lib/learning-switch.c

   1 /*
   2  * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
   3  *
   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:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  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.
  15  */
  16
  17 #include <config.h>
  18 #include "learning-switch.h"
  19
  20 #include <errno.h>
  21 #include <inttypes.h>
  22 #include <netinet/in.h>
  23 #include <stdlib.h>
  24 #include <time.h>
  25
  26 #include "byte-order.h"
  27 #include "classifier.h"
  28 #include "flow.h"
  29 #include "hmap.h"
  30 #include "mac-learning.h"
  31 #include "ofpbuf.h"
  32 #include "ofp-actions.h"
  33 #include "ofp-errors.h"
  34 #include "ofp-msgs.h"
  35 #include "ofp-parse.h"
  36 #include "ofp-print.h"
  37 #include "ofp-util.h"
  38 #include "openflow/openflow.h"
  39 #include "poll-loop.h"
  40 #include "rconn.h"
  41 #include "shash.h"
  42 #include "simap.h"
  43 #include "timeval.h"
  44 #include "vconn.h"
  45 #include "vlog.h"
  46
  47 VLOG_DEFINE_THIS_MODULE(learning_switch);
  48
  49 struct lswitch_port {
  50     struct hmap_node hmap_node; /* Hash node for port number. */
  51     ofp_port_t port_no;         /* OpenFlow port number. */
  52     uint32_t queue_id;          /* OpenFlow queue number. */
  53 };
  54
  55 enum lswitch_state {
  56     S_CONNECTING,               /* Waiting for connection to complete. */
  57     S_FEATURES_REPLY,           /* Waiting for features reply. */
  58     S_SWITCHING,                /* Switching flows. */
  59 };
  60
  61 struct lswitch {
  62     struct rconn *rconn;
  63     enum lswitch_state state;
  64
  65     /* If nonnegative, the switch sets up flows that expire after the given
  66      * number of seconds (or never expire, if the value is OFP_FLOW_PERMANENT).
  67      * Otherwise, the switch processes every packet. */
  68     int max_idle;
  69
  70     enum ofputil_protocol protocol;
  71     unsigned long long int datapath_id;
  72     struct mac_learning *ml;    /* NULL to act as hub instead of switch. */
  73     struct flow_wildcards wc;   /* Wildcards to apply to flows. */
  74     bool action_normal;         /* Use OFPP_NORMAL? */
  75
  76     /* Queue distribution. */
  77     uint32_t default_queue;     /* Default OpenFlow queue, or UINT32_MAX. */
  78     struct hmap queue_numbers;  /* Map from port number to lswitch_port. */
  79     struct shash queue_names;   /* Map from port name to lswitch_port. */
  80
  81     /* Number of outgoing queued packets on the rconn. */
  82     struct rconn_packet_counter *queued;
  83
  84     /* If true, do not reply to any messages from the switch (for debugging
  85      * fail-open mode). */
  86     bool mute;
  87
  88     /* Optional "flow mod" requests to send to the switch at connection time,
  89      * to set up the flow table. */
  90     const struct ofputil_flow_mod *default_flows;
  91     size_t n_default_flows;
  92     enum ofputil_protocol usable_protocols;
  93 };
  94
  95 /* The log messages here could actually be useful in debugging, so keep the
  96  * rate limit relatively high. */
  97 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
  98
  99 static void queue_tx(struct lswitch *, struct ofpbuf *);
 100 static void send_features_request(struct lswitch *);
 101
 102 static void lswitch_process_packet(struct lswitch *, const struct ofpbuf *);
 103 static enum ofperr process_switch_features(struct lswitch *,
 104                                            struct ofp_header *);
 105 static void process_packet_in(struct lswitch *, const struct ofp_header *);
 106 static void process_echo_request(struct lswitch *, const struct ofp_header *);
 107
 108 /* Creates and returns a new learning switch whose configuration is given by
 109  * 'cfg'.
