X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=lib%2Frconn.c;h=1129a3b7a69207e6da6a873284a68ef57fd72904;hb=a644168568471eb33f8d1e60e872859f4e5bbc69;hp=a27e4321d261bea5d95f40f0a1e3fe05b46a6a14;hpb=8398cf7efc3979cd6c8152915a584c13873ee322;p=openvswitch

diff --git a/lib/rconn.c b/lib/rconn.c
index a27e4321..1129a3b7 100644
--- a/lib/rconn.c
+++ b/lib/rconn.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2008, 2009 Nicira Networks.
+ * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira, Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -22,6 +22,7 @@
 #include <stdlib.h>
 #include <string.h>
 #include "coverage.h"
+#include "ofp-util.h"
 #include "ofpbuf.h"
 #include "openflow/openflow.h"
 #include "poll-loop.h"
@@ -29,10 +30,15 @@
 #include "timeval.h"
 #include "util.h"
 #include "vconn.h"
-
-#define THIS_MODULE VLM_rconn
 #include "vlog.h"
 
+VLOG_DEFINE_THIS_MODULE(rconn);
+
+COVERAGE_DEFINE(rconn_discarded);
+COVERAGE_DEFINE(rconn_overflow);
+COVERAGE_DEFINE(rconn_queued);
+COVERAGE_DEFINE(rconn_sent);
+
 #define STATES                                  \
     STATE(VOID, 1 << 0)                         \
     STATE(BACKOFF, 1 << 1)                      \
@@ -64,18 +70,21 @@ struct rconn {
     time_t state_entered;
 
     struct vconn *vconn;
-    char *name;
+    char *name;                 /* Human-readable descriptive name. */
+    char *target;               /* vconn name, passed to vconn_open(). */
     bool reliable;
 
-    struct ovs_queue txq;
+    struct list txq;            /* Contains "struct ofpbuf"s. */
 
     int backoff;
     int max_backoff;
     time_t backoff_deadline;
     time_t last_received;
     time_t last_connected;
+    time_t last_disconnected;
     unsigned int packets_sent;
     unsigned int seqno;
+    int last_error;
 
     /* In S_ACTIVE and S_IDLE, probably_admitted reports whether we believe
      * that the peer has made a (positive) admission control decision on our
@@ -95,16 +104,6 @@ struct rconn {
     time_t creation_time;
     unsigned long int total_time_connected;
 
-    /* If we can't connect to the peer, it could be for any number of reasons.
-     * Usually, one would assume it is because the peer is not running or
-     * because the network is partitioned.  But it could also be because the
-     * network topology has changed, in which case the upper layer will need to
-     * reassess it (in particular, obtain a new IP address via DHCP and find
-     * the new location of the controller).  We set this flag when we suspect
-     * that this could be the case. */
-    bool questionable_connectivity;
-    time_t last_questioned;
-
     /* Throughout this file, "probe" is shorthand for "inactivity probe".
      * When nothing has been received from the peer for a while, we send out
      * an echo request as an inactivity probe packet.  We should receive back
@@ -120,8 +119,9 @@ struct rconn {
      *
      * We don't cache the local port, because that changes from one connection
      * attempt to the next. */
-    uint32_t local_ip, remote_ip;
-    uint16_t remote_port;
+    ovs_be32 local_ip, remote_ip;
+    ovs_be16 remote_port;
+    uint8_t dscp;
 
     /* Messages sent or received are copied to the monitor connections. */
 #define MAX_MONITORS 8
@@ -133,33 +133,17 @@ static unsigned int elapsed_in_this_state(const struct rconn *);
 static unsigned int timeout(const struct rconn *);
 static bool timed_out(const struct rconn *);
 static void state_transition(struct rconn *, enum state);
-static void set_vconn_name(struct rconn *, const char *name);
+static void rconn_set_target__(struct rconn *,
+                               const char *target, const char *name);
 static int try_send(struct rconn *);
-static int reconnect(struct rconn *);
+static void reconnect(struct rconn *);
+static void report_error(struct rconn *, int error);
 static void disconnect(struct rconn *, int error);
 static void flush_queue(struct rconn *);
-static void question_connectivity(struct rconn *);
 static void copy_to_monitor(struct rconn *, const struct ofpbuf *);
 static bool is_connected_state(enum state);
 static bool is_admitted_msg(const struct ofpbuf *);
-
-/* Creates a new rconn, connects it (reliably) to 'name', and returns it. */
-struct rconn *
-rconn_new(const char *name, int inactivity_probe_interval, int max_backoff)
-{
-    struct rconn *rc = rconn_create(inactivity_probe_interval, max_backoff);
-    rconn_connect(rc, name);
-    return rc;
-}
-
-/* Creates a new rconn, connects it (unreliably) to 'vconn', and returns it. */
-struct rconn *
-rconn_new_from_vconn(const char *name, struct vconn *vconn) 
-{
-    struct rconn *rc = rconn_create(60, 0);
-    rconn_connect_unreliably(rc, name, vconn);
-    return rc;
-}
+static bool rconn_logging_connection_attempts__(const struct rconn *);
 
