]>
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 | |
8cd4882f | 92 | * anything internal to the rconn (or ofproto for that matter). */ |
064af421 BP |
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 | ||
19d1ab55 BP |
114 | /* When we create a vconn we obtain these values, to save them past the end |
115 | * of the vconn's lifetime. Otherwise, in-band control will only allow | |
116 | * traffic when a vconn is actually open, but it is nice to allow ARP to | |
117 | * complete even between connection attempts, and it is also polite to | |
118 | * allow traffic from other switches to go through to the controller | |
119 | * whether or not we are connected. | |
120 | * | |
121 | * We don't cache the local port, because that changes from one connection | |
122 | * attempt to the next. */ | |
123 | uint32_t local_ip, remote_ip; | |
124 | uint16_t remote_port; | |
125 | ||
064af421 BP |
126 | /* Messages sent or received are copied to the monitor connections. */ |
127 | #define MAX_MONITORS 8 | |
128 | struct vconn *monitors[8]; | |
129 | size_t n_monitors; | |
130 | }; | |
131 | ||
132 | static unsigned int elapsed_in_this_state(const struct rconn *); | |
133 | static unsigned int timeout(const struct rconn *); | |
134 | static bool timed_out(const struct rconn *); | |
135 | static void state_transition(struct rconn *, enum state); | |
19d1ab55 | 136 | static void set_vconn_name(struct rconn *, const char *name); |
064af421 BP |
137 | static int try_send(struct rconn *); |
138 | static int reconnect(struct rconn *); | |
139 | static void disconnect(struct rconn *, int error); | |
140 | static void flush_queue(struct rconn *); | |
141 | static void question_connectivity(struct rconn *); | |
142 | static void copy_to_monitor(struct rconn *, const struct ofpbuf *); | |
143 | static bool is_connected_state(enum state); | |
144 | static bool is_admitted_msg(const struct ofpbuf *); | |
145 | ||
146 | /* Creates a new rconn, connects it (reliably) to 'name', and returns it. */ | |
147 | struct rconn * | |
148 | rconn_new(const char *name, int inactivity_probe_interval, int max_backoff) | |
149 | { | |
150 | struct rconn *rc = rconn_create(inactivity_probe_interval, max_backoff); | |
151 | rconn_connect(rc, name); | |
152 | return rc; | |
153 | } | |
154 | ||
155 | /* Creates a new rconn, connects it (unreliably) to 'vconn', and returns it. */ | |
156 | struct rconn * | |
157 | rconn_new_from_vconn(const char *name, struct vconn *vconn) | |
158 | { | |
159 | struct rconn *rc = rconn_create(60, 0); | |
160 | rconn_connect_unreliably(rc, name, vconn); | |
161 | return rc; | |
162 | } | |
163 | ||
164 | /* Creates and returns a new rconn. | |
165 | * | |
166 | * 'probe_interval' is a number of seconds. If the interval passes once | |
167 | * without an OpenFlow message being received from the peer, the rconn sends | |
168 | * out an "echo request" message. If the interval passes again without a | |
169 | * message being received, the rconn disconnects and re-connects to the peer. | |
170 | * Setting 'probe_interval' to 0 disables this behavior. | |
171 | * | |
172 | * 'max_backoff' is the maximum number of seconds between attempts to connect | |
173 | * to the peer. The actual interval starts at 1 second and doubles on each | |
174 | * failure until it reaches 'max_backoff'. If 0 is specified, the default of | |
c9aaa877 | 175 | * 8 seconds is used. */ |
064af421 BP |
176 | struct rconn * |
177 | rconn_create(int probe_interval, int max_backoff) | |
178 | { | |
179 | struct rconn *rc = xcalloc(1, sizeof *rc); | |
180 | ||
181 | rc->state = S_VOID; | |
182 | rc->state_entered = time_now(); | |
183 | ||
184 | rc->vconn = NULL; | |
185 | rc->name = xstrdup("void"); | |
186 | rc->reliable = false; | |
187 | ||
188 | queue_init(&rc->txq); | |
189 | ||
190 | rc->backoff = 0; | |
c9aaa877 | 191 | rc->max_backoff = max_backoff ? max_backoff : 8; |
064af421 BP |
192 | rc->backoff_deadline = TIME_MIN; |
193 | rc->last_received = time_now(); | |
194 | rc->last_connected = time_now(); | |
195 | rc->seqno = 0; | |
