ofproto/ofproto-dpif.c

   1 /*
   2  * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 Nicira, Inc.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18 #include <errno.h>
  19
  20 #include "bfd.h"
  21 #include "bond.h"
  22 #include "bundle.h"
  23 #include "byte-order.h"
  24 #include "connectivity.h"
  25 #include "connmgr.h"
  26 #include "coverage.h"
  27 #include "cfm.h"
  28 #include "dpif.h"
  29 #include "fail-open.h"
  30 #include "guarded-list.h"
  31 #include "hmapx.h"
  32 #include "lacp.h"
  33 #include "learn.h"
  34 #include "mac-learning.h"
  35 #include "mcast-snooping.h"
  36 #include "multipath.h"
  37 #include "netdev-vport.h"
  38 #include "netdev.h"
  39 #include "netlink.h"
  40 #include "nx-match.h"
  41 #include "odp-util.h"
  42 #include "odp-execute.h"
  43 #include "ofproto/ofproto-dpif.h"
  44 #include "ofproto/ofproto-provider.h"
  45 #include "ofproto-dpif-ipfix.h"
  46 #include "ofproto-dpif-mirror.h"
  47 #include "ofproto-dpif-monitor.h"
  48 #include "ofproto-dpif-rid.h"
  49 #include "ofproto-dpif-sflow.h"
  50 #include "ofproto-dpif-upcall.h"
  51 #include "ofproto-dpif-xlate.h"
  52 #include "openvswitch/ofp-actions.h"
  53 #include "openvswitch/dynamic-string.h"
  54 #include "openvswitch/meta-flow.h"
  55 #include "openvswitch/ofp-parse.h"
  56 #include "openvswitch/ofp-print.h"
  57 #include "openvswitch/ofp-util.h"
  58 #include "openvswitch/ofpbuf.h"
  59 #include "openvswitch/vlog.h"
  60 #include "ovs-lldp.h"
  61 #include "ovs-rcu.h"
  62 #include "ovs-router.h"
  63 #include "poll-loop.h"
  64 #include "seq.h"
  65 #include "simap.h"
  66 #include "smap.h"
  67 #include "timer.h"
  68 #include "tunnel.h"
  69 #include "unaligned.h"
  70 #include "unixctl.h"
  71 #include "vlan-bitmap.h"
  72
  73 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  74
  75 COVERAGE_DEFINE(ofproto_dpif_expired);
  76 COVERAGE_DEFINE(packet_in_overflow);
  77
  78 struct flow_miss;
  79
  80 struct rule_dpif {
  81     struct rule up;
  82
  83     /* These statistics:
  84      *
  85      *   - Do include packets and bytes from datapath flows which have not
  86      *   recently been processed by a revalidator. */
  87     struct ovs_mutex stats_mutex;
  88     struct dpif_flow_stats stats OVS_GUARDED;
  89
  90    /* In non-NULL, will point to a new rule (for which a reference is held) to
  91     * which all the stats updates should be forwarded. This exists only
  92     * transitionally when flows are replaced.
  93     *
  94     * Protected by stats_mutex.  If both 'rule->stats_mutex' and
  95     * 'rule->new_rule->stats_mutex' must be held together, acquire them in that
  96     * order, */
  97     struct rule_dpif *new_rule OVS_GUARDED;
  98
  99     /* If non-zero then the recirculation id that has
 100      * been allocated for use with this rule.
 101      * The recirculation id and associated internal flow should
 102      * be freed when the rule is freed */
 103     uint32_t recirc_id;
 104 };
 105
 106 /* RULE_CAST() depends on this. */
 107 BUILD_ASSERT_DECL(offsetof(struct rule_dpif, up) == 0);
 108
 109 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
 110                            long long int *used);
 111 static struct rule_dpif *rule_dpif_cast(const struct rule *);
 112 static void rule_expire(struct rule_dpif *, long long now);
 113
 114 struct group_dpif {
 115     struct ofgroup up;
 116
 117     /* These statistics:
 118      *
 119      *   - Do include packets and bytes from datapath flows which have not
 120      *   recently been processed by a revalidator. */
 121     struct ovs_mutex stats_mutex;
 122     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 123     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 124 };
 125
 126 struct ofbundle {
 127     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
 128     struct ofproto_dpif *ofproto; /* Owning ofproto. */
 129     void *aux;                  /* Key supplied by ofproto's client. */
 130     char *name;                 /* Identifier for log messages. */
 131
 132     /* Configuration. */
 133     struct ovs_list ports;      /* Contains "struct ofport"s. */
 134     enum port_vlan_mode vlan_mode; /* VLAN mode */
 135     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
 136     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
 137                                  * NULL if all VLANs are trunked. */
 138     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 139     struct bond *bond;          /* Nonnull iff more than one port. */
 140     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 141
 142     /* Status. */
 143     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 144 };
 145
 146 static void bundle_remove(struct ofport *);
 147 static void bundle_update(struct ofbundle *);
 148 static void bundle_destroy(struct ofbundle *);
 149 static void bundle_del_port(struct ofport_dpif *);
 150 static void bundle_run(struct ofbundle *);
 151 static void bundle_wait(struct ofbundle *);
 152 static void bundle_flush_macs(struct ofbundle *, bool);
 153 static void bundle_move(struct ofbundle *, struct ofbundle *);
 154
 155 static void stp_run(struct ofproto_dpif *ofproto);
 156 static void stp_wait(struct ofproto_dpif *ofproto);
 157 static int set_stp_port(struct ofport *,
 158                         const struct ofproto_port_stp_settings *);
 159
 160 static void rstp_run(struct ofproto_dpif *ofproto);
 161 static void set_rstp_port(struct ofport *,
 162                          const struct ofproto_port_rstp_settings *);
 163
 164 struct ofport_dpif {
 165     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 166     struct ofport up;
 167
 168     odp_port_t odp_port;
 169     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 170     struct ovs_list bundle_node;/* In struct ofbundle's "ports" list. */
 171     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 172     struct bfd *bfd;            /* BFD, if any. */
 173     struct lldp *lldp;          /* lldp, if any. */
 174     bool may_enable;            /* May be enabled in bonds. */
 175     bool is_tunnel;             /* This port is a tunnel. */
 176     bool is_layer3;             /* This is a layer 3 port. */
 177     long long int carrier_seq;  /* Carrier status changes. */
 178     struct ofport_dpif *peer;   /* Peer if patch port. */
 179
 180     /* Spanning tree. */
 181     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 182     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 183     long long int stp_state_entered;
 184
 185     /* Rapid Spanning Tree. */
 186     struct rstp_port *rstp_port; /* Rapid Spanning Tree Protocol, if any. */
 187     enum rstp_state rstp_state; /* Always RSTP_DISABLED if RSTP not in use. */
 188
 189     /* Queue to DSCP mapping. */
 190     struct ofproto_port_queue *qdscp;
 191     size_t n_qdscp;
 192 };
 193
 194 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 195                                        ofp_port_t);
 196
 197 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 198                                        odp_port_t);
 199
 200 static struct ofport_dpif *
 201 ofport_dpif_cast(const struct ofport *ofport)
 202 {
 203     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 204 }
 205
 206 static void port_run(struct ofport_dpif *);
 207 static int set_bfd(struct ofport *, const struct smap *);
 208 static int set_cfm(struct ofport *, const struct cfm_settings *);
 209 static int set_lldp(struct ofport *ofport_, const struct smap *cfg);
 210 static void ofport_update_peer(struct ofport_dpif *);
 211
 212 /* Reasons that we might need to revalidate every datapath flow, and
 213  * corresponding coverage counters.
 214  *
 215  * A value of 0 means that there is no need to revalidate.
 216  *
 217  * It would be nice to have some cleaner way to integrate with coverage
 218  * counters, but with only a few reasons I guess this is good enough for
 219  * now. */
 220 enum revalidate_reason {
 221     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 222     REV_STP,                   /* Spanning tree protocol port status change. */
 223     REV_RSTP,                  /* RSTP port status change. */
 224     REV_BOND,                  /* Bonding changed. */
 225     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 226     REV_FLOW_TABLE,            /* Flow table changed. */
 227     REV_MAC_LEARNING,          /* Mac learning changed. */
 228     REV_MCAST_SNOOPING,        /* Multicast snooping changed. */
 229 };
 230 COVERAGE_DEFINE(rev_reconfigure);
 231 COVERAGE_DEFINE(rev_stp);
 232 COVERAGE_DEFINE(rev_rstp);
 233 COVERAGE_DEFINE(rev_bond);
 234 COVERAGE_DEFINE(rev_port_toggled);
 235 COVERAGE_DEFINE(rev_flow_table);
 236 COVERAGE_DEFINE(rev_mac_learning);
 237 COVERAGE_DEFINE(rev_mcast_snooping);
 238
 239 /* All datapaths of a given type share a single dpif backer instance. */
 240 struct dpif_backer {
 241     char *type;
 242     int refcount;
 243     struct dpif *dpif;
 244     struct udpif *udpif;
 245
 246     struct ovs_rwlock odp_to_ofport_lock;
 247     struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
 248
 249     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 250
 251     enum revalidate_reason need_revalidate; /* Revalidate all flows. */
 252
 253     bool recv_set_enable; /* Enables or disables receiving packets. */
 254
 255     /* Version string of the datapath stored in OVSDB. */
 256     char *dp_version_string;
 257
 258     /* Datapath feature support. */
 259     struct dpif_backer_support support;
 260     struct atomic_count tnl_count;
 261 };
 262
 263 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 264 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 265
 266 struct ofproto_dpif {
 267     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 268     struct ofproto up;
 269     struct dpif_backer *backer;
 270
 271     /* Unique identifier for this instantiation of this bridge in this running
 272      * process.  */
 273     struct uuid uuid;
 274
 275     ATOMIC(cls_version_t) tables_version;  /* For classifier lookups. */
 276
 277     uint64_t dump_seq; /* Last read of udpif_dump_seq(). */
 278
 279     /* Special OpenFlow rules. */
 280     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 281     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 282     struct rule_dpif *drop_frags_rule; /* Used in OFPUTIL_FRAG_DROP mode. */
 283
 284     /* Bridging. */
 285     struct netflow *netflow;
 286     struct dpif_sflow *sflow;
 287     struct dpif_ipfix *ipfix;
 288     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 289     struct mac_learning *ml;
 290     struct mcast_snooping *ms;
 291     bool has_bonded_bundles;
 292     bool lacp_enabled;
 293     struct mbridge *mbridge;
 294
 295     struct ovs_mutex stats_mutex;
 296     struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
 297                                             * consumed in userspace. */
 298
 299     /* Spanning tree. */
 300     struct stp *stp;
 301     long long int stp_last_tick;
 302
 303     /* Rapid Spanning Tree. */
 304     struct rstp *rstp;
 305     long long int rstp_last_tick;
 306
 307     /* Ports. */
 308     struct sset ports;             /* Set of standard port names. */
 309     struct sset ghost_ports;       /* Ports with no datapath port. */
 310     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 311     int port_poll_errno;           /* Last errno for port_poll() reply. */
 312     uint64_t change_seq;           /* Connectivity status changes. */
 313
 314     /* Work queues. */
 315     struct guarded_list ams;      /* Contains "struct ofproto_async_msgs"s. */
 316     struct seq *ams_seq;          /* For notifying 'ams' reception. */
 317     uint64_t ams_seqno;
 318 };
 319
 320 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 321 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 322
 323 static bool ofproto_use_tnl_push_pop = true;
 324 static void ofproto_unixctl_init(void);
 325
 326 static inline struct ofproto_dpif *
 327 ofproto_dpif_cast(const struct ofproto *ofproto)
 328 {
 329     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 330     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 331 }
 332
 333 bool
 334 ofproto_dpif_get_enable_ufid(const struct dpif_backer *backer)
 335 {
 336     return backer->support.ufid;
 337 }
 338
 339 struct dpif_backer_support *
 340 ofproto_dpif_get_support(const struct ofproto_dpif *ofproto)
 341 {
 342     return &ofproto->backer->support;
 343 }
 344
 345 static void ofproto_trace(struct ofproto_dpif *, struct flow *,
 346                           const struct dp_packet *packet,
 347                           const struct ofpact[], size_t ofpacts_len,
 348                           struct ds *);
 349
 350 /* Global variables. */
 351 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 352
 353 /* Initial mappings of port to bridge mappings. */
 354 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 355
 356 /* Executes 'fm'.  The caller retains ownership of 'fm' and everything in
 357  * it. */
 358 void
 359 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 360                       const struct ofputil_flow_mod *fm)
 361 {
 362     struct ofproto_flow_mod ofm;
 363
 364     /* Multiple threads may do this for the same 'fm' at the same time.
 365      * Allocate ofproto_flow_mod with execution context from stack.
 366      *
 367      * Note: This copy could be avoided by making ofproto_flow_mod more
 368      * complex, but that may not be desireable, and a learn action is not that
 369      * fast to begin with. */
 370     ofm.fm = *fm;
 371     ofproto_flow_mod(&ofproto->up, &ofm);
 372 }
 373
 374 /* Appends 'am' to the queue of asynchronous messages to be sent to the
 375  * controller.  Takes ownership of 'am' and any data it points to. */
 376 void
 377 ofproto_dpif_send_async_msg(struct ofproto_dpif *ofproto,
 378                             struct ofproto_async_msg *am)
 379 {
 380     if (!guarded_list_push_back(&ofproto->ams, &am->list_node, 1024)) {
 381         COVERAGE_INC(packet_in_overflow);
 382         ofproto_async_msg_free(am);
 383     }
 384
 385     /* Wakes up main thread for packet-in I/O. */
 386     seq_change(ofproto->ams_seq);
 387 }
 388
 389 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
 390  * packet rather than to send the packet to the controller.
 391  *
 392  * This function returns false to indicate that a packet_in message
 393  * for a "table-miss" should be sent to at least one controller.
 394  * False otherwise. */
 395 bool
 396 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif *ofproto)
 397 {
 398     return connmgr_wants_packet_in_on_miss(ofproto->up.connmgr);
 399 }
 400 \f
 401 /* Factory functions. */
 402
 403 static void
 404 init(const struct shash *iface_hints)
 405 {
 406     struct shash_node *node;
 407
 408     /* Make a local copy, since we don't own 'iface_hints' elements. */
 409     SHASH_FOR_EACH(node, iface_hints) {
 410         const struct iface_hint *orig_hint = node->data;
 411         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 412
 413         new_hint->br_name = xstrdup(orig_hint->br_name);
 414         new_hint->br_type = xstrdup(orig_hint->br_type);
 415         new_hint->ofp_port = orig_hint->ofp_port;
 416
 417         shash_add(&init_ofp_ports, node->name, new_hint);
 418     }
 419
 420     ofproto_unixctl_init();
 421     udpif_init();
 422 }
 423
 424 static void
 425 enumerate_types(struct sset *types)
 426 {
 427     dp_enumerate_types(types);
 428 }
 429
 430 static int
 431 enumerate_names(const char *type, struct sset *names)
 432 {
 433     struct ofproto_dpif *ofproto;
 434
 435     sset_clear(names);
 436     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 437         if (strcmp(type, ofproto->up.type)) {
 438             continue;
 439         }
 440         sset_add(names, ofproto->up.name);
 441     }
 442
 443     return 0;
 444 }
 445
 446 static int
 447 del(const char *type, const char *name)
 448 {
 449     struct dpif *dpif;
 450     int error;
 451
 452     error = dpif_open(name, type, &dpif);
 453     if (!error) {
 454         error = dpif_delete(dpif);
 455         dpif_close(dpif);
 456     }
 457     return error;
 458 }
 459 \f
 460 static const char *
 461 port_open_type(const char *datapath_type, const char *port_type)
 462 {
 463     return dpif_port_open_type(datapath_type, port_type);
 464 }
 465
 466 /* Type functions. */
 467
 468 static void process_dpif_port_changes(struct dpif_backer *);
 469 static void process_dpif_all_ports_changed(struct dpif_backer *);
 470 static void process_dpif_port_change(struct dpif_backer *,
 471                                      const char *devname);
 472 static void process_dpif_port_error(struct dpif_backer *, int error);
 473
 474 static struct ofproto_dpif *
 475 lookup_ofproto_dpif_by_port_name(const char *name)
 476 {
 477     struct ofproto_dpif *ofproto;
 478
 479     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 480         if (sset_contains(&ofproto->ports, name)) {
 481             return ofproto;
 482         }
 483     }
 484
 485     return NULL;
 486 }
 487
 488 bool
 489 ofproto_dpif_backer_enabled(struct dpif_backer* backer)
 490 {
 491     return backer->recv_set_enable;
 492 }
 493
 494 static int
 495 type_run(const char *type)
 496 {
 497     struct dpif_backer *backer;
 498
 499     backer = shash_find_data(&all_dpif_backers, type);
 500     if (!backer) {
 501         /* This is not necessarily a problem, since backers are only
 502          * created on demand. */
 503         return 0;
 504     }
 505
 506     /* This must be called before dpif_run() */
 507     dpif_poll_threads_set(backer->dpif, pmd_cpu_mask);
 508
 509     if (dpif_run(backer->dpif)) {
 510         backer->need_revalidate = REV_RECONFIGURE;
 511     }
 512
 513     udpif_run(backer->udpif);
 514
 515     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 516      * and the configuration has now changed to "false", enable receiving
 517      * packets from the datapath. */
 518     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 519         int error;
 520
 521         backer->recv_set_enable = true;
 522
 523         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 524         if (error) {
 525             VLOG_ERR("Failed to enable receiving packets in dpif.");
 526             return error;
 527         }
 528         dpif_flow_flush(backer->dpif);
 529         backer->need_revalidate = REV_RECONFIGURE;
 530     }
 531
 532     if (backer->recv_set_enable) {
 533         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 534     }
 535
 536     if (backer->need_revalidate) {
 537         struct ofproto_dpif *ofproto;
 538         struct simap_node *node;
 539         struct simap tmp_backers;
 540
 541         /* Handle tunnel garbage collection. */
 542         simap_init(&tmp_backers);
 543         simap_swap(&backer->tnl_backers, &tmp_backers);
 544
 545         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 546             struct ofport_dpif *iter;
 547
 548             if (backer != ofproto->backer) {
 549                 continue;
 550             }
 551
 552             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 553                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 554                 const char *dp_port;
 555
 556                 if (!iter->is_tunnel) {
 557                     continue;
 558                 }
 559
 560                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 561                                                      namebuf, sizeof namebuf);
 562                 node = simap_find(&tmp_backers, dp_port);
 563                 if (node) {
 564                     simap_put(&backer->tnl_backers, dp_port, node->data);
 565                     simap_delete(&tmp_backers, node);
 566                     node = simap_find(&backer->tnl_backers, dp_port);
 567                 } else {
 568                     node = simap_find(&backer->tnl_backers, dp_port);
 569                     if (!node) {
 570                         odp_port_t odp_port = ODPP_NONE;
 571
 572                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 573                                            &odp_port)) {
 574                             simap_put(&backer->tnl_backers, dp_port,
 575                                       odp_to_u32(odp_port));
 576                             node = simap_find(&backer->tnl_backers, dp_port);
 577                         }
 578                     }
 579                 }
 580
 581                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 582                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 583                                          iter->odp_port,
 584                                          ovs_native_tunneling_is_on(ofproto), dp_port)) {
 585                     backer->need_revalidate = REV_RECONFIGURE;
 586                 }
 587             }
 588         }
 589
 590         SIMAP_FOR_EACH (node, &tmp_backers) {
 591             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 592         }
 593         simap_destroy(&tmp_backers);
 594
 595         switch (backer->need_revalidate) {
 596         case REV_RECONFIGURE:    COVERAGE_INC(rev_reconfigure);    break;
 597         case REV_STP:            COVERAGE_INC(rev_stp);            break;
 598         case REV_RSTP:           COVERAGE_INC(rev_rstp);           break;
 599         case REV_BOND:           COVERAGE_INC(rev_bond);           break;
 600         case REV_PORT_TOGGLED:   COVERAGE_INC(rev_port_toggled);   break;
 601         case REV_FLOW_TABLE:     COVERAGE_INC(rev_flow_table);     break;
 602         case REV_MAC_LEARNING:   COVERAGE_INC(rev_mac_learning);   break;
 603         case REV_MCAST_SNOOPING: COVERAGE_INC(rev_mcast_snooping); break;
 604         }
 605         backer->need_revalidate = 0;
 606
 607         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 608             struct ofport_dpif *ofport;
 609             struct ofbundle *bundle;
 610
 611             if (ofproto->backer != backer) {
 612                 continue;
 613             }
 614
 615             xlate_txn_start();
 616             xlate_ofproto_set(ofproto, ofproto->up.name,
 617                               ofproto->backer->dpif, ofproto->ml,
 618                               ofproto->stp, ofproto->rstp, ofproto->ms,
 619                               ofproto->mbridge, ofproto->sflow, ofproto->ipfix,
 620                               ofproto->netflow,
 621                               ofproto->up.forward_bpdu,
 622                               connmgr_has_in_band(ofproto->up.connmgr),
 623                               &ofproto->backer->support);
 624
 625             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 626                 xlate_bundle_set(ofproto, bundle, bundle->name,
 627                                  bundle->vlan_mode, bundle->vlan,
 628                                  bundle->trunks, bundle->use_priority_tags,
 629                                  bundle->bond, bundle->lacp,
 630                                  bundle->floodable);
 631             }
 632
 633             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 634                 int stp_port = ofport->stp_port
 635                     ? stp_port_no(ofport->stp_port)
 636                     : -1;
 637                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 638                                  ofport->up.ofp_port, ofport->odp_port,
 639                                  ofport->up.netdev, ofport->cfm, ofport->bfd,
 640                                  ofport->lldp, ofport->peer, stp_port,
 641                                  ofport->rstp_port, ofport->qdscp,
 642                                  ofport->n_qdscp, ofport->up.pp.config,
 643                                  ofport->up.pp.state, ofport->is_tunnel,
 644                                  ofport->may_enable);
 645             }
 646             xlate_txn_commit();
 647         }
 648
 649         udpif_revalidate(backer->udpif);
 650     }
 651
 652     process_dpif_port_changes(backer);
 653
 654     return 0;
 655 }
 656
 657 /* Check for and handle port changes in 'backer''s dpif. */
 658 static void
 659 process_dpif_port_changes(struct dpif_backer *backer)
 660 {
 661     for (;;) {
 662         char *devname;
 663         int error;
 664
 665         error = dpif_port_poll(backer->dpif, &devname);
 666         switch (error) {
 667         case EAGAIN:
 668             return;
 669
 670         case ENOBUFS:
 671             process_dpif_all_ports_changed(backer);
 672             break;
 673
 674         case 0:
 675             process_dpif_port_change(backer, devname);
 676             free(devname);
 677             break;
 678
 679         default:
 680             process_dpif_port_error(backer, error);
 681             break;
 682         }
 683     }
 684 }
 685
 686 static void
 687 process_dpif_all_ports_changed(struct dpif_backer *backer)
 688 {
 689     struct ofproto_dpif *ofproto;
 690     struct dpif_port dpif_port;
 691     struct dpif_port_dump dump;
 692     struct sset devnames;
 693     const char *devname;
 694
 695     sset_init(&devnames);
 696     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 697         if (ofproto->backer == backer) {
 698             struct ofport *ofport;
 699
 700             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 701                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 702             }
 703         }
 704     }
 705     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 706         sset_add(&devnames, dpif_port.name);
 707     }
 708
 709     SSET_FOR_EACH (devname, &devnames) {
 710         process_dpif_port_change(backer, devname);
 711     }
 712     sset_destroy(&devnames);
 713 }
 714
 715 static void
 716 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 717 {
 718     struct ofproto_dpif *ofproto;
 719     struct dpif_port port;
 720
 721     /* Don't report on the datapath's device. */
 722     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 723         return;
 724     }
 725
 726     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 727                    &all_ofproto_dpifs) {
 728         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 729             return;
 730         }
 731     }
 732
 733     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 734     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 735         /* The port was removed.  If we know the datapath,
 736          * report it through poll_set().  If we don't, it may be
 737          * notifying us of a removal we initiated, so ignore it.
