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;
83 time_t last_connected;
84 time_t last_disconnected;
85 unsigned int packets_sent;
89 /* In S_ACTIVE and S_IDLE, probably_admitted reports whether we believe
90 * that the peer has made a (positive) admission control decision on our
91 * connection. If we have not yet been (probably) admitted, then the
92 * connection does not reset the timer used for deciding whether the switch
93 * should go into fail-open mode.
95 * last_admitted reports the last time we believe such a positive admission
96 * control decision was made. */
97 bool probably_admitted;
100 /* These values are simply for statistics reporting, not used directly by
101 * anything internal to the rconn (or ofproto for that matter). */
102 unsigned int packets_received;
103 unsigned int n_attempted_connections, n_successful_connections;
104 time_t creation_time;
105 unsigned long int total_time_connected;
107 /* Throughout this file, "probe" is shorthand for "inactivity probe". When
108 * no activity has been observed from the peer for a while, we send out an
109 * echo request as an inactivity probe packet. We should receive back a
112 * "Activity" is defined as either receiving an OpenFlow message from the
113 * peer or successfully sending a message that had been in 'txq'. */
114 int probe_interval; /* Secs of inactivity before sending probe. */
115 time_t last_activity; /* Last time we saw some activity. */
117 /* When we create a vconn we obtain these values, to save them past the end
118 * of the vconn's lifetime. Otherwise, in-band control will only allow
119 * traffic when a vconn is actually open, but it is nice to allow ARP to
120 * complete even between connection attempts, and it is also polite to
121 * allow traffic from other switches to go through to the controller
122 * whether or not we are connected.
124 * We don't cache the local port, because that changes from one connection
125 * attempt to the next. */
126 ovs_be32 local_ip, remote_ip;
127 ovs_be16 remote_port;
130 /* Messages sent or received are copied to the monitor connections. */
131 #define MAX_MONITORS 8
132 struct vconn *monitors[8];
135 uint32_t allowed_versions;
138 uint32_t rconn_get_allowed_versions(const struct rconn *rconn)
140 return rconn->allowed_versions;
143 static unsigned int elapsed_in_this_state(const struct rconn *);
144 static unsigned int timeout(const struct rconn *);
145 static bool timed_out(const struct rconn *);
146 static void state_transition(struct rconn *, enum state);
147 static void rconn_set_target__(struct rconn *,
148 const char *target, const char *name);
149 static int try_send(struct rconn *);
150 static void reconnect(struct rconn *);
151 static void report_error(struct rconn *, int error);
152 static void disconnect(struct rconn *, int error);
153 static void flush_queue(struct rconn *);
154 static void copy_to_monitor(struct rconn *, const struct ofpbuf *);
155 static bool is_connected_state(enum state);
156 static bool is_admitted_msg(const struct ofpbuf *);
157 static bool rconn_logging_connection_attempts__(const struct rconn *);
159 /* Creates and returns a new rconn.
161 * 'probe_interval' is a number of seconds. If the interval passes once
162 * without an OpenFlow message being received from the peer, the rconn sends
163 * out an "echo request" message. If the interval passes again without a
164 * message being received, the rconn disconnects and re-connects to the peer.
165 * Setting 'probe_interval' to 0 disables this behavior.
167 * 'max_backoff' is the maximum number of seconds between attempts to connect
168 * to the peer. The actual interval starts at 1 second and doubles on each
169 * failure until it reaches 'max_backoff'. If 0 is specified, the default of
172 * The new rconn is initially unconnected. Use rconn_connect() or
173 * rconn_connect_unreliably() to connect it.
175 * Connections made by the rconn will automatically negotiate an OpenFlow
176 * protocol version acceptable to both peers on the connection. The version
177 * negotiated will be one of those in the 'allowed_versions' bitmap: version
178 * 'x' is allowed if allowed_versions & (1 << x) is nonzero. (The underlying
179 * vconn will treat an 'allowed_versions' of 0 as OFPUTIL_DEFAULT_VERSIONS.)