196 | ||
197 | rc->packets_sent = 0; | |
198 | ||
199 | rc->probably_admitted = false; | |
200 | rc->last_admitted = time_now(); | |
201 | ||
202 | rc->packets_received = 0; | |
203 | rc->n_attempted_connections = 0; | |
204 | rc->n_successful_connections = 0; | |
205 | rc->creation_time = time_now(); | |
206 | rc->total_time_connected = 0; | |
207 | ||
208 | rc->questionable_connectivity = false; | |
209 | rc->last_questioned = time_now(); | |
210 | ||
211 | rconn_set_probe_interval(rc, probe_interval); | |
212 | ||
213 | rc->n_monitors = 0; | |
214 | ||
215 | return rc; | |
216 | } | |
217 | ||
218 | void | |
219 | rconn_set_max_backoff(struct rconn *rc, int max_backoff) | |
220 | { | |
221 | rc->max_backoff = MAX(1, max_backoff); | |
222 | if (rc->state == S_BACKOFF && rc->backoff > max_backoff) { | |
223 | rc->backoff = max_backoff; | |
224 | if (rc->backoff_deadline > time_now() + max_backoff) { | |
225 | rc->backoff_deadline = time_now() + max_backoff; | |
226 | } | |
227 | } | |
228 | } | |
229 | ||
230 | int | |
231 | rconn_get_max_backoff(const struct rconn *rc) | |
232 | { | |
233 | return rc->max_backoff; | |
234 | } | |
235 | ||
236 | void | |
237 | rconn_set_probe_interval(struct rconn *rc, int probe_interval) | |
238 | { | |
239 | rc->probe_interval = probe_interval ? MAX(5, probe_interval) : 0; | |
240 | } | |
241 | ||
242 | int | |
243 | rconn_get_probe_interval(const struct rconn *rc) | |
244 | { | |
245 | return rc->probe_interval; | |
246 | } | |
247 | ||
248 | int | |
249 | rconn_connect(struct rconn *rc, const char *name) | |
250 | { | |
251 | rconn_disconnect(rc); | |
19d1ab55 | 252 | set_vconn_name(rc, name); |
064af421 BP |
253 | rc->reliable = true; |
254 | return reconnect(rc); | |
255 | } | |
256 | ||
257 | void | |
258 | rconn_connect_unreliably(struct rconn *rc, | |
259 | const char *name, struct vconn *vconn) | |
260 | { | |
261 | assert(vconn != NULL); | |
262 | rconn_disconnect(rc); | |
19d1ab55 | 263 | set_vconn_name(rc, name); |
064af421 BP |
264 | rc->reliable = false; |
265 | rc->vconn = vconn; | |
266 | rc->last_connected = time_now(); | |
267 | state_transition(rc, S_ACTIVE); | |
268 | } | |
269 | ||
270 | /* If 'rc' is connected, forces it to drop the connection and reconnect. */ | |
271 | void | |
272 | rconn_reconnect(struct rconn *rc) | |
273 | { | |
274 | if (rc->state & (S_ACTIVE | S_IDLE)) { | |
275 | disconnect(rc, 0); | |
276 | } | |
277 | } | |
278 | ||
279 | void | |
280 | rconn_disconnect(struct rconn *rc) | |
281 | { | |
282 | if (rc->state != S_VOID) { | |
283 | if (rc->vconn) { | |
284 | vconn_close(rc->vconn); | |
285 | rc->vconn = NULL; | |
286 | } | |
19d1ab55 | 287 | set_vconn_name(rc, "void"); |
064af421 BP |
288 | rc->reliable = false; |
289 | ||
290 | rc->backoff = 0; | |
291 | rc->backoff_deadline = TIME_MIN; | |
292 | ||
293 | state_transition(rc, S_VOID); | |
294 | } | |
295 | } | |
296 | ||
297 | /* Disconnects 'rc' and frees the underlying storage. */ | |
298 | void | |
299 | rconn_destroy(struct rconn *rc) | |
300 | { | |
301 | if (rc) { | |
302 | size_t i; | |
303 | ||
304 | free(rc->name); | |
305 | vconn_close(rc->vconn); | |
306 | flush_queue(rc); | |
307 | queue_destroy(&rc->txq); | |
308 | for (i = 0; i < rc->n_monitors; i++) { | |
309 | vconn_close(rc->monitors[i]); | |
310 | } | |
311 | free(rc); | |
312 | } | |
313 | } | |
314 | ||
315 | static unsigned int | |
316 | timeout_VOID(const struct rconn *rc UNUSED) | |
317 | { | |
318 | return UINT_MAX; | |
319 | } | |
320 | ||
321 | static void | |
322 | run_VOID(struct rconn *rc UNUSED) | |
323 | { | |
324 | /* Nothing to do. */ | |
325 | } | |
326 | ||
327 | static int | |
328 | reconnect(struct rconn *rc) | |
329 | { | |
330 | int retval; | |
331 | ||
332 | VLOG_INFO("%s: connecting...", rc->name); | |
333 | rc->n_attempted_connections++; | |
334 | retval = vconn_open(rc->name, OFP_VERSION, &rc->vconn); | |
335 | if (!retval) { | |
19d1ab55 BP |
336 | rc->remote_ip = vconn_get_remote_ip(rc->vconn); |
337 | rc->local_ip = vconn_get_local_ip(rc->vconn); | |
