lib/learning-switch.c

   1 /*
   2  * Copyright (c) 2008, 2009, 2010, 2011, 2012 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-parse.h"
  35 #include "ofp-print.h"
  36 #include "ofp-util.h"
  37 #include "openflow/openflow.h"
  38 #include "poll-loop.h"
  39 #include "rconn.h"
  40 #include "shash.h"
  41 #include "simap.h"
  42 #include "timeval.h"
  43 #include "vconn.h"
  44 #include "vlog.h"
  45
  46 VLOG_DEFINE_THIS_MODULE(learning_switch);
  47
  48 struct lswitch_port {
  49     struct hmap_node hmap_node; /* Hash node for port number. */
  50     uint16_t port_no;           /* OpenFlow port number, in host byte order. */
  51     uint32_t queue_id;          /* OpenFlow queue number. */
  52 };
  53
  54 struct lswitch {
  55     /* If nonnegative, the switch sets up flows that expire after the given
  56      * number of seconds (or never expire, if the value is OFP_FLOW_PERMANENT).
  57      * Otherwise, the switch processes every packet. */
  58     int max_idle;
  59
  60     enum ofputil_protocol protocol;
  61     unsigned long long int datapath_id;
  62     time_t last_features_request;
  63     struct mac_learning *ml;    /* NULL to act as hub instead of switch. */
  64     struct flow_wildcards wc;   /* Wildcards to apply to flows. */
  65     bool action_normal;         /* Use OFPP_NORMAL? */
  66
  67     /* Queue distribution. */
  68     uint32_t default_queue;     /* Default OpenFlow queue, or UINT32_MAX. */
  69     struct hmap queue_numbers;  /* Map from port number to lswitch_port. */
  70     struct shash queue_names;   /* Map from port name to lswitch_port. */
  71
  72     /* Number of outgoing queued packets on the rconn. */
  73     struct rconn_packet_counter *queued;
  74 };
  75
  76 /* The log messages here could actually be useful in debugging, so keep the
  77  * rate limit relatively high. */
  78 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
  79
  80 static void queue_tx(struct lswitch *, struct rconn *, struct ofpbuf *);
  81 static void send_features_request(struct lswitch *, struct rconn *);
  82
  83 static enum ofperr process_switch_features(struct lswitch *,
  84                                            struct ofp_switch_features *);
  85 static void process_packet_in(struct lswitch *, struct rconn *,
  86                               const struct ofp_header *);
  87 static void process_echo_request(struct lswitch *, struct rconn *,
  88                                  const struct ofp_header *);
  89
  90 /* Creates and returns a new learning switch whose configuration is given by
  91  * 'cfg'.
  92  *
  93  * 'rconn' is used to send out an OpenFlow features request. */
  94 struct lswitch *
  95 lswitch_create(struct rconn *rconn, const struct lswitch_config *cfg)
  96 {
  97     enum ofputil_protocol protocol;
  98     struct lswitch *sw;
  99
 100     sw = xzalloc(sizeof *sw);
 101     sw->max_idle = cfg->max_idle;
 102     sw->datapath_id = 0;
 103     sw->last_features_request = time_now() - 1;
 104     sw->ml = (cfg->mode == LSW_LEARN
 105               ? mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME)
 106               : NULL);
 107     sw->action_normal = cfg->mode == LSW_NORMAL;
 108
 109     flow_wildcards_init_exact(&sw->wc);
 110     if (cfg->wildcards) {
 111         uint32_t ofpfw;
 112
 113         if (cfg->wildcards == UINT32_MAX) {
 114             /* Try to wildcard as many fields as possible, but we cannot
 115              * wildcard all fields.  We need in_port to detect moves.  We need
