2 * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira, Inc.
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:
8 * http://www.apache.org/licenses/LICENSE-2.0
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.
28 #include "openflow/openflow.h"
29 #include "poll-loop.h"
36 VLOG_DEFINE_THIS_MODULE(rconn);
38 COVERAGE_DEFINE(rconn_discarded);
39 COVERAGE_DEFINE(rconn_overflow);
40 COVERAGE_DEFINE(rconn_queued);
41 COVERAGE_DEFINE(rconn_sent);
45 STATE(BACKOFF, 1 << 1) \
46 STATE(CONNECTING, 1 << 2) \
47 STATE(ACTIVE, 1 << 3) \
50 #define STATE(NAME, VALUE) S_##NAME = VALUE,
56 state_name(enum state state)
59 #define STATE(NAME, VALUE) case S_##NAME: return #NAME;
66 /* A reliable connection to an OpenFlow switch or controller.
68 * See the large comment in rconn.h for more information. */
74 char *name; /* Human-readable descriptive name. */
75 char *target; /* vconn name, passed to vconn_open(). */
78 struct list txq; /* Contains "struct ofpbuf"s. */
82 time_t backoff_deadline;
84 time_t last_connected;
85 time_t last_disconnected;
86 unsigned int packets_sent;
90 /* In S_ACTIVE and S_IDLE, probably_admitted reports whether we believe
91 * that the peer has made a (positive) admission control decision on our
92 * connection. If we have not yet been (probably) admitted, then the
93 * connection does not reset the timer used for deciding whether the switch
94 * should go into fail-open mode.
96 * last_admitted reports the last time we believe such a positive admission
97 * control decision was made. */
98 bool probably_admitted;
101 /* These values are simply for statistics reporting, not used directly by
102 * anything internal to the rconn (or ofproto for that matter). */
103 unsigned int packets_received;
104 unsigned int n_attempted_connections, n_successful_connections;
105 time_t creation_time;
106 unsigned long int total_time_connected;
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
112 int probe_interval; /* Secs of inactivity before sending probe. */
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.
121 * We don't cache the local port, because that changes from one connection
122 * attempt to the next. */
123 ovs_be32 local_ip, remote_ip;
124 ovs_be16 remote_port;
127 /* Messages sent or received are copied to the monitor connections. */
128 #define MAX_MONITORS 8
129 struct vconn *monitors[8];
133 static unsigned int elapsed_in_this_state(const struct rconn *);
134 static unsigned int timeout(const struct rconn *);
135 static bool timed_out(const struct rconn *);
136 static void state_transition(struct rconn *, enum state);
137 static void rconn_set_target__(struct rconn *,
138 const char *target, const char *name);
139 static int try_send(struct rconn *);
140 static void reconnect(struct rconn *);
141 static void report_error(struct rconn *, int error);
142 static void disconnect(struct rconn *, int error);
143 static void flush_queue(struct rconn *);
144 static void copy_to_monitor(struct rconn *, const struct ofpbuf *);
145 static bool is_connected_state(enum state);
146 static bool is_admitted_msg(const struct ofpbuf *);
147 static bool rconn_logging_connection_attempts__(const struct rconn *);
149 /* Creates and returns a new rconn.
151 * 'probe_interval' is a number of seconds. If the interval passes once
152 * without an OpenFlow message being received from the peer, the rconn sends
153 * out an "echo request" message. If the interval passes again without a
154 * message being received, the rconn disconnects and re-connects to the peer.
155 * Setting 'probe_interval' to 0 disables this behavior.