182 rconn_create(int probe_interval, int max_backoff, uint8_t dscp,
183 uint32_t allowed_versions)
185 struct rconn *rc = xzalloc(sizeof *rc);
188 rc->state_entered = time_now();
191 rc->name = xstrdup("void");
192 rc->target = xstrdup("void");
193 rc->reliable = false;
198 rc->max_backoff = max_backoff ? max_backoff : 8;
199 rc->backoff_deadline = TIME_MIN;
200 rc->last_connected = TIME_MIN;
201 rc->last_disconnected = TIME_MIN;
204 rc->packets_sent = 0;
206 rc->probably_admitted = false;
207 rc->last_admitted = time_now();
209 rc->packets_received = 0;
210 rc->n_attempted_connections = 0;
211 rc->n_successful_connections = 0;
212 rc->creation_time = time_now();
213 rc->total_time_connected = 0;
215 rc->last_activity = time_now();
217 rconn_set_probe_interval(rc, probe_interval);
218 rconn_set_dscp(rc, dscp);
221 rc->allowed_versions = allowed_versions;
227 rconn_set_max_backoff(struct rconn *rc, int max_backoff)
229 rc->max_backoff = MAX(1, max_backoff);
230 if (rc->state == S_BACKOFF && rc->backoff > max_backoff) {
231 rc->backoff = max_backoff;
232 if (rc->backoff_deadline > time_now() + max_backoff) {
233 rc->backoff_deadline = time_now() + max_backoff;
239 rconn_get_max_backoff(const struct rconn *rc)
241 return rc->max_backoff;
245 rconn_set_dscp(struct rconn *rc, uint8_t dscp)
251 rconn_get_dscp(const struct rconn *rc)
257 rconn_set_probe_interval(struct rconn *rc, int probe_interval)
259 rc->probe_interval = probe_interval ? MAX(5, probe_interval) : 0;
263 rconn_get_probe_interval(const struct rconn *rc)
265 return rc->probe_interval;
268 /* Drops any existing connection on 'rc', then sets up 'rc' to connect to
269 * 'target' and reconnect as needed. 'target' should be a remote OpenFlow
270 * target in a form acceptable to vconn_open().
272 * If 'name' is nonnull, then it is used in log messages in place of 'target'.
273 * It should presumably give more information to a human reader than 'target',
274 * but it need not be acceptable to vconn_open(). */
276 rconn_connect(struct rconn *rc, const char *target, const char *name)
278 rconn_disconnect(rc);
279 rconn_set_target__(rc, target, name);
284 /* Drops any existing connection on 'rc', then configures 'rc' to use
285 * 'vconn'. If the connection on 'vconn' drops, 'rc' will not reconnect on it
288 * By default, the target obtained from vconn_get_name(vconn) is used in log
289 * messages. If 'name' is nonnull, then it is used instead. It should
290 * presumably give more information to a human reader than the target, but it
291 * need not be acceptable to vconn_open(). */
293 rconn_connect_unreliably(struct rconn *rc,
294 struct vconn *vconn, const char *name)
296 assert(vconn != NULL);
297 rconn_disconnect(rc);
298 rconn_set_target__(rc, vconn_get_name(vconn), name);
299 rc->reliable = false;
301 rc->last_connected = time_now();
302 state_transition(rc, S_ACTIVE);
305 /* If 'rc' is connected, forces it to drop the connection and reconnect. */
307 rconn_reconnect(struct rconn *rc)
309 if (rc->state & (S_ACTIVE | S_IDLE)) {
310 VLOG_INFO("%s: disconnecting", rc->name);
316 rconn_disconnect(struct rconn *rc)
318 if (rc->state != S_VOID) {
320 vconn_close(rc->vconn);
323 rconn_set_target__(rc, "void", NULL);
324 rc->reliable = false;
327 rc->backoff_deadline = TIME_MIN;
329 state_transition(rc, S_VOID);
333 /* Disconnects 'rc' and frees the underlying storage. */
335 rconn_destroy(struct rconn *rc)