 738          * If there's a pending ENOBUFS, let it stand, since
 739          * everything will be reevaluated. */
 740         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 741             sset_add(&ofproto->port_poll_set, devname);
 742             ofproto->port_poll_errno = 0;
 743         }
 744     } else if (!ofproto) {
 745         /* The port was added, but we don't know with which
 746          * ofproto we should associate it.  Delete it. */
 747         dpif_port_del(backer->dpif, port.port_no);
 748     } else {
 749         struct ofport_dpif *ofport;
 750
 751         ofport = ofport_dpif_cast(shash_find_data(
 752                                       &ofproto->up.port_by_name, devname));
 753         if (ofport
 754             && ofport->odp_port != port.port_no
 755             && !odp_port_to_ofport(backer, port.port_no))
 756         {
 757             /* 'ofport''s datapath port number has changed from
 758              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 759              * structures to match. */
 760             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 761             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 762             ofport->odp_port = port.port_no;
 763             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 764                         hash_odp_port(port.port_no));
 765             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 766             backer->need_revalidate = REV_RECONFIGURE;
 767         }
 768     }
 769     dpif_port_destroy(&port);
 770 }
 771
 772 /* Propagate 'error' to all ofprotos based on 'backer'. */
 773 static void
 774 process_dpif_port_error(struct dpif_backer *backer, int error)
 775 {
 776     struct ofproto_dpif *ofproto;
 777
 778     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 779         if (ofproto->backer == backer) {
 780             sset_clear(&ofproto->port_poll_set);
 781             ofproto->port_poll_errno = error;
 782         }
 783     }
 784 }
 785
 786 static void
 787 type_wait(const char *type)
 788 {
 789     struct dpif_backer *backer;
 790
 791     backer = shash_find_data(&all_dpif_backers, type);
 792     if (!backer) {
 793         /* This is not necessarily a problem, since backers are only
 794          * created on demand. */
 795         return;
 796     }
 797
 798     dpif_wait(backer->dpif);
 799 }
 800 \f
 801 /* Basic life-cycle. */
 802
 803 static int add_internal_flows(struct ofproto_dpif *);
 804
 805 static struct ofproto *
 806 alloc(void)
 807 {
 808     struct ofproto_dpif *ofproto = xzalloc(sizeof *ofproto);
 809     return &ofproto->up;
 810 }
 811
 812 static void
 813 dealloc(struct ofproto *ofproto_)
 814 {
 815     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
 816     free(ofproto);
 817 }
 818
 819 static void
 820 close_dpif_backer(struct dpif_backer *backer)
 821 {
 822     ovs_assert(backer->refcount > 0);
 823
 824     if (--backer->refcount) {
 825         return;
 826     }
 827
 828     udpif_destroy(backer->udpif);
 829
 830     simap_destroy(&backer->tnl_backers);
 831     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
 832     hmap_destroy(&backer->odp_to_ofport_map);
 833     shash_find_and_delete(&all_dpif_backers, backer->type);
 834     free(backer->type);
 835     free(backer->dp_version_string);
 836     dpif_close(backer->dpif);
 837     free(backer);
 838 }
 839
 840 /* Datapath port slated for removal from datapath. */
 841 struct odp_garbage {
 842     struct ovs_list list_node;
 843     odp_port_t odp_port;
 844 };
 845
 846 static bool check_variable_length_userdata(struct dpif_backer *backer);
 847 static void check_support(struct dpif_backer *backer);
 848
 849 static int
 850 open_dpif_backer(const char *type, struct dpif_backer **backerp)
 851 {
 852     struct dpif_backer *backer;
 853     struct dpif_port_dump port_dump;
 854     struct dpif_port port;
 855     struct shash_node *node;
 856     struct ovs_list garbage_list;
 857     struct odp_garbage *garbage;
 858
 859     struct sset names;
 860     char *backer_name;
 861     const char *name;
 862     int error;
 863
 864     backer = shash_find_data(&all_dpif_backers, type);
 865     if (backer) {
 866         backer->refcount++;
 867         *backerp = backer;
 868         return 0;
 869     }
 870
 871     backer_name = xasprintf("ovs-%s", type);
 872
 873     /* Remove any existing datapaths, since we assume we're the only
 874      * userspace controlling the datapath. */
 875     sset_init(&names);
 876     dp_enumerate_names(type, &names);
 877     SSET_FOR_EACH(name, &names) {
 878         struct dpif *old_dpif;
 879
 880         /* Don't remove our backer if it exists. */
 881         if (!strcmp(name, backer_name)) {
 882             continue;
 883         }
 884
 885         if (dpif_open(name, type, &old_dpif)) {
 886             VLOG_WARN("couldn't open old datapath %s to remove it", name);
 887         } else {
 888             dpif_delete(old_dpif);
 889             dpif_close(old_dpif);
 890         }
 891     }
 892     sset_destroy(&names);
 893
 894     backer = xmalloc(sizeof *backer);
 895
 896     error = dpif_create_and_open(backer_name, type, &backer->dpif);
 897     free(backer_name);
 898     if (error) {
 899         VLOG_ERR("failed to open datapath of type %s: %s", type,
 900                  ovs_strerror(error));
 901         free(backer);
 902         return error;
 903     }
 904     backer->udpif = udpif_create(backer, backer->dpif);
 905
 906     backer->type = xstrdup(type);
 907     backer->refcount = 1;
 908     hmap_init(&backer->odp_to_ofport_map);
 909     ovs_rwlock_init(&backer->odp_to_ofport_lock);
 910     backer->need_revalidate = 0;
 911     simap_init(&backer->tnl_backers);
 912     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
 913     *backerp = backer;
 914
 915     if (backer->recv_set_enable) {
 916         dpif_flow_flush(backer->dpif);
 917     }
 918
 919     /* Loop through the ports already on the datapath and remove any
 920      * that we don't need anymore. */
 921     ovs_list_init(&garbage_list);
 922     dpif_port_dump_start(&port_dump, backer->dpif);
 923     while (dpif_port_dump_next(&port_dump, &port)) {
 924         node = shash_find(&init_ofp_ports, port.name);
 925         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
 926             garbage = xmalloc(sizeof *garbage);
 927             garbage->odp_port = port.port_no;
 928             ovs_list_push_front(&garbage_list, &garbage->list_node);
 929         }
 930     }
 931     dpif_port_dump_done(&port_dump);
 932
 933     LIST_FOR_EACH_POP (garbage, list_node, &garbage_list) {
 934         dpif_port_del(backer->dpif, garbage->odp_port);
 935         free(garbage);
 936     }
 937
 938     shash_add(&all_dpif_backers, type, backer);
 939
 940     check_support(backer);
 941     atomic_count_init(&backer->tnl_count, 0);
 942
 943     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 944     if (error) {
 945         VLOG_ERR("failed to listen on datapath of type %s: %s",
 946                  type, ovs_strerror(error));
 947         close_dpif_backer(backer);
 948         return error;
 949     }
 950
 951     if (backer->recv_set_enable) {
 952         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 953     }
 954
 955     /* This check fails if performed before udpif threads have been set,
 956      * as the kernel module checks that the 'pid' in userspace action
 957      * is non-zero. */
 958     backer->support.variable_length_userdata
 959         = check_variable_length_userdata(backer);
 960     backer->dp_version_string = dpif_get_dp_version(backer->dpif);
 961
 962     return error;
 963 }
 964
 965 bool
 966 ovs_native_tunneling_is_on(struct ofproto_dpif *ofproto)
 967 {
 968     return ofproto_use_tnl_push_pop && ofproto->backer->support.tnl_push_pop &&
 969            atomic_count_get(&ofproto->backer->tnl_count);
 970 }
 971
 972 /* Tests whether 'backer''s datapath supports recirculation.  Only newer
 973  * datapaths support OVS_KEY_ATTR_RECIRC_ID in keys.  We need to disable some
 974  * features on older datapaths that don't support this feature.
 975  *
 976  * Returns false if 'backer' definitely does not support recirculation, true if
 977  * it seems to support recirculation or if at least the error we get is
 978  * ambiguous. */
 979 static bool
 980 check_recirc(struct dpif_backer *backer)
 981 {
 982     struct flow flow;
 983     struct odputil_keybuf keybuf;
 984     struct ofpbuf key;
 985     bool enable_recirc;
 986     struct odp_flow_key_parms odp_parms = {
 987         .flow = &flow,
 988         .support = {
 989             .recirc = true,
 990         },
 991     };
 992
 993     memset(&flow, 0, sizeof flow);
 994     flow.recirc_id = 1;
 995     flow.dp_hash = 1;
 996
 997     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
 998     odp_flow_key_from_flow(&odp_parms, &key);
 999     enable_recirc = dpif_probe_feature(backer->dpif, "recirculation", &key,
1000                                        NULL);
1001
1002     if (enable_recirc) {
1003         VLOG_INFO("%s: Datapath supports recirculation",
1004                   dpif_name(backer->dpif));
1005     } else {
1006         VLOG_INFO("%s: Datapath does not support recirculation",
1007                   dpif_name(backer->dpif));
1008     }
1009
1010     return enable_recirc;
1011 }
1012
1013 /* Tests whether 'dpif' supports unique flow ids. We can skip serializing
1014  * some flow attributes for datapaths that support this feature.
1015  *
1016  * Returns true if 'dpif' supports UFID for flow operations.
1017  * Returns false if  'dpif' does not support UFID. */
1018 static bool
1019 check_ufid(struct dpif_backer *backer)
1020 {
1021     struct flow flow;
1022     struct odputil_keybuf keybuf;
1023     struct ofpbuf key;
1024     ovs_u128 ufid;
1025     bool enable_ufid;
1026     struct odp_flow_key_parms odp_parms = {
1027         .flow = &flow,
1028     };
1029
1030     memset(&flow, 0, sizeof flow);
1031     flow.dl_type = htons(0x1234);
1032
1033     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1034     odp_flow_key_from_flow(&odp_parms, &key);
1035     dpif_flow_hash(backer->dpif, key.data, key.size, &ufid);
1036
1037     enable_ufid = dpif_probe_feature(backer->dpif, "UFID", &key, &ufid);
1038
1039     if (enable_ufid) {
1040         VLOG_INFO("%s: Datapath supports unique flow ids",
1041                   dpif_name(backer->dpif));
1042     } else {
1043         VLOG_INFO("%s: Datapath does not support unique flow ids",
1044                   dpif_name(backer->dpif));
1045     }
1046     return enable_ufid;
1047 }
1048
1049 /* Tests whether 'backer''s datapath supports variable-length
1050  * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.  We need
1051  * to disable some features on older datapaths that don't support this
1052  * feature.
1053  *
1054  * Returns false if 'backer' definitely does not support variable-length
1055  * userdata, true if it seems to support them or if at least the error we get
1056  * is ambiguous. */
1057 static bool
1058 check_variable_length_userdata(struct dpif_backer *backer)
1059 {
1060     struct eth_header *eth;
1061     struct ofpbuf actions;
1062     struct dpif_execute execute;
1063     struct dp_packet packet;
1064     struct flow flow;
1065     size_t start;
1066     int error;
1067
1068     /* Compose a userspace action that will cause an ERANGE error on older
1069      * datapaths that don't support variable-length userdata.
1070      *
1071      * We really test for using userdata longer than 8 bytes, but older
1072      * datapaths accepted these, silently truncating the userdata to 8 bytes.
1073      * The same older datapaths rejected userdata shorter than 8 bytes, so we
1074      * test for that instead as a proxy for longer userdata support. */
1075     ofpbuf_init(&actions, 64);
1076     start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
1077     nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
1078                    dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
1079     nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
1080     nl_msg_end_nested(&actions, start);
1081
1082     /* Compose a dummy ethernet packet. */
1083     dp_packet_init(&packet, ETH_HEADER_LEN);
1084     eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1085     eth->eth_type = htons(0x1234);
1086
1087     flow_extract(&packet, &flow);
1088
1089     /* Execute the actions.  On older datapaths this fails with ERANGE, on
1090      * newer datapaths it succeeds. */
1091     execute.actions = actions.data;
1092     execute.actions_len = actions.size;
1093     execute.packet = &packet;
1094     execute.flow = &flow;
1095     execute.needs_help = false;
1096     execute.probe = true;
1097     execute.mtu = 0;
1098
1099     error = dpif_execute(backer->dpif, &execute);
1100
1101     dp_packet_uninit(&packet);
1102     ofpbuf_uninit(&actions);
1103
1104     switch (error) {
1105     case 0:
1106         return true;
1107
1108     case ERANGE:
1109         /* Variable-length userdata is not supported. */
1110         VLOG_WARN("%s: datapath does not support variable-length userdata "
1111                   "feature (needs Linux 3.10+ or kernel module from OVS "
1112                   "1..11+).  The NXAST_SAMPLE action will be ignored.",
1113                   dpif_name(backer->dpif));
1114         return false;
1115
1116     default:
1117         /* Something odd happened.  We're not sure whether variable-length
1118          * userdata is supported.  Default to "yes". */
1119         VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1120                   dpif_name(backer->dpif), ovs_strerror(error));
1121         return true;
1122     }
1123 }
1124
1125 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1126  *
1127  * Returns the number of elements in a struct flow's mpls_lse field
1128  * if the datapath supports at least that many entries in an
1129  * MPLS label stack.
1130  * Otherwise returns the number of MPLS push actions supported by
1131  * the datapath. */
1132 static size_t
1133 check_max_mpls_depth(struct dpif_backer *backer)
1134 {
1135     struct flow flow;
1136     int n;
1137
1138     for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
1139         struct odputil_keybuf keybuf;
1140         struct ofpbuf key;
1141         struct odp_flow_key_parms odp_parms = {
1142             .flow = &flow,
1143         };
1144
1145         memset(&flow, 0, sizeof flow);
1146         flow.dl_type = htons(ETH_TYPE_MPLS);
1147         flow_set_mpls_bos(&flow, n, 1);
1148
1149         ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1150         odp_flow_key_from_flow(&odp_parms, &key);
1151         if (!dpif_probe_feature(backer->dpif, "MPLS", &key, NULL)) {
1152             break;
1153         }
1154     }
1155
1156     VLOG_INFO("%s: MPLS label stack length probed as %d",
1157               dpif_name(backer->dpif), n);
1158     return n;
1159 }
1160
1161 /* Tests whether 'backer''s datapath supports masked data in
1162  * OVS_ACTION_ATTR_SET actions.  We need to disable some features on older
1163  * datapaths that don't support this feature. */
1164 static bool
1165 check_masked_set_action(struct dpif_backer *backer)
1166 {
1167     struct eth_header *eth;
1168     struct ofpbuf actions;
1169     struct dpif_execute execute;
1170     struct dp_packet packet;
1171     struct flow flow;
1172     int error;
1173     struct ovs_key_ethernet key, mask;
1174
1175     /* Compose a set action that will cause an EINVAL error on older
1176      * datapaths that don't support masked set actions.
1177      * Avoid using a full mask, as it could be translated to a non-masked
1178      * set action instead. */
1179     ofpbuf_init(&actions, 64);
1180     memset(&key, 0x53, sizeof key);
1181     memset(&mask, 0x7f, sizeof mask);
1182     commit_masked_set_action(&actions, OVS_KEY_ATTR_ETHERNET, &key, &mask,
1183                              sizeof key);
1184
1185     /* Compose a dummy ethernet packet. */
1186     dp_packet_init(&packet, ETH_HEADER_LEN);
1187     eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1188     eth->eth_type = htons(0x1234);
1189
1190     flow_extract(&packet, &flow);
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.flow = &flow;
1198     execute.needs_help = false;
1199     execute.probe = true;
1200     execute.mtu = 0;
1201
1202     error = dpif_execute(backer->dpif, &execute);
1203
1204     dp_packet_uninit(&packet);
1205     ofpbuf_uninit(&actions);
1206
1207     if (error) {
1208         /* Masked set action is not supported. */
1209         VLOG_INFO("%s: datapath does not support masked set action feature.",
1210                   dpif_name(backer->dpif));
1211     }
1212     return !error;
1213 }
1214
1215 /* Tests whether 'backer''s datapath supports truncation of a packet in
1216  * OVS_ACTION_ATTR_TRUNC.  We need to disable some features on older
1217  * datapaths that don't support this feature. */
1218 static bool
1219 check_trunc_action(struct dpif_backer *backer)
1220 {
1221     struct eth_header *eth;
1222     struct ofpbuf actions;
1223     struct dpif_execute execute;
1224     struct dp_packet packet;
1225     struct ovs_action_trunc *trunc;
1226     struct flow flow;
1227     int error;
1228
1229     /* Compose an action with output(port:1,
1230      *              max_len:OVS_ACTION_OUTPUT_MIN + 1).
1231      * This translates to one truncate action and one output action. */
1232     ofpbuf_init(&actions, 64);
1233     trunc = nl_msg_put_unspec_uninit(&actions,
1234                             OVS_ACTION_ATTR_TRUNC, sizeof *trunc);
1235
1236     trunc->max_len = ETH_HEADER_LEN + 1;
1237     nl_msg_put_odp_port(&actions, OVS_ACTION_ATTR_OUTPUT, u32_to_odp(1));
1238
1239     /* Compose a dummy Ethernet packet. */
1240     dp_packet_init(&packet, ETH_HEADER_LEN);
1241     eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1242     eth->eth_type = htons(0x1234);
1243
1244     flow_extract(&packet, &flow);
1245
1246     /* Execute the actions.  On older datapaths this fails with EINVAL, on
1247      * newer datapaths it succeeds. */
1248     execute.actions = actions.data;
1249     execute.actions_len = actions.size;
1250     execute.packet = &packet;
1251     execute.flow = &flow;
1252     execute.needs_help = false;
1253     execute.probe = true;
1254     execute.mtu = 0;
1255
1256     error = dpif_execute(backer->dpif, &execute);
1257
1258     dp_packet_uninit(&packet);
1259     ofpbuf_uninit(&actions);
1260
1261     if (error) {
1262         VLOG_INFO("%s: Datapath does not support truncate action",
1263                   dpif_name(backer->dpif));
1264     } else {
1265         VLOG_INFO("%s: Datapath supports truncate action",
1266                   dpif_name(backer->dpif));
1267     }
1268
1269     return !error;
1270 }
1271
1272 #define CHECK_FEATURE__(NAME, SUPPORT, FIELD, VALUE)                        \
1273 static bool                                                                 \
1274 check_##NAME(struct dpif_backer *backer)                                    \
1275 {                                                                           \
1276     struct flow flow;                                                       \
1277     struct odputil_keybuf keybuf;                                           \
1278     struct ofpbuf key;                                                      \
1279     bool enable;                                                            \
1280     struct odp_flow_key_parms odp_parms = {                                 \
1281         .flow = &flow,                                                      \
1282         .support = {                                                        \
1283             .SUPPORT = true,                                                \
1284         },                                                                  \
1285     };                                                                      \
1286                                                                             \
1287     memset(&flow, 0, sizeof flow);                                          \
1288     flow.FIELD = VALUE;                                                     \
1289                                                                             \
1290     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);                         \
1291     odp_flow_key_from_flow(&odp_parms, &key);                               \
1292     enable = dpif_probe_feature(backer->dpif, #NAME, &key, NULL);           \
1293                                                                             \
1294     if (enable) {                                                           \
1295         VLOG_INFO("%s: Datapath supports "#NAME, dpif_name(backer->dpif));  \
1296     } else {                                                                \
1297         VLOG_INFO("%s: Datapath does not support "#NAME,                    \
1298                   dpif_name(backer->dpif));                                 \
1299     }                                                                       \
1300                                                                             \
1301     return enable;                                                          \
1302 }
1303 #define CHECK_FEATURE(FIELD) CHECK_FEATURE__(FIELD, FIELD, FIELD, 1)
1304
1305 CHECK_FEATURE(ct_state)
1306 CHECK_FEATURE(ct_zone)
1307 CHECK_FEATURE(ct_mark)
1308 CHECK_FEATURE__(ct_label, ct_label, ct_label.u64.lo, 1)
1309 CHECK_FEATURE__(ct_state_nat, ct_state, ct_state, CS_TRACKED|CS_SRC_NAT)
1310
1311 #undef CHECK_FEATURE
1312 #undef CHECK_FEATURE__
1313
1314 static void
1315 check_support(struct dpif_backer *backer)
1316 {
1317     /* This feature needs to be tested after udpif threads are set. */
1318     backer->support.variable_length_userdata = false;
1319
1320     backer->support.odp.recirc = check_recirc(backer);
1321     backer->support.odp.max_mpls_depth = check_max_mpls_depth(backer);
1322     backer->support.masked_set_action = check_masked_set_action(backer);
1323     backer->support.trunc = check_trunc_action(backer);
1324     backer->support.ufid = check_ufid(backer);
1325     backer->support.tnl_push_pop = dpif_supports_tnl_push_pop(backer->dpif);
1326
1327     backer->support.odp.ct_state = check_ct_state(backer);
1328     backer->support.odp.ct_zone = check_ct_zone(backer);
1329     backer->support.odp.ct_mark = check_ct_mark(backer);
1330     backer->support.odp.ct_label = check_ct_label(backer);
1331
1332     backer->support.odp.ct_state_nat = check_ct_state_nat(backer);
1333 }
1334
1335 static int
1336 construct(struct ofproto *ofproto_)
1337 {
1338     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1339     struct shash_node *node, *next;
1340     int error;
1341
1342     /* Tunnel module can get used right after the udpif threads are running. */
1343     ofproto_tunnel_init();
1344
1345     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1346     if (error) {
1347         return error;
1348     }
1349
1350     uuid_generate(&ofproto->uuid);
1351     atomic_init(&ofproto->tables_version, CLS_MIN_VERSION);
1352     ofproto->netflow = NULL;
1353     ofproto->sflow = NULL;
1354     ofproto->ipfix = NULL;
1355     ofproto->stp = NULL;
1356     ofproto->rstp = NULL;
1357     ofproto->dump_seq = 0;
1358     hmap_init(&ofproto->bundles);
1359     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1360     ofproto->ms = NULL;
1361     ofproto->mbridge = mbridge_create();
1362     ofproto->has_bonded_bundles = false;
1363     ofproto->lacp_enabled = false;
1364     ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1365
1366     guarded_list_init(&ofproto->ams);
1367
1368     sset_init(&ofproto->ports);
1369     sset_init(&ofproto->ghost_ports);
1370     sset_init(&ofproto->port_poll_set);
1371     ofproto->port_poll_errno = 0;
1372     ofproto->change_seq = 0;
1373     ofproto->ams_seq = seq_create();
1374     ofproto->ams_seqno = seq_read(ofproto->ams_seq);
1375
1376
1377     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1378         struct iface_hint *iface_hint = node->data;
1379
1380         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1381             /* Check if the datapath already has this port. */
1382             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1383                 sset_add(&ofproto->ports, node->name);
1384             }
1385
1386             free(iface_hint->br_name);
1387             free(iface_hint->br_type);
1388             free(iface_hint);
1389             shash_delete(&init_ofp_ports, node);
1390         }
1391     }
1392
1393     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1394                 hash_string(ofproto->up.name, 0));
1395     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1396
1397     ofproto_init_tables(ofproto_, N_TABLES);
1398     error = add_internal_flows(ofproto);
1399
1400     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1401
1402     return error;
1403 }
1404
1405 static int
1406 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1407                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1408 {
1409     struct match match;
1410     int error;
1411     struct rule *rule;
1412
1413     match_init_catchall(&match);
1414     match_set_reg(&match, 0, id);
1415
1416     error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, 0, ofpacts,
1417                                            &rule);
1418     *rulep = error ? NULL : rule_dpif_cast(rule);
1419
1420     return error;
1421 }
1422
1423 static int
1424 add_internal_flows(struct ofproto_dpif *ofproto)
1425 {
1426     struct ofpact_controller *controller;
1427     uint64_t ofpacts_stub[128 / 8];
1428     struct ofpbuf ofpacts;
1429     struct rule *unused_rulep OVS_UNUSED;
1430     struct match match;
1431     int error;
1432     int id;
1433
1434     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1435     id = 1;
1436
1437     controller = ofpact_put_CONTROLLER(&ofpacts);
1438     controller->max_len = UINT16_MAX;
1439     controller->controller_id = 0;
1440     controller->reason = OFPR_IMPLICIT_MISS;
1441     ofpact_finish_CONTROLLER(&ofpacts, &controller);
1442
1443     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1444                                    &ofproto->miss_rule);
1445     if (error) {
1446         return error;
1447     }
1448
1449     ofpbuf_clear(&ofpacts);
1450     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1451                                    &ofproto->no_packet_in_rule);
1452     if (error) {
1453         return error;
1454     }
1455
1456     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1457                                    &ofproto->drop_frags_rule);
1458     if (error) {
1459         return error;
1460     }
1461
1462     /* Drop any run away non-recirc rule lookups. Recirc_id has to be
1463      * zero when reaching this rule.
