2 * Copyright (c) 2008, 2009, 2010, 2011, 2012 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
101 ? mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME)
103 sw->action_normal = cfg->mode == LSW_NORMAL;
105 flow_wildcards_init_exact(&sw->wc);
106 if (cfg->wildcards) {
109 if (cfg->wildcards == UINT32_MAX) {
110 /* Try to wildcard as many fields as possible, but we cannot
111 * wildcard all fields. We need in_port to detect moves. We need
112 * Ethernet source and dest and VLAN VID to do L2 learning. */
113 ofpfw = (OFPFW_DL_TYPE | OFPFW_DL_VLAN_PCP
114 | OFPFW_NW_SRC_ALL | OFPFW_NW_DST_ALL
115 | OFPFW_NW_TOS | OFPFW_NW_PROTO
116 | OFPFW_TP_SRC | OFPFW_TP_DST);
118 ofpfw = cfg->wildcards;
121 ofputil_wildcard_from_openflow(ofpfw, &sw->wc);
124 sw->default_queue = cfg->default_queue;
125 hmap_init(&sw->queue_numbers);
126 shash_init(&sw->queue_names);
127 if (cfg->port_queues) {
128 struct shash_node *node;
130 SHASH_FOR_EACH (node, cfg->port_queues) {
131 struct lswitch_port *port = xmalloc(sizeof *port);
132 hmap_node_nullify(&port->hmap_node);
133 port->queue_id = (uintptr_t) node->data;
134 shash_add(&sw->queue_names, node->name, port);
138 sw->queued = rconn_packet_counter_create();
139 send_features_request(sw, rconn);
141 if (cfg->default_flows) {
142 enum ofputil_protocol usable_protocols;
143 enum ofputil_protocol protocol;
144 struct ofpbuf *msg = NULL;
149 /* Figure out the initial protocol on the connection. */
150 ofp_version = rconn_get_version(rconn);
151 protocol = ofputil_protocol_from_ofp_version(ofp_version);
153 /* If the initial protocol isn't good enough for default_flows, then
154 * pick one that will work and encode messages to set up that
157 * This could be improved by actually negotiating a mutually acceptable
158 * flow format with the switch, but that would require an asynchronous
159 * state machine. This version ought to work fine in practice. */
160 usable_protocols = ofputil_flow_mod_usable_protocols(
161 cfg->default_flows, cfg->n_default_flows);
162 if (!(protocol & usable_protocols)) {
163 enum ofputil_protocol want = rightmost_1bit(usable_protocols);
165 msg = ofputil_encode_set_protocol(protocol, want, &protocol);
169 error = rconn_send(rconn, msg, NULL);
173 for (i = 0; !error && i < cfg->n_default_flows; i++) {
174 msg = ofputil_encode_flow_mod(&cfg->default_flows[i], protocol);
175 error = rconn_send(rconn, msg, NULL);
179 VLOG_INFO_RL(&rl, "%s: failed to queue default flows (%s)",
180 rconn_get_name(rconn), strerror(error));
190 lswitch_destroy(struct lswitch *sw)
193 struct lswitch_port *node, *next;
195 HMAP_FOR_EACH_SAFE (node, next, hmap_node, &sw->queue_numbers) {
196 hmap_remove(&sw->queue_numbers, &node->hmap_node);
199 shash_destroy(&sw->queue_names);
200 mac_learning_destroy(sw->ml);
201 rconn_packet_counter_destroy(sw->queued);
206 /* Takes care of necessary 'sw' activity, except for receiving packets (which
207 * the caller must do). */
209 lswitch_run(struct lswitch *sw)
212 mac_learning_run(sw->ml, NULL);
217 lswitch_wait(struct lswitch *sw)
220 mac_learning_wait(sw->ml);
224 /* Processes 'msg', which should be an OpenFlow received on 'rconn', according
225 * to the learning switch state in 'sw'. The most likely result of processing
226 * is that flow-setup and packet-out OpenFlow messages will be sent out on
229 lswitch_process_packet(struct lswitch *sw, struct rconn *rconn,
230 const struct ofpbuf *msg)
232 const struct ofp_header *oh = msg->data;
233 const struct ofputil_msg_type *type;
235 if (sw->datapath_id == 0
236 && oh->type != OFPT_ECHO_REQUEST
237 && oh->type != OFPT_FEATURES_REPLY) {
238 send_features_request(sw, rconn);
242 ofputil_decode_msg_type(oh, &type);
243 switch (ofputil_msg_type_code(type)) {
244 case OFPUTIL_OFPT_ECHO_REQUEST:
245 process_echo_request(sw, rconn, msg->data);
248 case OFPUTIL_OFPT_FEATURES_REPLY:
249 process_switch_features(sw, msg->data);
252 case OFPUTIL_OFPT_PACKET_IN:
253 process_packet_in(sw, rconn, msg->data);
256 case OFPUTIL_OFPT_FLOW_REMOVED:
260 case OFPUTIL_MSG_INVALID:
261 case OFPUTIL_OFPT_HELLO:
262 case OFPUTIL_OFPT_ERROR:
263 case OFPUTIL_OFPT_ECHO_REPLY:
264 case OFPUTIL_OFPT_FEATURES_REQUEST:
265 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
266 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
267 case OFPUTIL_OFPT_SET_CONFIG:
268 case OFPUTIL_OFPT_PORT_STATUS:
269 case OFPUTIL_OFPT_PACKET_OUT:
270 case OFPUTIL_OFPT_FLOW_MOD:
271 case OFPUTIL_OFPT_PORT_MOD:
272 case OFPUTIL_OFPT_BARRIER_REQUEST:
273 case OFPUTIL_OFPT_BARRIER_REPLY:
274 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
275 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
276 case OFPUTIL_OFPST_DESC_REQUEST:
277 case OFPUTIL_OFPST_FLOW_REQUEST:
278 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
279 case OFPUTIL_OFPST_TABLE_REQUEST:
280 case OFPUTIL_OFPST_PORT_REQUEST:
281 case OFPUTIL_OFPST_QUEUE_REQUEST:
282 case OFPUTIL_OFPST_DESC_REPLY:
283 case OFPUTIL_OFPST_FLOW_REPLY:
284 case OFPUTIL_OFPST_QUEUE_REPLY:
285 case OFPUTIL_OFPST_PORT_REPLY:
286 case OFPUTIL_OFPST_TABLE_REPLY:
287 case OFPUTIL_OFPST_AGGREGATE_REPLY:
288 case OFPUTIL_NXT_ROLE_REQUEST:
289 case OFPUTIL_NXT_ROLE_REPLY:
290 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
291 case OFPUTIL_NXT_SET_FLOW_FORMAT:
292 case OFPUTIL_NXT_SET_PACKET_IN_FORMAT:
293 case OFPUTIL_NXT_PACKET_IN:
294 case OFPUTIL_NXT_FLOW_MOD:
295 case OFPUTIL_NXT_FLOW_REMOVED:
296 case OFPUTIL_NXT_FLOW_AGE:
297 case OFPUTIL_NXT_SET_ASYNC_CONFIG:
298 case OFPUTIL_NXT_SET_CONTROLLER_ID:
299 case OFPUTIL_NXST_FLOW_REQUEST:
300 case OFPUTIL_NXST_AGGREGATE_REQUEST:
301 case OFPUTIL_NXST_FLOW_REPLY:
302 case OFPUTIL_NXST_AGGREGATE_REPLY:
304 if (VLOG_IS_DBG_ENABLED()) {
305 char *s = ofp_to_string(msg->data, msg->size, 2);
306 VLOG_DBG_RL(&rl, "%016llx: OpenFlow packet ignored: %s",
314 send_features_request(struct lswitch *sw, struct rconn *rconn)
316 time_t now = time_now();
317 if (now >= sw->last_features_request + 1) {
319 struct ofp_switch_config *osc;
321 /* Send OFPT_FEATURES_REQUEST. */
322 make_openflow(sizeof(struct ofp_header), OFPT_FEATURES_REQUEST, &b);
323 queue_tx(sw, rconn, b);
325 /* Send OFPT_SET_CONFIG. */
326 osc = make_openflow(sizeof *osc, OFPT_SET_CONFIG, &b);
327 osc->miss_send_len = htons(OFP_DEFAULT_MISS_SEND_LEN);
328 queue_tx(sw, rconn, b);
330 sw->last_features_request = now;
335 queue_tx(struct lswitch *sw, struct rconn *rconn, struct ofpbuf *b)
337 int retval = rconn_send_with_limit(rconn, b, sw->queued, 10);
338 if (retval && retval != ENOTCONN) {
339 if (retval == EAGAIN) {
340 VLOG_INFO_RL(&rl, "%016llx: %s: tx queue overflow",
341 sw->datapath_id, rconn_get_name(rconn));
343 VLOG_WARN_RL(&rl, "%016llx: %s: send: %s",
344 sw->datapath_id, rconn_get_name(rconn),
351 process_switch_features(struct lswitch *sw, struct ofp_switch_features *osf)
356 sw->datapath_id = ntohll(osf->datapath_id);
358 n_ports = (ntohs(osf->header.length) - sizeof *osf) / sizeof *osf->ports;
359 for (i = 0; i < n_ports; i++) {
360 struct ofp_phy_port *opp = &osf->ports[i];
361 struct lswitch_port *lp;
363 opp->name[OFP_MAX_PORT_NAME_LEN - 1] = '\0';
364 lp = shash_find_data(&sw->queue_names, opp->name);
365 if (lp && hmap_node_is_null(&lp->hmap_node)) {
366 lp->port_no = ntohs(opp->port_no);
367 hmap_insert(&sw->queue_numbers, &lp->hmap_node,
368 hash_int(lp->port_no, 0));
374 lswitch_choose_destination(struct lswitch *sw, const struct flow *flow)
378 /* Learn the source MAC. */
379 if (mac_learning_may_learn(sw->ml, flow->dl_src, 0)) {
380 struct mac_entry *mac = mac_learning_insert(sw->ml, flow->dl_src, 0);
381 if (mac_entry_is_new(mac) || mac->port.i != flow->in_port) {
