ofproto/ofproto-dpif.c

   1 /*
   2  * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015 Nicira, Inc.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18
  19 #include "ofproto/ofproto-dpif.h"
  20 #include "ofproto/ofproto-provider.h"
  21
  22 #include <errno.h>
  23
  24 #include "bfd.h"
  25 #include "bond.h"
  26 #include "bundle.h"
  27 #include "byte-order.h"
  28 #include "connectivity.h"
  29 #include "connmgr.h"
  30 #include "coverage.h"
  31 #include "cfm.h"
  32 #include "ovs-lldp.h"
  33 #include "dpif.h"
  34 #include "dynamic-string.h"
  35 #include "fail-open.h"
  36 #include "guarded-list.h"
  37 #include "hmapx.h"
  38 #include "lacp.h"
  39 #include "learn.h"
  40 #include "mac-learning.h"
  41 #include "mcast-snooping.h"
  42 #include "meta-flow.h"
  43 #include "multipath.h"
  44 #include "netdev-vport.h"
  45 #include "netdev.h"
  46 #include "netlink.h"
  47 #include "nx-match.h"
  48 #include "odp-util.h"
  49 #include "odp-execute.h"
  50 #include "ofp-util.h"
  51 #include "ofpbuf.h"
  52 #include "ofp-actions.h"
  53 #include "ofp-parse.h"
  54 #include "ofp-print.h"
  55 #include "ofproto-dpif-ipfix.h"
  56 #include "ofproto-dpif-mirror.h"
  57 #include "ofproto-dpif-monitor.h"
  58 #include "ofproto-dpif-rid.h"
  59 #include "ofproto-dpif-sflow.h"
  60 #include "ofproto-dpif-upcall.h"
  61 #include "ofproto-dpif-xlate.h"
  62 #include "poll-loop.h"
  63 #include "ovs-rcu.h"
  64 #include "ovs-router.h"
  65 #include "seq.h"
  66 #include "simap.h"
  67 #include "smap.h"
  68 #include "timer.h"
  69 #include "tunnel.h"
  70 #include "unaligned.h"
  71 #include "unixctl.h"
  72 #include "vlan-bitmap.h"
  73 #include "openvswitch/vlog.h"
  74
  75 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  76
  77 COVERAGE_DEFINE(ofproto_dpif_expired);
  78 COVERAGE_DEFINE(packet_in_overflow);
  79
  80 struct flow_miss;
  81
  82 struct rule_dpif {
  83     struct rule up;
  84
  85     /* These statistics:
  86      *
  87      *   - Do include packets and bytes from datapath flows which have not
  88      *   recently been processed by a revalidator. */
  89     struct ovs_mutex stats_mutex;
  90     struct dpif_flow_stats stats OVS_GUARDED;
  91
  92    /* In non-NULL, will point to a new rule (for which a reference is held) to
  93     * which all the stats updates should be forwarded. This exists only
  94     * transitionally when flows are replaced.
  95     *
  96     * Protected by stats_mutex.  If both 'rule->stats_mutex' and
  97     * 'rule->new_rule->stats_mutex' must be held together, acquire them in that
  98     * order, */
  99     struct rule_dpif *new_rule OVS_GUARDED;
 100
 101     /* If non-zero then the recirculation id that has
 102      * been allocated for use with this rule.
 103      * The recirculation id and associated internal flow should
 104      * be freed when the rule is freed */
 105     uint32_t recirc_id;
 106 };
 107
 108 /* RULE_CAST() depends on this. */
 109 BUILD_ASSERT_DECL(offsetof(struct rule_dpif, up) == 0);
 110
 111 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
 112                            long long int *used);
 113 static struct rule_dpif *rule_dpif_cast(const struct rule *);
 114 static void rule_expire(struct rule_dpif *);
 115
 116 struct group_dpif {
 117     struct ofgroup up;
 118
 119     /* These statistics:
 120      *
 121      *   - Do include packets and bytes from datapath flows which have not
 122      *   recently been processed by a revalidator. */
 123     struct ovs_mutex stats_mutex;
 124     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 125     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 126 };
 127
 128 struct ofbundle {
 129     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
 130     struct ofproto_dpif *ofproto; /* Owning ofproto. */
 131     void *aux;                  /* Key supplied by ofproto's client. */
 132     char *name;                 /* Identifier for log messages. */
 133
 134     /* Configuration. */
 135     struct ovs_list ports;      /* Contains "struct ofport"s. */
 136     enum port_vlan_mode vlan_mode; /* VLAN mode */
 137     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
 138     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
 139                                  * NULL if all VLANs are trunked. */
 140     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 141     struct bond *bond;          /* Nonnull iff more than one port. */
 142     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 143
 144     /* Status. */
 145     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 146 };
 147
 148 static void bundle_remove(struct ofport *);
 149 static void bundle_update(struct ofbundle *);
 150 static void bundle_destroy(struct ofbundle *);
 151 static void bundle_del_port(struct ofport_dpif *);
 152 static void bundle_run(struct ofbundle *);
 153 static void bundle_wait(struct ofbundle *);
 154 static void bundle_flush_macs(struct ofbundle *, bool);
 155 static void bundle_move(struct ofbundle *, struct ofbundle *);
 156
 157 static void stp_run(struct ofproto_dpif *ofproto);
 158 static void stp_wait(struct ofproto_dpif *ofproto);
 159 static int set_stp_port(struct ofport *,
 160                         const struct ofproto_port_stp_settings *);
 161
 162 static void rstp_run(struct ofproto_dpif *ofproto);
 163 static void set_rstp_port(struct ofport *,
 164                          const struct ofproto_port_rstp_settings *);
 165
 166 struct ofport_dpif {
 167     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 168     struct ofport up;
 169
 170     odp_port_t odp_port;
 171     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 172     struct ovs_list bundle_node;/* In struct ofbundle's "ports" list. */
 173     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 174     struct bfd *bfd;            /* BFD, if any. */
 175     struct lldp *lldp;          /* lldp, if any. */
 176     bool may_enable;            /* May be enabled in bonds. */
 177     bool is_tunnel;             /* This port is a tunnel. */
 178     bool is_layer3;             /* This is a layer 3 port. */
 179     long long int carrier_seq;  /* Carrier status changes. */
 180     struct ofport_dpif *peer;   /* Peer if patch port. */
 181
 182     /* Spanning tree. */
 183     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 184     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 185     long long int stp_state_entered;
 186
 187     /* Rapid Spanning Tree. */
 188     struct rstp_port *rstp_port; /* Rapid Spanning Tree Protocol, if any. */
 189     enum rstp_state rstp_state; /* Always RSTP_DISABLED if RSTP not in use. */
 190
 191     /* Queue to DSCP mapping. */
 192     struct ofproto_port_queue *qdscp;
 193     size_t n_qdscp;
 194
 195     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 196      *
 197      * This is deprecated.  It is only for compatibility with broken device
 198      * drivers in old versions of Linux that do not properly support VLANs when
 199      * VLAN devices are not used.  When broken device drivers are no longer in
 200      * widespread use, we will delete these interfaces. */
 201     ofp_port_t realdev_ofp_port;
 202     int vlandev_vid;
 203 };
 204
 205 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 206  *
 207  * This is deprecated.  It is only for compatibility with broken device drivers
 208  * in old versions of Linux that do not properly support VLANs when VLAN
 209  * devices are not used.  When broken device drivers are no longer in
 210  * widespread use, we will delete these interfaces. */
 211 struct vlan_splinter {
 212     struct hmap_node realdev_vid_node;
 213     struct hmap_node vlandev_node;
 214     ofp_port_t realdev_ofp_port;
 215     ofp_port_t vlandev_ofp_port;
 216     int vid;
 217 };
 218
 219 static void vsp_remove(struct ofport_dpif *);
 220 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 221
 222 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 223                                        ofp_port_t);
 224
 225 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 226                                        odp_port_t);
 227
 228 static struct ofport_dpif *
 229 ofport_dpif_cast(const struct ofport *ofport)
 230 {
 231     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 232 }
 233
 234 static void port_run(struct ofport_dpif *);
 235 static int set_bfd(struct ofport *, const struct smap *);
 236 static int set_cfm(struct ofport *, const struct cfm_settings *);
 237 static int set_lldp(struct ofport *ofport_, const struct smap *cfg);
 238 static void ofport_update_peer(struct ofport_dpif *);
 239
 240 /* Reasons that we might need to revalidate every datapath flow, and
 241  * corresponding coverage counters.
