/*
- * 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 <time.h>
+#include "byte-order.h"
+#include "classifier.h"
#include "flow.h"
+#include "hmap.h"
#include "mac-learning.h"
#include "ofpbuf.h"
+#include "ofp-actions.h"
+#include "ofp-errors.h"
#include "ofp-parse.h"
#include "ofp-print.h"
#include "ofp-util.h"
#include "openflow/openflow.h"
#include "poll-loop.h"
-#include "queue.h"
#include "rconn.h"
-#include "stp.h"
+#include "shash.h"
+#include "simap.h"
#include "timeval.h"
#include "vconn.h"
#include "vlog.h"
-#include "xtoxll.h"
-VLOG_DEFINE_THIS_MODULE(learning_switch)
+VLOG_DEFINE_THIS_MODULE(learning_switch);
-enum port_state {
- P_DISABLED = 1 << 0,
- P_LISTENING = 1 << 1,
- P_LEARNING = 1 << 2,
- P_FORWARDING = 1 << 3,
- P_BLOCKING = 1 << 4
+struct lswitch_port {
+ struct hmap_node hmap_node; /* Hash node for port number. */
+ uint16_t port_no; /* OpenFlow port number, in host byte order. */
+ uint32_t queue_id; /* OpenFlow queue number. */
};
struct lswitch {
* Otherwise, the switch processes every packet. */
int max_idle;
+ enum ofputil_protocol protocol;
unsigned long long int datapath_id;
- uint32_t capabilities;
time_t last_features_request;
struct mac_learning *ml; /* NULL to act as hub instead of switch. */
- uint32_t wildcards; /* Wildcards to apply to flows. */
+ struct flow_wildcards wc; /* Wildcards to apply to flows. */
bool action_normal; /* Use OFPP_NORMAL? */
- uint32_t queue; /* OpenFlow queue to use, or UINT32_MAX. */
+
+ /* Queue distribution. */
+ uint32_t default_queue; /* Default OpenFlow queue, or UINT32_MAX. */
+ struct hmap queue_numbers; /* Map from port number to lswitch_port. */
+ struct shash queue_names; /* Map from port name to lswitch_port. */
/* Number of outgoing queued packets on the rconn. */
struct rconn_packet_counter *queued;
-
- /* Spanning tree protocol implementation.
- *
- * We implement STP states by, whenever a port's STP state changes,
- * querying all the flows on the switch and then deleting any of them that
- * are inappropriate for a port's STP state. */
- long long int next_query; /* Next time at which to query all flows. */
- long long int last_query; /* Last time we sent a query. */
- long long int last_reply; /* Last time we received a query reply. */
- unsigned int port_states[STP_MAX_PORTS];
- uint32_t query_xid; /* XID used for query. */
- int n_flows, n_no_recv, n_no_send;
};
/* The log messages here could actually be useful in debugging, so keep the
static void queue_tx(struct lswitch *, struct rconn *, struct ofpbuf *);
static void send_features_request(struct lswitch *, struct rconn *);
-static void send_default_flows(struct lswitch *sw, struct rconn *rconn,
- FILE *default_flows);
-static void schedule_query(struct lswitch *, long long int delay);
-static bool may_learn(const struct lswitch *, uint16_t port_no);
-static bool may_recv(const struct lswitch *, uint16_t port_no,
- bool any_actions);
-static bool may_send(const struct lswitch *, uint16_t port_no);
-
-typedef void packet_handler_func(struct lswitch *, struct rconn *, void *);
-static packet_handler_func process_switch_features;
-static packet_handler_func process_packet_in;
-static packet_handler_func process_echo_request;
-static packet_handler_func process_port_status;
-static packet_handler_func process_phy_port;
-static packet_handler_func process_stats_reply;
-
-/* Creates and returns a new learning switch.
- *
- * If 'learn_macs' is true, the new switch will learn the ports on which MAC
- * addresses appear. Otherwise, the new switch will flood all packets.
- *
- * If 'max_idle' is nonnegative, the new switch will set up flows that expire
- * after the given number of seconds (or never expire, if 'max_idle' is
- * OFP_FLOW_PERMANENT). Otherwise, the new switch will process every packet.
- *
- * The caller may provide the file stream 'default_flows' that defines
- * default flows that should be pushed when a switch connects. Each
- * line is a flow entry in the format described for "add-flows" command
- * in the Flow Syntax section of the ovs-ofct(8) man page. The caller
- * is responsible for closing the stream.