1464      *
1465      * (priority=2), recirc_id=0, actions=drop
1466      */
1467     ofpbuf_clear(&ofpacts);
1468     match_init_catchall(&match);
1469     match_set_recirc_id(&match, 0);
1470     error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, 0, &ofpacts,
1471                                            &unused_rulep);
1472     return error;
1473 }
1474
1475 static void
1476 destruct(struct ofproto *ofproto_)
1477 {
1478     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1479     struct ofproto_async_msg *am;
1480     struct rule_dpif *rule;
1481     struct oftable *table;
1482     struct ovs_list ams;
1483
1484     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1485     xlate_txn_start();
1486     xlate_remove_ofproto(ofproto);
1487     xlate_txn_commit();
1488
1489     /* Ensure that the upcall processing threads have no remaining references
1490      * to the ofproto or anything in it. */
1491     udpif_synchronize(ofproto->backer->udpif);
1492
1493     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1494
1495     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1496         CLS_FOR_EACH (rule, up.cr, &table->cls) {
1497             ofproto_rule_delete(&ofproto->up, &rule->up);
1498         }
1499     }
1500     ofproto_group_delete_all(&ofproto->up);
1501
1502     guarded_list_pop_all(&ofproto->ams, &ams);
1503     LIST_FOR_EACH_POP (am, list_node, &ams) {
1504         ofproto_async_msg_free(am);
1505     }
1506     guarded_list_destroy(&ofproto->ams);
1507
1508     recirc_free_ofproto(ofproto, ofproto->up.name);
1509
1510     mbridge_unref(ofproto->mbridge);
1511
1512     netflow_unref(ofproto->netflow);
1513     dpif_sflow_unref(ofproto->sflow);
1514     dpif_ipfix_unref(ofproto->ipfix);
1515     hmap_destroy(&ofproto->bundles);
1516     mac_learning_unref(ofproto->ml);
1517     mcast_snooping_unref(ofproto->ms);
1518
1519     sset_destroy(&ofproto->ports);
1520     sset_destroy(&ofproto->ghost_ports);
1521     sset_destroy(&ofproto->port_poll_set);
1522
1523     ovs_mutex_destroy(&ofproto->stats_mutex);
1524
1525     seq_destroy(ofproto->ams_seq);
1526
1527     close_dpif_backer(ofproto->backer);
1528 }
1529
1530 static int
1531 run(struct ofproto *ofproto_)
1532 {
1533     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1534     uint64_t new_seq, new_dump_seq;
1535
1536     if (mbridge_need_revalidate(ofproto->mbridge)) {
1537         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1538         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1539         mac_learning_flush(ofproto->ml);
1540         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1541         mcast_snooping_mdb_flush(ofproto->ms);
1542     }
1543
1544     /* Always updates the ofproto->ams_seqno to avoid frequent wakeup during
1545      * flow restore.  Even though nothing is processed during flow restore,
1546      * all queued 'ams' will be handled immediately when flow restore
1547      * completes. */
1548     ofproto->ams_seqno = seq_read(ofproto->ams_seq);
1549
1550     /* Do not perform any periodic activity required by 'ofproto' while
1551      * waiting for flow restore to complete. */
1552     if (!ofproto_get_flow_restore_wait()) {
1553         struct ofproto_async_msg *am;
1554         struct ovs_list ams;
1555
1556         guarded_list_pop_all(&ofproto->ams, &ams);
1557         LIST_FOR_EACH_POP (am, list_node, &ams) {
1558             connmgr_send_async_msg(ofproto->up.connmgr, am);
1559             ofproto_async_msg_free(am);
1560         }
1561     }
1562
1563     if (ofproto->netflow) {
1564         netflow_run(ofproto->netflow);
1565     }
1566     if (ofproto->sflow) {
1567         dpif_sflow_run(ofproto->sflow);
1568     }
1569     if (ofproto->ipfix) {
1570         dpif_ipfix_run(ofproto->ipfix);
1571     }
1572
1573     new_seq = seq_read(connectivity_seq_get());
1574     if (ofproto->change_seq != new_seq) {
1575         struct ofport_dpif *ofport;
1576
1577         HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1578             port_run(ofport);
1579         }
1580
1581         ofproto->change_seq = new_seq;
1582     }
1583     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1584         struct ofbundle *bundle;
1585
1586         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1587             bundle_run(bundle);
1588         }
1589     }
1590
1591     stp_run(ofproto);
1592     rstp_run(ofproto);
1593     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1594     if (mac_learning_run(ofproto->ml)) {
1595         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1596     }
1597     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1598
1599     if (mcast_snooping_run(ofproto->ms)) {
1600         ofproto->backer->need_revalidate = REV_MCAST_SNOOPING;
1601     }
1602
1603     new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1604     if (ofproto->dump_seq != new_dump_seq) {
1605         struct rule *rule, *next_rule;
1606         long long now = time_msec();
1607
1608         /* We know stats are relatively fresh, so now is a good time to do some
1609          * periodic work. */
1610         ofproto->dump_seq = new_dump_seq;
1611
1612         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1613          * has passed. */
1614         ovs_mutex_lock(&ofproto_mutex);
1615         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1616                             &ofproto->up.expirable) {
1617             rule_expire(rule_dpif_cast(rule), now);
1618         }
1619         ovs_mutex_unlock(&ofproto_mutex);
1620
1621         /* All outstanding data in existing flows has been accounted, so it's a
1622          * good time to do bond rebalancing. */
1623         if (ofproto->has_bonded_bundles) {
1624             struct ofbundle *bundle;
1625
1626             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1627                 if (bundle->bond) {
1628                     bond_rebalance(bundle->bond);
1629                 }
1630             }
1631         }
1632     }
1633     return 0;
1634 }
1635
1636 static void
1637 ofproto_dpif_wait(struct ofproto *ofproto_)
1638 {
1639     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1640
1641     if (ofproto_get_flow_restore_wait()) {
1642         return;
1643     }
1644
1645     if (ofproto->sflow) {
1646         dpif_sflow_wait(ofproto->sflow);
1647     }
1648     if (ofproto->ipfix) {
1649         dpif_ipfix_wait(ofproto->ipfix);
1650     }
1651     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1652         struct ofbundle *bundle;
1653
1654         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1655             bundle_wait(bundle);
1656         }
1657     }
1658     if (ofproto->netflow) {
1659         netflow_wait(ofproto->netflow);
1660     }
1661     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1662     mac_learning_wait(ofproto->ml);
1663     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1664     mcast_snooping_wait(ofproto->ms);
1665     stp_wait(ofproto);
1666     if (ofproto->backer->need_revalidate) {
1667         poll_immediate_wake();
1668     }
1669
1670     seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1671     seq_wait(ofproto->ams_seq, ofproto->ams_seqno);
1672 }
1673
1674 static void
1675 type_get_memory_usage(const char *type, struct simap *usage)
1676 {
1677     struct dpif_backer *backer;
1678
1679     backer = shash_find_data(&all_dpif_backers, type);
1680     if (backer) {
1681         udpif_get_memory_usage(backer->udpif, usage);
1682     }
1683 }
1684
1685 static void
1686 flush(struct ofproto *ofproto_)
1687 {
1688     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1689     struct dpif_backer *backer = ofproto->backer;
1690
1691     if (backer) {
1692         udpif_flush(backer->udpif);
1693     }
1694 }
1695
1696 static void
1697 query_tables(struct ofproto *ofproto,
1698              struct ofputil_table_features *features,
1699              struct ofputil_table_stats *stats)
1700 {
1701     strcpy(features->name, "classifier");
1702
1703     if (stats) {
1704         int i;
1705
1706         for (i = 0; i < ofproto->n_tables; i++) {
1707             unsigned long missed, matched;
1708
1709             atomic_read_relaxed(&ofproto->tables[i].n_matched, &matched);
1710             atomic_read_relaxed(&ofproto->tables[i].n_missed, &missed);
1711
1712             stats[i].matched_count = matched;
1713             stats[i].lookup_count = matched + missed;
1714         }
1715     }
1716 }
1717
1718 static void
1719 set_tables_version(struct ofproto *ofproto_, cls_version_t version)
1720 {
1721     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1722
1723     atomic_store_relaxed(&ofproto->tables_version, version);
1724     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
1725 }
1726
1727 static struct ofport *
1728 port_alloc(void)
1729 {
1730     struct ofport_dpif *port = xzalloc(sizeof *port);
1731     return &port->up;
1732 }
1733
1734 static void
1735 port_dealloc(struct ofport *port_)
1736 {
1737     struct ofport_dpif *port = ofport_dpif_cast(port_);
1738     free(port);
1739 }
1740
1741 static int
1742 port_construct(struct ofport *port_)
1743 {
1744     struct ofport_dpif *port = ofport_dpif_cast(port_);
1745     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1746     const struct netdev *netdev = port->up.netdev;
1747     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1748     const char *dp_port_name;
1749     struct dpif_port dpif_port;
1750     int error;
1751
1752     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1753     port->bundle = NULL;
1754     port->cfm = NULL;
1755     port->bfd = NULL;
1756     port->lldp = NULL;
1757     port->may_enable = false;
1758     port->stp_port = NULL;
1759     port->stp_state = STP_DISABLED;
1760     port->rstp_port = NULL;
1761     port->rstp_state = RSTP_DISABLED;
1762     port->is_tunnel = false;
1763     port->peer = NULL;
1764     port->qdscp = NULL;
1765     port->n_qdscp = 0;
1766     port->carrier_seq = netdev_get_carrier_resets(netdev);
1767     port->is_layer3 = netdev_vport_is_layer3(netdev);
1768
1769     if (netdev_vport_is_patch(netdev)) {
1770         /* By bailing out here, we don't submit the port to the sFlow module
1771          * to be considered for counter polling export.  This is correct
1772          * because the patch port represents an interface that sFlow considers
1773          * to be "internal" to the switch as a whole, and therefore not a
1774          * candidate for counter polling. */
1775         port->odp_port = ODPP_NONE;
1776         ofport_update_peer(port);
1777         return 0;
1778     }
1779
1780     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1781     error = dpif_port_query_by_name(ofproto->backer->dpif, dp_port_name,
1782                                     &dpif_port);
1783     if (error) {
1784         return error;
1785     }
1786
1787     port->odp_port = dpif_port.port_no;
1788
1789     if (netdev_get_tunnel_config(netdev)) {
1790         atomic_count_inc(&ofproto->backer->tnl_count);
1791         error = tnl_port_add(port, port->up.netdev, port->odp_port,
1792                              ovs_native_tunneling_is_on(ofproto), dp_port_name);
1793         if (error) {
1794             atomic_count_dec(&ofproto->backer->tnl_count);
1795             dpif_port_destroy(&dpif_port);
1796             return error;
1797         }
1798
1799         port->is_tunnel = true;
1800         if (ofproto->ipfix) {
1801            dpif_ipfix_add_tunnel_port(ofproto->ipfix, port_, port->odp_port);
1802         }
1803     } else {
1804         /* Sanity-check that a mapping doesn't already exist.  This
1805          * shouldn't happen for non-tunnel ports. */
1806         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1807             VLOG_ERR("port %s already has an OpenFlow port number",
1808                      dpif_port.name);
1809             dpif_port_destroy(&dpif_port);
1810             return EBUSY;
1811         }
1812
1813         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1814         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1815                     hash_odp_port(port->odp_port));
1816         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1817     }
1818     dpif_port_destroy(&dpif_port);
1819
1820     if (ofproto->sflow) {
1821         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1822     }
1823
1824     return 0;
1825 }
1826
1827 static void
1828 port_destruct(struct ofport *port_, bool del)
1829 {
1830     struct ofport_dpif *port = ofport_dpif_cast(port_);
1831     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1832     const char *devname = netdev_get_name(port->up.netdev);
1833     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1834     const char *dp_port_name;
1835
1836     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1837     xlate_txn_start();
1838     xlate_ofport_remove(port);
1839     xlate_txn_commit();
1840
1841     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1842                                               sizeof namebuf);
1843     if (del && dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1844         /* The underlying device is still there, so delete it.  This
1845          * happens when the ofproto is being destroyed, since the caller
1846          * assumes that removal of attached ports will happen as part of
1847          * destruction. */
1848         if (!port->is_tunnel) {
1849             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1850         }
1851     }
1852
1853     if (port->peer) {
1854         port->peer->peer = NULL;
1855         port->peer = NULL;
1856     }
1857
1858     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1859         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1860         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1861         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1862     }
1863
1864     if (port->is_tunnel) {
1865         atomic_count_dec(&ofproto->backer->tnl_count);
1866     }
1867
1868     if (port->is_tunnel && ofproto->ipfix) {
1869        dpif_ipfix_del_tunnel_port(ofproto->ipfix, port->odp_port);
1870     }
1871
1872     tnl_port_del(port);
1873     sset_find_and_delete(&ofproto->ports, devname);
1874     sset_find_and_delete(&ofproto->ghost_ports, devname);
1875     bundle_remove(port_);
1876     set_cfm(port_, NULL);
1877     set_bfd(port_, NULL);
1878     set_lldp(port_, NULL);
1879     if (port->stp_port) {
1880         stp_port_disable(port->stp_port);
1881     }
1882     set_rstp_port(port_, NULL);
1883     if (ofproto->sflow) {
1884         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1885     }
1886
1887     free(port->qdscp);
1888 }
1889
1890 static void
1891 port_modified(struct ofport *port_)
1892 {
1893     struct ofport_dpif *port = ofport_dpif_cast(port_);
1894     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1895     const char *dp_port_name;
1896     struct netdev *netdev = port->up.netdev;
1897
1898     if (port->bundle && port->bundle->bond) {
1899         bond_slave_set_netdev(port->bundle->bond, port, netdev);
1900     }
1901
1902     if (port->cfm) {
1903         cfm_set_netdev(port->cfm, netdev);
1904     }
1905
1906     if (port->bfd) {
1907         bfd_set_netdev(port->bfd, netdev);
1908     }
1909
1910     ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1911                                      port->lldp, &port->up.pp.hw_addr);
1912
1913     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1914
1915     if (port->is_tunnel) {
1916         struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1917
1918         if (tnl_port_reconfigure(port, netdev, port->odp_port,
1919                                  ovs_native_tunneling_is_on(ofproto),
1920                                  dp_port_name)) {
1921             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1922         }
1923     }
1924
1925     ofport_update_peer(port);
1926 }
1927
1928 static void
1929 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1930 {
1931     struct ofport_dpif *port = ofport_dpif_cast(port_);
1932     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1933     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1934
1935     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1936                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1937                    OFPUTIL_PC_NO_PACKET_IN)) {
1938         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1939
1940         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1941             bundle_update(port->bundle);
1942         }
1943     }
1944 }
1945
1946 static int
1947 set_sflow(struct ofproto *ofproto_,
1948           const struct ofproto_sflow_options *sflow_options)
1949 {
1950     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1951     struct dpif_sflow *ds = ofproto->sflow;
1952
1953     if (sflow_options) {
1954         uint32_t old_probability = ds ? dpif_sflow_get_probability(ds) : 0;
1955         if (!ds) {
1956             struct ofport_dpif *ofport;
1957
1958             ds = ofproto->sflow = dpif_sflow_create();
1959             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1960                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1961             }
1962         }
1963         dpif_sflow_set_options(ds, sflow_options);
1964         if (dpif_sflow_get_probability(ds) != old_probability) {
1965             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1966         }
1967     } else {
1968         if (ds) {
1969             dpif_sflow_unref(ds);
1970             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1971             ofproto->sflow = NULL;
1972         }
1973     }
1974     return 0;
1975 }
1976
1977 static int
1978 set_ipfix(
1979     struct ofproto *ofproto_,
1980     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1981     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1982     size_t n_flow_exporters_options)
1983 {
1984     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1985     struct dpif_ipfix *di = ofproto->ipfix;
1986     bool has_options = bridge_exporter_options || flow_exporters_options;
1987     bool new_di = false;
1988
1989     if (has_options && !di) {
1990         di = ofproto->ipfix = dpif_ipfix_create();
1991         new_di = true;
1992     }
1993
1994     if (di) {
1995         /* Call set_options in any case to cleanly flush the flow
1996          * caches in the last exporters that are to be destroyed. */
1997         dpif_ipfix_set_options(
1998             di, bridge_exporter_options, flow_exporters_options,
1999             n_flow_exporters_options);
2000
2001         /* Add tunnel ports only when a new ipfix created */
2002         if (new_di == true) {
2003             struct ofport_dpif *ofport;
2004             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
2005                 if (ofport->is_tunnel == true) {
2006                     dpif_ipfix_add_tunnel_port(di, &ofport->up, ofport->odp_port);
2007                 }
2008             }
2009         }
2010
2011         if (!has_options) {
2012             dpif_ipfix_unref(di);
2013             ofproto->ipfix = NULL;
2014         }
2015     }
2016
2017     return 0;
2018 }
2019
2020 static int
2021 get_ipfix_stats(const struct ofproto *ofproto_,
2022                 bool bridge_ipfix,
2023                 struct ovs_list *replies)
2024 {
2025     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2026     struct dpif_ipfix *di = ofproto->ipfix;
2027
2028     if (!di) {
2029         return OFPERR_NXST_NOT_CONFIGURED;
2030     }
2031
2032     return dpif_ipfix_get_stats(di, bridge_ipfix, replies);
2033 }
2034
2035 static int
2036 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
2037 {
2038     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2039     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2040     struct cfm *old = ofport->cfm;
2041     int error = 0;
2042
2043     if (s) {
2044         if (!ofport->cfm) {
2045             ofport->cfm = cfm_create(ofport->up.netdev);
2046         }
2047
2048         if (cfm_configure(ofport->cfm, s)) {
2049             error = 0;
2050             goto out;
2051         }
2052
2053         error = EINVAL;
2054     }
2055     cfm_unref(ofport->cfm);
2056     ofport->cfm = NULL;
2057 out:
2058     if (ofport->cfm != old) {
2059         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2060     }
2061     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
2062                                      ofport->lldp, &ofport->up.pp.hw_addr);
2063     return error;
2064 }
2065
2066 static bool
2067 cfm_status_changed(struct ofport *ofport_)
2068 {
2069     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2070
2071     return ofport->cfm ? cfm_check_status_change(ofport->cfm) : true;
2072 }
2073
2074 static int
2075 get_cfm_status(const struct ofport *ofport_,
2076                struct cfm_status *status)
2077 {
2078     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2079     int ret = 0;
2080
2081     if (ofport->cfm) {
2082         cfm_get_status(ofport->cfm, status);
2083     } else {
2084         ret = ENOENT;
2085     }
2086
2087     return ret;
2088 }
2089
2090 static int
2091 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2092 {
2093     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2094     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2095     struct bfd *old;
2096
2097     old = ofport->bfd;
2098     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2099                                 cfg, ofport->up.netdev);
