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-actions.h"
33 #include "ofp-errors.h"
35 #include "ofp-parse.h"
36 #include "ofp-print.h"
38 #include "openflow/openflow.h"
39 #include "poll-loop.h"
47 VLOG_DEFINE_THIS_MODULE(learning_switch);
50 struct hmap_node hmap_node; /* Hash node for port number. */
51 uint16_t port_no; /* OpenFlow port number, in host byte order. */
52 uint32_t queue_id; /* OpenFlow queue number. */
56 /* If nonnegative, the switch sets up flows that expire after the given
57 * number of seconds (or never expire, if the value is OFP_FLOW_PERMANENT).
58 * Otherwise, the switch processes every packet. */
61 enum ofputil_protocol protocol;
62 unsigned long long int datapath_id;
63 time_t last_features_request;
64 struct mac_learning *ml; /* NULL to act as hub instead of switch. */
65 struct flow_wildcards wc; /* Wildcards to apply to flows. */
66 bool action_normal; /* Use OFPP_NORMAL? */
68 /* Queue distribution. */
69 uint32_t default_queue; /* Default OpenFlow queue, or UINT32_MAX. */
70 struct hmap queue_numbers; /* Map from port number to lswitch_port. */
71 struct shash queue_names; /* Map from port name to lswitch_port. */
73 /* Number of outgoing queued packets on the rconn. */
74 struct rconn_packet_counter *queued;
77 /* The log messages here could actually be useful in debugging, so keep the
78 * rate limit relatively high. */
79 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
81 static void queue_tx(struct lswitch *, struct rconn *, struct ofpbuf *);
82 static void send_features_request(struct lswitch *, struct rconn *);
84 static enum ofperr process_switch_features(struct lswitch *,
86 static void process_packet_in(struct lswitch *, struct rconn *,
87 const struct ofp_header *);
88 static void process_echo_request(struct lswitch *, struct rconn *,
89 const struct ofp_header *);
91 /* Creates and returns a new learning switch whose configuration is given by
94 * 'rconn' is used to send out an OpenFlow features request. */
96 lswitch_create(struct rconn *rconn, const struct lswitch_config *cfg)
98 enum ofputil_protocol protocol;
101 sw = xzalloc(sizeof *sw);
102 sw->max_idle = cfg->max_idle;
104 sw->last_features_request = time_now() - 1;
105 sw->ml = (cfg->mode == LSW_LEARN
106 ? mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME)
108 sw->action_normal = cfg->mode == LSW_NORMAL;
110 flow_wildcards_init_exact(&sw->wc);
111 if (cfg->wildcards) {
114 if (cfg->wildcards == UINT32_MAX) {
115 /* Try to wildcard as many fields as possible, but we cannot
116 * wildcard all fields. We need in_port to detect moves. We need
117 * Ethernet source and dest and VLAN VID to do L2 learning. */
118 ofpfw = (OFPFW10_DL_TYPE | OFPFW10_DL_VLAN_PCP
119 | OFPFW10_NW_SRC_ALL | OFPFW10_NW_DST_ALL
120 | OFPFW10_NW_TOS | OFPFW10_NW_PROTO
121 | OFPFW10_TP_SRC | OFPFW10_TP_DST);
123 ofpfw = cfg->wildcards;
126 ofputil_wildcard_from_ofpfw10(ofpfw, &sw->wc);
129 sw->default_queue = cfg->default_queue;
130 hmap_init(&sw->queue_numbers);
131 shash_init(&sw->queue_names);