+
+static enum ofperr process_switch_features(struct lswitch *,
+ struct ofp_switch_features *);
+static void process_packet_in(struct lswitch *, struct rconn *,
+ const struct ofp_header *);
+static void process_echo_request(struct lswitch *, struct rconn *,
+ const struct ofp_header *);
+
+/* Creates and returns a new learning switch whose configuration is given by
+ * 'cfg'.
*
* 'rconn' is used to send out an OpenFlow features request. */
struct lswitch *
-lswitch_create(struct rconn *rconn, bool learn_macs,
- bool exact_flows, int max_idle, bool action_normal,
- FILE *default_flows)
+lswitch_create(struct rconn *rconn, const struct lswitch_config *cfg)
{
+ enum ofputil_protocol protocol;
struct lswitch *sw;
- size_t i;
sw = xzalloc(sizeof *sw);
- sw->max_idle = max_idle;
+ sw->max_idle = cfg->max_idle;
sw->datapath_id = 0;
sw->last_features_request = time_now() - 1;
- sw->ml = learn_macs ? mac_learning_create() : NULL;
- sw->action_normal = action_normal;
- if (exact_flows) {
- /* Exact match. */
- sw->wildcards = 0;
- } else {
- /* We cannot wildcard all fields.
- * We need in_port to detect moves.
- * We need both SA and DA to do learning. */
- sw->wildcards = (OFPFW_DL_TYPE | OFPFW_NW_SRC_MASK | OFPFW_NW_DST_MASK
- | OFPFW_NW_PROTO | OFPFW_TP_SRC | OFPFW_TP_DST);
+ sw->ml = (cfg->mode == LSW_LEARN
+ ? mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME)
+ : NULL);
+ sw->action_normal = cfg->mode == LSW_NORMAL;
+
+ flow_wildcards_init_exact(&sw->wc);
+ if (cfg->wildcards) {
+ uint32_t ofpfw;
+
+ if (cfg->wildcards == UINT32_MAX) {
+ /* Try to wildcard as many fields as possible, but we cannot
+ * wildcard all fields. We need in_port to detect moves. We need
+ * Ethernet source and dest and VLAN VID to do L2 learning. */
+ ofpfw = (OFPFW10_DL_TYPE | OFPFW10_DL_VLAN_PCP
+ | OFPFW10_NW_SRC_ALL | OFPFW10_NW_DST_ALL
+ | OFPFW10_NW_TOS | OFPFW10_NW_PROTO
+ | OFPFW10_TP_SRC | OFPFW10_TP_DST);
+ } else {
+ ofpfw = cfg->wildcards;
+ }
+
+ ofputil_wildcard_from_ofpfw10(ofpfw, &sw->wc);
}
- sw->queue = UINT32_MAX;
- sw->queued = rconn_packet_counter_create();
- sw->next_query = LLONG_MIN;
- sw->last_query = LLONG_MIN;
- sw->last_reply = LLONG_MIN;
- for (i = 0; i < STP_MAX_PORTS; i++) {
- sw->port_states[i] = P_DISABLED;
+
+ sw->default_queue = cfg->default_queue;
+ hmap_init(&sw->queue_numbers);
+ shash_init(&sw->queue_names);
+ if (cfg->port_queues) {
+ struct simap_node *node;
+
+ SIMAP_FOR_EACH (node, cfg->port_queues) {
+ struct lswitch_port *port = xmalloc(sizeof *port);
+ hmap_node_nullify(&port->hmap_node);
+ port->queue_id = node->data;
+ shash_add(&sw->queue_names, node->name, port);
+ }
}
+
+ sw->queued = rconn_packet_counter_create();
send_features_request(sw, rconn);
- if (default_flows) {
- send_default_flows(sw, rconn, default_flows);
+
+ protocol = ofputil_protocol_from_ofp_version(rconn_get_version(rconn));
+ if (cfg->default_flows) {
+ enum ofputil_protocol usable_protocols;
+ struct ofpbuf *msg = NULL;
+ int error = 0;
+ size_t i;
+
+ /* If the initial protocol isn't good enough for default_flows, then
+ * pick one that will work and encode messages to set up that
+ * protocol.
