ofproto/ofproto-dpif.c

   1 /*
   2  * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 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 <errno.h>
  19
  20 #include "bfd.h"
  21 #include "bond.h"
  22 #include "bundle.h"
  23 #include "byte-order.h"
  24 #include "connectivity.h"
  25 #include "connmgr.h"
  26 #include "coverage.h"
  27 #include "cfm.h"
  28 #include "dpif.h"
  29 #include "fail-open.h"
  30 #include "guarded-list.h"
  31 #include "hmapx.h"
  32 #include "lacp.h"
  33 #include "learn.h"
  34 #include "mac-learning.h"
  35 #include "mcast-snooping.h"
  36 #include "multipath.h"
  37 #include "netdev-vport.h"
  38 #include "netdev.h"
  39 #include "netlink.h"
  40 #include "nx-match.h"
  41 #include "odp-util.h"
  42 #include "odp-execute.h"
  43 #include "ofproto/ofproto-dpif.h"
  44 #include "ofproto/ofproto-provider.h"
  45 #include "ofproto-dpif-ipfix.h"
  46 #include "ofproto-dpif-mirror.h"
  47 #include "ofproto-dpif-monitor.h"
  48 #include "ofproto-dpif-rid.h"
  49 #include "ofproto-dpif-sflow.h"
  50 #include "ofproto-dpif-upcall.h"
  51 #include "ofproto-dpif-xlate.h"
  52 #include "openvswitch/ofp-actions.h"
  53 #include "openvswitch/dynamic-string.h"
  54 #include "openvswitch/meta-flow.h"
  55 #include "openvswitch/ofp-parse.h"
  56 #include "openvswitch/ofp-print.h"
  57 #include "openvswitch/ofp-util.h"
  58 #include "openvswitch/ofpbuf.h"
  59 #include "openvswitch/vlog.h"
  60 #include "ovs-lldp.h"
  61 #include "ovs-rcu.h"
  62 #include "ovs-router.h"
  63 #include "poll-loop.h"
  64 #include "seq.h"
  65 #include "simap.h"
  66 #include "smap.h"
  67 #include "timer.h"
  68 #include "tunnel.h"
  69 #include "unaligned.h"
  70 #include "unixctl.h"
  71 #include "vlan-bitmap.h"
  72
  73 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  74
  75 COVERAGE_DEFINE(ofproto_dpif_expired);
  76 COVERAGE_DEFINE(packet_in_overflow);
  77
  78 struct flow_miss;
  79
  80 struct rule_dpif {
  81     struct rule up;
  82
  83     /* These statistics:
  84      *
  85      *   - Do include packets and bytes from datapath flows which have not
  86      *   recently been processed by a revalidator. */
  87     struct ovs_mutex stats_mutex;
  88     struct dpif_flow_stats stats OVS_GUARDED;
  89
  90    /* In non-NULL, will point to a new rule (for which a reference is held) to
  91     * which all the stats updates should be forwarded. This exists only
  92     * transitionally when flows are replaced.
  93     *
  94     * Protected by stats_mutex.  If both 'rule->stats_mutex' and
  95     * 'rule->new_rule->stats_mutex' must be held together, acquire them in that
  96     * order, */
  97     struct rule_dpif *new_rule OVS_GUARDED;
  98
  99     /* If non-zero then the recirculation id that has
 100      * been allocated for use with this rule.
 101      * The recirculation id and associated internal flow should
 102      * be freed when the rule is freed */
 103     uint32_t recirc_id;
 104 };
 105
 106 /* RULE_CAST() depends on this. */
 107 BUILD_ASSERT_DECL(offsetof(struct rule_dpif, up) == 0);
 108
 109 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
 110                            long long int *used);
 111 static struct rule_dpif *rule_dpif_cast(const struct rule *);
 112 static void rule_expire(struct rule_dpif *, long long now);
 113
 114 struct group_dpif {
 115     struct ofgroup up;
 116
 117     /* These statistics:
 118      *
 119      *   - Do include packets and bytes from datapath flows which have not
 120      *   recently been processed by a revalidator. */
 121     struct ovs_mutex stats_mutex;
 122     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 123     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 124 };
 125
 126 struct ofbundle {
 127     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
 128     struct ofproto_dpif *ofproto; /* Owning ofproto. */
 129     void *aux;                  /* Key supplied by ofproto's client. */
 130     char *name;                 /* Identifier for log messages. */
 131
 132     /* Configuration. */
 133     struct ovs_list ports;      /* Contains "struct ofport"s. */
 134     enum port_vlan_mode vlan_mode; /* VLAN mode */
 135     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
 136     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
 137                                  * NULL if all VLANs are trunked. */
 138     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 139     struct bond *bond;          /* Nonnull iff more than one port. */
 140     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 141
 142     /* Status. */
 143     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 144 };
 145
 146 static void bundle_remove(struct ofport *);
 147 static void bundle_update(struct ofbundle *);
 148 static void bundle_destroy(struct ofbundle *);
 149 static void bundle_del_port(struct ofport_dpif *);
 150 static void bundle_run(struct ofbundle *);
 151 static void bundle_wait(struct ofbundle *);
 152 static void bundle_flush_macs(struct ofbundle *, bool);
 153 static void bundle_move(struct ofbundle *, struct ofbundle *);
 154
 155 static void stp_run(struct ofproto_dpif *ofproto);
 156 static void stp_wait(struct ofproto_dpif *ofproto);
 157 static int set_stp_port(struct ofport *,
 158                         const struct ofproto_port_stp_settings *);
 159
 160 static void rstp_run(struct ofproto_dpif *ofproto);
 161 static void set_rstp_port(struct ofport *,
 162                          const struct ofproto_port_rstp_settings *);
 163
 164 struct ofport_dpif {
 165     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 166     struct ofport up;
 167
 168     odp_port_t odp_port;
 169     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 170     struct ovs_list bundle_node;/* In struct ofbundle's "ports" list. */
 171     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 172     struct bfd *bfd;            /* BFD, if any. */
 173     struct lldp *lldp;          /* lldp, if any. */
 174     bool may_enable;            /* May be enabled in bonds. */
 175     bool is_tunnel;             /* This port is a tunnel. */
 176     bool is_layer3;             /* This is a layer 3 port. */
 177     long long int carrier_seq;  /* Carrier status changes. */
 178     struct ofport_dpif *peer;   /* Peer if patch port. */
 179
 180     /* Spanning tree. */
 181     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 182     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 183     long long int stp_state_entered;
 184
 185     /* Rapid Spanning Tree. */
 186     struct rstp_port *rstp_port; /* Rapid Spanning Tree Protocol, if any. */
 187     enum rstp_state rstp_state; /* Always RSTP_DISABLED if RSTP not in use. */
 188
 189     /* Queue to DSCP mapping. */
 190     struct ofproto_port_queue *qdscp;
 191     size_t n_qdscp;
 192 };
 193
 194 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 195                                        ofp_port_t);
 196
 197 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 198                                        odp_port_t);
 199
 200 static struct ofport_dpif *
 201 ofport_dpif_cast(const struct ofport *ofport)
 202 {
 203     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 204 }
 205
 206 static void port_run(struct ofport_dpif *);
 207 static int set_bfd(struct ofport *, const struct smap *);
 208 static int set_cfm(struct ofport *, const struct cfm_settings *);
 209 static int set_lldp(struct ofport *ofport_, const struct smap *cfg);
 210 static void ofport_update_peer(struct ofport_dpif *);
 211
 212 /* Reasons that we might need to revalidate every datapath flow, and
 213  * corresponding coverage counters.
 214  *
 215  * A value of 0 means that there is no need to revalidate.
 216  *
 217  * It would be nice to have some cleaner way to integrate with coverage
 218  * counters, but with only a few reasons I guess this is good enough for
 219  * now. */
 220 enum revalidate_reason {
 221     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 222     REV_STP,                   /* Spanning tree protocol port status change. */
 223     REV_RSTP,                  /* RSTP port status change. */
 224     REV_BOND,                  /* Bonding changed. */
 225     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 226     REV_FLOW_TABLE,            /* Flow table changed. */
 227     REV_MAC_LEARNING,          /* Mac learning changed. */
 228     REV_MCAST_SNOOPING,        /* Multicast snooping changed. */
 229 };
 230 COVERAGE_DEFINE(rev_reconfigure);
 231 COVERAGE_DEFINE(rev_stp);
 232 COVERAGE_DEFINE(rev_rstp);
 233 COVERAGE_DEFINE(rev_bond);
 234 COVERAGE_DEFINE(rev_port_toggled);
 235 COVERAGE_DEFINE(rev_flow_table);
 236 COVERAGE_DEFINE(rev_mac_learning);
 237 COVERAGE_DEFINE(rev_mcast_snooping);
 238
 239 /* All datapaths of a given type share a single dpif backer instance. */
 240 struct dpif_backer {
 241     char *type;
 242     int refcount;
 243     struct dpif *dpif;
 244     struct udpif *udpif;
 245
 246     struct ovs_rwlock odp_to_ofport_lock;
 247     struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
 248
 249     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 250
 251     enum revalidate_reason need_revalidate; /* Revalidate all flows. */
 252
 253     bool recv_set_enable; /* Enables or disables receiving packets. */
 254
 255     /* Version string of the datapath stored in OVSDB. */
 256     char *dp_version_string;
 257
 258     /* Datapath feature support. */
 259     struct dpif_backer_support support;
 260     struct atomic_count tnl_count;
 261 };
 262
 263 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 264 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 265
 266 struct ofproto_dpif {
 267     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 268     struct ofproto up;
 269     struct dpif_backer *backer;
 270
 271     /* Unique identifier for this instantiation of this bridge in this running
 272      * process.  */
 273     struct uuid uuid;
 274
 275     ATOMIC(cls_version_t) tables_version;  /* For classifier lookups. */
 276
 277     uint64_t dump_seq; /* Last read of udpif_dump_seq(). */
 278
 279     /* Special OpenFlow rules. */
 280     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 281     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 282     struct rule_dpif *drop_frags_rule; /* Used in OFPUTIL_FRAG_DROP mode. */
 283
 284     /* Bridging. */
 285     struct netflow *netflow;
 286     struct dpif_sflow *sflow;
 287     struct dpif_ipfix *ipfix;
 288     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 289     struct mac_learning *ml;
 290     struct mcast_snooping *ms;
 291     bool has_bonded_bundles;
 292     bool lacp_enabled;
 293     struct mbridge *mbridge;
 294
 295     struct ovs_mutex stats_mutex;
 296     struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
 297                                             * consumed in userspace. */
 298
 299     /* Spanning tree. */
 300     struct stp *stp;
 301     long long int stp_last_tick;
 302
 303     /* Rapid Spanning Tree. */
 304     struct rstp *rstp;
 305     long long int rstp_last_tick;
 306
 307     /* Ports. */
 308     struct sset ports;             /* Set of standard port names. */
 309     struct sset ghost_ports;       /* Ports with no datapath port. */
 310     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 311     int port_poll_errno;           /* Last errno for port_poll() reply. */
 312     uint64_t change_seq;           /* Connectivity status changes. */
 313
 314     /* Work queues. */
 315     struct guarded_list ams;      /* Contains "struct ofproto_async_msgs"s. */
 316     struct seq *ams_seq;          /* For notifying 'ams' reception. */
 317     uint64_t ams_seqno;
 318 };
 319
 320 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 321 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 322
 323 static bool ofproto_use_tnl_push_pop = true;
 324 static void ofproto_unixctl_init(void);
 325
 326 static inline struct ofproto_dpif *
 327 ofproto_dpif_cast(const struct ofproto *ofproto)
 328 {
 329     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 330     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 331 }
 332
 333 bool
 334 ofproto_dpif_get_enable_ufid(const struct dpif_backer *backer)
 335 {
 336     return backer->support.ufid;
 337 }
 338
 339 struct dpif_backer_support *
 340 ofproto_dpif_get_support(const struct ofproto_dpif *ofproto)
 341 {
 342     return &ofproto->backer->support;
 343 }
 344
 345 static void ofproto_trace(struct ofproto_dpif *, struct flow *,
 346                           const struct dp_packet *packet,
 347                           const struct ofpact[], size_t ofpacts_len,
 348                           struct ds *);
 349
 350 /* Global variables. */
 351 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 352
 353 /* Initial mappings of port to bridge mappings. */
 354 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 355
 356 /* Executes 'fm'.  The caller retains ownership of 'fm' and everything in
 357  * it. */
 358 void
 359 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 360                       const struct ofputil_flow_mod *fm)
 361 {
 362     struct ofproto_flow_mod ofm;
 363
 364     /* Multiple threads may do this for the same 'fm' at the same time.
 365      * Allocate ofproto_flow_mod with execution context from stack.
 366      *
 367      * Note: This copy could be avoided by making ofproto_flow_mod more
 368      * complex, but that may not be desireable, and a learn action is not that
 369      * fast to begin with. */
 370     ofm.fm = *fm;
 371     ofproto_flow_mod(&ofproto->up, &ofm);
 372 }
 373
 374 /* Appends 'am' to the queue of asynchronous messages to be sent to the
 375  * controller.  Takes ownership of 'am' and any data it points to. */
 376 void
 377 ofproto_dpif_send_async_msg(struct ofproto_dpif *ofproto,
 378                             struct ofproto_async_msg *am)
 379 {
 380     if (!guarded_list_push_back(&ofproto->ams, &am->list_node, 1024)) {
 381         COVERAGE_INC(packet_in_overflow);
 382         ofproto_async_msg_free(am);
 383     }
 384
 385     /* Wakes up main thread for packet-in I/O. */
 386     seq_change(ofproto->ams_seq);
 387 }
 388
 389 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
 390  * packet rather than to send the packet to the controller.
 391  *
 392  * This function returns false to indicate that a packet_in message
 393  * for a "table-miss" should be sent to at least one controller.
 394  * False otherwise. */
 395 bool
 396 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif *ofproto)
 397 {
 398     return connmgr_wants_packet_in_on_miss(ofproto->up.connmgr);
 399 }
 400 \f
 401 /* Factory functions. */
 402
 403 static void
 404 init(const struct shash *iface_hints)
 405 {
 406     struct shash_node *node;
 407
 408     /* Make a local copy, since we don't own 'iface_hints' elements. */
 409     SHASH_FOR_EACH(node, iface_hints) {
 410         const struct iface_hint *orig_hint = node->data;
 411         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 412
 413         new_hint->br_name = xstrdup(orig_hint->br_name);
 414         new_hint->br_type = xstrdup(orig_hint->br_type);
 415         new_hint->ofp_port = orig_hint->ofp_port;
 416
 417         shash_add(&init_ofp_ports, node->name, new_hint);
 418     }
 419
 420     ofproto_unixctl_init();
 421     udpif_init();
 422 }
 423
 424 static void
 425 enumerate_types(struct sset *types)
 426 {
 427     dp_enumerate_types(types);
 428 }
 429
 430 static int
 431 enumerate_names(const char *type, struct sset *names)
 432 {
 433     struct ofproto_dpif *ofproto;
 434
 435     sset_clear(names);
 436     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 437         if (strcmp(type, ofproto->up.type)) {
 438             continue;
 439         }
 440         sset_add(names, ofproto->up.name);
 441     }
 442
 443     return 0;
 444 }
 445
 446 static int
 447 del(const char *type, const char *name)
 448 {
 449     struct dpif *dpif;
 450     int error;
 451
 452     error = dpif_open(name, type, &dpif);
 453     if (!error) {
 454         error = dpif_delete(dpif);
 455         dpif_close(dpif);
 456     }
 457     return error;
 458 }
 459 \f
 460 static const char *
 461 port_open_type(const char *datapath_type, const char *port_type)
 462 {
 463     return dpif_port_open_type(datapath_type, port_type);
 464 }
 465
 466 /* Type functions. */
 467
 468 static void process_dpif_port_changes(struct dpif_backer *);
 469 static void process_dpif_all_ports_changed(struct dpif_backer *);
 470 static void process_dpif_port_change(struct dpif_backer *,
 471                                      const char *devname);
 472 static void process_dpif_port_error(struct dpif_backer *, int error);
 473
 474 static struct ofproto_dpif *
 475 lookup_ofproto_dpif_by_port_name(const char *name)
 476 {
 477     struct ofproto_dpif *ofproto;
 478
 479     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 480         if (sset_contains(&ofproto->ports, name)) {
 481             return ofproto;
 482         }
 483     }
 484
 485     return NULL;
 486 }
 487
 488 bool
 489 ofproto_dpif_backer_enabled(struct dpif_backer* backer)
 490 {
 491     return backer->recv_set_enable;
 492 }
 493
 494 static int
 495 type_run(const char *type)
 496 {
 497     struct dpif_backer *backer;
 498
 499     backer = shash_find_data(&all_dpif_backers, type);
 500     if (!backer) {
 501         /* This is not necessarily a problem, since backers are only
 502          * created on demand. */
 503         return 0;
 504     }
 505
 506     /* This must be called before dpif_run() */
 507     dpif_poll_threads_set(backer->dpif, pmd_cpu_mask);
 508
 509     if (dpif_run(backer->dpif)) {
 510         backer->need_revalidate = REV_RECONFIGURE;
 511     }
 512
 513     udpif_run(backer->udpif);
 514
 515     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 516      * and the configuration has now changed to "false", enable receiving
 517      * packets from the datapath. */
 518     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 519         int error;
 520
 521         backer->recv_set_enable = true;
 522
 523         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 524         if (error) {
 525             VLOG_ERR("Failed to enable receiving packets in dpif.");
 526             return error;
 527         }
 528         dpif_flow_flush(backer->dpif);
 529         backer->need_revalidate = REV_RECONFIGURE;
 530     }
 531
 532     if (backer->recv_set_enable) {
 533         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 534     }
 535
 536     if (backer->need_revalidate) {
 537         struct ofproto_dpif *ofproto;
 538         struct simap_node *node;
 539         struct simap tmp_backers;
 540
 541         /* Handle tunnel garbage collection. */
 542         simap_init(&tmp_backers);
 543         simap_swap(&backer->tnl_backers, &tmp_backers);
 544
 545         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 546             struct ofport_dpif *iter;
 547
 548             if (backer != ofproto->backer) {
 549                 continue;
 550             }
 551
 552             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 553                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 554                 const char *dp_port;
 555
 556                 if (!iter->is_tunnel) {
 557                     continue;
 558                 }
 559
 560                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 561                                                      namebuf, sizeof namebuf);
 562                 node = simap_find(&tmp_backers, dp_port);
 563                 if (node) {
 564                     simap_put(&backer->tnl_backers, dp_port, node->data);
 565                     simap_delete(&tmp_backers, node);
 566                     node = simap_find(&backer->tnl_backers, dp_port);
 567                 } else {
 568                     node = simap_find(&backer->tnl_backers, dp_port);
 569                     if (!node) {
 570                         odp_port_t odp_port = ODPP_NONE;
 571
 572                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 573                                            &odp_port)) {
 574                             simap_put(&backer->tnl_backers, dp_port,
 575                                       odp_to_u32(odp_port));
 576                             node = simap_find(&backer->tnl_backers, dp_port);
 577                         }
 578                     }
 579                 }
 580
 581                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 582                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 583                                          iter->odp_port,
 584                                          ovs_native_tunneling_is_on(ofproto), dp_port)) {
 585                     backer->need_revalidate = REV_RECONFIGURE;
 586                 }
 587             }
 588         }
 589
 590         SIMAP_FOR_EACH (node, &tmp_backers) {
 591             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 592         }
 593         simap_destroy(&tmp_backers);
 594
 595         switch (backer->need_revalidate) {
 596         case REV_RECONFIGURE:    COVERAGE_INC(rev_reconfigure);    break;
 597         case REV_STP:            COVERAGE_INC(rev_stp);            break;
 598         case REV_RSTP:           COVERAGE_INC(rev_rstp);           break;
 599         case REV_BOND:           COVERAGE_INC(rev_bond);           break;
 600         case REV_PORT_TOGGLED:   COVERAGE_INC(rev_port_toggled);   break;
 601         case REV_FLOW_TABLE:     COVERAGE_INC(rev_flow_table);     break;
 602         case REV_MAC_LEARNING:   COVERAGE_INC(rev_mac_learning);   break;
 603         case REV_MCAST_SNOOPING: COVERAGE_INC(rev_mcast_snooping); break;
 604         }
 605         backer->need_revalidate = 0;
 606
 607         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 608             struct ofport_dpif *ofport;
 609             struct ofbundle *bundle;
 610
 611             if (ofproto->backer != backer) {
 612                 continue;
 613             }
 614
 615             xlate_txn_start();
 616             xlate_ofproto_set(ofproto, ofproto->up.name,
 617                               ofproto->backer->dpif, ofproto->ml,
 618                               ofproto->stp, ofproto->rstp, ofproto->ms,
 619                               ofproto->mbridge, ofproto->sflow, ofproto->ipfix,
 620                               ofproto->netflow,
 621                               ofproto->up.forward_bpdu,
 622                               connmgr_has_in_band(ofproto->up.connmgr),
 623                               &ofproto->backer->support);
 624
 625             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 626                 xlate_bundle_set(ofproto, bundle, bundle->name,
 627                                  bundle->vlan_mode, bundle->vlan,
 628                                  bundle->trunks, bundle->use_priority_tags,
 629                                  bundle->bond, bundle->lacp,
 630                                  bundle->floodable);
 631             }
 632
 633             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 634                 int stp_port = ofport->stp_port
 635                     ? stp_port_no(ofport->stp_port)
 636                     : -1;
 637                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 638                                  ofport->up.ofp_port, ofport->odp_port,
 639                                  ofport->up.netdev, ofport->cfm, ofport->bfd,
 640                                  ofport->lldp, ofport->peer, stp_port,
 641                                  ofport->rstp_port, ofport->qdscp,
 642                                  ofport->n_qdscp, ofport->up.pp.config,
 643                                  ofport->up.pp.state, ofport->is_tunnel,
 644                                  ofport->may_enable);
 645             }
 646             xlate_txn_commit();
 647         }
 648
 649         udpif_revalidate(backer->udpif);
 650     }
 651
 652     process_dpif_port_changes(backer);
 653
 654     return 0;
 655 }
 656
 657 /* Check for and handle port changes in 'backer''s dpif. */
 658 static void
 659 process_dpif_port_changes(struct dpif_backer *backer)
 660 {
 661     for (;;) {
 662         char *devname;
 663         int error;
 664
 665         error = dpif_port_poll(backer->dpif, &devname);
 666         switch (error) {
 667         case EAGAIN:
 668             return;
 669
 670         case ENOBUFS:
 671             process_dpif_all_ports_changed(backer);
 672             break;
 673
 674         case 0:
 675             process_dpif_port_change(backer, devname);
 676             free(devname);
 677             break;
 678
 679         default:
 680             process_dpif_port_error(backer, error);
 681             break;
 682         }
 683     }
 684 }
 685
 686 static void
 687 process_dpif_all_ports_changed(struct dpif_backer *backer)
 688 {
 689     struct ofproto_dpif *ofproto;
 690     struct dpif_port dpif_port;
 691     struct dpif_port_dump dump;
 692     struct sset devnames;
 693     const char *devname;
 694
 695     sset_init(&devnames);
 696     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 697         if (ofproto->backer == backer) {
 698             struct ofport *ofport;
 699
 700             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 701                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 702             }
 703         }
 704     }
 705     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 706         sset_add(&devnames, dpif_port.name);
 707     }
 708
 709     SSET_FOR_EACH (devname, &devnames) {
 710         process_dpif_port_change(backer, devname);
 711     }
 712     sset_destroy(&devnames);
 713 }
 714
 715 static void
 716 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 717 {
 718     struct ofproto_dpif *ofproto;
 719     struct dpif_port port;
 720
 721     /* Don't report on the datapath's device. */
 722     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 723         return;
 724     }
 725
 726     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 727                    &all_ofproto_dpifs) {
 728         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 729             return;
 730         }
 731     }
 732
 733     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 734     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 735         /* The port was removed.  If we know the datapath,
 736          * report it through poll_set().  If we don't, it may be
 737          * notifying us of a removal we initiated, so ignore it.
