lib/learning-switch.c

   1 /*
   2  * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira Networks.
   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-errors.h"
  33 #include "ofp-parse.h"
  34 #include "ofp-print.h"
  35 #include "ofp-util.h"
  36 #include "openflow/openflow.h"
  37 #include "poll-loop.h"
  38 #include "rconn.h"
  39 #include "shash.h"
  40 #include "timeval.h"
  41 #include "vconn.h"
  42 #include "vlog.h"
  43
  44 VLOG_DEFINE_THIS_MODULE(learning_switch);
  45
  46 struct lswitch_port {
  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. */
  50 };
  51
  52 struct lswitch {
  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. */
  56     int max_idle;
  57
  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? */
  63
  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. */
  68
  69     /* Number of outgoing queued packets on the rconn. */
  70     struct rconn_packet_counter *queued;
  71 };
  72
  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);
  76
  77 static void queue_tx(struct lswitch *, struct rconn *, struct ofpbuf *);
  78 static void send_features_request(struct lswitch *, struct rconn *);
  79
  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 *);
  86
  87 /* Creates and returns a new learning switch whose configuration is given by
  88  * 'cfg'.
  89  *
  90  * 'rconn' is used to send out an OpenFlow features request. */
  91 struct lswitch *
  92 lswitch_create(struct rconn *rconn, const struct lswitch_config *cfg)
  93 {
  94     struct lswitch *sw;
  95
  96     sw = xzalloc(sizeof *sw);
  97     sw->max_idle = cfg->max_idle;
  98     sw->datapath_id = 0;
  99     sw->last_features_request = time_now() - 1;
 100     sw->ml = (cfg->mode == LSW_LEARN
 101               ? mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME)
 102               : NULL);
 103     sw->action_normal = cfg->mode == LSW_NORMAL;
 104
 105     flow_wildcards_init_exact(&sw->wc);
 106     if (cfg->wildcards) {
 107         uint32_t ofpfw;
 108
 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);
 117         } else {
 118             ofpfw = cfg->wildcards;
 119         }
 120
 121         ofputil_wildcard_from_openflow(ofpfw, &sw->wc);
 122     }
 123
 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;
 129
 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);
 135         }
 136     }
 137
 138     sw->queued = rconn_packet_counter_create();
 139     send_features_request(sw, rconn);
 140
 141     if (cfg->default_flows) {
 142         const struct ofpbuf *b;
 143
 144         LIST_FOR_EACH (b, list_node, cfg->default_flows) {
 145             struct ofpbuf *copy = ofpbuf_clone(b);
 146             int error = rconn_send(rconn, copy, NULL);
 147             if (error) {
 148                 VLOG_INFO_RL(&rl, "%s: failed to queue default flows (%s)",
 149                              rconn_get_name(rconn), strerror(error));
 150                 ofpbuf_delete(copy);
 151                 break;
 152             }
 153         }
 154     }
 155
 156     return sw;
 157 }
 158
 159 /* Destroys 'sw'. */
 160 void
 161 lswitch_destroy(struct lswitch *sw)
 162 {
 163     if (sw) {
 164         struct lswitch_port *node, *next;
 165
 166         HMAP_FOR_EACH_SAFE (node, next, hmap_node, &sw->queue_numbers) {
 167             hmap_remove(&sw->queue_numbers, &node->hmap_node);
 168             free(node);
 169         }
 170         shash_destroy(&sw->queue_names);
 171         mac_learning_destroy(sw->ml);
 172         rconn_packet_counter_destroy(sw->queued);
 173         free(sw);
 174     }
 175 }
 176
 177 /* Takes care of necessary 'sw' activity, except for receiving packets (which
 178  * the caller must do). */
 179 void
 180 lswitch_run(struct lswitch *sw)
 181 {
 182     if (sw->ml) {
 183         mac_learning_run(sw->ml, NULL);
 184     }
 185 }
 186
 187 void
