ofproto/ofproto-dpif.c

   1 /*
   2  * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015 Nicira, Inc.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18
  19 #include "ofproto/ofproto-dpif.h"
  20 #include "ofproto/ofproto-provider.h"
  21
  22 #include <errno.h>
  23
  24 #include "bfd.h"
  25 #include "bond.h"
  26 #include "bundle.h"
  27 #include "byte-order.h"
  28 #include "connectivity.h"
  29 #include "connmgr.h"
  30 #include "coverage.h"
  31 #include "cfm.h"
  32 #include "ovs-lldp.h"
  33 #include "dpif.h"
  34 #include "dynamic-string.h"
  35 #include "fail-open.h"
  36 #include "guarded-list.h"
  37 #include "hmapx.h"
  38 #include "lacp.h"
  39 #include "learn.h"
  40 #include "mac-learning.h"
  41 #include "mcast-snooping.h"
  42 #include "meta-flow.h"
  43 #include "multipath.h"
  44 #include "netdev-vport.h"
  45 #include "netdev.h"
  46 #include "netlink.h"
  47 #include "nx-match.h"
  48 #include "odp-util.h"
  49 #include "odp-execute.h"
  50 #include "ofp-util.h"
  51 #include "ofpbuf.h"
  52 #include "ofp-actions.h"
  53 #include "ofp-parse.h"
  54 #include "ofp-print.h"
  55 #include "ofproto-dpif-ipfix.h"
  56 #include "ofproto-dpif-mirror.h"
  57 #include "ofproto-dpif-monitor.h"
  58 #include "ofproto-dpif-rid.h"
  59 #include "ofproto-dpif-sflow.h"
  60 #include "ofproto-dpif-upcall.h"
  61 #include "ofproto-dpif-xlate.h"
  62 #include "poll-loop.h"
  63 #include "ovs-rcu.h"
  64 #include "ovs-router.h"
  65 #include "seq.h"
  66 #include "simap.h"
  67 #include "smap.h"
  68 #include "timer.h"
  69 #include "tunnel.h"
  70 #include "unaligned.h"
  71 #include "unixctl.h"
  72 #include "vlan-bitmap.h"
  73 #include "openvswitch/vlog.h"
  74
  75 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  76
  77 COVERAGE_DEFINE(ofproto_dpif_expired);
  78 COVERAGE_DEFINE(packet_in_overflow);
  79
  80 struct flow_miss;
  81
  82 struct rule_dpif {
  83     struct rule up;
  84
  85     /* These statistics:
  86      *
  87      *   - Do include packets and bytes from datapath flows which have not
  88      *   recently been processed by a revalidator. */
  89     struct ovs_mutex stats_mutex;
  90     struct dpif_flow_stats stats OVS_GUARDED;
  91
  92    /* In non-NULL, will point to a new rule (for which a reference is held) to
  93     * which all the stats updates should be forwarded. This exists only
  94     * transitionally when flows are replaced.
  95     *
  96     * Protected by stats_mutex.  If both 'rule->stats_mutex' and
  97     * 'rule->new_rule->stats_mutex' must be held together, acquire them in that
  98     * order, */
  99     struct rule_dpif *new_rule OVS_GUARDED;
 100
 101     /* If non-zero then the recirculation id that has
 102      * been allocated for use with this rule.
 103      * The recirculation id and associated internal flow should
 104      * be freed when the rule is freed */
 105     uint32_t recirc_id;
 106 };
 107
 108 /* RULE_CAST() depends on this. */
 109 BUILD_ASSERT_DECL(offsetof(struct rule_dpif, up) == 0);
 110
 111 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
 112                            long long int *used);
 113 static struct rule_dpif *rule_dpif_cast(const struct rule *);
 114 static void rule_expire(struct rule_dpif *);
 115
 116 struct group_dpif {
 117     struct ofgroup up;
 118
 119     /* These statistics:
 120      *
 121      *   - Do include packets and bytes from datapath flows which have not
 122      *   recently been processed by a revalidator. */
 123     struct ovs_mutex stats_mutex;
 124     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 125     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 126 };
 127
 128 struct ofbundle {
 129     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
 130     struct ofproto_dpif *ofproto; /* Owning ofproto. */
 131     void *aux;                  /* Key supplied by ofproto's client. */
 132     char *name;                 /* Identifier for log messages. */
 133
 134     /* Configuration. */
 135     struct ovs_list ports;      /* Contains "struct ofport"s. */
 136     enum port_vlan_mode vlan_mode; /* VLAN mode */
 137     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
 138     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
 139                                  * NULL if all VLANs are trunked. */
 140     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 141     struct bond *bond;          /* Nonnull iff more than one port. */
 142     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 143
 144     /* Status. */
 145     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 146 };
 147
 148 static void bundle_remove(struct ofport *);
 149 static void bundle_update(struct ofbundle *);
 150 static void bundle_destroy(struct ofbundle *);
 151 static void bundle_del_port(struct ofport_dpif *);
 152 static void bundle_run(struct ofbundle *);
 153 static void bundle_wait(struct ofbundle *);
 154 static void bundle_flush_macs(struct ofbundle *, bool);
 155 static void bundle_move(struct ofbundle *, struct ofbundle *);
 156
 157 static void stp_run(struct ofproto_dpif *ofproto);
 158 static void stp_wait(struct ofproto_dpif *ofproto);
 159 static int set_stp_port(struct ofport *,
 160                         const struct ofproto_port_stp_settings *);
 161
 162 static void rstp_run(struct ofproto_dpif *ofproto);
 163 static void set_rstp_port(struct ofport *,
 164                          const struct ofproto_port_rstp_settings *);
 165
 166 struct ofport_dpif {
 167     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 168     struct ofport up;
 169
 170     odp_port_t odp_port;
 171     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 172     struct ovs_list bundle_node;/* In struct ofbundle's "ports" list. */
 173     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 174     struct bfd *bfd;            /* BFD, if any. */
 175     struct lldp *lldp;          /* lldp, if any. */
 176     bool may_enable;            /* May be enabled in bonds. */
 177     bool is_tunnel;             /* This port is a tunnel. */
 178     bool is_layer3;             /* This is a layer 3 port. */
 179     long long int carrier_seq;  /* Carrier status changes. */
 180     struct ofport_dpif *peer;   /* Peer if patch port. */
 181
 182     /* Spanning tree. */
 183     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 184     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 185     long long int stp_state_entered;
 186
 187     /* Rapid Spanning Tree. */
 188     struct rstp_port *rstp_port; /* Rapid Spanning Tree Protocol, if any. */
 189     enum rstp_state rstp_state; /* Always RSTP_DISABLED if RSTP not in use. */
 190
 191     /* Queue to DSCP mapping. */
 192     struct ofproto_port_queue *qdscp;
 193     size_t n_qdscp;
 194
 195     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 196      *
 197      * This is deprecated.  It is only for compatibility with broken device
 198      * drivers in old versions of Linux that do not properly support VLANs when
 199      * VLAN devices are not used.  When broken device drivers are no longer in
 200      * widespread use, we will delete these interfaces. */
 201     ofp_port_t realdev_ofp_port;
 202     int vlandev_vid;
 203 };
 204
 205 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 206  *
 207  * This is deprecated.  It is only for compatibility with broken device drivers
 208  * in old versions of Linux that do not properly support VLANs when VLAN
 209  * devices are not used.  When broken device drivers are no longer in
 210  * widespread use, we will delete these interfaces. */
 211 struct vlan_splinter {
 212     struct hmap_node realdev_vid_node;
 213     struct hmap_node vlandev_node;
 214     ofp_port_t realdev_ofp_port;
 215     ofp_port_t vlandev_ofp_port;
 216     int vid;
 217 };
 218
 219 static void vsp_remove(struct ofport_dpif *);
 220 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 221
 222 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 223                                        ofp_port_t);
 224
 225 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 226                                        odp_port_t);
 227
 228 static struct ofport_dpif *
 229 ofport_dpif_cast(const struct ofport *ofport)
 230 {
 231     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 232 }
 233
 234 static void port_run(struct ofport_dpif *);
 235 static int set_bfd(struct ofport *, const struct smap *);
 236 static int set_cfm(struct ofport *, const struct cfm_settings *);
 237 static int set_lldp(struct ofport *ofport_, const struct smap *cfg);
 238 static void ofport_update_peer(struct ofport_dpif *);
 239
 240 /* Reasons that we might need to revalidate every datapath flow, and
 241  * corresponding coverage counters.
 242  *
 243  * A value of 0 means that there is no need to revalidate.
 244  *
 245  * It would be nice to have some cleaner way to integrate with coverage
 246  * counters, but with only a few reasons I guess this is good enough for
 247  * now. */
 248 enum revalidate_reason {
 249     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 250     REV_STP,                   /* Spanning tree protocol port status change. */
 251     REV_RSTP,                  /* RSTP port status change. */
 252     REV_BOND,                  /* Bonding changed. */
 253     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 254     REV_FLOW_TABLE,            /* Flow table changed. */
 255     REV_MAC_LEARNING,          /* Mac learning changed. */
 256     REV_MCAST_SNOOPING,        /* Multicast snooping changed. */
 257 };
 258 COVERAGE_DEFINE(rev_reconfigure);
 259 COVERAGE_DEFINE(rev_stp);
 260 COVERAGE_DEFINE(rev_rstp);
 261 COVERAGE_DEFINE(rev_bond);
 262 COVERAGE_DEFINE(rev_port_toggled);
 263 COVERAGE_DEFINE(rev_flow_table);
 264 COVERAGE_DEFINE(rev_mac_learning);
 265 COVERAGE_DEFINE(rev_mcast_snooping);
 266
 267 /* All datapaths of a given type share a single dpif backer instance. */
 268 struct dpif_backer {
 269     char *type;
 270     int refcount;
 271     struct dpif *dpif;
 272     struct udpif *udpif;
 273
 274     struct ovs_rwlock odp_to_ofport_lock;
 275     struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
 276
 277     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 278
 279     enum revalidate_reason need_revalidate; /* Revalidate all flows. */
 280
 281     bool recv_set_enable; /* Enables or disables receiving packets. */
 282
 283     /* Version string of the datapath stored in OVSDB. */
 284     char *dp_version_string;
 285
 286     /* Datapath feature support. */
 287     struct dpif_backer_support support;
 288     struct atomic_count tnl_count;
 289 };
 290
 291 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 292 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 293
 294 struct ofproto_dpif {
 295     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 296     struct ofproto up;
 297     struct dpif_backer *backer;
 298
 299     atomic_llong tables_version;  /* Version # to use in classifier lookups. */
 300
 301     uint64_t dump_seq; /* Last read of udpif_dump_seq(). */
 302
 303     /* Special OpenFlow rules. */
 304     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 305     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 306     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 307
 308     /* Bridging. */
 309     struct netflow *netflow;
 310     struct dpif_sflow *sflow;
 311     struct dpif_ipfix *ipfix;
 312     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 313     struct mac_learning *ml;
 314     struct mcast_snooping *ms;
 315     bool has_bonded_bundles;
 316     bool lacp_enabled;
 317     struct mbridge *mbridge;
 318
 319     struct ovs_mutex stats_mutex;
 320     struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
 321                                             * consumed in userspace. */
 322
 323     /* Spanning tree. */
 324     struct stp *stp;
 325     long long int stp_last_tick;
 326
 327     /* Rapid Spanning Tree. */
 328     struct rstp *rstp;
 329     long long int rstp_last_tick;
 330
 331     /* VLAN splinters. */
 332     struct ovs_mutex vsp_mutex;
 333     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 334     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 335
 336     /* Ports. */
 337     struct sset ports;             /* Set of standard port names. */
 338     struct sset ghost_ports;       /* Ports with no datapath port. */
 339     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 340     int port_poll_errno;           /* Last errno for port_poll() reply. */
 341     uint64_t change_seq;           /* Connectivity status changes. */
 342
 343     /* Work queues. */
 344     struct guarded_list pins;      /* Contains "struct ofputil_packet_in"s. */
 345     struct seq *pins_seq;          /* For notifying 'pins' reception. */
 346     uint64_t pins_seqno;
 347 };
 348
 349 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 350 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 351
 352 static bool ofproto_use_tnl_push_pop = true;
 353 static void ofproto_unixctl_init(void);
 354
 355 static inline struct ofproto_dpif *
 356 ofproto_dpif_cast(const struct ofproto *ofproto)
 357 {
 358     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 359     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 360 }
 361
 362 size_t
 363 ofproto_dpif_get_max_mpls_depth(const struct ofproto_dpif *ofproto)
 364 {
 365     return ofproto->backer->support.max_mpls_depth;
 366 }
 367
 368 bool
 369 ofproto_dpif_get_enable_recirc(const struct ofproto_dpif *ofproto)
 370 {
 371     return ofproto->backer->support.recirc;
 372 }
 373
 374 bool
 375 ofproto_dpif_get_enable_ufid(struct dpif_backer *backer)
 376 {
 377     return backer->support.ufid;
 378 }
 379
 380 static void ofproto_trace(struct ofproto_dpif *, struct flow *,
 381                           const struct dp_packet *packet,
 382                           const struct ofpact[], size_t ofpacts_len,
 383                           struct ds *);
 384
 385 /* Global variables. */
 386 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 387
 388 /* Initial mappings of port to bridge mappings. */
 389 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 390
 391 /* Executes 'fm'.  The caller retains ownership of 'fm' and everything in
 392  * it. */
 393 void
 394 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 395                       struct ofputil_flow_mod *fm)
 396 {
 397     ofproto_flow_mod(&ofproto->up, fm);
 398 }
 399
 400 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 401  * Takes ownership of 'pin' and pin->packet. */
 402 void
 403 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 404                             struct ofproto_packet_in *pin)
 405 {
 406     if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
 407         COVERAGE_INC(packet_in_overflow);
 408         free(CONST_CAST(void *, pin->up.packet));
 409         free(pin);
 410     }
 411
 412     /* Wakes up main thread for packet-in I/O. */
 413     seq_change(ofproto->pins_seq);
 414 }
 415
 416 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
 417  * packet rather than to send the packet to the controller.
 418  *
 419  * This function returns false to indicate that a packet_in message
 420  * for a "table-miss" should be sent to at least one controller.
 421  * False otherwise. */
 422 bool
 423 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif *ofproto)
 424 {
 425     return connmgr_wants_packet_in_on_miss(ofproto->up.connmgr);
 426 }
 427 \f
 428 /* Factory functions. */
 429
 430 static void
 431 init(const struct shash *iface_hints)
 432 {
 433     struct shash_node *node;
 434
 435     /* Make a local copy, since we don't own 'iface_hints' elements. */
 436     SHASH_FOR_EACH(node, iface_hints) {
 437         const struct iface_hint *orig_hint = node->data;
 438         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 439
 440         new_hint->br_name = xstrdup(orig_hint->br_name);
 441         new_hint->br_type = xstrdup(orig_hint->br_type);
 442         new_hint->ofp_port = orig_hint->ofp_port;
 443
 444         shash_add(&init_ofp_ports, node->name, new_hint);
 445     }
 446 }
 447
 448 static void
 449 enumerate_types(struct sset *types)
 450 {
 451     dp_enumerate_types(types);
 452 }
 453
 454 static int
 455 enumerate_names(const char *type, struct sset *names)
 456 {
 457     struct ofproto_dpif *ofproto;
 458
 459     sset_clear(names);
 460     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 461         if (strcmp(type, ofproto->up.type)) {
 462             continue;
 463         }
 464         sset_add(names, ofproto->up.name);
 465     }
 466
 467     return 0;
 468 }
 469
 470 static int
 471 del(const char *type, const char *name)
 472 {
 473     struct dpif *dpif;
 474     int error;
 475
 476     error = dpif_open(name, type, &dpif);
 477     if (!error) {
 478         error = dpif_delete(dpif);
 479         dpif_close(dpif);
 480     }
 481     return error;
 482 }
 483 \f
 484 static const char *
 485 port_open_type(const char *datapath_type, const char *port_type)
 486 {
 487     return dpif_port_open_type(datapath_type, port_type);
 488 }
 489
 490 /* Type functions. */
 491
 492 static void process_dpif_port_changes(struct dpif_backer *);
 493 static void process_dpif_all_ports_changed(struct dpif_backer *);
 494 static void process_dpif_port_change(struct dpif_backer *,
 495                                      const char *devname);
 496 static void process_dpif_port_error(struct dpif_backer *, int error);
 497
 498 static struct ofproto_dpif *
 499 lookup_ofproto_dpif_by_port_name(const char *name)
 500 {
 501     struct ofproto_dpif *ofproto;
 502
 503     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 504         if (sset_contains(&ofproto->ports, name)) {
 505             return ofproto;
 506         }
 507     }
 508
 509     return NULL;
 510 }
 511
 512 static int
 513 type_run(const char *type)
 514 {
 515     struct dpif_backer *backer;
 516
 517     backer = shash_find_data(&all_dpif_backers, type);
 518     if (!backer) {
 519         /* This is not necessarily a problem, since backers are only
 520          * created on demand. */
 521         return 0;
 522     }
 523
 524
 525     if (dpif_run(backer->dpif)) {
 526         backer->need_revalidate = REV_RECONFIGURE;
 527     }
 528
 529     udpif_run(backer->udpif);
 530
 531     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 532      * and the configuration has now changed to "false", enable receiving
 533      * packets from the datapath. */
 534     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 535         int error;
 536
 537         backer->recv_set_enable = true;
 538
 539         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 540         if (error) {
 541             VLOG_ERR("Failed to enable receiving packets in dpif.");
 542             return error;
 543         }
 544         dpif_flow_flush(backer->dpif);
 545         backer->need_revalidate = REV_RECONFIGURE;
 546     }
 547
 548     if (backer->recv_set_enable) {
 549         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 550     }
 551
 552     dpif_poll_threads_set(backer->dpif, n_dpdk_rxqs, pmd_cpu_mask);
 553
 554     if (backer->need_revalidate) {
 555         struct ofproto_dpif *ofproto;
 556         struct simap_node *node;
 557         struct simap tmp_backers;
 558
 559         /* Handle tunnel garbage collection. */
 560         simap_init(&tmp_backers);
 561         simap_swap(&backer->tnl_backers, &tmp_backers);
 562
 563         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 564             struct ofport_dpif *iter;
 565
 566             if (backer != ofproto->backer) {
 567                 continue;
 568             }
 569
 570             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 571                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 572                 const char *dp_port;
 573
 574                 if (!iter->is_tunnel) {
 575                     continue;
 576                 }
 577
 578                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 579                                                      namebuf, sizeof namebuf);
 580                 node = simap_find(&tmp_backers, dp_port);
 581                 if (node) {
 582                     simap_put(&backer->tnl_backers, dp_port, node->data);
 583                     simap_delete(&tmp_backers, node);
 584                     node = simap_find(&backer->tnl_backers, dp_port);
 585                 } else {
 586                     node = simap_find(&backer->tnl_backers, dp_port);
 587                     if (!node) {
 588                         odp_port_t odp_port = ODPP_NONE;
 589
 590                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 591                                            &odp_port)) {
 592                             simap_put(&backer->tnl_backers, dp_port,
 593                                       odp_to_u32(odp_port));
 594                             node = simap_find(&backer->tnl_backers, dp_port);
 595                         }
 596                     }
 597                 }
 598
 599                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 600                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 601                                          iter->odp_port,
 602                                          ovs_native_tunneling_is_on(ofproto), dp_port)) {
 603                     backer->need_revalidate = REV_RECONFIGURE;
 604                 }
 605             }
 606         }
 607
 608         SIMAP_FOR_EACH (node, &tmp_backers) {
 609             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 610         }
 611         simap_destroy(&tmp_backers);
 612
 613         switch (backer->need_revalidate) {
 614         case REV_RECONFIGURE:    COVERAGE_INC(rev_reconfigure);    break;
 615         case REV_STP:            COVERAGE_INC(rev_stp);            break;
 616         case REV_RSTP:           COVERAGE_INC(rev_rstp);           break;
 617         case REV_BOND:           COVERAGE_INC(rev_bond);           break;
 618         case REV_PORT_TOGGLED:   COVERAGE_INC(rev_port_toggled);   break;
 619         case REV_FLOW_TABLE:     COVERAGE_INC(rev_flow_table);     break;
 620         case REV_MAC_LEARNING:   COVERAGE_INC(rev_mac_learning);   break;
 621         case REV_MCAST_SNOOPING: COVERAGE_INC(rev_mcast_snooping); break;
 622         }
 623         backer->need_revalidate = 0;
 624
 625         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 626             struct ofport_dpif *ofport;
 627             struct ofbundle *bundle;
 628
 629             if (ofproto->backer != backer) {
 630                 continue;
 631             }
 632
 633             xlate_txn_start();
 634             xlate_ofproto_set(ofproto, ofproto->up.name,
 635                               ofproto->backer->dpif, ofproto->ml,
 636                               ofproto->stp, ofproto->rstp, ofproto->ms,
 637                               ofproto->mbridge, ofproto->sflow, ofproto->ipfix,
 638                               ofproto->netflow,
 639                               ofproto->up.forward_bpdu,
 640                               connmgr_has_in_band(ofproto->up.connmgr),
 641                               &ofproto->backer->support);
 642
 643             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 644                 xlate_bundle_set(ofproto, bundle, bundle->name,
 645                                  bundle->vlan_mode, bundle->vlan,
 646                                  bundle->trunks, bundle->use_priority_tags,
 647                                  bundle->bond, bundle->lacp,
 648                                  bundle->floodable);
 649             }
 650
 651             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 652                 int stp_port = ofport->stp_port
 653                     ? stp_port_no(ofport->stp_port)
 654                     : -1;
 655                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 656                                  ofport->up.ofp_port, ofport->odp_port,
 657                                  ofport->up.netdev, ofport->cfm, ofport->bfd,
 658                                  ofport->lldp, ofport->peer, stp_port,
 659                                  ofport->rstp_port, ofport->qdscp,
 660                                  ofport->n_qdscp, ofport->up.pp.config,
 661                                  ofport->up.pp.state, ofport->is_tunnel,
 662                                  ofport->may_enable);
 663             }
 664             xlate_txn_commit();
 665         }
 666
 667         udpif_revalidate(backer->udpif);
 668     }
 669
 670     process_dpif_port_changes(backer);
 671
 672     return 0;
 673 }
 674
 675 /* Check for and handle port changes in 'backer''s dpif. */
 676 static void
 677 process_dpif_port_changes(struct dpif_backer *backer)
 678 {
 679     for (;;) {
 680         char *devname;
 681         int error;
 682
 683         error = dpif_port_poll(backer->dpif, &devname);
 684         switch (error) {
 685         case EAGAIN:
 686             return;
 687
 688         case ENOBUFS:
 689             process_dpif_all_ports_changed(backer);
 690             break;
 691
 692         case 0:
 693             process_dpif_port_change(backer, devname);
 694             free(devname);
 695             break;
 696
 697         default:
 698             process_dpif_port_error(backer, error);
 699             break;
 700         }
 701     }
 702 }
 703
 704 static void
 705 process_dpif_all_ports_changed(struct dpif_backer *backer)
 706 {
 707     struct ofproto_dpif *ofproto;
 708     struct dpif_port dpif_port;
 709     struct dpif_port_dump dump;
 710     struct sset devnames;
 711     const char *devname;
 712
 713     sset_init(&devnames);
 714     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 715         if (ofproto->backer == backer) {
 716             struct ofport *ofport;
 717
 718             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 719                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 720             }
 721         }
 722     }
 723     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 724         sset_add(&devnames, dpif_port.name);
 725     }
 726
 727     SSET_FOR_EACH (devname, &devnames) {
 728         process_dpif_port_change(backer, devname);
 729     }
 730     sset_destroy(&devnames);
 731 }
 732
 733 static void
 734 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 735 {
 736     struct ofproto_dpif *ofproto;
 737     struct dpif_port port;
 738
 739     /* Don't report on the datapath's device. */
 740     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 741         return;
 742     }
 743
 744     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 745                    &all_ofproto_dpifs) {
 746         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 747             return;
 748         }
 749     }
 750
 751     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 752     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 753         /* The port was removed.  If we know the datapath,
 754          * report it through poll_set().  If we don't, it may be
 755          * notifying us of a removal we initiated, so ignore it.