 /* Creates and returns a new rconn.
  *
@@ -172,26 +156,31 @@ rconn_new_from_vconn(const char *name, struct vconn *vconn)
  * 'max_backoff' is the maximum number of seconds between attempts to connect
  * to the peer.  The actual interval starts at 1 second and doubles on each
  * failure until it reaches 'max_backoff'.  If 0 is specified, the default of
- * 8 seconds is used. */
+ * 8 seconds is used.
+ *
+ * The new rconn is initially unconnected.  Use rconn_connect() or
+ * rconn_connect_unreliably() to connect it. */
 struct rconn *
-rconn_create(int probe_interval, int max_backoff)
+rconn_create(int probe_interval, int max_backoff, uint8_t dscp)
 {
-    struct rconn *rc = xcalloc(1, sizeof *rc);
+    struct rconn *rc = xzalloc(sizeof *rc);
 
     rc->state = S_VOID;
     rc->state_entered = time_now();
 
     rc->vconn = NULL;
     rc->name = xstrdup("void");
+    rc->target = xstrdup("void");
     rc->reliable = false;
 
-    queue_init(&rc->txq);
+    list_init(&rc->txq);
 
     rc->backoff = 0;
     rc->max_backoff = max_backoff ? max_backoff : 8;
     rc->backoff_deadline = TIME_MIN;
     rc->last_received = time_now();
-    rc->last_connected = time_now();
+    rc->last_connected = TIME_MIN;
+    rc->last_disconnected = TIME_MIN;
     rc->seqno = 0;
 
     rc->packets_sent = 0;
@@ -205,10 +194,8 @@ rconn_create(int probe_interval, int max_backoff)
     rc->creation_time = time_now();
     rc->total_time_connected = 0;
 
-    rc->questionable_connectivity = false;
-    rc->last_questioned = time_now();
-
     rconn_set_probe_interval(rc, probe_interval);
+    rconn_set_dscp(rc, dscp);
 
     rc->n_monitors = 0;
 
@@ -233,6 +220,18 @@ rconn_get_max_backoff(const struct rconn *rc)
     return rc->max_backoff;
 }
 
+void
+rconn_set_dscp(struct rconn *rc, uint8_t dscp)
+{
+    rc->dscp = dscp;
+}
+
+uint8_t
+rconn_get_dscp(const struct rconn *rc)
+{
+    return rc->dscp;
+}
+
 void
 rconn_set_probe_interval(struct rconn *rc, int probe_interval)
 {
@@ -245,22 +244,37 @@ rconn_get_probe_interval(const struct rconn *rc)
     return rc->probe_interval;
 }
 
-int
-rconn_connect(struct rconn *rc, const char *name)
+/* Drops any existing connection on 'rc', then sets up 'rc' to connect to
+ * 'target' and reconnect as needed.  'target' should be a remote OpenFlow
+ * target in a form acceptable to vconn_open().
+ *
+ * If 'name' is nonnull, then it is used in log messages in place of 'target'.
+ * It should presumably give more information to a human reader than 'target',
+ * but it need not be acceptable to vconn_open(). */
+void
+rconn_connect(struct rconn *rc, const char *target, const char *name)
 {
     rconn_disconnect(rc);
-    set_vconn_name(rc, name);
+    rconn_set_target__(rc, target, name);
     rc->reliable = true;
-    return reconnect(rc);
+    reconnect(rc);
 }
 
