357 | run_CONNECTING(struct rconn *rc) | |
358 | { | |
359 | int retval = vconn_connect(rc->vconn); | |
360 | if (!retval) { | |
361 | VLOG_INFO("%s: connected", rc->name); | |
362 | rc->n_successful_connections++; | |
363 | state_transition(rc, S_ACTIVE); | |
364 | rc->last_connected = rc->state_entered; | |
365 | } else if (retval != EAGAIN) { | |
366 | VLOG_INFO("%s: connection failed (%s)", rc->name, strerror(retval)); | |
367 | disconnect(rc, retval); | |
368 | } else if (timed_out(rc)) { | |
369 | VLOG_INFO("%s: connection timed out", rc->name); | |
370 | rc->backoff_deadline = TIME_MAX; /* Prevent resetting backoff. */ | |
371 | disconnect(rc, 0); | |
372 | } | |
373 | } | |
374 | ||
375 | static void | |
376 | do_tx_work(struct rconn *rc) | |
377 | { | |
378 | if (!rc->txq.n) { | |
379 | return; | |
380 | } | |
381 | while (rc->txq.n > 0) { | |
382 | int error = try_send(rc); | |
383 | if (error) { | |
384 | break; | |
385 | } | |
386 | } | |
387 | if (!rc->txq.n) { | |
388 | poll_immediate_wake(); | |
389 | } | |
390 | } | |
391 | ||
392 | static unsigned int | |
393 | timeout_ACTIVE(const struct rconn *rc) | |
394 | { | |
395 | if (rc->probe_interval) { | |
396 | unsigned int base = MAX(rc->last_received, rc->state_entered); | |
397 | unsigned int arg = base + rc->probe_interval - rc->state_entered; | |
398 | return arg; | |
399 | } | |
400 | return UINT_MAX; | |
401 | } | |
402 | ||
403 | static void | |
404 | run_ACTIVE(struct rconn *rc) | |
405 | { | |
406 | if (timed_out(rc)) { | |
407 | unsigned int base = MAX(rc->last_received, rc->state_entered); | |
408 | VLOG_DBG("%s: idle %u seconds, sending inactivity probe", | |
409 | rc->name, (unsigned int) (time_now() - base)); | |
410 | ||
411 | /* Ordering is important here: rconn_send() can transition to BACKOFF, | |
412 | * and we don't want to transition back to IDLE if so, because then we | |
413 | * can end up queuing a packet with vconn == NULL and then *boom*. */ | |
414 | state_transition(rc, S_IDLE); | |
415 | rconn_send(rc, make_echo_request(), NULL); | |
416 | return; | |
417 | } | |
418 | ||
419 | do_tx_work(rc); | |
420 | } | |
421 | ||
422 | static unsigned int | |
423 | timeout_IDLE(const struct rconn *rc) | |
424 | { | |
425 | return rc->probe_interval; | |
426 | } | |
427 | ||
428 | static void | |
429 | run_IDLE(struct rconn *rc) | |
430 | { | |
431 | if (timed_out(rc)) { | |
432 | question_connectivity(rc); | |
433 | VLOG_ERR("%s: no response to inactivity probe after %u " | |
434 | "seconds, disconnecting", | |
435 | rc->name, elapsed_in_this_state(rc)); | |
436 | disconnect(rc, 0); | |
437 | } else { | |
438 | do_tx_work(rc); | |
439 | } | |
440 | } | |
441 | ||
442 | /* Performs whatever activities are necessary to maintain 'rc': if 'rc' is | |
443 | * disconnected, attempts to (re)connect, backing off as necessary; if 'rc' is | |
444 | * connected, attempts to send packets in the send queue, if any. */ | |
445 | void | |
446 | rconn_run(struct rconn *rc) | |
447 | { | |
448 | int old_state; | |
449 | do { | |
450 | old_state = rc->state; | |
451 | switch (rc->state) { | |
452 | #define STATE(NAME, VALUE) case S_##NAME: run_##NAME(rc); break; | |
453 | STATES | |
454 | #undef STATE | |
455 | default: | |