 756          * If there's a pending ENOBUFS, let it stand, since
 757          * everything will be reevaluated. */
 758         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 759             sset_add(&ofproto->port_poll_set, devname);
 760             ofproto->port_poll_errno = 0;
 761         }
 762     } else if (!ofproto) {
 763         /* The port was added, but we don't know with which
 764          * ofproto we should associate it.  Delete it. */
 765         dpif_port_del(backer->dpif, port.port_no);
 766     } else {
 767         struct ofport_dpif *ofport;
 768
 769         ofport = ofport_dpif_cast(shash_find_data(
 770                                       &ofproto->up.port_by_name, devname));
 771         if (ofport
 772             && ofport->odp_port != port.port_no
 773             && !odp_port_to_ofport(backer, port.port_no))
 774         {
 775             /* 'ofport''s datapath port number has changed from
 776              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 777              * structures to match. */
 778             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 779             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 780             ofport->odp_port = port.port_no;
 781             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 782                         hash_odp_port(port.port_no));
 783             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 784             backer->need_revalidate = REV_RECONFIGURE;
 785         }
 786     }
 787     dpif_port_destroy(&port);
 788 }
 789
 790 /* Propagate 'error' to all ofprotos based on 'backer'. */
 791 static void
 792 process_dpif_port_error(struct dpif_backer *backer, int error)
 793 {
 794     struct ofproto_dpif *ofproto;
 795
 796     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 797         if (ofproto->backer == backer) {
 798             sset_clear(&ofproto->port_poll_set);
 799             ofproto->port_poll_errno = error;
 800         }
 801     }
 802 }
 803
 804 static void
 805 type_wait(const char *type)
 806 {
 807     struct dpif_backer *backer;
 808
 809     backer = shash_find_data(&all_dpif_backers, type);
 810     if (!backer) {
 811         /* This is not necessarily a problem, since backers are only
 812          * created on demand. */
 813         return;
 814     }
 815
 816     dpif_wait(backer->dpif);
 817 }
 818 \f
 819 /* Basic life-cycle. */
 820
 821 static int add_internal_flows(struct ofproto_dpif *);
 822
 823 static struct ofproto *
 824 alloc(void)
 825 {
 826     struct ofproto_dpif *ofproto = xzalloc(sizeof *ofproto);
 827     return &ofproto->up;
 828 }
 829
 830 static void
 831 dealloc(struct ofproto *ofproto_)
 832 {
 833     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
 834     free(ofproto);
 835 }
 836
 837 static void
 838 close_dpif_backer(struct dpif_backer *backer)
 839 {
 840     ovs_assert(backer->refcount > 0);
 841
 842     if (--backer->refcount) {
 843         return;
 844     }
 845
 846     udpif_destroy(backer->udpif);
 847
 848     simap_destroy(&backer->tnl_backers);
 849     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
 850     hmap_destroy(&backer->odp_to_ofport_map);
 851     shash_find_and_delete(&all_dpif_backers, backer->type);
 852     free(backer->type);
 853     free(backer->dp_version_string);
 854     dpif_close(backer->dpif);
 855     free(backer);
 856 }
 857
 858 /* Datapath port slated for removal from datapath. */
 859 struct odp_garbage {
 860     struct ovs_list list_node;
 861     odp_port_t odp_port;
 862 };
 863
 864 static bool check_variable_length_userdata(struct dpif_backer *backer);
 865 static void check_support(struct dpif_backer *backer);
 866
 867 static int
 868 open_dpif_backer(const char *type, struct dpif_backer **backerp)
 869 {
 870     struct dpif_backer *backer;
 871     struct dpif_port_dump port_dump;
 872     struct dpif_port port;
 873     struct shash_node *node;
 874     struct ovs_list garbage_list;
 875     struct odp_garbage *garbage;
 876
 877     struct sset names;
 878     char *backer_name;
 879     const char *name;
 880     int error;
 881
 882     recirc_init();
 883
 884     backer = shash_find_data(&all_dpif_backers, type);
 885     if (backer) {
 886         backer->refcount++;
 887         *backerp = backer;
 888         return 0;
 889     }
 890
 891     backer_name = xasprintf("ovs-%s", type);
 892
 893     /* Remove any existing datapaths, since we assume we're the only
 894      * userspace controlling the datapath. */
 895     sset_init(&names);
 896     dp_enumerate_names(type, &names);
 897     SSET_FOR_EACH(name, &names) {
 898         struct dpif *old_dpif;
 899
 900         /* Don't remove our backer if it exists. */
 901         if (!strcmp(name, backer_name)) {
 902             continue;
 903         }
 904
 905         if (dpif_open(name, type, &old_dpif)) {
 906             VLOG_WARN("couldn't open old datapath %s to remove it", name);
 907         } else {
 908             dpif_delete(old_dpif);
 909             dpif_close(old_dpif);
 910         }
 911     }
 912     sset_destroy(&names);
 913
 914     backer = xmalloc(sizeof *backer);
 915
 916     error = dpif_create_and_open(backer_name, type, &backer->dpif);
 917     free(backer_name);
 918     if (error) {
 919         VLOG_ERR("failed to open datapath of type %s: %s", type,
 920                  ovs_strerror(error));
 921         free(backer);
 922         return error;
 923     }
 924     backer->udpif = udpif_create(backer, backer->dpif);
 925
 926     backer->type = xstrdup(type);
 927     backer->refcount = 1;
 928     hmap_init(&backer->odp_to_ofport_map);
 929     ovs_rwlock_init(&backer->odp_to_ofport_lock);
 930     backer->need_revalidate = 0;
 931     simap_init(&backer->tnl_backers);
 932     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
 933     *backerp = backer;
 934
 935     if (backer->recv_set_enable) {
 936         dpif_flow_flush(backer->dpif);
 937     }
 938
 939     /* Loop through the ports already on the datapath and remove any
 940      * that we don't need anymore. */
 941     list_init(&garbage_list);
 942     dpif_port_dump_start(&port_dump, backer->dpif);
 943     while (dpif_port_dump_next(&port_dump, &port)) {
 944         node = shash_find(&init_ofp_ports, port.name);
 945         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
 946             garbage = xmalloc(sizeof *garbage);
 947             garbage->odp_port = port.port_no;
 948             list_push_front(&garbage_list, &garbage->list_node);
 949         }
 950     }
 951     dpif_port_dump_done(&port_dump);
 952
 953     LIST_FOR_EACH_POP (garbage, list_node, &garbage_list) {
 954         dpif_port_del(backer->dpif, garbage->odp_port);
 955         free(garbage);
 956     }
 957
 958     shash_add(&all_dpif_backers, type, backer);
 959
 960     check_support(backer);
 961     atomic_count_init(&backer->tnl_count, 0);
 962
 963     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 964     if (error) {
 965         VLOG_ERR("failed to listen on datapath of type %s: %s",
 966                  type, ovs_strerror(error));
 967         close_dpif_backer(backer);
 968         return error;
 969     }
 970
 971     if (backer->recv_set_enable) {
 972         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 973     }
 974
 975     /* This check fails if performed before udpif threads have been set,
 976      * as the kernel module checks that the 'pid' in userspace action
 977      * is non-zero. */
 978     backer->support.variable_length_userdata
 979         = check_variable_length_userdata(backer);
 980     backer->dp_version_string = dpif_get_dp_version(backer->dpif);
 981
 982     return error;
 983 }
 984
 985 bool
 986 ovs_native_tunneling_is_on(struct ofproto_dpif *ofproto)
 987 {
 988     return ofproto_use_tnl_push_pop && ofproto->backer->support.tnl_push_pop &&
 989            atomic_count_get(&ofproto->backer->tnl_count);
 990 }
 991
 992 /* Tests whether 'backer''s datapath supports recirculation.  Only newer
 993  * datapaths support OVS_KEY_ATTR_RECIRC_ID in keys.  We need to disable some
 994  * features on older datapaths that don't support this feature.
 995  *
 996  * Returns false if 'backer' definitely does not support recirculation, true if
 997  * it seems to support recirculation or if at least the error we get is
 998  * ambiguous. */
 999 static bool
1000 check_recirc(struct dpif_backer *backer)
1001 {
1002     struct flow flow;
1003     struct odputil_keybuf keybuf;
1004     struct ofpbuf key;
1005     bool enable_recirc;
1006
1007     memset(&flow, 0, sizeof flow);
1008     flow.recirc_id = 1;
1009     flow.dp_hash = 1;
1010
1011     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1012     odp_flow_key_from_flow(&key, &flow, NULL, 0, true);
1013     enable_recirc = dpif_probe_feature(backer->dpif, "recirculation", &key,
1014                                        NULL);
1015
1016     if (enable_recirc) {
1017         VLOG_INFO("%s: Datapath supports recirculation",
1018                   dpif_name(backer->dpif));
1019     } else {
1020         VLOG_INFO("%s: Datapath does not support recirculation",
1021                   dpif_name(backer->dpif));
1022     }
1023
1024     return enable_recirc;
1025 }
1026
1027 /* Tests whether 'dpif' supports unique flow ids. We can skip serializing
1028  * some flow attributes for datapaths that support this feature.
1029  *
1030  * Returns true if 'dpif' supports UFID for flow operations.
1031  * Returns false if  'dpif' does not support UFID. */
1032 static bool
1033 check_ufid(struct dpif_backer *backer)
1034 {
1035     struct flow flow;
1036     struct odputil_keybuf keybuf;
1037     struct ofpbuf key;
1038     ovs_u128 ufid;
1039     bool enable_ufid;
1040
1041     memset(&flow, 0, sizeof flow);
1042     flow.dl_type = htons(0x1234);
1043
1044     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1045     odp_flow_key_from_flow(&key, &flow, NULL, 0, true);
1046     dpif_flow_hash(backer->dpif, key.data, key.size, &ufid);
1047
1048     enable_ufid = dpif_probe_feature(backer->dpif, "UFID", &key, &ufid);
1049
1050     if (enable_ufid) {
1051         VLOG_INFO("%s: Datapath supports unique flow ids",
1052                   dpif_name(backer->dpif));
1053     } else {
1054         VLOG_INFO("%s: Datapath does not support unique flow ids",
1055                   dpif_name(backer->dpif));
1056     }
1057     return enable_ufid;
1058 }
1059
1060 /* Tests whether 'backer''s datapath supports variable-length
1061  * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.  We need
1062  * to disable some features on older datapaths that don't support this
1063  * feature.
1064  *
1065  * Returns false if 'backer' definitely does not support variable-length
1066  * userdata, true if it seems to support them or if at least the error we get
1067  * is ambiguous. */
1068 static bool
1069 check_variable_length_userdata(struct dpif_backer *backer)
1070 {
1071     struct eth_header *eth;
1072     struct ofpbuf actions;
1073     struct dpif_execute execute;
1074     struct dp_packet packet;
1075     size_t start;
1076     int error;
1077
1078     /* Compose a userspace action that will cause an ERANGE error on older
1079      * datapaths that don't support variable-length userdata.
1080      *
1081      * We really test for using userdata longer than 8 bytes, but older
1082      * datapaths accepted these, silently truncating the userdata to 8 bytes.
1083      * The same older datapaths rejected userdata shorter than 8 bytes, so we
1084      * test for that instead as a proxy for longer userdata support. */
1085     ofpbuf_init(&actions, 64);
1086     start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
1087     nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
1088                    dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
1089     nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
1090     nl_msg_end_nested(&actions, start);
1091
1092     /* Compose a dummy ethernet packet. */
1093     dp_packet_init(&packet, ETH_HEADER_LEN);
1094     eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1095     eth->eth_type = htons(0x1234);
1096
1097     /* Execute the actions.  On older datapaths this fails with ERANGE, on
1098      * newer datapaths it succeeds. */
1099     execute.actions = actions.data;
1100     execute.actions_len = actions.size;
1101     execute.packet = &packet;
1102     execute.needs_help = false;
1103     execute.probe = true;
1104
1105     error = dpif_execute(backer->dpif, &execute);
1106
1107     dp_packet_uninit(&packet);
1108     ofpbuf_uninit(&actions);
1109
1110     switch (error) {
1111     case 0:
1112         return true;
1113
1114     case ERANGE:
1115         /* Variable-length userdata is not supported. */
1116         VLOG_WARN("%s: datapath does not support variable-length userdata "
1117                   "feature (needs Linux 3.10+ or kernel module from OVS "
1118                   "1..11+).  The NXAST_SAMPLE action will be ignored.",
1119                   dpif_name(backer->dpif));
1120         return false;
1121
1122     default:
1123         /* Something odd happened.  We're not sure whether variable-length
1124          * userdata is supported.  Default to "yes". */
1125         VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1126                   dpif_name(backer->dpif), ovs_strerror(error));
1127         return true;
1128     }
1129 }
1130
1131 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1132  *
1133  * Returns the number of elements in a struct flow's mpls_lse field
1134  * if the datapath supports at least that many entries in an
1135  * MPLS label stack.
1136  * Otherwise returns the number of MPLS push actions supported by
1137  * the datapath. */
1138 static size_t
1139 check_max_mpls_depth(struct dpif_backer *backer)
1140 {
1141     struct flow flow;
1142     int n;
1143
1144     for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
1145         struct odputil_keybuf keybuf;
1146         struct ofpbuf key;
1147
1148         memset(&flow, 0, sizeof flow);
1149         flow.dl_type = htons(ETH_TYPE_MPLS);
1150         flow_set_mpls_bos(&flow, n, 1);
1151
1152         ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1153         odp_flow_key_from_flow(&key, &flow, NULL, 0, false);
1154         if (!dpif_probe_feature(backer->dpif, "MPLS", &key, NULL)) {
1155             break;
1156         }
1157     }
1158
1159     VLOG_INFO("%s: MPLS label stack length probed as %d",
1160               dpif_name(backer->dpif), n);
1161     return n;
1162 }
1163
1164 /* Tests whether 'backer''s datapath supports masked data in
1165  * OVS_ACTION_ATTR_SET actions.  We need to disable some features on older
1166  * datapaths that don't support this feature. */
1167 static bool
1168 check_masked_set_action(struct dpif_backer *backer)
1169 {
1170     struct eth_header *eth;
1171     struct ofpbuf actions;
1172     struct dpif_execute execute;
1173     struct dp_packet packet;
1174     int error;
1175     struct ovs_key_ethernet key, mask;
1176
1177     /* Compose a set action that will cause an EINVAL error on older
1178      * datapaths that don't support masked set actions.
1179      * Avoid using a full mask, as it could be translated to a non-masked
1180      * set action instead. */
1181     ofpbuf_init(&actions, 64);
1182     memset(&key, 0x53, sizeof key);
1183     memset(&mask, 0x7f, sizeof mask);
1184     commit_masked_set_action(&actions, OVS_KEY_ATTR_ETHERNET, &key, &mask,
1185                              sizeof key);
1186
1187     /* Compose a dummy ethernet packet. */
1188     dp_packet_init(&packet, ETH_HEADER_LEN);
1189     eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1190     eth->eth_type = htons(0x1234);
1191
1192     /* Execute the actions.  On older datapaths this fails with EINVAL, on
1193      * newer datapaths it succeeds. */
1194     execute.actions = actions.data;
1195     execute.actions_len = actions.size;
1196     execute.packet = &packet;
1197     execute.needs_help = false;
1198     execute.probe = true;
1199
1200     error = dpif_execute(backer->dpif, &execute);
1201
1202     dp_packet_uninit(&packet);
1203     ofpbuf_uninit(&actions);
1204
1205     if (error) {
1206         /* Masked set action is not supported. */
1207         VLOG_INFO("%s: datapath does not support masked set action feature.",
1208                   dpif_name(backer->dpif));
1209     }
1210     return !error;
1211 }
1212
1213 static void
1214 check_support(struct dpif_backer *backer)
1215 {
1216     /* This feature needs to be tested after udpif threads are set. */
1217     backer->support.variable_length_userdata = false;
1218
1219     backer->support.recirc = check_recirc(backer);
1220     backer->support.max_mpls_depth = check_max_mpls_depth(backer);
1221     backer->support.masked_set_action = check_masked_set_action(backer);
1222     backer->support.ufid = check_ufid(backer);
1223     backer->support.tnl_push_pop = dpif_supports_tnl_push_pop(backer->dpif);
1224 }
1225
1226 static int
1227 construct(struct ofproto *ofproto_)
1228 {
1229     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1230     struct shash_node *node, *next;
1231     int error;
1232
1233     /* Tunnel module can get used right after the udpif threads are running. */
1234     ofproto_tunnel_init();
1235
1236     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1237     if (error) {
1238         return error;
1239     }
1240
1241     atomic_init(&ofproto->tables_version, CLS_MIN_VERSION);
1242     ofproto->netflow = NULL;
1243     ofproto->sflow = NULL;
1244     ofproto->ipfix = NULL;
1245     ofproto->stp = NULL;
1246     ofproto->rstp = NULL;
1247     ofproto->dump_seq = 0;
1248     hmap_init(&ofproto->bundles);
1249     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1250     ofproto->ms = NULL;
1251     ofproto->mbridge = mbridge_create();
1252     ofproto->has_bonded_bundles = false;
1253     ofproto->lacp_enabled = false;
1254     ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1255     ovs_mutex_init(&ofproto->vsp_mutex);
1256
1257     guarded_list_init(&ofproto->pins);
1258
1259     ofproto_unixctl_init();
1260
1261     hmap_init(&ofproto->vlandev_map);
1262     hmap_init(&ofproto->realdev_vid_map);
1263
1264     sset_init(&ofproto->ports);
1265     sset_init(&ofproto->ghost_ports);
1266     sset_init(&ofproto->port_poll_set);
1267     ofproto->port_poll_errno = 0;
1268     ofproto->change_seq = 0;
1269     ofproto->pins_seq = seq_create();
1270     ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1271
1272
1273     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1274         struct iface_hint *iface_hint = node->data;
1275
1276         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1277             /* Check if the datapath already has this port. */
1278             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1279                 sset_add(&ofproto->ports, node->name);
1280             }
1281
1282             free(iface_hint->br_name);
1283             free(iface_hint->br_type);
1284             free(iface_hint);
1285             shash_delete(&init_ofp_ports, node);
1286         }
1287     }
1288
1289     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1290                 hash_string(ofproto->up.name, 0));
1291     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1292
1293     ofproto_init_tables(ofproto_, N_TABLES);
1294     error = add_internal_flows(ofproto);
1295
1296     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1297
1298     return error;
1299 }
1300
1301 static int
1302 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1303                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1304 {
1305     struct match match;
1306     int error;
1307     struct rule *rule;
1308
1309     match_init_catchall(&match);
1310     match_set_reg(&match, 0, id);
1311
1312     error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, 0, ofpacts,
1313                                            &rule);
1314     *rulep = error ? NULL : rule_dpif_cast(rule);
1315
1316     return error;
1317 }
1318
1319 static int
1320 add_internal_flows(struct ofproto_dpif *ofproto)
1321 {
1322     struct ofpact_controller *controller;
1323     uint64_t ofpacts_stub[128 / 8];
1324     struct ofpbuf ofpacts;
1325     struct rule *unused_rulep OVS_UNUSED;
1326     struct match match;
1327     int error;
1328     int id;
1329
1330     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1331     id = 1;
1332
1333     controller = ofpact_put_CONTROLLER(&ofpacts);
1334     controller->max_len = UINT16_MAX;
1335     controller->controller_id = 0;
1336     controller->reason = OFPR_NO_MATCH;
1337     ofpact_pad(&ofpacts);
1338
1339     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1340                                    &ofproto->miss_rule);
1341     if (error) {
1342         return error;
1343     }
1344
1345     ofpbuf_clear(&ofpacts);
1346     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1347                                    &ofproto->no_packet_in_rule);
1348     if (error) {
1349         return error;
1350     }
1351
1352     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1353                                    &ofproto->drop_frags_rule);
1354     if (error) {
1355         return error;
1356     }
1357
1358     /* Drop any run away non-recirc rule lookups. Recirc_id has to be
1359      * zero when reaching this rule.