+ *
+ * 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(
+ cfg->default_flows, cfg->n_default_flows);
+ if (!(protocol & usable_protocols)) {
+ enum ofputil_protocol want = rightmost_1bit(usable_protocols);
+ while (!error) {
+ msg = ofputil_encode_set_protocol(protocol, want, &protocol);
+ if (!msg) {
+ break;
+ }
+ error = rconn_send(rconn, msg, NULL);
+ }
+ }
+
+ for (i = 0; !error && i < cfg->n_default_flows; i++) {
+ msg = ofputil_encode_flow_mod(&cfg->default_flows[i], protocol);
+ error = rconn_send(rconn, msg, NULL);
+ }
+
+ if (error) {
+ VLOG_INFO_RL(&rl, "%s: failed to queue default flows (%s)",
+ rconn_get_name(rconn), strerror(error));
+ }
}
+ sw->protocol = protocol;
+
return sw;
}
lswitch_destroy(struct lswitch *sw)
{
if (sw) {
+ struct lswitch_port *node, *next;
+
+ HMAP_FOR_EACH_SAFE (node, next, hmap_node, &sw->queue_numbers) {
+ hmap_remove(&sw->queue_numbers, &node->hmap_node);
+ free(node);
+ }
+ shash_destroy(&sw->queue_names);
mac_learning_destroy(sw->ml);
rconn_packet_counter_destroy(sw->queued);
free(sw);
}
}
-/* Sets 'queue' as the OpenFlow queue used by packets and flows set up by 'sw'.
- * Specify UINT32_MAX to avoid specifying a particular queue, which is also the
- * default if this function is never called for 'sw'. */
-void
-lswitch_set_queue(struct lswitch *sw, uint32_t queue)
-{
- sw->queue = queue;
-}
-
/* Takes care of necessary 'sw' activity, except for receiving packets (which
* the caller must do). */
void
-lswitch_run(struct lswitch *sw, struct rconn *rconn)
+lswitch_run(struct lswitch *sw)
{
- long long int now = time_msec();
-
if (sw->ml) {
mac_learning_run(sw->ml, NULL);
}
-
- /* If we're waiting for more replies, keeping waiting for up to 10 s. */
- if (sw->last_reply != LLONG_MIN) {
- if (now - sw->last_reply > 10000) {
- VLOG_ERR_RL(&rl, "%016llx: No more flow stat replies last 10 s",
- sw->datapath_id);
- sw->last_reply = LLONG_MIN;
- sw->last_query = LLONG_MIN;
- schedule_query(sw, 0);
- } else {
- return;
- }
- }
-
- /* If we're waiting for any reply at all, keep waiting for up to 10 s. */
- if (sw->last_query != LLONG_MIN) {
- if (now - sw->last_query > 10000) {
- VLOG_ERR_RL(&rl, "%016llx: No flow stat replies in last 10 s",
- sw->datapath_id);
- sw->last_query = LLONG_MIN;
- schedule_query(sw, 0);
- } else {
- return;
- }
- }
-
- /* If it's time to send another query, do so. */
- if (sw->next_query != LLONG_MIN && now >= sw->next_query) {
- sw->next_query = LLONG_MIN;
- if (!rconn_is_connected(rconn)) {
- schedule_query(sw, 1000);
- } else {
- struct ofp_stats_request *osr;
- struct ofp_flow_stats_request *ofsr;
- struct ofpbuf *b;
- int error;
-
- VLOG_DBG("%016llx: Sending flow stats request to implement STP",
- sw->datapath_id);
-
- sw->last_query = now;
- sw->query_xid = random_uint32();
- sw->n_flows = 0;
- sw->n_no_recv = 0;
- sw->n_no_send = 0;
- osr = make_openflow_xid(sizeof *osr + sizeof *ofsr,
- OFPT_STATS_REQUEST, sw->query_xid, &b);
- osr->type = htons(OFPST_FLOW);
- osr->flags = htons(0);
- ofsr = (struct ofp_flow_stats_request *) osr->body;
- ofsr->match.wildcards = htonl(OFPFW_ALL);
- ofsr->table_id = 0xff;
- ofsr->out_port = htons(OFPP_NONE);
-
- error = rconn_send(rconn, b, NULL);
- if (error) {
- VLOG_WARN_RL(&rl, "%016llx: sending flow stats request "
- "failed: %s", sw->datapath_id, strerror(error));
