ofproto/ofproto-dpif.c

   1 /*
   2  * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015 Nicira, Inc.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18
  19 #include "ofproto/ofproto-dpif.h"
  20 #include "ofproto/ofproto-provider.h"
  21
  22 #include <errno.h>
  23
  24 #include "bfd.h"
  25 #include "bond.h"
  26 #include "bundle.h"
  27 #include "byte-order.h"
  28 #include "connectivity.h"
  29 #include "connmgr.h"
  30 #include "coverage.h"
  31 #include "cfm.h"
  32 #include "dpif.h"
  33 #include "dynamic-string.h"
  34 #include "fail-open.h"
  35 #include "guarded-list.h"
  36 #include "hmapx.h"
  37 #include "lacp.h"
  38 #include "learn.h"
  39 #include "mac-learning.h"
  40 #include "mcast-snooping.h"
  41 #include "meta-flow.h"
  42 #include "multipath.h"
  43 #include "netdev-vport.h"
  44 #include "netdev.h"
  45 #include "netlink.h"
  46 #include "nx-match.h"
  47 #include "odp-util.h"
  48 #include "odp-execute.h"
  49 #include "ofp-util.h"
  50 #include "ofpbuf.h"
  51 #include "ofp-actions.h"
  52 #include "ofp-parse.h"
  53 #include "ofp-print.h"
  54 #include "ofproto-dpif-ipfix.h"
  55 #include "ofproto-dpif-mirror.h"
  56 #include "ofproto-dpif-monitor.h"
  57 #include "ofproto-dpif-rid.h"
  58 #include "ofproto-dpif-sflow.h"
  59 #include "ofproto-dpif-upcall.h"
  60 #include "ofproto-dpif-xlate.h"
  61 #include "poll-loop.h"
  62 #include "ovs-rcu.h"
  63 #include "ovs-router.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 #include "openvswitch/vlog.h"
  73
  74 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  75
  76 COVERAGE_DEFINE(ofproto_dpif_expired);
  77 COVERAGE_DEFINE(packet_in_overflow);
  78
  79 struct flow_miss;
  80
  81 struct rule_dpif {
  82     struct rule up;
  83
  84     /* These statistics:
  85      *
  86      *   - Do include packets and bytes from datapath flows which have not
  87      *   recently been processed by a revalidator. */
  88     struct ovs_mutex stats_mutex;
  89     struct dpif_flow_stats stats OVS_GUARDED;
  90
  91     /* If non-zero then the recirculation id that has
  92      * been allocated for use with this rule.
  93      * The recirculation id and associated internal flow should
  94      * be freed when the rule is freed */
  95     uint32_t recirc_id;
  96 };
  97
  98 /* RULE_CAST() depends on this. */
  99 BUILD_ASSERT_DECL(offsetof(struct rule_dpif, up) == 0);
 100
 101 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
 102                            long long int *used);
 103 static struct rule_dpif *rule_dpif_cast(const struct rule *);
 104 static void rule_expire(struct rule_dpif *);
 105
 106 struct group_dpif {
 107     struct ofgroup up;
 108
 109     /* These statistics:
 110      *
 111      *   - Do include packets and bytes from datapath flows which have not
 112      *   recently been processed by a revalidator. */
 113     struct ovs_mutex stats_mutex;
 114     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 115     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 116 };
 117
 118 struct ofbundle {
 119     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
 120     struct ofproto_dpif *ofproto; /* Owning ofproto. */
 121     void *aux;                  /* Key supplied by ofproto's client. */
 122     char *name;                 /* Identifier for log messages. */
 123
 124     /* Configuration. */
 125     struct ovs_list ports;      /* Contains "struct ofport"s. */
 126     enum port_vlan_mode vlan_mode; /* VLAN mode */
 127     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
 128     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
 129                                  * NULL if all VLANs are trunked. */
 130     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 131     struct bond *bond;          /* Nonnull iff more than one port. */
 132     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 133
 134     /* Status. */
 135     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 136 };
 137
 138 static void bundle_remove(struct ofport *);
 139 static void bundle_update(struct ofbundle *);
 140 static void bundle_destroy(struct ofbundle *);
 141 static void bundle_del_port(struct ofport_dpif *);
 142 static void bundle_run(struct ofbundle *);
 143 static void bundle_wait(struct ofbundle *);
 144 static void bundle_flush_macs(struct ofbundle *, bool);
 145 static void bundle_move(struct ofbundle *, struct ofbundle *);
 146
 147 static void stp_run(struct ofproto_dpif *ofproto);
 148 static void stp_wait(struct ofproto_dpif *ofproto);
 149 static int set_stp_port(struct ofport *,
 150                         const struct ofproto_port_stp_settings *);
 151
 152 static void rstp_run(struct ofproto_dpif *ofproto);
 153 static void set_rstp_port(struct ofport *,
 154                          const struct ofproto_port_rstp_settings *);
 155
 156 struct ofport_dpif {
 157     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 158     struct ofport up;
 159
 160     odp_port_t odp_port;
 161     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 162     struct ovs_list bundle_node;/* In struct ofbundle's "ports" list. */
 163     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 164     struct bfd *bfd;            /* BFD, if any. */
 165     bool may_enable;            /* May be enabled in bonds. */
 166     bool is_tunnel;             /* This port is a tunnel. */
 167     bool is_layer3;             /* This is a layer 3 port. */
 168     long long int carrier_seq;  /* Carrier status changes. */
 169     struct ofport_dpif *peer;   /* Peer if patch port. */
 170
 171     /* Spanning tree. */
 172     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 173     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 174     long long int stp_state_entered;
 175
 176     /* Rapid Spanning Tree. */
 177     struct rstp_port *rstp_port; /* Rapid Spanning Tree Protocol, if any. */
 178     enum rstp_state rstp_state; /* Always RSTP_DISABLED if RSTP not in use. */
 179
 180     /* Queue to DSCP mapping. */
 181     struct ofproto_port_queue *qdscp;
 182     size_t n_qdscp;
 183
 184     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 185      *
 186      * This is deprecated.  It is only for compatibility with broken device
 187      * drivers in old versions of Linux that do not properly support VLANs when
 188      * VLAN devices are not used.  When broken device drivers are no longer in
 189      * widespread use, we will delete these interfaces. */
 190     ofp_port_t realdev_ofp_port;
 191     int vlandev_vid;
 192 };
 193
 194 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 195  *
 196  * This is deprecated.  It is only for compatibility with broken device drivers
 197  * in old versions of Linux that do not properly support VLANs when VLAN
 198  * devices are not used.  When broken device drivers are no longer in
 199  * widespread use, we will delete these interfaces. */
 200 struct vlan_splinter {
 201     struct hmap_node realdev_vid_node;
 202     struct hmap_node vlandev_node;
 203     ofp_port_t realdev_ofp_port;
 204     ofp_port_t vlandev_ofp_port;
 205     int vid;
 206 };
 207
 208 static void vsp_remove(struct ofport_dpif *);
 209 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 210
 211 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 212                                        ofp_port_t);
 213
 214 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 215                                        odp_port_t);
 216
 217 static struct ofport_dpif *
 218 ofport_dpif_cast(const struct ofport *ofport)
 219 {
 220     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 221 }
 222
 223 static void port_run(struct ofport_dpif *);
 224 static int set_bfd(struct ofport *, const struct smap *);
 225 static int set_cfm(struct ofport *, const struct cfm_settings *);
 226 static void ofport_update_peer(struct ofport_dpif *);
 227
 228 /* Reasons that we might need to revalidate every datapath flow, and
 229  * corresponding coverage counters.
 230  *
 231  * A value of 0 means that there is no need to revalidate.
 232  *
 233  * It would be nice to have some cleaner way to integrate with coverage
 234  * counters, but with only a few reasons I guess this is good enough for
 235  * now. */
 236 enum revalidate_reason {
 237     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 238     REV_STP,                   /* Spanning tree protocol port status change. */
 239     REV_RSTP,                  /* RSTP port status change. */
 240     REV_BOND,                  /* Bonding changed. */
 241     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 242     REV_FLOW_TABLE,            /* Flow table changed. */
 243     REV_MAC_LEARNING,          /* Mac learning changed. */
 244     REV_MCAST_SNOOPING,        /* Multicast snooping changed. */
 245 };
 246 COVERAGE_DEFINE(rev_reconfigure);
 247 COVERAGE_DEFINE(rev_stp);
 248 COVERAGE_DEFINE(rev_rstp);
 249 COVERAGE_DEFINE(rev_bond);
 250 COVERAGE_DEFINE(rev_port_toggled);
 251 COVERAGE_DEFINE(rev_flow_table);
 252 COVERAGE_DEFINE(rev_mac_learning);
 253 COVERAGE_DEFINE(rev_mcast_snooping);
 254
 255 /* Stores mapping between 'recirc_id' and 'ofproto-dpif'. */
 256 struct dpif_backer_recirc_node {
 257     struct cmap_node cmap_node;
 258     struct ofproto_dpif *ofproto;
 259     uint32_t recirc_id;
 260 };
 261
 262 /* All datapaths of a given type share a single dpif backer instance. */
 263 struct dpif_backer {
 264     char *type;
 265     int refcount;
 266     struct dpif *dpif;
 267     struct udpif *udpif;
 268
 269     struct ovs_rwlock odp_to_ofport_lock;
 270     struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
 271
 272     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 273
 274     enum revalidate_reason need_revalidate; /* Revalidate all flows. */
 275
 276     bool recv_set_enable; /* Enables or disables receiving packets. */
 277
 278     /* Recirculation. */
 279     struct recirc_id_pool *rid_pool;       /* Recirculation ID pool. */
 280     struct cmap recirc_map;         /* Map of 'recirc_id's to 'ofproto's. */
 281     struct ovs_mutex recirc_mutex;  /* Protects 'recirc_map'. */
 282     bool enable_recirc;   /* True if the datapath supports recirculation */
 283
 284     /* True if the datapath supports unique flow identifiers */
 285     bool enable_ufid;
 286
 287     /* True if the datapath supports variable-length
 288      * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
 289      * False if the datapath supports only 8-byte (or shorter) userdata. */
 290     bool variable_length_userdata;
 291
 292     /* True if the datapath supports masked data in OVS_ACTION_ATTR_SET
 293      * actions. */
 294     bool masked_set_action;
 295
 296     /* Maximum number of MPLS label stack entries that the datapath supports
 297      * in a match */
 298     size_t max_mpls_depth;
 299
 300     /* Version string of the datapath stored in OVSDB. */
 301     char *dp_version_string;
 302
 303     /* True if the datapath supports tnl_push and pop actions. */
 304     bool enable_tnl_push_pop;
 305     struct atomic_count tnl_count;
 306 };
 307
 308 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 309 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 310
 311 struct ofproto_dpif {
 312     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 313     struct ofproto up;
 314     struct dpif_backer *backer;
 315
 316     uint64_t dump_seq; /* Last read of udpif_dump_seq(). */
 317
 318     /* Special OpenFlow rules. */
 319     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 320     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 321     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 322
 323     /* Bridging. */
 324     struct netflow *netflow;
 325     struct dpif_sflow *sflow;
 326     struct dpif_ipfix *ipfix;
 327     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 328     struct mac_learning *ml;
 329     struct mcast_snooping *ms;
 330     bool has_bonded_bundles;
 331     bool lacp_enabled;
 332     struct mbridge *mbridge;
 333
 334     struct ovs_mutex stats_mutex;
 335     struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
 336                                             * consumed in userspace. */
 337
 338     /* Spanning tree. */
 339     struct stp *stp;
 340     long long int stp_last_tick;
 341
 342     /* Rapid Spanning Tree. */
 343     struct rstp *rstp;
 344     long long int rstp_last_tick;
 345
 346     /* VLAN splinters. */
 347     struct ovs_mutex vsp_mutex;
 348     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 349     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 350
 351     /* Ports. */
 352     struct sset ports;             /* Set of standard port names. */
 353     struct sset ghost_ports;       /* Ports with no datapath port. */
 354     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 355     int port_poll_errno;           /* Last errno for port_poll() reply. */
 356     uint64_t change_seq;           /* Connectivity status changes. */
 357
 358     /* Work queues. */
 359     struct guarded_list pins;      /* Contains "struct ofputil_packet_in"s. */
 360     struct seq *pins_seq;          /* For notifying 'pins' reception. */
 361     uint64_t pins_seqno;
 362 };
 363
 364 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 365 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 366
 367 static bool ofproto_use_tnl_push_pop = true;
 368 static void ofproto_unixctl_init(void);
 369
 370 static inline struct ofproto_dpif *
 371 ofproto_dpif_cast(const struct ofproto *ofproto)
 372 {
 373     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 374     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 375 }
 376
 377 size_t
 378 ofproto_dpif_get_max_mpls_depth(const struct ofproto_dpif *ofproto)
 379 {
 380     return ofproto->backer->max_mpls_depth;
 381 }
 382
 383 bool
 384 ofproto_dpif_get_enable_recirc(const struct ofproto_dpif *ofproto)
 385 {
 386     return ofproto->backer->enable_recirc;
 387 }
 388
 389 bool
 390 ofproto_dpif_get_enable_ufid(struct dpif_backer *backer)
 391 {
 392     return backer->enable_ufid;
 393 }
 394
 395 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
 396                                         ofp_port_t ofp_port);
 397 static void ofproto_trace(struct ofproto_dpif *, struct flow *,
 398                           const struct ofpbuf *packet,
 399                           const struct ofpact[], size_t ofpacts_len,
 400                           struct ds *);
 401
 402 /* Global variables. */
 403 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 404
 405 /* Initial mappings of port to bridge mappings. */
 406 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 407
 408 /* Executes 'fm'.  The caller retains ownership of 'fm' and everything in
 409  * it. */
 410 void
 411 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 412                       struct ofputil_flow_mod *fm)
 413 {
 414     ofproto_flow_mod(&ofproto->up, fm);
 415 }
 416
 417 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 418  * Takes ownership of 'pin' and pin->packet. */
 419 void
 420 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 421                             struct ofproto_packet_in *pin)
 422 {
 423     if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
 424         COVERAGE_INC(packet_in_overflow);
 425         free(CONST_CAST(void *, pin->up.packet));
 426         free(pin);
 427     }
 428
 429     /* Wakes up main thread for packet-in I/O. */
 430     seq_change(ofproto->pins_seq);
 431 }
 432
 433 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
 434  * packet rather than to send the packet to the controller.
 435  *
 436  * This function returns false to indicate that a packet_in message
 437  * for a "table-miss" should be sent to at least one controller.
 438  * False otherwise. */
 439 bool
 440 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif *ofproto)
 441 {
 442     return connmgr_wants_packet_in_on_miss(ofproto->up.connmgr);
 443 }
 444 \f
 445 /* Factory functions. */
 446
 447 static void
 448 init(const struct shash *iface_hints)
 449 {
 450     struct shash_node *node;
 451
 452     /* Make a local copy, since we don't own 'iface_hints' elements. */
 453     SHASH_FOR_EACH(node, iface_hints) {
 454         const struct iface_hint *orig_hint = node->data;
 455         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 456
 457         new_hint->br_name = xstrdup(orig_hint->br_name);
 458         new_hint->br_type = xstrdup(orig_hint->br_type);
 459         new_hint->ofp_port = orig_hint->ofp_port;
 460
 461         shash_add(&init_ofp_ports, node->name, new_hint);
 462     }
 463 }
 464
 465 static void
 466 enumerate_types(struct sset *types)
 467 {
 468     dp_enumerate_types(types);
 469 }
 470
 471 static int
 472 enumerate_names(const char *type, struct sset *names)
 473 {
 474     struct ofproto_dpif *ofproto;
 475
 476     sset_clear(names);
 477     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 478         if (strcmp(type, ofproto->up.type)) {
 479             continue;
 480         }
 481         sset_add(names, ofproto->up.name);
 482     }
 483
 484     return 0;
 485 }
 486
 487 static int
 488 del(const char *type, const char *name)
 489 {
 490     struct dpif *dpif;
 491     int error;
 492
 493     error = dpif_open(name, type, &dpif);
 494     if (!error) {
 495         error = dpif_delete(dpif);
 496         dpif_close(dpif);
 497     }
 498     return error;
 499 }
 500 \f
 501 static const char *
 502 port_open_type(const char *datapath_type, const char *port_type)
 503 {
 504     return dpif_port_open_type(datapath_type, port_type);
 505 }
 506
 507 /* Type functions. */
 508
 509 static void process_dpif_port_changes(struct dpif_backer *);
 510 static void process_dpif_all_ports_changed(struct dpif_backer *);
 511 static void process_dpif_port_change(struct dpif_backer *,
 512                                      const char *devname);
 513 static void process_dpif_port_error(struct dpif_backer *, int error);
 514
 515 static struct ofproto_dpif *
 516 lookup_ofproto_dpif_by_port_name(const char *name)
 517 {
 518     struct ofproto_dpif *ofproto;
 519
 520     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 521         if (sset_contains(&ofproto->ports, name)) {
 522             return ofproto;
 523         }
 524     }
 525
 526     return NULL;
 527 }
 528
 529 static int
 530 type_run(const char *type)
 531 {
 532     struct dpif_backer *backer;
 533
 534     backer = shash_find_data(&all_dpif_backers, type);
 535     if (!backer) {
 536         /* This is not necessarily a problem, since backers are only
 537          * created on demand. */
 538         return 0;
 539     }
 540
 541
 542     if (dpif_run(backer->dpif)) {
 543         backer->need_revalidate = REV_RECONFIGURE;
 544     }
 545
 546     udpif_run(backer->udpif);
 547
 548     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 549      * and the configuration has now changed to "false", enable receiving
 550      * packets from the datapath. */
 551     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 552         int error;
 553
 554         backer->recv_set_enable = true;
 555
 556         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 557         if (error) {
 558             VLOG_ERR("Failed to enable receiving packets in dpif.");
 559             return error;
 560         }
 561         dpif_flow_flush(backer->dpif);
 562         backer->need_revalidate = REV_RECONFIGURE;
 563     }
 564
 565     if (backer->recv_set_enable) {
 566         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 567     }
 568
 569     dpif_poll_threads_set(backer->dpif, n_dpdk_rxqs, pmd_cpu_mask);
 570
 571     if (backer->need_revalidate) {
 572         struct ofproto_dpif *ofproto;
 573         struct simap_node *node;
 574         struct simap tmp_backers;
 575
 576         /* Handle tunnel garbage collection. */
 577         simap_init(&tmp_backers);
 578         simap_swap(&backer->tnl_backers, &tmp_backers);
 579
 580         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 581             struct ofport_dpif *iter;
 582
 583             if (backer != ofproto->backer) {
 584                 continue;
 585             }
 586
 587             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 588                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 589                 const char *dp_port;
 590
 591                 if (!iter->is_tunnel) {
 592                     continue;
 593                 }
 594
 595                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 596                                                      namebuf, sizeof namebuf);
 597                 node = simap_find(&tmp_backers, dp_port);
 598                 if (node) {
 599                     simap_put(&backer->tnl_backers, dp_port, node->data);
 600                     simap_delete(&tmp_backers, node);
 601                     node = simap_find(&backer->tnl_backers, dp_port);
 602                 } else {
 603                     node = simap_find(&backer->tnl_backers, dp_port);
 604                     if (!node) {
 605                         odp_port_t odp_port = ODPP_NONE;
 606
 607                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 608                                            &odp_port)) {
 609                             simap_put(&backer->tnl_backers, dp_port,
 610                                       odp_to_u32(odp_port));
 611                             node = simap_find(&backer->tnl_backers, dp_port);
 612                         }
 613                     }
 614                 }
 615
 616                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 617                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 618                                          iter->odp_port,
 619                                          ovs_native_tunneling_is_on(ofproto), dp_port)) {
 620                     backer->need_revalidate = REV_RECONFIGURE;
 621                 }
 622             }
 623         }
 624
 625         SIMAP_FOR_EACH (node, &tmp_backers) {
 626             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 627         }
 628         simap_destroy(&tmp_backers);
 629
 630         switch (backer->need_revalidate) {
 631         case REV_RECONFIGURE:    COVERAGE_INC(rev_reconfigure);    break;
 632         case REV_STP:            COVERAGE_INC(rev_stp);            break;
 633         case REV_RSTP:           COVERAGE_INC(rev_rstp);           break;
 634         case REV_BOND:           COVERAGE_INC(rev_bond);           break;
 635         case REV_PORT_TOGGLED:   COVERAGE_INC(rev_port_toggled);   break;
 636         case REV_FLOW_TABLE:     COVERAGE_INC(rev_flow_table);     break;
 637         case REV_MAC_LEARNING:   COVERAGE_INC(rev_mac_learning);   break;
 638         case REV_MCAST_SNOOPING: COVERAGE_INC(rev_mcast_snooping); break;
 639         }
 640         backer->need_revalidate = 0;
 641
 642         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 643             struct ofport_dpif *ofport;
 644             struct ofbundle *bundle;
 645
 646             if (ofproto->backer != backer) {
 647                 continue;
 648             }
 649
 650             xlate_txn_start();
 651             xlate_ofproto_set(ofproto, ofproto->up.name,
 652                               ofproto->backer->dpif, ofproto->ml,
 653                               ofproto->stp, ofproto->rstp, ofproto->ms,
 654                               ofproto->mbridge, ofproto->sflow, ofproto->ipfix,
 655                               ofproto->netflow,
 656                               ofproto->up.forward_bpdu,
 657                               connmgr_has_in_band(ofproto->up.connmgr),
 658                               ofproto->backer->enable_recirc,
 659                               ofproto->backer->variable_length_userdata,
 660                               ofproto->backer->max_mpls_depth,
 661                               ofproto->backer->masked_set_action);
 662
 663             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 664                 xlate_bundle_set(ofproto, bundle, bundle->name,
 665                                  bundle->vlan_mode, bundle->vlan,
 666                                  bundle->trunks, bundle->use_priority_tags,
 667                                  bundle->bond, bundle->lacp,
 668                                  bundle->floodable);
 669             }
 670
 671             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 672                 int stp_port = ofport->stp_port
 673                     ? stp_port_no(ofport->stp_port)
 674                     : -1;
 675                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 676                                  ofport->up.ofp_port, ofport->odp_port,
 677                                  ofport->up.netdev, ofport->cfm,
 678                                  ofport->bfd, ofport->peer, stp_port,
 679                                  ofport->rstp_port, ofport->qdscp,
 680                                  ofport->n_qdscp, ofport->up.pp.config,
 681                                  ofport->up.pp.state, ofport->is_tunnel,
 682                                  ofport->may_enable);
 683             }
 684             xlate_txn_commit();
 685         }
 686
 687         udpif_revalidate(backer->udpif);
 688     }
 689
 690     process_dpif_port_changes(backer);
 691
 692     return 0;
 693 }
 694
 695 /* Check for and handle port changes in 'backer''s dpif. */
 696 static void
 697 process_dpif_port_changes(struct dpif_backer *backer)
 698 {
 699     for (;;) {
 700         char *devname;
 701         int error;
 702
 703         error = dpif_port_poll(backer->dpif, &devname);
 704         switch (error) {
 705         case EAGAIN:
 706             return;
 707
 708         case ENOBUFS:
 709             process_dpif_all_ports_changed(backer);
 710             break;
 711
 712         case 0:
 713             process_dpif_port_change(backer, devname);
 714             free(devname);
 715             break;
 716
 717         default:
 718             process_dpif_port_error(backer, error);
 719             break;
 720         }
 721     }
 722 }
 723
 724 static void
 725 process_dpif_all_ports_changed(struct dpif_backer *backer)
 726 {
 727     struct ofproto_dpif *ofproto;
 728     struct dpif_port dpif_port;
 729     struct dpif_port_dump dump;
 730     struct sset devnames;
 731     const char *devname;
 732
 733     sset_init(&devnames);
 734     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 735         if (ofproto->backer == backer) {
 736             struct ofport *ofport;
 737
 738             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 739                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 740             }
 741         }
 742     }
 743     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 744         sset_add(&devnames, dpif_port.name);
 745     }
 746
 747     SSET_FOR_EACH (devname, &devnames) {
 748         process_dpif_port_change(backer, devname);
 749     }
 750     sset_destroy(&devnames);
 751 }
 752
 753 static void
 754 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 755 {
 756     struct ofproto_dpif *ofproto;
 757     struct dpif_port port;
 758
 759     /* Don't report on the datapath's device. */
 760     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 761         return;
 762     }
 763
 764     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 765                    &all_ofproto_dpifs) {
 766         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 767             return;
 768         }
 769     }
 770
 771     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 772     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 773         /* The port was removed.  If we know the datapath,
 774          * report it through poll_set().  If we don't, it may be
 775          * notifying us of a removal we initiated, so ignore it.