132 if (cfg->port_queues) {
133 struct simap_node *node;
135 SIMAP_FOR_EACH (node, cfg->port_queues) {
136 struct lswitch_port *port = xmalloc(sizeof *port);
137 hmap_node_nullify(&port->hmap_node);
138 port->queue_id = node->data;
139 shash_add(&sw->queue_names, node->name, port);
143 sw->queued = rconn_packet_counter_create();
144 send_features_request(sw, rconn);
146 protocol = ofputil_protocol_from_ofp_version(rconn_get_version(rconn));
147 if (cfg->default_flows) {
148 enum ofputil_protocol usable_protocols;
149 struct ofpbuf *msg = NULL;
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));
183 sw->protocol = protocol;
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)
236 if (ofptype_pull(&type, &b)) {
240 if (sw->datapath_id == 0
241 && type != OFPTYPE_ECHO_REQUEST
242 && type != OFPTYPE_FEATURES_REPLY) {
243 send_features_request(sw, rconn);
248 case OFPTYPE_ECHO_REQUEST:
249 process_echo_request(sw, rconn, msg->data);
252 case OFPTYPE_FEATURES_REPLY:
253 process_switch_features(sw, msg->data);
256 case OFPTYPE_PACKET_IN:
257 process_packet_in(sw, rconn, msg->data);
260 case OFPTYPE_FLOW_REMOVED:
266 case OFPTYPE_ECHO_REPLY:
267 case OFPTYPE_FEATURES_REQUEST:
268 case OFPTYPE_GET_CONFIG_REQUEST:
269 case OFPTYPE_GET_CONFIG_REPLY:
270 case OFPTYPE_SET_CONFIG:
271 case OFPTYPE_PORT_STATUS:
272 case OFPTYPE_PACKET_OUT:
273 case OFPTYPE_FLOW_MOD:
274 case OFPTYPE_PORT_MOD:
275 case OFPTYPE_BARRIER_REQUEST:
276 case OFPTYPE_BARRIER_REPLY:
277 case OFPTYPE_DESC_STATS_REQUEST:
278 case OFPTYPE_DESC_STATS_REPLY:
279 case OFPTYPE_FLOW_STATS_REQUEST:
280 case OFPTYPE_FLOW_STATS_REPLY:
281 case OFPTYPE_AGGREGATE_STATS_REQUEST:
282 case OFPTYPE_AGGREGATE_STATS_REPLY:
283 case OFPTYPE_TABLE_STATS_REQUEST:
284 case OFPTYPE_TABLE_STATS_REPLY:
285 case OFPTYPE_PORT_STATS_REQUEST:
286 case OFPTYPE_PORT_STATS_REPLY:
287 case OFPTYPE_QUEUE_STATS_REQUEST:
288 case OFPTYPE_QUEUE_STATS_REPLY:
289 case OFPTYPE_PORT_DESC_STATS_REQUEST:
290 case OFPTYPE_PORT_DESC_STATS_REPLY:
291 case OFPTYPE_ROLE_REQUEST:
292 case OFPTYPE_ROLE_REPLY:
293 case OFPTYPE_SET_FLOW_FORMAT:
294 case OFPTYPE_FLOW_MOD_TABLE_ID:
295 case OFPTYPE_SET_PACKET_IN_FORMAT:
296 case OFPTYPE_FLOW_AGE:
297 case OFPTYPE_SET_ASYNC_CONFIG:
298 case OFPTYPE_SET_CONTROLLER_ID:
299 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
300 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
301 case OFPTYPE_FLOW_MONITOR_CANCEL:
302 case OFPTYPE_FLOW_MONITOR_PAUSED:
303 case OFPTYPE_FLOW_MONITOR_RESUMED:
305 if (VLOG_IS_DBG_ENABLED()) {
306 char *s = ofp_to_string(msg->data, msg->size, 2);
307 VLOG_DBG_RL(&rl, "%016llx: OpenFlow packet ignored: %s",
315 send_features_request(struct lswitch *sw, struct rconn *rconn)
317 time_t now = time_now();
318 if (now >= sw->last_features_request + 1) {
320 struct ofp_switch_config *osc;
322 /* Send OFPT_FEATURES_REQUEST. */
323 b = ofpraw_alloc(OFPRAW_OFPT_FEATURES_REQUEST, OFP10_VERSION, 0);
324 queue_tx(sw, rconn, b);
326 /* Send OFPT_SET_CONFIG. */
327 b = ofpraw_alloc(OFPRAW_OFPT_SET_CONFIG, OFP10_VERSION, sizeof *osc);