342 vconn_close(rc->vconn);
344 ofpbuf_list_delete(&rc->txq);
345 for (i = 0; i < rc->n_monitors; i++) {
346 vconn_close(rc->monitors[i]);
353 timeout_VOID(const struct rconn *rc OVS_UNUSED)
359 run_VOID(struct rconn *rc OVS_UNUSED)
365 reconnect(struct rconn *rc)
369 if (rconn_logging_connection_attempts__(rc)) {
370 VLOG_INFO("%s: connecting...", rc->name);
372 rc->n_attempted_connections++;
373 retval = vconn_open(rc->target, rc->allowed_versions,
374 &rc->vconn, rc->dscp);
376 rc->remote_ip = vconn_get_remote_ip(rc->vconn);
377 rc->local_ip = vconn_get_local_ip(rc->vconn);
378 rc->remote_port = vconn_get_remote_port(rc->vconn);
379 rc->backoff_deadline = time_now() + rc->backoff;
380 state_transition(rc, S_CONNECTING);
382 VLOG_WARN("%s: connection failed (%s)", rc->name, strerror(retval));
383 rc->backoff_deadline = TIME_MAX; /* Prevent resetting backoff. */
384 disconnect(rc, retval);
389 timeout_BACKOFF(const struct rconn *rc)
395 run_BACKOFF(struct rconn *rc)
403 timeout_CONNECTING(const struct rconn *rc)
405 return MAX(1, rc->backoff);
409 run_CONNECTING(struct rconn *rc)
411 int retval = vconn_connect(rc->vconn);
413 VLOG_INFO("%s: connected", rc->name);
414 rc->n_successful_connections++;
415 state_transition(rc, S_ACTIVE);
416 rc->last_connected = rc->state_entered;
417 } else if (retval != EAGAIN) {
418 if (rconn_logging_connection_attempts__(rc)) {
419 VLOG_INFO("%s: connection failed (%s)",
420 rc->name, strerror(retval));
422 disconnect(rc, retval);
423 } else if (timed_out(rc)) {
424 if (rconn_logging_connection_attempts__(rc)) {
425 VLOG_INFO("%s: connection timed out", rc->name);
427 rc->backoff_deadline = TIME_MAX; /* Prevent resetting backoff. */
428 disconnect(rc, ETIMEDOUT);
433 do_tx_work(struct rconn *rc)
435 if (list_is_empty(&rc->txq)) {
438 while (!list_is_empty(&rc->txq)) {
439 int error = try_send(rc);
443 rc->last_activity = time_now();
445 if (list_is_empty(&rc->txq)) {
446 poll_immediate_wake();
451 timeout_ACTIVE(const struct rconn *rc)
453 if (rc->probe_interval) {
454 unsigned int base = MAX(rc->last_activity, rc->state_entered);
455 unsigned int arg = base + rc->probe_interval - rc->state_entered;
462 run_ACTIVE(struct rconn *rc)
465 unsigned int base = MAX(rc->last_activity, rc->state_entered);
468 VLOG_DBG("%s: idle %u seconds, sending inactivity probe",
469 rc->name, (unsigned int) (time_now() - base));
471 version = rconn_get_version(rc);
472 assert(version >= 0 && version <= 0xff);
474 /* Ordering is important here: rconn_send() can transition to BACKOFF,
475 * and we don't want to transition back to IDLE if so, because then we
476 * can end up queuing a packet with vconn == NULL and then *boom*. */
477 state_transition(rc, S_IDLE);
478 rconn_send(rc, make_echo_request(version), NULL);
486 timeout_IDLE(const struct rconn *rc)
488 return rc->probe_interval;
492 run_IDLE(struct rconn *rc)
495 VLOG_ERR("%s: no response to inactivity probe after %u "
496 "seconds, disconnecting",
497 rc->name, elapsed_in_this_state(rc));
498 disconnect(rc, ETIMEDOUT);
504 /* Performs whatever activities are necessary to maintain 'rc': if 'rc' is
505 * disconnected, attempts to (re)connect, backing off as necessary; if 'rc' is
506 * connected, attempts to send packets in the send queue, if any. */
508 rconn_run(struct rconn *rc)
514 vconn_run(rc->vconn);
516 for (i = 0; i < rc->n_monitors; i++) {