157 * 'max_backoff' is the maximum number of seconds between attempts to connect
158 * to the peer. The actual interval starts at 1 second and doubles on each
159 * failure until it reaches 'max_backoff'. If 0 is specified, the default of
162 * The new rconn is initially unconnected. Use rconn_connect() or
163 * rconn_connect_unreliably() to connect it. */
165 rconn_create(int probe_interval, int max_backoff, uint8_t dscp)
167 struct rconn *rc = xzalloc(sizeof *rc);
170 rc->state_entered = time_now();
173 rc->name = xstrdup("void");
174 rc->target = xstrdup("void");
175 rc->reliable = false;
180 rc->max_backoff = max_backoff ? max_backoff : 8;
181 rc->backoff_deadline = TIME_MIN;
182 rc->last_received = time_now();
183 rc->last_connected = TIME_MIN;
184 rc->last_disconnected = TIME_MIN;
187 rc->packets_sent = 0;
189 rc->probably_admitted = false;
190 rc->last_admitted = time_now();
192 rc->packets_received = 0;
193 rc->n_attempted_connections = 0;
194 rc->n_successful_connections = 0;
195 rc->creation_time = time_now();
196 rc->total_time_connected = 0;
198 rconn_set_probe_interval(rc, probe_interval);
199 rconn_set_dscp(rc, dscp);
207 rconn_set_max_backoff(struct rconn *rc, int max_backoff)
209 rc->max_backoff = MAX(1, max_backoff);
210 if (rc->state == S_BACKOFF && rc->backoff > max_backoff) {
211 rc->backoff = max_backoff;
212 if (rc->backoff_deadline > time_now() + max_backoff) {
213 rc->backoff_deadline = time_now() + max_backoff;
219 rconn_get_max_backoff(const struct rconn *rc)
221 return rc->max_backoff;
225 rconn_set_dscp(struct rconn *rc, uint8_t dscp)
231 rconn_get_dscp(const struct rconn *rc)
237 rconn_set_probe_interval(struct rconn *rc, int probe_interval)
239 rc->probe_interval = probe_interval ? MAX(5, probe_interval) : 0;
243 rconn_get_probe_interval(const struct rconn *rc)
245 return rc->probe_interval;
248 /* Drops any existing connection on 'rc', then sets up 'rc' to connect to
249 * 'target' and reconnect as needed. 'target' should be a remote OpenFlow
250 * target in a form acceptable to vconn_open().
252 * If 'name' is nonnull, then it is used in log messages in place of 'target'.
253 * It should presumably give more information to a human reader than 'target',
254 * but it need not be acceptable to vconn_open(). */
256 rconn_connect(struct rconn *rc, const char *target, const char *name)
258 rconn_disconnect(rc);
259 rconn_set_target__(rc, target, name);
264 /* Drops any existing connection on 'rc', then configures 'rc' to use
265 * 'vconn'. If the connection on 'vconn' drops, 'rc' will not reconnect on it
268 * By default, the target obtained from vconn_get_name(vconn) is used in log
269 * messages. If 'name' is nonnull, then it is used instead. It should
270 * presumably give more information to a human reader than the target, but it
271 * need not be acceptable to vconn_open(). */
273 rconn_connect_unreliably(struct rconn *rc,
274 struct vconn *vconn, const char *name)
276 assert(vconn != NULL);
277 rconn_disconnect(rc);
278 rconn_set_target__(rc, vconn_get_name(vconn), name);
279 rc->reliable = false;
281 rc->last_connected = time_now();
282 state_transition(rc, S_ACTIVE);
285 /* If 'rc' is connected, forces it to drop the connection and reconnect. */
287 rconn_reconnect(struct rconn *rc)
289 if (rc->state & (S_ACTIVE | S_IDLE)) {
290 VLOG_INFO("%s: disconnecting", rc->name);
296 rconn_disconnect(struct rconn *rc)
298 if (rc->state != S_VOID) {