 776          * If there's a pending ENOBUFS, let it stand, since
 777          * everything will be reevaluated. */
 778         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 779             sset_add(&ofproto->port_poll_set, devname);
 780             ofproto->port_poll_errno = 0;
 781         }
 782     } else if (!ofproto) {
 783         /* The port was added, but we don't know with which
 784          * ofproto we should associate it.  Delete it. */
 785         dpif_port_del(backer->dpif, port.port_no);
 786     } else {
 787         struct ofport_dpif *ofport;
 788
 789         ofport = ofport_dpif_cast(shash_find_data(
 790                                       &ofproto->up.port_by_name, devname));
 791         if (ofport
 792             && ofport->odp_port != port.port_no
 793             && !odp_port_to_ofport(backer, port.port_no))
 794         {
 795             /* 'ofport''s datapath port number has changed from
 796              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 797              * structures to match. */
 798             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 799             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 800             ofport->odp_port = port.port_no;
 801             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 802                         hash_odp_port(port.port_no));
 803             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 804             backer->need_revalidate = REV_RECONFIGURE;
 805         }
 806     }
 807     dpif_port_destroy(&port);
 808 }
 809
 810 /* Propagate 'error' to all ofprotos based on 'backer'. */
 811 static void
 812 process_dpif_port_error(struct dpif_backer *backer, int error)
 813 {
 814     struct ofproto_dpif *ofproto;
 815
 816     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 817         if (ofproto->backer == backer) {
 818             sset_clear(&ofproto->port_poll_set);
 819             ofproto->port_poll_errno = error;
 820         }
 821     }
 822 }
 823
 824 static void
 825 type_wait(const char *type)
 826 {
 827     struct dpif_backer *backer;
 828
 829     backer = shash_find_data(&all_dpif_backers, type);
 830     if (!backer) {
 831         /* This is not necessarily a problem, since backers are only
 832          * created on demand. */
 833         return;
 834     }
 835
 836     dpif_wait(backer->dpif);
 837 }
 838 \f
 839 /* Basic life-cycle. */
 840
 841 static int add_internal_flows(struct ofproto_dpif *);
 842
 843 static struct ofproto *
 844 alloc(void)
 845 {
 846     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
 847     return &ofproto->up;
 848 }
 849
 850 static void
 851 dealloc(struct ofproto *ofproto_)
 852 {
 853     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
 854     free(ofproto);
 855 }
 856
 857 /* Called when 'ofproto' is destructed.  Checks for and clears any
 858  * recirc_id leak. */
 859 static void
 860 dpif_backer_recirc_clear_ofproto(struct dpif_backer *backer,
 861                                  struct ofproto_dpif *ofproto)
 862 {
 863     struct dpif_backer_recirc_node *node;
 864
 865     ovs_mutex_lock(&backer->recirc_mutex);
 866     CMAP_FOR_EACH (node, cmap_node, &backer->recirc_map) {
 867         if (node->ofproto == ofproto) {
 868             VLOG_ERR("recirc_id %"PRIu32", not freed when ofproto (%s) "
 869                      "is destructed", node->recirc_id, ofproto->up.name);
 870             cmap_remove(&backer->recirc_map, &node->cmap_node,
 871                         node->recirc_id);
 872             ovsrcu_postpone(free, node);
 873         }
 874     }
 875     ovs_mutex_unlock(&backer->recirc_mutex);
 876 }
 877
 878 static void
 879 close_dpif_backer(struct dpif_backer *backer)
 880 {
 881     ovs_assert(backer->refcount > 0);
 882
 883     if (--backer->refcount) {
 884         return;
 885     }
 886
 887     udpif_destroy(backer->udpif);
 888
 889     simap_destroy(&backer->tnl_backers);
 890     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
 891     hmap_destroy(&backer->odp_to_ofport_map);
 892     shash_find_and_delete(&all_dpif_backers, backer->type);
 893     recirc_id_pool_destroy(backer->rid_pool);
 894     cmap_destroy(&backer->recirc_map);
 895     ovs_mutex_destroy(&backer->recirc_mutex);
 896     free(backer->type);
 897     free(backer->dp_version_string);
 898     dpif_close(backer->dpif);
 899     free(backer);
 900 }
 901
 902 /* Datapath port slated for removal from datapath. */
 903 struct odp_garbage {
 904     struct ovs_list list_node;
 905     odp_port_t odp_port;
 906 };
 907
 908 static bool check_variable_length_userdata(struct dpif_backer *backer);
 909 static size_t check_max_mpls_depth(struct dpif_backer *backer);
 910 static bool check_recirc(struct dpif_backer *backer);
 911 static bool check_ufid(struct dpif_backer *backer);
 912 static bool check_masked_set_action(struct dpif_backer *backer);
 913
 914 static int
 915 open_dpif_backer(const char *type, struct dpif_backer **backerp)
 916 {
 917     struct dpif_backer *backer;
 918     struct dpif_port_dump port_dump;
 919     struct dpif_port port;
 920     struct shash_node *node;
 921     struct ovs_list garbage_list;
 922     struct odp_garbage *garbage, *next;
 923
 924     struct sset names;
 925     char *backer_name;
 926     const char *name;
 927     int error;
 928
 929     backer = shash_find_data(&all_dpif_backers, type);
 930     if (backer) {
 931         backer->refcount++;
 932         *backerp = backer;
 933         return 0;
 934     }
 935
 936     backer_name = xasprintf("ovs-%s", type);
 937
 938     /* Remove any existing datapaths, since we assume we're the only
 939      * userspace controlling the datapath. */
 940     sset_init(&names);
 941     dp_enumerate_names(type, &names);
 942     SSET_FOR_EACH(name, &names) {
 943         struct dpif *old_dpif;
 944
 945         /* Don't remove our backer if it exists. */
 946         if (!strcmp(name, backer_name)) {
 947             continue;
 948         }
 949
 950         if (dpif_open(name, type, &old_dpif)) {
 951             VLOG_WARN("couldn't open old datapath %s to remove it", name);
 952         } else {
 953             dpif_delete(old_dpif);
 954             dpif_close(old_dpif);
 955         }
 956     }
 957     sset_destroy(&names);
 958
 959     backer = xmalloc(sizeof *backer);
 960
 961     error = dpif_create_and_open(backer_name, type, &backer->dpif);
 962     free(backer_name);
 963     if (error) {
 964         VLOG_ERR("failed to open datapath of type %s: %s", type,
 965                  ovs_strerror(error));
 966         free(backer);
 967         return error;
 968     }
 969     backer->udpif = udpif_create(backer, backer->dpif);
 970
 971     backer->type = xstrdup(type);
 972     backer->refcount = 1;
 973     hmap_init(&backer->odp_to_ofport_map);
 974     ovs_rwlock_init(&backer->odp_to_ofport_lock);
 975     backer->need_revalidate = 0;
 976     simap_init(&backer->tnl_backers);
 977     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
 978     *backerp = backer;
 979
 980     if (backer->recv_set_enable) {
 981         dpif_flow_flush(backer->dpif);
 982     }
 983
 984     /* Loop through the ports already on the datapath and remove any
 985      * that we don't need anymore. */
 986     list_init(&garbage_list);
 987     dpif_port_dump_start(&port_dump, backer->dpif);
 988     while (dpif_port_dump_next(&port_dump, &port)) {
 989         node = shash_find(&init_ofp_ports, port.name);
 990         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
 991             garbage = xmalloc(sizeof *garbage);
 992             garbage->odp_port = port.port_no;
 993             list_push_front(&garbage_list, &garbage->list_node);
 994         }
 995     }
 996     dpif_port_dump_done(&port_dump);
 997
 998     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
 999         dpif_port_del(backer->dpif, garbage->odp_port);
1000         list_remove(&garbage->list_node);
1001         free(garbage);
1002     }
1003
1004     shash_add(&all_dpif_backers, type, backer);
1005
1006     backer->enable_recirc = check_recirc(backer);
1007     backer->max_mpls_depth = check_max_mpls_depth(backer);
1008     backer->masked_set_action = check_masked_set_action(backer);
1009     backer->enable_ufid = check_ufid(backer);
1010     backer->rid_pool = recirc_id_pool_create();
1011     ovs_mutex_init(&backer->recirc_mutex);
1012     cmap_init(&backer->recirc_map);
1013
1014     backer->enable_tnl_push_pop = dpif_supports_tnl_push_pop(backer->dpif);
1015     atomic_count_init(&backer->tnl_count, 0);
1016
1017     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1018     if (error) {
1019         VLOG_ERR("failed to listen on datapath of type %s: %s",
1020                  type, ovs_strerror(error));
1021         close_dpif_backer(backer);
1022         return error;
1023     }
1024
1025     if (backer->recv_set_enable) {
1026         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
1027     }
1028
1029     /* This check fails if performed before udpif threads have been set,
1030      * as the kernel module checks that the 'pid' in userspace action
1031      * is non-zero. */
1032     backer->variable_length_userdata = check_variable_length_userdata(backer);
1033     backer->dp_version_string = dpif_get_dp_version(backer->dpif);
1034
1035     return error;
1036 }
1037
1038 bool
1039 ovs_native_tunneling_is_on(struct ofproto_dpif *ofproto)
1040 {
1041     return ofproto_use_tnl_push_pop && ofproto->backer->enable_tnl_push_pop &&
1042            atomic_count_get(&ofproto->backer->tnl_count);
1043 }
1044
1045 /* Tests whether 'backer''s datapath supports recirculation.  Only newer
1046  * datapaths support OVS_KEY_ATTR_RECIRC_ID in keys.  We need to disable some
1047  * features on older datapaths that don't support this feature.
1048  *
1049  * Returns false if 'backer' definitely does not support recirculation, true if
1050  * it seems to support recirculation or if at least the error we get is
1051  * ambiguous. */
1052 static bool
1053 check_recirc(struct dpif_backer *backer)
1054 {
1055     struct flow flow;
1056     struct odputil_keybuf keybuf;
1057     struct ofpbuf key;
1058     bool enable_recirc;
1059
1060     memset(&flow, 0, sizeof flow);
1061     flow.recirc_id = 1;
1062     flow.dp_hash = 1;
1063
1064     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1065     odp_flow_key_from_flow(&key, &flow, NULL, 0, true);
1066     enable_recirc = dpif_probe_feature(backer->dpif, "recirculation", &key,
1067                                        NULL);
1068
1069     if (enable_recirc) {
1070         VLOG_INFO("%s: Datapath supports recirculation",
1071                   dpif_name(backer->dpif));
1072     } else {
1073         VLOG_INFO("%s: Datapath does not support recirculation",
1074                   dpif_name(backer->dpif));
1075     }
1076
1077     return enable_recirc;
1078 }
1079
1080 /* Tests whether 'dpif' supports userspace flow ids. We can skip serializing
1081  * some flow attributes for datapaths that support this feature.
1082  *
1083  * Returns true if 'dpif' supports UFID for flow operations.
1084  * Returns false if  'dpif' does not support UFID. */
1085 static bool
1086 check_ufid(struct dpif_backer *backer)
1087 {
1088     struct flow flow;
1089     struct odputil_keybuf keybuf;
1090     struct ofpbuf key;
1091     ovs_u128 ufid;
1092     bool enable_ufid;
1093
1094     memset(&flow, 0, sizeof flow);
1095     flow.dl_type = htons(0x1234);
1096
1097     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1098     odp_flow_key_from_flow(&key, &flow, NULL, 0, true);
1099     dpif_flow_hash(backer->dpif, ofpbuf_data(&key), ofpbuf_size(&key), &ufid);
1100
1101     enable_ufid = dpif_probe_feature(backer->dpif, "UFID", &key, &ufid);
1102
1103     if (enable_ufid) {
1104         VLOG_INFO("%s: Datapath supports userspace flow ids",
1105                   dpif_name(backer->dpif));
1106     } else {
1107         VLOG_INFO("%s: Datapath does not support userspace flow ids",
1108                   dpif_name(backer->dpif));
1109     }
1110     return enable_ufid;
1111 }
1112
1113 /* Tests whether 'backer''s datapath supports variable-length
1114  * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.  We need
1115  * to disable some features on older datapaths that don't support this
1116  * feature.
1117  *
1118  * Returns false if 'backer' definitely does not support variable-length
1119  * userdata, true if it seems to support them or if at least the error we get
1120  * is ambiguous. */
1121 static bool
1122 check_variable_length_userdata(struct dpif_backer *backer)
1123 {
1124     struct eth_header *eth;
1125     struct ofpbuf actions;
1126     struct dpif_execute execute;
1127     struct ofpbuf packet;
1128     size_t start;
1129     int error;
1130
1131     /* Compose a userspace action that will cause an ERANGE error on older
1132      * datapaths that don't support variable-length userdata.
1133      *
1134      * We really test for using userdata longer than 8 bytes, but older
1135      * datapaths accepted these, silently truncating the userdata to 8 bytes.
1136      * The same older datapaths rejected userdata shorter than 8 bytes, so we
1137      * test for that instead as a proxy for longer userdata support. */
1138     ofpbuf_init(&actions, 64);
1139     start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
1140     nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
1141                    dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
1142     nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
1143     nl_msg_end_nested(&actions, start);
1144
1145     /* Compose a dummy ethernet packet. */
1146     ofpbuf_init(&packet, ETH_HEADER_LEN);
1147     eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
1148     eth->eth_type = htons(0x1234);
1149
1150     /* Execute the actions.  On older datapaths this fails with ERANGE, on
1151      * newer datapaths it succeeds. */
1152     execute.actions = ofpbuf_data(&actions);
1153     execute.actions_len = ofpbuf_size(&actions);
1154     execute.packet = &packet;
1155     execute.md = PKT_METADATA_INITIALIZER(0);
1156     execute.needs_help = false;
1157     execute.probe = true;
1158
1159     error = dpif_execute(backer->dpif, &execute);
1160
1161     ofpbuf_uninit(&packet);
1162     ofpbuf_uninit(&actions);
1163
1164     switch (error) {
1165     case 0:
1166         return true;
1167
1168     case ERANGE:
1169         /* Variable-length userdata is not supported. */
1170         VLOG_WARN("%s: datapath does not support variable-length userdata "
1171                   "feature (needs Linux 3.10+ or kernel module from OVS "
1172                   "1..11+).  The NXAST_SAMPLE action will be ignored.",
1173                   dpif_name(backer->dpif));
1174         return false;
1175
1176     default:
1177         /* Something odd happened.  We're not sure whether variable-length
1178          * userdata is supported.  Default to "yes". */
1179         VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1180                   dpif_name(backer->dpif), ovs_strerror(error));
1181         return true;
1182     }
1183 }
1184
1185 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1186  *
1187  * Returns the number of elements in a struct flow's mpls_lse field
1188  * if the datapath supports at least that many entries in an
1189  * MPLS label stack.
1190  * Otherwise returns the number of MPLS push actions supported by
1191  * the datapath. */
1192 static size_t
1193 check_max_mpls_depth(struct dpif_backer *backer)
1194 {
1195     struct flow flow;
1196     int n;
1197
1198     for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
1199         struct odputil_keybuf keybuf;
1200         struct ofpbuf key;
1201
1202         memset(&flow, 0, sizeof flow);
1203         flow.dl_type = htons(ETH_TYPE_MPLS);
1204         flow_set_mpls_bos(&flow, n, 1);
1205
1206         ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1207         odp_flow_key_from_flow(&key, &flow, NULL, 0, false);
1208         if (!dpif_probe_feature(backer->dpif, "MPLS", &key, NULL)) {
1209             break;
1210         }
1211     }
1212
1213     VLOG_INFO("%s: MPLS label stack length probed as %d",
1214               dpif_name(backer->dpif), n);
1215     return n;
1216 }
1217
1218 /* Tests whether 'backer''s datapath supports masked data in
1219  * OVS_ACTION_ATTR_SET actions.  We need to disable some features on older
1220  * datapaths that don't support this feature. */
1221 static bool
1222 check_masked_set_action(struct dpif_backer *backer)
1223 {
1224     struct eth_header *eth;
1225     struct ofpbuf actions;
1226     struct dpif_execute execute;
1227     struct ofpbuf packet;
1228     int error;
1229     struct ovs_key_ethernet key, mask;
1230
1231     /* Compose a set action that will cause an EINVAL error on older
1232      * datapaths that don't support masked set actions.
1233      * Avoid using a full mask, as it could be translated to a non-masked
1234      * set action instead. */
1235     ofpbuf_init(&actions, 64);
1236     memset(&key, 0x53, sizeof key);
1237     memset(&mask, 0x7f, sizeof mask);
1238     commit_masked_set_action(&actions, OVS_KEY_ATTR_ETHERNET, &key, &mask,
1239                              sizeof key);
1240
1241     /* Compose a dummy ethernet packet. */
1242     ofpbuf_init(&packet, ETH_HEADER_LEN);
1243     eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
1244     eth->eth_type = htons(0x1234);
1245
1246     /* Execute the actions.  On older datapaths this fails with EINVAL, on
1247      * newer datapaths it succeeds. */
1248     execute.actions = ofpbuf_data(&actions);
1249     execute.actions_len = ofpbuf_size(&actions);
1250     execute.packet = &packet;
1251     execute.md = PKT_METADATA_INITIALIZER(0);
1252     execute.needs_help = false;
1253     execute.probe = true;
1254
1255     error = dpif_execute(backer->dpif, &execute);
1256
1257     ofpbuf_uninit(&packet);
1258     ofpbuf_uninit(&actions);
1259
1260     if (error) {
1261         /* Masked set action is not supported. */
1262         VLOG_INFO("%s: datapath does not support masked set action feature.",
1263                   dpif_name(backer->dpif));
1264     }
1265     return !error;
1266 }
1267
1268 static int
1269 construct(struct ofproto *ofproto_)
1270 {
1271     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1272     struct shash_node *node, *next;
1273     int error;
1274
1275     /* Tunnel module can get used right after the udpif threads are running. */
1276     ofproto_tunnel_init();
1277
1278     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1279     if (error) {
1280         return error;
1281     }
1282
1283     ofproto->netflow = NULL;
1284     ofproto->sflow = NULL;
1285     ofproto->ipfix = NULL;
1286     ofproto->stp = NULL;
1287     ofproto->rstp = NULL;
1288     ofproto->dump_seq = 0;
1289     hmap_init(&ofproto->bundles);
1290     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1291     ofproto->ms = NULL;
1292     ofproto->mbridge = mbridge_create();
1293     ofproto->has_bonded_bundles = false;
1294     ofproto->lacp_enabled = false;
1295     ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1296     ovs_mutex_init(&ofproto->vsp_mutex);
1297
1298     guarded_list_init(&ofproto->pins);
1299
1300     ofproto_unixctl_init();
1301
1302     hmap_init(&ofproto->vlandev_map);
1303     hmap_init(&ofproto->realdev_vid_map);
1304
1305     sset_init(&ofproto->ports);
1306     sset_init(&ofproto->ghost_ports);
1307     sset_init(&ofproto->port_poll_set);
1308     ofproto->port_poll_errno = 0;
1309     ofproto->change_seq = 0;
1310     ofproto->pins_seq = seq_create();
1311     ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1312
1313
1314     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1315         struct iface_hint *iface_hint = node->data;
1316
1317         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1318             /* Check if the datapath already has this port. */
1319             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1320                 sset_add(&ofproto->ports, node->name);
1321             }
1322
1323             free(iface_hint->br_name);
1324             free(iface_hint->br_type);
1325             free(iface_hint);
1326             shash_delete(&init_ofp_ports, node);
1327         }
1328     }
1329
1330     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1331                 hash_string(ofproto->up.name, 0));
1332     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1333
1334     ofproto_init_tables(ofproto_, N_TABLES);
1335     error = add_internal_flows(ofproto);
1336
1337     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1338
1339     return error;
1340 }
1341
1342 static int
1343 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1344                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1345 {
1346     struct match match;
1347     int error;
1348     struct rule *rule;
1349
1350     match_init_catchall(&match);
1351     match_set_reg(&match, 0, id);
1352
1353     error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, 0, ofpacts,
1354                                            &rule);
1355     *rulep = error ? NULL : rule_dpif_cast(rule);
1356
1357     return error;
1358 }
1359
1360 static int
1361 add_internal_flows(struct ofproto_dpif *ofproto)
1362 {
1363     struct ofpact_controller *controller;
1364     uint64_t ofpacts_stub[128 / 8];
1365     struct ofpbuf ofpacts;
1366     struct rule *unused_rulep OVS_UNUSED;
1367     struct ofpact_resubmit *resubmit;
1368     struct match match;
1369     int error;
1370     int id;
1371
1372     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1373     id = 1;
1374
1375     controller = ofpact_put_CONTROLLER(&ofpacts);
1376     controller->max_len = UINT16_MAX;
1377     controller->controller_id = 0;
1378     controller->reason = OFPR_NO_MATCH;
1379     ofpact_pad(&ofpacts);
1380
1381     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1382                                    &ofproto->miss_rule);
1383     if (error) {
1384         return error;
1385     }
1386
1387     ofpbuf_clear(&ofpacts);
1388     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1389                                    &ofproto->no_packet_in_rule);
1390     if (error) {
1391         return error;
1392     }
1393
1394     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1395                                    &ofproto->drop_frags_rule);
1396     if (error) {
1397         return error;
1398     }
1399
1400     /* Drop any run away non-recirc rule lookups. Recirc_id has to be
1401      * zero when reaching this rule.
