ofproto/ofproto-dpif.c

   1 /*
   2  * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 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     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1276     if (error) {
1277         return error;
1278     }
1279
1280     ofproto->netflow = NULL;
1281     ofproto->sflow = NULL;
1282     ofproto->ipfix = NULL;
1283     ofproto->stp = NULL;
1284     ofproto->rstp = NULL;
1285     ofproto->dump_seq = 0;
1286     hmap_init(&ofproto->bundles);
1287     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1288     ofproto->ms = NULL;
1289     ofproto->mbridge = mbridge_create();
1290     ofproto->has_bonded_bundles = false;
1291     ofproto->lacp_enabled = false;
1292     ofproto_tunnel_init();
1293     ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1294     ovs_mutex_init(&ofproto->vsp_mutex);
1295
1296     guarded_list_init(&ofproto->pins);
1297
1298     ofproto_unixctl_init();
1299
1300     hmap_init(&ofproto->vlandev_map);
1301     hmap_init(&ofproto->realdev_vid_map);
1302
1303     sset_init(&ofproto->ports);
1304     sset_init(&ofproto->ghost_ports);
1305     sset_init(&ofproto->port_poll_set);
1306     ofproto->port_poll_errno = 0;
1307     ofproto->change_seq = 0;
1308     ofproto->pins_seq = seq_create();
1309     ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1310
1311
1312     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1313         struct iface_hint *iface_hint = node->data;
1314
1315         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1316             /* Check if the datapath already has this port. */
1317             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1318                 sset_add(&ofproto->ports, node->name);
1319             }
1320
1321             free(iface_hint->br_name);
1322             free(iface_hint->br_type);
1323             free(iface_hint);
1324             shash_delete(&init_ofp_ports, node);
1325         }
1326     }
1327
1328     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1329                 hash_string(ofproto->up.name, 0));
1330     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1331
1332     ofproto_init_tables(ofproto_, N_TABLES);
1333     error = add_internal_flows(ofproto);
1334
1335     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1336
1337     return error;
1338 }
1339
1340 static int
1341 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1342                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1343 {
1344     struct match match;
1345     int error;
1346     struct rule *rule;
1347
1348     match_init_catchall(&match);
1349     match_set_reg(&match, 0, id);
1350
1351     error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, 0, ofpacts,
1352                                            &rule);
1353     *rulep = error ? NULL : rule_dpif_cast(rule);
1354
1355     return error;
1356 }
1357
1358 static int
1359 add_internal_flows(struct ofproto_dpif *ofproto)
1360 {
1361     struct ofpact_controller *controller;
1362     uint64_t ofpacts_stub[128 / 8];
1363     struct ofpbuf ofpacts;
1364     struct rule *unused_rulep OVS_UNUSED;
1365     struct ofpact_resubmit *resubmit;
1366     struct match match;
1367     int error;
1368     int id;
1369
1370     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1371     id = 1;
1372
1373     controller = ofpact_put_CONTROLLER(&ofpacts);
1374     controller->max_len = UINT16_MAX;
1375     controller->controller_id = 0;
1376     controller->reason = OFPR_NO_MATCH;
1377     ofpact_pad(&ofpacts);
1378
1379     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1380                                    &ofproto->miss_rule);
1381     if (error) {
1382         return error;
1383     }
1384
1385     ofpbuf_clear(&ofpacts);
1386     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1387                                    &ofproto->no_packet_in_rule);
1388     if (error) {
1389         return error;
1390     }
1391
1392     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1393                                    &ofproto->drop_frags_rule);
1394     if (error) {
1395         return error;
1396     }
1397
1398     /* Drop any run away non-recirc rule lookups. Recirc_id has to be
1399      * zero when reaching this rule.
1400      *
1401      * (priority=2), recirc_id=0, actions=drop
1402      */
1403     ofpbuf_clear(&ofpacts);
1404     match_init_catchall(&match);
1405     match_set_recirc_id(&match, 0);
1406     error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, 0, &ofpacts,
1407                                            &unused_rulep);
1408     if (error) {
1409         return error;
1410     }
1411
1412     /* Continue rule lookups for not-matched recirc rules from table 0.
1413      *
1414      * (priority=1), actions=resubmit(, 0)
1415      */
1416     resubmit = ofpact_put_RESUBMIT(&ofpacts);
1417     resubmit->in_port = OFPP_IN_PORT;
1418     resubmit->table_id = 0;
1419
1420     match_init_catchall(&match);
1421     error = ofproto_dpif_add_internal_flow(ofproto, &match, 1, 0, &ofpacts,
1422                                            &unused_rulep);
1423
1424     return error;
1425 }
1426
1427 static void
1428 destruct(struct ofproto *ofproto_)
1429 {
1430     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1431     struct ofproto_packet_in *pin, *next_pin;
1432     struct rule_dpif *rule;
1433     struct oftable *table;
1434     struct ovs_list pins;
1435
1436     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1437     xlate_txn_start();
1438     xlate_remove_ofproto(ofproto);
1439     xlate_txn_commit();
1440
1441     /* Ensure that the upcall processing threads have no remaining references
1442      * to the ofproto or anything in it. */
1443     udpif_synchronize(ofproto->backer->udpif);
1444
1445     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1446
1447     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1448         CLS_FOR_EACH (rule, up.cr, &table->cls) {
1449             ofproto_rule_delete(&ofproto->up, &rule->up);
1450         }
1451     }
1452
1453     guarded_list_pop_all(&ofproto->pins, &pins);
1454     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1455         list_remove(&pin->list_node);
1456         free(CONST_CAST(void *, pin->up.packet));
1457         free(pin);
1458     }
1459     guarded_list_destroy(&ofproto->pins);
1460
1461     dpif_backer_recirc_clear_ofproto(ofproto->backer, ofproto);
1462
1463     mbridge_unref(ofproto->mbridge);
1464
1465     netflow_unref(ofproto->netflow);
1466     dpif_sflow_unref(ofproto->sflow);
1467     dpif_ipfix_unref(ofproto->ipfix);
1468     hmap_destroy(&ofproto->bundles);
1469     mac_learning_unref(ofproto->ml);
1470     mcast_snooping_unref(ofproto->ms);
1471
1472     hmap_destroy(&ofproto->vlandev_map);
1473     hmap_destroy(&ofproto->realdev_vid_map);
1474
1475     sset_destroy(&ofproto->ports);
1476     sset_destroy(&ofproto->ghost_ports);
1477     sset_destroy(&ofproto->port_poll_set);
1478
1479     ovs_mutex_destroy(&ofproto->stats_mutex);
1480     ovs_mutex_destroy(&ofproto->vsp_mutex);
1481
1482     seq_destroy(ofproto->pins_seq);
1483
1484     close_dpif_backer(ofproto->backer);
1485 }
1486
1487 static int
1488 run(struct ofproto *ofproto_)
1489 {
1490     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1491     uint64_t new_seq, new_dump_seq;
1492
1493     if (mbridge_need_revalidate(ofproto->mbridge)) {
1494         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1495         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1496         mac_learning_flush(ofproto->ml);
1497         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1498         mcast_snooping_mdb_flush(ofproto->ms);
1499     }
1500
1501     /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
1502      * flow restore.  Even though nothing is processed during flow restore,
1503      * all queued 'pins' will be handled immediately when flow restore
1504      * completes. */
1505     ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1506
1507     /* Do not perform any periodic activity required by 'ofproto' while
1508      * waiting for flow restore to complete. */
1509     if (!ofproto_get_flow_restore_wait()) {
1510         struct ofproto_packet_in *pin, *next_pin;
1511         struct ovs_list pins;
1512
1513         guarded_list_pop_all(&ofproto->pins, &pins);
1514         LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1515             connmgr_send_packet_in(ofproto->up.connmgr, pin);
1516             list_remove(&pin->list_node);
1517             free(CONST_CAST(void *, pin->up.packet));
1518             free(pin);
1519         }
1520     }
1521
1522     if (ofproto->netflow) {
1523         netflow_run(ofproto->netflow);
1524     }
1525     if (ofproto->sflow) {
1526         dpif_sflow_run(ofproto->sflow);
1527     }
1528     if (ofproto->ipfix) {
1529         dpif_ipfix_run(ofproto->ipfix);
1530     }
1531
1532     new_seq = seq_read(connectivity_seq_get());
1533     if (ofproto->change_seq != new_seq) {
1534         struct ofport_dpif *ofport;
1535
1536         HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1537             port_run(ofport);
1538         }
1539
1540         ofproto->change_seq = new_seq;
1541     }
1542     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1543         struct ofbundle *bundle;
1544
1545         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1546             bundle_run(bundle);
1547         }
1548     }
1549
1550     stp_run(ofproto);
1551     rstp_run(ofproto);
1552     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1553     if (mac_learning_run(ofproto->ml)) {
1554         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1555     }
1556     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1557
1558     if (mcast_snooping_run(ofproto->ms)) {
1559         ofproto->backer->need_revalidate = REV_MCAST_SNOOPING;
1560     }
1561
1562     new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1563     if (ofproto->dump_seq != new_dump_seq) {
1564         struct rule *rule, *next_rule;
1565
1566         /* We know stats are relatively fresh, so now is a good time to do some
1567          * periodic work. */
1568         ofproto->dump_seq = new_dump_seq;
1569
1570         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1571          * has passed. */
1572         ovs_mutex_lock(&ofproto_mutex);
1573         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1574                             &ofproto->up.expirable) {
1575             rule_expire(rule_dpif_cast(rule));
1576         }
1577         ovs_mutex_unlock(&ofproto_mutex);
1578
1579         /* All outstanding data in existing flows has been accounted, so it's a
1580          * good time to do bond rebalancing. */
1581         if (ofproto->has_bonded_bundles) {
1582             struct ofbundle *bundle;
1583
1584             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1585                 if (bundle->bond) {
1586                     bond_rebalance(bundle->bond);
1587                 }
1588             }
1589         }
1590     }
1591     return 0;
1592 }
1593
1594 static void
1595 wait(struct ofproto *ofproto_)
1596 {
1597     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1598
1599     if (ofproto_get_flow_restore_wait()) {
1600         return;
1601     }
1602
1603     if (ofproto->sflow) {
1604         dpif_sflow_wait(ofproto->sflow);
1605     }
1606     if (ofproto->ipfix) {
1607         dpif_ipfix_wait(ofproto->ipfix);
1608     }
1609     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1610         struct ofbundle *bundle;
1611
1612         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1613             bundle_wait(bundle);
1614         }
1615     }
1616     if (ofproto->netflow) {
1617         netflow_wait(ofproto->netflow);
1618     }
1619     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1620     mac_learning_wait(ofproto->ml);
1621     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1622     mcast_snooping_wait(ofproto->ms);
1623     stp_wait(ofproto);
1624     if (ofproto->backer->need_revalidate) {
1625         /* Shouldn't happen, but if it does just go around again. */
1626         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1627         poll_immediate_wake();
1628     }
1629
1630     seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1631     seq_wait(ofproto->pins_seq, ofproto->pins_seqno);
1632 }
1633
1634 static void
1635 type_get_memory_usage(const char *type, struct simap *usage)
1636 {
1637     struct dpif_backer *backer;
1638
1639     backer = shash_find_data(&all_dpif_backers, type);
1640     if (backer) {
1641         udpif_get_memory_usage(backer->udpif, usage);
1642     }
1643 }
1644
1645 static void
1646 flush(struct ofproto *ofproto_)
1647 {
1648     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1649     struct dpif_backer *backer = ofproto->backer;
1650
1651     if (backer) {
1652         udpif_flush(backer->udpif);
1653     }
1654 }
1655
1656 static void
1657 query_tables(struct ofproto *ofproto,
1658              struct ofputil_table_features *features,
1659              struct ofputil_table_stats *stats)
1660 {
1661     strcpy(features->name, "classifier");
1662
1663     if (stats) {
1664         int i;
1665
1666         for (i = 0; i < ofproto->n_tables; i++) {
1667             unsigned long missed, matched;
1668
1669             atomic_read_relaxed(&ofproto->tables[i].n_matched, &matched);
1670             atomic_read_relaxed(&ofproto->tables[i].n_missed, &missed);
1671
1672             stats[i].matched_count = matched;
1673             stats[i].lookup_count = matched + missed;
1674         }
1675     }
1676 }
1677
1678 static struct ofport *
1679 port_alloc(void)
1680 {
1681     struct ofport_dpif *port = xmalloc(sizeof *port);
1682     return &port->up;
1683 }
1684
1685 static void
1686 port_dealloc(struct ofport *port_)
1687 {
1688     struct ofport_dpif *port = ofport_dpif_cast(port_);
1689     free(port);
1690 }
1691
1692 static int
1693 port_construct(struct ofport *port_)
1694 {
1695     struct ofport_dpif *port = ofport_dpif_cast(port_);
1696     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1697     const struct netdev *netdev = port->up.netdev;
1698     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1699     struct dpif_port dpif_port;
1700     int error;
1701
1702     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1703     port->bundle = NULL;
1704     port->cfm = NULL;
1705     port->bfd = NULL;
1706     port->may_enable = false;
1707     port->stp_port = NULL;
1708     port->stp_state = STP_DISABLED;
1709     port->rstp_port = NULL;
1710     port->rstp_state = RSTP_DISABLED;
1711     port->is_tunnel = false;
1712     port->peer = NULL;
1713     port->qdscp = NULL;
1714     port->n_qdscp = 0;
1715     port->realdev_ofp_port = 0;
1716     port->vlandev_vid = 0;
1717     port->carrier_seq = netdev_get_carrier_resets(netdev);
1718     port->is_layer3 = netdev_vport_is_layer3(netdev);
1719
1720     if (netdev_vport_is_patch(netdev)) {
1721         /* By bailing out here, we don't submit the port to the sFlow module
1722          * to be considered for counter polling export.  This is correct
1723          * because the patch port represents an interface that sFlow considers
1724          * to be "internal" to the switch as a whole, and therefore not an
1725          * candidate for counter polling. */
1726         port->odp_port = ODPP_NONE;
1727         ofport_update_peer(port);
1728         return 0;
1729     }
1730
1731     error = dpif_port_query_by_name(ofproto->backer->dpif,
1732                                     netdev_vport_get_dpif_port(netdev, namebuf,
1733                                                                sizeof namebuf),
1734                                     &dpif_port);
1735     if (error) {
1736         return error;
1737     }
1738
1739     port->odp_port = dpif_port.port_no;
1740
1741     if (netdev_get_tunnel_config(netdev)) {
1742         atomic_count_inc(&ofproto->backer->tnl_count);
1743         tnl_port_add(port, port->up.netdev, port->odp_port,
1744                      ovs_native_tunneling_is_on(ofproto), namebuf);
1745         port->is_tunnel = true;
1746         if (ofproto->ipfix) {
1747            dpif_ipfix_add_tunnel_port(ofproto->ipfix, port_, port->odp_port);
1748         }
1749     } else {
1750         /* Sanity-check that a mapping doesn't already exist.  This
1751          * shouldn't happen for non-tunnel ports. */
1752         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1753             VLOG_ERR("port %s already has an OpenFlow port number",
1754                      dpif_port.name);
1755             dpif_port_destroy(&dpif_port);
1756             return EBUSY;
1757         }
1758
1759         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1760         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1761                     hash_odp_port(port->odp_port));
1762         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1763     }
1764     dpif_port_destroy(&dpif_port);
1765
1766     if (ofproto->sflow) {
1767         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1768     }
1769
1770     return 0;
1771 }
1772
1773 static void
1774 port_destruct(struct ofport *port_)
1775 {
1776     struct ofport_dpif *port = ofport_dpif_cast(port_);
1777     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1778     const char *devname = netdev_get_name(port->up.netdev);
1779     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1780     const char *dp_port_name;
1781
1782     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1783     xlate_txn_start();
1784     xlate_ofport_remove(port);
1785     xlate_txn_commit();
1786
1787     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1788                                               sizeof namebuf);
1789     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1790         /* The underlying device is still there, so delete it.  This
1791          * happens when the ofproto is being destroyed, since the caller
1792          * assumes that removal of attached ports will happen as part of
1793          * destruction. */
1794         if (!port->is_tunnel) {
1795             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1796         }
1797     }
1798
1799     if (port->peer) {
1800         port->peer->peer = NULL;
1801         port->peer = NULL;
1802     }
1803
1804     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1805         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1806         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1807         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1808     }
1809
1810     if (port->is_tunnel) {
1811         atomic_count_dec(&ofproto->backer->tnl_count);
1812     }
1813
1814     if (port->is_tunnel && ofproto->ipfix) {
1815        dpif_ipfix_del_tunnel_port(ofproto->ipfix, port->odp_port);
1816     }
1817
1818     tnl_port_del(port);
1819     sset_find_and_delete(&ofproto->ports, devname);
1820     sset_find_and_delete(&ofproto->ghost_ports, devname);
1821     bundle_remove(port_);
1822     set_cfm(port_, NULL);
1823     set_bfd(port_, NULL);
1824     if (port->stp_port) {
1825         stp_port_disable(port->stp_port);
1826     }
1827     set_rstp_port(port_, NULL);
1828     if (ofproto->sflow) {
1829         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1830     }
1831
1832     free(port->qdscp);
1833 }
1834
1835 static void
1836 port_modified(struct ofport *port_)
1837 {
1838     struct ofport_dpif *port = ofport_dpif_cast(port_);
1839     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1840     struct netdev *netdev = port->up.netdev;
1841
1842     if (port->bundle && port->bundle->bond) {
1843         bond_slave_set_netdev(port->bundle->bond, port, netdev);
1844     }
1845
1846     if (port->cfm) {
1847         cfm_set_netdev(port->cfm, netdev);
1848     }
1849
1850     if (port->bfd) {
1851         bfd_set_netdev(port->bfd, netdev);
1852     }
1853
1854     ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1855                                      port->up.pp.hw_addr);
1856
1857     netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1858
1859     if (port->is_tunnel) {
1860         struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1861
1862         if (tnl_port_reconfigure(port, netdev, port->odp_port,
1863                                  ovs_native_tunneling_is_on(ofproto), namebuf)) {
1864             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1865         }
1866     }
1867
1868     ofport_update_peer(port);
1869 }
1870
1871 static void
1872 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1873 {
1874     struct ofport_dpif *port = ofport_dpif_cast(port_);
1875     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1876     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1877
1878     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1879                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1880                    OFPUTIL_PC_NO_PACKET_IN)) {
1881         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1882
1883         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1884             bundle_update(port->bundle);
1885         }
1886     }
1887 }
1888
1889 static int
1890 set_sflow(struct ofproto *ofproto_,
1891           const struct ofproto_sflow_options *sflow_options)
1892 {
1893     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1894     struct dpif_sflow *ds = ofproto->sflow;
1895
1896     if (sflow_options) {
1897         uint32_t old_probability = ds ? dpif_sflow_get_probability(ds) : 0;
1898         if (!ds) {
1899             struct ofport_dpif *ofport;
1900
1901             ds = ofproto->sflow = dpif_sflow_create();
1902             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1903                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1904             }
1905         }
1906         dpif_sflow_set_options(ds, sflow_options);
1907         if (dpif_sflow_get_probability(ds) != old_probability) {
1908             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1909         }
1910     } else {
1911         if (ds) {
1912             dpif_sflow_unref(ds);
1913             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1914             ofproto->sflow = NULL;
1915         }
1916     }
1917     return 0;
1918 }