300 vconn_close(rc->vconn);
303 rconn_set_target__(rc, "void", NULL);
304 rc->reliable = false;
307 rc->backoff_deadline = TIME_MIN;
309 state_transition(rc, S_VOID);
313 /* Disconnects 'rc' and frees the underlying storage. */
315 rconn_destroy(struct rconn *rc)
322 vconn_close(rc->vconn);
324 ofpbuf_list_delete(&rc->txq);
325 for (i = 0; i < rc->n_monitors; i++) {
326 vconn_close(rc->monitors[i]);
333 timeout_VOID(const struct rconn *rc OVS_UNUSED)
339 run_VOID(struct rconn *rc OVS_UNUSED)
345 reconnect(struct rconn *rc)
349 if (rconn_logging_connection_attempts__(rc)) {
350 VLOG_INFO("%s: connecting...", rc->name);
352 rc->n_attempted_connections++;
353 retval = vconn_open(rc->target, OFP10_VERSION, &rc->vconn, rc->dscp);
355 rc->remote_ip = vconn_get_remote_ip(rc->vconn);
356 rc->local_ip = vconn_get_local_ip(rc->vconn);
357 rc->remote_port = vconn_get_remote_port(rc->vconn);
358 rc->backoff_deadline = time_now() + rc->backoff;
359 state_transition(rc, S_CONNECTING);
361 VLOG_WARN("%s: connection failed (%s)", rc->name, strerror(retval));
362 rc->backoff_deadline = TIME_MAX; /* Prevent resetting backoff. */
363 disconnect(rc, retval);
368 timeout_BACKOFF(const struct rconn *rc)
374 run_BACKOFF(struct rconn *rc)
382 timeout_CONNECTING(const struct rconn *rc)
384 return MAX(1, rc->backoff);
388 run_CONNECTING(struct rconn *rc)
390 int retval = vconn_connect(rc->vconn);
392 VLOG_INFO("%s: connected", rc->name);
393 rc->n_successful_connections++;
394 state_transition(rc, S_ACTIVE);
395 rc->last_connected = rc->state_entered;
396 } else if (retval != EAGAIN) {
397 if (rconn_logging_connection_attempts__(rc)) {
398 VLOG_INFO("%s: connection failed (%s)",
399 rc->name, strerror(retval));
401 disconnect(rc, retval);
402 } else if (timed_out(rc)) {
403 if (rconn_logging_connection_attempts__(rc)) {
404 VLOG_INFO("%s: connection timed out", rc->name);
406 rc->backoff_deadline = TIME_MAX; /* Prevent resetting backoff. */
407 disconnect(rc, ETIMEDOUT);
412 do_tx_work(struct rconn *rc)
414 if (list_is_empty(&rc->txq)) {
417 while (!list_is_empty(&rc->txq)) {
418 int error = try_send(rc);
423 if (list_is_empty(&rc->txq)) {
424 poll_immediate_wake();
429 timeout_ACTIVE(const struct rconn *rc)
431 if (rc->probe_interval) {
432 unsigned int base = MAX(rc->last_received, rc->state_entered);
433 unsigned int arg = base + rc->probe_interval - rc->state_entered;
440 run_ACTIVE(struct rconn *rc)
443 unsigned int base = MAX(rc->last_received, rc->state_entered);
444 VLOG_DBG("%s: idle %u seconds, sending inactivity probe",
445 rc->name, (unsigned int) (time_now() - base));
447 /* Ordering is important here: rconn_send() can transition to BACKOFF,
448 * and we don't want to transition back to IDLE if so, because then we
449 * can end up queuing a packet with vconn == NULL and then *boom*. */
450 state_transition(rc, S_IDLE);
451 rconn_send(rc, make_echo_request(), NULL);
459 timeout_IDLE(const struct rconn *rc)
461 return rc->probe_interval;
465 run_IDLE(struct rconn *rc)
468 VLOG_ERR("%s: no response to inactivity probe after %u "
469 "seconds, disconnecting",
470 rc->name, elapsed_in_this_state(rc));
471 disconnect(rc, ETIMEDOUT);
477 /* Performs whatever activities are necessary to maintain 'rc': if 'rc' is
478 * disconnected, attempts to (re)connect, backing off as necessary; if 'rc' is
479 * connected, attempts to send packets in the send queue, if any. */
481 rconn_run(struct rconn *rc)
487 vconn_run(rc->vconn);