517 vconn_run(rc->monitors[i]);
521 old_state = rc->state;
523 #define STATE(NAME, VALUE) case S_##NAME: run_##NAME(rc); break;
529 } while (rc->state != old_state);
532 /* Causes the next call to poll_block() to wake up when rconn_run() should be
535 rconn_run_wait(struct rconn *rc)
541 vconn_run_wait(rc->vconn);
542 if ((rc->state & (S_ACTIVE | S_IDLE)) && !list_is_empty(&rc->txq)) {
543 vconn_wait(rc->vconn, WAIT_SEND);
546 for (i = 0; i < rc->n_monitors; i++) {
547 vconn_run_wait(rc->monitors[i]);
551 if (timeo != UINT_MAX) {
552 long long int expires = sat_add(rc->state_entered, timeo);
553 poll_timer_wait_until(expires * 1000);
557 /* Attempts to receive a packet from 'rc'. If successful, returns the packet;
558 * otherwise, returns a null pointer. The caller is responsible for freeing
559 * the packet (with ofpbuf_delete()). */
561 rconn_recv(struct rconn *rc)
563 if (rc->state & (S_ACTIVE | S_IDLE)) {
564 struct ofpbuf *buffer;
565 int error = vconn_recv(rc->vconn, &buffer);
567 copy_to_monitor(rc, buffer);
568 if (rc->probably_admitted || is_admitted_msg(buffer)
569 || time_now() - rc->last_connected >= 30) {
570 rc->probably_admitted = true;
571 rc->last_admitted = time_now();
573 rc->last_activity = time_now();
574 rc->packets_received++;
575 if (rc->state == S_IDLE) {
576 state_transition(rc, S_ACTIVE);
579 } else if (error != EAGAIN) {
580 report_error(rc, error);
581 disconnect(rc, error);
587 /* Causes the next call to poll_block() to wake up when a packet may be ready
588 * to be received by vconn_recv() on 'rc'. */
590 rconn_recv_wait(struct rconn *rc)
593 vconn_wait(rc->vconn, WAIT_RECV);
597 /* Sends 'b' on 'rc'. Returns 0 if successful, or ENOTCONN if 'rc' is not
598 * currently connected. Takes ownership of 'b'.
600 * If 'counter' is non-null, then 'counter' will be incremented while the
601 * packet is in flight, then decremented when it has been sent (or discarded
602 * due to disconnection). Because 'b' may be sent (or discarded) before this
603 * function returns, the caller may not be able to observe any change in
606 * There is no rconn_send_wait() function: an rconn has a send queue that it
607 * takes care of sending if you call rconn_run(), which will have the side
608 * effect of waking up poll_block(). */
610 rconn_send(struct rconn *rc, struct ofpbuf *b,
611 struct rconn_packet_counter *counter)
613 if (rconn_is_connected(rc)) {
614 COVERAGE_INC(rconn_queued);
615 copy_to_monitor(rc, b);
616 b->private_p = counter;
618 rconn_packet_counter_inc(counter, b->size);
620 list_push_back(&rc->txq, &b->list_node);
622 /* If the queue was empty before we added 'b', try to send some
623 * packets. (But if the queue had packets in it, it's because the
624 * vconn is backlogged and there's no point in stuffing more into it
625 * now. We'll get back to that in rconn_run().) */
626 if (rc->txq.next == &b->list_node) {
636 /* Sends 'b' on 'rc'. Increments 'counter' while the packet is in flight; it
637 * will be decremented when it has been sent (or discarded due to
638 * disconnection). Returns 0 if successful, EAGAIN if 'counter->n' is already
639 * at least as large as 'queue_limit', or ENOTCONN if 'rc' is not currently
640 * connected. Regardless of return value, 'b' is destroyed.
642 * Because 'b' may be sent (or discarded) before this function returns, the
643 * caller may not be able to observe any change in 'counter'.