 110  *
 111  * 'rconn' is used to send out an OpenFlow features request. */
 112 struct lswitch *
 113 lswitch_create(struct rconn *rconn, const struct lswitch_config *cfg)
 114 {
 115     struct lswitch *sw;
 116     uint32_t ofpfw;
 117
 118     sw = xzalloc(sizeof *sw);
 119     sw->rconn = rconn;
 120     sw->state = S_CONNECTING;
 121     sw->max_idle = cfg->max_idle;
 122     sw->datapath_id = 0;
 123     sw->ml = (cfg->mode == LSW_LEARN
 124               ? mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME)
 125               : NULL);
 126     sw->action_normal = cfg->mode == LSW_NORMAL;
 127
 128     switch (cfg->wildcards) {
 129     case 0:
 130         ofpfw = 0;
 131         break;
 132
 133     case UINT32_MAX:
 134         /* Try to wildcard as many fields as possible, but we cannot
 135          * wildcard all fields.  We need in_port to detect moves.  We need
 136          * Ethernet source and dest and VLAN VID to do L2 learning. */
 137         ofpfw = (OFPFW10_DL_TYPE | OFPFW10_DL_VLAN_PCP
 138                  | OFPFW10_NW_SRC_ALL | OFPFW10_NW_DST_ALL
 139                  | OFPFW10_NW_TOS | OFPFW10_NW_PROTO
 140                  | OFPFW10_TP_SRC | OFPFW10_TP_DST);
 141         break;
 142
 143     default:
 144         ofpfw = cfg->wildcards;
 145         break;
 146     }
 147     ofputil_wildcard_from_ofpfw10(ofpfw, &sw->wc);
 148
 149     sw->default_queue = cfg->default_queue;
 150     hmap_init(&sw->queue_numbers);
 151     shash_init(&sw->queue_names);
 152     if (cfg->port_queues) {
 153         struct simap_node *node;
 154
 155         SIMAP_FOR_EACH (node, cfg->port_queues) {
 156             struct lswitch_port *port = xmalloc(sizeof *port);
 157             hmap_node_nullify(&port->hmap_node);
 158             port->queue_id = node->data;
 159             shash_add(&sw->queue_names, node->name, port);
 160         }
 161     }
 162
 163     sw->default_flows = cfg->default_flows;
 164     sw->n_default_flows = cfg->n_default_flows;
 165     sw->usable_protocols = cfg->usable_protocols;
 166
 167     sw->queued = rconn_packet_counter_create();
 168
 169     return sw;
 170 }
 171
 172 static void
 173 lswitch_handshake(struct lswitch *sw)
 174 {
 175     enum ofputil_protocol protocol;
 176     enum ofp_version version;
 177
 178     send_features_request(sw);
 179
 180     version = rconn_get_version(sw->rconn);
 181     protocol = ofputil_protocol_from_ofp_version(version);
 182     if (version >= OFP13_VERSION) {
 183         /* OpenFlow 1.3 and later by default drop packets that miss in the flow
 184          * table.  Set up a flow to send packets to the controller by
 185          * default. */
 186         struct ofputil_flow_mod fm;
 187         struct ofpact_output output;
 188         struct ofpbuf *msg;
 189         int error;
 190
 191         ofpact_init_OUTPUT(&output);
 192         output.port = OFPP_CONTROLLER;
 193         output.max_len = OFP_DEFAULT_MISS_SEND_LEN;
 194
 195         match_init_catchall(&fm.match);
 196         fm.priority = 0;
 197         fm.cookie = 0;
 198         fm.cookie_mask = 0;
 199         fm.new_cookie = 0;
 200         fm.modify_cookie = false;
 201         fm.table_id = 0;
 202         fm.command = OFPFC_ADD;
 203         fm.idle_timeout = 0;
 204         fm.hard_timeout = 0;
 205         fm.importance = 0;
 206         fm.buffer_id = UINT32_MAX;
 207         fm.out_port = OFPP_NONE;
 208         fm.out_group = OFPG_ANY;
 209         fm.flags = 0;
 210         fm.ofpacts = &output.ofpact;
 211         fm.ofpacts_len = sizeof output;
 212         fm.delete_reason = 0;
 213
 214         msg = ofputil_encode_flow_mod(&fm, protocol);
 215         error = rconn_send(sw->rconn, msg, NULL);
 216         if (error) {
 217             VLOG_INFO_RL(&rl, "%s: failed to add default flow (%s)",
 218                          rconn_get_name(sw->rconn), ovs_strerror(error));
 219         }
 220     }
 221     if (sw->default_flows) {
 222         struct ofpbuf *msg = NULL;
 223         int error = 0;
 224         size_t i;
 225
 226         /* If the initial protocol isn't good enough for default_flows, then
 227          * pick one that will work and encode messages to set up that
 228          * protocol.