1360      *
1361      * (priority=2), recirc_id=0, actions=drop
1362      */
1363     ofpbuf_clear(&ofpacts);
1364     match_init_catchall(&match);
1365     match_set_recirc_id(&match, 0);
1366     error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, 0, &ofpacts,
1367                                            &unused_rulep);
1368     return error;
1369 }
1370
1371 static void
1372 destruct(struct ofproto *ofproto_)
1373 {
1374     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1375     struct ofproto_packet_in *pin;
1376     struct rule_dpif *rule;
1377     struct oftable *table;
1378     struct ovs_list pins;
1379
1380     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1381     xlate_txn_start();
1382     xlate_remove_ofproto(ofproto);
1383     xlate_txn_commit();
1384
1385     /* Ensure that the upcall processing threads have no remaining references
1386      * to the ofproto or anything in it. */
1387     udpif_synchronize(ofproto->backer->udpif);
1388
1389     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1390
1391     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1392         CLS_FOR_EACH (rule, up.cr, &table->cls) {
1393             ofproto_rule_delete(&ofproto->up, &rule->up);
1394         }
1395     }
1396
1397     guarded_list_pop_all(&ofproto->pins, &pins);
1398     LIST_FOR_EACH_POP (pin, list_node, &pins) {
1399         free(CONST_CAST(void *, pin->up.packet));
1400         free(pin);
1401     }
1402     guarded_list_destroy(&ofproto->pins);
1403
1404     recirc_free_ofproto(ofproto, ofproto->up.name);
1405
1406     mbridge_unref(ofproto->mbridge);
1407
1408     netflow_unref(ofproto->netflow);
1409     dpif_sflow_unref(ofproto->sflow);
1410     dpif_ipfix_unref(ofproto->ipfix);
1411     hmap_destroy(&ofproto->bundles);
1412     mac_learning_unref(ofproto->ml);
1413     mcast_snooping_unref(ofproto->ms);
1414
1415     hmap_destroy(&ofproto->vlandev_map);
1416     hmap_destroy(&ofproto->realdev_vid_map);
1417
1418     sset_destroy(&ofproto->ports);
1419     sset_destroy(&ofproto->ghost_ports);
1420     sset_destroy(&ofproto->port_poll_set);
1421
1422     ovs_mutex_destroy(&ofproto->stats_mutex);
1423     ovs_mutex_destroy(&ofproto->vsp_mutex);
1424
1425     seq_destroy(ofproto->pins_seq);
1426
1427     close_dpif_backer(ofproto->backer);
1428 }
1429
1430 static int
1431 run(struct ofproto *ofproto_)
1432 {
1433     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1434     uint64_t new_seq, new_dump_seq;
1435
1436     if (mbridge_need_revalidate(ofproto->mbridge)) {
1437         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1438         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1439         mac_learning_flush(ofproto->ml);
1440         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1441         mcast_snooping_mdb_flush(ofproto->ms);
1442     }
1443
1444     /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
1445      * flow restore.  Even though nothing is processed during flow restore,
1446      * all queued 'pins' will be handled immediately when flow restore
1447      * completes. */
1448     ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1449
1450     /* Do not perform any periodic activity required by 'ofproto' while
1451      * waiting for flow restore to complete. */
1452     if (!ofproto_get_flow_restore_wait()) {
1453         struct ofproto_packet_in *pin;
1454         struct ovs_list pins;
1455
1456         guarded_list_pop_all(&ofproto->pins, &pins);
1457         LIST_FOR_EACH_POP (pin, list_node, &pins) {
1458             connmgr_send_packet_in(ofproto->up.connmgr, pin);
1459             free(CONST_CAST(void *, pin->up.packet));
1460             free(pin);
1461         }
1462     }
1463
1464     if (ofproto->netflow) {
1465         netflow_run(ofproto->netflow);
1466     }
1467     if (ofproto->sflow) {
1468         dpif_sflow_run(ofproto->sflow);
1469     }
1470     if (ofproto->ipfix) {
1471         dpif_ipfix_run(ofproto->ipfix);
1472     }
1473
1474     new_seq = seq_read(connectivity_seq_get());
1475     if (ofproto->change_seq != new_seq) {
1476         struct ofport_dpif *ofport;
1477
1478         HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1479             port_run(ofport);
1480         }
1481
1482         ofproto->change_seq = new_seq;
1483     }
1484     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1485         struct ofbundle *bundle;
1486
1487         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1488             bundle_run(bundle);
1489         }
1490     }
1491
1492     stp_run(ofproto);
1493     rstp_run(ofproto);
1494     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1495     if (mac_learning_run(ofproto->ml)) {
1496         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1497     }
1498     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1499
1500     if (mcast_snooping_run(ofproto->ms)) {
1501         ofproto->backer->need_revalidate = REV_MCAST_SNOOPING;
1502     }
1503
1504     new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1505     if (ofproto->dump_seq != new_dump_seq) {
1506         struct rule *rule, *next_rule;
1507
1508         /* We know stats are relatively fresh, so now is a good time to do some
1509          * periodic work. */
1510         ofproto->dump_seq = new_dump_seq;
1511
1512         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1513          * has passed. */
1514         ovs_mutex_lock(&ofproto_mutex);
1515         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1516                             &ofproto->up.expirable) {
1517             rule_expire(rule_dpif_cast(rule));
1518         }
1519         ovs_mutex_unlock(&ofproto_mutex);
1520
1521         /* All outstanding data in existing flows has been accounted, so it's a
1522          * good time to do bond rebalancing. */
1523         if (ofproto->has_bonded_bundles) {
1524             struct ofbundle *bundle;
1525
1526             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1527                 if (bundle->bond) {
1528                     bond_rebalance(bundle->bond);
1529                 }
1530             }
1531         }
1532     }
1533     return 0;
1534 }
1535
1536 static void
1537 wait(struct ofproto *ofproto_)
1538 {
1539     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1540
1541     if (ofproto_get_flow_restore_wait()) {
1542         return;
1543     }
1544
1545     if (ofproto->sflow) {
1546         dpif_sflow_wait(ofproto->sflow);
1547     }
1548     if (ofproto->ipfix) {
1549         dpif_ipfix_wait(ofproto->ipfix);
1550     }
1551     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1552         struct ofbundle *bundle;
1553
1554         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1555             bundle_wait(bundle);
1556         }
1557     }
1558     if (ofproto->netflow) {
1559         netflow_wait(ofproto->netflow);
1560     }
1561     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1562     mac_learning_wait(ofproto->ml);
1563     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1564     mcast_snooping_wait(ofproto->ms);
1565     stp_wait(ofproto);
1566     if (ofproto->backer->need_revalidate) {
1567         /* Shouldn't happen, but if it does just go around again. */
1568         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1569         poll_immediate_wake();
1570     }
1571
1572     seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1573     seq_wait(ofproto->pins_seq, ofproto->pins_seqno);
1574 }
1575
1576 static void
1577 type_get_memory_usage(const char *type, struct simap *usage)
1578 {
1579     struct dpif_backer *backer;
1580
1581     backer = shash_find_data(&all_dpif_backers, type);
1582     if (backer) {
1583         udpif_get_memory_usage(backer->udpif, usage);
1584     }
1585 }
1586
1587 static void
1588 flush(struct ofproto *ofproto_)
1589 {
1590     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1591     struct dpif_backer *backer = ofproto->backer;
1592
1593     if (backer) {
1594         udpif_flush(backer->udpif);
1595     }
1596 }
1597
1598 static void
1599 query_tables(struct ofproto *ofproto,
1600              struct ofputil_table_features *features,
1601              struct ofputil_table_stats *stats)
1602 {
1603     strcpy(features->name, "classifier");
1604
1605     if (stats) {
1606         int i;
1607
1608         for (i = 0; i < ofproto->n_tables; i++) {
1609             unsigned long missed, matched;
1610
1611             atomic_read_relaxed(&ofproto->tables[i].n_matched, &matched);
1612             atomic_read_relaxed(&ofproto->tables[i].n_missed, &missed);
1613
1614             stats[i].matched_count = matched;
1615             stats[i].lookup_count = matched + missed;
1616         }
1617     }
1618 }
1619
1620 static void
1621 set_tables_version(struct ofproto *ofproto_, long long version)
1622 {
1623     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1624
1625     atomic_store_relaxed(&ofproto->tables_version, version);
1626 }
1627
1628
1629 static struct ofport *
1630 port_alloc(void)
1631 {
1632     struct ofport_dpif *port = xzalloc(sizeof *port);
1633     return &port->up;
1634 }
1635
1636 static void
1637 port_dealloc(struct ofport *port_)
1638 {
1639     struct ofport_dpif *port = ofport_dpif_cast(port_);
1640     free(port);
1641 }
1642
1643 static int
1644 port_construct(struct ofport *port_)
1645 {
1646     struct ofport_dpif *port = ofport_dpif_cast(port_);
1647     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1648     const struct netdev *netdev = port->up.netdev;
1649     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1650     const char *dp_port_name;
1651     struct dpif_port dpif_port;
1652     int error;
1653
1654     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1655     port->bundle = NULL;
1656     port->cfm = NULL;
1657     port->bfd = NULL;
1658     port->lldp = NULL;
1659     port->may_enable = false;
1660     port->stp_port = NULL;
1661     port->stp_state = STP_DISABLED;
1662     port->rstp_port = NULL;
1663     port->rstp_state = RSTP_DISABLED;
1664     port->is_tunnel = false;
1665     port->peer = NULL;
1666     port->qdscp = NULL;
1667     port->n_qdscp = 0;
1668     port->realdev_ofp_port = 0;
1669     port->vlandev_vid = 0;
1670     port->carrier_seq = netdev_get_carrier_resets(netdev);
1671     port->is_layer3 = netdev_vport_is_layer3(netdev);
1672
1673     if (netdev_vport_is_patch(netdev)) {
1674         /* By bailing out here, we don't submit the port to the sFlow module
1675          * to be considered for counter polling export.  This is correct
1676          * because the patch port represents an interface that sFlow considers
1677          * to be "internal" to the switch as a whole, and therefore not a
1678          * candidate for counter polling. */
1679         port->odp_port = ODPP_NONE;
1680         ofport_update_peer(port);
1681         return 0;
1682     }
1683
1684     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1685     error = dpif_port_query_by_name(ofproto->backer->dpif, dp_port_name,
1686                                     &dpif_port);
1687     if (error) {
1688         return error;
1689     }
1690
1691     port->odp_port = dpif_port.port_no;
1692
1693     if (netdev_get_tunnel_config(netdev)) {
1694         atomic_count_inc(&ofproto->backer->tnl_count);
1695         error = tnl_port_add(port, port->up.netdev, port->odp_port,
1696                              ovs_native_tunneling_is_on(ofproto), dp_port_name);
1697         if (error) {
1698             atomic_count_dec(&ofproto->backer->tnl_count);
1699             dpif_port_destroy(&dpif_port);
1700             return error;
1701         }
1702
1703         port->is_tunnel = true;
1704         if (ofproto->ipfix) {
1705            dpif_ipfix_add_tunnel_port(ofproto->ipfix, port_, port->odp_port);
1706         }
1707     } else {
1708         /* Sanity-check that a mapping doesn't already exist.  This
1709          * shouldn't happen for non-tunnel ports. */
1710         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1711             VLOG_ERR("port %s already has an OpenFlow port number",
1712                      dpif_port.name);
1713             dpif_port_destroy(&dpif_port);
1714             return EBUSY;
1715         }
1716
1717         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1718         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1719                     hash_odp_port(port->odp_port));
1720         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1721     }
1722     dpif_port_destroy(&dpif_port);
1723
1724     if (ofproto->sflow) {
1725         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1726     }
1727
1728     return 0;
1729 }
1730
1731 static void
1732 port_destruct(struct ofport *port_)
1733 {
1734     struct ofport_dpif *port = ofport_dpif_cast(port_);
1735     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1736     const char *devname = netdev_get_name(port->up.netdev);
1737     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1738     const char *dp_port_name;
1739
1740     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1741     xlate_txn_start();
1742     xlate_ofport_remove(port);
1743     xlate_txn_commit();
1744
1745     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1746                                               sizeof namebuf);
1747     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1748         /* The underlying device is still there, so delete it.  This
1749          * happens when the ofproto is being destroyed, since the caller
1750          * assumes that removal of attached ports will happen as part of
1751          * destruction. */
1752         if (!port->is_tunnel) {
1753             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1754         }
1755     }
1756
1757     if (port->peer) {
1758         port->peer->peer = NULL;
1759         port->peer = NULL;
1760     }
1761
1762     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1763         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1764         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1765         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1766     }
1767
1768     if (port->is_tunnel) {
1769         atomic_count_dec(&ofproto->backer->tnl_count);
1770     }
1771
1772     if (port->is_tunnel && ofproto->ipfix) {
1773        dpif_ipfix_del_tunnel_port(ofproto->ipfix, port->odp_port);
1774     }
1775
1776     tnl_port_del(port);
1777     sset_find_and_delete(&ofproto->ports, devname);
1778     sset_find_and_delete(&ofproto->ghost_ports, devname);
1779     bundle_remove(port_);
1780     set_cfm(port_, NULL);
1781     set_bfd(port_, NULL);
1782     set_lldp(port_, NULL);
1783     if (port->stp_port) {
1784         stp_port_disable(port->stp_port);
1785     }
1786     set_rstp_port(port_, NULL);
1787     if (ofproto->sflow) {
1788         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1789     }
1790
1791     free(port->qdscp);
1792 }
1793
1794 static void
1795 port_modified(struct ofport *port_)
1796 {
1797     struct ofport_dpif *port = ofport_dpif_cast(port_);
1798     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1799     const char *dp_port_name;
1800     struct netdev *netdev = port->up.netdev;
1801
1802     if (port->bundle && port->bundle->bond) {
1803         bond_slave_set_netdev(port->bundle->bond, port, netdev);
1804     }
1805
1806     if (port->cfm) {
1807         cfm_set_netdev(port->cfm, netdev);
1808     }
1809
1810     if (port->bfd) {
1811         bfd_set_netdev(port->bfd, netdev);
1812     }
1813
1814     ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1815                                      port->lldp, port->up.pp.hw_addr);
1816
1817     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1818
1819     if (port->is_tunnel) {
1820         struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1821
1822         if (tnl_port_reconfigure(port, netdev, port->odp_port,
1823                                  ovs_native_tunneling_is_on(ofproto),
1824                                  dp_port_name)) {
1825             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1826         }
1827     }
1828
1829     ofport_update_peer(port);
1830 }
1831
1832 static void
1833 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1834 {
1835     struct ofport_dpif *port = ofport_dpif_cast(port_);
1836     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1837     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1838
1839     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1840                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1841                    OFPUTIL_PC_NO_PACKET_IN)) {
1842         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1843
1844         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1845             bundle_update(port->bundle);
1846         }
1847     }
1848 }
1849
1850 static int
1851 set_sflow(struct ofproto *ofproto_,
1852           const struct ofproto_sflow_options *sflow_options)
1853 {
1854     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1855     struct dpif_sflow *ds = ofproto->sflow;
1856
1857     if (sflow_options) {
1858         uint32_t old_probability = ds ? dpif_sflow_get_probability(ds) : 0;
1859         if (!ds) {
1860             struct ofport_dpif *ofport;
1861
1862             ds = ofproto->sflow = dpif_sflow_create();
1863             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1864                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1865             }
1866         }
1867         dpif_sflow_set_options(ds, sflow_options);
1868         if (dpif_sflow_get_probability(ds) != old_probability) {
1869             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1870         }
1871     } else {
1872         if (ds) {
1873             dpif_sflow_unref(ds);
1874             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1875             ofproto->sflow = NULL;
1876         }
1877     }
1878     return 0;
1879 }
1880
1881 static int
1882 set_ipfix(
1883     struct ofproto *ofproto_,
1884     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1885     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1886     size_t n_flow_exporters_options)
1887 {
1888     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1889     struct dpif_ipfix *di = ofproto->ipfix;
1890     bool has_options = bridge_exporter_options || flow_exporters_options;
1891     bool new_di = false;
1892
1893     if (has_options && !di) {
1894         di = ofproto->ipfix = dpif_ipfix_create();
1895         new_di = true;
1896     }
1897
1898     if (di) {
1899         /* Call set_options in any case to cleanly flush the flow
1900          * caches in the last exporters that are to be destroyed. */
1901         dpif_ipfix_set_options(
1902             di, bridge_exporter_options, flow_exporters_options,
1903             n_flow_exporters_options);
1904
1905         /* Add tunnel ports only when a new ipfix created */
1906         if (new_di == true) {
1907             struct ofport_dpif *ofport;
1908             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1909                 if (ofport->is_tunnel == true) {
1910                     dpif_ipfix_add_tunnel_port(di, &ofport->up, ofport->odp_port);
1911                 }
1912             }
1913         }
1914
1915         if (!has_options) {
1916             dpif_ipfix_unref(di);
1917             ofproto->ipfix = NULL;
1918         }
1919     }
1920
1921     return 0;
1922 }
1923
1924 static int
1925 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1926 {
1927     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1928     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1929     struct cfm *old = ofport->cfm;
1930     int error = 0;
1931
1932     if (s) {
1933         if (!ofport->cfm) {
1934             ofport->cfm = cfm_create(ofport->up.netdev);
1935         }
1936
1937         if (cfm_configure(ofport->cfm, s)) {
1938             error = 0;
1939             goto out;
1940         }
1941
1942         error = EINVAL;
1943     }
1944     cfm_unref(ofport->cfm);
1945     ofport->cfm = NULL;
1946 out:
1947     if (ofport->cfm != old) {
1948         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1949     }
1950     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1951                                      ofport->lldp, ofport->up.pp.hw_addr);
1952     return error;
1953 }
1954
1955 static bool
1956 cfm_status_changed(struct ofport *ofport_)
1957 {
1958     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1959
1960     return ofport->cfm ? cfm_check_status_change(ofport->cfm) : true;
1961 }
1962
1963 static int
1964 get_cfm_status(const struct ofport *ofport_,
1965                struct cfm_status *status)
1966 {
1967     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1968     int ret = 0;
1969
1970     if (ofport->cfm) {
1971         cfm_get_status(ofport->cfm, status);
1972     } else {
1973         ret = ENOENT;
1974     }
1975
1976     return ret;
1977 }
1978
1979 static int
1980 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1981 {
1982     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1983     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1984     struct bfd *old;
1985
1986     old = ofport->bfd;
1987     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1988                                 cfg, ofport->up.netdev);
1989     if (ofport->bfd != old) {
1990         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1991     }
1992     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1993                                      ofport->lldp, ofport->up.pp.hw_addr);
1994     return 0;
1995 }
1996
1997 static bool
1998 bfd_status_changed(struct ofport *ofport_)
1999 {
2000     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2001
2002     return ofport->bfd ? bfd_check_status_change(ofport->bfd) : true;
2003 }
2004
2005 static int
2006 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2007 {
2008     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2009     int ret = 0;
2010
2011     if (ofport->bfd) {
2012         bfd_get_status(ofport->bfd, smap);
2013     } else {
2014         ret = ENOENT;
2015     }
2016
2017     return ret;
2018 }
2019
2020 static int
2021 set_lldp(struct ofport *ofport_,
2022          const struct smap *cfg)
2023 {
2024     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2025     int error = 0;
2026
2027     if (cfg) {
2028         if (!ofport->lldp) {
2029             struct ofproto_dpif *ofproto;
2030
2031             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2032             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2033             ofport->lldp = lldp_create(ofport->up.netdev, ofport_->mtu, cfg);
2034         }
2035
2036         if (!lldp_configure(ofport->lldp, cfg)) {
2037             error = EINVAL;
2038         }
2039     }
2040     if (error) {
2041         lldp_unref(ofport->lldp);
2042         ofport->lldp = NULL;
2043     }
2044
2045     ofproto_dpif_monitor_port_update(ofport,
2046                                      ofport->bfd,
2047                                      ofport->cfm,
2048                                      ofport->lldp,
2049                                      ofport->up.pp.hw_addr);
2050     return error;
2051 }
2052
2053 static bool
2054 get_lldp_status(const struct ofport *ofport_,
2055                struct lldp_status *status OVS_UNUSED)
2056 {
2057     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2058
2059     return ofport->lldp ? true : false;
2060 }
2061
2062 static int
2063 set_aa(struct ofproto *ofproto OVS_UNUSED,
2064        const struct aa_settings *s)
2065 {
2066     return aa_configure(s);
2067 }
2068
2069 static int
2070 aa_mapping_set(struct ofproto *ofproto_ OVS_UNUSED, void *aux,
2071                const struct aa_mapping_settings *s)
2072 {
2073     return aa_mapping_register(aux, s);
2074 }
2075
2076 static int
2077 aa_mapping_unset(struct ofproto *ofproto OVS_UNUSED, void *aux)
2078 {
2079     return aa_mapping_unregister(aux);
2080 }
2081
2082 static int
2083 aa_vlan_get_queued(struct ofproto *ofproto OVS_UNUSED, struct ovs_list *list)
2084 {
2085     return aa_get_vlan_queued(list);
2086 }
2087
2088 static unsigned int
2089 aa_vlan_get_queue_size(struct ofproto *ofproto OVS_UNUSED)
2090 {
2091     return aa_get_vlan_queue_size();
2092 }
2093
2094 \f
2095 /* Spanning Tree. */
2096
2097 /* Called while rstp_mutex is held. */
2098 static void
2099 rstp_send_bpdu_cb(struct dp_packet *pkt, void *ofport_, void *ofproto_)
2100 {
2101     struct ofproto_dpif *ofproto = ofproto_;
2102     struct ofport_dpif *ofport = ofport_;
2103     struct eth_header *eth = dp_packet_l2(pkt);
2104
2105     netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2106     if (eth_addr_is_zero(eth->eth_src)) {
2107         VLOG_WARN_RL(&rl, "%s port %d: cannot send RSTP BPDU on a port which "
2108                      "does not have a configured source MAC address.",
2109                      ofproto->up.name, ofp_to_u16(ofport->up.ofp_port));
2110     } else {
2111         ofproto_dpif_send_packet(ofport, pkt);
2112     }
2113     dp_packet_delete(pkt);
2114 }
2115
2116 static void
2117 send_bpdu_cb(struct dp_packet *pkt, int port_num, void *ofproto_)
2118 {
2119     struct ofproto_dpif *ofproto = ofproto_;
2120     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2121     struct ofport_dpif *ofport;
2122
2123     ofport = stp_port_get_aux(sp);
2124     if (!ofport) {
2125         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2126                      ofproto->up.name, port_num);
2127     } else {
2128         struct eth_header *eth = dp_packet_l2(pkt);
2129
2130         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2131         if (eth_addr_is_zero(eth->eth_src)) {
2132             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2133                          "with unknown MAC", ofproto->up.name, port_num);
2134         } else {
2135             ofproto_dpif_send_packet(ofport, pkt);
2136         }
2137     }
2138     dp_packet_delete(pkt);
2139 }
2140
2141 /* Configure RSTP on 'ofproto_' using the settings defined in 's'. */
2142 static void
2143 set_rstp(struct ofproto *ofproto_, const struct ofproto_rstp_settings *s)
2144 {
2145     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2146
2147     /* Only revalidate flows if the configuration changed. */
2148     if (!s != !ofproto->rstp) {
2149         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2150     }
2151
2152     if (s) {
2153         if (!ofproto->rstp) {
2154             ofproto->rstp = rstp_create(ofproto_->name, s->address,
2155                                         rstp_send_bpdu_cb, ofproto);
2156             ofproto->rstp_last_tick = time_msec();
2157         }
2158         rstp_set_bridge_address(ofproto->rstp, s->address);
2159         rstp_set_bridge_priority(ofproto->rstp, s->priority);
2160         rstp_set_bridge_ageing_time(ofproto->rstp, s->ageing_time);
2161         rstp_set_bridge_force_protocol_version(ofproto->rstp,
2162                                                s->force_protocol_version);
2163         rstp_set_bridge_max_age(ofproto->rstp, s->bridge_max_age);
2164         rstp_set_bridge_forward_delay(ofproto->rstp, s->bridge_forward_delay);
2165         rstp_set_bridge_transmit_hold_count(ofproto->rstp,
2166                                             s->transmit_hold_count);
2167     } else {
2168         struct ofport *ofport;
2169         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2170             set_rstp_port(ofport, NULL);
2171         }
2172         rstp_unref(ofproto->rstp);
2173         ofproto->rstp = NULL;
2174     }
2175 }
2176
2177 static void
2178 get_rstp_status(struct ofproto *ofproto_, struct ofproto_rstp_status *s)
2179 {
2180     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2181
2182     if (ofproto->rstp) {
2183         s->enabled = true;
2184         s->root_id = rstp_get_root_id(ofproto->rstp);
2185         s->bridge_id = rstp_get_bridge_id(ofproto->rstp);
2186         s->designated_id = rstp_get_designated_id(ofproto->rstp);
2187         s->root_path_cost = rstp_get_root_path_cost(ofproto->rstp);
2188         s->designated_port_id = rstp_get_designated_port_id(ofproto->rstp);
2189         s->bridge_port_id = rstp_get_bridge_port_id(ofproto->rstp);
2190     } else {
2191         s->enabled = false;
2192     }
2193 }
2194
2195 static void
2196 update_rstp_port_state(struct ofport_dpif *ofport)
2197 {
2198     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2199     enum rstp_state state;
2200
2201     /* Figure out new state. */
2202     state = ofport->rstp_port ? rstp_port_get_state(ofport->rstp_port)
2203         : RSTP_DISABLED;
2204
2205     /* Update state. */
2206     if (ofport->rstp_state != state) {
2207         enum ofputil_port_state of_state;
2208         bool fwd_change;
2209
2210         VLOG_DBG("port %s: RSTP state changed from %s to %s",
2211                  netdev_get_name(ofport->up.netdev),
2212                  rstp_state_name(ofport->rstp_state),
2213                  rstp_state_name(state));
2214
2215         if (rstp_learn_in_state(ofport->rstp_state)
2216             != rstp_learn_in_state(state)) {
2217             /* XXX: Learning action flows should also be flushed. */
2218             if (ofport->bundle) {
2219                 if (!rstp_shift_root_learned_address(ofproto->rstp)
2220                     || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2221                     bundle_flush_macs(ofport->bundle, false);
2222                 }
2223             }
2224         }
2225         fwd_change = rstp_forward_in_state(ofport->rstp_state)
2226             != rstp_forward_in_state(state);
2227
2228         ofproto->backer->need_revalidate = REV_RSTP;
2229         ofport->rstp_state = state;
2230
2231         if (fwd_change && ofport->bundle) {
2232             bundle_update(ofport->bundle);
2233         }
2234
2235         /* Update the RSTP state bits in the OpenFlow port description. */
2236         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2237         of_state |= (state == RSTP_LEARNING ? OFPUTIL_PS_STP_LEARN
2238                 : state == RSTP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2239                 : state == RSTP_DISCARDING ?  OFPUTIL_PS_STP_LISTEN
2240                 : 0);
2241         ofproto_port_set_state(&ofport->up, of_state);
2242     }
2243 }
2244
2245 static void
2246 rstp_run(struct ofproto_dpif *ofproto)
2247 {
2248     if (ofproto->rstp) {
2249         long long int now = time_msec();
2250         long long int elapsed = now - ofproto->rstp_last_tick;
2251         struct rstp_port *rp;
2252         struct ofport_dpif *ofport;
2253
2254         /* Every second, decrease the values of the timers. */
2255         if (elapsed >= 1000) {
2256             rstp_tick_timers(ofproto->rstp);
2257             ofproto->rstp_last_tick = now;
2258         }
2259         rp = NULL;
2260         while ((ofport = rstp_get_next_changed_port_aux(ofproto->rstp, &rp))) {
2261             update_rstp_port_state(ofport);
2262         }
2263         rp = NULL;
2264         ofport = NULL;
2265         /* FIXME: This check should be done on-event (i.e., when setting
2266          * p->fdb_flush) and not periodically.