- ofpbuf_delete(b);
- schedule_query(sw, 1000);
- }
- }
- }
-}
-
-static void
-wait_timeout(long long int started)
-{
- poll_timer_wait_until(started + 10000);
}
void
if (sw->ml) {
mac_learning_wait(sw->ml);
}
-
- if (sw->last_reply != LLONG_MIN) {
- wait_timeout(sw->last_reply);
- } else if (sw->last_query != LLONG_MIN) {
- wait_timeout(sw->last_query);
- }
}
/* Processes 'msg', which should be an OpenFlow received on 'rconn', according
lswitch_process_packet(struct lswitch *sw, struct rconn *rconn,
const struct ofpbuf *msg)
{
- struct processor {
- uint8_t type;
- size_t min_size;
- packet_handler_func *handler;
- };
- static const struct processor processors[] = {
- {
- OFPT_ECHO_REQUEST,
- sizeof(struct ofp_header),
- process_echo_request
- },
- {
- OFPT_FEATURES_REPLY,
- sizeof(struct ofp_switch_features),
- process_switch_features
- },
- {
- OFPT_PACKET_IN,
- offsetof(struct ofp_packet_in, data),
- process_packet_in
- },
- {
- OFPT_PORT_STATUS,
- sizeof(struct ofp_port_status),
- process_port_status
- },
- {
- OFPT_STATS_REPLY,
- offsetof(struct ofp_stats_reply, body),
- process_stats_reply
- },
- {
- OFPT_FLOW_REMOVED,
- sizeof(struct ofp_flow_removed),
- NULL
- },
- };
- const size_t n_processors = ARRAY_SIZE(processors);
- const struct processor *p;
- struct ofp_header *oh;
-
- oh = msg->data;
+ const struct ofp_header *oh = msg->data;
+ const struct ofputil_msg_type *type;
+
if (sw->datapath_id == 0
&& oh->type != OFPT_ECHO_REQUEST
&& oh->type != OFPT_FEATURES_REPLY) {
return;
}
- for (p = processors; p < &processors[n_processors]; p++) {
- if (oh->type == p->type) {
- if (msg->size < p->min_size) {
- VLOG_WARN_RL(&rl, "%016llx: %s: too short (%zu bytes) for "
- "type %"PRIu8" (min %zu)", sw->datapath_id,
- rconn_get_name(rconn), msg->size, oh->type,
- p->min_size);
- return;
- }
- if (p->handler) {
- (p->handler)(sw, rconn, msg->data);
- }
- return;
+ ofputil_decode_msg_type(oh, &type);
+ switch (ofputil_msg_type_code(type)) {
+ case OFPUTIL_OFPT_ECHO_REQUEST:
+ process_echo_request(sw, rconn, msg->data);
+ break;
+
+ case OFPUTIL_OFPT_FEATURES_REPLY:
+ process_switch_features(sw, msg->data);
+ break;
+
+ case OFPUTIL_OFPT_PACKET_IN:
+ case OFPUTIL_NXT_PACKET_IN:
+ process_packet_in(sw, rconn, msg->data);
+ break;
+
+ case OFPUTIL_OFPT_FLOW_REMOVED:
+ /* Nothing to do. */
+ break;
+
+ case OFPUTIL_MSG_INVALID:
+ case OFPUTIL_OFPT_HELLO:
+ case OFPUTIL_OFPT_ERROR:
+ case OFPUTIL_OFPT_ECHO_REPLY:
+ case OFPUTIL_OFPT_FEATURES_REQUEST:
+ case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
+ case OFPUTIL_OFPT_GET_CONFIG_REPLY:
+ case OFPUTIL_OFPT_SET_CONFIG:
+ case OFPUTIL_OFPT_PORT_STATUS:
+ case OFPUTIL_OFPT_PACKET_OUT:
+ case OFPUTIL_OFPT_FLOW_MOD:
+ case OFPUTIL_OFPT_PORT_MOD:
+ case OFPUTIL_OFPT_BARRIER_REQUEST:
+ case OFPUTIL_OFPT_BARRIER_REPLY:
+ case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
+ case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
+ case OFPUTIL_OFPST_DESC_REQUEST:
+ case OFPUTIL_OFPST_FLOW_REQUEST:
+ case OFPUTIL_OFPST_AGGREGATE_REQUEST:
+ case OFPUTIL_OFPST_TABLE_REQUEST:
+ case OFPUTIL_OFPST_PORT_REQUEST:
+ case OFPUTIL_OFPST_QUEUE_REQUEST:
+ case OFPUTIL_OFPST_PORT_DESC_REQUEST:
+ case OFPUTIL_OFPST_DESC_REPLY:
+ case OFPUTIL_OFPST_FLOW_REPLY:
+ case OFPUTIL_OFPST_QUEUE_REPLY:
+ case OFPUTIL_OFPST_PORT_REPLY:
+ case OFPUTIL_OFPST_TABLE_REPLY:
+ case OFPUTIL_OFPST_AGGREGATE_REPLY:
+ case OFPUTIL_OFPST_PORT_DESC_REPLY:
+ case OFPUTIL_NXT_ROLE_REQUEST:
+ case OFPUTIL_NXT_ROLE_REPLY:
+ case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
+ case OFPUTIL_NXT_SET_FLOW_FORMAT:
+ case OFPUTIL_NXT_SET_PACKET_IN_FORMAT:
+ case OFPUTIL_NXT_FLOW_MOD:
+ case OFPUTIL_NXT_FLOW_REMOVED:
+ case OFPUTIL_NXT_FLOW_AGE:
+ case OFPUTIL_NXT_SET_ASYNC_CONFIG:
+ case OFPUTIL_NXT_SET_CONTROLLER_ID:
+ case OFPUTIL_NXST_FLOW_REQUEST:
+ case OFPUTIL_NXST_AGGREGATE_REQUEST:
+ case OFPUTIL_NXST_FLOW_REPLY:
+ case OFPUTIL_NXST_AGGREGATE_REPLY:
+ 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);
}
}
- if (VLOG_IS_DBG_ENABLED()) {
- char *p = ofp_to_string(msg->data, msg->size, 2);
- VLOG_DBG_RL(&rl, "%016llx: OpenFlow packet ignored: %s",
- sw->datapath_id, p);
- free(p);
- }
}
\f
static void
}
}
-static void
-send_default_flows(struct lswitch *sw, struct rconn *rconn,
- FILE *default_flows)
-{
- char line[1024];
-
- while (fgets(line, sizeof line, default_flows)) {
- struct ofpbuf *b;
- struct ofp_flow_mod *ofm;
- uint16_t priority, idle_timeout, hard_timeout;
- uint64_t cookie;
- struct ofp_match match;
-
- char *comment;
-
- /* Delete comments. */
- comment = strchr(line, '#');
- if (comment) {
- *comment = '\0';
- }
-
- /* Drop empty lines. */
- if (line[strspn(line, " \t\n")] == '\0') {
- continue;
- }
-
- /* Parse and send. str_to_flow() will expand and reallocate the data
- * in 'buffer', so we can't keep pointers to across the str_to_flow()
- * call. */
- make_openflow(sizeof *ofm, OFPT_FLOW_MOD, &b);
- parse_ofp_str(line, &match, b,
- NULL, NULL, &priority, &idle_timeout, &hard_timeout,
- &cookie);
- ofm = b->data;
- ofm->match = match;
- ofm->command = htons(OFPFC_ADD);
- ofm->cookie = htonll(cookie);
- ofm->idle_timeout = htons(idle_timeout);
- ofm->hard_timeout = htons(hard_timeout);
- ofm->buffer_id = htonl(UINT32_MAX);
- ofm->priority = htons(priority);
-
- update_openflow_length(b);
- queue_tx(sw, rconn, b);
- }
-}
-
static void
queue_tx(struct lswitch *sw, struct rconn *rconn, struct ofpbuf *b)
{
}
}
-static void
-schedule_query(struct lswitch *sw, long long int delay)
+static enum ofperr
+process_switch_features(struct lswitch *sw, struct ofp_switch_features *osf)
{
- long long int now = time_msec();
- if (sw->next_query == LLONG_MIN || sw->next_query > now + delay) {
- sw->next_query = now + delay;
- }
-}
-
-static void
-process_switch_features(struct lswitch *sw, struct rconn *rconn, void *osf_)
-{
- struct ofp_switch_features *osf = osf_;
- size_t n_ports = ((ntohs(osf->header.length)
- - offsetof(struct ofp_switch_features, ports))
- / sizeof *osf->ports);
- size_t i;
-
- sw->datapath_id = ntohll(osf->datapath_id);
- sw->capabilities = ntohl(osf->capabilities);
- for (i = 0; i < n_ports; i++) {
- process_phy_port(sw, rconn, &osf->ports[i]);
- }
- if (sw->capabilities & OFPC_STP) {
- schedule_query(sw, 1000);
+ struct ofputil_switch_features features;
+ struct ofputil_phy_port port;
+ enum ofperr error;
+ struct ofpbuf b;
+
+ error = ofputil_decode_switch_features(osf, &features, &b);
+ if (error) {
+ VLOG_ERR("received invalid switch feature reply (%s)",
+ ofperr_to_string(error));
+ return error;
+ }
+
+ sw->datapath_id = features.datapath_id;
+