+/* Drops any existing connection on 'rc', then configures 'rc' to use
+ * 'vconn'.  If the connection on 'vconn' drops, 'rc' will not reconnect on it
+ * own.
+ *
+ * By default, the target obtained from vconn_get_name(vconn) is used in log
+ * messages.  If 'name' is nonnull, then it is used instead.  It should
+ * presumably give more information to a human reader than the target, but it
+ * need not be acceptable to vconn_open(). */
 void
 rconn_connect_unreliably(struct rconn *rc,
-                         const char *name, struct vconn *vconn)
+                         struct vconn *vconn, const char *name)
 {
     assert(vconn != NULL);
     rconn_disconnect(rc);
-    set_vconn_name(rc, name);
+    rconn_set_target__(rc, vconn_get_name(vconn), name);
     rc->reliable = false;
     rc->vconn = vconn;
     rc->last_connected = time_now();
@@ -272,6 +286,7 @@ void
 rconn_reconnect(struct rconn *rc)
 {
     if (rc->state & (S_ACTIVE | S_IDLE)) {
+        VLOG_INFO("%s: disconnecting", rc->name);
         disconnect(rc, 0);
     }
 }
@@ -284,7 +299,7 @@ rconn_disconnect(struct rconn *rc)
             vconn_close(rc->vconn);
             rc->vconn = NULL;
         }
-        set_vconn_name(rc, "void");
+        rconn_set_target__(rc, "void", NULL);
         rc->reliable = false;
 
         rc->backoff = 0;
@@ -302,9 +317,10 @@ rconn_destroy(struct rconn *rc)
         size_t i;
 
         free(rc->name);
+        free(rc->target);
         vconn_close(rc->vconn);
         flush_queue(rc);
-        queue_destroy(&rc->txq);
+        ofpbuf_list_delete(&rc->txq);
         for (i = 0; i < rc->n_monitors; i++) {
             vconn_close(rc->monitors[i]);
         }
@@ -313,25 +329,27 @@ rconn_destroy(struct rconn *rc)
 }
 
 static unsigned int
-timeout_VOID(const struct rconn *rc UNUSED)
+timeout_VOID(const struct rconn *rc OVS_UNUSED)
 {
     return UINT_MAX;
 }
 
 static void
-run_VOID(struct rconn *rc UNUSED)
+run_VOID(struct rconn *rc OVS_UNUSED)
 {
     /* Nothing to do. */
 }
 
-static int
+static void
 reconnect(struct rconn *rc)
 {
     int retval;
 
-    VLOG_INFO("%s: connecting...", rc->name);
+    if (rconn_logging_connection_attempts__(rc)) {
+        VLOG_INFO("%s: connecting...", rc->name);
+    }
     rc->n_attempted_connections++;
-    retval = vconn_open(rc->name, OFP_VERSION, &rc->vconn);
+    retval = vconn_open(rc->target, OFP10_VERSION, &rc->vconn, rc->dscp);
     if (!retval) {
         rc->remote_ip = vconn_get_remote_ip(rc->vconn);
         rc->local_ip = vconn_get_local_ip(rc->vconn);
@@ -341,9 +359,8 @@ reconnect(struct rconn *rc)
     } else {
         VLOG_WARN("%s: connection failed (%s)", rc->name, strerror(retval));
         rc->backoff_deadline = TIME_MAX; /* Prevent resetting backoff. */
-        disconnect(rc, 0);
+        disconnect(rc, retval);
     }
-    return retval;
 }
 
 static unsigned int
@@ -376,28 +393,33 @@ run_CONNECTING(struct rconn *rc)
         state_transition(rc, S_ACTIVE);
         rc->last_connected = rc->state_entered;
     } else if (retval != EAGAIN) {
-        VLOG_INFO("%s: connection failed (%s)", rc->name, strerror(retval));
+        if (rconn_logging_connection_attempts__(rc)) {
+            VLOG_INFO("%s: connection failed (%s)",
+                      rc->name, strerror(retval));
+        }
         disconnect(rc, retval);
     } else if (timed_out(rc)) {
-        VLOG_INFO("%s: connection timed out", rc->name);
+        if (rconn_logging_connection_attempts__(rc)) {
+            VLOG_INFO("%s: connection timed out", rc->name);
+        }
         rc->backoff_deadline = TIME_MAX; /* Prevent resetting backoff. */
-        disconnect(rc, 0);
+        disconnect(rc, ETIMEDOUT);
     }
 }
 