338 | rc->remote_port = vconn_get_remote_port(rc->vconn); | |
064af421 BP |
339 | rc->backoff_deadline = time_now() + rc->backoff; |
340 | state_transition(rc, S_CONNECTING); | |
341 | } else { | |
342 | VLOG_WARN("%s: connection failed (%s)", rc->name, strerror(retval)); | |
343 | rc->backoff_deadline = TIME_MAX; /* Prevent resetting backoff. */ | |
344 | disconnect(rc, 0); | |
345 | } | |
346 | return retval; | |
347 | } | |
348 | ||
349 | static unsigned int | |
350 | timeout_BACKOFF(const struct rconn *rc) | |
351 | { | |
352 | return rc->backoff; | |
353 | } | |
354 | ||
355 | static void | |
356 | run_BACKOFF(struct rconn *rc) | |
357 | { | |
358 | if (timed_out(rc)) { | |
359 | reconnect(rc); | |
360 | } | |
361 | } | |
362 | ||
363 | static unsigned int | |
364 | timeout_CONNECTING(const struct rconn *rc) | |
365 | { | |
366 | return MAX(1, rc->backoff); | |
367 | } | |
368 | ||
369 | static void | |
370 | run_CONNECTING(struct rconn *rc) | |
371 | { | |
372 | int retval = vconn_connect(rc->vconn); | |
373 | if (!retval) { | |
374 | VLOG_INFO("%s: connected", rc->name); | |
375 | rc->n_successful_connections++; | |
376 | state_transition(rc, S_ACTIVE); | |
377 | rc->last_connected = rc->state_entered; | |
378 | } else if (retval != EAGAIN) { | |
379 | VLOG_INFO("%s: connection failed (%s)", rc->name, strerror(retval)); | |
380 | disconnect(rc, retval); | |
381 | } else if (timed_out(rc)) { | |
382 | VLOG_INFO("%s: connection timed out", rc->name); | |
383 | rc->backoff_deadline = TIME_MAX; /* Prevent resetting backoff. */ | |
384 | disconnect(rc, 0); | |
385 | } | |
386 | } | |
387 | ||
388 | static void | |
389 | do_tx_work(struct rconn *rc) | |
390 | { | |
391 | if (!rc->txq.n) { | |
392 | return; | |
393 | } | |
394 | while (rc->txq.n > 0) { | |
395 | int error = try_send(rc); | |
396 | if (error) { | |
397 | break; | |
398 | } | |
399 | } | |
400 | if (!rc->txq.n) { | |
401 | poll_immediate_wake(); | |
402 | } | |
403 | } | |
404 | ||
405 | static unsigned int | |
406 | timeout_ACTIVE(const struct rconn *rc) | |
407 | { | |
408 | if (rc->probe_interval) { | |
409 | unsigned int base = MAX(rc->last_received, rc->state_entered); | |
410 | unsigned int arg = base + rc->probe_interval - rc->state_entered; | |
411 | return arg; | |
412 | } | |
413 | return UINT_MAX; | |
414 | } | |
415 | ||
416 | static void | |
417 | run_ACTIVE(struct rconn *rc) | |
418 | { | |
419 | if (timed_out(rc)) { | |
420 | unsigned int base = MAX(rc->last_received, rc->state_entered); | |
421 | VLOG_DBG("%s: idle %u seconds, sending inactivity probe", | |
422 | rc->name, (unsigned int) (time_now() - base)); | |
423 | ||
424 | /* Ordering is important here: rconn_send() can transition to BACKOFF, | |
425 | * and we don't want to transition back to IDLE if so, because then we | |
426 | * can end up queuing a packet with vconn == NULL and then *boom*. */ | |
427 | state_transition(rc, S_IDLE); | |
428 | rconn_send(rc, make_echo_request(), NULL); | |
429 | return; | |
430 | } | |
431 | ||
432 | do_tx_work(rc); | |
433 | } | |
434 | ||
435 | static unsigned int | |
436 | timeout_IDLE(const struct rconn *rc) | |
437 | { | |
438 | return rc->probe_interval; | |
439 | } | |
440 | ||
441 | static void | |
442 | run_IDLE(struct rconn *rc) | |
443 | { | |
444 | if (timed_out(rc)) { | |
445 | question_connectivity(rc); | |
446 | VLOG_ERR("%s: no response to inactivity probe after %u " | |
447 | "seconds, disconnecting", | |
448 | rc->name, elapsed_in_this_state(rc)); | |
449 | disconnect(rc, 0); | |
450 | } else { | |
451 | do_tx_work(rc); | |
452 | } | |
453 | } | |
454 | ||
455 | /* Performs whatever activities are necessary to maintain 'rc': if 'rc' is | |
456 | * disconnected, attempts to (re)connect, backing off as necessary; if 'rc' is | |
457 | * connected, attempts to send packets in the send queue, if any. */ | |
458 | void | |
459 | rconn_run(struct rconn *rc) | |
460 | { | |
461 | int old_state; | |
462 | do { | |
463 | old_state = rc->state; | |
464 | switch (rc->state) { | |
465 | #define STATE(NAME, VALUE) case S_##NAME: run_##NAME(rc); break; | |