+ while (!ofputil_pull_phy_port(osf->header.version, &b, &port)) {
+ struct lswitch_port *lp = shash_find_data(&sw->queue_names, port.name);
+ 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));
+ }
}
+ return 0;
}
static uint16_t
-lswitch_choose_destination(struct lswitch *sw, const flow_t *flow)
+lswitch_choose_destination(struct lswitch *sw, const struct flow *flow)
{
uint16_t out_port;
/* Learn the source MAC. */
- if (may_learn(sw, flow->in_port) && sw->ml) {
- if (mac_learning_learn(sw->ml, flow->dl_src, 0, flow->in_port,
- GRAT_ARP_LOCK_NONE)) {
+ 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);
}
}
return OFPP_NONE;
}
- if (!may_recv(sw, flow->in_port, false)) {
- /* STP prevents receiving anything on this port. */
- return OFPP_NONE;
- }
-
out_port = OFPP_FLOOD;
if (sw->ml) {
- int learned_port = mac_learning_lookup(sw->ml, flow->dl_dst, 0, NULL);
- if (learned_port >= 0 && may_send(sw, learned_port)) {
- out_port = learned_port;
+ struct mac_entry *mac;
+
+ mac = mac_learning_lookup(sw->ml, flow->dl_dst, 0, NULL);
+ if (mac) {
+ out_port = mac->port.i;
if (out_port == flow->in_port) {
/* Don't send a packet back out its input port. */
return OFPP_NONE;
return out_port;
}
+static uint32_t
+get_queue_id(const struct lswitch *sw, uint16_t in_port)
+{
+ const struct lswitch_port *port;
+
+ HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_int(in_port, 0),
+ &sw->queue_numbers) {
+ if (port->port_no == in_port) {
+ return port->queue_id;
+ }
+ }
+
+ return sw->default_queue;
+}
+
static void
-process_packet_in(struct lswitch *sw, struct rconn *rconn, void *opi_)
+process_packet_in(struct lswitch *sw, struct rconn *rconn,
+ const struct ofp_header *oh)
{
- struct ofp_packet_in *opi = opi_;
- uint16_t in_port = ntohs(opi->in_port);
+ struct ofputil_packet_in pi;
+ uint32_t queue_id;
uint16_t out_port;
- struct ofp_action_header actions[2];
- size_t actions_len;
+ uint64_t ofpacts_stub[64 / 8];
+ struct ofpbuf ofpacts;
+
+ struct ofputil_packet_out po;
+ enum ofperr error;
- size_t pkt_ofs, pkt_len;
struct ofpbuf pkt;
- flow_t flow;
+ struct flow flow;
+
+ error = ofputil_decode_packet_in(&pi, oh);
+ if (error) {
+ VLOG_WARN_RL(&rl, "failed to decode packet-in: %s",
+ ofperr_to_string(error));
+ return;
+ }
+
+ /* Ignore packets sent via output to OFPP_CONTROLLER. This library never
+ * uses such an action. You never know what experiments might be going on,
+ * though, and it seems best not to interfere with them. */
+ if (pi.reason != OFPR_NO_MATCH) {
+ return;
+ }
/* Extract flow data from 'opi' into 'flow'. */
- pkt_ofs = offsetof(struct ofp_packet_in, data);
- pkt_len = ntohs(opi->header.length) - pkt_ofs;
- pkt.data = opi->data;
- pkt.size = pkt_len;
- flow_extract(&pkt, 0, in_port, &flow);
+ ofpbuf_use_const(&pkt, pi.packet, pi.packet_len);
+ flow_extract(&pkt, 0, pi.fmd.tun_id, pi.fmd.in_port, &flow);
/* Choose output port. */
out_port = lswitch_choose_destination(sw, &flow);
/* Make actions. */
- memset(actions, 0, sizeof actions);
+ queue_id = get_queue_id(sw, pi.fmd.in_port);
+ ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
if (out_port == OFPP_NONE) {
- actions_len = 0;
- } else if (sw->queue == UINT32_MAX || out_port >= OFPP_MAX) {
- struct ofp_action_output *oao = (struct ofp_action_output *) actions;
- oao->type = htons(OFPAT_OUTPUT);