 static void
 do_tx_work(struct rconn *rc)
 {
-    if (!rc->txq.n) {
+    if (list_is_empty(&rc->txq)) {
         return;
     }
-    while (rc->txq.n > 0) {
+    while (!list_is_empty(&rc->txq)) {
         int error = try_send(rc);
         if (error) {
             break;
         }
     }
-    if (!rc->txq.n) {
+    if (list_is_empty(&rc->txq)) {
         poll_immediate_wake();
     }
 }
@@ -442,11 +464,10 @@ static void
 run_IDLE(struct rconn *rc)
 {
     if (timed_out(rc)) {
-        question_connectivity(rc);
         VLOG_ERR("%s: no response to inactivity probe after %u "
                  "seconds, disconnecting",
                  rc->name, elapsed_in_this_state(rc));
-        disconnect(rc, 0);
+        disconnect(rc, ETIMEDOUT);
     } else {
         do_tx_work(rc);
     }
@@ -459,6 +480,15 @@ void
 rconn_run(struct rconn *rc)
 {
     int old_state;
+    size_t i;
+
+    if (rc->vconn) {
+        vconn_run(rc->vconn);
+    }
+    for (i = 0; i < rc->n_monitors; i++) {
+        vconn_run(rc->monitors[i]);
+    }
+
     do {
         old_state = rc->state;
         switch (rc->state) {
@@ -476,15 +506,23 @@ rconn_run(struct rconn *rc)
 void
 rconn_run_wait(struct rconn *rc)
 {
-    unsigned int timeo = timeout(rc);
-    if (timeo != UINT_MAX) {
-        unsigned int expires = sat_add(rc->state_entered, timeo);
-        unsigned int remaining = sat_sub(expires, time_now());
-        poll_timer_wait(sat_mul(remaining, 1000));
+    unsigned int timeo;
+    size_t i;
+
+    if (rc->vconn) {
+        vconn_run_wait(rc->vconn);
+        if ((rc->state & (S_ACTIVE | S_IDLE)) && !list_is_empty(&rc->txq)) {
+            vconn_wait(rc->vconn, WAIT_SEND);
+        }
+    }
+    for (i = 0; i < rc->n_monitors; i++) {
+        vconn_run_wait(rc->monitors[i]);
     }
 
-    if ((rc->state & (S_ACTIVE | S_IDLE)) && rc->txq.n) {
-        vconn_wait(rc->vconn, WAIT_SEND);
+    timeo = timeout(rc);
+    if (timeo != UINT_MAX) {
+        long long int expires = sat_add(rc->state_entered, timeo);
+        poll_timer_wait_until(expires * 1000);
     }
 }
 
@@ -499,7 +537,7 @@ rconn_recv(struct rconn *rc)
         int error = vconn_recv(rc->vconn, &buffer);
         if (!error) {
             copy_to_monitor(rc, buffer);
-            if (is_admitted_msg(buffer)
+            if (rc->probably_admitted || is_admitted_msg(buffer)
                 || time_now() - rc->last_connected >= 30) {
                 rc->probably_admitted = true;
                 rc->last_admitted = time_now();
@@ -511,6 +549,7 @@ rconn_recv(struct rconn *rc)
             }
             return buffer;
         } else if (error != EAGAIN) {
+            report_error(rc, error);
             disconnect(rc, error);
         }
     }
@@ -527,9 +566,8 @@ rconn_recv_wait(struct rconn *rc)
     }
 }
 
-/* Sends 'b' on 'rc'.  Returns 0 if successful (in which case 'b' is
- * destroyed), or ENOTCONN if 'rc' is not currently connected (in which case
- * the caller retains ownership of 'b').
+/* Sends 'b' on 'rc'.  Returns 0 if successful, or ENOTCONN if 'rc' is not
+ * currently connected.  Takes ownership of 'b'.
  *
  * If 'counter' is non-null, then 'counter' will be incremented while the
  * packet is in flight, then decremented when it has been sent (or discarded
@@ -547,21 +585,22 @@ rconn_send(struct rconn *rc, struct ofpbuf *b,
     if (rconn_is_connected(rc)) {
         COVERAGE_INC(rconn_queued);
         copy_to_monitor(rc, b);
-        b->private = counter;
+        b->private_p = counter;
         if (counter) {
-            rconn_packet_counter_inc(counter);
+            rconn_packet_counter_inc(counter, b->size);
         }
-        queue_push_tail(&rc->txq, b);
+        list_push_back(&rc->txq, &b->list_node);
 