1919
1920 static int
1921 set_ipfix(
1922     struct ofproto *ofproto_,
1923     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1924     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1925     size_t n_flow_exporters_options)
1926 {
1927     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1928     struct dpif_ipfix *di = ofproto->ipfix;
1929     bool has_options = bridge_exporter_options || flow_exporters_options;
1930     bool new_di = false;
1931
1932     if (has_options && !di) {
1933         di = ofproto->ipfix = dpif_ipfix_create();
1934         new_di = true;
1935     }
1936
1937     if (di) {
1938         /* Call set_options in any case to cleanly flush the flow
1939          * caches in the last exporters that are to be destroyed. */
1940         dpif_ipfix_set_options(
1941             di, bridge_exporter_options, flow_exporters_options,
1942             n_flow_exporters_options);
1943
1944         /* Add tunnel ports only when a new ipfix created */
1945         if (new_di == true) {
1946             struct ofport_dpif *ofport;
1947             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1948                 if (ofport->is_tunnel == true) {
1949                     dpif_ipfix_add_tunnel_port(di, &ofport->up, ofport->odp_port);
1950                 }
1951             }
1952         }
1953
1954         if (!has_options) {
1955             dpif_ipfix_unref(di);
1956             ofproto->ipfix = NULL;
1957         }
1958     }
1959
1960     return 0;
1961 }
1962
1963 static int
1964 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1965 {
1966     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1967     int error = 0;
1968
1969     if (s) {
1970         if (!ofport->cfm) {
1971             struct ofproto_dpif *ofproto;
1972
1973             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1974             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1975             ofport->cfm = cfm_create(ofport->up.netdev);
1976         }
1977
1978         if (cfm_configure(ofport->cfm, s)) {
1979             error = 0;
1980             goto out;
1981         }
1982
1983         error = EINVAL;
1984     }
1985     cfm_unref(ofport->cfm);
1986     ofport->cfm = NULL;
1987 out:
1988     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1989                                      ofport->up.pp.hw_addr);
1990     return error;
1991 }
1992
1993 static bool
1994 cfm_status_changed(struct ofport *ofport_)
1995 {
1996     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1997
1998     return ofport->cfm ? cfm_check_status_change(ofport->cfm) : true;
1999 }
2000
2001 static int
2002 get_cfm_status(const struct ofport *ofport_,
2003                struct cfm_status *status)
2004 {
2005     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2006     int ret = 0;
2007
2008     if (ofport->cfm) {
2009         cfm_get_status(ofport->cfm, status);
2010     } else {
2011         ret = ENOENT;
2012     }
2013
2014     return ret;
2015 }
2016
2017 static int
2018 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2019 {
2020     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2021     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2022     struct bfd *old;
2023
2024     old = ofport->bfd;
2025     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2026                                 cfg, ofport->up.netdev);
2027     if (ofport->bfd != old) {
2028         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2029     }
2030     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
2031                                      ofport->up.pp.hw_addr);
2032     return 0;
2033 }
2034
2035 static bool
2036 bfd_status_changed(struct ofport *ofport_)
2037 {
2038     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2039
2040     return ofport->bfd ? bfd_check_status_change(ofport->bfd) : true;
2041 }
2042
2043 static int
2044 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2045 {
2046     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2047     int ret = 0;
2048
2049     if (ofport->bfd) {
2050         bfd_get_status(ofport->bfd, smap);
2051     } else {
2052         ret = ENOENT;
2053     }
2054
2055     return ret;
2056 }
2057 \f
2058 /* Spanning Tree. */
2059
2060 /* Called while rstp_mutex is held. */
2061 static void
2062 rstp_send_bpdu_cb(struct ofpbuf *pkt, void *ofport_, void *ofproto_)
2063 {
2064     struct ofproto_dpif *ofproto = ofproto_;
2065     struct ofport_dpif *ofport = ofport_;
2066     struct eth_header *eth = ofpbuf_l2(pkt);
2067
2068     netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2069     if (eth_addr_is_zero(eth->eth_src)) {
2070         VLOG_WARN_RL(&rl, "%s port %d: cannot send RSTP BPDU on a port which "
2071                      "does not have a configured source MAC address.",
2072                      ofproto->up.name, ofp_to_u16(ofport->up.ofp_port));
2073     } else {
2074         ofproto_dpif_send_packet(ofport, pkt);
2075     }
2076     ofpbuf_delete(pkt);
2077 }
2078
2079 static void
2080 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2081 {
2082     struct ofproto_dpif *ofproto = ofproto_;
2083     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2084     struct ofport_dpif *ofport;
2085
2086     ofport = stp_port_get_aux(sp);
2087     if (!ofport) {
2088         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2089                      ofproto->up.name, port_num);
2090     } else {
2091         struct eth_header *eth = ofpbuf_l2(pkt);
2092
2093         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2094         if (eth_addr_is_zero(eth->eth_src)) {
2095             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2096                          "with unknown MAC", ofproto->up.name, port_num);
2097         } else {
2098             ofproto_dpif_send_packet(ofport, pkt);
2099         }
2100     }
2101     ofpbuf_delete(pkt);
2102 }
2103
2104 /* Configure RSTP on 'ofproto_' using the settings defined in 's'. */
2105 static void
2106 set_rstp(struct ofproto *ofproto_, const struct ofproto_rstp_settings *s)
2107 {
2108     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2109
2110     /* Only revalidate flows if the configuration changed. */
2111     if (!s != !ofproto->rstp) {
2112         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2113     }
2114
2115     if (s) {
2116         if (!ofproto->rstp) {
2117             ofproto->rstp = rstp_create(ofproto_->name, s->address,
2118                                         rstp_send_bpdu_cb, ofproto);
2119             ofproto->rstp_last_tick = time_msec();
2120         }
2121         rstp_set_bridge_address(ofproto->rstp, s->address);
2122         rstp_set_bridge_priority(ofproto->rstp, s->priority);
2123         rstp_set_bridge_ageing_time(ofproto->rstp, s->ageing_time);
2124         rstp_set_bridge_force_protocol_version(ofproto->rstp,
2125                                                s->force_protocol_version);
2126         rstp_set_bridge_max_age(ofproto->rstp, s->bridge_max_age);
2127         rstp_set_bridge_forward_delay(ofproto->rstp, s->bridge_forward_delay);
2128         rstp_set_bridge_transmit_hold_count(ofproto->rstp,
2129                                             s->transmit_hold_count);
2130     } else {
2131         struct ofport *ofport;
2132         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2133             set_rstp_port(ofport, NULL);
2134         }
2135         rstp_unref(ofproto->rstp);
2136         ofproto->rstp = NULL;
2137     }
2138 }
2139
2140 static void
2141 get_rstp_status(struct ofproto *ofproto_, struct ofproto_rstp_status *s)
2142 {
2143     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2144
2145     if (ofproto->rstp) {
2146         s->enabled = true;
2147         s->root_id = rstp_get_root_id(ofproto->rstp);
2148         s->bridge_id = rstp_get_bridge_id(ofproto->rstp);
2149         s->designated_id = rstp_get_designated_id(ofproto->rstp);
2150         s->root_path_cost = rstp_get_root_path_cost(ofproto->rstp);
2151         s->designated_port_id = rstp_get_designated_port_id(ofproto->rstp);
2152         s->bridge_port_id = rstp_get_bridge_port_id(ofproto->rstp);
2153     } else {
2154         s->enabled = false;
2155     }
2156 }
2157
2158 static void
2159 update_rstp_port_state(struct ofport_dpif *ofport)
2160 {
2161     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2162     enum rstp_state state;
2163
2164     /* Figure out new state. */
2165     state = ofport->rstp_port ? rstp_port_get_state(ofport->rstp_port)
2166         : RSTP_DISABLED;
2167
2168     /* Update state. */
2169     if (ofport->rstp_state != state) {
2170         enum ofputil_port_state of_state;
2171         bool fwd_change;
2172
2173         VLOG_DBG("port %s: RSTP state changed from %s to %s",
2174                  netdev_get_name(ofport->up.netdev),
2175                  rstp_state_name(ofport->rstp_state),
2176                  rstp_state_name(state));
2177
2178         if (rstp_learn_in_state(ofport->rstp_state)
2179             != rstp_learn_in_state(state)) {
2180             /* XXX: Learning action flows should also be flushed. */
2181             if (ofport->bundle) {
2182                 if (!rstp_shift_root_learned_address(ofproto->rstp)
2183                     || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2184                     bundle_flush_macs(ofport->bundle, false);
2185                 }
2186             }
2187         }
2188         fwd_change = rstp_forward_in_state(ofport->rstp_state)
2189             != rstp_forward_in_state(state);
2190
2191         ofproto->backer->need_revalidate = REV_RSTP;
2192         ofport->rstp_state = state;
2193
2194         if (fwd_change && ofport->bundle) {
2195             bundle_update(ofport->bundle);
2196         }
2197
2198         /* Update the RSTP state bits in the OpenFlow port description. */
2199         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2200         of_state |= (state == RSTP_LEARNING ? OFPUTIL_PS_STP_LEARN
2201                 : state == RSTP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2202                 : state == RSTP_DISCARDING ?  OFPUTIL_PS_STP_LISTEN
2203                 : 0);
2204         ofproto_port_set_state(&ofport->up, of_state);
2205     }
2206 }
2207
2208 static void
2209 rstp_run(struct ofproto_dpif *ofproto)
2210 {
2211     if (ofproto->rstp) {
2212         long long int now = time_msec();
2213         long long int elapsed = now - ofproto->rstp_last_tick;
2214         struct rstp_port *rp;
2215         struct ofport_dpif *ofport;
2216
2217         /* Every second, decrease the values of the timers. */
2218         if (elapsed >= 1000) {
2219             rstp_tick_timers(ofproto->rstp);
2220             ofproto->rstp_last_tick = now;
2221         }
2222         rp = NULL;
2223         while ((ofport = rstp_get_next_changed_port_aux(ofproto->rstp, &rp))) {
2224             update_rstp_port_state(ofport);
2225         }
2226         rp = NULL;
2227         ofport = NULL;
2228         /* FIXME: This check should be done on-event (i.e., when setting
2229          * p->fdb_flush) and not periodically.
2230          */
2231         while ((ofport = rstp_check_and_reset_fdb_flush(ofproto->rstp, &rp))) {
2232             if (!rstp_shift_root_learned_address(ofproto->rstp)
2233                 || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2234                 bundle_flush_macs(ofport->bundle, false);
2235             }
2236         }
2237
2238         if (rstp_shift_root_learned_address(ofproto->rstp)) {
2239             bundle_move(((struct ofport_dpif *)rstp_get_old_root_aux(ofproto->rstp))->bundle,
2240                         ((struct ofport_dpif *)rstp_get_new_root_aux(ofproto->rstp))->bundle);
2241             rstp_reset_root_changed(ofproto->rstp);
2242         }
2243     }
2244 }
2245
2246 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2247 static int
2248 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2249 {
2250     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2251
2252     /* Only revalidate flows if the configuration changed. */
2253     if (!s != !ofproto->stp) {
2254         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2255     }
2256
2257     if (s) {
2258         if (!ofproto->stp) {
2259             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2260                                       send_bpdu_cb, ofproto);
2261             ofproto->stp_last_tick = time_msec();
2262         }
2263
2264         stp_set_bridge_id(ofproto->stp, s->system_id);
2265         stp_set_bridge_priority(ofproto->stp, s->priority);
2266         stp_set_hello_time(ofproto->stp, s->hello_time);
2267         stp_set_max_age(ofproto->stp, s->max_age);
2268         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2269     }  else {
2270         struct ofport *ofport;
2271
2272         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2273             set_stp_port(ofport, NULL);
2274         }
2275
2276         stp_unref(ofproto->stp);
2277         ofproto->stp = NULL;
2278     }
2279
2280     return 0;
2281 }
2282
2283 static int
2284 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2285 {
2286     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2287
2288     if (ofproto->stp) {
2289         s->enabled = true;
2290         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2291         s->designated_root = stp_get_designated_root(ofproto->stp);
2292         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2293     } else {
2294         s->enabled = false;
2295     }
2296
2297     return 0;
2298 }
2299
2300 static void
2301 update_stp_port_state(struct ofport_dpif *ofport)
2302 {
2303     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2304     enum stp_state state;
2305
2306     /* Figure out new state. */
2307     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2308                              : STP_DISABLED;
2309
2310     /* Update state. */
2311     if (ofport->stp_state != state) {
2312         enum ofputil_port_state of_state;
2313         bool fwd_change;
2314
2315         VLOG_DBG("port %s: STP state changed from %s to %s",
2316                  netdev_get_name(ofport->up.netdev),
2317                  stp_state_name(ofport->stp_state),
2318                  stp_state_name(state));
2319         if (stp_learn_in_state(ofport->stp_state)
2320                 != stp_learn_in_state(state)) {
2321             /* xxx Learning action flows should also be flushed. */
2322             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2323             mac_learning_flush(ofproto->ml);
2324             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2325             mcast_snooping_mdb_flush(ofproto->ms);
2326         }
2327         fwd_change = stp_forward_in_state(ofport->stp_state)
2328                         != stp_forward_in_state(state);
2329
2330         ofproto->backer->need_revalidate = REV_STP;
2331         ofport->stp_state = state;
2332         ofport->stp_state_entered = time_msec();
2333
2334         if (fwd_change && ofport->bundle) {
2335             bundle_update(ofport->bundle);
2336         }
2337
2338         /* Update the STP state bits in the OpenFlow port description. */
2339         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2340         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2341                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2342                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2343                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2344                      : 0);
2345         ofproto_port_set_state(&ofport->up, of_state);
2346     }
2347 }
2348
2349 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2350  * caller is responsible for assigning STP port numbers and ensuring
2351  * there are no duplicates. */
2352 static int
2353 set_stp_port(struct ofport *ofport_,
2354              const struct ofproto_port_stp_settings *s)
2355 {
2356     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2357     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2358     struct stp_port *sp = ofport->stp_port;
2359
2360     if (!s || !s->enable) {
2361         if (sp) {
2362             ofport->stp_port = NULL;
2363             stp_port_disable(sp);
2364             update_stp_port_state(ofport);
2365         }
2366         return 0;
2367     } else if (sp && stp_port_no(sp) != s->port_num
2368                && ofport == stp_port_get_aux(sp)) {
2369         /* The port-id changed, so disable the old one if it's not
2370          * already in use by another port. */
2371         stp_port_disable(sp);
2372     }
2373
2374     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2375
2376     /* Set name before enabling the port so that debugging messages can print
2377      * the name. */
2378     stp_port_set_name(sp, netdev_get_name(ofport->up.netdev));
2379     stp_port_enable(sp);
2380
2381     stp_port_set_aux(sp, ofport);
2382     stp_port_set_priority(sp, s->priority);
2383     stp_port_set_path_cost(sp, s->path_cost);
2384
2385     update_stp_port_state(ofport);
2386
2387     return 0;
2388 }
2389
2390 static int
2391 get_stp_port_status(struct ofport *ofport_,
2392                     struct ofproto_port_stp_status *s)
2393 {
2394     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2395     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2396     struct stp_port *sp = ofport->stp_port;
2397
2398     if (!ofproto->stp || !sp) {
2399         s->enabled = false;
2400         return 0;
2401     }
2402
2403     s->enabled = true;
2404     s->port_id = stp_port_get_id(sp);
2405     s->state = stp_port_get_state(sp);
2406     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2407     s->role = stp_port_get_role(sp);
2408
2409     return 0;
2410 }
2411
2412 static int
2413 get_stp_port_stats(struct ofport *ofport_,
2414                    struct ofproto_port_stp_stats *s)
2415 {
2416     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2417     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2418     struct stp_port *sp = ofport->stp_port;
2419
2420     if (!ofproto->stp || !sp) {
2421         s->enabled = false;
2422         return 0;
2423     }
2424
2425     s->enabled = true;
2426     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2427
2428     return 0;
2429 }
2430
2431 static void
2432 stp_run(struct ofproto_dpif *ofproto)
2433 {
2434     if (ofproto->stp) {
2435         long long int now = time_msec();
2436         long long int elapsed = now - ofproto->stp_last_tick;
2437         struct stp_port *sp;
2438
2439         if (elapsed > 0) {
2440             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2441             ofproto->stp_last_tick = now;
2442         }
2443         while (stp_get_changed_port(ofproto->stp, &sp)) {
2444             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2445
2446             if (ofport) {
2447                 update_stp_port_state(ofport);
2448             }
2449         }
2450
2451         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2452             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2453             mac_learning_flush(ofproto->ml);
2454             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2455             mcast_snooping_mdb_flush(ofproto->ms);
2456         }
2457     }
2458 }
2459
2460 static void
2461 stp_wait(struct ofproto_dpif *ofproto)
2462 {
2463     if (ofproto->stp) {
2464         poll_timer_wait(1000);
2465     }
2466 }
2467
2468 /* Configures RSTP on 'ofport_' using the settings defined in 's'.  The
2469  * caller is responsible for assigning RSTP port numbers and ensuring
2470  * there are no duplicates. */
2471 static void
2472 set_rstp_port(struct ofport *ofport_,
2473               const struct ofproto_port_rstp_settings *s)
2474 {
2475     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2476     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2477     struct rstp_port *rp = ofport->rstp_port;
2478
2479     if (!s || !s->enable) {
2480         if (rp) {
2481             rstp_port_unref(rp);
2482             ofport->rstp_port = NULL;
2483             update_rstp_port_state(ofport);
2484         }
2485         return;
2486     }
2487
2488     /* Check if need to add a new port. */
2489     if (!rp) {
2490         rp = ofport->rstp_port = rstp_add_port(ofproto->rstp);
2491     }
2492
2493     rstp_port_set(rp, s->port_num, s->priority, s->path_cost,
2494                   s->admin_edge_port, s->auto_edge,
2495                   s->admin_p2p_mac_state, s->admin_port_state, s->mcheck,
2496                   ofport);
2497     update_rstp_port_state(ofport);
2498     /* Synchronize operational status. */
2499     rstp_port_set_mac_operational(rp, ofport->may_enable);
2500 }
2501
2502 static void
2503 get_rstp_port_status(struct ofport *ofport_,
2504         struct ofproto_port_rstp_status *s)
2505 {
2506     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2507     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2508     struct rstp_port *rp = ofport->rstp_port;
2509
2510     if (!ofproto->rstp || !rp) {
2511         s->enabled = false;
2512         return;
2513     }
2514
2515     s->enabled = true;
2516     rstp_port_get_status(rp, &s->port_id, &s->state, &s->role,
2517                          &s->designated_bridge_id, &s->designated_port_id,
2518                          &s->designated_path_cost, &s->tx_count,
2519                          &s->rx_count, &s->error_count, &s->uptime);
2520 }
2521
2522 \f
2523 static int
2524 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2525            size_t n_qdscp)
2526 {
2527     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2528     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2529
2530     if (ofport->n_qdscp != n_qdscp
2531         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2532                               n_qdscp * sizeof *qdscp))) {
2533         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2534         free(ofport->qdscp);
2535         ofport->qdscp = n_qdscp
2536             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2537             : NULL;
2538         ofport->n_qdscp = n_qdscp;
2539     }
2540
2541     return 0;
2542 }
2543 \f
2544 /* Bundles. */
2545
2546 /* Expires all MAC learning entries associated with 'bundle' and forces its
2547  * ofproto to revalidate every flow.