456 | NOT_REACHED(); | |
457 | } | |
458 | } while (rc->state != old_state); | |
459 | } | |
460 | ||
461 | /* Causes the next call to poll_block() to wake up when rconn_run() should be | |
462 | * called on 'rc'. */ | |
463 | void | |
464 | rconn_run_wait(struct rconn *rc) | |
465 | { | |
466 | unsigned int timeo = timeout(rc); | |
467 | if (timeo != UINT_MAX) { | |
468 | unsigned int expires = sat_add(rc->state_entered, timeo); | |
469 | unsigned int remaining = sat_sub(expires, time_now()); | |
470 | poll_timer_wait(sat_mul(remaining, 1000)); | |
471 | } | |
472 | ||
473 | if ((rc->state & (S_ACTIVE | S_IDLE)) && rc->txq.n) { | |
474 | vconn_wait(rc->vconn, WAIT_SEND); | |
475 | } | |
476 | } | |
477 | ||
478 | /* Attempts to receive a packet from 'rc'. If successful, returns the packet; | |
479 | * otherwise, returns a null pointer. The caller is responsible for freeing | |
480 | * the packet (with ofpbuf_delete()). */ | |
481 | struct ofpbuf * | |
482 | rconn_recv(struct rconn *rc) | |
483 | { | |
484 | if (rc->state & (S_ACTIVE | S_IDLE)) { | |
485 | struct ofpbuf *buffer; | |
486 | int error = vconn_recv(rc->vconn, &buffer); | |
487 | if (!error) { | |
488 | copy_to_monitor(rc, buffer); | |
489 | if (is_admitted_msg(buffer) | |
490 | || time_now() - rc->last_connected >= 30) { | |
491 | rc->probably_admitted = true; | |
492 | rc->last_admitted = time_now(); | |
493 | } | |
494 | rc->last_received = time_now(); | |
495 | rc->packets_received++; | |
496 | if (rc->state == S_IDLE) { | |
497 | state_transition(rc, S_ACTIVE); | |
498 | } | |
499 | return buffer; | |
500 | } else if (error != EAGAIN) { | |
501 | disconnect(rc, error); | |
502 | } | |
503 | } | |
504 | return NULL; | |
505 | } | |
506 | ||
507 | /* Causes the next call to poll_block() to wake up when a packet may be ready | |
508 | * to be received by vconn_recv() on 'rc'. */ | |
509 | void | |
510 | rconn_recv_wait(struct rconn *rc) | |
511 | { | |
512 | if (rc->vconn) { | |
513 | vconn_wait(rc->vconn, WAIT_RECV); | |
514 | } | |
515 | } | |
516 | ||
517 | /* Sends 'b' on 'rc'. Returns 0 if successful (in which case 'b' is | |
518 | * destroyed), or ENOTCONN if 'rc' is not currently connected (in which case | |
519 | * the caller retains ownership of 'b'). | |
520 | * | |
521 | * If 'counter' is non-null, then 'counter' will be incremented while the | |
522 | * packet is in flight, then decremented when it has been sent (or discarded | |
523 | * due to disconnection). Because 'b' may be sent (or discarded) before this | |
524 | * function returns, the caller may not be able to observe any change in | |
525 | * 'counter'. | |
526 | * | |
527 | * There is no rconn_send_wait() function: an rconn has a send queue that it | |
528 | * takes care of sending if you call rconn_run(), which will have the side | |
529 | * effect of waking up poll_block(). */ | |
530 | int | |
531 | rconn_send(struct rconn *rc, struct ofpbuf *b, | |
532 | struct rconn_packet_counter *counter) | |
533 | { | |
534 | if (rconn_is_connected(rc)) { | |
535 | COVERAGE_INC(rconn_queued); | |
536 | copy_to_monitor(rc, b); | |
537 | b->private = counter; | |
538 | if (counter) { | |
539 | rconn_packet_counter_inc(counter); | |
540 | } | |
541 | queue_push_tail(&rc->txq, b); | |