         /* If the queue was empty before we added 'b', try to send some
          * packets.  (But if the queue had packets in it, it's because the
          * vconn is backlogged and there's no point in stuffing more into it
          * now.  We'll get back to that in rconn_run().) */
-        if (rc->txq.n == 1) {
+        if (rc->txq.next == &b->list_node) {
             try_send(rc);
         }
         return 0;
     } else {
+        ofpbuf_delete(b);
         return ENOTCONN;
     }
 }
@@ -583,10 +622,11 @@ rconn_send_with_limit(struct rconn *rc, struct ofpbuf *b,
                       struct rconn_packet_counter *counter, int queue_limit)
 {
     int retval;
-    retval = counter->n >= queue_limit ? EAGAIN : rconn_send(rc, b, counter);
+    retval = (counter->n_packets >= queue_limit
+              ? EAGAIN
+              : rconn_send(rc, b, counter));
     if (retval) {
         COVERAGE_INC(rconn_overflow);
-        ofpbuf_delete(b);
     }
     return retval;
 }
@@ -615,13 +655,31 @@ rconn_add_monitor(struct rconn *rc, struct vconn *vconn)
     }
 }
 
-/* Returns 'rc''s name (the 'name' argument passed to rconn_new()). */
+/* Returns 'rc''s name.  This is a name for human consumption, appropriate for
+ * use in log messages.  It is not necessarily a name that may be passed
+ * directly to, e.g., vconn_open(). */
 const char *
 rconn_get_name(const struct rconn *rc)
 {
     return rc->name;
 }
 
+/* Sets 'rc''s name to 'new_name'. */
+void
+rconn_set_name(struct rconn *rc, const char *new_name)
+{
+    free(rc->name);
+    rc->name = xstrdup(new_name);
+}
+
+/* Returns 'rc''s target.  This is intended to be a string that may be passed
+ * directly to, e.g., vconn_open(). */
+const char *
+rconn_get_target(const struct rconn *rc)
+{
+    return rc->target;
+}
+
 /* Returns true if 'rconn' is connected or in the process of reconnecting,
  * false if 'rconn' is disconnected and will not reconnect on its own. */
 bool
@@ -637,64 +695,63 @@ rconn_is_connected(const struct rconn *rconn)
     return is_connected_state(rconn->state);
 }
 
-/* Returns 0 if 'rconn' is connected.  Otherwise, if 'rconn' is in a "failure
- * mode" (that is, it is not connected), returns the number of seconds that it
- * has been in failure mode, ignoring any times that it connected but the
- * controller's admission control policy caused it to be quickly
- * disconnected. */
+/* Returns true if 'rconn' is connected and thought to have been accepted by
+ * the peer's admission-control policy. */
+bool
+rconn_is_admitted(const struct rconn *rconn)
+{
+    return (rconn_is_connected(rconn)
+            && rconn->last_admitted >= rconn->last_connected);
+}
+
+/* Returns 0 if 'rconn' is currently connected and considered to have been
+ * accepted by the peer's admission-control policy, otherwise the number of
+ * seconds since 'rconn' was last in such a state. */
 int
 rconn_failure_duration(const struct rconn *rconn)
 {
-    return rconn_is_connected(rconn) ? 0 : time_now() - rconn->last_admitted;
+    return rconn_is_admitted(rconn) ? 0 : time_now() - rconn->last_admitted;
 }
 
 /* Returns the IP address of the peer, or 0 if the peer's IP address is not
  * known. */
-uint32_t
-rconn_get_remote_ip(const struct rconn *rconn) 
+ovs_be32
+rconn_get_remote_ip(const struct rconn *rconn)
 {
     return rconn->remote_ip;
 }
 
 /* Returns the transport port of the peer, or 0 if the peer's port is not
  * known. */
-uint16_t
-rconn_get_remote_port(const struct rconn *rconn) 
+ovs_be16
+rconn_get_remote_port(const struct rconn *rconn)
 {
     return rconn->remote_port;
 }
 