1402      *
1403      * (priority=2), recirc_id=0, actions=drop
1404      */
1405     ofpbuf_clear(&ofpacts);
1406     match_init_catchall(&match);
1407     match_set_recirc_id(&match, 0);
1408     error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, 0, &ofpacts,
1409                                            &unused_rulep);
1410     if (error) {
1411         return error;
1412     }
1413
1414     /* Continue rule lookups for not-matched recirc rules from table 0.
1415      *
1416      * (priority=1), actions=resubmit(, 0)
1417      */
1418     resubmit = ofpact_put_RESUBMIT(&ofpacts);
1419     resubmit->in_port = OFPP_IN_PORT;
1420     resubmit->table_id = 0;
1421
1422     match_init_catchall(&match);
1423     error = ofproto_dpif_add_internal_flow(ofproto, &match, 1, 0, &ofpacts,
1424                                            &unused_rulep);
1425
1426     return error;
1427 }
1428
1429 static void
1430 destruct(struct ofproto *ofproto_)
1431 {
1432     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1433     struct ofproto_packet_in *pin, *next_pin;
1434     struct rule_dpif *rule;
1435     struct oftable *table;
1436     struct ovs_list pins;
1437
1438     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1439     xlate_txn_start();
1440     xlate_remove_ofproto(ofproto);
1441     xlate_txn_commit();
1442
1443     /* Ensure that the upcall processing threads have no remaining references
1444      * to the ofproto or anything in it. */
1445     udpif_synchronize(ofproto->backer->udpif);
1446
1447     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1448
1449     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1450         CLS_FOR_EACH (rule, up.cr, &table->cls) {
1451             ofproto_rule_delete(&ofproto->up, &rule->up);
1452         }
1453     }
1454
1455     guarded_list_pop_all(&ofproto->pins, &pins);
1456     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1457         list_remove(&pin->list_node);
1458         free(CONST_CAST(void *, pin->up.packet));
1459         free(pin);
1460     }
1461     guarded_list_destroy(&ofproto->pins);
1462
1463     dpif_backer_recirc_clear_ofproto(ofproto->backer, ofproto);
1464
1465     mbridge_unref(ofproto->mbridge);
1466
1467     netflow_unref(ofproto->netflow);
1468     dpif_sflow_unref(ofproto->sflow);
1469     dpif_ipfix_unref(ofproto->ipfix);
1470     hmap_destroy(&ofproto->bundles);
1471     mac_learning_unref(ofproto->ml);
1472     mcast_snooping_unref(ofproto->ms);
1473
1474     hmap_destroy(&ofproto->vlandev_map);
1475     hmap_destroy(&ofproto->realdev_vid_map);
1476
1477     sset_destroy(&ofproto->ports);
1478     sset_destroy(&ofproto->ghost_ports);
1479     sset_destroy(&ofproto->port_poll_set);
1480
1481     ovs_mutex_destroy(&ofproto->stats_mutex);
1482     ovs_mutex_destroy(&ofproto->vsp_mutex);
1483
1484     seq_destroy(ofproto->pins_seq);
1485
1486     close_dpif_backer(ofproto->backer);
1487 }
1488
1489 static int
1490 run(struct ofproto *ofproto_)
1491 {
1492     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1493     uint64_t new_seq, new_dump_seq;
1494
1495     if (mbridge_need_revalidate(ofproto->mbridge)) {
1496         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1497         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1498         mac_learning_flush(ofproto->ml);
1499         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1500         mcast_snooping_mdb_flush(ofproto->ms);
1501     }
1502
1503     /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
1504      * flow restore.  Even though nothing is processed during flow restore,
1505      * all queued 'pins' will be handled immediately when flow restore
1506      * completes. */
1507     ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1508
1509     /* Do not perform any periodic activity required by 'ofproto' while
1510      * waiting for flow restore to complete. */
1511     if (!ofproto_get_flow_restore_wait()) {
1512         struct ofproto_packet_in *pin, *next_pin;
1513         struct ovs_list pins;
1514
1515         guarded_list_pop_all(&ofproto->pins, &pins);
1516         LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1517             connmgr_send_packet_in(ofproto->up.connmgr, pin);
1518             list_remove(&pin->list_node);
1519             free(CONST_CAST(void *, pin->up.packet));
1520             free(pin);
1521         }
1522     }
1523
1524     if (ofproto->netflow) {
1525         netflow_run(ofproto->netflow);
1526     }
1527     if (ofproto->sflow) {
1528         dpif_sflow_run(ofproto->sflow);
1529     }
1530     if (ofproto->ipfix) {
1531         dpif_ipfix_run(ofproto->ipfix);
1532     }
1533
1534     new_seq = seq_read(connectivity_seq_get());
1535     if (ofproto->change_seq != new_seq) {
1536         struct ofport_dpif *ofport;
1537
1538         HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1539             port_run(ofport);
1540         }
1541
1542         ofproto->change_seq = new_seq;
1543     }
1544     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1545         struct ofbundle *bundle;
1546
1547         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1548             bundle_run(bundle);
1549         }
1550     }
1551
1552     stp_run(ofproto);
1553     rstp_run(ofproto);
1554     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1555     if (mac_learning_run(ofproto->ml)) {
1556         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1557     }
1558     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1559
1560     if (mcast_snooping_run(ofproto->ms)) {
1561         ofproto->backer->need_revalidate = REV_MCAST_SNOOPING;
1562     }
1563
1564     new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1565     if (ofproto->dump_seq != new_dump_seq) {
1566         struct rule *rule, *next_rule;
1567
1568         /* We know stats are relatively fresh, so now is a good time to do some
1569          * periodic work. */
1570         ofproto->dump_seq = new_dump_seq;
1571
1572         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1573          * has passed. */
1574         ovs_mutex_lock(&ofproto_mutex);
1575         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1576                             &ofproto->up.expirable) {
1577             rule_expire(rule_dpif_cast(rule));
1578         }
1579         ovs_mutex_unlock(&ofproto_mutex);
1580
1581         /* All outstanding data in existing flows has been accounted, so it's a
1582          * good time to do bond rebalancing. */
1583         if (ofproto->has_bonded_bundles) {
1584             struct ofbundle *bundle;
1585
1586             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1587                 if (bundle->bond) {
1588                     bond_rebalance(bundle->bond);
1589                 }
1590             }
1591         }
1592     }
1593     return 0;
1594 }
1595
1596 static void
1597 wait(struct ofproto *ofproto_)
1598 {
1599     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1600
1601     if (ofproto_get_flow_restore_wait()) {
1602         return;
1603     }
1604
1605     if (ofproto->sflow) {
1606         dpif_sflow_wait(ofproto->sflow);
1607     }
1608     if (ofproto->ipfix) {
1609         dpif_ipfix_wait(ofproto->ipfix);
1610     }
1611     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1612         struct ofbundle *bundle;
1613
1614         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1615             bundle_wait(bundle);
1616         }
1617     }
1618     if (ofproto->netflow) {
1619         netflow_wait(ofproto->netflow);
1620     }
1621     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1622     mac_learning_wait(ofproto->ml);
1623     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1624     mcast_snooping_wait(ofproto->ms);
1625     stp_wait(ofproto);
1626     if (ofproto->backer->need_revalidate) {
1627         /* Shouldn't happen, but if it does just go around again. */
1628         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1629         poll_immediate_wake();
1630     }
1631
1632     seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1633     seq_wait(ofproto->pins_seq, ofproto->pins_seqno);
1634 }
1635
1636 static void
1637 type_get_memory_usage(const char *type, struct simap *usage)
1638 {
1639     struct dpif_backer *backer;
1640
1641     backer = shash_find_data(&all_dpif_backers, type);
1642     if (backer) {
1643         udpif_get_memory_usage(backer->udpif, usage);
1644     }
1645 }
1646
1647 static void
1648 flush(struct ofproto *ofproto_)
1649 {
1650     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1651     struct dpif_backer *backer = ofproto->backer;
1652
1653     if (backer) {
1654         udpif_flush(backer->udpif);
1655     }
1656 }
1657
1658 static void
1659 query_tables(struct ofproto *ofproto,
1660              struct ofputil_table_features *features,
1661              struct ofputil_table_stats *stats)
1662 {
1663     strcpy(features->name, "classifier");
1664
1665     if (stats) {
1666         int i;
1667
1668         for (i = 0; i < ofproto->n_tables; i++) {
1669             unsigned long missed, matched;
1670
1671             atomic_read_relaxed(&ofproto->tables[i].n_matched, &matched);
1672             atomic_read_relaxed(&ofproto->tables[i].n_missed, &missed);
1673
1674             stats[i].matched_count = matched;
1675             stats[i].lookup_count = matched + missed;
1676         }
1677     }
1678 }
1679
1680 static struct ofport *
1681 port_alloc(void)
1682 {
1683     struct ofport_dpif *port = xmalloc(sizeof *port);
1684     return &port->up;
1685 }
1686
1687 static void
1688 port_dealloc(struct ofport *port_)
1689 {
1690     struct ofport_dpif *port = ofport_dpif_cast(port_);
1691     free(port);
1692 }
1693
1694 static int
1695 port_construct(struct ofport *port_)
1696 {
1697     struct ofport_dpif *port = ofport_dpif_cast(port_);
1698     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1699     const struct netdev *netdev = port->up.netdev;
1700     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1701     struct dpif_port dpif_port;
1702     int error;
1703
1704     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1705     port->bundle = NULL;
1706     port->cfm = NULL;
1707     port->bfd = NULL;
1708     port->may_enable = false;
1709     port->stp_port = NULL;
1710     port->stp_state = STP_DISABLED;
1711     port->rstp_port = NULL;
1712     port->rstp_state = RSTP_DISABLED;
1713     port->is_tunnel = false;
1714     port->peer = NULL;
1715     port->qdscp = NULL;
1716     port->n_qdscp = 0;
1717     port->realdev_ofp_port = 0;
1718     port->vlandev_vid = 0;
1719     port->carrier_seq = netdev_get_carrier_resets(netdev);
1720     port->is_layer3 = netdev_vport_is_layer3(netdev);
1721
1722     if (netdev_vport_is_patch(netdev)) {
1723         /* By bailing out here, we don't submit the port to the sFlow module
1724          * to be considered for counter polling export.  This is correct
1725          * because the patch port represents an interface that sFlow considers
1726          * to be "internal" to the switch as a whole, and therefore not an
1727          * candidate for counter polling. */
1728         port->odp_port = ODPP_NONE;
1729         ofport_update_peer(port);
1730         return 0;
1731     }
1732
1733     error = dpif_port_query_by_name(ofproto->backer->dpif,
1734                                     netdev_vport_get_dpif_port(netdev, namebuf,
1735                                                                sizeof namebuf),
1736                                     &dpif_port);
1737     if (error) {
1738         return error;
1739     }
1740
1741     port->odp_port = dpif_port.port_no;
1742
1743     if (netdev_get_tunnel_config(netdev)) {
1744         atomic_count_inc(&ofproto->backer->tnl_count);
1745         tnl_port_add(port, port->up.netdev, port->odp_port,
1746                      ovs_native_tunneling_is_on(ofproto), namebuf);
1747         port->is_tunnel = true;
1748         if (ofproto->ipfix) {
1749            dpif_ipfix_add_tunnel_port(ofproto->ipfix, port_, port->odp_port);
1750         }
1751     } else {
1752         /* Sanity-check that a mapping doesn't already exist.  This
1753          * shouldn't happen for non-tunnel ports. */
1754         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1755             VLOG_ERR("port %s already has an OpenFlow port number",
1756                      dpif_port.name);
1757             dpif_port_destroy(&dpif_port);
1758             return EBUSY;
1759         }
1760
1761         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1762         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1763                     hash_odp_port(port->odp_port));
1764         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1765     }
1766     dpif_port_destroy(&dpif_port);
1767
1768     if (ofproto->sflow) {
1769         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1770     }
1771
1772     return 0;
1773 }
1774
1775 static void
1776 port_destruct(struct ofport *port_)
1777 {
1778     struct ofport_dpif *port = ofport_dpif_cast(port_);
1779     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1780     const char *devname = netdev_get_name(port->up.netdev);
1781     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1782     const char *dp_port_name;
1783
1784     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1785     xlate_txn_start();
1786     xlate_ofport_remove(port);
1787     xlate_txn_commit();
1788
1789     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1790                                               sizeof namebuf);
1791     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1792         /* The underlying device is still there, so delete it.  This
1793          * happens when the ofproto is being destroyed, since the caller
1794          * assumes that removal of attached ports will happen as part of
1795          * destruction. */
1796         if (!port->is_tunnel) {
1797             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1798         }
1799     }
1800
1801     if (port->peer) {
1802         port->peer->peer = NULL;
1803         port->peer = NULL;
1804     }
1805
1806     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1807         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1808         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1809         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1810     }
1811
1812     if (port->is_tunnel) {
1813         atomic_count_dec(&ofproto->backer->tnl_count);
1814     }
1815
1816     if (port->is_tunnel && ofproto->ipfix) {
1817        dpif_ipfix_del_tunnel_port(ofproto->ipfix, port->odp_port);
1818     }
1819
1820     tnl_port_del(port);
1821     sset_find_and_delete(&ofproto->ports, devname);
1822     sset_find_and_delete(&ofproto->ghost_ports, devname);
1823     bundle_remove(port_);
1824     set_cfm(port_, NULL);
1825     set_bfd(port_, NULL);
1826     if (port->stp_port) {
1827         stp_port_disable(port->stp_port);
1828     }
1829     set_rstp_port(port_, NULL);
1830     if (ofproto->sflow) {
1831         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1832     }
1833
1834     free(port->qdscp);
1835 }
1836
1837 static void
1838 port_modified(struct ofport *port_)
1839 {
1840     struct ofport_dpif *port = ofport_dpif_cast(port_);
1841     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1842     struct netdev *netdev = port->up.netdev;
1843
1844     if (port->bundle && port->bundle->bond) {
1845         bond_slave_set_netdev(port->bundle->bond, port, netdev);
1846     }
1847
1848     if (port->cfm) {
1849         cfm_set_netdev(port->cfm, netdev);
1850     }
1851
1852     if (port->bfd) {
1853         bfd_set_netdev(port->bfd, netdev);
1854     }
1855
1856     ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1857                                      port->up.pp.hw_addr);
1858
1859     netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1860
1861     if (port->is_tunnel) {
1862         struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1863
1864         if (tnl_port_reconfigure(port, netdev, port->odp_port,
1865                                  ovs_native_tunneling_is_on(ofproto), namebuf)) {
1866             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1867         }
1868     }
1869
1870     ofport_update_peer(port);
1871 }
1872
1873 static void
1874 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1875 {
1876     struct ofport_dpif *port = ofport_dpif_cast(port_);
1877     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1878     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1879
1880     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1881                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1882                    OFPUTIL_PC_NO_PACKET_IN)) {
1883         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1884
1885         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1886             bundle_update(port->bundle);
1887         }
1888     }
1889 }
1890
1891 static int
1892 set_sflow(struct ofproto *ofproto_,
1893           const struct ofproto_sflow_options *sflow_options)
1894 {
1895     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1896     struct dpif_sflow *ds = ofproto->sflow;
1897
1898     if (sflow_options) {
1899         uint32_t old_probability = ds ? dpif_sflow_get_probability(ds) : 0;
1900         if (!ds) {
1901             struct ofport_dpif *ofport;
1902
1903             ds = ofproto->sflow = dpif_sflow_create();
1904             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1905                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1906             }
1907         }
1908         dpif_sflow_set_options(ds, sflow_options);
1909         if (dpif_sflow_get_probability(ds) != old_probability) {
1910             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1911         }
1912     } else {
1913         if (ds) {
1914             dpif_sflow_unref(ds);
1915             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1916             ofproto->sflow = NULL;
1917         }
1918     }
1919     return 0;
1920 }
1921
1922 static int
1923 set_ipfix(
1924     struct ofproto *ofproto_,
1925     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1926     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1927     size_t n_flow_exporters_options)
1928 {
1929     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1930     struct dpif_ipfix *di = ofproto->ipfix;
1931     bool has_options = bridge_exporter_options || flow_exporters_options;
1932     bool new_di = false;
1933
1934     if (has_options && !di) {
1935         di = ofproto->ipfix = dpif_ipfix_create();
1936         new_di = true;
1937     }
1938
1939     if (di) {
1940         /* Call set_options in any case to cleanly flush the flow
1941          * caches in the last exporters that are to be destroyed. */
1942         dpif_ipfix_set_options(
1943             di, bridge_exporter_options, flow_exporters_options,
1944             n_flow_exporters_options);
1945
1946         /* Add tunnel ports only when a new ipfix created */
1947         if (new_di == true) {
1948             struct ofport_dpif *ofport;
1949             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1950                 if (ofport->is_tunnel == true) {
1951                     dpif_ipfix_add_tunnel_port(di, &ofport->up, ofport->odp_port);
1952                 }
1953             }
1954         }
1955
1956         if (!has_options) {
1957             dpif_ipfix_unref(di);
1958             ofproto->ipfix = NULL;
1959         }
1960     }
1961
1962     return 0;
1963 }
1964
1965 static int
1966 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1967 {
1968     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1969     int error = 0;
1970
1971     if (s) {
1972         if (!ofport->cfm) {
1973             struct ofproto_dpif *ofproto;
1974
1975             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1976             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1977             ofport->cfm = cfm_create(ofport->up.netdev);
1978         }
1979
1980         if (cfm_configure(ofport->cfm, s)) {
1981             error = 0;
1982             goto out;
1983         }
1984
1985         error = EINVAL;
1986     }
1987     cfm_unref(ofport->cfm);
1988     ofport->cfm = NULL;
1989 out:
1990     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1991                                      ofport->up.pp.hw_addr);
1992     return error;
1993 }
1994
1995 static bool
1996 cfm_status_changed(struct ofport *ofport_)
1997 {
1998     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1999
2000     return ofport->cfm ? cfm_check_status_change(ofport->cfm) : true;
2001 }
2002
2003 static int
2004 get_cfm_status(const struct ofport *ofport_,
2005                struct cfm_status *status)
2006 {
2007     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2008     int ret = 0;
2009
2010     if (ofport->cfm) {
2011         cfm_get_status(ofport->cfm, status);
2012     } else {
2013         ret = ENOENT;
2014     }
2015
2016     return ret;
2017 }
2018
2019 static int
2020 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2021 {
2022     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2023     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2024     struct bfd *old;
2025
2026     old = ofport->bfd;
2027     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2028                                 cfg, ofport->up.netdev);
2029     if (ofport->bfd != old) {
2030         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2031     }
2032     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
2033                                      ofport->up.pp.hw_addr);
2034     return 0;
2035 }
2036
2037 static bool
2038 bfd_status_changed(struct ofport *ofport_)
2039 {
2040     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2041
2042     return ofport->bfd ? bfd_check_status_change(ofport->bfd) : true;
2043 }
2044
2045 static int
2046 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2047 {
2048     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2049     int ret = 0;
2050
2051     if (ofport->bfd) {
2052         bfd_get_status(ofport->bfd, smap);
2053     } else {
2054         ret = ENOENT;
2055     }
2056
2057     return ret;
2058 }
2059 \f
2060 /* Spanning Tree. */
2061
2062 /* Called while rstp_mutex is held. */
2063 static void
2064 rstp_send_bpdu_cb(struct ofpbuf *pkt, void *ofport_, void *ofproto_)
2065 {
2066     struct ofproto_dpif *ofproto = ofproto_;
2067     struct ofport_dpif *ofport = ofport_;
2068     struct eth_header *eth = ofpbuf_l2(pkt);
2069
2070     netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2071     if (eth_addr_is_zero(eth->eth_src)) {
2072         VLOG_WARN_RL(&rl, "%s port %d: cannot send RSTP BPDU on a port which "
2073                      "does not have a configured source MAC address.",
2074                      ofproto->up.name, ofp_to_u16(ofport->up.ofp_port));
2075     } else {
2076         ofproto_dpif_send_packet(ofport, pkt);
2077     }
2078     ofpbuf_delete(pkt);
2079 }
2080
2081 static void
2082 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2083 {
2084     struct ofproto_dpif *ofproto = ofproto_;
2085     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2086     struct ofport_dpif *ofport;
2087
2088     ofport = stp_port_get_aux(sp);
2089     if (!ofport) {
2090         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2091                      ofproto->up.name, port_num);
2092     } else {
2093         struct eth_header *eth = ofpbuf_l2(pkt);
2094
2095         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2096         if (eth_addr_is_zero(eth->eth_src)) {
2097             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2098                          "with unknown MAC", ofproto->up.name, port_num);
2099         } else {
2100             ofproto_dpif_send_packet(ofport, pkt);
2101         }
2102     }
2103     ofpbuf_delete(pkt);
2104 }
2105
2106 /* Configure RSTP on 'ofproto_' using the settings defined in 's'. */
2107 static void
2108 set_rstp(struct ofproto *ofproto_, const struct ofproto_rstp_settings *s)
2109 {
2110     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2111
2112     /* Only revalidate flows if the configuration changed. */
2113     if (!s != !ofproto->rstp) {
2114         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2115     }
2116
2117     if (s) {
2118         if (!ofproto->rstp) {
2119             ofproto->rstp = rstp_create(ofproto_->name, s->address,
2120                                         rstp_send_bpdu_cb, ofproto);
2121             ofproto->rstp_last_tick = time_msec();
2122         }
2123         rstp_set_bridge_address(ofproto->rstp, s->address);
2124         rstp_set_bridge_priority(ofproto->rstp, s->priority);
2125         rstp_set_bridge_ageing_time(ofproto->rstp, s->ageing_time);
2126         rstp_set_bridge_force_protocol_version(ofproto->rstp,
2127                                                s->force_protocol_version);
2128         rstp_set_bridge_max_age(ofproto->rstp, s->bridge_max_age);
2129         rstp_set_bridge_forward_delay(ofproto->rstp, s->bridge_forward_delay);
2130         rstp_set_bridge_transmit_hold_count(ofproto->rstp,
2131                                             s->transmit_hold_count);
2132     } else {
2133         struct ofport *ofport;
2134         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2135             set_rstp_port(ofport, NULL);
2136         }
2137         rstp_unref(ofproto->rstp);
2138         ofproto->rstp = NULL;
2139     }
2140 }
2141
2142 static void
2143 get_rstp_status(struct ofproto *ofproto_, struct ofproto_rstp_status *s)
2144 {
2145     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2146
2147     if (ofproto->rstp) {
2148         s->enabled = true;
2149         s->root_id = rstp_get_root_id(ofproto->rstp);
2150         s->bridge_id = rstp_get_bridge_id(ofproto->rstp);
2151         s->designated_id = rstp_get_designated_id(ofproto->rstp);
2152         s->root_path_cost = rstp_get_root_path_cost(ofproto->rstp);
2153         s->designated_port_id = rstp_get_designated_port_id(ofproto->rstp);
2154         s->bridge_port_id = rstp_get_bridge_port_id(ofproto->rstp);
2155     } else {
2156         s->enabled = false;
2157     }
2158 }
2159
2160 static void
2161 update_rstp_port_state(struct ofport_dpif *ofport)
2162 {
2163     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2164     enum rstp_state state;
2165
2166     /* Figure out new state. */
2167     state = ofport->rstp_port ? rstp_port_get_state(ofport->rstp_port)
2168         : RSTP_DISABLED;
2169
2170     /* Update state. */
2171     if (ofport->rstp_state != state) {
2172         enum ofputil_port_state of_state;
2173         bool fwd_change;
2174
2175         VLOG_DBG("port %s: RSTP state changed from %s to %s",
2176                  netdev_get_name(ofport->up.netdev),
2177                  rstp_state_name(ofport->rstp_state),
2178                  rstp_state_name(state));
2179
2180         if (rstp_learn_in_state(ofport->rstp_state)
2181             != rstp_learn_in_state(state)) {
2182             /* XXX: Learning action flows should also be flushed. */
2183             if (ofport->bundle) {
2184                 if (!rstp_shift_root_learned_address(ofproto->rstp)
2185                     || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2186                     bundle_flush_macs(ofport->bundle, false);
2187                 }
2188             }
2189         }
2190         fwd_change = rstp_forward_in_state(ofport->rstp_state)
2191             != rstp_forward_in_state(state);
2192
2193         ofproto->backer->need_revalidate = REV_RSTP;
2194         ofport->rstp_state = state;
2195
2196         if (fwd_change && ofport->bundle) {
2197             bundle_update(ofport->bundle);
2198         }
2199
2200         /* Update the RSTP state bits in the OpenFlow port description. */
2201         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2202         of_state |= (state == RSTP_LEARNING ? OFPUTIL_PS_STP_LEARN
2203                 : state == RSTP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2204                 : state == RSTP_DISCARDING ?  OFPUTIL_PS_STP_LISTEN
2205                 : 0);
2206         ofproto_port_set_state(&ofport->up, of_state);
2207     }
2208 }
2209
2210 static void
2211 rstp_run(struct ofproto_dpif *ofproto)
2212 {
2213     if (ofproto->rstp) {
2214         long long int now = time_msec();
2215         long long int elapsed = now - ofproto->rstp_last_tick;
2216         struct rstp_port *rp;
2217         struct ofport_dpif *ofport;
2218
2219         /* Every second, decrease the values of the timers. */
2220         if (elapsed >= 1000) {
2221             rstp_tick_timers(ofproto->rstp);
2222             ofproto->rstp_last_tick = now;
2223         }
2224         rp = NULL;
2225         while ((ofport = rstp_get_next_changed_port_aux(ofproto->rstp, &rp))) {
2226             update_rstp_port_state(ofport);
2227         }
2228         rp = NULL;
2229         ofport = NULL;
2230         /* FIXME: This check should be done on-event (i.e., when setting
2231          * p->fdb_flush) and not periodically.
