lib/learning-switch.c

   1 /*
   2  * Copyright (c) 2008, 2009, 2010 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 "flow.h"
  27 #include "mac-learning.h"
  28 #include "ofpbuf.h"
  29 #include "ofp-parse.h"
  30 #include "ofp-print.h"
  31 #include "ofp-util.h"
  32 #include "openflow/openflow.h"
  33 #include "poll-loop.h"
  34 #include "queue.h"
  35 #include "rconn.h"
  36 #include "timeval.h"
  37 #include "vconn.h"
  38 #include "vlog.h"
  39 #include "xtoxll.h"
  40
  41 VLOG_DEFINE_THIS_MODULE(learning_switch)
  42
  43 struct lswitch {
  44     /* If nonnegative, the switch sets up flows that expire after the given
  45      * number of seconds (or never expire, if the value is OFP_FLOW_PERMANENT).
  46      * Otherwise, the switch processes every packet. */
  47     int max_idle;
  48
  49     unsigned long long int datapath_id;
  50     time_t last_features_request;
  51     struct mac_learning *ml;    /* NULL to act as hub instead of switch. */
  52     uint32_t wildcards;         /* Wildcards to apply to flows. */
  53     bool action_normal;         /* Use OFPP_NORMAL? */
  54     uint32_t queue;             /* OpenFlow queue to use, or UINT32_MAX. */
  55
  56     /* Number of outgoing queued packets on the rconn. */
  57     struct rconn_packet_counter *queued;
  58 };
  59
  60 /* The log messages here could actually be useful in debugging, so keep the
  61  * rate limit relatively high. */
  62 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
  63
  64 static void queue_tx(struct lswitch *, struct rconn *, struct ofpbuf *);
  65 static void send_features_request(struct lswitch *, struct rconn *);
  66 static void send_default_flows(struct lswitch *sw, struct rconn *rconn,
  67                                FILE *default_flows);
  68
  69 typedef void packet_handler_func(struct lswitch *, struct rconn *, void *);
  70 static packet_handler_func process_switch_features;
  71 static packet_handler_func process_packet_in;
  72 static packet_handler_func process_echo_request;
  73
  74 /* Creates and returns a new learning switch.
  75  *
  76  * If 'learn_macs' is true, the new switch will learn the ports on which MAC
  77  * addresses appear.  Otherwise, the new switch will flood all packets.
  78  *
  79  * If 'max_idle' is nonnegative, the new switch will set up flows that expire
  80  * after the given number of seconds (or never expire, if 'max_idle' is
  81  * OFP_FLOW_PERMANENT).  Otherwise, the new switch will process every packet.
  82  *
  83  * The caller may provide the file stream 'default_flows' that defines
  84  * default flows that should be pushed when a switch connects.  Each
  85  * line is a flow entry in the format described for "add-flows" command
  86  * in the Flow Syntax section of the ovs-ofct(8) man page.  The caller
  87  * is responsible for closing the stream.
  88  *
  89  * 'rconn' is used to send out an OpenFlow features request. */
  90 struct lswitch *
  91 lswitch_create(struct rconn *rconn, bool learn_macs,
  92                bool exact_flows, int max_idle, bool action_normal,
  93                FILE *default_flows)
  94 {
  95     struct lswitch *sw;
  96
  97     sw = xzalloc(sizeof *sw);
  98     sw->max_idle = max_idle;
  99     sw->datapath_id = 0;
 100     sw->last_features_request = time_now() - 1;
 101     sw->ml = learn_macs ? mac_learning_create() : NULL;
 102     sw->action_normal = action_normal;
 103     if (exact_flows) {
 104         /* Exact match. */
 105         sw->wildcards = 0;
 106     } else {
 107         /* We cannot wildcard all fields.
 108          * We need in_port to detect moves.
 109          * We need both SA and DA to do learning. */
 110         sw->wildcards = (OFPFW_DL_TYPE | OFPFW_NW_SRC_MASK | OFPFW_NW_DST_MASK
 111                          | OFPFW_NW_PROTO | OFPFW_TP_SRC | OFPFW_TP_DST);
 112     }
 113     sw->queue = UINT32_MAX;
 114     sw->queued = rconn_packet_counter_create();
 115     send_features_request(sw, rconn);
 116     if (default_flows) {
 117         send_default_flows(sw, rconn, default_flows);
 118     }
 119     return sw;
 120 }
 121
 122 /* Destroys 'sw'. */
 123 void
 124 lswitch_destroy(struct lswitch *sw)
 125 {
 126     if (sw) {
 127         mac_learning_destroy(sw->ml);
 128         rconn_packet_counter_destroy(sw->queued);
 129         free(sw);
 130     }
 131 }
 132
 133 /* Sets 'queue' as the OpenFlow queue used by packets and flows set up by 'sw'.