 /* Returns the IP address used to connect to the peer, or 0 if the
- * connection is not an IP-based protocol or if its IP address is not 
+ * connection is not an IP-based protocol or if its IP address is not
  * known. */
-uint32_t
-rconn_get_local_ip(const struct rconn *rconn) 
+ovs_be32
+rconn_get_local_ip(const struct rconn *rconn)
 {
     return rconn->local_ip;
 }
 
 /* Returns the transport port used to connect to the peer, or 0 if the
  * connection does not contain a port or if the port is not known. */
-uint16_t
-rconn_get_local_port(const struct rconn *rconn) 
+ovs_be16
+rconn_get_local_port(const struct rconn *rconn)
 {
     return rconn->vconn ? vconn_get_local_port(rconn->vconn) : 0;
 }
 
-/* If 'rconn' can't connect to the peer, it could be for any number of reasons.
- * Usually, one would assume it is because the peer is not running or because
- * the network is partitioned.  But it could also be because the network
- * topology has changed, in which case the upper layer will need to reassess it
- * (in particular, obtain a new IP address via DHCP and find the new location
- * of the controller).  When this appears that this might be the case, this
- * function returns true.  It also clears the questionability flag and prevents
- * it from being set again for some time. */
-bool
-rconn_is_connectivity_questionable(struct rconn *rconn)
+/* Returns the OpenFlow version negotiated with the peer, or -1 if there is
+ * currently no connection or if version negotiation is not yet complete. */
+int
+rconn_get_version(const struct rconn *rconn)
 {
-    bool questionable = rconn->questionable_connectivity;
-    rconn->questionable_connectivity = false;
-    return questionable;
+    return rconn->vconn ? vconn_get_version(rconn->vconn) : -1;
 }
 
 /* Returns the total number of packets successfully received by the underlying
@@ -729,13 +786,21 @@ rconn_get_successful_connections(const struct rconn *rc)
 }
 
 /* Returns the time at which the last successful connection was made by
- * 'rc'. */
+ * 'rc'. Returns TIME_MIN if never connected. */
 time_t
 rconn_get_last_connection(const struct rconn *rc)
 {
     return rc->last_connected;
 }
 
+/* Returns the time at which 'rc' was last disconnected. Returns TIME_MIN
+ * if never disconnected. */
+time_t
+rconn_get_last_disconnect(const struct rconn *rc)
+{
+    return rc->last_disconnected;
+}
+
 /* Returns the time at which the last OpenFlow message was received by 'rc'.
  * If no packets have been received on 'rc', returns the time at which 'rc'
  * was created. */
@@ -782,12 +847,34 @@ rconn_get_connection_seqno(const struct rconn *rc)
 {
     return rc->seqno;
 }
+
+/* Returns a value that explains why 'rc' last disconnected:
+ *
+ *   - 0 means that the last disconnection was caused by a call to
+ *     rconn_disconnect(), or that 'rc' is new and has not yet completed its
+ *     initial connection or connection attempt.
+ *
+ *   - EOF means that the connection was closed in the normal way by the peer.
+ *
+ *   - A positive integer is an errno value that represents the error.
+ */
+int
+rconn_get_last_error(const struct rconn *rc)
+{
+    return rc->last_error;
+}
+
+/* Returns the number of messages queued for transmission on 'rc'. */
+unsigned int
+rconn_count_txqlen(const struct rconn *rc)
+{
+    return list_size(&rc->txq);
+}
 
 struct rconn_packet_counter *
 rconn_packet_counter_create(void)
 {
-    struct rconn_packet_counter *c = xmalloc(sizeof *c);
-    c->n = 0;
+    struct rconn_packet_counter *c = xzalloc(sizeof *c);
     c->ref_cnt = 1;
     return c;
 }
@@ -797,35 +884,47 @@ rconn_packet_counter_destroy(struct rconn_packet_counter *c)
 {
     if (c) {
         assert(c->ref_cnt > 0);
-        if (!--c->ref_cnt && !c->n) {
+        if (!--c->ref_cnt && !c->n_packets) {
             free(c);
         }
     }
 }
 