382 VLOG_DBG_RL(&rl, "%016llx: learned that "ETH_ADDR_FMT" is on "
383 "port %"PRIu16, sw->datapath_id,
384 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
386 mac->port.i = flow->in_port;
387 mac_learning_changed(sw->ml, mac);
391 /* Drop frames for reserved multicast addresses. */
392 if (eth_addr_is_reserved(flow->dl_dst)) {
396 out_port = OFPP_FLOOD;
398 struct mac_entry *mac;
400 mac = mac_learning_lookup(sw->ml, flow->dl_dst, 0, NULL);
402 out_port = mac->port.i;
403 if (out_port == flow->in_port) {
404 /* Don't send a packet back out its input port. */
410 /* Check if we need to use "NORMAL" action. */
411 if (sw->action_normal && out_port != OFPP_FLOOD) {
419 get_queue_id(const struct lswitch *sw, uint16_t in_port)
421 const struct lswitch_port *port;
423 HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_int(in_port, 0),
424 &sw->queue_numbers) {
425 if (port->port_no == in_port) {
426 return port->queue_id;
430 return sw->default_queue;
434 process_packet_in(struct lswitch *sw, struct rconn *rconn,
435 const struct ofp_packet_in *opi)
437 uint16_t in_port = ntohs(opi->in_port);
441 struct ofp_action_header actions[2];
444 struct ofputil_packet_out po;
446 size_t pkt_ofs, pkt_len;
450 /* Ignore packets sent via output to OFPP_CONTROLLER. This library never
451 * uses such an action. You never know what experiments might be going on,
452 * though, and it seems best not to interfere with them. */
453 if (opi->reason != OFPR_NO_MATCH) {
457 /* Extract flow data from 'opi' into 'flow'. */
458 pkt_ofs = offsetof(struct ofp_packet_in, data);
459 pkt_len = ntohs(opi->header.length) - pkt_ofs;
460 ofpbuf_use_const(&pkt, opi->data, pkt_len);
461 flow_extract(&pkt, 0, 0, in_port, &flow);
463 /* Choose output port. */
464 out_port = lswitch_choose_destination(sw, &flow);
467 queue_id = get_queue_id(sw, in_port);
468 if (out_port == OFPP_NONE) {
470 } else if (queue_id == UINT32_MAX || out_port >= OFPP_MAX) {
471 struct ofp_action_output oao;
473 memset(&oao, 0, sizeof oao);
474 oao.type = htons(OFPAT10_OUTPUT);
475 oao.len = htons(sizeof oao);
476 oao.port = htons(out_port);
478 memcpy(actions, &oao, sizeof oao);
479 actions_len = sizeof oao;
481 struct ofp_action_enqueue oae;
483 memset(&oae, 0, sizeof oae);
484 oae.type = htons(OFPAT10_ENQUEUE);
485 oae.len = htons(sizeof oae);
486 oae.port = htons(out_port);
487 oae.queue_id = htonl(queue_id);
489 memcpy(actions, &oae, sizeof oae);
490 actions_len = sizeof oae;
492 assert(actions_len <= sizeof actions);
494 /* Prepare packet_out in case we need one. */
495 po.buffer_id = ntohl(opi->buffer_id);
496 if (po.buffer_id == UINT32_MAX) {
497 po.packet = pkt.data;
498 po.packet_len = pkt.size;
503 po.in_port = in_port;
504 po.actions = (union ofp_action *) actions;
505 po.n_actions = actions_len / sizeof *actions;
507 /* Send the packet, and possibly the whole flow, to the output port. */
508 if (sw->max_idle >= 0 && (!sw->ml || out_port != OFPP_FLOOD)) {
509 struct ofpbuf *buffer;
510 struct cls_rule rule;
512 /* The output port is known, or we always flood everything, so add a
514 cls_rule_init(&flow, &sw->wc, 0, &rule);
515 buffer = make_add_flow(&rule, ntohl(opi->buffer_id),
516 sw->max_idle, actions_len);
517 ofpbuf_put(buffer, actions, actions_len);
518 queue_tx(sw, rconn, buffer);
520 /* If the switch didn't buffer the packet, we need to send a copy. */
521 if (ntohl(opi->buffer_id) == UINT32_MAX && actions_len > 0) {
522 queue_tx(sw, rconn, ofputil_encode_packet_out(&po));
525 /* We don't know that MAC, or we don't set up flows. Send along the
526 * packet without setting up a flow. */
527 if (ntohl(opi->buffer_id) != UINT32_MAX || actions_len > 0) {
528 queue_tx(sw, rconn, ofputil_encode_packet_out(&po));
534 process_echo_request(struct lswitch *sw, struct rconn *rconn,
535 const struct ofp_header *rq)
537 queue_tx(sw, rconn, make_echo_reply(rq));