2 * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
18 #include "learning-switch.h"
22 #include <netinet/in.h>
26 #include "byte-order.h"
27 #include "classifier.h"
30 #include "mac-learning.h"
32 #include "ofp-errors.h"
33 #include "ofp-parse.h"
34 #include "ofp-print.h"
36 #include "openflow/openflow.h"
37 #include "poll-loop.h"
44 VLOG_DEFINE_THIS_MODULE(learning_switch);
47 struct hmap_node hmap_node; /* Hash node for port number. */
48 uint16_t port_no; /* OpenFlow port number, in host byte order. */
49 uint32_t queue_id; /* OpenFlow queue number. */
53 /* If nonnegative, the switch sets up flows that expire after the given
54 * number of seconds (or never expire, if the value is OFP_FLOW_PERMANENT).
55 * Otherwise, the switch processes every packet. */
58 unsigned long long int datapath_id;
59 time_t last_features_request;
60 struct mac_learning *ml; /* NULL to act as hub instead of switch. */
61 struct flow_wildcards wc; /* Wildcards to apply to flows. */
62 bool action_normal; /* Use OFPP_NORMAL? */
64 /* Queue distribution. */
65 uint32_t default_queue; /* Default OpenFlow queue, or UINT32_MAX. */
66 struct hmap queue_numbers; /* Map from port number to lswitch_port. */
67 struct shash queue_names; /* Map from port name to lswitch_port. */
69 /* Number of outgoing queued packets on the rconn. */
70 struct rconn_packet_counter *queued;
73 /* The log messages here could actually be useful in debugging, so keep the
74 * rate limit relatively high. */
75 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
77 static void queue_tx(struct lswitch *, struct rconn *, struct ofpbuf *);
78 static void send_features_request(struct lswitch *, struct rconn *);
80 static void process_switch_features(struct lswitch *,
81 struct ofp_switch_features *);
82 static void process_packet_in(struct lswitch *, struct rconn *,
83 const struct ofp_packet_in *);
84 static void process_echo_request(struct lswitch *, struct rconn *,
85 const struct ofp_header *);
87 /* Creates and returns a new learning switch whose configuration is given by
90 * 'rconn' is used to send out an OpenFlow features request. */
92 lswitch_create(struct rconn *rconn, const struct lswitch_config *cfg)
96 sw = xzalloc(sizeof *sw);
97 sw->max_idle = cfg->max_idle;
99 sw->last_features_request = time_now() - 1;
100 sw->ml = cfg->mode == LSW_LEARN ? mac_learning_create() : NULL;
101 sw->action_normal = cfg->mode == LSW_NORMAL;
103 flow_wildcards_init_exact(&sw->wc);
104 if (cfg->wildcards) {
107 if (cfg->wildcards == UINT32_MAX) {
108 /* Try to wildcard as many fields as possible, but we cannot
109 * wildcard all fields. We need in_port to detect moves. We need
110 * Ethernet source and dest and VLAN VID to do L2 learning. */
111 ofpfw = (OFPFW_DL_TYPE | OFPFW_DL_VLAN_PCP
112 | OFPFW_NW_SRC_ALL | OFPFW_NW_DST_ALL
113 | OFPFW_NW_TOS | OFPFW_NW_PROTO
114 | OFPFW_TP_SRC | OFPFW_TP_DST);
116 ofpfw = cfg->wildcards;
119 ofputil_wildcard_from_openflow(ofpfw, &sw->wc);
122 sw->default_queue = cfg->default_queue;
123 hmap_init(&sw->queue_numbers);
124 shash_init(&sw->queue_names);
125 if (cfg->port_queues) {