645 * There is no rconn_send_wait() function: an rconn has a send queue that it
646 * takes care of sending if you call rconn_run(), which will have the side
647 * effect of waking up poll_block(). */
649 rconn_send_with_limit(struct rconn *rc, struct ofpbuf *b,
650 struct rconn_packet_counter *counter, int queue_limit)
653 retval = (counter->n_packets >= queue_limit
655 : rconn_send(rc, b, counter));
657 COVERAGE_INC(rconn_overflow);
662 /* Returns the total number of packets successfully sent on the underlying
663 * vconn. A packet is not counted as sent while it is still queued in the
664 * rconn, only when it has been successfuly passed to the vconn. */
666 rconn_packets_sent(const struct rconn *rc)
668 return rc->packets_sent;
671 /* Adds 'vconn' to 'rc' as a monitoring connection, to which all messages sent
672 * and received on 'rconn' will be copied. 'rc' takes ownership of 'vconn'. */
674 rconn_add_monitor(struct rconn *rc, struct vconn *vconn)
676 if (rc->n_monitors < ARRAY_SIZE(rc->monitors)) {
677 int version = vconn_get_version(rc->vconn);
679 /* Override the allowed versions of the snoop vconn so that
680 * only the version of the controller connection is allowed.
681 * This is because the snoop will see the same messages as the
683 vconn_set_allowed_versions(vconn, 1u << version);
685 VLOG_INFO("new monitor connection from %s", vconn_get_name(vconn));
686 rc->monitors[rc->n_monitors++] = vconn;
688 VLOG_DBG("too many monitor connections, discarding %s",
689 vconn_get_name(vconn));
694 /* Returns 'rc''s name. This is a name for human consumption, appropriate for
695 * use in log messages. It is not necessarily a name that may be passed
696 * directly to, e.g., vconn_open(). */
698 rconn_get_name(const struct rconn *rc)
703 /* Sets 'rc''s name to 'new_name'. */
705 rconn_set_name(struct rconn *rc, const char *new_name)
708 rc->name = xstrdup(new_name);
711 /* Returns 'rc''s target. This is intended to be a string that may be passed
712 * directly to, e.g., vconn_open(). */
714 rconn_get_target(const struct rconn *rc)
719 /* Returns true if 'rconn' is connected or in the process of reconnecting,
720 * false if 'rconn' is disconnected and will not reconnect on its own. */
722 rconn_is_alive(const struct rconn *rconn)
724 return rconn->state != S_VOID;
727 /* Returns true if 'rconn' is connected, false otherwise. */
729 rconn_is_connected(const struct rconn *rconn)
731 return is_connected_state(rconn->state);
734 /* Returns true if 'rconn' is connected and thought to have been accepted by
735 * the peer's admission-control policy. */
737 rconn_is_admitted(const struct rconn *rconn)
739 return (rconn_is_connected(rconn)
740 && rconn->last_admitted >= rconn->last_connected);
743 /* Returns 0 if 'rconn' is currently connected and considered to have been
744 * accepted by the peer's admission-control policy, otherwise the number of
745 * seconds since 'rconn' was last in such a state. */
747 rconn_failure_duration(const struct rconn *rconn)
749 return rconn_is_admitted(rconn) ? 0 : time_now() - rconn->last_admitted;
752 /* Returns the IP address of the peer, or 0 if the peer's IP address is not
755 rconn_get_remote_ip(const struct rconn *rconn)
757 return rconn->remote_ip;
760 /* Returns the transport port of the peer, or 0 if the peer's port is not
763 rconn_get_remote_port(const struct rconn *rconn)
765 return rconn->remote_port;
768 /* Returns the IP address used to connect to the peer, or 0 if the
769 * connection is not an IP-based protocol or if its IP address is not
772 rconn_get_local_ip(const struct rconn *rconn)
774 return rconn->local_ip;
777 /* Returns the transport port used to connect to the peer, or 0 if the
778 * connection does not contain a port or if the port is not known. */
780 rconn_get_local_port(const struct rconn *rconn)
782 return rconn->vconn ? vconn_get_local_port(rconn->vconn) : 0;
785 /* Returns the OpenFlow version negotiated with the peer, or -1 if there is
786 * currently no connection or if version negotiation is not yet complete. */
788 rconn_get_version(const struct rconn *rconn)
790 return rconn->vconn ? vconn_get_version(rconn->vconn) : -1;