2232          */
2233         while ((ofport = rstp_check_and_reset_fdb_flush(ofproto->rstp, &rp))) {
2234             if (!rstp_shift_root_learned_address(ofproto->rstp)
2235                 || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2236                 bundle_flush_macs(ofport->bundle, false);
2237             }
2238         }
2239
2240         if (rstp_shift_root_learned_address(ofproto->rstp)) {
2241             bundle_move(((struct ofport_dpif *)rstp_get_old_root_aux(ofproto->rstp))->bundle,
2242                         ((struct ofport_dpif *)rstp_get_new_root_aux(ofproto->rstp))->bundle);
2243             rstp_reset_root_changed(ofproto->rstp);
2244         }
2245     }
2246 }
2247
2248 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2249 static int
2250 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2251 {
2252     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2253
2254     /* Only revalidate flows if the configuration changed. */
2255     if (!s != !ofproto->stp) {
2256         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2257     }
2258
2259     if (s) {
2260         if (!ofproto->stp) {
2261             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2262                                       send_bpdu_cb, ofproto);
2263             ofproto->stp_last_tick = time_msec();
2264         }
2265
2266         stp_set_bridge_id(ofproto->stp, s->system_id);
2267         stp_set_bridge_priority(ofproto->stp, s->priority);
2268         stp_set_hello_time(ofproto->stp, s->hello_time);
2269         stp_set_max_age(ofproto->stp, s->max_age);
2270         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2271     }  else {
2272         struct ofport *ofport;
2273
2274         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2275             set_stp_port(ofport, NULL);
2276         }
2277
2278         stp_unref(ofproto->stp);
2279         ofproto->stp = NULL;
2280     }
2281
2282     return 0;
2283 }
2284
2285 static int
2286 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2287 {
2288     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2289
2290     if (ofproto->stp) {
2291         s->enabled = true;
2292         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2293         s->designated_root = stp_get_designated_root(ofproto->stp);
2294         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2295     } else {
2296         s->enabled = false;
2297     }
2298
2299     return 0;
2300 }
2301
2302 static void
2303 update_stp_port_state(struct ofport_dpif *ofport)
2304 {
2305     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2306     enum stp_state state;
2307
2308     /* Figure out new state. */
2309     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2310                              : STP_DISABLED;
2311
2312     /* Update state. */
2313     if (ofport->stp_state != state) {
2314         enum ofputil_port_state of_state;
2315         bool fwd_change;
2316
2317         VLOG_DBG("port %s: STP state changed from %s to %s",
2318                  netdev_get_name(ofport->up.netdev),
2319                  stp_state_name(ofport->stp_state),
2320                  stp_state_name(state));
2321         if (stp_learn_in_state(ofport->stp_state)
2322                 != stp_learn_in_state(state)) {
2323             /* xxx Learning action flows should also be flushed. */
2324             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2325             mac_learning_flush(ofproto->ml);
2326             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2327             mcast_snooping_mdb_flush(ofproto->ms);
2328         }
2329         fwd_change = stp_forward_in_state(ofport->stp_state)
2330                         != stp_forward_in_state(state);
2331
2332         ofproto->backer->need_revalidate = REV_STP;
2333         ofport->stp_state = state;
2334         ofport->stp_state_entered = time_msec();
2335
2336         if (fwd_change && ofport->bundle) {
2337             bundle_update(ofport->bundle);
2338         }
2339
2340         /* Update the STP state bits in the OpenFlow port description. */
2341         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2342         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2343                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2344                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2345                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2346                      : 0);
2347         ofproto_port_set_state(&ofport->up, of_state);
2348     }
2349 }
2350
2351 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2352  * caller is responsible for assigning STP port numbers and ensuring
2353  * there are no duplicates. */
2354 static int
2355 set_stp_port(struct ofport *ofport_,
2356              const struct ofproto_port_stp_settings *s)
2357 {
2358     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2359     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2360     struct stp_port *sp = ofport->stp_port;
2361
2362     if (!s || !s->enable) {
2363         if (sp) {
2364             ofport->stp_port = NULL;
2365             stp_port_disable(sp);
2366             update_stp_port_state(ofport);
2367         }
2368         return 0;
2369     } else if (sp && stp_port_no(sp) != s->port_num
2370                && ofport == stp_port_get_aux(sp)) {
2371         /* The port-id changed, so disable the old one if it's not
2372          * already in use by another port. */
2373         stp_port_disable(sp);
2374     }
2375
2376     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2377
2378     /* Set name before enabling the port so that debugging messages can print
2379      * the name. */
2380     stp_port_set_name(sp, netdev_get_name(ofport->up.netdev));
2381     stp_port_enable(sp);
2382
2383     stp_port_set_aux(sp, ofport);
2384     stp_port_set_priority(sp, s->priority);
2385     stp_port_set_path_cost(sp, s->path_cost);
2386
2387     update_stp_port_state(ofport);
2388
2389     return 0;
2390 }
2391
2392 static int
2393 get_stp_port_status(struct ofport *ofport_,
2394                     struct ofproto_port_stp_status *s)
2395 {
2396     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2397     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2398     struct stp_port *sp = ofport->stp_port;
2399
2400     if (!ofproto->stp || !sp) {
2401         s->enabled = false;
2402         return 0;
2403     }
2404
2405     s->enabled = true;
2406     s->port_id = stp_port_get_id(sp);
2407     s->state = stp_port_get_state(sp);
2408     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2409     s->role = stp_port_get_role(sp);
2410
2411     return 0;
2412 }
2413
2414 static int
2415 get_stp_port_stats(struct ofport *ofport_,
2416                    struct ofproto_port_stp_stats *s)
2417 {
2418     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2419     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2420     struct stp_port *sp = ofport->stp_port;
2421
2422     if (!ofproto->stp || !sp) {
2423         s->enabled = false;
2424         return 0;
2425     }
2426
2427     s->enabled = true;
2428     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2429
2430     return 0;
2431 }
2432
2433 static void
2434 stp_run(struct ofproto_dpif *ofproto)
2435 {
2436     if (ofproto->stp) {
2437         long long int now = time_msec();
2438         long long int elapsed = now - ofproto->stp_last_tick;
2439         struct stp_port *sp;
2440
2441         if (elapsed > 0) {
2442             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2443             ofproto->stp_last_tick = now;
2444         }
2445         while (stp_get_changed_port(ofproto->stp, &sp)) {
2446             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2447
2448             if (ofport) {
2449                 update_stp_port_state(ofport);
2450             }
2451         }
2452
2453         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2454             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2455             mac_learning_flush(ofproto->ml);
2456             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2457             mcast_snooping_mdb_flush(ofproto->ms);
2458         }
2459     }
2460 }
2461
2462 static void
2463 stp_wait(struct ofproto_dpif *ofproto)
2464 {
2465     if (ofproto->stp) {
2466         poll_timer_wait(1000);
2467     }
2468 }
2469
2470 /* Configures RSTP on 'ofport_' using the settings defined in 's'.  The
2471  * caller is responsible for assigning RSTP port numbers and ensuring
2472  * there are no duplicates. */
2473 static void
2474 set_rstp_port(struct ofport *ofport_,
2475               const struct ofproto_port_rstp_settings *s)
2476 {
2477     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2478     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2479     struct rstp_port *rp = ofport->rstp_port;
2480
2481     if (!s || !s->enable) {
2482         if (rp) {
2483             rstp_port_unref(rp);
2484             ofport->rstp_port = NULL;
2485             update_rstp_port_state(ofport);
2486         }
2487         return;
2488     }
2489
2490     /* Check if need to add a new port. */
2491     if (!rp) {
2492         rp = ofport->rstp_port = rstp_add_port(ofproto->rstp);
2493     }
2494
2495     rstp_port_set(rp, s->port_num, s->priority, s->path_cost,
2496                   s->admin_edge_port, s->auto_edge,
2497                   s->admin_p2p_mac_state, s->admin_port_state, s->mcheck,
2498                   ofport);
2499     update_rstp_port_state(ofport);
2500     /* Synchronize operational status. */
2501     rstp_port_set_mac_operational(rp, ofport->may_enable);
2502 }
2503
2504 static void
2505 get_rstp_port_status(struct ofport *ofport_,
2506         struct ofproto_port_rstp_status *s)
2507 {
2508     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2509     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2510     struct rstp_port *rp = ofport->rstp_port;
2511
2512     if (!ofproto->rstp || !rp) {
2513         s->enabled = false;
2514         return;
2515     }
2516
2517     s->enabled = true;
2518     rstp_port_get_status(rp, &s->port_id, &s->state, &s->role,
2519                          &s->designated_bridge_id, &s->designated_port_id,
2520                          &s->designated_path_cost, &s->tx_count,
2521                          &s->rx_count, &s->error_count, &s->uptime);
2522 }
2523
2524 \f
2525 static int
2526 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2527            size_t n_qdscp)
2528 {
2529     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2530     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2531
2532     if (ofport->n_qdscp != n_qdscp
2533         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2534                               n_qdscp * sizeof *qdscp))) {
2535         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2536         free(ofport->qdscp);
2537         ofport->qdscp = n_qdscp
2538             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2539             : NULL;
2540         ofport->n_qdscp = n_qdscp;
2541     }
2542
2543     return 0;
2544 }
2545 \f
2546 /* Bundles. */
2547
2548 /* Expires all MAC learning entries associated with 'bundle' and forces its
2549  * ofproto to revalidate every flow.
2550  *
2551  * Normally MAC learning entries are removed only from the ofproto associated
2552  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2553  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2554  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2555  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2556  * with the host from which it migrated. */
2557 static void
2558 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2559 {
2560     struct ofproto_dpif *ofproto = bundle->ofproto;
2561     struct mac_learning *ml = ofproto->ml;
2562     struct mac_entry *mac, *next_mac;
2563
2564     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2565     ovs_rwlock_wrlock(&ml->rwlock);
2566     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2567         if (mac_entry_get_port(ml, mac) == bundle) {
2568             if (all_ofprotos) {
2569                 struct ofproto_dpif *o;
2570
2571                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2572                     if (o != ofproto) {
2573                         struct mac_entry *e;
2574
2575                         ovs_rwlock_wrlock(&o->ml->rwlock);
2576                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2577                         if (e) {
2578                             mac_learning_expire(o->ml, e);
2579                         }
2580                         ovs_rwlock_unlock(&o->ml->rwlock);
2581                     }
2582                 }
2583             }
2584
2585             mac_learning_expire(ml, mac);
2586         }
2587     }
2588     ovs_rwlock_unlock(&ml->rwlock);
2589 }
2590
2591 static void
2592 bundle_move(struct ofbundle *old, struct ofbundle *new)
2593 {
2594     struct ofproto_dpif *ofproto = old->ofproto;
2595     struct mac_learning *ml = ofproto->ml;
2596     struct mac_entry *mac, *next_mac;
2597
2598     ovs_assert(new->ofproto == old->ofproto);
2599
2600     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2601     ovs_rwlock_wrlock(&ml->rwlock);
2602     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2603         if (mac_entry_get_port(ml, mac) == old) {
2604             mac_entry_set_port(ml, mac, new);
2605         }
2606     }
2607     ovs_rwlock_unlock(&ml->rwlock);
2608 }
2609
2610 static struct ofbundle *
2611 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2612 {
2613     struct ofbundle *bundle;
2614
2615     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2616                              &ofproto->bundles) {
2617         if (bundle->aux == aux) {
2618             return bundle;
2619         }
2620     }
2621     return NULL;
2622 }
2623
2624 static void
2625 bundle_update(struct ofbundle *bundle)
2626 {
2627     struct ofport_dpif *port;
2628
2629     bundle->floodable = true;
2630     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2631         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2632             || port->is_layer3
2633             || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2634             || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2635             bundle->floodable = false;
2636             break;
2637         }
2638     }
2639 }
2640
2641 static void
2642 bundle_del_port(struct ofport_dpif *port)
2643 {
2644     struct ofbundle *bundle = port->bundle;
2645
2646     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2647
2648     list_remove(&port->bundle_node);
2649     port->bundle = NULL;
2650
2651     if (bundle->lacp) {
2652         lacp_slave_unregister(bundle->lacp, port);
2653     }
2654     if (bundle->bond) {
2655         bond_slave_unregister(bundle->bond, port);
2656     }
2657
2658     bundle_update(bundle);
2659 }
2660
2661 static bool
2662 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2663                 struct lacp_slave_settings *lacp)
2664 {
2665     struct ofport_dpif *port;
2666
2667     port = get_ofp_port(bundle->ofproto, ofp_port);
2668     if (!port) {
2669         return false;
2670     }
2671
2672     if (port->bundle != bundle) {
2673         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2674         if (port->bundle) {
2675             bundle_remove(&port->up);
2676         }
2677
2678         port->bundle = bundle;
2679         list_push_back(&bundle->ports, &port->bundle_node);
2680         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2681             || port->is_layer3
2682             || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2683             || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2684             bundle->floodable = false;
2685         }
2686     }
2687     if (lacp) {
2688         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2689         lacp_slave_register(bundle->lacp, port, lacp);
2690     }
2691
2692     return true;
2693 }
2694
2695 static void
2696 bundle_destroy(struct ofbundle *bundle)
2697 {
2698     struct ofproto_dpif *ofproto;
2699     struct ofport_dpif *port, *next_port;
2700
2701     if (!bundle) {
2702         return;
2703     }
2704
2705     ofproto = bundle->ofproto;
2706     mbridge_unregister_bundle(ofproto->mbridge, bundle);
2707
2708     xlate_txn_start();
2709     xlate_bundle_remove(bundle);
2710     xlate_txn_commit();
2711
2712     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2713         bundle_del_port(port);
2714     }
2715
2716     bundle_flush_macs(bundle, true);
2717     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2718     free(bundle->name);
2719     free(bundle->trunks);
2720     lacp_unref(bundle->lacp);
2721     bond_unref(bundle->bond);
2722     free(bundle);
2723 }
2724
2725 static int
2726 bundle_set(struct ofproto *ofproto_, void *aux,
2727            const struct ofproto_bundle_settings *s)
2728 {
2729     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2730     bool need_flush = false;
2731     struct ofport_dpif *port;
2732     struct ofbundle *bundle;
2733     unsigned long *trunks;
2734     int vlan;
2735     size_t i;
2736     bool ok;
2737
2738     if (!s) {
2739         bundle_destroy(bundle_lookup(ofproto, aux));
2740         return 0;
2741     }
2742
2743     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2744     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2745
2746     bundle = bundle_lookup(ofproto, aux);
2747     if (!bundle) {
2748         bundle = xmalloc(sizeof *bundle);
2749
2750         bundle->ofproto = ofproto;
2751         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2752                     hash_pointer(aux, 0));
2753         bundle->aux = aux;
2754         bundle->name = NULL;
2755
2756         list_init(&bundle->ports);
2757         bundle->vlan_mode = PORT_VLAN_TRUNK;
2758         bundle->vlan = -1;
2759         bundle->trunks = NULL;
2760         bundle->use_priority_tags = s->use_priority_tags;
2761         bundle->lacp = NULL;
2762         bundle->bond = NULL;
2763
2764         bundle->floodable = true;
2765         mbridge_register_bundle(ofproto->mbridge, bundle);
2766     }
2767
2768     if (!bundle->name || strcmp(s->name, bundle->name)) {
2769         free(bundle->name);
2770         bundle->name = xstrdup(s->name);
2771     }
2772
2773     /* LACP. */
2774     if (s->lacp) {
2775         ofproto->lacp_enabled = true;
2776         if (!bundle->lacp) {
2777             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2778             bundle->lacp = lacp_create();
2779         }
2780         lacp_configure(bundle->lacp, s->lacp);
2781     } else {
2782         lacp_unref(bundle->lacp);
2783         bundle->lacp = NULL;
2784     }
2785
2786     /* Update set of ports. */
2787     ok = true;
2788     for (i = 0; i < s->n_slaves; i++) {
2789         if (!bundle_add_port(bundle, s->slaves[i],
2790                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2791             ok = false;
2792         }
2793     }
2794     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2795         struct ofport_dpif *next_port;
2796
2797         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2798             for (i = 0; i < s->n_slaves; i++) {
2799                 if (s->slaves[i] == port->up.ofp_port) {
2800                     goto found;
2801                 }
2802             }
2803
2804             bundle_del_port(port);
2805         found: ;
2806         }
2807     }
2808     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2809
2810     if (list_is_empty(&bundle->ports)) {
2811         bundle_destroy(bundle);
2812         return EINVAL;
2813     }
2814
2815     /* Set VLAN tagging mode */
2816     if (s->vlan_mode != bundle->vlan_mode
2817         || s->use_priority_tags != bundle->use_priority_tags) {
2818         bundle->vlan_mode = s->vlan_mode;
2819         bundle->use_priority_tags = s->use_priority_tags;
2820         need_flush = true;
2821     }
2822
2823     /* Set VLAN tag. */
2824     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2825             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2826             : 0);
2827     if (vlan != bundle->vlan) {
2828         bundle->vlan = vlan;
2829         need_flush = true;
2830     }
2831
2832     /* Get trunked VLANs. */
2833     switch (s->vlan_mode) {
2834     case PORT_VLAN_ACCESS:
2835         trunks = NULL;
2836         break;
2837
2838     case PORT_VLAN_TRUNK:
2839         trunks = CONST_CAST(unsigned long *, s->trunks);
2840         break;
2841
2842     case PORT_VLAN_NATIVE_UNTAGGED:
2843     case PORT_VLAN_NATIVE_TAGGED:
2844         if (vlan != 0 && (!s->trunks
2845                           || !bitmap_is_set(s->trunks, vlan)
2846                           || bitmap_is_set(s->trunks, 0))) {
2847             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2848             if (s->trunks) {
2849                 trunks = bitmap_clone(s->trunks, 4096);
2850             } else {
2851                 trunks = bitmap_allocate1(4096);
2852             }
2853             bitmap_set1(trunks, vlan);
2854             bitmap_set0(trunks, 0);
2855         } else {
2856             trunks = CONST_CAST(unsigned long *, s->trunks);
2857         }
2858         break;
2859
2860     default:
2861         OVS_NOT_REACHED();
2862     }
2863     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2864         free(bundle->trunks);
2865         if (trunks == s->trunks) {
2866             bundle->trunks = vlan_bitmap_clone(trunks);
2867         } else {
2868             bundle->trunks = trunks;
2869             trunks = NULL;
2870         }
2871         need_flush = true;
2872     }
2873     if (trunks != s->trunks) {
2874         free(trunks);
2875     }
2876
2877     /* Bonding. */
2878     if (!list_is_short(&bundle->ports)) {
2879         bundle->ofproto->has_bonded_bundles = true;
2880         if (bundle->bond) {
2881             if (bond_reconfigure(bundle->bond, s->bond)) {
2882                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2883             }
2884         } else {
2885             bundle->bond = bond_create(s->bond, ofproto);
2886             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2887         }
2888
2889         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2890             bond_slave_register(bundle->bond, port,
2891                                 port->up.ofp_port, port->up.netdev);
2892         }
2893     } else {
2894         bond_unref(bundle->bond);
2895         bundle->bond = NULL;
2896     }
2897
2898     /* If we changed something that would affect MAC learning, un-learn
2899      * everything on this port and force flow revalidation. */
2900     if (need_flush) {
2901         bundle_flush_macs(bundle, false);
2902     }
2903
2904     return 0;
2905 }
2906
2907 static void
2908 bundle_remove(struct ofport *port_)
2909 {
2910     struct ofport_dpif *port = ofport_dpif_cast(port_);
2911     struct ofbundle *bundle = port->bundle;
2912
2913     if (bundle) {
2914         bundle_del_port(port);
2915         if (list_is_empty(&bundle->ports)) {
2916             bundle_destroy(bundle);
2917         } else if (list_is_short(&bundle->ports)) {
2918             bond_unref(bundle->bond);
2919             bundle->bond = NULL;
2920         }
2921     }
2922 }
2923
2924 static void
2925 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2926 {
2927     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2928     struct ofport_dpif *port = port_;
2929     uint8_t ea[ETH_ADDR_LEN];
2930     int error;
2931
2932     error = netdev_get_etheraddr(port->up.netdev, ea);
2933     if (!error) {
2934         struct ofpbuf packet;
2935         void *packet_pdu;
2936
2937         ofpbuf_init(&packet, 0);
2938         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2939                                  pdu_size);
2940         memcpy(packet_pdu, pdu, pdu_size);
2941
2942         ofproto_dpif_send_packet(port, &packet);
2943         ofpbuf_uninit(&packet);
2944     } else {
2945         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2946                     "%s (%s)", port->bundle->name,
2947                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2948     }
2949 }
2950
2951 static void
2952 bundle_send_learning_packets(struct ofbundle *bundle)
2953 {
2954     struct ofproto_dpif *ofproto = bundle->ofproto;
2955     struct ofpbuf *learning_packet;
2956     int error, n_packets, n_errors;