2548  *
2549  * Normally MAC learning entries are removed only from the ofproto associated
2550  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2551  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2552  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2553  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2554  * with the host from which it migrated. */
2555 static void
2556 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2557 {
2558     struct ofproto_dpif *ofproto = bundle->ofproto;
2559     struct mac_learning *ml = ofproto->ml;
2560     struct mac_entry *mac, *next_mac;
2561
2562     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2563     ovs_rwlock_wrlock(&ml->rwlock);
2564     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2565         if (mac->port.p == bundle) {
2566             if (all_ofprotos) {
2567                 struct ofproto_dpif *o;
2568
2569                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2570                     if (o != ofproto) {
2571                         struct mac_entry *e;
2572
2573                         ovs_rwlock_wrlock(&o->ml->rwlock);
2574                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2575                         if (e) {
2576                             mac_learning_expire(o->ml, e);
2577                         }
2578                         ovs_rwlock_unlock(&o->ml->rwlock);
2579                     }
2580                 }
2581             }
2582
2583             mac_learning_expire(ml, mac);
2584         }
2585     }
2586     ovs_rwlock_unlock(&ml->rwlock);
2587 }
2588
2589 static void
2590 bundle_move(struct ofbundle *old, struct ofbundle *new)
2591 {
2592     struct ofproto_dpif *ofproto = old->ofproto;
2593     struct mac_learning *ml = ofproto->ml;
2594     struct mac_entry *mac, *next_mac;
2595
2596     ovs_assert(new->ofproto == old->ofproto);
2597
2598     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2599     ovs_rwlock_wrlock(&ml->rwlock);
2600     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2601         if (mac->port.p == old) {
2602             mac->port.p = new;
2603         }
2604     }
2605     ovs_rwlock_unlock(&ml->rwlock);
2606 }
2607
2608 static struct ofbundle *
2609 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2610 {
2611     struct ofbundle *bundle;
2612
2613     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2614                              &ofproto->bundles) {
2615         if (bundle->aux == aux) {
2616             return bundle;
2617         }
2618     }
2619     return NULL;
2620 }
2621
2622 static void
2623 bundle_update(struct ofbundle *bundle)
2624 {
2625     struct ofport_dpif *port;
2626
2627     bundle->floodable = true;
2628     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2629         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2630             || port->is_layer3
2631             || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2632             || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2633             bundle->floodable = false;
2634             break;
2635         }
2636     }
2637 }
2638
2639 static void
2640 bundle_del_port(struct ofport_dpif *port)
2641 {
2642     struct ofbundle *bundle = port->bundle;
2643
2644     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2645
2646     list_remove(&port->bundle_node);
2647     port->bundle = NULL;
2648
2649     if (bundle->lacp) {
2650         lacp_slave_unregister(bundle->lacp, port);
2651     }
2652     if (bundle->bond) {
2653         bond_slave_unregister(bundle->bond, port);
2654     }
2655
2656     bundle_update(bundle);
2657 }
2658
2659 static bool
2660 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2661                 struct lacp_slave_settings *lacp)
2662 {
2663     struct ofport_dpif *port;
2664
2665     port = get_ofp_port(bundle->ofproto, ofp_port);
2666     if (!port) {
2667         return false;
2668     }
2669
2670     if (port->bundle != bundle) {
2671         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2672         if (port->bundle) {
2673             bundle_remove(&port->up);
2674         }
2675
2676         port->bundle = bundle;
2677         list_push_back(&bundle->ports, &port->bundle_node);
2678         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2679             || port->is_layer3
2680             || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2681             || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2682             bundle->floodable = false;
2683         }
2684     }
2685     if (lacp) {
2686         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2687         lacp_slave_register(bundle->lacp, port, lacp);
2688     }
2689
2690     return true;
2691 }
2692
2693 static void
2694 bundle_destroy(struct ofbundle *bundle)
2695 {
2696     struct ofproto_dpif *ofproto;
2697     struct ofport_dpif *port, *next_port;
2698
2699     if (!bundle) {
2700         return;
2701     }
2702
2703     ofproto = bundle->ofproto;
2704     mbridge_unregister_bundle(ofproto->mbridge, bundle);
2705
2706     xlate_txn_start();
2707     xlate_bundle_remove(bundle);
2708     xlate_txn_commit();
2709
2710     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2711         bundle_del_port(port);
2712     }
2713
2714     bundle_flush_macs(bundle, true);
2715     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2716     free(bundle->name);
2717     free(bundle->trunks);
2718     lacp_unref(bundle->lacp);
2719     bond_unref(bundle->bond);
2720     free(bundle);
2721 }
2722
2723 static int
2724 bundle_set(struct ofproto *ofproto_, void *aux,
2725            const struct ofproto_bundle_settings *s)
2726 {
2727     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2728     bool need_flush = false;
2729     struct ofport_dpif *port;
2730     struct ofbundle *bundle;
2731     unsigned long *trunks;
2732     int vlan;
2733     size_t i;
2734     bool ok;
2735
2736     if (!s) {
2737         bundle_destroy(bundle_lookup(ofproto, aux));
2738         return 0;
2739     }
2740
2741     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2742     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2743
2744     bundle = bundle_lookup(ofproto, aux);
2745     if (!bundle) {
2746         bundle = xmalloc(sizeof *bundle);
2747
2748         bundle->ofproto = ofproto;
2749         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2750                     hash_pointer(aux, 0));
2751         bundle->aux = aux;
2752         bundle->name = NULL;
2753
2754         list_init(&bundle->ports);
2755         bundle->vlan_mode = PORT_VLAN_TRUNK;
2756         bundle->vlan = -1;
2757         bundle->trunks = NULL;
2758         bundle->use_priority_tags = s->use_priority_tags;
2759         bundle->lacp = NULL;
2760         bundle->bond = NULL;
2761
2762         bundle->floodable = true;
2763         mbridge_register_bundle(ofproto->mbridge, bundle);
2764     }
2765
2766     if (!bundle->name || strcmp(s->name, bundle->name)) {
2767         free(bundle->name);
2768         bundle->name = xstrdup(s->name);
2769     }
2770
2771     /* LACP. */
2772     if (s->lacp) {
2773         ofproto->lacp_enabled = true;
2774         if (!bundle->lacp) {
2775             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2776             bundle->lacp = lacp_create();
2777         }
2778         lacp_configure(bundle->lacp, s->lacp);
2779     } else {
2780         lacp_unref(bundle->lacp);
2781         bundle->lacp = NULL;
2782     }
2783
2784     /* Update set of ports. */
2785     ok = true;
2786     for (i = 0; i < s->n_slaves; i++) {
2787         if (!bundle_add_port(bundle, s->slaves[i],
2788                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2789             ok = false;
2790         }
2791     }
2792     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2793         struct ofport_dpif *next_port;
2794
2795         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2796             for (i = 0; i < s->n_slaves; i++) {
2797                 if (s->slaves[i] == port->up.ofp_port) {
2798                     goto found;
2799                 }
2800             }
2801
2802             bundle_del_port(port);
2803         found: ;
2804         }
2805     }
2806     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2807
2808     if (list_is_empty(&bundle->ports)) {
2809         bundle_destroy(bundle);
2810         return EINVAL;
2811     }
2812
2813     /* Set VLAN tagging mode */
2814     if (s->vlan_mode != bundle->vlan_mode
2815         || s->use_priority_tags != bundle->use_priority_tags) {
2816         bundle->vlan_mode = s->vlan_mode;
2817         bundle->use_priority_tags = s->use_priority_tags;
2818         need_flush = true;
2819     }
2820
2821     /* Set VLAN tag. */
2822     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2823             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2824             : 0);
2825     if (vlan != bundle->vlan) {
2826         bundle->vlan = vlan;
2827         need_flush = true;
2828     }
2829
2830     /* Get trunked VLANs. */
2831     switch (s->vlan_mode) {
2832     case PORT_VLAN_ACCESS:
2833         trunks = NULL;
2834         break;
2835
2836     case PORT_VLAN_TRUNK:
2837         trunks = CONST_CAST(unsigned long *, s->trunks);
2838         break;
2839
2840     case PORT_VLAN_NATIVE_UNTAGGED:
2841     case PORT_VLAN_NATIVE_TAGGED:
2842         if (vlan != 0 && (!s->trunks
2843                           || !bitmap_is_set(s->trunks, vlan)
2844                           || bitmap_is_set(s->trunks, 0))) {
2845             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2846             if (s->trunks) {
2847                 trunks = bitmap_clone(s->trunks, 4096);
2848             } else {
2849                 trunks = bitmap_allocate1(4096);
2850             }
2851             bitmap_set1(trunks, vlan);
2852             bitmap_set0(trunks, 0);
2853         } else {
2854             trunks = CONST_CAST(unsigned long *, s->trunks);
2855         }
2856         break;
2857
2858     default:
2859         OVS_NOT_REACHED();
2860     }
2861     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2862         free(bundle->trunks);
2863         if (trunks == s->trunks) {
2864             bundle->trunks = vlan_bitmap_clone(trunks);
2865         } else {
2866             bundle->trunks = trunks;
2867             trunks = NULL;
2868         }
2869         need_flush = true;
2870     }
2871     if (trunks != s->trunks) {
2872         free(trunks);
2873     }
2874
2875     /* Bonding. */
2876     if (!list_is_short(&bundle->ports)) {
2877         bundle->ofproto->has_bonded_bundles = true;
2878         if (bundle->bond) {
2879             if (bond_reconfigure(bundle->bond, s->bond)) {
2880                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2881             }
2882         } else {
2883             bundle->bond = bond_create(s->bond, ofproto);
2884             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2885         }
2886
2887         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2888             bond_slave_register(bundle->bond, port,
2889                                 port->up.ofp_port, port->up.netdev);
2890         }
2891     } else {
2892         bond_unref(bundle->bond);
2893         bundle->bond = NULL;
2894     }
2895
2896     /* If we changed something that would affect MAC learning, un-learn
2897      * everything on this port and force flow revalidation. */
2898     if (need_flush) {
2899         bundle_flush_macs(bundle, false);
2900     }
2901
2902     return 0;
2903 }
2904
2905 static void
2906 bundle_remove(struct ofport *port_)
2907 {
2908     struct ofport_dpif *port = ofport_dpif_cast(port_);
2909     struct ofbundle *bundle = port->bundle;
2910
2911     if (bundle) {
2912         bundle_del_port(port);
2913         if (list_is_empty(&bundle->ports)) {
2914             bundle_destroy(bundle);
2915         } else if (list_is_short(&bundle->ports)) {
2916             bond_unref(bundle->bond);
2917             bundle->bond = NULL;
2918         }
2919     }
2920 }
2921
2922 static void
2923 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2924 {
2925     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2926     struct ofport_dpif *port = port_;
2927     uint8_t ea[ETH_ADDR_LEN];
2928     int error;
2929
2930     error = netdev_get_etheraddr(port->up.netdev, ea);
2931     if (!error) {
2932         struct ofpbuf packet;
2933         void *packet_pdu;
2934
2935         ofpbuf_init(&packet, 0);
2936         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2937                                  pdu_size);
2938         memcpy(packet_pdu, pdu, pdu_size);
2939
2940         ofproto_dpif_send_packet(port, &packet);
2941         ofpbuf_uninit(&packet);
2942     } else {
2943         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2944                     "%s (%s)", port->bundle->name,
2945                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2946     }
2947 }
2948
2949 static void
2950 bundle_send_learning_packets(struct ofbundle *bundle)
2951 {
2952     struct ofproto_dpif *ofproto = bundle->ofproto;
2953     struct ofpbuf *learning_packet;
2954     int error, n_packets, n_errors;
2955     struct mac_entry *e;
2956     struct ovs_list packets;
2957
2958     list_init(&packets);
2959     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2960     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2961         if (e->port.p != bundle) {
2962             void *port_void;
2963
2964             learning_packet = bond_compose_learning_packet(bundle->bond,
2965                                                            e->mac, e->vlan,
2966                                                            &port_void);
2967             /* Temporarily use 'frame' as a private pointer (see below). */
2968             ovs_assert(learning_packet->frame == ofpbuf_data(learning_packet));
2969             learning_packet->frame = port_void;
2970             list_push_back(&packets, &learning_packet->list_node);
2971         }
2972     }
2973     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2974
2975     error = n_packets = n_errors = 0;
2976     LIST_FOR_EACH (learning_packet, list_node, &packets) {
2977         int ret;
2978         void *port_void = learning_packet->frame;
2979
2980         /* Restore 'frame'. */
2981         learning_packet->frame = ofpbuf_data(learning_packet);
2982         ret = ofproto_dpif_send_packet(port_void, learning_packet);
2983         if (ret) {
2984             error = ret;
2985             n_errors++;
2986         }
2987         n_packets++;
2988     }
2989     ofpbuf_list_delete(&packets);
2990
2991     if (n_errors) {
2992         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2993         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2994                      "packets, last error was: %s",
2995                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2996     } else {
2997         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2998                  bundle->name, n_packets);
2999     }
3000 }
3001
3002 static void
3003 bundle_run(struct ofbundle *bundle)
3004 {
3005     if (bundle->lacp) {
3006         lacp_run(bundle->lacp, send_pdu_cb);
3007     }
3008     if (bundle->bond) {
3009         struct ofport_dpif *port;
3010
3011         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
3012             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
3013         }
3014
3015         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
3016             bundle->ofproto->backer->need_revalidate = REV_BOND;
3017         }
3018
3019         if (bond_should_send_learning_packets(bundle->bond)) {
3020             bundle_send_learning_packets(bundle);
3021         }
3022     }
3023 }
3024
3025 static void
3026 bundle_wait(struct ofbundle *bundle)
3027 {
3028     if (bundle->lacp) {
3029         lacp_wait(bundle->lacp);
3030     }
3031     if (bundle->bond) {
3032         bond_wait(bundle->bond);
3033     }
3034 }
3035 \f
3036 /* Mirrors. */
3037
3038 static int
3039 mirror_set__(struct ofproto *ofproto_, void *aux,
3040              const struct ofproto_mirror_settings *s)
3041 {
3042     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3043     struct ofbundle **srcs, **dsts;
3044     int error;
3045     size_t i;
3046
3047     if (!s) {
3048         mirror_destroy(ofproto->mbridge, aux);
3049         return 0;
3050     }
3051
3052     srcs = xmalloc(s->n_srcs * sizeof *srcs);
3053     dsts = xmalloc(s->n_dsts * sizeof *dsts);
3054
3055     for (i = 0; i < s->n_srcs; i++) {
3056         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
3057     }
3058
3059     for (i = 0; i < s->n_dsts; i++) {
3060         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
3061     }
3062
3063     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
3064                        s->n_dsts, s->src_vlans,
3065                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
3066     free(srcs);
3067     free(dsts);
3068     return error;
3069 }
3070
3071 static int
3072 mirror_get_stats__(struct ofproto *ofproto, void *aux,
3073                    uint64_t *packets, uint64_t *bytes)
3074 {
3075     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
3076                             bytes);
3077 }
3078
3079 static int
3080 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3081 {
3082     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3083     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3084     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3085         mac_learning_flush(ofproto->ml);
3086     }
3087     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3088     return 0;
3089 }
3090
3091 static bool
3092 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3093 {
3094     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3095     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3096     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
3097 }
3098
3099 static void
3100 forward_bpdu_changed(struct ofproto *ofproto_)
3101 {
3102     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3103     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3104 }
3105
3106 static void
3107 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3108                      size_t max_entries)
3109 {
3110     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3111     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3112     mac_learning_set_idle_time(ofproto->ml, idle_time);
3113     mac_learning_set_max_entries(ofproto->ml, max_entries);
3114     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3115 }
3116
3117 /* Configures multicast snooping on 'ofport' using the settings
3118  * defined in 's'. */
3119 static int