 188 lswitch_wait(struct lswitch *sw)
 189 {
 190     if (sw->ml) {
 191         mac_learning_wait(sw->ml);
 192     }
 193 }
 194
 195 /* Processes 'msg', which should be an OpenFlow received on 'rconn', according
 196  * to the learning switch state in 'sw'.  The most likely result of processing
 197  * is that flow-setup and packet-out OpenFlow messages will be sent out on
 198  * 'rconn'.  */
 199 void
 200 lswitch_process_packet(struct lswitch *sw, struct rconn *rconn,
 201                        const struct ofpbuf *msg)
 202 {
 203     const struct ofp_header *oh = msg->data;
 204     const struct ofputil_msg_type *type;
 205
 206     if (sw->datapath_id == 0
 207         && oh->type != OFPT_ECHO_REQUEST
 208         && oh->type != OFPT_FEATURES_REPLY) {
 209         send_features_request(sw, rconn);
 210         return;
 211     }
 212
 213     ofputil_decode_msg_type(oh, &type);
 214     switch (ofputil_msg_type_code(type)) {
 215     case OFPUTIL_OFPT_ECHO_REQUEST:
 216         process_echo_request(sw, rconn, msg->data);
 217         break;
 218
 219     case OFPUTIL_OFPT_FEATURES_REPLY:
 220         process_switch_features(sw, msg->data);
 221         break;
 222
 223     case OFPUTIL_OFPT_PACKET_IN:
 224         process_packet_in(sw, rconn, msg->data);
 225         break;
 226
 227     case OFPUTIL_OFPT_FLOW_REMOVED:
 228         /* Nothing to do. */
 229         break;
 230
 231     case OFPUTIL_MSG_INVALID:
 232     case OFPUTIL_OFPT_HELLO:
 233     case OFPUTIL_OFPT_ERROR:
 234     case OFPUTIL_OFPT_ECHO_REPLY:
 235     case OFPUTIL_OFPT_FEATURES_REQUEST:
 236     case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
 237     case OFPUTIL_OFPT_GET_CONFIG_REPLY:
 238     case OFPUTIL_OFPT_SET_CONFIG:
 239     case OFPUTIL_OFPT_PORT_STATUS:
 240     case OFPUTIL_OFPT_PACKET_OUT:
 241     case OFPUTIL_OFPT_FLOW_MOD:
 242     case OFPUTIL_OFPT_PORT_MOD:
 243     case OFPUTIL_OFPT_BARRIER_REQUEST:
 244     case OFPUTIL_OFPT_BARRIER_REPLY:
 245     case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
 246     case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
 247     case OFPUTIL_OFPST_DESC_REQUEST:
 248     case OFPUTIL_OFPST_FLOW_REQUEST:
 249     case OFPUTIL_OFPST_AGGREGATE_REQUEST:
 250     case OFPUTIL_OFPST_TABLE_REQUEST:
 251     case OFPUTIL_OFPST_PORT_REQUEST:
 252     case OFPUTIL_OFPST_QUEUE_REQUEST:
 253     case OFPUTIL_OFPST_DESC_REPLY:
 254     case OFPUTIL_OFPST_FLOW_REPLY:
 255     case OFPUTIL_OFPST_QUEUE_REPLY:
 256     case OFPUTIL_OFPST_PORT_REPLY:
 257     case OFPUTIL_OFPST_TABLE_REPLY:
 258     case OFPUTIL_OFPST_AGGREGATE_REPLY:
 259     case OFPUTIL_NXT_ROLE_REQUEST:
 260     case OFPUTIL_NXT_ROLE_REPLY:
 261     case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
 262     case OFPUTIL_NXT_SET_FLOW_FORMAT:
 263     case OFPUTIL_NXT_SET_PACKET_IN_FORMAT:
 264     case OFPUTIL_NXT_PACKET_IN:
 265     case OFPUTIL_NXT_FLOW_MOD:
 266     case OFPUTIL_NXT_FLOW_REMOVED:
 267     case OFPUTIL_NXT_FLOW_AGE:
 268     case OFPUTIL_NXT_SET_ASYNC_CONFIG:
 269     case OFPUTIL_NXT_SET_CONTROLLER_ID:
 270     case OFPUTIL_NXST_FLOW_REQUEST:
 271     case OFPUTIL_NXST_AGGREGATE_REQUEST:
 272     case OFPUTIL_NXST_FLOW_REPLY:
 273     case OFPUTIL_NXST_AGGREGATE_REPLY:
 274     default:
 275         if (VLOG_IS_DBG_ENABLED()) {
 276             char *s = ofp_to_string(msg->data, msg->size, 2);
 277             VLOG_DBG_RL(&rl, "%016llx: OpenFlow packet ignored: %s",
 278                         sw->datapath_id, s);
 279             free(s);
 280         }
 281     }
 282 }
 283 \f
 284 static void
 285 send_features_request(struct lswitch *sw, struct rconn *rconn)
 286 {
 287     time_t now = time_now();
 288     if (now >= sw->last_features_request + 1) {
 289         struct ofpbuf *b;
 290         struct ofp_switch_config *osc;
 291
 292         /* Send OFPT_FEATURES_REQUEST. */
