lib/learning-switch.c

   1 /*
   2  * Copyright (c) 2008, 2009, 2010, 2011 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-parse.h"
  33 #include "ofp-print.h"
  34 #include "ofp-util.h"
  35 #include "openflow/openflow.h"
  36 #include "poll-loop.h"
  37 #include "rconn.h"
  38 #include "shash.h"
  39 #include "timeval.h"
  40 #include "vconn.h"
  41 #include "vlog.h"
  42
  43 VLOG_DEFINE_THIS_MODULE(learning_switch);
  44
  45 struct lswitch_port {
  46     struct hmap_node hmap_node; /* Hash node for port number. */
  47     uint16_t port_no;           /* OpenFlow port number, in host byte order. */
  48     uint32_t queue_id;          /* OpenFlow queue number. */
  49 };
  50
  51 struct lswitch {
  52     /* If nonnegative, the switch sets up flows that expire after the given
  53      * number of seconds (or never expire, if the value is OFP_FLOW_PERMANENT).
  54      * Otherwise, the switch processes every packet. */
  55     int max_idle;
  56
  57     unsigned long long int datapath_id;
  58     time_t last_features_request;
  59     struct mac_learning *ml;    /* NULL to act as hub instead of switch. */
  60     struct flow_wildcards wc;   /* Wildcards to apply to flows. */
  61     bool action_normal;         /* Use OFPP_NORMAL? */
  62
  63     /* Queue distribution. */
  64     uint32_t default_queue;     /* Default OpenFlow queue, or UINT32_MAX. */
  65     struct hmap queue_numbers;  /* Map from port number to lswitch_port. */
  66     struct shash queue_names;   /* Map from port name to lswitch_port. */
  67
  68     /* Number of outgoing queued packets on the rconn. */
  69     struct rconn_packet_counter *queued;
  70 };
  71
  72 /* The log messages here could actually be useful in debugging, so keep the
  73  * rate limit relatively high. */
  74 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
  75
  76 static void queue_tx(struct lswitch *, struct rconn *, struct ofpbuf *);
  77 static void send_features_request(struct lswitch *, struct rconn *);
  78
  79 static void process_switch_features(struct lswitch *,
  80                                     struct ofp_switch_features *);
  81 static void process_packet_in(struct lswitch *, struct rconn *,
  82                               const struct ofp_packet_in *);
  83 static void process_echo_request(struct lswitch *, struct rconn *,
  84                                  const struct ofp_header *);
  85
  86 /* Creates and returns a new learning switch whose configuration is given by
  87  * 'cfg'.
  88  *
  89  * 'rconn' is used to send out an OpenFlow features request. */
  90 struct lswitch *
  91 lswitch_create(struct rconn *rconn, const struct lswitch_config *cfg)
  92 {
  93     struct lswitch *sw;
  94
  95     sw = xzalloc(sizeof *sw);
  96     sw->max_idle = cfg->max_idle;
  97     sw->datapath_id = 0;
  98     sw->last_features_request = time_now() - 1;
  99     sw->ml = cfg->mode == LSW_LEARN ? mac_learning_create() : NULL;
 100     sw->action_normal = cfg->mode == LSW_NORMAL;
 101
 102     flow_wildcards_init_exact(&sw->wc);
 103     if (!cfg->exact_flows) {
 104         /* We cannot wildcard all fields.
 105          * We need in_port to detect moves.