 242  *
 243  * A value of 0 means that there is no need to revalidate.
 244  *
 245  * It would be nice to have some cleaner way to integrate with coverage
 246  * counters, but with only a few reasons I guess this is good enough for
 247  * now. */
 248 enum revalidate_reason {
 249     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 250     REV_STP,                   /* Spanning tree protocol port status change. */
 251     REV_RSTP,                  /* RSTP port status change. */
 252     REV_BOND,                  /* Bonding changed. */
 253     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 254     REV_FLOW_TABLE,            /* Flow table changed. */
 255     REV_MAC_LEARNING,          /* Mac learning changed. */
 256     REV_MCAST_SNOOPING,        /* Multicast snooping changed. */
 257 };
 258 COVERAGE_DEFINE(rev_reconfigure);
 259 COVERAGE_DEFINE(rev_stp);
 260 COVERAGE_DEFINE(rev_rstp);
 261 COVERAGE_DEFINE(rev_bond);
 262 COVERAGE_DEFINE(rev_port_toggled);
 263 COVERAGE_DEFINE(rev_flow_table);
 264 COVERAGE_DEFINE(rev_mac_learning);
 265 COVERAGE_DEFINE(rev_mcast_snooping);
 266
 267 /* All datapaths of a given type share a single dpif backer instance. */
 268 struct dpif_backer {
 269     char *type;
 270     int refcount;
 271     struct dpif *dpif;
 272     struct udpif *udpif;
 273
 274     struct ovs_rwlock odp_to_ofport_lock;
 275     struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
 276
 277     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 278
 279     enum revalidate_reason need_revalidate; /* Revalidate all flows. */
 280
 281     bool recv_set_enable; /* Enables or disables receiving packets. */
 282
 283     /* Version string of the datapath stored in OVSDB. */
 284     char *dp_version_string;
 285
 286     /* Datapath feature support. */
 287     struct dpif_backer_support support;
 288     struct atomic_count tnl_count;
 289 };
 290
 291 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 292 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 293
 294 struct ofproto_dpif {
 295     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 296     struct ofproto up;
 297     struct dpif_backer *backer;
 298
 299     ATOMIC(cls_version_t) tables_version;  /* For classifier lookups. */
 300
 301     uint64_t dump_seq; /* Last read of udpif_dump_seq(). */
 302
 303     /* Special OpenFlow rules. */
 304     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 305     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 306     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 307
 308     /* Bridging. */
 309     struct netflow *netflow;
 310     struct dpif_sflow *sflow;
 311     struct dpif_ipfix *ipfix;
 312     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 313     struct mac_learning *ml;
 314     struct mcast_snooping *ms;
 315     bool has_bonded_bundles;
 316     bool lacp_enabled;
 317     struct mbridge *mbridge;
 318
 319     struct ovs_mutex stats_mutex;
 320     struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
 321                                             * consumed in userspace. */
 322
 323     /* Spanning tree. */
 324     struct stp *stp;
 325     long long int stp_last_tick;
 326
 327     /* Rapid Spanning Tree. */
 328     struct rstp *rstp;
 329     long long int rstp_last_tick;
 330
 331     /* VLAN splinters. */
 332     struct ovs_mutex vsp_mutex;
 333     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 334     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 335
 336     /* Ports. */
 337     struct sset ports;             /* Set of standard port names. */
 338     struct sset ghost_ports;       /* Ports with no datapath port. */
 339     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 340     int port_poll_errno;           /* Last errno for port_poll() reply. */
 341     uint64_t change_seq;           /* Connectivity status changes. */
 342
 343     /* Work queues. */
 344     struct guarded_list pins;      /* Contains "struct ofputil_packet_in"s. */
 345     struct seq *pins_seq;          /* For notifying 'pins' reception. */
 346     uint64_t pins_seqno;
 347 };
 348
 349 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 350 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 351
 352 static bool ofproto_use_tnl_push_pop = true;
 353 static void ofproto_unixctl_init(void);
 354
 355 static inline struct ofproto_dpif *
 356 ofproto_dpif_cast(const struct ofproto *ofproto)
 357 {
 358     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 359     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 360 }
 361
 362 bool
 363 ofproto_dpif_get_enable_ufid(const struct dpif_backer *backer)
 364 {
 365     return backer->support.ufid;
 366 }
 367
 368 struct dpif_backer_support *
 369 ofproto_dpif_get_support(const struct ofproto_dpif *ofproto)
 370 {
 371     return &ofproto->backer->support;
 372 }
 373
 374 static void ofproto_trace(struct ofproto_dpif *, struct flow *,
 375                           const struct dp_packet *packet,
 376                           const struct ofpact[], size_t ofpacts_len,
 377                           struct ds *);
 378
 379 /* Global variables. */
 380 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 381
 382 /* Initial mappings of port to bridge mappings. */
 383 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 384
 385 /* Executes 'fm'.  The caller retains ownership of 'fm' and everything in
 386  * it. */
 387 void
 388 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 389                       const struct ofputil_flow_mod *fm)
 390 {
 391     struct ofproto_flow_mod ofm;
 392
 393     /* Multiple threads may do this for the same 'fm' at the same time.
 394      * Allocate ofproto_flow_mod with execution context from stack.
 395      *
 396      * Note: This copy could be avoided by making ofproto_flow_mod more
 397      * complex, but that may not be desireable, and a learn action is not that
 398      * fast to begin with. */
 399     ofm.fm = *fm;
 400     ofproto_flow_mod(&ofproto->up, &ofm);
 401 }
 402
 403 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 404  * Takes ownership of 'pin' and pin->packet. */
 405 void
 406 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 407                             struct ofproto_packet_in *pin)
 408 {
 409     if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
 410         COVERAGE_INC(packet_in_overflow);
 411         free(CONST_CAST(void *, pin->up.packet));
 412         free(pin);
 413     }
 414
 415     /* Wakes up main thread for packet-in I/O. */
 416     seq_change(ofproto->pins_seq);
 417 }
 418
 419 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
 420  * packet rather than to send the packet to the controller.
 421  *
 422  * This function returns false to indicate that a packet_in message
 423  * for a "table-miss" should be sent to at least one controller.