2267          */
2268         while ((ofport = rstp_check_and_reset_fdb_flush(ofproto->rstp, &rp))) {
2269             if (!rstp_shift_root_learned_address(ofproto->rstp)
2270                 || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2271                 bundle_flush_macs(ofport->bundle, false);
2272             }
2273         }
2274
2275         if (rstp_shift_root_learned_address(ofproto->rstp)) {
2276             bundle_move(((struct ofport_dpif *)rstp_get_old_root_aux(ofproto->rstp))->bundle,
2277                         ((struct ofport_dpif *)rstp_get_new_root_aux(ofproto->rstp))->bundle);
2278             rstp_reset_root_changed(ofproto->rstp);
2279         }
2280     }
2281 }
2282
2283 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2284 static int
2285 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2286 {
2287     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2288
2289     /* Only revalidate flows if the configuration changed. */
2290     if (!s != !ofproto->stp) {
2291         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2292     }
2293
2294     if (s) {
2295         if (!ofproto->stp) {
2296             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2297                                       send_bpdu_cb, ofproto);
2298             ofproto->stp_last_tick = time_msec();
2299         }
2300
2301         stp_set_bridge_id(ofproto->stp, s->system_id);
2302         stp_set_bridge_priority(ofproto->stp, s->priority);
2303         stp_set_hello_time(ofproto->stp, s->hello_time);
2304         stp_set_max_age(ofproto->stp, s->max_age);
2305         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2306     }  else {
2307         struct ofport *ofport;
2308
2309         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2310             set_stp_port(ofport, NULL);
2311         }
2312
2313         stp_unref(ofproto->stp);
2314         ofproto->stp = NULL;
2315     }
2316
2317     return 0;
2318 }
2319
2320 static int
2321 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2322 {
2323     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2324
2325     if (ofproto->stp) {
2326         s->enabled = true;
2327         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2328         s->designated_root = stp_get_designated_root(ofproto->stp);
2329         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2330     } else {
2331         s->enabled = false;
2332     }
2333
2334     return 0;
2335 }
2336
2337 static void
2338 update_stp_port_state(struct ofport_dpif *ofport)
2339 {
2340     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2341     enum stp_state state;
2342
2343     /* Figure out new state. */
2344     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2345                              : STP_DISABLED;
2346
2347     /* Update state. */
2348     if (ofport->stp_state != state) {
2349         enum ofputil_port_state of_state;
2350         bool fwd_change;
2351
2352         VLOG_DBG("port %s: STP state changed from %s to %s",
2353                  netdev_get_name(ofport->up.netdev),
2354                  stp_state_name(ofport->stp_state),
2355                  stp_state_name(state));
2356         if (stp_learn_in_state(ofport->stp_state)
2357                 != stp_learn_in_state(state)) {
2358             /* xxx Learning action flows should also be flushed. */
2359             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2360             mac_learning_flush(ofproto->ml);
2361             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2362             mcast_snooping_mdb_flush(ofproto->ms);
2363         }
2364         fwd_change = stp_forward_in_state(ofport->stp_state)
2365                         != stp_forward_in_state(state);
2366
2367         ofproto->backer->need_revalidate = REV_STP;
2368         ofport->stp_state = state;
2369         ofport->stp_state_entered = time_msec();
2370
2371         if (fwd_change && ofport->bundle) {
2372             bundle_update(ofport->bundle);
2373         }
2374
2375         /* Update the STP state bits in the OpenFlow port description. */
2376         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2377         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2378                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2379                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2380                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2381                      : 0);
2382         ofproto_port_set_state(&ofport->up, of_state);
2383     }
2384 }
2385
2386 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2387  * caller is responsible for assigning STP port numbers and ensuring
2388  * there are no duplicates. */
2389 static int
2390 set_stp_port(struct ofport *ofport_,
2391              const struct ofproto_port_stp_settings *s)
2392 {
2393     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2394     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2395     struct stp_port *sp = ofport->stp_port;
2396
2397     if (!s || !s->enable) {
2398         if (sp) {
2399             ofport->stp_port = NULL;
2400             stp_port_disable(sp);
2401             update_stp_port_state(ofport);
2402         }
2403         return 0;
2404     } else if (sp && stp_port_no(sp) != s->port_num
2405                && ofport == stp_port_get_aux(sp)) {
2406         /* The port-id changed, so disable the old one if it's not
2407          * already in use by another port. */
2408         stp_port_disable(sp);
2409     }
2410
2411     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2412
2413     /* Set name before enabling the port so that debugging messages can print
2414      * the name. */
2415     stp_port_set_name(sp, netdev_get_name(ofport->up.netdev));
2416     stp_port_enable(sp);
2417
2418     stp_port_set_aux(sp, ofport);
2419     stp_port_set_priority(sp, s->priority);
2420     stp_port_set_path_cost(sp, s->path_cost);
2421
2422     update_stp_port_state(ofport);
2423
2424     return 0;
2425 }
2426
2427 static int
2428 get_stp_port_status(struct ofport *ofport_,
2429                     struct ofproto_port_stp_status *s)
2430 {
2431     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2432     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2433     struct stp_port *sp = ofport->stp_port;
2434
2435     if (!ofproto->stp || !sp) {
2436         s->enabled = false;
2437         return 0;
2438     }
2439
2440     s->enabled = true;
2441     s->port_id = stp_port_get_id(sp);
2442     s->state = stp_port_get_state(sp);
2443     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2444     s->role = stp_port_get_role(sp);
2445
2446     return 0;
2447 }
2448
2449 static int
2450 get_stp_port_stats(struct ofport *ofport_,
2451                    struct ofproto_port_stp_stats *s)
2452 {
2453     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2454     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2455     struct stp_port *sp = ofport->stp_port;
2456
2457     if (!ofproto->stp || !sp) {
2458         s->enabled = false;
2459         return 0;
2460     }
2461
2462     s->enabled = true;
2463     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2464
2465     return 0;
2466 }
2467
2468 static void
2469 stp_run(struct ofproto_dpif *ofproto)
2470 {
2471     if (ofproto->stp) {
2472         long long int now = time_msec();
2473         long long int elapsed = now - ofproto->stp_last_tick;
2474         struct stp_port *sp;
2475
2476         if (elapsed > 0) {
2477             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2478             ofproto->stp_last_tick = now;
2479         }
2480         while (stp_get_changed_port(ofproto->stp, &sp)) {
2481             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2482
2483             if (ofport) {
2484                 update_stp_port_state(ofport);
2485             }
2486         }
2487
2488         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2489             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2490             mac_learning_flush(ofproto->ml);
2491             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2492             mcast_snooping_mdb_flush(ofproto->ms);
2493         }
2494     }
2495 }
2496
2497 static void
2498 stp_wait(struct ofproto_dpif *ofproto)
2499 {
2500     if (ofproto->stp) {
2501         poll_timer_wait(1000);
2502     }
2503 }
2504
2505 /* Configures RSTP on 'ofport_' using the settings defined in 's'.  The
2506  * caller is responsible for assigning RSTP port numbers and ensuring
2507  * there are no duplicates. */
2508 static void
2509 set_rstp_port(struct ofport *ofport_,
2510               const struct ofproto_port_rstp_settings *s)
2511 {
2512     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2513     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2514     struct rstp_port *rp = ofport->rstp_port;
2515
2516     if (!s || !s->enable) {
2517         if (rp) {
2518             rstp_port_unref(rp);
2519             ofport->rstp_port = NULL;
2520             update_rstp_port_state(ofport);
2521         }
2522         return;
2523     }
2524
2525     /* Check if need to add a new port. */
2526     if (!rp) {
2527         rp = ofport->rstp_port = rstp_add_port(ofproto->rstp);
2528     }
2529
2530     rstp_port_set(rp, s->port_num, s->priority, s->path_cost,
2531                   s->admin_edge_port, s->auto_edge,
2532                   s->admin_p2p_mac_state, s->admin_port_state, s->mcheck,
2533                   ofport);
2534     update_rstp_port_state(ofport);
2535     /* Synchronize operational status. */
2536     rstp_port_set_mac_operational(rp, ofport->may_enable);
2537 }
2538
2539 static void
2540 get_rstp_port_status(struct ofport *ofport_,
2541         struct ofproto_port_rstp_status *s)
2542 {
2543     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2544     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2545     struct rstp_port *rp = ofport->rstp_port;
2546
2547     if (!ofproto->rstp || !rp) {
2548         s->enabled = false;
2549         return;
2550     }
2551
2552     s->enabled = true;
2553     rstp_port_get_status(rp, &s->port_id, &s->state, &s->role,
2554                          &s->designated_bridge_id, &s->designated_port_id,
2555                          &s->designated_path_cost, &s->tx_count,
2556                          &s->rx_count, &s->error_count, &s->uptime);
2557 }
2558
2559 \f
2560 static int
2561 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2562            size_t n_qdscp)
2563 {
2564     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2565     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2566
2567     if (ofport->n_qdscp != n_qdscp
2568         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2569                               n_qdscp * sizeof *qdscp))) {
2570         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2571         free(ofport->qdscp);
2572         ofport->qdscp = n_qdscp
2573             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2574             : NULL;
2575         ofport->n_qdscp = n_qdscp;
2576     }
2577
2578     return 0;
2579 }
2580 \f
2581 /* Bundles. */
2582
2583 /* Expires all MAC learning entries associated with 'bundle' and forces its
2584  * ofproto to revalidate every flow.
2585  *
2586  * Normally MAC learning entries are removed only from the ofproto associated
2587  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2588  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2589  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2590  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2591  * with the host from which it migrated. */
2592 static void
2593 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2594 {
2595     struct ofproto_dpif *ofproto = bundle->ofproto;
2596     struct mac_learning *ml = ofproto->ml;
2597     struct mac_entry *mac, *next_mac;
2598
2599     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2600     ovs_rwlock_wrlock(&ml->rwlock);
2601     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2602         if (mac_entry_get_port(ml, mac) == bundle) {
2603             if (all_ofprotos) {
2604                 struct ofproto_dpif *o;
2605
2606                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2607                     if (o != ofproto) {
2608                         struct mac_entry *e;
2609
2610                         ovs_rwlock_wrlock(&o->ml->rwlock);
2611                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2612                         if (e) {
2613                             mac_learning_expire(o->ml, e);
2614                         }
2615                         ovs_rwlock_unlock(&o->ml->rwlock);
2616                     }
2617                 }
2618             }
2619
2620             mac_learning_expire(ml, mac);
2621         }
2622     }
2623     ovs_rwlock_unlock(&ml->rwlock);
2624 }
2625
2626 static void
2627 bundle_move(struct ofbundle *old, struct ofbundle *new)
2628 {
2629     struct ofproto_dpif *ofproto = old->ofproto;
2630     struct mac_learning *ml = ofproto->ml;
2631     struct mac_entry *mac, *next_mac;
2632
2633     ovs_assert(new->ofproto == old->ofproto);
2634
2635     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2636     ovs_rwlock_wrlock(&ml->rwlock);
2637     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2638         if (mac_entry_get_port(ml, mac) == old) {
2639             mac_entry_set_port(ml, mac, new);
2640         }
2641     }
2642     ovs_rwlock_unlock(&ml->rwlock);
2643 }
2644
2645 static struct ofbundle *
2646 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2647 {
2648     struct ofbundle *bundle;
2649
2650     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2651                              &ofproto->bundles) {
2652         if (bundle->aux == aux) {
2653             return bundle;
2654         }
2655     }
2656     return NULL;
2657 }
2658
2659 static void
2660 bundle_update(struct ofbundle *bundle)
2661 {
2662     struct ofport_dpif *port;
2663
2664     bundle->floodable = true;
2665     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2666         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2667             || port->is_layer3
2668             || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2669             || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2670             bundle->floodable = false;
2671             break;
2672         }
2673     }
2674 }
2675
2676 static void
2677 bundle_del_port(struct ofport_dpif *port)
2678 {
2679     struct ofbundle *bundle = port->bundle;
2680
2681     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2682
2683     list_remove(&port->bundle_node);
2684     port->bundle = NULL;
2685
2686     if (bundle->lacp) {
2687         lacp_slave_unregister(bundle->lacp, port);
2688     }
2689     if (bundle->bond) {
2690         bond_slave_unregister(bundle->bond, port);
2691     }
2692
2693     bundle_update(bundle);
2694 }
2695
2696 static bool
2697 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2698                 struct lacp_slave_settings *lacp)
2699 {
2700     struct ofport_dpif *port;
2701
2702     port = ofp_port_to_ofport(bundle->ofproto, ofp_port);
2703     if (!port) {
2704         return false;
2705     }
2706
2707     if (port->bundle != bundle) {
2708         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2709         if (port->bundle) {
2710             bundle_remove(&port->up);
2711         }
2712
2713         port->bundle = bundle;
2714         list_push_back(&bundle->ports, &port->bundle_node);
2715         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2716             || port->is_layer3
2717             || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2718             || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2719             bundle->floodable = false;
2720         }
2721     }
2722     if (lacp) {
2723         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2724         lacp_slave_register(bundle->lacp, port, lacp);
2725     }
2726
2727     return true;
2728 }
2729
2730 static void
2731 bundle_destroy(struct ofbundle *bundle)
2732 {
2733     struct ofproto_dpif *ofproto;
2734     struct ofport_dpif *port, *next_port;
2735
2736     if (!bundle) {
2737         return;
2738     }
2739
2740     ofproto = bundle->ofproto;
2741     mbridge_unregister_bundle(ofproto->mbridge, bundle);
2742
2743     xlate_txn_start();
2744     xlate_bundle_remove(bundle);
2745     xlate_txn_commit();
2746
2747     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2748         bundle_del_port(port);
2749     }
2750
2751     bundle_flush_macs(bundle, true);
2752     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2753     free(bundle->name);
2754     free(bundle->trunks);
2755     lacp_unref(bundle->lacp);
2756     bond_unref(bundle->bond);
2757     free(bundle);
2758 }
2759
2760 static int
2761 bundle_set(struct ofproto *ofproto_, void *aux,
2762            const struct ofproto_bundle_settings *s)
2763 {
2764     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2765     bool need_flush = false;
2766     struct ofport_dpif *port;
2767     struct ofbundle *bundle;
2768     unsigned long *trunks;
2769     int vlan;
2770     size_t i;
2771     bool ok;
2772
2773     if (!s) {
2774         bundle_destroy(bundle_lookup(ofproto, aux));
2775         return 0;
2776     }
2777
2778     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2779     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2780
2781     bundle = bundle_lookup(ofproto, aux);
2782     if (!bundle) {
2783         bundle = xmalloc(sizeof *bundle);
2784
2785         bundle->ofproto = ofproto;
2786         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2787                     hash_pointer(aux, 0));
2788         bundle->aux = aux;
2789         bundle->name = NULL;
2790
2791         list_init(&bundle->ports);
2792         bundle->vlan_mode = PORT_VLAN_TRUNK;
2793         bundle->vlan = -1;
2794         bundle->trunks = NULL;
2795         bundle->use_priority_tags = s->use_priority_tags;
2796         bundle->lacp = NULL;
2797         bundle->bond = NULL;
2798
2799         bundle->floodable = true;
2800         mbridge_register_bundle(ofproto->mbridge, bundle);
2801     }
2802
2803     if (!bundle->name || strcmp(s->name, bundle->name)) {
2804         free(bundle->name);
2805         bundle->name = xstrdup(s->name);
2806     }
2807
2808     /* LACP. */
2809     if (s->lacp) {
2810         ofproto->lacp_enabled = true;
2811         if (!bundle->lacp) {
2812             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2813             bundle->lacp = lacp_create();
2814         }
2815         lacp_configure(bundle->lacp, s->lacp);
2816     } else {
2817         lacp_unref(bundle->lacp);
2818         bundle->lacp = NULL;
2819     }
2820
2821     /* Update set of ports. */
2822     ok = true;
2823     for (i = 0; i < s->n_slaves; i++) {
2824         if (!bundle_add_port(bundle, s->slaves[i],
2825                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2826             ok = false;
2827         }
2828     }
2829     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2830         struct ofport_dpif *next_port;
2831
2832         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2833             for (i = 0; i < s->n_slaves; i++) {
2834                 if (s->slaves[i] == port->up.ofp_port) {
2835                     goto found;
2836                 }
2837             }
2838
2839             bundle_del_port(port);
2840         found: ;
2841         }
2842     }
2843     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2844
2845     if (list_is_empty(&bundle->ports)) {
2846         bundle_destroy(bundle);
2847         return EINVAL;
2848     }
2849
2850     /* Set VLAN tagging mode */
2851     if (s->vlan_mode != bundle->vlan_mode
2852         || s->use_priority_tags != bundle->use_priority_tags) {
2853         bundle->vlan_mode = s->vlan_mode;
2854         bundle->use_priority_tags = s->use_priority_tags;
2855         need_flush = true;
2856     }
2857
2858     /* Set VLAN tag. */
2859     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2860             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2861             : 0);
2862     if (vlan != bundle->vlan) {
2863         bundle->vlan = vlan;
2864         need_flush = true;
2865     }
2866
2867     /* Get trunked VLANs. */
2868     switch (s->vlan_mode) {
2869     case PORT_VLAN_ACCESS:
2870         trunks = NULL;
2871         break;
2872
2873     case PORT_VLAN_TRUNK:
2874         trunks = CONST_CAST(unsigned long *, s->trunks);
2875         break;
2876
2877     case PORT_VLAN_NATIVE_UNTAGGED:
2878     case PORT_VLAN_NATIVE_TAGGED:
2879         if (vlan != 0 && (!s->trunks
2880                           || !bitmap_is_set(s->trunks, vlan)
2881                           || bitmap_is_set(s->trunks, 0))) {
2882             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2883             if (s->trunks) {
2884                 trunks = bitmap_clone(s->trunks, 4096);
2885             } else {
2886                 trunks = bitmap_allocate1(4096);
2887             }
2888             bitmap_set1(trunks, vlan);
2889             bitmap_set0(trunks, 0);
2890         } else {
2891             trunks = CONST_CAST(unsigned long *, s->trunks);
2892         }
2893         break;
2894
2895     default:
2896         OVS_NOT_REACHED();
2897     }
2898     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2899         free(bundle->trunks);
2900         if (trunks == s->trunks) {
2901             bundle->trunks = vlan_bitmap_clone(trunks);
2902         } else {
2903             bundle->trunks = trunks;
2904             trunks = NULL;
2905         }
2906         need_flush = true;
2907     }
2908     if (trunks != s->trunks) {
2909         free(trunks);
2910     }
2911
2912     /* Bonding. */
2913     if (!list_is_short(&bundle->ports)) {
2914         bundle->ofproto->has_bonded_bundles = true;
2915         if (bundle->bond) {
2916             if (bond_reconfigure(bundle->bond, s->bond)) {
2917                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2918             }
2919         } else {
2920             bundle->bond = bond_create(s->bond, ofproto);
2921             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2922         }
2923
2924         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2925             bond_slave_register(bundle->bond, port,
2926                                 port->up.ofp_port, port->up.netdev);
2927         }
2928     } else {
2929         bond_unref(bundle->bond);
2930         bundle->bond = NULL;
2931     }
2932
2933     /* If we changed something that would affect MAC learning, un-learn
2934      * everything on this port and force flow revalidation. */
2935     if (need_flush) {
2936         bundle_flush_macs(bundle, false);
2937     }
2938
2939     return 0;
2940 }
2941
2942 static void
2943 bundle_remove(struct ofport *port_)
2944 {
2945     struct ofport_dpif *port = ofport_dpif_cast(port_);
2946     struct ofbundle *bundle = port->bundle;
2947
2948     if (bundle) {
2949         bundle_del_port(port);
2950         if (list_is_empty(&bundle->ports)) {
2951             bundle_destroy(bundle);
2952         } else if (list_is_short(&bundle->ports)) {
2953             bond_unref(bundle->bond);
2954             bundle->bond = NULL;
2955         }
2956     }
2957 }
2958
2959 static void
2960 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2961 {
2962     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2963     struct ofport_dpif *port = port_;
2964     uint8_t ea[ETH_ADDR_LEN];
2965     int error;
2966
2967     error = netdev_get_etheraddr(port->up.netdev, ea);
2968     if (!error) {
2969         struct dp_packet packet;
2970         void *packet_pdu;
2971
2972         dp_packet_init(&packet, 0);
2973         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2974                                  pdu_size);
2975         memcpy(packet_pdu, pdu, pdu_size);
2976
2977         ofproto_dpif_send_packet(port, &packet);
2978         dp_packet_uninit(&packet);
2979     } else {
2980         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2981                     "%s (%s)", port->bundle->name,
2982                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2983     }
2984 }
2985
2986 static void
2987 bundle_send_learning_packets(struct ofbundle *bundle)
2988 {
2989     struct ofproto_dpif *ofproto = bundle->ofproto;
2990     int error, n_packets, n_errors;
2991     struct mac_entry *e;
2992     struct pkt_list {
2993         struct ovs_list list_node;
2994         struct ofport_dpif *port;