 116              * Ethernet source and dest and VLAN VID to do L2 learning. */
 117             ofpfw = (OFPFW10_DL_TYPE | OFPFW10_DL_VLAN_PCP
 118                      | OFPFW10_NW_SRC_ALL | OFPFW10_NW_DST_ALL
 119                      | OFPFW10_NW_TOS | OFPFW10_NW_PROTO
 120                      | OFPFW10_TP_SRC | OFPFW10_TP_DST);
 121         } else {
 122             ofpfw = cfg->wildcards;
 123         }
 124
 125         ofputil_wildcard_from_ofpfw10(ofpfw, &sw->wc);
 126     }
 127
 128     sw->default_queue = cfg->default_queue;
 129     hmap_init(&sw->queue_numbers);
 130     shash_init(&sw->queue_names);
 131     if (cfg->port_queues) {
 132         struct simap_node *node;
 133
 134         SIMAP_FOR_EACH (node, cfg->port_queues) {
 135             struct lswitch_port *port = xmalloc(sizeof *port);
 136             hmap_node_nullify(&port->hmap_node);
 137             port->queue_id = node->data;
 138             shash_add(&sw->queue_names, node->name, port);
 139         }
 140     }
 141
 142     sw->queued = rconn_packet_counter_create();
 143     send_features_request(sw, rconn);
 144
 145     protocol = ofputil_protocol_from_ofp_version(rconn_get_version(rconn));
 146     if (cfg->default_flows) {
 147         enum ofputil_protocol usable_protocols;
 148         struct ofpbuf *msg = NULL;
 149         int error = 0;
 150         size_t i;
 151
 152         /* If the initial protocol isn't good enough for default_flows, then
 153          * pick one that will work and encode messages to set up that
 154          * protocol.
 155          *
 156          * This could be improved by actually negotiating a mutually acceptable
 157          * flow format with the switch, but that would require an asynchronous
 158          * state machine.  This version ought to work fine in practice. */
 159         usable_protocols = ofputil_flow_mod_usable_protocols(
 160             cfg->default_flows, cfg->n_default_flows);
 161         if (!(protocol & usable_protocols)) {
 162             enum ofputil_protocol want = rightmost_1bit(usable_protocols);
 163             while (!error) {
 164                 msg = ofputil_encode_set_protocol(protocol, want, &protocol);
 165                 if (!msg) {
 166                     break;
 167                 }
 168                 error = rconn_send(rconn, msg, NULL);
 169             }
 170         }
 171
 172         for (i = 0; !error && i < cfg->n_default_flows; i++) {
 173             msg = ofputil_encode_flow_mod(&cfg->default_flows[i], protocol);
 174             error = rconn_send(rconn, msg, NULL);
 175         }
 176
 177         if (error) {
 178             VLOG_INFO_RL(&rl, "%s: failed to queue default flows (%s)",
 179                          rconn_get_name(rconn), strerror(error));
 180         }
 181     }
 182     sw->protocol = protocol;
 183
 184     return sw;
 185 }
 186
 187 /* Destroys 'sw'. */
 188 void
 189 lswitch_destroy(struct lswitch *sw)
 190 {
 191     if (sw) {
 192         struct lswitch_port *node, *next;
 193
 194         HMAP_FOR_EACH_SAFE (node, next, hmap_node, &sw->queue_numbers) {
 195             hmap_remove(&sw->queue_numbers, &node->hmap_node);
 196             free(node);
 197         }
 198         shash_destroy(&sw->queue_names);
 199         mac_learning_destroy(sw->ml);
 200         rconn_packet_counter_destroy(sw->queued);
 201         free(sw);
 202     }
 203 }
 204
 205 /* Takes care of necessary 'sw' activity, except for receiving packets (which
 206  * the caller must do). */
 207 void
 208 lswitch_run(struct lswitch *sw)
 209 {
