2 * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira, Inc.
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"
45 VLOG_DEFINE_THIS_MODULE(learning_switch);
48 struct hmap_node hmap_node; /* Hash node for port number. */
49 uint16_t port_no; /* OpenFlow port number, in host byte order. */
50 uint32_t queue_id; /* OpenFlow queue number. */
54 /* If nonnegative, the switch sets up flows that expire after the given
55 * number of seconds (or never expire, if the value is OFP_FLOW_PERMANENT).
56 * Otherwise, the switch processes every packet. */
59 unsigned long long int datapath_id;
60 time_t last_features_request;
61 struct mac_learning *ml; /* NULL to act as hub instead of switch. */
62 struct flow_wildcards wc; /* Wildcards to apply to flows. */
63 bool action_normal; /* Use OFPP_NORMAL? */
65 /* Queue distribution. */
66 uint32_t default_queue; /* Default OpenFlow queue, or UINT32_MAX. */
67 struct hmap queue_numbers; /* Map from port number to lswitch_port. */
68 struct shash queue_names; /* Map from port name to lswitch_port. */
70 /* Number of outgoing queued packets on the rconn. */
71 struct rconn_packet_counter *queued;
74 /* The log messages here could actually be useful in debugging, so keep the
75 * rate limit relatively high. */
76 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
78 static void queue_tx(struct lswitch *, struct rconn *, struct ofpbuf *);
79 static void send_features_request(struct lswitch *, struct rconn *);
81 static enum ofperr process_switch_features(struct lswitch *,
82 struct ofp_switch_features *);
83 static void process_packet_in(struct lswitch *, struct rconn *,
84 const struct ofp_packet_in *);
85 static void process_echo_request(struct lswitch *, struct rconn *,
86 const struct ofp_header *);
88 /* Creates and returns a new learning switch whose configuration is given by
91 * 'rconn' is used to send out an OpenFlow features request. */
93 lswitch_create(struct rconn *rconn, const struct lswitch_config *cfg)
97 sw = xzalloc(sizeof *sw);
98 sw->max_idle = cfg->max_idle;
100 sw->last_features_request = time_now() - 1;
101 sw->ml = (cfg->mode == LSW_LEARN
102 ? mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME)
104 sw->action_normal = cfg->mode == LSW_NORMAL;
106 flow_wildcards_init_exact(&sw->wc);
107 if (cfg->wildcards) {
110 if (cfg->wildcards == UINT32_MAX) {
111 /* Try to wildcard as many fields as possible, but we cannot
112 * wildcard all fields. We need in_port to detect moves. We need
113 * Ethernet source and dest and VLAN VID to do L2 learning. */
114 ofpfw = (OFPFW10_DL_TYPE | OFPFW10_DL_VLAN_PCP
115 | OFPFW10_NW_SRC_ALL | OFPFW10_NW_DST_ALL
116 | OFPFW10_NW_TOS | OFPFW10_NW_PROTO
117 | OFPFW10_TP_SRC | OFPFW10_TP_DST);
119 ofpfw = cfg->wildcards;
122 ofputil_wildcard_from_ofpfw10(ofpfw, &sw->wc);
125 sw->default_queue = cfg->default_queue;
126 hmap_init(&sw->queue_numbers);
127 shash_init(&sw->queue_names);
128 if (cfg->port_queues) {
129 struct simap_node *node;
131 SIMAP_FOR_EACH (node, cfg->port_queues) {
132 struct lswitch_port *port = xmalloc(sizeof *port);
133 hmap_node_nullify(&port->hmap_node);
134 port->queue_id = node->data;
135 shash_add(&sw->queue_names, node->name, port);
139 sw->queued = rconn_packet_counter_create();
140 send_features_request(sw, rconn);
142 if (cfg->default_flows) {
143 enum ofputil_protocol usable_protocols;
144 enum ofputil_protocol protocol;
145 struct ofpbuf *msg = NULL;
150 /* Figure out the initial protocol on the connection. */
151 ofp_version = rconn_get_version(rconn);
152 protocol = ofputil_protocol_from_ofp_version(ofp_version);
154 /* If the initial protocol isn't good enough for default_flows, then
155 * pick one that will work and encode messages to set up that
158 * This could be improved by actually negotiating a mutually acceptable