3120 set_mcast_snooping(struct ofproto *ofproto_,
3121                    const struct ofproto_mcast_snooping_settings *s)
3122 {
3123     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3124
3125     /* Only revalidate flows if the configuration changed. */
3126     if (!s != !ofproto->ms) {
3127         ofproto->backer->need_revalidate = REV_RECONFIGURE;
3128     }
3129
3130     if (s) {
3131         if (!ofproto->ms) {
3132             ofproto->ms = mcast_snooping_create();
3133         }
3134
3135         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3136         mcast_snooping_set_idle_time(ofproto->ms, s->idle_time);
3137         mcast_snooping_set_max_entries(ofproto->ms, s->max_entries);
3138         if (mcast_snooping_set_flood_unreg(ofproto->ms, s->flood_unreg)) {
3139             ofproto->backer->need_revalidate = REV_RECONFIGURE;
3140         }
3141         ovs_rwlock_unlock(&ofproto->ms->rwlock);
3142     } else {
3143         mcast_snooping_unref(ofproto->ms);
3144         ofproto->ms = NULL;
3145     }
3146
3147     return 0;
3148 }
3149
3150 /* Configures multicast snooping port's flood setting on 'ofproto'. */
3151 static int
3152 set_mcast_snooping_port(struct ofproto *ofproto_, void *aux, bool flood)
3153 {
3154     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3155     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3156
3157     if (ofproto->ms) {
3158         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3159         mcast_snooping_set_port_flood(ofproto->ms, bundle->vlan, bundle,
3160                                       flood);
3161         ovs_rwlock_unlock(&ofproto->ms->rwlock);
3162     }
3163     return 0;
3164 }
3165
3166 \f
3167 /* Ports. */
3168
3169 static struct ofport_dpif *
3170 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
3171 {
3172     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3173     return ofport ? ofport_dpif_cast(ofport) : NULL;
3174 }
3175
3176 static void
3177 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3178                             struct ofproto_port *ofproto_port,
3179                             struct dpif_port *dpif_port)
3180 {
3181     ofproto_port->name = dpif_port->name;
3182     ofproto_port->type = dpif_port->type;
3183     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3184 }
3185
3186 static void
3187 ofport_update_peer(struct ofport_dpif *ofport)
3188 {
3189     const struct ofproto_dpif *ofproto;
3190     struct dpif_backer *backer;
3191     char *peer_name;
3192
3193     if (!netdev_vport_is_patch(ofport->up.netdev)) {
3194         return;
3195     }
3196
3197     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
3198     backer->need_revalidate = REV_RECONFIGURE;
3199
3200     if (ofport->peer) {
3201         ofport->peer->peer = NULL;
3202         ofport->peer = NULL;
3203     }
3204
3205     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
3206     if (!peer_name) {
3207         return;
3208     }
3209
3210     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3211         struct ofport *peer_ofport;
3212         struct ofport_dpif *peer;
3213         char *peer_peer;
3214
3215         if (ofproto->backer != backer) {
3216             continue;
3217         }
3218
3219         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
3220         if (!peer_ofport) {
3221             continue;
3222         }
3223
3224         peer = ofport_dpif_cast(peer_ofport);
3225         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
3226         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
3227                                  peer_peer)) {
3228             ofport->peer = peer;
3229             ofport->peer->peer = ofport;
3230         }
3231         free(peer_peer);
3232
3233         break;
3234     }
3235     free(peer_name);
3236 }
3237
3238 static void
3239 port_run(struct ofport_dpif *ofport)
3240 {
3241     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3242     bool carrier_changed = carrier_seq != ofport->carrier_seq;
3243     bool enable = netdev_get_carrier(ofport->up.netdev);
3244     bool cfm_enable = false;
3245     bool bfd_enable = false;
3246
3247     ofport->carrier_seq = carrier_seq;
3248
3249     if (ofport->cfm) {
3250         int cfm_opup = cfm_get_opup(ofport->cfm);
3251
3252         cfm_enable = !cfm_get_fault(ofport->cfm);
3253
3254         if (cfm_opup >= 0) {
3255             cfm_enable = cfm_enable && cfm_opup;
3256         }
3257     }
3258
3259     if (ofport->bfd) {
3260         bfd_enable = bfd_forwarding(ofport->bfd);
3261     }
3262
3263     if (ofport->bfd || ofport->cfm) {
3264         enable = enable && (cfm_enable || bfd_enable);
3265     }
3266
3267     if (ofport->bundle) {
3268         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3269         if (carrier_changed) {
3270             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3271         }
3272     }
3273
3274     if (ofport->may_enable != enable) {
3275         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3276
3277         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3278
3279         if (ofport->rstp_port) {
3280             rstp_port_set_mac_operational(ofport->rstp_port, enable);
3281         }
3282     }
3283
3284     ofport->may_enable = enable;
3285 }
3286
3287 static int
3288 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3289                    struct ofproto_port *ofproto_port)
3290 {
3291     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3292     struct dpif_port dpif_port;
3293     int error;
3294
3295     if (sset_contains(&ofproto->ghost_ports, devname)) {
3296         const char *type = netdev_get_type_from_name(devname);
3297
3298         /* We may be called before ofproto->up.port_by_name is populated with
3299          * the appropriate ofport.  For this reason, we must get the name and
3300          * type from the netdev layer directly. */
3301         if (type) {
3302             const struct ofport *ofport;
3303
3304             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3305             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3306             ofproto_port->name = xstrdup(devname);
3307             ofproto_port->type = xstrdup(type);
3308             return 0;
3309         }
3310         return ENODEV;
3311     }
3312
3313     if (!sset_contains(&ofproto->ports, devname)) {
3314         return ENODEV;
3315     }
3316     error = dpif_port_query_by_name(ofproto->backer->dpif,
3317                                     devname, &dpif_port);
3318     if (!error) {
3319         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3320     }
3321     return error;
3322 }
3323
3324 static int
3325 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3326 {
3327     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3328     const char *devname = netdev_get_name(netdev);
3329     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3330     const char *dp_port_name;
3331
3332     if (netdev_vport_is_patch(netdev)) {
3333         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3334         return 0;
3335     }
3336
3337     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3338     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3339         odp_port_t port_no = ODPP_NONE;
3340         int error;
3341
3342         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3343         if (error) {
3344             return error;
3345         }
3346         if (netdev_get_tunnel_config(netdev)) {
3347             simap_put(&ofproto->backer->tnl_backers,
3348                       dp_port_name, odp_to_u32(port_no));
3349         }
3350     }
3351
3352     if (netdev_get_tunnel_config(netdev)) {
3353         sset_add(&ofproto->ghost_ports, devname);
3354     } else {
3355         sset_add(&ofproto->ports, devname);
3356     }
3357     return 0;
3358 }
3359
3360 static int
3361 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3362 {
3363     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3364     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3365     int error = 0;
3366
3367     if (!ofport) {
3368         return 0;
3369     }
3370
3371     sset_find_and_delete(&ofproto->ghost_ports,
3372                          netdev_get_name(ofport->up.netdev));
3373     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3374     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3375         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3376         if (!error) {
3377             /* The caller is going to close ofport->up.netdev.  If this is a
3378              * bonded port, then the bond is using that netdev, so remove it
3379              * from the bond.  The client will need to reconfigure everything
3380              * after deleting ports, so then the slave will get re-added. */
3381             bundle_remove(&ofport->up);
3382         }
3383     }
3384     return error;
3385 }
3386
3387 static int
3388 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3389 {
3390     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3391     int error;
3392
3393     error = netdev_get_stats(ofport->up.netdev, stats);
3394
3395     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3396         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3397
3398         ovs_mutex_lock(&ofproto->stats_mutex);
3399         /* ofproto->stats.tx_packets represents packets that we created
3400          * internally and sent to some port (e.g. packets sent with
3401          * ofproto_dpif_send_packet()).  Account for them as if they had
3402          * come from OFPP_LOCAL and got forwarded. */
3403
3404         if (stats->rx_packets != UINT64_MAX) {
3405             stats->rx_packets += ofproto->stats.tx_packets;
3406         }
3407
3408         if (stats->rx_bytes != UINT64_MAX) {
3409             stats->rx_bytes += ofproto->stats.tx_bytes;
3410         }
3411
3412         /* ofproto->stats.rx_packets represents packets that were received on
3413          * some port and we processed internally and dropped (e.g. STP).
3414          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3415
3416         if (stats->tx_packets != UINT64_MAX) {
3417             stats->tx_packets += ofproto->stats.rx_packets;
3418         }
3419
3420         if (stats->tx_bytes != UINT64_MAX) {
3421             stats->tx_bytes += ofproto->stats.rx_bytes;
3422         }
3423         ovs_mutex_unlock(&ofproto->stats_mutex);
3424     }
3425
3426     return error;
3427 }
3428
3429 static int
3430 port_get_lacp_stats(const struct ofport *ofport_, struct lacp_slave_stats *stats)
3431 {
3432     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3433     if (ofport->bundle && ofport->bundle->lacp) {
3434         if (lacp_get_slave_stats(ofport->bundle->lacp, ofport, stats)) {
3435             return 0;
3436         }
3437     }
3438     return -1;
3439 }
3440
3441 struct port_dump_state {
3442     uint32_t bucket;
3443     uint32_t offset;
3444     bool ghost;
3445
3446     struct ofproto_port port;
3447     bool has_port;
3448 };
3449
3450 static int
3451 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3452 {
3453     *statep = xzalloc(sizeof(struct port_dump_state));
3454     return 0;
3455 }
3456
3457 static int
3458 port_dump_next(const struct ofproto *ofproto_, void *state_,
3459                struct ofproto_port *port)
3460 {
3461     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3462     struct port_dump_state *state = state_;
3463     const struct sset *sset;
3464     struct sset_node *node;
3465
3466     if (state->has_port) {
3467         ofproto_port_destroy(&state->port);
3468         state->has_port = false;
3469     }
3470     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3471     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3472         int error;
3473
3474         error = port_query_by_name(ofproto_, node->name, &state->port);
3475         if (!error) {
3476             *port = state->port;
3477             state->has_port = true;
3478             return 0;
3479         } else if (error != ENODEV) {
3480             return error;
3481         }
3482     }
3483
3484     if (!state->ghost) {
3485         state->ghost = true;
3486         state->bucket = 0;
3487         state->offset = 0;
3488         return port_dump_next(ofproto_, state_, port);
3489     }
3490
3491     return EOF;
3492 }
3493
3494 static int
3495 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3496 {
3497     struct port_dump_state *state = state_;
3498
3499     if (state->has_port) {
3500         ofproto_port_destroy(&state->port);
3501     }
3502     free(state);
3503     return 0;
3504 }
3505
3506 static int
3507 port_poll(const struct ofproto *ofproto_, char **devnamep)
3508 {
3509     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3510
3511     if (ofproto->port_poll_errno) {
3512         int error = ofproto->port_poll_errno;
3513         ofproto->port_poll_errno = 0;
3514         return error;
3515     }
3516
3517     if (sset_is_empty(&ofproto->port_poll_set)) {
3518         return EAGAIN;
3519     }
3520
3521     *devnamep = sset_pop(&ofproto->port_poll_set);
3522     return 0;
3523 }
3524
3525 static void
3526 port_poll_wait(const struct ofproto *ofproto_)
3527 {
3528     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3529     dpif_port_poll_wait(ofproto->backer->dpif);
3530 }
3531
3532 static int
3533 port_is_lacp_current(const struct ofport *ofport_)
3534 {
3535     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3536     return (ofport->bundle && ofport->bundle->lacp
3537             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3538             : -1);
3539 }
3540 \f
3541 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3542  * then delete it entirely. */
3543 static void
3544 rule_expire(struct rule_dpif *rule)
3545     OVS_REQUIRES(ofproto_mutex)
3546 {
3547     uint16_t hard_timeout, idle_timeout;
3548     long long int now = time_msec();
3549     int reason = -1;
3550
3551     hard_timeout = rule->up.hard_timeout;
3552     idle_timeout = rule->up.idle_timeout;
3553
3554     /* Has 'rule' expired? */
3555     if (hard_timeout) {
3556         long long int modified;
3557
3558         ovs_mutex_lock(&rule->up.mutex);
3559         modified = rule->up.modified;
3560         ovs_mutex_unlock(&rule->up.mutex);
3561
3562         if (now > modified + hard_timeout * 1000) {
3563             reason = OFPRR_HARD_TIMEOUT;
3564         }
3565     }
3566
3567     if (reason < 0 && idle_timeout) {
3568         long long int used;
3569
3570         ovs_mutex_lock(&rule->stats_mutex);
3571         used = rule->stats.used;
3572         ovs_mutex_unlock(&rule->stats_mutex);
3573
3574         if (now > used + idle_timeout * 1000) {
3575             reason = OFPRR_IDLE_TIMEOUT;
3576         }
3577     }
3578
3579     if (reason >= 0) {
3580         COVERAGE_INC(ofproto_dpif_expired);
3581         ofproto_rule_expire(&rule->up, reason);
3582     }
3583 }
3584
3585 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3586  * 'flow' must reflect the data in 'packet'. */
3587 int
3588 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3589                              const struct flow *flow,
3590                              struct rule_dpif *rule,
3591                              const struct ofpact *ofpacts, size_t ofpacts_len,
3592                              struct ofpbuf *packet)
3593 {
3594     struct dpif_flow_stats stats;
3595     struct xlate_out xout;
3596     struct xlate_in xin;
3597     ofp_port_t in_port;
3598     struct dpif_execute execute;
3599     int error;
3600
3601     ovs_assert((rule != NULL) != (ofpacts != NULL));
3602
3603     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3604
3605     if (rule) {
3606         rule_dpif_credit_stats(rule, &stats);
3607     }
3608
3609     xlate_in_init(&xin, ofproto, flow, flow->in_port.ofp_port, rule,
3610                   stats.tcp_flags, packet);
3611     xin.ofpacts = ofpacts;
3612     xin.ofpacts_len = ofpacts_len;
3613     xin.resubmit_stats = &stats;
3614     xlate_actions(&xin, &xout);
3615
3616     execute.actions = ofpbuf_data(xout.odp_actions);
3617     execute.actions_len = ofpbuf_size(xout.odp_actions);
3618
3619     execute.packet = packet;
3620     execute.md = pkt_metadata_from_flow(flow);
3621     execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3622     execute.probe = false;
3623
3624     /* Fix up in_port. */
3625     in_port = flow->in_port.ofp_port;
3626     if (in_port == OFPP_NONE) {
3627         in_port = OFPP_LOCAL;
3628     }
3629     execute.md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3630
3631     error = dpif_execute(ofproto->backer->dpif, &execute);
3632
3633     xlate_out_uninit(&xout);
3634
3635     return error;
3636 }
3637
3638 void
3639 rule_dpif_credit_stats(struct rule_dpif *rule,
3640                        const struct dpif_flow_stats *stats)
3641 {
3642     ovs_mutex_lock(&rule->stats_mutex);
3643     rule->stats.n_packets += stats->n_packets;
3644     rule->stats.n_bytes += stats->n_bytes;
3645     rule->stats.used = MAX(rule->stats.used, stats->used);
3646     ovs_mutex_unlock(&rule->stats_mutex);
3647 }
3648
3649 ovs_be64
3650 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3651     OVS_REQUIRES(rule->up.mutex)
3652 {
3653     return rule->up.flow_cookie;
3654 }
3655
3656 void
3657 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3658                      uint16_t hard_timeout)
3659 {
3660     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3661 }
3662
3663 /* Returns 'rule''s actions.  The returned actions are RCU-protected, and can
3664  * be read until the calling thread quiesces. */
3665 const struct rule_actions *
3666 rule_dpif_get_actions(const struct rule_dpif *rule)
3667 {
3668     return rule_get_actions(&rule->up);
3669 }
3670
3671 /* Sets 'rule''s recirculation id. */
3672 static void
3673 rule_dpif_set_recirc_id(struct rule_dpif *rule, uint32_t id)
3674     OVS_REQUIRES(rule->up.mutex)
3675 {
3676     ovs_assert(!rule->recirc_id);
3677     rule->recirc_id = id;
3678 }
3679
3680 /* Returns 'rule''s recirculation id. */
3681 uint32_t
3682 rule_dpif_get_recirc_id(struct rule_dpif *rule)
3683     OVS_REQUIRES(rule->up.mutex)
3684 {
3685     if (!rule->recirc_id) {
3686         struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3687
3688         rule_dpif_set_recirc_id(rule, ofproto_dpif_alloc_recirc_id(ofproto));
3689     }
3690     return rule->recirc_id;
3691 }
3692
3693 /* Sets 'rule''s recirculation id. */
3694 void
3695 rule_set_recirc_id(struct rule *rule_, uint32_t id)
3696 {
3697     struct rule_dpif *rule = rule_dpif_cast(rule_);
3698
3699     ovs_mutex_lock(&rule->up.mutex);
3700     rule_dpif_set_recirc_id(rule, id);
3701     ovs_mutex_unlock(&rule->up.mutex);
3702 }
3703
3704 /* Lookup 'flow' in table 0 of 'ofproto''s classifier.
