2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 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.
27 #include "openflow/openflow.h"
28 #include "poll-loop.h"
35 VLOG_DEFINE_THIS_MODULE(rconn);
37 COVERAGE_DEFINE(rconn_discarded);
38 COVERAGE_DEFINE(rconn_overflow);
39 COVERAGE_DEFINE(rconn_queued);
40 COVERAGE_DEFINE(rconn_sent);
44 STATE(BACKOFF, 1 << 1) \
45 STATE(CONNECTING, 1 << 2) \
46 STATE(ACTIVE, 1 << 3) \
49 #define STATE(NAME, VALUE) S_##NAME = VALUE,
55 state_name(enum state state)
58 #define STATE(NAME, VALUE) case S_##NAME: return #NAME;
65 /* A reliable connection to an OpenFlow switch or controller.
67 * See the large comment in rconn.h for more information. */
73 char *name; /* Human-readable descriptive name. */
74 char *target; /* vconn name, passed to vconn_open(). */
77 struct list txq; /* Contains "struct ofpbuf"s. */
81 time_t backoff_deadline;
82 time_t last_connected;
83 time_t last_disconnected;
84 unsigned int packets_sent;
88 /* In S_ACTIVE and S_IDLE, probably_admitted reports whether we believe
89 * that the peer has made a (positive) admission control decision on our
90 * connection. If we have not yet been (probably) admitted, then the
91 * connection does not reset the timer used for deciding whether the switch
92 * should go into fail-open mode.
94 * last_admitted reports the last time we believe such a positive admission
95 * control decision was made. */
96 bool probably_admitted;
99 /* These values are simply for statistics reporting, not used directly by
100 * anything internal to the rconn (or ofproto for that matter). */
101 unsigned int packets_received;
102 unsigned int n_attempted_connections, n_successful_connections;
103 time_t creation_time;
104 unsigned long int total_time_connected;
106 /* Throughout this file, "probe" is shorthand for "inactivity probe". When
107 * no activity has been observed from the peer for a while, we send out an
108 * echo request as an inactivity probe packet. We should receive back a
111 * "Activity" is defined as either receiving an OpenFlow message from the
112 * peer or successfully sending a message that had been in 'txq'. */
113 int probe_interval; /* Secs of inactivity before sending probe. */
114 time_t last_activity; /* Last time we saw some activity. */
116 /* When we create a vconn we obtain these values, to save them past the end
117 * of the vconn's lifetime. Otherwise, in-band control will only allow
118 * traffic when a vconn is actually open, but it is nice to allow ARP to
119 * complete even between connection attempts, and it is also polite to
120 * allow traffic from other switches to go through to the controller
121 * whether or not we are connected.
123 * We don't cache the local port, because that changes from one connection
124 * attempt to the next. */
125 ovs_be32 local_ip, remote_ip;
126 ovs_be16 remote_port;
129 /* Messages sent or received are copied to the monitor connections. */
130 #define MAX_MONITORS 8
131 struct vconn *monitors[8];
134 uint32_t allowed_versions;
137 uint32_t rconn_get_allowed_versions(const struct rconn *rconn)
139 return rconn->allowed_versions;
142 static unsigned int elapsed_in_this_state(const struct rconn *);
143 static unsigned int timeout(const struct rconn *);
144 static bool timed_out(const struct rconn *);
145 static void state_transition(struct rconn *, enum state);
146 static void rconn_set_target__(struct rconn *,