159 * flow format with the switch, but that would require an asynchronous
160 * state machine. This version ought to work fine in practice. */
161 usable_protocols = ofputil_flow_mod_usable_protocols(
162 cfg->default_flows, cfg->n_default_flows);
163 if (!(protocol & usable_protocols)) {
164 enum ofputil_protocol want = rightmost_1bit(usable_protocols);
166 msg = ofputil_encode_set_protocol(protocol, want, &protocol);
170 error = rconn_send(rconn, msg, NULL);
174 for (i = 0; !error && i < cfg->n_default_flows; i++) {
175 msg = ofputil_encode_flow_mod(&cfg->default_flows[i], protocol);
176 error = rconn_send(rconn, msg, NULL);
180 VLOG_INFO_RL(&rl, "%s: failed to queue default flows (%s)",
181 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_PORT_DESC_REQUEST:
283 case OFPUTIL_OFPST_DESC_REPLY:
284 case OFPUTIL_OFPST_FLOW_REPLY:
285 case OFPUTIL_OFPST_QUEUE_REPLY:
286 case OFPUTIL_OFPST_PORT_REPLY:
287 case OFPUTIL_OFPST_TABLE_REPLY:
288 case OFPUTIL_OFPST_AGGREGATE_REPLY:
289 case OFPUTIL_OFPST_PORT_DESC_REPLY:
290 case OFPUTIL_NXT_ROLE_REQUEST:
291 case OFPUTIL_NXT_ROLE_REPLY:
292 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
293 case OFPUTIL_NXT_SET_FLOW_FORMAT:
294 case OFPUTIL_NXT_SET_PACKET_IN_FORMAT:
295 case OFPUTIL_NXT_PACKET_IN:
296 case OFPUTIL_NXT_FLOW_MOD:
297 case OFPUTIL_NXT_FLOW_REMOVED:
298 case OFPUTIL_NXT_FLOW_AGE:
299 case OFPUTIL_NXT_SET_ASYNC_CONFIG:
300 case OFPUTIL_NXT_SET_CONTROLLER_ID:
301 case OFPUTIL_NXST_FLOW_REQUEST:
302 case OFPUTIL_NXST_AGGREGATE_REQUEST:
303 case OFPUTIL_NXST_FLOW_REPLY:
304 case OFPUTIL_NXST_AGGREGATE_REPLY:
306 if (VLOG_IS_DBG_ENABLED()) {
307 char *s = ofp_to_string(msg->data, msg->size, 2);
308 VLOG_DBG_RL(&rl, "%016llx: OpenFlow packet ignored: %s",
316 send_features_request(struct lswitch *sw, struct rconn *rconn)
318 time_t now = time_now();
319 if (now >= sw->last_features_request + 1) {
321 struct ofp_switch_config *osc;
323 /* Send OFPT_FEATURES_REQUEST. */
324 make_openflow(sizeof(struct ofp_header), OFPT_FEATURES_REQUEST, &b);
325 queue_tx(sw, rconn, b);
327 /* Send OFPT_SET_CONFIG. */
328 osc = make_openflow(sizeof *osc, OFPT_SET_CONFIG, &b);
329 osc->miss_send_len = htons(OFP_DEFAULT_MISS_SEND_LEN);
330 queue_tx(sw, rconn, b);
332 sw->last_features_request = now;
337 queue_tx(struct lswitch *sw, struct rconn *rconn, struct ofpbuf *b)
339 int retval = rconn_send_with_limit(rconn, b, sw->queued, 10);
340 if (retval && retval != ENOTCONN) {
341 if (retval == EAGAIN) {
342 VLOG_INFO_RL(&rl, "%016llx: %s: tx queue overflow",
343 sw->datapath_id, rconn_get_name(rconn));
345 VLOG_WARN_RL(&rl, "%016llx: %s: send: %s",
346 sw->datapath_id, rconn_get_name(rconn),
353 process_switch_features(struct lswitch *sw, struct ofp_switch_features *osf)
355 struct ofputil_switch_features features;
356 struct ofputil_phy_port port;
360 error = ofputil_decode_switch_features(osf, &features, &b);
362 VLOG_ERR("received invalid switch feature reply (%s)",
363 ofperr_to_string(error));
367 sw->datapath_id = features.datapath_id;
369 while (!ofputil_pull_phy_port(osf->header.version, &b, &port)) {
370 struct lswitch_port *lp = shash_find_data(&sw->queue_names, port.name);
371 if (lp && hmap_node_is_null(&lp->hmap_node)) {
372 lp->port_no = port.port_no;
373 hmap_insert(&sw->queue_numbers, &lp->hmap_node,
374 hash_int(lp->port_no, 0));
381 lswitch_choose_destination(struct lswitch *sw, const struct flow *flow)
385 /* Learn the source MAC. */
386 if (mac_learning_may_learn(sw->ml, flow->dl_src, 0)) {
387 struct mac_entry *mac = mac_learning_insert(sw->ml, flow->dl_src, 0);
388 if (mac_entry_is_new(mac) || mac->port.i != flow->in_port) {
389 VLOG_DBG_RL(&rl, "%016llx: learned that "ETH_ADDR_FMT" is on "
390 "port %"PRIu16, sw->datapath_id,
