ofproto/in-band.c

   1 /*
   2  * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015 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 "in-band.h"
  19 #include <arpa/inet.h>
  20 #include <errno.h>
  21 #include <inttypes.h>
  22 #include <sys/socket.h>
  23 #include <net/if.h>
  24 #include <string.h>
  25 #include <stdlib.h>
  26 #include "classifier.h"
  27 #include "dhcp.h"
  28 #include "flow.h"
  29 #include "netdev.h"
  30 #include "netlink.h"
  31 #include "odp-util.h"
  32 #include "ofp-actions.h"
  33 #include "ofproto.h"
  34 #include "ofpbuf.h"
  35 #include "ofproto-provider.h"
  36 #include "openflow/openflow.h"
  37 #include "packets.h"
  38 #include "poll-loop.h"
  39 #include "timeval.h"
  40 #include "openvswitch/vlog.h"
  41
  42 VLOG_DEFINE_THIS_MODULE(in_band);
  43
  44 /* Priorities used in classifier for in-band rules.  These values are higher
  45  * than any that may be set with OpenFlow, and "18" kind of looks like "IB".
  46  * The ordering of priorities is not important because all of the rules set up
  47  * by in-band control have the same action.  The only reason to use more than
  48  * one priority is to make the kind of flow easier to see during debugging. */
  49 enum {
  50     /* One set per bridge. */
  51     IBR_FROM_LOCAL_DHCP = 180000, /* (a) From local port, DHCP. */
  52     IBR_TO_LOCAL_ARP,             /* (b) To local port, ARP. */
  53     IBR_FROM_LOCAL_ARP,           /* (c) From local port, ARP. */
  54
  55     /* One set per unique next-hop MAC. */
  56     IBR_TO_NEXT_HOP_ARP,          /* (d) To remote MAC, ARP. */
  57     IBR_FROM_NEXT_HOP_ARP,        /* (e) From remote MAC, ARP. */
  58
  59     /* One set per unique remote IP address. */
  60     IBR_TO_REMOTE_ARP,            /* (f) To remote IP, ARP. */
  61     IBR_FROM_REMOTE_ARP,          /* (g) From remote IP, ARP. */
  62
  63     /* One set per unique remote (IP,port) pair. */
  64     IBR_TO_REMOTE_TCP,            /* (h) To remote IP, TCP port. */
  65     IBR_FROM_REMOTE_TCP           /* (i) From remote IP, TCP port. */
  66 };
  67
  68 /* Track one remote IP and next hop information. */
  69 struct in_band_remote {
  70     struct sockaddr_in remote_addr; /* IP address, in network byte order. */
  71     uint8_t remote_mac[ETH_ADDR_LEN]; /* Next-hop MAC, all-zeros if unknown. */
  72     uint8_t last_remote_mac[ETH_ADDR_LEN]; /* Previous nonzero next-hop MAC. */
  73     struct netdev *remote_netdev; /* Device to send to next-hop MAC. */
  74 };
  75
  76 /* What to do to an in_band_rule. */
  77 enum in_band_op {
  78     ADD,                       /* Add the rule to ofproto's flow table. */
  79     DEL                        /* Delete the rule from ofproto's flow table. */
  80 };
  81
  82 /* A rule to add to or delete from ofproto's flow table.  */
  83 struct in_band_rule {
  84     struct hmap_node hmap_node; /* In struct in_band's "rules" hmap. */
  85     struct match match;
  86     int priority;
  87     enum in_band_op op;
  88 };
  89
  90 struct in_band {
  91     struct ofproto *ofproto;
  92     int queue_id;
  93
  94     /* Remote information. */
  95     time_t next_remote_refresh; /* Refresh timer. */
  96     struct in_band_remote *remotes;
  97     size_t n_remotes;
  98
  99     /* Local information. */
 100     time_t next_local_refresh;       /* Refresh timer. */
 101     uint8_t local_mac[ETH_ADDR_LEN]; /* Current MAC. */
 102     struct netdev *local_netdev;     /* Local port's network device. */
 103
 104     /* Flow tracking. */
 105     struct hmap rules;          /* Contains "struct in_band_rule"s. */
 106 };
 107
 108 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 60);
 109
 110 static int
 111 refresh_remote(struct in_band *ib, struct in_band_remote *r)
 112 {
 113     struct in_addr next_hop_inaddr;
 114     char *next_hop_dev;
 115     int retval;
 116
 117     /* Find the next-hop IP address. */
 118     memset(r->remote_mac, 0, sizeof r->remote_mac);
 119     retval = netdev_get_next_hop(ib->local_netdev, &r->remote_addr.sin_addr,