147 const char *target, const char *name);
148 static int try_send(struct rconn *);
149 static void reconnect(struct rconn *);
150 static void report_error(struct rconn *, int error);
151 static void disconnect(struct rconn *, int error);
152 static void flush_queue(struct rconn *);
153 static void close_monitor(struct rconn *, size_t idx, int retval);
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 ovs_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, 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)",
383 rc->name, ovs_strerror(retval));
384 rc->backoff_deadline = TIME_MAX; /* Prevent resetting backoff. */
385 disconnect(rc, retval);
390 timeout_BACKOFF(const struct rconn *rc)
396 run_BACKOFF(struct rconn *rc)
404 timeout_CONNECTING(const struct rconn *rc)
406 return MAX(1, rc->backoff);
410 run_CONNECTING(struct rconn *rc)
412 int retval = vconn_connect(rc->vconn);
414 VLOG_INFO("%s: connected", rc->name);
415 rc->n_successful_connections++;
416 state_transition(rc, S_ACTIVE);
417 rc->last_connected = rc->state_entered;
418 } else if (retval != EAGAIN) {
419 if (rconn_logging_connection_attempts__(rc)) {
420 VLOG_INFO("%s: connection failed (%s)",
421 rc->name, ovs_strerror(retval));
423 disconnect(rc, retval);
424 } else if (timed_out(rc)) {
425 if (rconn_logging_connection_attempts__(rc)) {
426 VLOG_INFO("%s: connection timed out", rc->name);
428 rc->backoff_deadline = TIME_MAX; /* Prevent resetting backoff. */
429 disconnect(rc, ETIMEDOUT);
434 do_tx_work(struct rconn *rc)
436 if (list_is_empty(&rc->txq)) {
439 while (!list_is_empty(&rc->txq)) {
440 int error = try_send(rc);
444 rc->last_activity = time_now();
446 if (list_is_empty(&rc->txq)) {
447 poll_immediate_wake();
452 timeout_ACTIVE(const struct rconn *rc)
454 if (rc->probe_interval) {
455 unsigned int base = MAX(rc->last_activity, rc->state_entered);
456 unsigned int arg = base + rc->probe_interval - rc->state_entered;
463 run_ACTIVE(struct rconn *rc)
466 unsigned int base = MAX(rc->last_activity, rc->state_entered);
469 VLOG_DBG("%s: idle %u seconds, sending inactivity probe",
470 rc->name, (unsigned int) (time_now() - base));
472 version = rconn_get_version(rc);
473 ovs_assert(version >= 0 && version <= 0xff);
475 /* Ordering is important here: rconn_send() can transition to BACKOFF,
476 * and we don't want to transition back to IDLE if so, because then we
477 * can end up queuing a packet with vconn == NULL and then *boom*. */
478 state_transition(rc, S_IDLE);
479 rconn_send(rc, make_echo_request(version), NULL);
487 timeout_IDLE(const struct rconn *rc)
489 return rc->probe_interval;
493 run_IDLE(struct rconn *rc)
496 VLOG_ERR("%s: no response to inactivity probe after %u "
497 "seconds, disconnecting",
498 rc->name, elapsed_in_this_state(rc));
499 disconnect(rc, ETIMEDOUT);
505 /* Performs whatever activities are necessary to maintain 'rc': if 'rc' is
506 * disconnected, attempts to (re)connect, backing off as necessary; if 'rc' is
507 * connected, attempts to send packets in the send queue, if any. */
509 rconn_run(struct rconn *rc)
515 vconn_run(rc->vconn);
517 for (i = 0; i < rc->n_monitors; ) {
521 vconn_run(rc->monitors[i]);
523 /* Drain any stray message that came in on the monitor connection. */
524 retval = vconn_recv(rc->monitors[i], &msg);
527 } else if (retval != EAGAIN) {
528 close_monitor(rc, i, retval);
