/*
- * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira, Inc.
+ * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 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 "byte-order.h"
#include "classifier.h"
+#include "dp-packet.h"
#include "flow.h"
#include "hmap.h"
#include "mac-learning.h"
#include "shash.h"
#include "simap.h"
#include "timeval.h"
-#include "vconn.h"
-#include "vlog.h"
+#include "openvswitch/vconn.h"
+#include "openvswitch/vlog.h"
VLOG_DEFINE_THIS_MODULE(learning_switch);
struct lswitch_port {
struct hmap_node hmap_node; /* Hash node for port number. */
- uint16_t port_no; /* OpenFlow port number, in host byte order. */
+ ofp_port_t port_no; /* OpenFlow port number. */
uint32_t queue_id; /* OpenFlow queue number. */
};
* to set up the flow table. */
const struct ofputil_flow_mod *default_flows;
size_t n_default_flows;
+ enum ofputil_protocol usable_protocols;
};
/* The log messages here could actually be useful in debugging, so keep the
static void process_packet_in(struct lswitch *, const struct ofp_header *);
static void process_echo_request(struct lswitch *, const struct ofp_header *);
+static ofp_port_t get_mac_entry_ofp_port(const struct mac_learning *ml,
+ const struct mac_entry *)
+ OVS_REQ_RDLOCK(ml->rwlock);
+static void set_mac_entry_ofp_port(struct mac_learning *ml,
+ struct mac_entry *, ofp_port_t)
+ OVS_REQ_WRLOCK(ml->rwlock);
+
/* Creates and returns a new learning switch whose configuration is given by
* 'cfg'.
*
sw->default_flows = cfg->default_flows;
sw->n_default_flows = cfg->n_default_flows;
+ sw->usable_protocols = cfg->usable_protocols;
sw->queued = rconn_packet_counter_create();
lswitch_handshake(struct lswitch *sw)
{
enum ofputil_protocol protocol;
+ enum ofp_version version;
send_features_request(sw);
- protocol = ofputil_protocol_from_ofp_version(rconn_get_version(sw->rconn));
+ version = rconn_get_version(sw->rconn);
+ protocol = ofputil_protocol_from_ofp_version(version);
+ if (version >= OFP13_VERSION) {
+ /* OpenFlow 1.3 and later by default drop packets that miss in the flow
+ * table. Set up a flow to send packets to the controller by
+ * default. */
+ struct ofpact_output output;
+ struct ofpbuf *msg;
+ int error;
+
+ ofpact_init_OUTPUT(&output);
+ output.port = OFPP_CONTROLLER;
+ output.max_len = OFP_DEFAULT_MISS_SEND_LEN;
+
+ struct ofputil_flow_mod fm = {
+ .match = MATCH_CATCHALL_INITIALIZER,
+ .priority = 0,
+ .table_id = 0,
+ .command = OFPFC_ADD,
+ .buffer_id = UINT32_MAX,
+ .out_port = OFPP_NONE,
+ .out_group = OFPG_ANY,
+ .ofpacts = &output.ofpact,
+ .ofpacts_len = sizeof output,
+ };
+
+ msg = ofputil_encode_flow_mod(&fm, protocol);
+ error = rconn_send(sw->rconn, msg, NULL);
+ if (error) {
+ VLOG_INFO_RL(&rl, "%s: failed to add default flow (%s)",
+ rconn_get_name(sw->rconn), ovs_strerror(error));
+ }
+ }
if (sw->default_flows) {
- enum ofputil_protocol usable_protocols;
struct ofpbuf *msg = NULL;
int error = 0;
size_t i;
* This could be improved by actually negotiating a mutually acceptable
* flow format with the switch, but that would require an asynchronous
* state machine. This version ought to work fine in practice. */
- usable_protocols = ofputil_flow_mod_usable_protocols(
- sw->default_flows, sw->n_default_flows);
- if (!(protocol & usable_protocols)) {
- enum ofputil_protocol want = rightmost_1bit(usable_protocols);
+ if (!(protocol & sw->usable_protocols)) {
+ enum ofputil_protocol want = rightmost_1bit(sw->usable_protocols);