489 for (i = 0; i < rc->n_monitors; i++) {
490 vconn_run(rc->monitors[i]);
494 old_state = rc->state;
496 #define STATE(NAME, VALUE) case S_##NAME: run_##NAME(rc); break;
502 } while (rc->state != old_state);
505 /* Causes the next call to poll_block() to wake up when rconn_run() should be
508 rconn_run_wait(struct rconn *rc)
514 vconn_run_wait(rc->vconn);
515 if ((rc->state & (S_ACTIVE | S_IDLE)) && !list_is_empty(&rc->txq)) {
516 vconn_wait(rc->vconn, WAIT_SEND);
519 for (i = 0; i < rc->n_monitors; i++) {
520 vconn_run_wait(rc->monitors[i]);
524 if (timeo != UINT_MAX) {
525 long long int expires = sat_add(rc->state_entered, timeo);
526 poll_timer_wait_until(expires * 1000);
530 /* Attempts to receive a packet from 'rc'. If successful, returns the packet;
531 * otherwise, returns a null pointer. The caller is responsible for freeing
532 * the packet (with ofpbuf_delete()). */
534 rconn_recv(struct rconn *rc)
536 if (rc->state & (S_ACTIVE | S_IDLE)) {
537 struct ofpbuf *buffer;
538 int error = vconn_recv(rc->vconn, &buffer);
540 copy_to_monitor(rc, buffer);
541 if (rc->probably_admitted || is_admitted_msg(buffer)
542 || time_now() - rc->last_connected >= 30) {
543 rc->probably_admitted = true;
544 rc->last_admitted = time_now();
546 rc->last_received = time_now();
547 rc->packets_received++;
548 if (rc->state == S_IDLE) {
549 state_transition(rc, S_ACTIVE);
552 } else if (error != EAGAIN) {
553 report_error(rc, error);
554 disconnect(rc, error);
560 /* Causes the next call to poll_block() to wake up when a packet may be ready
561 * to be received by vconn_recv() on 'rc'. */
563 rconn_recv_wait(struct rconn *rc)
566 vconn_wait(rc->vconn, WAIT_RECV);
570 /* Sends 'b' on 'rc'. Returns 0 if successful, or ENOTCONN if 'rc' is not
571 * currently connected. Takes ownership of 'b'.
573 * If 'counter' is non-null, then 'counter' will be incremented while the
574 * packet is in flight, then decremented when it has been sent (or discarded
575 * due to disconnection). Because 'b' may be sent (or discarded) before this
576 * function returns, the caller may not be able to observe any change in
579 * There is no rconn_send_wait() function: an rconn has a send queue that it
580 * takes care of sending if you call rconn_run(), which will have the side
581 * effect of waking up poll_block(). */
583 rconn_send(struct rconn *rc, struct ofpbuf *b,
584 struct rconn_packet_counter *counter)
586 if (rconn_is_connected(rc)) {
587 COVERAGE_INC(rconn_queued);
588 copy_to_monitor(rc, b);
589 b->private_p = counter;
591 rconn_packet_counter_inc(counter, b->size);
593 list_push_back(&rc->txq, &b->list_node);
595 /* If the queue was empty before we added 'b', try to send some
596 * packets. (But if the queue had packets in it, it's because the
597 * vconn is backlogged and there's no point in stuffing more into it
598 * now. We'll get back to that in rconn_run().) */
599 if (rc->txq.next == &b->list_node) {
609 /* Sends 'b' on 'rc'. Increments 'counter' while the packet is in flight; it
610 * will be decremented when it has been sent (or discarded due to
611 * disconnection). Returns 0 if successful, EAGAIN if 'counter->n' is already
612 * at least as large as 'queue_limit', or ENOTCONN if 'rc' is not currently
613 * connected. Regardless of return value, 'b' is destroyed.
615 * Because 'b' may be sent (or discarded) before this function returns, the
616 * caller may not be able to observe any change in 'counter'.
