/*
- * Copyright (c) 2008, 2009 Nicira Networks.
+ * Copyright (c) 2008, 2009, 2010 Nicira Networks.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
time_t last_connected;
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
* a response. */
int probe_interval; /* Secs of inactivity before sending probe. */
+ /* When we create a vconn we obtain these values, to save them past the end
+ * of the vconn's lifetime. Otherwise, in-band control will only allow
+ * traffic when a vconn is actually open, but it is nice to allow ARP to
+ * complete even between connection attempts, and it is also polite to
+ * allow traffic from other switches to go through to the controller
+ * whether or not we are connected.
+ *
+ * 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;
+
/* Messages sent or received are copied to the monitor connections. */
#define MAX_MONITORS 8
struct vconn *monitors[8];
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 int try_send(struct rconn *);
static int 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 *);
* '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
- * 60 seconds is used. */
+ * 8 seconds is used. */
struct rconn *
rconn_create(int probe_interval, int max_backoff)
{
- struct rconn *rc = xcalloc(1, sizeof *rc);
+ struct rconn *rc = xzalloc(sizeof *rc);
rc->state = S_VOID;
rc->state_entered = time_now();
queue_init(&rc->txq);
rc->backoff = 0;
- rc->max_backoff = max_backoff ? max_backoff : 60;
+ rc->max_backoff = max_backoff ? max_backoff : 8;
rc->backoff_deadline = TIME_MIN;
rc->last_received = time_now();
rc->last_connected = time_now();
rconn_connect(struct rconn *rc, const char *name)
{
rconn_disconnect(rc);
- free(rc->name);
- rc->name = xstrdup(name);
+ set_vconn_name(rc, name);
rc->reliable = true;
return reconnect(rc);
}
{
assert(vconn != NULL);
rconn_disconnect(rc);
- free(rc->name);
- rc->name = xstrdup(name);
+ set_vconn_name(rc, name);
rc->reliable = false;
rc->vconn = vconn;
rc->last_connected = time_now();
rconn_reconnect(struct rconn *rc)
{
if (rc->state & (S_ACTIVE | S_IDLE)) {
+ VLOG_INFO("%s: disconnecting", rc->name);
disconnect(rc, 0);
}
}
vconn_close(rc->vconn);
rc->vconn = NULL;
}
- free(rc->name);
- rc->name = xstrdup("void");
+ set_vconn_name(rc, "void");
rc->reliable = false;
rc->backoff = 0;
}
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. */
}
rc->n_attempted_connections++;
retval = vconn_open(rc->name, OFP_VERSION, &rc->vconn);
if (!retval) {
+ rc->remote_ip = vconn_get_remote_ip(rc->vconn);
+ rc->local_ip = vconn_get_local_ip(rc->vconn);
+ rc->remote_port = vconn_get_remote_port(rc->vconn);
rc->backoff_deadline = time_now() + rc->backoff;
state_transition(rc, S_CONNECTING);
} 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;
}
} else if (timed_out(rc)) {
VLOG_INFO("%s: connection timed out", rc->name);
rc->backoff_deadline = TIME_MAX; /* Prevent resetting backoff. */
- disconnect(rc, 0);
+ disconnect(rc, ETIMEDOUT);
}
}
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);
}
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) {
void
rconn_run_wait(struct rconn *rc)
{
- unsigned int timeo = timeout(rc);
+ unsigned int timeo;
+ size_t i;
+
+ if (rc->vconn) {
+ vconn_run_wait(rc->vconn);
+ }
+ for (i = 0; i < rc->n_monitors; i++) {
+ vconn_run_wait(rc->monitors[i]);
+ }
+
+ timeo = timeout(rc);
if (timeo != UINT_MAX) {
unsigned int expires = sat_add(rc->state_entered, timeo);
unsigned int remaining = sat_sub(expires, time_now());
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();
}
return buffer;
} else if (error != EAGAIN) {
+ report_error(rc, error);
disconnect(rc, error);
}
}
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);
}
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 is not connected over
- * an IP-based protocol or if its IP address is not known. */
+/* 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)
{
- return rconn->vconn ? vconn_get_remote_ip(rconn->vconn) : 0;
+ return rconn->remote_ip;
}
-/* Returns the transport port of the peer, or 0 if the peer does not
- * contain a port or if the port is not known. */
+/* 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)
{
- return rconn->vconn ? vconn_get_remote_port(rconn->vconn) : 0;
+ return rconn->remote_port;
}
/* Returns the IP address used to connect to the peer, or 0 if the
uint32_t
rconn_get_local_ip(const struct rconn *rconn)
{
- return rconn->vconn ? vconn_get_local_ip(rconn->vconn) : 0;
+ return rconn->local_ip;
}
/* Returns the transport port used to connect to the peer, or 0 if the
{
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;
+}
\f
struct rconn_packet_counter *
rconn_packet_counter_create(void)
}
}
\f
+/* 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. */
+static void
+set_vconn_name(struct rconn *rc, const char *name)
+{
+ free(rc->name);
+ rc->name = xstrdup(name);
+ rc->local_ip = 0;
+ rc->remote_ip = 0;
+ rc->remote_port = 0;
+}
+
/* Tries to send a packet from 'rc''s send buffer. Returns 0 if successful,
* otherwise a positive errno value. */
static int
{
int retval = 0;
struct ofpbuf *next = rc->txq.head->next;
- struct rconn_packet_counter *counter = rc->txq.head->private;
+ struct rconn_packet_counter *counter = rc->txq.head->private_p;
retval = vconn_send(rc->vconn, rc->txq.head);
if (retval) {
if (retval != EAGAIN) {
+ report_error(rc, retval);
disconnect(rc, retval);
}
return retval;
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);
- }
vconn_close(rc->vconn);
rc->vconn = NULL;
flush_queue(rc);
}
while (rc->txq.n > 0) {
struct ofpbuf *b = queue_pop_head(&rc->txq);
- struct rconn_packet_counter *counter = b->private;
+ struct rconn_packet_counter *counter = b->private_p;
if (counter) {
rconn_packet_counter_dec(counter);
}
git://git.onelab.eu / sliver-openvswitch.git / blobdiff