542 | ||
543 | /* If the queue was empty before we added 'b', try to send some | |
544 | * packets. (But if the queue had packets in it, it's because the | |
545 | * vconn is backlogged and there's no point in stuffing more into it | |
546 | * now. We'll get back to that in rconn_run().) */ | |
547 | if (rc->txq.n == 1) { | |
548 | try_send(rc); | |
549 | } | |
550 | return 0; | |
551 | } else { | |
552 | return ENOTCONN; | |
553 | } | |
554 | } | |
555 | ||
556 | /* Sends 'b' on 'rc'. Increments 'counter' while the packet is in flight; it | |
557 | * will be decremented when it has been sent (or discarded due to | |
558 | * disconnection). Returns 0 if successful, EAGAIN if 'counter->n' is already | |
559 | * at least as large as 'queue_limit', or ENOTCONN if 'rc' is not currently | |
560 | * connected. Regardless of return value, 'b' is destroyed. | |
561 | * | |
562 | * Because 'b' may be sent (or discarded) before this function returns, the | |
563 | * caller may not be able to observe any change in 'counter'. | |
564 | * | |
565 | * There is no rconn_send_wait() function: an rconn has a send queue that it | |
566 | * takes care of sending if you call rconn_run(), which will have the side | |
567 | * effect of waking up poll_block(). */ | |
568 | int | |
569 | rconn_send_with_limit(struct rconn *rc, struct ofpbuf *b, | |
570 | struct rconn_packet_counter *counter, int queue_limit) | |
571 | { | |
572 | int retval; | |
573 | retval = counter->n >= queue_limit ? EAGAIN : rconn_send(rc, b, counter); | |
574 | if (retval) { | |
575 | COVERAGE_INC(rconn_overflow); | |
576 | ofpbuf_delete(b); | |
577 | } | |
578 | return retval; | |
579 | } | |
580 | ||
581 | /* Returns the total number of packets successfully sent on the underlying | |
582 | * vconn. A packet is not counted as sent while it is still queued in the | |
583 | * rconn, only when it has been successfuly passed to the vconn. */ | |
584 | unsigned int | |
585 | rconn_packets_sent(const struct rconn *rc) | |
586 | { | |
587 | return rc->packets_sent; | |
588 | } | |
589 | ||
590 | /* Adds 'vconn' to 'rc' as a monitoring connection, to which all messages sent | |
591 | * and received on 'rconn' will be copied. 'rc' takes ownership of 'vconn'. */ | |
592 | void | |
593 | rconn_add_monitor(struct rconn *rc, struct vconn *vconn) | |
594 | { | |
595 | if (rc->n_monitors < ARRAY_SIZE(rc->monitors)) { | |
596 | VLOG_INFO("new monitor connection from %s", vconn_get_name(vconn)); | |
597 | rc->monitors[rc->n_monitors++] = vconn; | |
598 | } else { | |
599 | VLOG_DBG("too many monitor connections, discarding %s", | |
600 | vconn_get_name(vconn)); | |
601 | vconn_close(vconn); | |
602 | } | |
603 | } | |
604 | ||
605 | /* Returns 'rc''s name (the 'name' argument passed to rconn_new()). */ | |
606 | const char * | |
607 | rconn_get_name(const struct rconn *rc) | |
608 | { | |
609 | return rc->name; | |
610 | } | |
611 | ||
612 | /* Returns true if 'rconn' is connected or in the process of reconnecting, | |
613 | * false if 'rconn' is disconnected and will not reconnect on its own. */ | |
614 | bool | |
615 | rconn_is_alive(const struct rconn *rconn) | |
616 | { | |
617 | return rconn->state != S_VOID; | |
618 | } | |
619 | ||