2995         struct dp_packet *pkt;
2996     } *pkt_node;
2997     struct ovs_list packets;
2998
2999     list_init(&packets);
3000     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3001     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3002         if (mac_entry_get_port(ofproto->ml, e) != bundle) {
3003             pkt_node = xmalloc(sizeof *pkt_node);
3004             pkt_node->pkt = bond_compose_learning_packet(bundle->bond,
3005                                                          e->mac, e->vlan,
3006                                                          (void **)&pkt_node->port);
3007             list_push_back(&packets, &pkt_node->list_node);
3008         }
3009     }
3010     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3011
3012     error = n_packets = n_errors = 0;
3013     LIST_FOR_EACH_POP (pkt_node, list_node, &packets) {
3014         int ret;
3015
3016         ret = ofproto_dpif_send_packet(pkt_node->port, pkt_node->pkt);
3017         dp_packet_delete(pkt_node->pkt);
3018         free(pkt_node);
3019         if (ret) {
3020             error = ret;
3021             n_errors++;
3022         }
3023         n_packets++;
3024     }
3025
3026     if (n_errors) {
3027         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3028         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3029                      "packets, last error was: %s",
3030                      bundle->name, n_errors, n_packets, ovs_strerror(error));
3031     } else {
3032         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3033                  bundle->name, n_packets);
3034     }
3035 }
3036
3037 static void
3038 bundle_run(struct ofbundle *bundle)
3039 {
3040     if (bundle->lacp) {
3041         lacp_run(bundle->lacp, send_pdu_cb);
3042     }
3043     if (bundle->bond) {
3044         struct ofport_dpif *port;
3045
3046         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
3047             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
3048         }
3049
3050         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
3051             bundle->ofproto->backer->need_revalidate = REV_BOND;
3052         }
3053
3054         if (bond_should_send_learning_packets(bundle->bond)) {
3055             bundle_send_learning_packets(bundle);
3056         }
3057     }
3058 }
3059
3060 static void
3061 bundle_wait(struct ofbundle *bundle)
3062 {
3063     if (bundle->lacp) {
3064         lacp_wait(bundle->lacp);
3065     }
3066     if (bundle->bond) {
3067         bond_wait(bundle->bond);
3068     }
3069 }
3070 \f
3071 /* Mirrors. */
3072
3073 static int
3074 mirror_set__(struct ofproto *ofproto_, void *aux,
3075              const struct ofproto_mirror_settings *s)
3076 {
3077     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3078     struct ofbundle **srcs, **dsts;
3079     int error;
3080     size_t i;
3081
3082     if (!s) {
3083         mirror_destroy(ofproto->mbridge, aux);
3084         return 0;
3085     }
3086
3087     srcs = xmalloc(s->n_srcs * sizeof *srcs);
3088     dsts = xmalloc(s->n_dsts * sizeof *dsts);
3089
3090     for (i = 0; i < s->n_srcs; i++) {
3091         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
3092     }
3093
3094     for (i = 0; i < s->n_dsts; i++) {
3095         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
3096     }
3097
3098     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
3099                        s->n_dsts, s->src_vlans,
3100                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
3101     free(srcs);
3102     free(dsts);
3103     return error;
3104 }
3105
3106 static int
3107 mirror_get_stats__(struct ofproto *ofproto, void *aux,
3108                    uint64_t *packets, uint64_t *bytes)
3109 {
3110     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
3111                             bytes);
3112 }
3113
3114 static int
3115 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3116 {
3117     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3118     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3119     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3120         mac_learning_flush(ofproto->ml);
3121     }
3122     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3123     return 0;
3124 }
3125
3126 static bool
3127 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3128 {
3129     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3130     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3131     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
3132 }
3133
3134 static void
3135 forward_bpdu_changed(struct ofproto *ofproto_)
3136 {
3137     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3138     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3139 }
3140
3141 static void
3142 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3143                      size_t max_entries)
3144 {
3145     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3146     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3147     mac_learning_set_idle_time(ofproto->ml, idle_time);
3148     mac_learning_set_max_entries(ofproto->ml, max_entries);
3149     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3150 }
3151
3152 /* Configures multicast snooping on 'ofport' using the settings
3153  * defined in 's'. */
3154 static int
3155 set_mcast_snooping(struct ofproto *ofproto_,
3156                    const struct ofproto_mcast_snooping_settings *s)
3157 {
3158     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3159
3160     /* Only revalidate flows if the configuration changed. */
3161     if (!s != !ofproto->ms) {
3162         ofproto->backer->need_revalidate = REV_RECONFIGURE;
3163     }
3164
3165     if (s) {
3166         if (!ofproto->ms) {
3167             ofproto->ms = mcast_snooping_create();
3168         }
3169
3170         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3171         mcast_snooping_set_idle_time(ofproto->ms, s->idle_time);
3172         mcast_snooping_set_max_entries(ofproto->ms, s->max_entries);
3173         if (mcast_snooping_set_flood_unreg(ofproto->ms, s->flood_unreg)) {
3174             ofproto->backer->need_revalidate = REV_RECONFIGURE;
3175         }
3176         ovs_rwlock_unlock(&ofproto->ms->rwlock);
3177     } else {
3178         mcast_snooping_unref(ofproto->ms);
3179         ofproto->ms = NULL;
3180     }
3181
3182     return 0;
3183 }
3184
3185 /* Configures multicast snooping port's flood settings on 'ofproto'. */
3186 static int
3187 set_mcast_snooping_port(struct ofproto *ofproto_, void *aux,
3188                         const struct ofproto_mcast_snooping_port_settings *s)
3189 {
3190     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3191     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3192
3193     if (ofproto->ms && s) {
3194         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3195         mcast_snooping_set_port_flood(ofproto->ms, bundle, s->flood);
3196         mcast_snooping_set_port_flood_reports(ofproto->ms, bundle,
3197                                               s->flood_reports);
3198         ovs_rwlock_unlock(&ofproto->ms->rwlock);
3199     }
3200     return 0;
3201 }
3202
3203 \f
3204 /* Ports. */
3205
3206 struct ofport_dpif *
3207 ofp_port_to_ofport(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
3208 {
3209     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3210     return ofport ? ofport_dpif_cast(ofport) : NULL;
3211 }
3212
3213 static void
3214 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3215                             struct ofproto_port *ofproto_port,
3216                             struct dpif_port *dpif_port)
3217 {
3218     ofproto_port->name = dpif_port->name;
3219     ofproto_port->type = dpif_port->type;
3220     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3221 }
3222
3223 static void
3224 ofport_update_peer(struct ofport_dpif *ofport)
3225 {
3226     const struct ofproto_dpif *ofproto;
3227     struct dpif_backer *backer;
3228     char *peer_name;
3229
3230     if (!netdev_vport_is_patch(ofport->up.netdev)) {
3231         return;
3232     }
3233
3234     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
3235     backer->need_revalidate = REV_RECONFIGURE;
3236
3237     if (ofport->peer) {
3238         ofport->peer->peer = NULL;
3239         ofport->peer = NULL;
3240     }
3241
3242     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
3243     if (!peer_name) {
3244         return;
3245     }
3246
3247     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3248         struct ofport *peer_ofport;
3249         struct ofport_dpif *peer;
3250         char *peer_peer;
3251
3252         if (ofproto->backer != backer) {
3253             continue;
3254         }
3255
3256         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
3257         if (!peer_ofport) {
3258             continue;
3259         }
3260
3261         peer = ofport_dpif_cast(peer_ofport);
3262         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
3263         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
3264                                  peer_peer)) {
3265             ofport->peer = peer;
3266             ofport->peer->peer = ofport;
3267         }
3268         free(peer_peer);
3269
3270         break;
3271     }
3272     free(peer_name);
3273 }
3274
3275 static void
3276 port_run(struct ofport_dpif *ofport)
3277 {
3278     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3279     bool carrier_changed = carrier_seq != ofport->carrier_seq;
3280     bool enable = netdev_get_carrier(ofport->up.netdev);
3281     bool cfm_enable = false;
3282     bool bfd_enable = false;
3283
3284     ofport->carrier_seq = carrier_seq;
3285
3286     if (ofport->cfm) {
3287         int cfm_opup = cfm_get_opup(ofport->cfm);
3288
3289         cfm_enable = !cfm_get_fault(ofport->cfm);
3290
3291         if (cfm_opup >= 0) {
3292             cfm_enable = cfm_enable && cfm_opup;
3293         }
3294     }
3295
3296     if (ofport->bfd) {
3297         bfd_enable = bfd_forwarding(ofport->bfd);
3298     }
3299
3300     if (ofport->bfd || ofport->cfm) {
3301         enable = enable && (cfm_enable || bfd_enable);
3302     }
3303
3304     if (ofport->bundle) {
3305         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3306         if (carrier_changed) {
3307             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3308         }
3309     }
3310
3311     if (ofport->may_enable != enable) {
3312         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3313
3314         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3315
3316         if (ofport->rstp_port) {
3317             rstp_port_set_mac_operational(ofport->rstp_port, enable);
3318         }
3319     }
3320
3321     ofport->may_enable = enable;
3322 }
3323
3324 static int
3325 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3326                    struct ofproto_port *ofproto_port)
3327 {
3328     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3329     struct dpif_port dpif_port;
3330     int error;
3331
3332     if (sset_contains(&ofproto->ghost_ports, devname)) {
3333         const char *type = netdev_get_type_from_name(devname);
3334
3335         /* We may be called before ofproto->up.port_by_name is populated with
3336          * the appropriate ofport.  For this reason, we must get the name and
3337          * type from the netdev layer directly. */
3338         if (type) {
3339             const struct ofport *ofport;
3340
3341             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3342             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3343             ofproto_port->name = xstrdup(devname);
3344             ofproto_port->type = xstrdup(type);
3345             return 0;
3346         }
3347         return ENODEV;
3348     }
3349
3350     if (!sset_contains(&ofproto->ports, devname)) {
3351         return ENODEV;
3352     }
3353     error = dpif_port_query_by_name(ofproto->backer->dpif,
3354                                     devname, &dpif_port);
3355     if (!error) {
3356         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3357     }
3358     return error;
3359 }
3360
3361 static int
3362 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3363 {
3364     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3365     const char *devname = netdev_get_name(netdev);
3366     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3367     const char *dp_port_name;
3368
3369     if (netdev_vport_is_patch(netdev)) {
3370         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3371         return 0;
3372     }
3373
3374     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3375     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3376         odp_port_t port_no = ODPP_NONE;
3377         int error;
3378
3379         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3380         if (error) {
3381             return error;
3382         }
3383         if (netdev_get_tunnel_config(netdev)) {
3384             simap_put(&ofproto->backer->tnl_backers,
3385                       dp_port_name, odp_to_u32(port_no));
3386         }
3387     }
3388
3389     if (netdev_get_tunnel_config(netdev)) {
3390         sset_add(&ofproto->ghost_ports, devname);
3391     } else {
3392         sset_add(&ofproto->ports, devname);
3393     }
3394     return 0;
3395 }
3396
3397 static int
3398 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3399 {
3400     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3401     struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
3402     int error = 0;
3403
3404     if (!ofport) {
3405         return 0;
3406     }
3407
3408     sset_find_and_delete(&ofproto->ghost_ports,
3409                          netdev_get_name(ofport->up.netdev));
3410     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3411     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3412         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3413         if (!error) {
3414             /* The caller is going to close ofport->up.netdev.  If this is a
3415              * bonded port, then the bond is using that netdev, so remove it
3416              * from the bond.  The client will need to reconfigure everything
3417              * after deleting ports, so then the slave will get re-added. */
3418             bundle_remove(&ofport->up);
3419         }
3420     }
3421     return error;
3422 }
3423
3424 static int
3425 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3426 {
3427     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3428     int error;
3429
3430     error = netdev_get_stats(ofport->up.netdev, stats);
3431
3432     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3433         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3434
3435         ovs_mutex_lock(&ofproto->stats_mutex);
3436         /* ofproto->stats.tx_packets represents packets that we created
3437          * internally and sent to some port (e.g. packets sent with
3438          * ofproto_dpif_send_packet()).  Account for them as if they had
3439          * come from OFPP_LOCAL and got forwarded. */
3440
3441         if (stats->rx_packets != UINT64_MAX) {
3442             stats->rx_packets += ofproto->stats.tx_packets;
3443         }
3444
3445         if (stats->rx_bytes != UINT64_MAX) {
3446             stats->rx_bytes += ofproto->stats.tx_bytes;
3447         }
3448
3449         /* ofproto->stats.rx_packets represents packets that were received on
3450          * some port and we processed internally and dropped (e.g. STP).
3451          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3452
3453         if (stats->tx_packets != UINT64_MAX) {
3454             stats->tx_packets += ofproto->stats.rx_packets;
3455         }
3456
3457         if (stats->tx_bytes != UINT64_MAX) {
3458             stats->tx_bytes += ofproto->stats.rx_bytes;
3459         }
3460         ovs_mutex_unlock(&ofproto->stats_mutex);
3461     }
3462
3463     return error;
3464 }
3465
3466 static int
3467 port_get_lacp_stats(const struct ofport *ofport_, struct lacp_slave_stats *stats)
3468 {
3469     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3470     if (ofport->bundle && ofport->bundle->lacp) {
3471         if (lacp_get_slave_stats(ofport->bundle->lacp, ofport, stats)) {
3472             return 0;
3473         }
3474     }
3475     return -1;
3476 }
3477
3478 struct port_dump_state {
3479     uint32_t bucket;
3480     uint32_t offset;
3481     bool ghost;
3482
3483     struct ofproto_port port;
3484     bool has_port;
3485 };
3486
3487 static int
3488 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3489 {
3490     *statep = xzalloc(sizeof(struct port_dump_state));
3491     return 0;
3492 }
3493
3494 static int
3495 port_dump_next(const struct ofproto *ofproto_, void *state_,
3496                struct ofproto_port *port)
3497 {
3498     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3499     struct port_dump_state *state = state_;
3500     const struct sset *sset;
3501     struct sset_node *node;
3502
3503     if (state->has_port) {
3504         ofproto_port_destroy(&state->port);
3505         state->has_port = false;
3506     }
3507     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3508     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3509         int error;
3510
3511         error = port_query_by_name(ofproto_, node->name, &state->port);
3512         if (!error) {
3513             *port = state->port;
3514             state->has_port = true;
3515             return 0;
3516         } else if (error != ENODEV) {
3517             return error;
3518         }
3519     }
3520
3521     if (!state->ghost) {
3522         state->ghost = true;
3523         state->bucket = 0;
3524         state->offset = 0;
3525         return port_dump_next(ofproto_, state_, port);
3526     }
3527
3528     return EOF;
3529 }
3530
3531 static int
3532 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3533 {
3534     struct port_dump_state *state = state_;
3535
3536     if (state->has_port) {
3537         ofproto_port_destroy(&state->port);
3538     }
3539     free(state);
3540     return 0;
3541 }
3542
3543 static int
3544 port_poll(const struct ofproto *ofproto_, char **devnamep)
3545 {
3546     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3547
3548     if (ofproto->port_poll_errno) {
3549         int error = ofproto->port_poll_errno;
3550         ofproto->port_poll_errno = 0;
3551         return error;
3552     }
3553
3554     if (sset_is_empty(&ofproto->port_poll_set)) {
3555         return EAGAIN;
3556     }
3557
3558     *devnamep = sset_pop(&ofproto->port_poll_set);
3559     return 0;
3560 }
3561
3562 static void
3563 port_poll_wait(const struct ofproto *ofproto_)
3564 {
3565     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3566     dpif_port_poll_wait(ofproto->backer->dpif);
3567 }
3568
3569 static int
3570 port_is_lacp_current(const struct ofport *ofport_)
3571 {
3572     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3573     return (ofport->bundle && ofport->bundle->lacp
3574             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3575             : -1);
3576 }
3577 \f
3578 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3579  * then delete it entirely. */
3580 static void
3581 rule_expire(struct rule_dpif *rule)
3582     OVS_REQUIRES(ofproto_mutex)
3583 {
3584     uint16_t hard_timeout, idle_timeout;
3585     long long int now = time_msec();
3586     int reason = -1;
3587
3588     hard_timeout = rule->up.hard_timeout;
3589     idle_timeout = rule->up.idle_timeout;
3590
3591     /* Has 'rule' expired? */
3592     if (hard_timeout) {
3593         long long int modified;
3594
3595         ovs_mutex_lock(&rule->up.mutex);
3596         modified = rule->up.modified;
3597         ovs_mutex_unlock(&rule->up.mutex);
3598
3599         if (now > modified + hard_timeout * 1000) {
3600             reason = OFPRR_HARD_TIMEOUT;
3601         }
3602     }
3603
3604     if (reason < 0 && idle_timeout) {
3605         long long int used;
3606
3607         ovs_mutex_lock(&rule->stats_mutex);
3608         used = rule->stats.used;
3609         ovs_mutex_unlock(&rule->stats_mutex);
3610
3611         if (now > used + idle_timeout * 1000) {
3612             reason = OFPRR_IDLE_TIMEOUT;
3613         }
3614     }
3615
3616     if (reason >= 0) {
3617         COVERAGE_INC(ofproto_dpif_expired);
3618         ofproto_rule_expire(&rule->up, reason);
3619     }
3620 }
3621
3622 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3623  * 'flow' must reflect the data in 'packet'. */
3624 int
3625 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3626                              const struct flow *flow,
3627                              struct rule_dpif *rule,
3628                              const struct ofpact *ofpacts, size_t ofpacts_len,
3629                              struct dp_packet *packet)
3630 {
3631     struct dpif_flow_stats stats;
3632     struct xlate_out xout;
3633     struct xlate_in xin;
3634     ofp_port_t in_port;
3635     struct dpif_execute execute;
3636     int error;
3637
3638     ovs_assert((rule != NULL) != (ofpacts != NULL));
3639
3640     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3641
3642     if (rule) {
3643         rule_dpif_credit_stats(rule, &stats);
3644     }
3645
3646     xlate_in_init(&xin, ofproto, flow, flow->in_port.ofp_port, rule,
3647                   stats.tcp_flags, packet);
3648     xin.ofpacts = ofpacts;
3649     xin.ofpacts_len = ofpacts_len;
3650     xin.resubmit_stats = &stats;
3651     xlate_actions(&xin, &xout);
3652
3653     execute.actions = xout.odp_actions->data;
3654     execute.actions_len = xout.odp_actions->size;
3655
3656     pkt_metadata_from_flow(&packet->md, flow);
3657     execute.packet = packet;
3658     execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3659     execute.probe = false;
3660
3661     /* Fix up in_port. */
3662     in_port = flow->in_port.ofp_port;
3663     if (in_port == OFPP_NONE) {
3664         in_port = OFPP_LOCAL;
3665     }
3666     execute.packet->md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3667
3668     error = dpif_execute(ofproto->backer->dpif, &execute);
3669
3670     xlate_out_uninit(&xout);
3671
3672     return error;
3673 }
3674
3675 void
3676 rule_dpif_credit_stats(struct rule_dpif *rule,
3677                        const struct dpif_flow_stats *stats)
3678 {
3679     ovs_mutex_lock(&rule->stats_mutex);
3680     if (OVS_UNLIKELY(rule->new_rule)) {
3681         rule_dpif_credit_stats(rule->new_rule, stats);
3682     } else {
3683         rule->stats.n_packets += stats->n_packets;
3684         rule->stats.n_bytes += stats->n_bytes;
3685         rule->stats.used = MAX(rule->stats.used, stats->used);
3686     }
3687     ovs_mutex_unlock(&rule->stats_mutex);
3688 }
3689
3690 ovs_be64
3691 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3692     OVS_REQUIRES(rule->up.mutex)
3693 {
3694     return rule->up.flow_cookie;
3695 }
3696
3697 void
3698 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3699                      uint16_t hard_timeout)
3700 {
3701     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3702 }
3703
3704 /* Returns 'rule''s actions.  The returned actions are RCU-protected, and can
3705  * be read until the calling thread quiesces. */
3706 const struct rule_actions *
3707 rule_dpif_get_actions(const struct rule_dpif *rule)
3708 {
3709     return rule_get_actions(&rule->up);
3710 }
3711
3712 /* Sets 'rule''s recirculation id. */
3713 static void
3714 rule_dpif_set_recirc_id(struct rule_dpif *rule, uint32_t id)
3715     OVS_REQUIRES(rule->up.mutex)
3716 {
3717     ovs_assert(!rule->recirc_id || rule->recirc_id == id);
3718     if (rule->recirc_id == id) {
3719         /* Release the new reference to the same id. */
3720         recirc_free_id(id);
3721     } else {
3722         rule->recirc_id = id;
3723     }
3724 }
3725
3726 /* Sets 'rule''s recirculation id. */
3727 void
3728 rule_set_recirc_id(struct rule *rule_, uint32_t id)
3729 {
3730     struct rule_dpif *rule = rule_dpif_cast(rule_);
3731
3732     ovs_mutex_lock(&rule->up.mutex);
3733     rule_dpif_set_recirc_id(rule, id);
3734     ovs_mutex_unlock(&rule->up.mutex);
3735 }
3736
3737 long long
3738 ofproto_dpif_get_tables_version(struct ofproto_dpif *ofproto OVS_UNUSED)
3739 {
3740     long long version;
3741
3742     atomic_read_relaxed(&ofproto->tables_version, &version);
3743
3744     return version;
3745 }
3746
3747 /* The returned rule (if any) is valid at least until the next RCU quiescent
3748  * period.  If the rule needs to stay around longer, a non-zero 'take_ref'
3749  * must be passed in to cause a reference to be taken on it.