 424  * False otherwise. */
 425 bool
 426 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif *ofproto)
 427 {
 428     return connmgr_wants_packet_in_on_miss(ofproto->up.connmgr);
 429 }
 430 \f
 431 /* Factory functions. */
 432
 433 static void
 434 init(const struct shash *iface_hints)
 435 {
 436     struct shash_node *node;
 437
 438     /* Make a local copy, since we don't own 'iface_hints' elements. */
 439     SHASH_FOR_EACH(node, iface_hints) {
 440         const struct iface_hint *orig_hint = node->data;
 441         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 442
 443         new_hint->br_name = xstrdup(orig_hint->br_name);
 444         new_hint->br_type = xstrdup(orig_hint->br_type);
 445         new_hint->ofp_port = orig_hint->ofp_port;
 446
 447         shash_add(&init_ofp_ports, node->name, new_hint);
 448     }
 449
 450     ofproto_unixctl_init();
 451     udpif_init();
 452 }
 453
 454 static void
 455 enumerate_types(struct sset *types)
 456 {
 457     dp_enumerate_types(types);
 458 }
 459
 460 static int
 461 enumerate_names(const char *type, struct sset *names)
 462 {
 463     struct ofproto_dpif *ofproto;
 464
 465     sset_clear(names);
 466     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 467         if (strcmp(type, ofproto->up.type)) {
 468             continue;
 469         }
 470         sset_add(names, ofproto->up.name);
 471     }
 472
 473     return 0;
 474 }
 475
 476 static int
 477 del(const char *type, const char *name)
 478 {
 479     struct dpif *dpif;
 480     int error;
 481
 482     error = dpif_open(name, type, &dpif);
 483     if (!error) {
 484         error = dpif_delete(dpif);
 485         dpif_close(dpif);
 486     }
 487     return error;
 488 }
 489 \f
 490 static const char *
 491 port_open_type(const char *datapath_type, const char *port_type)
 492 {
 493     return dpif_port_open_type(datapath_type, port_type);
 494 }
 495
 496 /* Type functions. */
 497
 498 static void process_dpif_port_changes(struct dpif_backer *);
 499 static void process_dpif_all_ports_changed(struct dpif_backer *);
 500 static void process_dpif_port_change(struct dpif_backer *,
 501                                      const char *devname);
 502 static void process_dpif_port_error(struct dpif_backer *, int error);
 503
 504 static struct ofproto_dpif *
 505 lookup_ofproto_dpif_by_port_name(const char *name)
 506 {
 507     struct ofproto_dpif *ofproto;
 508
 509     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 510         if (sset_contains(&ofproto->ports, name)) {
 511             return ofproto;
 512         }
 513     }
 514
 515     return NULL;
 516 }
 517
 518 bool
 519 ofproto_dpif_backer_enabled(struct dpif_backer* backer)
 520 {
 521     return backer->recv_set_enable;
 522 }
 523
 524 static int
 525 type_run(const char *type)
 526 {
 527     struct dpif_backer *backer;
 528
 529     backer = shash_find_data(&all_dpif_backers, type);
 530     if (!backer) {
 531         /* This is not necessarily a problem, since backers are only
 532          * created on demand. */
 533         return 0;
 534     }
 535
 536
 537     if (dpif_run(backer->dpif)) {
 538         backer->need_revalidate = REV_RECONFIGURE;
 539     }
 540
 541     udpif_run(backer->udpif);
 542
 543     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 544      * and the configuration has now changed to "false", enable receiving
 545      * packets from the datapath. */
 546     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 547         int error;
 548
 549         backer->recv_set_enable = true;
 550
 551         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 552         if (error) {
 553             VLOG_ERR("Failed to enable receiving packets in dpif.");
 554             return error;
 555         }
 556         dpif_flow_flush(backer->dpif);
 557         backer->need_revalidate = REV_RECONFIGURE;
 558     }
 559
 560     if (backer->recv_set_enable) {
 561         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 562     }
 563
 564     dpif_poll_threads_set(backer->dpif, n_dpdk_rxqs, pmd_cpu_mask);
 565
 566     if (backer->need_revalidate) {
 567         struct ofproto_dpif *ofproto;
 568         struct simap_node *node;
 569         struct simap tmp_backers;
 570
 571         /* Handle tunnel garbage collection. */
 572         simap_init(&tmp_backers);
 573         simap_swap(&backer->tnl_backers, &tmp_backers);
 574
 575         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 576             struct ofport_dpif *iter;
 577
 578             if (backer != ofproto->backer) {
 579                 continue;
 580             }
 581
 582             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 583                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 584                 const char *dp_port;
 585
 586                 if (!iter->is_tunnel) {
 587                     continue;
 588                 }
 589
 590                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 591                                                      namebuf, sizeof namebuf);
 592                 node = simap_find(&tmp_backers, dp_port);
 593                 if (node) {
 594                     simap_put(&backer->tnl_backers, dp_port, node->data);
 595                     simap_delete(&tmp_backers, node);
 596                     node = simap_find(&backer->tnl_backers, dp_port);
 597                 } else {
 598                     node = simap_find(&backer->tnl_backers, dp_port);
 599                     if (!node) {
 600                         odp_port_t odp_port = ODPP_NONE;
 601
 602                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 603                                            &odp_port)) {
 604                             simap_put(&backer->tnl_backers, dp_port,
 605                                       odp_to_u32(odp_port));
 606                             node = simap_find(&backer->tnl_backers, dp_port);
 607                         }
 608                     }
 609                 }
 610
 611                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 612                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 613                                          iter->odp_port,
 614                                          ovs_native_tunneling_is_on(ofproto), dp_port)) {
 615                     backer->need_revalidate = REV_RECONFIGURE;
 616                 }
 617             }
 618         }
 619
 620         SIMAP_FOR_EACH (node, &tmp_backers) {
 621             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 622         }
 623         simap_destroy(&tmp_backers);
 624
 625         switch (backer->need_revalidate) {
 626         case REV_RECONFIGURE:    COVERAGE_INC(rev_reconfigure);    break;
 627         case REV_STP:            COVERAGE_INC(rev_stp);            break;
 628         case REV_RSTP:           COVERAGE_INC(rev_rstp);           break;
 629         case REV_BOND:           COVERAGE_INC(rev_bond);           break;
 630         case REV_PORT_TOGGLED:   COVERAGE_INC(rev_port_toggled);   break;
 631         case REV_FLOW_TABLE:     COVERAGE_INC(rev_flow_table);     break;
 632         case REV_MAC_LEARNING:   COVERAGE_INC(rev_mac_learning);   break;
 633         case REV_MCAST_SNOOPING: COVERAGE_INC(rev_mcast_snooping); break;
 634         }
 635         backer->need_revalidate = 0;
 636
 637         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 638             struct ofport_dpif *ofport;
 639             struct ofbundle *bundle;
 640
 641             if (ofproto->backer != backer) {
 642                 continue;
 643             }
 644
 645             xlate_txn_start();
 646             xlate_ofproto_set(ofproto, ofproto->up.name,
 647                               ofproto->backer->dpif, ofproto->ml,
 648                               ofproto->stp, ofproto->rstp, ofproto->ms,
 649                               ofproto->mbridge, ofproto->sflow, ofproto->ipfix,
 650                               ofproto->netflow,
 651                               ofproto->up.forward_bpdu,
 652                               connmgr_has_in_band(ofproto->up.connmgr),
 653                               &ofproto->backer->support);
 654
 655             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 656                 xlate_bundle_set(ofproto, bundle, bundle->name,
 657                                  bundle->vlan_mode, bundle->vlan,
 658                                  bundle->trunks, bundle->use_priority_tags,
 659                                  bundle->bond, bundle->lacp,
 660                                  bundle->floodable);
 661             }
 662