620 | /* Returns true if 'rconn' is connected, false otherwise. */ | |
621 | bool | |
622 | rconn_is_connected(const struct rconn *rconn) | |
623 | { | |
624 | return is_connected_state(rconn->state); | |
625 | } | |
626 | ||
627 | /* Returns 0 if 'rconn' is connected. Otherwise, if 'rconn' is in a "failure | |
628 | * mode" (that is, it is not connected), returns the number of seconds that it | |
629 | * has been in failure mode, ignoring any times that it connected but the | |
630 | * controller's admission control policy caused it to be quickly | |
631 | * disconnected. */ | |
632 | int | |
633 | rconn_failure_duration(const struct rconn *rconn) | |
634 | { | |
635 | return rconn_is_connected(rconn) ? 0 : time_now() - rconn->last_admitted; | |
636 | } | |
637 | ||
638 | /* Returns the IP address of the peer, or 0 if the peer is not connected over | |
639 | * an IP-based protocol or if its IP address is not known. */ | |
640 | uint32_t | |
193456d5 | 641 | rconn_get_remote_ip(const struct rconn *rconn) |
064af421 | 642 | { |
193456d5 JP |
643 | return rconn->vconn ? vconn_get_remote_ip(rconn->vconn) : 0; |
644 | } | |
645 | ||
646 | /* Returns the transport port of the peer, or 0 if the peer does not | |
647 | * contain a port or if the port is not known. */ | |
648 | uint16_t | |
649 | rconn_get_remote_port(const struct rconn *rconn) | |
650 | { | |
651 | return rconn->vconn ? vconn_get_remote_port(rconn->vconn) : 0; | |
652 | } | |
653 | ||
654 | /* Returns the IP address used to connect to the peer, or 0 if the | |
655 | * connection is not an IP-based protocol or if its IP address is not | |
656 | * known. */ | |
657 | uint32_t | |
658 | rconn_get_local_ip(const struct rconn *rconn) | |
659 | { | |
660 | return rconn->vconn ? vconn_get_local_ip(rconn->vconn) : 0; | |
661 | } | |
662 | ||
663 | /* Returns the transport port used to connect to the peer, or 0 if the | |
664 | * connection does not contain a port or if the port is not known. */ | |
665 | uint16_t | |
666 | rconn_get_local_port(const struct rconn *rconn) | |
667 | { | |
668 | return rconn->vconn ? vconn_get_local_port(rconn->vconn) : 0; | |
064af421 BP |
669 | } |
670 | ||
671 | /* If 'rconn' can't connect to the peer, it could be for any number of reasons. | |
672 | * Usually, one would assume it is because the peer is not running or because | |
673 | * the network is partitioned. But it could also be because the network | |
674 | * topology has changed, in which case the upper layer will need to reassess it | |
675 | * (in particular, obtain a new IP address via DHCP and find the new location | |
676 | * of the controller). When this appears that this might be the case, this | |
677 | * function returns true. It also clears the questionability flag and prevents | |
678 | * it from being set again for some time. */ | |
679 | bool | |
680 | rconn_is_connectivity_questionable(struct rconn *rconn) | |
681 | { | |
682 | bool questionable = rconn->questionable_connectivity; | |
683 | rconn->questionable_connectivity = false; | |
684 | return questionable; | |
685 | } | |
686 | ||
687 | /* Returns the total number of packets successfully received by the underlying | |
688 | * vconn. */ | |
689 | unsigned int | |