3750  *
3751  * 'flow' is non-const to allow for temporary modifications during the lookup.
3752  * Any changes are restored before returning. */
3753 static struct rule_dpif *
3754 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, long long version,
3755                           uint8_t table_id, struct flow *flow,
3756                           struct flow_wildcards *wc, bool take_ref)
3757 {
3758     struct classifier *cls = &ofproto->up.tables[table_id].cls;
3759     const struct cls_rule *cls_rule;
3760     struct rule_dpif *rule;
3761
3762     do {
3763         cls_rule = classifier_lookup(cls, version, flow, wc);
3764
3765         rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3766
3767         /* Try again if the rule was released before we get the reference. */
3768     } while (rule && take_ref && !rule_dpif_try_ref(rule));
3769
3770     return rule;
3771 }
3772
3773 /* Look up 'flow' in 'ofproto''s classifier version 'version', starting from
3774  * table '*table_id'.  Returns the rule that was found, which may be one of the
3775  * special rules according to packet miss hadling.  If 'may_packet_in' is
3776  * false, returning of the miss_rule (which issues packet ins for the
3777  * controller) is avoided.  Updates 'wc', if nonnull, to reflect the fields
3778  * that were used during the lookup.
3779  *
3780  * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3781  * if none is found then the table miss configuration for that table is
3782  * honored, which can result in additional lookups in other OpenFlow tables.
3783  * In this case the function updates '*table_id' to reflect the final OpenFlow
3784  * table that was searched.
3785  *
3786  * If 'honor_table_miss' is false, then only one table lookup occurs, in
3787  * '*table_id'.
3788  *
3789  * The rule is returned in '*rule', which is valid at least until the next
3790  * RCU quiescent period.  If the '*rule' needs to stay around longer,
3791  * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3792  * on it before this returns.
3793  *
3794  * 'in_port' allows the lookup to take place as if the in port had the value
3795  * 'in_port'.  This is needed for resubmit action support.
3796  *
3797  * 'flow' is non-const to allow for temporary modifications during the lookup.
3798  * Any changes are restored before returning. */
3799 struct rule_dpif *
3800 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto, long long version,
3801                             struct flow *flow, struct flow_wildcards *wc,
3802                             bool take_ref, const struct dpif_flow_stats *stats,
3803                             uint8_t *table_id, ofp_port_t in_port,
3804                             bool may_packet_in, bool honor_table_miss)
3805 {
3806     ovs_be16 old_tp_src = flow->tp_src, old_tp_dst = flow->tp_dst;
3807     ofp_port_t old_in_port = flow->in_port.ofp_port;
3808     enum ofputil_table_miss miss_config;
3809     struct rule_dpif *rule;
3810     uint8_t next_id;
3811
3812     /* We always unwildcard nw_frag (for IP), so they
3813      * need not be unwildcarded here. */
3814     if (flow->nw_frag & FLOW_NW_FRAG_ANY
3815         && ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3816         if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3817             /* We must pretend that transport ports are unavailable. */
3818             flow->tp_src = htons(0);
3819             flow->tp_dst = htons(0);
3820         } else {
3821             /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM).
3822              * Use the drop_frags_rule (which cannot disappear). */
3823             rule = ofproto->drop_frags_rule;
3824             if (take_ref) {
3825                 rule_dpif_ref(rule);
3826             }
3827             if (stats) {
3828                 struct oftable *tbl = &ofproto->up.tables[*table_id];
3829                 unsigned long orig;
3830
3831                 atomic_add_relaxed(&tbl->n_matched, stats->n_packets, &orig);
3832             }
3833             return rule;
3834         }
3835     }
3836
3837     /* Look up a flow with 'in_port' as the input port.  Then restore the
3838      * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
3839      * have surprising behavior). */
3840     flow->in_port.ofp_port = in_port;
3841
3842     /* Our current implementation depends on n_tables == N_TABLES, and
3843      * TBL_INTERNAL being the last table. */
3844     BUILD_ASSERT_DECL(N_TABLES == TBL_INTERNAL + 1);
3845
3846     miss_config = OFPUTIL_TABLE_MISS_CONTINUE;
3847
3848     for (next_id = *table_id;
3849          next_id < ofproto->up.n_tables;
3850          next_id++, next_id += (next_id == TBL_INTERNAL))
3851     {
3852         *table_id = next_id;
3853         rule = rule_dpif_lookup_in_table(ofproto, version, next_id, flow, wc,
3854                                          take_ref);
3855         if (stats) {
3856             struct oftable *tbl = &ofproto->up.tables[next_id];
3857             unsigned long orig;
3858
3859             atomic_add_relaxed(rule ? &tbl->n_matched : &tbl->n_missed,
3860                                stats->n_packets, &orig);
3861         }
3862         if (rule) {
3863             goto out;   /* Match. */
3864         }
3865         if (honor_table_miss) {
3866             miss_config = ofproto_table_get_miss_config(&ofproto->up,
3867                                                         *table_id);
3868             if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE) {
3869                 continue;
3870             }
3871         }
3872         break;
3873     }
3874     /* Miss. */
3875     rule = ofproto->no_packet_in_rule;
3876     if (may_packet_in) {
3877         if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE
3878             || miss_config == OFPUTIL_TABLE_MISS_CONTROLLER) {
3879             struct ofport_dpif *port;
3880
3881             port = ofp_port_to_ofport(ofproto, old_in_port);
3882             if (!port) {
3883                 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3884                              old_in_port);
3885             } else if (!(port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN)) {
3886                 rule = ofproto->miss_rule;
3887             }
3888         } else if (miss_config == OFPUTIL_TABLE_MISS_DEFAULT &&
3889                    connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3890             rule = ofproto->miss_rule;
3891         }
3892     }
3893     if (take_ref) {
3894         rule_dpif_ref(rule);
3895     }
3896 out:
3897     /* Restore port numbers, as they may have been modified above. */
3898     flow->tp_src = old_tp_src;
3899     flow->tp_dst = old_tp_dst;
3900     /* Restore the old in port. */
3901     flow->in_port.ofp_port = old_in_port;
3902
3903     return rule;
3904 }
3905
3906 static void
3907 complete_operation(struct rule_dpif *rule)
3908     OVS_REQUIRES(ofproto_mutex)
3909 {
3910     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3911
3912     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3913 }
3914
3915 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3916 {
3917     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3918 }
3919
3920 static struct rule *
3921 rule_alloc(void)
3922 {
3923     struct rule_dpif *rule = xzalloc(sizeof *rule);
3924     return &rule->up;
3925 }
3926
3927 static void
3928 rule_dealloc(struct rule *rule_)
3929 {
3930     struct rule_dpif *rule = rule_dpif_cast(rule_);
3931     free(rule);
3932 }
3933
3934 static enum ofperr
3935 rule_construct(struct rule *rule_)
3936     OVS_NO_THREAD_SAFETY_ANALYSIS
3937 {
3938     struct rule_dpif *rule = rule_dpif_cast(rule_);
3939     ovs_mutex_init_adaptive(&rule->stats_mutex);
3940     rule->stats.n_packets = 0;
3941     rule->stats.n_bytes = 0;
3942     rule->stats.used = rule->up.modified;
3943     rule->recirc_id = 0;
3944     rule->new_rule = NULL;
3945
3946     return 0;
3947 }
3948
3949 static void
3950 rule_insert(struct rule *rule_, struct rule *old_rule_, bool forward_stats)
3951     OVS_REQUIRES(ofproto_mutex)
3952 {
3953     struct rule_dpif *rule = rule_dpif_cast(rule_);
3954
3955     if (old_rule_ && forward_stats) {
3956         struct rule_dpif *old_rule = rule_dpif_cast(old_rule_);
3957
3958         ovs_assert(!old_rule->new_rule);
3959
3960         /* Take a reference to the new rule, and refer all stats updates from
3961          * the old rule to the new rule. */
3962         rule_dpif_ref(rule);
3963
3964         ovs_mutex_lock(&old_rule->stats_mutex);
3965         ovs_mutex_lock(&rule->stats_mutex);
3966         old_rule->new_rule = rule;       /* Forward future stats. */
3967         rule->stats = old_rule->stats;   /* Transfer stats to the new rule. */
3968         ovs_mutex_unlock(&rule->stats_mutex);
3969         ovs_mutex_unlock(&old_rule->stats_mutex);
3970     }
3971
3972     complete_operation(rule);
3973 }
3974
3975 static void
3976 rule_delete(struct rule *rule_)
3977     OVS_REQUIRES(ofproto_mutex)
3978 {
3979     struct rule_dpif *rule = rule_dpif_cast(rule_);
3980     complete_operation(rule);
3981 }
3982
3983 static void
3984 rule_destruct(struct rule *rule_)
3985     OVS_NO_THREAD_SAFETY_ANALYSIS
3986 {
3987     struct rule_dpif *rule = rule_dpif_cast(rule_);
3988
3989     ovs_mutex_destroy(&rule->stats_mutex);
3990     /* Release reference to the new rule, if any. */
3991     if (rule->new_rule) {
3992         rule_dpif_unref(rule->new_rule);
3993     }
3994     if (rule->recirc_id) {
3995         recirc_free_id(rule->recirc_id);
3996     }
3997 }
3998
3999 static void
4000 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
4001                long long int *used)
4002 {
4003     struct rule_dpif *rule = rule_dpif_cast(rule_);
4004
4005     ovs_mutex_lock(&rule->stats_mutex);
4006     if (OVS_UNLIKELY(rule->new_rule)) {
4007         rule_get_stats(&rule->new_rule->up, packets, bytes, used);
4008     } else {
4009         *packets = rule->stats.n_packets;
4010         *bytes = rule->stats.n_bytes;
4011         *used = rule->stats.used;
4012     }
4013     ovs_mutex_unlock(&rule->stats_mutex);
4014 }
4015
4016 static void
4017 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4018                   struct dp_packet *packet)
4019 {
4020     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4021
4022     ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
4023 }
4024
4025 static enum ofperr
4026 rule_execute(struct rule *rule, const struct flow *flow,
4027              struct dp_packet *packet)
4028 {
4029     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4030     dp_packet_delete(packet);
4031     return 0;
4032 }
4033
4034 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4035 {
4036     return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4037 }
4038
4039 static struct ofgroup *
4040 group_alloc(void)
4041 {
4042     struct group_dpif *group = xzalloc(sizeof *group);
4043     return &group->up;
4044 }
4045
4046 static void
4047 group_dealloc(struct ofgroup *group_)
4048 {
4049     struct group_dpif *group = group_dpif_cast(group_);
4050     free(group);
4051 }
4052
4053 static void
4054 group_construct_stats(struct group_dpif *group)
4055     OVS_REQUIRES(group->stats_mutex)
4056 {
4057     struct ofputil_bucket *bucket;
4058     const struct ovs_list *buckets;
4059
4060     group->packet_count = 0;
4061     group->byte_count = 0;
4062
4063     group_dpif_get_buckets(group, &buckets);
4064     LIST_FOR_EACH (bucket, list_node, buckets) {
4065         bucket->stats.packet_count = 0;
4066         bucket->stats.byte_count = 0;
4067     }
4068 }
4069
4070 void
4071 group_dpif_credit_stats(struct group_dpif *group,
4072                         struct ofputil_bucket *bucket,
4073                         const struct dpif_flow_stats *stats)
4074 {
4075     ovs_mutex_lock(&group->stats_mutex);
4076     group->packet_count += stats->n_packets;
4077     group->byte_count += stats->n_bytes;
4078     if (bucket) {
4079         bucket->stats.packet_count += stats->n_packets;
4080         bucket->stats.byte_count += stats->n_bytes;
4081     } else { /* Credit to all buckets */
4082         const struct ovs_list *buckets;
4083
4084         group_dpif_get_buckets(group, &buckets);
4085         LIST_FOR_EACH (bucket, list_node, buckets) {
4086             bucket->stats.packet_count += stats->n_packets;
4087             bucket->stats.byte_count += stats->n_bytes;
4088         }
4089     }
4090     ovs_mutex_unlock(&group->stats_mutex);
4091 }
4092
4093 static enum ofperr
4094 group_construct(struct ofgroup *group_)
4095 {
4096     struct group_dpif *group = group_dpif_cast(group_);
4097     const struct ofputil_bucket *bucket;
4098
4099     /* Prevent group chaining because our locking structure makes it hard to
4100      * implement deadlock-free.  (See xlate_group_resource_check().) */
4101     LIST_FOR_EACH (bucket, list_node, &group->up.buckets) {
4102         const struct ofpact *a;
4103
4104         OFPACT_FOR_EACH (a, bucket->ofpacts, bucket->ofpacts_len) {
4105             if (a->type == OFPACT_GROUP) {
4106                 return OFPERR_OFPGMFC_CHAINING_UNSUPPORTED;
4107             }
4108         }
4109     }
4110
4111     ovs_mutex_init_adaptive(&group->stats_mutex);
4112     ovs_mutex_lock(&group->stats_mutex);
4113     group_construct_stats(group);
4114     ovs_mutex_unlock(&group->stats_mutex);
4115     return 0;
4116 }
4117
4118 static void
4119 group_destruct(struct ofgroup *group_)
4120 {
4121     struct group_dpif *group = group_dpif_cast(group_);
4122     ovs_mutex_destroy(&group->stats_mutex);
4123 }
4124
4125 static enum ofperr
4126 group_modify(struct ofgroup *group_)
4127 {
4128     struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
4129
4130     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4131
4132     return 0;
4133 }
4134
4135 static enum ofperr
4136 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4137 {
4138     struct group_dpif *group = group_dpif_cast(group_);
4139     struct ofputil_bucket *bucket;
4140     const struct ovs_list *buckets;
4141     struct bucket_counter *bucket_stats;
4142
4143     ovs_mutex_lock(&group->stats_mutex);
4144     ogs->packet_count = group->packet_count;
4145     ogs->byte_count = group->byte_count;
4146
4147     group_dpif_get_buckets(group, &buckets);
4148     bucket_stats = ogs->bucket_stats;
4149     LIST_FOR_EACH (bucket, list_node, buckets) {
4150         bucket_stats->packet_count = bucket->stats.packet_count;
4151         bucket_stats->byte_count = bucket->stats.byte_count;
4152         bucket_stats++;
4153     }
4154     ovs_mutex_unlock(&group->stats_mutex);
4155
4156     return 0;
4157 }
4158
4159 /* If the group exists, this function increments the groups's reference count.
4160  *
4161  * Make sure to call group_dpif_unref() after no longer needing to maintain
4162  * a reference to the group. */
4163 bool
4164 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4165                   struct group_dpif **group)
4166 {
4167     struct ofgroup *ofgroup;
4168     bool found;
4169
4170     found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4171     *group = found ?  group_dpif_cast(ofgroup) : NULL;
4172
4173     return found;
4174 }
4175
4176 void
4177 group_dpif_get_buckets(const struct group_dpif *group,
4178                        const struct ovs_list **buckets)
4179 {
4180     *buckets = &group->up.buckets;
4181 }
4182
4183 enum ofp11_group_type
4184 group_dpif_get_type(const struct group_dpif *group)
4185 {
4186     return group->up.type;
4187 }
4188
4189 const char *
4190 group_dpif_get_selection_method(const struct group_dpif *group)
4191 {
4192     return group->up.props.selection_method;
4193 }
4194 \f
4195 /* Sends 'packet' out 'ofport'.
4196  * May modify 'packet'.
4197  * Returns 0 if successful, otherwise a positive errno value. */
4198 int
4199 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct dp_packet *packet)
4200 {
4201     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4202     int error;
4203
4204     error = xlate_send_packet(ofport, packet);
4205
4206     ovs_mutex_lock(&ofproto->stats_mutex);
4207     ofproto->stats.tx_packets++;
4208     ofproto->stats.tx_bytes += dp_packet_size(packet);
4209     ovs_mutex_unlock(&ofproto->stats_mutex);
4210     return error;
4211 }
4212
4213 uint64_t
4214 group_dpif_get_selection_method_param(const struct group_dpif *group)
4215 {
4216     return group->up.props.selection_method_param;
4217 }
4218
4219 const struct field_array *
4220 group_dpif_get_fields(const struct group_dpif *group)
4221 {
4222     return &group->up.props.fields;
4223 }
4224 \f
4225 /* Return the version string of the datapath that backs up
4226  * this 'ofproto'.