793 /* Returns the total number of packets successfully received by the underlying
796 rconn_packets_received(const struct rconn *rc)
798 return rc->packets_received;
801 /* Returns a string representing the internal state of 'rc'. The caller must
802 * not modify or free the string. */
804 rconn_get_state(const struct rconn *rc)
806 return state_name(rc->state);
809 /* Returns the time at which the last successful connection was made by
810 * 'rc'. Returns TIME_MIN if never connected. */
812 rconn_get_last_connection(const struct rconn *rc)
814 return rc->last_connected;
817 /* Returns the time at which 'rc' was last disconnected. Returns TIME_MIN
818 * if never disconnected. */
820 rconn_get_last_disconnect(const struct rconn *rc)
822 return rc->last_disconnected;
825 /* Returns 'rc''s current connection sequence number, a number that changes
826 * every time that 'rconn' connects or disconnects. */
828 rconn_get_connection_seqno(const struct rconn *rc)
833 /* Returns a value that explains why 'rc' last disconnected:
835 * - 0 means that the last disconnection was caused by a call to
836 * rconn_disconnect(), or that 'rc' is new and has not yet completed its
837 * initial connection or connection attempt.
839 * - EOF means that the connection was closed in the normal way by the peer.
841 * - A positive integer is an errno value that represents the error.
844 rconn_get_last_error(const struct rconn *rc)
846 return rc->last_error;
849 /* Returns the number of messages queued for transmission on 'rc'. */
851 rconn_count_txqlen(const struct rconn *rc)
853 return list_size(&rc->txq);
856 struct rconn_packet_counter *
857 rconn_packet_counter_create(void)
859 struct rconn_packet_counter *c = xzalloc(sizeof *c);
865 rconn_packet_counter_destroy(struct rconn_packet_counter *c)
868 assert(c->ref_cnt > 0);
869 if (!--c->ref_cnt && !c->n_packets) {
876 rconn_packet_counter_inc(struct rconn_packet_counter *c, unsigned int n_bytes)
879 c->n_bytes += n_bytes;
883 rconn_packet_counter_dec(struct rconn_packet_counter *c, unsigned int n_bytes)
885 assert(c->n_packets > 0);
886 assert(c->n_bytes >= n_bytes);
888 c->n_bytes -= n_bytes;
898 /* Set rc->target and rc->name to 'target' and 'name', respectively. If 'name'
899 * is null, 'target' is used.
901 * Also, clear out the cached IP address and port information, since changing
902 * the target also likely changes these values. */
904 rconn_set_target__(struct rconn *rc, const char *target, const char *name)
907 rc->name = xstrdup(name ? name : target);
909 rc->target = xstrdup(target);
915 /* Tries to send a packet from 'rc''s send buffer. Returns 0 if successful,
916 * otherwise a positive errno value. */
918 try_send(struct rconn *rc)
920 struct ofpbuf *msg = ofpbuf_from_list(rc->txq.next);
921 unsigned int n_bytes = msg->size;
922 struct rconn_packet_counter *counter = msg->private_p;
925 /* Eagerly remove 'msg' from the txq. We can't remove it from the list
926 * after sending, if sending is successful, because it is then owned by the
927 * vconn, which might have freed it already. */
928 list_remove(&msg->list_node);
930 retval = vconn_send(rc->vconn, msg);
932 list_push_front(&rc->txq, &msg->list_node);
933 if (retval != EAGAIN) {
934 report_error(rc, retval);
935 disconnect(rc, retval);
939 COVERAGE_INC(rconn_sent);
942 rconn_packet_counter_dec(counter, n_bytes);
947 /* Reports that 'error' caused 'rc' to disconnect. 'error' may be a positive
948 * errno value, or it may be EOF to indicate that the connection was closed
951 report_error(struct rconn *rc, int error)
954 /* If 'rc' isn't reliable, then we don't really expect this connection
955 * to last forever anyway (probably it's a connection that we received
956 * via accept()), so use DBG level to avoid cluttering the logs. */
957 enum vlog_level level = rc->reliable ? VLL_INFO : VLL_DBG;
958 VLOG(level, "%s: connection closed by peer", rc->name);
960 VLOG_WARN("%s: connection dropped (%s)", rc->name, strerror(error));
964 /* Disconnects 'rc' and records 'error' as the error that caused 'rc''s last
967 * - 0 means that this disconnection is due to a request by 'rc''s client,
968 * not due to any kind of network error.