2957     struct mac_entry *e;
2958     struct ovs_list packets;
2959
2960     list_init(&packets);
2961     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2962     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2963         if (mac_entry_get_port(ofproto->ml, e) != bundle) {
2964             void *port_void;
2965
2966             learning_packet = bond_compose_learning_packet(bundle->bond,
2967                                                            e->mac, e->vlan,
2968                                                            &port_void);
2969             /* Temporarily use 'frame' as a private pointer (see below). */
2970             ovs_assert(learning_packet->frame == ofpbuf_data(learning_packet));
2971             learning_packet->frame = port_void;
2972             list_push_back(&packets, &learning_packet->list_node);
2973         }
2974     }
2975     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2976
2977     error = n_packets = n_errors = 0;
2978     LIST_FOR_EACH (learning_packet, list_node, &packets) {
2979         int ret;
2980         void *port_void = learning_packet->frame;
2981
2982         /* Restore 'frame'. */
2983         learning_packet->frame = ofpbuf_data(learning_packet);
2984         ret = ofproto_dpif_send_packet(port_void, learning_packet);
2985         if (ret) {
2986             error = ret;
2987             n_errors++;
2988         }
2989         n_packets++;
2990     }
2991     ofpbuf_list_delete(&packets);
2992
2993     if (n_errors) {
2994         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2995         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2996                      "packets, last error was: %s",
2997                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2998     } else {
2999         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3000                  bundle->name, n_packets);
3001     }
3002 }
3003
3004 static void
3005 bundle_run(struct ofbundle *bundle)
3006 {
3007     if (bundle->lacp) {
3008         lacp_run(bundle->lacp, send_pdu_cb);
3009     }
3010     if (bundle->bond) {
3011         struct ofport_dpif *port;
3012
3013         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
3014             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
3015         }
3016
3017         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
3018             bundle->ofproto->backer->need_revalidate = REV_BOND;
3019         }
3020
3021         if (bond_should_send_learning_packets(bundle->bond)) {
3022             bundle_send_learning_packets(bundle);
3023         }
3024     }
3025 }
3026
3027 static void
3028 bundle_wait(struct ofbundle *bundle)
3029 {
3030     if (bundle->lacp) {
3031         lacp_wait(bundle->lacp);
3032     }
3033     if (bundle->bond) {
3034         bond_wait(bundle->bond);
3035     }
3036 }
3037 \f
3038 /* Mirrors. */
3039
3040 static int
3041 mirror_set__(struct ofproto *ofproto_, void *aux,
3042              const struct ofproto_mirror_settings *s)
3043 {
3044     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3045     struct ofbundle **srcs, **dsts;
3046     int error;
3047     size_t i;
3048
3049     if (!s) {
3050         mirror_destroy(ofproto->mbridge, aux);
3051         return 0;
3052     }
3053
3054     srcs = xmalloc(s->n_srcs * sizeof *srcs);
3055     dsts = xmalloc(s->n_dsts * sizeof *dsts);
3056
3057     for (i = 0; i < s->n_srcs; i++) {
3058         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
3059     }
3060
3061     for (i = 0; i < s->n_dsts; i++) {
3062         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
3063     }
3064
3065     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
3066                        s->n_dsts, s->src_vlans,
3067                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
3068     free(srcs);
3069     free(dsts);
3070     return error;
3071 }
3072
3073 static int
3074 mirror_get_stats__(struct ofproto *ofproto, void *aux,
3075                    uint64_t *packets, uint64_t *bytes)
3076 {
3077     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
3078                             bytes);
3079 }
3080
3081 static int
3082 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3083 {
3084     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3085     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3086     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3087         mac_learning_flush(ofproto->ml);
3088     }
3089     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3090     return 0;
3091 }
3092
3093 static bool
3094 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3095 {
3096     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3097     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3098     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
3099 }
3100
3101 static void
3102 forward_bpdu_changed(struct ofproto *ofproto_)
3103 {
3104     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3105     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3106 }
3107
3108 static void
3109 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3110                      size_t max_entries)
3111 {
3112     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3113     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3114     mac_learning_set_idle_time(ofproto->ml, idle_time);
3115     mac_learning_set_max_entries(ofproto->ml, max_entries);
3116     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3117 }
3118
3119 /* Configures multicast snooping on 'ofport' using the settings
3120  * defined in 's'. */
3121 static int
3122 set_mcast_snooping(struct ofproto *ofproto_,
3123                    const struct ofproto_mcast_snooping_settings *s)
3124 {
3125     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3126
3127     /* Only revalidate flows if the configuration changed. */
3128     if (!s != !ofproto->ms) {
3129         ofproto->backer->need_revalidate = REV_RECONFIGURE;
3130     }
3131
3132     if (s) {
3133         if (!ofproto->ms) {
3134             ofproto->ms = mcast_snooping_create();
3135         }
3136
3137         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3138         mcast_snooping_set_idle_time(ofproto->ms, s->idle_time);
3139         mcast_snooping_set_max_entries(ofproto->ms, s->max_entries);
3140         if (mcast_snooping_set_flood_unreg(ofproto->ms, s->flood_unreg)) {
3141             ofproto->backer->need_revalidate = REV_RECONFIGURE;
3142         }
3143         ovs_rwlock_unlock(&ofproto->ms->rwlock);
3144     } else {
3145         mcast_snooping_unref(ofproto->ms);
3146         ofproto->ms = NULL;
3147     }
3148
3149     return 0;
3150 }
3151
3152 /* Configures multicast snooping port's flood settings on 'ofproto'. */
3153 static int
3154 set_mcast_snooping_port(struct ofproto *ofproto_, void *aux,
3155                         const struct ofproto_mcast_snooping_port_settings *s)
3156 {
3157     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3158     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3159
3160     if (ofproto->ms && s) {
3161         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3162         mcast_snooping_set_port_flood(ofproto->ms, bundle, s->flood);
3163         mcast_snooping_set_port_flood_reports(ofproto->ms, bundle,
3164                                               s->flood_reports);
3165         ovs_rwlock_unlock(&ofproto->ms->rwlock);
3166     }
3167     return 0;
3168 }
3169
3170 \f
3171 /* Ports. */
3172
3173 static struct ofport_dpif *
3174 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
3175 {
3176     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3177     return ofport ? ofport_dpif_cast(ofport) : NULL;
3178 }
3179
3180 static void
3181 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3182                             struct ofproto_port *ofproto_port,
3183                             struct dpif_port *dpif_port)
3184 {
3185     ofproto_port->name = dpif_port->name;
3186     ofproto_port->type = dpif_port->type;
3187     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3188 }
3189
3190 static void
3191 ofport_update_peer(struct ofport_dpif *ofport)
3192 {
3193     const struct ofproto_dpif *ofproto;
3194     struct dpif_backer *backer;
3195     char *peer_name;
3196
3197     if (!netdev_vport_is_patch(ofport->up.netdev)) {
3198         return;
3199     }
3200
3201     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
3202     backer->need_revalidate = REV_RECONFIGURE;
3203
3204     if (ofport->peer) {
3205         ofport->peer->peer = NULL;
3206         ofport->peer = NULL;
3207     }
3208
3209     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
3210     if (!peer_name) {
3211         return;
3212     }
3213
3214     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3215         struct ofport *peer_ofport;
3216         struct ofport_dpif *peer;
3217         char *peer_peer;
3218
3219         if (ofproto->backer != backer) {
3220             continue;
3221         }
3222
3223         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
3224         if (!peer_ofport) {
3225             continue;
3226         }
3227
3228         peer = ofport_dpif_cast(peer_ofport);
3229         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
3230         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
3231                                  peer_peer)) {
3232             ofport->peer = peer;
3233             ofport->peer->peer = ofport;
3234         }
3235         free(peer_peer);
3236
3237         break;
3238     }
3239     free(peer_name);
3240 }
3241
3242 static void
3243 port_run(struct ofport_dpif *ofport)
3244 {
3245     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3246     bool carrier_changed = carrier_seq != ofport->carrier_seq;
3247     bool enable = netdev_get_carrier(ofport->up.netdev);
3248     bool cfm_enable = false;
3249     bool bfd_enable = false;
3250
3251     ofport->carrier_seq = carrier_seq;
3252
3253     if (ofport->cfm) {
3254         int cfm_opup = cfm_get_opup(ofport->cfm);
3255
3256         cfm_enable = !cfm_get_fault(ofport->cfm);
3257
3258         if (cfm_opup >= 0) {
3259             cfm_enable = cfm_enable && cfm_opup;
3260         }
3261     }
3262
3263     if (ofport->bfd) {
3264         bfd_enable = bfd_forwarding(ofport->bfd);
3265     }
3266
3267     if (ofport->bfd || ofport->cfm) {
3268         enable = enable && (cfm_enable || bfd_enable);
3269     }
3270
3271     if (ofport->bundle) {
3272         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3273         if (carrier_changed) {
3274             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3275         }
3276     }
3277
3278     if (ofport->may_enable != enable) {
3279         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3280
3281         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3282
3283         if (ofport->rstp_port) {
3284             rstp_port_set_mac_operational(ofport->rstp_port, enable);
3285         }
3286     }
3287
3288     ofport->may_enable = enable;
3289 }
3290
3291 static int
3292 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3293                    struct ofproto_port *ofproto_port)
3294 {
3295     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3296     struct dpif_port dpif_port;
3297     int error;
3298
3299     if (sset_contains(&ofproto->ghost_ports, devname)) {
3300         const char *type = netdev_get_type_from_name(devname);
3301
3302         /* We may be called before ofproto->up.port_by_name is populated with
3303          * the appropriate ofport.  For this reason, we must get the name and
3304          * type from the netdev layer directly. */
3305         if (type) {
3306             const struct ofport *ofport;
3307
3308             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3309             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3310             ofproto_port->name = xstrdup(devname);
3311             ofproto_port->type = xstrdup(type);
3312             return 0;
3313         }
3314         return ENODEV;
3315     }
3316
3317     if (!sset_contains(&ofproto->ports, devname)) {
3318         return ENODEV;
3319     }
3320     error = dpif_port_query_by_name(ofproto->backer->dpif,
3321                                     devname, &dpif_port);
3322     if (!error) {
3323         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3324     }
3325     return error;
3326 }
3327
3328 static int
3329 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3330 {
3331     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3332     const char *devname = netdev_get_name(netdev);
3333     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3334     const char *dp_port_name;
3335
3336     if (netdev_vport_is_patch(netdev)) {
3337         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3338         return 0;
3339     }
3340
3341     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3342     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3343         odp_port_t port_no = ODPP_NONE;
3344         int error;
3345
3346         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3347         if (error) {
3348             return error;
3349         }
3350         if (netdev_get_tunnel_config(netdev)) {
3351             simap_put(&ofproto->backer->tnl_backers,
3352                       dp_port_name, odp_to_u32(port_no));
3353         }
3354     }
3355
3356     if (netdev_get_tunnel_config(netdev)) {
3357         sset_add(&ofproto->ghost_ports, devname);
3358     } else {
3359         sset_add(&ofproto->ports, devname);
3360     }
3361     return 0;
3362 }
3363
3364 static int
3365 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3366 {
3367     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3368     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3369     int error = 0;
3370
3371     if (!ofport) {
3372         return 0;
3373     }
3374
3375     sset_find_and_delete(&ofproto->ghost_ports,
3376                          netdev_get_name(ofport->up.netdev));
3377     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3378     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3379         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3380         if (!error) {
3381             /* The caller is going to close ofport->up.netdev.  If this is a
3382              * bonded port, then the bond is using that netdev, so remove it
3383              * from the bond.  The client will need to reconfigure everything
3384              * after deleting ports, so then the slave will get re-added. */
3385             bundle_remove(&ofport->up);
3386         }
3387     }
3388     return error;
3389 }
3390
3391 static int
3392 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3393 {
3394     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3395     int error;
3396
3397     error = netdev_get_stats(ofport->up.netdev, stats);
3398
3399     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3400         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3401
3402         ovs_mutex_lock(&ofproto->stats_mutex);
3403         /* ofproto->stats.tx_packets represents packets that we created
3404          * internally and sent to some port (e.g. packets sent with
3405          * ofproto_dpif_send_packet()).  Account for them as if they had
3406          * come from OFPP_LOCAL and got forwarded. */
3407
3408         if (stats->rx_packets != UINT64_MAX) {
3409             stats->rx_packets += ofproto->stats.tx_packets;
3410         }
3411
3412         if (stats->rx_bytes != UINT64_MAX) {
3413             stats->rx_bytes += ofproto->stats.tx_bytes;
3414         }
3415
3416         /* ofproto->stats.rx_packets represents packets that were received on
3417          * some port and we processed internally and dropped (e.g. STP).
3418          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3419
3420         if (stats->tx_packets != UINT64_MAX) {
3421             stats->tx_packets += ofproto->stats.rx_packets;
3422         }
3423
3424         if (stats->tx_bytes != UINT64_MAX) {
3425             stats->tx_bytes += ofproto->stats.rx_bytes;
3426         }
3427         ovs_mutex_unlock(&ofproto->stats_mutex);
3428     }
3429
3430     return error;
3431 }
3432
3433 static int
3434 port_get_lacp_stats(const struct ofport *ofport_, struct lacp_slave_stats *stats)
3435 {
3436     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3437     if (ofport->bundle && ofport->bundle->lacp) {
3438         if (lacp_get_slave_stats(ofport->bundle->lacp, ofport, stats)) {
3439             return 0;
3440         }
3441     }
3442     return -1;
3443 }
3444
3445 struct port_dump_state {
3446     uint32_t bucket;
3447     uint32_t offset;
3448     bool ghost;
3449
3450     struct ofproto_port port;
3451     bool has_port;
3452 };
3453
3454 static int
3455 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3456 {
3457     *statep = xzalloc(sizeof(struct port_dump_state));
3458     return 0;
3459 }
3460
3461 static int
3462 port_dump_next(const struct ofproto *ofproto_, void *state_,
3463                struct ofproto_port *port)
3464 {
3465     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3466     struct port_dump_state *state = state_;
3467     const struct sset *sset;
3468     struct sset_node *node;
3469
3470     if (state->has_port) {
3471         ofproto_port_destroy(&state->port);
3472         state->has_port = false;
3473     }
3474     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3475     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3476         int error;
3477
3478         error = port_query_by_name(ofproto_, node->name, &state->port);
3479         if (!error) {
3480             *port = state->port;
3481             state->has_port = true;
3482             return 0;
3483         } else if (error != ENODEV) {
3484             return error;
3485         }
3486     }
3487
3488     if (!state->ghost) {
3489         state->ghost = true;
3490         state->bucket = 0;
3491         state->offset = 0;
3492         return port_dump_next(ofproto_, state_, port);
3493     }
3494
3495     return EOF;
3496 }
3497
3498 static int
3499 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3500 {
3501     struct port_dump_state *state = state_;
3502
3503     if (state->has_port) {
3504         ofproto_port_destroy(&state->port);
3505     }
3506     free(state);
3507     return 0;
3508 }
3509
3510 static int
3511 port_poll(const struct ofproto *ofproto_, char **devnamep)
3512 {
3513     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3514
3515     if (ofproto->port_poll_errno) {
3516         int error = ofproto->port_poll_errno;
3517         ofproto->port_poll_errno = 0;
3518         return error;
3519     }
3520
3521     if (sset_is_empty(&ofproto->port_poll_set)) {
3522         return EAGAIN;
3523     }
3524
3525     *devnamep = sset_pop(&ofproto->port_poll_set);
3526     return 0;
3527 }
3528
3529 static void
3530 port_poll_wait(const struct ofproto *ofproto_)
3531 {
3532     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3533     dpif_port_poll_wait(ofproto->backer->dpif);
3534 }
3535
3536 static int
3537 port_is_lacp_current(const struct ofport *ofport_)
3538 {
3539     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3540     return (ofport->bundle && ofport->bundle->lacp
3541             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3542             : -1);
3543 }
3544 \f
3545 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3546  * then delete it entirely. */
3547 static void
3548 rule_expire(struct rule_dpif *rule)
3549     OVS_REQUIRES(ofproto_mutex)
3550 {
3551     uint16_t hard_timeout, idle_timeout;
3552     long long int now = time_msec();
3553     int reason = -1;
3554
3555     hard_timeout = rule->up.hard_timeout;
3556     idle_timeout = rule->up.idle_timeout;
3557
3558     /* Has 'rule' expired? */
3559     if (hard_timeout) {
3560         long long int modified;
3561
3562         ovs_mutex_lock(&rule->up.mutex);
3563         modified = rule->up.modified;
3564         ovs_mutex_unlock(&rule->up.mutex);
3565
3566         if (now > modified + hard_timeout * 1000) {
3567             reason = OFPRR_HARD_TIMEOUT;
3568         }
3569     }
3570
3571     if (reason < 0 && idle_timeout) {
3572         long long int used;
3573
3574         ovs_mutex_lock(&rule->stats_mutex);
3575         used = rule->stats.used;
3576         ovs_mutex_unlock(&rule->stats_mutex);
3577
3578         if (now > used + idle_timeout * 1000) {
3579             reason = OFPRR_IDLE_TIMEOUT;
3580         }
3581     }
3582
3583     if (reason >= 0) {
3584         COVERAGE_INC(ofproto_dpif_expired);
3585         ofproto_rule_expire(&rule->up, reason);
3586     }
3587 }
3588
3589 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3590  * 'flow' must reflect the data in 'packet'. */
3591 int
3592 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3593                              const struct flow *flow,
3594                              struct rule_dpif *rule,
3595                              const struct ofpact *ofpacts, size_t ofpacts_len,
3596                              struct ofpbuf *packet)
3597 {
3598     struct dpif_flow_stats stats;
3599     struct xlate_out xout;
3600     struct xlate_in xin;
3601     ofp_port_t in_port;
3602     struct dpif_execute execute;
3603     int error;
3604
3605     ovs_assert((rule != NULL) != (ofpacts != NULL));
3606
3607     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3608
3609     if (rule) {
3610         rule_dpif_credit_stats(rule, &stats);
3611     }
3612
3613     xlate_in_init(&xin, ofproto, flow, flow->in_port.ofp_port, rule,
3614                   stats.tcp_flags, packet);
3615     xin.ofpacts = ofpacts;
3616     xin.ofpacts_len = ofpacts_len;
3617     xin.resubmit_stats = &stats;
3618     xlate_actions(&xin, &xout);
3619
3620     execute.actions = ofpbuf_data(xout.odp_actions);
3621     execute.actions_len = ofpbuf_size(xout.odp_actions);
3622
3623     execute.packet = packet;
3624     execute.md = pkt_metadata_from_flow(flow);
3625     execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3626     execute.probe = false;
3627
3628     /* Fix up in_port. */
3629     in_port = flow->in_port.ofp_port;
3630     if (in_port == OFPP_NONE) {
3631         in_port = OFPP_LOCAL;
3632     }
3633     execute.md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3634
3635     error = dpif_execute(ofproto->backer->dpif, &execute);
3636
3637     xlate_out_uninit(&xout);
3638
3639     return error;
3640 }
3641
3642 void
3643 rule_dpif_credit_stats(struct rule_dpif *rule,
3644                        const struct dpif_flow_stats *stats)
3645 {
3646     ovs_mutex_lock(&rule->stats_mutex);
3647     rule->stats.n_packets += stats->n_packets;
3648     rule->stats.n_bytes += stats->n_bytes;
3649     rule->stats.used = MAX(rule->stats.used, stats->used);
3650     ovs_mutex_unlock(&rule->stats_mutex);
3651 }
3652
3653 ovs_be64
3654 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3655     OVS_REQUIRES(rule->up.mutex)
3656 {
3657     return rule->up.flow_cookie;
3658 }
3659
3660 void
3661 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3662                      uint16_t hard_timeout)
3663 {
3664     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3665 }
3666
3667 /* Returns 'rule''s actions.  The returned actions are RCU-protected, and can
3668  * be read until the calling thread quiesces. */
3669 const struct rule_actions *
3670 rule_dpif_get_actions(const struct rule_dpif *rule)
3671 {
3672     return rule_get_actions(&rule->up);
3673 }
3674
3675 /* Sets 'rule''s recirculation id. */
3676 static void
3677 rule_dpif_set_recirc_id(struct rule_dpif *rule, uint32_t id)
3678     OVS_REQUIRES(rule->up.mutex)
3679 {
3680     ovs_assert(!rule->recirc_id);
3681     rule->recirc_id = id;
3682 }
3683
3684 /* Returns 'rule''s recirculation id. */
3685 uint32_t
3686 rule_dpif_get_recirc_id(struct rule_dpif *rule)
3687     OVS_REQUIRES(rule->up.mutex)
3688 {
3689     if (!rule->recirc_id) {
3690         struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3691
3692         rule_dpif_set_recirc_id(rule, ofproto_dpif_alloc_recirc_id(ofproto));
3693     }
3694     return rule->recirc_id;
3695 }
3696
3697 /* Sets 'rule''s recirculation id. */
3698 void
3699 rule_set_recirc_id(struct rule *rule_, uint32_t id)
3700 {
3701     struct rule_dpif *rule = rule_dpif_cast(rule_);
3702
3703     ovs_mutex_lock(&rule->up.mutex);
3704     rule_dpif_set_recirc_id(rule, id);
3705     ovs_mutex_unlock(&rule->up.mutex);
3706 }
3707
3708 /* Lookup 'flow' in table 0 of 'ofproto''s classifier.