 738          * If there's a pending ENOBUFS, let it stand, since
 739          * everything will be reevaluated. */
 740         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 741             sset_add(&ofproto->port_poll_set, devname);
 742             ofproto->port_poll_errno = 0;
 743         }
 744     } else if (!ofproto) {
 745         /* The port was added, but we don't know with which
 746          * ofproto we should associate it.  Delete it. */
 747         dpif_port_del(backer->dpif, port.port_no);
 748     } else {
 749         struct ofport_dpif *ofport;
 750
 751         ofport = ofport_dpif_cast(shash_find_data(
 752                                       &ofproto->up.port_by_name, devname));
 753         if (ofport
 754             && ofport->odp_port != port.port_no
 755             && !odp_port_to_ofport(backer, port.port_no))
 756         {
 757             /* 'ofport''s datapath port number has changed from
 758              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 759              * structures to match. */
 760             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 761             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 762             ofport->odp_port = port.port_no;
 763             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 764                         hash_odp_port(port.port_no));
 765             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 766             backer->need_revalidate = REV_RECONFIGURE;
 767         }
 768     }
 769     dpif_port_destroy(&port);
 770 }
 771
 772 /* Propagate 'error' to all ofprotos based on 'backer'. */
 773 static void
 774 process_dpif_port_error(struct dpif_backer *backer, int error)
 775 {
 776     struct ofproto_dpif *ofproto;
 777
 778     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 779         if (ofproto->backer == backer) {
 780             sset_clear(&ofproto->port_poll_set);
 781             ofproto->port_poll_errno = error;
 782         }
 783     }
 784 }
 785
 786 static void
 787 type_wait(const char *type)
 788 {
 789     struct dpif_backer *backer;
 790
 791     backer = shash_find_data(&all_dpif_backers, type);
 792     if (!backer) {
 793         /* This is not necessarily a problem, since backers are only
 794          * created on demand. */
 795         return;
 796     }
 797
 798     dpif_wait(backer->dpif);
 799 }
 800 \f
 801 /* Basic life-cycle. */
 802
 803 static int add_internal_flows(struct ofproto_dpif *);
 804
 805 static struct ofproto *
 806 alloc(void)
 807 {
 808     struct ofproto_dpif *ofproto = xzalloc(sizeof *ofproto);
 809     return &ofproto->up;
 810 }
 811
 812 static void
 813 dealloc(struct ofproto *ofproto_)
 814 {
 815     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
 816     free(ofproto);
 817 }
 818
 819 static void
 820 close_dpif_backer(struct dpif_backer *backer)
 821 {
 822     ovs_assert(backer->refcount > 0);
 823
 824     if (--backer->refcount) {
 825         return;
 826     }
 827
 828     udpif_destroy(backer->udpif);
 829
 830     simap_destroy(&backer->tnl_backers);
 831     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
 832     hmap_destroy(&backer->odp_to_ofport_map);
 833     shash_find_and_delete(&all_dpif_backers, backer->type);
 834     free(backer->type);
 835     free(backer->dp_version_string);
 836     dpif_close(backer->dpif);
 837     free(backer);
 838 }
 839
 840 /* Datapath port slated for removal from datapath. */
 841 struct odp_garbage {
 842     struct ovs_list list_node;
 843     odp_port_t odp_port;
 844 };
 845
 846 static bool check_variable_length_userdata(struct dpif_backer *backer);
 847 static void check_support(struct dpif_backer *backer);
 848
 849 static int
 850 open_dpif_backer(const char *type, struct dpif_backer **backerp)
 851 {
 852     struct dpif_backer *backer;
 853     struct dpif_port_dump port_dump;
 854     struct dpif_port port;
 855     struct shash_node *node;
 856     struct ovs_list garbage_list;
 857     struct odp_garbage *garbage;
 858
 859     struct sset names;
 860     char *backer_name;
 861     const char *name;
 862     int error;
 863
 864     backer = shash_find_data(&all_dpif_backers, type);
 865     if (backer) {
 866         backer->refcount++;
 867         *backerp = backer;
 868         return 0;
 869     }
 870
 871     backer_name = xasprintf("ovs-%s", type);
 872
 873     /* Remove any existing datapaths, since we assume we're the only
 874      * userspace controlling the datapath. */
 875     sset_init(&names);
 876     dp_enumerate_names(type, &names);
 877     SSET_FOR_EACH(name, &names) {
 878         struct dpif *old_dpif;
 879
 880         /* Don't remove our backer if it exists. */
 881         if (!strcmp(name, backer_name)) {
 882             continue;
 883         }
 884
 885         if (dpif_open(name, type, &old_dpif)) {
 886             VLOG_WARN("couldn't open old datapath %s to remove it", name);
 887         } else {
 888             dpif_delete(old_dpif);
 889             dpif_close(old_dpif);
 890         }
 891     }
 892     sset_destroy(&names);
 893
 894     backer = xmalloc(sizeof *backer);
 895
 896     error = dpif_create_and_open(backer_name, type, &backer->dpif);
 897     free(backer_name);
 898     if (error) {
 899         VLOG_ERR("failed to open datapath of type %s: %s", type,
 900                  ovs_strerror(error));
 901         free(backer);
 902         return error;
 903     }
 904     backer->udpif = udpif_create(backer, backer->dpif);
 905
 906     backer->type = xstrdup(type);
 907     backer->refcount = 1;
 908     hmap_init(&backer->odp_to_ofport_map);
 909     ovs_rwlock_init(&backer->odp_to_ofport_lock);
 910     backer->need_revalidate = 0;
 911     simap_init(&backer->tnl_backers);
 912     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
 913     *backerp = backer;
 914
 915     if (backer->recv_set_enable) {
 916         dpif_flow_flush(backer->dpif);
 917     }
 918
 919     /* Loop through the ports already on the datapath and remove any
 920      * that we don't need anymore. */
 921     ovs_list_init(&garbage_list);
 922     dpif_port_dump_start(&port_dump, backer->dpif);
 923     while (dpif_port_dump_next(&port_dump, &port)) {
 924         node = shash_find(&init_ofp_ports, port.name);
 925         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
 926             garbage = xmalloc(sizeof *garbage);
 927             garbage->odp_port = port.port_no;
 928             ovs_list_push_front(&garbage_list, &garbage->list_node);
 929         }
 930     }
 931     dpif_port_dump_done(&port_dump);
 932
 933     LIST_FOR_EACH_POP (garbage, list_node, &garbage_list) {
 934         dpif_port_del(backer->dpif, garbage->odp_port);
 935         free(garbage);
 936     }
 937
 938     shash_add(&all_dpif_backers, type, backer);
 939
 940     check_support(backer);
 941     atomic_count_init(&backer->tnl_count, 0);
 942
 943     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 944     if (error) {
 945         VLOG_ERR("failed to listen on datapath of type %s: %s",
 946                  type, ovs_strerror(error));
 947         close_dpif_backer(backer);
 948         return error;
 949     }
 950
 951     if (backer->recv_set_enable) {
 952         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 953     }
 954
 955     /* This check fails if performed before udpif threads have been set,
 956      * as the kernel module checks that the 'pid' in userspace action
 957      * is non-zero. */
 958     backer->support.variable_length_userdata
 959         = check_variable_length_userdata(backer);
 960     backer->dp_version_string = dpif_get_dp_version(backer->dpif);
 961
 962     return error;
 963 }
 964
 965 bool
 966 ovs_native_tunneling_is_on(struct ofproto_dpif *ofproto)
 967 {
 968     return ofproto_use_tnl_push_pop && ofproto->backer->support.tnl_push_pop &&
 969            atomic_count_get(&ofproto->backer->tnl_count);
 970 }
 971
 972 /* Tests whether 'backer''s datapath supports recirculation.  Only newer
 973  * datapaths support OVS_KEY_ATTR_RECIRC_ID in keys.  We need to disable some
 974  * features on older datapaths that don't support this feature.
 975  *
 976  * Returns false if 'backer' definitely does not support recirculation, true if
 977  * it seems to support recirculation or if at least the error we get is
 978  * ambiguous. */
 979 static bool
 980 check_recirc(struct dpif_backer *backer)
 981 {
 982     struct flow flow;
 983     struct odputil_keybuf keybuf;
 984     struct ofpbuf key;
 985     bool enable_recirc;
 986     struct odp_flow_key_parms odp_parms = {
 987         .flow = &flow,
 988         .support = {
 989             .recirc = true,
 990         },
 991     };
 992
 993     memset(&flow, 0, sizeof flow);
 994     flow.recirc_id = 1;
 995     flow.dp_hash = 1;
 996
 997     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
 998     odp_flow_key_from_flow(&odp_parms, &key);
 999     enable_recirc = dpif_probe_feature(backer->dpif, "recirculation", &key,
1000                                        NULL);
1001
1002     if (enable_recirc) {
1003         VLOG_INFO("%s: Datapath supports recirculation",
1004                   dpif_name(backer->dpif));
1005     } else {
1006         VLOG_INFO("%s: Datapath does not support recirculation",
1007                   dpif_name(backer->dpif));
1008     }
1009
1010     return enable_recirc;
1011 }
1012
1013 /* Tests whether 'dpif' supports unique flow ids. We can skip serializing
1014  * some flow attributes for datapaths that support this feature.
1015  *
1016  * Returns true if 'dpif' supports UFID for flow operations.
1017  * Returns false if  'dpif' does not support UFID. */
1018 static bool
1019 check_ufid(struct dpif_backer *backer)
1020 {
1021     struct flow flow;
1022     struct odputil_keybuf keybuf;
1023     struct ofpbuf key;
1024     ovs_u128 ufid;
1025     bool enable_ufid;
1026     struct odp_flow_key_parms odp_parms = {
1027         .flow = &flow,
1028     };
1029
1030     memset(&flow, 0, sizeof flow);
1031     flow.dl_type = htons(0x1234);
1032
1033     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1034     odp_flow_key_from_flow(&odp_parms, &key);
1035     dpif_flow_hash(backer->dpif, key.data, key.size, &ufid);
1036
1037     enable_ufid = dpif_probe_feature(backer->dpif, "UFID", &key, &ufid);
1038
1039     if (enable_ufid) {
1040         VLOG_INFO("%s: Datapath supports unique flow ids",
1041                   dpif_name(backer->dpif));
1042     } else {
1043         VLOG_INFO("%s: Datapath does not support unique flow ids",
1044                   dpif_name(backer->dpif));
1045     }
1046     return enable_ufid;
1047 }
1048
1049 /* Tests whether 'backer''s datapath supports variable-length
1050  * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.  We need
1051  * to disable some features on older datapaths that don't support this
1052  * feature.
1053  *
1054  * Returns false if 'backer' definitely does not support variable-length
1055  * userdata, true if it seems to support them or if at least the error we get
1056  * is ambiguous. */
1057 static bool
1058 check_variable_length_userdata(struct dpif_backer *backer)
1059 {
1060     struct eth_header *eth;
1061     struct ofpbuf actions;
1062     struct dpif_execute execute;
1063     struct dp_packet packet;
1064     struct flow flow;
1065     size_t start;
1066     int error;
1067
1068     /* Compose a userspace action that will cause an ERANGE error on older
1069      * datapaths that don't support variable-length userdata.
1070      *
1071      * We really test for using userdata longer than 8 bytes, but older
1072      * datapaths accepted these, silently truncating the userdata to 8 bytes.
1073      * The same older datapaths rejected userdata shorter than 8 bytes, so we
1074      * test for that instead as a proxy for longer userdata support. */
1075     ofpbuf_init(&actions, 64);
1076     start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
1077     nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
1078                    dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
1079     nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
1080     nl_msg_end_nested(&actions, start);
1081
1082     /* Compose a dummy ethernet packet. */
1083     dp_packet_init(&packet, ETH_HEADER_LEN);
1084     eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1085     eth->eth_type = htons(0x1234);
1086
1087     flow_extract(&packet, &flow);
1088
1089     /* Execute the actions.  On older datapaths this fails with ERANGE, on
1090      * newer datapaths it succeeds. */
1091     execute.actions = actions.data;
1092     execute.actions_len = actions.size;
1093     execute.packet = &packet;
1094     execute.flow = &flow;
1095     execute.needs_help = false;
1096     execute.probe = true;
1097     execute.mtu = 0;
1098
1099     error = dpif_execute(backer->dpif, &execute);
1100
1101     dp_packet_uninit(&packet);
1102     ofpbuf_uninit(&actions);
1103
1104     switch (error) {
1105     case 0:
1106         return true;
1107
1108     case ERANGE:
1109         /* Variable-length userdata is not supported. */
1110         VLOG_WARN("%s: datapath does not support variable-length userdata "
1111                   "feature (needs Linux 3.10+ or kernel module from OVS "
1112                   "1..11+).  The NXAST_SAMPLE action will be ignored.",
1113                   dpif_name(backer->dpif));
1114         return false;
1115
1116     default:
1117         /* Something odd happened.  We're not sure whether variable-length
1118          * userdata is supported.  Default to "yes". */
1119         VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1120                   dpif_name(backer->dpif), ovs_strerror(error));
1121         return true;
1122     }
1123 }
1124
1125 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1126  *
1127  * Returns the number of elements in a struct flow's mpls_lse field
1128  * if the datapath supports at least that many entries in an
1129  * MPLS label stack.
1130  * Otherwise returns the number of MPLS push actions supported by
1131  * the datapath. */
1132 static size_t
1133 check_max_mpls_depth(struct dpif_backer *backer)
1134 {
1135     struct flow flow;
1136     int n;
1137
1138     for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
1139         struct odputil_keybuf keybuf;
1140         struct ofpbuf key;
1141         struct odp_flow_key_parms odp_parms = {
1142             .flow = &flow,
1143         };
1144
1145         memset(&flow, 0, sizeof flow);
1146         flow.dl_type = htons(ETH_TYPE_MPLS);
1147         flow_set_mpls_bos(&flow, n, 1);
1148
1149         ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1150         odp_flow_key_from_flow(&odp_parms, &key);
1151         if (!dpif_probe_feature(backer->dpif, "MPLS", &key, NULL)) {
1152             break;
1153         }
1154     }
1155
1156     VLOG_INFO("%s: MPLS label stack length probed as %d",
1157               dpif_name(backer->dpif), n);
1158     return n;
1159 }
1160
1161 /* Tests whether 'backer''s datapath supports masked data in
1162  * OVS_ACTION_ATTR_SET actions.  We need to disable some features on older
1163  * datapaths that don't support this feature. */
1164 static bool
1165 check_masked_set_action(struct dpif_backer *backer)
1166 {
1167     struct eth_header *eth;
1168     struct ofpbuf actions;
1169     struct dpif_execute execute;
1170     struct dp_packet packet;
1171     struct flow flow;
1172     int error;
1173     struct ovs_key_ethernet key, mask;
1174
1175     /* Compose a set action that will cause an EINVAL error on older
1176      * datapaths that don't support masked set actions.
1177      * Avoid using a full mask, as it could be translated to a non-masked
1178      * set action instead. */
1179     ofpbuf_init(&actions, 64);
1180     memset(&key, 0x53, sizeof key);
1181     memset(&mask, 0x7f, sizeof mask);
1182     commit_masked_set_action(&actions, OVS_KEY_ATTR_ETHERNET, &key, &mask,
1183                              sizeof key);
1184
1185     /* Compose a dummy ethernet packet. */
1186     dp_packet_init(&packet, ETH_HEADER_LEN);
1187     eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1188     eth->eth_type = htons(0x1234);
1189
1190     flow_extract(&packet, &flow);
1191
1192     /* Execute the actions.  On older datapaths this fails with EINVAL, on
1193      * newer datapaths it succeeds. */
1194     execute.actions = actions.data;
1195     execute.actions_len = actions.size;
1196     execute.packet = &packet;
1197     execute.flow = &flow;
1198     execute.needs_help = false;
1199     execute.probe = true;
1200     execute.mtu = 0;
1201
1202     error = dpif_execute(backer->dpif, &execute);
1203
1204     dp_packet_uninit(&packet);
1205     ofpbuf_uninit(&actions);
1206
1207     if (error) {
1208         /* Masked set action is not supported. */
1209         VLOG_INFO("%s: datapath does not support masked set action feature.",
1210                   dpif_name(backer->dpif));
1211     }
1212     return !error;
1213 }
1214
1215 #define CHECK_FEATURE__(NAME, SUPPORT, FIELD, VALUE)                        \
1216 static bool                                                                 \
1217 check_##NAME(struct dpif_backer *backer)                                    \
1218 {                                                                           \
1219     struct flow flow;                                                       \
1220     struct odputil_keybuf keybuf;                                           \
1221     struct ofpbuf key;                                                      \
1222     bool enable;                                                            \
1223     struct odp_flow_key_parms odp_parms = {                                 \
1224         .flow = &flow,                                                      \
1225         .support = {                                                        \
1226             .SUPPORT = true,                                                \
1227         },                                                                  \
1228     };                                                                      \
1229                                                                             \
1230     memset(&flow, 0, sizeof flow);                                          \
1231     flow.FIELD = VALUE;                                                     \
1232                                                                             \
1233     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);                         \
1234     odp_flow_key_from_flow(&odp_parms, &key);                               \
1235     enable = dpif_probe_feature(backer->dpif, #NAME, &key, NULL);           \
1236                                                                             \
1237     if (enable) {                                                           \
1238         VLOG_INFO("%s: Datapath supports "#NAME, dpif_name(backer->dpif));  \
1239     } else {                                                                \
1240         VLOG_INFO("%s: Datapath does not support "#NAME,                    \
1241                   dpif_name(backer->dpif));                                 \
1242     }                                                                       \
1243                                                                             \
1244     return enable;                                                          \
1245 }
1246 #define CHECK_FEATURE(FIELD) CHECK_FEATURE__(FIELD, FIELD, FIELD, 1)
1247
1248 CHECK_FEATURE(ct_state)
1249 CHECK_FEATURE(ct_zone)
1250 CHECK_FEATURE(ct_mark)
1251 CHECK_FEATURE__(ct_label, ct_label, ct_label.u64.lo, 1)
1252 CHECK_FEATURE__(ct_state_nat, ct_state, ct_state, CS_TRACKED|CS_SRC_NAT)
1253
1254 #undef CHECK_FEATURE
1255 #undef CHECK_FEATURE__
1256
1257 static void
1258 check_support(struct dpif_backer *backer)
1259 {
1260     /* This feature needs to be tested after udpif threads are set. */
1261     backer->support.variable_length_userdata = false;
1262
1263     backer->support.odp.recirc = check_recirc(backer);
1264     backer->support.odp.max_mpls_depth = check_max_mpls_depth(backer);
1265     backer->support.masked_set_action = check_masked_set_action(backer);
1266     backer->support.ufid = check_ufid(backer);
1267     backer->support.tnl_push_pop = dpif_supports_tnl_push_pop(backer->dpif);
1268
1269     backer->support.odp.ct_state = check_ct_state(backer);
1270     backer->support.odp.ct_zone = check_ct_zone(backer);
1271     backer->support.odp.ct_mark = check_ct_mark(backer);
1272     backer->support.odp.ct_label = check_ct_label(backer);
1273
1274     backer->support.odp.ct_state_nat = check_ct_state_nat(backer);
1275 }
1276
1277 static int
1278 construct(struct ofproto *ofproto_)
1279 {
1280     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1281     struct shash_node *node, *next;
1282     int error;
1283
1284     /* Tunnel module can get used right after the udpif threads are running. */
1285     ofproto_tunnel_init();
1286
1287     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1288     if (error) {
1289         return error;
1290     }
1291
1292     uuid_generate(&ofproto->uuid);
1293     atomic_init(&ofproto->tables_version, CLS_MIN_VERSION);
1294     ofproto->netflow = NULL;
1295     ofproto->sflow = NULL;
1296     ofproto->ipfix = NULL;
1297     ofproto->stp = NULL;
1298     ofproto->rstp = NULL;
1299     ofproto->dump_seq = 0;
1300     hmap_init(&ofproto->bundles);
1301     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1302     ofproto->ms = NULL;
1303     ofproto->mbridge = mbridge_create();
1304     ofproto->has_bonded_bundles = false;
1305     ofproto->lacp_enabled = false;
1306     ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1307
1308     guarded_list_init(&ofproto->ams);
1309
1310     sset_init(&ofproto->ports);
1311     sset_init(&ofproto->ghost_ports);
1312     sset_init(&ofproto->port_poll_set);
1313     ofproto->port_poll_errno = 0;
1314     ofproto->change_seq = 0;
1315     ofproto->ams_seq = seq_create();
1316     ofproto->ams_seqno = seq_read(ofproto->ams_seq);
1317
1318
1319     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1320         struct iface_hint *iface_hint = node->data;
1321
1322         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1323             /* Check if the datapath already has this port. */
1324             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1325                 sset_add(&ofproto->ports, node->name);
1326             }
1327
1328             free(iface_hint->br_name);
1329             free(iface_hint->br_type);
1330             free(iface_hint);
1331             shash_delete(&init_ofp_ports, node);
1332         }
1333     }
1334
1335     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1336                 hash_string(ofproto->up.name, 0));
1337     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1338
1339     ofproto_init_tables(ofproto_, N_TABLES);
1340     error = add_internal_flows(ofproto);
1341
1342     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1343
1344     return error;
1345 }
1346
1347 static int
1348 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1349                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1350 {
1351     struct match match;
1352     int error;
1353     struct rule *rule;
1354
1355     match_init_catchall(&match);
1356     match_set_reg(&match, 0, id);
1357
1358     error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, 0, ofpacts,
1359                                            &rule);
1360     *rulep = error ? NULL : rule_dpif_cast(rule);
1361
1362     return error;
1363 }
1364
1365 static int
1366 add_internal_flows(struct ofproto_dpif *ofproto)
1367 {
1368     struct ofpact_controller *controller;
1369     uint64_t ofpacts_stub[128 / 8];
1370     struct ofpbuf ofpacts;
1371     struct rule *unused_rulep OVS_UNUSED;
1372     struct match match;
1373     int error;
1374     int id;
1375
1376     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1377     id = 1;
1378
1379     controller = ofpact_put_CONTROLLER(&ofpacts);
1380     controller->max_len = UINT16_MAX;
1381     controller->controller_id = 0;
1382     controller->reason = OFPR_IMPLICIT_MISS;
1383     ofpact_finish_CONTROLLER(&ofpacts, &controller);
1384
1385     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1386                                    &ofproto->miss_rule);
1387     if (error) {
1388         return error;
1389     }
1390
1391     ofpbuf_clear(&ofpacts);
1392     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1393                                    &ofproto->no_packet_in_rule);
1394     if (error) {
1395         return error;
1396     }
1397
1398     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1399                                    &ofproto->drop_frags_rule);
1400     if (error) {
1401         return error;
1402     }
1403
1404     /* Drop any run away non-recirc rule lookups. Recirc_id has to be
1405      * zero when reaching this rule.