 210     if (sw->ml) {
 211         mac_learning_run(sw->ml, NULL);
 212     }
 213 }
 214
 215 void
 216 lswitch_wait(struct lswitch *sw)
 217 {
 218     if (sw->ml) {
 219         mac_learning_wait(sw->ml);
 220     }
 221 }
 222
 223 /* Processes 'msg', which should be an OpenFlow received on 'rconn', according
 224  * to the learning switch state in 'sw'.  The most likely result of processing
 225  * is that flow-setup and packet-out OpenFlow messages will be sent out on
 226  * 'rconn'.  */
 227 void
 228 lswitch_process_packet(struct lswitch *sw, struct rconn *rconn,
 229                        const struct ofpbuf *msg)
 230 {
 231     const struct ofp_header *oh = msg->data;
 232     const struct ofputil_msg_type *type;
 233
 234     if (sw->datapath_id == 0
 235         && oh->type != OFPT_ECHO_REQUEST
 236         && oh->type != OFPT_FEATURES_REPLY) {
 237         send_features_request(sw, rconn);
 238         return;
 239     }
 240
 241     ofputil_decode_msg_type(oh, &type);
 242     switch (ofputil_msg_type_code(type)) {
 243     case OFPUTIL_OFPT_ECHO_REQUEST:
 244         process_echo_request(sw, rconn, msg->data);
 245         break;
 246
 247     case OFPUTIL_OFPT_FEATURES_REPLY:
 248         process_switch_features(sw, msg->data);
 249         break;
 250
 251     case OFPUTIL_OFPT_PACKET_IN:
 252     case OFPUTIL_NXT_PACKET_IN:
 253         process_packet_in(sw, rconn, msg->data);
 254         break;
 255
 256     case OFPUTIL_OFPT_FLOW_REMOVED:
 257         /* Nothing to do. */
 258         break;
 259
 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_FLOW_MOD:
 296     case OFPUTIL_NXT_FLOW_REMOVED:
 297     case OFPUTIL_NXT_FLOW_AGE:
 298     case OFPUTIL_NXT_FLOW_MONITOR_CANCEL:
 299     case OFPUTIL_NXT_FLOW_MONITOR_PAUSED:
 300     case OFPUTIL_NXT_FLOW_MONITOR_RESUMED:
 301     case OFPUTIL_NXT_SET_ASYNC_CONFIG:
 302     case OFPUTIL_NXT_SET_CONTROLLER_ID:
 303     case OFPUTIL_NXST_FLOW_REQUEST:
 304     case OFPUTIL_NXST_AGGREGATE_REQUEST:
 305     case OFPUTIL_NXST_FLOW_MONITOR_REQUEST:
 306     case OFPUTIL_NXST_FLOW_REPLY:
 307     case OFPUTIL_NXST_AGGREGATE_REPLY:
 308     case OFPUTIL_NXST_FLOW_MONITOR_REPLY:
 309     default:
 310         if (VLOG_IS_DBG_ENABLED()) {
 311             char *s = ofp_to_string(msg->data, msg->size, 2);
 312             VLOG_DBG_RL(&rl, "%016llx: OpenFlow packet ignored: %s",
 313                         sw->datapath_id, s);
 314             free(s);
 315         }
 316     }
 317 }
 318 \f
 319 static void
 320 send_features_request(struct lswitch *sw, struct rconn *rconn)
 321 {
 322     time_t now = time_now();
 323     if (now >= sw->last_features_request + 1) {
 324         struct ofpbuf *b;
 325         struct ofp_switch_config *osc;
 326
 327         /* Send OFPT_FEATURES_REQUEST. */
 328         make_openflow(sizeof(struct ofp_header), OFPT_FEATURES_REQUEST, &b);
 329         queue_tx(sw, rconn, b);
 330
 331         /* Send OFPT_SET_CONFIG. */
 332         osc = make_openflow(sizeof *osc, OFPT_SET_CONFIG, &b);
 333         osc->miss_send_len = htons(OFP_DEFAULT_MISS_SEND_LEN);
 334         queue_tx(sw, rconn, b);
 335
 336         sw->last_features_request = now;
 337     }
 338 }
 339
 340 static void
 341 queue_tx(struct lswitch *sw, struct rconn *rconn, struct ofpbuf *b)
 342 {
 343     int retval = rconn_send_with_limit(rconn, b, sw->queued, 10);
 344     if (retval && retval != ENOTCONN) {
 345         if (retval == EAGAIN) {
 346             VLOG_INFO_RL(&rl, "%016llx: %s: tx queue overflow",
 347                          sw->datapath_id, rconn_get_name(rconn));