3705  * If 'wc' is non-null, sets the fields that were relevant as part of
3706  * the lookup. Returns the table id where a match or miss occurred via
3707  * 'table_id'.  This will be zero unless there was a miss and
3708  * OFPTC11_TABLE_MISS_CONTINUE is in effect for the sequence of tables
3709  * where misses occur, or TBL_INTERNAL if the rule has a non-zero
3710  * recirculation ID, and a match was found in the internal table, or if
3711  * there was no match and one of the special rules (drop_frags_rule,
3712  * miss_rule, or no_packet_in_rule) was returned.
3713  *
3714  * The return value is the found rule, which is valid at least until the next
3715  * RCU quiescent period.  If the rule needs to stay around longer,
3716  * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3717  * on it before this returns. */
3718 struct rule_dpif *
3719 rule_dpif_lookup(struct ofproto_dpif *ofproto, struct flow *flow,
3720                  struct flow_wildcards *wc, bool take_ref,
3721                  const struct dpif_flow_stats *stats, uint8_t *table_id)
3722 {
3723     *table_id = 0;
3724
3725     if (ofproto_dpif_get_enable_recirc(ofproto)) {
3726         /* Always exactly match recirc_id since datapath supports
3727          * recirculation.  */
3728         if (wc) {
3729             wc->masks.recirc_id = UINT32_MAX;
3730         }
3731         *table_id = rule_dpif_lookup_get_init_table_id(flow);
3732     }
3733
3734     return rule_dpif_lookup_from_table(ofproto, flow, wc, take_ref, stats,
3735                                        table_id, flow->in_port.ofp_port, true,
3736                                        true);
3737 }
3738
3739 /* The returned rule (if any) is valid at least until the next RCU quiescent
3740  * period.  If the rule needs to stay around longer, a non-zero 'take_ref'
3741  * must be passed in to cause a reference to be taken on it.
3742  *
3743  * 'flow' is non-const to allow for temporary modifications during the lookup.
3744  * Any changes are restored before returning. */
3745 static struct rule_dpif *
3746 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, uint8_t table_id,
3747                           struct flow *flow, struct flow_wildcards *wc,
3748                           bool take_ref)
3749 {
3750     struct classifier *cls = &ofproto->up.tables[table_id].cls;
3751     const struct cls_rule *cls_rule;
3752     struct rule_dpif *rule;
3753
3754     do {
3755         cls_rule = classifier_lookup(cls, flow, wc);
3756
3757         rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3758
3759         /* Try again if the rule was released before we get the reference. */
3760     } while (rule && take_ref && !rule_dpif_try_ref(rule));
3761
3762     return rule;
3763 }
3764
3765 /* Look up 'flow' in 'ofproto''s classifier starting from table '*table_id'.
3766  * Returns the rule that was found, which may be one of the special rules
3767  * according to packet miss hadling.  If 'may_packet_in' is false, returning of
3768  * the miss_rule (which issues packet ins for the controller) is avoided.
3769  * Updates 'wc', if nonnull, to reflect the fields that were used during the
3770  * lookup.
3771  *
3772  * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3773  * if none is found then the table miss configuration for that table is
3774  * honored, which can result in additional lookups in other OpenFlow tables.
3775  * In this case the function updates '*table_id' to reflect the final OpenFlow
3776  * table that was searched.
3777  *
3778  * If 'honor_table_miss' is false, then only one table lookup occurs, in
3779  * '*table_id'.
3780  *
3781  * The rule is returned in '*rule', which is valid at least until the next
3782  * RCU quiescent period.  If the '*rule' needs to stay around longer,
3783  * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3784  * on it before this returns.
3785  *
3786  * 'in_port' allows the lookup to take place as if the in port had the value
3787  * 'in_port'.  This is needed for resubmit action support.
3788  *
3789  * 'flow' is non-const to allow for temporary modifications during the lookup.
3790  * Any changes are restored before returning. */
3791 struct rule_dpif *
3792 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto, struct flow *flow,
3793                             struct flow_wildcards *wc, bool take_ref,
3794                             const struct dpif_flow_stats *stats,
3795                             uint8_t *table_id, ofp_port_t in_port,
3796                             bool may_packet_in, bool honor_table_miss)
3797 {
3798     ovs_be16 old_tp_src = flow->tp_src, old_tp_dst = flow->tp_dst;
3799     ofp_port_t old_in_port = flow->in_port.ofp_port;
3800     enum ofputil_table_miss miss_config;
3801     struct rule_dpif *rule;
3802     uint8_t next_id;
3803
3804     /* We always unwildcard nw_frag (for IP), so they
3805      * need not be unwildcarded here. */
3806     if (flow->nw_frag & FLOW_NW_FRAG_ANY
3807         && ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3808         if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3809             /* We must pretend that transport ports are unavailable. */
3810             flow->tp_src = htons(0);
3811             flow->tp_dst = htons(0);
3812         } else {
3813             /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM).
3814              * Use the drop_frags_rule (which cannot disappear). */
3815             rule = ofproto->drop_frags_rule;
3816             if (take_ref) {
3817                 rule_dpif_ref(rule);
3818             }
3819             if (stats) {
3820                 struct oftable *tbl = &ofproto->up.tables[*table_id];
3821                 unsigned long orig;
3822
3823                 atomic_add_relaxed(&tbl->n_matched, stats->n_packets, &orig);
3824             }
3825             return rule;
3826         }
3827     }
3828
3829     /* Look up a flow with 'in_port' as the input port.  Then restore the
3830      * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
3831      * have surprising behavior). */
3832     flow->in_port.ofp_port = in_port;
3833
3834     /* Our current implementation depends on n_tables == N_TABLES, and
3835      * TBL_INTERNAL being the last table. */
3836     BUILD_ASSERT_DECL(N_TABLES == TBL_INTERNAL + 1);
3837
3838     miss_config = OFPUTIL_TABLE_MISS_CONTINUE;
3839
3840     for (next_id = *table_id;
3841          next_id < ofproto->up.n_tables;
3842          next_id++, next_id += (next_id == TBL_INTERNAL))
3843     {
3844         *table_id = next_id;
3845         rule = rule_dpif_lookup_in_table(ofproto, next_id, flow, wc, take_ref);
3846         if (stats) {
3847             struct oftable *tbl = &ofproto->up.tables[next_id];
3848             unsigned long orig;
3849
3850             atomic_add_relaxed(rule ? &tbl->n_matched : &tbl->n_missed,
3851                                stats->n_packets, &orig);
3852         }
3853         if (rule) {
3854             goto out;   /* Match. */
3855         }
3856         if (honor_table_miss) {
3857             miss_config = ofproto_table_get_miss_config(&ofproto->up,
3858                                                         *table_id);
3859             if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE) {
3860                 continue;
3861             }
3862         }
3863         break;
3864     }
3865     /* Miss. */
3866     rule = ofproto->no_packet_in_rule;
3867     if (may_packet_in) {
3868         if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE
3869             || miss_config == OFPUTIL_TABLE_MISS_CONTROLLER) {
3870             struct ofport_dpif *port;
3871
3872             port = get_ofp_port(ofproto, old_in_port);
3873             if (!port) {
3874                 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3875                              old_in_port);
3876             } else if (!(port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN)) {
3877                 rule = ofproto->miss_rule;
3878             }
3879         } else if (miss_config == OFPUTIL_TABLE_MISS_DEFAULT &&
3880                    connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3881             rule = ofproto->miss_rule;
3882         }
3883     }
3884     if (take_ref) {
3885         rule_dpif_ref(rule);
3886     }
3887 out:
3888     /* Restore port numbers, as they may have been modified above. */
3889     flow->tp_src = old_tp_src;
3890     flow->tp_dst = old_tp_dst;
3891     /* Restore the old in port. */
3892     flow->in_port.ofp_port = old_in_port;
3893
3894     return rule;
3895 }
3896
3897 static void
3898 complete_operation(struct rule_dpif *rule)
3899     OVS_REQUIRES(ofproto_mutex)
3900 {
3901     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3902
3903     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3904 }
3905
3906 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3907 {
3908     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3909 }
3910
3911 static struct rule *
3912 rule_alloc(void)
3913 {
3914     struct rule_dpif *rule = xmalloc(sizeof *rule);
3915     return &rule->up;
3916 }
3917
3918 static void
3919 rule_dealloc(struct rule *rule_)
3920 {
3921     struct rule_dpif *rule = rule_dpif_cast(rule_);
3922     free(rule);
3923 }
3924
3925 static enum ofperr
3926 rule_construct(struct rule *rule_)
3927     OVS_NO_THREAD_SAFETY_ANALYSIS
3928 {
3929     struct rule_dpif *rule = rule_dpif_cast(rule_);
3930     ovs_mutex_init_adaptive(&rule->stats_mutex);
3931     rule->stats.n_packets = 0;
3932     rule->stats.n_bytes = 0;
3933     rule->stats.used = rule->up.modified;
3934     rule->recirc_id = 0;
3935
3936     return 0;
3937 }
3938
3939 static enum ofperr
3940 rule_insert(struct rule *rule_)
3941     OVS_REQUIRES(ofproto_mutex)
3942 {
3943     struct rule_dpif *rule = rule_dpif_cast(rule_);
3944     complete_operation(rule);
3945     return 0;
3946 }
3947
3948 static void
3949 rule_delete(struct rule *rule_)
3950     OVS_REQUIRES(ofproto_mutex)
3951 {
3952     struct rule_dpif *rule = rule_dpif_cast(rule_);
3953     complete_operation(rule);
3954 }
3955
3956 static void
3957 rule_destruct(struct rule *rule_)
3958 {
3959     struct rule_dpif *rule = rule_dpif_cast(rule_);
3960
3961     ovs_mutex_destroy(&rule->stats_mutex);
3962     if (rule->recirc_id) {
3963         struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3964
3965         ofproto_dpif_free_recirc_id(ofproto, rule->recirc_id);
3966     }
3967 }
3968
3969 static void
3970 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
3971                long long int *used)
3972 {
3973     struct rule_dpif *rule = rule_dpif_cast(rule_);
3974
3975     ovs_mutex_lock(&rule->stats_mutex);
3976     *packets = rule->stats.n_packets;
3977     *bytes = rule->stats.n_bytes;
3978     *used = rule->stats.used;
3979     ovs_mutex_unlock(&rule->stats_mutex);
3980 }
3981
3982 static void
3983 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
3984                   struct ofpbuf *packet)
3985 {
3986     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3987
3988     ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
3989 }
3990
3991 static enum ofperr
3992 rule_execute(struct rule *rule, const struct flow *flow,
3993              struct ofpbuf *packet)
3994 {
3995     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
3996     ofpbuf_delete(packet);
3997     return 0;
3998 }
3999
4000 static void
4001 rule_modify_actions(struct rule *rule_, bool reset_counters)
4002     OVS_REQUIRES(ofproto_mutex)
4003 {
4004     struct rule_dpif *rule = rule_dpif_cast(rule_);
4005
4006     if (reset_counters) {
4007         ovs_mutex_lock(&rule->stats_mutex);
4008         rule->stats.n_packets = 0;
4009         rule->stats.n_bytes = 0;
4010         ovs_mutex_unlock(&rule->stats_mutex);
4011     }
4012
4013     complete_operation(rule);
4014 }
4015
4016 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4017 {
4018     return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4019 }
4020
4021 static struct ofgroup *
4022 group_alloc(void)
4023 {
4024     struct group_dpif *group = xzalloc(sizeof *group);
4025     return &group->up;
4026 }
4027
4028 static void
4029 group_dealloc(struct ofgroup *group_)
4030 {
4031     struct group_dpif *group = group_dpif_cast(group_);
4032     free(group);
4033 }
4034
4035 static void
4036 group_construct_stats(struct group_dpif *group)
4037     OVS_REQUIRES(group->stats_mutex)
4038 {
4039     struct ofputil_bucket *bucket;
4040     const struct ovs_list *buckets;
4041
4042     group->packet_count = 0;
4043     group->byte_count = 0;
4044
4045     group_dpif_get_buckets(group, &buckets);
4046     LIST_FOR_EACH (bucket, list_node, buckets) {
4047         bucket->stats.packet_count = 0;
4048         bucket->stats.byte_count = 0;
4049     }
4050 }
4051
4052 void
4053 group_dpif_credit_stats(struct group_dpif *group,
4054                         struct ofputil_bucket *bucket,
4055                         const struct dpif_flow_stats *stats)
4056 {
4057     ovs_mutex_lock(&group->stats_mutex);
4058     group->packet_count += stats->n_packets;
4059     group->byte_count += stats->n_bytes;
4060     if (bucket) {
4061         bucket->stats.packet_count += stats->n_packets;
4062         bucket->stats.byte_count += stats->n_bytes;
4063     } else { /* Credit to all buckets */
4064         const struct ovs_list *buckets;
4065
4066         group_dpif_get_buckets(group, &buckets);
4067         LIST_FOR_EACH (bucket, list_node, buckets) {
4068             bucket->stats.packet_count += stats->n_packets;
4069             bucket->stats.byte_count += stats->n_bytes;
4070         }
4071     }
4072     ovs_mutex_unlock(&group->stats_mutex);
4073 }
4074
4075 static enum ofperr
4076 group_construct(struct ofgroup *group_)
4077 {
4078     struct group_dpif *group = group_dpif_cast(group_);
4079     const struct ofputil_bucket *bucket;
4080
4081     /* Prevent group chaining because our locking structure makes it hard to
4082      * implement deadlock-free.  (See xlate_group_resource_check().) */
4083     LIST_FOR_EACH (bucket, list_node, &group->up.buckets) {
4084         const struct ofpact *a;
4085
4086         OFPACT_FOR_EACH (a, bucket->ofpacts, bucket->ofpacts_len) {
4087             if (a->type == OFPACT_GROUP) {
4088                 return OFPERR_OFPGMFC_CHAINING_UNSUPPORTED;
4089             }
4090         }
4091     }
4092
4093     ovs_mutex_init_adaptive(&group->stats_mutex);
4094     ovs_mutex_lock(&group->stats_mutex);
4095     group_construct_stats(group);
4096     ovs_mutex_unlock(&group->stats_mutex);
4097     return 0;
4098 }
4099
4100 static void
4101 group_destruct(struct ofgroup *group_)
4102 {
4103     struct group_dpif *group = group_dpif_cast(group_);
4104     ovs_mutex_destroy(&group->stats_mutex);
4105 }
4106
4107 static enum ofperr
4108 group_modify(struct ofgroup *group_)
4109 {
4110     struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
4111
4112     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4113
4114     return 0;
4115 }
4116
4117 static enum ofperr
4118 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4119 {
4120     struct group_dpif *group = group_dpif_cast(group_);
4121     struct ofputil_bucket *bucket;
4122     const struct ovs_list *buckets;
4123     struct bucket_counter *bucket_stats;
4124
4125     ovs_mutex_lock(&group->stats_mutex);
4126     ogs->packet_count = group->packet_count;
4127     ogs->byte_count = group->byte_count;
4128
4129     group_dpif_get_buckets(group, &buckets);
4130     bucket_stats = ogs->bucket_stats;
4131     LIST_FOR_EACH (bucket, list_node, buckets) {
4132         bucket_stats->packet_count = bucket->stats.packet_count;
4133         bucket_stats->byte_count = bucket->stats.byte_count;
4134         bucket_stats++;
4135     }
4136     ovs_mutex_unlock(&group->stats_mutex);
4137
4138     return 0;
4139 }
4140
4141 /* If the group exists, this function increments the groups's reference count.
