/*
- * Copyright (c) 2008, 2009, 2010 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.
#include <stdlib.h>
#include <string.h>
#include "coverage.h"
+#include "ofp-msgs.h"
#include "ofp-util.h"
#include "ofpbuf.h"
#include "openflow/openflow.h"
#include "vconn.h"
#include "vlog.h"
-VLOG_DEFINE_THIS_MODULE(rconn)
+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) \
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;
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
- * a response. */
+ /* Throughout this file, "probe" is shorthand for "inactivity probe". When
+ * no activity has been observed from the peer for a while, we send out an
+ * echo request as an inactivity probe packet. We should receive back a
+ * response.
+ *
+ * "Activity" is defined as either receiving an OpenFlow message from the
+ * peer or successfully sending a message that had been in 'txq'. */
int probe_interval; /* Secs of inactivity before sending probe. */
+ time_t last_activity; /* Last time we saw some activity. */
/* 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
*
* 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
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 *);
* 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 = xzalloc(sizeof *rc);
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;
rc->creation_time = time_now();
rc->total_time_connected = 0;
- rc->questionable_connectivity = false;
- rc->last_questioned = time_now();
+ rc->last_activity = time_now();
rconn_set_probe_interval(rc, probe_interval);
+ rconn_set_dscp(rc, dscp);
rc->n_monitors = 0;
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)
{
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]);
}
VLOG_INFO("%s: connecting...", rc->name);
}
rc->n_attempted_connections++;
- retval = vconn_open(rc->target, OFP_VERSION, &rc->vconn);
+ retval = vconn_open(rc->target, 0, &rc->vconn, rc->dscp);
if (!retval) {
rc->remote_ip = vconn_get_remote_ip(rc->vconn);
rc->local_ip = vconn_get_local_ip(rc->vconn);
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;
}
+ rc->last_activity = time_now();
}
- if (!rc->txq.n) {
+ if (list_is_empty(&rc->txq)) {
poll_immediate_wake();
}
}
timeout_ACTIVE(const struct rconn *rc)
{
if (rc->probe_interval) {
- unsigned int base = MAX(rc->last_received, rc->state_entered);
+ unsigned int base = MAX(rc->last_activity, rc->state_entered);
unsigned int arg = base + rc->probe_interval - rc->state_entered;
return arg;
}
run_ACTIVE(struct rconn *rc)
{
if (timed_out(rc)) {
- unsigned int base = MAX(rc->last_received, rc->state_entered);
+ unsigned int base = MAX(rc->last_activity, rc->state_entered);
+ int version;
+
VLOG_DBG("%s: idle %u seconds, sending inactivity probe",
rc->name, (unsigned int) (time_now() - base));
+ version = rconn_get_version(rc);
+ assert(version >= 0 && version <= 0xff);
+
/* Ordering is important here: rconn_send() can transition to BACKOFF,
* and we don't want to transition back to IDLE if so, because then we
* can end up queuing a packet with vconn == NULL and then *boom*. */
state_transition(rc, S_IDLE);
- rconn_send(rc, make_echo_request(), NULL);
+ rconn_send(rc, make_echo_request(version), NULL);
return;
}
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));
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]);
long long int expires = sat_add(rc->state_entered, timeo);
poll_timer_wait_until(expires * 1000);
}
-
- if ((rc->state & (S_ACTIVE | S_IDLE)) && rc->txq.n) {
- vconn_wait(rc->vconn, WAIT_SEND);
- }
}
/* Attempts to receive a packet from 'rc'. If successful, returns the packet;
rc->probably_admitted = true;
rc->last_admitted = time_now();
}
- rc->last_received = time_now();
+ rc->last_activity = time_now();
rc->packets_received++;
if (rc->state == S_IDLE) {
state_transition(rc, S_ACTIVE);
}
}
-/* 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
copy_to_monitor(rc, b);
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;
}
}
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;
}
/* Returns the IP address of the peer, or 0 if the peer's IP address is not
* known. */
-uint32_t
+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
+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
* known. */
-uint32_t
+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
+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
return state_name(rc->state);
}
-/* Returns the number of connection attempts made by 'rc', including any
- * ongoing attempt that has not yet succeeded or failed. */
-unsigned int
-rconn_get_attempted_connections(const struct rconn *rc)
-{
- return rc->n_attempted_connections;
-}
-
-/* Returns the number of successful connection attempts made by 'rc'. */
-unsigned int
-rconn_get_successful_connections(const struct rconn *rc)
-{
- return rc->n_successful_connections;
-}
-
/* 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 the last OpenFlow message was received by 'rc'.