 120                                  &next_hop_inaddr, &next_hop_dev);
 121     if (retval) {
 122         VLOG_WARN("cannot find route for controller ("IP_FMT"): %s",
 123                   IP_ARGS(r->remote_addr.sin_addr.s_addr),
 124                   ovs_strerror(retval));
 125         return 1;
 126     }
 127     if (!next_hop_inaddr.s_addr) {
 128         next_hop_inaddr = r->remote_addr.sin_addr;
 129     }
 130
 131     /* Open the next-hop network device. */
 132     if (!r->remote_netdev
 133         || strcmp(netdev_get_name(r->remote_netdev), next_hop_dev))
 134     {
 135         netdev_close(r->remote_netdev);
 136
 137         retval = netdev_open(next_hop_dev, "system", &r->remote_netdev);
 138         if (retval) {
 139             VLOG_WARN_RL(&rl, "cannot open netdev %s (next hop "
 140                          "to controller "IP_FMT"): %s",
 141                          next_hop_dev, IP_ARGS(r->remote_addr.sin_addr.s_addr),
 142                          ovs_strerror(retval));
 143             free(next_hop_dev);
 144             return 1;
 145         }
 146     }
 147     free(next_hop_dev);
 148
 149     /* Look up the MAC address of the next-hop IP address. */
 150     retval = netdev_arp_lookup(r->remote_netdev, next_hop_inaddr.s_addr,
 151                                r->remote_mac);
 152     if (retval) {
 153         VLOG_DBG_RL(&rl, "cannot look up remote MAC address ("IP_FMT"): %s",
 154                     IP_ARGS(next_hop_inaddr.s_addr), ovs_strerror(retval));
 155     }
 156
 157     /* If we don't have a MAC address, then refresh quickly, since we probably
 158      * will get a MAC address soon (via ARP).  Otherwise, we can afford to wait
 159      * a little while. */
 160     return eth_addr_is_zero(r->remote_mac) ? 1 : 10;
 161 }
 162
 163 static bool
 164 refresh_remotes(struct in_band *ib)
 165 {
 166     struct in_band_remote *r;
 167     bool any_changes;
 168
 169     if (time_now() < ib->next_remote_refresh) {
 170         return false;
 171     }
 172
 173     any_changes = false;
 174     ib->next_remote_refresh = TIME_MAX;
 175     for (r = ib->remotes; r < &ib->remotes[ib->n_remotes]; r++) {
 176         uint8_t old_remote_mac[ETH_ADDR_LEN];
 177         time_t next_refresh;
 178
 179         /* Save old MAC. */
 180         memcpy(old_remote_mac, r->remote_mac, ETH_ADDR_LEN);
 181
 182         /* Refresh remote information. */
 183         next_refresh = refresh_remote(ib, r) + time_now();
 184         ib->next_remote_refresh = MIN(ib->next_remote_refresh, next_refresh);
 185
 186         /* If the MAC changed, log the changes. */
 187         if (!eth_addr_equals(r->remote_mac, old_remote_mac)) {
 188             any_changes = true;
 189             if (!eth_addr_is_zero(r->remote_mac)
 190                 && !eth_addr_equals(r->last_remote_mac, r->remote_mac)) {
 191                 VLOG_DBG("remote MAC address changed from "ETH_ADDR_FMT
 192                          " to "ETH_ADDR_FMT,
 193                          ETH_ADDR_ARGS(r->last_remote_mac),
 194                          ETH_ADDR_ARGS(r->remote_mac));
 195                 memcpy(r->last_remote_mac, r->remote_mac, ETH_ADDR_LEN);
 196             }
 197         }
 198     }
 199
 200     return any_changes;
 201 }
 202
 203 /* Refreshes the MAC address of the local port into ib->local_mac, if it is due
 204  * for a refresh.  Returns true if anything changed, otherwise false.  */
 205 static bool
 206 refresh_local(struct in_band *ib)
 207 {
 208     uint8_t ea[ETH_ADDR_LEN];
 209     time_t now;
 210
 211     now = time_now();
 212     if (now < ib->next_local_refresh) {
 213         return false;
 214     }
 215     ib->next_local_refresh = now + 1;
 216
 217     if (netdev_get_etheraddr(ib->local_netdev, ea)
 218         || eth_addr_equals(ea, ib->local_mac)) {
 219         return false;
 220     }
 221
 222     memcpy(ib->local_mac, ea, ETH_ADDR_LEN);
 223     return true;
 224 }
 225
 226 /* Returns true if packets in 'flow' should be directed to the local port.