126 struct shash_node *node;
128 SHASH_FOR_EACH (node, cfg->port_queues) {
129 struct lswitch_port *port = xmalloc(sizeof *port);
130 hmap_node_nullify(&port->hmap_node);
131 port->queue_id = (uintptr_t) node->data;
132 shash_add(&sw->queue_names, node->name, port);
136 sw->queued = rconn_packet_counter_create();
137 send_features_request(sw, rconn);
139 if (cfg->default_flows) {
140 const struct ofpbuf *b;
142 LIST_FOR_EACH (b, list_node, cfg->default_flows) {
143 struct ofpbuf *copy = ofpbuf_clone(b);
144 int error = rconn_send(rconn, copy, NULL);
146 VLOG_INFO_RL(&rl, "%s: failed to queue default flows (%s)",
147 rconn_get_name(rconn), strerror(error));
159 lswitch_destroy(struct lswitch *sw)
162 struct lswitch_port *node, *next;
164 HMAP_FOR_EACH_SAFE (node, next, hmap_node, &sw->queue_numbers) {
165 hmap_remove(&sw->queue_numbers, &node->hmap_node);
168 shash_destroy(&sw->queue_names);
169 mac_learning_destroy(sw->ml);
170 rconn_packet_counter_destroy(sw->queued);
175 /* Takes care of necessary 'sw' activity, except for receiving packets (which
176 * the caller must do). */
178 lswitch_run(struct lswitch *sw)
181 mac_learning_run(sw->ml, NULL);
186 lswitch_wait(struct lswitch *sw)
189 mac_learning_wait(sw->ml);
193 /* Processes 'msg', which should be an OpenFlow received on 'rconn', according
194 * to the learning switch state in 'sw'. The most likely result of processing
195 * is that flow-setup and packet-out OpenFlow messages will be sent out on
198 lswitch_process_packet(struct lswitch *sw, struct rconn *rconn,
199 const struct ofpbuf *msg)
201 const struct ofp_header *oh = msg->data;
202 const struct ofputil_msg_type *type;
204 if (sw->datapath_id == 0
205 && oh->type != OFPT_ECHO_REQUEST
206 && oh->type != OFPT_FEATURES_REPLY) {
207 send_features_request(sw, rconn);
211 ofputil_decode_msg_type(oh, &type);
212 switch (ofputil_msg_type_code(type)) {
213 case OFPUTIL_OFPT_ECHO_REQUEST:
214 process_echo_request(sw, rconn, msg->data);
217 case OFPUTIL_OFPT_FEATURES_REPLY:
218 process_switch_features(sw, msg->data);
221 case OFPUTIL_OFPT_PACKET_IN:
222 process_packet_in(sw, rconn, msg->data);
225 case OFPUTIL_OFPT_FLOW_REMOVED:
229 case OFPUTIL_MSG_INVALID:
230 case OFPUTIL_OFPT_HELLO:
231 case OFPUTIL_OFPT_ERROR:
232 case OFPUTIL_OFPT_ECHO_REPLY:
233 case OFPUTIL_OFPT_FEATURES_REQUEST:
234 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
235 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
236 case OFPUTIL_OFPT_SET_CONFIG:
237 case OFPUTIL_OFPT_PORT_STATUS:
238 case OFPUTIL_OFPT_PACKET_OUT:
239 case OFPUTIL_OFPT_FLOW_MOD:
240 case OFPUTIL_OFPT_PORT_MOD:
241 case OFPUTIL_OFPT_BARRIER_REQUEST:
242 case OFPUTIL_OFPT_BARRIER_REPLY:
243 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
244 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
245 case OFPUTIL_OFPST_DESC_REQUEST:
246 case OFPUTIL_OFPST_FLOW_REQUEST:
247 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
248 case OFPUTIL_OFPST_TABLE_REQUEST:
249 case OFPUTIL_OFPST_PORT_REQUEST:
250 case OFPUTIL_OFPST_QUEUE_REQUEST:
251 case OFPUTIL_OFPST_DESC_REPLY:
252 case OFPUTIL_OFPST_FLOW_REPLY:
253 case OFPUTIL_OFPST_QUEUE_REPLY:
254 case OFPUTIL_OFPST_PORT_REPLY:
255 case OFPUTIL_OFPST_TABLE_REPLY:
256 case OFPUTIL_OFPST_AGGREGATE_REPLY:
257 case OFPUTIL_NXT_ROLE_REQUEST:
258 case OFPUTIL_NXT_ROLE_REPLY:
259 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
260 case OFPUTIL_NXT_SET_FLOW_FORMAT:
261 case OFPUTIL_NXT_SET_PACKET_IN_FORMAT:
262 case OFPUTIL_NXT_PACKET_IN:
263 case OFPUTIL_NXT_FLOW_MOD:
264 case OFPUTIL_NXT_FLOW_REMOVED:
265 case OFPUTIL_NXST_FLOW_REQUEST:
266 case OFPUTIL_NXST_AGGREGATE_REQUEST:
267 case OFPUTIL_NXST_FLOW_REPLY:
268 case OFPUTIL_NXST_AGGREGATE_REPLY:
270 if (VLOG_IS_DBG_ENABLED()) {
271 char *s = ofp_to_string(msg->data, msg->size, 2);
272 VLOG_DBG_RL(&rl, "%016llx: OpenFlow packet ignored: %s",
280 send_features_request(struct lswitch *sw, struct rconn *rconn)
282 time_t now = time_now();
283 if (now >= sw->last_features_request + 1) {
285 struct ofp_switch_config *osc;
287 /* Send OFPT_FEATURES_REQUEST. */
288 make_openflow(sizeof(struct ofp_header), OFPT_FEATURES_REQUEST, &b);
289 queue_tx(sw, rconn, b);
291 /* Send OFPT_SET_CONFIG. */
292 osc = make_openflow(sizeof *osc, OFPT_SET_CONFIG, &b);
293 osc->miss_send_len = htons(OFP_DEFAULT_MISS_SEND_LEN);
294 queue_tx(sw, rconn, b);
296 sw->last_features_request = now;
301 queue_tx(struct lswitch *sw, struct rconn *rconn, struct ofpbuf *b)
303 int retval = rconn_send_with_limit(rconn, b, sw->queued, 10);
304 if (retval && retval != ENOTCONN) {
305 if (retval == EAGAIN) {
306 VLOG_INFO_RL(&rl, "%016llx: %s: tx queue overflow",
307 sw->datapath_id, rconn_get_name(rconn));
309 VLOG_WARN_RL(&rl, "%016llx: %s: send: %s",
310 sw->datapath_id, rconn_get_name(rconn),
317 process_switch_features(struct lswitch *sw, struct ofp_switch_features *osf)
322 sw->datapath_id = ntohll(osf->datapath_id);
324 n_ports = (ntohs(osf->header.length) - sizeof *osf) / sizeof *osf->ports;
325 for (i = 0; i < n_ports; i++) {
326 struct ofp_phy_port *opp = &osf->ports[i];
327 struct lswitch_port *lp;
329 opp->name[OFP_MAX_PORT_NAME_LEN - 1] = '\0';
330 lp = shash_find_data(&sw->queue_names, opp->name);
331 if (lp && hmap_node_is_null(&lp->hmap_node)) {
332 lp->port_no = ntohs(opp->port_no);
333 hmap_insert(&sw->queue_numbers, &lp->hmap_node,
334 hash_int(lp->port_no, 0));
340 lswitch_choose_destination(struct lswitch *sw, const struct flow *flow)
344 /* Learn the source MAC. */
345 if (mac_learning_may_learn(sw->ml, flow->dl_src, 0)) {
346 struct mac_entry *mac = mac_learning_insert(sw->ml, flow->dl_src, 0);
347 if (mac_entry_is_new(mac) || mac->port.i != flow->in_port) {
348 VLOG_DBG_RL(&rl, "%016llx: learned that "ETH_ADDR_FMT" is on "
349 "port %"PRIu16, sw->datapath_id,
350 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
352 mac->port.i = flow->in_port;
353 mac_learning_changed(sw->ml, mac);
357 /* Drop frames for reserved multicast addresses. */
358 if (eth_addr_is_reserved(flow->dl_dst)) {
362 out_port = OFPP_FLOOD;
364 struct mac_entry *mac;
366 mac = mac_learning_lookup(sw->ml, flow->dl_dst, 0, NULL);
368 out_port = mac->port.i;
369 if (out_port == flow->in_port) {