2100     if (ofport->bfd != old) {
2101         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2102     }
2103     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
2104                                      ofport->lldp, &ofport->up.pp.hw_addr);
2105     return 0;
2106 }
2107
2108 static bool
2109 bfd_status_changed(struct ofport *ofport_)
2110 {
2111     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2112
2113     return ofport->bfd ? bfd_check_status_change(ofport->bfd) : true;
2114 }
2115
2116 static int
2117 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2118 {
2119     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2120     int ret = 0;
2121
2122     if (ofport->bfd) {
2123         bfd_get_status(ofport->bfd, smap);
2124     } else {
2125         ret = ENOENT;
2126     }
2127
2128     return ret;
2129 }
2130
2131 static int
2132 set_lldp(struct ofport *ofport_,
2133          const struct smap *cfg)
2134 {
2135     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2136     int error = 0;
2137
2138     if (cfg) {
2139         if (!ofport->lldp) {
2140             struct ofproto_dpif *ofproto;
2141
2142             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2143             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2144             ofport->lldp = lldp_create(ofport->up.netdev, ofport_->mtu, cfg);
2145         }
2146
2147         if (!lldp_configure(ofport->lldp, cfg)) {
2148             error = EINVAL;
2149         }
2150     }
2151     if (error) {
2152         lldp_unref(ofport->lldp);
2153         ofport->lldp = NULL;
2154     }
2155
2156     ofproto_dpif_monitor_port_update(ofport,
2157                                      ofport->bfd,
2158                                      ofport->cfm,
2159                                      ofport->lldp,
2160                                      &ofport->up.pp.hw_addr);
2161     return error;
2162 }
2163
2164 static bool
2165 get_lldp_status(const struct ofport *ofport_,
2166                struct lldp_status *status OVS_UNUSED)
2167 {
2168     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2169
2170     return ofport->lldp ? true : false;
2171 }
2172
2173 static int
2174 set_aa(struct ofproto *ofproto OVS_UNUSED,
2175        const struct aa_settings *s)
2176 {
2177     return aa_configure(s);
2178 }
2179
2180 static int
2181 aa_mapping_set(struct ofproto *ofproto_ OVS_UNUSED, void *aux,
2182                const struct aa_mapping_settings *s)
2183 {
2184     return aa_mapping_register(aux, s);
2185 }
2186
2187 static int
2188 aa_mapping_unset(struct ofproto *ofproto OVS_UNUSED, void *aux)
2189 {
2190     return aa_mapping_unregister(aux);
2191 }
2192
2193 static int
2194 aa_vlan_get_queued(struct ofproto *ofproto OVS_UNUSED, struct ovs_list *list)
2195 {
2196     return aa_get_vlan_queued(list);
2197 }
2198
2199 static unsigned int
2200 aa_vlan_get_queue_size(struct ofproto *ofproto OVS_UNUSED)
2201 {
2202     return aa_get_vlan_queue_size();
2203 }
2204
2205 \f
2206 /* Spanning Tree. */
2207
2208 /* Called while rstp_mutex is held. */
2209 static void
2210 rstp_send_bpdu_cb(struct dp_packet *pkt, void *ofport_, void *ofproto_)
2211 {
2212     struct ofproto_dpif *ofproto = ofproto_;
2213     struct ofport_dpif *ofport = ofport_;
2214     struct eth_header *eth = dp_packet_l2(pkt);
2215
2216     netdev_get_etheraddr(ofport->up.netdev, &eth->eth_src);
2217     if (eth_addr_is_zero(eth->eth_src)) {
2218         VLOG_WARN_RL(&rl, "%s port %d: cannot send RSTP BPDU on a port which "
2219                      "does not have a configured source MAC address.",
2220                      ofproto->up.name, ofp_to_u16(ofport->up.ofp_port));
2221     } else {
2222         ofproto_dpif_send_packet(ofport, pkt);
2223     }
2224     dp_packet_delete(pkt);
2225 }
2226
2227 static void
2228 send_bpdu_cb(struct dp_packet *pkt, int port_num, void *ofproto_)
2229 {
2230     struct ofproto_dpif *ofproto = ofproto_;
2231     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2232     struct ofport_dpif *ofport;
2233
2234     ofport = stp_port_get_aux(sp);
2235     if (!ofport) {
2236         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2237                      ofproto->up.name, port_num);
2238     } else {
2239         struct eth_header *eth = dp_packet_l2(pkt);
2240
2241         netdev_get_etheraddr(ofport->up.netdev, &eth->eth_src);
2242         if (eth_addr_is_zero(eth->eth_src)) {
2243             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2244                          "with unknown MAC", ofproto->up.name, port_num);
2245         } else {
2246             ofproto_dpif_send_packet(ofport, pkt);
2247         }
2248     }
2249     dp_packet_delete(pkt);
2250 }
2251
2252 /* Configure RSTP on 'ofproto_' using the settings defined in 's'. */
2253 static void
2254 set_rstp(struct ofproto *ofproto_, const struct ofproto_rstp_settings *s)
2255 {
2256     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2257
2258     /* Only revalidate flows if the configuration changed. */
2259     if (!s != !ofproto->rstp) {
2260         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2261     }
2262
2263     if (s) {
2264         if (!ofproto->rstp) {
2265             ofproto->rstp = rstp_create(ofproto_->name, s->address,
2266                                         rstp_send_bpdu_cb, ofproto);
2267             ofproto->rstp_last_tick = time_msec();
2268         }
2269         rstp_set_bridge_address(ofproto->rstp, s->address);
2270         rstp_set_bridge_priority(ofproto->rstp, s->priority);
2271         rstp_set_bridge_ageing_time(ofproto->rstp, s->ageing_time);
2272         rstp_set_bridge_force_protocol_version(ofproto->rstp,
2273                                                s->force_protocol_version);
2274         rstp_set_bridge_max_age(ofproto->rstp, s->bridge_max_age);
2275         rstp_set_bridge_forward_delay(ofproto->rstp, s->bridge_forward_delay);
2276         rstp_set_bridge_transmit_hold_count(ofproto->rstp,
2277                                             s->transmit_hold_count);
2278     } else {
2279         struct ofport *ofport;
2280         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2281             set_rstp_port(ofport, NULL);
2282         }
2283         rstp_unref(ofproto->rstp);
2284         ofproto->rstp = NULL;
2285     }
2286 }
2287
2288 static void
2289 get_rstp_status(struct ofproto *ofproto_, struct ofproto_rstp_status *s)
2290 {
2291     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2292
2293     if (ofproto->rstp) {
2294         s->enabled = true;
2295         s->root_id = rstp_get_root_id(ofproto->rstp);
2296         s->bridge_id = rstp_get_bridge_id(ofproto->rstp);
2297         s->designated_id = rstp_get_designated_id(ofproto->rstp);
2298         s->root_path_cost = rstp_get_root_path_cost(ofproto->rstp);
2299         s->designated_port_id = rstp_get_designated_port_id(ofproto->rstp);
2300         s->bridge_port_id = rstp_get_bridge_port_id(ofproto->rstp);
2301     } else {
2302         s->enabled = false;
2303     }
2304 }
2305
2306 static void
2307 update_rstp_port_state(struct ofport_dpif *ofport)
2308 {
2309     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2310     enum rstp_state state;
2311
2312     /* Figure out new state. */
2313     state = ofport->rstp_port ? rstp_port_get_state(ofport->rstp_port)
2314         : RSTP_DISABLED;
2315
2316     /* Update state. */
2317     if (ofport->rstp_state != state) {
2318         enum ofputil_port_state of_state;
2319         bool fwd_change;
2320
2321         VLOG_DBG("port %s: RSTP state changed from %s to %s",
2322                  netdev_get_name(ofport->up.netdev),
2323                  rstp_state_name(ofport->rstp_state),
2324                  rstp_state_name(state));
2325
2326         if (rstp_learn_in_state(ofport->rstp_state)
2327             != rstp_learn_in_state(state)) {
2328             /* XXX: Learning action flows should also be flushed. */
2329             if (ofport->bundle) {
2330                 if (!rstp_shift_root_learned_address(ofproto->rstp)
2331                     || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2332                     bundle_flush_macs(ofport->bundle, false);
2333                 }
2334             }
2335         }
2336         fwd_change = rstp_forward_in_state(ofport->rstp_state)
2337             != rstp_forward_in_state(state);
2338
2339         ofproto->backer->need_revalidate = REV_RSTP;
2340         ofport->rstp_state = state;
2341
2342         if (fwd_change && ofport->bundle) {
2343             bundle_update(ofport->bundle);
2344         }
2345
2346         /* Update the RSTP state bits in the OpenFlow port description. */
2347         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2348         of_state |= (state == RSTP_LEARNING ? OFPUTIL_PS_STP_LEARN
2349                 : state == RSTP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2350                 : state == RSTP_DISCARDING ?  OFPUTIL_PS_STP_LISTEN
2351                 : 0);
2352         ofproto_port_set_state(&ofport->up, of_state);
2353     }
2354 }
2355
2356 static void
2357 rstp_run(struct ofproto_dpif *ofproto)
2358 {
2359     if (ofproto->rstp) {
2360         long long int now = time_msec();
2361         long long int elapsed = now - ofproto->rstp_last_tick;
2362         struct rstp_port *rp;
2363         struct ofport_dpif *ofport;
2364
2365         /* Every second, decrease the values of the timers. */
2366         if (elapsed >= 1000) {
2367             rstp_tick_timers(ofproto->rstp);
2368             ofproto->rstp_last_tick = now;
2369         }
2370         rp = NULL;
2371         while ((ofport = rstp_get_next_changed_port_aux(ofproto->rstp, &rp))) {
2372             update_rstp_port_state(ofport);
2373         }
2374         rp = NULL;
2375         ofport = NULL;
2376         /* FIXME: This check should be done on-event (i.e., when setting
2377          * p->fdb_flush) and not periodically.
2378          */
2379         while ((ofport = rstp_check_and_reset_fdb_flush(ofproto->rstp, &rp))) {
2380             if (!rstp_shift_root_learned_address(ofproto->rstp)
2381                 || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2382                 bundle_flush_macs(ofport->bundle, false);
2383             }
2384         }
2385
2386         if (rstp_shift_root_learned_address(ofproto->rstp)) {
2387             struct ofport_dpif *old_root_aux =
2388                 (struct ofport_dpif *)rstp_get_old_root_aux(ofproto->rstp);
2389             struct ofport_dpif *new_root_aux =
2390                 (struct ofport_dpif *)rstp_get_new_root_aux(ofproto->rstp);
2391             if (old_root_aux != NULL && new_root_aux != NULL) {
2392                 bundle_move(old_root_aux->bundle, new_root_aux->bundle);
2393                 rstp_reset_root_changed(ofproto->rstp);
2394             }
2395         }
2396     }
2397 }
2398
2399 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2400 static int
2401 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2402 {
2403     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2404
2405     /* Only revalidate flows if the configuration changed. */
2406     if (!s != !ofproto->stp) {
2407         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2408     }
2409
2410     if (s) {
2411         if (!ofproto->stp) {
2412             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2413                                       send_bpdu_cb, ofproto);
2414             ofproto->stp_last_tick = time_msec();
2415         }
2416
2417         stp_set_bridge_id(ofproto->stp, s->system_id);
2418         stp_set_bridge_priority(ofproto->stp, s->priority);
2419         stp_set_hello_time(ofproto->stp, s->hello_time);
2420         stp_set_max_age(ofproto->stp, s->max_age);
2421         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2422     }  else {
2423         struct ofport *ofport;
2424
2425         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2426             set_stp_port(ofport, NULL);
2427         }
2428
2429         stp_unref(ofproto->stp);
2430         ofproto->stp = NULL;
2431     }
2432
2433     return 0;
2434 }
2435
2436 static int
2437 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2438 {
2439     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2440
2441     if (ofproto->stp) {
2442         s->enabled = true;
2443         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2444         s->designated_root = stp_get_designated_root(ofproto->stp);
2445         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2446     } else {
2447         s->enabled = false;
2448     }
2449
2450     return 0;
2451 }
2452
2453 static void
2454 update_stp_port_state(struct ofport_dpif *ofport)
2455 {
2456     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2457     enum stp_state state;
2458
2459     /* Figure out new state. */
2460     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2461                              : STP_DISABLED;
2462
2463     /* Update state. */
2464     if (ofport->stp_state != state) {
2465         enum ofputil_port_state of_state;
2466         bool fwd_change;
2467
2468         VLOG_DBG("port %s: STP state changed from %s to %s",
2469                  netdev_get_name(ofport->up.netdev),
2470                  stp_state_name(ofport->stp_state),
2471                  stp_state_name(state));
2472         if (stp_learn_in_state(ofport->stp_state)
2473                 != stp_learn_in_state(state)) {
2474             /* xxx Learning action flows should also be flushed. */
2475             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2476             mac_learning_flush(ofproto->ml);
2477             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2478             mcast_snooping_mdb_flush(ofproto->ms);
2479         }
2480         fwd_change = stp_forward_in_state(ofport->stp_state)
2481                         != stp_forward_in_state(state);
2482
2483         ofproto->backer->need_revalidate = REV_STP;
2484         ofport->stp_state = state;
2485         ofport->stp_state_entered = time_msec();
2486
2487         if (fwd_change && ofport->bundle) {
2488             bundle_update(ofport->bundle);
2489         }
2490
2491         /* Update the STP state bits in the OpenFlow port description. */
2492         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2493         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2494                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2495                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2496                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2497                      : 0);
2498         ofproto_port_set_state(&ofport->up, of_state);
2499     }
2500 }
2501
2502 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2503  * caller is responsible for assigning STP port numbers and ensuring
2504  * there are no duplicates. */
2505 static int
2506 set_stp_port(struct ofport *ofport_,
2507              const struct ofproto_port_stp_settings *s)
2508 {
2509     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2510     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2511     struct stp_port *sp = ofport->stp_port;
2512
2513     if (!s || !s->enable) {
2514         if (sp) {
2515             ofport->stp_port = NULL;
2516             stp_port_disable(sp);
2517             update_stp_port_state(ofport);
2518         }
2519         return 0;
2520     } else if (sp && stp_port_no(sp) != s->port_num
2521                && ofport == stp_port_get_aux(sp)) {
2522         /* The port-id changed, so disable the old one if it's not
2523          * already in use by another port. */
2524         stp_port_disable(sp);
2525     }
2526
2527     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2528
2529     /* Set name before enabling the port so that debugging messages can print
2530      * the name. */
2531     stp_port_set_name(sp, netdev_get_name(ofport->up.netdev));
2532     stp_port_enable(sp);
2533
2534     stp_port_set_aux(sp, ofport);
2535     stp_port_set_priority(sp, s->priority);
2536     stp_port_set_path_cost(sp, s->path_cost);
2537
2538     update_stp_port_state(ofport);
2539
2540     return 0;
2541 }
2542
2543 static int
2544 get_stp_port_status(struct ofport *ofport_,
2545                     struct ofproto_port_stp_status *s)
2546 {
2547     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2548     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2549     struct stp_port *sp = ofport->stp_port;
2550
2551     if (!ofproto->stp || !sp) {
2552         s->enabled = false;
2553         return 0;
2554     }
2555
2556     s->enabled = true;
2557     s->port_id = stp_port_get_id(sp);
2558     s->state = stp_port_get_state(sp);
2559     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2560     s->role = stp_port_get_role(sp);
2561
2562     return 0;
2563 }
2564
2565 static int
2566 get_stp_port_stats(struct ofport *ofport_,
2567                    struct ofproto_port_stp_stats *s)
2568 {
2569     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2570     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2571     struct stp_port *sp = ofport->stp_port;
2572
2573     if (!ofproto->stp || !sp) {
2574         s->enabled = false;
2575         return 0;
2576     }
2577
2578     s->enabled = true;
2579     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2580
2581     return 0;
2582 }
2583
2584 static void
2585 stp_run(struct ofproto_dpif *ofproto)
2586 {
2587     if (ofproto->stp) {
2588         long long int now = time_msec();
2589         long long int elapsed = now - ofproto->stp_last_tick;
2590         struct stp_port *sp;
2591
2592         if (elapsed > 0) {
2593             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2594             ofproto->stp_last_tick = now;
2595         }
2596         while (stp_get_changed_port(ofproto->stp, &sp)) {
2597             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2598
2599             if (ofport) {
2600                 update_stp_port_state(ofport);
2601             }
2602         }
2603
2604         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2605             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2606             mac_learning_flush(ofproto->ml);
2607             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2608             mcast_snooping_mdb_flush(ofproto->ms);
2609         }
2610     }
2611 }
2612
2613 static void
2614 stp_wait(struct ofproto_dpif *ofproto)
2615 {
2616     if (ofproto->stp) {
2617         poll_timer_wait(1000);
2618     }
2619 }
2620
2621 /* Configures RSTP on 'ofport_' using the settings defined in 's'.  The
2622  * caller is responsible for assigning RSTP port numbers and ensuring
2623  * there are no duplicates. */
2624 static void
2625 set_rstp_port(struct ofport *ofport_,
2626               const struct ofproto_port_rstp_settings *s)
2627 {
2628     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2629     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2630     struct rstp_port *rp = ofport->rstp_port;
2631
2632     if (!s || !s->enable) {
2633         if (rp) {
2634             rstp_port_set_aux(rp, NULL);
2635             rstp_port_set_state(rp, RSTP_DISABLED);
2636             rstp_port_set_mac_operational(rp, false);
2637             ofport->rstp_port = NULL;
2638             rstp_port_unref(rp);
2639             update_rstp_port_state(ofport);
2640         }
2641         return;
2642     }
2643
2644     /* Check if need to add a new port. */
2645     if (!rp) {
2646         rp = ofport->rstp_port = rstp_add_port(ofproto->rstp);
2647     }
2648
2649     rstp_port_set(rp, s->port_num, s->priority, s->path_cost,
2650                   s->admin_edge_port, s->auto_edge,
2651                   s->admin_p2p_mac_state, s->admin_port_state, s->mcheck,
2652                   ofport);
2653     update_rstp_port_state(ofport);
2654     /* Synchronize operational status. */
2655     rstp_port_set_mac_operational(rp, ofport->may_enable);
2656 }
2657
2658 static void
2659 get_rstp_port_status(struct ofport *ofport_,
2660         struct ofproto_port_rstp_status *s)
2661 {
2662     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2663     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2664     struct rstp_port *rp = ofport->rstp_port;
2665
2666     if (!ofproto->rstp || !rp) {
2667         s->enabled = false;
2668         return;
2669     }
2670
2671     s->enabled = true;
2672     rstp_port_get_status(rp, &s->port_id, &s->state, &s->role,
2673                          &s->designated_bridge_id, &s->designated_port_id,
2674                          &s->designated_path_cost, &s->tx_count,
2675                          &s->rx_count, &s->error_count, &s->uptime);
2676 }
2677
2678 \f
2679 static int
2680 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2681            size_t n_qdscp)
2682 {
2683     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2684     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2685
2686     if (ofport->n_qdscp != n_qdscp
2687         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2688                               n_qdscp * sizeof *qdscp))) {
2689         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2690         free(ofport->qdscp);
2691         ofport->qdscp = n_qdscp
2692             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2693             : NULL;
2694         ofport->n_qdscp = n_qdscp;
2695     }
2696
2697     return 0;
2698 }
2699 \f
2700 /* Bundles. */
2701
2702 /* Expires all MAC learning entries associated with 'bundle' and forces its
2703  * ofproto to revalidate every flow.