535 old_state = rc->state;
537 #define STATE(NAME, VALUE) case S_##NAME: run_##NAME(rc); break;
543 } while (rc->state != old_state);
546 /* Causes the next call to poll_block() to wake up when rconn_run() should be
549 rconn_run_wait(struct rconn *rc)
555 vconn_run_wait(rc->vconn);
556 if ((rc->state & (S_ACTIVE | S_IDLE)) && !list_is_empty(&rc->txq)) {
557 vconn_wait(rc->vconn, WAIT_SEND);
560 for (i = 0; i < rc->n_monitors; i++) {
561 vconn_run_wait(rc->monitors[i]);
562 vconn_recv_wait(rc->monitors[i]);
566 if (timeo != UINT_MAX) {
567 long long int expires = sat_add(rc->state_entered, timeo);
568 poll_timer_wait_until(expires * 1000);
572 /* Attempts to receive a packet from 'rc'. If successful, returns the packet;
573 * otherwise, returns a null pointer. The caller is responsible for freeing
574 * the packet (with ofpbuf_delete()). */
576 rconn_recv(struct rconn *rc)
578 if (rc->state & (S_ACTIVE | S_IDLE)) {
579 struct ofpbuf *buffer;
580 int error = vconn_recv(rc->vconn, &buffer);
582 copy_to_monitor(rc, buffer);
583 if (rc->probably_admitted || is_admitted_msg(buffer)
584 || time_now() - rc->last_connected >= 30) {
585 rc->probably_admitted = true;
586 rc->last_admitted = time_now();
588 rc->last_activity = time_now();
589 rc->packets_received++;
590 if (rc->state == S_IDLE) {
591 state_transition(rc, S_ACTIVE);
594 } else if (error != EAGAIN) {
595 report_error(rc, error);
596 disconnect(rc, error);
602 /* Causes the next call to poll_block() to wake up when a packet may be ready
603 * to be received by vconn_recv() on 'rc'. */
605 rconn_recv_wait(struct rconn *rc)
608 vconn_wait(rc->vconn, WAIT_RECV);
612 /* Sends 'b' on 'rc'. Returns 0 if successful, or ENOTCONN if 'rc' is not
613 * currently connected. Takes ownership of 'b'.
615 * If 'counter' is non-null, then 'counter' will be incremented while the
616 * packet is in flight, then decremented when it has been sent (or discarded
617 * due to disconnection). Because 'b' may be sent (or discarded) before this
618 * function returns, the caller may not be able to observe any change in
621 * There is no rconn_send_wait() function: an rconn has a send queue that it
622 * takes care of sending if you call rconn_run(), which will have the side
623 * effect of waking up poll_block(). */
625 rconn_send(struct rconn *rc, struct ofpbuf *b,
626 struct rconn_packet_counter *counter)
628 if (rconn_is_connected(rc)) {
629 COVERAGE_INC(rconn_queued);
630 copy_to_monitor(rc, b);
631 b->private_p = counter;
633 rconn_packet_counter_inc(counter, b->size);
635 list_push_back(&rc->txq, &b->list_node);
637 /* If the queue was empty before we added 'b', try to send some
638 * packets. (But if the queue had packets in it, it's because the
639 * vconn is backlogged and there's no point in stuffing more into it
640 * now. We'll get back to that in rconn_run().) */
641 if (rc->txq.next == &b->list_node) {
651 /* Sends 'b' on 'rc'. Increments 'counter' while the packet is in flight; it
652 * will be decremented when it has been sent (or discarded due to
653 * disconnection). Returns 0 if successful, EAGAIN if 'counter->n' is already
654 * at least as large as 'queue_limit', or ENOTCONN if 'rc' is not currently
655 * connected. Regardless of return value, 'b' is destroyed.
657 * Because 'b' may be sent (or discarded) before this function returns, the
658 * caller may not be able to observe any change in 'counter'.