328 osc = ofpbuf_put_uninit(b, sizeof *osc);
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_header *oh)
355 struct ofputil_switch_features features;
356 struct ofputil_phy_port port;
360 error = ofputil_decode_switch_features(oh, &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(oh->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_header *oh)
444 struct ofputil_packet_in pi;
448 uint64_t ofpacts_stub[64 / 8];
449 struct ofpbuf ofpacts;
451 struct ofputil_packet_out po;
457 error = ofputil_decode_packet_in(&pi, oh);
459 VLOG_WARN_RL(&rl, "failed to decode packet-in: %s",
460 ofperr_to_string(error));
464 /* Ignore packets sent via output to OFPP_CONTROLLER. This library never
465 * uses such an action. You never know what experiments might be going on,
466 * though, and it seems best not to interfere with them. */
467 if (pi.reason != OFPR_NO_MATCH) {
471 /* Extract flow data from 'opi' into 'flow'. */
472 ofpbuf_use_const(&pkt, pi.packet, pi.packet_len);
473 flow_extract(&pkt, 0, pi.fmd.tun_id, pi.fmd.in_port, &flow);
475 /* Choose output port. */
476 out_port = lswitch_choose_destination(sw, &flow);
479 queue_id = get_queue_id(sw, pi.fmd.in_port);
480 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
481 if (out_port == OFPP_NONE) {
483 } else if (queue_id == UINT32_MAX || out_port >= OFPP_MAX) {
484 ofpact_put_OUTPUT(&ofpacts)->port = out_port;
486 struct ofpact_enqueue *enqueue = ofpact_put_ENQUEUE(&ofpacts);
487 enqueue->port = out_port;
488 enqueue->queue = queue_id;
490 ofpact_pad(&ofpacts);
492 /* Prepare packet_out in case we need one. */
493 po.buffer_id = pi.buffer_id;
494 if (po.buffer_id == UINT32_MAX) {
495 po.packet = pkt.data;
496 po.packet_len = pkt.size;
501 po.in_port = pi.fmd.in_port;
502 po.ofpacts = ofpacts.data;
503 po.ofpacts_len = ofpacts.size;
505 /* Send the packet, and possibly the whole flow, to the output port. */
506 if (sw->max_idle >= 0 && (!sw->ml || out_port != OFPP_FLOOD)) {
507 struct ofputil_flow_mod fm;
508 struct ofpbuf *buffer;
510 /* The output port is known, or we always flood everything, so add a
512 memset(&fm, 0, sizeof fm);
513 cls_rule_init(&flow, &sw->wc, 0, &fm.cr);
515 fm.command = OFPFC_ADD;
516 fm.idle_timeout = sw->max_idle;
517 fm.buffer_id = pi.buffer_id;
518 fm.out_port = OFPP_NONE;
519 fm.ofpacts = ofpacts.data;
520 fm.ofpacts_len = ofpacts.size;
521 buffer = ofputil_encode_flow_mod(&fm, sw->protocol);
523 queue_tx(sw, rconn, buffer);
525 /* If the switch didn't buffer the packet, we need to send a copy. */
526 if (pi.buffer_id == UINT32_MAX && out_port != OFPP_NONE) {
527 queue_tx(sw, rconn, ofputil_encode_packet_out(&po));
530 /* We don't know that MAC, or we don't set up flows. Send along the
531 * packet without setting up a flow. */
532 if (pi.buffer_id != UINT32_MAX || out_port != OFPP_NONE) {
533 queue_tx(sw, rconn, ofputil_encode_packet_out(&po));
539 process_echo_request(struct lswitch *sw, struct rconn *rconn,
540 const struct ofp_header *rq)
542 queue_tx(sw, rconn, make_echo_reply(rq));