466 | STATES | |
467 | #undef STATE | |
468 | default: | |
469 | NOT_REACHED(); | |
470 | } | |
471 | } while (rc->state != old_state); | |
472 | } | |
473 | ||
474 | /* Causes the next call to poll_block() to wake up when rconn_run() should be | |
475 | * called on 'rc'. */ | |
476 | void | |
477 | rconn_run_wait(struct rconn *rc) | |
478 | { | |
479 | unsigned int timeo = timeout(rc); | |
480 | if (timeo != UINT_MAX) { | |
481 | unsigned int expires = sat_add(rc->state_entered, timeo); | |
482 | unsigned int remaining = sat_sub(expires, time_now()); | |
483 | poll_timer_wait(sat_mul(remaining, 1000)); | |
484 | } | |
485 | ||
486 | if ((rc->state & (S_ACTIVE | S_IDLE)) && rc->txq.n) { | |
487 | vconn_wait(rc->vconn, WAIT_SEND); | |
488 | } | |
489 | } | |
490 | ||
491 | /* Attempts to receive a packet from 'rc'. If successful, returns the packet; | |
492 | * otherwise, returns a null pointer. The caller is responsible for freeing | |
493 | * the packet (with ofpbuf_delete()). */ | |
494 | struct ofpbuf * | |
495 | rconn_recv(struct rconn *rc) | |
496 | { | |
497 | if (rc->state & (S_ACTIVE | S_IDLE)) { | |
498 | struct ofpbuf *buffer; | |
499 | int error = vconn_recv(rc->vconn, &buffer); | |
500 | if (!error) { | |
501 | copy_to_monitor(rc, buffer); | |
7778bd15 | 502 | if (rc->probably_admitted || is_admitted_msg(buffer) |
064af421 BP |
503 | || time_now() - rc->last_connected >= 30) { |
504 | rc->probably_admitted = true; | |
505 | rc->last_admitted = time_now(); | |
506 | } | |
507 | rc->last_received = time_now(); | |
508 | rc->packets_received++; | |
509 | if (rc->state == S_IDLE) { | |
510 | state_transition(rc, S_ACTIVE); | |
511 | } | |
512 | return buffer; | |
513 | } else if (error != EAGAIN) { | |
514 | disconnect(rc, error); | |
515 | } | |
516 | } | |
517 | return NULL; | |
518 | } | |
519 | ||
520 | /* Causes the next call to poll_block() to wake up when a packet may be ready | |
521 | * to be received by vconn_recv() on 'rc'. */ | |
522 | void | |
523 | rconn_recv_wait(struct rconn *rc) | |
524 | { | |
525 | if (rc->vconn) { | |
526 | vconn_wait(rc->vconn, WAIT_RECV); | |
527 | } | |
528 | } | |
529 | ||
530 | /* Sends 'b' on 'rc'. Returns 0 if successful (in which case 'b' is | |
531 | * destroyed), or ENOTCONN if 'rc' is not currently connected (in which case | |
532 | * the caller retains ownership of 'b'). | |
533 | * | |
534 | * If 'counter' is non-null, then 'counter' will be incremented while the | |
535 | * packet is in flight, then decremented when it has been sent (or discarded | |
536 | * due to disconnection). Because 'b' may be sent (or discarded) before this | |
537 | * function returns, the caller may not be able to observe any change in | |
538 | * 'counter'. | |
539 | * | |
540 | * There is no rconn_send_wait() function: an rconn has a send queue that it | |
541 | * takes care of sending if you call rconn_run(), which will have the side | |
542 | * effect of waking up poll_block(). */ | |
543 | int | |
544 | rconn_send(struct rconn *rc, struct ofpbuf *b, | |
545 | struct rconn_packet_counter *counter) | |
546 | { | |
547 | if (rconn_is_connected(rc)) { | |
548 | COVERAGE_INC(rconn_queued); | |
549 | copy_to_monitor(rc, b); | |
d45e9c65 | 550 | b->private_p = counter; |
064af421 BP |
551 | if (counter) { |
552 | rconn_packet_counter_inc(counter); | |
553 | } | |
554 | queue_push_tail(&rc->txq, b); | |
555 | ||
556 | /* If the queue was empty before we added 'b', try to send some | |
557 | * packets. (But if the queue had packets in it, it's because the | |
558 | * vconn is backlogged and there's no point in stuffing more into it | |
559 | * now. We'll get back to that in rconn_run().) */ | |
560 | if (rc->txq.n == 1) { | |
561 | try_send(rc); | |
562 | } | |
563 | return 0; | |
564 | } else { | |
565 | return ENOTCONN; | |
566 | } | |
567 | } | |
568 | ||
569 | /* Sends 'b' on 'rc'. Increments 'counter' while the packet is in flight; it | |
570 | * will be decremented when it has been sent (or discarded due to | |