660 * There is no rconn_send_wait() function: an rconn has a send queue that it
661 * takes care of sending if you call rconn_run(), which will have the side
662 * effect of waking up poll_block(). */
664 rconn_send_with_limit(struct rconn *rc, struct ofpbuf *b,
665 struct rconn_packet_counter *counter, int queue_limit)
667 if (counter->n_packets < queue_limit) {
668 return rconn_send(rc, b, counter);
670 COVERAGE_INC(rconn_overflow);
676 /* Returns the total number of packets successfully sent on the underlying
677 * vconn. A packet is not counted as sent while it is still queued in the
678 * rconn, only when it has been successfuly passed to the vconn. */
680 rconn_packets_sent(const struct rconn *rc)
682 return rc->packets_sent;
685 /* Adds 'vconn' to 'rc' as a monitoring connection, to which all messages sent
686 * and received on 'rconn' will be copied. 'rc' takes ownership of 'vconn'. */
688 rconn_add_monitor(struct rconn *rc, struct vconn *vconn)
690 if (rc->n_monitors < ARRAY_SIZE(rc->monitors)) {
691 VLOG_INFO("new monitor connection from %s", vconn_get_name(vconn));
692 rc->monitors[rc->n_monitors++] = vconn;
694 VLOG_DBG("too many monitor connections, discarding %s",
695 vconn_get_name(vconn));
700 /* Returns 'rc''s name. This is a name for human consumption, appropriate for
701 * use in log messages. It is not necessarily a name that may be passed
702 * directly to, e.g., vconn_open(). */
704 rconn_get_name(const struct rconn *rc)
709 /* Sets 'rc''s name to 'new_name'. */
711 rconn_set_name(struct rconn *rc, const char *new_name)
714 rc->name = xstrdup(new_name);
717 /* Returns 'rc''s target. This is intended to be a string that may be passed
718 * directly to, e.g., vconn_open(). */
720 rconn_get_target(const struct rconn *rc)
725 /* Returns true if 'rconn' is connected or in the process of reconnecting,
726 * false if 'rconn' is disconnected and will not reconnect on its own. */
728 rconn_is_alive(const struct rconn *rconn)
730 return rconn->state != S_VOID;
733 /* Returns true if 'rconn' is connected, false otherwise. */
735 rconn_is_connected(const struct rconn *rconn)
737 return is_connected_state(rconn->state);
740 /* Returns true if 'rconn' is connected and thought to have been accepted by
741 * the peer's admission-control policy. */
743 rconn_is_admitted(const struct rconn *rconn)
745 return (rconn_is_connected(rconn)
746 && rconn->last_admitted >= rconn->last_connected);
749 /* Returns 0 if 'rconn' is currently connected and considered to have been
750 * accepted by the peer's admission-control policy, otherwise the number of
751 * seconds since 'rconn' was last in such a state. */
753 rconn_failure_duration(const struct rconn *rconn)
755 return rconn_is_admitted(rconn) ? 0 : time_now() - rconn->last_admitted;
758 /* Returns the IP address of the peer, or 0 if the peer's IP address is not
761 rconn_get_remote_ip(const struct rconn *rconn)
763 return rconn->remote_ip;
766 /* Returns the transport port of the peer, or 0 if the peer's port is not
769 rconn_get_remote_port(const struct rconn *rconn)
771 return rconn->remote_port;
774 /* Returns the IP address used to connect to the peer, or 0 if the
775 * connection is not an IP-based protocol or if its IP address is not
778 rconn_get_local_ip(const struct rconn *rconn)
780 return rconn->local_ip;
783 /* Returns the transport port used to connect to the peer, or 0 if the
784 * connection does not contain a port or if the port is not known. */
786 rconn_get_local_port(const struct rconn *rconn)
788 return rconn->vconn ? vconn_get_local_port(rconn->vconn) : 0;
791 /* Returns the OpenFlow version negotiated with the peer, or -1 if there is
792 * currently no connection or if version negotiation is not yet complete. */
794 rconn_get_version(const struct rconn *rconn)
796 return rconn->vconn ? vconn_get_version(rconn->vconn) : -1;
799 /* Returns the total number of packets successfully received by the underlying
802 rconn_packets_received(const struct rconn *rc)
804 return rc->packets_received;
807 /* Returns a string representing the internal state of 'rc'. The caller must
808 * not modify or free the string. */
810 rconn_get_state(const struct rconn *rc)
812 return state_name(rc->state);
815 /* Returns the time at which the last successful connection was made by
816 * 'rc'. Returns TIME_MIN if never connected. */
818 rconn_get_last_connection(const struct rconn *rc)
820 return rc->last_connected;
823 /* Returns the time at which 'rc' was last disconnected. Returns TIME_MIN
824 * if never disconnected. */
826 rconn_get_last_disconnect(const struct rconn *rc)
828 return rc->last_disconnected;
831 /* Returns 'rc''s current connection sequence number, a number that changes
832 * every time that 'rconn' connects or disconnects. */
834 rconn_get_connection_seqno(const struct rconn *rc)
839 /* Returns a value that explains why 'rc' last disconnected:
841 * - 0 means that the last disconnection was caused by a call to
842 * rconn_disconnect(), or that 'rc' is new and has not yet completed its
843 * initial connection or connection attempt.