 663             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 664                 int stp_port = ofport->stp_port
 665                     ? stp_port_no(ofport->stp_port)
 666                     : -1;
 667                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 668                                  ofport->up.ofp_port, ofport->odp_port,
 669                                  ofport->up.netdev, ofport->cfm, ofport->bfd,
 670                                  ofport->lldp, ofport->peer, stp_port,
 671                                  ofport->rstp_port, ofport->qdscp,
 672                                  ofport->n_qdscp, ofport->up.pp.config,
 673                                  ofport->up.pp.state, ofport->is_tunnel,
 674                                  ofport->may_enable);
 675             }
 676             xlate_txn_commit();
 677         }
 678
 679         udpif_revalidate(backer->udpif);
 680     }
 681
 682     process_dpif_port_changes(backer);
 683
 684     return 0;
 685 }
 686
 687 /* Check for and handle port changes in 'backer''s dpif. */
 688 static void
 689 process_dpif_port_changes(struct dpif_backer *backer)
 690 {
 691     for (;;) {
 692         char *devname;
 693         int error;
 694
 695         error = dpif_port_poll(backer->dpif, &devname);
 696         switch (error) {
 697         case EAGAIN:
 698             return;
 699
 700         case ENOBUFS:
 701             process_dpif_all_ports_changed(backer);
 702             break;
 703
 704         case 0:
 705             process_dpif_port_change(backer, devname);
 706             free(devname);
 707             break;
 708
 709         default:
 710             process_dpif_port_error(backer, error);
 711             break;
 712         }
 713     }
 714 }
 715
 716 static void
 717 process_dpif_all_ports_changed(struct dpif_backer *backer)
 718 {
 719     struct ofproto_dpif *ofproto;
 720     struct dpif_port dpif_port;
 721     struct dpif_port_dump dump;
 722     struct sset devnames;
 723     const char *devname;
 724
 725     sset_init(&devnames);
 726     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 727         if (ofproto->backer == backer) {
 728             struct ofport *ofport;
 729
 730             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 731                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 732             }
 733         }
 734     }
 735     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 736         sset_add(&devnames, dpif_port.name);
 737     }
 738
 739     SSET_FOR_EACH (devname, &devnames) {
 740         process_dpif_port_change(backer, devname);
 741     }
 742     sset_destroy(&devnames);
 743 }
 744
 745 static void
 746 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 747 {
 748     struct ofproto_dpif *ofproto;
 749     struct dpif_port port;
 750
 751     /* Don't report on the datapath's device. */
 752     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 753         return;
 754     }
 755
 756     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 757                    &all_ofproto_dpifs) {
 758         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 759             return;
 760         }
 761     }
 762
 763     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 764     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 765         /* The port was removed.  If we know the datapath,
 766          * report it through poll_set().  If we don't, it may be
 767          * notifying us of a removal we initiated, so ignore it.
 768          * If there's a pending ENOBUFS, let it stand, since
 769          * everything will be reevaluated. */
 770         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 771             sset_add(&ofproto->port_poll_set, devname);
 772             ofproto->port_poll_errno = 0;
 773         }
 774     } else if (!ofproto) {
 775         /* The port was added, but we don't know with which
 776          * ofproto we should associate it.  Delete it. */
 777         dpif_port_del(backer->dpif, port.port_no);
 778     } else {
 779         struct ofport_dpif *ofport;
 780
 781         ofport = ofport_dpif_cast(shash_find_data(
 782                                       &ofproto->up.port_by_name, devname));
 783         if (ofport
 784             && ofport->odp_port != port.port_no
 785             && !odp_port_to_ofport(backer, port.port_no))
 786         {
 787             /* 'ofport''s datapath port number has changed from
 788              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 789              * structures to match. */
 790             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 791             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 792             ofport->odp_port = port.port_no;
 793             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 794                         hash_odp_port(port.port_no));
 795             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 796             backer->need_revalidate = REV_RECONFIGURE;
 797         }
 798     }
 799     dpif_port_destroy(&port);
 800 }
 801
 802 /* Propagate 'error' to all ofprotos based on 'backer'. */
 803 static void
 804 process_dpif_port_error(struct dpif_backer *backer, int error)
 805 {
 806     struct ofproto_dpif *ofproto;
 807
 808     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 809         if (ofproto->backer == backer) {
 810             sset_clear(&ofproto->port_poll_set);
 811             ofproto->port_poll_errno = error;
 812         }
 813     }
 814 }
 815
 816 static void
 817 type_wait(const char *type)
 818 {
 819     struct dpif_backer *backer;
 820
 821     backer = shash_find_data(&all_dpif_backers, type);
 822     if (!backer) {
 823         /* This is not necessarily a problem, since backers are only
 824          * created on demand. */
 825         return;
 826     }
 827
 828     dpif_wait(backer->dpif);
 829 }
 830 \f
 831 /* Basic life-cycle. */
 832
 833 static int add_internal_flows(struct ofproto_dpif *);
 834
 835 static struct ofproto *
 836 alloc(void)
 837 {
 838     struct ofproto_dpif *ofproto = xzalloc(sizeof *ofproto);
 839     return &ofproto->up;
 840 }
 841
 842 static void
 843 dealloc(struct ofproto *ofproto_)
 844 {
 845     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
 846     free(ofproto);
 847 }
 848
 849 static void
 850 close_dpif_backer(struct dpif_backer *backer)
 851 {
 852     ovs_assert(backer->refcount > 0);
 853
 854     if (--backer->refcount) {
 855         return;
 856     }
 857
 858     udpif_destroy(backer->udpif);
 859
 860     simap_destroy(&backer->tnl_backers);
 861     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
 862     hmap_destroy(&backer->odp_to_ofport_map);
 863     shash_find_and_delete(&all_dpif_backers, backer->type);
 864     free(backer->type);
 865     free(backer->dp_version_string);
 866     dpif_close(backer->dpif);
 867     free(backer);
 868 }
 869
 870 /* Datapath port slated for removal from datapath. */
 871 struct odp_garbage {
 872     struct ovs_list list_node;
 873     odp_port_t odp_port;
 874 };
 875
 876 static bool check_variable_length_userdata(struct dpif_backer *backer);
 877 static void check_support(struct dpif_backer *backer);
 878
 879 static int
 880 open_dpif_backer(const char *type, struct dpif_backer **backerp)
 881 {
 882     struct dpif_backer *backer;
 883     struct dpif_port_dump port_dump;
 884     struct dpif_port port;
 885     struct shash_node *node;
 886     struct ovs_list garbage_list;
 887     struct odp_garbage *garbage;
 888
 889     struct sset names;
 890     char *backer_name;
 891     const char *name;
 892     int error;
 893
 894     recirc_init();
 895
 896     backer = shash_find_data(&all_dpif_backers, type);
 897     if (backer) {
 898         backer->refcount++;
 899         *backerp = backer;
 900         return 0;
 901     }
 902
 903     backer_name = xasprintf("ovs-%s", type);
 904
 905     /* Remove any existing datapaths, since we assume we're the only
 906      * userspace controlling the datapath. */
 907     sset_init(&names);
 908     dp_enumerate_names(type, &names);
 909     SSET_FOR_EACH(name, &names) {
 910         struct dpif *old_dpif;
 911
 912         /* Don't remove our backer if it exists. */
 913         if (!strcmp(name, backer_name)) {
 914             continue;
 915         }
 916
 917         if (dpif_open(name, type, &old_dpif)) {
 918             VLOG_WARN("couldn't open old datapath %s to remove it", name);
 919         } else {
 920             dpif_delete(old_dpif);
 921             dpif_close(old_dpif);
 922         }
 923     }
 924     sset_destroy(&names);
 925
 926     backer = xmalloc(sizeof *backer);
 927
 928     error = dpif_create_and_open(backer_name, type, &backer->dpif);