while (!error) {
msg = ofputil_encode_set_protocol(protocol, want, &protocol);
if (!msg) {
error = rconn_send(sw->rconn, msg, NULL);
}
}
- if (protocol & usable_protocols) {
+ if (protocol & sw->usable_protocols) {
for (i = 0; !error && i < sw->n_default_flows; i++) {
msg = ofputil_encode_flow_mod(&sw->default_flows[i], protocol);
error = rconn_send(sw->rconn, msg, NULL);
if (error) {
VLOG_INFO_RL(&rl, "%s: failed to queue default flows (%s)",
- rconn_get_name(sw->rconn), strerror(error));
+ rconn_get_name(sw->rconn), ovs_strerror(error));
}
} else {
VLOG_INFO_RL(&rl, "%s: failed to set usable protocol",
free(node);
}
shash_destroy(&sw->queue_names);
- mac_learning_destroy(sw->ml);
+ mac_learning_unref(sw->ml);
rconn_packet_counter_destroy(sw->queued);
free(sw);
}
int i;
if (sw->ml) {
- mac_learning_run(sw->ml, NULL);
+ ovs_rwlock_wrlock(&sw->ml->rwlock);
+ mac_learning_run(sw->ml);
+ ovs_rwlock_unlock(&sw->ml->rwlock);
}
rconn_run(sw->rconn);
lswitch_wait(struct lswitch *sw)
{
if (sw->ml) {
+ ovs_rwlock_rdlock(&sw->ml->rwlock);
mac_learning_wait(sw->ml);
+ ovs_rwlock_unlock(&sw->ml->rwlock);
}
rconn_run_wait(sw->rconn);
rconn_recv_wait(sw->rconn);
return;
}
- switch (type) {
- case OFPTYPE_ECHO_REQUEST:
+ if (type == OFPTYPE_ECHO_REQUEST) {
process_echo_request(sw, msg->data);
- break;
-
- case OFPTYPE_FEATURES_REPLY:
+ } else if (type == OFPTYPE_FEATURES_REPLY) {
if (sw->state == S_FEATURES_REPLY) {
if (!process_switch_features(sw, msg->data)) {
sw->state = S_SWITCHING;
rconn_disconnect(sw->rconn);
}
}
- break;
-
- case OFPTYPE_PACKET_IN:
+ } else if (type == OFPTYPE_PACKET_IN) {
process_packet_in(sw, msg->data);
- break;
-
- case OFPTYPE_FLOW_REMOVED:
+ } else if (type == OFPTYPE_FLOW_REMOVED) {
/* Nothing to do. */
- break;
-
- case OFPTYPE_HELLO:
- case OFPTYPE_ERROR:
- case OFPTYPE_ECHO_REPLY:
- case OFPTYPE_FEATURES_REQUEST:
- case OFPTYPE_GET_CONFIG_REQUEST:
- case OFPTYPE_GET_CONFIG_REPLY:
- case OFPTYPE_SET_CONFIG:
- 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:
- if (VLOG_IS_DBG_ENABLED()) {
- char *s = ofp_to_string(msg->data, msg->size, 2);
- VLOG_DBG_RL(&rl, "%016llx: OpenFlow packet ignored: %s",
- sw->datapath_id, s);
- free(s);
- }
+ } else if (VLOG_IS_DBG_ENABLED()) {
+ char *s = ofp_to_string(msg->data, msg->size, 2);
+ VLOG_DBG_RL(&rl, "%016llx: OpenFlow packet ignored: %s",
+ sw->datapath_id, s);
+ free(s);
}
}
\f
send_features_request(struct lswitch *sw)
{
struct ofpbuf *b;
- struct ofp_switch_config *osc;
int ofp_version = rconn_get_version(sw->rconn);
- assert(ofp_version > 0 && ofp_version < 0xff);
+ ovs_assert(ofp_version > 0 && ofp_version < 0xff);
/* Send OFPT_FEATURES_REQUEST. */
b = ofpraw_alloc(OFPRAW_OFPT_FEATURES_REQUEST, ofp_version, 0);
queue_tx(sw, b);
/* Send OFPT_SET_CONFIG. */
- b = ofpraw_alloc(OFPRAW_OFPT_SET_CONFIG, ofp_version, sizeof *osc);
- osc = ofpbuf_put_zeros(b, sizeof *osc);
- osc->miss_send_len = htons(OFP_DEFAULT_MISS_SEND_LEN);
- queue_tx(sw, b);
+ struct ofputil_switch_config config = {
+ .miss_send_len = OFP_DEFAULT_MISS_SEND_LEN
+ };
+ queue_tx(sw, ofputil_encode_set_config(&config, ofp_version));
}
static void
} else {
VLOG_WARN_RL(&rl, "%016llx: %s: send: %s",
sw->datapath_id, rconn_get_name(sw->rconn),
- strerror(retval));
+ ovs_strerror(retval));
}
}
}
{
struct ofputil_switch_features features;
struct ofputil_phy_port port;
- enum ofperr error;
- struct ofpbuf b;