2704  *
2705  * Normally MAC learning entries are removed only from the ofproto associated
2706  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2707  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2708  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2709  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2710  * with the host from which it migrated. */
2711 static void
2712 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2713 {
2714     struct ofproto_dpif *ofproto = bundle->ofproto;
2715     struct mac_learning *ml = ofproto->ml;
2716     struct mac_entry *mac, *next_mac;
2717
2718     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2719     ovs_rwlock_wrlock(&ml->rwlock);
2720     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2721         if (mac_entry_get_port(ml, mac) == bundle) {
2722             if (all_ofprotos) {
2723                 struct ofproto_dpif *o;
2724
2725                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2726                     if (o != ofproto) {
2727                         struct mac_entry *e;
2728
2729                         ovs_rwlock_wrlock(&o->ml->rwlock);
2730                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2731                         if (e) {
2732                             mac_learning_expire(o->ml, e);
2733                         }
2734                         ovs_rwlock_unlock(&o->ml->rwlock);
2735                     }
2736                 }
2737             }
2738
2739             mac_learning_expire(ml, mac);
2740         }
2741     }
2742     ovs_rwlock_unlock(&ml->rwlock);
2743 }
2744
2745 static void
2746 bundle_move(struct ofbundle *old, struct ofbundle *new)
2747 {
2748     struct ofproto_dpif *ofproto = old->ofproto;
2749     struct mac_learning *ml = ofproto->ml;
2750     struct mac_entry *mac, *next_mac;
2751
2752     ovs_assert(new->ofproto == old->ofproto);
2753
2754     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2755     ovs_rwlock_wrlock(&ml->rwlock);
2756     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2757         if (mac_entry_get_port(ml, mac) == old) {
2758             mac_entry_set_port(ml, mac, new);
2759         }
2760     }
2761     ovs_rwlock_unlock(&ml->rwlock);
2762 }
2763
2764 static struct ofbundle *
2765 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2766 {
2767     struct ofbundle *bundle;
2768
2769     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2770                              &ofproto->bundles) {
2771         if (bundle->aux == aux) {
2772             return bundle;
2773         }
2774     }
2775     return NULL;
2776 }
2777
2778 static void
2779 bundle_update(struct ofbundle *bundle)
2780 {
2781     struct ofport_dpif *port;
2782
2783     bundle->floodable = true;
2784     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2785         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2786             || port->is_layer3
2787             || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2788             || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2789             bundle->floodable = false;
2790             break;
2791         }
2792     }
2793 }
2794
2795 static void
2796 bundle_del_port(struct ofport_dpif *port)
2797 {
2798     struct ofbundle *bundle = port->bundle;
2799
2800     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2801
2802     ovs_list_remove(&port->bundle_node);
2803     port->bundle = NULL;
2804
2805     if (bundle->lacp) {
2806         lacp_slave_unregister(bundle->lacp, port);
2807     }
2808     if (bundle->bond) {
2809         bond_slave_unregister(bundle->bond, port);
2810     }
2811
2812     bundle_update(bundle);
2813 }
2814
2815 static bool
2816 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2817                 struct lacp_slave_settings *lacp)
2818 {
2819     struct ofport_dpif *port;
2820
2821     port = ofp_port_to_ofport(bundle->ofproto, ofp_port);
2822     if (!port) {
2823         return false;
2824     }
2825
2826     if (port->bundle != bundle) {
2827         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2828         if (port->bundle) {
2829             bundle_remove(&port->up);
2830         }
2831
2832         port->bundle = bundle;
2833         ovs_list_push_back(&bundle->ports, &port->bundle_node);
2834         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2835             || port->is_layer3
2836             || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2837             || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2838             bundle->floodable = false;
2839         }
2840     }
2841     if (lacp) {
2842         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2843         lacp_slave_register(bundle->lacp, port, lacp);
2844     }
2845
2846     return true;
2847 }
2848
2849 static void
2850 bundle_destroy(struct ofbundle *bundle)
2851 {
2852     struct ofproto_dpif *ofproto;
2853     struct ofport_dpif *port, *next_port;
2854
2855     if (!bundle) {
2856         return;
2857     }
2858
2859     ofproto = bundle->ofproto;
2860     mbridge_unregister_bundle(ofproto->mbridge, bundle);
2861
2862     xlate_txn_start();
2863     xlate_bundle_remove(bundle);
2864     xlate_txn_commit();
2865
2866     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2867         bundle_del_port(port);
2868     }
2869
2870     bundle_flush_macs(bundle, true);
2871     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2872     free(bundle->name);
2873     free(bundle->trunks);
2874     lacp_unref(bundle->lacp);
2875     bond_unref(bundle->bond);
2876     free(bundle);
2877 }
2878
2879 static int
2880 bundle_set(struct ofproto *ofproto_, void *aux,
2881            const struct ofproto_bundle_settings *s)
2882 {
2883     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2884     bool need_flush = false;
2885     struct ofport_dpif *port;
2886     struct ofbundle *bundle;
2887     unsigned long *trunks;
2888     int vlan;
2889     size_t i;
2890     bool ok;
2891
2892     if (!s) {
2893         bundle_destroy(bundle_lookup(ofproto, aux));
2894         return 0;
2895     }
2896
2897     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2898     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2899
2900     bundle = bundle_lookup(ofproto, aux);
2901     if (!bundle) {
2902         bundle = xmalloc(sizeof *bundle);
2903
2904         bundle->ofproto = ofproto;
2905         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2906                     hash_pointer(aux, 0));
2907         bundle->aux = aux;
2908         bundle->name = NULL;
2909
2910         ovs_list_init(&bundle->ports);
2911         bundle->vlan_mode = PORT_VLAN_TRUNK;
2912         bundle->vlan = -1;
2913         bundle->trunks = NULL;
2914         bundle->use_priority_tags = s->use_priority_tags;
2915         bundle->lacp = NULL;
2916         bundle->bond = NULL;
2917
2918         bundle->floodable = true;
2919         mbridge_register_bundle(ofproto->mbridge, bundle);
2920     }
2921
2922     if (!bundle->name || strcmp(s->name, bundle->name)) {
2923         free(bundle->name);
2924         bundle->name = xstrdup(s->name);
2925     }
2926
2927     /* LACP. */
2928     if (s->lacp) {
2929         ofproto->lacp_enabled = true;
2930         if (!bundle->lacp) {
2931             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2932             bundle->lacp = lacp_create();
2933         }
2934         lacp_configure(bundle->lacp, s->lacp);
2935     } else {
2936         lacp_unref(bundle->lacp);
2937         bundle->lacp = NULL;
2938     }
2939
2940     /* Update set of ports. */
2941     ok = true;
2942     for (i = 0; i < s->n_slaves; i++) {
2943         if (!bundle_add_port(bundle, s->slaves[i],
2944                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2945             ok = false;
2946         }
2947     }
2948     if (!ok || ovs_list_size(&bundle->ports) != s->n_slaves) {
2949         struct ofport_dpif *next_port;
2950
2951         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2952             for (i = 0; i < s->n_slaves; i++) {
2953                 if (s->slaves[i] == port->up.ofp_port) {
2954                     goto found;
2955                 }
2956             }
2957
2958             bundle_del_port(port);
2959         found: ;
2960         }
2961     }
2962     ovs_assert(ovs_list_size(&bundle->ports) <= s->n_slaves);
2963
2964     if (ovs_list_is_empty(&bundle->ports)) {
2965         bundle_destroy(bundle);
2966         return EINVAL;
2967     }
2968
2969     /* Set VLAN tagging mode */
2970     if (s->vlan_mode != bundle->vlan_mode
2971         || s->use_priority_tags != bundle->use_priority_tags) {
2972         bundle->vlan_mode = s->vlan_mode;
2973         bundle->use_priority_tags = s->use_priority_tags;
2974         need_flush = true;
2975     }
2976
2977     /* Set VLAN tag. */
2978     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2979             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2980             : 0);
2981     if (vlan != bundle->vlan) {
2982         bundle->vlan = vlan;
2983         need_flush = true;
2984     }
2985
2986     /* Get trunked VLANs. */
2987     switch (s->vlan_mode) {
2988     case PORT_VLAN_ACCESS:
2989         trunks = NULL;
2990         break;
2991
2992     case PORT_VLAN_TRUNK:
2993         trunks = CONST_CAST(unsigned long *, s->trunks);
2994         break;
2995
2996     case PORT_VLAN_NATIVE_UNTAGGED:
2997     case PORT_VLAN_NATIVE_TAGGED:
2998         if (vlan != 0 && (!s->trunks
2999                           || !bitmap_is_set(s->trunks, vlan)
3000                           || bitmap_is_set(s->trunks, 0))) {
3001             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
3002             if (s->trunks) {
3003                 trunks = bitmap_clone(s->trunks, 4096);
3004             } else {
3005                 trunks = bitmap_allocate1(4096);
3006             }
3007             bitmap_set1(trunks, vlan);
3008             bitmap_set0(trunks, 0);
3009         } else {
3010             trunks = CONST_CAST(unsigned long *, s->trunks);
3011         }
3012         break;
3013
3014     default:
3015         OVS_NOT_REACHED();
3016     }
3017     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
3018         free(bundle->trunks);
3019         if (trunks == s->trunks) {
3020             bundle->trunks = vlan_bitmap_clone(trunks);
3021         } else {
3022             bundle->trunks = trunks;
3023             trunks = NULL;
3024         }
3025         need_flush = true;
3026     }
3027     if (trunks != s->trunks) {
3028         free(trunks);
3029     }
3030
3031     /* Bonding. */
3032     if (!ovs_list_is_short(&bundle->ports)) {
3033         bundle->ofproto->has_bonded_bundles = true;
3034         if (bundle->bond) {
3035             if (bond_reconfigure(bundle->bond, s->bond)) {
3036                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3037             }
3038         } else {
3039             bundle->bond = bond_create(s->bond, ofproto);
3040             ofproto->backer->need_revalidate = REV_RECONFIGURE;
3041         }
3042
3043         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
3044             bond_slave_register(bundle->bond, port,
3045                                 port->up.ofp_port, port->up.netdev);
3046         }
3047     } else {
3048         bond_unref(bundle->bond);
3049         bundle->bond = NULL;
3050     }
3051
3052     /* If we changed something that would affect MAC learning, un-learn
3053      * everything on this port and force flow revalidation. */
3054     if (need_flush) {
3055         bundle_flush_macs(bundle, false);
3056     }
3057
3058     return 0;
3059 }
3060
3061 static void
3062 bundle_remove(struct ofport *port_)
3063 {
3064     struct ofport_dpif *port = ofport_dpif_cast(port_);
3065     struct ofbundle *bundle = port->bundle;
3066
3067     if (bundle) {
3068         bundle_del_port(port);
3069         if (ovs_list_is_empty(&bundle->ports)) {
3070             bundle_destroy(bundle);
3071         } else if (ovs_list_is_short(&bundle->ports)) {
3072             bond_unref(bundle->bond);
3073             bundle->bond = NULL;
3074         }
3075     }
3076 }
3077
3078 static void
3079 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
3080 {
3081     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3082     struct ofport_dpif *port = port_;
3083     struct eth_addr ea;
3084     int error;
3085
3086     error = netdev_get_etheraddr(port->up.netdev, &ea);
3087     if (!error) {
3088         struct dp_packet packet;
3089         void *packet_pdu;
3090
3091         dp_packet_init(&packet, 0);
3092         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
3093                                  pdu_size);
3094         memcpy(packet_pdu, pdu, pdu_size);
3095
3096         ofproto_dpif_send_packet(port, &packet);
3097         dp_packet_uninit(&packet);
3098     } else {
3099         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
3100                     "%s (%s)", port->bundle->name,
3101                     netdev_get_name(port->up.netdev), ovs_strerror(error));
3102     }
3103 }
3104
3105 static void
3106 bundle_send_learning_packets(struct ofbundle *bundle)
3107 {
3108     struct ofproto_dpif *ofproto = bundle->ofproto;
3109     int error, n_packets, n_errors;
3110     struct mac_entry *e;
3111     struct pkt_list {
3112         struct ovs_list list_node;
3113         struct ofport_dpif *port;
3114         struct dp_packet *pkt;
3115     } *pkt_node;
3116     struct ovs_list packets;
3117
3118     ovs_list_init(&packets);
3119     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3120     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3121         if (mac_entry_get_port(ofproto->ml, e) != bundle) {
3122             pkt_node = xmalloc(sizeof *pkt_node);
3123             pkt_node->pkt = bond_compose_learning_packet(bundle->bond,
3124                                                          e->mac, e->vlan,
3125                                                          (void **)&pkt_node->port);
3126             ovs_list_push_back(&packets, &pkt_node->list_node);
3127         }
3128     }
3129     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3130
3131     error = n_packets = n_errors = 0;
3132     LIST_FOR_EACH_POP (pkt_node, list_node, &packets) {
3133         int ret;
3134
3135         ret = ofproto_dpif_send_packet(pkt_node->port, pkt_node->pkt);
3136         dp_packet_delete(pkt_node->pkt);
3137         free(pkt_node);
3138         if (ret) {
3139             error = ret;
3140             n_errors++;
3141         }
3142         n_packets++;
3143     }
3144
3145     if (n_errors) {
3146         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3147         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3148                      "packets, last error was: %s",
3149                      bundle->name, n_errors, n_packets, ovs_strerror(error));
3150     } else {
3151         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3152                  bundle->name, n_packets);
3153     }
3154 }
3155
3156 static void
3157 bundle_run(struct ofbundle *bundle)
3158 {
3159     if (bundle->lacp) {
3160         lacp_run(bundle->lacp, send_pdu_cb);
3161     }
3162     if (bundle->bond) {
3163         struct ofport_dpif *port;
3164
3165         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
3166             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
3167         }
3168
3169         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
3170             bundle->ofproto->backer->need_revalidate = REV_BOND;
3171         }
3172
3173         if (bond_should_send_learning_packets(bundle->bond)) {
3174             bundle_send_learning_packets(bundle);
3175         }
3176     }
3177 }
3178
3179 static void
3180 bundle_wait(struct ofbundle *bundle)
3181 {
3182     if (bundle->lacp) {
3183         lacp_wait(bundle->lacp);
3184     }
3185     if (bundle->bond) {
3186         bond_wait(bundle->bond);
3187     }
3188 }
3189 \f
3190 /* Mirrors. */
3191
3192 static int
3193 mirror_set__(struct ofproto *ofproto_, void *aux,
3194              const struct ofproto_mirror_settings *s)
3195 {
3196     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3197     struct ofbundle **srcs, **dsts;
3198     int error;
3199     size_t i;
3200
3201     if (!s) {
3202         mirror_destroy(ofproto->mbridge, aux);
3203         return 0;
3204     }
3205
3206     srcs = xmalloc(s->n_srcs * sizeof *srcs);
3207     dsts = xmalloc(s->n_dsts * sizeof *dsts);
3208
3209     for (i = 0; i < s->n_srcs; i++) {
3210         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
3211     }
3212
3213     for (i = 0; i < s->n_dsts; i++) {
3214         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
3215     }
3216
3217     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
3218                        s->n_dsts, s->src_vlans,
3219                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
3220     free(srcs);
3221     free(dsts);
3222     return error;
3223 }
3224
3225 static int
3226 mirror_get_stats__(struct ofproto *ofproto, void *aux,
3227                    uint64_t *packets, uint64_t *bytes)
3228 {
3229     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
3230                             bytes);
3231 }
3232
3233 static int
3234 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3235 {
3236     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3237     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3238     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3239         mac_learning_flush(ofproto->ml);
3240     }
3241     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3242     return 0;
3243 }
3244
3245 static bool
3246 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3247 {
3248     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3249     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3250     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
3251 }
3252
3253 static void
3254 forward_bpdu_changed(struct ofproto *ofproto_)
3255 {
3256     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3257     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3258 }
3259
3260 static void
3261 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3262                      size_t max_entries)
3263 {
3264     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3265     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3266     mac_learning_set_idle_time(ofproto->ml, idle_time);
3267     mac_learning_set_max_entries(ofproto->ml, max_entries);
3268     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3269 }
3270
3271 /* Configures multicast snooping on 'ofport' using the settings
3272  * defined in 's'. */
3273 static int
3274 set_mcast_snooping(struct ofproto *ofproto_,
3275                    const struct ofproto_mcast_snooping_settings *s)
3276 {
3277     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3278
3279     /* Only revalidate flows if the configuration changed. */
3280     if (!s != !ofproto->ms) {
3281         ofproto->backer->need_revalidate = REV_RECONFIGURE;
3282     }
3283
3284     if (s) {
3285         if (!ofproto->ms) {
3286             ofproto->ms = mcast_snooping_create();
3287         }
3288
3289         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3290         mcast_snooping_set_idle_time(ofproto->ms, s->idle_time);
3291         mcast_snooping_set_max_entries(ofproto->ms, s->max_entries);
3292         if (mcast_snooping_set_flood_unreg(ofproto->ms, s->flood_unreg)) {
3293             ofproto->backer->need_revalidate = REV_RECONFIGURE;
3294         }
3295         ovs_rwlock_unlock(&ofproto->ms->rwlock);
3296     } else {
3297         mcast_snooping_unref(ofproto->ms);
3298         ofproto->ms = NULL;
3299     }
3300
3301     return 0;
3302 }
3303
3304 /* Configures multicast snooping port's flood settings on 'ofproto'. */
3305 static int
3306 set_mcast_snooping_port(struct ofproto *ofproto_, void *aux,
3307                         const struct ofproto_mcast_snooping_port_settings *s)
3308 {
3309     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3310     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3311
3312     if (ofproto->ms && s) {
3313         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3314         mcast_snooping_set_port_flood(ofproto->ms, bundle, s->flood);
3315         mcast_snooping_set_port_flood_reports(ofproto->ms, bundle,
3316                                               s->flood_reports);
3317         ovs_rwlock_unlock(&ofproto->ms->rwlock);
3318     }
3319     return 0;
3320 }
3321
3322 \f
3323 /* Ports. */
3324
3325 struct ofport_dpif *
3326 ofp_port_to_ofport(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
3327 {
3328     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3329     return ofport ? ofport_dpif_cast(ofport) : NULL;
3330 }
3331
3332 static void
3333 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3334                             struct ofproto_port *ofproto_port,
3335                             struct dpif_port *dpif_port)
3336 {
3337     ofproto_port->name = dpif_port->name;
3338     ofproto_port->type = dpif_port->type;
3339     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3340 }
3341
3342 static void
3343 ofport_update_peer(struct ofport_dpif *ofport)
3344 {
3345     const struct ofproto_dpif *ofproto;
3346     struct dpif_backer *backer;
3347     char *peer_name;
3348
3349     if (!netdev_vport_is_patch(ofport->up.netdev)) {
3350         return;
3351     }
3352
3353     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
3354     backer->need_revalidate = REV_RECONFIGURE;
3355
3356     if (ofport->peer) {
3357         ofport->peer->peer = NULL;
3358         ofport->peer = NULL;
3359     }
3360
3361     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
3362     if (!peer_name) {
3363         return;
3364     }
3365
3366     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3367         struct ofport *peer_ofport;
3368         struct ofport_dpif *peer;
3369         char *peer_peer;
3370
3371         if (ofproto->backer != backer) {
3372             continue;
3373         }
3374
3375         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
3376         if (!peer_ofport) {
3377             continue;
3378         }
3379
3380         peer = ofport_dpif_cast(peer_ofport);
3381         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
3382         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
3383                                  peer_peer)) {
3384             ofport->peer = peer;
3385             ofport->peer->peer = ofport;
3386         }
3387         free(peer_peer);
3388
3389         break;
3390     }
3391     free(peer_name);
3392 }
3393
3394 static void
3395 port_run(struct ofport_dpif *ofport)
3396 {
3397     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3398     bool carrier_changed = carrier_seq != ofport->carrier_seq;
3399     bool enable = netdev_get_carrier(ofport->up.netdev);
3400     bool cfm_enable = false;
3401     bool bfd_enable = false;
3402
3403     ofport->carrier_seq = carrier_seq;
3404
3405     if (ofport->cfm) {
3406         int cfm_opup = cfm_get_opup(ofport->cfm);
3407
3408         cfm_enable = !cfm_get_fault(ofport->cfm);
3409
3410         if (cfm_opup >= 0) {
3411             cfm_enable = cfm_enable && cfm_opup;
3412         }
3413     }
3414
3415     if (ofport->bfd) {
3416         bfd_enable = bfd_forwarding(ofport->bfd);
3417     }
3418
3419     if (ofport->bfd || ofport->cfm) {
3420         enable = enable && (cfm_enable || bfd_enable);
3421     }
3422
3423     if (ofport->bundle) {
3424         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3425         if (carrier_changed) {
3426             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3427         }
3428     }
3429
3430     if (ofport->may_enable != enable) {
3431         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3432
3433         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3434
3435         if (ofport->rstp_port) {
3436             rstp_port_set_mac_operational(ofport->rstp_port, enable);
3437         }
3438     }
3439
3440     ofport->may_enable = enable;
3441 }
3442
3443 static int
3444 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3445                    struct ofproto_port *ofproto_port)
3446 {
3447     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3448     struct dpif_port dpif_port;
3449     int error;
3450
3451     if (sset_contains(&ofproto->ghost_ports, devname)) {
3452         const char *type = netdev_get_type_from_name(devname);
3453
3454         /* We may be called before ofproto->up.port_by_name is populated with
3455          * the appropriate ofport.  For this reason, we must get the name and
3456          * type from the netdev layer directly. */
3457         if (type) {
3458             const struct ofport *ofport;
3459
3460             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3461             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3462             ofproto_port->name = xstrdup(devname);
3463             ofproto_port->type = xstrdup(type);
3464             return 0;
3465         }
3466         return ENODEV;
3467     }
3468
3469     if (!sset_contains(&ofproto->ports, devname)) {
3470         return ENODEV;
3471     }
3472     error = dpif_port_query_by_name(ofproto->backer->dpif,
3473                                     devname, &dpif_port);
3474     if (!error) {
3475         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3476     }
3477     return error;
3478 }
3479
3480 static int
3481 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3482 {
3483     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3484     const char *devname = netdev_get_name(netdev);
3485     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3486     const char *dp_port_name;
3487
3488     if (netdev_vport_is_patch(netdev)) {
3489         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3490         return 0;
3491     }
3492
3493     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3494     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3495         odp_port_t port_no = ODPP_NONE;
3496         int error;
3497
3498         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3499         if (error) {
3500             return error;
3501         }
3502         if (netdev_get_tunnel_config(netdev)) {
3503             simap_put(&ofproto->backer->tnl_backers,
3504                       dp_port_name, odp_to_u32(port_no));
3505         }
3506     }
3507
3508     if (netdev_get_tunnel_config(netdev)) {
3509         sset_add(&ofproto->ghost_ports, devname);
3510     } else {
3511         sset_add(&ofproto->ports, devname);
3512     }
3513     return 0;
3514 }
3515
3516 static int
3517 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3518 {
3519     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3520     struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
3521     int error = 0;
3522
3523     if (!ofport) {
3524         return 0;
3525     }
3526
3527     sset_find_and_delete(&ofproto->ghost_ports,
3528                          netdev_get_name(ofport->up.netdev));
3529     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3530     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3531         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3532         if (!error) {
3533             /* The caller is going to close ofport->up.netdev.  If this is a
3534              * bonded port, then the bond is using that netdev, so remove it
3535              * from the bond.  The client will need to reconfigure everything
3536              * after deleting ports, so then the slave will get re-added. */
3537             bundle_remove(&ofport->up);
3538         }
3539     }
3540     return error;
3541 }
3542
3543 static int
3544 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3545 {
3546     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3547     int error;
3548
3549     error = netdev_get_stats(ofport->up.netdev, stats);
3550
3551     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3552         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3553
3554         ovs_mutex_lock(&ofproto->stats_mutex);
3555         /* ofproto->stats.tx_packets represents packets that we created
3556          * internally and sent to some port (e.g. packets sent with
3557          * ofproto_dpif_send_packet()).  Account for them as if they had
3558          * come from OFPP_LOCAL and got forwarded. */
3559
3560         if (stats->rx_packets != UINT64_MAX) {
3561             stats->rx_packets += ofproto->stats.tx_packets;
3562         }
3563
3564         if (stats->rx_bytes != UINT64_MAX) {
3565             stats->rx_bytes += ofproto->stats.tx_bytes;
3566         }
3567
3568         /* ofproto->stats.rx_packets represents packets that were received on
3569          * some port and we processed internally and dropped (e.g. STP).
