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