 229          *
 230          * This could be improved by actually negotiating a mutually acceptable
 231          * flow format with the switch, but that would require an asynchronous
 232          * state machine.  This version ought to work fine in practice. */
 233         if (!(protocol & sw->usable_protocols)) {
 234             enum ofputil_protocol want = rightmost_1bit(sw->usable_protocols);
 235             while (!error) {
 236                 msg = ofputil_encode_set_protocol(protocol, want, &protocol);
 237                 if (!msg) {
 238                     break;
 239                 }
 240                 error = rconn_send(sw->rconn, msg, NULL);
 241             }
 242         }
 243         if (protocol & sw->usable_protocols) {
 244             for (i = 0; !error && i < sw->n_default_flows; i++) {
 245                 msg = ofputil_encode_flow_mod(&sw->default_flows[i], protocol);
 246                 error = rconn_send(sw->rconn, msg, NULL);
 247             }
 248
 249             if (error) {
 250                 VLOG_INFO_RL(&rl, "%s: failed to queue default flows (%s)",
 251                              rconn_get_name(sw->rconn), ovs_strerror(error));
 252             }
 253         } else {
 254             VLOG_INFO_RL(&rl, "%s: failed to set usable protocol",
 255                          rconn_get_name(sw->rconn));
 256         }
 257     }
 258     sw->protocol = protocol;
 259 }
 260
 261 bool
 262 lswitch_is_alive(const struct lswitch *sw)
 263 {
 264     return rconn_is_alive(sw->rconn);
 265 }
 266
 267 /* Destroys 'sw'. */
 268 void
 269 lswitch_destroy(struct lswitch *sw)
 270 {
 271     if (sw) {
 272         struct lswitch_port *node, *next;
 273
 274         rconn_destroy(sw->rconn);
 275         HMAP_FOR_EACH_SAFE (node, next, hmap_node, &sw->queue_numbers) {
 276             hmap_remove(&sw->queue_numbers, &node->hmap_node);
 277             free(node);
 278         }
 279         shash_destroy(&sw->queue_names);
 280         mac_learning_unref(sw->ml);
 281         rconn_packet_counter_destroy(sw->queued);
 282         free(sw);
 283     }
 284 }
 285
 286 /* Takes care of necessary 'sw' activity, except for receiving packets (which
 287  * the caller must do). */
 288 void
 289 lswitch_run(struct lswitch *sw)
 290 {
 291     int i;
 292
 293     if (sw->ml) {
 294         ovs_rwlock_wrlock(&sw->ml->rwlock);
 295         mac_learning_run(sw->ml);
 296         ovs_rwlock_unlock(&sw->ml->rwlock);
 297     }
 298
 299     rconn_run(sw->rconn);
 300
 301     if (sw->state == S_CONNECTING) {
 302         if (rconn_get_version(sw->rconn) != -1) {
 303             lswitch_handshake(sw);
 304             sw->state = S_FEATURES_REPLY;
 305         }
 306         return;
 307     }
 308
 309     for (i = 0; i < 50; i++) {
 310         struct ofpbuf *msg;
 311
 312         msg = rconn_recv(sw->rconn);
 313         if (!msg) {
 314             break;
 315         }
 316
 317         if (!sw->mute) {
 318             lswitch_process_packet(sw, msg);
 319         }
 320         ofpbuf_delete(msg);
 321     }
 322 }
 323
 324 void
 325 lswitch_wait(struct lswitch *sw)
 326 {
 327     if (sw->ml) {
 328         ovs_rwlock_rdlock(&sw->ml->rwlock);
 329         mac_learning_wait(sw->ml);
 330         ovs_rwlock_unlock(&sw->ml->rwlock);
 331     }