 929     free(backer_name);
 930     if (error) {
 931         VLOG_ERR("failed to open datapath of type %s: %s", type,
 932                  ovs_strerror(error));
 933         free(backer);
 934         return error;
 935     }
 936     backer->udpif = udpif_create(backer, backer->dpif);
 937
 938     backer->type = xstrdup(type);
 939     backer->refcount = 1;
 940     hmap_init(&backer->odp_to_ofport_map);
 941     ovs_rwlock_init(&backer->odp_to_ofport_lock);
 942     backer->need_revalidate = 0;
 943     simap_init(&backer->tnl_backers);
 944     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
 945     *backerp = backer;
 946
 947     if (backer->recv_set_enable) {
 948         dpif_flow_flush(backer->dpif);
 949     }
 950
 951     /* Loop through the ports already on the datapath and remove any
 952      * that we don't need anymore. */
 953     list_init(&garbage_list);
 954     dpif_port_dump_start(&port_dump, backer->dpif);
 955     while (dpif_port_dump_next(&port_dump, &port)) {
 956         node = shash_find(&init_ofp_ports, port.name);
 957         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
 958             garbage = xmalloc(sizeof *garbage);
 959             garbage->odp_port = port.port_no;
 960             list_push_front(&garbage_list, &garbage->list_node);
 961         }
 962     }
 963     dpif_port_dump_done(&port_dump);
 964
 965     LIST_FOR_EACH_POP (garbage, list_node, &garbage_list) {
 966         dpif_port_del(backer->dpif, garbage->odp_port);
 967         free(garbage);
 968     }
 969
 970     shash_add(&all_dpif_backers, type, backer);
 971
 972     check_support(backer);
 973     atomic_count_init(&backer->tnl_count, 0);
 974
 975     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 976     if (error) {
 977         VLOG_ERR("failed to listen on datapath of type %s: %s",
 978                  type, ovs_strerror(error));
 979         close_dpif_backer(backer);
 980         return error;
 981     }
 982
 983     if (backer->recv_set_enable) {
 984         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 985     }
 986
 987     /* This check fails if performed before udpif threads have been set,
 988      * as the kernel module checks that the 'pid' in userspace action
 989      * is non-zero. */
 990     backer->support.variable_length_userdata
 991         = check_variable_length_userdata(backer);
 992     backer->dp_version_string = dpif_get_dp_version(backer->dpif);
 993
 994     return error;
 995 }
 996
 997 bool
 998 ovs_native_tunneling_is_on(struct ofproto_dpif *ofproto)
 999 {
1000     return ofproto_use_tnl_push_pop && ofproto->backer->support.tnl_push_pop &&
1001            atomic_count_get(&ofproto->backer->tnl_count);
1002 }
1003
1004 /* Tests whether 'backer''s datapath supports recirculation.  Only newer
1005  * datapaths support OVS_KEY_ATTR_RECIRC_ID in keys.  We need to disable some
1006  * features on older datapaths that don't support this feature.
1007  *
1008  * Returns false if 'backer' definitely does not support recirculation, true if
1009  * it seems to support recirculation or if at least the error we get is
1010  * ambiguous. */
1011 static bool
1012 check_recirc(struct dpif_backer *backer)
1013 {
1014     struct flow flow;
1015     struct odputil_keybuf keybuf;
1016     struct ofpbuf key;
1017     bool enable_recirc;
1018     struct odp_flow_key_parms odp_parms = {
1019         .flow = &flow,
1020         .support = {
1021             .recirc = true,
1022         },
1023     };
1024
1025     memset(&flow, 0, sizeof flow);
1026     flow.recirc_id = 1;
1027     flow.dp_hash = 1;
1028
1029     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1030     odp_flow_key_from_flow(&odp_parms, &key);
1031     enable_recirc = dpif_probe_feature(backer->dpif, "recirculation", &key,
1032                                        NULL);
1033
1034     if (enable_recirc) {
1035         VLOG_INFO("%s: Datapath supports recirculation",
1036                   dpif_name(backer->dpif));
1037     } else {
1038         VLOG_INFO("%s: Datapath does not support recirculation",
1039                   dpif_name(backer->dpif));
1040     }
1041
1042     return enable_recirc;
1043 }
1044
1045 /* Tests whether 'dpif' supports unique flow ids. We can skip serializing
1046  * some flow attributes for datapaths that support this feature.
1047  *
1048  * Returns true if 'dpif' supports UFID for flow operations.
1049  * Returns false if  'dpif' does not support UFID. */
1050 static bool
1051 check_ufid(struct dpif_backer *backer)
1052 {
1053     struct flow flow;
1054     struct odputil_keybuf keybuf;
1055     struct ofpbuf key;
1056     ovs_u128 ufid;
1057     bool enable_ufid;
1058     struct odp_flow_key_parms odp_parms = {
1059         .flow = &flow,
1060     };
1061
1062     memset(&flow, 0, sizeof flow);
1063     flow.dl_type = htons(0x1234);
1064
1065     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1066     odp_flow_key_from_flow(&odp_parms, &key);
1067     dpif_flow_hash(backer->dpif, key.data, key.size, &ufid);
1068
1069     enable_ufid = dpif_probe_feature(backer->dpif, "UFID", &key, &ufid);
1070
1071     if (enable_ufid) {
1072         VLOG_INFO("%s: Datapath supports unique flow ids",
1073                   dpif_name(backer->dpif));
1074     } else {
1075         VLOG_INFO("%s: Datapath does not support unique flow ids",
1076                   dpif_name(backer->dpif));
1077     }
1078     return enable_ufid;
1079 }
1080
1081 /* Tests whether 'backer''s datapath supports variable-length
1082  * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.  We need
1083  * to disable some features on older datapaths that don't support this
1084  * feature.
1085  *
1086  * Returns false if 'backer' definitely does not support variable-length
1087  * userdata, true if it seems to support them or if at least the error we get
1088  * is ambiguous. */
1089 static bool
1090 check_variable_length_userdata(struct dpif_backer *backer)
1091 {
1092     struct eth_header *eth;
1093     struct ofpbuf actions;
1094     struct dpif_execute execute;
1095     struct dp_packet packet;
1096     size_t start;
1097     int error;
1098
1099     /* Compose a userspace action that will cause an ERANGE error on older
1100      * datapaths that don't support variable-length userdata.
1101      *
1102      * We really test for using userdata longer than 8 bytes, but older
1103      * datapaths accepted these, silently truncating the userdata to 8 bytes.
1104      * The same older datapaths rejected userdata shorter than 8 bytes, so we
1105      * test for that instead as a proxy for longer userdata support. */
1106     ofpbuf_init(&actions, 64);
1107     start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
1108     nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
1109                    dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
1110     nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
1111     nl_msg_end_nested(&actions, start);
1112
1113     /* Compose a dummy ethernet packet. */
1114     dp_packet_init(&packet, ETH_HEADER_LEN);
1115     eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1116     eth->eth_type = htons(0x1234);
1117
1118     /* Execute the actions.  On older datapaths this fails with ERANGE, on
1119      * newer datapaths it succeeds. */
1120     execute.actions = actions.data;
1121     execute.actions_len = actions.size;
1122     execute.packet = &packet;
1123     execute.needs_help = false;
1124     execute.probe = true;
1125     execute.mtu = 0;
1126
1127     error = dpif_execute(backer->dpif, &execute);
1128
1129     dp_packet_uninit(&packet);
1130     ofpbuf_uninit(&actions);
1131
1132     switch (error) {
1133     case 0:
1134         return true;
1135
1136     case ERANGE:
1137         /* Variable-length userdata is not supported. */
1138         VLOG_WARN("%s: datapath does not support variable-length userdata "
1139                   "feature (needs Linux 3.10+ or kernel module from OVS "
1140                   "1..11+).  The NXAST_SAMPLE action will be ignored.",
1141                   dpif_name(backer->dpif));
1142         return false;
1143
1144     default:
1145         /* Something odd happened.  We're not sure whether variable-length
1146          * userdata is supported.  Default to "yes". */
1147         VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1148                   dpif_name(backer->dpif), ovs_strerror(error));
1149         return true;
1150     }
1151 }
1152
1153 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1154  *
1155  * Returns the number of elements in a struct flow's mpls_lse field
1156  * if the datapath supports at least that many entries in an
1157  * MPLS label stack.