3709  * If 'wc' is non-null, sets the fields that were relevant as part of
3710  * the lookup. Returns the table id where a match or miss occurred via
3711  * 'table_id'.  This will be zero unless there was a miss and
3712  * OFPTC11_TABLE_MISS_CONTINUE is in effect for the sequence of tables
3713  * where misses occur, or TBL_INTERNAL if the rule has a non-zero
3714  * recirculation ID, and a match was found in the internal table, or if
3715  * there was no match and one of the special rules (drop_frags_rule,
3716  * miss_rule, or no_packet_in_rule) was returned.
3717  *
3718  * The return value is the found rule, which is valid at least until the next
3719  * RCU quiescent period.  If the rule needs to stay around longer,
3720  * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3721  * on it before this returns. */
3722 struct rule_dpif *
3723 rule_dpif_lookup(struct ofproto_dpif *ofproto, struct flow *flow,
3724                  struct flow_wildcards *wc, bool take_ref,
3725                  const struct dpif_flow_stats *stats, uint8_t *table_id)
3726 {
3727     *table_id = 0;
3728
3729     if (ofproto_dpif_get_enable_recirc(ofproto)) {
3730         /* Always exactly match recirc_id since datapath supports
3731          * recirculation.  */
3732         if (wc) {
3733             wc->masks.recirc_id = UINT32_MAX;
3734         }
3735         *table_id = rule_dpif_lookup_get_init_table_id(flow);
3736     }
3737
3738     return rule_dpif_lookup_from_table(ofproto, flow, wc, take_ref, stats,
3739                                        table_id, flow->in_port.ofp_port, true,
3740                                        true);
3741 }
3742
3743 /* The returned rule (if any) is valid at least until the next RCU quiescent
3744  * period.  If the rule needs to stay around longer, a non-zero 'take_ref'
3745  * must be passed in to cause a reference to be taken on it.
3746  *
3747  * 'flow' is non-const to allow for temporary modifications during the lookup.
3748  * Any changes are restored before returning. */
3749 static struct rule_dpif *
3750 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, uint8_t table_id,
3751                           struct flow *flow, struct flow_wildcards *wc,
3752                           bool take_ref)
3753 {
3754     struct classifier *cls = &ofproto->up.tables[table_id].cls;
3755     const struct cls_rule *cls_rule;
3756     struct rule_dpif *rule;
3757
3758     do {
3759         cls_rule = classifier_lookup(cls, flow, wc);
3760
3761         rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3762
3763         /* Try again if the rule was released before we get the reference. */
3764     } while (rule && take_ref && !rule_dpif_try_ref(rule));
3765
3766     return rule;
3767 }
3768
3769 /* Look up 'flow' in 'ofproto''s classifier starting from table '*table_id'.
3770  * Returns the rule that was found, which may be one of the special rules
3771  * according to packet miss hadling.  If 'may_packet_in' is false, returning of
3772  * the miss_rule (which issues packet ins for the controller) is avoided.
3773  * Updates 'wc', if nonnull, to reflect the fields that were used during the
3774  * lookup.
3775  *
3776  * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3777  * if none is found then the table miss configuration for that table is
3778  * honored, which can result in additional lookups in other OpenFlow tables.
3779  * In this case the function updates '*table_id' to reflect the final OpenFlow
3780  * table that was searched.
3781  *
3782  * If 'honor_table_miss' is false, then only one table lookup occurs, in
3783  * '*table_id'.
3784  *
3785  * The rule is returned in '*rule', which is valid at least until the next
3786  * RCU quiescent period.  If the '*rule' needs to stay around longer,
3787  * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3788  * on it before this returns.
3789  *
3790  * 'in_port' allows the lookup to take place as if the in port had the value
3791  * 'in_port'.  This is needed for resubmit action support.
3792  *
3793  * 'flow' is non-const to allow for temporary modifications during the lookup.
3794  * Any changes are restored before returning. */
3795 struct rule_dpif *
3796 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto, struct flow *flow,
3797                             struct flow_wildcards *wc, bool take_ref,
3798                             const struct dpif_flow_stats *stats,
3799                             uint8_t *table_id, ofp_port_t in_port,
3800                             bool may_packet_in, bool honor_table_miss)
3801 {
3802     ovs_be16 old_tp_src = flow->tp_src, old_tp_dst = flow->tp_dst;
3803     ofp_port_t old_in_port = flow->in_port.ofp_port;
3804     enum ofputil_table_miss miss_config;
3805     struct rule_dpif *rule;
3806     uint8_t next_id;
3807
3808     /* We always unwildcard nw_frag (for IP), so they
3809      * need not be unwildcarded here. */
3810     if (flow->nw_frag & FLOW_NW_FRAG_ANY
3811         && ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3812         if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3813             /* We must pretend that transport ports are unavailable. */
3814             flow->tp_src = htons(0);
3815             flow->tp_dst = htons(0);
3816         } else {
3817             /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM).
3818              * Use the drop_frags_rule (which cannot disappear). */
3819             rule = ofproto->drop_frags_rule;
3820             if (take_ref) {
3821                 rule_dpif_ref(rule);
3822             }
3823             if (stats) {
3824                 struct oftable *tbl = &ofproto->up.tables[*table_id];
3825                 unsigned long orig;
3826
3827                 atomic_add_relaxed(&tbl->n_matched, stats->n_packets, &orig);
3828             }
3829             return rule;
3830         }
3831     }
3832
3833     /* Look up a flow with 'in_port' as the input port.  Then restore the
3834      * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
3835      * have surprising behavior). */
3836     flow->in_port.ofp_port = in_port;
3837
3838     /* Our current implementation depends on n_tables == N_TABLES, and
3839      * TBL_INTERNAL being the last table. */
3840     BUILD_ASSERT_DECL(N_TABLES == TBL_INTERNAL + 1);
3841
3842     miss_config = OFPUTIL_TABLE_MISS_CONTINUE;
3843
3844     for (next_id = *table_id;
3845          next_id < ofproto->up.n_tables;
3846          next_id++, next_id += (next_id == TBL_INTERNAL))
3847     {
3848         *table_id = next_id;
3849         rule = rule_dpif_lookup_in_table(ofproto, next_id, flow, wc, take_ref);
3850         if (stats) {
3851             struct oftable *tbl = &ofproto->up.tables[next_id];
3852             unsigned long orig;
3853
3854             atomic_add_relaxed(rule ? &tbl->n_matched : &tbl->n_missed,
3855                                stats->n_packets, &orig);
3856         }
3857         if (rule) {
3858             goto out;   /* Match. */
3859         }
3860         if (honor_table_miss) {
3861             miss_config = ofproto_table_get_miss_config(&ofproto->up,
3862                                                         *table_id);
3863             if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE) {
3864                 continue;
3865             }
3866         }
3867         break;
3868     }
3869     /* Miss. */
3870     rule = ofproto->no_packet_in_rule;
3871     if (may_packet_in) {
3872         if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE
3873             || miss_config == OFPUTIL_TABLE_MISS_CONTROLLER) {
3874             struct ofport_dpif *port;
3875
3876             port = get_ofp_port(ofproto, old_in_port);
3877             if (!port) {
3878                 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3879                              old_in_port);
3880             } else if (!(port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN)) {
3881                 rule = ofproto->miss_rule;
3882             }
3883         } else if (miss_config == OFPUTIL_TABLE_MISS_DEFAULT &&
3884                    connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3885             rule = ofproto->miss_rule;
3886         }
3887     }
3888     if (take_ref) {
3889         rule_dpif_ref(rule);
3890     }
3891 out:
3892     /* Restore port numbers, as they may have been modified above. */
3893     flow->tp_src = old_tp_src;
3894     flow->tp_dst = old_tp_dst;
3895     /* Restore the old in port. */
3896     flow->in_port.ofp_port = old_in_port;
3897
3898     return rule;
3899 }
3900
3901 static void
3902 complete_operation(struct rule_dpif *rule)
3903     OVS_REQUIRES(ofproto_mutex)
3904 {
3905     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3906
3907     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3908 }
3909
3910 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3911 {
3912     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3913 }
3914
3915 static struct rule *
3916 rule_alloc(void)
3917 {
3918     struct rule_dpif *rule = xmalloc(sizeof *rule);
3919     return &rule->up;
3920 }
3921
3922 static void
3923 rule_dealloc(struct rule *rule_)
3924 {
3925     struct rule_dpif *rule = rule_dpif_cast(rule_);
3926     free(rule);
3927 }
3928
3929 static enum ofperr
3930 rule_construct(struct rule *rule_)
3931     OVS_NO_THREAD_SAFETY_ANALYSIS
3932 {
3933     struct rule_dpif *rule = rule_dpif_cast(rule_);
3934     ovs_mutex_init_adaptive(&rule->stats_mutex);
3935     rule->stats.n_packets = 0;
3936     rule->stats.n_bytes = 0;
3937     rule->stats.used = rule->up.modified;
3938     rule->recirc_id = 0;
3939
3940     return 0;
3941 }
3942
3943 static enum ofperr
3944 rule_insert(struct rule *rule_)
3945     OVS_REQUIRES(ofproto_mutex)
3946 {
3947     struct rule_dpif *rule = rule_dpif_cast(rule_);
3948     complete_operation(rule);
3949     return 0;
3950 }
3951
3952 static void
3953 rule_delete(struct rule *rule_)
3954     OVS_REQUIRES(ofproto_mutex)
3955 {
3956     struct rule_dpif *rule = rule_dpif_cast(rule_);
3957     complete_operation(rule);
3958 }
3959
3960 static void
3961 rule_destruct(struct rule *rule_)
3962 {
3963     struct rule_dpif *rule = rule_dpif_cast(rule_);
3964
3965     ovs_mutex_destroy(&rule->stats_mutex);
3966     if (rule->recirc_id) {
3967         struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3968
3969         ofproto_dpif_free_recirc_id(ofproto, rule->recirc_id);
3970     }
3971 }
3972
3973 static void
3974 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
3975                long long int *used)
3976 {
3977     struct rule_dpif *rule = rule_dpif_cast(rule_);
3978
3979     ovs_mutex_lock(&rule->stats_mutex);
3980     *packets = rule->stats.n_packets;
3981     *bytes = rule->stats.n_bytes;
3982     *used = rule->stats.used;
3983     ovs_mutex_unlock(&rule->stats_mutex);
3984 }
3985
3986 static void
3987 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
3988                   struct ofpbuf *packet)
3989 {
3990     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3991
3992     ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
3993 }
3994
3995 static enum ofperr
3996 rule_execute(struct rule *rule, const struct flow *flow,
3997              struct ofpbuf *packet)
3998 {
3999     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4000     ofpbuf_delete(packet);
4001     return 0;
4002 }
4003
4004 static void
4005 rule_modify_actions(struct rule *rule_, bool reset_counters)
4006     OVS_REQUIRES(ofproto_mutex)
4007 {
4008     struct rule_dpif *rule = rule_dpif_cast(rule_);
4009
4010     if (reset_counters) {
4011         ovs_mutex_lock(&rule->stats_mutex);
4012         rule->stats.n_packets = 0;
4013         rule->stats.n_bytes = 0;
4014         ovs_mutex_unlock(&rule->stats_mutex);
4015     }
4016
4017     complete_operation(rule);
4018 }
4019
4020 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4021 {
4022     return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4023 }
4024
4025 static struct ofgroup *
4026 group_alloc(void)
4027 {
4028     struct group_dpif *group = xzalloc(sizeof *group);
4029     return &group->up;
4030 }
4031
4032 static void
4033 group_dealloc(struct ofgroup *group_)
4034 {
4035     struct group_dpif *group = group_dpif_cast(group_);
4036     free(group);
4037 }
4038
4039 static void
4040 group_construct_stats(struct group_dpif *group)
4041     OVS_REQUIRES(group->stats_mutex)
4042 {
4043     struct ofputil_bucket *bucket;
4044     const struct ovs_list *buckets;
4045
4046     group->packet_count = 0;
4047     group->byte_count = 0;
4048
4049     group_dpif_get_buckets(group, &buckets);
4050     LIST_FOR_EACH (bucket, list_node, buckets) {
4051         bucket->stats.packet_count = 0;
4052         bucket->stats.byte_count = 0;
4053     }
4054 }
4055
4056 void
4057 group_dpif_credit_stats(struct group_dpif *group,
4058                         struct ofputil_bucket *bucket,
4059                         const struct dpif_flow_stats *stats)
4060 {
4061     ovs_mutex_lock(&group->stats_mutex);
4062     group->packet_count += stats->n_packets;
4063     group->byte_count += stats->n_bytes;
4064     if (bucket) {
4065         bucket->stats.packet_count += stats->n_packets;
4066         bucket->stats.byte_count += stats->n_bytes;
4067     } else { /* Credit to all buckets */
4068         const struct ovs_list *buckets;
4069
4070         group_dpif_get_buckets(group, &buckets);
4071         LIST_FOR_EACH (bucket, list_node, buckets) {
4072             bucket->stats.packet_count += stats->n_packets;
4073             bucket->stats.byte_count += stats->n_bytes;
4074         }
4075     }
4076     ovs_mutex_unlock(&group->stats_mutex);
4077 }
4078
4079 static enum ofperr
4080 group_construct(struct ofgroup *group_)
4081 {
4082     struct group_dpif *group = group_dpif_cast(group_);
4083     const struct ofputil_bucket *bucket;
4084
4085     /* Prevent group chaining because our locking structure makes it hard to
4086      * implement deadlock-free.  (See xlate_group_resource_check().) */
4087     LIST_FOR_EACH (bucket, list_node, &group->up.buckets) {
4088         const struct ofpact *a;
4089
4090         OFPACT_FOR_EACH (a, bucket->ofpacts, bucket->ofpacts_len) {
4091             if (a->type == OFPACT_GROUP) {
4092                 return OFPERR_OFPGMFC_CHAINING_UNSUPPORTED;
4093             }
4094         }
4095     }
4096
4097     ovs_mutex_init_adaptive(&group->stats_mutex);
4098     ovs_mutex_lock(&group->stats_mutex);
4099     group_construct_stats(group);
4100     ovs_mutex_unlock(&group->stats_mutex);
4101     return 0;
4102 }
4103
4104 static void
4105 group_destruct(struct ofgroup *group_)
4106 {
4107     struct group_dpif *group = group_dpif_cast(group_);
4108     ovs_mutex_destroy(&group->stats_mutex);
4109 }
4110
4111 static enum ofperr
4112 group_modify(struct ofgroup *group_)
4113 {
4114     struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
4115
4116     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4117
4118     return 0;
4119 }
4120
4121 static enum ofperr
4122 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4123 {
4124     struct group_dpif *group = group_dpif_cast(group_);
4125     struct ofputil_bucket *bucket;
4126     const struct ovs_list *buckets;
4127     struct bucket_counter *bucket_stats;
4128
4129     ovs_mutex_lock(&group->stats_mutex);
4130     ogs->packet_count = group->packet_count;
4131     ogs->byte_count = group->byte_count;
4132
4133     group_dpif_get_buckets(group, &buckets);
4134     bucket_stats = ogs->bucket_stats;
4135     LIST_FOR_EACH (bucket, list_node, buckets) {
4136         bucket_stats->packet_count = bucket->stats.packet_count;
4137         bucket_stats->byte_count = bucket->stats.byte_count;
4138         bucket_stats++;
4139     }
4140     ovs_mutex_unlock(&group->stats_mutex);
4141
4142     return 0;
4143 }
4144
4145 /* If the group exists, this function increments the groups's reference count.
4146  *
4147  * Make sure to call group_dpif_unref() after no longer needing to maintain
4148  * a reference to the group. */
4149 bool
4150 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4151                   struct group_dpif **group)
4152 {
4153     struct ofgroup *ofgroup;
4154     bool found;
4155
4156     found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4157     *group = found ?  group_dpif_cast(ofgroup) : NULL;
4158
4159     return found;
4160 }
4161
4162 void
4163 group_dpif_get_buckets(const struct group_dpif *group,
4164                        const struct ovs_list **buckets)
4165 {
4166     *buckets = &group->up.buckets;
4167 }
4168
4169 enum ofp11_group_type
4170 group_dpif_get_type(const struct group_dpif *group)
4171 {
4172     return group->up.type;
4173 }
4174 \f
4175 /* Sends 'packet' out 'ofport'.
4176  * May modify 'packet'.
4177  * Returns 0 if successful, otherwise a positive errno value. */
4178 int
4179 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4180 {
4181     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4182     int error;
4183
4184     error = xlate_send_packet(ofport, packet);
4185
4186     ovs_mutex_lock(&ofproto->stats_mutex);
4187     ofproto->stats.tx_packets++;
4188     ofproto->stats.tx_bytes += ofpbuf_size(packet);
4189     ovs_mutex_unlock(&ofproto->stats_mutex);
4190     return error;
4191 }
4192 \f
4193 /* Return the version string of the datapath that backs up
4194  * this 'ofproto'.