 void
-rconn_packet_counter_inc(struct rconn_packet_counter *c)
+rconn_packet_counter_inc(struct rconn_packet_counter *c, unsigned int n_bytes)
 {
-    c->n++;
+    c->n_packets++;
+    c->n_bytes += n_bytes;
 }
 
 void
-rconn_packet_counter_dec(struct rconn_packet_counter *c)
+rconn_packet_counter_dec(struct rconn_packet_counter *c, unsigned int n_bytes)
 {
-    assert(c->n > 0);
-    if (!--c->n && !c->ref_cnt) {
-        free(c);
+    assert(c->n_packets > 0);
+    assert(c->n_bytes >= n_bytes);
+
+    c->n_bytes -= n_bytes;
+    c->n_packets--;
+    if (!c->n_packets) {
+        assert(!c->n_bytes);
+        if (!c->ref_cnt) {
+            free(c);
+        }
     }
 }
 
-/* Set the name of the remote vconn to 'name' and clear out the cached IP
- * address and port information, since changing the name also likely changes
- * these values. */
+/* Set rc->target and rc->name to 'target' and 'name', respectively.  If 'name'
+ * is null, 'target' is used.
+ *
+ * Also, clear out the cached IP address and port information, since changing
+ * the target also likely changes these values. */
 static void
-set_vconn_name(struct rconn *rc, const char *name)
+rconn_set_target__(struct rconn *rc, const char *target, const char *name)
 {
     free(rc->name);
-    rc->name = xstrdup(name);
+    rc->name = xstrdup(name ? name : target);
+    free(rc->target);
+    rc->target = xstrdup(target);
     rc->local_ip = 0;
     rc->remote_ip = 0;
     rc->remote_port = 0;
@@ -836,12 +935,21 @@ set_vconn_name(struct rconn *rc, const char *name)
 static int
 try_send(struct rconn *rc)
 {
-    int retval = 0;
-    struct ofpbuf *next = rc->txq.head->next;
-    struct rconn_packet_counter *counter = rc->txq.head->private;
-    retval = vconn_send(rc->vconn, rc->txq.head);
+    struct ofpbuf *msg = ofpbuf_from_list(rc->txq.next);
+    unsigned int n_bytes = msg->size;
+    struct rconn_packet_counter *counter = msg->private_p;
+    int retval;
+
+    /* Eagerly remove 'msg' from the txq.  We can't remove it from the list
+     * after sending, if sending is successful, because it is then owned by the
+     * vconn, which might have freed it already. */
+    list_remove(&msg->list_node);
+
+    retval = vconn_send(rc->vconn, msg);
     if (retval) {
+        list_push_front(&rc->txq, &msg->list_node);
         if (retval != EAGAIN) {
+            report_error(rc, retval);
             disconnect(rc, retval);
         }
         return retval;
@@ -849,32 +957,47 @@ try_send(struct rconn *rc)
     COVERAGE_INC(rconn_sent);
     rc->packets_sent++;
     if (counter) {
-        rconn_packet_counter_dec(counter);
+        rconn_packet_counter_dec(counter, n_bytes);
     }
-    queue_advance_head(&rc->txq, next);
     return 0;
 }
 
-/* Disconnects 'rc'.  'error' is used only for logging purposes.  If it is
- * nonzero, then it should be EOF to indicate the connection was closed by the
- * peer in a normal fashion or a positive errno value. */
+/* Reports that 'error' caused 'rc' to disconnect.  'error' may be a positive
+ * errno value, or it may be EOF to indicate that the connection was closed
+ * normally. */
+static void
+report_error(struct rconn *rc, int error)
+{
+    if (error == EOF) {
+        /* If 'rc' isn't reliable, then we don't really expect this connection
+         * to last forever anyway (probably it's a connection that we received
+         * via accept()), so use DBG level to avoid cluttering the logs. */
+        enum vlog_level level = rc->reliable ? VLL_INFO : VLL_DBG;
+        VLOG(level, "%s: connection closed by peer", rc->name);
+    } else {
+        VLOG_WARN("%s: connection dropped (%s)", rc->name, strerror(error));
+    }
+}
+
+/* Disconnects 'rc' and records 'error' as the error that caused 'rc''s last
+ * disconnection:
+ *
+ *   - 0 means that this disconnection is due to a request by 'rc''s client,
+ *     not due to any kind of network error.
+ *
+ *   - EOF means that the connection was closed in the normal way by the peer.
+ *
+ *   - A positive integer is an errno value that represents the error.
+ */
 static void
 disconnect(struct rconn *rc, int error)
 {
+    rc->last_error = error;
     if (rc->reliable) {
         time_t now = time_now();
 