 106          * We need both SA and DA to do learning. */
 107         sw->wc.wildcards = (FWW_DL_TYPE | FWW_NW_PROTO
 108                             | FWW_TP_SRC | FWW_TP_DST);
 109         sw->wc.nw_src_mask = htonl(0);
 110         sw->wc.nw_dst_mask = htonl(0);
 111     }
 112
 113     sw->default_queue = cfg->default_queue;
 114     hmap_init(&sw->queue_numbers);
 115     shash_init(&sw->queue_names);
 116     if (cfg->port_queues) {
 117         struct shash_node *node;
 118
 119         SHASH_FOR_EACH (node, cfg->port_queues) {
 120             struct lswitch_port *port = xmalloc(sizeof *port);
 121             hmap_node_nullify(&port->hmap_node);
 122             port->queue_id = (uintptr_t) node->data;
 123             shash_add(&sw->queue_names, node->name, port);
 124         }
 125     }
 126
 127     sw->queued = rconn_packet_counter_create();
 128     send_features_request(sw, rconn);
 129
 130     if (cfg->default_flows) {
 131         const struct ofpbuf *b;
 132
 133         LIST_FOR_EACH (b, list_node, cfg->default_flows) {
 134             queue_tx(sw, rconn, ofpbuf_clone(b));
 135         }
 136     }
 137
 138     return sw;
 139 }
 140
 141 /* Destroys 'sw'. */
 142 void
 143 lswitch_destroy(struct lswitch *sw)
 144 {
 145     if (sw) {
 146         struct lswitch_port *node, *next;
 147
 148         HMAP_FOR_EACH_SAFE (node, next, hmap_node, &sw->queue_numbers) {
 149             hmap_remove(&sw->queue_numbers, &node->hmap_node);
 150             free(node);
 151         }
 152         shash_destroy(&sw->queue_names);
 153         mac_learning_destroy(sw->ml);
 154         rconn_packet_counter_destroy(sw->queued);
 155         free(sw);
 156     }
 157 }
 158
 159 /* Takes care of necessary 'sw' activity, except for receiving packets (which
 160  * the caller must do). */
 161 void
 162 lswitch_run(struct lswitch *sw)
 163 {
 164     if (sw->ml) {
 165         mac_learning_run(sw->ml, NULL);
 166     }
 167 }
 168
 169 void
 170 lswitch_wait(struct lswitch *sw)
 171 {
 172     if (sw->ml) {
 173         mac_learning_wait(sw->ml);
 174     }
 175 }
 176
 177 /* Processes 'msg', which should be an OpenFlow received on 'rconn', according
 178  * to the learning switch state in 'sw'.  The most likely result of processing
 179  * is that flow-setup and packet-out OpenFlow messages will be sent out on
 180  * 'rconn'.  */
 181 void
 182 lswitch_process_packet(struct lswitch *sw, struct rconn *rconn,
 183                        const struct ofpbuf *msg)
 184 {
 185     const struct ofp_header *oh = msg->data;
 186     const struct ofputil_msg_type *type;
 187
 188     if (sw->datapath_id == 0
 189         && oh->type != OFPT_ECHO_REQUEST
 190         && oh->type != OFPT_FEATURES_REPLY) {
 191         send_features_request(sw, rconn);
 192         return;
 193     }
 194
 195     ofputil_decode_msg_type(oh, &type);
 196     switch (ofputil_msg_type_code(type)) {
 197     case OFPUTIL_OFPT_ECHO_REQUEST:
 198         process_echo_request(sw, rconn, msg->data);
 199         break;
 200
 201     case OFPUTIL_OFPT_FEATURES_REPLY:
 202         process_switch_features(sw, msg->data);
 203         break;
 204
 205     case OFPUTIL_OFPT_PACKET_IN:
 206         process_packet_in(sw, rconn, msg->data);
 207         break;
 208
 209     case OFPUTIL_OFPT_FLOW_REMOVED:
 210         /* Nothing to do. */
 211         break;
 212
 213     case OFPUTIL_INVALID:
 214     case OFPUTIL_OFPT_HELLO:
 215     case OFPUTIL_OFPT_ERROR:
 216     case OFPUTIL_OFPT_ECHO_REPLY:
 217     case OFPUTIL_OFPT_FEATURES_REQUEST:
 218     case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
 219     case OFPUTIL_OFPT_GET_CONFIG_REPLY:
 220     case OFPUTIL_OFPT_SET_CONFIG:
 221     case OFPUTIL_OFPT_PORT_STATUS:
 222     case OFPUTIL_OFPT_PACKET_OUT:
 223     case OFPUTIL_OFPT_FLOW_MOD:
 224     case OFPUTIL_OFPT_PORT_MOD:
 225     case OFPUTIL_OFPT_BARRIER_REQUEST:
 226     case OFPUTIL_OFPT_BARRIER_REPLY:
 227     case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
 228     case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
 229     case OFPUTIL_OFPST_DESC_REQUEST:
 230     case OFPUTIL_OFPST_FLOW_REQUEST:
 231     case OFPUTIL_OFPST_AGGREGATE_REQUEST:
 232     case OFPUTIL_OFPST_TABLE_REQUEST:
 233     case OFPUTIL_OFPST_PORT_REQUEST:
 234     case OFPUTIL_OFPST_QUEUE_REQUEST:
 235     case OFPUTIL_OFPST_DESC_REPLY:
 236     case OFPUTIL_OFPST_FLOW_REPLY:
 237     case OFPUTIL_OFPST_QUEUE_REPLY:
 238     case OFPUTIL_OFPST_PORT_REPLY:
 239     case OFPUTIL_OFPST_TABLE_REPLY:
 240     case OFPUTIL_OFPST_AGGREGATE_REPLY:
 241     case OFPUTIL_NXT_ROLE_REQUEST:
 242     case OFPUTIL_NXT_ROLE_REPLY:
 243     case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
 244     case OFPUTIL_NXT_SET_FLOW_FORMAT:
 245     case OFPUTIL_NXT_FLOW_MOD:
 246     case OFPUTIL_NXT_FLOW_REMOVED:
 247     case OFPUTIL_NXST_FLOW_REQUEST:
 248     case OFPUTIL_NXST_AGGREGATE_REQUEST:
 249     case OFPUTIL_NXST_FLOW_REPLY:
 250     case OFPUTIL_NXST_AGGREGATE_REPLY:
 251     default:
 252         if (VLOG_IS_DBG_ENABLED()) {
 253             char *s = ofp_to_string(msg->data, msg->size, 2);
 254             VLOG_DBG_RL(&rl, "%016llx: OpenFlow packet ignored: %s",
 255                         sw->datapath_id, s);
 256             free(s);
 257         }
 258     }
 259 }
 260 \f
 261 static void
 262 send_features_request(struct lswitch *sw, struct rconn *rconn)
 263 {
 264     time_t now = time_now();
 265     if (now >= sw->last_features_request + 1) {
 266         struct ofpbuf *b;
 267         struct ofp_switch_config *osc;
 268
 269         /* Send OFPT_FEATURES_REQUEST. */
 270         make_openflow(sizeof(struct ofp_header), OFPT_FEATURES_REQUEST, &b);
 271         queue_tx(sw, rconn, b);
 272
 273         /* Send OFPT_SET_CONFIG. */
 274         osc = make_openflow(sizeof *osc, OFPT_SET_CONFIG, &b);
 275         osc->miss_send_len = htons(OFP_DEFAULT_MISS_SEND_LEN);
 276         queue_tx(sw, rconn, b);
 277
 278         sw->last_features_request = now;
 279     }
 280 }
 281
 282 static void
 283 queue_tx(struct lswitch *sw, struct rconn *rconn, struct ofpbuf *b)
 284 {
 285     int retval = rconn_send_with_limit(rconn, b, sw->queued, 10);
 286     if (retval && retval != ENOTCONN) {
 287         if (retval == EAGAIN) {
 288             VLOG_INFO_RL(&rl, "%016llx: %s: tx queue overflow",
 289                          sw->datapath_id, rconn_get_name(rconn));
 290         } else {
 291             VLOG_WARN_RL(&rl, "%016llx: %s: send: %s",
 292                          sw->datapath_id, rconn_get_name(rconn),
 293                          strerror(retval));
 294         }
 295     }
 296 }
 297
 298 static void
 299 process_switch_features(struct lswitch *sw, struct ofp_switch_features *osf)
 300 {
 301     size_t n_ports;
 302     size_t i;
 303
 304     sw->datapath_id = ntohll(osf->datapath_id);
 305
 306     n_ports = (ntohs(osf->header.length) - sizeof *osf) / sizeof *osf->ports;
 307     for (i = 0; i < n_ports; i++) {
 308         struct ofp_phy_port *opp = &osf->ports[i];
 309         struct lswitch_port *lp;
 310
 311         opp->name[OFP_MAX_PORT_NAME_LEN - 1] = '\0';
 312         lp = shash_find_data(&sw->queue_names, opp->name);
 313         if (lp && hmap_node_is_null(&lp->hmap_node)) {
 314             lp->port_no = ntohs(opp->port_no);
 315             hmap_insert(&sw->queue_numbers, &lp->hmap_node,
 316                         hash_int(lp->port_no, 0));
 317         }
 318     }
 319 }
 320
 321 static uint16_t
 322 lswitch_choose_destination(struct lswitch *sw, const struct flow *flow)
 323 {
 324     uint16_t out_port;
 325
 326     /* Learn the source MAC. */
 327     if (mac_learning_may_learn(sw->ml, flow->dl_src, 0)) {
 328         struct mac_entry *mac = mac_learning_insert(sw->ml, flow->dl_src, 0);
 329         if (mac_entry_is_new(mac) || mac->port.i != flow->in_port) {
 330             VLOG_DBG_RL(&rl, "%016llx: learned that "ETH_ADDR_FMT" is on "
 331                         "port %"PRIu16, sw->datapath_id,
 332                         ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
 333
 334             mac->port.i = flow->in_port;
 335             mac_learning_changed(sw->ml, mac);
 336         }
 337     }
 338
 339     /* Drop frames for reserved multicast addresses. */
 340     if (eth_addr_is_reserved(flow->dl_dst)) {
 341         return OFPP_NONE;
 342     }
 343
 344     out_port = OFPP_FLOOD;
 345     if (sw->ml) {
 346         struct mac_entry *mac;
 347
 348         mac = mac_learning_lookup(sw->ml, flow->dl_dst, 0, NULL);
 349         if (mac) {
 350             out_port = mac->port.i;
 351             if (out_port == flow->in_port) {
 352                 /* Don't send a packet back out its input port. */
 353                 return OFPP_NONE;
 354             }