571 | * disconnection). Returns 0 if successful, EAGAIN if 'counter->n' is already | |
572 | * at least as large as 'queue_limit', or ENOTCONN if 'rc' is not currently | |
573 | * connected. Regardless of return value, 'b' is destroyed. | |
574 | * | |
575 | * Because 'b' may be sent (or discarded) before this function returns, the | |
576 | * caller may not be able to observe any change in 'counter'. | |
577 | * | |
578 | * There is no rconn_send_wait() function: an rconn has a send queue that it | |
579 | * takes care of sending if you call rconn_run(), which will have the side | |
580 | * effect of waking up poll_block(). */ | |
581 | int | |
582 | rconn_send_with_limit(struct rconn *rc, struct ofpbuf *b, | |
583 | struct rconn_packet_counter *counter, int queue_limit) | |
584 | { | |
585 | int retval; | |
586 | retval = counter->n >= queue_limit ? EAGAIN : rconn_send(rc, b, counter); | |
587 | if (retval) { | |
588 | COVERAGE_INC(rconn_overflow); | |
589 | ofpbuf_delete(b); | |
590 | } | |
591 | return retval; | |
592 | } | |
593 | ||
594 | /* Returns the total number of packets successfully sent on the underlying | |
595 | * vconn. A packet is not counted as sent while it is still queued in the | |
596 | * rconn, only when it has been successfuly passed to the vconn. */ | |
597 | unsigned int | |
598 | rconn_packets_sent(const struct rconn *rc) | |
599 | { | |
600 | return rc->packets_sent; | |
601 | } | |
602 | ||
603 | /* Adds 'vconn' to 'rc' as a monitoring connection, to which all messages sent | |
604 | * and received on 'rconn' will be copied. 'rc' takes ownership of 'vconn'. */ | |
605 | void | |
606 | rconn_add_monitor(struct rconn *rc, struct vconn *vconn) | |
607 | { | |
608 | if (rc->n_monitors < ARRAY_SIZE(rc->monitors)) { | |
609 | VLOG_INFO("new monitor connection from %s", vconn_get_name(vconn)); | |
610 | rc->monitors[rc->n_monitors++] = vconn; | |
611 | } else { | |
612 | VLOG_DBG("too many monitor connections, discarding %s", | |
613 | vconn_get_name(vconn)); | |
614 | vconn_close(vconn); | |
615 | } | |
616 | } | |
617 | ||
618 | /* Returns 'rc''s name (the 'name' argument passed to rconn_new()). */ | |
619 | const char * | |
620 | rconn_get_name(const struct rconn *rc) | |
621 | { | |
622 | return rc->name; | |
623 | } | |
624 | ||
625 | /* Returns true if 'rconn' is connected or in the process of reconnecting, | |
626 | * false if 'rconn' is disconnected and will not reconnect on its own. */ | |
627 | bool | |
628 | rconn_is_alive(const struct rconn *rconn) | |
629 | { | |
630 | return rconn->state != S_VOID; | |
631 | } | |
632 | ||
633 | /* Returns true if 'rconn' is connected, false otherwise. */ | |
634 | bool | |
635 | rconn_is_connected(const struct rconn *rconn) | |
636 | { | |
637 | return is_connected_state(rconn->state); | |
638 | } | |
639 | ||
7778bd15 BP |
640 | /* Returns true if 'rconn' is connected and thought to have been accepted by |
641 | * the peer's admission-control policy. */ | |
642 | bool | |
643 | rconn_is_admitted(const struct rconn *rconn) | |
644 | { | |
645 | return (rconn_is_connected(rconn) | |
646 | && rconn->last_admitted >= rconn->last_connected); | |
647 | } | |
648 | ||
649 | /* Returns 0 if 'rconn' is currently connected and considered to have been | |
650 | * accepted by the peer's admission-control policy, otherwise the number of | |
651 | * seconds since 'rconn' was last in such a state. */ | |
064af421 BP |
652 | int |
653 | rconn_failure_duration(const struct rconn *rconn) | |
654 | { | |
7778bd15 | 655 | return rconn_is_admitted(rconn) ? 0 : time_now() - rconn->last_admitted; |
064af421 BP |
656 | } |
657 | ||
19d1ab55 BP |
658 | /* Returns the IP address of the peer, or 0 if the peer's IP address is not |
659 | * known. */ | |
064af421 | 660 | uint32_t |
193456d5 | 661 | rconn_get_remote_ip(const struct rconn *rconn) |
064af421 | 662 | { |
19d1ab55 | 663 | return rconn->remote_ip; |
193456d5 JP |
664 | } |
665 | ||
19d1ab55 BP |
666 | /* Returns the transport port of the peer, or 0 if the peer's port is not |
667 | * known. */ | |
193456d5 JP |
668 | uint16_t |