970 * - EOF means that the connection was closed in the normal way by the peer.
972 * - A positive integer is an errno value that represents the error.
975 disconnect(struct rconn *rc, int error)
977 rc->last_error = error;
979 time_t now = time_now();
981 if (rc->state & (S_CONNECTING | S_ACTIVE | S_IDLE)) {
982 rc->last_disconnected = now;
983 vconn_close(rc->vconn);
988 if (now >= rc->backoff_deadline) {
990 } else if (rc->backoff < rc->max_backoff / 2) {
991 rc->backoff = MAX(1, 2 * rc->backoff);
992 VLOG_INFO("%s: waiting %d seconds before reconnect",
993 rc->name, rc->backoff);
995 if (rconn_logging_connection_attempts__(rc)) {
996 VLOG_INFO("%s: continuing to retry connections in the "
997 "background but suppressing further logging",
1000 rc->backoff = rc->max_backoff;
1002 rc->backoff_deadline = now + rc->backoff;
1003 state_transition(rc, S_BACKOFF);
1005 rc->last_disconnected = time_now();
1006 rconn_disconnect(rc);
1010 /* Drops all the packets from 'rc''s send queue and decrements their queue
1013 flush_queue(struct rconn *rc)
1015 if (list_is_empty(&rc->txq)) {
1018 while (!list_is_empty(&rc->txq)) {
1019 struct ofpbuf *b = ofpbuf_from_list(list_pop_front(&rc->txq));
1020 struct rconn_packet_counter *counter = b->private_p;
1022 rconn_packet_counter_dec(counter, b->size);
1024 COVERAGE_INC(rconn_discarded);
1027 poll_immediate_wake();
1031 elapsed_in_this_state(const struct rconn *rc)
1033 return time_now() - rc->state_entered;
1037 timeout(const struct rconn *rc)
1039 switch (rc->state) {
1040 #define STATE(NAME, VALUE) case S_##NAME: return timeout_##NAME(rc);
1049 timed_out(const struct rconn *rc)
1051 return time_now() >= sat_add(rc->state_entered, timeout(rc));
1055 state_transition(struct rconn *rc, enum state state)
1057 rc->seqno += (rc->state == S_ACTIVE) != (state == S_ACTIVE);
1058 if (is_connected_state(state) && !is_connected_state(rc->state)) {
1059 rc->probably_admitted = false;
1061 if (rconn_is_connected(rc)) {
1062 rc->total_time_connected += elapsed_in_this_state(rc);
1064 VLOG_DBG("%s: entering %s", rc->name, state_name(state));
1066 rc->state_entered = time_now();
1070 copy_to_monitor(struct rconn *rc, const struct ofpbuf *b)
1072 struct ofpbuf *clone = NULL;
1076 for (i = 0; i < rc->n_monitors; ) {
1077 struct vconn *vconn = rc->monitors[i];
1080 clone = ofpbuf_clone(b);
1082 retval = vconn_send(vconn, clone);
1085 } else if (retval != EAGAIN) {
1086 VLOG_DBG("%s: closing monitor connection to %s: %s",
1087 rconn_get_name(rc), vconn_get_name(vconn),
1089 rc->monitors[i] = rc->monitors[--rc->n_monitors];
1094 ofpbuf_delete(clone);
1098 is_connected_state(enum state state)
1100 return (state & (S_ACTIVE | S_IDLE)) != 0;