1406      *
1407      * (priority=2), recirc_id=0, actions=drop
1408      */
1409     ofpbuf_clear(&ofpacts);
1410     match_init_catchall(&match);
1411     match_set_recirc_id(&match, 0);
1412     error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, 0, &ofpacts,
1413                                            &unused_rulep);
1414     return error;
1415 }
1416
1417 static void
1418 destruct(struct ofproto *ofproto_)
1419 {
1420     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1421     struct ofproto_async_msg *am;
1422     struct rule_dpif *rule;
1423     struct oftable *table;
1424     struct ovs_list ams;
1425
1426     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1427     xlate_txn_start();
1428     xlate_remove_ofproto(ofproto);
1429     xlate_txn_commit();
1430
1431     /* Ensure that the upcall processing threads have no remaining references
1432      * to the ofproto or anything in it. */
1433     udpif_synchronize(ofproto->backer->udpif);
1434
1435     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1436
1437     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1438         CLS_FOR_EACH (rule, up.cr, &table->cls) {
1439             ofproto_rule_delete(&ofproto->up, &rule->up);
1440         }
1441     }
1442     ofproto_group_delete_all(&ofproto->up);
1443
1444     guarded_list_pop_all(&ofproto->ams, &ams);
1445     LIST_FOR_EACH_POP (am, list_node, &ams) {
1446         ofproto_async_msg_free(am);
1447     }
1448     guarded_list_destroy(&ofproto->ams);
1449
1450     recirc_free_ofproto(ofproto, ofproto->up.name);
1451
1452     mbridge_unref(ofproto->mbridge);
1453
1454     netflow_unref(ofproto->netflow);
1455     dpif_sflow_unref(ofproto->sflow);
1456     dpif_ipfix_unref(ofproto->ipfix);
1457     hmap_destroy(&ofproto->bundles);
1458     mac_learning_unref(ofproto->ml);
1459     mcast_snooping_unref(ofproto->ms);
1460
1461     sset_destroy(&ofproto->ports);
1462     sset_destroy(&ofproto->ghost_ports);
1463     sset_destroy(&ofproto->port_poll_set);
1464
1465     ovs_mutex_destroy(&ofproto->stats_mutex);
1466
1467     seq_destroy(ofproto->ams_seq);
1468
1469     close_dpif_backer(ofproto->backer);
1470 }
1471
1472 static int
1473 run(struct ofproto *ofproto_)
1474 {
1475     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1476     uint64_t new_seq, new_dump_seq;
1477
1478     if (mbridge_need_revalidate(ofproto->mbridge)) {
1479         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1480         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1481         mac_learning_flush(ofproto->ml);
1482         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1483         mcast_snooping_mdb_flush(ofproto->ms);
1484     }
1485
1486     /* Always updates the ofproto->ams_seqno to avoid frequent wakeup during
1487      * flow restore.  Even though nothing is processed during flow restore,
1488      * all queued 'ams' will be handled immediately when flow restore
1489      * completes. */
1490     ofproto->ams_seqno = seq_read(ofproto->ams_seq);
1491
1492     /* Do not perform any periodic activity required by 'ofproto' while
1493      * waiting for flow restore to complete. */
1494     if (!ofproto_get_flow_restore_wait()) {
1495         struct ofproto_async_msg *am;
1496         struct ovs_list ams;
1497
1498         guarded_list_pop_all(&ofproto->ams, &ams);
1499         LIST_FOR_EACH_POP (am, list_node, &ams) {
1500             connmgr_send_async_msg(ofproto->up.connmgr, am);
1501             ofproto_async_msg_free(am);
1502         }
1503     }
1504
1505     if (ofproto->netflow) {
1506         netflow_run(ofproto->netflow);
1507     }
1508     if (ofproto->sflow) {
1509         dpif_sflow_run(ofproto->sflow);
1510     }
1511     if (ofproto->ipfix) {
1512         dpif_ipfix_run(ofproto->ipfix);
1513     }
1514
1515     new_seq = seq_read(connectivity_seq_get());
1516     if (ofproto->change_seq != new_seq) {
1517         struct ofport_dpif *ofport;
1518
1519         HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1520             port_run(ofport);
1521         }
1522
1523         ofproto->change_seq = new_seq;
1524     }
1525     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1526         struct ofbundle *bundle;
1527
1528         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1529             bundle_run(bundle);
1530         }
1531     }
1532
1533     stp_run(ofproto);
1534     rstp_run(ofproto);
1535     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1536     if (mac_learning_run(ofproto->ml)) {
1537         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1538     }
1539     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1540
1541     if (mcast_snooping_run(ofproto->ms)) {
1542         ofproto->backer->need_revalidate = REV_MCAST_SNOOPING;
1543     }
1544
1545     new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1546     if (ofproto->dump_seq != new_dump_seq) {
1547         struct rule *rule, *next_rule;
1548         long long now = time_msec();
1549
1550         /* We know stats are relatively fresh, so now is a good time to do some
1551          * periodic work. */
1552         ofproto->dump_seq = new_dump_seq;
1553
1554         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1555          * has passed. */
1556         ovs_mutex_lock(&ofproto_mutex);
1557         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1558                             &ofproto->up.expirable) {
1559             rule_expire(rule_dpif_cast(rule), now);
1560         }
1561         ovs_mutex_unlock(&ofproto_mutex);
1562
1563         /* All outstanding data in existing flows has been accounted, so it's a
1564          * good time to do bond rebalancing. */
1565         if (ofproto->has_bonded_bundles) {
1566             struct ofbundle *bundle;
1567
1568             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1569                 if (bundle->bond) {
1570                     bond_rebalance(bundle->bond);
1571                 }
1572             }
1573         }
1574     }
1575     return 0;
1576 }
1577
1578 static void
1579 ofproto_dpif_wait(struct ofproto *ofproto_)
1580 {
1581     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1582
1583     if (ofproto_get_flow_restore_wait()) {
1584         return;
1585     }
1586
1587     if (ofproto->sflow) {
1588         dpif_sflow_wait(ofproto->sflow);
1589     }
1590     if (ofproto->ipfix) {
1591         dpif_ipfix_wait(ofproto->ipfix);
1592     }
1593     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1594         struct ofbundle *bundle;
1595
1596         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1597             bundle_wait(bundle);
1598         }
1599     }
1600     if (ofproto->netflow) {
1601         netflow_wait(ofproto->netflow);
1602     }
1603     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1604     mac_learning_wait(ofproto->ml);
1605     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1606     mcast_snooping_wait(ofproto->ms);
1607     stp_wait(ofproto);
1608     if (ofproto->backer->need_revalidate) {
1609         poll_immediate_wake();
1610     }
1611
1612     seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1613     seq_wait(ofproto->ams_seq, ofproto->ams_seqno);
1614 }
1615
1616 static void
1617 type_get_memory_usage(const char *type, struct simap *usage)
1618 {
1619     struct dpif_backer *backer;
1620
1621     backer = shash_find_data(&all_dpif_backers, type);
1622     if (backer) {
1623         udpif_get_memory_usage(backer->udpif, usage);
1624     }
1625 }
1626
1627 static void
1628 flush(struct ofproto *ofproto_)
1629 {
1630     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1631     struct dpif_backer *backer = ofproto->backer;
1632
1633     if (backer) {
1634         udpif_flush(backer->udpif);
1635     }
1636 }
1637
1638 static void
1639 query_tables(struct ofproto *ofproto,
1640              struct ofputil_table_features *features,
1641              struct ofputil_table_stats *stats)
1642 {
1643     strcpy(features->name, "classifier");
1644
1645     if (stats) {
1646         int i;
1647
1648         for (i = 0; i < ofproto->n_tables; i++) {
1649             unsigned long missed, matched;
1650
1651             atomic_read_relaxed(&ofproto->tables[i].n_matched, &matched);
1652             atomic_read_relaxed(&ofproto->tables[i].n_missed, &missed);
1653
1654             stats[i].matched_count = matched;
1655             stats[i].lookup_count = matched + missed;
1656         }
1657     }
1658 }
1659
1660 static void
1661 set_tables_version(struct ofproto *ofproto_, cls_version_t version)
1662 {
1663     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1664
1665     atomic_store_relaxed(&ofproto->tables_version, version);
1666     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
1667 }
1668
1669 static struct ofport *
1670 port_alloc(void)
1671 {
1672     struct ofport_dpif *port = xzalloc(sizeof *port);
1673     return &port->up;
1674 }
1675
1676 static void
1677 port_dealloc(struct ofport *port_)
1678 {
1679     struct ofport_dpif *port = ofport_dpif_cast(port_);
1680     free(port);
1681 }
1682
1683 static int
1684 port_construct(struct ofport *port_)
1685 {
1686     struct ofport_dpif *port = ofport_dpif_cast(port_);
1687     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1688     const struct netdev *netdev = port->up.netdev;
1689     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1690     const char *dp_port_name;
1691     struct dpif_port dpif_port;
1692     int error;
1693
1694     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1695     port->bundle = NULL;
1696     port->cfm = NULL;
1697     port->bfd = NULL;
1698     port->lldp = NULL;
1699     port->may_enable = false;
1700     port->stp_port = NULL;
1701     port->stp_state = STP_DISABLED;
1702     port->rstp_port = NULL;
1703     port->rstp_state = RSTP_DISABLED;
1704     port->is_tunnel = false;
1705     port->peer = NULL;
1706     port->qdscp = NULL;
1707     port->n_qdscp = 0;
1708     port->carrier_seq = netdev_get_carrier_resets(netdev);
1709     port->is_layer3 = netdev_vport_is_layer3(netdev);
1710
1711     if (netdev_vport_is_patch(netdev)) {
1712         /* By bailing out here, we don't submit the port to the sFlow module
1713          * to be considered for counter polling export.  This is correct
1714          * because the patch port represents an interface that sFlow considers
1715          * to be "internal" to the switch as a whole, and therefore not a
1716          * candidate for counter polling. */
1717         port->odp_port = ODPP_NONE;
1718         ofport_update_peer(port);
1719         return 0;
1720     }
1721
1722     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1723     error = dpif_port_query_by_name(ofproto->backer->dpif, dp_port_name,
1724                                     &dpif_port);
1725     if (error) {
1726         return error;
1727     }
1728
1729     port->odp_port = dpif_port.port_no;
1730
1731     if (netdev_get_tunnel_config(netdev)) {
1732         atomic_count_inc(&ofproto->backer->tnl_count);
1733         error = tnl_port_add(port, port->up.netdev, port->odp_port,
1734                              ovs_native_tunneling_is_on(ofproto), dp_port_name);
1735         if (error) {
1736             atomic_count_dec(&ofproto->backer->tnl_count);
1737             dpif_port_destroy(&dpif_port);
1738             return error;
1739         }
1740
1741         port->is_tunnel = true;
1742         if (ofproto->ipfix) {
1743            dpif_ipfix_add_tunnel_port(ofproto->ipfix, port_, port->odp_port);
1744         }
1745     } else {
1746         /* Sanity-check that a mapping doesn't already exist.  This
1747          * shouldn't happen for non-tunnel ports. */
1748         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1749             VLOG_ERR("port %s already has an OpenFlow port number",
1750                      dpif_port.name);
1751             dpif_port_destroy(&dpif_port);
1752             return EBUSY;
1753         }
1754
1755         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1756         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1757                     hash_odp_port(port->odp_port));
1758         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1759     }
1760     dpif_port_destroy(&dpif_port);
1761
1762     if (ofproto->sflow) {
1763         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1764     }
1765
1766     return 0;
1767 }
1768
1769 static void
1770 port_destruct(struct ofport *port_, bool del)
1771 {
1772     struct ofport_dpif *port = ofport_dpif_cast(port_);
1773     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1774     const char *devname = netdev_get_name(port->up.netdev);
1775     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1776     const char *dp_port_name;
1777
1778     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1779     xlate_txn_start();
1780     xlate_ofport_remove(port);
1781     xlate_txn_commit();
1782
1783     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1784                                               sizeof namebuf);
1785     if (del && dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1786         /* The underlying device is still there, so delete it.  This
1787          * happens when the ofproto is being destroyed, since the caller
1788          * assumes that removal of attached ports will happen as part of
1789          * destruction. */
1790         if (!port->is_tunnel) {
1791             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1792         }
1793     }
1794
1795     if (port->peer) {
1796         port->peer->peer = NULL;
1797         port->peer = NULL;
1798     }
1799
1800     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1801         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1802         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1803         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1804     }
1805
1806     if (port->is_tunnel) {
1807         atomic_count_dec(&ofproto->backer->tnl_count);
1808     }
1809
1810     if (port->is_tunnel && ofproto->ipfix) {
1811        dpif_ipfix_del_tunnel_port(ofproto->ipfix, port->odp_port);
1812     }
1813
1814     tnl_port_del(port);
1815     sset_find_and_delete(&ofproto->ports, devname);
1816     sset_find_and_delete(&ofproto->ghost_ports, devname);
1817     bundle_remove(port_);
1818     set_cfm(port_, NULL);
1819     set_bfd(port_, NULL);
1820     set_lldp(port_, NULL);
1821     if (port->stp_port) {
1822         stp_port_disable(port->stp_port);
1823     }
1824     set_rstp_port(port_, NULL);
1825     if (ofproto->sflow) {
1826         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1827     }
1828
1829     free(port->qdscp);
1830 }
1831
1832 static void
1833 port_modified(struct ofport *port_)
1834 {
1835     struct ofport_dpif *port = ofport_dpif_cast(port_);
1836     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1837     const char *dp_port_name;
1838     struct netdev *netdev = port->up.netdev;
1839
1840     if (port->bundle && port->bundle->bond) {
1841         bond_slave_set_netdev(port->bundle->bond, port, netdev);
1842     }
1843
1844     if (port->cfm) {
1845         cfm_set_netdev(port->cfm, netdev);
1846     }
1847
1848     if (port->bfd) {
1849         bfd_set_netdev(port->bfd, netdev);
1850     }
1851
1852     ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1853                                      port->lldp, &port->up.pp.hw_addr);
1854
1855     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1856
1857     if (port->is_tunnel) {
1858         struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1859
1860         if (tnl_port_reconfigure(port, netdev, port->odp_port,
1861                                  ovs_native_tunneling_is_on(ofproto),
1862                                  dp_port_name)) {
1863             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1864         }
1865     }
1866
1867     ofport_update_peer(port);
1868 }
1869
1870 static void
1871 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1872 {
1873     struct ofport_dpif *port = ofport_dpif_cast(port_);
1874     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1875     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1876
1877     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1878                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1879                    OFPUTIL_PC_NO_PACKET_IN)) {
1880         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1881
1882         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1883             bundle_update(port->bundle);
1884         }
1885     }
1886 }
1887
1888 static int
1889 set_sflow(struct ofproto *ofproto_,
1890           const struct ofproto_sflow_options *sflow_options)
1891 {
1892     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1893     struct dpif_sflow *ds = ofproto->sflow;
1894
1895     if (sflow_options) {
1896         uint32_t old_probability = ds ? dpif_sflow_get_probability(ds) : 0;
1897         if (!ds) {
1898             struct ofport_dpif *ofport;
1899
1900             ds = ofproto->sflow = dpif_sflow_create();
1901             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1902                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1903             }
1904         }
1905         dpif_sflow_set_options(ds, sflow_options);
1906         if (dpif_sflow_get_probability(ds) != old_probability) {
1907             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1908         }
1909     } else {
1910         if (ds) {
1911             dpif_sflow_unref(ds);
1912             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1913             ofproto->sflow = NULL;
1914         }
1915     }
1916     return 0;
1917 }
1918
1919 static int
1920 set_ipfix(
1921     struct ofproto *ofproto_,
1922     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1923     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1924     size_t n_flow_exporters_options)
1925 {
1926     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1927     struct dpif_ipfix *di = ofproto->ipfix;
1928     bool has_options = bridge_exporter_options || flow_exporters_options;
1929     bool new_di = false;
1930
1931     if (has_options && !di) {
1932         di = ofproto->ipfix = dpif_ipfix_create();
1933         new_di = true;
1934     }
1935
1936     if (di) {
1937         /* Call set_options in any case to cleanly flush the flow
1938          * caches in the last exporters that are to be destroyed. */
1939         dpif_ipfix_set_options(
1940             di, bridge_exporter_options, flow_exporters_options,
1941             n_flow_exporters_options);
1942
1943         /* Add tunnel ports only when a new ipfix created */
1944         if (new_di == true) {
1945             struct ofport_dpif *ofport;
1946             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1947                 if (ofport->is_tunnel == true) {
1948                     dpif_ipfix_add_tunnel_port(di, &ofport->up, ofport->odp_port);
1949                 }
1950             }
1951         }
1952
1953         if (!has_options) {
1954             dpif_ipfix_unref(di);
1955             ofproto->ipfix = NULL;
1956         }
1957     }
1958
1959     return 0;
1960 }
1961
1962 static int
1963 get_ipfix_stats(const struct ofproto *ofproto_,
1964                 bool bridge_ipfix,
1965                 struct ovs_list *replies)
1966 {
1967     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1968     struct dpif_ipfix *di = ofproto->ipfix;
1969
1970     if (!di) {
1971         return OFPERR_NXST_NOT_CONFIGURED;
1972     }
1973
1974     return dpif_ipfix_get_stats(di, bridge_ipfix, replies);
1975 }
1976
1977 static int
1978 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1979 {
1980     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1981     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1982     struct cfm *old = ofport->cfm;
1983     int error = 0;
1984
1985     if (s) {
1986         if (!ofport->cfm) {
1987             ofport->cfm = cfm_create(ofport->up.netdev);
1988         }
1989
1990         if (cfm_configure(ofport->cfm, s)) {
1991             error = 0;
1992             goto out;
1993         }
1994
1995         error = EINVAL;
1996     }
1997     cfm_unref(ofport->cfm);
1998     ofport->cfm = NULL;
1999 out:
2000     if (ofport->cfm != old) {
2001         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2002     }
2003     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
2004                                      ofport->lldp, &ofport->up.pp.hw_addr);
2005     return error;
2006 }
2007
2008 static bool
2009 cfm_status_changed(struct ofport *ofport_)
2010 {
2011     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2012
2013     return ofport->cfm ? cfm_check_status_change(ofport->cfm) : true;
2014 }
2015
2016 static int
2017 get_cfm_status(const struct ofport *ofport_,
2018                struct cfm_status *status)
2019 {
2020     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2021     int ret = 0;
2022
2023     if (ofport->cfm) {
2024         cfm_get_status(ofport->cfm, status);
2025     } else {
2026         ret = ENOENT;
2027     }
2028
2029     return ret;
2030 }
2031
2032 static int
2033 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2034 {
2035     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2036     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2037     struct bfd *old;
2038
2039     old = ofport->bfd;
2040     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2041                                 cfg, ofport->up.netdev);