- error = ofputil_decode_switch_features(oh, &features, &b);
+ struct ofpbuf b = ofpbuf_const_initializer(oh, ntohs(oh->length));
+ enum ofperr error = ofputil_pull_switch_features(&b, &features);
if (error) {
VLOG_ERR("received invalid switch feature reply (%s)",
ofperr_to_string(error));
if (lp && hmap_node_is_null(&lp->hmap_node)) {
lp->port_no = port.port_no;
hmap_insert(&sw->queue_numbers, &lp->hmap_node,
- hash_int(lp->port_no, 0));
+ hash_ofp_port(lp->port_no));
}
}
return 0;
}
-static uint16_t
+static ofp_port_t
lswitch_choose_destination(struct lswitch *sw, const struct flow *flow)
{
- uint16_t out_port;
+ ofp_port_t out_port;
/* Learn the source MAC. */
- if (mac_learning_may_learn(sw->ml, flow->dl_src, 0)) {
- struct mac_entry *mac = mac_learning_insert(sw->ml, flow->dl_src, 0);
- if (mac_entry_is_new(mac) || mac->port.i != flow->in_port) {
- VLOG_DBG_RL(&rl, "%016llx: learned that "ETH_ADDR_FMT" is on "
- "port %"PRIu16, sw->datapath_id,
- ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
-
- mac->port.i = flow->in_port;
- mac_learning_changed(sw->ml, mac);
+ if (sw->ml) {
+ ovs_rwlock_wrlock(&sw->ml->rwlock);
+ if (mac_learning_may_learn(sw->ml, flow->dl_src, 0)) {
+ struct mac_entry *mac = mac_learning_insert(sw->ml, flow->dl_src,
+ 0);
+ if (get_mac_entry_ofp_port(sw->ml, mac)
+ != flow->in_port.ofp_port) {
+ VLOG_DBG_RL(&rl, "%016llx: learned that "ETH_ADDR_FMT" is on "
+ "port %"PRIu16, sw->datapath_id,
+ ETH_ADDR_ARGS(flow->dl_src),
+ flow->in_port.ofp_port);
+
+ set_mac_entry_ofp_port(sw->ml, mac, flow->in_port.ofp_port);
+ }
}
+ ovs_rwlock_unlock(&sw->ml->rwlock);
}
/* Drop frames for reserved multicast addresses. */
if (sw->ml) {
struct mac_entry *mac;
- mac = mac_learning_lookup(sw->ml, flow->dl_dst, 0, NULL);
+ ovs_rwlock_rdlock(&sw->ml->rwlock);
+ mac = mac_learning_lookup(sw->ml, flow->dl_dst, 0);
if (mac) {
- out_port = mac->port.i;
- if (out_port == flow->in_port) {
+ out_port = get_mac_entry_ofp_port(sw->ml, mac);
+ if (out_port == flow->in_port.ofp_port) {
/* Don't send a packet back out its input port. */
+ ovs_rwlock_unlock(&sw->ml->rwlock);
return OFPP_NONE;
}
}
+ ovs_rwlock_unlock(&sw->ml->rwlock);
}
/* Check if we need to use "NORMAL" action. */
}
static uint32_t
-get_queue_id(const struct lswitch *sw, uint16_t in_port)
+get_queue_id(const struct lswitch *sw, ofp_port_t in_port)
{
const struct lswitch_port *port;
- HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_int(in_port, 0),
+ HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_ofp_port(in_port),
&sw->queue_numbers) {
if (port->port_no == in_port) {
return port->queue_id;
process_packet_in(struct lswitch *sw, const struct ofp_header *oh)
{
struct ofputil_packet_in pi;
+ uint32_t buffer_id;
uint32_t queue_id;
- uint16_t out_port;
+ ofp_port_t out_port;
uint64_t ofpacts_stub[64 / 8];
struct ofpbuf ofpacts;
struct ofputil_packet_out po;
enum ofperr error;
- struct ofpbuf pkt;
+ struct dp_packet pkt;
struct flow flow;
- error = ofputil_decode_packet_in(&pi, oh);
+ error = ofputil_decode_packet_in(oh, true, &pi, NULL, &buffer_id, NULL);
if (error) {
VLOG_WARN_RL(&rl, "failed to decode packet-in: %s",
ofperr_to_string(error));
return;
}
- /* Extract flow data from 'opi' into 'flow'. */
- ofpbuf_use_const(&pkt, pi.packet, pi.packet_len);
- flow_extract(&pkt, 0, 0, NULL, pi.fmd.in_port, &flow);
- flow.tunnel.tun_id = pi.fmd.tun_id;
+ /* Extract flow data from 'pi' into 'flow'. */