 227  * (This keeps the flow table from preventing DHCP replies from being seen by
 228  * the local port.) */
 229 bool
 230 in_band_must_output_to_local_port(const struct flow *flow)
 231 {
 232     return (flow->dl_type == htons(ETH_TYPE_IP)
 233             && flow->nw_proto == IPPROTO_UDP
 234             && flow->tp_src == htons(DHCP_SERVER_PORT)
 235             && flow->tp_dst == htons(DHCP_CLIENT_PORT));
 236 }
 237
 238 /* Returns the number of in-band rules currently installed in the flow
 239  * table. */
 240 int
 241 in_band_count_rules(const struct in_band *ib)
 242 {
 243     return hmap_count(&ib->rules);
 244 }
 245
 246 static void
 247 add_rule(struct in_band *ib, const struct match *match, int priority)
 248 {
 249     uint32_t hash = match_hash(match, 0);
 250     struct in_band_rule *rule;
 251
 252     HMAP_FOR_EACH_WITH_HASH (rule, hmap_node, hash, &ib->rules) {
 253         if (match_equal(&rule->match, match)) {
 254             rule->op = ADD;
 255             return;
 256         }
 257     }
 258
 259     rule = xmalloc(sizeof *rule);
 260     rule->match = *match;
 261     rule->priority = priority;
 262     rule->op = ADD;
 263     hmap_insert(&ib->rules, &rule->hmap_node, hash);
 264 }
 265
 266 static void
 267 update_rules(struct in_band *ib)
 268 {
 269     struct in_band_rule *ib_rule;
 270     struct in_band_remote *r;
 271     struct match match;
 272
 273     /* Mark all the existing rules for deletion.  (Afterward we will re-add any
 274      * rules that are still valid.) */
 275     HMAP_FOR_EACH (ib_rule, hmap_node, &ib->rules) {
 276         ib_rule->op = DEL;
 277     }
 278
 279     if (ib->n_remotes && !eth_addr_is_zero(ib->local_mac)) {
 280         /* (a) Allow DHCP requests sent from the local port. */
 281         match_init_catchall(&match);
 282         match_set_in_port(&match, OFPP_LOCAL);
 283         match_set_dl_type(&match, htons(ETH_TYPE_IP));
 284         match_set_dl_src(&match, ib->local_mac);
 285         match_set_nw_proto(&match, IPPROTO_UDP);
 286         match_set_tp_src(&match, htons(DHCP_CLIENT_PORT));
 287         match_set_tp_dst(&match, htons(DHCP_SERVER_PORT));
 288         add_rule(ib, &match, IBR_FROM_LOCAL_DHCP);
 289
 290         /* (b) Allow ARP replies to the local port's MAC address. */
 291         match_init_catchall(&match);
 292         match_set_dl_type(&match, htons(ETH_TYPE_ARP));
 293         match_set_dl_dst(&match, ib->local_mac);
 294         match_set_nw_proto(&match, ARP_OP_REPLY);
 295         add_rule(ib, &match, IBR_TO_LOCAL_ARP);
 296
 297         /* (c) Allow ARP requests from the local port's MAC address.  */
 298         match_init_catchall(&match);
 299         match_set_dl_type(&match, htons(ETH_TYPE_ARP));
 300         match_set_dl_src(&match, ib->local_mac);