 293         make_openflow(sizeof(struct ofp_header), OFPT_FEATURES_REQUEST, &b);
 294         queue_tx(sw, rconn, b);
 295
 296         /* Send OFPT_SET_CONFIG. */
 297         osc = make_openflow(sizeof *osc, OFPT_SET_CONFIG, &b);
 298         osc->miss_send_len = htons(OFP_DEFAULT_MISS_SEND_LEN);
 299         queue_tx(sw, rconn, b);
 300
 301         sw->last_features_request = now;
 302     }
 303 }
 304
 305 static void
 306 queue_tx(struct lswitch *sw, struct rconn *rconn, struct ofpbuf *b)
 307 {
 308     int retval = rconn_send_with_limit(rconn, b, sw->queued, 10);
 309     if (retval && retval != ENOTCONN) {
 310         if (retval == EAGAIN) {
 311             VLOG_INFO_RL(&rl, "%016llx: %s: tx queue overflow",
 312                          sw->datapath_id, rconn_get_name(rconn));
 313         } else {
 314             VLOG_WARN_RL(&rl, "%016llx: %s: send: %s",
 315                          sw->datapath_id, rconn_get_name(rconn),
 316                          strerror(retval));
 317         }
 318     }
 319 }
 320
 321 static void
 322 process_switch_features(struct lswitch *sw, struct ofp_switch_features *osf)
 323 {
 324     size_t n_ports;
 325     size_t i;
 326
 327     sw->datapath_id = ntohll(osf->datapath_id);
 328
 329     n_ports = (ntohs(osf->header.length) - sizeof *osf) / sizeof *osf->ports;
 330     for (i = 0; i < n_ports; i++) {
 331         struct ofp_phy_port *opp = &osf->ports[i];
 332         struct lswitch_port *lp;
 333
 334         opp->name[OFP_MAX_PORT_NAME_LEN - 1] = '\0';
 335         lp = shash_find_data(&sw->queue_names, opp->name);
 336         if (lp && hmap_node_is_null(&lp->hmap_node)) {
 337             lp->port_no = ntohs(opp->port_no);
 338             hmap_insert(&sw->queue_numbers, &lp->hmap_node,
 339                         hash_int(lp->port_no, 0));
 340         }
 341     }
 342 }
 343
 344 static uint16_t
 345 lswitch_choose_destination(struct lswitch *sw, const struct flow *flow)
 346 {
 347     uint16_t out_port;
 348
 349     /* Learn the source MAC. */
 350     if (mac_learning_may_learn(sw->ml, flow->dl_src, 0)) {
 351         struct mac_entry *mac = mac_learning_insert(sw->ml, flow->dl_src, 0);
 352         if (mac_entry_is_new(mac) || mac->port.i != flow->in_port) {
 353             VLOG_DBG_RL(&rl, "%016llx: learned that "ETH_ADDR_FMT" is on "
 354                         "port %"PRIu16, sw->datapath_id,
 355                         ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
 356
 357             mac->port.i = flow->in_port;
 358             mac_learning_changed(sw->ml, mac);
 359         }
 360     }
 361
 362     /* Drop frames for reserved multicast addresses. */
 363     if (eth_addr_is_reserved(flow->dl_dst)) {
 364         return OFPP_NONE;
 365     }
 366
 367     out_port = OFPP_FLOOD;
 368     if (sw->ml) {
 369         struct mac_entry *mac;
 370
 371         mac = mac_learning_lookup(sw->ml, flow->dl_dst, 0, NULL);
 372         if (mac) {
 373             out_port = mac->port.i;
 374             if (out_port == flow->in_port) {
 375                 /* Don't send a packet back out its input port. */
 376                 return OFPP_NONE;
 377             }
 378         }
 379     }
 380
 381     /* Check if we need to use "NORMAL" action. */
 382     if (sw->action_normal && out_port != OFPP_FLOOD) {
 383         return OFPP_NORMAL;
 384     }
 385
 386     return out_port;
 387 }
 388
 389 static uint32_t
 390 get_queue_id(const struct lswitch *sw, uint16_t in_port)
 391 {
 392     const struct lswitch_port *port;
 393
 394     HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_int(in_port, 0),
 395                              &sw->queue_numbers) {
 396         if (port->port_no == in_port) {
 397             return port->queue_id;
 398         }
 399     }
 400
 401     return sw->default_queue;
 402 }
 403
 404 static void
 405 process_packet_in(struct lswitch *sw, struct rconn *rconn,