4195  */
4196 static const char *
4197 get_datapath_version(const struct ofproto *ofproto_)
4198 {
4199     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4200
4201     return ofproto->backer->dp_version_string;
4202 }
4203
4204 static bool
4205 set_frag_handling(struct ofproto *ofproto_,
4206                   enum ofp_config_flags frag_handling)
4207 {
4208     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4209     if (frag_handling != OFPC_FRAG_REASM) {
4210         ofproto->backer->need_revalidate = REV_RECONFIGURE;
4211         return true;
4212     } else {
4213         return false;
4214     }
4215 }
4216
4217 static enum ofperr
4218 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
4219            const struct flow *flow,
4220            const struct ofpact *ofpacts, size_t ofpacts_len)
4221 {
4222     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4223
4224     ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
4225                                  ofpacts_len, packet);
4226     return 0;
4227 }
4228 \f
4229 /* NetFlow. */
4230
4231 static int
4232 set_netflow(struct ofproto *ofproto_,
4233             const struct netflow_options *netflow_options)
4234 {
4235     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4236
4237     if (netflow_options) {
4238         if (!ofproto->netflow) {
4239             ofproto->netflow = netflow_create();
4240             ofproto->backer->need_revalidate = REV_RECONFIGURE;
4241         }
4242         return netflow_set_options(ofproto->netflow, netflow_options);
4243     } else if (ofproto->netflow) {
4244         ofproto->backer->need_revalidate = REV_RECONFIGURE;
4245         netflow_unref(ofproto->netflow);
4246         ofproto->netflow = NULL;
4247     }
4248
4249     return 0;
4250 }
4251
4252 static void
4253 get_netflow_ids(const struct ofproto *ofproto_,
4254                 uint8_t *engine_type, uint8_t *engine_id)
4255 {
4256     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4257
4258     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
4259 }
4260 \f
4261 static struct ofproto_dpif *
4262 ofproto_dpif_lookup(const char *name)
4263 {
4264     struct ofproto_dpif *ofproto;
4265
4266     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
4267                              hash_string(name, 0), &all_ofproto_dpifs) {
4268         if (!strcmp(ofproto->up.name, name)) {
4269             return ofproto;
4270         }
4271     }
4272     return NULL;
4273 }
4274
4275 static void
4276 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
4277                           const char *argv[], void *aux OVS_UNUSED)
4278 {
4279     struct ofproto_dpif *ofproto;
4280
4281     if (argc > 1) {
4282         ofproto = ofproto_dpif_lookup(argv[1]);
4283         if (!ofproto) {
4284             unixctl_command_reply_error(conn, "no such bridge");
4285             return;
4286         }
4287         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4288         mac_learning_flush(ofproto->ml);
4289         ovs_rwlock_unlock(&ofproto->ml->rwlock);
4290     } else {
4291         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4292             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4293             mac_learning_flush(ofproto->ml);
4294             ovs_rwlock_unlock(&ofproto->ml->rwlock);
4295         }
4296     }
4297
4298     unixctl_command_reply(conn, "table successfully flushed");
4299 }
4300
4301 static void
4302 ofproto_unixctl_mcast_snooping_flush(struct unixctl_conn *conn, int argc,
4303                                      const char *argv[], void *aux OVS_UNUSED)
4304 {
4305     struct ofproto_dpif *ofproto;
4306
4307     if (argc > 1) {
4308         ofproto = ofproto_dpif_lookup(argv[1]);
4309         if (!ofproto) {
4310             unixctl_command_reply_error(conn, "no such bridge");
4311             return;
4312         }
4313
4314         if (!mcast_snooping_enabled(ofproto->ms)) {
4315             unixctl_command_reply_error(conn, "multicast snooping is disabled");
4316             return;
4317         }
4318         mcast_snooping_mdb_flush(ofproto->ms);
4319     } else {
4320         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4321             if (!mcast_snooping_enabled(ofproto->ms)) {
4322                 continue;
4323             }
4324             mcast_snooping_mdb_flush(ofproto->ms);
4325         }
4326     }
4327
4328     unixctl_command_reply(conn, "table successfully flushed");
4329 }
4330
4331 static struct ofport_dpif *
4332 ofbundle_get_a_port(const struct ofbundle *bundle)
4333 {
4334     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
4335                         bundle_node);
4336 }
4337
4338 static void
4339 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4340                          const char *argv[], void *aux OVS_UNUSED)
4341 {
4342     struct ds ds = DS_EMPTY_INITIALIZER;
4343     const struct ofproto_dpif *ofproto;
4344     const struct mac_entry *e;
4345
4346     ofproto = ofproto_dpif_lookup(argv[1]);
4347     if (!ofproto) {
4348         unixctl_command_reply_error(conn, "no such bridge");
4349         return;
4350     }
4351
4352     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
4353     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
4354     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
4355         struct ofbundle *bundle = mac_entry_get_port(ofproto->ml, e);
4356         char name[OFP_MAX_PORT_NAME_LEN];
4357
4358         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4359                                name, sizeof name);
4360         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
4361                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
4362                       mac_entry_age(ofproto->ml, e));
4363     }
4364     ovs_rwlock_unlock(&ofproto->ml->rwlock);
4365     unixctl_command_reply(conn, ds_cstr(&ds));
4366     ds_destroy(&ds);
4367 }
4368
4369 static void
4370 ofproto_unixctl_mcast_snooping_show(struct unixctl_conn *conn,
4371                                     int argc OVS_UNUSED,
4372                                     const char *argv[],
4373                                     void *aux OVS_UNUSED)
4374 {
4375     struct ds ds = DS_EMPTY_INITIALIZER;
4376     const struct ofproto_dpif *ofproto;
4377     const struct ofbundle *bundle;
4378     const struct mcast_group *grp;
4379     struct mcast_group_bundle *b;
4380     struct mcast_mrouter_bundle *mrouter;
4381
4382     ofproto = ofproto_dpif_lookup(argv[1]);
4383     if (!ofproto) {
4384         unixctl_command_reply_error(conn, "no such bridge");
4385         return;
4386     }
4387
4388     if (!mcast_snooping_enabled(ofproto->ms)) {
4389         unixctl_command_reply_error(conn, "multicast snooping is disabled");
4390         return;
4391     }
4392
4393     ds_put_cstr(&ds, " port  VLAN  GROUP                Age\n");
4394     ovs_rwlock_rdlock(&ofproto->ms->rwlock);
4395     LIST_FOR_EACH (grp, group_node, &ofproto->ms->group_lru) {
4396         LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
4397             char name[OFP_MAX_PORT_NAME_LEN];
4398
4399             bundle = b->port;
4400             ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4401                                    name, sizeof name);
4402             ds_put_format(&ds, "%5s  %4d  "IP_FMT"         %3d\n",
4403                           name, grp->vlan, IP_ARGS(grp->ip4),
4404                           mcast_bundle_age(ofproto->ms, b));
4405         }
4406     }
4407
4408     /* ports connected to multicast routers */
4409     LIST_FOR_EACH(mrouter, mrouter_node, &ofproto->ms->mrouter_lru) {
4410         char name[OFP_MAX_PORT_NAME_LEN];
4411
4412         bundle = mrouter->port;
4413         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4414                                name, sizeof name);
4415             ds_put_format(&ds, "%5s  %4d  querier             %3d\n",
4416                       name, mrouter->vlan,
4417                       mcast_mrouter_age(ofproto->ms, mrouter));
4418     }
4419     ovs_rwlock_unlock(&ofproto->ms->rwlock);
4420     unixctl_command_reply(conn, ds_cstr(&ds));
4421     ds_destroy(&ds);
4422 }
4423
4424 struct trace_ctx {
4425     struct xlate_out xout;
4426     struct xlate_in xin;
4427     const struct flow *key;
4428     struct flow flow;
4429     struct flow_wildcards wc;
4430     struct ds *result;
4431 };
4432
4433 static void
4434 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
4435 {
4436     const struct rule_actions *actions;
4437     ovs_be64 cookie;
4438
4439     ds_put_char_multiple(result, '\t', level);
4440     if (!rule) {
4441         ds_put_cstr(result, "No match\n");
4442         return;
4443     }
4444
4445     ovs_mutex_lock(&rule->up.mutex);
4446     cookie = rule->up.flow_cookie;
4447     ovs_mutex_unlock(&rule->up.mutex);
4448
4449     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
4450                   rule ? rule->up.table_id : 0, ntohll(cookie));
4451     cls_rule_format(&rule->up.cr, result);
4452     ds_put_char(result, '\n');
4453
4454     actions = rule_dpif_get_actions(rule);
4455
4456     ds_put_char_multiple(result, '\t', level);
4457     ds_put_cstr(result, "OpenFlow actions=");
4458     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
4459     ds_put_char(result, '\n');
4460 }
4461
4462 static void
4463 trace_format_flow(struct ds *result, int level, const char *title,
4464                   struct trace_ctx *trace)
4465 {
4466     ds_put_char_multiple(result, '\t', level);
4467     ds_put_format(result, "%s: ", title);
4468     /* Do not report unchanged flows for resubmits. */
4469     if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
4470         || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
4471         ds_put_cstr(result, "unchanged");
4472     } else {
4473         flow_format(result, &trace->xin.flow);
4474         trace->flow = trace->xin.flow;
4475     }
4476     ds_put_char(result, '\n');
4477 }
4478
4479 static void
4480 trace_format_regs(struct ds *result, int level, const char *title,
4481                   struct trace_ctx *trace)
4482 {
4483     size_t i;
4484
4485     ds_put_char_multiple(result, '\t', level);
4486     ds_put_format(result, "%s:", title);
4487     for (i = 0; i < FLOW_N_REGS; i++) {
4488         ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
4489     }
4490     ds_put_char(result, '\n');
4491 }
4492
4493 static void
4494 trace_format_odp(struct ds *result, int level, const char *title,
4495                  struct trace_ctx *trace)
4496 {
4497     struct ofpbuf *odp_actions = trace->xout.odp_actions;
4498
4499     ds_put_char_multiple(result, '\t', level);
4500     ds_put_format(result, "%s: ", title);
4501     format_odp_actions(result, ofpbuf_data(odp_actions),
4502                                ofpbuf_size(odp_actions));
4503     ds_put_char(result, '\n');
4504 }
4505
4506 static void
4507 trace_format_megaflow(struct ds *result, int level, const char *title,
4508                       struct trace_ctx *trace)
4509 {
4510     struct match match;
4511
4512     ds_put_char_multiple(result, '\t', level);
4513     ds_put_format(result, "%s: ", title);
4514     flow_wildcards_or(&trace->wc, &trace->xout.wc, &trace->wc);
4515     match_init(&match, trace->key, &trace->wc);
4516     match_format(&match, result, OFP_DEFAULT_PRIORITY);
4517     ds_put_char(result, '\n');
4518 }
4519
4520 static void trace_report(struct xlate_in *xin, const char *s, int recurse);
4521
4522 static void
4523 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
4524 {
4525     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4526     struct ds *result = trace->result;
4527
4528     if (!recurse) {
4529         if (rule == xin->ofproto->miss_rule) {
4530             trace_report(xin, "No match, flow generates \"packet in\"s.",
4531                          recurse);
4532         } else if (rule == xin->ofproto->no_packet_in_rule) {
4533             trace_report(xin, "No match, packets dropped because "
4534                          "OFPPC_NO_PACKET_IN is set on in_port.", recurse);
4535         } else if (rule == xin->ofproto->drop_frags_rule) {
4536             trace_report(xin, "Packets dropped because they are IP "
4537                          "fragments and the fragment handling mode is "
4538                          "\"drop\".", recurse);
4539         }
4540     }
4541
4542     ds_put_char(result, '\n');
4543     if (recurse) {
4544         trace_format_flow(result, recurse, "Resubmitted flow", trace);
4545         trace_format_regs(result, recurse, "Resubmitted regs", trace);
4546         trace_format_odp(result,  recurse, "Resubmitted  odp", trace);
4547         trace_format_megaflow(result, recurse, "Resubmitted megaflow", trace);
4548     }
4549     trace_format_rule(result, recurse, rule);
4550 }
4551
4552 static void
4553 trace_report(struct xlate_in *xin, const char *s, int recurse)
4554 {
4555     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4556     struct ds *result = trace->result;
4557
4558     ds_put_char_multiple(result, '\t', recurse);
4559     ds_put_cstr(result, s);
4560     ds_put_char(result, '\n');
4561 }
4562
4563 /* Parses the 'argc' elements of 'argv', ignoring argv[0].  The following
4564  * forms are supported:
4565  *
4566  *     - [dpname] odp_flow [-generate | packet]
4567  *     - bridge br_flow [-generate | packet]
4568  *
4569  * On success, initializes '*ofprotop' and 'flow' and returns NULL.  On failure
4570  * returns a nonnull malloced error message. */
4571 static char * OVS_WARN_UNUSED_RESULT
4572 parse_flow_and_packet(int argc, const char *argv[],
4573                       struct ofproto_dpif **ofprotop, struct flow *flow,
4574                       struct ofpbuf **packetp)
4575 {
4576     const struct dpif_backer *backer = NULL;
4577     const char *error = NULL;
4578     char *m_err = NULL;
4579     struct simap port_names = SIMAP_INITIALIZER(&port_names);
4580     struct ofpbuf *packet;
4581     struct ofpbuf odp_key;
4582     struct ofpbuf odp_mask;
4583
4584     ofpbuf_init(&odp_key, 0);
4585     ofpbuf_init(&odp_mask, 0);
4586
4587     /* Handle "-generate" or a hex string as the last argument. */
4588     if (!strcmp(argv[argc - 1], "-generate")) {
4589         packet = ofpbuf_new(0);
4590         argc--;
4591     } else {
4592         error = eth_from_hex(argv[argc - 1], &packet);
4593         if (!error) {
4594             argc--;
4595         } else if (argc == 4) {
4596             /* The 3-argument form must end in "-generate' or a hex string. */
4597             goto exit;
4598         }
4599         error = NULL;
4600     }
4601
4602     /* odp_flow can have its in_port specified as a name instead of port no.
4603      * We do not yet know whether a given flow is a odp_flow or a br_flow.
4604      * But, to know whether a flow is odp_flow through odp_flow_from_string(),
4605      * we need to create a simap of name to port no. */
4606     if (argc == 3) {
4607         const char *dp_type;
4608         if (!strncmp(argv[1], "ovs-", 4)) {
4609             dp_type = argv[1] + 4;
4610         } else {
4611             dp_type = argv[1];
4612         }
4613         backer = shash_find_data(&all_dpif_backers, dp_type);
4614     } else if (argc == 2) {
4615         struct shash_node *node;
4616         if (shash_count(&all_dpif_backers) == 1) {
4617             node = shash_first(&all_dpif_backers);
4618             backer = node->data;
4619         }
4620     } else {
4621         error = "Syntax error";
4622         goto exit;
4623     }
4624     if (backer && backer->dpif) {
4625         struct dpif_port dpif_port;
4626         struct dpif_port_dump port_dump;
4627         DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4628             simap_put(&port_names, dpif_port.name,
4629                       odp_to_u32(dpif_port.port_no));
4630         }
4631     }
4632
4633     /* Parse the flow and determine whether a datapath or
4634      * bridge is specified. If function odp_flow_key_from_string()
4635      * returns 0, the flow is a odp_flow. If function
4636      * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4637     if (!odp_flow_from_string(argv[argc - 1], &port_names,
4638                               &odp_key, &odp_mask)) {
4639         if (!backer) {
4640             error = "Cannot find the datapath";
4641             goto exit;
4642         }
4643
4644         if (odp_flow_key_to_flow(ofpbuf_data(&odp_key), ofpbuf_size(&odp_key),
4645                                  flow) == ODP_FIT_ERROR) {
4646             error = "Failed to parse flow key";
4647             goto exit;
4648         }
4649
4650         *ofprotop = xlate_lookup_ofproto(backer, flow,
4651                                          &flow->in_port.ofp_port);
4652         if (*ofprotop == NULL) {
4653             error = "Invalid datapath flow";
4654             goto exit;
4655         }
4656
4657         vsp_adjust_flow(*ofprotop, flow, NULL);
4658
4659     } else {
4660         char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4661
4662         if (err) {
4663             m_err = xasprintf("Bad flow syntax: %s", err);
4664             free(err);
4665             goto exit;
4666         } else {
4667             if (argc != 3) {
4668                 error = "Must specify bridge name";
4669                 goto exit;
4670             }
4671
4672             *ofprotop = ofproto_dpif_lookup(argv[1]);
4673             if (!*ofprotop) {
4674                 error = "Unknown bridge name";
4675                 goto exit;
4676             }
4677         }
4678     }
4679
4680     /* Generate a packet, if requested. */
4681     if (packet) {
4682         if (!ofpbuf_size(packet)) {
4683             flow_compose(packet, flow);
4684         } else {
4685             struct pkt_metadata md = pkt_metadata_from_flow(flow);
4686
4687             /* Use the metadata from the flow and the packet argument
4688              * to reconstruct the flow. */
4689             flow_extract(packet, &md, flow);
4690         }
4691     }
4692
4693 exit:
4694     if (error && !m_err) {
4695         m_err = xstrdup(error);
4696     }
4697     if (m_err) {
4698         ofpbuf_delete(packet);
4699         packet = NULL;
4700     }
4701     *packetp = packet;
4702     ofpbuf_uninit(&odp_key);
4703     ofpbuf_uninit(&odp_mask);
4704     simap_destroy(&port_names);
4705     return m_err;
4706 }
4707
4708 static void
4709 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4710                       void *aux OVS_UNUSED)
4711 {
4712     struct ofproto_dpif *ofproto;
4713     struct ofpbuf *packet;
4714     char *error;
4715     struct flow flow;
4716
4717     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4718     if (!error) {
4719         struct ds result;
4720
4721         ds_init(&result);
4722         ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4723         unixctl_command_reply(conn, ds_cstr(&result));
4724         ds_destroy(&result);
4725         ofpbuf_delete(packet);
4726     } else {
4727         unixctl_command_reply_error(conn, error);
4728         free(error);
4729     }
4730 }
4731
4732 static void
4733 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4734                               const char *argv[], void *aux OVS_UNUSED)
4735 {
4736     enum ofputil_protocol usable_protocols;
4737     struct ofproto_dpif *ofproto;
4738     bool enforce_consistency;
4739     struct ofpbuf ofpacts;
4740     struct ofpbuf *packet;
4741     struct ds result;
4742     struct flow flow;
4743     uint16_t in_port;
4744
4745     /* Three kinds of error return values! */
4746     enum ofperr retval;
4747     char *error;
4748
4749     packet = NULL;
4750     ds_init(&result);
4751     ofpbuf_init(&ofpacts, 0);
4752
4753     /* Parse actions. */
4754     error = ofpacts_parse_actions(argv[--argc], &ofpacts, &usable_protocols);
4755     if (error) {
4756         unixctl_command_reply_error(conn, error);
4757         free(error);
4758         goto exit;
4759     }
4760
4761     /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4762      * consistency between the flow and the actions.  With -consistent, we
4763      * enforce consistency even for a flow supported in OpenFlow 1.0. */
4764     if (!strcmp(argv[1], "-consistent")) {
4765         enforce_consistency = true;
4766         argv++;
4767         argc--;
4768     } else {
4769         enforce_consistency = false;
4770     }
4771
4772     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4773     if (error) {
4774         unixctl_command_reply_error(conn, error);
4775         free(error);
4776         goto exit;
4777     }
4778
4779     /* Do the same checks as handle_packet_out() in ofproto.c.
4780      *
4781      * We pass a 'table_id' of 0 to ofpacts_check(), which isn't
4782      * strictly correct because these actions aren't in any table, but it's OK
4783      * because it 'table_id' is used only to check goto_table instructions, but
4784      * packet-outs take a list of actions and therefore it can't include
4785      * instructions.
4786      *
4787      * We skip the "meter" check here because meter is an instruction, not an
4788      * action, and thus cannot appear in ofpacts. */
4789     in_port = ofp_to_u16(flow.in_port.ofp_port);
4790     if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
4791         unixctl_command_reply_error(conn, "invalid in_port");
4792         goto exit;
4793     }
4794     if (enforce_consistency) {
4795         retval = ofpacts_check_consistency(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts),
4796                                            &flow, u16_to_ofp(ofproto->up.max_ports),
4797                                            0, 0, usable_protocols);
4798     } else {
4799         retval = ofpacts_check(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &flow,
4800                                u16_to_ofp(ofproto->up.max_ports), 0, 0,
4801                                &usable_protocols);
4802     }
4803
4804     if (retval) {
4805         ds_clear(&result);
4806         ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
4807         unixctl_command_reply_error(conn, ds_cstr(&result));
4808         goto exit;
4809     }
4810
4811     ofproto_trace(ofproto, &flow, packet,
4812                   ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &result);
4813     unixctl_command_reply(conn, ds_cstr(&result));
4814
4815 exit:
4816     ds_destroy(&result);
4817     ofpbuf_delete(packet);
4818     ofpbuf_uninit(&ofpacts);
4819 }
4820
4821 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4822  * description of the results to 'ds'.
4823  *
4824  * The trace follows a packet with the specified 'flow' through the flow
4825  * table.  'packet' may be nonnull to trace an actual packet, with consequent
4826  * side effects (if it is nonnull then its flow must be 'flow').