 134  * Specify UINT32_MAX to avoid specifying a particular queue, which is also the
 135  * default if this function is never called for 'sw'.  */
 136 void
 137 lswitch_set_queue(struct lswitch *sw, uint32_t queue)
 138 {
 139     sw->queue = queue;
 140 }
 141
 142 /* Takes care of necessary 'sw' activity, except for receiving packets (which
 143  * the caller must do). */
 144 void
 145 lswitch_run(struct lswitch *sw)
 146 {
 147     if (sw->ml) {
 148         mac_learning_run(sw->ml, NULL);
 149     }
 150 }
 151
 152 void
 153 lswitch_wait(struct lswitch *sw)
 154 {
 155     if (sw->ml) {
 156         mac_learning_wait(sw->ml);
 157     }
 158 }
 159
 160 /* Processes 'msg', which should be an OpenFlow received on 'rconn', according
 161  * to the learning switch state in 'sw'.  The most likely result of processing
 162  * is that flow-setup and packet-out OpenFlow messages will be sent out on
 163  * 'rconn'.  */
 164 void
 165 lswitch_process_packet(struct lswitch *sw, struct rconn *rconn,
 166                        const struct ofpbuf *msg)
 167 {
 168     struct processor {
 169         uint8_t type;
 170         size_t min_size;
 171         packet_handler_func *handler;
 172     };
 173     static const struct processor processors[] = {
 174         {
 175             OFPT_ECHO_REQUEST,
 176             sizeof(struct ofp_header),
 177             process_echo_request
 178         },
 179         {
 180             OFPT_FEATURES_REPLY,
 181             sizeof(struct ofp_switch_features),
 182             process_switch_features
 183         },
 184         {
 185             OFPT_PACKET_IN,
 186             offsetof(struct ofp_packet_in, data),
 187             process_packet_in
 188         },
 189         {
 190             OFPT_FLOW_REMOVED,
 191             sizeof(struct ofp_flow_removed),
 192             NULL
 193         },
 194     };
 195     const size_t n_processors = ARRAY_SIZE(processors);
 196     const struct processor *p;
 197     struct ofp_header *oh;
 198
 199     oh = msg->data;
 200     if (sw->datapath_id == 0
 201         && oh->type != OFPT_ECHO_REQUEST
 202         && oh->type != OFPT_FEATURES_REPLY) {
 203         send_features_request(sw, rconn);
 204         return;
 205     }
 206
 207     for (p = processors; p < &processors[n_processors]; p++) {
 208         if (oh->type == p->type) {
 209             if (msg->size < p->min_size) {
 210                 VLOG_WARN_RL(&rl, "%016llx: %s: too short (%zu bytes) for "
 211                              "type %"PRIu8" (min %zu)", sw->datapath_id,
 212                              rconn_get_name(rconn), msg->size, oh->type,
 213                              p->min_size);
 214                 return;
 215             }
 216             if (p->handler) {
 217                 (p->handler)(sw, rconn, msg->data);
 218             }
 219             return;
 220         }
 221     }
 222     if (VLOG_IS_DBG_ENABLED()) {
 223         char *s = ofp_to_string(msg->data, msg->size, 2);
 224         VLOG_DBG_RL(&rl, "%016llx: OpenFlow packet ignored: %s",
 225                     sw->datapath_id, s);
 226         free(s);
 227     }
 228 }
 229 \f
 230 static void
 231 send_features_request(struct lswitch *sw, struct rconn *rconn)
 232 {
 233     time_t now = time_now();
 234     if (now >= sw->last_features_request + 1) {
 235         struct ofpbuf *b;
 236         struct ofp_switch_config *osc;
 237
 238         /* Send OFPT_FEATURES_REQUEST. */
 239         make_openflow(sizeof(struct ofp_header), OFPT_FEATURES_REQUEST, &b);
 240         queue_tx(sw, rconn, b);
 241
 242         /* Send OFPT_SET_CONFIG. */
 243         osc = make_openflow(sizeof *osc, OFPT_SET_CONFIG, &b);
 244         osc->miss_send_len = htons(OFP_DEFAULT_MISS_SEND_LEN);
 245         queue_tx(sw, rconn, b);
 246