2042     if (ofport->bfd != old) {
2043         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2044     }
2045     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
2046                                      ofport->lldp, &ofport->up.pp.hw_addr);
2047     return 0;
2048 }
2049
2050 static bool
2051 bfd_status_changed(struct ofport *ofport_)
2052 {
2053     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2054
2055     return ofport->bfd ? bfd_check_status_change(ofport->bfd) : true;
2056 }
2057
2058 static int
2059 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2060 {
2061     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2062     int ret = 0;
2063
2064     if (ofport->bfd) {
2065         bfd_get_status(ofport->bfd, smap);
2066     } else {
2067         ret = ENOENT;
2068     }
2069
2070     return ret;
2071 }
2072
2073 static int
2074 set_lldp(struct ofport *ofport_,
2075          const struct smap *cfg)
2076 {
2077     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2078     int error = 0;
2079
2080     if (cfg) {
2081         if (!ofport->lldp) {
2082             struct ofproto_dpif *ofproto;
2083
2084             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2085             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2086             ofport->lldp = lldp_create(ofport->up.netdev, ofport_->mtu, cfg);
2087         }
2088
2089         if (!lldp_configure(ofport->lldp, cfg)) {
2090             error = EINVAL;
2091         }
2092     }
2093     if (error) {
2094         lldp_unref(ofport->lldp);
2095         ofport->lldp = NULL;
2096     }
2097
2098     ofproto_dpif_monitor_port_update(ofport,
2099                                      ofport->bfd,
2100                                      ofport->cfm,
2101                                      ofport->lldp,
2102                                      &ofport->up.pp.hw_addr);
2103     return error;
2104 }
2105
2106 static bool
2107 get_lldp_status(const struct ofport *ofport_,
2108                struct lldp_status *status OVS_UNUSED)
2109 {
2110     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2111
2112     return ofport->lldp ? true : false;
2113 }
2114
2115 static int
2116 set_aa(struct ofproto *ofproto OVS_UNUSED,
2117        const struct aa_settings *s)
2118 {
2119     return aa_configure(s);
2120 }
2121
2122 static int
2123 aa_mapping_set(struct ofproto *ofproto_ OVS_UNUSED, void *aux,
2124                const struct aa_mapping_settings *s)
2125 {
2126     return aa_mapping_register(aux, s);
2127 }
2128
2129 static int
2130 aa_mapping_unset(struct ofproto *ofproto OVS_UNUSED, void *aux)
2131 {
2132     return aa_mapping_unregister(aux);
2133 }
2134
2135 static int
2136 aa_vlan_get_queued(struct ofproto *ofproto OVS_UNUSED, struct ovs_list *list)
2137 {
2138     return aa_get_vlan_queued(list);
2139 }
2140
2141 static unsigned int
2142 aa_vlan_get_queue_size(struct ofproto *ofproto OVS_UNUSED)
2143 {
2144     return aa_get_vlan_queue_size();
2145 }
2146
2147 \f
2148 /* Spanning Tree. */
2149
2150 /* Called while rstp_mutex is held. */
2151 static void
2152 rstp_send_bpdu_cb(struct dp_packet *pkt, void *ofport_, void *ofproto_)
2153 {
2154     struct ofproto_dpif *ofproto = ofproto_;
2155     struct ofport_dpif *ofport = ofport_;
2156     struct eth_header *eth = dp_packet_l2(pkt);
2157
2158     netdev_get_etheraddr(ofport->up.netdev, &eth->eth_src);
2159     if (eth_addr_is_zero(eth->eth_src)) {
2160         VLOG_WARN_RL(&rl, "%s port %d: cannot send RSTP BPDU on a port which "
2161                      "does not have a configured source MAC address.",
2162                      ofproto->up.name, ofp_to_u16(ofport->up.ofp_port));
2163     } else {
2164         ofproto_dpif_send_packet(ofport, pkt);
2165     }
2166     dp_packet_delete(pkt);
2167 }
2168
2169 static void
2170 send_bpdu_cb(struct dp_packet *pkt, int port_num, void *ofproto_)
2171 {
2172     struct ofproto_dpif *ofproto = ofproto_;
2173     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2174     struct ofport_dpif *ofport;
2175
2176     ofport = stp_port_get_aux(sp);
2177     if (!ofport) {
2178         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2179                      ofproto->up.name, port_num);
2180     } else {
2181         struct eth_header *eth = dp_packet_l2(pkt);
2182
2183         netdev_get_etheraddr(ofport->up.netdev, &eth->eth_src);
2184         if (eth_addr_is_zero(eth->eth_src)) {
2185             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2186                          "with unknown MAC", ofproto->up.name, port_num);
2187         } else {
2188             ofproto_dpif_send_packet(ofport, pkt);
2189         }
2190     }
2191     dp_packet_delete(pkt);
2192 }
2193
2194 /* Configure RSTP on 'ofproto_' using the settings defined in 's'. */
2195 static void
2196 set_rstp(struct ofproto *ofproto_, const struct ofproto_rstp_settings *s)
2197 {
2198     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2199
2200     /* Only revalidate flows if the configuration changed. */
2201     if (!s != !ofproto->rstp) {
2202         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2203     }
2204
2205     if (s) {
2206         if (!ofproto->rstp) {
2207             ofproto->rstp = rstp_create(ofproto_->name, s->address,
2208                                         rstp_send_bpdu_cb, ofproto);
2209             ofproto->rstp_last_tick = time_msec();
2210         }
2211         rstp_set_bridge_address(ofproto->rstp, s->address);
2212         rstp_set_bridge_priority(ofproto->rstp, s->priority);
2213         rstp_set_bridge_ageing_time(ofproto->rstp, s->ageing_time);
2214         rstp_set_bridge_force_protocol_version(ofproto->rstp,
2215                                                s->force_protocol_version);
2216         rstp_set_bridge_max_age(ofproto->rstp, s->bridge_max_age);
2217         rstp_set_bridge_forward_delay(ofproto->rstp, s->bridge_forward_delay);
2218         rstp_set_bridge_transmit_hold_count(ofproto->rstp,
2219                                             s->transmit_hold_count);
2220     } else {
2221         struct ofport *ofport;
2222         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2223             set_rstp_port(ofport, NULL);
2224         }
2225         rstp_unref(ofproto->rstp);
2226         ofproto->rstp = NULL;
2227     }
2228 }
2229
2230 static void
2231 get_rstp_status(struct ofproto *ofproto_, struct ofproto_rstp_status *s)
2232 {
2233     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2234
2235     if (ofproto->rstp) {
2236         s->enabled = true;
2237         s->root_id = rstp_get_root_id(ofproto->rstp);
2238         s->bridge_id = rstp_get_bridge_id(ofproto->rstp);
2239         s->designated_id = rstp_get_designated_id(ofproto->rstp);
2240         s->root_path_cost = rstp_get_root_path_cost(ofproto->rstp);
2241         s->designated_port_id = rstp_get_designated_port_id(ofproto->rstp);
2242         s->bridge_port_id = rstp_get_bridge_port_id(ofproto->rstp);
2243     } else {
2244         s->enabled = false;
2245     }
2246 }
2247
2248 static void
2249 update_rstp_port_state(struct ofport_dpif *ofport)
2250 {
2251     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2252     enum rstp_state state;
2253
2254     /* Figure out new state. */
2255     state = ofport->rstp_port ? rstp_port_get_state(ofport->rstp_port)
2256         : RSTP_DISABLED;
2257
2258     /* Update state. */
2259     if (ofport->rstp_state != state) {
2260         enum ofputil_port_state of_state;
2261         bool fwd_change;
2262
2263         VLOG_DBG("port %s: RSTP state changed from %s to %s",
2264                  netdev_get_name(ofport->up.netdev),
2265                  rstp_state_name(ofport->rstp_state),
2266                  rstp_state_name(state));
2267
2268         if (rstp_learn_in_state(ofport->rstp_state)
2269             != rstp_learn_in_state(state)) {
2270             /* XXX: Learning action flows should also be flushed. */
2271             if (ofport->bundle) {
2272                 if (!rstp_shift_root_learned_address(ofproto->rstp)
2273                     || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2274                     bundle_flush_macs(ofport->bundle, false);
2275                 }
2276             }
2277         }
2278         fwd_change = rstp_forward_in_state(ofport->rstp_state)
2279             != rstp_forward_in_state(state);
2280
2281         ofproto->backer->need_revalidate = REV_RSTP;
2282         ofport->rstp_state = state;
2283
2284         if (fwd_change && ofport->bundle) {
2285             bundle_update(ofport->bundle);
2286         }
2287
2288         /* Update the RSTP state bits in the OpenFlow port description. */
2289         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2290         of_state |= (state == RSTP_LEARNING ? OFPUTIL_PS_STP_LEARN
2291                 : state == RSTP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2292                 : state == RSTP_DISCARDING ?  OFPUTIL_PS_STP_LISTEN
2293                 : 0);
2294         ofproto_port_set_state(&ofport->up, of_state);
2295     }
2296 }
2297
2298 static void
2299 rstp_run(struct ofproto_dpif *ofproto)
2300 {
2301     if (ofproto->rstp) {
2302         long long int now = time_msec();
2303         long long int elapsed = now - ofproto->rstp_last_tick;
2304         struct rstp_port *rp;
2305         struct ofport_dpif *ofport;
2306
2307         /* Every second, decrease the values of the timers. */
2308         if (elapsed >= 1000) {
2309             rstp_tick_timers(ofproto->rstp);
2310             ofproto->rstp_last_tick = now;
2311         }
2312         rp = NULL;
2313         while ((ofport = rstp_get_next_changed_port_aux(ofproto->rstp, &rp))) {
2314             update_rstp_port_state(ofport);
2315         }
2316         rp = NULL;
2317         ofport = NULL;
2318         /* FIXME: This check should be done on-event (i.e., when setting
2319          * p->fdb_flush) and not periodically.
2320          */
2321         while ((ofport = rstp_check_and_reset_fdb_flush(ofproto->rstp, &rp))) {
2322             if (!rstp_shift_root_learned_address(ofproto->rstp)
2323                 || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2324                 bundle_flush_macs(ofport->bundle, false);
2325             }
2326         }
2327
2328         if (rstp_shift_root_learned_address(ofproto->rstp)) {
2329             struct ofport_dpif *old_root_aux =
2330                 (struct ofport_dpif *)rstp_get_old_root_aux(ofproto->rstp);
2331             struct ofport_dpif *new_root_aux =
2332                 (struct ofport_dpif *)rstp_get_new_root_aux(ofproto->rstp);
2333             if (old_root_aux != NULL && new_root_aux != NULL) {
2334                 bundle_move(old_root_aux->bundle, new_root_aux->bundle);
2335                 rstp_reset_root_changed(ofproto->rstp);
2336             }
2337         }
2338     }
2339 }
2340
2341 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2342 static int
2343 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2344 {
2345     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2346
2347     /* Only revalidate flows if the configuration changed. */
2348     if (!s != !ofproto->stp) {
2349         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2350     }
2351
2352     if (s) {
2353         if (!ofproto->stp) {
2354             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2355                                       send_bpdu_cb, ofproto);
2356             ofproto->stp_last_tick = time_msec();
2357         }
2358
2359         stp_set_bridge_id(ofproto->stp, s->system_id);
2360         stp_set_bridge_priority(ofproto->stp, s->priority);
2361         stp_set_hello_time(ofproto->stp, s->hello_time);
2362         stp_set_max_age(ofproto->stp, s->max_age);
2363         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2364     }  else {
2365         struct ofport *ofport;
2366
2367         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2368             set_stp_port(ofport, NULL);
2369         }
2370
2371         stp_unref(ofproto->stp);
2372         ofproto->stp = NULL;
2373     }
2374
2375     return 0;
2376 }
2377
2378 static int
2379 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2380 {
2381     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2382
2383     if (ofproto->stp) {
2384         s->enabled = true;
2385         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2386         s->designated_root = stp_get_designated_root(ofproto->stp);
2387         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2388     } else {
2389         s->enabled = false;
2390     }
2391
2392     return 0;
2393 }
2394
2395 static void
2396 update_stp_port_state(struct ofport_dpif *ofport)
2397 {
2398     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2399     enum stp_state state;
2400
2401     /* Figure out new state. */
2402     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2403                              : STP_DISABLED;
2404
2405     /* Update state. */
2406     if (ofport->stp_state != state) {
2407         enum ofputil_port_state of_state;
2408         bool fwd_change;
2409
2410         VLOG_DBG("port %s: STP state changed from %s to %s",
2411                  netdev_get_name(ofport->up.netdev),
2412                  stp_state_name(ofport->stp_state),
2413                  stp_state_name(state));
2414         if (stp_learn_in_state(ofport->stp_state)
2415                 != stp_learn_in_state(state)) {
2416             /* xxx Learning action flows should also be flushed. */
2417             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2418             mac_learning_flush(ofproto->ml);
2419             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2420             mcast_snooping_mdb_flush(ofproto->ms);
2421         }
2422         fwd_change = stp_forward_in_state(ofport->stp_state)
2423                         != stp_forward_in_state(state);
2424
2425         ofproto->backer->need_revalidate = REV_STP;
2426         ofport->stp_state = state;
2427         ofport->stp_state_entered = time_msec();
2428
2429         if (fwd_change && ofport->bundle) {
2430             bundle_update(ofport->bundle);
2431         }
2432
2433         /* Update the STP state bits in the OpenFlow port description. */
2434         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2435         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2436                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2437                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2438                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2439                      : 0);
2440         ofproto_port_set_state(&ofport->up, of_state);
2441     }
2442 }
2443
2444 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2445  * caller is responsible for assigning STP port numbers and ensuring
2446  * there are no duplicates. */
2447 static int
2448 set_stp_port(struct ofport *ofport_,
2449              const struct ofproto_port_stp_settings *s)
2450 {
2451     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2452     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2453     struct stp_port *sp = ofport->stp_port;
2454
2455     if (!s || !s->enable) {
2456         if (sp) {
2457             ofport->stp_port = NULL;
2458             stp_port_disable(sp);
2459             update_stp_port_state(ofport);
2460         }
2461         return 0;
2462     } else if (sp && stp_port_no(sp) != s->port_num
2463                && ofport == stp_port_get_aux(sp)) {
2464         /* The port-id changed, so disable the old one if it's not
2465          * already in use by another port. */
2466         stp_port_disable(sp);
2467     }
2468
2469     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2470
2471     /* Set name before enabling the port so that debugging messages can print
2472      * the name. */
2473     stp_port_set_name(sp, netdev_get_name(ofport->up.netdev));
2474     stp_port_enable(sp);
2475
2476     stp_port_set_aux(sp, ofport);
2477     stp_port_set_priority(sp, s->priority);
2478     stp_port_set_path_cost(sp, s->path_cost);
2479
2480     update_stp_port_state(ofport);
2481
2482     return 0;
2483 }
2484
2485 static int
2486 get_stp_port_status(struct ofport *ofport_,
2487                     struct ofproto_port_stp_status *s)
2488 {
2489     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2490     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2491     struct stp_port *sp = ofport->stp_port;
2492
2493     if (!ofproto->stp || !sp) {
2494         s->enabled = false;
2495         return 0;
2496     }
2497
2498     s->enabled = true;
2499     s->port_id = stp_port_get_id(sp);
2500     s->state = stp_port_get_state(sp);
2501     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2502     s->role = stp_port_get_role(sp);
2503
2504     return 0;
2505 }
2506
2507 static int
2508 get_stp_port_stats(struct ofport *ofport_,
2509                    struct ofproto_port_stp_stats *s)
2510 {
2511     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2512     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2513     struct stp_port *sp = ofport->stp_port;
2514
2515     if (!ofproto->stp || !sp) {
2516         s->enabled = false;
2517         return 0;
2518     }
2519
2520     s->enabled = true;
2521     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2522
2523     return 0;
2524 }
2525
2526 static void
2527 stp_run(struct ofproto_dpif *ofproto)
2528 {
2529     if (ofproto->stp) {
2530         long long int now = time_msec();
2531         long long int elapsed = now - ofproto->stp_last_tick;
2532         struct stp_port *sp;
2533
2534         if (elapsed > 0) {
2535             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2536             ofproto->stp_last_tick = now;
2537         }
2538         while (stp_get_changed_port(ofproto->stp, &sp)) {
2539             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2540
2541             if (ofport) {
2542                 update_stp_port_state(ofport);
2543             }
2544         }
2545
2546         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2547             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2548             mac_learning_flush(ofproto->ml);
2549             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2550             mcast_snooping_mdb_flush(ofproto->ms);
2551         }
2552     }
2553 }
2554
2555 static void
2556 stp_wait(struct ofproto_dpif *ofproto)
2557 {
2558     if (ofproto->stp) {
2559         poll_timer_wait(1000);
2560     }
2561 }
2562
2563 /* Configures RSTP on 'ofport_' using the settings defined in 's'.  The
2564  * caller is responsible for assigning RSTP port numbers and ensuring
2565  * there are no duplicates. */
2566 static void
2567 set_rstp_port(struct ofport *ofport_,
2568               const struct ofproto_port_rstp_settings *s)
2569 {
2570     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2571     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2572     struct rstp_port *rp = ofport->rstp_port;
2573
2574     if (!s || !s->enable) {
2575         if (rp) {
2576             rstp_port_set_aux(rp, NULL);
2577             rstp_port_set_state(rp, RSTP_DISABLED);
2578             rstp_port_set_mac_operational(rp, false);
2579             ofport->rstp_port = NULL;
2580             rstp_port_unref(rp);
2581             update_rstp_port_state(ofport);
2582         }
2583         return;
2584     }
2585
2586     /* Check if need to add a new port. */
2587     if (!rp) {
2588         rp = ofport->rstp_port = rstp_add_port(ofproto->rstp);
2589     }
2590
2591     rstp_port_set(rp, s->port_num, s->priority, s->path_cost,
2592                   s->admin_edge_port, s->auto_edge,
2593                   s->admin_p2p_mac_state, s->admin_port_state, s->mcheck,
2594                   ofport);
2595     update_rstp_port_state(ofport);
2596     /* Synchronize operational status. */
2597     rstp_port_set_mac_operational(rp, ofport->may_enable);
2598 }
2599
2600 static void
2601 get_rstp_port_status(struct ofport *ofport_,
2602         struct ofproto_port_rstp_status *s)
2603 {
2604     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2605     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2606     struct rstp_port *rp = ofport->rstp_port;
2607
2608     if (!ofproto->rstp || !rp) {
2609         s->enabled = false;
2610         return;
2611     }
2612
2613     s->enabled = true;
2614     rstp_port_get_status(rp, &s->port_id, &s->state, &s->role,
2615                          &s->designated_bridge_id, &s->designated_port_id,
2616                          &s->designated_path_cost, &s->tx_count,
2617                          &s->rx_count, &s->error_count, &s->uptime);
2618 }
2619
2620 \f
2621 static int
2622 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2623            size_t n_qdscp)
2624 {
2625     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2626     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2627
2628     if (ofport->n_qdscp != n_qdscp
2629         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2630                               n_qdscp * sizeof *qdscp))) {
2631         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2632         free(ofport->qdscp);
2633         ofport->qdscp = n_qdscp
2634             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2635             : NULL;
2636         ofport->n_qdscp = n_qdscp;
2637     }
2638
2639     return 0;
2640 }
2641 \f
2642 /* Bundles. */
2643
2644 /* Expires all MAC learning entries associated with 'bundle' and forces its
2645  * ofproto to revalidate every flow.