 332     rconn_run_wait(sw->rconn);
 333     rconn_recv_wait(sw->rconn);
 334 }
 335
 336 /* Processes 'msg', which should be an OpenFlow received on 'rconn', according
 337  * to the learning switch state in 'sw'.  The most likely result of processing
 338  * is that flow-setup and packet-out OpenFlow messages will be sent out on
 339  * 'rconn'.  */
 340 static void
 341 lswitch_process_packet(struct lswitch *sw, const struct ofpbuf *msg)
 342 {
 343     enum ofptype type;
 344     struct ofpbuf b;
 345
 346     b = *msg;
 347     if (ofptype_pull(&type, &b)) {
 348         return;
 349     }
 350
 351     if (sw->state == S_FEATURES_REPLY
 352         && type != OFPTYPE_ECHO_REQUEST
 353         && type != OFPTYPE_FEATURES_REPLY) {
 354         return;
 355     }
 356
 357     switch (type) {
 358     case OFPTYPE_ECHO_REQUEST:
 359         process_echo_request(sw, ofpbuf_data(msg));
 360         break;
 361
 362     case OFPTYPE_FEATURES_REPLY:
 363         if (sw->state == S_FEATURES_REPLY) {
 364             if (!process_switch_features(sw, ofpbuf_data(msg))) {
 365                 sw->state = S_SWITCHING;
 366             } else {
 367                 rconn_disconnect(sw->rconn);
 368             }
 369         }
 370         break;
 371
 372     case OFPTYPE_PACKET_IN:
 373         process_packet_in(sw, ofpbuf_data(msg));
 374         break;
 375
 376     case OFPTYPE_FLOW_REMOVED:
 377         /* Nothing to do. */
 378         break;
 379
 380     case OFPTYPE_HELLO:
 381     case OFPTYPE_ERROR:
 382     case OFPTYPE_ECHO_REPLY:
 383     case OFPTYPE_FEATURES_REQUEST:
 384     case OFPTYPE_GET_CONFIG_REQUEST:
 385     case OFPTYPE_GET_CONFIG_REPLY:
 386     case OFPTYPE_SET_CONFIG:
 387     case OFPTYPE_PORT_STATUS:
 388     case OFPTYPE_PACKET_OUT:
 389     case OFPTYPE_FLOW_MOD:
 390     case OFPTYPE_GROUP_MOD:
 391     case OFPTYPE_PORT_MOD:
 392     case OFPTYPE_TABLE_MOD:
 393     case OFPTYPE_BARRIER_REQUEST:
 394     case OFPTYPE_BARRIER_REPLY:
 395     case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
 396     case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
 397     case OFPTYPE_DESC_STATS_REQUEST:
 398     case OFPTYPE_DESC_STATS_REPLY:
 399     case OFPTYPE_FLOW_STATS_REQUEST:
 400     case OFPTYPE_FLOW_STATS_REPLY:
 401     case OFPTYPE_AGGREGATE_STATS_REQUEST:
 402     case OFPTYPE_AGGREGATE_STATS_REPLY:
 403     case OFPTYPE_TABLE_STATS_REQUEST:
 404     case OFPTYPE_TABLE_STATS_REPLY:
 405     case OFPTYPE_PORT_STATS_REQUEST:
 406     case OFPTYPE_PORT_STATS_REPLY:
 407     case OFPTYPE_QUEUE_STATS_REQUEST:
 408     case OFPTYPE_QUEUE_STATS_REPLY:
 409     case OFPTYPE_PORT_DESC_STATS_REQUEST:
 410     case OFPTYPE_PORT_DESC_STATS_REPLY:
 411     case OFPTYPE_ROLE_REQUEST:
 412     case OFPTYPE_ROLE_REPLY:
 413     case OFPTYPE_ROLE_STATUS:
 414     case OFPTYPE_SET_FLOW_FORMAT:
 415     case OFPTYPE_FLOW_MOD_TABLE_ID:
 416     case OFPTYPE_SET_PACKET_IN_FORMAT:
 417     case OFPTYPE_FLOW_AGE:
 418     case OFPTYPE_SET_CONTROLLER_ID:
 419     case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
 420     case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
 421     case OFPTYPE_FLOW_MONITOR_CANCEL:
 422     case OFPTYPE_FLOW_MONITOR_PAUSED:
 423     case OFPTYPE_FLOW_MONITOR_RESUMED:
 424     case OFPTYPE_GET_ASYNC_REQUEST:
 425     case OFPTYPE_GET_ASYNC_REPLY:
 426     case OFPTYPE_SET_ASYNC_CONFIG:
 427     case OFPTYPE_METER_MOD:
 428     case OFPTYPE_GROUP_STATS_REQUEST:
 429     case OFPTYPE_GROUP_STATS_REPLY:
 430     case OFPTYPE_GROUP_DESC_STATS_REQUEST:
 431     case OFPTYPE_GROUP_DESC_STATS_REPLY:
 432     case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
 433     case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
 434     case OFPTYPE_METER_STATS_REQUEST:
 435     case OFPTYPE_METER_STATS_REPLY:
 436     case OFPTYPE_METER_CONFIG_STATS_REQUEST:
 437     case OFPTYPE_METER_CONFIG_STATS_REPLY:
 438     case OFPTYPE_METER_FEATURES_STATS_REQUEST:
 439     case OFPTYPE_METER_FEATURES_STATS_REPLY:
 440     case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
 441     case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
 442     case OFPTYPE_BUNDLE_CONTROL:
 443     case OFPTYPE_BUNDLE_ADD_MESSAGE:
 444     default:
 445         if (VLOG_IS_DBG_ENABLED()) {
 446             char *s = ofp_to_string(ofpbuf_data(msg), ofpbuf_size(msg), 2);
 447             VLOG_DBG_RL(&rl, "%016llx: OpenFlow packet ignored: %s",
 448                         sw->datapath_id, s);
 449             free(s);
 450         }
 451     }
 452 }
 453 \f
 454 static void
 455 send_features_request(struct lswitch *sw)
 456 {
 457     struct ofpbuf *b;
 458     struct ofp_switch_config *osc;
 459     int ofp_version = rconn_get_version(sw->rconn);
 460
 461     ovs_assert(ofp_version > 0 && ofp_version < 0xff);
 462
 463     /* Send OFPT_FEATURES_REQUEST. */
 464     b = ofpraw_alloc(OFPRAW_OFPT_FEATURES_REQUEST, ofp_version, 0);
 465     queue_tx(sw, b);
 466
 467     /* Send OFPT_SET_CONFIG. */
 468     b = ofpraw_alloc(OFPRAW_OFPT_SET_CONFIG, ofp_version, sizeof *osc);
 469     osc = ofpbuf_put_zeros(b, sizeof *osc);
 470     osc->miss_send_len = htons(OFP_DEFAULT_MISS_SEND_LEN);
 471     queue_tx(sw, b);
 472 }
 473
 474 static void
 475 queue_tx(struct lswitch *sw, struct ofpbuf *b)
 476 {
 477     int retval = rconn_send_with_limit(sw->rconn, b, sw->queued, 10);
 478     if (retval && retval != ENOTCONN) {
 479         if (retval == EAGAIN) {
 480             VLOG_INFO_RL(&rl, "%016llx: %s: tx queue overflow",
 481                          sw->datapath_id, rconn_get_name(sw->rconn));
 482         } else {
 483             VLOG_WARN_RL(&rl, "%016llx: %s: send: %s",
 484                          sw->datapath_id, rconn_get_name(sw->rconn),
 485                          ovs_strerror(retval));
 486         }
 487     }
 488 }
 489
 490 static enum ofperr
 491 process_switch_features(struct lswitch *sw, struct ofp_header *oh)
 492 {
 493     struct ofputil_switch_features features;
 494     struct ofputil_phy_port port;
 495     enum ofperr error;
 496     struct ofpbuf b;
 497
 498     error = ofputil_decode_switch_features(oh, &features, &b);
 499     if (error) {
 500         VLOG_ERR("received invalid switch feature reply (%s)",
 501                  ofperr_to_string(error));
 502         return error;
 503     }
 504
 505     sw->datapath_id = features.datapath_id;
 506
 507     while (!ofputil_pull_phy_port(oh->version, &b, &port)) {
 508         struct lswitch_port *lp = shash_find_data(&sw->queue_names, port.name);
 509         if (lp && hmap_node_is_null(&lp->hmap_node)) {
 510             lp->port_no = port.port_no;