 247         sw->last_features_request = now;
 248     }
 249 }
 250
 251 static void
 252 send_default_flows(struct lswitch *sw, struct rconn *rconn,
 253                    FILE *default_flows)
 254 {
 255     char line[1024];
 256
 257     while (fgets(line, sizeof line, default_flows)) {
 258         struct ofpbuf *b;
 259         struct ofp_flow_mod *ofm;
 260         uint16_t priority, idle_timeout, hard_timeout;
 261         uint64_t cookie;
 262         struct ofp_match match;
 263
 264         char *comment;
 265
 266         /* Delete comments. */
 267         comment = strchr(line, '#');
 268         if (comment) {
 269             *comment = '\0';
 270         }
 271
 272         /* Drop empty lines. */
 273         if (line[strspn(line, " \t\n")] == '\0') {
 274             continue;
 275         }
 276
 277         /* Parse and send.  str_to_flow() will expand and reallocate the data
 278          * in 'buffer', so we can't keep pointers to across the str_to_flow()
 279          * call. */
 280         make_openflow(sizeof *ofm, OFPT_FLOW_MOD, &b);
 281         parse_ofp_str(line, &match, b,
 282                       NULL, NULL, &priority, &idle_timeout, &hard_timeout,
 283                       &cookie);
 284         ofm = b->data;
 285         ofm->match = match;
 286         ofm->command = htons(OFPFC_ADD);
 287         ofm->cookie = htonll(cookie);
 288         ofm->idle_timeout = htons(idle_timeout);
 289         ofm->hard_timeout = htons(hard_timeout);
 290         ofm->buffer_id = htonl(UINT32_MAX);
 291         ofm->priority = htons(priority);
 292
 293         update_openflow_length(b);
 294         queue_tx(sw, rconn, b);
 295     }
 296 }
 297
 298 static void
 299 queue_tx(struct lswitch *sw, struct rconn *rconn, struct ofpbuf *b)
 300 {
 301     int retval = rconn_send_with_limit(rconn, b, sw->queued, 10);
 302     if (retval && retval != ENOTCONN) {
 303         if (retval == EAGAIN) {
 304             VLOG_INFO_RL(&rl, "%016llx: %s: tx queue overflow",
 305                          sw->datapath_id, rconn_get_name(rconn));
 306         } else {
 307             VLOG_WARN_RL(&rl, "%016llx: %s: send: %s",
 308                          sw->datapath_id, rconn_get_name(rconn),
 309                          strerror(retval));
 310         }
 311     }
 312 }
 313
 314 static void
 315 process_switch_features(struct lswitch *sw, struct rconn *rconn OVS_UNUSED,
 316                         void *osf_)
 317 {
 318     struct ofp_switch_features *osf = osf_;
 319
 320     sw->datapath_id = ntohll(osf->datapath_id);
 321 }
 322
 323 static uint16_t
 324 lswitch_choose_destination(struct lswitch *sw, const flow_t *flow)
 325 {
 326     uint16_t out_port;
 327
 328     /* Learn the source MAC. */
 329     if (sw->ml) {
 330         if (mac_learning_learn(sw->ml, flow->dl_src, 0, flow->in_port,
 331                                GRAT_ARP_LOCK_NONE)) {
 332             VLOG_DBG_RL(&rl, "%016llx: learned that "ETH_ADDR_FMT" is on "
 333                         "port %"PRIu16, sw->datapath_id,
 334                         ETH_ADDR_ARGS(flow->dl_src), flow->in_port);
 335         }
 336     }
 337
 338     /* Drop frames for reserved multicast addresses. */
 339     if (eth_addr_is_reserved(flow->dl_dst)) {
 340         return OFPP_NONE;
 341     }
 342
 343     out_port = OFPP_FLOOD;
 344     if (sw->ml) {
 345         int learned_port = mac_learning_lookup(sw->ml, flow->dl_dst, 0, NULL);
 346         if (learned_port >= 0) {
 347             out_port = learned_port;
 348             if (out_port == flow->in_port) {
 349                 /* Don't send a packet back out its input port. */
 350                 return OFPP_NONE;
 351             }
 352         }
 353     }
 354
 355     /* Check if we need to use "NORMAL" action. */