1158  * Otherwise returns the number of MPLS push actions supported by
1159  * the datapath. */
1160 static size_t
1161 check_max_mpls_depth(struct dpif_backer *backer)
1162 {
1163     struct flow flow;
1164     int n;
1165
1166     for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
1167         struct odputil_keybuf keybuf;
1168         struct ofpbuf key;
1169         struct odp_flow_key_parms odp_parms = {
1170             .flow = &flow,
1171         };
1172
1173         memset(&flow, 0, sizeof flow);
1174         flow.dl_type = htons(ETH_TYPE_MPLS);
1175         flow_set_mpls_bos(&flow, n, 1);
1176
1177         ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1178         odp_flow_key_from_flow(&odp_parms, &key);
1179         if (!dpif_probe_feature(backer->dpif, "MPLS", &key, NULL)) {
1180             break;
1181         }
1182     }
1183
1184     VLOG_INFO("%s: MPLS label stack length probed as %d",
1185               dpif_name(backer->dpif), n);
1186     return n;
1187 }
1188
1189 /* Tests whether 'backer''s datapath supports masked data in
1190  * OVS_ACTION_ATTR_SET actions.  We need to disable some features on older
1191  * datapaths that don't support this feature. */
1192 static bool
1193 check_masked_set_action(struct dpif_backer *backer)
1194 {
1195     struct eth_header *eth;
1196     struct ofpbuf actions;
1197     struct dpif_execute execute;
1198     struct dp_packet packet;
1199     int error;
1200     struct ovs_key_ethernet key, mask;
1201
1202     /* Compose a set action that will cause an EINVAL error on older
1203      * datapaths that don't support masked set actions.
1204      * Avoid using a full mask, as it could be translated to a non-masked
1205      * set action instead. */
1206     ofpbuf_init(&actions, 64);
1207     memset(&key, 0x53, sizeof key);
1208     memset(&mask, 0x7f, sizeof mask);
1209     commit_masked_set_action(&actions, OVS_KEY_ATTR_ETHERNET, &key, &mask,
1210                              sizeof key);
1211
1212     /* Compose a dummy ethernet packet. */
1213     dp_packet_init(&packet, ETH_HEADER_LEN);
1214     eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1215     eth->eth_type = htons(0x1234);
1216
1217     /* Execute the actions.  On older datapaths this fails with EINVAL, on
1218      * newer datapaths it succeeds. */
1219     execute.actions = actions.data;
1220     execute.actions_len = actions.size;
1221     execute.packet = &packet;
1222     execute.needs_help = false;
1223     execute.probe = true;
1224     execute.mtu = 0;
1225
1226     error = dpif_execute(backer->dpif, &execute);
1227
1228     dp_packet_uninit(&packet);
1229     ofpbuf_uninit(&actions);
1230
1231     if (error) {
1232         /* Masked set action is not supported. */
1233         VLOG_INFO("%s: datapath does not support masked set action feature.",
1234                   dpif_name(backer->dpif));
1235     }
1236     return !error;
1237 }
1238
1239 #define CHECK_FEATURE__(NAME, SUPPORT, FIELD, VALUE)                        \
1240 static bool                                                                 \
1241 check_##NAME(struct dpif_backer *backer)                                    \
1242 {                                                                           \
1243     struct flow flow;                                                       \
1244     struct odputil_keybuf keybuf;                                           \
1245     struct ofpbuf key;                                                      \
1246     bool enable;                                                            \
1247     struct odp_flow_key_parms odp_parms = {                                 \
1248         .flow = &flow,                                                      \
1249         .support = {                                                        \
1250             .SUPPORT = true,                                                \
1251         },                                                                  \
1252     };                                                                      \
1253                                                                             \
1254     memset(&flow, 0, sizeof flow);                                          \
1255     flow.FIELD = VALUE;                                                     \
1256                                                                             \
1257     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);                         \
1258     odp_flow_key_from_flow(&odp_parms, &key);                               \
1259     enable = dpif_probe_feature(backer->dpif, #NAME, &key, NULL);           \
1260                                                                             \
1261     if (enable) {                                                           \
1262         VLOG_INFO("%s: Datapath supports "#NAME, dpif_name(backer->dpif));  \
1263     } else {                                                                \
1264         VLOG_INFO("%s: Datapath does not support "#NAME,                    \
1265                   dpif_name(backer->dpif));                                 \
1266     }                                                                       \
1267                                                                             \
1268     return enable;                                                          \
1269 }
1270 #define CHECK_FEATURE(FIELD) CHECK_FEATURE__(FIELD, FIELD, FIELD, 1)
1271
1272 CHECK_FEATURE(ct_state)
1273 CHECK_FEATURE(ct_zone)
1274 CHECK_FEATURE(ct_mark)
1275 CHECK_FEATURE__(ct_label, ct_label, ct_label.u64.lo, 1)
1276 CHECK_FEATURE__(ct_state_nat, ct_state, ct_state, CS_TRACKED|CS_SRC_NAT)
1277
1278 #undef CHECK_FEATURE
1279 #undef CHECK_FEATURE__
1280
1281 static void
1282 check_support(struct dpif_backer *backer)
1283 {
1284     /* This feature needs to be tested after udpif threads are set. */
1285     backer->support.variable_length_userdata = false;
1286
1287     backer->support.odp.recirc = check_recirc(backer);
1288     backer->support.odp.max_mpls_depth = check_max_mpls_depth(backer);
1289     backer->support.masked_set_action = check_masked_set_action(backer);
1290     backer->support.ufid = check_ufid(backer);
1291     backer->support.tnl_push_pop = dpif_supports_tnl_push_pop(backer->dpif);
1292
1293     backer->support.odp.ct_state = check_ct_state(backer);
1294     backer->support.odp.ct_zone = check_ct_zone(backer);
1295     backer->support.odp.ct_mark = check_ct_mark(backer);
1296     backer->support.odp.ct_label = check_ct_label(backer);
1297
1298     backer->support.odp.ct_state_nat = check_ct_state_nat(backer);
1299 }
1300
1301 static int
1302 construct(struct ofproto *ofproto_)
1303 {
1304     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1305     struct shash_node *node, *next;
1306     int error;
1307
1308     /* Tunnel module can get used right after the udpif threads are running. */
1309     ofproto_tunnel_init();
1310
1311     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1312     if (error) {
1313         return error;
1314     }
1315
1316     atomic_init(&ofproto->tables_version, CLS_MIN_VERSION);
1317     ofproto->netflow = NULL;
1318     ofproto->sflow = NULL;
1319     ofproto->ipfix = NULL;
1320     ofproto->stp = NULL;
1321     ofproto->rstp = NULL;
1322     ofproto->dump_seq = 0;
1323     hmap_init(&ofproto->bundles);
1324     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1325     ofproto->ms = NULL;
1326     ofproto->mbridge = mbridge_create();
1327     ofproto->has_bonded_bundles = false;
1328     ofproto->lacp_enabled = false;
1329     ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1330     ovs_mutex_init(&ofproto->vsp_mutex);
1331
1332     guarded_list_init(&ofproto->pins);
1333
1334     hmap_init(&ofproto->vlandev_map);
1335     hmap_init(&ofproto->realdev_vid_map);
1336
1337     sset_init(&ofproto->ports);
1338     sset_init(&ofproto->ghost_ports);
1339     sset_init(&ofproto->port_poll_set);
1340     ofproto->port_poll_errno = 0;
1341     ofproto->change_seq = 0;
1342     ofproto->pins_seq = seq_create();
1343     ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1344
1345
1346     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1347         struct iface_hint *iface_hint = node->data;