4227  */
4228 static const char *
4229 get_datapath_version(const struct ofproto *ofproto_)
4230 {
4231     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4232
4233     return ofproto->backer->dp_version_string;
4234 }
4235
4236 static bool
4237 set_frag_handling(struct ofproto *ofproto_,
4238                   enum ofp_config_flags frag_handling)
4239 {
4240     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4241     if (frag_handling != OFPC_FRAG_REASM) {
4242         ofproto->backer->need_revalidate = REV_RECONFIGURE;
4243         return true;
4244     } else {
4245         return false;
4246     }
4247 }
4248
4249 static enum ofperr
4250 packet_out(struct ofproto *ofproto_, struct dp_packet *packet,
4251            const struct flow *flow,
4252            const struct ofpact *ofpacts, size_t ofpacts_len)
4253 {
4254     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4255
4256     ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
4257                                  ofpacts_len, packet);
4258     return 0;
4259 }
4260 \f
4261 /* NetFlow. */
4262
4263 static int
4264 set_netflow(struct ofproto *ofproto_,
4265             const struct netflow_options *netflow_options)
4266 {
4267     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4268
4269     if (netflow_options) {
4270         if (!ofproto->netflow) {
4271             ofproto->netflow = netflow_create();
4272             ofproto->backer->need_revalidate = REV_RECONFIGURE;
4273         }
4274         return netflow_set_options(ofproto->netflow, netflow_options);
4275     } else if (ofproto->netflow) {
4276         ofproto->backer->need_revalidate = REV_RECONFIGURE;
4277         netflow_unref(ofproto->netflow);
4278         ofproto->netflow = NULL;
4279     }
4280
4281     return 0;
4282 }
4283
4284 static void
4285 get_netflow_ids(const struct ofproto *ofproto_,
4286                 uint8_t *engine_type, uint8_t *engine_id)
4287 {
4288     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4289
4290     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
4291 }
4292 \f
4293 static struct ofproto_dpif *
4294 ofproto_dpif_lookup(const char *name)
4295 {
4296     struct ofproto_dpif *ofproto;
4297
4298     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
4299                              hash_string(name, 0), &all_ofproto_dpifs) {
4300         if (!strcmp(ofproto->up.name, name)) {
4301             return ofproto;
4302         }
4303     }
4304     return NULL;
4305 }
4306
4307 static void
4308 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
4309                           const char *argv[], void *aux OVS_UNUSED)
4310 {
4311     struct ofproto_dpif *ofproto;
4312
4313     if (argc > 1) {
4314         ofproto = ofproto_dpif_lookup(argv[1]);
4315         if (!ofproto) {
4316             unixctl_command_reply_error(conn, "no such bridge");
4317             return;
4318         }
4319         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4320         mac_learning_flush(ofproto->ml);
4321         ovs_rwlock_unlock(&ofproto->ml->rwlock);
4322     } else {
4323         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4324             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4325             mac_learning_flush(ofproto->ml);
4326             ovs_rwlock_unlock(&ofproto->ml->rwlock);
4327         }
4328     }
4329
4330     unixctl_command_reply(conn, "table successfully flushed");
4331 }
4332
4333 static void
4334 ofproto_unixctl_mcast_snooping_flush(struct unixctl_conn *conn, int argc,
4335                                      const char *argv[], void *aux OVS_UNUSED)
4336 {
4337     struct ofproto_dpif *ofproto;
4338
4339     if (argc > 1) {
4340         ofproto = ofproto_dpif_lookup(argv[1]);
4341         if (!ofproto) {
4342             unixctl_command_reply_error(conn, "no such bridge");
4343             return;
4344         }
4345
4346         if (!mcast_snooping_enabled(ofproto->ms)) {
4347             unixctl_command_reply_error(conn, "multicast snooping is disabled");
4348             return;
4349         }
4350         mcast_snooping_mdb_flush(ofproto->ms);
4351     } else {
4352         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4353             if (!mcast_snooping_enabled(ofproto->ms)) {
4354                 continue;
4355             }
4356             mcast_snooping_mdb_flush(ofproto->ms);
4357         }
4358     }
4359
4360     unixctl_command_reply(conn, "table successfully flushed");
4361 }
4362
4363 static struct ofport_dpif *
4364 ofbundle_get_a_port(const struct ofbundle *bundle)
4365 {
4366     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
4367                         bundle_node);
4368 }
4369
4370 static void
4371 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4372                          const char *argv[], void *aux OVS_UNUSED)
4373 {
4374     struct ds ds = DS_EMPTY_INITIALIZER;
4375     const struct ofproto_dpif *ofproto;
4376     const struct mac_entry *e;
4377
4378     ofproto = ofproto_dpif_lookup(argv[1]);
4379     if (!ofproto) {
4380         unixctl_command_reply_error(conn, "no such bridge");
4381         return;
4382     }
4383
4384     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
4385     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
4386     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
4387         struct ofbundle *bundle = mac_entry_get_port(ofproto->ml, e);
4388         char name[OFP_MAX_PORT_NAME_LEN];
4389
4390         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4391                                name, sizeof name);
4392         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
4393                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
4394                       mac_entry_age(ofproto->ml, e));
4395     }
4396     ovs_rwlock_unlock(&ofproto->ml->rwlock);
4397     unixctl_command_reply(conn, ds_cstr(&ds));
4398     ds_destroy(&ds);
4399 }
4400
4401 static void
4402 ofproto_unixctl_mcast_snooping_show(struct unixctl_conn *conn,
4403                                     int argc OVS_UNUSED,
4404                                     const char *argv[],
4405                                     void *aux OVS_UNUSED)
4406 {
4407     struct ds ds = DS_EMPTY_INITIALIZER;
4408     const struct ofproto_dpif *ofproto;
4409     const struct ofbundle *bundle;
4410     const struct mcast_group *grp;
4411     struct mcast_group_bundle *b;
4412     struct mcast_mrouter_bundle *mrouter;
4413
4414     ofproto = ofproto_dpif_lookup(argv[1]);
4415     if (!ofproto) {
4416         unixctl_command_reply_error(conn, "no such bridge");
4417         return;
4418     }
4419
4420     if (!mcast_snooping_enabled(ofproto->ms)) {
4421         unixctl_command_reply_error(conn, "multicast snooping is disabled");
4422         return;
4423     }
4424
4425     ds_put_cstr(&ds, " port  VLAN  GROUP                Age\n");
4426     ovs_rwlock_rdlock(&ofproto->ms->rwlock);
4427     LIST_FOR_EACH (grp, group_node, &ofproto->ms->group_lru) {
4428         LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
4429             char name[OFP_MAX_PORT_NAME_LEN];
4430
4431             bundle = b->port;
4432             ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4433                                    name, sizeof name);
4434             ds_put_format(&ds, "%5s  %4d  "IP_FMT"         %3d\n",
4435                           name, grp->vlan, IP_ARGS(grp->ip4),
4436                           mcast_bundle_age(ofproto->ms, b));
4437         }
4438     }
4439
4440     /* ports connected to multicast routers */
4441     LIST_FOR_EACH(mrouter, mrouter_node, &ofproto->ms->mrouter_lru) {
4442         char name[OFP_MAX_PORT_NAME_LEN];
4443
4444         bundle = mrouter->port;
4445         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4446                                name, sizeof name);
4447             ds_put_format(&ds, "%5s  %4d  querier             %3d\n",
4448                       name, mrouter->vlan,
4449                       mcast_mrouter_age(ofproto->ms, mrouter));
4450     }
4451     ovs_rwlock_unlock(&ofproto->ms->rwlock);
4452     unixctl_command_reply(conn, ds_cstr(&ds));
4453     ds_destroy(&ds);
4454 }
4455
4456 struct trace_ctx {
4457     struct xlate_out xout;
4458     struct xlate_in xin;
4459     const struct flow *key;
4460     struct flow flow;
4461     struct flow_wildcards wc;
4462     struct ds *result;
4463 };
4464
4465 static void
4466 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
4467 {
4468     const struct rule_actions *actions;
4469     ovs_be64 cookie;
4470
4471     ds_put_char_multiple(result, '\t', level);
4472     if (!rule) {
4473         ds_put_cstr(result, "No match\n");
4474         return;
4475     }
4476
4477     ovs_mutex_lock(&rule->up.mutex);
4478     cookie = rule->up.flow_cookie;
4479     ovs_mutex_unlock(&rule->up.mutex);
4480
4481     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
4482                   rule ? rule->up.table_id : 0, ntohll(cookie));
4483     cls_rule_format(&rule->up.cr, result);
4484     ds_put_char(result, '\n');
4485
4486     actions = rule_dpif_get_actions(rule);
4487
4488     ds_put_char_multiple(result, '\t', level);
4489     ds_put_cstr(result, "OpenFlow actions=");
4490     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
4491     ds_put_char(result, '\n');
4492 }
4493
4494 static void
4495 trace_format_flow(struct ds *result, int level, const char *title,
4496                   struct trace_ctx *trace)
4497 {
4498     ds_put_char_multiple(result, '\t', level);
4499     ds_put_format(result, "%s: ", title);
4500     /* Do not report unchanged flows for resubmits. */
4501     if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
4502         || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
4503         ds_put_cstr(result, "unchanged");
4504     } else {
4505         flow_format(result, &trace->xin.flow);
4506         trace->flow = trace->xin.flow;
4507     }
4508     ds_put_char(result, '\n');
4509 }
4510
4511 static void
4512 trace_format_regs(struct ds *result, int level, const char *title,
4513                   struct trace_ctx *trace)
4514 {
4515     size_t i;
4516
4517     ds_put_char_multiple(result, '\t', level);
4518     ds_put_format(result, "%s:", title);
4519     for (i = 0; i < FLOW_N_REGS; i++) {
4520         ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
4521     }
4522     ds_put_char(result, '\n');
4523 }
4524
4525 static void
4526 trace_format_odp(struct ds *result, int level, const char *title,
4527                  struct trace_ctx *trace)
4528 {
4529     struct ofpbuf *odp_actions = trace->xout.odp_actions;
4530
4531     ds_put_char_multiple(result, '\t', level);
4532     ds_put_format(result, "%s: ", title);
4533     format_odp_actions(result, odp_actions->data, odp_actions->size);
4534     ds_put_char(result, '\n');
4535 }
4536
4537 static void
4538 trace_format_megaflow(struct ds *result, int level, const char *title,
4539                       struct trace_ctx *trace)
4540 {
4541     struct match match;
4542
4543     ds_put_char_multiple(result, '\t', level);
4544     ds_put_format(result, "%s: ", title);
4545     flow_wildcards_or(&trace->wc, &trace->xout.wc, &trace->wc);
4546     match_init(&match, trace->key, &trace->wc);
4547     match_format(&match, result, OFP_DEFAULT_PRIORITY);
4548     ds_put_char(result, '\n');
4549 }
4550
4551 static void trace_report(struct xlate_in *xin, const char *s, int recurse);
4552
4553 static void
4554 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
4555 {
4556     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4557     struct ds *result = trace->result;
4558
4559     if (!recurse) {
4560         if (rule == xin->ofproto->miss_rule) {
4561             trace_report(xin, "No match, flow generates \"packet in\"s.",
4562                          recurse);
4563         } else if (rule == xin->ofproto->no_packet_in_rule) {
4564             trace_report(xin, "No match, packets dropped because "
4565                          "OFPPC_NO_PACKET_IN is set on in_port.", recurse);
4566         } else if (rule == xin->ofproto->drop_frags_rule) {
4567             trace_report(xin, "Packets dropped because they are IP "
4568                          "fragments and the fragment handling mode is "
4569                          "\"drop\".", recurse);
4570         }
4571     }
4572
4573     ds_put_char(result, '\n');
4574     if (recurse) {
4575         trace_format_flow(result, recurse, "Resubmitted flow", trace);
4576         trace_format_regs(result, recurse, "Resubmitted regs", trace);
4577         trace_format_odp(result,  recurse, "Resubmitted  odp", trace);
4578         trace_format_megaflow(result, recurse, "Resubmitted megaflow", trace);
4579     }
4580     trace_format_rule(result, recurse, rule);
4581 }
4582
4583 static void
4584 trace_report(struct xlate_in *xin, const char *s, int recurse)
4585 {
4586     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4587     struct ds *result = trace->result;
4588
4589     ds_put_char_multiple(result, '\t', recurse);
4590     ds_put_cstr(result, s);
4591     ds_put_char(result, '\n');
4592 }
4593
4594 /* Parses the 'argc' elements of 'argv', ignoring argv[0].  The following
4595  * forms are supported:
4596  *
4597  *     - [dpname] odp_flow [-generate | packet]
4598  *     - bridge br_flow [-generate | packet]
4599  *
4600  * On success, initializes '*ofprotop' and 'flow' and returns NULL.  On failure
4601  * returns a nonnull malloced error message. */
4602 static char * OVS_WARN_UNUSED_RESULT
4603 parse_flow_and_packet(int argc, const char *argv[],
4604                       struct ofproto_dpif **ofprotop, struct flow *flow,
4605                       struct dp_packet **packetp)
4606 {
4607     const struct dpif_backer *backer = NULL;
4608     const char *error = NULL;
4609     char *m_err = NULL;
4610     struct simap port_names = SIMAP_INITIALIZER(&port_names);
4611     struct dp_packet *packet;
4612     struct ofpbuf odp_key;
4613     struct ofpbuf odp_mask;
4614
4615     ofpbuf_init(&odp_key, 0);
4616     ofpbuf_init(&odp_mask, 0);
4617
4618     /* Handle "-generate" or a hex string as the last argument. */
4619     if (!strcmp(argv[argc - 1], "-generate")) {
4620         packet = dp_packet_new(0);
4621         argc--;
4622     } else {
4623         error = eth_from_hex(argv[argc - 1], &packet);
4624         if (!error) {
4625             argc--;
4626         } else if (argc == 4) {
4627             /* The 3-argument form must end in "-generate' or a hex string. */
4628             goto exit;
4629         }
4630         error = NULL;
4631     }
4632
4633     /* odp_flow can have its in_port specified as a name instead of port no.
4634      * We do not yet know whether a given flow is a odp_flow or a br_flow.
4635      * But, to know whether a flow is odp_flow through odp_flow_from_string(),
4636      * we need to create a simap of name to port no. */
4637     if (argc == 3) {
4638         const char *dp_type;
4639         if (!strncmp(argv[1], "ovs-", 4)) {
4640             dp_type = argv[1] + 4;
4641         } else {
4642             dp_type = argv[1];
4643         }
4644         backer = shash_find_data(&all_dpif_backers, dp_type);
4645     } else if (argc == 2) {
4646         struct shash_node *node;
4647         if (shash_count(&all_dpif_backers) == 1) {
4648             node = shash_first(&all_dpif_backers);
4649             backer = node->data;
4650         }
4651     } else {
4652         error = "Syntax error";
4653         goto exit;
4654     }
4655     if (backer && backer->dpif) {
4656         struct dpif_port dpif_port;
4657         struct dpif_port_dump port_dump;
4658         DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4659             simap_put(&port_names, dpif_port.name,
4660                       odp_to_u32(dpif_port.port_no));
4661         }
4662     }
4663
4664     /* Parse the flow and determine whether a datapath or
4665      * bridge is specified. If function odp_flow_key_from_string()
4666      * returns 0, the flow is a odp_flow. If function
4667      * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4668     if (!odp_flow_from_string(argv[argc - 1], &port_names,
4669                               &odp_key, &odp_mask)) {
4670         if (!backer) {
4671             error = "Cannot find the datapath";
4672             goto exit;
4673         }
4674
4675         if (odp_flow_key_to_flow(odp_key.data, odp_key.size, flow) == ODP_FIT_ERROR) {
4676             error = "Failed to parse datapath flow key";
4677             goto exit;
4678         }
4679
4680         *ofprotop = xlate_lookup_ofproto(backer, flow,
4681                                          &flow->in_port.ofp_port);
4682         if (*ofprotop == NULL) {
4683             error = "Invalid datapath flow";
4684             goto exit;
4685         }
4686
4687         vsp_adjust_flow(*ofprotop, flow, NULL);
4688
4689     } else {
4690         char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4691
4692         if (err) {
4693             m_err = xasprintf("Bad openflow flow syntax: %s", err);
4694             free(err);
4695             goto exit;
4696         } else {
4697             if (argc != 3) {
4698                 error = "Must specify bridge name";
4699                 goto exit;
4700             }
4701
4702             *ofprotop = ofproto_dpif_lookup(argv[1]);
4703             if (!*ofprotop) {
4704                 error = "Unknown bridge name";
4705                 goto exit;
4706             }
4707         }
4708     }
4709
4710     /* Generate a packet, if requested. */
4711     if (packet) {
4712         if (!dp_packet_size(packet)) {
4713             flow_compose(packet, flow);
4714         } else {
4715             /* Use the metadata from the flow and the packet argument
4716              * to reconstruct the flow. */
4717             pkt_metadata_from_flow(&packet->md, flow);
4718             flow_extract(packet, flow);
4719         }
4720     }
4721
4722 exit:
4723     if (error && !m_err) {
4724         m_err = xstrdup(error);
4725     }
4726     if (m_err) {
4727         dp_packet_delete(packet);
4728         packet = NULL;
4729     }
4730     *packetp = packet;
4731     ofpbuf_uninit(&odp_key);
4732     ofpbuf_uninit(&odp_mask);
4733     simap_destroy(&port_names);
4734     return m_err;
4735 }
4736
4737 static void
4738 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4739                       void *aux OVS_UNUSED)
4740 {
4741     struct ofproto_dpif *ofproto;
4742     struct dp_packet *packet;
4743     char *error;
4744     struct flow flow;
4745
4746     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4747     if (!error) {
4748         struct ds result;
4749
4750         ds_init(&result);
4751         ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4752         unixctl_command_reply(conn, ds_cstr(&result));
4753         ds_destroy(&result);
4754         dp_packet_delete(packet);
4755     } else {
4756         unixctl_command_reply_error(conn, error);
4757         free(error);
4758     }
4759 }
4760
4761 static void
4762 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4763                               const char *argv[], void *aux OVS_UNUSED)
4764 {
4765     enum ofputil_protocol usable_protocols;
4766     struct ofproto_dpif *ofproto;
4767     bool enforce_consistency;
4768     struct ofpbuf ofpacts;
4769     struct dp_packet *packet;
4770     struct ds result;
4771     struct flow flow;
4772     uint16_t in_port;
4773
4774     /* Three kinds of error return values! */
4775     enum ofperr retval;
4776     char *error;
4777
4778     packet = NULL;
4779     ds_init(&result);
4780     ofpbuf_init(&ofpacts, 0);
4781
4782     /* Parse actions. */
4783     error = ofpacts_parse_actions(argv[--argc], &ofpacts, &usable_protocols);
4784     if (error) {
4785         unixctl_command_reply_error(conn, error);
4786         free(error);
4787         goto exit;
4788     }
4789
4790     /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4791      * consistency between the flow and the actions.  With -consistent, we
4792      * enforce consistency even for a flow supported in OpenFlow 1.0. */
4793     if (!strcmp(argv[1], "-consistent")) {
4794         enforce_consistency = true;
4795         argv++;
4796         argc--;
4797     } else {
4798         enforce_consistency = false;
4799     }
4800
4801     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4802     if (error) {
4803         unixctl_command_reply_error(conn, error);
4804         free(error);
4805         goto exit;
4806     }
4807
4808     /* Do the same checks as handle_packet_out() in ofproto.c.
4809      *
4810      * We pass a 'table_id' of 0 to ofpacts_check(), which isn't
4811      * strictly correct because these actions aren't in any table, but it's OK
4812      * because it 'table_id' is used only to check goto_table instructions, but
4813      * packet-outs take a list of actions and therefore it can't include
4814      * instructions.
4815      *
4816      * We skip the "meter" check here because meter is an instruction, not an
4817      * action, and thus cannot appear in ofpacts. */
4818     in_port = ofp_to_u16(flow.in_port.ofp_port);
4819     if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
4820         unixctl_command_reply_error(conn, "invalid in_port");
4821         goto exit;
4822     }
4823     if (enforce_consistency) {
4824         retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size,
4825                                            &flow, u16_to_ofp(ofproto->up.max_ports),
4826                                            0, 0, usable_protocols);
4827     } else {
4828         retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow,
4829                                u16_to_ofp(ofproto->up.max_ports), 0, 0,
4830                                &usable_protocols);
4831     }
4832
4833     if (retval) {
4834         ds_clear(&result);
4835         ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
4836         unixctl_command_reply_error(conn, ds_cstr(&result));
4837         goto exit;
4838     }
4839
4840     ofproto_trace(ofproto, &flow, packet,
4841                   ofpacts.data, ofpacts.size, &result);
4842     unixctl_command_reply(conn, ds_cstr(&result));
4843
4844 exit:
4845     ds_destroy(&result);
4846     dp_packet_delete(packet);
4847     ofpbuf_uninit(&ofpacts);
4848 }
4849
4850 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4851  * description of the results to 'ds'.
4852  *
4853  * The trace follows a packet with the specified 'flow' through the flow
4854  * table.  'packet' may be nonnull to trace an actual packet, with consequent
4855  * side effects (if it is nonnull then its flow must be 'flow').