         if (rc->state & (S_CONNECTING | S_ACTIVE | S_IDLE)) {
-            if (error > 0) {
-                VLOG_WARN("%s: connection dropped (%s)",
-                          rc->name, strerror(error));
-            } else if (error == EOF) {
-                if (rc->reliable) {
-                    VLOG_INFO("%s: connection closed by peer", rc->name);
-                }
-            } else {
-                VLOG_INFO("%s: connection dropped", rc->name);
-            }
+            rc->last_disconnected = now;
             vconn_close(rc->vconn);
             rc->vconn = NULL;
             flush_queue(rc);
@@ -882,17 +1005,22 @@ disconnect(struct rconn *rc, int error)
 
         if (now >= rc->backoff_deadline) {
             rc->backoff = 1;
-        } else {
-            rc->backoff = MIN(rc->max_backoff, MAX(1, 2 * rc->backoff));
-            VLOG_INFO("%s: waiting %d seconds before reconnect\n",
+        } else if (rc->backoff < rc->max_backoff / 2) {
+            rc->backoff = MAX(1, 2 * rc->backoff);
+            VLOG_INFO("%s: waiting %d seconds before reconnect",
                       rc->name, rc->backoff);
+        } else {
+            if (rconn_logging_connection_attempts__(rc)) {
+                VLOG_INFO("%s: continuing to retry connections in the "
+                          "background but suppressing further logging",
+                          rc->name);
+            }
+            rc->backoff = rc->max_backoff;
         }
         rc->backoff_deadline = now + rc->backoff;
         state_transition(rc, S_BACKOFF);
-        if (now - rc->last_connected > 60) {
-            question_connectivity(rc);
-        }
     } else {
+        rc->last_disconnected = time_now();
         rconn_disconnect(rc);
     }
 }
@@ -902,14 +1030,14 @@ disconnect(struct rconn *rc, int error)
 static void
 flush_queue(struct rconn *rc)
 {
-    if (!rc->txq.n) {
+    if (list_is_empty(&rc->txq)) {
         return;
     }
-    while (rc->txq.n > 0) {
-        struct ofpbuf *b = queue_pop_head(&rc->txq);
-        struct rconn_packet_counter *counter = b->private;
+    while (!list_is_empty(&rc->txq)) {
+        struct ofpbuf *b = ofpbuf_from_list(list_pop_front(&rc->txq));
+        struct rconn_packet_counter *counter = b->private_p;
         if (counter) {
-            rconn_packet_counter_dec(counter);
+            rconn_packet_counter_dec(counter, b->size);
         }
         COVERAGE_INC(rconn_discarded);
         ofpbuf_delete(b);
@@ -956,16 +1084,6 @@ state_transition(struct rconn *rc, enum state state)
     rc->state_entered = time_now();
 }
 
-static void
-question_connectivity(struct rconn *rc) 
-{
-    time_t now = time_now();
-    if (now - rc->last_questioned > 60) {
-        rc->questionable_connectivity = true;
-        rc->last_questioned = now;
-    }
-}
-
 static void
 copy_to_monitor(struct rconn *rc, const struct ofpbuf *b)
 {
@@ -995,7 +1113,7 @@ copy_to_monitor(struct rconn *rc, const struct ofpbuf *b)
 }
 
 static bool
-is_connected_state(enum state state) 
+is_connected_state(enum state state)
 {
     return (state & (S_ACTIVE | S_IDLE)) != 0;
 }
@@ -1017,3 +1135,12 @@ is_admitted_msg(const struct ofpbuf *b)
                                 (1u << OFPT_GET_CONFIG_REPLY) |
                                 (1u << OFPT_SET_CONFIG)));
 }
+
+/* Returns true if 'rc' is currently logging information about connection
+ * attempts, false if logging should be suppressed because 'rc' hasn't
+ * successuflly connected in too long. */
+static bool
+rconn_logging_connection_attempts__(const struct rconn *rc)
+{
+    return rc->backoff < rc->max_backoff;
+}