845 * - EOF means that the connection was closed in the normal way by the peer.
847 * - A positive integer is an errno value that represents the error.
850 rconn_get_last_error(const struct rconn *rc)
852 return rc->last_error;
855 /* Returns the number of messages queued for transmission on 'rc'. */
857 rconn_count_txqlen(const struct rconn *rc)
859 return list_size(&rc->txq);
862 struct rconn_packet_counter *
863 rconn_packet_counter_create(void)
865 struct rconn_packet_counter *c = xzalloc(sizeof *c);
871 rconn_packet_counter_destroy(struct rconn_packet_counter *c)
874 ovs_assert(c->ref_cnt > 0);
875 if (!--c->ref_cnt && !c->n_packets) {
882 rconn_packet_counter_inc(struct rconn_packet_counter *c, unsigned int n_bytes)
885 c->n_bytes += n_bytes;
889 rconn_packet_counter_dec(struct rconn_packet_counter *c, unsigned int n_bytes)
891 ovs_assert(c->n_packets > 0);
892 ovs_assert(c->n_bytes >= n_bytes);
894 c->n_bytes -= n_bytes;
897 ovs_assert(!c->n_bytes);
904 /* Set rc->target and rc->name to 'target' and 'name', respectively. If 'name'
905 * is null, 'target' is used.
907 * Also, clear out the cached IP address and port information, since changing
908 * the target also likely changes these values. */
910 rconn_set_target__(struct rconn *rc, const char *target, const char *name)
913 rc->name = xstrdup(name ? name : target);
915 rc->target = xstrdup(target);
921 /* Tries to send a packet from 'rc''s send buffer. Returns 0 if successful,
922 * otherwise a positive errno value. */
924 try_send(struct rconn *rc)
926 struct ofpbuf *msg = ofpbuf_from_list(rc->txq.next);
927 unsigned int n_bytes = msg->size;
928 struct rconn_packet_counter *counter = msg->private_p;
931 /* Eagerly remove 'msg' from the txq. We can't remove it from the list
932 * after sending, if sending is successful, because it is then owned by the
933 * vconn, which might have freed it already. */
934 list_remove(&msg->list_node);
936 retval = vconn_send(rc->vconn, msg);
938 list_push_front(&rc->txq, &msg->list_node);
939 if (retval != EAGAIN) {
940 report_error(rc, retval);
941 disconnect(rc, retval);
945 COVERAGE_INC(rconn_sent);
948 rconn_packet_counter_dec(counter, n_bytes);
953 /* Reports that 'error' caused 'rc' to disconnect. 'error' may be a positive
954 * errno value, or it may be EOF to indicate that the connection was closed
957 report_error(struct rconn *rc, int error)
960 /* If 'rc' isn't reliable, then we don't really expect this connection
961 * to last forever anyway (probably it's a connection that we received
962 * via accept()), so use DBG level to avoid cluttering the logs. */
963 enum vlog_level level = rc->reliable ? VLL_INFO : VLL_DBG;
964 VLOG(level, "%s: connection closed by peer", rc->name);
966 VLOG_WARN("%s: connection dropped (%s)",
967 rc->name, ovs_strerror(error));
971 /* Disconnects 'rc' and records 'error' as the error that caused 'rc''s last
974 * - 0 means that this disconnection is due to a request by 'rc''s client,
975 * not due to any kind of network error.