669 | rconn_get_remote_port(const struct rconn *rconn) | |
670 | { | |
19d1ab55 | 671 | return rconn->remote_port; |
193456d5 JP |
672 | } |
673 | ||
674 | /* Returns the IP address used to connect to the peer, or 0 if the | |
675 | * connection is not an IP-based protocol or if its IP address is not | |
676 | * known. */ | |
677 | uint32_t | |
678 | rconn_get_local_ip(const struct rconn *rconn) | |
679 | { | |
19d1ab55 | 680 | return rconn->local_ip; |
193456d5 JP |
681 | } |
682 | ||
683 | /* Returns the transport port used to connect to the peer, or 0 if the | |
684 | * connection does not contain a port or if the port is not known. */ | |
685 | uint16_t | |
686 | rconn_get_local_port(const struct rconn *rconn) | |
687 | { | |
688 | return rconn->vconn ? vconn_get_local_port(rconn->vconn) : 0; | |
064af421 BP |
689 | } |
690 | ||
691 | /* If 'rconn' can't connect to the peer, it could be for any number of reasons. | |
692 | * Usually, one would assume it is because the peer is not running or because | |
693 | * the network is partitioned. But it could also be because the network | |
694 | * topology has changed, in which case the upper layer will need to reassess it | |
695 | * (in particular, obtain a new IP address via DHCP and find the new location | |
696 | * of the controller). When this appears that this might be the case, this | |
697 | * function returns true. It also clears the questionability flag and prevents | |
698 | * it from being set again for some time. */ | |
699 | bool | |
700 | rconn_is_connectivity_questionable(struct rconn *rconn) | |
701 | { | |
702 | bool questionable = rconn->questionable_connectivity; | |
703 | rconn->questionable_connectivity = false; | |
704 | return questionable; | |
705 | } | |
706 | ||
707 | /* Returns the total number of packets successfully received by the underlying | |
708 | * vconn. */ | |
709 | unsigned int | |
710 | rconn_packets_received(const struct rconn *rc) | |
711 | { | |
712 | return rc->packets_received; | |
713 | } | |
714 | ||
715 | /* Returns a string representing the internal state of 'rc'. The caller must | |
716 | * not modify or free the string. */ | |
717 | const char * | |
718 | rconn_get_state(const struct rconn *rc) | |
719 | { | |
720 | return state_name(rc->state); | |
721 | } | |
722 | ||
723 | /* Returns the number of connection attempts made by 'rc', including any | |
724 | * ongoing attempt that has not yet succeeded or failed. */ | |
725 | unsigned int | |
726 | rconn_get_attempted_connections(const struct rconn *rc) | |
727 | { | |
728 | return rc->n_attempted_connections; | |
729 | } | |
730 | ||
731 | /* Returns the number of successful connection attempts made by 'rc'. */ | |
732 | unsigned int | |
733 | rconn_get_successful_connections(const struct rconn *rc) | |
734 | { | |
735 | return rc->n_successful_connections; | |
736 | } | |
737 | ||
738 | /* Returns the time at which the last successful connection was made by | |
739 | * 'rc'. */ | |
740 | time_t | |
741 | rconn_get_last_connection(const struct rconn *rc) | |
742 | { | |
743 | return rc->last_connected; | |
744 | } | |
745 | ||
7df824b7 BP |
746 | /* Returns the time at which the last OpenFlow message was received by 'rc'. |
747 | * If no packets have been received on 'rc', returns the time at which 'rc' | |
748 | * was created. */ | |
749 | time_t | |
750 | rconn_get_last_received(const struct rconn *rc) | |
751 | { | |
752 | return rc->last_received; | |
753 | } | |
754 | ||
064af421 BP |
755 | /* Returns the time at which 'rc' was created. */ |
756 | time_t | |
757 | rconn_get_creation_time(const struct rconn *rc) | |
758 | { | |
759 | return rc->creation_time; | |
760 | } | |
761 | ||
762 | /* Returns the approximate number of seconds that 'rc' has been connected. */ | |
763 | unsigned long int | |
764 | rconn_get_total_time_connected(const struct rconn *rc) | |
765 | { | |
766 | return (rc->total_time_connected | |
767 | + (rconn_is_connected(rc) ? elapsed_in_this_state(rc) : 0)); | |
768 | } | |
769 | ||