3570          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3571
3572         if (stats->tx_packets != UINT64_MAX) {
3573             stats->tx_packets += ofproto->stats.rx_packets;
3574         }
3575
3576         if (stats->tx_bytes != UINT64_MAX) {
3577             stats->tx_bytes += ofproto->stats.rx_bytes;
3578         }
3579         ovs_mutex_unlock(&ofproto->stats_mutex);
3580     }
3581
3582     return error;
3583 }
3584
3585 static int
3586 port_get_lacp_stats(const struct ofport *ofport_, struct lacp_slave_stats *stats)
3587 {
3588     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3589     if (ofport->bundle && ofport->bundle->lacp) {
3590         if (lacp_get_slave_stats(ofport->bundle->lacp, ofport, stats)) {
3591             return 0;
3592         }
3593     }
3594     return -1;
3595 }
3596
3597 struct port_dump_state {
3598     struct sset_position pos;
3599     bool ghost;
3600
3601     struct ofproto_port port;
3602     bool has_port;
3603 };
3604
3605 static int
3606 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3607 {
3608     *statep = xzalloc(sizeof(struct port_dump_state));
3609     return 0;
3610 }
3611
3612 static int
3613 port_dump_next(const struct ofproto *ofproto_, void *state_,
3614                struct ofproto_port *port)
3615 {
3616     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3617     struct port_dump_state *state = state_;
3618     const struct sset *sset;
3619     struct sset_node *node;
3620
3621     if (state->has_port) {
3622         ofproto_port_destroy(&state->port);
3623         state->has_port = false;
3624     }
3625     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3626     while ((node = sset_at_position(sset, &state->pos))) {
3627         int error;
3628
3629         error = port_query_by_name(ofproto_, node->name, &state->port);
3630         if (!error) {
3631             *port = state->port;
3632             state->has_port = true;
3633             return 0;
3634         } else if (error != ENODEV) {
3635             return error;
3636         }
3637     }
3638
3639     if (!state->ghost) {
3640         state->ghost = true;
3641         memset(&state->pos, 0, sizeof state->pos);
3642         return port_dump_next(ofproto_, state_, port);
3643     }
3644
3645     return EOF;
3646 }
3647
3648 static int
3649 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3650 {
3651     struct port_dump_state *state = state_;
3652
3653     if (state->has_port) {
3654         ofproto_port_destroy(&state->port);
3655     }
3656     free(state);
3657     return 0;
3658 }
3659
3660 static int
3661 port_poll(const struct ofproto *ofproto_, char **devnamep)
3662 {
3663     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3664
3665     if (ofproto->port_poll_errno) {
3666         int error = ofproto->port_poll_errno;
3667         ofproto->port_poll_errno = 0;
3668         return error;
3669     }
3670
3671     if (sset_is_empty(&ofproto->port_poll_set)) {
3672         return EAGAIN;
3673     }
3674
3675     *devnamep = sset_pop(&ofproto->port_poll_set);
3676     return 0;
3677 }
3678
3679 static void
3680 port_poll_wait(const struct ofproto *ofproto_)
3681 {
3682     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3683     dpif_port_poll_wait(ofproto->backer->dpif);
3684 }
3685
3686 static int
3687 port_is_lacp_current(const struct ofport *ofport_)
3688 {
3689     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3690     return (ofport->bundle && ofport->bundle->lacp
3691             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3692             : -1);
3693 }
3694 \f
3695 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3696  * then delete it entirely. */
3697 static void
3698 rule_expire(struct rule_dpif *rule, long long now)
3699     OVS_REQUIRES(ofproto_mutex)
3700 {
3701     uint16_t hard_timeout, idle_timeout;
3702     int reason = -1;
3703
3704     hard_timeout = rule->up.hard_timeout;
3705     idle_timeout = rule->up.idle_timeout;
3706
3707     /* Has 'rule' expired? */
3708     if (hard_timeout) {
3709         long long int modified;
3710
3711         ovs_mutex_lock(&rule->up.mutex);
3712         modified = rule->up.modified;
3713         ovs_mutex_unlock(&rule->up.mutex);
3714
3715         if (now > modified + hard_timeout * 1000) {
3716             reason = OFPRR_HARD_TIMEOUT;
3717         }
3718     }
3719
3720     if (reason < 0 && idle_timeout) {
3721         long long int used;
3722
3723         ovs_mutex_lock(&rule->stats_mutex);
3724         used = rule->stats.used;
3725         ovs_mutex_unlock(&rule->stats_mutex);
3726
3727         if (now > used + idle_timeout * 1000) {
3728             reason = OFPRR_IDLE_TIMEOUT;
3729         }
3730     }
3731
3732     if (reason >= 0) {
3733         COVERAGE_INC(ofproto_dpif_expired);
3734         ofproto_rule_expire(&rule->up, reason);
3735     }
3736 }
3737
3738 static void
3739 ofproto_dpif_set_packet_odp_port(const struct ofproto_dpif *ofproto,
3740                                  ofp_port_t in_port, struct dp_packet *packet)
3741 {
3742     if (in_port == OFPP_NONE) {
3743         in_port = OFPP_LOCAL;
3744     }
3745     packet->md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3746 }
3747
3748 int
3749 ofproto_dpif_execute_actions__(struct ofproto_dpif *ofproto,
3750                                const struct flow *flow,
3751                                struct rule_dpif *rule,
3752                                const struct ofpact *ofpacts, size_t ofpacts_len,
3753                                int indentation, int depth, int resubmits,
3754                                struct dp_packet *packet)
3755 {
3756     struct dpif_flow_stats stats;
3757     struct xlate_out xout;
3758     struct xlate_in xin;
3759     struct dpif_execute execute;
3760     int error;
3761
3762     ovs_assert((rule != NULL) != (ofpacts != NULL));
3763
3764     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3765
3766     if (rule) {
3767         rule_dpif_credit_stats(rule, &stats);
3768     }
3769
3770     uint64_t odp_actions_stub[1024 / 8];
3771     struct ofpbuf odp_actions = OFPBUF_STUB_INITIALIZER(odp_actions_stub);
3772     xlate_in_init(&xin, ofproto, flow, flow->in_port.ofp_port, rule,
3773                   stats.tcp_flags, packet, NULL, &odp_actions);
3774     xin.ofpacts = ofpacts;
3775     xin.ofpacts_len = ofpacts_len;
3776     xin.resubmit_stats = &stats;
3777     xin.indentation = indentation;
3778     xin.depth = depth;
3779     xin.resubmits = resubmits;
3780     if (xlate_actions(&xin, &xout) != XLATE_OK) {
3781         error = EINVAL;
3782         goto out;
3783     }
3784
3785     execute.actions = odp_actions.data;
3786     execute.actions_len = odp_actions.size;
3787
3788     pkt_metadata_from_flow(&packet->md, flow);
3789     execute.packet = packet;
3790     execute.flow = flow;
3791     execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3792     execute.probe = false;
3793     execute.mtu = 0;
3794
3795     /* Fix up in_port. */
3796     ofproto_dpif_set_packet_odp_port(ofproto, flow->in_port.ofp_port, packet);
3797
3798     error = dpif_execute(ofproto->backer->dpif, &execute);
3799 out:
3800     xlate_out_uninit(&xout);
3801     ofpbuf_uninit(&odp_actions);
3802
3803     return error;
3804 }
3805
3806 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3807  * 'flow' must reflect the data in 'packet'. */
3808 int
3809 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3810                              const struct flow *flow,
3811                              struct rule_dpif *rule,
3812                              const struct ofpact *ofpacts, size_t ofpacts_len,
3813                              struct dp_packet *packet)
3814 {
3815     return ofproto_dpif_execute_actions__(ofproto, flow, rule, ofpacts,
3816                                           ofpacts_len, 0, 0, 0, packet);
3817 }
3818
3819 void
3820 rule_dpif_credit_stats(struct rule_dpif *rule,
3821                        const struct dpif_flow_stats *stats)
3822 {
3823     ovs_mutex_lock(&rule->stats_mutex);
3824     if (OVS_UNLIKELY(rule->new_rule)) {
3825         rule_dpif_credit_stats(rule->new_rule, stats);
3826     } else {
3827         rule->stats.n_packets += stats->n_packets;
3828         rule->stats.n_bytes += stats->n_bytes;
3829         rule->stats.used = MAX(rule->stats.used, stats->used);
3830     }
3831     ovs_mutex_unlock(&rule->stats_mutex);
3832 }
3833
3834 ovs_be64
3835 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3836     OVS_REQUIRES(rule->up.mutex)
3837 {
3838     return rule->up.flow_cookie;
3839 }
3840
3841 void
3842 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3843                      uint16_t hard_timeout)
3844 {
3845     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3846 }
3847
3848 /* Returns 'rule''s actions.  The returned actions are RCU-protected, and can
3849  * be read until the calling thread quiesces. */
3850 const struct rule_actions *
3851 rule_dpif_get_actions(const struct rule_dpif *rule)
3852 {
3853     return rule_get_actions(&rule->up);
3854 }
3855
3856 /* Sets 'rule''s recirculation id. */
3857 static void
3858 rule_dpif_set_recirc_id(struct rule_dpif *rule, uint32_t id)
3859     OVS_REQUIRES(rule->up.mutex)
3860 {
3861     ovs_assert(!rule->recirc_id || rule->recirc_id == id);
3862     if (rule->recirc_id == id) {
3863         /* Release the new reference to the same id. */
3864         recirc_free_id(id);
3865     } else {
3866         rule->recirc_id = id;
3867     }
3868 }
3869
3870 /* Sets 'rule''s recirculation id. */
3871 void
3872 rule_set_recirc_id(struct rule *rule_, uint32_t id)
3873 {
3874     struct rule_dpif *rule = rule_dpif_cast(rule_);
3875
3876     ovs_mutex_lock(&rule->up.mutex);
3877     rule_dpif_set_recirc_id(rule, id);
3878     ovs_mutex_unlock(&rule->up.mutex);
3879 }
3880
3881 cls_version_t
3882 ofproto_dpif_get_tables_version(struct ofproto_dpif *ofproto OVS_UNUSED)
3883 {
3884     cls_version_t version;
3885
3886     atomic_read_relaxed(&ofproto->tables_version, &version);
3887
3888     return version;
3889 }
3890
3891 /* The returned rule (if any) is valid at least until the next RCU quiescent
3892  * period.  If the rule needs to stay around longer, the caller should take
3893  * a reference.
3894  *
3895  * 'flow' is non-const to allow for temporary modifications during the lookup.
3896  * Any changes are restored before returning. */
3897 static struct rule_dpif *
3898 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, cls_version_t version,
3899                           uint8_t table_id, struct flow *flow,
3900                           struct flow_wildcards *wc)
3901 {
3902     struct classifier *cls = &ofproto->up.tables[table_id].cls;
3903     return rule_dpif_cast(rule_from_cls_rule(classifier_lookup(cls, version,
3904                                                                flow, wc)));
3905 }
3906
3907 /* Look up 'flow' in 'ofproto''s classifier version 'version', starting from
3908  * table '*table_id'.  Returns the rule that was found, which may be one of the
3909  * special rules according to packet miss hadling.  If 'may_packet_in' is
3910  * false, returning of the miss_rule (which issues packet ins for the
3911  * controller) is avoided.  Updates 'wc', if nonnull, to reflect the fields
3912  * that were used during the lookup.
3913  *
3914  * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3915  * if none is found then the table miss configuration for that table is
3916  * honored, which can result in additional lookups in other OpenFlow tables.
3917  * In this case the function updates '*table_id' to reflect the final OpenFlow
3918  * table that was searched.
3919  *
3920  * If 'honor_table_miss' is false, then only one table lookup occurs, in
3921  * '*table_id'.
3922  *
3923  * The rule is returned in '*rule', which is valid at least until the next
3924  * RCU quiescent period.  If the '*rule' needs to stay around longer, the
3925  * caller must take a reference.
3926  *
3927  * 'in_port' allows the lookup to take place as if the in port had the value
3928  * 'in_port'.  This is needed for resubmit action support.
3929  *
3930  * 'flow' is non-const to allow for temporary modifications during the lookup.
3931  * Any changes are restored before returning. */
3932 struct rule_dpif *
3933 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3934                             cls_version_t version, struct flow *flow,
3935                             struct flow_wildcards *wc,
3936                             const struct dpif_flow_stats *stats,
3937                             uint8_t *table_id, ofp_port_t in_port,
3938                             bool may_packet_in, bool honor_table_miss)
3939 {
3940     ovs_be16 old_tp_src = flow->tp_src, old_tp_dst = flow->tp_dst;
3941     ofp_port_t old_in_port = flow->in_port.ofp_port;
3942     enum ofputil_table_miss miss_config;
3943     struct rule_dpif *rule;
3944     uint8_t next_id;
3945
3946     /* We always unwildcard nw_frag (for IP), so they
3947      * need not be unwildcarded here. */
3948     if (flow->nw_frag & FLOW_NW_FRAG_ANY
3949         && ofproto->up.frag_handling != OFPUTIL_FRAG_NX_MATCH) {
3950         if (ofproto->up.frag_handling == OFPUTIL_FRAG_NORMAL) {
3951             /* We must pretend that transport ports are unavailable. */
3952             flow->tp_src = htons(0);
3953             flow->tp_dst = htons(0);
3954         } else {
3955             /* Must be OFPUTIL_FRAG_DROP (we don't have OFPUTIL_FRAG_REASM).
3956              * Use the drop_frags_rule (which cannot disappear). */
3957             rule = ofproto->drop_frags_rule;
3958             if (stats) {
3959                 struct oftable *tbl = &ofproto->up.tables[*table_id];
3960                 unsigned long orig;
3961
3962                 atomic_add_relaxed(&tbl->n_matched, stats->n_packets, &orig);
3963             }
3964             return rule;
3965         }
3966     }
3967
3968     /* Look up a flow with 'in_port' as the input port.  Then restore the
3969      * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
3970      * have surprising behavior). */
3971     flow->in_port.ofp_port = in_port;
3972
3973     /* Our current implementation depends on n_tables == N_TABLES, and
3974      * TBL_INTERNAL being the last table. */
3975     BUILD_ASSERT_DECL(N_TABLES == TBL_INTERNAL + 1);
3976
3977     miss_config = OFPUTIL_TABLE_MISS_CONTINUE;
3978
3979     for (next_id = *table_id;
3980          next_id < ofproto->up.n_tables;
3981          next_id++, next_id += (next_id == TBL_INTERNAL))
3982     {
3983         *table_id = next_id;
3984         rule = rule_dpif_lookup_in_table(ofproto, version, next_id, flow, wc);
3985         if (stats) {
3986             struct oftable *tbl = &ofproto->up.tables[next_id];
3987             unsigned long orig;
3988
3989             atomic_add_relaxed(rule ? &tbl->n_matched : &tbl->n_missed,
3990                                stats->n_packets, &orig);
3991         }
3992         if (rule) {
3993             goto out;   /* Match. */
3994         }
3995         if (honor_table_miss) {
3996             miss_config = ofproto_table_get_miss_config(&ofproto->up,
3997                                                         *table_id);
3998             if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE) {
3999                 continue;
4000             }
4001         }
4002         break;
4003     }
4004     /* Miss. */
4005     rule = ofproto->no_packet_in_rule;
4006     if (may_packet_in) {
4007         if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE
4008             || miss_config == OFPUTIL_TABLE_MISS_CONTROLLER) {
4009             struct ofport_dpif *port;
4010
4011             port = ofp_port_to_ofport(ofproto, old_in_port);
4012             if (!port) {
4013                 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4014                              old_in_port);
4015             } else if (!(port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN)) {
4016                 rule = ofproto->miss_rule;
4017             }
4018         } else if (miss_config == OFPUTIL_TABLE_MISS_DEFAULT &&
4019                    connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
4020             rule = ofproto->miss_rule;
4021         }
4022     }
4023 out:
4024     /* Restore port numbers, as they may have been modified above. */
4025     flow->tp_src = old_tp_src;
4026     flow->tp_dst = old_tp_dst;
4027     /* Restore the old in port. */
4028     flow->in_port.ofp_port = old_in_port;
4029
4030     return rule;
4031 }
4032
4033 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4034 {
4035     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4036 }
4037
4038 static struct rule *
4039 rule_alloc(void)
4040 {
4041     struct rule_dpif *rule = xzalloc(sizeof *rule);
4042     return &rule->up;
4043 }
4044
4045 static void
4046 rule_dealloc(struct rule *rule_)
4047 {
4048     struct rule_dpif *rule = rule_dpif_cast(rule_);
4049     free(rule);
4050 }
4051
4052 static enum ofperr
4053 check_mask(struct ofproto_dpif *ofproto, const struct miniflow *flow)
4054 {
4055     const struct odp_support *support;
4056     uint16_t ct_state, ct_zone;
4057     ovs_u128 ct_label;
4058     uint32_t ct_mark;
4059
4060     support = &ofproto_dpif_get_support(ofproto)->odp;
4061     ct_state = MINIFLOW_GET_U16(flow, ct_state);
4062     if (support->ct_state && support->ct_zone && support->ct_mark
4063         && support->ct_label && support->ct_state_nat) {
4064         return ct_state & CS_UNSUPPORTED_MASK ? OFPERR_OFPBMC_BAD_MASK : 0;
4065     }
4066
4067     ct_zone = MINIFLOW_GET_U16(flow, ct_zone);
4068     ct_mark = MINIFLOW_GET_U32(flow, ct_mark);
4069     ct_label = MINIFLOW_GET_U128(flow, ct_label);
4070
4071     if ((ct_state && !support->ct_state)
4072         || (ct_state & CS_UNSUPPORTED_MASK)
4073         || ((ct_state & (CS_SRC_NAT | CS_DST_NAT)) && !support->ct_state_nat)
4074         || (ct_zone && !support->ct_zone)
4075         || (ct_mark && !support->ct_mark)
4076         || (!ovs_u128_is_zero(ct_label) && !support->ct_label)) {
4077         return OFPERR_OFPBMC_BAD_MASK;
4078     }
4079
4080     return 0;
4081 }
4082
4083 static enum ofperr
4084 check_actions(const struct ofproto_dpif *ofproto,
4085               const struct rule_actions *const actions)
4086 {
4087     const struct ofpact *ofpact;
4088
4089     OFPACT_FOR_EACH (ofpact, actions->ofpacts, actions->ofpacts_len) {
4090         const struct odp_support *support;
4091         const struct ofpact_conntrack *ct;
4092         const struct ofpact *a;
4093
4094         if (ofpact->type != OFPACT_CT) {
4095             continue;
4096         }
4097
4098         ct = CONTAINER_OF(ofpact, struct ofpact_conntrack, ofpact);
4099         support = &ofproto_dpif_get_support(ofproto)->odp;
4100
4101         if (!support->ct_state) {
4102             return OFPERR_OFPBAC_BAD_TYPE;
4103         }
4104         if ((ct->zone_imm || ct->zone_src.field) && !support->ct_zone) {
4105             return OFPERR_OFPBAC_BAD_ARGUMENT;
4106         }
4107
4108         OFPACT_FOR_EACH(a, ct->actions, ofpact_ct_get_action_len(ct)) {
4109             const struct mf_field *dst = ofpact_get_mf_dst(a);
4110
4111             if (a->type == OFPACT_NAT && !support->ct_state_nat) {
4112                 /* The backer doesn't seem to support the NAT bits in
4113                  * 'ct_state': assume that it doesn't support the NAT
4114                  * action. */
4115                 return OFPERR_OFPBAC_BAD_TYPE;
4116             }
4117             if (dst && ((dst->id == MFF_CT_MARK && !support->ct_mark)
4118                         || (dst->id == MFF_CT_LABEL && !support->ct_label))) {
4119                 return OFPERR_OFPBAC_BAD_SET_ARGUMENT;
4120             }
4121         }
4122     }
4123
4124     return 0;
4125 }
4126
4127 static enum ofperr
4128 rule_check(struct rule *rule)
4129 {
4130     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->ofproto);
4131     enum ofperr err;
4132
4133     err = check_mask(ofproto, &rule->cr.match.mask->masks);
4134     if (err) {
4135         return err;
4136     }
4137     return check_actions(ofproto, rule->actions);
4138 }
4139
4140 static enum ofperr
4141 rule_construct(struct rule *rule_)
4142     OVS_NO_THREAD_SAFETY_ANALYSIS
4143 {
4144     struct rule_dpif *rule = rule_dpif_cast(rule_);
4145     int error;
4146
4147     error = rule_check(rule_);
4148     if (error) {
4149         return error;
4150     }
4151
4152     ovs_mutex_init_adaptive(&rule->stats_mutex);
4153     rule->stats.n_packets = 0;
4154     rule->stats.n_bytes = 0;
4155     rule->stats.used = rule->up.modified;
4156     rule->recirc_id = 0;
4157     rule->new_rule = NULL;
4158
4159     return 0;
4160 }
4161
4162 static void
4163 rule_insert(struct rule *rule_, struct rule *old_rule_, bool forward_stats)
4164     OVS_REQUIRES(ofproto_mutex)
4165 {
4166     struct rule_dpif *rule = rule_dpif_cast(rule_);
4167
4168     if (old_rule_ && forward_stats) {
4169         struct rule_dpif *old_rule = rule_dpif_cast(old_rule_);
4170
4171         ovs_assert(!old_rule->new_rule);
4172
4173         /* Take a reference to the new rule, and refer all stats updates from
4174          * the old rule to the new rule. */
4175         rule_dpif_ref(rule);
4176
4177         ovs_mutex_lock(&old_rule->stats_mutex);
4178         ovs_mutex_lock(&rule->stats_mutex);
4179         old_rule->new_rule = rule;       /* Forward future stats. */
4180         rule->stats = old_rule->stats;   /* Transfer stats to the new rule. */
4181         ovs_mutex_unlock(&rule->stats_mutex);
4182         ovs_mutex_unlock(&old_rule->stats_mutex);
4183     }
4184 }
4185
4186 static void
4187 rule_destruct(struct rule *rule_)
4188     OVS_NO_THREAD_SAFETY_ANALYSIS
4189 {
4190     struct rule_dpif *rule = rule_dpif_cast(rule_);
4191
4192     ovs_mutex_destroy(&rule->stats_mutex);
4193     /* Release reference to the new rule, if any. */
4194     if (rule->new_rule) {
4195         rule_dpif_unref(rule->new_rule);
4196     }
4197     if (rule->recirc_id) {
4198         recirc_free_id(rule->recirc_id);
4199     }
4200 }
4201
4202 static void
4203 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
4204                long long int *used)
4205 {
4206     struct rule_dpif *rule = rule_dpif_cast(rule_);
4207
4208     ovs_mutex_lock(&rule->stats_mutex);
4209     if (OVS_UNLIKELY(rule->new_rule)) {
4210         rule_get_stats(&rule->new_rule->up, packets, bytes, used);
4211     } else {
4212         *packets = rule->stats.n_packets;
4213         *bytes = rule->stats.n_bytes;
4214         *used = rule->stats.used;
4215     }
4216     ovs_mutex_unlock(&rule->stats_mutex);
4217 }
4218
4219 static void
4220 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4221                   struct dp_packet *packet)
4222 {
4223     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4224
4225     ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
4226 }
4227
4228 static enum ofperr
4229 rule_execute(struct rule *rule, const struct flow *flow,
4230              struct dp_packet *packet)
4231 {
4232     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4233     dp_packet_delete(packet);
4234     return 0;
4235 }
4236
4237 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4238 {
4239     return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4240 }
4241
4242 static struct ofgroup *
4243 group_alloc(void)
4244 {
4245     struct group_dpif *group = xzalloc(sizeof *group);
4246     return &group->up;
4247 }
4248
4249 static void
4250 group_dealloc(struct ofgroup *group_)
4251 {
4252     struct group_dpif *group = group_dpif_cast(group_);
4253     free(group);
4254 }
4255
4256 static void
4257 group_construct_stats(struct group_dpif *group)
4258     OVS_REQUIRES(group->stats_mutex)
4259 {
4260     struct ofputil_bucket *bucket;
4261     const struct ovs_list *buckets;
4262
4263     group->packet_count = 0;
4264     group->byte_count = 0;
4265
4266     group_dpif_get_buckets(group, &buckets);
4267     LIST_FOR_EACH (bucket, list_node, buckets) {
4268         bucket->stats.packet_count = 0;
4269         bucket->stats.byte_count = 0;
4270     }
4271 }
4272
4273 void
4274 group_dpif_credit_stats(struct group_dpif *group,
4275                         struct ofputil_bucket *bucket,
4276                         const struct dpif_flow_stats *stats)
4277 {
4278     ovs_mutex_lock(&group->stats_mutex);
4279     group->packet_count += stats->n_packets;
4280     group->byte_count += stats->n_bytes;
4281     if (bucket) {
4282         bucket->stats.packet_count += stats->n_packets;
4283         bucket->stats.byte_count += stats->n_bytes;
4284     } else { /* Credit to all buckets */
4285         const struct ovs_list *buckets;
4286
4287         group_dpif_get_buckets(group, &buckets);
4288         LIST_FOR_EACH (bucket, list_node, buckets) {
4289             bucket->stats.packet_count += stats->n_packets;
4290             bucket->stats.byte_count += stats->n_bytes;
4291         }
4292     }
4293     ovs_mutex_unlock(&group->stats_mutex);
4294 }
4295
4296 static enum ofperr
4297 group_construct(struct ofgroup *group_)
4298 {
4299     struct group_dpif *group = group_dpif_cast(group_);
4300
4301     ovs_mutex_init_adaptive(&group->stats_mutex);
4302     ovs_mutex_lock(&group->stats_mutex);
4303     group_construct_stats(group);
4304     ovs_mutex_unlock(&group->stats_mutex);
4305     return 0;
4306 }
4307
4308 static void
4309 group_destruct(struct ofgroup *group_)
4310 {
4311     struct group_dpif *group = group_dpif_cast(group_);
4312     ovs_mutex_destroy(&group->stats_mutex);
4313 }
4314
4315 static enum ofperr
4316 group_modify(struct ofgroup *group_)
4317 {
4318     struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
4319
4320     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4321
4322     return 0;
4323 }
4324
4325 static enum ofperr
4326 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4327 {
4328     struct group_dpif *group = group_dpif_cast(group_);
4329     struct ofputil_bucket *bucket;
4330     const struct ovs_list *buckets;
4331     struct bucket_counter *bucket_stats;
4332
4333     ovs_mutex_lock(&group->stats_mutex);
4334     ogs->packet_count = group->packet_count;
4335     ogs->byte_count = group->byte_count;
4336
4337     group_dpif_get_buckets(group, &buckets);
4338     bucket_stats = ogs->bucket_stats;
4339     LIST_FOR_EACH (bucket, list_node, buckets) {
4340         bucket_stats->packet_count = bucket->stats.packet_count;
4341         bucket_stats->byte_count = bucket->stats.byte_count;
4342         bucket_stats++;
4343     }
4344     ovs_mutex_unlock(&group->stats_mutex);
4345
4346     return 0;
4347 }
4348
4349 /* If the group exists, this function increments the groups's reference count.