2646  *
2647  * Normally MAC learning entries are removed only from the ofproto associated
2648  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2649  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2650  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2651  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2652  * with the host from which it migrated. */
2653 static void
2654 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2655 {
2656     struct ofproto_dpif *ofproto = bundle->ofproto;
2657     struct mac_learning *ml = ofproto->ml;
2658     struct mac_entry *mac, *next_mac;
2659
2660     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2661     ovs_rwlock_wrlock(&ml->rwlock);
2662     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2663         if (mac_entry_get_port(ml, mac) == bundle) {
2664             if (all_ofprotos) {
2665                 struct ofproto_dpif *o;
2666
2667                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2668                     if (o != ofproto) {
2669                         struct mac_entry *e;
2670
2671                         ovs_rwlock_wrlock(&o->ml->rwlock);
2672                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2673                         if (e) {
2674                             mac_learning_expire(o->ml, e);
2675                         }
2676                         ovs_rwlock_unlock(&o->ml->rwlock);
2677                     }
2678                 }
2679             }
2680
2681             mac_learning_expire(ml, mac);
2682         }
2683     }
2684     ovs_rwlock_unlock(&ml->rwlock);
2685 }
2686
2687 static void
2688 bundle_move(struct ofbundle *old, struct ofbundle *new)
2689 {
2690     struct ofproto_dpif *ofproto = old->ofproto;
2691     struct mac_learning *ml = ofproto->ml;
2692     struct mac_entry *mac, *next_mac;
2693
2694     ovs_assert(new->ofproto == old->ofproto);
2695
2696     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2697     ovs_rwlock_wrlock(&ml->rwlock);
2698     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2699         if (mac_entry_get_port(ml, mac) == old) {
2700             mac_entry_set_port(ml, mac, new);
2701         }
2702     }
2703     ovs_rwlock_unlock(&ml->rwlock);
2704 }
2705
2706 static struct ofbundle *
2707 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2708 {
2709     struct ofbundle *bundle;
2710
2711     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2712                              &ofproto->bundles) {
2713         if (bundle->aux == aux) {
2714             return bundle;
2715         }
2716     }
2717     return NULL;
2718 }
2719
2720 static void
2721 bundle_update(struct ofbundle *bundle)
2722 {
2723     struct ofport_dpif *port;
2724
2725     bundle->floodable = true;
2726     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2727         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2728             || port->is_layer3
2729             || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2730             || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2731             bundle->floodable = false;
2732             break;
2733         }
2734     }
2735 }
2736
2737 static void
2738 bundle_del_port(struct ofport_dpif *port)
2739 {
2740     struct ofbundle *bundle = port->bundle;
2741
2742     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2743
2744     ovs_list_remove(&port->bundle_node);
2745     port->bundle = NULL;
2746
2747     if (bundle->lacp) {
2748         lacp_slave_unregister(bundle->lacp, port);
2749     }
2750     if (bundle->bond) {
2751         bond_slave_unregister(bundle->bond, port);
2752     }
2753
2754     bundle_update(bundle);
2755 }
2756
2757 static bool
2758 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2759                 struct lacp_slave_settings *lacp)
2760 {
2761     struct ofport_dpif *port;
2762
2763     port = ofp_port_to_ofport(bundle->ofproto, ofp_port);
2764     if (!port) {
2765         return false;
2766     }
2767
2768     if (port->bundle != bundle) {
2769         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2770         if (port->bundle) {
2771             bundle_remove(&port->up);
2772         }
2773
2774         port->bundle = bundle;
2775         ovs_list_push_back(&bundle->ports, &port->bundle_node);
2776         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2777             || port->is_layer3
2778             || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2779             || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2780             bundle->floodable = false;
2781         }
2782     }
2783     if (lacp) {
2784         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2785         lacp_slave_register(bundle->lacp, port, lacp);
2786     }
2787
2788     return true;
2789 }
2790
2791 static void
2792 bundle_destroy(struct ofbundle *bundle)
2793 {
2794     struct ofproto_dpif *ofproto;
2795     struct ofport_dpif *port, *next_port;
2796
2797     if (!bundle) {
2798         return;
2799     }
2800
2801     ofproto = bundle->ofproto;
2802     mbridge_unregister_bundle(ofproto->mbridge, bundle);
2803
2804     xlate_txn_start();
2805     xlate_bundle_remove(bundle);
2806     xlate_txn_commit();
2807
2808     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2809         bundle_del_port(port);
2810     }
2811
2812     bundle_flush_macs(bundle, true);
2813     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2814     free(bundle->name);
2815     free(bundle->trunks);
2816     lacp_unref(bundle->lacp);
2817     bond_unref(bundle->bond);
2818     free(bundle);
2819 }
2820
2821 static int
2822 bundle_set(struct ofproto *ofproto_, void *aux,
2823            const struct ofproto_bundle_settings *s)
2824 {
2825     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2826     bool need_flush = false;
2827     struct ofport_dpif *port;
2828     struct ofbundle *bundle;
2829     unsigned long *trunks;
2830     int vlan;
2831     size_t i;
2832     bool ok;
2833
2834     if (!s) {
2835         bundle_destroy(bundle_lookup(ofproto, aux));
2836         return 0;
2837     }
2838
2839     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2840     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2841
2842     bundle = bundle_lookup(ofproto, aux);
2843     if (!bundle) {
2844         bundle = xmalloc(sizeof *bundle);
2845
2846         bundle->ofproto = ofproto;
2847         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2848                     hash_pointer(aux, 0));
2849         bundle->aux = aux;
2850         bundle->name = NULL;
2851
2852         ovs_list_init(&bundle->ports);
2853         bundle->vlan_mode = PORT_VLAN_TRUNK;
2854         bundle->vlan = -1;
2855         bundle->trunks = NULL;
2856         bundle->use_priority_tags = s->use_priority_tags;
2857         bundle->lacp = NULL;
2858         bundle->bond = NULL;
2859
2860         bundle->floodable = true;
2861         mbridge_register_bundle(ofproto->mbridge, bundle);
2862     }
2863
2864     if (!bundle->name || strcmp(s->name, bundle->name)) {
2865         free(bundle->name);
2866         bundle->name = xstrdup(s->name);
2867     }
2868
2869     /* LACP. */
2870     if (s->lacp) {
2871         ofproto->lacp_enabled = true;
2872         if (!bundle->lacp) {
2873             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2874             bundle->lacp = lacp_create();
2875         }
2876         lacp_configure(bundle->lacp, s->lacp);
2877     } else {
2878         lacp_unref(bundle->lacp);
2879         bundle->lacp = NULL;
2880     }
2881
2882     /* Update set of ports. */
2883     ok = true;
2884     for (i = 0; i < s->n_slaves; i++) {
2885         if (!bundle_add_port(bundle, s->slaves[i],
2886                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2887             ok = false;
2888         }
2889     }
2890     if (!ok || ovs_list_size(&bundle->ports) != s->n_slaves) {
2891         struct ofport_dpif *next_port;
2892
2893         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2894             for (i = 0; i < s->n_slaves; i++) {
2895                 if (s->slaves[i] == port->up.ofp_port) {
2896                     goto found;
2897                 }
2898             }
2899
2900             bundle_del_port(port);
2901         found: ;
2902         }
2903     }
2904     ovs_assert(ovs_list_size(&bundle->ports) <= s->n_slaves);
2905
2906     if (ovs_list_is_empty(&bundle->ports)) {
2907         bundle_destroy(bundle);
2908         return EINVAL;
2909     }
2910
2911     /* Set VLAN tagging mode */
2912     if (s->vlan_mode != bundle->vlan_mode
2913         || s->use_priority_tags != bundle->use_priority_tags) {
2914         bundle->vlan_mode = s->vlan_mode;
2915         bundle->use_priority_tags = s->use_priority_tags;
2916         need_flush = true;
2917     }
2918
2919     /* Set VLAN tag. */
2920     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2921             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2922             : 0);
2923     if (vlan != bundle->vlan) {
2924         bundle->vlan = vlan;
2925         need_flush = true;
2926     }
2927
2928     /* Get trunked VLANs. */
2929     switch (s->vlan_mode) {
2930     case PORT_VLAN_ACCESS:
2931         trunks = NULL;
2932         break;
2933
2934     case PORT_VLAN_TRUNK:
2935         trunks = CONST_CAST(unsigned long *, s->trunks);
2936         break;
2937
2938     case PORT_VLAN_NATIVE_UNTAGGED:
2939     case PORT_VLAN_NATIVE_TAGGED:
2940         if (vlan != 0 && (!s->trunks
2941                           || !bitmap_is_set(s->trunks, vlan)
2942                           || bitmap_is_set(s->trunks, 0))) {
2943             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2944             if (s->trunks) {
2945                 trunks = bitmap_clone(s->trunks, 4096);
2946             } else {
2947                 trunks = bitmap_allocate1(4096);
2948             }
2949             bitmap_set1(trunks, vlan);
2950             bitmap_set0(trunks, 0);
2951         } else {
2952             trunks = CONST_CAST(unsigned long *, s->trunks);
2953         }
2954         break;
2955
2956     default:
2957         OVS_NOT_REACHED();
2958     }
2959     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2960         free(bundle->trunks);
2961         if (trunks == s->trunks) {
2962             bundle->trunks = vlan_bitmap_clone(trunks);
2963         } else {
2964             bundle->trunks = trunks;
2965             trunks = NULL;
2966         }
2967         need_flush = true;
2968     }
2969     if (trunks != s->trunks) {
2970         free(trunks);
2971     }
2972
2973     /* Bonding. */
2974     if (!ovs_list_is_short(&bundle->ports)) {
2975         bundle->ofproto->has_bonded_bundles = true;
2976         if (bundle->bond) {
2977             if (bond_reconfigure(bundle->bond, s->bond)) {
2978                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2979             }
2980         } else {
2981             bundle->bond = bond_create(s->bond, ofproto);
2982             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2983         }
2984
2985         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2986             bond_slave_register(bundle->bond, port,
2987                                 port->up.ofp_port, port->up.netdev);
2988         }
2989     } else {
2990         bond_unref(bundle->bond);
2991         bundle->bond = NULL;
2992     }
2993
2994     /* If we changed something that would affect MAC learning, un-learn
2995      * everything on this port and force flow revalidation. */
2996     if (need_flush) {
2997         bundle_flush_macs(bundle, false);
2998     }
2999
3000     return 0;
3001 }
3002
3003 static void
3004 bundle_remove(struct ofport *port_)
3005 {
3006     struct ofport_dpif *port = ofport_dpif_cast(port_);
3007     struct ofbundle *bundle = port->bundle;
3008
3009     if (bundle) {
3010         bundle_del_port(port);
3011         if (ovs_list_is_empty(&bundle->ports)) {
3012             bundle_destroy(bundle);
3013         } else if (ovs_list_is_short(&bundle->ports)) {
3014             bond_unref(bundle->bond);
3015             bundle->bond = NULL;
3016         }
3017     }
3018 }
3019
3020 static void
3021 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
3022 {
3023     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3024     struct ofport_dpif *port = port_;
3025     struct eth_addr ea;
3026     int error;
3027
3028     error = netdev_get_etheraddr(port->up.netdev, &ea);
3029     if (!error) {
3030         struct dp_packet packet;
3031         void *packet_pdu;
3032
3033         dp_packet_init(&packet, 0);
3034         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
3035                                  pdu_size);
3036         memcpy(packet_pdu, pdu, pdu_size);
3037
3038         ofproto_dpif_send_packet(port, &packet);
3039         dp_packet_uninit(&packet);
3040     } else {
3041         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
3042                     "%s (%s)", port->bundle->name,
3043                     netdev_get_name(port->up.netdev), ovs_strerror(error));
3044     }
3045 }
3046
3047 static void
3048 bundle_send_learning_packets(struct ofbundle *bundle)
3049 {
3050     struct ofproto_dpif *ofproto = bundle->ofproto;
3051     int error, n_packets, n_errors;
3052     struct mac_entry *e;
3053     struct pkt_list {
3054         struct ovs_list list_node;
3055         struct ofport_dpif *port;
3056         struct dp_packet *pkt;
3057     } *pkt_node;
3058     struct ovs_list packets;
3059
3060     ovs_list_init(&packets);
3061     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3062     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3063         if (mac_entry_get_port(ofproto->ml, e) != bundle) {
3064             pkt_node = xmalloc(sizeof *pkt_node);
3065             pkt_node->pkt = bond_compose_learning_packet(bundle->bond,
3066                                                          e->mac, e->vlan,
3067                                                          (void **)&pkt_node->port);
3068             ovs_list_push_back(&packets, &pkt_node->list_node);
3069         }
3070     }
3071     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3072
3073     error = n_packets = n_errors = 0;
3074     LIST_FOR_EACH_POP (pkt_node, list_node, &packets) {
3075         int ret;
3076
3077         ret = ofproto_dpif_send_packet(pkt_node->port, pkt_node->pkt);
3078         dp_packet_delete(pkt_node->pkt);
3079         free(pkt_node);
3080         if (ret) {
3081             error = ret;
3082             n_errors++;
3083         }
3084         n_packets++;
3085     }
3086
3087     if (n_errors) {
3088         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3089         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3090                      "packets, last error was: %s",
3091                      bundle->name, n_errors, n_packets, ovs_strerror(error));
3092     } else {
3093         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3094                  bundle->name, n_packets);
3095     }
3096 }
3097
3098 static void
3099 bundle_run(struct ofbundle *bundle)
3100 {
3101     if (bundle->lacp) {
3102         lacp_run(bundle->lacp, send_pdu_cb);
3103     }
3104     if (bundle->bond) {
3105         struct ofport_dpif *port;
3106
3107         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
3108             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
3109         }
3110
3111         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
3112             bundle->ofproto->backer->need_revalidate = REV_BOND;
3113         }
3114
3115         if (bond_should_send_learning_packets(bundle->bond)) {
3116             bundle_send_learning_packets(bundle);
3117         }
3118     }
3119 }
3120
3121 static void
3122 bundle_wait(struct ofbundle *bundle)
3123 {
3124     if (bundle->lacp) {
3125         lacp_wait(bundle->lacp);
3126     }
3127     if (bundle->bond) {
3128         bond_wait(bundle->bond);
3129     }
3130 }
3131 \f
3132 /* Mirrors. */
3133
3134 static int
3135 mirror_set__(struct ofproto *ofproto_, void *aux,
3136              const struct ofproto_mirror_settings *s)
3137 {
3138     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3139     struct ofbundle **srcs, **dsts;
3140     int error;
3141     size_t i;
3142
3143     if (!s) {
3144         mirror_destroy(ofproto->mbridge, aux);
3145         return 0;
3146     }
3147
3148     srcs = xmalloc(s->n_srcs * sizeof *srcs);
3149     dsts = xmalloc(s->n_dsts * sizeof *dsts);
3150
3151     for (i = 0; i < s->n_srcs; i++) {
3152         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
3153     }
3154
3155     for (i = 0; i < s->n_dsts; i++) {
3156         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
3157     }
3158
3159     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
3160                        s->n_dsts, s->src_vlans,
3161                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
3162     free(srcs);
3163     free(dsts);
3164     return error;
3165 }
3166
3167 static int
3168 mirror_get_stats__(struct ofproto *ofproto, void *aux,
3169                    uint64_t *packets, uint64_t *bytes)
3170 {
3171     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
3172                             bytes);
3173 }
3174
3175 static int
3176 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3177 {
3178     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3179     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3180     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3181         mac_learning_flush(ofproto->ml);
3182     }
3183     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3184     return 0;
3185 }
3186
3187 static bool
3188 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3189 {
3190     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3191     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3192     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
3193 }
3194
3195 static void
3196 forward_bpdu_changed(struct ofproto *ofproto_)
3197 {
3198     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3199     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3200 }
3201
3202 static void
3203 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3204                      size_t max_entries)
3205 {
3206     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3207     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3208     mac_learning_set_idle_time(ofproto->ml, idle_time);
3209     mac_learning_set_max_entries(ofproto->ml, max_entries);
3210     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3211 }
3212
3213 /* Configures multicast snooping on 'ofport' using the settings
3214  * defined in 's'. */
3215 static int
3216 set_mcast_snooping(struct ofproto *ofproto_,
3217                    const struct ofproto_mcast_snooping_settings *s)
3218 {
3219     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3220
3221     /* Only revalidate flows if the configuration changed. */
3222     if (!s != !ofproto->ms) {
3223         ofproto->backer->need_revalidate = REV_RECONFIGURE;
3224     }
3225
3226     if (s) {
3227         if (!ofproto->ms) {
3228             ofproto->ms = mcast_snooping_create();
3229         }
3230
3231         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3232         mcast_snooping_set_idle_time(ofproto->ms, s->idle_time);
3233         mcast_snooping_set_max_entries(ofproto->ms, s->max_entries);
3234         if (mcast_snooping_set_flood_unreg(ofproto->ms, s->flood_unreg)) {
3235             ofproto->backer->need_revalidate = REV_RECONFIGURE;
3236         }
3237         ovs_rwlock_unlock(&ofproto->ms->rwlock);
3238     } else {
3239         mcast_snooping_unref(ofproto->ms);
3240         ofproto->ms = NULL;
3241     }
3242
3243     return 0;
3244 }
3245
3246 /* Configures multicast snooping port's flood settings on 'ofproto'. */
3247 static int
3248 set_mcast_snooping_port(struct ofproto *ofproto_, void *aux,
3249                         const struct ofproto_mcast_snooping_port_settings *s)
3250 {
3251     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3252     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3253
3254     if (ofproto->ms && s) {
3255         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3256         mcast_snooping_set_port_flood(ofproto->ms, bundle, s->flood);
3257         mcast_snooping_set_port_flood_reports(ofproto->ms, bundle,
3258                                               s->flood_reports);
3259         ovs_rwlock_unlock(&ofproto->ms->rwlock);
3260     }
3261     return 0;
3262 }
3263
3264 \f
3265 /* Ports. */
3266
3267 struct ofport_dpif *
3268 ofp_port_to_ofport(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
3269 {
3270     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3271     return ofport ? ofport_dpif_cast(ofport) : NULL;
3272 }
3273
3274 static void
3275 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3276                             struct ofproto_port *ofproto_port,
3277                             struct dpif_port *dpif_port)
3278 {
3279     ofproto_port->name = dpif_port->name;
3280     ofproto_port->type = dpif_port->type;
3281     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3282 }
3283
3284 static void
3285 ofport_update_peer(struct ofport_dpif *ofport)
3286 {
3287     const struct ofproto_dpif *ofproto;
3288     struct dpif_backer *backer;
3289     char *peer_name;
3290
3291     if (!netdev_vport_is_patch(ofport->up.netdev)) {
3292         return;
3293     }
3294
3295     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
3296     backer->need_revalidate = REV_RECONFIGURE;
3297
3298     if (ofport->peer) {
3299         ofport->peer->peer = NULL;
3300         ofport->peer = NULL;
3301     }
3302
3303     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
3304     if (!peer_name) {
3305         return;
3306     }
3307
3308     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3309         struct ofport *peer_ofport;
3310         struct ofport_dpif *peer;
3311         char *peer_peer;
3312
3313         if (ofproto->backer != backer) {
3314             continue;
3315         }
3316
3317         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
3318         if (!peer_ofport) {
3319             continue;
3320         }
3321
3322         peer = ofport_dpif_cast(peer_ofport);
3323         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
3324         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
3325                                  peer_peer)) {
3326             ofport->peer = peer;
3327             ofport->peer->peer = ofport;
3328         }
3329         free(peer_peer);
3330
3331         break;
3332     }
3333     free(peer_name);
3334 }
3335
3336 static void
3337 port_run(struct ofport_dpif *ofport)
3338 {
3339     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3340     bool carrier_changed = carrier_seq != ofport->carrier_seq;
3341     bool enable = netdev_get_carrier(ofport->up.netdev);
3342     bool cfm_enable = false;
3343     bool bfd_enable = false;
3344
3345     ofport->carrier_seq = carrier_seq;
3346
3347     if (ofport->cfm) {
3348         int cfm_opup = cfm_get_opup(ofport->cfm);
3349
3350         cfm_enable = !cfm_get_fault(ofport->cfm);
3351
3352         if (cfm_opup >= 0) {
3353             cfm_enable = cfm_enable && cfm_opup;
3354         }
3355     }
3356
3357     if (ofport->bfd) {
3358         bfd_enable = bfd_forwarding(ofport->bfd);
3359     }
3360
3361     if (ofport->bfd || ofport->cfm) {
3362         enable = enable && (cfm_enable || bfd_enable);
3363     }
3364
3365     if (ofport->bundle) {
3366         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3367         if (carrier_changed) {
3368             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3369         }
3370     }
3371
3372     if (ofport->may_enable != enable) {
3373         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3374
3375         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3376
3377         if (ofport->rstp_port) {
3378             rstp_port_set_mac_operational(ofport->rstp_port, enable);
3379         }
3380     }
3381
3382     ofport->may_enable = enable;
3383 }
3384
3385 static int
3386 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3387                    struct ofproto_port *ofproto_port)
3388 {
3389     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3390     struct dpif_port dpif_port;
3391     int error;
3392
3393     if (sset_contains(&ofproto->ghost_ports, devname)) {
3394         const char *type = netdev_get_type_from_name(devname);
3395
3396         /* We may be called before ofproto->up.port_by_name is populated with
3397          * the appropriate ofport.  For this reason, we must get the name and
3398          * type from the netdev layer directly. */
3399         if (type) {
3400             const struct ofport *ofport;
3401
3402             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3403             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3404             ofproto_port->name = xstrdup(devname);
3405             ofproto_port->type = xstrdup(type);
3406             return 0;
3407         }
3408         return ENODEV;
3409     }
3410
3411     if (!sset_contains(&ofproto->ports, devname)) {
3412         return ENODEV;
3413     }
3414     error = dpif_port_query_by_name(ofproto->backer->dpif,
3415                                     devname, &dpif_port);
3416     if (!error) {
3417         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3418     }
3419     return error;
3420 }
3421
3422 static int
3423 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3424 {
3425     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3426     const char *devname = netdev_get_name(netdev);
3427     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3428     const char *dp_port_name;
3429
3430     if (netdev_vport_is_patch(netdev)) {
3431         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3432         return 0;
3433     }
3434
3435     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3436     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3437         odp_port_t port_no = ODPP_NONE;
3438         int error;
3439
3440         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3441         if (error) {
3442             return error;
3443         }
3444         if (netdev_get_tunnel_config(netdev)) {
3445             simap_put(&ofproto->backer->tnl_backers,
3446                       dp_port_name, odp_to_u32(port_no));
3447         }
3448     }
3449
3450     if (netdev_get_tunnel_config(netdev)) {
3451         sset_add(&ofproto->ghost_ports, devname);
3452     } else {
3453         sset_add(&ofproto->ports, devname);
3454     }
3455     return 0;
3456 }
3457
3458 static int
3459 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3460 {
3461     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3462     struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
3463     int error = 0;
3464
3465     if (!ofport) {
3466         return 0;
3467     }
3468
3469     sset_find_and_delete(&ofproto->ghost_ports,
3470                          netdev_get_name(ofport->up.netdev));
3471     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3472     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3473         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3474         if (!error) {
3475             /* The caller is going to close ofport->up.netdev.  If this is a
3476              * bonded port, then the bond is using that netdev, so remove it
3477              * from the bond.  The client will need to reconfigure everything
3478              * after deleting ports, so then the slave will get re-added. */
3479             bundle_remove(&ofport->up);
3480         }
3481     }
3482     return error;
3483 }
3484
3485 static int
3486 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3487 {
3488     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3489     int error;
3490
3491     error = netdev_get_stats(ofport->up.netdev, stats);
3492
3493     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3494         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3495
3496         ovs_mutex_lock(&ofproto->stats_mutex);
3497         /* ofproto->stats.tx_packets represents packets that we created
3498          * internally and sent to some port (e.g. packets sent with
3499          * ofproto_dpif_send_packet()).  Account for them as if they had
3500          * come from OFPP_LOCAL and got forwarded. */
3501
3502         if (stats->rx_packets != UINT64_MAX) {
3503             stats->rx_packets += ofproto->stats.tx_packets;
3504         }
3505
3506         if (stats->rx_bytes != UINT64_MAX) {
3507             stats->rx_bytes += ofproto->stats.tx_bytes;
3508         }
3509
3510         /* ofproto->stats.rx_packets represents packets that were received on
3511          * some port and we processed internally and dropped (e.g. STP).