+ dp_packet_use_const(&pkt, pi.packet, pi.packet_len);
+ flow_extract(&pkt, &flow);
+ flow.in_port.ofp_port = pi.flow_metadata.flow.in_port.ofp_port;
+ flow.tunnel.tun_id = pi.flow_metadata.flow.tunnel.tun_id;
/* Choose output port. */
out_port = lswitch_choose_destination(sw, &flow);
/* Make actions. */
- queue_id = get_queue_id(sw, pi.fmd.in_port);
+ queue_id = get_queue_id(sw, pi.flow_metadata.flow.in_port.ofp_port);
ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
if (out_port == OFPP_NONE) {
/* No actions. */
- } else if (queue_id == UINT32_MAX || out_port >= OFPP_MAX) {
+ } else if (queue_id == UINT32_MAX
+ || ofp_to_u16(out_port) >= ofp_to_u16(OFPP_MAX)) {
ofpact_put_OUTPUT(&ofpacts)->port = out_port;
} else {
struct ofpact_enqueue *enqueue = ofpact_put_ENQUEUE(&ofpacts);
enqueue->port = out_port;
enqueue->queue = queue_id;
}
- ofpact_pad(&ofpacts);
/* Prepare packet_out in case we need one. */
- po.buffer_id = pi.buffer_id;
- if (po.buffer_id == UINT32_MAX) {
- po.packet = pkt.data;
- po.packet_len = pkt.size;
+ po.buffer_id = buffer_id;
+ if (buffer_id == UINT32_MAX) {
+ po.packet = dp_packet_data(&pkt);
+ po.packet_len = dp_packet_size(&pkt);
} else {
po.packet = NULL;
po.packet_len = 0;
}
- po.in_port = pi.fmd.in_port;
+ po.in_port = pi.flow_metadata.flow.in_port.ofp_port;
po.ofpacts = ofpacts.data;
po.ofpacts_len = ofpacts.size;
/* Send the packet, and possibly the whole flow, to the output port. */
if (sw->max_idle >= 0 && (!sw->ml || out_port != OFPP_FLOOD)) {
- struct ofputil_flow_mod fm;
- struct ofpbuf *buffer;
-
/* The output port is known, or we always flood everything, so add a
* new flow. */
- memset(&fm, 0, sizeof fm);
+ struct ofputil_flow_mod fm = {
+ .priority = 1, /* Must be > 0 because of table-miss flow entry. */
+ .table_id = 0xff,
+ .command = OFPFC_ADD,
+ .idle_timeout = sw->max_idle,
+ .buffer_id = buffer_id,
+ .out_port = OFPP_NONE,
+ .ofpacts = ofpacts.data,
+ .ofpacts_len = ofpacts.size,
+ };
match_init(&fm.match, &flow, &sw->wc);
ofputil_normalize_match_quiet(&fm.match);
- fm.priority = 0;
- fm.table_id = 0xff;
- fm.command = OFPFC_ADD;
- fm.idle_timeout = sw->max_idle;
- fm.buffer_id = pi.buffer_id;
- fm.out_port = OFPP_NONE;
- fm.ofpacts = ofpacts.data;
- fm.ofpacts_len = ofpacts.size;
- buffer = ofputil_encode_flow_mod(&fm, sw->protocol);
+
+ struct ofpbuf *buffer = ofputil_encode_flow_mod(&fm, sw->protocol);
queue_tx(sw, buffer);
/* If the switch didn't buffer the packet, we need to send a copy. */
- if (pi.buffer_id == UINT32_MAX && out_port != OFPP_NONE) {
+ if (buffer_id == UINT32_MAX && out_port != OFPP_NONE) {
queue_tx(sw, ofputil_encode_packet_out(&po, sw->protocol));
}
} else {
/* We don't know that MAC, or we don't set up flows. Send along the
* packet without setting up a flow. */
- if (pi.buffer_id != UINT32_MAX || out_port != OFPP_NONE) {
+ if (buffer_id != UINT32_MAX || out_port != OFPP_NONE) {
queue_tx(sw, ofputil_encode_packet_out(&po, sw->protocol));
}
}
{
queue_tx(sw, make_echo_reply(rq));
}
+
+static ofp_port_t
+get_mac_entry_ofp_port(const struct mac_learning *ml,
+ const struct mac_entry *e)
+ OVS_REQ_RDLOCK(ml->rwlock)
+{
+ void *port = mac_entry_get_port(ml, e);
+ return (OVS_FORCE ofp_port_t) (uintptr_t) port;
+}
+
+static void
+set_mac_entry_ofp_port(struct mac_learning *ml,
+ struct mac_entry *e, ofp_port_t ofp_port)
+ OVS_REQ_WRLOCK(ml->rwlock)
+{
+ mac_entry_set_port(ml, e, (void *) (OVS_FORCE uintptr_t) ofp_port);
+}