4350  *
4351  * Make sure to call group_dpif_unref() after no longer needing to maintain
4352  * a reference to the group. */
4353 bool
4354 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4355                   struct group_dpif **group)
4356 {
4357     struct ofgroup *ofgroup;
4358     bool found;
4359
4360     found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4361     *group = found ?  group_dpif_cast(ofgroup) : NULL;
4362
4363     return found;
4364 }
4365
4366 void
4367 group_dpif_get_buckets(const struct group_dpif *group,
4368                        const struct ovs_list **buckets)
4369 {
4370     *buckets = &group->up.buckets;
4371 }
4372
4373 enum ofp11_group_type
4374 group_dpif_get_type(const struct group_dpif *group)
4375 {
4376     return group->up.type;
4377 }
4378
4379 const char *
4380 group_dpif_get_selection_method(const struct group_dpif *group)
4381 {
4382     return group->up.props.selection_method;
4383 }
4384 \f
4385 /* Sends 'packet' out 'ofport'.
4386  * May modify 'packet'.
4387  * Returns 0 if successful, otherwise a positive errno value. */
4388 int
4389 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct dp_packet *packet)
4390 {
4391     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4392     int error;
4393
4394     error = xlate_send_packet(ofport, packet);
4395
4396     ovs_mutex_lock(&ofproto->stats_mutex);
4397     ofproto->stats.tx_packets++;
4398     ofproto->stats.tx_bytes += dp_packet_size(packet);
4399     ovs_mutex_unlock(&ofproto->stats_mutex);
4400     return error;
4401 }
4402
4403 uint64_t
4404 group_dpif_get_selection_method_param(const struct group_dpif *group)
4405 {
4406     return group->up.props.selection_method_param;
4407 }
4408
4409 const struct field_array *
4410 group_dpif_get_fields(const struct group_dpif *group)
4411 {
4412     return &group->up.props.fields;
4413 }
4414 \f
4415 /* Return the version string of the datapath that backs up
4416  * this 'ofproto'.
4417  */
4418 static const char *
4419 get_datapath_version(const struct ofproto *ofproto_)
4420 {
4421     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4422
4423     return ofproto->backer->dp_version_string;
4424 }
4425
4426 static bool
4427 set_frag_handling(struct ofproto *ofproto_,
4428                   enum ofputil_frag_handling frag_handling)
4429 {
4430     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4431     if (frag_handling != OFPUTIL_FRAG_REASM) {
4432         ofproto->backer->need_revalidate = REV_RECONFIGURE;
4433         return true;
4434     } else {
4435         return false;
4436     }
4437 }
4438
4439 static enum ofperr
4440 packet_out(struct ofproto *ofproto_, struct dp_packet *packet,
4441            const struct flow *flow,
4442            const struct ofpact *ofpacts, size_t ofpacts_len)
4443 {
4444     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4445
4446     ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
4447                                  ofpacts_len, packet);
4448     return 0;
4449 }
4450
4451 static enum ofperr
4452 nxt_resume(struct ofproto *ofproto_,
4453            const struct ofputil_packet_in_private *pin)
4454 {
4455     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4456
4457     /* Translate pin into datapath actions. */
4458     uint64_t odp_actions_stub[1024 / 8];
4459     struct ofpbuf odp_actions = OFPBUF_STUB_INITIALIZER(odp_actions_stub);
4460     enum slow_path_reason slow;
4461     enum ofperr error = xlate_resume(ofproto, pin, &odp_actions, &slow);
4462
4463     /* Steal 'pin->packet' and put it into a dp_packet. */
4464     struct dp_packet packet;
4465     dp_packet_init(&packet, pin->public.packet_len);
4466     dp_packet_put(&packet, pin->public.packet, pin->public.packet_len);
4467
4468     pkt_metadata_from_flow(&packet.md, &pin->public.flow_metadata.flow);
4469
4470     /* Fix up in_port. */
4471     ofproto_dpif_set_packet_odp_port(ofproto,
4472                                      pin->public.flow_metadata.flow.in_port.ofp_port,
4473                                      &packet);
4474
4475     struct flow headers;
4476     flow_extract(&packet, &headers);
4477
4478     /* Execute the datapath actions on the packet. */
4479     struct dpif_execute execute = {
4480         .actions = odp_actions.data,
4481         .actions_len = odp_actions.size,
4482         .needs_help = (slow & SLOW_ACTION) != 0,
4483         .packet = &packet,
4484         .flow = &headers,
4485     };
4486     dpif_execute(ofproto->backer->dpif, &execute);
4487
4488     /* Clean up. */
4489     ofpbuf_uninit(&odp_actions);
4490     dp_packet_uninit(&packet);
4491
4492     return error;
4493 }
4494 \f
4495 /* NetFlow. */
4496
4497 static int
4498 set_netflow(struct ofproto *ofproto_,
4499             const struct netflow_options *netflow_options)
4500 {
4501     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4502
4503     if (netflow_options) {
4504         if (!ofproto->netflow) {
4505             ofproto->netflow = netflow_create();
4506             ofproto->backer->need_revalidate = REV_RECONFIGURE;
4507         }
4508         return netflow_set_options(ofproto->netflow, netflow_options);
4509     } else if (ofproto->netflow) {
4510         ofproto->backer->need_revalidate = REV_RECONFIGURE;
4511         netflow_unref(ofproto->netflow);
4512         ofproto->netflow = NULL;
4513     }
4514
4515     return 0;
4516 }
4517
4518 static void
4519 get_netflow_ids(const struct ofproto *ofproto_,
4520                 uint8_t *engine_type, uint8_t *engine_id)
4521 {
4522     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4523
4524     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
4525 }
4526 \f
4527 static struct ofproto_dpif *
4528 ofproto_dpif_lookup(const char *name)
4529 {
4530     struct ofproto_dpif *ofproto;
4531
4532     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
4533                              hash_string(name, 0), &all_ofproto_dpifs) {
4534         if (!strcmp(ofproto->up.name, name)) {
4535             return ofproto;
4536         }
4537     }
4538     return NULL;
4539 }
4540
4541 static void
4542 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
4543                           const char *argv[], void *aux OVS_UNUSED)
4544 {
4545     struct ofproto_dpif *ofproto;
4546
4547     if (argc > 1) {
4548         ofproto = ofproto_dpif_lookup(argv[1]);
4549         if (!ofproto) {
4550             unixctl_command_reply_error(conn, "no such bridge");
4551             return;
4552         }
4553         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4554         mac_learning_flush(ofproto->ml);
4555         ovs_rwlock_unlock(&ofproto->ml->rwlock);
4556     } else {
4557         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4558             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4559             mac_learning_flush(ofproto->ml);
4560             ovs_rwlock_unlock(&ofproto->ml->rwlock);
4561         }
4562     }
4563
4564     unixctl_command_reply(conn, "table successfully flushed");
4565 }
4566
4567 static void
4568 ofproto_unixctl_mcast_snooping_flush(struct unixctl_conn *conn, int argc,
4569                                      const char *argv[], void *aux OVS_UNUSED)
4570 {
4571     struct ofproto_dpif *ofproto;
4572
4573     if (argc > 1) {
4574         ofproto = ofproto_dpif_lookup(argv[1]);
4575         if (!ofproto) {
4576             unixctl_command_reply_error(conn, "no such bridge");
4577             return;
4578         }
4579
4580         if (!mcast_snooping_enabled(ofproto->ms)) {
4581             unixctl_command_reply_error(conn, "multicast snooping is disabled");
4582             return;
4583         }
4584         mcast_snooping_mdb_flush(ofproto->ms);
4585     } else {
4586         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4587             if (!mcast_snooping_enabled(ofproto->ms)) {
4588                 continue;
4589             }
4590             mcast_snooping_mdb_flush(ofproto->ms);
4591         }
4592     }
4593
4594     unixctl_command_reply(conn, "table successfully flushed");
4595 }
4596
4597 static struct ofport_dpif *
4598 ofbundle_get_a_port(const struct ofbundle *bundle)
4599 {
4600     return CONTAINER_OF(ovs_list_front(&bundle->ports), struct ofport_dpif,
4601                         bundle_node);
4602 }
4603
4604 static void
4605 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4606                          const char *argv[], void *aux OVS_UNUSED)
4607 {
4608     struct ds ds = DS_EMPTY_INITIALIZER;
4609     const struct ofproto_dpif *ofproto;
4610     const struct mac_entry *e;
4611
4612     ofproto = ofproto_dpif_lookup(argv[1]);
4613     if (!ofproto) {
4614         unixctl_command_reply_error(conn, "no such bridge");
4615         return;
4616     }
4617
4618     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
4619     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
4620     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
4621         struct ofbundle *bundle = mac_entry_get_port(ofproto->ml, e);
4622         char name[OFP_MAX_PORT_NAME_LEN];
4623
4624         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4625                                name, sizeof name);
4626         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
4627                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
4628                       mac_entry_age(ofproto->ml, e));
4629     }
4630     ovs_rwlock_unlock(&ofproto->ml->rwlock);
4631     unixctl_command_reply(conn, ds_cstr(&ds));
4632     ds_destroy(&ds);
4633 }
4634
4635 static void
4636 ofproto_unixctl_mcast_snooping_show(struct unixctl_conn *conn,
4637                                     int argc OVS_UNUSED,
4638                                     const char *argv[],
4639                                     void *aux OVS_UNUSED)
4640 {
4641     struct ds ds = DS_EMPTY_INITIALIZER;
4642     const struct ofproto_dpif *ofproto;
4643     const struct ofbundle *bundle;
4644     const struct mcast_group *grp;
4645     struct mcast_group_bundle *b;
4646     struct mcast_mrouter_bundle *mrouter;
4647
4648     ofproto = ofproto_dpif_lookup(argv[1]);
4649     if (!ofproto) {
4650         unixctl_command_reply_error(conn, "no such bridge");
4651         return;
4652     }
4653
4654     if (!mcast_snooping_enabled(ofproto->ms)) {
4655         unixctl_command_reply_error(conn, "multicast snooping is disabled");
4656         return;
4657     }
4658
4659     ds_put_cstr(&ds, " port  VLAN  GROUP                Age\n");
4660     ovs_rwlock_rdlock(&ofproto->ms->rwlock);
4661     LIST_FOR_EACH (grp, group_node, &ofproto->ms->group_lru) {
4662         LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
4663             char name[OFP_MAX_PORT_NAME_LEN];
4664
4665             bundle = b->port;
4666             ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4667                                    name, sizeof name);
4668             ds_put_format(&ds, "%5s  %4d  ", name, grp->vlan);
4669             ipv6_format_mapped(&grp->addr, &ds);
4670             ds_put_format(&ds, "         %3d\n",
4671                           mcast_bundle_age(ofproto->ms, b));
4672         }
4673     }
4674
4675     /* ports connected to multicast routers */
4676     LIST_FOR_EACH(mrouter, mrouter_node, &ofproto->ms->mrouter_lru) {
4677         char name[OFP_MAX_PORT_NAME_LEN];
4678
4679         bundle = mrouter->port;
4680         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4681                                name, sizeof name);
4682         ds_put_format(&ds, "%5s  %4d  querier             %3d\n",
4683                       name, mrouter->vlan,
4684                       mcast_mrouter_age(ofproto->ms, mrouter));
4685     }
4686     ovs_rwlock_unlock(&ofproto->ms->rwlock);
4687     unixctl_command_reply(conn, ds_cstr(&ds));
4688     ds_destroy(&ds);
4689 }
4690
4691 struct trace_ctx {
4692     struct xlate_out xout;
4693     struct xlate_in xin;
4694     const struct flow *key;
4695     struct flow flow;
4696     struct ds *result;
4697     struct flow_wildcards wc;
4698     struct ofpbuf odp_actions;
4699 };
4700
4701 static void
4702 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
4703 {
4704     const struct rule_actions *actions;
4705     ovs_be64 cookie;
4706
4707     ds_put_char_multiple(result, '\t', level);
4708     if (!rule) {
4709         ds_put_cstr(result, "No match\n");
4710         return;
4711     }
4712
4713     ovs_mutex_lock(&rule->up.mutex);
4714     cookie = rule->up.flow_cookie;
4715     ovs_mutex_unlock(&rule->up.mutex);
4716
4717     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
4718                   rule ? rule->up.table_id : 0, ntohll(cookie));
4719     cls_rule_format(&rule->up.cr, result);
4720     ds_put_char(result, '\n');
4721
4722     actions = rule_dpif_get_actions(rule);
4723
4724     ds_put_char_multiple(result, '\t', level);
4725     ds_put_cstr(result, "OpenFlow actions=");
4726     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
4727     ds_put_char(result, '\n');
4728 }
4729
4730 static void
4731 trace_format_flow(struct ds *result, int level, const char *title,
4732                   struct trace_ctx *trace)
4733 {
4734     ds_put_char_multiple(result, '\t', level);
4735     ds_put_format(result, "%s: ", title);
4736     /* Do not report unchanged flows for resubmits. */
4737     if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
4738         || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
4739         ds_put_cstr(result, "unchanged");
4740     } else {
4741         flow_format(result, &trace->xin.flow);
4742         trace->flow = trace->xin.flow;
4743     }
4744     ds_put_char(result, '\n');
4745 }
4746
4747 static void
4748 trace_format_regs(struct ds *result, int level, const char *title,
4749                   struct trace_ctx *trace)
4750 {
4751     size_t i;
4752
4753     ds_put_char_multiple(result, '\t', level);
4754     ds_put_format(result, "%s:", title);
4755     for (i = 0; i < FLOW_N_REGS; i++) {
4756         ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
4757     }
4758     ds_put_char(result, '\n');
4759 }
4760
4761 static void
4762 trace_format_odp(struct ds *result, int level, const char *title,
4763                  struct trace_ctx *trace)
4764 {
4765     struct ofpbuf *odp_actions = &trace->odp_actions;
4766
4767     ds_put_char_multiple(result, '\t', level);
4768     ds_put_format(result, "%s: ", title);
4769     format_odp_actions(result, odp_actions->data, odp_actions->size);
4770     ds_put_char(result, '\n');
4771 }
4772
4773 static void
4774 trace_format_megaflow(struct ds *result, int level, const char *title,
4775                       struct trace_ctx *trace)
4776 {
4777     struct match match;
4778
4779     ds_put_char_multiple(result, '\t', level);
4780     ds_put_format(result, "%s: ", title);
4781     match_init(&match, trace->key, &trace->wc);
4782     match_format(&match, result, OFP_DEFAULT_PRIORITY);
4783     ds_put_char(result, '\n');
4784 }
4785
4786 static void trace_report(struct xlate_in *, int indentation,
4787                          const char *format, ...)
4788     OVS_PRINTF_FORMAT(3, 4);
4789 static void trace_report_valist(struct xlate_in *, int indentation,
4790                                 const char *format, va_list args)
4791     OVS_PRINTF_FORMAT(3, 0);
4792
4793 static void
4794 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int indentation)
4795 {
4796     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4797     struct ds *result = trace->result;
4798
4799     if (!indentation) {
4800         if (rule == xin->ofproto->miss_rule) {
4801             trace_report(xin, indentation,
4802                          "No match, flow generates \"packet in\"s.");
4803         } else if (rule == xin->ofproto->no_packet_in_rule) {
4804             trace_report(xin, indentation, "No match, packets dropped because "
4805                          "OFPPC_NO_PACKET_IN is set on in_port.");
4806         } else if (rule == xin->ofproto->drop_frags_rule) {
4807             trace_report(xin, indentation,
4808                          "Packets dropped because they are IP fragments and "
4809                          "the fragment handling mode is \"drop\".");
4810         }
4811     }
4812
4813     ds_put_char(result, '\n');
4814     if (indentation) {
4815         trace_format_flow(result, indentation, "Resubmitted flow", trace);
4816         trace_format_regs(result, indentation, "Resubmitted regs", trace);
4817         trace_format_odp(result,  indentation, "Resubmitted  odp", trace);
4818         trace_format_megaflow(result, indentation, "Resubmitted megaflow",
4819                               trace);
4820     }
4821     trace_format_rule(result, indentation, rule);
4822 }
4823
4824 static void
4825 trace_report_valist(struct xlate_in *xin, int indentation,
4826                     const char *format, va_list args)
4827 {
4828     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4829     struct ds *result = trace->result;
4830
4831     ds_put_char_multiple(result, '\t', indentation);
4832     ds_put_format_valist(result, format, args);
4833     ds_put_char(result, '\n');
4834 }
4835
4836 static void
4837 trace_report(struct xlate_in *xin, int indentation, const char *format, ...)
4838 {
4839     va_list args;
4840
4841     va_start(args, format);
4842     trace_report_valist(xin, indentation, format, args);
4843     va_end(args);
4844 }
4845
4846 /* Parses the 'argc' elements of 'argv', ignoring argv[0].  The following
4847  * forms are supported:
4848  *
4849  *     - [dpname] odp_flow [-generate | packet]
4850  *     - bridge br_flow [-generate | packet]
4851  *
4852  * On success, initializes '*ofprotop' and 'flow' and returns NULL.  On failure
4853  * returns a nonnull malloced error message. */
4854 static char * OVS_WARN_UNUSED_RESULT
4855 parse_flow_and_packet(int argc, const char *argv[],
4856                       struct ofproto_dpif **ofprotop, struct flow *flow,
4857                       struct dp_packet **packetp)
4858 {
4859     const struct dpif_backer *backer = NULL;
4860     const char *error = NULL;
4861     char *m_err = NULL;
4862     struct simap port_names = SIMAP_INITIALIZER(&port_names);
4863     struct dp_packet *packet;
4864     struct ofpbuf odp_key;
4865     struct ofpbuf odp_mask;
4866
4867     ofpbuf_init(&odp_key, 0);
4868     ofpbuf_init(&odp_mask, 0);
4869
4870     /* Handle "-generate" or a hex string as the last argument. */
4871     if (!strcmp(argv[argc - 1], "-generate")) {
4872         packet = dp_packet_new(0);
4873         argc--;
4874     } else {
4875         error = eth_from_hex(argv[argc - 1], &packet);
4876         if (!error) {
4877             argc--;
4878         } else if (argc == 4) {
4879             /* The 3-argument form must end in "-generate' or a hex string. */
4880             goto exit;
4881         }
4882         error = NULL;
4883     }
4884
4885     /* odp_flow can have its in_port specified as a name instead of port no.
4886      * We do not yet know whether a given flow is a odp_flow or a br_flow.
4887      * But, to know whether a flow is odp_flow through odp_flow_from_string(),
4888      * we need to create a simap of name to port no. */
4889     if (argc == 3) {
4890         const char *dp_type;
4891         if (!strncmp(argv[1], "ovs-", 4)) {
4892             dp_type = argv[1] + 4;
4893         } else {
4894             dp_type = argv[1];
4895         }
4896         backer = shash_find_data(&all_dpif_backers, dp_type);
4897     } else if (argc == 2) {
4898         struct shash_node *node;
4899         if (shash_count(&all_dpif_backers) == 1) {
4900             node = shash_first(&all_dpif_backers);
4901             backer = node->data;
4902         }
4903     } else {
4904         error = "Syntax error";
4905         goto exit;
4906     }
4907     if (backer && backer->dpif) {
4908         struct dpif_port dpif_port;
4909         struct dpif_port_dump port_dump;
4910         DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4911             simap_put(&port_names, dpif_port.name,
4912                       odp_to_u32(dpif_port.port_no));
4913         }
4914     }
4915
4916     /* Parse the flow and determine whether a datapath or
4917      * bridge is specified. If function odp_flow_key_from_string()
4918      * returns 0, the flow is a odp_flow. If function
4919      * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4920     if (!odp_flow_from_string(argv[argc - 1], &port_names,
4921                               &odp_key, &odp_mask)) {
4922         if (!backer) {
4923             error = "Cannot find the datapath";
4924             goto exit;
4925         }
4926
4927         if (odp_flow_key_to_flow(odp_key.data, odp_key.size, flow) == ODP_FIT_ERROR) {
4928             error = "Failed to parse datapath flow key";
4929             goto exit;
4930         }
4931
4932         *ofprotop = xlate_lookup_ofproto(backer, flow,
4933                                          &flow->in_port.ofp_port);
4934         if (*ofprotop == NULL) {
4935             error = "Invalid datapath flow";
4936             goto exit;
4937         }
4938     } else {
4939         char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4940
4941         if (err) {
4942             m_err = xasprintf("Bad openflow flow syntax: %s", err);
4943             free(err);
4944             goto exit;
4945         } else {
4946             if (argc != 3) {
4947                 error = "Must specify bridge name";
4948                 goto exit;
4949             }
4950
4951             *ofprotop = ofproto_dpif_lookup(argv[1]);
4952             if (!*ofprotop) {
4953                 error = "Unknown bridge name";
4954                 goto exit;
4955             }
4956         }
4957     }
4958
4959     /* Generate a packet, if requested. */
4960     if (packet) {
4961         if (!dp_packet_size(packet)) {
4962             flow_compose(packet, flow);
4963         } else {
4964             /* Use the metadata from the flow and the packet argument
4965              * to reconstruct the flow. */
4966             pkt_metadata_from_flow(&packet->md, flow);
4967             flow_extract(packet, flow);
4968         }
4969     }
4970
4971 exit:
4972     if (error && !m_err) {
4973         m_err = xstrdup(error);
4974     }
4975     if (m_err) {
4976         dp_packet_delete(packet);
4977         packet = NULL;
4978     }
4979     *packetp = packet;
4980     ofpbuf_uninit(&odp_key);
4981     ofpbuf_uninit(&odp_mask);
4982     simap_destroy(&port_names);
4983     return m_err;
4984 }
4985
4986 static void
4987 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4988                       void *aux OVS_UNUSED)
4989 {
4990     struct ofproto_dpif *ofproto;
4991     struct dp_packet *packet;
4992     char *error;
4993     struct flow flow;
4994
4995     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4996     if (!error) {
4997         struct ds result;
4998
4999         ds_init(&result);
5000         ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
5001         unixctl_command_reply(conn, ds_cstr(&result));
5002         ds_destroy(&result);
5003         dp_packet_delete(packet);
5004     } else {
5005         unixctl_command_reply_error(conn, error);
5006         free(error);
5007     }
5008 }
5009
5010 static void
5011 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
5012                               const char *argv[], void *aux OVS_UNUSED)
5013 {
5014     enum ofputil_protocol usable_protocols;
5015     struct ofproto_dpif *ofproto;
5016     bool enforce_consistency;
5017     struct ofpbuf ofpacts;
5018     struct dp_packet *packet;
5019     struct ds result;
5020     struct flow flow;
5021     uint16_t in_port;
5022
5023     /* Three kinds of error return values! */
5024     enum ofperr retval;
5025     char *error;
5026
5027     packet = NULL;
5028     ds_init(&result);
5029     ofpbuf_init(&ofpacts, 0);
5030
5031     /* Parse actions. */
5032     error = ofpacts_parse_actions(argv[--argc], &ofpacts, &usable_protocols);
5033     if (error) {
5034         unixctl_command_reply_error(conn, error);
5035         free(error);
5036         goto exit;
5037     }
5038
5039     /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
5040      * consistency between the flow and the actions.  With -consistent, we
5041      * enforce consistency even for a flow supported in OpenFlow 1.0. */
5042     if (!strcmp(argv[1], "-consistent")) {
5043         enforce_consistency = true;
5044         argv++;
5045         argc--;
5046     } else {
5047         enforce_consistency = false;
5048     }
5049
5050     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
5051     if (error) {
5052         unixctl_command_reply_error(conn, error);
5053         free(error);
5054         goto exit;
5055     }
5056
5057     /* Do the same checks as handle_packet_out() in ofproto.c.