- oao->len = htons(sizeof *oao);
- oao->port = htons(out_port);
- actions_len = sizeof *oao;
+ /* No actions. */
+ } else if (queue_id == UINT32_MAX || out_port >= 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;
} else {
- struct ofp_action_enqueue *oae = (struct ofp_action_enqueue *) actions;
- oae->type = htons(OFPAT_ENQUEUE);
- oae->len = htons(sizeof *oae);
- oae->port = htons(out_port);
- oae->queue_id = htonl(sw->queue);
- actions_len = sizeof *oae;
+ po.packet = NULL;
+ po.packet_len = 0;
}
- assert(actions_len <= sizeof actions);
+ po.in_port = pi.fmd.in_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;
- struct ofp_flow_mod *ofm;
/* The output port is known, or we always flood everything, so add a
* new flow. */
- buffer = make_add_flow(&flow, ntohl(opi->buffer_id),
- sw->max_idle, actions_len);
- ofpbuf_put(buffer, actions, actions_len);
- ofm = buffer->data;
- ofm->match.wildcards = htonl(sw->wildcards);
+ memset(&fm, 0, sizeof fm);
+ cls_rule_init(&flow, &sw->wc, 0, &fm.cr);
+ 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);
+
queue_tx(sw, rconn, buffer);
/* If the switch didn't buffer the packet, we need to send a copy. */
- if (ntohl(opi->buffer_id) == UINT32_MAX && actions_len > 0) {
- queue_tx(sw, rconn,
- make_packet_out(&pkt, UINT32_MAX, in_port,
- actions, actions_len / sizeof *actions));
+ if (pi.buffer_id == UINT32_MAX && out_port != OFPP_NONE) {
+ queue_tx(sw, rconn, ofputil_encode_packet_out(&po));
}
} else {
/* We don't know that MAC, or we don't set up flows. Send along the
* packet without setting up a flow. */
- if (ntohl(opi->buffer_id) != UINT32_MAX || actions_len > 0) {
- queue_tx(sw, rconn,
- make_packet_out(&pkt, ntohl(opi->buffer_id), in_port,
- actions, actions_len / sizeof *actions));
+ if (pi.buffer_id != UINT32_MAX || out_port != OFPP_NONE) {
+ queue_tx(sw, rconn, ofputil_encode_packet_out(&po));
}
}
}
static void
-process_echo_request(struct lswitch *sw, struct rconn *rconn, void *rq_)
+process_echo_request(struct lswitch *sw, struct rconn *rconn,
+ const struct ofp_header *rq)
{
- struct ofp_header *rq = rq_;
queue_tx(sw, rconn, make_echo_reply(rq));
}
-
-static void
-process_port_status(struct lswitch *sw, struct rconn *rconn, void *ops_)
-{
- struct ofp_port_status *ops = ops_;
- process_phy_port(sw, rconn, &ops->desc);
-}
-
-static void
-process_phy_port(struct lswitch *sw, struct rconn *rconn OVS_UNUSED,
- void *opp_)
-{
- const struct ofp_phy_port *opp = opp_;
- uint16_t port_no = ntohs(opp->port_no);
- if (sw->capabilities & OFPC_STP && port_no < STP_MAX_PORTS) {
- uint32_t config = ntohl(opp->config);
- uint32_t state = ntohl(opp->state);
- unsigned int *port_state = &sw->port_states[port_no];
- unsigned int new_port_state;
-
- if (!(config & (OFPPC_NO_STP | OFPPC_PORT_DOWN))
- && !(state & OFPPS_LINK_DOWN))
- {
- switch (state & OFPPS_STP_MASK) {
- case OFPPS_STP_LISTEN:
- new_port_state = P_LISTENING;
- break;
- case OFPPS_STP_LEARN:
- new_port_state = P_LEARNING;
- break;
- case OFPPS_STP_FORWARD:
- new_port_state = P_FORWARDING;
- break;
- case OFPPS_STP_BLOCK:
- new_port_state = P_BLOCKING;
- break;
- default:
- new_port_state = P_DISABLED;
- break;
- }
- } else {
- new_port_state = P_FORWARDING;
- }
- if (*port_state != new_port_state) {
- *port_state = new_port_state;