977 * - EOF means that the connection was closed in the normal way by the peer.
979 * - A positive integer is an errno value that represents the error.
982 disconnect(struct rconn *rc, int error)
984 rc->last_error = error;
986 time_t now = time_now();
988 if (rc->state & (S_CONNECTING | S_ACTIVE | S_IDLE)) {
989 rc->last_disconnected = now;
990 vconn_close(rc->vconn);
995 if (now >= rc->backoff_deadline) {
997 } else if (rc->backoff < rc->max_backoff / 2) {
998 rc->backoff = MAX(1, 2 * rc->backoff);
999 VLOG_INFO("%s: waiting %d seconds before reconnect",
1000 rc->name, rc->backoff);
1002 if (rconn_logging_connection_attempts__(rc)) {
1003 VLOG_INFO("%s: continuing to retry connections in the "
1004 "background but suppressing further logging",
1007 rc->backoff = rc->max_backoff;
1009 rc->backoff_deadline = now + rc->backoff;
1010 state_transition(rc, S_BACKOFF);
1012 rc->last_disconnected = time_now();
1013 rconn_disconnect(rc);
1017 /* Drops all the packets from 'rc''s send queue and decrements their queue
1020 flush_queue(struct rconn *rc)
1022 if (list_is_empty(&rc->txq)) {
1025 while (!list_is_empty(&rc->txq)) {
1026 struct ofpbuf *b = ofpbuf_from_list(list_pop_front(&rc->txq));
1027 struct rconn_packet_counter *counter = b->private_p;
1029 rconn_packet_counter_dec(counter, b->size);
1031 COVERAGE_INC(rconn_discarded);
1034 poll_immediate_wake();
1038 elapsed_in_this_state(const struct rconn *rc)
1040 return time_now() - rc->state_entered;
1044 timeout(const struct rconn *rc)
1046 switch (rc->state) {
1047 #define STATE(NAME, VALUE) case S_##NAME: return timeout_##NAME(rc);
1056 timed_out(const struct rconn *rc)
1058 return time_now() >= sat_add(rc->state_entered, timeout(rc));
1062 state_transition(struct rconn *rc, enum state state)
1064 rc->seqno += (rc->state == S_ACTIVE) != (state == S_ACTIVE);
1065 if (is_connected_state(state) && !is_connected_state(rc->state)) {
1066 rc->probably_admitted = false;
1068 if (rconn_is_connected(rc)) {
1069 rc->total_time_connected += elapsed_in_this_state(rc);
1071 VLOG_DBG("%s: entering %s", rc->name, state_name(state));
1073 rc->state_entered = time_now();
1077 close_monitor(struct rconn *rc, size_t idx, int retval)
1079 VLOG_DBG("%s: closing monitor connection to %s: %s",
1080 rconn_get_name(rc), vconn_get_name(rc->monitors[idx]),
1081 ovs_retval_to_string(retval));
1082 rc->monitors[idx] = rc->monitors[--rc->n_monitors];
1086 copy_to_monitor(struct rconn *rc, const struct ofpbuf *b)
1088 struct ofpbuf *clone = NULL;
1092 for (i = 0; i < rc->n_monitors; ) {
1093 struct vconn *vconn = rc->monitors[i];
1096 clone = ofpbuf_clone(b);
1098 retval = vconn_send(vconn, clone);
1101 } else if (retval != EAGAIN) {
1102 close_monitor(rc, i, retval);
1107 ofpbuf_delete(clone);
1111 is_connected_state(enum state state)
1113 return (state & (S_ACTIVE | S_IDLE)) != 0;
1117 is_admitted_msg(const struct ofpbuf *b)