618 * There is no rconn_send_wait() function: an rconn has a send queue that it
619 * takes care of sending if you call rconn_run(), which will have the side
620 * effect of waking up poll_block(). */
622 rconn_send_with_limit(struct rconn *rc, struct ofpbuf *b,
623 struct rconn_packet_counter *counter, int queue_limit)
626 retval = (counter->n_packets >= queue_limit
628 : rconn_send(rc, b, counter));
630 COVERAGE_INC(rconn_overflow);
635 /* Returns the total number of packets successfully sent on the underlying
636 * vconn. A packet is not counted as sent while it is still queued in the
637 * rconn, only when it has been successfuly passed to the vconn. */
639 rconn_packets_sent(const struct rconn *rc)
641 return rc->packets_sent;
644 /* Adds 'vconn' to 'rc' as a monitoring connection, to which all messages sent
645 * and received on 'rconn' will be copied. 'rc' takes ownership of 'vconn'. */
647 rconn_add_monitor(struct rconn *rc, struct vconn *vconn)
649 if (rc->n_monitors < ARRAY_SIZE(rc->monitors)) {
650 VLOG_INFO("new monitor connection from %s", vconn_get_name(vconn));
651 rc->monitors[rc->n_monitors++] = vconn;
653 VLOG_DBG("too many monitor connections, discarding %s",
654 vconn_get_name(vconn));
659 /* Returns 'rc''s name. This is a name for human consumption, appropriate for
660 * use in log messages. It is not necessarily a name that may be passed
661 * directly to, e.g., vconn_open(). */
663 rconn_get_name(const struct rconn *rc)
668 /* Sets 'rc''s name to 'new_name'. */
670 rconn_set_name(struct rconn *rc, const char *new_name)
673 rc->name = xstrdup(new_name);
676 /* Returns 'rc''s target. This is intended to be a string that may be passed
677 * directly to, e.g., vconn_open(). */
679 rconn_get_target(const struct rconn *rc)
684 /* Returns true if 'rconn' is connected or in the process of reconnecting,
685 * false if 'rconn' is disconnected and will not reconnect on its own. */
687 rconn_is_alive(const struct rconn *rconn)
689 return rconn->state != S_VOID;
692 /* Returns true if 'rconn' is connected, false otherwise. */
694 rconn_is_connected(const struct rconn *rconn)
696 return is_connected_state(rconn->state);
699 /* Returns true if 'rconn' is connected and thought to have been accepted by
700 * the peer's admission-control policy. */
702 rconn_is_admitted(const struct rconn *rconn)
704 return (rconn_is_connected(rconn)
705 && rconn->last_admitted >= rconn->last_connected);
708 /* Returns 0 if 'rconn' is currently connected and considered to have been
709 * accepted by the peer's admission-control policy, otherwise the number of
710 * seconds since 'rconn' was last in such a state. */
712 rconn_failure_duration(const struct rconn *rconn)
714 return rconn_is_admitted(rconn) ? 0 : time_now() - rconn->last_admitted;
717 /* Returns the IP address of the peer, or 0 if the peer's IP address is not
720 rconn_get_remote_ip(const struct rconn *rconn)
722 return rconn->remote_ip;
725 /* Returns the transport port of the peer, or 0 if the peer's port is not
728 rconn_get_remote_port(const struct rconn *rconn)
730 return rconn->remote_port;
733 /* Returns the IP address used to connect to the peer, or 0 if the
734 * connection is not an IP-based protocol or if its IP address is not
737 rconn_get_local_ip(const struct rconn *rconn)
739 return rconn->local_ip;
742 /* Returns the transport port used to connect to the peer, or 0 if the
743 * connection does not contain a port or if the port is not known. */
745 rconn_get_local_port(const struct rconn *rconn)
747 return rconn->vconn ? vconn_get_local_port(rconn->vconn) : 0;
750 /* Returns the OpenFlow version negotiated with the peer, or -1 if there is
751 * currently no connection or if version negotiation is not yet complete. */
753 rconn_get_version(const struct rconn *rconn)
755 return rconn->vconn ? vconn_get_version(rconn->vconn) : -1;
758 /* Returns the total number of packets successfully received by the underlying
761 rconn_packets_received(const struct rconn *rc)
763 return rc->packets_received;
766 /* Returns a string representing the internal state of 'rc'. The caller must
767 * not modify or free the string. */
769 rconn_get_state(const struct rconn *rc)
771 return state_name(rc->state);
774 /* Returns the number of connection attempts made by 'rc', including any
775 * ongoing attempt that has not yet succeeded or failed. */
777 rconn_get_attempted_connections(const struct rconn *rc)
779 return rc->n_attempted_connections;
782 /* Returns the number of successful connection attempts made by 'rc'. */
784 rconn_get_successful_connections(const struct rconn *rc)
786 return rc->n_successful_connections;
789 /* Returns the time at which the last successful connection was made by
790 * 'rc'. Returns TIME_MIN if never connected. */
792 rconn_get_last_connection(const struct rconn *rc)
794 return rc->last_connected;
797 /* Returns the time at which 'rc' was last disconnected. Returns TIME_MIN
798 * if never disconnected. */
800 rconn_get_last_disconnect(const struct rconn *rc)
802 return rc->last_disconnected;
805 /* Returns the time at which the last OpenFlow message was received by 'rc'.