 301         match_set_nw_proto(&match, ARP_OP_REQUEST);
 302         add_rule(ib, &match, IBR_FROM_LOCAL_ARP);
 303     }
 304
 305     for (r = ib->remotes; r < &ib->remotes[ib->n_remotes]; r++) {
 306         const uint8_t *remote_mac = r->remote_mac;
 307
 308         if (eth_addr_is_zero(remote_mac)) {
 309             continue;
 310         }
 311
 312         /* (d) Allow ARP replies to the next hop's MAC address. */
 313         match_init_catchall(&match);
 314         match_set_dl_type(&match, htons(ETH_TYPE_ARP));
 315         match_set_dl_dst(&match, remote_mac);
 316         match_set_nw_proto(&match, ARP_OP_REPLY);
 317         add_rule(ib, &match, IBR_TO_NEXT_HOP_ARP);
 318
 319         /* (e) Allow ARP requests from the next hop's MAC address. */
 320         match_init_catchall(&match);
 321         match_set_dl_type(&match, htons(ETH_TYPE_ARP));
 322         match_set_dl_src(&match, remote_mac);
 323         match_set_nw_proto(&match, ARP_OP_REQUEST);
 324         add_rule(ib, &match, IBR_FROM_NEXT_HOP_ARP);
 325     }
 326
 327     for (r = ib->remotes; r < &ib->remotes[ib->n_remotes]; r++) {
 328         const struct sockaddr_in *a = &r->remote_addr;
 329
 330         /* (f) Allow ARP replies containing the remote's IP address as a
 331          * target. */
 332         match_init_catchall(&match);
 333         match_set_dl_type(&match, htons(ETH_TYPE_ARP));
 334         match_set_nw_proto(&match, ARP_OP_REPLY);
 335         match_set_nw_dst(&match, a->sin_addr.s_addr);
 336         add_rule(ib, &match, IBR_TO_REMOTE_ARP);
 337
 338         /* (g) Allow ARP requests containing the remote's IP address as a
 339          * source. */
 340         match_init_catchall(&match);
 341         match_set_dl_type(&match, htons(ETH_TYPE_ARP));
 342         match_set_nw_proto(&match, ARP_OP_REQUEST);
 343         match_set_nw_src(&match, a->sin_addr.s_addr);
 344         add_rule(ib, &match, IBR_FROM_REMOTE_ARP);
 345
 346         /* (h) Allow TCP traffic to the remote's IP and port. */
 347         match_init_catchall(&match);
 348         match_set_dl_type(&match, htons(ETH_TYPE_IP));
 349         match_set_nw_proto(&match, IPPROTO_TCP);
 350         match_set_nw_dst(&match, a->sin_addr.s_addr);
 351         match_set_tp_dst(&match, a->sin_port);
 352         add_rule(ib, &match, IBR_TO_REMOTE_TCP);
 353
 354         /* (i) Allow TCP traffic from the remote's IP and port. */
 355         match_init_catchall(&match);
 356         match_set_dl_type(&match, htons(ETH_TYPE_IP));
 357         match_set_nw_proto(&match, IPPROTO_TCP);
 358         match_set_nw_src(&match, a->sin_addr.s_addr);
 359         match_set_tp_src(&match, a->sin_port);
 360         add_rule(ib, &match, IBR_FROM_REMOTE_TCP);
 361     }
 362 }
 363
 364 /* Updates the OpenFlow flow table for the current state of in-band control.