4142  *
4143  * Make sure to call group_dpif_unref() after no longer needing to maintain
4144  * a reference to the group. */
4145 bool
4146 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4147                   struct group_dpif **group)
4148 {
4149     struct ofgroup *ofgroup;
4150     bool found;
4151
4152     found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4153     *group = found ?  group_dpif_cast(ofgroup) : NULL;
4154
4155     return found;
4156 }
4157
4158 void
4159 group_dpif_get_buckets(const struct group_dpif *group,
4160                        const struct ovs_list **buckets)
4161 {
4162     *buckets = &group->up.buckets;
4163 }
4164
4165 enum ofp11_group_type
4166 group_dpif_get_type(const struct group_dpif *group)
4167 {
4168     return group->up.type;
4169 }
4170 \f
4171 /* Sends 'packet' out 'ofport'.
4172  * May modify 'packet'.
4173  * Returns 0 if successful, otherwise a positive errno value. */
4174 int
4175 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4176 {
4177     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4178     int error;
4179
4180     error = xlate_send_packet(ofport, packet);
4181
4182     ovs_mutex_lock(&ofproto->stats_mutex);
4183     ofproto->stats.tx_packets++;
4184     ofproto->stats.tx_bytes += ofpbuf_size(packet);
4185     ovs_mutex_unlock(&ofproto->stats_mutex);
4186     return error;
4187 }
4188 \f
4189 /* Return the version string of the datapath that backs up
4190  * this 'ofproto'.
4191  */
4192 static const char *
4193 get_datapath_version(const struct ofproto *ofproto_)
4194 {
4195     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4196
4197     return ofproto->backer->dp_version_string;
4198 }
4199
4200 static bool
4201 set_frag_handling(struct ofproto *ofproto_,
4202                   enum ofp_config_flags frag_handling)
4203 {
4204     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4205     if (frag_handling != OFPC_FRAG_REASM) {
4206         ofproto->backer->need_revalidate = REV_RECONFIGURE;
4207         return true;
4208     } else {
4209         return false;
4210     }
4211 }
4212
4213 static enum ofperr
4214 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
4215            const struct flow *flow,
4216            const struct ofpact *ofpacts, size_t ofpacts_len)
4217 {
4218     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4219
4220     ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
4221                                  ofpacts_len, packet);
4222     return 0;
4223 }
4224 \f
4225 /* NetFlow. */
4226
4227 static int
4228 set_netflow(struct ofproto *ofproto_,
4229             const struct netflow_options *netflow_options)
4230 {
4231     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4232
4233     if (netflow_options) {
4234         if (!ofproto->netflow) {
4235             ofproto->netflow = netflow_create();
4236             ofproto->backer->need_revalidate = REV_RECONFIGURE;
4237         }
4238         return netflow_set_options(ofproto->netflow, netflow_options);
4239     } else if (ofproto->netflow) {
4240         ofproto->backer->need_revalidate = REV_RECONFIGURE;
4241         netflow_unref(ofproto->netflow);
4242         ofproto->netflow = NULL;
4243     }
4244
4245     return 0;
4246 }
4247
4248 static void
4249 get_netflow_ids(const struct ofproto *ofproto_,
4250                 uint8_t *engine_type, uint8_t *engine_id)
4251 {
4252     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4253
4254     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
4255 }
4256 \f
4257 static struct ofproto_dpif *
4258 ofproto_dpif_lookup(const char *name)
4259 {
4260     struct ofproto_dpif *ofproto;
4261
4262     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
4263                              hash_string(name, 0), &all_ofproto_dpifs) {
4264         if (!strcmp(ofproto->up.name, name)) {
4265             return ofproto;
4266         }
4267     }
4268     return NULL;
4269 }
4270
4271 static void
4272 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
4273                           const char *argv[], void *aux OVS_UNUSED)
4274 {
4275     struct ofproto_dpif *ofproto;
4276
4277     if (argc > 1) {
4278         ofproto = ofproto_dpif_lookup(argv[1]);
4279         if (!ofproto) {
4280             unixctl_command_reply_error(conn, "no such bridge");
4281             return;
4282         }
4283         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4284         mac_learning_flush(ofproto->ml);
4285         ovs_rwlock_unlock(&ofproto->ml->rwlock);
4286     } else {
4287         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4288             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4289             mac_learning_flush(ofproto->ml);
4290             ovs_rwlock_unlock(&ofproto->ml->rwlock);
4291         }
4292     }
4293
4294     unixctl_command_reply(conn, "table successfully flushed");
4295 }
4296
4297 static void
4298 ofproto_unixctl_mcast_snooping_flush(struct unixctl_conn *conn, int argc,
4299                                      const char *argv[], void *aux OVS_UNUSED)
4300 {
4301     struct ofproto_dpif *ofproto;
4302
4303     if (argc > 1) {
4304         ofproto = ofproto_dpif_lookup(argv[1]);
4305         if (!ofproto) {
4306             unixctl_command_reply_error(conn, "no such bridge");
4307             return;
4308         }
4309
4310         if (!mcast_snooping_enabled(ofproto->ms)) {
4311             unixctl_command_reply_error(conn, "multicast snooping is disabled");
4312             return;
4313         }
4314         mcast_snooping_mdb_flush(ofproto->ms);
4315     } else {
4316         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4317             if (!mcast_snooping_enabled(ofproto->ms)) {
4318                 continue;
4319             }
4320             mcast_snooping_mdb_flush(ofproto->ms);
4321         }
4322     }
4323
4324     unixctl_command_reply(conn, "table successfully flushed");
4325 }
4326
4327 static struct ofport_dpif *
4328 ofbundle_get_a_port(const struct ofbundle *bundle)
4329 {
4330     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
4331                         bundle_node);
4332 }
4333
4334 static void
4335 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4336                          const char *argv[], void *aux OVS_UNUSED)
4337 {
4338     struct ds ds = DS_EMPTY_INITIALIZER;
4339     const struct ofproto_dpif *ofproto;
4340     const struct mac_entry *e;
4341
4342     ofproto = ofproto_dpif_lookup(argv[1]);
4343     if (!ofproto) {
4344         unixctl_command_reply_error(conn, "no such bridge");
4345         return;
4346     }
4347
4348     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
4349     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
4350     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
4351         struct ofbundle *bundle = e->port.p;
4352         char name[OFP_MAX_PORT_NAME_LEN];
4353
4354         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4355                                name, sizeof name);
4356         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
4357                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
4358                       mac_entry_age(ofproto->ml, e));
4359     }
4360     ovs_rwlock_unlock(&ofproto->ml->rwlock);
4361     unixctl_command_reply(conn, ds_cstr(&ds));
4362     ds_destroy(&ds);
4363 }
4364
4365 static void
4366 ofproto_unixctl_mcast_snooping_show(struct unixctl_conn *conn,
4367                                     int argc OVS_UNUSED,
4368                                     const char *argv[],
4369                                     void *aux OVS_UNUSED)
4370 {
4371     struct ds ds = DS_EMPTY_INITIALIZER;
4372     const struct ofproto_dpif *ofproto;
4373     const struct ofbundle *bundle;
4374     const struct mcast_group *grp;
4375     struct mcast_group_bundle *b;
4376     struct mcast_mrouter_bundle *mrouter;
4377
4378     ofproto = ofproto_dpif_lookup(argv[1]);
4379     if (!ofproto) {
4380         unixctl_command_reply_error(conn, "no such bridge");
4381         return;
4382     }
4383
4384     if (!mcast_snooping_enabled(ofproto->ms)) {
4385         unixctl_command_reply_error(conn, "multicast snooping is disabled");
4386         return;
4387     }
4388
4389     ds_put_cstr(&ds, " port  VLAN  GROUP                Age\n");
4390     ovs_rwlock_rdlock(&ofproto->ms->rwlock);
4391     LIST_FOR_EACH (grp, group_node, &ofproto->ms->group_lru) {
4392         LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
4393             char name[OFP_MAX_PORT_NAME_LEN];
4394
4395             bundle = b->port;
4396             ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4397                                    name, sizeof name);
4398             ds_put_format(&ds, "%5s  %4d  "IP_FMT"         %3d\n",
4399                           name, grp->vlan, IP_ARGS(grp->ip4),
4400                           mcast_bundle_age(ofproto->ms, b));
4401         }
4402     }
4403
4404     /* ports connected to multicast routers */
4405     LIST_FOR_EACH(mrouter, mrouter_node, &ofproto->ms->mrouter_lru) {
4406         char name[OFP_MAX_PORT_NAME_LEN];
4407
4408         bundle = mrouter->port;
4409         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4410                                name, sizeof name);
4411             ds_put_format(&ds, "%5s  %4d  querier             %3d\n",
4412                       name, mrouter->vlan,
4413                       mcast_mrouter_age(ofproto->ms, mrouter));
4414     }
4415     ovs_rwlock_unlock(&ofproto->ms->rwlock);
4416     unixctl_command_reply(conn, ds_cstr(&ds));
4417     ds_destroy(&ds);
4418 }
4419
4420 struct trace_ctx {
4421     struct xlate_out xout;
4422     struct xlate_in xin;
4423     const struct flow *key;
4424     struct flow flow;
4425     struct flow_wildcards wc;
4426     struct ds *result;
4427 };
4428
4429 static void
4430 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
4431 {
4432     const struct rule_actions *actions;
4433     ovs_be64 cookie;
4434
4435     ds_put_char_multiple(result, '\t', level);
4436     if (!rule) {
4437         ds_put_cstr(result, "No match\n");
4438         return;
4439     }
4440
4441     ovs_mutex_lock(&rule->up.mutex);
4442     cookie = rule->up.flow_cookie;
4443     ovs_mutex_unlock(&rule->up.mutex);
4444
4445     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
4446                   rule ? rule->up.table_id : 0, ntohll(cookie));
4447     cls_rule_format(&rule->up.cr, result);
4448     ds_put_char(result, '\n');
4449
4450     actions = rule_dpif_get_actions(rule);
4451
4452     ds_put_char_multiple(result, '\t', level);
4453     ds_put_cstr(result, "OpenFlow actions=");
4454     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
4455     ds_put_char(result, '\n');
4456 }
4457
4458 static void
4459 trace_format_flow(struct ds *result, int level, const char *title,
4460                   struct trace_ctx *trace)
4461 {
4462     ds_put_char_multiple(result, '\t', level);
4463     ds_put_format(result, "%s: ", title);
4464     /* Do not report unchanged flows for resubmits. */
4465     if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
4466         || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
4467         ds_put_cstr(result, "unchanged");
4468     } else {
4469         flow_format(result, &trace->xin.flow);
4470         trace->flow = trace->xin.flow;
4471     }
4472     ds_put_char(result, '\n');
4473 }
4474
4475 static void
4476 trace_format_regs(struct ds *result, int level, const char *title,
4477                   struct trace_ctx *trace)
4478 {
4479     size_t i;
4480
4481     ds_put_char_multiple(result, '\t', level);
4482     ds_put_format(result, "%s:", title);
4483     for (i = 0; i < FLOW_N_REGS; i++) {
4484         ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
4485     }
4486     ds_put_char(result, '\n');
4487 }
4488
4489 static void
4490 trace_format_odp(struct ds *result, int level, const char *title,
4491                  struct trace_ctx *trace)
4492 {
4493     struct ofpbuf *odp_actions = trace->xout.odp_actions;
4494
4495     ds_put_char_multiple(result, '\t', level);
4496     ds_put_format(result, "%s: ", title);
4497     format_odp_actions(result, ofpbuf_data(odp_actions),
4498                                ofpbuf_size(odp_actions));
4499     ds_put_char(result, '\n');
4500 }
4501
4502 static void
4503 trace_format_megaflow(struct ds *result, int level, const char *title,
4504                       struct trace_ctx *trace)
4505 {
4506     struct match match;
4507
4508     ds_put_char_multiple(result, '\t', level);
4509     ds_put_format(result, "%s: ", title);
4510     flow_wildcards_or(&trace->wc, &trace->xout.wc, &trace->wc);
4511     match_init(&match, trace->key, &trace->wc);
4512     match_format(&match, result, OFP_DEFAULT_PRIORITY);
4513     ds_put_char(result, '\n');
4514 }
4515
4516 static void trace_report(struct xlate_in *xin, const char *s, int recurse);
4517
4518 static void
4519 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
4520 {
4521     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4522     struct ds *result = trace->result;
4523
4524     if (!recurse) {
4525         if (rule == xin->ofproto->miss_rule) {
4526             trace_report(xin, "No match, flow generates \"packet in\"s.",
4527                          recurse);
4528         } else if (rule == xin->ofproto->no_packet_in_rule) {
4529             trace_report(xin, "No match, packets dropped because "
4530                          "OFPPC_NO_PACKET_IN is set on in_port.", recurse);
4531         } else if (rule == xin->ofproto->drop_frags_rule) {
4532             trace_report(xin, "Packets dropped because they are IP "
4533                          "fragments and the fragment handling mode is "
4534                          "\"drop\".", recurse);
4535         }
4536     }
4537
4538     ds_put_char(result, '\n');
4539     if (recurse) {
4540         trace_format_flow(result, recurse, "Resubmitted flow", trace);
4541         trace_format_regs(result, recurse, "Resubmitted regs", trace);
4542         trace_format_odp(result,  recurse, "Resubmitted  odp", trace);
4543         trace_format_megaflow(result, recurse, "Resubmitted megaflow", trace);
4544     }
4545     trace_format_rule(result, recurse, rule);
4546 }
4547
4548 static void
4549 trace_report(struct xlate_in *xin, const char *s, int recurse)
4550 {
4551     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4552     struct ds *result = trace->result;
4553
4554     ds_put_char_multiple(result, '\t', recurse);
4555     ds_put_cstr(result, s);
4556     ds_put_char(result, '\n');
4557 }
4558
4559 /* Parses the 'argc' elements of 'argv', ignoring argv[0].  The following
4560  * forms are supported:
4561  *
4562  *     - [dpname] odp_flow [-generate | packet]
4563  *     - bridge br_flow [-generate | packet]
4564  *
4565  * On success, initializes '*ofprotop' and 'flow' and returns NULL.  On failure
4566  * returns a nonnull malloced error message. */
4567 static char * OVS_WARN_UNUSED_RESULT
4568 parse_flow_and_packet(int argc, const char *argv[],
4569                       struct ofproto_dpif **ofprotop, struct flow *flow,
4570                       struct ofpbuf **packetp)
4571 {
4572     const struct dpif_backer *backer = NULL;
4573     const char *error = NULL;
4574     char *m_err = NULL;
4575     struct simap port_names = SIMAP_INITIALIZER(&port_names);
4576     struct ofpbuf *packet;
4577     struct ofpbuf odp_key;
4578     struct ofpbuf odp_mask;
4579
4580     ofpbuf_init(&odp_key, 0);
4581     ofpbuf_init(&odp_mask, 0);
4582
4583     /* Handle "-generate" or a hex string as the last argument. */
4584     if (!strcmp(argv[argc - 1], "-generate")) {
4585         packet = ofpbuf_new(0);
4586         argc--;
4587     } else {
4588         error = eth_from_hex(argv[argc - 1], &packet);
4589         if (!error) {
4590             argc--;
4591         } else if (argc == 4) {
4592             /* The 3-argument form must end in "-generate' or a hex string. */
4593             goto exit;
4594         }
4595         error = NULL;
4596     }
4597
4598     /* odp_flow can have its in_port specified as a name instead of port no.