770 | /* Returns the current amount of backoff, in seconds. This is the amount of | |
771 | * time after which the rconn will transition from BACKOFF to CONNECTING. */ | |
772 | int | |
773 | rconn_get_backoff(const struct rconn *rc) | |
774 | { | |
775 | return rc->backoff; | |
776 | } | |
777 | ||
778 | /* Returns the number of seconds spent in this state so far. */ | |
779 | unsigned int | |
780 | rconn_get_state_elapsed(const struct rconn *rc) | |
781 | { | |
782 | return elapsed_in_this_state(rc); | |
783 | } | |
784 | ||
785 | /* Returns 'rc''s current connection sequence number, a number that changes | |
786 | * every time that 'rconn' connects or disconnects. */ | |
787 | unsigned int | |
788 | rconn_get_connection_seqno(const struct rconn *rc) | |
789 | { | |
790 | return rc->seqno; | |
791 | } | |
792 | \f | |
793 | struct rconn_packet_counter * | |
794 | rconn_packet_counter_create(void) | |
795 | { | |
796 | struct rconn_packet_counter *c = xmalloc(sizeof *c); | |
797 | c->n = 0; | |
798 | c->ref_cnt = 1; | |
799 | return c; | |
800 | } | |
801 | ||
802 | void | |
803 | rconn_packet_counter_destroy(struct rconn_packet_counter *c) | |
804 | { | |
805 | if (c) { | |
806 | assert(c->ref_cnt > 0); | |
807 | if (!--c->ref_cnt && !c->n) { | |
808 | free(c); | |
809 | } | |
810 | } | |
811 | } | |
812 | ||
813 | void | |
814 | rconn_packet_counter_inc(struct rconn_packet_counter *c) | |
815 | { | |
816 | c->n++; | |
817 | } | |
818 | ||
819 | void | |
820 | rconn_packet_counter_dec(struct rconn_packet_counter *c) | |
821 | { | |
822 | assert(c->n > 0); | |
823 | if (!--c->n && !c->ref_cnt) { | |
824 | free(c); | |
825 | } | |
826 | } | |
827 | \f | |
19d1ab55 BP |
828 | /* Set the name of the remote vconn to 'name' and clear out the cached IP |
829 | * address and port information, since changing the name also likely changes | |
830 | * these values. */ | |
831 | static void | |
832 | set_vconn_name(struct rconn *rc, const char *name) | |
833 | { | |
834 | free(rc->name); | |
835 | rc->name = xstrdup(name); | |
836 | rc->local_ip = 0; | |
837 | rc->remote_ip = 0; | |
838 | rc->remote_port = 0; | |
839 | } | |
840 | ||
064af421 BP |
841 | /* Tries to send a packet from 'rc''s send buffer. Returns 0 if successful, |
842 | * otherwise a positive errno value. */ | |
843 | static int | |
844 | try_send(struct rconn *rc) | |
845 | { | |
846 | int retval = 0; | |
847 | struct ofpbuf *next = rc->txq.head->next; | |
d45e9c65 | 848 | struct rconn_packet_counter *counter = rc->txq.head->private_p; |
064af421 BP |
849 | retval = vconn_send(rc->vconn, rc->txq.head); |
850 | if (retval) { | |
851 | if (retval != EAGAIN) { | |
852 | disconnect(rc, retval); | |
853 | } | |
854 | return retval; | |
855 | } | |
856 | COVERAGE_INC(rconn_sent); | |
857 | rc->packets_sent++; | |
858 | if (counter) { | |
859 | rconn_packet_counter_dec(counter); | |
860 | } | |
861 | queue_advance_head(&rc->txq, next); | |
862 | return 0; | |
863 | } | |
864 | ||
865 | /* Disconnects 'rc'. 'error' is used only for logging purposes. If it is | |
866 | * nonzero, then it should be EOF to indicate the connection was closed by the | |
867 | * peer in a normal fashion or a positive errno value. */ | |
868 | static void | |
869 | disconnect(struct rconn *rc, int error) | |
870 | { | |
871 | if (rc->reliable) { | |
872 | time_t now = time_now(); | |
873 | ||
874 | if (rc->state & (S_CONNECTING | S_ACTIVE | S_IDLE)) { | |
875 | if (error > 0) { | |
876 | VLOG_WARN("%s: connection dropped (%s)", | |
877 | rc->name, strerror(error)); | |
878 | } else if (error == EOF) { | |
879 | if (rc->reliable) { | |
880 | VLOG_INFO("%s: connection closed by peer", rc->name); | |
881 | } | |
882 | } else { | |
883 | VLOG_INFO("%s: connection dropped", rc->name); | |
884 | } | |
885 | vconn_close(rc->vconn); | |
886 | rc->vconn = NULL; | |
887 | flush_queue(rc); | |
888 | } | |
889 | ||
890 | if (now >= rc->backoff_deadline) { | |
891 | rc->backoff = 1; | |
892 | } else { | |
893 | rc->backoff = MIN(rc->max_backoff, MAX(1, 2 * rc->backoff)); | |
894 | VLOG_INFO("%s: waiting %d seconds before reconnect\n", | |