391 ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
393 mac->port.i = flow->in_port;
394 mac_learning_changed(sw->ml, mac);
398 /* Drop frames for reserved multicast addresses. */
399 if (eth_addr_is_reserved(flow->dl_dst)) {
403 out_port = OFPP_FLOOD;
405 struct mac_entry *mac;
407 mac = mac_learning_lookup(sw->ml, flow->dl_dst, 0, NULL);
409 out_port = mac->port.i;
410 if (out_port == flow->in_port) {
411 /* Don't send a packet back out its input port. */
417 /* Check if we need to use "NORMAL" action. */
418 if (sw->action_normal && out_port != OFPP_FLOOD) {
426 get_queue_id(const struct lswitch *sw, uint16_t in_port)
428 const struct lswitch_port *port;
430 HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_int(in_port, 0),
431 &sw->queue_numbers) {
432 if (port->port_no == in_port) {
433 return port->queue_id;
437 return sw->default_queue;
441 process_packet_in(struct lswitch *sw, struct rconn *rconn,
442 const struct ofp_packet_in *opi)
444 uint16_t in_port = ntohs(opi->in_port);
448 struct ofp_action_header actions[2];
451 struct ofputil_packet_out po;
453 size_t pkt_ofs, pkt_len;
457 /* Ignore packets sent via output to OFPP_CONTROLLER. This library never
458 * uses such an action. You never know what experiments might be going on,
459 * though, and it seems best not to interfere with them. */
460 if (opi->reason != OFPR_NO_MATCH) {
464 /* Extract flow data from 'opi' into 'flow'. */
465 pkt_ofs = offsetof(struct ofp_packet_in, data);
466 pkt_len = ntohs(opi->header.length) - pkt_ofs;
467 ofpbuf_use_const(&pkt, opi->data, pkt_len);
468 flow_extract(&pkt, 0, 0, in_port, &flow);
470 /* Choose output port. */
471 out_port = lswitch_choose_destination(sw, &flow);
474 queue_id = get_queue_id(sw, in_port);
475 if (out_port == OFPP_NONE) {
477 } else if (queue_id == UINT32_MAX || out_port >= OFPP_MAX) {
478 struct ofp_action_output oao;
480 memset(&oao, 0, sizeof oao);
481 oao.type = htons(OFPAT10_OUTPUT);
482 oao.len = htons(sizeof oao);
483 oao.port = htons(out_port);
485 memcpy(actions, &oao, sizeof oao);
486 actions_len = sizeof oao;
488 struct ofp_action_enqueue oae;
490 memset(&oae, 0, sizeof oae);
491 oae.type = htons(OFPAT10_ENQUEUE);
492 oae.len = htons(sizeof oae);
493 oae.port = htons(out_port);
494 oae.queue_id = htonl(queue_id);
496 memcpy(actions, &oae, sizeof oae);
497 actions_len = sizeof oae;
499 assert(actions_len <= sizeof actions);
501 /* Prepare packet_out in case we need one. */
502 po.buffer_id = ntohl(opi->buffer_id);
503 if (po.buffer_id == UINT32_MAX) {
504 po.packet = pkt.data;
505 po.packet_len = pkt.size;
510 po.in_port = in_port;
511 po.actions = (union ofp_action *) actions;
512 po.n_actions = actions_len / sizeof *actions;
514 /* Send the packet, and possibly the whole flow, to the output port. */
515 if (sw->max_idle >= 0 && (!sw->ml || out_port != OFPP_FLOOD)) {
516 struct ofpbuf *buffer;
517 struct cls_rule rule;
519 /* The output port is known, or we always flood everything, so add a
521 cls_rule_init(&flow, &sw->wc, 0, &rule);
522 buffer = make_add_flow(&rule, ntohl(opi->buffer_id),
523 sw->max_idle, actions_len);
524 ofpbuf_put(buffer, actions, actions_len);
525 queue_tx(sw, rconn, buffer);
527 /* If the switch didn't buffer the packet, we need to send a copy. */
528 if (ntohl(opi->buffer_id) == UINT32_MAX && actions_len > 0) {
529 queue_tx(sw, rconn, ofputil_encode_packet_out(&po));
532 /* We don't know that MAC, or we don't set up flows. Send along the
533 * packet without setting up a flow. */
534 if (ntohl(opi->buffer_id) != UINT32_MAX || actions_len > 0) {
535 queue_tx(sw, rconn, ofputil_encode_packet_out(&po));
541 process_echo_request(struct lswitch *sw, struct rconn *rconn,
542 const struct ofp_header *rq)
544 queue_tx(sw, rconn, make_echo_reply(rq));