4599      * We do not yet know whether a given flow is a odp_flow or a br_flow.
4600      * But, to know whether a flow is odp_flow through odp_flow_from_string(),
4601      * we need to create a simap of name to port no. */
4602     if (argc == 3) {
4603         const char *dp_type;
4604         if (!strncmp(argv[1], "ovs-", 4)) {
4605             dp_type = argv[1] + 4;
4606         } else {
4607             dp_type = argv[1];
4608         }
4609         backer = shash_find_data(&all_dpif_backers, dp_type);
4610     } else if (argc == 2) {
4611         struct shash_node *node;
4612         if (shash_count(&all_dpif_backers) == 1) {
4613             node = shash_first(&all_dpif_backers);
4614             backer = node->data;
4615         }
4616     } else {
4617         error = "Syntax error";
4618         goto exit;
4619     }
4620     if (backer && backer->dpif) {
4621         struct dpif_port dpif_port;
4622         struct dpif_port_dump port_dump;
4623         DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4624             simap_put(&port_names, dpif_port.name,
4625                       odp_to_u32(dpif_port.port_no));
4626         }
4627     }
4628
4629     /* Parse the flow and determine whether a datapath or
4630      * bridge is specified. If function odp_flow_key_from_string()
4631      * returns 0, the flow is a odp_flow. If function
4632      * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4633     if (!odp_flow_from_string(argv[argc - 1], &port_names,
4634                               &odp_key, &odp_mask)) {
4635         if (!backer) {
4636             error = "Cannot find the datapath";
4637             goto exit;
4638         }
4639
4640         if (odp_flow_key_to_flow(ofpbuf_data(&odp_key), ofpbuf_size(&odp_key),
4641                                  flow) == ODP_FIT_ERROR) {
4642             error = "Failed to parse flow key";
4643             goto exit;
4644         }
4645
4646         *ofprotop = xlate_lookup_ofproto(backer, flow,
4647                                          &flow->in_port.ofp_port);
4648         if (*ofprotop == NULL) {
4649             error = "Invalid datapath flow";
4650             goto exit;
4651         }
4652
4653         vsp_adjust_flow(*ofprotop, flow, NULL);
4654
4655     } else {
4656         char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4657
4658         if (err) {
4659             m_err = xasprintf("Bad flow syntax: %s", err);
4660             free(err);
4661             goto exit;
4662         } else {
4663             if (argc != 3) {
4664                 error = "Must specify bridge name";
4665                 goto exit;
4666             }
4667
4668             *ofprotop = ofproto_dpif_lookup(argv[1]);
4669             if (!*ofprotop) {
4670                 error = "Unknown bridge name";
4671                 goto exit;
4672             }
4673         }
4674     }
4675
4676     /* Generate a packet, if requested. */
4677     if (packet) {
4678         if (!ofpbuf_size(packet)) {
4679             flow_compose(packet, flow);
4680         } else {
4681             struct pkt_metadata md = pkt_metadata_from_flow(flow);
4682
4683             /* Use the metadata from the flow and the packet argument
4684              * to reconstruct the flow. */
4685             flow_extract(packet, &md, flow);
4686         }
4687     }
4688
4689 exit:
4690     if (error && !m_err) {
4691         m_err = xstrdup(error);
4692     }
4693     if (m_err) {
4694         ofpbuf_delete(packet);
4695         packet = NULL;
4696     }
4697     *packetp = packet;
4698     ofpbuf_uninit(&odp_key);
4699     ofpbuf_uninit(&odp_mask);
4700     simap_destroy(&port_names);
4701     return m_err;
4702 }
4703
4704 static void
4705 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4706                       void *aux OVS_UNUSED)
4707 {
4708     struct ofproto_dpif *ofproto;
4709     struct ofpbuf *packet;
4710     char *error;
4711     struct flow flow;
4712
4713     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4714     if (!error) {
4715         struct ds result;
4716
4717         ds_init(&result);
4718         ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4719         unixctl_command_reply(conn, ds_cstr(&result));
4720         ds_destroy(&result);
4721         ofpbuf_delete(packet);
4722     } else {
4723         unixctl_command_reply_error(conn, error);
4724         free(error);
4725     }
4726 }
4727
4728 static void
4729 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4730                               const char *argv[], void *aux OVS_UNUSED)
4731 {
4732     enum ofputil_protocol usable_protocols;
4733     struct ofproto_dpif *ofproto;
4734     bool enforce_consistency;
4735     struct ofpbuf ofpacts;
4736     struct ofpbuf *packet;
4737     struct ds result;
4738     struct flow flow;
4739     uint16_t in_port;
4740
4741     /* Three kinds of error return values! */
4742     enum ofperr retval;
4743     char *error;
4744
4745     packet = NULL;
4746     ds_init(&result);
4747     ofpbuf_init(&ofpacts, 0);
4748
4749     /* Parse actions. */
4750     error = ofpacts_parse_actions(argv[--argc], &ofpacts, &usable_protocols);
4751     if (error) {
4752         unixctl_command_reply_error(conn, error);
4753         free(error);
4754         goto exit;
4755     }
4756
4757     /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4758      * consistency between the flow and the actions.  With -consistent, we
4759      * enforce consistency even for a flow supported in OpenFlow 1.0. */
4760     if (!strcmp(argv[1], "-consistent")) {
4761         enforce_consistency = true;
4762         argv++;
4763         argc--;
4764     } else {
4765         enforce_consistency = false;
4766     }
4767
4768     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4769     if (error) {
4770         unixctl_command_reply_error(conn, error);
4771         free(error);
4772         goto exit;
4773     }
4774
4775     /* Do the same checks as handle_packet_out() in ofproto.c.
4776      *
4777      * We pass a 'table_id' of 0 to ofpacts_check(), which isn't
4778      * strictly correct because these actions aren't in any table, but it's OK
4779      * because it 'table_id' is used only to check goto_table instructions, but
4780      * packet-outs take a list of actions and therefore it can't include
4781      * instructions.
4782      *
4783      * We skip the "meter" check here because meter is an instruction, not an
4784      * action, and thus cannot appear in ofpacts. */
4785     in_port = ofp_to_u16(flow.in_port.ofp_port);
4786     if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
4787         unixctl_command_reply_error(conn, "invalid in_port");
4788         goto exit;
4789     }
4790     if (enforce_consistency) {
4791         retval = ofpacts_check_consistency(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts),
4792                                            &flow, u16_to_ofp(ofproto->up.max_ports),
4793                                            0, 0, usable_protocols);
4794     } else {
4795         retval = ofpacts_check(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &flow,
4796                                u16_to_ofp(ofproto->up.max_ports), 0, 0,
4797                                &usable_protocols);
4798     }
4799
4800     if (retval) {
4801         ds_clear(&result);
4802         ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
4803         unixctl_command_reply_error(conn, ds_cstr(&result));
4804         goto exit;
4805     }
4806
4807     ofproto_trace(ofproto, &flow, packet,
4808                   ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &result);
4809     unixctl_command_reply(conn, ds_cstr(&result));
4810
4811 exit:
4812     ds_destroy(&result);
4813     ofpbuf_delete(packet);
4814     ofpbuf_uninit(&ofpacts);
4815 }
4816
4817 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4818  * description of the results to 'ds'.
4819  *
4820  * The trace follows a packet with the specified 'flow' through the flow
4821  * table.  'packet' may be nonnull to trace an actual packet, with consequent
4822  * side effects (if it is nonnull then its flow must be 'flow').
4823  *
4824  * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4825  * trace, otherwise the actions are determined by a flow table lookup. */
4826 static void
4827 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
4828               const struct ofpbuf *packet,
4829               const struct ofpact ofpacts[], size_t ofpacts_len,
4830               struct ds *ds)
4831 {
4832     struct trace_ctx trace;
4833
4834     ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
4835     ds_put_cstr(ds, "Flow: ");
4836     flow_format(ds, flow);
4837     ds_put_char(ds, '\n');
4838
4839     flow_wildcards_init_catchall(&trace.wc);
4840
4841     trace.result = ds;
4842     trace.key = flow; /* Original flow key, used for megaflow. */
4843     trace.flow = *flow; /* May be modified by actions. */
4844     xlate_in_init(&trace.xin, ofproto, flow, flow->in_port.ofp_port, NULL,
4845                   ntohs(flow->tcp_flags), packet);
4846     trace.xin.ofpacts = ofpacts;
4847     trace.xin.ofpacts_len = ofpacts_len;
4848     trace.xin.resubmit_hook = trace_resubmit;
4849     trace.xin.report_hook = trace_report;
4850
4851     xlate_actions(&trace.xin, &trace.xout);
4852
4853     ds_put_char(ds, '\n');
4854     trace_format_flow(ds, 0, "Final flow", &trace);
4855     trace_format_megaflow(ds, 0, "Megaflow", &trace);
4856
4857     ds_put_cstr(ds, "Datapath actions: ");
4858     format_odp_actions(ds, ofpbuf_data(trace.xout.odp_actions),
4859                        ofpbuf_size(trace.xout.odp_actions));
4860
4861     if (trace.xout.slow) {
4862         enum slow_path_reason slow;
4863
4864         ds_put_cstr(ds, "\nThis flow is handled by the userspace "
4865                     "slow path because it:");
4866
4867         slow = trace.xout.slow;
4868         while (slow) {
4869             enum slow_path_reason bit = rightmost_1bit(slow);
4870
4871             ds_put_format(ds, "\n\t- %s.",
4872                           slow_path_reason_to_explanation(bit));
4873
4874             slow &= ~bit;
4875         }
4876     }
4877
4878     xlate_out_uninit(&trace.xout);
4879 }
4880
4881 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
4882  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
4883  * to destroy 'ofproto_shash' and free the returned value. */
4884 static const struct shash_node **
4885 get_ofprotos(struct shash *ofproto_shash)
4886 {
4887     const struct ofproto_dpif *ofproto;
4888
4889     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4890         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
4891         shash_add_nocopy(ofproto_shash, name, ofproto);
4892     }
4893
4894     return shash_sort(ofproto_shash);
4895 }
4896
4897 static void
4898 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
4899                               const char *argv[] OVS_UNUSED,
4900                               void *aux OVS_UNUSED)
4901 {
4902     struct ds ds = DS_EMPTY_INITIALIZER;
4903     struct shash ofproto_shash;
4904     const struct shash_node **sorted_ofprotos;
4905     int i;
4906
4907     shash_init(&ofproto_shash);
4908     sorted_ofprotos = get_ofprotos(&ofproto_shash);
4909     for (i = 0; i < shash_count(&ofproto_shash); i++) {
4910         const struct shash_node *node = sorted_ofprotos[i];
4911         ds_put_format(&ds, "%s\n", node->name);
4912     }
4913
4914     shash_destroy(&ofproto_shash);
4915     free(sorted_ofprotos);
4916
4917     unixctl_command_reply(conn, ds_cstr(&ds));
4918     ds_destroy(&ds);
4919 }
4920
4921 static void
4922 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
4923 {
4924     const struct shash_node **ofprotos;
4925     struct dpif_dp_stats dp_stats;
4926     struct shash ofproto_shash;
4927     size_t i;
4928
4929     dpif_get_dp_stats(backer->dpif, &dp_stats);
4930
4931     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
4932                   dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
4933
4934     shash_init(&ofproto_shash);
4935     ofprotos = get_ofprotos(&ofproto_shash);
4936     for (i = 0; i < shash_count(&ofproto_shash); i++) {
4937         struct ofproto_dpif *ofproto = ofprotos[i]->data;
4938         const struct shash_node **ports;
4939         size_t j;
4940
4941         if (ofproto->backer != backer) {
4942             continue;
4943         }
4944
4945         ds_put_format(ds, "\t%s:\n", ofproto->up.name);
4946
4947         ports = shash_sort(&ofproto->up.port_by_name);
4948         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
4949             const struct shash_node *node = ports[j];
4950             struct ofport *ofport = node->data;
4951             struct smap config;
4952             odp_port_t odp_port;
4953
4954             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
4955                           ofport->ofp_port);
4956
4957             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
4958             if (odp_port != ODPP_NONE) {
4959                 ds_put_format(ds, "%"PRIu32":", odp_port);
4960             } else {
4961                 ds_put_cstr(ds, "none:");
4962             }
4963
4964             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
4965
4966             smap_init(&config);
4967             if (!netdev_get_config(ofport->netdev, &config)) {
4968                 const struct smap_node **nodes;
4969                 size_t i;
4970
4971                 nodes = smap_sort(&config);
4972                 for (i = 0; i < smap_count(&config); i++) {
4973                     const struct smap_node *node = nodes[i];
4974                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
4975                                   node->key, node->value);
4976                 }
4977                 free(nodes);
4978             }
4979             smap_destroy(&config);
4980
4981             ds_put_char(ds, ')');
4982             ds_put_char(ds, '\n');
4983         }
4984         free(ports);
4985     }
4986     shash_destroy(&ofproto_shash);
4987     free(ofprotos);
4988 }
4989
4990 static void
4991 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4992                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4993 {
4994     struct ds ds = DS_EMPTY_INITIALIZER;
4995     const struct shash_node **backers;
4996     int i;
4997
4998     backers = shash_sort(&all_dpif_backers);
4999     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5000         dpif_show_backer(backers[i]->data, &ds);
5001     }
5002     free(backers);
5003
5004     unixctl_command_reply(conn, ds_cstr(&ds));
5005     ds_destroy(&ds);
5006 }
5007
5008 static void
5009 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5010                                 int argc OVS_UNUSED, const char *argv[],
5011                                 void *aux OVS_UNUSED)
5012 {
5013     const struct ofproto_dpif *ofproto;
5014
5015     struct ds ds = DS_EMPTY_INITIALIZER;
5016     bool verbosity = false;
5017
5018     struct dpif_port dpif_port;
5019     struct dpif_port_dump port_dump;
5020     struct hmap portno_names;
5021
5022     struct dpif_flow_dump *flow_dump;
5023     struct dpif_flow_dump_thread *flow_dump_thread;
5024     struct dpif_flow f;
5025     int error;
5026
5027     ofproto = ofproto_dpif_lookup(argv[argc - 1]);
5028     if (!ofproto) {
5029         unixctl_command_reply_error(conn, "no such bridge");
5030         return;
5031     }
5032
5033     if (argc > 2 && !strcmp(argv[1], "-m")) {
5034         verbosity = true;
5035     }
5036
5037     hmap_init(&portno_names);
5038     DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
5039         odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
5040     }
5041
5042     ds_init(&ds);
5043     flow_dump = dpif_flow_dump_create(ofproto->backer->dpif, false);
5044     flow_dump_thread = dpif_flow_dump_thread_create(flow_dump);
5045     while (dpif_flow_dump_next(flow_dump_thread, &f, 1)) {
5046         struct flow flow;
5047
5048         if (odp_flow_key_to_flow(f.key, f.key_len, &flow) == ODP_FIT_ERROR
5049             || xlate_lookup_ofproto(ofproto->backer, &flow, NULL) != ofproto) {
5050             continue;
5051         }
5052
5053         if (verbosity) {
5054             odp_format_ufid(&f.ufid, &ds);
5055             ds_put_cstr(&ds, " ");
5056         }
5057         odp_flow_format(f.key, f.key_len, f.mask, f.mask_len,
5058                         &portno_names, &ds, verbosity);
5059         ds_put_cstr(&ds, ", ");
5060         dpif_flow_stats_format(&f.stats, &ds);
5061         ds_put_cstr(&ds, ", actions:");
5062         format_odp_actions(&ds, f.actions, f.actions_len);
5063         ds_put_char(&ds, '\n');
5064     }
5065     dpif_flow_dump_thread_destroy(flow_dump_thread);
5066     error = dpif_flow_dump_destroy(flow_dump);
5067
5068     if (error) {
5069         ds_clear(&ds);
5070         ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
5071         unixctl_command_reply_error(conn, ds_cstr(&ds));
5072     } else {
5073         unixctl_command_reply(conn, ds_cstr(&ds));
5074     }
5075     odp_portno_names_destroy(&portno_names);
5076     hmap_destroy(&portno_names);
5077     ds_destroy(&ds);
5078 }
5079
5080 static void
5081 ofproto_revalidate_all_backers(void)
5082 {
5083     const struct shash_node **backers;
5084     int i;
5085
5086     backers = shash_sort(&all_dpif_backers);
5087     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5088         struct dpif_backer *backer = backers[i]->data;
5089         backer->need_revalidate = REV_RECONFIGURE;
5090     }
5091     free(backers);
5092 }
5093
5094 static void
5095 disable_tnl_push_pop(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5096                      const char *argv[], void *aux OVS_UNUSED)
5097 {
5098     if (!strcasecmp(argv[1], "off")) {
5099         ofproto_use_tnl_push_pop = false;
5100         unixctl_command_reply(conn, "Tunnel push-pop off");
5101         ofproto_revalidate_all_backers();
5102     } else if (!strcasecmp(argv[1], "on")) {
5103         ofproto_use_tnl_push_pop = true;
5104         unixctl_command_reply(conn, "Tunnel push-pop on");
5105         ofproto_revalidate_all_backers();
5106     }
5107 }
5108
5109 static void
5110 ofproto_unixctl_init(void)
5111 {
5112     static bool registered;
5113     if (registered) {
5114         return;
5115     }
5116     registered = true;
5117
5118     unixctl_command_register(
5119         "ofproto/trace",
5120         "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
5121         1, 3, ofproto_unixctl_trace, NULL);
5122     unixctl_command_register(
5123         "ofproto/trace-packet-out",
5124         "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
5125         2, 6, ofproto_unixctl_trace_actions, NULL);
5126     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5127                              ofproto_unixctl_fdb_flush, NULL);
5128     unixctl_command_register("fdb/show", "bridge", 1, 1,
5129                              ofproto_unixctl_fdb_show, NULL);
5130     unixctl_command_register("mdb/flush", "[bridge]", 0, 1,
5131                              ofproto_unixctl_mcast_snooping_flush, NULL);
5132     unixctl_command_register("mdb/show", "bridge", 1, 1,
5133                              ofproto_unixctl_mcast_snooping_show, NULL);
5134     unixctl_command_register("dpif/dump-dps", "", 0, 0,
5135                              ofproto_unixctl_dpif_dump_dps, NULL);
5136     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5137                              NULL);
5138     unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
5139                              ofproto_unixctl_dpif_dump_flows, NULL);
5140
5141     unixctl_command_register("ofproto/tnl-push-pop", "[on]|[off]", 1, 1,
5142                              disable_tnl_push_pop, NULL);
5143 }
5144
5145 /* Returns true if 'table' is the table used for internal rules,
5146  * false otherwise. */
5147 bool
5148 table_is_internal(uint8_t table_id)
5149 {
5150     return table_id == TBL_INTERNAL;
5151 }
5152 \f
5153 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5154  *
5155  * This is deprecated.  It is only for compatibility with broken device drivers
5156  * in old versions of Linux that do not properly support VLANs when VLAN
5157  * devices are not used.  When broken device drivers are no longer in
5158  * widespread use, we will delete these interfaces. */
5159
5160 static int
5161 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5162 {
5163     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5164     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5165
5166     if (realdev_ofp_port == ofport->realdev_ofp_port
5167         && vid == ofport->vlandev_vid) {
5168         return 0;
5169     }
5170
5171     ofproto->backer->need_revalidate = REV_RECONFIGURE;
5172
5173     if (ofport->realdev_ofp_port) {
5174         vsp_remove(ofport);
5175     }
5176     if (realdev_ofp_port && ofport->bundle) {
5177         /* vlandevs are enslaved to their realdevs, so they are not allowed to
5178          * themselves be part of a bundle. */
5179         bundle_set(ofport_->ofproto, ofport->bundle, NULL);
5180     }
5181
5182     ofport->realdev_ofp_port = realdev_ofp_port;
5183     ofport->vlandev_vid = vid;
5184
5185     if (realdev_ofp_port) {
5186         vsp_add(ofport, realdev_ofp_port, vid);
5187     }
5188
5189     return 0;
5190 }
5191
5192 static uint32_t
5193 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
5194 {
5195     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
5196 }
5197
5198 bool
5199 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
5200     OVS_EXCLUDED(ofproto->vsp_mutex)
5201 {
5202     /* hmap_is_empty is thread safe. */
5203     return !hmap_is_empty(&ofproto->realdev_vid_map);
5204 }
5205
5206
5207 static ofp_port_t
5208 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
5209                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5210     OVS_REQUIRES(ofproto->vsp_mutex)
5211 {
5212     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5213         int vid = vlan_tci_to_vid(vlan_tci);
5214         const struct vlan_splinter *vsp;
5215
5216         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5217                                  hash_realdev_vid(realdev_ofp_port, vid),
5218                                  &ofproto->realdev_vid_map) {
5219             if (vsp->realdev_ofp_port == realdev_ofp_port
5220                 && vsp->vid == vid) {
5221                 return vsp->vlandev_ofp_port;
5222             }
5223         }
5224     }
5225     return realdev_ofp_port;
5226 }
5227
5228 /* Returns the OFP port number of the Linux VLAN device that corresponds to
5229  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
5230  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
5231  * 'vlan_tci' 9, it would return the port number of eth0.9.