 355         }
 356     }
 357
 358     /* Check if we need to use "NORMAL" action. */
 359     if (sw->action_normal && out_port != OFPP_FLOOD) {
 360         return OFPP_NORMAL;
 361     }
 362
 363     return out_port;
 364 }
 365
 366 static uint32_t
 367 get_queue_id(const struct lswitch *sw, uint16_t in_port)
 368 {
 369     const struct lswitch_port *port;
 370
 371     HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_int(in_port, 0),
 372                              &sw->queue_numbers) {
 373         if (port->port_no == in_port) {
 374             return port->queue_id;
 375         }
 376     }
 377
 378     return sw->default_queue;
 379 }
 380
 381 static void
 382 process_packet_in(struct lswitch *sw, struct rconn *rconn,
 383                   const struct ofp_packet_in *opi)
 384 {
 385     uint16_t in_port = ntohs(opi->in_port);
 386     uint32_t queue_id;
 387     uint16_t out_port;
 388
 389     struct ofp_action_header actions[2];
 390     size_t actions_len;
 391
 392     size_t pkt_ofs, pkt_len;
 393     struct ofpbuf pkt;
 394     struct flow flow;
 395
 396     /* Ignore packets sent via output to OFPP_CONTROLLER.  This library never
 397      * uses such an action.  You never know what experiments might be going on,
 398      * though, and it seems best not to interfere with them. */
 399     if (opi->reason != OFPR_NO_MATCH) {
 400         return;
 401     }
 402
 403     /* Extract flow data from 'opi' into 'flow'. */
 404     pkt_ofs = offsetof(struct ofp_packet_in, data);
 405     pkt_len = ntohs(opi->header.length) - pkt_ofs;
 406     ofpbuf_use_const(&pkt, opi->data, pkt_len);
 407     flow_extract(&pkt, 0, in_port, &flow);
 408
 409     /* Choose output port. */
 410     out_port = lswitch_choose_destination(sw, &flow);
 411
 412     /* Make actions. */
 413     queue_id = get_queue_id(sw, in_port);
 414     if (out_port == OFPP_NONE) {
 415         actions_len = 0;
 416     } else if (queue_id == UINT32_MAX || out_port >= OFPP_MAX) {
 417         struct ofp_action_output oao;
 418
 419         memset(&oao, 0, sizeof oao);
 420         oao.type = htons(OFPAT_OUTPUT);
 421         oao.len = htons(sizeof oao);
 422         oao.port = htons(out_port);
 423
 424         memcpy(actions, &oao, sizeof oao);
 425         actions_len = sizeof oao;
 426     } else {
 427         struct ofp_action_enqueue oae;
 428
 429         memset(&oae, 0, sizeof oae);
 430         oae.type = htons(OFPAT_ENQUEUE);
 431         oae.len = htons(sizeof oae);
 432         oae.port = htons(out_port);
 433         oae.queue_id = htonl(queue_id);
 434
 435         memcpy(actions, &oae, sizeof oae);
 436         actions_len = sizeof oae;
 437     }
 438     assert(actions_len <= sizeof actions);
 439
 440     /* Send the packet, and possibly the whole flow, to the output port. */
 441     if (sw->max_idle >= 0 && (!sw->ml || out_port != OFPP_FLOOD)) {
 442         struct ofpbuf *buffer;
 443         struct cls_rule rule;
 444
 445         /* The output port is known, or we always flood everything, so add a
 446          * new flow. */
 447         cls_rule_init(&flow, &sw->wc, 0, &rule);
 448         buffer = make_add_flow(&rule, ntohl(opi->buffer_id),
 449                                sw->max_idle, actions_len);
 450         ofpbuf_put(buffer, actions, actions_len);
 451         queue_tx(sw, rconn, buffer);
 452
 453         /* If the switch didn't buffer the packet, we need to send a copy. */
 454         if (ntohl(opi->buffer_id) == UINT32_MAX && actions_len > 0) {
 455             queue_tx(sw, rconn,
 456                      make_packet_out(&pkt, UINT32_MAX, in_port,
 457                                      actions, actions_len / sizeof *actions));
 458         }
 459     } else {
 460         /* We don't know that MAC, or we don't set up flows.  Send along the
 461          * packet without setting up a flow. */
 462         if (ntohl(opi->buffer_id) != UINT32_MAX || actions_len > 0) {
 463             queue_tx(sw, rconn,
 464                      make_packet_out(&pkt, ntohl(opi->buffer_id), in_port,
 465                                      actions, actions_len / sizeof *actions));
 466         }
 467     }
 468 }
 469
 470 static void
 471 process_echo_request(struct lswitch *sw, struct rconn *rconn,
 472                      const struct ofp_header *rq)
 473 {
 474     queue_tx(sw, rconn, make_echo_reply(rq));
 475 }