4827  *
4828  * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4829  * trace, otherwise the actions are determined by a flow table lookup. */
4830 static void
4831 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
4832               const struct ofpbuf *packet,
4833               const struct ofpact ofpacts[], size_t ofpacts_len,
4834               struct ds *ds)
4835 {
4836     struct trace_ctx trace;
4837
4838     ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
4839     ds_put_cstr(ds, "Flow: ");
4840     flow_format(ds, flow);
4841     ds_put_char(ds, '\n');
4842
4843     flow_wildcards_init_catchall(&trace.wc);
4844
4845     trace.result = ds;
4846     trace.key = flow; /* Original flow key, used for megaflow. */
4847     trace.flow = *flow; /* May be modified by actions. */
4848     xlate_in_init(&trace.xin, ofproto, flow, flow->in_port.ofp_port, NULL,
4849                   ntohs(flow->tcp_flags), packet);
4850     trace.xin.ofpacts = ofpacts;
4851     trace.xin.ofpacts_len = ofpacts_len;
4852     trace.xin.resubmit_hook = trace_resubmit;
4853     trace.xin.report_hook = trace_report;
4854
4855     xlate_actions(&trace.xin, &trace.xout);
4856
4857     ds_put_char(ds, '\n');
4858     trace_format_flow(ds, 0, "Final flow", &trace);
4859     trace_format_megaflow(ds, 0, "Megaflow", &trace);
4860
4861     ds_put_cstr(ds, "Datapath actions: ");
4862     format_odp_actions(ds, ofpbuf_data(trace.xout.odp_actions),
4863                        ofpbuf_size(trace.xout.odp_actions));
4864
4865     if (trace.xout.slow) {
4866         enum slow_path_reason slow;
4867
4868         ds_put_cstr(ds, "\nThis flow is handled by the userspace "
4869                     "slow path because it:");
4870
4871         slow = trace.xout.slow;
4872         while (slow) {
4873             enum slow_path_reason bit = rightmost_1bit(slow);
4874
4875             ds_put_format(ds, "\n\t- %s.",
4876                           slow_path_reason_to_explanation(bit));
4877
4878             slow &= ~bit;
4879         }
4880     }
4881
4882     xlate_out_uninit(&trace.xout);
4883 }
4884
4885 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
4886  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
4887  * to destroy 'ofproto_shash' and free the returned value. */
4888 static const struct shash_node **
4889 get_ofprotos(struct shash *ofproto_shash)
4890 {
4891     const struct ofproto_dpif *ofproto;
4892
4893     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4894         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
4895         shash_add_nocopy(ofproto_shash, name, ofproto);
4896     }
4897
4898     return shash_sort(ofproto_shash);
4899 }
4900
4901 static void
4902 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
4903                               const char *argv[] OVS_UNUSED,
4904                               void *aux OVS_UNUSED)
4905 {
4906     struct ds ds = DS_EMPTY_INITIALIZER;
4907     struct shash ofproto_shash;
4908     const struct shash_node **sorted_ofprotos;
4909     int i;
4910
4911     shash_init(&ofproto_shash);
4912     sorted_ofprotos = get_ofprotos(&ofproto_shash);
4913     for (i = 0; i < shash_count(&ofproto_shash); i++) {
4914         const struct shash_node *node = sorted_ofprotos[i];
4915         ds_put_format(&ds, "%s\n", node->name);
4916     }
4917
4918     shash_destroy(&ofproto_shash);
4919     free(sorted_ofprotos);
4920
4921     unixctl_command_reply(conn, ds_cstr(&ds));
4922     ds_destroy(&ds);
4923 }
4924
4925 static void
4926 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
4927 {
4928     const struct shash_node **ofprotos;
4929     struct dpif_dp_stats dp_stats;
4930     struct shash ofproto_shash;
4931     size_t i;
4932
4933     dpif_get_dp_stats(backer->dpif, &dp_stats);
4934
4935     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
4936                   dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
4937
4938     shash_init(&ofproto_shash);
4939     ofprotos = get_ofprotos(&ofproto_shash);
4940     for (i = 0; i < shash_count(&ofproto_shash); i++) {
4941         struct ofproto_dpif *ofproto = ofprotos[i]->data;
4942         const struct shash_node **ports;
4943         size_t j;
4944
4945         if (ofproto->backer != backer) {
4946             continue;
4947         }
4948
4949         ds_put_format(ds, "\t%s:\n", ofproto->up.name);
4950
4951         ports = shash_sort(&ofproto->up.port_by_name);
4952         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
4953             const struct shash_node *node = ports[j];
4954             struct ofport *ofport = node->data;
4955             struct smap config;
4956             odp_port_t odp_port;
4957
4958             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
4959                           ofport->ofp_port);
4960
4961             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
4962             if (odp_port != ODPP_NONE) {
4963                 ds_put_format(ds, "%"PRIu32":", odp_port);
4964             } else {
4965                 ds_put_cstr(ds, "none:");
4966             }
4967
4968             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
4969
4970             smap_init(&config);
4971             if (!netdev_get_config(ofport->netdev, &config)) {
4972                 const struct smap_node **nodes;
4973                 size_t i;
4974
4975                 nodes = smap_sort(&config);
4976                 for (i = 0; i < smap_count(&config); i++) {
4977                     const struct smap_node *node = nodes[i];
4978                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
4979                                   node->key, node->value);
4980                 }
4981                 free(nodes);
4982             }
4983             smap_destroy(&config);
4984
4985             ds_put_char(ds, ')');
4986             ds_put_char(ds, '\n');
4987         }
4988         free(ports);
4989     }
4990     shash_destroy(&ofproto_shash);
4991     free(ofprotos);
4992 }
4993
4994 static void
4995 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4996                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4997 {
4998     struct ds ds = DS_EMPTY_INITIALIZER;
4999     const struct shash_node **backers;
5000     int i;
5001
5002     backers = shash_sort(&all_dpif_backers);
5003     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5004         dpif_show_backer(backers[i]->data, &ds);
5005     }
5006     free(backers);
5007
5008     unixctl_command_reply(conn, ds_cstr(&ds));
5009     ds_destroy(&ds);
5010 }
5011
5012 static void
5013 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5014                                 int argc OVS_UNUSED, const char *argv[],
5015                                 void *aux OVS_UNUSED)
5016 {
5017     const struct ofproto_dpif *ofproto;
5018
5019     struct ds ds = DS_EMPTY_INITIALIZER;
5020     bool verbosity = false;
5021
5022     struct dpif_port dpif_port;
5023     struct dpif_port_dump port_dump;
5024     struct hmap portno_names;
5025
5026     struct dpif_flow_dump *flow_dump;
5027     struct dpif_flow_dump_thread *flow_dump_thread;
5028     struct dpif_flow f;
5029     int error;
5030
5031     ofproto = ofproto_dpif_lookup(argv[argc - 1]);
5032     if (!ofproto) {
5033         unixctl_command_reply_error(conn, "no such bridge");
5034         return;
5035     }
5036
5037     if (argc > 2 && !strcmp(argv[1], "-m")) {
5038         verbosity = true;
5039     }
5040
5041     hmap_init(&portno_names);
5042     DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
5043         odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
5044     }
5045
5046     ds_init(&ds);
5047     flow_dump = dpif_flow_dump_create(ofproto->backer->dpif, false);
5048     flow_dump_thread = dpif_flow_dump_thread_create(flow_dump);
5049     while (dpif_flow_dump_next(flow_dump_thread, &f, 1)) {
5050         struct flow flow;
5051
5052         if (odp_flow_key_to_flow(f.key, f.key_len, &flow) == ODP_FIT_ERROR
5053             || xlate_lookup_ofproto(ofproto->backer, &flow, NULL) != ofproto) {
5054             continue;
5055         }
5056
5057         if (verbosity) {
5058             odp_format_ufid(&f.ufid, &ds);
5059             ds_put_cstr(&ds, " ");
5060         }
5061         odp_flow_format(f.key, f.key_len, f.mask, f.mask_len,
5062                         &portno_names, &ds, verbosity);
5063         ds_put_cstr(&ds, ", ");
5064         dpif_flow_stats_format(&f.stats, &ds);
5065         ds_put_cstr(&ds, ", actions:");
5066         format_odp_actions(&ds, f.actions, f.actions_len);
5067         ds_put_char(&ds, '\n');
5068     }
5069     dpif_flow_dump_thread_destroy(flow_dump_thread);
5070     error = dpif_flow_dump_destroy(flow_dump);
5071
5072     if (error) {
5073         ds_clear(&ds);
5074         ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
5075         unixctl_command_reply_error(conn, ds_cstr(&ds));
5076     } else {
5077         unixctl_command_reply(conn, ds_cstr(&ds));
5078     }
5079     odp_portno_names_destroy(&portno_names);
5080     hmap_destroy(&portno_names);
5081     ds_destroy(&ds);
5082 }
5083
5084 static void
5085 ofproto_revalidate_all_backers(void)
5086 {
5087     const struct shash_node **backers;
5088     int i;
5089
5090     backers = shash_sort(&all_dpif_backers);
5091     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5092         struct dpif_backer *backer = backers[i]->data;
5093         backer->need_revalidate = REV_RECONFIGURE;
5094     }
5095     free(backers);
5096 }
5097
5098 static void
5099 disable_tnl_push_pop(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5100                      const char *argv[], void *aux OVS_UNUSED)
5101 {
5102     if (!strcasecmp(argv[1], "off")) {
5103         ofproto_use_tnl_push_pop = false;
5104         unixctl_command_reply(conn, "Tunnel push-pop off");
5105         ofproto_revalidate_all_backers();
5106     } else if (!strcasecmp(argv[1], "on")) {
5107         ofproto_use_tnl_push_pop = true;
5108         unixctl_command_reply(conn, "Tunnel push-pop on");
5109         ofproto_revalidate_all_backers();
5110     }
5111 }
5112
5113 static void
5114 ofproto_unixctl_init(void)
5115 {
5116     static bool registered;
5117     if (registered) {
5118         return;
5119     }
5120     registered = true;
5121
5122     unixctl_command_register(
5123         "ofproto/trace",
5124         "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
5125         1, 3, ofproto_unixctl_trace, NULL);
5126     unixctl_command_register(
5127         "ofproto/trace-packet-out",
5128         "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
5129         2, 6, ofproto_unixctl_trace_actions, NULL);
5130     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5131                              ofproto_unixctl_fdb_flush, NULL);
5132     unixctl_command_register("fdb/show", "bridge", 1, 1,
5133                              ofproto_unixctl_fdb_show, NULL);
5134     unixctl_command_register("mdb/flush", "[bridge]", 0, 1,
5135                              ofproto_unixctl_mcast_snooping_flush, NULL);
5136     unixctl_command_register("mdb/show", "bridge", 1, 1,
5137                              ofproto_unixctl_mcast_snooping_show, NULL);
5138     unixctl_command_register("dpif/dump-dps", "", 0, 0,
5139                              ofproto_unixctl_dpif_dump_dps, NULL);
5140     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5141                              NULL);
5142     unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
5143                              ofproto_unixctl_dpif_dump_flows, NULL);
5144
5145     unixctl_command_register("ofproto/tnl-push-pop", "[on]|[off]", 1, 1,
5146                              disable_tnl_push_pop, NULL);
5147 }
5148
5149 /* Returns true if 'table' is the table used for internal rules,
5150  * false otherwise. */
5151 bool
5152 table_is_internal(uint8_t table_id)
5153 {
5154     return table_id == TBL_INTERNAL;
5155 }
5156 \f
5157 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5158  *
5159  * This is deprecated.  It is only for compatibility with broken device drivers
5160  * in old versions of Linux that do not properly support VLANs when VLAN
5161  * devices are not used.  When broken device drivers are no longer in
5162  * widespread use, we will delete these interfaces. */
5163
5164 static int
5165 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5166 {
5167     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5168     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5169
5170     if (realdev_ofp_port == ofport->realdev_ofp_port
5171         && vid == ofport->vlandev_vid) {
5172         return 0;
5173     }
5174
5175     ofproto->backer->need_revalidate = REV_RECONFIGURE;
5176
5177     if (ofport->realdev_ofp_port) {
5178         vsp_remove(ofport);
5179     }
5180     if (realdev_ofp_port && ofport->bundle) {
5181         /* vlandevs are enslaved to their realdevs, so they are not allowed to
5182          * themselves be part of a bundle. */
5183         bundle_set(ofport_->ofproto, ofport->bundle, NULL);
5184     }
5185
5186     ofport->realdev_ofp_port = realdev_ofp_port;
5187     ofport->vlandev_vid = vid;
5188
5189     if (realdev_ofp_port) {
5190         vsp_add(ofport, realdev_ofp_port, vid);
5191     }
5192
5193     return 0;
5194 }
5195
5196 static uint32_t
5197 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
5198 {
5199     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
5200 }
5201
5202 bool
5203 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
5204     OVS_EXCLUDED(ofproto->vsp_mutex)
5205 {
5206     /* hmap_is_empty is thread safe. */
5207     return !hmap_is_empty(&ofproto->realdev_vid_map);
5208 }
5209
5210
5211 static ofp_port_t
5212 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
5213                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5214     OVS_REQUIRES(ofproto->vsp_mutex)
5215 {
5216     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5217         int vid = vlan_tci_to_vid(vlan_tci);
5218         const struct vlan_splinter *vsp;
5219
5220         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5221                                  hash_realdev_vid(realdev_ofp_port, vid),
5222                                  &ofproto->realdev_vid_map) {
5223             if (vsp->realdev_ofp_port == realdev_ofp_port
5224                 && vsp->vid == vid) {
5225                 return vsp->vlandev_ofp_port;
5226             }
5227         }
5228     }
5229     return realdev_ofp_port;
5230 }
5231
5232 /* Returns the OFP port number of the Linux VLAN device that corresponds to
5233  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
5234  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
5235  * 'vlan_tci' 9, it would return the port number of eth0.9.
5236  *
5237  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
5238  * function just returns its 'realdev_ofp_port' argument. */
5239 ofp_port_t
5240 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5241                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5242     OVS_EXCLUDED(ofproto->vsp_mutex)
5243 {
5244     ofp_port_t ret;
5245
5246     /* hmap_is_empty is thread safe, see if we can return immediately. */
5247     if (hmap_is_empty(&ofproto->realdev_vid_map)) {
5248         return realdev_ofp_port;
5249     }
5250     ovs_mutex_lock(&ofproto->vsp_mutex);
5251     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
5252     ovs_mutex_unlock(&ofproto->vsp_mutex);
5253     return ret;
5254 }
5255
5256 static struct vlan_splinter *
5257 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
5258 {
5259     struct vlan_splinter *vsp;
5260
5261     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
5262                              hash_ofp_port(vlandev_ofp_port),
5263                              &ofproto->vlandev_map) {
5264         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5265             return vsp;
5266         }
5267     }
5268
5269     return NULL;
5270 }
5271
5272 /* Returns the OpenFlow port number of the "real" device underlying the Linux
5273  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
5274  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
5275  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
5276  * eth0 and store 9 in '*vid'.
5277  *
5278  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
5279  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
5280  * always does.*/
5281 static ofp_port_t
5282 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5283                        ofp_port_t vlandev_ofp_port, int *vid)
5284     OVS_REQUIRES(ofproto->vsp_mutex)
5285 {
5286     if (!hmap_is_empty(&ofproto->vlandev_map)) {
5287         const struct vlan_splinter *vsp;
5288
5289         vsp = vlandev_find(ofproto, vlandev_ofp_port);
5290         if (vsp) {
5291             if (vid) {
5292                 *vid = vsp->vid;
5293             }
5294             return vsp->realdev_ofp_port;
5295         }
5296     }
5297     return 0;
5298 }
5299
5300 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
5301  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
5302  * 'flow->in_port' to the "real" device backing the VLAN device, sets
5303  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Optionally pushes the
5304  * appropriate VLAN on 'packet' if provided.  Otherwise (which is always the
5305  * case unless VLAN splinters are enabled), returns false without making any
5306  * changes. */
5307 bool
5308 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow,
5309                 struct ofpbuf *packet)
5310     OVS_EXCLUDED(ofproto->vsp_mutex)
5311 {
5312     ofp_port_t realdev;
5313     int vid;
5314
5315     /* hmap_is_empty is thread safe. */
5316     if (hmap_is_empty(&ofproto->vlandev_map)) {
5317         return false;
5318     }
5319
5320     ovs_mutex_lock(&ofproto->vsp_mutex);
5321     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
5322     ovs_mutex_unlock(&ofproto->vsp_mutex);
5323     if (!realdev) {
5324         return false;
5325     }
5326
5327     /* Cause the flow to be processed as if it came in on the real device with
5328      * the VLAN device's VLAN ID. */
5329     flow->in_port.ofp_port = realdev;
5330     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
5331
5332     if (packet) {
5333         /* Make the packet resemble the flow, so that it gets sent to an
5334          * OpenFlow controller properly, so that it looks correct for sFlow,
5335          * and so that flow_extract() will get the correct vlan_tci if it is
5336          * called on 'packet'. */
5337         eth_push_vlan(packet, htons(ETH_TYPE_VLAN), flow->vlan_tci);
5338     }
5339
5340     return true;
5341 }
5342
5343 static void
5344 vsp_remove(struct ofport_dpif *port)
5345 {
5346     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5347     struct vlan_splinter *vsp;
5348
5349     ovs_mutex_lock(&ofproto->vsp_mutex);
5350     vsp = vlandev_find(ofproto, port->up.ofp_port);
5351     if (vsp) {
5352         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5353         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
5354         free(vsp);
5355
5356         port->realdev_ofp_port = 0;
5357     } else {
5358         VLOG_ERR("missing vlan device record");
5359     }
5360     ovs_mutex_unlock(&ofproto->vsp_mutex);
5361 }
5362
5363 static void
5364 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
5365 {
5366     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5367
5368     ovs_mutex_lock(&ofproto->vsp_mutex);
5369     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
5370         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
5371             == realdev_ofp_port)) {
5372         struct vlan_splinter *vsp;
5373
5374         vsp = xmalloc(sizeof *vsp);
5375         vsp->realdev_ofp_port = realdev_ofp_port;
5376         vsp->vlandev_ofp_port = port->up.ofp_port;
5377         vsp->vid = vid;
5378
5379         port->realdev_ofp_port = realdev_ofp_port;
5380
5381         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
5382                     hash_ofp_port(port->up.ofp_port));
5383         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
5384                     hash_realdev_vid(realdev_ofp_port, vid));
5385     } else {
5386         VLOG_ERR("duplicate vlan device record");
5387     }
5388     ovs_mutex_unlock(&ofproto->vsp_mutex);
5389 }
5390
5391 static odp_port_t
5392 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
5393 {
5394     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
5395     return ofport ? ofport->odp_port : ODPP_NONE;
5396 }
5397
5398 struct ofport_dpif *
5399 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
5400 {
5401     struct ofport_dpif *port;
5402
5403     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
5404     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
5405                              &backer->odp_to_ofport_map) {
5406         if (port->odp_port == odp_port) {
5407             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5408             return port;
5409         }
5410     }
5411
5412     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5413     return NULL;
5414 }
5415
5416 static ofp_port_t
5417 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
5418 {
5419     struct ofport_dpif *port;
5420
5421     port = odp_port_to_ofport(ofproto->backer, odp_port);
5422     if (port && &ofproto->up == port->up.ofproto) {
5423         return port->up.ofp_port;
5424     } else {
5425         return OFPP_NONE;
5426     }
5427 }
5428
5429 struct ofproto_dpif *
5430 ofproto_dpif_recirc_get_ofproto(const struct dpif_backer *backer,
5431                                 uint32_t recirc_id)
5432 {
5433     struct dpif_backer_recirc_node *node;
5434
5435     node = CONTAINER_OF(cmap_find(&backer->recirc_map, recirc_id),
5436                         struct dpif_backer_recirc_node, cmap_node);
5437
5438     return node ? node->ofproto : NULL;
5439 }
5440
5441 uint32_t
5442 ofproto_dpif_alloc_recirc_id(struct ofproto_dpif *ofproto)
5443 {
5444     struct dpif_backer *backer = ofproto->backer;
5445     uint32_t recirc_id = recirc_id_alloc(backer->rid_pool);
5446
5447     if (recirc_id) {
5448         struct dpif_backer_recirc_node *node = xmalloc(sizeof *node);
5449
5450         node->recirc_id = recirc_id;
5451         node->ofproto = ofproto;
5452
5453         ovs_mutex_lock(&backer->recirc_mutex);
5454         cmap_insert(&backer->recirc_map, &node->cmap_node, node->recirc_id);
5455         ovs_mutex_unlock(&backer->recirc_mutex);
5456     }
5457
5458     return recirc_id;
5459 }
5460
5461 void
5462 ofproto_dpif_free_recirc_id(struct ofproto_dpif *ofproto, uint32_t recirc_id)
5463 {
5464     struct dpif_backer *backer = ofproto->backer;
5465     struct dpif_backer_recirc_node *node;
5466
5467     node = CONTAINER_OF(cmap_find(&backer->recirc_map, recirc_id),
5468                         struct dpif_backer_recirc_node, cmap_node);
5469     if (node) {
5470         ovs_mutex_lock(&backer->recirc_mutex);
5471         cmap_remove(&backer->recirc_map, &node->cmap_node, node->recirc_id);
5472         ovs_mutex_unlock(&backer->recirc_mutex);
5473         recirc_id_free(backer->rid_pool, node->recirc_id);
5474
5475         /* 'recirc_id' should never be freed by non-owning 'ofproto'. */
5476         ovs_assert(node->ofproto == ofproto);
5477
5478         /* RCU postpone the free, since other threads may be referring
5479          * to 'node' at same time. */
5480         ovsrcu_postpone(free, node);
5481     }
5482 }
5483
5484 int
5485 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
5486                                const struct match *match, int priority,
5487                                uint16_t idle_timeout,
5488                                const struct ofpbuf *ofpacts,
5489                                struct rule **rulep)
5490 {
5491     struct ofputil_flow_mod fm;
5492     struct rule_dpif *rule;
5493     int error;
5494
5495     fm.match = *match;
5496     fm.priority = priority;
5497     fm.new_cookie = htonll(0);
5498     fm.cookie = htonll(0);
5499     fm.cookie_mask = htonll(0);
5500     fm.modify_cookie = false;
5501     fm.table_id = TBL_INTERNAL;
5502     fm.command = OFPFC_ADD;
5503     fm.idle_timeout = idle_timeout;
5504     fm.hard_timeout = 0;
5505     fm.importance = 0;
5506     fm.buffer_id = 0;
5507     fm.out_port = 0;
5508     fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5509     fm.ofpacts = ofpbuf_data(ofpacts);
5510     fm.ofpacts_len = ofpbuf_size(ofpacts);
5511
5512     error = ofproto_flow_mod(&ofproto->up, &fm);
5513     if (error) {
5514         VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
5515                     ofperr_to_string(error));
5516         *rulep = NULL;
5517         return error;
5518     }
5519
5520     rule = rule_dpif_lookup_in_table(ofproto, TBL_INTERNAL, &fm.match.flow,
5521                                      &fm.match.wc, false);
5522     if (rule) {
5523         *rulep = &rule->up;
5524     } else {
5525         OVS_NOT_REACHED();
5526     }
5527     return 0;
5528 }
5529
5530 int
5531 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
5532                                   struct match *match, int priority)
5533 {
5534     struct ofputil_flow_mod fm;
5535     int error;
5536
5537     fm.match = *match;
5538     fm.priority = priority;
5539     fm.new_cookie = htonll(0);
5540     fm.cookie = htonll(0);
5541     fm.cookie_mask = htonll(0);
5542     fm.modify_cookie = false;
5543     fm.table_id = TBL_INTERNAL;
5544     fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5545     fm.command = OFPFC_DELETE_STRICT;
5546
5547     error = ofproto_flow_mod(&ofproto->up, &fm);
5548     if (error) {
5549         VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
5550                     ofperr_to_string(error));
5551         return error;
5552     }
5553
5554     return 0;
5555 }
5556
5557 const struct ofproto_class ofproto_dpif_class = {
5558     init,
5559     enumerate_types,
5560     enumerate_names,
5561     del,
5562     port_open_type,
5563     type_run,
5564     type_wait,
5565     alloc,
5566     construct,
5567     destruct,
5568     dealloc,
5569     run,
5570     wait,
5571     NULL,                       /* get_memory_usage. */
5572     type_get_memory_usage,
5573     flush,
5574     query_tables,
5575     port_alloc,
5576     port_construct,
5577     port_destruct,
5578     port_dealloc,
5579     port_modified,
5580     port_reconfigured,
5581     port_query_by_name,
5582     port_add,
5583     port_del,
5584     port_get_stats,
5585     port_dump_start,
5586     port_dump_next,
5587     port_dump_done,
5588     port_poll,
5589     port_poll_wait,
5590     port_is_lacp_current,
5591     port_get_lacp_stats,
5592     NULL,                       /* rule_choose_table */
5593     rule_alloc,
5594     rule_construct,
5595     rule_insert,
5596     rule_delete,
5597     rule_destruct,
5598     rule_dealloc,
5599     rule_get_stats,
5600     rule_execute,
5601     NULL,                       /* rule_premodify_actions */
5602     rule_modify_actions,
5603     set_frag_handling,
5604     packet_out,
5605     set_netflow,
5606     get_netflow_ids,
5607     set_sflow,
5608     set_ipfix,
5609     set_cfm,
5610     cfm_status_changed,
5611     get_cfm_status,
5612     set_bfd,
5613     bfd_status_changed,
5614     get_bfd_status,
5615     set_stp,
5616     get_stp_status,
5617     set_stp_port,
5618     get_stp_port_status,
5619     get_stp_port_stats,
5620     set_rstp,
5621     get_rstp_status,
5622     set_rstp_port,
5623     get_rstp_port_status,
5624     set_queues,
5625     bundle_set,
5626     bundle_remove,
5627     mirror_set__,
5628     mirror_get_stats__,
5629     set_flood_vlans,
5630     is_mirror_output_bundle,
5631     forward_bpdu_changed,
5632     set_mac_table_config,
5633     set_mcast_snooping,
5634     set_mcast_snooping_port,
5635     set_realdev,
5636     NULL,                       /* meter_get_features */
5637     NULL,                       /* meter_set */
5638     NULL,                       /* meter_get */
5639     NULL,                       /* meter_del */
5640     group_alloc,                /* group_alloc */
5641     group_construct,            /* group_construct */
5642     group_destruct,             /* group_destruct */
5643     group_dealloc,              /* group_dealloc */
5644     group_modify,               /* group_modify */
5645     group_get_stats,            /* group_get_stats */
5646     get_datapath_version,       /* get_datapath_version */
5647 };