- * If no packets have been received on 'rc', returns the time at which 'rc'
- * was created. */
-time_t
-rconn_get_last_received(const struct rconn *rc)
-{
- return rc->last_received;
-}
-
-/* Returns the time at which 'rc' was created. */
+/* Returns the time at which 'rc' was last disconnected. Returns TIME_MIN
+ * if never disconnected. */
time_t
-rconn_get_creation_time(const struct rconn *rc)
-{
- return rc->creation_time;
-}
-
-/* Returns the approximate number of seconds that 'rc' has been connected. */
-unsigned long int
-rconn_get_total_time_connected(const struct rconn *rc)
-{
- return (rc->total_time_connected
- + (rconn_is_connected(rc) ? elapsed_in_this_state(rc) : 0));
-}
-
-/* Returns the current amount of backoff, in seconds. This is the amount of
- * time after which the rconn will transition from BACKOFF to CONNECTING. */
-int
-rconn_get_backoff(const struct rconn *rc)
-{
- return rc->backoff;
-}
-
-/* Returns the number of seconds spent in this state so far. */
-unsigned int
-rconn_get_state_elapsed(const struct rconn *rc)
+rconn_get_last_disconnect(const struct rconn *rc)
{
- return elapsed_in_this_state(rc);
+ return rc->last_disconnected;
}
/* Returns 'rc''s current connection sequence number, a number that changes
{
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);
+}
\f
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;
}
{
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);
+ }
}
}
\f
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_p;
- 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);
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;
}
time_t now = time_now();
if (rc->state & (S_CONNECTING | S_ACTIVE | S_IDLE)) {
+ rc->last_disconnected = now;
vconn_close(rc->vconn);
rc->vconn = NULL;
flush_queue(rc);
}
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);
}
}
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);
+ 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);
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)
{
static bool
is_admitted_msg(const struct ofpbuf *b)
{
- struct ofp_header *oh = b->data;
- uint8_t type = oh->type;
- return !(type < 32
- && (1u << type) & ((1u << OFPT_HELLO) |
- (1u << OFPT_ERROR) |
- (1u << OFPT_ECHO_REQUEST) |
- (1u << OFPT_ECHO_REPLY) |
- (1u << OFPT_VENDOR) |
- (1u << OFPT_FEATURES_REQUEST) |
- (1u << OFPT_FEATURES_REPLY) |
- (1u << OFPT_GET_CONFIG_REQUEST) |
- (1u << OFPT_GET_CONFIG_REPLY) |
- (1u << OFPT_SET_CONFIG)));
+ enum ofptype type;
+ enum ofperr error;
+
+ error = ofptype_decode(&type, b->data);
+ if (error) {
+ return false;
+ }
+
+ switch (type) {
+ case OFPTYPE_HELLO:
+ case OFPTYPE_ERROR:
+ case OFPTYPE_ECHO_REQUEST:
+ case OFPTYPE_ECHO_REPLY:
+ case OFPTYPE_FEATURES_REQUEST:
+ case OFPTYPE_FEATURES_REPLY:
+ case OFPTYPE_GET_CONFIG_REQUEST:
+ case OFPTYPE_GET_CONFIG_REPLY:
+ case OFPTYPE_SET_CONFIG:
+ return false;
+
+ case OFPTYPE_PACKET_IN:
+ case OFPTYPE_FLOW_REMOVED:
+ case OFPTYPE_PORT_STATUS:
+ case OFPTYPE_PACKET_OUT:
+ case OFPTYPE_FLOW_MOD:
+ case OFPTYPE_PORT_MOD:
+ case OFPTYPE_BARRIER_REQUEST:
+ case OFPTYPE_BARRIER_REPLY:
+ case OFPTYPE_DESC_STATS_REQUEST:
+ case OFPTYPE_DESC_STATS_REPLY:
+ case OFPTYPE_FLOW_STATS_REQUEST:
+ case OFPTYPE_FLOW_STATS_REPLY:
+ case OFPTYPE_AGGREGATE_STATS_REQUEST:
+ case OFPTYPE_AGGREGATE_STATS_REPLY:
+ case OFPTYPE_TABLE_STATS_REQUEST:
+ case OFPTYPE_TABLE_STATS_REPLY:
+ case OFPTYPE_PORT_STATS_REQUEST:
+ case OFPTYPE_PORT_STATS_REPLY:
+ case OFPTYPE_QUEUE_STATS_REQUEST:
+ case OFPTYPE_QUEUE_STATS_REPLY:
+ case OFPTYPE_PORT_DESC_STATS_REQUEST:
+ case OFPTYPE_PORT_DESC_STATS_REPLY:
+ case OFPTYPE_ROLE_REQUEST:
+ case OFPTYPE_ROLE_REPLY:
+ case OFPTYPE_SET_FLOW_FORMAT:
+ case OFPTYPE_FLOW_MOD_TABLE_ID:
+ case OFPTYPE_SET_PACKET_IN_FORMAT:
+ case OFPTYPE_FLOW_AGE:
+ case OFPTYPE_SET_ASYNC_CONFIG:
+ case OFPTYPE_SET_CONTROLLER_ID:
+ case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
+ case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
+ case OFPTYPE_FLOW_MONITOR_CANCEL:
+ case OFPTYPE_FLOW_MONITOR_PAUSED:
+ case OFPTYPE_FLOW_MONITOR_RESUMED:
+ default:
+ return true;
+ }
}
/* Returns true if 'rc' is currently logging information about connection