4856  *
4857  * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4858  * trace, otherwise the actions are determined by a flow table lookup. */
4859 static void
4860 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
4861               const struct dp_packet *packet,
4862               const struct ofpact ofpacts[], size_t ofpacts_len,
4863               struct ds *ds)
4864 {
4865     struct trace_ctx trace;
4866
4867     ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
4868     ds_put_cstr(ds, "Flow: ");
4869     flow_format(ds, flow);
4870     ds_put_char(ds, '\n');
4871
4872     flow_wildcards_init_catchall(&trace.wc);
4873
4874     trace.result = ds;
4875     trace.key = flow; /* Original flow key, used for megaflow. */
4876     trace.flow = *flow; /* May be modified by actions. */
4877     xlate_in_init(&trace.xin, ofproto, flow, flow->in_port.ofp_port, NULL,
4878                   ntohs(flow->tcp_flags), packet);
4879     trace.xin.ofpacts = ofpacts;
4880     trace.xin.ofpacts_len = ofpacts_len;
4881     trace.xin.resubmit_hook = trace_resubmit;
4882     trace.xin.report_hook = trace_report;
4883
4884     xlate_actions(&trace.xin, &trace.xout);
4885
4886     ds_put_char(ds, '\n');
4887     trace_format_flow(ds, 0, "Final flow", &trace);
4888     trace_format_megaflow(ds, 0, "Megaflow", &trace);
4889
4890     ds_put_cstr(ds, "Datapath actions: ");
4891     format_odp_actions(ds, trace.xout.odp_actions->data,
4892                        trace.xout.odp_actions->size);
4893
4894     if (trace.xout.slow) {
4895         enum slow_path_reason slow;
4896
4897         ds_put_cstr(ds, "\nThis flow is handled by the userspace "
4898                     "slow path because it:");
4899
4900         slow = trace.xout.slow;
4901         while (slow) {
4902             enum slow_path_reason bit = rightmost_1bit(slow);
4903
4904             ds_put_format(ds, "\n\t- %s.",
4905                           slow_path_reason_to_explanation(bit));
4906
4907             slow &= ~bit;
4908         }
4909     }
4910
4911     xlate_out_uninit(&trace.xout);
4912 }
4913
4914 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
4915  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
4916  * to destroy 'ofproto_shash' and free the returned value. */
4917 static const struct shash_node **
4918 get_ofprotos(struct shash *ofproto_shash)
4919 {
4920     const struct ofproto_dpif *ofproto;
4921
4922     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4923         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
4924         shash_add_nocopy(ofproto_shash, name, ofproto);
4925     }
4926
4927     return shash_sort(ofproto_shash);
4928 }
4929
4930 static void
4931 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
4932                               const char *argv[] OVS_UNUSED,
4933                               void *aux OVS_UNUSED)
4934 {
4935     struct ds ds = DS_EMPTY_INITIALIZER;
4936     struct shash ofproto_shash;
4937     const struct shash_node **sorted_ofprotos;
4938     int i;
4939
4940     shash_init(&ofproto_shash);
4941     sorted_ofprotos = get_ofprotos(&ofproto_shash);
4942     for (i = 0; i < shash_count(&ofproto_shash); i++) {
4943         const struct shash_node *node = sorted_ofprotos[i];
4944         ds_put_format(&ds, "%s\n", node->name);
4945     }
4946
4947     shash_destroy(&ofproto_shash);
4948     free(sorted_ofprotos);
4949
4950     unixctl_command_reply(conn, ds_cstr(&ds));
4951     ds_destroy(&ds);
4952 }
4953
4954 static void
4955 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
4956 {
4957     const struct shash_node **ofprotos;
4958     struct dpif_dp_stats dp_stats;
4959     struct shash ofproto_shash;
4960     size_t i;
4961
4962     dpif_get_dp_stats(backer->dpif, &dp_stats);
4963
4964     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
4965                   dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
4966
4967     shash_init(&ofproto_shash);
4968     ofprotos = get_ofprotos(&ofproto_shash);
4969     for (i = 0; i < shash_count(&ofproto_shash); i++) {
4970         struct ofproto_dpif *ofproto = ofprotos[i]->data;
4971         const struct shash_node **ports;
4972         size_t j;
4973
4974         if (ofproto->backer != backer) {
4975             continue;
4976         }
4977
4978         ds_put_format(ds, "\t%s:\n", ofproto->up.name);
4979
4980         ports = shash_sort(&ofproto->up.port_by_name);
4981         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
4982             const struct shash_node *node = ports[j];
4983             struct ofport *ofport = node->data;
4984             struct smap config;
4985             odp_port_t odp_port;
4986
4987             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
4988                           ofport->ofp_port);
4989
4990             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
4991             if (odp_port != ODPP_NONE) {
4992                 ds_put_format(ds, "%"PRIu32":", odp_port);
4993             } else {
4994                 ds_put_cstr(ds, "none:");
4995             }
4996
4997             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
4998
4999             smap_init(&config);
5000             if (!netdev_get_config(ofport->netdev, &config)) {
5001                 const struct smap_node **nodes;
5002                 size_t i;
5003
5004                 nodes = smap_sort(&config);
5005                 for (i = 0; i < smap_count(&config); i++) {
5006                     const struct smap_node *node = nodes[i];
5007                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5008                                   node->key, node->value);
5009                 }
5010                 free(nodes);
5011             }
5012             smap_destroy(&config);
5013
5014             ds_put_char(ds, ')');
5015             ds_put_char(ds, '\n');
5016         }
5017         free(ports);
5018     }
5019     shash_destroy(&ofproto_shash);
5020     free(ofprotos);
5021 }
5022
5023 static void
5024 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5025                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5026 {
5027     struct ds ds = DS_EMPTY_INITIALIZER;
5028     const struct shash_node **backers;
5029     int i;
5030
5031     backers = shash_sort(&all_dpif_backers);
5032     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5033         dpif_show_backer(backers[i]->data, &ds);
5034     }
5035     free(backers);
5036
5037     unixctl_command_reply(conn, ds_cstr(&ds));
5038     ds_destroy(&ds);
5039 }
5040
5041 static void
5042 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5043                                 int argc OVS_UNUSED, const char *argv[],
5044                                 void *aux OVS_UNUSED)
5045 {
5046     const struct ofproto_dpif *ofproto;
5047
5048     struct ds ds = DS_EMPTY_INITIALIZER;
5049     bool verbosity = false;
5050
5051     struct dpif_port dpif_port;
5052     struct dpif_port_dump port_dump;
5053     struct hmap portno_names;
5054
5055     struct dpif_flow_dump *flow_dump;
5056     struct dpif_flow_dump_thread *flow_dump_thread;
5057     struct dpif_flow f;
5058     int error;
5059
5060     ofproto = ofproto_dpif_lookup(argv[argc - 1]);
5061     if (!ofproto) {
5062         unixctl_command_reply_error(conn, "no such bridge");
5063         return;
5064     }
5065
5066     if (argc > 2 && !strcmp(argv[1], "-m")) {
5067         verbosity = true;
5068     }
5069
5070     hmap_init(&portno_names);
5071     DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
5072         odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
5073     }
5074
5075     ds_init(&ds);
5076     flow_dump = dpif_flow_dump_create(ofproto->backer->dpif, false);
5077     flow_dump_thread = dpif_flow_dump_thread_create(flow_dump);
5078     while (dpif_flow_dump_next(flow_dump_thread, &f, 1)) {
5079         struct flow flow;
5080
5081         if (odp_flow_key_to_flow(f.key, f.key_len, &flow) == ODP_FIT_ERROR
5082             || xlate_lookup_ofproto(ofproto->backer, &flow, NULL) != ofproto) {
5083             continue;
5084         }
5085
5086         if (verbosity) {
5087             odp_format_ufid(&f.ufid, &ds);
5088             ds_put_cstr(&ds, " ");
5089         }
5090         odp_flow_format(f.key, f.key_len, f.mask, f.mask_len,
5091                         &portno_names, &ds, verbosity);
5092         ds_put_cstr(&ds, ", ");
5093         dpif_flow_stats_format(&f.stats, &ds);
5094         ds_put_cstr(&ds, ", actions:");
5095         format_odp_actions(&ds, f.actions, f.actions_len);
5096         ds_put_char(&ds, '\n');
5097     }
5098     dpif_flow_dump_thread_destroy(flow_dump_thread);
5099     error = dpif_flow_dump_destroy(flow_dump);
5100
5101     if (error) {
5102         ds_clear(&ds);
5103         ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
5104         unixctl_command_reply_error(conn, ds_cstr(&ds));
5105     } else {
5106         unixctl_command_reply(conn, ds_cstr(&ds));
5107     }
5108     odp_portno_names_destroy(&portno_names);
5109     hmap_destroy(&portno_names);
5110     ds_destroy(&ds);
5111 }
5112
5113 static void
5114 ofproto_revalidate_all_backers(void)
5115 {
5116     const struct shash_node **backers;
5117     int i;
5118
5119     backers = shash_sort(&all_dpif_backers);
5120     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5121         struct dpif_backer *backer = backers[i]->data;
5122         backer->need_revalidate = REV_RECONFIGURE;
5123     }
5124     free(backers);
5125 }
5126
5127 static void
5128 disable_tnl_push_pop(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5129                      const char *argv[], void *aux OVS_UNUSED)
5130 {
5131     if (!strcasecmp(argv[1], "off")) {
5132         ofproto_use_tnl_push_pop = false;
5133         unixctl_command_reply(conn, "Tunnel push-pop off");
5134         ofproto_revalidate_all_backers();
5135     } else if (!strcasecmp(argv[1], "on")) {
5136         ofproto_use_tnl_push_pop = true;
5137         unixctl_command_reply(conn, "Tunnel push-pop on");
5138         ofproto_revalidate_all_backers();
5139     }
5140 }
5141
5142 static void
5143 ofproto_unixctl_init(void)
5144 {
5145     static bool registered;
5146     if (registered) {
5147         return;
5148     }
5149     registered = true;
5150
5151     unixctl_command_register(
5152         "ofproto/trace",
5153         "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
5154         1, 3, ofproto_unixctl_trace, NULL);
5155     unixctl_command_register(
5156         "ofproto/trace-packet-out",
5157         "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
5158         2, 6, ofproto_unixctl_trace_actions, NULL);
5159     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5160                              ofproto_unixctl_fdb_flush, NULL);
5161     unixctl_command_register("fdb/show", "bridge", 1, 1,
5162                              ofproto_unixctl_fdb_show, NULL);
5163     unixctl_command_register("mdb/flush", "[bridge]", 0, 1,
5164                              ofproto_unixctl_mcast_snooping_flush, NULL);
5165     unixctl_command_register("mdb/show", "bridge", 1, 1,
5166                              ofproto_unixctl_mcast_snooping_show, NULL);
5167     unixctl_command_register("dpif/dump-dps", "", 0, 0,
5168                              ofproto_unixctl_dpif_dump_dps, NULL);
5169     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5170                              NULL);
5171     unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
5172                              ofproto_unixctl_dpif_dump_flows, NULL);
5173
5174     unixctl_command_register("ofproto/tnl-push-pop", "[on]|[off]", 1, 1,
5175                              disable_tnl_push_pop, NULL);
5176 }
5177
5178 /* Returns true if 'table' is the table used for internal rules,
5179  * false otherwise. */
5180 bool
5181 table_is_internal(uint8_t table_id)
5182 {
5183     return table_id == TBL_INTERNAL;
5184 }
5185 \f
5186 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5187  *
5188  * This is deprecated.  It is only for compatibility with broken device drivers
5189  * in old versions of Linux that do not properly support VLANs when VLAN
5190  * devices are not used.  When broken device drivers are no longer in
5191  * widespread use, we will delete these interfaces. */
5192
5193 static int
5194 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5195 {
5196     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5197     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5198
5199     if (realdev_ofp_port == ofport->realdev_ofp_port
5200         && vid == ofport->vlandev_vid) {
5201         return 0;
5202     }
5203
5204     ofproto->backer->need_revalidate = REV_RECONFIGURE;
5205
5206     if (ofport->realdev_ofp_port) {
5207         vsp_remove(ofport);
5208     }
5209     if (realdev_ofp_port && ofport->bundle) {
5210         /* vlandevs are enslaved to their realdevs, so they are not allowed to
5211          * themselves be part of a bundle. */
5212         bundle_set(ofport_->ofproto, ofport->bundle, NULL);
5213     }
5214
5215     ofport->realdev_ofp_port = realdev_ofp_port;
5216     ofport->vlandev_vid = vid;
5217
5218     if (realdev_ofp_port) {
5219         vsp_add(ofport, realdev_ofp_port, vid);
5220     }
5221
5222     return 0;
5223 }
5224
5225 static uint32_t
5226 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
5227 {
5228     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
5229 }
5230
5231 bool
5232 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
5233     OVS_EXCLUDED(ofproto->vsp_mutex)
5234 {
5235     /* hmap_is_empty is thread safe. */
5236     return !hmap_is_empty(&ofproto->realdev_vid_map);
5237 }
5238
5239
5240 static ofp_port_t
5241 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
5242                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5243     OVS_REQUIRES(ofproto->vsp_mutex)
5244 {
5245     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5246         int vid = vlan_tci_to_vid(vlan_tci);
5247         const struct vlan_splinter *vsp;
5248
5249         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5250                                  hash_realdev_vid(realdev_ofp_port, vid),
5251                                  &ofproto->realdev_vid_map) {
5252             if (vsp->realdev_ofp_port == realdev_ofp_port
5253                 && vsp->vid == vid) {
5254                 return vsp->vlandev_ofp_port;
5255             }
5256         }
5257     }
5258     return realdev_ofp_port;
5259 }
5260
5261 /* Returns the OFP port number of the Linux VLAN device that corresponds to
5262  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
5263  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
5264  * 'vlan_tci' 9, it would return the port number of eth0.9.
5265  *
5266  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
5267  * function just returns its 'realdev_ofp_port' argument. */
5268 ofp_port_t
5269 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5270                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5271     OVS_EXCLUDED(ofproto->vsp_mutex)
5272 {
5273     ofp_port_t ret;
5274
5275     /* hmap_is_empty is thread safe, see if we can return immediately. */
5276     if (hmap_is_empty(&ofproto->realdev_vid_map)) {
5277         return realdev_ofp_port;
5278     }
5279     ovs_mutex_lock(&ofproto->vsp_mutex);
5280     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
5281     ovs_mutex_unlock(&ofproto->vsp_mutex);
5282     return ret;
5283 }
5284
5285 static struct vlan_splinter *
5286 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
5287 {
5288     struct vlan_splinter *vsp;
5289
5290     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
5291                              hash_ofp_port(vlandev_ofp_port),
5292                              &ofproto->vlandev_map) {
5293         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5294             return vsp;
5295         }
5296     }
5297
5298     return NULL;
5299 }
5300
5301 /* Returns the OpenFlow port number of the "real" device underlying the Linux
5302  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
5303  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
5304  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
5305  * eth0 and store 9 in '*vid'.
5306  *
5307  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
5308  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
5309  * always does.*/
5310 static ofp_port_t
5311 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5312                        ofp_port_t vlandev_ofp_port, int *vid)
5313     OVS_REQUIRES(ofproto->vsp_mutex)
5314 {
5315     if (!hmap_is_empty(&ofproto->vlandev_map)) {
5316         const struct vlan_splinter *vsp;
5317
5318         vsp = vlandev_find(ofproto, vlandev_ofp_port);
5319         if (vsp) {
5320             if (vid) {
5321                 *vid = vsp->vid;
5322             }
5323             return vsp->realdev_ofp_port;
5324         }
5325     }
5326     return 0;
5327 }
5328
5329 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
5330  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
5331  * 'flow->in_port' to the "real" device backing the VLAN device, sets
5332  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Optionally pushes the
5333  * appropriate VLAN on 'packet' if provided.  Otherwise (which is always the
5334  * case unless VLAN splinters are enabled), returns false without making any
5335  * changes. */
5336 bool
5337 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow,
5338                 struct dp_packet *packet)
5339     OVS_EXCLUDED(ofproto->vsp_mutex)
5340 {
5341     ofp_port_t realdev;
5342     int vid;
5343
5344     /* hmap_is_empty is thread safe. */
5345     if (hmap_is_empty(&ofproto->vlandev_map)) {
5346         return false;
5347     }
5348
5349     ovs_mutex_lock(&ofproto->vsp_mutex);
5350     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
5351     ovs_mutex_unlock(&ofproto->vsp_mutex);
5352     if (!realdev) {
5353         return false;
5354     }
5355
5356     /* Cause the flow to be processed as if it came in on the real device with
5357      * the VLAN device's VLAN ID. */
5358     flow->in_port.ofp_port = realdev;
5359     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
5360
5361     if (packet) {
5362         /* Make the packet resemble the flow, so that it gets sent to an
5363          * OpenFlow controller properly, so that it looks correct for sFlow,
5364          * and so that flow_extract() will get the correct vlan_tci if it is
5365          * called on 'packet'. */
5366         eth_push_vlan(packet, htons(ETH_TYPE_VLAN), flow->vlan_tci);
5367     }
5368
5369     return true;
5370 }
5371
5372 static void
5373 vsp_remove(struct ofport_dpif *port)
5374 {
5375     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5376     struct vlan_splinter *vsp;
5377
5378     ovs_mutex_lock(&ofproto->vsp_mutex);
5379     vsp = vlandev_find(ofproto, port->up.ofp_port);
5380     if (vsp) {
5381         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5382         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
5383         free(vsp);
5384
5385         port->realdev_ofp_port = 0;
5386     } else {
5387         VLOG_ERR("missing vlan device record");
5388     }
5389     ovs_mutex_unlock(&ofproto->vsp_mutex);
5390 }
5391
5392 static void
5393 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
5394 {
5395     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5396
5397     ovs_mutex_lock(&ofproto->vsp_mutex);
5398     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
5399         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
5400             == realdev_ofp_port)) {
5401         struct vlan_splinter *vsp;
5402
5403         vsp = xmalloc(sizeof *vsp);
5404         vsp->realdev_ofp_port = realdev_ofp_port;
5405         vsp->vlandev_ofp_port = port->up.ofp_port;
5406         vsp->vid = vid;
5407
5408         port->realdev_ofp_port = realdev_ofp_port;
5409
5410         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
5411                     hash_ofp_port(port->up.ofp_port));
5412         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
5413                     hash_realdev_vid(realdev_ofp_port, vid));
5414     } else {
5415         VLOG_ERR("duplicate vlan device record");
5416     }
5417     ovs_mutex_unlock(&ofproto->vsp_mutex);
5418 }
5419
5420 static odp_port_t
5421 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
5422 {
5423     const struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
5424     return ofport ? ofport->odp_port : ODPP_NONE;
5425 }
5426
5427 struct ofport_dpif *
5428 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
5429 {
5430     struct ofport_dpif *port;
5431
5432     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
5433     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
5434                              &backer->odp_to_ofport_map) {
5435         if (port->odp_port == odp_port) {
5436             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5437             return port;
5438         }
5439     }
5440
5441     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5442     return NULL;
5443 }
5444
5445 static ofp_port_t
5446 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
5447 {
5448     struct ofport_dpif *port;
5449
5450     port = odp_port_to_ofport(ofproto->backer, odp_port);
5451     if (port && &ofproto->up == port->up.ofproto) {
5452         return port->up.ofp_port;
5453     } else {
5454         return OFPP_NONE;
5455     }
5456 }
5457
5458 int
5459 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
5460                                const struct match *match, int priority,
5461                                uint16_t idle_timeout,
5462                                const struct ofpbuf *ofpacts,
5463                                struct rule **rulep)
5464 {
5465     struct ofputil_flow_mod fm;
5466     struct rule_dpif *rule;
5467     int error;
5468
5469     fm.match = *match;
5470     fm.priority = priority;
5471     fm.new_cookie = htonll(0);
5472     fm.cookie = htonll(0);
5473     fm.cookie_mask = htonll(0);
5474     fm.modify_cookie = false;
5475     fm.table_id = TBL_INTERNAL;
5476     fm.command = OFPFC_ADD;
5477     fm.idle_timeout = idle_timeout;
5478     fm.hard_timeout = 0;
5479     fm.importance = 0;
5480     fm.buffer_id = 0;
5481     fm.out_port = 0;
5482     fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5483     fm.ofpacts = ofpacts->data;
5484     fm.ofpacts_len = ofpacts->size;
5485
5486     error = ofproto_flow_mod(&ofproto->up, &fm);
5487     if (error) {
5488         VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
5489                     ofperr_to_string(error));
5490         *rulep = NULL;
5491         return error;
5492     }
5493
5494     rule = rule_dpif_lookup_in_table(ofproto,
5495                                      ofproto_dpif_get_tables_version(ofproto),
5496                                      TBL_INTERNAL, &fm.match.flow,
5497                                      &fm.match.wc, false);
5498     if (rule) {
5499         *rulep = &rule->up;
5500     } else {
5501         OVS_NOT_REACHED();
5502     }
5503     return 0;
5504 }
5505
5506 int
5507 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
5508                                   struct match *match, int priority)
5509 {
5510     struct ofputil_flow_mod fm;
5511     int error;
5512
5513     fm.match = *match;
5514     fm.priority = priority;
5515     fm.new_cookie = htonll(0);
5516     fm.cookie = htonll(0);
5517     fm.cookie_mask = htonll(0);
5518     fm.modify_cookie = false;
5519     fm.table_id = TBL_INTERNAL;
5520     fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5521     fm.command = OFPFC_DELETE_STRICT;
5522
5523     error = ofproto_flow_mod(&ofproto->up, &fm);
5524     if (error) {
5525         VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
5526                     ofperr_to_string(error));
5527         return error;
5528     }
5529
5530     return 0;
5531 }
5532
5533 const struct ofproto_class ofproto_dpif_class = {
5534     init,
5535     enumerate_types,
5536     enumerate_names,
5537     del,
5538     port_open_type,
5539     type_run,
5540     type_wait,
5541     alloc,
5542     construct,
5543     destruct,
5544     dealloc,
5545     run,
5546     wait,
5547     NULL,                       /* get_memory_usage. */
5548     type_get_memory_usage,
5549     flush,
5550     query_tables,
5551     set_tables_version,
5552     port_alloc,
5553     port_construct,
5554     port_destruct,
5555     port_dealloc,
5556     port_modified,
5557     port_reconfigured,
5558     port_query_by_name,
5559     port_add,
5560     port_del,
5561     port_get_stats,
5562     port_dump_start,
5563     port_dump_next,
5564     port_dump_done,
5565     port_poll,
5566     port_poll_wait,
5567     port_is_lacp_current,
5568     port_get_lacp_stats,
5569     NULL,                       /* rule_choose_table */
5570     rule_alloc,
5571     rule_construct,
5572     rule_insert,
5573     rule_delete,
5574     rule_destruct,
5575     rule_dealloc,
5576     rule_get_stats,
5577     rule_execute,
5578     set_frag_handling,
5579     packet_out,
5580     set_netflow,
5581     get_netflow_ids,
5582     set_sflow,
5583     set_ipfix,
5584     set_cfm,
5585     cfm_status_changed,
5586     get_cfm_status,
5587     set_lldp,
5588     get_lldp_status,
5589     set_aa,
5590     aa_mapping_set,
5591     aa_mapping_unset,
5592     aa_vlan_get_queued,
5593     aa_vlan_get_queue_size,
5594     set_bfd,
5595     bfd_status_changed,
5596     get_bfd_status,
5597     set_stp,
5598     get_stp_status,
5599     set_stp_port,
5600     get_stp_port_status,
5601     get_stp_port_stats,
5602     set_rstp,
5603     get_rstp_status,
5604     set_rstp_port,
5605     get_rstp_port_status,
5606     set_queues,
5607     bundle_set,
5608     bundle_remove,
5609     mirror_set__,
5610     mirror_get_stats__,
5611     set_flood_vlans,
5612     is_mirror_output_bundle,
5613     forward_bpdu_changed,
5614     set_mac_table_config,
5615     set_mcast_snooping,
5616     set_mcast_snooping_port,
5617     set_realdev,
5618     NULL,                       /* meter_get_features */
5619     NULL,                       /* meter_set */
5620     NULL,                       /* meter_get */
5621     NULL,                       /* meter_del */
5622     group_alloc,                /* group_alloc */
5623     group_construct,            /* group_construct */
5624     group_destruct,             /* group_destruct */
5625     group_dealloc,              /* group_dealloc */
5626     group_modify,               /* group_modify */
5627     group_get_stats,            /* group_get_stats */
5628     get_datapath_version,       /* get_datapath_version */
5629 };