1348
1349         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1350             /* Check if the datapath already has this port. */
1351             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1352                 sset_add(&ofproto->ports, node->name);
1353             }
1354
1355             free(iface_hint->br_name);
1356             free(iface_hint->br_type);
1357             free(iface_hint);
1358             shash_delete(&init_ofp_ports, node);
1359         }
1360     }
1361
1362     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1363                 hash_string(ofproto->up.name, 0));
1364     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1365
1366     ofproto_init_tables(ofproto_, N_TABLES);
1367     error = add_internal_flows(ofproto);
1368
1369     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1370
1371     return error;
1372 }
1373
1374 static int
1375 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1376                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1377 {
1378     struct match match;
1379     int error;
1380     struct rule *rule;
1381
1382     match_init_catchall(&match);
1383     match_set_reg(&match, 0, id);
1384
1385     error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, 0, ofpacts,
1386                                            &rule);
1387     *rulep = error ? NULL : rule_dpif_cast(rule);
1388
1389     return error;
1390 }
1391
1392 static int
1393 add_internal_flows(struct ofproto_dpif *ofproto)
1394 {
1395     struct ofpact_controller *controller;
1396     uint64_t ofpacts_stub[128 / 8];
1397     struct ofpbuf ofpacts;
1398     struct rule *unused_rulep OVS_UNUSED;
1399     struct match match;
1400     int error;
1401     int id;
1402
1403     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1404     id = 1;
1405
1406     controller = ofpact_put_CONTROLLER(&ofpacts);
1407     controller->max_len = UINT16_MAX;
1408     controller->controller_id = 0;
1409     controller->reason = OFPR_NO_MATCH;
1410     ofpact_pad(&ofpacts);
1411
1412     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1413                                    &ofproto->miss_rule);
1414     if (error) {
1415         return error;
1416     }
1417
1418     ofpbuf_clear(&ofpacts);
1419     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1420                                    &ofproto->no_packet_in_rule);
1421     if (error) {
1422         return error;
1423     }
1424
1425     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1426                                    &ofproto->drop_frags_rule);
1427     if (error) {
1428         return error;
1429     }
1430
1431     /* Drop any run away non-recirc rule lookups. Recirc_id has to be
1432      * zero when reaching this rule.
1433      *
1434      * (priority=2), recirc_id=0, actions=drop
1435      */
1436     ofpbuf_clear(&ofpacts);
1437     match_init_catchall(&match);
1438     match_set_recirc_id(&match, 0);
1439     error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, 0, &ofpacts,
1440                                            &unused_rulep);
1441     return error;
1442 }
1443
1444 static void
1445 destruct(struct ofproto *ofproto_)
1446 {
1447     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1448     struct ofproto_packet_in *pin;
1449     struct rule_dpif *rule;
1450     struct oftable *table;
1451     struct ovs_list pins;
1452
1453     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1454     xlate_txn_start();
1455     xlate_remove_ofproto(ofproto);
1456     xlate_txn_commit();
1457
1458     /* Ensure that the upcall processing threads have no remaining references
1459      * to the ofproto or anything in it. */
1460     udpif_synchronize(ofproto->backer->udpif);
1461
1462     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1463
1464     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1465         CLS_FOR_EACH (rule, up.cr, &table->cls) {
1466             ofproto_rule_delete(&ofproto->up, &rule->up);
1467         }
1468     }
1469     ofproto_group_delete_all(&ofproto->up);
1470
1471     guarded_list_pop_all(&ofproto->pins, &pins);
1472     LIST_FOR_EACH_POP (pin, list_node, &pins) {
1473         free(CONST_CAST(void *, pin->up.packet));
1474         free(pin);
1475     }
1476     guarded_list_destroy(&ofproto->pins);
1477
1478     recirc_free_ofproto(ofproto, ofproto->up.name);
1479
1480     mbridge_unref(ofproto->mbridge);
1481
1482     netflow_unref(ofproto->netflow);
1483     dpif_sflow_unref(ofproto->sflow);
1484     dpif_ipfix_unref(ofproto->ipfix);
1485     hmap_destroy(&ofproto->bundles);
1486     mac_learning_unref(ofproto->ml);
1487     mcast_snooping_unref(ofproto->ms);
1488
1489     hmap_destroy(&ofproto->vlandev_map);
1490     hmap_destroy(&ofproto->realdev_vid_map);
1491
1492     sset_destroy(&ofproto->ports);
1493     sset_destroy(&ofproto->ghost_ports);
1494     sset_destroy(&ofproto->port_poll_set);
1495
1496     ovs_mutex_destroy(&ofproto->stats_mutex);
1497     ovs_mutex_destroy(&ofproto->vsp_mutex);
1498
1499     seq_destroy(ofproto->pins_seq);
1500
1501     close_dpif_backer(ofproto->backer);
1502 }
1503
1504 static int
1505 run(struct ofproto *ofproto_)
1506 {
1507     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1508     uint64_t new_seq, new_dump_seq;
1509
1510     if (mbridge_need_revalidate(ofproto->mbridge)) {
1511         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1512         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1513         mac_learning_flush(ofproto->ml);
1514         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1515         mcast_snooping_mdb_flush(ofproto->ms);
1516     }
1517
1518     /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
1519      * flow restore.  Even though nothing is processed during flow restore,
1520      * all queued 'pins' will be handled immediately when flow restore
1521      * completes. */
1522     ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1523
1524     /* Do not perform any periodic activity required by 'ofproto' while
1525      * waiting for flow restore to complete. */
1526     if (!ofproto_get_flow_restore_wait()) {
1527         struct ofproto_packet_in *pin;
1528         struct ovs_list pins;
1529
1530         guarded_list_pop_all(&ofproto->pins, &pins);
1531         LIST_FOR_EACH_POP (pin, list_node, &pins) {
1532             connmgr_send_packet_in(ofproto->up.connmgr, pin);
1533             free(CONST_CAST(void *, pin->up.packet));
1534             free(pin);
1535         }
1536     }
1537
1538     if (ofproto->netflow) {
1539         netflow_run(ofproto->netflow);
1540     }
1541     if (ofproto->sflow) {
1542         dpif_sflow_run(ofproto->sflow);
1543     }
1544     if (ofproto->ipfix) {
1545         dpif_ipfix_run(ofproto->ipfix);
1546     }
1547
1548     new_seq = seq_read(connectivity_seq_get());
1549     if (ofproto->change_seq != new_seq) {
1550         struct ofport_dpif *ofport;
1551
1552         HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1553             port_run(ofport);
1554         }
1555
1556         ofproto->change_seq = new_seq;
1557     }
1558     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1559         struct ofbundle *bundle;
1560
1561         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1562             bundle_run(bundle);
1563         }
1564     }
1565
1566     stp_run(ofproto);
1567     rstp_run(ofproto);
1568     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1569     if (mac_learning_run(ofproto->ml)) {
1570         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1571     }
1572     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1573
1574     if (mcast_snooping_run(ofproto->ms)) {
1575         ofproto->backer->need_revalidate = REV_MCAST_SNOOPING;
1576     }
1577
1578     new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1579     if (ofproto->dump_seq != new_dump_seq) {
1580         struct rule *rule, *next_rule;
1581
1582         /* We know stats are relatively fresh, so now is a good time to do some
1583          * periodic work. */
1584         ofproto->dump_seq = new_dump_seq;
1585
1586         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1587          * has passed. */