 511             hmap_insert(&sw->queue_numbers, &lp->hmap_node,
 512                         hash_ofp_port(lp->port_no));
 513         }
 514     }
 515     return 0;
 516 }
 517
 518 static ofp_port_t
 519 lswitch_choose_destination(struct lswitch *sw, const struct flow *flow)
 520 {
 521     ofp_port_t out_port;
 522
 523     /* Learn the source MAC. */
 524     if (sw->ml) {
 525         ovs_rwlock_wrlock(&sw->ml->rwlock);
 526         if (mac_learning_may_learn(sw->ml, flow->dl_src, 0)) {
 527             struct mac_entry *mac = mac_learning_insert(sw->ml, flow->dl_src,
 528                                                         0);
 529             if (mac->port.ofp_port != flow->in_port.ofp_port) {
 530                 VLOG_DBG_RL(&rl, "%016llx: learned that "ETH_ADDR_FMT" is on "
 531                             "port %"PRIu16, sw->datapath_id,
 532                             ETH_ADDR_ARGS(flow->dl_src),
 533                             flow->in_port.ofp_port);
 534
 535                 mac->port.ofp_port = flow->in_port.ofp_port;
 536                 mac_learning_changed(sw->ml);
 537             }
 538         }
 539         ovs_rwlock_unlock(&sw->ml->rwlock);
 540     }
 541
 542     /* Drop frames for reserved multicast addresses. */
 543     if (eth_addr_is_reserved(flow->dl_dst)) {
 544         return OFPP_NONE;
 545     }
 546
 547     out_port = OFPP_FLOOD;
 548     if (sw->ml) {
 549         struct mac_entry *mac;
 550
 551         ovs_rwlock_rdlock(&sw->ml->rwlock);
 552         mac = mac_learning_lookup(sw->ml, flow->dl_dst, 0);
 553         if (mac) {
 554             out_port = mac->port.ofp_port;
 555             if (out_port == flow->in_port.ofp_port) {
 556                 /* Don't send a packet back out its input port. */
 557                 ovs_rwlock_unlock(&sw->ml->rwlock);
 558                 return OFPP_NONE;
 559             }
 560         }
 561         ovs_rwlock_unlock(&sw->ml->rwlock);
 562     }
 563
 564     /* Check if we need to use "NORMAL" action. */
 565     if (sw->action_normal && out_port != OFPP_FLOOD) {
 566         return OFPP_NORMAL;
 567     }
 568
 569     return out_port;
 570 }
 571
 572 static uint32_t
 573 get_queue_id(const struct lswitch *sw, ofp_port_t in_port)
 574 {
 575     const struct lswitch_port *port;
 576
 577     HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_ofp_port(in_port),
 578                              &sw->queue_numbers) {
 579         if (port->port_no == in_port) {
 580             return port->queue_id;
 581         }
 582     }
 583
 584     return sw->default_queue;
 585 }
 586
 587 static void
 588 process_packet_in(struct lswitch *sw, const struct ofp_header *oh)
 589 {
 590     struct ofputil_packet_in pi;
 591     uint32_t queue_id;
 592     ofp_port_t out_port;
 593
 594     uint64_t ofpacts_stub[64 / 8];
 595     struct ofpbuf ofpacts;
 596
 597     struct ofputil_packet_out po;
 598     enum ofperr error;
 599
 600     struct ofpbuf pkt;
 601     struct flow flow;
 602
 603     error = ofputil_decode_packet_in(&pi, oh);
 604     if (error) {
 605         VLOG_WARN_RL(&rl, "failed to decode packet-in: %s",
 606                      ofperr_to_string(error));
 607         return;
 608     }
 609
 610     /* Ignore packets sent via output to OFPP_CONTROLLER.  This library never