3512          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3513
3514         if (stats->tx_packets != UINT64_MAX) {
3515             stats->tx_packets += ofproto->stats.rx_packets;
3516         }
3517
3518         if (stats->tx_bytes != UINT64_MAX) {
3519             stats->tx_bytes += ofproto->stats.rx_bytes;
3520         }
3521         ovs_mutex_unlock(&ofproto->stats_mutex);
3522     }
3523
3524     return error;
3525 }
3526
3527 static int
3528 port_get_lacp_stats(const struct ofport *ofport_, struct lacp_slave_stats *stats)
3529 {
3530     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3531     if (ofport->bundle && ofport->bundle->lacp) {
3532         if (lacp_get_slave_stats(ofport->bundle->lacp, ofport, stats)) {
3533             return 0;
3534         }
3535     }
3536     return -1;
3537 }
3538
3539 struct port_dump_state {
3540     struct sset_position pos;
3541     bool ghost;
3542
3543     struct ofproto_port port;
3544     bool has_port;
3545 };
3546
3547 static int
3548 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3549 {
3550     *statep = xzalloc(sizeof(struct port_dump_state));
3551     return 0;
3552 }
3553
3554 static int
3555 port_dump_next(const struct ofproto *ofproto_, void *state_,
3556                struct ofproto_port *port)
3557 {
3558     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3559     struct port_dump_state *state = state_;
3560     const struct sset *sset;
3561     struct sset_node *node;
3562
3563     if (state->has_port) {
3564         ofproto_port_destroy(&state->port);
3565         state->has_port = false;
3566     }
3567     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3568     while ((node = sset_at_position(sset, &state->pos))) {
3569         int error;
3570
3571         error = port_query_by_name(ofproto_, node->name, &state->port);
3572         if (!error) {
3573             *port = state->port;
3574             state->has_port = true;
3575             return 0;
3576         } else if (error != ENODEV) {
3577             return error;
3578         }
3579     }
3580
3581     if (!state->ghost) {
3582         state->ghost = true;
3583         memset(&state->pos, 0, sizeof state->pos);
3584         return port_dump_next(ofproto_, state_, port);
3585     }
3586
3587     return EOF;
3588 }
3589
3590 static int
3591 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3592 {
3593     struct port_dump_state *state = state_;
3594
3595     if (state->has_port) {
3596         ofproto_port_destroy(&state->port);
3597     }
3598     free(state);
3599     return 0;
3600 }
3601
3602 static int
3603 port_poll(const struct ofproto *ofproto_, char **devnamep)
3604 {
3605     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3606
3607     if (ofproto->port_poll_errno) {
3608         int error = ofproto->port_poll_errno;
3609         ofproto->port_poll_errno = 0;
3610         return error;
3611     }
3612
3613     if (sset_is_empty(&ofproto->port_poll_set)) {
3614         return EAGAIN;
3615     }
3616
3617     *devnamep = sset_pop(&ofproto->port_poll_set);
3618     return 0;
3619 }
3620
3621 static void
3622 port_poll_wait(const struct ofproto *ofproto_)
3623 {
3624     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3625     dpif_port_poll_wait(ofproto->backer->dpif);
3626 }
3627
3628 static int
3629 port_is_lacp_current(const struct ofport *ofport_)
3630 {
3631     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3632     return (ofport->bundle && ofport->bundle->lacp
3633             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3634             : -1);
3635 }
3636 \f
3637 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3638  * then delete it entirely. */
3639 static void
3640 rule_expire(struct rule_dpif *rule, long long now)
3641     OVS_REQUIRES(ofproto_mutex)
3642 {
3643     uint16_t hard_timeout, idle_timeout;
3644     int reason = -1;
3645
3646     hard_timeout = rule->up.hard_timeout;
3647     idle_timeout = rule->up.idle_timeout;
3648
3649     /* Has 'rule' expired? */
3650     if (hard_timeout) {
3651         long long int modified;
3652
3653         ovs_mutex_lock(&rule->up.mutex);
3654         modified = rule->up.modified;
3655         ovs_mutex_unlock(&rule->up.mutex);
3656
3657         if (now > modified + hard_timeout * 1000) {
3658             reason = OFPRR_HARD_TIMEOUT;
3659         }
3660     }
3661
3662     if (reason < 0 && idle_timeout) {
3663         long long int used;
3664
3665         ovs_mutex_lock(&rule->stats_mutex);
3666         used = rule->stats.used;
3667         ovs_mutex_unlock(&rule->stats_mutex);
3668
3669         if (now > used + idle_timeout * 1000) {
3670             reason = OFPRR_IDLE_TIMEOUT;
3671         }
3672     }
3673
3674     if (reason >= 0) {
3675         COVERAGE_INC(ofproto_dpif_expired);
3676         ofproto_rule_expire(&rule->up, reason);
3677     }
3678 }
3679
3680 static void
3681 ofproto_dpif_set_packet_odp_port(const struct ofproto_dpif *ofproto,
3682                                  ofp_port_t in_port, struct dp_packet *packet)
3683 {
3684     if (in_port == OFPP_NONE) {
3685         in_port = OFPP_LOCAL;
3686     }
3687     packet->md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3688 }
3689
3690 int
3691 ofproto_dpif_execute_actions__(struct ofproto_dpif *ofproto,
3692                                const struct flow *flow,
3693                                struct rule_dpif *rule,
3694                                const struct ofpact *ofpacts, size_t ofpacts_len,
3695                                int indentation, int depth, int resubmits,
3696                                struct dp_packet *packet)
3697 {
3698     struct dpif_flow_stats stats;
3699     struct xlate_out xout;
3700     struct xlate_in xin;
3701     struct dpif_execute execute;
3702     int error;
3703
3704     ovs_assert((rule != NULL) != (ofpacts != NULL));
3705
3706     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3707
3708     if (rule) {
3709         rule_dpif_credit_stats(rule, &stats);
3710     }
3711
3712     uint64_t odp_actions_stub[1024 / 8];
3713     struct ofpbuf odp_actions = OFPBUF_STUB_INITIALIZER(odp_actions_stub);
3714     xlate_in_init(&xin, ofproto, flow, flow->in_port.ofp_port, rule,
3715                   stats.tcp_flags, packet, NULL, &odp_actions);
3716     xin.ofpacts = ofpacts;
3717     xin.ofpacts_len = ofpacts_len;
3718     xin.resubmit_stats = &stats;
3719     xin.indentation = indentation;
3720     xin.depth = depth;
3721     xin.resubmits = resubmits;
3722     if (xlate_actions(&xin, &xout) != XLATE_OK) {
3723         error = EINVAL;
3724         goto out;
3725     }
3726
3727     execute.actions = odp_actions.data;
3728     execute.actions_len = odp_actions.size;
3729
3730     pkt_metadata_from_flow(&packet->md, flow);
3731     execute.packet = packet;
3732     execute.flow = flow;
3733     execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3734     execute.probe = false;
3735     execute.mtu = 0;
3736
3737     /* Fix up in_port. */
3738     ofproto_dpif_set_packet_odp_port(ofproto, flow->in_port.ofp_port, packet);
3739
3740     error = dpif_execute(ofproto->backer->dpif, &execute);
3741 out:
3742     xlate_out_uninit(&xout);
3743     ofpbuf_uninit(&odp_actions);
3744
3745     return error;
3746 }
3747
3748 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3749  * 'flow' must reflect the data in 'packet'. */
3750 int
3751 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3752                              const struct flow *flow,
3753                              struct rule_dpif *rule,
3754                              const struct ofpact *ofpacts, size_t ofpacts_len,
3755                              struct dp_packet *packet)
3756 {
3757     return ofproto_dpif_execute_actions__(ofproto, flow, rule, ofpacts,
3758                                           ofpacts_len, 0, 0, 0, packet);
3759 }
3760
3761 void
3762 rule_dpif_credit_stats(struct rule_dpif *rule,
3763                        const struct dpif_flow_stats *stats)
3764 {
3765     ovs_mutex_lock(&rule->stats_mutex);
3766     if (OVS_UNLIKELY(rule->new_rule)) {
3767         rule_dpif_credit_stats(rule->new_rule, stats);
3768     } else {
3769         rule->stats.n_packets += stats->n_packets;
3770         rule->stats.n_bytes += stats->n_bytes;
3771         rule->stats.used = MAX(rule->stats.used, stats->used);
3772     }
3773     ovs_mutex_unlock(&rule->stats_mutex);
3774 }
3775
3776 ovs_be64
3777 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3778     OVS_REQUIRES(rule->up.mutex)
3779 {
3780     return rule->up.flow_cookie;
3781 }
3782
3783 void
3784 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3785                      uint16_t hard_timeout)
3786 {
3787     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3788 }
3789
3790 /* Returns 'rule''s actions.  The returned actions are RCU-protected, and can
3791  * be read until the calling thread quiesces. */
3792 const struct rule_actions *
3793 rule_dpif_get_actions(const struct rule_dpif *rule)
3794 {
3795     return rule_get_actions(&rule->up);
3796 }
3797
3798 /* Sets 'rule''s recirculation id. */
3799 static void
3800 rule_dpif_set_recirc_id(struct rule_dpif *rule, uint32_t id)
3801     OVS_REQUIRES(rule->up.mutex)
3802 {
3803     ovs_assert(!rule->recirc_id || rule->recirc_id == id);
3804     if (rule->recirc_id == id) {
3805         /* Release the new reference to the same id. */
3806         recirc_free_id(id);
3807     } else {
3808         rule->recirc_id = id;
3809     }
3810 }
3811
3812 /* Sets 'rule''s recirculation id. */
3813 void
3814 rule_set_recirc_id(struct rule *rule_, uint32_t id)
3815 {
3816     struct rule_dpif *rule = rule_dpif_cast(rule_);
3817
3818     ovs_mutex_lock(&rule->up.mutex);
3819     rule_dpif_set_recirc_id(rule, id);
3820     ovs_mutex_unlock(&rule->up.mutex);
3821 }
3822
3823 cls_version_t
3824 ofproto_dpif_get_tables_version(struct ofproto_dpif *ofproto OVS_UNUSED)
3825 {
3826     cls_version_t version;
3827
3828     atomic_read_relaxed(&ofproto->tables_version, &version);
3829
3830     return version;
3831 }
3832
3833 /* The returned rule (if any) is valid at least until the next RCU quiescent
3834  * period.  If the rule needs to stay around longer, the caller should take
3835  * a reference.
3836  *
3837  * 'flow' is non-const to allow for temporary modifications during the lookup.
3838  * Any changes are restored before returning. */
3839 static struct rule_dpif *
3840 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, cls_version_t version,
3841                           uint8_t table_id, struct flow *flow,
3842                           struct flow_wildcards *wc)
3843 {
3844     struct classifier *cls = &ofproto->up.tables[table_id].cls;
3845     return rule_dpif_cast(rule_from_cls_rule(classifier_lookup(cls, version,
3846                                                                flow, wc)));
3847 }
3848
3849 /* Look up 'flow' in 'ofproto''s classifier version 'version', starting from
3850  * table '*table_id'.  Returns the rule that was found, which may be one of the
3851  * special rules according to packet miss hadling.  If 'may_packet_in' is
3852  * false, returning of the miss_rule (which issues packet ins for the
3853  * controller) is avoided.  Updates 'wc', if nonnull, to reflect the fields
3854  * that were used during the lookup.
3855  *
3856  * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3857  * if none is found then the table miss configuration for that table is
3858  * honored, which can result in additional lookups in other OpenFlow tables.
3859  * In this case the function updates '*table_id' to reflect the final OpenFlow
3860  * table that was searched.
3861  *
3862  * If 'honor_table_miss' is false, then only one table lookup occurs, in
3863  * '*table_id'.
3864  *
3865  * The rule is returned in '*rule', which is valid at least until the next
3866  * RCU quiescent period.  If the '*rule' needs to stay around longer, the
3867  * caller must take a reference.
3868  *
3869  * 'in_port' allows the lookup to take place as if the in port had the value
3870  * 'in_port'.  This is needed for resubmit action support.
3871  *
3872  * 'flow' is non-const to allow for temporary modifications during the lookup.
3873  * Any changes are restored before returning. */
3874 struct rule_dpif *
3875 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3876                             cls_version_t version, struct flow *flow,
3877                             struct flow_wildcards *wc,
3878                             const struct dpif_flow_stats *stats,
3879                             uint8_t *table_id, ofp_port_t in_port,
3880                             bool may_packet_in, bool honor_table_miss)
3881 {
3882     ovs_be16 old_tp_src = flow->tp_src, old_tp_dst = flow->tp_dst;
3883     ofp_port_t old_in_port = flow->in_port.ofp_port;
3884     enum ofputil_table_miss miss_config;
3885     struct rule_dpif *rule;
3886     uint8_t next_id;
3887
3888     /* We always unwildcard nw_frag (for IP), so they
3889      * need not be unwildcarded here. */
3890     if (flow->nw_frag & FLOW_NW_FRAG_ANY
3891         && ofproto->up.frag_handling != OFPUTIL_FRAG_NX_MATCH) {
3892         if (ofproto->up.frag_handling == OFPUTIL_FRAG_NORMAL) {
3893             /* We must pretend that transport ports are unavailable. */
3894             flow->tp_src = htons(0);
3895             flow->tp_dst = htons(0);
3896         } else {
3897             /* Must be OFPUTIL_FRAG_DROP (we don't have OFPUTIL_FRAG_REASM).
3898              * Use the drop_frags_rule (which cannot disappear). */
3899             rule = ofproto->drop_frags_rule;
3900             if (stats) {
3901                 struct oftable *tbl = &ofproto->up.tables[*table_id];
3902                 unsigned long orig;
3903
3904                 atomic_add_relaxed(&tbl->n_matched, stats->n_packets, &orig);
3905             }
3906             return rule;
3907         }
3908     }
3909
3910     /* Look up a flow with 'in_port' as the input port.  Then restore the
3911      * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
3912      * have surprising behavior). */
3913     flow->in_port.ofp_port = in_port;
3914
3915     /* Our current implementation depends on n_tables == N_TABLES, and
3916      * TBL_INTERNAL being the last table. */
3917     BUILD_ASSERT_DECL(N_TABLES == TBL_INTERNAL + 1);
3918
3919     miss_config = OFPUTIL_TABLE_MISS_CONTINUE;
3920
3921     for (next_id = *table_id;
3922          next_id < ofproto->up.n_tables;
3923          next_id++, next_id += (next_id == TBL_INTERNAL))
3924     {
3925         *table_id = next_id;
3926         rule = rule_dpif_lookup_in_table(ofproto, version, next_id, flow, wc);
3927         if (stats) {
3928             struct oftable *tbl = &ofproto->up.tables[next_id];
3929             unsigned long orig;
3930
3931             atomic_add_relaxed(rule ? &tbl->n_matched : &tbl->n_missed,
3932                                stats->n_packets, &orig);
3933         }
3934         if (rule) {
3935             goto out;   /* Match. */
3936         }
3937         if (honor_table_miss) {
3938             miss_config = ofproto_table_get_miss_config(&ofproto->up,
3939                                                         *table_id);
3940             if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE) {
3941                 continue;
3942             }
3943         }
3944         break;
3945     }
3946     /* Miss. */
3947     rule = ofproto->no_packet_in_rule;
3948     if (may_packet_in) {
3949         if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE
3950             || miss_config == OFPUTIL_TABLE_MISS_CONTROLLER) {
3951             struct ofport_dpif *port;
3952
3953             port = ofp_port_to_ofport(ofproto, old_in_port);
3954             if (!port) {
3955                 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3956                              old_in_port);
3957             } else if (!(port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN)) {
3958                 rule = ofproto->miss_rule;
3959             }
3960         } else if (miss_config == OFPUTIL_TABLE_MISS_DEFAULT &&
3961                    connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3962             rule = ofproto->miss_rule;
3963         }
3964     }
3965 out:
3966     /* Restore port numbers, as they may have been modified above. */
3967     flow->tp_src = old_tp_src;
3968     flow->tp_dst = old_tp_dst;
3969     /* Restore the old in port. */
3970     flow->in_port.ofp_port = old_in_port;
3971
3972     return rule;
3973 }
3974
3975 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3976 {
3977     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3978 }
3979
3980 static struct rule *
3981 rule_alloc(void)
3982 {
3983     struct rule_dpif *rule = xzalloc(sizeof *rule);
3984     return &rule->up;
3985 }
3986
3987 static void
3988 rule_dealloc(struct rule *rule_)
3989 {
3990     struct rule_dpif *rule = rule_dpif_cast(rule_);
3991     free(rule);
3992 }
3993
3994 static enum ofperr
3995 check_mask(struct ofproto_dpif *ofproto, const struct miniflow *flow)
3996 {
3997     const struct odp_support *support;
3998     uint16_t ct_state, ct_zone;
3999     ovs_u128 ct_label;
4000     uint32_t ct_mark;
4001
4002     support = &ofproto_dpif_get_support(ofproto)->odp;
4003     ct_state = MINIFLOW_GET_U16(flow, ct_state);
4004     if (support->ct_state && support->ct_zone && support->ct_mark
4005         && support->ct_label && support->ct_state_nat) {
4006         return ct_state & CS_UNSUPPORTED_MASK ? OFPERR_OFPBMC_BAD_MASK : 0;
4007     }
4008
4009     ct_zone = MINIFLOW_GET_U16(flow, ct_zone);
4010     ct_mark = MINIFLOW_GET_U32(flow, ct_mark);
4011     ct_label = MINIFLOW_GET_U128(flow, ct_label);
4012
4013     if ((ct_state && !support->ct_state)
4014         || (ct_state & CS_UNSUPPORTED_MASK)
4015         || ((ct_state & (CS_SRC_NAT | CS_DST_NAT)) && !support->ct_state_nat)
4016         || (ct_zone && !support->ct_zone)
4017         || (ct_mark && !support->ct_mark)
4018         || (!ovs_u128_is_zero(ct_label) && !support->ct_label)) {
4019         return OFPERR_OFPBMC_BAD_MASK;
4020     }
4021
4022     return 0;
4023 }
4024
4025 static enum ofperr
4026 check_actions(const struct ofproto_dpif *ofproto,
4027               const struct rule_actions *const actions)
4028 {
4029     const struct ofpact *ofpact;
4030
4031     OFPACT_FOR_EACH (ofpact, actions->ofpacts, actions->ofpacts_len) {
4032         const struct odp_support *support;
4033         const struct ofpact_conntrack *ct;
4034         const struct ofpact *a;
4035
4036         if (ofpact->type != OFPACT_CT) {
4037             continue;
4038         }
4039
4040         ct = CONTAINER_OF(ofpact, struct ofpact_conntrack, ofpact);
4041         support = &ofproto_dpif_get_support(ofproto)->odp;
4042
4043         if (!support->ct_state) {
4044             return OFPERR_OFPBAC_BAD_TYPE;
4045         }
4046         if ((ct->zone_imm || ct->zone_src.field) && !support->ct_zone) {
4047             return OFPERR_OFPBAC_BAD_ARGUMENT;
4048         }
4049
4050         OFPACT_FOR_EACH(a, ct->actions, ofpact_ct_get_action_len(ct)) {
4051             const struct mf_field *dst = ofpact_get_mf_dst(a);
4052
4053             if (a->type == OFPACT_NAT && !support->ct_state_nat) {
4054                 /* The backer doesn't seem to support the NAT bits in
4055                  * 'ct_state': assume that it doesn't support the NAT
4056                  * action. */
4057                 return OFPERR_OFPBAC_BAD_TYPE;
4058             }
4059             if (dst && ((dst->id == MFF_CT_MARK && !support->ct_mark)
4060                         || (dst->id == MFF_CT_LABEL && !support->ct_label))) {
4061                 return OFPERR_OFPBAC_BAD_SET_ARGUMENT;
4062             }
4063         }
4064     }
4065
4066     return 0;
4067 }
4068
4069 static enum ofperr
4070 rule_check(struct rule *rule)
4071 {
4072     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->ofproto);
4073     enum ofperr err;
4074
4075     err = check_mask(ofproto, &rule->cr.match.mask->masks);
4076     if (err) {
4077         return err;
4078     }
4079     return check_actions(ofproto, rule->actions);
4080 }
4081
4082 static enum ofperr
4083 rule_construct(struct rule *rule_)
4084     OVS_NO_THREAD_SAFETY_ANALYSIS
4085 {
4086     struct rule_dpif *rule = rule_dpif_cast(rule_);
4087     int error;
4088
4089     error = rule_check(rule_);
4090     if (error) {
4091         return error;
4092     }
4093
4094     ovs_mutex_init_adaptive(&rule->stats_mutex);
4095     rule->stats.n_packets = 0;
4096     rule->stats.n_bytes = 0;
4097     rule->stats.used = rule->up.modified;
4098     rule->recirc_id = 0;
4099     rule->new_rule = NULL;
4100
4101     return 0;
4102 }
4103
4104 static void
4105 rule_insert(struct rule *rule_, struct rule *old_rule_, bool forward_stats)
4106     OVS_REQUIRES(ofproto_mutex)
4107 {
4108     struct rule_dpif *rule = rule_dpif_cast(rule_);
4109
4110     if (old_rule_ && forward_stats) {
4111         struct rule_dpif *old_rule = rule_dpif_cast(old_rule_);
4112
4113         ovs_assert(!old_rule->new_rule);
4114
4115         /* Take a reference to the new rule, and refer all stats updates from
4116          * the old rule to the new rule. */
4117         rule_dpif_ref(rule);
4118
4119         ovs_mutex_lock(&old_rule->stats_mutex);
4120         ovs_mutex_lock(&rule->stats_mutex);
4121         old_rule->new_rule = rule;       /* Forward future stats. */
4122         rule->stats = old_rule->stats;   /* Transfer stats to the new rule. */
4123         ovs_mutex_unlock(&rule->stats_mutex);
4124         ovs_mutex_unlock(&old_rule->stats_mutex);
4125     }
4126 }
4127
4128 static void
4129 rule_destruct(struct rule *rule_)
4130     OVS_NO_THREAD_SAFETY_ANALYSIS
4131 {
4132     struct rule_dpif *rule = rule_dpif_cast(rule_);
4133
4134     ovs_mutex_destroy(&rule->stats_mutex);
4135     /* Release reference to the new rule, if any. */
4136     if (rule->new_rule) {
4137         rule_dpif_unref(rule->new_rule);
4138     }
4139     if (rule->recirc_id) {
4140         recirc_free_id(rule->recirc_id);
4141     }
4142 }
4143
4144 static void
4145 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
4146                long long int *used)
4147 {
4148     struct rule_dpif *rule = rule_dpif_cast(rule_);
4149
4150     ovs_mutex_lock(&rule->stats_mutex);
4151     if (OVS_UNLIKELY(rule->new_rule)) {
4152         rule_get_stats(&rule->new_rule->up, packets, bytes, used);
4153     } else {
4154         *packets = rule->stats.n_packets;
4155         *bytes = rule->stats.n_bytes;
4156         *used = rule->stats.used;
4157     }
4158     ovs_mutex_unlock(&rule->stats_mutex);
4159 }
4160
4161 static void
4162 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4163                   struct dp_packet *packet)
4164 {
4165     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4166
4167     ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
4168 }
4169
4170 static enum ofperr
4171 rule_execute(struct rule *rule, const struct flow *flow,
4172              struct dp_packet *packet)
4173 {
4174     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4175     dp_packet_delete(packet);
4176     return 0;
4177 }
4178
4179 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4180 {
4181     return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4182 }
4183
4184 static struct ofgroup *
4185 group_alloc(void)
4186 {
4187     struct group_dpif *group = xzalloc(sizeof *group);
4188     return &group->up;
4189 }
4190
4191 static void
4192 group_dealloc(struct ofgroup *group_)
4193 {
4194     struct group_dpif *group = group_dpif_cast(group_);
4195     free(group);
4196 }
4197
4198 static void
4199 group_construct_stats(struct group_dpif *group)
4200     OVS_REQUIRES(group->stats_mutex)
4201 {
4202     struct ofputil_bucket *bucket;
4203     const struct ovs_list *buckets;
4204
4205     group->packet_count = 0;
4206     group->byte_count = 0;
4207
4208     group_dpif_get_buckets(group, &buckets);
4209     LIST_FOR_EACH (bucket, list_node, buckets) {
4210         bucket->stats.packet_count = 0;
4211         bucket->stats.byte_count = 0;
4212     }
4213 }
4214
4215 void
4216 group_dpif_credit_stats(struct group_dpif *group,
4217                         struct ofputil_bucket *bucket,
4218                         const struct dpif_flow_stats *stats)
4219 {
4220     ovs_mutex_lock(&group->stats_mutex);
4221     group->packet_count += stats->n_packets;
4222     group->byte_count += stats->n_bytes;
4223     if (bucket) {
4224         bucket->stats.packet_count += stats->n_packets;
4225         bucket->stats.byte_count += stats->n_bytes;
4226     } else { /* Credit to all buckets */
4227         const struct ovs_list *buckets;
4228
4229         group_dpif_get_buckets(group, &buckets);
4230         LIST_FOR_EACH (bucket, list_node, buckets) {
4231             bucket->stats.packet_count += stats->n_packets;
4232             bucket->stats.byte_count += stats->n_bytes;
4233         }
4234     }
4235     ovs_mutex_unlock(&group->stats_mutex);
4236 }
4237
4238 static enum ofperr
4239 group_construct(struct ofgroup *group_)
4240 {
4241     struct group_dpif *group = group_dpif_cast(group_);
4242
4243     ovs_mutex_init_adaptive(&group->stats_mutex);
4244     ovs_mutex_lock(&group->stats_mutex);
4245     group_construct_stats(group);
4246     ovs_mutex_unlock(&group->stats_mutex);
4247     return 0;
4248 }
4249
4250 static void
4251 group_destruct(struct ofgroup *group_)
4252 {
4253     struct group_dpif *group = group_dpif_cast(group_);
4254     ovs_mutex_destroy(&group->stats_mutex);
4255 }
4256
4257 static enum ofperr
4258 group_modify(struct ofgroup *group_)
4259 {
4260     struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
4261
4262     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4263
4264     return 0;
4265 }
4266
4267 static enum ofperr
4268 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4269 {
4270     struct group_dpif *group = group_dpif_cast(group_);
4271     struct ofputil_bucket *bucket;
4272     const struct ovs_list *buckets;
4273     struct bucket_counter *bucket_stats;
4274
4275     ovs_mutex_lock(&group->stats_mutex);
4276     ogs->packet_count = group->packet_count;
4277     ogs->byte_count = group->byte_count;
4278
4279     group_dpif_get_buckets(group, &buckets);
4280     bucket_stats = ogs->bucket_stats;
4281     LIST_FOR_EACH (bucket, list_node, buckets) {
4282         bucket_stats->packet_count = bucket->stats.packet_count;
4283         bucket_stats->byte_count = bucket->stats.byte_count;
4284         bucket_stats++;
4285     }
4286     ovs_mutex_unlock(&group->stats_mutex);
4287
4288     return 0;
4289 }
4290
4291 /* If the group exists, this function increments the groups's reference count.