 365  * Returns true ordinarily.  Returns false if no remotes are configured on 'ib'
 366  * and 'ib' doesn't have any rules left to remove from the OpenFlow flow
 367  * table.  Thus, a false return value means that the caller can destroy 'ib'
 368  * without leaving extra flows hanging around in the flow table. */
 369 bool
 370 in_band_run(struct in_band *ib)
 371 {
 372     uint64_t ofpacts_stub[128 / 8];
 373     struct ofpbuf ofpacts;
 374
 375     struct in_band_rule *rule, *next;
 376
 377     ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
 378
 379     if (ib->queue_id >= 0) {
 380         ofpact_put_SET_QUEUE(&ofpacts)->queue_id = ib->queue_id;
 381     }
 382     ofpact_put_OUTPUT(&ofpacts)->port = OFPP_NORMAL;
 383
 384     refresh_local(ib);
 385     refresh_remotes(ib);
 386
 387     update_rules(ib);
 388
 389     HMAP_FOR_EACH_SAFE (rule, next, hmap_node, &ib->rules) {
 390         switch (rule->op) {
 391         case ADD:
 392             ofproto_add_flow(ib->ofproto, &rule->match, rule->priority,
 393                              ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts));
 394             break;
 395
 396         case DEL:
 397             ofproto_delete_flow(ib->ofproto, &rule->match, rule->priority);
 398             hmap_remove(&ib->rules, &rule->hmap_node);
 399             free(rule);
 400             break;
 401         }
 402     }
 403
 404     ofpbuf_uninit(&ofpacts);
 405
 406     return ib->n_remotes || !hmap_is_empty(&ib->rules);
 407 }
 408
 409 void
 410 in_band_wait(struct in_band *in_band)
 411 {
 412     long long int wakeup
 413             = MIN(in_band->next_remote_refresh, in_band->next_local_refresh);
 414     poll_timer_wait_until(wakeup * 1000);
 415 }
 416
 417 int
 418 in_band_create(struct ofproto *ofproto, const char *local_name,
 419                struct in_band **in_bandp)
 420 {
 421     struct in_band *in_band;
 422     struct netdev *local_netdev;
 423     int error;
 424
 425     *in_bandp = NULL;
 426     error = netdev_open(local_name, "internal", &local_netdev);
 427     if (error) {
 428         VLOG_ERR("failed to initialize in-band control: cannot open "
 429                  "datapath local port %s (%s)",
 430                  local_name, ovs_strerror(error));
 431         return error;
 432     }
 433
 434     in_band = xzalloc(sizeof *in_band);
 435     in_band->ofproto = ofproto;
 436     in_band->queue_id = -1;
 437     in_band->next_remote_refresh = TIME_MIN;
 438     in_band->next_local_refresh = TIME_MIN;
 439     in_band->local_netdev = local_netdev;
 440     hmap_init(&in_band->rules);
 441
 442     *in_bandp = in_band;
 443
 444     return 0;
 445 }
 446
 447 void
 448 in_band_destroy(struct in_band *ib)
 449 {
 450     if (ib) {
 451         struct in_band_rule *rule, *next;
 452
 453         HMAP_FOR_EACH_SAFE (rule, next, hmap_node, &ib->rules) {
 454             hmap_remove(&ib->rules, &rule->hmap_node);
 455             free(rule);
 456         }
 457         hmap_destroy(&ib->rules);
 458         in_band_set_remotes(ib, NULL, 0);
 459         netdev_close(ib->local_netdev);
 460         free(ib);
 461     }
 462 }
 463
 464 static bool
 465 any_addresses_changed(struct in_band *ib,
 466                       const struct sockaddr_in *addresses, size_t n)
 467 {
 468     size_t i;
 469
 470     if (n != ib->n_remotes) {
 471         return true;
 472     }
 473
 474     for (i = 0; i < n; i++) {
 475         const struct sockaddr_in *old = &ib->remotes[i].remote_addr;
 476         const struct sockaddr_in *new = &addresses[i];
 477
 478         if (old->sin_addr.s_addr != new->sin_addr.s_addr ||
 479             old->sin_port != new->sin_port) {
 480             return true;
 481         }
 482     }
 483
 484     return false;
 485 }
 486
 487 void
 488 in_band_set_remotes(struct in_band *ib,
 489                     const struct sockaddr_in *addresses, size_t n)
 490 {
 491     size_t i;
 492
 493     if (!any_addresses_changed(ib, addresses, n)) {
 494         return;
 495     }
 496
 497     /* Clear old remotes. */
 498     for (i = 0; i < ib->n_remotes; i++) {
 499         netdev_close(ib->remotes[i].remote_netdev);
 500     }
 501     free(ib->remotes);
 502
 503     /* Set up new remotes. */
 504     ib->remotes = n ? xzalloc(n * sizeof *ib->remotes) : NULL;
 505     ib->n_remotes = n;
 506     for (i = 0; i < n; i++) {
 507         ib->remotes[i].remote_addr = addresses[i];
 508     }
 509
 510     /* Force refresh in next call to in_band_run(). */
 511     ib->next_remote_refresh = TIME_MIN;
 512 }
 513
 514 /* Sets the OpenFlow queue used by flows set up by 'ib' to 'queue_id'.  If
 515  * 'queue_id' is negative, 'ib' will not set any queue (which is also the
 516  * default). */
 517 void
 518 in_band_set_queue(struct in_band *ib, int queue_id)
 519 {
 520     ib->queue_id = queue_id;
 521 }
 522