 611      * uses such an action.  You never know what experiments might be going on,
 612      * though, and it seems best not to interfere with them. */
 613     if (pi.reason != OFPR_NO_MATCH) {
 614         return;
 615     }
 616
 617     /* Extract flow data from 'opi' into 'flow'. */
 618     ofpbuf_use_const(&pkt, pi.packet, pi.packet_len);
 619     flow_extract(&pkt, NULL, &flow);
 620     flow.in_port.ofp_port = pi.fmd.in_port;
 621     flow.tunnel.tun_id = pi.fmd.tun_id;
 622
 623     /* Choose output port. */
 624     out_port = lswitch_choose_destination(sw, &flow);
 625
 626     /* Make actions. */
 627     queue_id = get_queue_id(sw, pi.fmd.in_port);
 628     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
 629     if (out_port == OFPP_NONE) {
 630         /* No actions. */
 631     } else if (queue_id == UINT32_MAX
 632                || ofp_to_u16(out_port) >= ofp_to_u16(OFPP_MAX)) {
 633         ofpact_put_OUTPUT(&ofpacts)->port = out_port;
 634     } else {
 635         struct ofpact_enqueue *enqueue = ofpact_put_ENQUEUE(&ofpacts);
 636         enqueue->port = out_port;
 637         enqueue->queue = queue_id;
 638     }
 639     ofpact_pad(&ofpacts);
 640
 641     /* Prepare packet_out in case we need one. */
 642     po.buffer_id = pi.buffer_id;
 643     if (po.buffer_id == UINT32_MAX) {
 644         po.packet = ofpbuf_data(&pkt);
 645         po.packet_len = ofpbuf_size(&pkt);
 646     } else {
 647         po.packet = NULL;
 648         po.packet_len = 0;
 649     }
 650     po.in_port = pi.fmd.in_port;
 651     po.ofpacts = ofpbuf_data(&ofpacts);
 652     po.ofpacts_len = ofpbuf_size(&ofpacts);
 653
 654     /* Send the packet, and possibly the whole flow, to the output port. */
 655     if (sw->max_idle >= 0 && (!sw->ml || out_port != OFPP_FLOOD)) {
 656         struct ofputil_flow_mod fm;
 657         struct ofpbuf *buffer;
 658
 659         /* The output port is known, or we always flood everything, so add a
 660          * new flow. */
 661         memset(&fm, 0, sizeof fm);
 662         match_init(&fm.match, &flow, &sw->wc);
 663         ofputil_normalize_match_quiet(&fm.match);
 664         fm.priority = 1; /* Must be > 0 because of table-miss flow entry. */
 665         fm.table_id = 0xff;
 666         fm.command = OFPFC_ADD;
 667         fm.idle_timeout = sw->max_idle;
 668         fm.buffer_id = pi.buffer_id;
 669         fm.out_port = OFPP_NONE;
 670         fm.ofpacts = ofpbuf_data(&ofpacts);
 671         fm.ofpacts_len = ofpbuf_size(&ofpacts);
 672         buffer = ofputil_encode_flow_mod(&fm, sw->protocol);
 673
 674         queue_tx(sw, buffer);
 675
 676         /* If the switch didn't buffer the packet, we need to send a copy. */
 677         if (pi.buffer_id == UINT32_MAX && out_port != OFPP_NONE) {
 678             queue_tx(sw, ofputil_encode_packet_out(&po, sw->protocol));
 679         }
 680     } else {
 681         /* We don't know that MAC, or we don't set up flows.  Send along the
 682          * packet without setting up a flow. */
 683         if (pi.buffer_id != UINT32_MAX || out_port != OFPP_NONE) {
 684             queue_tx(sw, ofputil_encode_packet_out(&po, sw->protocol));
 685         }
 686     }
 687 }
 688
 689 static void
 690 process_echo_request(struct lswitch *sw, const struct ofp_header *rq)
 691 {
 692     queue_tx(sw, make_echo_reply(rq));
 693 }