1122 error = ofptype_decode(&type, b->data);
1130 case OFPTYPE_ECHO_REQUEST:
1131 case OFPTYPE_ECHO_REPLY:
1132 case OFPTYPE_FEATURES_REQUEST:
1133 case OFPTYPE_FEATURES_REPLY:
1134 case OFPTYPE_GET_CONFIG_REQUEST:
1135 case OFPTYPE_GET_CONFIG_REPLY:
1136 case OFPTYPE_SET_CONFIG:
1137 /* FIXME: Change the following once they are implemented: */
1138 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
1139 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
1140 case OFPTYPE_GET_ASYNC_REQUEST:
1141 case OFPTYPE_GET_ASYNC_REPLY:
1142 case OFPTYPE_GROUP_STATS_REQUEST:
1143 case OFPTYPE_GROUP_STATS_REPLY:
1144 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
1145 case OFPTYPE_GROUP_DESC_STATS_REPLY:
1146 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
1147 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
1148 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
1149 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
1152 case OFPTYPE_PACKET_IN:
1153 case OFPTYPE_FLOW_REMOVED:
1154 case OFPTYPE_PORT_STATUS:
1155 case OFPTYPE_PACKET_OUT:
1156 case OFPTYPE_FLOW_MOD:
1157 case OFPTYPE_GROUP_MOD:
1158 case OFPTYPE_PORT_MOD:
1159 case OFPTYPE_TABLE_MOD:
1160 case OFPTYPE_METER_MOD:
1161 case OFPTYPE_BARRIER_REQUEST:
1162 case OFPTYPE_BARRIER_REPLY:
1163 case OFPTYPE_DESC_STATS_REQUEST:
1164 case OFPTYPE_DESC_STATS_REPLY:
1165 case OFPTYPE_FLOW_STATS_REQUEST:
1166 case OFPTYPE_FLOW_STATS_REPLY:
1167 case OFPTYPE_AGGREGATE_STATS_REQUEST:
1168 case OFPTYPE_AGGREGATE_STATS_REPLY:
1169 case OFPTYPE_TABLE_STATS_REQUEST:
1170 case OFPTYPE_TABLE_STATS_REPLY:
1171 case OFPTYPE_PORT_STATS_REQUEST:
1172 case OFPTYPE_PORT_STATS_REPLY:
1173 case OFPTYPE_QUEUE_STATS_REQUEST:
1174 case OFPTYPE_QUEUE_STATS_REPLY:
1175 case OFPTYPE_PORT_DESC_STATS_REQUEST:
1176 case OFPTYPE_PORT_DESC_STATS_REPLY:
1177 case OFPTYPE_METER_STATS_REQUEST:
1178 case OFPTYPE_METER_STATS_REPLY:
1179 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
1180 case OFPTYPE_METER_CONFIG_STATS_REPLY:
1181 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
1182 case OFPTYPE_METER_FEATURES_STATS_REPLY:
1183 case OFPTYPE_ROLE_REQUEST:
1184 case OFPTYPE_ROLE_REPLY:
1185 case OFPTYPE_SET_FLOW_FORMAT:
1186 case OFPTYPE_FLOW_MOD_TABLE_ID:
1187 case OFPTYPE_SET_PACKET_IN_FORMAT:
1188 case OFPTYPE_FLOW_AGE:
1189 case OFPTYPE_SET_ASYNC_CONFIG:
1190 case OFPTYPE_SET_CONTROLLER_ID:
1191 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
1192 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
1193 case OFPTYPE_FLOW_MONITOR_CANCEL:
1194 case OFPTYPE_FLOW_MONITOR_PAUSED:
1195 case OFPTYPE_FLOW_MONITOR_RESUMED:
1201 /* Returns true if 'rc' is currently logging information about connection
1202 * attempts, false if logging should be suppressed because 'rc' hasn't
1203 * successuflly connected in too long. */
1205 rconn_logging_connection_attempts__(const struct rconn *rc)
1207 return rc->backoff < rc->max_backoff;