5232  *
5233  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
5234  * function just returns its 'realdev_ofp_port' argument. */
5235 ofp_port_t
5236 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5237                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5238     OVS_EXCLUDED(ofproto->vsp_mutex)
5239 {
5240     ofp_port_t ret;
5241
5242     /* hmap_is_empty is thread safe, see if we can return immediately. */
5243     if (hmap_is_empty(&ofproto->realdev_vid_map)) {
5244         return realdev_ofp_port;
5245     }
5246     ovs_mutex_lock(&ofproto->vsp_mutex);
5247     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
5248     ovs_mutex_unlock(&ofproto->vsp_mutex);
5249     return ret;
5250 }
5251
5252 static struct vlan_splinter *
5253 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
5254 {
5255     struct vlan_splinter *vsp;
5256
5257     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
5258                              hash_ofp_port(vlandev_ofp_port),
5259                              &ofproto->vlandev_map) {
5260         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5261             return vsp;
5262         }
5263     }
5264
5265     return NULL;
5266 }
5267
5268 /* Returns the OpenFlow port number of the "real" device underlying the Linux
5269  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
5270  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
5271  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
5272  * eth0 and store 9 in '*vid'.
5273  *
5274  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
5275  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
5276  * always does.*/
5277 static ofp_port_t
5278 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5279                        ofp_port_t vlandev_ofp_port, int *vid)
5280     OVS_REQUIRES(ofproto->vsp_mutex)
5281 {
5282     if (!hmap_is_empty(&ofproto->vlandev_map)) {
5283         const struct vlan_splinter *vsp;
5284
5285         vsp = vlandev_find(ofproto, vlandev_ofp_port);
5286         if (vsp) {
5287             if (vid) {
5288                 *vid = vsp->vid;
5289             }
5290             return vsp->realdev_ofp_port;
5291         }
5292     }
5293     return 0;
5294 }
5295
5296 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
5297  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
5298  * 'flow->in_port' to the "real" device backing the VLAN device, sets
5299  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Optionally pushes the
5300  * appropriate VLAN on 'packet' if provided.  Otherwise (which is always the
5301  * case unless VLAN splinters are enabled), returns false without making any
5302  * changes. */
5303 bool
5304 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow,
5305                 struct ofpbuf *packet)
5306     OVS_EXCLUDED(ofproto->vsp_mutex)
5307 {
5308     ofp_port_t realdev;
5309     int vid;
5310
5311     /* hmap_is_empty is thread safe. */
5312     if (hmap_is_empty(&ofproto->vlandev_map)) {
5313         return false;
5314     }
5315
5316     ovs_mutex_lock(&ofproto->vsp_mutex);
5317     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
5318     ovs_mutex_unlock(&ofproto->vsp_mutex);
5319     if (!realdev) {
5320         return false;
5321     }
5322
5323     /* Cause the flow to be processed as if it came in on the real device with
5324      * the VLAN device's VLAN ID. */
5325     flow->in_port.ofp_port = realdev;
5326     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
5327
5328     if (packet) {
5329         /* Make the packet resemble the flow, so that it gets sent to an
5330          * OpenFlow controller properly, so that it looks correct for sFlow,
5331          * and so that flow_extract() will get the correct vlan_tci if it is
5332          * called on 'packet'. */
5333         eth_push_vlan(packet, htons(ETH_TYPE_VLAN), flow->vlan_tci);
5334     }
5335
5336     return true;
5337 }
5338
5339 static void
5340 vsp_remove(struct ofport_dpif *port)
5341 {
5342     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5343     struct vlan_splinter *vsp;
5344
5345     ovs_mutex_lock(&ofproto->vsp_mutex);
5346     vsp = vlandev_find(ofproto, port->up.ofp_port);
5347     if (vsp) {
5348         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5349         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
5350         free(vsp);
5351
5352         port->realdev_ofp_port = 0;
5353     } else {
5354         VLOG_ERR("missing vlan device record");
5355     }
5356     ovs_mutex_unlock(&ofproto->vsp_mutex);
5357 }
5358
5359 static void
5360 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
5361 {
5362     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5363
5364     ovs_mutex_lock(&ofproto->vsp_mutex);
5365     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
5366         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
5367             == realdev_ofp_port)) {
5368         struct vlan_splinter *vsp;
5369
5370         vsp = xmalloc(sizeof *vsp);
5371         vsp->realdev_ofp_port = realdev_ofp_port;
5372         vsp->vlandev_ofp_port = port->up.ofp_port;
5373         vsp->vid = vid;
5374
5375         port->realdev_ofp_port = realdev_ofp_port;
5376
5377         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
5378                     hash_ofp_port(port->up.ofp_port));
5379         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
5380                     hash_realdev_vid(realdev_ofp_port, vid));
5381     } else {
5382         VLOG_ERR("duplicate vlan device record");
5383     }
5384     ovs_mutex_unlock(&ofproto->vsp_mutex);
5385 }
5386
5387 static odp_port_t
5388 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
5389 {
5390     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
5391     return ofport ? ofport->odp_port : ODPP_NONE;
5392 }
5393
5394 struct ofport_dpif *
5395 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
5396 {
5397     struct ofport_dpif *port;
5398
5399     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
5400     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
5401                              &backer->odp_to_ofport_map) {
5402         if (port->odp_port == odp_port) {
5403             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5404             return port;
5405         }
5406     }
5407
5408     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5409     return NULL;
5410 }
5411
5412 static ofp_port_t
5413 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
5414 {
5415     struct ofport_dpif *port;
5416
5417     port = odp_port_to_ofport(ofproto->backer, odp_port);
5418     if (port && &ofproto->up == port->up.ofproto) {
5419         return port->up.ofp_port;
5420     } else {
5421         return OFPP_NONE;
5422     }
5423 }
5424
5425 struct ofproto_dpif *
5426 ofproto_dpif_recirc_get_ofproto(const struct dpif_backer *backer,
5427                                 uint32_t recirc_id)
5428 {
5429     struct dpif_backer_recirc_node *node;
5430
5431     node = CONTAINER_OF(cmap_find(&backer->recirc_map, recirc_id),
5432                         struct dpif_backer_recirc_node, cmap_node);
5433
5434     return node ? node->ofproto : NULL;
5435 }
5436
5437 uint32_t
5438 ofproto_dpif_alloc_recirc_id(struct ofproto_dpif *ofproto)
5439 {
5440     struct dpif_backer *backer = ofproto->backer;
5441     uint32_t recirc_id = recirc_id_alloc(backer->rid_pool);
5442
5443     if (recirc_id) {
5444         struct dpif_backer_recirc_node *node = xmalloc(sizeof *node);
5445
5446         node->recirc_id = recirc_id;
5447         node->ofproto = ofproto;
5448
5449         ovs_mutex_lock(&backer->recirc_mutex);
5450         cmap_insert(&backer->recirc_map, &node->cmap_node, node->recirc_id);
5451         ovs_mutex_unlock(&backer->recirc_mutex);
5452     }
5453
5454     return recirc_id;
5455 }
5456
5457 void
5458 ofproto_dpif_free_recirc_id(struct ofproto_dpif *ofproto, uint32_t recirc_id)
5459 {
5460     struct dpif_backer *backer = ofproto->backer;
5461     struct dpif_backer_recirc_node *node;
5462
5463     node = CONTAINER_OF(cmap_find(&backer->recirc_map, recirc_id),
5464                         struct dpif_backer_recirc_node, cmap_node);
5465     if (node) {
5466         ovs_mutex_lock(&backer->recirc_mutex);
5467         cmap_remove(&backer->recirc_map, &node->cmap_node, node->recirc_id);
5468         ovs_mutex_unlock(&backer->recirc_mutex);
5469         recirc_id_free(backer->rid_pool, node->recirc_id);
5470
5471         /* 'recirc_id' should never be freed by non-owning 'ofproto'. */
5472         ovs_assert(node->ofproto == ofproto);
5473
5474         /* RCU postpone the free, since other threads may be referring
5475          * to 'node' at same time. */
5476         ovsrcu_postpone(free, node);
5477     }
5478 }
5479
5480 int
5481 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
5482                                const struct match *match, int priority,
5483                                uint16_t idle_timeout,
5484                                const struct ofpbuf *ofpacts,
5485                                struct rule **rulep)
5486 {
5487     struct ofputil_flow_mod fm;
5488     struct rule_dpif *rule;
5489     int error;
5490
5491     fm.match = *match;
5492     fm.priority = priority;
5493     fm.new_cookie = htonll(0);
5494     fm.cookie = htonll(0);
5495     fm.cookie_mask = htonll(0);
5496     fm.modify_cookie = false;
5497     fm.table_id = TBL_INTERNAL;
5498     fm.command = OFPFC_ADD;
5499     fm.idle_timeout = idle_timeout;
5500     fm.hard_timeout = 0;
5501     fm.importance = 0;
5502     fm.buffer_id = 0;
5503     fm.out_port = 0;
5504     fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5505     fm.ofpacts = ofpbuf_data(ofpacts);
5506     fm.ofpacts_len = ofpbuf_size(ofpacts);
5507
5508     error = ofproto_flow_mod(&ofproto->up, &fm);
5509     if (error) {
5510         VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
5511                     ofperr_to_string(error));
5512         *rulep = NULL;
5513         return error;
5514     }
5515
5516     rule = rule_dpif_lookup_in_table(ofproto, TBL_INTERNAL, &fm.match.flow,
5517                                      &fm.match.wc, false);
5518     if (rule) {
5519         *rulep = &rule->up;
5520     } else {
5521         OVS_NOT_REACHED();
5522     }
5523     return 0;
5524 }
5525
5526 int
5527 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
5528                                   struct match *match, int priority)
5529 {
5530     struct ofputil_flow_mod fm;
5531     int error;
5532
5533     fm.match = *match;
5534     fm.priority = priority;
5535     fm.new_cookie = htonll(0);
5536     fm.cookie = htonll(0);
5537     fm.cookie_mask = htonll(0);
5538     fm.modify_cookie = false;
5539     fm.table_id = TBL_INTERNAL;
5540     fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5541     fm.command = OFPFC_DELETE_STRICT;
5542
5543     error = ofproto_flow_mod(&ofproto->up, &fm);
5544     if (error) {
5545         VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
5546                     ofperr_to_string(error));
5547         return error;
5548     }
5549
5550     return 0;
5551 }
5552
5553 const struct ofproto_class ofproto_dpif_class = {
5554     init,
5555     enumerate_types,
5556     enumerate_names,
5557     del,
5558     port_open_type,
5559     type_run,
5560     type_wait,
5561     alloc,
5562     construct,
5563     destruct,
5564     dealloc,
5565     run,
5566     wait,
5567     NULL,                       /* get_memory_usage. */
5568     type_get_memory_usage,
5569     flush,
5570     query_tables,
5571     port_alloc,
5572     port_construct,
5573     port_destruct,
5574     port_dealloc,
5575     port_modified,
5576     port_reconfigured,
5577     port_query_by_name,
5578     port_add,
5579     port_del,
5580     port_get_stats,
5581     port_dump_start,
5582     port_dump_next,
5583     port_dump_done,
5584     port_poll,
5585     port_poll_wait,
5586     port_is_lacp_current,
5587     port_get_lacp_stats,
5588     NULL,                       /* rule_choose_table */
5589     rule_alloc,
5590     rule_construct,
5591     rule_insert,
5592     rule_delete,
5593     rule_destruct,
5594     rule_dealloc,
5595     rule_get_stats,
5596     rule_execute,
5597     NULL,                       /* rule_premodify_actions */
5598     rule_modify_actions,
5599     set_frag_handling,
5600     packet_out,
5601     set_netflow,
5602     get_netflow_ids,
5603     set_sflow,
5604     set_ipfix,
5605     set_cfm,
5606     cfm_status_changed,
5607     get_cfm_status,
5608     set_bfd,
5609     bfd_status_changed,
5610     get_bfd_status,
5611     set_stp,
5612     get_stp_status,
5613     set_stp_port,
5614     get_stp_port_status,
5615     get_stp_port_stats,
5616     set_rstp,
5617     get_rstp_status,
5618     set_rstp_port,
5619     get_rstp_port_status,
5620     set_queues,
5621     bundle_set,
5622     bundle_remove,
5623     mirror_set__,
5624     mirror_get_stats__,
5625     set_flood_vlans,
5626     is_mirror_output_bundle,
5627     forward_bpdu_changed,
5628     set_mac_table_config,
5629     set_mcast_snooping,
5630     set_mcast_snooping_port,
5631     set_realdev,
5632     NULL,                       /* meter_get_features */
5633     NULL,                       /* meter_set */
5634     NULL,                       /* meter_get */
5635     NULL,                       /* meter_del */
5636     group_alloc,                /* group_alloc */
5637     group_construct,            /* group_construct */
5638     group_destruct,             /* group_destruct */
5639     group_dealloc,              /* group_dealloc */
5640     group_modify,               /* group_modify */
5641     group_get_stats,            /* group_get_stats */
5642     get_datapath_version,       /* get_datapath_version */
5643 };