806 * If no packets have been received on 'rc', returns the time at which 'rc'
809 rconn_get_last_received(const struct rconn *rc)
811 return rc->last_received;
814 /* Returns the time at which 'rc' was created. */
816 rconn_get_creation_time(const struct rconn *rc)
818 return rc->creation_time;
821 /* Returns the approximate number of seconds that 'rc' has been connected. */
823 rconn_get_total_time_connected(const struct rconn *rc)
825 return (rc->total_time_connected
826 + (rconn_is_connected(rc) ? elapsed_in_this_state(rc) : 0));
829 /* Returns the current amount of backoff, in seconds. This is the amount of
830 * time after which the rconn will transition from BACKOFF to CONNECTING. */
832 rconn_get_backoff(const struct rconn *rc)
837 /* Returns the number of seconds spent in this state so far. */
839 rconn_get_state_elapsed(const struct rconn *rc)
841 return elapsed_in_this_state(rc);
844 /* Returns 'rc''s current connection sequence number, a number that changes
845 * every time that 'rconn' connects or disconnects. */
847 rconn_get_connection_seqno(const struct rconn *rc)
852 /* Returns a value that explains why 'rc' last disconnected:
854 * - 0 means that the last disconnection was caused by a call to
855 * rconn_disconnect(), or that 'rc' is new and has not yet completed its
856 * initial connection or connection attempt.
858 * - EOF means that the connection was closed in the normal way by the peer.
860 * - A positive integer is an errno value that represents the error.
863 rconn_get_last_error(const struct rconn *rc)
865 return rc->last_error;
868 /* Returns the number of messages queued for transmission on 'rc'. */
870 rconn_count_txqlen(const struct rconn *rc)
872 return list_size(&rc->txq);
875 struct rconn_packet_counter *
876 rconn_packet_counter_create(void)
878 struct rconn_packet_counter *c = xzalloc(sizeof *c);
884 rconn_packet_counter_destroy(struct rconn_packet_counter *c)
887 assert(c->ref_cnt > 0);
888 if (!--c->ref_cnt && !c->n_packets) {
895 rconn_packet_counter_inc(struct rconn_packet_counter *c, unsigned int n_bytes)
898 c->n_bytes += n_bytes;
902 rconn_packet_counter_dec(struct rconn_packet_counter *c, unsigned int n_bytes)
904 assert(c->n_packets > 0);
905 assert(c->n_bytes >= n_bytes);
907 c->n_bytes -= n_bytes;
917 /* Set rc->target and rc->name to 'target' and 'name', respectively. If 'name'
918 * is null, 'target' is used.