370 /* Don't send a packet back out its input port. */
376 /* Check if we need to use "NORMAL" action. */
377 if (sw->action_normal && out_port != OFPP_FLOOD) {
385 get_queue_id(const struct lswitch *sw, uint16_t in_port)
387 const struct lswitch_port *port;
389 HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_int(in_port, 0),
390 &sw->queue_numbers) {
391 if (port->port_no == in_port) {
392 return port->queue_id;
396 return sw->default_queue;
400 process_packet_in(struct lswitch *sw, struct rconn *rconn,
401 const struct ofp_packet_in *opi)
403 uint16_t in_port = ntohs(opi->in_port);
407 struct ofp_action_header actions[2];
410 size_t pkt_ofs, pkt_len;
414 /* Ignore packets sent via output to OFPP_CONTROLLER. This library never
415 * uses such an action. You never know what experiments might be going on,
416 * though, and it seems best not to interfere with them. */
417 if (opi->reason != OFPR_NO_MATCH) {
421 /* Extract flow data from 'opi' into 'flow'. */
422 pkt_ofs = offsetof(struct ofp_packet_in, data);
423 pkt_len = ntohs(opi->header.length) - pkt_ofs;
424 ofpbuf_use_const(&pkt, opi->data, pkt_len);
425 flow_extract(&pkt, 0, 0, in_port, &flow);
427 /* Choose output port. */
428 out_port = lswitch_choose_destination(sw, &flow);
431 queue_id = get_queue_id(sw, in_port);
432 if (out_port == OFPP_NONE) {
434 } else if (queue_id == UINT32_MAX || out_port >= OFPP_MAX) {
435 struct ofp_action_output oao;
437 memset(&oao, 0, sizeof oao);
438 oao.type = htons(OFPAT_OUTPUT);
439 oao.len = htons(sizeof oao);
440 oao.port = htons(out_port);
442 memcpy(actions, &oao, sizeof oao);
443 actions_len = sizeof oao;
445 struct ofp_action_enqueue oae;
447 memset(&oae, 0, sizeof oae);
448 oae.type = htons(OFPAT_ENQUEUE);
449 oae.len = htons(sizeof oae);
450 oae.port = htons(out_port);
451 oae.queue_id = htonl(queue_id);
453 memcpy(actions, &oae, sizeof oae);
454 actions_len = sizeof oae;
456 assert(actions_len <= sizeof actions);
458 /* Send the packet, and possibly the whole flow, to the output port. */
459 if (sw->max_idle >= 0 && (!sw->ml || out_port != OFPP_FLOOD)) {
460 struct ofpbuf *buffer;
461 struct cls_rule rule;
463 /* The output port is known, or we always flood everything, so add a
465 cls_rule_init(&flow, &sw->wc, 0, &rule);
466 buffer = make_add_flow(&rule, ntohl(opi->buffer_id),
467 sw->max_idle, actions_len);
468 ofpbuf_put(buffer, actions, actions_len);
469 queue_tx(sw, rconn, buffer);
471 /* If the switch didn't buffer the packet, we need to send a copy. */
472 if (ntohl(opi->buffer_id) == UINT32_MAX && actions_len > 0) {
474 make_packet_out(&pkt, UINT32_MAX, in_port,
475 actions, actions_len / sizeof *actions));
478 /* We don't know that MAC, or we don't set up flows. Send along the
479 * packet without setting up a flow. */
480 if (ntohl(opi->buffer_id) != UINT32_MAX || actions_len > 0) {
482 make_packet_out(&pkt, ntohl(opi->buffer_id), in_port,
483 actions, actions_len / sizeof *actions));
489 process_echo_request(struct lswitch *sw, struct rconn *rconn,
490 const struct ofp_header *rq)
492 queue_tx(sw, rconn, make_echo_reply(rq));