895 | rc->name, rc->backoff); | |
896 | } | |
897 | rc->backoff_deadline = now + rc->backoff; | |
898 | state_transition(rc, S_BACKOFF); | |
899 | if (now - rc->last_connected > 60) { | |
900 | question_connectivity(rc); | |
901 | } | |
902 | } else { | |
903 | rconn_disconnect(rc); | |
904 | } | |
905 | } | |
906 | ||
907 | /* Drops all the packets from 'rc''s send queue and decrements their queue | |
908 | * counts. */ | |
909 | static void | |
910 | flush_queue(struct rconn *rc) | |
911 | { | |
912 | if (!rc->txq.n) { | |
913 | return; | |
914 | } | |
915 | while (rc->txq.n > 0) { | |
916 | struct ofpbuf *b = queue_pop_head(&rc->txq); | |
d45e9c65 | 917 | struct rconn_packet_counter *counter = b->private_p; |
064af421 BP |
918 | if (counter) { |
919 | rconn_packet_counter_dec(counter); | |
920 | } | |
921 | COVERAGE_INC(rconn_discarded); | |
922 | ofpbuf_delete(b); | |
923 | } | |
924 | poll_immediate_wake(); | |
925 | } | |
926 | ||
927 | static unsigned int | |
928 | elapsed_in_this_state(const struct rconn *rc) | |
929 | { | |
930 | return time_now() - rc->state_entered; | |
931 | } | |
932 | ||
933 | static unsigned int | |
934 | timeout(const struct rconn *rc) | |
935 | { | |
936 | switch (rc->state) { | |
937 | #define STATE(NAME, VALUE) case S_##NAME: return timeout_##NAME(rc); | |
938 | STATES | |
939 | #undef STATE | |
940 | default: | |
941 | NOT_REACHED(); | |
942 | } | |
943 | } | |
944 | ||
945 | static bool | |
946 | timed_out(const struct rconn *rc) | |
947 | { | |
948 | return time_now() >= sat_add(rc->state_entered, timeout(rc)); | |
949 | } | |
950 | ||
951 | static void | |
952 | state_transition(struct rconn *rc, enum state state) | |
953 | { | |
954 | rc->seqno += (rc->state == S_ACTIVE) != (state == S_ACTIVE); | |
955 | if (is_connected_state(state) && !is_connected_state(rc->state)) { | |
956 | rc->probably_admitted = false; | |
957 | } | |
958 | if (rconn_is_connected(rc)) { | |
959 | rc->total_time_connected += elapsed_in_this_state(rc); | |
960 | } | |
961 | VLOG_DBG("%s: entering %s", rc->name, state_name(state)); | |
962 | rc->state = state; | |
963 | rc->state_entered = time_now(); | |
964 | } | |
965 | ||
966 | static void | |
967 | question_connectivity(struct rconn *rc) | |
968 | { | |
969 | time_t now = time_now(); | |
970 | if (now - rc->last_questioned > 60) { | |
971 | rc->questionable_connectivity = true; | |
972 | rc->last_questioned = now; | |
973 | } | |
974 | } | |
975 | ||
976 | static void | |
977 | copy_to_monitor(struct rconn *rc, const struct ofpbuf *b) | |
978 | { | |
979 | struct ofpbuf *clone = NULL; | |
980 | int retval; | |
981 | size_t i; | |
982 | ||
983 | for (i = 0; i < rc->n_monitors; ) { | |
984 | struct vconn *vconn = rc->monitors[i]; | |
985 | ||
986 | if (!clone) { | |
987 | clone = ofpbuf_clone(b); | |
988 | } | |
989 | retval = vconn_send(vconn, clone); | |
990 | if (!retval) { | |
991 | clone = NULL; | |
992 | } else if (retval != EAGAIN) { | |
993 | VLOG_DBG("%s: closing monitor connection to %s: %s", | |
994 | rconn_get_name(rc), vconn_get_name(vconn), | |
995 | strerror(retval)); | |
996 | rc->monitors[i] = rc->monitors[--rc->n_monitors]; | |
997 | continue; | |
998 | } | |
999 | i++; | |
1000 | } | |
1001 | ofpbuf_delete(clone); | |
1002 | } | |
1003 | ||
1004 | static bool | |
1005 | is_connected_state(enum state state) | |
1006 | { | |
1007 | return (state & (S_ACTIVE | S_IDLE)) != 0; | |
1008 | } | |
1009 | ||
1010 | static bool | |
1011 | is_admitted_msg(const struct ofpbuf *b) | |
1012 | { | |
1013 | struct ofp_header *oh = b->data; | |
1014 | uint8_t type = oh->type; | |
1015 | return !(type < 32 | |
1016 | && (1u << type) & ((1u << OFPT_HELLO) | | |
1017 | (1u << OFPT_ERROR) | | |
1018 | (1u << OFPT_ECHO_REQUEST) | | |
1019 | (1u << OFPT_ECHO_REPLY) | | |
1020 | (1u << OFPT_VENDOR) | | |
1021 | (1u << OFPT_FEATURES_REQUEST) | | |
1022 | (1u << OFPT_FEATURES_REPLY) | | |
1023 | (1u << OFPT_GET_CONFIG_REQUEST) | | |
1024 | (1u << OFPT_GET_CONFIG_REPLY) | | |
1025 | (1u << OFPT_SET_CONFIG))); | |
1026 | } |