1104 is_admitted_msg(const struct ofpbuf *b)
1109 error = ofptype_decode(&type, b->data);
1117 case OFPTYPE_ECHO_REQUEST:
1118 case OFPTYPE_ECHO_REPLY:
1119 case OFPTYPE_FEATURES_REQUEST:
1120 case OFPTYPE_FEATURES_REPLY:
1121 case OFPTYPE_GET_CONFIG_REQUEST:
1122 case OFPTYPE_GET_CONFIG_REPLY:
1123 case OFPTYPE_SET_CONFIG:
1124 /* FIXME: Change the following once they are implemented: */
1125 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
1126 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
1127 case OFPTYPE_GET_ASYNC_REQUEST:
1128 case OFPTYPE_GET_ASYNC_REPLY:
1129 case OFPTYPE_METER_MOD:
1130 case OFPTYPE_GROUP_REQUEST:
1131 case OFPTYPE_GROUP_REPLY:
1132 case OFPTYPE_GROUP_DESC_REQUEST:
1133 case OFPTYPE_GROUP_DESC_REPLY:
1134 case OFPTYPE_GROUP_FEATURES_REQUEST:
1135 case OFPTYPE_GROUP_FEATURES_REPLY:
1136 case OFPTYPE_METER_REQUEST:
1137 case OFPTYPE_METER_REPLY:
1138 case OFPTYPE_METER_CONFIG_REQUEST:
1139 case OFPTYPE_METER_CONFIG_REPLY:
1140 case OFPTYPE_METER_FEATURES_REQUEST:
1141 case OFPTYPE_METER_FEATURES_REPLY:
1142 case OFPTYPE_TABLE_FEATURES_REQUEST:
1143 case OFPTYPE_TABLE_FEATURES_REPLY:
1146 case OFPTYPE_PACKET_IN:
1147 case OFPTYPE_FLOW_REMOVED:
1148 case OFPTYPE_PORT_STATUS:
1149 case OFPTYPE_PACKET_OUT:
1150 case OFPTYPE_FLOW_MOD:
1151 case OFPTYPE_PORT_MOD:
1152 case OFPTYPE_BARRIER_REQUEST:
1153 case OFPTYPE_BARRIER_REPLY:
1154 case OFPTYPE_DESC_STATS_REQUEST:
1155 case OFPTYPE_DESC_STATS_REPLY:
1156 case OFPTYPE_FLOW_STATS_REQUEST:
1157 case OFPTYPE_FLOW_STATS_REPLY:
1158 case OFPTYPE_AGGREGATE_STATS_REQUEST:
1159 case OFPTYPE_AGGREGATE_STATS_REPLY:
1160 case OFPTYPE_TABLE_STATS_REQUEST:
1161 case OFPTYPE_TABLE_STATS_REPLY:
1162 case OFPTYPE_PORT_STATS_REQUEST:
1163 case OFPTYPE_PORT_STATS_REPLY:
1164 case OFPTYPE_QUEUE_STATS_REQUEST:
1165 case OFPTYPE_QUEUE_STATS_REPLY:
1166 case OFPTYPE_PORT_DESC_STATS_REQUEST:
1167 case OFPTYPE_PORT_DESC_STATS_REPLY:
1168 case OFPTYPE_ROLE_REQUEST:
1169 case OFPTYPE_ROLE_REPLY:
1170 case OFPTYPE_SET_FLOW_FORMAT:
1171 case OFPTYPE_FLOW_MOD_TABLE_ID:
1172 case OFPTYPE_SET_PACKET_IN_FORMAT:
1173 case OFPTYPE_FLOW_AGE:
1174 case OFPTYPE_SET_ASYNC_CONFIG:
1175 case OFPTYPE_SET_CONTROLLER_ID:
1176 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
1177 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
1178 case OFPTYPE_FLOW_MONITOR_CANCEL:
1179 case OFPTYPE_FLOW_MONITOR_PAUSED:
1180 case OFPTYPE_FLOW_MONITOR_RESUMED:
1186 /* Returns true if 'rc' is currently logging information about connection
1187 * attempts, false if logging should be suppressed because 'rc' hasn't
1188 * successuflly connected in too long. */
1190 rconn_logging_connection_attempts__(const struct rconn *rc)
1192 return rc->backoff < rc->max_backoff;