 356     if (sw->action_normal && out_port != OFPP_FLOOD) {
 357         return OFPP_NORMAL;
 358     }
 359
 360     return out_port;
 361 }
 362
 363 static void
 364 process_packet_in(struct lswitch *sw, struct rconn *rconn, void *opi_)
 365 {
 366     struct ofp_packet_in *opi = opi_;
 367     uint16_t in_port = ntohs(opi->in_port);
 368     uint16_t out_port;
 369
 370     struct ofp_action_header actions[2];
 371     size_t actions_len;
 372
 373     size_t pkt_ofs, pkt_len;
 374     struct ofpbuf pkt;
 375     flow_t flow;
 376
 377     /* Ignore packets sent via output to OFPP_CONTROLLER.  This library never
 378      * uses such an action.  You never know what experiments might be going on,
 379      * though, and it seems best not to interfere with them. */
 380     if (opi->reason != OFPR_NO_MATCH) {
 381         return;
 382     }
 383
 384     /* Extract flow data from 'opi' into 'flow'. */
 385     pkt_ofs = offsetof(struct ofp_packet_in, data);
 386     pkt_len = ntohs(opi->header.length) - pkt_ofs;
 387     pkt.data = opi->data;
 388     pkt.size = pkt_len;
 389     flow_extract(&pkt, 0, in_port, &flow);
 390
 391     /* Choose output port. */
 392     out_port = lswitch_choose_destination(sw, &flow);
 393
 394     /* Make actions. */
 395     if (out_port == OFPP_NONE) {
 396         actions_len = 0;
 397     } else if (sw->queue == UINT32_MAX || out_port >= OFPP_MAX) {
 398         struct ofp_action_output oao;
 399
 400         memset(&oao, 0, sizeof oao);
 401         oao.type = htons(OFPAT_OUTPUT);
 402         oao.len = htons(sizeof oao);
 403         oao.port = htons(out_port);
 404
 405         memcpy(actions, &oao, sizeof oao);
 406         actions_len = sizeof oao;
 407     } else {
 408         struct ofp_action_enqueue oae;
 409
 410         memset(&oae, 0, sizeof oae);
 411         oae.type = htons(OFPAT_ENQUEUE);
 412         oae.len = htons(sizeof oae);
 413         oae.port = htons(out_port);
 414         oae.queue_id = htonl(sw->queue);
 415
 416         memcpy(actions, &oae, sizeof oae);
 417         actions_len = sizeof oae;
 418     }
 419     assert(actions_len <= sizeof actions);
 420
 421     /* Send the packet, and possibly the whole flow, to the output port. */
 422     if (sw->max_idle >= 0 && (!sw->ml || out_port != OFPP_FLOOD)) {
 423         struct ofpbuf *buffer;
 424         struct ofp_flow_mod *ofm;
 425
 426         /* The output port is known, or we always flood everything, so add a
 427          * new flow. */
 428         buffer = make_add_flow(&flow, ntohl(opi->buffer_id),
 429                                sw->max_idle, actions_len);
 430         ofpbuf_put(buffer, actions, actions_len);
 431         ofm = buffer->data;
 432         ofm->match.wildcards = htonl(sw->wildcards);
 433         queue_tx(sw, rconn, buffer);
 434
 435         /* If the switch didn't buffer the packet, we need to send a copy. */
 436         if (ntohl(opi->buffer_id) == UINT32_MAX && actions_len > 0) {
 437             queue_tx(sw, rconn,
 438                      make_packet_out(&pkt, UINT32_MAX, in_port,
 439                                      actions, actions_len / sizeof *actions));
 440         }
 441     } else {
 442         /* We don't know that MAC, or we don't set up flows.  Send along the
 443          * packet without setting up a flow. */
 444         if (ntohl(opi->buffer_id) != UINT32_MAX || actions_len > 0) {
 445             queue_tx(sw, rconn,
 446                      make_packet_out(&pkt, ntohl(opi->buffer_id), in_port,
 447                                      actions, actions_len / sizeof *actions));
 448         }
 449     }
 450 }
 451
 452 static void
 453 process_echo_request(struct lswitch *sw, struct rconn *rconn, void *rq_)
 454 {
 455     struct ofp_header *rq = rq_;
 456     queue_tx(sw, rconn, make_echo_reply(rq));
 457 }