690 | rconn_packets_received(const struct rconn *rc) | |
691 | { | |
692 | return rc->packets_received; | |
693 | } | |
694 | ||
695 | /* Returns a string representing the internal state of 'rc'. The caller must | |
696 | * not modify or free the string. */ | |
697 | const char * | |
698 | rconn_get_state(const struct rconn *rc) | |
699 | { | |
700 | return state_name(rc->state); | |
701 | } | |
702 | ||
703 | /* Returns the number of connection attempts made by 'rc', including any | |
704 | * ongoing attempt that has not yet succeeded or failed. */ | |
705 | unsigned int | |
706 | rconn_get_attempted_connections(const struct rconn *rc) | |
707 | { | |
708 | return rc->n_attempted_connections; | |
709 | } | |
710 | ||
711 | /* Returns the number of successful connection attempts made by 'rc'. */ | |
712 | unsigned int | |
713 | rconn_get_successful_connections(const struct rconn *rc) | |
714 | { | |
715 | return rc->n_successful_connections; | |
716 | } | |
717 | ||
718 | /* Returns the time at which the last successful connection was made by | |
719 | * 'rc'. */ | |
720 | time_t | |
721 | rconn_get_last_connection(const struct rconn *rc) | |
722 | { | |
723 | return rc->last_connected; | |
724 | } | |
725 | ||
7df824b7 BP |
726 | /* Returns the time at which the last OpenFlow message was received by 'rc'. |
727 | * If no packets have been received on 'rc', returns the time at which 'rc' | |
728 | * was created. */ | |
729 | time_t | |
730 | rconn_get_last_received(const struct rconn *rc) | |
731 | { | |
732 | return rc->last_received; | |
733 | } | |
734 | ||
064af421 BP |
735 | /* Returns the time at which 'rc' was created. */ |
736 | time_t | |
737 | rconn_get_creation_time(const struct rconn *rc) | |
738 | { | |
739 | return rc->creation_time; | |
740 | } | |
741 | ||
742 | /* Returns the approximate number of seconds that 'rc' has been connected. */ | |
743 | unsigned long int | |
744 | rconn_get_total_time_connected(const struct rconn *rc) | |
745 | { | |
746 | return (rc->total_time_connected | |
747 | + (rconn_is_connected(rc) ? elapsed_in_this_state(rc) : 0)); | |
748 | } | |
749 | ||
750 | /* Returns the current amount of backoff, in seconds. This is the amount of | |
751 | * time after which the rconn will transition from BACKOFF to CONNECTING. */ | |
752 | int | |
753 | rconn_get_backoff(const struct rconn *rc) | |
754 | { | |
755 | return rc->backoff; | |
756 | } | |
757 | ||
758 | /* Returns the number of seconds spent in this state so far. */ | |
759 | unsigned int | |
760 | rconn_get_state_elapsed(const struct rconn *rc) | |
761 | { | |
762 | return elapsed_in_this_state(rc); | |
763 | } | |
764 | ||
765 | /* Returns 'rc''s current connection sequence number, a number that changes | |
766 | * every time that 'rconn' connects or disconnects. */ | |
767 | unsigned int | |
768 | rconn_get_connection_seqno(const struct rconn *rc) | |
769 | { | |
770 | return rc->seqno; | |
771 | } | |
772 | \f | |
773 | struct rconn_packet_counter * | |
774 | rconn_packet_counter_create(void) | |
775 | { | |
776 | struct rconn_packet_counter *c = xmalloc(sizeof *c); | |
777 | c->n = 0; | |
778 | c->ref_cnt = 1; | |
779 | return c; | |
780 | } | |
781 | ||
782 | void | |