4292  *
4293  * Make sure to call group_dpif_unref() after no longer needing to maintain
4294  * a reference to the group. */
4295 bool
4296 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4297                   struct group_dpif **group)
4298 {
4299     struct ofgroup *ofgroup;
4300     bool found;
4301
4302     found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4303     *group = found ?  group_dpif_cast(ofgroup) : NULL;
4304
4305     return found;
4306 }
4307
4308 void
4309 group_dpif_get_buckets(const struct group_dpif *group,
4310                        const struct ovs_list **buckets)
4311 {
4312     *buckets = &group->up.buckets;
4313 }
4314
4315 enum ofp11_group_type
4316 group_dpif_get_type(const struct group_dpif *group)
4317 {
4318     return group->up.type;
4319 }
4320
4321 const char *
4322 group_dpif_get_selection_method(const struct group_dpif *group)
4323 {
4324     return group->up.props.selection_method;
4325 }
4326 \f
4327 /* Sends 'packet' out 'ofport'.
4328  * May modify 'packet'.
4329  * Returns 0 if successful, otherwise a positive errno value. */
4330 int
4331 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct dp_packet *packet)
4332 {
4333     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4334     int error;
4335
4336     error = xlate_send_packet(ofport, packet);
4337
4338     ovs_mutex_lock(&ofproto->stats_mutex);
4339     ofproto->stats.tx_packets++;
4340     ofproto->stats.tx_bytes += dp_packet_size(packet);
4341     ovs_mutex_unlock(&ofproto->stats_mutex);
4342     return error;
4343 }
4344
4345 uint64_t
4346 group_dpif_get_selection_method_param(const struct group_dpif *group)
4347 {
4348     return group->up.props.selection_method_param;
4349 }
4350
4351 const struct field_array *
4352 group_dpif_get_fields(const struct group_dpif *group)
4353 {
4354     return &group->up.props.fields;
4355 }
4356 \f
4357 /* Return the version string of the datapath that backs up
4358  * this 'ofproto'.
4359  */
4360 static const char *
4361 get_datapath_version(const struct ofproto *ofproto_)
4362 {
4363     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4364
4365     return ofproto->backer->dp_version_string;
4366 }
4367
4368 static bool
4369 set_frag_handling(struct ofproto *ofproto_,
4370                   enum ofputil_frag_handling frag_handling)
4371 {
4372     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4373     if (frag_handling != OFPUTIL_FRAG_REASM) {
4374         ofproto->backer->need_revalidate = REV_RECONFIGURE;
4375         return true;
4376     } else {
4377         return false;
4378     }
4379 }
4380
4381 static enum ofperr
4382 packet_out(struct ofproto *ofproto_, struct dp_packet *packet,
4383            const struct flow *flow,
4384            const struct ofpact *ofpacts, size_t ofpacts_len)
4385 {
4386     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4387
4388     ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
4389                                  ofpacts_len, packet);
4390     return 0;
4391 }
4392
4393 static enum ofperr
4394 nxt_resume(struct ofproto *ofproto_,
4395            const struct ofputil_packet_in_private *pin)
4396 {
4397     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4398
4399     /* Translate pin into datapath actions. */
4400     uint64_t odp_actions_stub[1024 / 8];
4401     struct ofpbuf odp_actions = OFPBUF_STUB_INITIALIZER(odp_actions_stub);
4402     enum slow_path_reason slow;
4403     enum ofperr error = xlate_resume(ofproto, pin, &odp_actions, &slow);
4404
4405     /* Steal 'pin->packet' and put it into a dp_packet. */
4406     struct dp_packet packet;
4407     dp_packet_init(&packet, pin->public.packet_len);
4408     dp_packet_put(&packet, pin->public.packet, pin->public.packet_len);
4409
4410     pkt_metadata_from_flow(&packet.md, &pin->public.flow_metadata.flow);
4411
4412     /* Fix up in_port. */
4413     ofproto_dpif_set_packet_odp_port(ofproto,
4414                                      pin->public.flow_metadata.flow.in_port.ofp_port,
4415                                      &packet);
4416
4417     struct flow headers;
4418     flow_extract(&packet, &headers);
4419
4420     /* Execute the datapath actions on the packet. */
4421     struct dpif_execute execute = {
4422         .actions = odp_actions.data,
4423         .actions_len = odp_actions.size,
4424         .needs_help = (slow & SLOW_ACTION) != 0,
4425         .packet = &packet,
4426         .flow = &headers,
4427     };
4428     dpif_execute(ofproto->backer->dpif, &execute);
4429
4430     /* Clean up. */
4431     ofpbuf_uninit(&odp_actions);
4432     dp_packet_uninit(&packet);
4433
4434     return error;
4435 }
4436 \f
4437 /* NetFlow. */
4438
4439 static int
4440 set_netflow(struct ofproto *ofproto_,
4441             const struct netflow_options *netflow_options)
4442 {
4443     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4444
4445     if (netflow_options) {
4446         if (!ofproto->netflow) {
4447             ofproto->netflow = netflow_create();
4448             ofproto->backer->need_revalidate = REV_RECONFIGURE;
4449         }
4450         return netflow_set_options(ofproto->netflow, netflow_options);
4451     } else if (ofproto->netflow) {
4452         ofproto->backer->need_revalidate = REV_RECONFIGURE;
4453         netflow_unref(ofproto->netflow);
4454         ofproto->netflow = NULL;
4455     }
4456
4457     return 0;
4458 }
4459
4460 static void
4461 get_netflow_ids(const struct ofproto *ofproto_,
4462                 uint8_t *engine_type, uint8_t *engine_id)
4463 {
4464     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4465
4466     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
4467 }
4468 \f
4469 static struct ofproto_dpif *
4470 ofproto_dpif_lookup(const char *name)
4471 {
4472     struct ofproto_dpif *ofproto;
4473
4474     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
4475                              hash_string(name, 0), &all_ofproto_dpifs) {
4476         if (!strcmp(ofproto->up.name, name)) {
4477             return ofproto;
4478         }
4479     }
4480     return NULL;
4481 }
4482
4483 static void
4484 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
4485                           const char *argv[], void *aux OVS_UNUSED)
4486 {
4487     struct ofproto_dpif *ofproto;
4488
4489     if (argc > 1) {
4490         ofproto = ofproto_dpif_lookup(argv[1]);
4491         if (!ofproto) {
4492             unixctl_command_reply_error(conn, "no such bridge");
4493             return;
4494         }
4495         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4496         mac_learning_flush(ofproto->ml);
4497         ovs_rwlock_unlock(&ofproto->ml->rwlock);
4498     } else {
4499         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4500             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4501             mac_learning_flush(ofproto->ml);
4502             ovs_rwlock_unlock(&ofproto->ml->rwlock);
4503         }
4504     }
4505
4506     unixctl_command_reply(conn, "table successfully flushed");
4507 }
4508
4509 static void
4510 ofproto_unixctl_mcast_snooping_flush(struct unixctl_conn *conn, int argc,
4511                                      const char *argv[], void *aux OVS_UNUSED)
4512 {
4513     struct ofproto_dpif *ofproto;
4514
4515     if (argc > 1) {
4516         ofproto = ofproto_dpif_lookup(argv[1]);
4517         if (!ofproto) {
4518             unixctl_command_reply_error(conn, "no such bridge");
4519             return;
4520         }
4521
4522         if (!mcast_snooping_enabled(ofproto->ms)) {
4523             unixctl_command_reply_error(conn, "multicast snooping is disabled");
4524             return;
4525         }
4526         mcast_snooping_mdb_flush(ofproto->ms);
4527     } else {
4528         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4529             if (!mcast_snooping_enabled(ofproto->ms)) {
4530                 continue;
4531             }
4532             mcast_snooping_mdb_flush(ofproto->ms);
4533         }
4534     }
4535
4536     unixctl_command_reply(conn, "table successfully flushed");
4537 }
4538
4539 static struct ofport_dpif *
4540 ofbundle_get_a_port(const struct ofbundle *bundle)
4541 {
4542     return CONTAINER_OF(ovs_list_front(&bundle->ports), struct ofport_dpif,
4543                         bundle_node);
4544 }
4545
4546 static void
4547 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4548                          const char *argv[], void *aux OVS_UNUSED)
4549 {
4550     struct ds ds = DS_EMPTY_INITIALIZER;
4551     const struct ofproto_dpif *ofproto;
4552     const struct mac_entry *e;
4553
4554     ofproto = ofproto_dpif_lookup(argv[1]);
4555     if (!ofproto) {
4556         unixctl_command_reply_error(conn, "no such bridge");
4557         return;
4558     }
4559
4560     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
4561     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
4562     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
4563         struct ofbundle *bundle = mac_entry_get_port(ofproto->ml, e);
4564         char name[OFP_MAX_PORT_NAME_LEN];
4565
4566         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4567                                name, sizeof name);
4568         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
4569                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
4570                       mac_entry_age(ofproto->ml, e));
4571     }
4572     ovs_rwlock_unlock(&ofproto->ml->rwlock);
4573     unixctl_command_reply(conn, ds_cstr(&ds));
4574     ds_destroy(&ds);
4575 }
4576
4577 static void
4578 ofproto_unixctl_mcast_snooping_show(struct unixctl_conn *conn,
4579                                     int argc OVS_UNUSED,
4580                                     const char *argv[],
4581                                     void *aux OVS_UNUSED)
4582 {
4583     struct ds ds = DS_EMPTY_INITIALIZER;
4584     const struct ofproto_dpif *ofproto;
4585     const struct ofbundle *bundle;
4586     const struct mcast_group *grp;
4587     struct mcast_group_bundle *b;
4588     struct mcast_mrouter_bundle *mrouter;
4589
4590     ofproto = ofproto_dpif_lookup(argv[1]);
4591     if (!ofproto) {
4592         unixctl_command_reply_error(conn, "no such bridge");
4593         return;
4594     }
4595
4596     if (!mcast_snooping_enabled(ofproto->ms)) {
4597         unixctl_command_reply_error(conn, "multicast snooping is disabled");
4598         return;
4599     }
4600
4601     ds_put_cstr(&ds, " port  VLAN  GROUP                Age\n");
4602     ovs_rwlock_rdlock(&ofproto->ms->rwlock);
4603     LIST_FOR_EACH (grp, group_node, &ofproto->ms->group_lru) {
4604         LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
4605             char name[OFP_MAX_PORT_NAME_LEN];
4606
4607             bundle = b->port;
4608             ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4609                                    name, sizeof name);
4610             ds_put_format(&ds, "%5s  %4d  ", name, grp->vlan);
4611             ipv6_format_mapped(&grp->addr, &ds);
4612             ds_put_format(&ds, "         %3d\n",
4613                           mcast_bundle_age(ofproto->ms, b));
4614         }
4615     }
4616
4617     /* ports connected to multicast routers */
4618     LIST_FOR_EACH(mrouter, mrouter_node, &ofproto->ms->mrouter_lru) {
4619         char name[OFP_MAX_PORT_NAME_LEN];
4620
4621         bundle = mrouter->port;
4622         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4623                                name, sizeof name);
4624         ds_put_format(&ds, "%5s  %4d  querier             %3d\n",
4625                       name, mrouter->vlan,
4626                       mcast_mrouter_age(ofproto->ms, mrouter));
4627     }
4628     ovs_rwlock_unlock(&ofproto->ms->rwlock);
4629     unixctl_command_reply(conn, ds_cstr(&ds));
4630     ds_destroy(&ds);
4631 }
4632
4633 struct trace_ctx {
4634     struct xlate_out xout;
4635     struct xlate_in xin;
4636     const struct flow *key;
4637     struct flow flow;
4638     struct ds *result;
4639     struct flow_wildcards wc;
4640     struct ofpbuf odp_actions;
4641 };
4642
4643 static void
4644 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
4645 {
4646     const struct rule_actions *actions;
4647     ovs_be64 cookie;
4648
4649     ds_put_char_multiple(result, '\t', level);
4650     if (!rule) {
4651         ds_put_cstr(result, "No match\n");
4652         return;
4653     }
4654
4655     ovs_mutex_lock(&rule->up.mutex);
4656     cookie = rule->up.flow_cookie;
4657     ovs_mutex_unlock(&rule->up.mutex);
4658
4659     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
4660                   rule ? rule->up.table_id : 0, ntohll(cookie));
4661     cls_rule_format(&rule->up.cr, result);
4662     ds_put_char(result, '\n');
4663
4664     actions = rule_dpif_get_actions(rule);
4665
4666     ds_put_char_multiple(result, '\t', level);
4667     ds_put_cstr(result, "OpenFlow actions=");
4668     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
4669     ds_put_char(result, '\n');
4670 }
4671
4672 static void
4673 trace_format_flow(struct ds *result, int level, const char *title,
4674                   struct trace_ctx *trace)
4675 {
4676     ds_put_char_multiple(result, '\t', level);
4677     ds_put_format(result, "%s: ", title);
4678     /* Do not report unchanged flows for resubmits. */
4679     if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
4680         || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
4681         ds_put_cstr(result, "unchanged");
4682     } else {
4683         flow_format(result, &trace->xin.flow);
4684         trace->flow = trace->xin.flow;
4685     }
4686     ds_put_char(result, '\n');
4687 }
4688
4689 static void
4690 trace_format_regs(struct ds *result, int level, const char *title,
4691                   struct trace_ctx *trace)
4692 {
4693     size_t i;
4694
4695     ds_put_char_multiple(result, '\t', level);
4696     ds_put_format(result, "%s:", title);
4697     for (i = 0; i < FLOW_N_REGS; i++) {
4698         ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
4699     }
4700     ds_put_char(result, '\n');
4701 }
4702
4703 static void
4704 trace_format_odp(struct ds *result, int level, const char *title,
4705                  struct trace_ctx *trace)
4706 {
4707     struct ofpbuf *odp_actions = &trace->odp_actions;
4708
4709     ds_put_char_multiple(result, '\t', level);
4710     ds_put_format(result, "%s: ", title);
4711     format_odp_actions(result, odp_actions->data, odp_actions->size);
4712     ds_put_char(result, '\n');
4713 }
4714
4715 static void
4716 trace_format_megaflow(struct ds *result, int level, const char *title,
4717                       struct trace_ctx *trace)
4718 {
4719     struct match match;
4720
4721     ds_put_char_multiple(result, '\t', level);
4722     ds_put_format(result, "%s: ", title);
4723     match_init(&match, trace->key, &trace->wc);
4724     match_format(&match, result, OFP_DEFAULT_PRIORITY);
4725     ds_put_char(result, '\n');
4726 }
4727
4728 static void trace_report(struct xlate_in *, int indentation,
4729                          const char *format, ...)
4730     OVS_PRINTF_FORMAT(3, 4);
4731 static void trace_report_valist(struct xlate_in *, int indentation,
4732                                 const char *format, va_list args)
4733     OVS_PRINTF_FORMAT(3, 0);
4734
4735 static void
4736 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int indentation)
4737 {
4738     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4739     struct ds *result = trace->result;
4740
4741     if (!indentation) {
4742         if (rule == xin->ofproto->miss_rule) {
4743             trace_report(xin, indentation,
4744                          "No match, flow generates \"packet in\"s.");
4745         } else if (rule == xin->ofproto->no_packet_in_rule) {
4746             trace_report(xin, indentation, "No match, packets dropped because "
4747                          "OFPPC_NO_PACKET_IN is set on in_port.");
4748         } else if (rule == xin->ofproto->drop_frags_rule) {
4749             trace_report(xin, indentation,
4750                          "Packets dropped because they are IP fragments and "
4751                          "the fragment handling mode is \"drop\".");
4752         }
4753     }
4754
4755     ds_put_char(result, '\n');
4756     if (indentation) {
4757         trace_format_flow(result, indentation, "Resubmitted flow", trace);
4758         trace_format_regs(result, indentation, "Resubmitted regs", trace);
4759         trace_format_odp(result,  indentation, "Resubmitted  odp", trace);
4760         trace_format_megaflow(result, indentation, "Resubmitted megaflow",
4761                               trace);
4762     }
4763     trace_format_rule(result, indentation, rule);
4764 }
4765
4766 static void
4767 trace_report_valist(struct xlate_in *xin, int indentation,
4768                     const char *format, va_list args)
4769 {
4770     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4771     struct ds *result = trace->result;
4772
4773     ds_put_char_multiple(result, '\t', indentation);
4774     ds_put_format_valist(result, format, args);
4775     ds_put_char(result, '\n');
4776 }
4777
4778 static void
4779 trace_report(struct xlate_in *xin, int indentation, const char *format, ...)
4780 {
4781     va_list args;
4782
4783     va_start(args, format);
4784     trace_report_valist(xin, indentation, format, args);
4785     va_end(args);
4786 }
4787
4788 /* Parses the 'argc' elements of 'argv', ignoring argv[0].  The following
4789  * forms are supported:
4790  *
4791  *     - [dpname] odp_flow [-generate | packet]
4792  *     - bridge br_flow [-generate | packet]
4793  *
4794  * On success, initializes '*ofprotop' and 'flow' and returns NULL.  On failure
4795  * returns a nonnull malloced error message. */
4796 static char * OVS_WARN_UNUSED_RESULT
4797 parse_flow_and_packet(int argc, const char *argv[],
4798                       struct ofproto_dpif **ofprotop, struct flow *flow,
4799                       struct dp_packet **packetp)
4800 {
4801     const struct dpif_backer *backer = NULL;
4802     const char *error = NULL;
4803     char *m_err = NULL;
4804     struct simap port_names = SIMAP_INITIALIZER(&port_names);
4805     struct dp_packet *packet;
4806     struct ofpbuf odp_key;
4807     struct ofpbuf odp_mask;
4808
4809     ofpbuf_init(&odp_key, 0);
4810     ofpbuf_init(&odp_mask, 0);
4811
4812     /* Handle "-generate" or a hex string as the last argument. */
4813     if (!strcmp(argv[argc - 1], "-generate")) {
4814         packet = dp_packet_new(0);
4815         argc--;
4816     } else {
4817         error = eth_from_hex(argv[argc - 1], &packet);
4818         if (!error) {
4819             argc--;
4820         } else if (argc == 4) {
4821             /* The 3-argument form must end in "-generate' or a hex string. */
4822             goto exit;
4823         }
4824         error = NULL;
4825     }
4826
4827     /* odp_flow can have its in_port specified as a name instead of port no.
4828      * We do not yet know whether a given flow is a odp_flow or a br_flow.