 406                   const struct ofp_packet_in *opi)
 407 {
 408     uint16_t in_port = ntohs(opi->in_port);
 409     uint32_t queue_id;
 410     uint16_t out_port;
 411
 412     struct ofp_action_header actions[2];
 413     size_t actions_len;
 414
 415     struct ofputil_packet_out po;
 416
 417     size_t pkt_ofs, pkt_len;
 418     struct ofpbuf pkt;
 419     struct flow flow;
 420
 421     /* Ignore packets sent via output to OFPP_CONTROLLER.  This library never
 422      * uses such an action.  You never know what experiments might be going on,
 423      * though, and it seems best not to interfere with them. */
 424     if (opi->reason != OFPR_NO_MATCH) {
 425         return;
 426     }
 427
 428     /* Extract flow data from 'opi' into 'flow'. */
 429     pkt_ofs = offsetof(struct ofp_packet_in, data);
 430     pkt_len = ntohs(opi->header.length) - pkt_ofs;
 431     ofpbuf_use_const(&pkt, opi->data, pkt_len);
 432     flow_extract(&pkt, 0, 0, in_port, &flow);
 433
 434     /* Choose output port. */
 435     out_port = lswitch_choose_destination(sw, &flow);
 436
 437     /* Make actions. */
 438     queue_id = get_queue_id(sw, in_port);
 439     if (out_port == OFPP_NONE) {
 440         actions_len = 0;
 441     } else if (queue_id == UINT32_MAX || out_port >= OFPP_MAX) {
 442         struct ofp_action_output oao;
 443
 444         memset(&oao, 0, sizeof oao);
 445         oao.type = htons(OFPAT_OUTPUT);
 446         oao.len = htons(sizeof oao);
 447         oao.port = htons(out_port);
 448
 449         memcpy(actions, &oao, sizeof oao);
 450         actions_len = sizeof oao;
 451     } else {
 452         struct ofp_action_enqueue oae;
 453
 454         memset(&oae, 0, sizeof oae);
 455         oae.type = htons(OFPAT_ENQUEUE);
 456         oae.len = htons(sizeof oae);
 457         oae.port = htons(out_port);
 458         oae.queue_id = htonl(queue_id);
 459
 460         memcpy(actions, &oae, sizeof oae);
 461         actions_len = sizeof oae;
 462     }
 463     assert(actions_len <= sizeof actions);
 464
 465     /* Prepare packet_out in case we need one. */
 466     po.buffer_id = ntohl(opi->buffer_id);
 467     if (po.buffer_id == UINT32_MAX) {
 468         po.packet = pkt.data;
 469         po.packet_len = pkt.size;
 470     } else {
 471         po.packet = NULL;
 472         po.packet_len = 0;
 473     }
 474     po.in_port = in_port;
 475     po.actions = (union ofp_action *) actions;
 476     po.n_actions = actions_len / sizeof *actions;
 477
 478     /* Send the packet, and possibly the whole flow, to the output port. */
 479     if (sw->max_idle >= 0 && (!sw->ml || out_port != OFPP_FLOOD)) {
 480         struct ofpbuf *buffer;
 481         struct cls_rule rule;
 482
 483         /* The output port is known, or we always flood everything, so add a
 484          * new flow. */
 485         cls_rule_init(&flow, &sw->wc, 0, &rule);
 486         buffer = make_add_flow(&rule, ntohl(opi->buffer_id),
 487                                sw->max_idle, actions_len);
 488         ofpbuf_put(buffer, actions, actions_len);
 489         queue_tx(sw, rconn, buffer);
 490
 491         /* If the switch didn't buffer the packet, we need to send a copy. */
 492         if (ntohl(opi->buffer_id) == UINT32_MAX && actions_len > 0) {
 493             queue_tx(sw, rconn, ofputil_encode_packet_out(&po));
 494         }
 495     } else {
 496         /* We don't know that MAC, or we don't set up flows.  Send along the
 497          * packet without setting up a flow. */
 498         if (ntohl(opi->buffer_id) != UINT32_MAX || actions_len > 0) {
 499             queue_tx(sw, rconn, ofputil_encode_packet_out(&po));
 500         }
 501     }
 502 }
 503
 504 static void
 505 process_echo_request(struct lswitch *sw, struct rconn *rconn,
 506                      const struct ofp_header *rq)
 507 {
 508     queue_tx(sw, rconn, make_echo_reply(rq));
 509 }