783 | rconn_packet_counter_destroy(struct rconn_packet_counter *c) | |
784 | { | |
785 | if (c) { | |
786 | assert(c->ref_cnt > 0); | |
787 | if (!--c->ref_cnt && !c->n) { | |
788 | free(c); | |
789 | } | |
790 | } | |
791 | } | |
792 | ||
793 | void | |
794 | rconn_packet_counter_inc(struct rconn_packet_counter *c) | |
795 | { | |
796 | c->n++; | |
797 | } | |
798 | ||
799 | void | |
800 | rconn_packet_counter_dec(struct rconn_packet_counter *c) | |
801 | { | |
802 | assert(c->n > 0); | |
803 | if (!--c->n && !c->ref_cnt) { | |
804 | free(c); | |
805 | } | |
806 | } | |
807 | \f | |
808 | /* Tries to send a packet from 'rc''s send buffer. Returns 0 if successful, | |
809 | * otherwise a positive errno value. */ | |
810 | static int | |
811 | try_send(struct rconn *rc) | |
812 | { | |
813 | int retval = 0; | |
814 | struct ofpbuf *next = rc->txq.head->next; | |
815 | struct rconn_packet_counter *counter = rc->txq.head->private; | |
816 | retval = vconn_send(rc->vconn, rc->txq.head); | |
817 | if (retval) { | |
818 | if (retval != EAGAIN) { | |
819 | disconnect(rc, retval); | |
820 | } | |
821 | return retval; | |
822 | } | |
823 | COVERAGE_INC(rconn_sent); | |
824 | rc->packets_sent++; | |
825 | if (counter) { | |
826 | rconn_packet_counter_dec(counter); | |
827 | } | |
828 | queue_advance_head(&rc->txq, next); | |
829 | return 0; | |
830 | } | |
831 | ||
832 | /* Disconnects 'rc'. 'error' is used only for logging purposes. If it is | |
833 | * nonzero, then it should be EOF to indicate the connection was closed by the | |
834 | * peer in a normal fashion or a positive errno value. */ | |
835 | static void | |
836 | disconnect(struct rconn *rc, int error) | |
837 | { | |
838 | if (rc->reliable) { | |
839 | time_t now = time_now(); | |
840 | ||
841 | if (rc->state & (S_CONNECTING | S_ACTIVE | S_IDLE)) { | |
842 | if (error > 0) { | |
843 | VLOG_WARN("%s: connection dropped (%s)", | |
844 | rc->name, strerror(error)); | |
845 | } else if (error == EOF) { | |
846 | if (rc->reliable) { | |
847 | VLOG_INFO("%s: connection closed by peer", rc->name); | |
848 | } | |
849 | } else { | |
850 | VLOG_INFO("%s: connection dropped", rc->name); | |
851 | } | |
852 | vconn_close(rc->vconn); | |
853 | rc->vconn = NULL; | |
854 | flush_queue(rc); | |
855 | } | |
856 | ||
857 | if (now >= rc->backoff_deadline) { | |
858 | rc->backoff = 1; | |
859 | } else { | |
860 | rc->backoff = MIN(rc->max_backoff, MAX(1, 2 * rc->backoff)); | |
861 | VLOG_INFO("%s: waiting %d seconds before reconnect\n", | |
862 | rc->name, rc->backoff); | |
863 | } | |
864 | rc->backoff_deadline = now + rc->backoff; | |
865 | state_transition(rc, S_BACKOFF); | |
866 | if (now - rc->last_connected > 60) { | |
867 | question_connectivity(rc); | |
868 | } | |
869 | } else { | |
870 | rconn_disconnect(rc); | |
871 | } | |
872 | } | |
873 | ||
874 | /* Drops all the packets from 'rc''s send queue and decrements their queue | |
875 | * counts. */ | |
876 | static void | |
877 | flush_queue(struct rconn *rc) | |
878 | { | |
879 | if (!rc->txq.n) { | |
880 | return; | |
881 | } | |
882 | while (rc->txq.n > 0) { | |
883 | struct ofpbuf *b = queue_pop_head(&rc->txq); | |
884 | struct rconn_packet_counter *counter = b->private; | |
885 | if (counter) { | |