 348         } else {
 349             VLOG_WARN_RL(&rl, "%016llx: %s: send: %s",
 350                          sw->datapath_id, rconn_get_name(rconn),
 351                          strerror(retval));
 352         }
 353     }
 354 }
 355
 356 static enum ofperr
 357 process_switch_features(struct lswitch *sw, struct ofp_switch_features *osf)
 358 {
 359     struct ofputil_switch_features features;
 360     struct ofputil_phy_port port;
 361     enum ofperr error;
 362     struct ofpbuf b;
 363
 364     error = ofputil_decode_switch_features(osf, &features, &b);
 365     if (error) {
 366         VLOG_ERR("received invalid switch feature reply (%s)",
 367                  ofperr_to_string(error));
 368         return error;
 369     }
 370
 371     sw->datapath_id = features.datapath_id;
 372
 373     while (!ofputil_pull_phy_port(osf->header.version, &b, &port)) {
 374         struct lswitch_port *lp = shash_find_data(&sw->queue_names, port.name);
 375         if (lp && hmap_node_is_null(&lp->hmap_node)) {
 376             lp->port_no = port.port_no;
 377             hmap_insert(&sw->queue_numbers, &lp->hmap_node,
 378                         hash_int(lp->port_no, 0));
 379         }
 380     }
 381     return 0;
 382 }
 383
 384 static uint16_t
 385 lswitch_choose_destination(struct lswitch *sw, const struct flow *flow)
 386 {
 387     uint16_t out_port;
 388
 389     /* Learn the source MAC. */
 390     if (mac_learning_may_learn(sw->ml, flow->dl_src, 0)) {
 391         struct mac_entry *mac = mac_learning_insert(sw->ml, flow->dl_src, 0);
 392         if (mac_entry_is_new(mac) || mac->port.i != flow->in_port) {
 393             VLOG_DBG_RL(&rl, "%016llx: learned that "ETH_ADDR_FMT" is on "
 394                         "port %"PRIu16, sw->datapath_id,
 395                         ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
 396
 397             mac->port.i = flow->in_port;
 398             mac_learning_changed(sw->ml, mac);
 399         }
 400     }
 401
 402     /* Drop frames for reserved multicast addresses. */
 403     if (eth_addr_is_reserved(flow->dl_dst)) {
 404         return OFPP_NONE;
 405     }
 406
 407     out_port = OFPP_FLOOD;
 408     if (sw->ml) {
 409         struct mac_entry *mac;
 410
 411         mac = mac_learning_lookup(sw->ml, flow->dl_dst, 0, NULL);
 412         if (mac) {
 413             out_port = mac->port.i;
 414             if (out_port == flow->in_port) {
 415                 /* Don't send a packet back out its input port. */
 416                 return OFPP_NONE;
 417             }
 418         }
 419     }
 420
 421     /* Check if we need to use "NORMAL" action. */
 422     if (sw->action_normal && out_port != OFPP_FLOOD) {
 423         return OFPP_NORMAL;
 424     }
 425
 426     return out_port;
 427 }
 428
 429 static uint32_t
 430 get_queue_id(const struct lswitch *sw, uint16_t in_port)
 431 {
 432     const struct lswitch_port *port;
 433
 434     HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_int(in_port, 0),
 435                              &sw->queue_numbers) {
 436         if (port->port_no == in_port) {
 437             return port->queue_id;
 438         }
 439     }
 440
 441     return sw->default_queue;
 442 }
 443
 444 static void
 445 process_packet_in(struct lswitch *sw, struct rconn *rconn,
 446                   const struct ofp_header *oh)
 447 {
 448     struct ofputil_packet_in pi;
 449     uint32_t queue_id;
 450     uint16_t out_port;
 451
 452     uint64_t ofpacts_stub[64 / 8];
 453     struct ofpbuf ofpacts;
