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