920 * Also, clear out the cached IP address and port information, since changing
921 * the target also likely changes these values. */
923 rconn_set_target__(struct rconn *rc, const char *target, const char *name)
926 rc->name = xstrdup(name ? name : target);
928 rc->target = xstrdup(target);
934 /* Tries to send a packet from 'rc''s send buffer. Returns 0 if successful,
935 * otherwise a positive errno value. */
937 try_send(struct rconn *rc)
939 struct ofpbuf *msg = ofpbuf_from_list(rc->txq.next);
940 unsigned int n_bytes = msg->size;
941 struct rconn_packet_counter *counter = msg->private_p;
944 /* Eagerly remove 'msg' from the txq. We can't remove it from the list
945 * after sending, if sending is successful, because it is then owned by the
946 * vconn, which might have freed it already. */
947 list_remove(&msg->list_node);
949 retval = vconn_send(rc->vconn, msg);
951 list_push_front(&rc->txq, &msg->list_node);
952 if (retval != EAGAIN) {
953 report_error(rc, retval);
954 disconnect(rc, retval);
958 COVERAGE_INC(rconn_sent);
961 rconn_packet_counter_dec(counter, n_bytes);
966 /* Reports that 'error' caused 'rc' to disconnect. 'error' may be a positive
967 * errno value, or it may be EOF to indicate that the connection was closed
970 report_error(struct rconn *rc, int error)
973 /* If 'rc' isn't reliable, then we don't really expect this connection
974 * to last forever anyway (probably it's a connection that we received
975 * via accept()), so use DBG level to avoid cluttering the logs. */
976 enum vlog_level level = rc->reliable ? VLL_INFO : VLL_DBG;
977 VLOG(level, "%s: connection closed by peer", rc->name);
979 VLOG_WARN("%s: connection dropped (%s)", rc->name, strerror(error));
983 /* Disconnects 'rc' and records 'error' as the error that caused 'rc''s last
986 * - 0 means that this disconnection is due to a request by 'rc''s client,
987 * not due to any kind of network error.
989 * - EOF means that the connection was closed in the normal way by the peer.
991 * - A positive integer is an errno value that represents the error.
994 disconnect(struct rconn *rc, int error)
996 rc->last_error = error;
998 time_t now = time_now();
1000 if (rc->state & (S_CONNECTING | S_ACTIVE | S_IDLE)) {
1001 rc->last_disconnected = now;
1002 vconn_close(rc->vconn);
1007 if (now >= rc->backoff_deadline) {
1009 } else if (rc->backoff < rc->max_backoff / 2) {
1010 rc->backoff = MAX(1, 2 * rc->backoff);
1011 VLOG_INFO("%s: waiting %d seconds before reconnect",
1012 rc->name, rc->backoff);
1014 if (rconn_logging_connection_attempts__(rc)) {
1015 VLOG_INFO("%s: continuing to retry connections in the "
1016 "background but suppressing further logging",
1019 rc->backoff = rc->max_backoff;
1021 rc->backoff_deadline = now + rc->backoff;
1022 state_transition(rc, S_BACKOFF);
1024 rc->last_disconnected = time_now();
1025 rconn_disconnect(rc);
1029 /* Drops all the packets from 'rc''s send queue and decrements their queue
1032 flush_queue(struct rconn *rc)
1034 if (list_is_empty(&rc->txq)) {
1037 while (!list_is_empty(&rc->txq)) {
1038 struct ofpbuf *b = ofpbuf_from_list(list_pop_front(&rc->txq));
1039 struct rconn_packet_counter *counter = b->private_p;
1041 rconn_packet_counter_dec(counter, b->size);
1043 COVERAGE_INC(rconn_discarded);