- schedule_query(sw, 1000);
- }
- }
-}
-
-static unsigned int
-get_port_state(const struct lswitch *sw, uint16_t port_no)
-{
- return (port_no >= STP_MAX_PORTS || !(sw->capabilities & OFPC_STP)
- ? P_FORWARDING
- : sw->port_states[port_no]);
-}
-
-static bool
-may_learn(const struct lswitch *sw, uint16_t port_no)
-{
- return get_port_state(sw, port_no) & (P_LEARNING | P_FORWARDING);
-}
-
-static bool
-may_recv(const struct lswitch *sw, uint16_t port_no, bool any_actions)
-{
- unsigned int state = get_port_state(sw, port_no);
- return !(any_actions
- ? state & (P_DISABLED | P_LISTENING | P_BLOCKING)
- : state & (P_DISABLED | P_LISTENING | P_BLOCKING | P_LEARNING));
-}
-
-static bool
-may_send(const struct lswitch *sw, uint16_t port_no)
-{
- return get_port_state(sw, port_no) & P_FORWARDING;
-}
-
-static void
-process_flow_stats(struct lswitch *sw, struct rconn *rconn,
- const struct ofp_flow_stats *ofs)
-{
- const char *end = (char *) ofs + ntohs(ofs->length);
- bool delete = false;
-
- /* Decide to delete the flow if it matches on an STP-disabled physical
- * port. But don't delete it if the flow just drops all received packets,
- * because that's a perfectly reasonable thing to do for disabled physical
- * ports. */
- if (!(ofs->match.wildcards & htonl(OFPFW_IN_PORT))) {
- if (!may_recv(sw, ntohs(ofs->match.in_port),
- end > (char *) ofs->actions)) {
- delete = true;
- sw->n_no_recv++;
- }
- }
-
- /* Decide to delete the flow if it forwards to an STP-disabled physical
- * port. */
- if (!delete) {
- const struct ofp_action_header *a;
- size_t len;
-
- for (a = ofs->actions; (char *) a < end; a += len / 8) {
- len = ntohs(a->len);
- if (len > end - (char *) a) {
- VLOG_DBG_RL(&rl, "%016llx: action exceeds available space "
- "(%zu > %td)",
- sw->datapath_id, len, end - (char *) a);
- break;
- } else if (len % 8) {
- VLOG_DBG_RL(&rl, "%016llx: action length (%zu) not multiple "
- "of 8 bytes", sw->datapath_id, len);
- break;
- }
-
- if (a->type == htons(OFPAT_OUTPUT)) {
- struct ofp_action_output *oao = (struct ofp_action_output *) a;
- if (!may_send(sw, ntohs(oao->port))) {
- delete = true;
- sw->n_no_send++;
- break;
- }
- }
- }
- }
-
- /* Delete the flow. */
- if (delete) {
- struct ofp_flow_mod *ofm;
- struct ofpbuf *b;
-
- ofm = make_openflow(offsetof(struct ofp_flow_mod, actions),
- OFPT_FLOW_MOD, &b);
- ofm->match = ofs->match;
- ofm->command = OFPFC_DELETE_STRICT;
- rconn_send(rconn, b, NULL);
- }
-}
-
-static void
-process_stats_reply(struct lswitch *sw, struct rconn *rconn, void *osr_)
-{
- struct ofp_stats_reply *osr = osr_;
- struct flow_stats_iterator i;
- const struct ofp_flow_stats *fs;
-
- if (sw->last_query == LLONG_MIN
- || osr->type != htons(OFPST_FLOW)
- || osr->header.xid != sw->query_xid) {
- return;
- }
- for (fs = flow_stats_first(&i, osr); fs; fs = flow_stats_next(&i)) {
- sw->n_flows++;
- process_flow_stats(sw, rconn, fs);
- }
- if (!(osr->flags & htons(OFPSF_REPLY_MORE))) {
- VLOG_DBG("%016llx: Deleted %d of %d received flows to "
- "implement STP, %d because of no-recv, %d because of "
- "no-send", sw->datapath_id,
- sw->n_no_recv + sw->n_no_send, sw->n_flows,
- sw->n_no_recv, sw->n_no_send);
- sw->last_query = LLONG_MIN;
- sw->last_reply = LLONG_MIN;
- } else {
- sw->last_reply = time_msec();
- }
-}
-
projects / cascardo / ovs.git / blobdiff