 454
 455     struct ofputil_packet_out po;
 456     enum ofperr error;
 457
 458     struct ofpbuf pkt;
 459     struct flow flow;
 460
 461     error = ofputil_decode_packet_in(&pi, oh);
 462     if (error) {
 463         VLOG_WARN_RL(&rl, "failed to decode packet-in: %s",
 464                      ofperr_to_string(error));
 465         return;
 466     }
 467
 468     /* Ignore packets sent via output to OFPP_CONTROLLER.  This library never
 469      * uses such an action.  You never know what experiments might be going on,
 470      * though, and it seems best not to interfere with them. */
 471     if (pi.reason != OFPR_NO_MATCH) {
 472         return;
 473     }
 474
 475     /* Extract flow data from 'opi' into 'flow'. */
 476     ofpbuf_use_const(&pkt, pi.packet, pi.packet_len);
 477     flow_extract(&pkt, 0, pi.fmd.tun_id, pi.fmd.in_port, &flow);
 478
 479     /* Choose output port. */
 480     out_port = lswitch_choose_destination(sw, &flow);
 481
 482     /* Make actions. */
 483     queue_id = get_queue_id(sw, pi.fmd.in_port);
 484     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
 485     if (out_port == OFPP_NONE) {
 486         /* No actions. */
 487     } else if (queue_id == UINT32_MAX || out_port >= OFPP_MAX) {
 488         ofpact_put_OUTPUT(&ofpacts)->port = out_port;
 489     } else {
 490         struct ofpact_enqueue *enqueue = ofpact_put_ENQUEUE(&ofpacts);
 491         enqueue->port = out_port;
 492         enqueue->queue = queue_id;
 493     }
 494     ofpact_pad(&ofpacts);
 495
 496     /* Prepare packet_out in case we need one. */
 497     po.buffer_id = pi.buffer_id;
 498     if (po.buffer_id == UINT32_MAX) {
 499         po.packet = pkt.data;
 500         po.packet_len = pkt.size;
 501     } else {
 502         po.packet = NULL;
 503         po.packet_len = 0;
 504     }
 505     po.in_port = pi.fmd.in_port;
 506     po.ofpacts = ofpacts.data;
 507     po.ofpacts_len = ofpacts.size;
 508
 509     /* Send the packet, and possibly the whole flow, to the output port. */
 510     if (sw->max_idle >= 0 && (!sw->ml || out_port != OFPP_FLOOD)) {
 511         struct ofputil_flow_mod fm;
 512         struct ofpbuf *buffer;
 513
 514         /* The output port is known, or we always flood everything, so add a
 515          * new flow. */
 516         memset(&fm, 0, sizeof fm);
 517         cls_rule_init(&flow, &sw->wc, 0, &fm.cr);
 518         fm.table_id = 0xff;
 519         fm.command = OFPFC_ADD;
 520         fm.idle_timeout = sw->max_idle;
 521         fm.buffer_id = pi.buffer_id;
 522         fm.out_port = OFPP_NONE;
 523         fm.ofpacts = ofpacts.data;
 524         fm.ofpacts_len = ofpacts.size;
 525         buffer = ofputil_encode_flow_mod(&fm, sw->protocol);
 526
 527         queue_tx(sw, rconn, buffer);
 528
 529         /* If the switch didn't buffer the packet, we need to send a copy. */
 530         if (pi.buffer_id == UINT32_MAX && out_port != OFPP_NONE) {
 531             queue_tx(sw, rconn, ofputil_encode_packet_out(&po));
 532         }
 533     } else {
 534         /* We don't know that MAC, or we don't set up flows.  Send along the
 535          * packet without setting up a flow. */
 536         if (pi.buffer_id != UINT32_MAX || out_port != OFPP_NONE) {
 537             queue_tx(sw, rconn, ofputil_encode_packet_out(&po));
 538         }
 539     }
 540 }
 541
 542 static void
 543 process_echo_request(struct lswitch *sw, struct rconn *rconn,
 544                      const struct ofp_header *rq)
 545 {
 546     queue_tx(sw, rconn, make_echo_reply(rq));
 547 }