1046 poll_immediate_wake();
1050 elapsed_in_this_state(const struct rconn *rc)
1052 return time_now() - rc->state_entered;
1056 timeout(const struct rconn *rc)
1058 switch (rc->state) {
1059 #define STATE(NAME, VALUE) case S_##NAME: return timeout_##NAME(rc);
1068 timed_out(const struct rconn *rc)
1070 return time_now() >= sat_add(rc->state_entered, timeout(rc));
1074 state_transition(struct rconn *rc, enum state state)
1076 rc->seqno += (rc->state == S_ACTIVE) != (state == S_ACTIVE);
1077 if (is_connected_state(state) && !is_connected_state(rc->state)) {
1078 rc->probably_admitted = false;
1080 if (rconn_is_connected(rc)) {
1081 rc->total_time_connected += elapsed_in_this_state(rc);
1083 VLOG_DBG("%s: entering %s", rc->name, state_name(state));
1085 rc->state_entered = time_now();
1089 copy_to_monitor(struct rconn *rc, const struct ofpbuf *b)
1091 struct ofpbuf *clone = NULL;
1095 for (i = 0; i < rc->n_monitors; ) {
1096 struct vconn *vconn = rc->monitors[i];
1099 clone = ofpbuf_clone(b);
1101 retval = vconn_send(vconn, clone);
1104 } else if (retval != EAGAIN) {
1105 VLOG_DBG("%s: closing monitor connection to %s: %s",
1106 rconn_get_name(rc), vconn_get_name(vconn),
1108 rc->monitors[i] = rc->monitors[--rc->n_monitors];
1113 ofpbuf_delete(clone);
1117 is_connected_state(enum state state)
1119 return (state & (S_ACTIVE | S_IDLE)) != 0;
1123 is_admitted_msg(const struct ofpbuf *b)
1128 error = ofptype_decode(&type, b->data);
1136 case OFPTYPE_ECHO_REQUEST:
1137 case OFPTYPE_ECHO_REPLY:
1138 case OFPTYPE_FEATURES_REQUEST:
1139 case OFPTYPE_FEATURES_REPLY:
1140 case OFPTYPE_GET_CONFIG_REQUEST:
1141 case OFPTYPE_GET_CONFIG_REPLY:
1142 case OFPTYPE_SET_CONFIG:
1145 case OFPTYPE_PACKET_IN:
1146 case OFPTYPE_FLOW_REMOVED:
1147 case OFPTYPE_PORT_STATUS:
1148 case OFPTYPE_PACKET_OUT:
1149 case OFPTYPE_FLOW_MOD:
1150 case OFPTYPE_PORT_MOD:
1151 case OFPTYPE_BARRIER_REQUEST:
1152 case OFPTYPE_BARRIER_REPLY:
1153 case OFPTYPE_DESC_STATS_REQUEST:
1154 case OFPTYPE_DESC_STATS_REPLY:
1155 case OFPTYPE_FLOW_STATS_REQUEST:
1156 case OFPTYPE_FLOW_STATS_REPLY:
1157 case OFPTYPE_AGGREGATE_STATS_REQUEST:
1158 case OFPTYPE_AGGREGATE_STATS_REPLY:
1159 case OFPTYPE_TABLE_STATS_REQUEST:
1160 case OFPTYPE_TABLE_STATS_REPLY:
1161 case OFPTYPE_PORT_STATS_REQUEST:
1162 case OFPTYPE_PORT_STATS_REPLY:
1163 case OFPTYPE_QUEUE_STATS_REQUEST:
1164 case OFPTYPE_QUEUE_STATS_REPLY:
1165 case OFPTYPE_PORT_DESC_STATS_REQUEST:
1166 case OFPTYPE_PORT_DESC_STATS_REPLY:
1167 case OFPTYPE_ROLE_REQUEST:
1168 case OFPTYPE_ROLE_REPLY:
1169 case OFPTYPE_SET_FLOW_FORMAT:
1170 case OFPTYPE_FLOW_MOD_TABLE_ID:
1171 case OFPTYPE_SET_PACKET_IN_FORMAT:
1172 case OFPTYPE_FLOW_AGE:
1173 case OFPTYPE_SET_ASYNC_CONFIG:
1174 case OFPTYPE_SET_CONTROLLER_ID:
1175 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
1176 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
1177 case OFPTYPE_FLOW_MONITOR_CANCEL:
1178 case OFPTYPE_FLOW_MONITOR_PAUSED:
1179 case OFPTYPE_FLOW_MONITOR_RESUMED:
1185 /* Returns true if 'rc' is currently logging information about connection
1186 * attempts, false if logging should be suppressed because 'rc' hasn't
1187 * successuflly connected in too long. */
1189 rconn_logging_connection_attempts__(const struct rconn *rc)
1191 return rc->backoff < rc->max_backoff;