886 | rconn_packet_counter_dec(counter); | |
887 | } | |
888 | COVERAGE_INC(rconn_discarded); | |
889 | ofpbuf_delete(b); | |
890 | } | |
891 | poll_immediate_wake(); | |
892 | } | |
893 | ||
894 | static unsigned int | |
895 | elapsed_in_this_state(const struct rconn *rc) | |
896 | { | |
897 | return time_now() - rc->state_entered; | |
898 | } | |
899 | ||
900 | static unsigned int | |
901 | timeout(const struct rconn *rc) | |
902 | { | |
903 | switch (rc->state) { | |
904 | #define STATE(NAME, VALUE) case S_##NAME: return timeout_##NAME(rc); | |
905 | STATES | |
906 | #undef STATE | |
907 | default: | |
908 | NOT_REACHED(); | |
909 | } | |
910 | } | |
911 | ||
912 | static bool | |
913 | timed_out(const struct rconn *rc) | |
914 | { | |
915 | return time_now() >= sat_add(rc->state_entered, timeout(rc)); | |
916 | } | |
917 | ||
918 | static void | |
919 | state_transition(struct rconn *rc, enum state state) | |
920 | { | |
921 | rc->seqno += (rc->state == S_ACTIVE) != (state == S_ACTIVE); | |
922 | if (is_connected_state(state) && !is_connected_state(rc->state)) { | |
923 | rc->probably_admitted = false; | |
924 | } | |
925 | if (rconn_is_connected(rc)) { | |
926 | rc->total_time_connected += elapsed_in_this_state(rc); | |
927 | } | |
928 | VLOG_DBG("%s: entering %s", rc->name, state_name(state)); | |
929 | rc->state = state; | |
930 | rc->state_entered = time_now(); | |
931 | } | |
932 | ||
933 | static void | |
934 | question_connectivity(struct rconn *rc) | |
935 | { | |
936 | time_t now = time_now(); | |
937 | if (now - rc->last_questioned > 60) { | |
938 | rc->questionable_connectivity = true; | |
939 | rc->last_questioned = now; | |
940 | } | |
941 | } | |
942 | ||
943 | static void | |
944 | copy_to_monitor(struct rconn *rc, const struct ofpbuf *b) | |
945 | { | |
946 | struct ofpbuf *clone = NULL; | |
947 | int retval; | |
948 | size_t i; | |
949 | ||
950 | for (i = 0; i < rc->n_monitors; ) { | |
951 | struct vconn *vconn = rc->monitors[i]; | |
952 | ||
953 | if (!clone) { | |
954 | clone = ofpbuf_clone(b); | |
955 | } | |
956 | retval = vconn_send(vconn, clone); | |
957 | if (!retval) { | |
958 | clone = NULL; | |
959 | } else if (retval != EAGAIN) { | |
960 | VLOG_DBG("%s: closing monitor connection to %s: %s", | |
961 | rconn_get_name(rc), vconn_get_name(vconn), | |
962 | strerror(retval)); | |
963 | rc->monitors[i] = rc->monitors[--rc->n_monitors]; | |
964 | continue; | |
965 | } | |
966 | i++; | |
967 | } | |
968 | ofpbuf_delete(clone); | |
969 | } | |
970 | ||
971 | static bool | |
972 | is_connected_state(enum state state) | |
973 | { | |
974 | return (state & (S_ACTIVE | S_IDLE)) != 0; | |
975 | } | |
976 | ||
977 | static bool | |
978 | is_admitted_msg(const struct ofpbuf *b) | |
979 | { | |
980 | struct ofp_header *oh = b->data; | |
981 | uint8_t type = oh->type; | |
982 | return !(type < 32 | |
983 | && (1u << type) & ((1u << OFPT_HELLO) | | |
984 | (1u << OFPT_ERROR) | | |
985 | (1u << OFPT_ECHO_REQUEST) | | |
986 | (1u << OFPT_ECHO_REPLY) | | |
987 | (1u << OFPT_VENDOR) | | |
988 | (1u << OFPT_FEATURES_REQUEST) | | |
989 | (1u << OFPT_FEATURES_REPLY) | | |
990 | (1u << OFPT_GET_CONFIG_REQUEST) | | |
991 | (1u << OFPT_GET_CONFIG_REPLY) | | |
992 | (1u << OFPT_SET_CONFIG))); | |
993 | } |