4829      * But, to know whether a flow is odp_flow through odp_flow_from_string(),
4830      * we need to create a simap of name to port no. */
4831     if (argc == 3) {
4832         const char *dp_type;
4833         if (!strncmp(argv[1], "ovs-", 4)) {
4834             dp_type = argv[1] + 4;
4835         } else {
4836             dp_type = argv[1];
4837         }
4838         backer = shash_find_data(&all_dpif_backers, dp_type);
4839     } else if (argc == 2) {
4840         struct shash_node *node;
4841         if (shash_count(&all_dpif_backers) == 1) {
4842             node = shash_first(&all_dpif_backers);
4843             backer = node->data;
4844         }
4845     } else {
4846         error = "Syntax error";
4847         goto exit;
4848     }
4849     if (backer && backer->dpif) {
4850         struct dpif_port dpif_port;
4851         struct dpif_port_dump port_dump;
4852         DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4853             simap_put(&port_names, dpif_port.name,
4854                       odp_to_u32(dpif_port.port_no));
4855         }
4856     }
4857
4858     /* Parse the flow and determine whether a datapath or
4859      * bridge is specified. If function odp_flow_key_from_string()
4860      * returns 0, the flow is a odp_flow. If function
4861      * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4862     if (!odp_flow_from_string(argv[argc - 1], &port_names,
4863                               &odp_key, &odp_mask)) {
4864         if (!backer) {
4865             error = "Cannot find the datapath";
4866             goto exit;
4867         }
4868
4869         if (odp_flow_key_to_flow(odp_key.data, odp_key.size, flow) == ODP_FIT_ERROR) {
4870             error = "Failed to parse datapath flow key";
4871             goto exit;
4872         }
4873
4874         *ofprotop = xlate_lookup_ofproto(backer, flow,
4875                                          &flow->in_port.ofp_port);
4876         if (*ofprotop == NULL) {
4877             error = "Invalid datapath flow";
4878             goto exit;
4879         }
4880     } else {
4881         char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4882
4883         if (err) {
4884             m_err = xasprintf("Bad openflow flow syntax: %s", err);
4885             free(err);
4886             goto exit;
4887         } else {
4888             if (argc != 3) {
4889                 error = "Must specify bridge name";
4890                 goto exit;
4891             }
4892
4893             *ofprotop = ofproto_dpif_lookup(argv[1]);
4894             if (!*ofprotop) {
4895                 error = "Unknown bridge name";
4896                 goto exit;
4897             }
4898         }
4899     }
4900
4901     /* Generate a packet, if requested. */
4902     if (packet) {
4903         if (!dp_packet_size(packet)) {
4904             flow_compose(packet, flow);
4905         } else {
4906             /* Use the metadata from the flow and the packet argument
4907              * to reconstruct the flow. */
4908             pkt_metadata_from_flow(&packet->md, flow);
4909             flow_extract(packet, flow);
4910         }
4911     }
4912
4913 exit:
4914     if (error && !m_err) {
4915         m_err = xstrdup(error);
4916     }
4917     if (m_err) {
4918         dp_packet_delete(packet);
4919         packet = NULL;
4920     }
4921     *packetp = packet;
4922     ofpbuf_uninit(&odp_key);
4923     ofpbuf_uninit(&odp_mask);
4924     simap_destroy(&port_names);
4925     return m_err;
4926 }
4927
4928 static void
4929 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4930                       void *aux OVS_UNUSED)
4931 {
4932     struct ofproto_dpif *ofproto;
4933     struct dp_packet *packet;
4934     char *error;
4935     struct flow flow;
4936
4937     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4938     if (!error) {
4939         struct ds result;
4940
4941         ds_init(&result);
4942         ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4943         unixctl_command_reply(conn, ds_cstr(&result));
4944         ds_destroy(&result);
4945         dp_packet_delete(packet);
4946     } else {
4947         unixctl_command_reply_error(conn, error);
4948         free(error);
4949     }
4950 }
4951
4952 static void
4953 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4954                               const char *argv[], void *aux OVS_UNUSED)
4955 {
4956     enum ofputil_protocol usable_protocols;
4957     struct ofproto_dpif *ofproto;
4958     bool enforce_consistency;
4959     struct ofpbuf ofpacts;
4960     struct dp_packet *packet;
4961     struct ds result;
4962     struct flow flow;
4963     uint16_t in_port;
4964
4965     /* Three kinds of error return values! */
4966     enum ofperr retval;
4967     char *error;
4968
4969     packet = NULL;
4970     ds_init(&result);
4971     ofpbuf_init(&ofpacts, 0);
4972
4973     /* Parse actions. */
4974     error = ofpacts_parse_actions(argv[--argc], &ofpacts, &usable_protocols);
4975     if (error) {
4976         unixctl_command_reply_error(conn, error);
4977         free(error);
4978         goto exit;
4979     }
4980
4981     /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4982      * consistency between the flow and the actions.  With -consistent, we
4983      * enforce consistency even for a flow supported in OpenFlow 1.0. */
4984     if (!strcmp(argv[1], "-consistent")) {
4985         enforce_consistency = true;
4986         argv++;
4987         argc--;
4988     } else {
4989         enforce_consistency = false;
4990     }
4991
4992     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4993     if (error) {
4994         unixctl_command_reply_error(conn, error);
4995         free(error);
4996         goto exit;
4997     }
4998
4999     /* Do the same checks as handle_packet_out() in ofproto.c.
5000      *
5001      * We pass a 'table_id' of 0 to ofpacts_check(), which isn't
5002      * strictly correct because these actions aren't in any table, but it's OK
5003      * because it 'table_id' is used only to check goto_table instructions, but
5004      * packet-outs take a list of actions and therefore it can't include
5005      * instructions.
5006      *
5007      * We skip the "meter" check here because meter is an instruction, not an
5008      * action, and thus cannot appear in ofpacts. */
5009     in_port = ofp_to_u16(flow.in_port.ofp_port);
5010     if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
5011         unixctl_command_reply_error(conn, "invalid in_port");
5012         goto exit;
5013     }
5014     if (enforce_consistency) {
5015         retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size, &flow,
5016                                            u16_to_ofp(ofproto->up.max_ports),
5017                                            0, ofproto->up.n_tables,
5018                                            usable_protocols);
5019     } else {
5020         retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow,
5021                                u16_to_ofp(ofproto->up.max_ports), 0,
5022                                ofproto->up.n_tables, &usable_protocols);
5023     }
5024     if (!retval) {
5025         retval = ofproto_check_ofpacts(&ofproto->up, ofpacts.data,
5026                                        ofpacts.size);
5027     }
5028
5029     if (retval) {
5030         ds_clear(&result);
5031         ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
5032         unixctl_command_reply_error(conn, ds_cstr(&result));
5033         goto exit;
5034     }
5035
5036     ofproto_trace(ofproto, &flow, packet,
5037                   ofpacts.data, ofpacts.size, &result);
5038     unixctl_command_reply(conn, ds_cstr(&result));
5039
5040 exit:
5041     ds_destroy(&result);
5042     dp_packet_delete(packet);
5043     ofpbuf_uninit(&ofpacts);
5044 }
5045
5046 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
5047  * description of the results to 'ds'.
5048  *
5049  * The trace follows a packet with the specified 'flow' through the flow
5050  * table.  'packet' may be nonnull to trace an actual packet, with consequent
5051  * side effects (if it is nonnull then its flow must be 'flow').
5052  *
5053  * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
5054  * trace, otherwise the actions are determined by a flow table lookup. */
5055 static void
5056 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
5057               const struct dp_packet *packet,
5058               const struct ofpact ofpacts[], size_t ofpacts_len,
5059               struct ds *ds)
5060 {
5061     struct trace_ctx trace;
5062     enum xlate_error error;
5063
5064     ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
5065     ds_put_cstr(ds, "Flow: ");
5066     flow_format(ds, flow);
5067     ds_put_char(ds, '\n');
5068
5069     ofpbuf_init(&trace.odp_actions, 0);
5070
5071     trace.result = ds;
5072     trace.key = flow; /* Original flow key, used for megaflow. */
5073     trace.flow = *flow; /* May be modified by actions. */
5074     xlate_in_init(&trace.xin, ofproto, flow, flow->in_port.ofp_port, NULL,
5075                   ntohs(flow->tcp_flags), packet, &trace.wc,
5076                   &trace.odp_actions);
5077     trace.xin.ofpacts = ofpacts;
5078     trace.xin.ofpacts_len = ofpacts_len;
5079     trace.xin.resubmit_hook = trace_resubmit;
5080     trace.xin.report_hook = trace_report_valist;
5081
5082     error = xlate_actions(&trace.xin, &trace.xout);
5083     ds_put_char(ds, '\n');
5084     trace.xin.flow.actset_output = 0;
5085     trace_format_flow(ds, 0, "Final flow", &trace);
5086     trace_format_megaflow(ds, 0, "Megaflow", &trace);
5087
5088     ds_put_cstr(ds, "Datapath actions: ");
5089     format_odp_actions(ds, trace.odp_actions.data, trace.odp_actions.size);
5090
5091     if (error != XLATE_OK) {
5092         ds_put_format(ds, "\nTranslation failed (%s), packet is dropped.\n",
5093                       xlate_strerror(error));
5094     } else if (trace.xout.slow) {
5095         enum slow_path_reason slow;
5096
5097         ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5098                     "slow path because it:");
5099
5100         slow = trace.xout.slow;
5101         while (slow) {
5102             enum slow_path_reason bit = rightmost_1bit(slow);
5103
5104             ds_put_format(ds, "\n\t- %s.",
5105                           slow_path_reason_to_explanation(bit));
5106
5107             slow &= ~bit;
5108         }
5109     }
5110
5111     xlate_out_uninit(&trace.xout);
5112     ofpbuf_uninit(&trace.odp_actions);
5113 }
5114
5115 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
5116  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
5117  * to destroy 'ofproto_shash' and free the returned value. */
5118 static const struct shash_node **
5119 get_ofprotos(struct shash *ofproto_shash)
5120 {
5121     const struct ofproto_dpif *ofproto;
5122
5123     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5124         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5125         shash_add_nocopy(ofproto_shash, name, ofproto);
5126     }
5127
5128     return shash_sort(ofproto_shash);
5129 }
5130
5131 static void
5132 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5133                               const char *argv[] OVS_UNUSED,
5134                               void *aux OVS_UNUSED)
5135 {
5136     struct ds ds = DS_EMPTY_INITIALIZER;
5137     struct shash ofproto_shash;
5138     const struct shash_node **sorted_ofprotos;
5139     int i;
5140
5141     shash_init(&ofproto_shash);
5142     sorted_ofprotos = get_ofprotos(&ofproto_shash);
5143     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5144         const struct shash_node *node = sorted_ofprotos[i];
5145         ds_put_format(&ds, "%s\n", node->name);
5146     }
5147
5148     shash_destroy(&ofproto_shash);
5149     free(sorted_ofprotos);
5150
5151     unixctl_command_reply(conn, ds_cstr(&ds));
5152     ds_destroy(&ds);
5153 }
5154
5155 static void
5156 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5157 {
5158     const struct shash_node **ofprotos;
5159     struct dpif_dp_stats dp_stats;
5160     struct shash ofproto_shash;
5161     size_t i;
5162
5163     dpif_get_dp_stats(backer->dpif, &dp_stats);
5164
5165     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5166                   dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
5167
5168     shash_init(&ofproto_shash);
5169     ofprotos = get_ofprotos(&ofproto_shash);
5170     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5171         struct ofproto_dpif *ofproto = ofprotos[i]->data;
5172         const struct shash_node **ports;
5173         size_t j;
5174
5175         if (ofproto->backer != backer) {
5176             continue;
5177         }
5178
5179         ds_put_format(ds, "\t%s:\n", ofproto->up.name);
5180
5181         ports = shash_sort(&ofproto->up.port_by_name);
5182         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5183             const struct shash_node *node = ports[j];
5184             struct ofport *ofport = node->data;
5185             struct smap config;
5186             odp_port_t odp_port;
5187
5188             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5189                           ofport->ofp_port);
5190
5191             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5192             if (odp_port != ODPP_NONE) {
5193                 ds_put_format(ds, "%"PRIu32":", odp_port);
5194             } else {
5195                 ds_put_cstr(ds, "none:");
5196             }
5197
5198             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5199
5200             smap_init(&config);
5201             if (!netdev_get_config(ofport->netdev, &config)) {
5202                 const struct smap_node **nodes;
5203                 size_t i;
5204
5205                 nodes = smap_sort(&config);
5206                 for (i = 0; i < smap_count(&config); i++) {
5207                     const struct smap_node *node = nodes[i];
5208                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5209                                   node->key, node->value);
5210                 }
5211                 free(nodes);
5212             }
5213             smap_destroy(&config);
5214
5215             ds_put_char(ds, ')');
5216             ds_put_char(ds, '\n');
5217         }
5218         free(ports);
5219     }
5220     shash_destroy(&ofproto_shash);
5221     free(ofprotos);
5222 }
5223
5224 static void
5225 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5226                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5227 {
5228     struct ds ds = DS_EMPTY_INITIALIZER;
5229     const struct shash_node **backers;
5230     int i;
5231
5232     backers = shash_sort(&all_dpif_backers);
5233     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5234         dpif_show_backer(backers[i]->data, &ds);
5235     }
5236     free(backers);
5237
5238     unixctl_command_reply(conn, ds_cstr(&ds));
5239     ds_destroy(&ds);
5240 }
5241
5242 static void
5243 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5244                                 int argc OVS_UNUSED, const char *argv[],
5245                                 void *aux OVS_UNUSED)
5246 {
5247     const struct ofproto_dpif *ofproto;
5248
5249     struct ds ds = DS_EMPTY_INITIALIZER;
5250     bool verbosity = false;
5251
5252     struct dpif_port dpif_port;
5253     struct dpif_port_dump port_dump;
5254     struct hmap portno_names;
5255
5256     struct dpif_flow_dump *flow_dump;
5257     struct dpif_flow_dump_thread *flow_dump_thread;
5258     struct dpif_flow f;
5259     int error;
5260
5261     ofproto = ofproto_dpif_lookup(argv[argc - 1]);
5262     if (!ofproto) {
5263         unixctl_command_reply_error(conn, "no such bridge");
5264         return;
5265     }
5266
5267     if (argc > 2 && !strcmp(argv[1], "-m")) {
5268         verbosity = true;
5269     }
5270
5271     hmap_init(&portno_names);
5272     DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
5273         odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
5274     }
5275
5276     ds_init(&ds);
5277     flow_dump = dpif_flow_dump_create(ofproto->backer->dpif, false);
5278     flow_dump_thread = dpif_flow_dump_thread_create(flow_dump);
5279     while (dpif_flow_dump_next(flow_dump_thread, &f, 1)) {
5280         struct flow flow;
5281
5282         if (odp_flow_key_to_flow(f.key, f.key_len, &flow) == ODP_FIT_ERROR
5283             || xlate_lookup_ofproto(ofproto->backer, &flow, NULL) != ofproto) {
5284             continue;
5285         }
5286
5287         if (verbosity) {
5288             odp_format_ufid(&f.ufid, &ds);
5289             ds_put_cstr(&ds, " ");
5290         }
5291         odp_flow_format(f.key, f.key_len, f.mask, f.mask_len,
5292                         &portno_names, &ds, verbosity);
5293         ds_put_cstr(&ds, ", ");
5294         dpif_flow_stats_format(&f.stats, &ds);
5295         ds_put_cstr(&ds, ", actions:");
5296         format_odp_actions(&ds, f.actions, f.actions_len);
5297         ds_put_char(&ds, '\n');
5298     }
5299     dpif_flow_dump_thread_destroy(flow_dump_thread);
5300     error = dpif_flow_dump_destroy(flow_dump);
5301
5302     if (error) {
5303         ds_clear(&ds);
5304         ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
5305         unixctl_command_reply_error(conn, ds_cstr(&ds));
5306     } else {
5307         unixctl_command_reply(conn, ds_cstr(&ds));
5308     }
5309     odp_portno_names_destroy(&portno_names);
5310     hmap_destroy(&portno_names);
5311     ds_destroy(&ds);
5312 }
5313
5314 static void
5315 ofproto_revalidate_all_backers(void)
5316 {
5317     const struct shash_node **backers;
5318     int i;
5319
5320     backers = shash_sort(&all_dpif_backers);
5321     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5322         struct dpif_backer *backer = backers[i]->data;
5323         backer->need_revalidate = REV_RECONFIGURE;
5324     }
5325     free(backers);
5326 }
5327
5328 static void
5329 disable_tnl_push_pop(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5330                      const char *argv[], void *aux OVS_UNUSED)
5331 {
5332     if (!strcasecmp(argv[1], "off")) {
5333         ofproto_use_tnl_push_pop = false;
5334         unixctl_command_reply(conn, "Tunnel push-pop off");
5335         ofproto_revalidate_all_backers();
5336     } else if (!strcasecmp(argv[1], "on")) {
5337         ofproto_use_tnl_push_pop = true;
5338         unixctl_command_reply(conn, "Tunnel push-pop on");
5339         ofproto_revalidate_all_backers();
5340     } else {
5341         unixctl_command_reply_error(conn, "Invalid argument");
5342     }
5343 }
5344
5345 static void
5346 ofproto_unixctl_init(void)
5347 {
5348     static bool registered;
5349     if (registered) {
5350         return;
5351     }
5352     registered = true;
5353
5354     unixctl_command_register(
5355         "ofproto/trace",
5356         "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
5357         1, 3, ofproto_unixctl_trace, NULL);
5358     unixctl_command_register(
5359         "ofproto/trace-packet-out",
5360         "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
5361         2, 6, ofproto_unixctl_trace_actions, NULL);
5362     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5363                              ofproto_unixctl_fdb_flush, NULL);
5364     unixctl_command_register("fdb/show", "bridge", 1, 1,
5365                              ofproto_unixctl_fdb_show, NULL);
5366     unixctl_command_register("mdb/flush", "[bridge]", 0, 1,
5367                              ofproto_unixctl_mcast_snooping_flush, NULL);
5368     unixctl_command_register("mdb/show", "bridge", 1, 1,
5369                              ofproto_unixctl_mcast_snooping_show, NULL);
5370     unixctl_command_register("dpif/dump-dps", "", 0, 0,
5371                              ofproto_unixctl_dpif_dump_dps, NULL);
5372     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5373                              NULL);
5374     unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
5375                              ofproto_unixctl_dpif_dump_flows, NULL);
5376
5377     unixctl_command_register("ofproto/tnl-push-pop", "[on]|[off]", 1, 1,
5378                              disable_tnl_push_pop, NULL);
5379 }
5380
5381 /* Returns true if 'table' is the table used for internal rules,
5382  * false otherwise. */
5383 bool
5384 table_is_internal(uint8_t table_id)
5385 {
5386     return table_id == TBL_INTERNAL;
5387 }
5388 \f
5389
5390 static odp_port_t
5391 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
5392 {
5393     const struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
5394     return ofport ? ofport->odp_port : ODPP_NONE;
5395 }
5396
5397 struct ofport_dpif *
5398 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
5399 {
5400     struct ofport_dpif *port;
5401
5402     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
5403     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
5404                              &backer->odp_to_ofport_map) {
5405         if (port->odp_port == odp_port) {
5406             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5407             return port;
5408         }
5409     }
5410
5411     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5412     return NULL;
5413 }
5414
5415 static ofp_port_t
5416 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
5417 {
5418     struct ofport_dpif *port;
5419
5420     port = odp_port_to_ofport(ofproto->backer, odp_port);
5421     if (port && &ofproto->up == port->up.ofproto) {
5422         return port->up.ofp_port;
5423     } else {
5424         return OFPP_NONE;
5425     }
5426 }
5427
5428 int
5429 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
5430                                const struct match *match, int priority,
5431                                uint16_t idle_timeout,
5432                                const struct ofpbuf *ofpacts,
5433                                struct rule **rulep)
5434 {
5435     struct ofproto_flow_mod ofm;
5436     struct rule_dpif *rule;
5437     int error;
5438
5439     ofm.fm = (struct ofputil_flow_mod) {
5440         .match = *match,
5441         .priority = priority,
5442         .table_id = TBL_INTERNAL,
5443         .command = OFPFC_ADD,
5444         .idle_timeout = idle_timeout,
5445         .flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY,
5446         .ofpacts = ofpacts->data,
5447         .ofpacts_len = ofpacts->size,
5448         .delete_reason = OVS_OFPRR_NONE,
5449     };
5450
5451     error = ofproto_flow_mod(&ofproto->up, &ofm);
5452     if (error) {
5453         VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
5454                     ofperr_to_string(error));
5455         *rulep = NULL;
5456         return error;
5457     }
5458
5459     rule = rule_dpif_lookup_in_table(ofproto,
5460                                      ofproto_dpif_get_tables_version(ofproto),
5461                                      TBL_INTERNAL, &ofm.fm.match.flow,
5462                                      &ofm.fm.match.wc);
5463     if (rule) {
5464         *rulep = &rule->up;
5465     } else {
5466         OVS_NOT_REACHED();
5467     }
5468     return 0;
5469 }
5470
5471 int
5472 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
5473                                   struct match *match, int priority)
5474 {
5475     struct ofproto_flow_mod ofm;
5476     int error;
5477
5478     ofm.fm = (struct ofputil_flow_mod) {
5479         .match = *match,
5480         .priority = priority,
5481         .table_id = TBL_INTERNAL,
5482         .flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY,
5483         .command = OFPFC_DELETE_STRICT,
5484     };
5485
5486     error = ofproto_flow_mod(&ofproto->up, &ofm);
5487     if (error) {
5488         VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
5489                     ofperr_to_string(error));
5490         return error;
5491     }
5492
5493     return 0;
5494 }
5495
5496 const struct uuid *
5497 ofproto_dpif_get_uuid(const struct ofproto_dpif *ofproto)
5498 {
5499     return &ofproto->uuid;
5500 }
5501
5502 const struct ofproto_class ofproto_dpif_class = {
5503     init,
5504     enumerate_types,
5505     enumerate_names,
5506     del,
5507     port_open_type,
5508     type_run,
5509     type_wait,
5510     alloc,
5511     construct,
5512     destruct,
5513     dealloc,
5514     run,
5515     ofproto_dpif_wait,
5516     NULL,                       /* get_memory_usage. */
5517     type_get_memory_usage,
5518     flush,
5519     query_tables,
5520     set_tables_version,
5521     port_alloc,
5522     port_construct,
5523     port_destruct,
5524     port_dealloc,
5525     port_modified,
5526     port_reconfigured,
5527     port_query_by_name,
5528     port_add,
5529     port_del,
5530     port_get_stats,
5531     port_dump_start,
5532     port_dump_next,
5533     port_dump_done,
5534     port_poll,
5535     port_poll_wait,
5536     port_is_lacp_current,
5537     port_get_lacp_stats,
5538     NULL,                       /* rule_choose_table */
5539     rule_alloc,
5540     rule_construct,
5541     rule_insert,
5542     NULL,                       /* rule_delete */
5543     rule_destruct,
5544     rule_dealloc,
5545     rule_get_stats,
5546     rule_execute,
5547     set_frag_handling,
5548     packet_out,
5549     nxt_resume,
5550     set_netflow,
5551     get_netflow_ids,
5552     set_sflow,
5553     set_ipfix,
5554     get_ipfix_stats,
5555     set_cfm,
5556     cfm_status_changed,
5557     get_cfm_status,
5558     set_lldp,
5559     get_lldp_status,
5560     set_aa,
5561     aa_mapping_set,
5562     aa_mapping_unset,
5563     aa_vlan_get_queued,
5564     aa_vlan_get_queue_size,
5565     set_bfd,
5566     bfd_status_changed,
5567     get_bfd_status,
5568     set_stp,
5569     get_stp_status,
5570     set_stp_port,
5571     get_stp_port_status,
5572     get_stp_port_stats,
5573     set_rstp,
5574     get_rstp_status,
5575     set_rstp_port,
5576     get_rstp_port_status,
5577     set_queues,
5578     bundle_set,
5579     bundle_remove,
5580     mirror_set__,
5581     mirror_get_stats__,
5582     set_flood_vlans,
5583     is_mirror_output_bundle,
5584     forward_bpdu_changed,
5585     set_mac_table_config,
5586     set_mcast_snooping,
5587     set_mcast_snooping_port,
5588     NULL,                       /* meter_get_features */
5589     NULL,                       /* meter_set */
5590     NULL,                       /* meter_get */
5591     NULL,                       /* meter_del */
5592     group_alloc,                /* group_alloc */
5593     group_construct,            /* group_construct */
5594     group_destruct,             /* group_destruct */
5595     group_dealloc,              /* group_dealloc */
5596     group_modify,               /* group_modify */
5597     group_get_stats,            /* group_get_stats */
5598     get_datapath_version,       /* get_datapath_version */
5599 };