5058      *
5059      * We pass a 'table_id' of 0 to ofpacts_check(), which isn't
5060      * strictly correct because these actions aren't in any table, but it's OK
5061      * because it 'table_id' is used only to check goto_table instructions, but
5062      * packet-outs take a list of actions and therefore it can't include
5063      * instructions.
5064      *
5065      * We skip the "meter" check here because meter is an instruction, not an
5066      * action, and thus cannot appear in ofpacts. */
5067     in_port = ofp_to_u16(flow.in_port.ofp_port);
5068     if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
5069         unixctl_command_reply_error(conn, "invalid in_port");
5070         goto exit;
5071     }
5072     if (enforce_consistency) {
5073         retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size, &flow,
5074                                            u16_to_ofp(ofproto->up.max_ports),
5075                                            0, ofproto->up.n_tables,
5076                                            usable_protocols);
5077     } else {
5078         retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow,
5079                                u16_to_ofp(ofproto->up.max_ports), 0,
5080                                ofproto->up.n_tables, &usable_protocols);
5081     }
5082     if (!retval) {
5083         retval = ofproto_check_ofpacts(&ofproto->up, ofpacts.data,
5084                                        ofpacts.size);
5085     }
5086
5087     if (retval) {
5088         ds_clear(&result);
5089         ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
5090         unixctl_command_reply_error(conn, ds_cstr(&result));
5091         goto exit;
5092     }
5093
5094     ofproto_trace(ofproto, &flow, packet,
5095                   ofpacts.data, ofpacts.size, &result);
5096     unixctl_command_reply(conn, ds_cstr(&result));
5097
5098 exit:
5099     ds_destroy(&result);
5100     dp_packet_delete(packet);
5101     ofpbuf_uninit(&ofpacts);
5102 }
5103
5104 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
5105  * description of the results to 'ds'.
5106  *
5107  * The trace follows a packet with the specified 'flow' through the flow
5108  * table.  'packet' may be nonnull to trace an actual packet, with consequent
5109  * side effects (if it is nonnull then its flow must be 'flow').
5110  *
5111  * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
5112  * trace, otherwise the actions are determined by a flow table lookup. */
5113 static void
5114 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
5115               const struct dp_packet *packet,
5116               const struct ofpact ofpacts[], size_t ofpacts_len,
5117               struct ds *ds)
5118 {
5119     struct trace_ctx trace;
5120     enum xlate_error error;
5121
5122     ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
5123     ds_put_cstr(ds, "Flow: ");
5124     flow_format(ds, flow);
5125     ds_put_char(ds, '\n');
5126
5127     ofpbuf_init(&trace.odp_actions, 0);
5128
5129     trace.result = ds;
5130     trace.key = flow; /* Original flow key, used for megaflow. */
5131     trace.flow = *flow; /* May be modified by actions. */
5132     xlate_in_init(&trace.xin, ofproto, flow, flow->in_port.ofp_port, NULL,
5133                   ntohs(flow->tcp_flags), packet, &trace.wc,
5134                   &trace.odp_actions);
5135     trace.xin.ofpacts = ofpacts;
5136     trace.xin.ofpacts_len = ofpacts_len;
5137     trace.xin.resubmit_hook = trace_resubmit;
5138     trace.xin.report_hook = trace_report_valist;
5139
5140     error = xlate_actions(&trace.xin, &trace.xout);
5141     ds_put_char(ds, '\n');
5142     trace.xin.flow.actset_output = 0;
5143     trace_format_flow(ds, 0, "Final flow", &trace);
5144     trace_format_megaflow(ds, 0, "Megaflow", &trace);
5145
5146     ds_put_cstr(ds, "Datapath actions: ");
5147     format_odp_actions(ds, trace.odp_actions.data, trace.odp_actions.size);
5148
5149     if (error != XLATE_OK) {
5150         ds_put_format(ds, "\nTranslation failed (%s), packet is dropped.\n",
5151                       xlate_strerror(error));
5152     } else if (trace.xout.slow) {
5153         enum slow_path_reason slow;
5154
5155         ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5156                     "slow path because it:");
5157
5158         slow = trace.xout.slow;
5159         while (slow) {
5160             enum slow_path_reason bit = rightmost_1bit(slow);
5161
5162             ds_put_format(ds, "\n\t- %s.",
5163                           slow_path_reason_to_explanation(bit));
5164
5165             slow &= ~bit;
5166         }
5167     }
5168
5169     xlate_out_uninit(&trace.xout);
5170     ofpbuf_uninit(&trace.odp_actions);
5171 }
5172
5173 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
5174  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
5175  * to destroy 'ofproto_shash' and free the returned value. */
5176 static const struct shash_node **
5177 get_ofprotos(struct shash *ofproto_shash)
5178 {
5179     const struct ofproto_dpif *ofproto;
5180
5181     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5182         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5183         shash_add_nocopy(ofproto_shash, name, ofproto);
5184     }
5185
5186     return shash_sort(ofproto_shash);
5187 }
5188
5189 static void
5190 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5191                               const char *argv[] OVS_UNUSED,
5192                               void *aux OVS_UNUSED)
5193 {
5194     struct ds ds = DS_EMPTY_INITIALIZER;
5195     struct shash ofproto_shash;
5196     const struct shash_node **sorted_ofprotos;
5197     int i;
5198
5199     shash_init(&ofproto_shash);
5200     sorted_ofprotos = get_ofprotos(&ofproto_shash);
5201     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5202         const struct shash_node *node = sorted_ofprotos[i];
5203         ds_put_format(&ds, "%s\n", node->name);
5204     }
5205
5206     shash_destroy(&ofproto_shash);
5207     free(sorted_ofprotos);
5208
5209     unixctl_command_reply(conn, ds_cstr(&ds));
5210     ds_destroy(&ds);
5211 }
5212
5213 static void
5214 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5215 {
5216     const struct shash_node **ofprotos;
5217     struct dpif_dp_stats dp_stats;
5218     struct shash ofproto_shash;
5219     size_t i;
5220
5221     dpif_get_dp_stats(backer->dpif, &dp_stats);
5222
5223     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5224                   dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
5225
5226     shash_init(&ofproto_shash);
5227     ofprotos = get_ofprotos(&ofproto_shash);
5228     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5229         struct ofproto_dpif *ofproto = ofprotos[i]->data;
5230         const struct shash_node **ports;
5231         size_t j;
5232
5233         if (ofproto->backer != backer) {
5234             continue;
5235         }
5236
5237         ds_put_format(ds, "\t%s:\n", ofproto->up.name);
5238
5239         ports = shash_sort(&ofproto->up.port_by_name);
5240         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5241             const struct shash_node *node = ports[j];
5242             struct ofport *ofport = node->data;
5243             struct smap config;
5244             odp_port_t odp_port;
5245
5246             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5247                           ofport->ofp_port);
5248
5249             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5250             if (odp_port != ODPP_NONE) {
5251                 ds_put_format(ds, "%"PRIu32":", odp_port);
5252             } else {
5253                 ds_put_cstr(ds, "none:");
5254             }
5255
5256             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5257
5258             smap_init(&config);
5259             if (!netdev_get_config(ofport->netdev, &config)) {
5260                 const struct smap_node **nodes;
5261                 size_t i;
5262
5263                 nodes = smap_sort(&config);
5264                 for (i = 0; i < smap_count(&config); i++) {
5265                     const struct smap_node *node = nodes[i];
5266                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5267                                   node->key, node->value);
5268                 }
5269                 free(nodes);
5270             }
5271             smap_destroy(&config);
5272
5273             ds_put_char(ds, ')');
5274             ds_put_char(ds, '\n');
5275         }
5276         free(ports);
5277     }
5278     shash_destroy(&ofproto_shash);
5279     free(ofprotos);
5280 }
5281
5282 static void
5283 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5284                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5285 {
5286     struct ds ds = DS_EMPTY_INITIALIZER;
5287     const struct shash_node **backers;
5288     int i;
5289
5290     backers = shash_sort(&all_dpif_backers);
5291     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5292         dpif_show_backer(backers[i]->data, &ds);
5293     }
5294     free(backers);
5295
5296     unixctl_command_reply(conn, ds_cstr(&ds));
5297     ds_destroy(&ds);
5298 }
5299
5300 static void
5301 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5302                                 int argc OVS_UNUSED, const char *argv[],
5303                                 void *aux OVS_UNUSED)
5304 {
5305     const struct ofproto_dpif *ofproto;
5306
5307     struct ds ds = DS_EMPTY_INITIALIZER;
5308     bool verbosity = false;
5309
5310     struct dpif_port dpif_port;
5311     struct dpif_port_dump port_dump;
5312     struct hmap portno_names;
5313
5314     struct dpif_flow_dump *flow_dump;
5315     struct dpif_flow_dump_thread *flow_dump_thread;
5316     struct dpif_flow f;
5317     int error;
5318
5319     ofproto = ofproto_dpif_lookup(argv[argc - 1]);
5320     if (!ofproto) {
5321         unixctl_command_reply_error(conn, "no such bridge");
5322         return;
5323     }
5324
5325     if (argc > 2 && !strcmp(argv[1], "-m")) {
5326         verbosity = true;
5327     }
5328
5329     hmap_init(&portno_names);
5330     DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
5331         odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
5332     }
5333
5334     ds_init(&ds);
5335     flow_dump = dpif_flow_dump_create(ofproto->backer->dpif, false);
5336     flow_dump_thread = dpif_flow_dump_thread_create(flow_dump);
5337     while (dpif_flow_dump_next(flow_dump_thread, &f, 1)) {
5338         struct flow flow;
5339
5340         if (odp_flow_key_to_flow(f.key, f.key_len, &flow) == ODP_FIT_ERROR
5341             || xlate_lookup_ofproto(ofproto->backer, &flow, NULL) != ofproto) {
5342             continue;
5343         }
5344
5345         if (verbosity) {
5346             odp_format_ufid(&f.ufid, &ds);
5347             ds_put_cstr(&ds, " ");
5348         }
5349         odp_flow_format(f.key, f.key_len, f.mask, f.mask_len,
5350                         &portno_names, &ds, verbosity);
5351         ds_put_cstr(&ds, ", ");
5352         dpif_flow_stats_format(&f.stats, &ds);
5353         ds_put_cstr(&ds, ", actions:");
5354         format_odp_actions(&ds, f.actions, f.actions_len);
5355         ds_put_char(&ds, '\n');
5356     }
5357     dpif_flow_dump_thread_destroy(flow_dump_thread);
5358     error = dpif_flow_dump_destroy(flow_dump);
5359
5360     if (error) {
5361         ds_clear(&ds);
5362         ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
5363         unixctl_command_reply_error(conn, ds_cstr(&ds));
5364     } else {
5365         unixctl_command_reply(conn, ds_cstr(&ds));
5366     }
5367     odp_portno_names_destroy(&portno_names);
5368     hmap_destroy(&portno_names);
5369     ds_destroy(&ds);
5370 }
5371
5372 static void
5373 ofproto_revalidate_all_backers(void)
5374 {
5375     const struct shash_node **backers;
5376     int i;
5377
5378     backers = shash_sort(&all_dpif_backers);
5379     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5380         struct dpif_backer *backer = backers[i]->data;
5381         backer->need_revalidate = REV_RECONFIGURE;
5382     }
5383     free(backers);
5384 }
5385
5386 static void
5387 disable_tnl_push_pop(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5388                      const char *argv[], void *aux OVS_UNUSED)
5389 {
5390     if (!strcasecmp(argv[1], "off")) {
5391         ofproto_use_tnl_push_pop = false;
5392         unixctl_command_reply(conn, "Tunnel push-pop off");
5393         ofproto_revalidate_all_backers();
5394     } else if (!strcasecmp(argv[1], "on")) {
5395         ofproto_use_tnl_push_pop = true;
5396         unixctl_command_reply(conn, "Tunnel push-pop on");
5397         ofproto_revalidate_all_backers();
5398     } else {
5399         unixctl_command_reply_error(conn, "Invalid argument");
5400     }
5401 }
5402
5403 static void
5404 disable_datapath_truncate(struct unixctl_conn *conn OVS_UNUSED,
5405                           int argc OVS_UNUSED,
5406                           const char *argv[] OVS_UNUSED,
5407                           void *aux OVS_UNUSED)
5408 {
5409     const struct shash_node **backers;
5410     int i;
5411
5412     backers = shash_sort(&all_dpif_backers);
5413     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5414         struct dpif_backer *backer = backers[i]->data;
5415         backer->support.trunc = false;
5416     }
5417     free(backers);
5418     unixctl_command_reply(conn, "Datapath truncate action diabled");
5419 }
5420
5421 static void
5422 ofproto_unixctl_init(void)
5423 {
5424     static bool registered;
5425     if (registered) {
5426         return;
5427     }
5428     registered = true;
5429
5430     unixctl_command_register(
5431         "ofproto/trace",
5432         "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
5433         1, 3, ofproto_unixctl_trace, NULL);
5434     unixctl_command_register(
5435         "ofproto/trace-packet-out",
5436         "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
5437         2, 6, ofproto_unixctl_trace_actions, NULL);
5438     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5439                              ofproto_unixctl_fdb_flush, NULL);
5440     unixctl_command_register("fdb/show", "bridge", 1, 1,
5441                              ofproto_unixctl_fdb_show, NULL);
5442     unixctl_command_register("mdb/flush", "[bridge]", 0, 1,
5443                              ofproto_unixctl_mcast_snooping_flush, NULL);
5444     unixctl_command_register("mdb/show", "bridge", 1, 1,
5445                              ofproto_unixctl_mcast_snooping_show, NULL);
5446     unixctl_command_register("dpif/dump-dps", "", 0, 0,
5447                              ofproto_unixctl_dpif_dump_dps, NULL);
5448     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5449                              NULL);
5450     unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
5451                              ofproto_unixctl_dpif_dump_flows, NULL);
5452
5453     unixctl_command_register("ofproto/tnl-push-pop", "[on]|[off]", 1, 1,
5454                              disable_tnl_push_pop, NULL);
5455
5456     unixctl_command_register("dpif/disable-truncate", "", 0, 0,
5457                              disable_datapath_truncate, NULL);
5458 }
5459
5460 /* Returns true if 'table' is the table used for internal rules,
5461  * false otherwise. */
5462 bool
5463 table_is_internal(uint8_t table_id)
5464 {
5465     return table_id == TBL_INTERNAL;
5466 }
5467 \f
5468
5469 static odp_port_t
5470 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
5471 {
5472     const struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
5473     return ofport ? ofport->odp_port : ODPP_NONE;
5474 }
5475
5476 struct ofport_dpif *
5477 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
5478 {
5479     struct ofport_dpif *port;
5480
5481     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
5482     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
5483                              &backer->odp_to_ofport_map) {
5484         if (port->odp_port == odp_port) {
5485             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5486             return port;
5487         }
5488     }
5489
5490     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5491     return NULL;
5492 }
5493
5494 static ofp_port_t
5495 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
5496 {
5497     struct ofport_dpif *port;
5498
5499     port = odp_port_to_ofport(ofproto->backer, odp_port);
5500     if (port && &ofproto->up == port->up.ofproto) {
5501         return port->up.ofp_port;
5502     } else {
5503         return OFPP_NONE;
5504     }
5505 }
5506
5507 int
5508 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
5509                                const struct match *match, int priority,
5510                                uint16_t idle_timeout,
5511                                const struct ofpbuf *ofpacts,
5512                                struct rule **rulep)
5513 {
5514     struct ofproto_flow_mod ofm;
5515     struct rule_dpif *rule;
5516     int error;
5517
5518     ofm.fm = (struct ofputil_flow_mod) {
5519         .match = *match,
5520         .priority = priority,
5521         .table_id = TBL_INTERNAL,
5522         .command = OFPFC_ADD,
5523         .idle_timeout = idle_timeout,
5524         .flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY,
5525         .ofpacts = ofpacts->data,
5526         .ofpacts_len = ofpacts->size,
5527         .delete_reason = OVS_OFPRR_NONE,
5528     };
5529
5530     error = ofproto_flow_mod(&ofproto->up, &ofm);
5531     if (error) {
5532         VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
5533                     ofperr_to_string(error));
5534         *rulep = NULL;
5535         return error;
5536     }
5537
5538     rule = rule_dpif_lookup_in_table(ofproto,
5539                                      ofproto_dpif_get_tables_version(ofproto),
5540                                      TBL_INTERNAL, &ofm.fm.match.flow,
5541                                      &ofm.fm.match.wc);
5542     if (rule) {
5543         *rulep = &rule->up;
5544     } else {
5545         OVS_NOT_REACHED();
5546     }
5547     return 0;
5548 }
5549
5550 int
5551 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
5552                                   struct match *match, int priority)
5553 {
5554     struct ofproto_flow_mod ofm;
5555     int error;
5556
5557     ofm.fm = (struct ofputil_flow_mod) {
5558         .match = *match,
5559         .priority = priority,
5560         .table_id = TBL_INTERNAL,
5561         .flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY,
5562         .command = OFPFC_DELETE_STRICT,
5563     };
5564
5565     error = ofproto_flow_mod(&ofproto->up, &ofm);
5566     if (error) {
5567         VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
5568                     ofperr_to_string(error));
5569         return error;
5570     }
5571
5572     return 0;
5573 }
5574
5575 const struct uuid *
5576 ofproto_dpif_get_uuid(const struct ofproto_dpif *ofproto)
5577 {
5578     return &ofproto->uuid;
5579 }
5580
5581 const struct ofproto_class ofproto_dpif_class = {
5582     init,
5583     enumerate_types,
5584     enumerate_names,
5585     del,
5586     port_open_type,
5587     type_run,
5588     type_wait,
5589     alloc,
5590     construct,
5591     destruct,
5592     dealloc,
5593     run,
5594     ofproto_dpif_wait,
5595     NULL,                       /* get_memory_usage. */
5596     type_get_memory_usage,
5597     flush,
5598     query_tables,
5599     set_tables_version,
5600     port_alloc,
5601     port_construct,
5602     port_destruct,
5603     port_dealloc,
5604     port_modified,
5605     port_reconfigured,
5606     port_query_by_name,
5607     port_add,
5608     port_del,
5609     port_get_stats,
5610     port_dump_start,
5611     port_dump_next,
5612     port_dump_done,
5613     port_poll,
5614     port_poll_wait,
5615     port_is_lacp_current,
5616     port_get_lacp_stats,
5617     NULL,                       /* rule_choose_table */
5618     rule_alloc,
5619     rule_construct,
5620     rule_insert,
5621     NULL,                       /* rule_delete */
5622     rule_destruct,
5623     rule_dealloc,
5624     rule_get_stats,
5625     rule_execute,
5626     set_frag_handling,
5627     packet_out,
5628     nxt_resume,
5629     set_netflow,
5630     get_netflow_ids,
5631     set_sflow,
5632     set_ipfix,
5633     get_ipfix_stats,
5634     set_cfm,
5635     cfm_status_changed,
5636     get_cfm_status,
5637     set_lldp,
5638     get_lldp_status,
5639     set_aa,
5640     aa_mapping_set,
5641     aa_mapping_unset,
5642     aa_vlan_get_queued,
5643     aa_vlan_get_queue_size,
5644     set_bfd,
5645     bfd_status_changed,
5646     get_bfd_status,
5647     set_stp,
5648     get_stp_status,
5649     set_stp_port,
5650     get_stp_port_status,
5651     get_stp_port_stats,
5652     set_rstp,
5653     get_rstp_status,
5654     set_rstp_port,
5655     get_rstp_port_status,
5656     set_queues,
5657     bundle_set,
5658     bundle_remove,
5659     mirror_set__,
5660     mirror_get_stats__,
5661     set_flood_vlans,
5662     is_mirror_output_bundle,
5663     forward_bpdu_changed,
5664     set_mac_table_config,
5665     set_mcast_snooping,
5666     set_mcast_snooping_port,
5667     NULL,                       /* meter_get_features */
5668     NULL,                       /* meter_set */
5669     NULL,                       /* meter_get */
5670     NULL,                       /* meter_del */
5671     group_alloc,                /* group_alloc */
5672     group_construct,            /* group_construct */
5673     group_destruct,             /* group_destruct */
5674     group_dealloc,              /* group_dealloc */
5675     group_modify,               /* group_modify */
5676     group_get_stats,            /* group_get_stats */
5677     get_datapath_version,       /* get_datapath_version */
5678 };