1588         ovs_mutex_lock(&ofproto_mutex);
1589         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1590                             &ofproto->up.expirable) {
1591             rule_expire(rule_dpif_cast(rule));
1592         }
1593         ovs_mutex_unlock(&ofproto_mutex);
1594
1595         /* All outstanding data in existing flows has been accounted, so it's a
1596          * good time to do bond rebalancing. */
1597         if (ofproto->has_bonded_bundles) {
1598             struct ofbundle *bundle;
1599
1600             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1601                 if (bundle->bond) {
1602                     bond_rebalance(bundle->bond);
1603                 }
1604             }
1605         }
1606     }
1607     return 0;
1608 }
1609
1610 static void
1611 wait(struct ofproto *ofproto_)
1612 {
1613     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1614
1615     if (ofproto_get_flow_restore_wait()) {
1616         return;
1617     }
1618
1619     if (ofproto->sflow) {
1620         dpif_sflow_wait(ofproto->sflow);
1621     }
1622     if (ofproto->ipfix) {
1623         dpif_ipfix_wait(ofproto->ipfix);
1624     }
1625     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1626         struct ofbundle *bundle;
1627
1628         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1629             bundle_wait(bundle);
1630         }
1631     }
1632     if (ofproto->netflow) {
1633         netflow_wait(ofproto->netflow);
1634     }
1635     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1636     mac_learning_wait(ofproto->ml);
1637     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1638     mcast_snooping_wait(ofproto->ms);
1639     stp_wait(ofproto);
1640     if (ofproto->backer->need_revalidate) {
1641         poll_immediate_wake();
1642     }
1643
1644     seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1645     seq_wait(ofproto->pins_seq, ofproto->pins_seqno);
1646 }
1647
1648 static void
1649 type_get_memory_usage(const char *type, struct simap *usage)
1650 {
1651     struct dpif_backer *backer;
1652
1653     backer = shash_find_data(&all_dpif_backers, type);
1654     if (backer) {
1655         udpif_get_memory_usage(backer->udpif, usage);
1656     }
1657 }
1658
1659 static void
1660 flush(struct ofproto *ofproto_)
1661 {
1662     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1663     struct dpif_backer *backer = ofproto->backer;
1664
1665     if (backer) {
1666         udpif_flush(backer->udpif);
1667     }
1668 }
1669
1670 static void
1671 query_tables(struct ofproto *ofproto,
1672              struct ofputil_table_features *features,
1673              struct ofputil_table_stats *stats)
1674 {
1675     strcpy(features->name, "classifier");
1676
1677     if (stats) {
1678         int i;
1679
1680         for (i = 0; i < ofproto->n_tables; i++) {
1681             unsigned long missed, matched;
1682
1683             atomic_read_relaxed(&ofproto->tables[i].n_matched, &matched);
1684             atomic_read_relaxed(&ofproto->tables[i].n_missed, &missed);
1685
1686             stats[i].matched_count = matched;
1687             stats[i].lookup_count = matched + missed;
1688         }
1689     }
1690 }
1691
1692 static void
1693 set_tables_version(struct ofproto *ofproto_, cls_version_t version)
1694 {
1695     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1696
1697     atomic_store_relaxed(&ofproto->tables_version, version);
1698 }
1699
1700
1701 static struct ofport *
1702 port_alloc(void)
1703 {
1704     struct ofport_dpif *port = xzalloc(sizeof *port);
1705     return &port->up;
1706 }
1707
1708 static void
1709 port_dealloc(struct ofport *port_)
1710 {
1711     struct ofport_dpif *port = ofport_dpif_cast(port_);
1712     free(port);
1713 }
1714
1715 static int
1716 port_construct(struct ofport *port_)
1717 {
1718     struct ofport_dpif *port = ofport_dpif_cast(port_);
1719     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1720     const struct netdev *netdev = port->up.netdev;
1721     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1722     const char *dp_port_name;
1723     struct dpif_port dpif_port;
1724     int error;
1725
1726     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1727     port->bundle = NULL;
1728     port->cfm = NULL;
1729     port->bfd = NULL;
1730     port->lldp = NULL;
1731     port->may_enable = false;
1732     port->stp_port = NULL;
1733     port->stp_state = STP_DISABLED;
1734     port->rstp_port = NULL;
1735     port->rstp_state = RSTP_DISABLED;
1736     port->is_tunnel = false;
1737     port->peer = NULL;
1738     port->qdscp = NULL;
1739     port->n_qdscp = 0;
1740     port->realdev_ofp_port = 0;
1741     port->vlandev_vid = 0;
1742     port->carrier_seq = netdev_get_carrier_resets(netdev);
1743     port->is_layer3 = netdev_vport_is_layer3(netdev);
1744
1745     if (netdev_vport_is_patch(netdev)) {
1746         /* By bailing out here, we don't submit the port to the sFlow module
1747          * to be considered for counter polling export.  This is correct
1748          * because the patch port represents an interface that sFlow considers
1749          * to be "internal" to the switch as a whole, and therefore not a
1750          * candidate for counter polling. */
1751         port->odp_port = ODPP_NONE;
1752         ofport_update_peer(port);
1753         return 0;
1754     }
1755
1756     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1757     error = dpif_port_query_by_name(ofproto->backer->dpif, dp_port_name,
1758                                     &dpif_port);
1759     if (error) {
1760         return error;
1761     }
1762
1763     port->odp_port = dpif_port.port_no;
1764
1765     if (netdev_get_tunnel_config(netdev)) {
1766         atomic_count_inc(&ofproto->backer->tnl_count);
1767         error = tnl_port_add(port, port->up.netdev, port->odp_port,
1768                              ovs_native_tunneling_is_on(ofproto), dp_port_name);
1769         if (error) {
1770             atomic_count_dec(&ofproto->backer->tnl_count);
1771             dpif_port_destroy(&dpif_port);
1772             return error;
1773         }
1774
1775         port->is_tunnel = true;
1776         if (ofproto->ipfix) {
1777            dpif_ipfix_add_tunnel_port(ofproto->ipfix, port_, port->odp_port);
1778         }
1779     } else {
1780         /* Sanity-check that a mapping doesn't already exist.  This
1781          * shouldn't happen for non-tunnel ports. */
1782         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1783             VLOG_ERR("port %s already has an OpenFlow port number",
1784                      dpif_port.name);
1785             dpif_port_destroy(&dpif_port);
1786             return EBUSY;
1787         }
1788
1789         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1790         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1791                     hash_odp_port(port->odp_port));
1792         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1793     }
1794     dpif_port_destroy(&dpif_port);
1795
1796     if (ofproto->sflow) {
1797         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1798     }
1799
1800     return 0;
1801 }
1802
1803 static void
1804 port_destruct(struct ofport *port_)
1805 {
1806     struct ofport_dpif *port = ofport_dpif_cast(port_);
1807     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1808     const char *devname = netdev_get_name(port->up.netdev);
1809     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1810     const char *dp_port_name;
1811
1812     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1813     xlate_txn_start();
1814     xlate_ofport_remove(port);
1815     xlate_txn_commit();
1816
1817     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1818                                               sizeof namebuf);
1819     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1820         /* The underlying device is still there, so delete it.  This
1821          * happens when the ofproto is being destroyed, since the caller
1822          * assumes that removal of attached ports will happen as part of
1823          * destruction. */
1824         if (!port->is_tunnel) {
1825             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1826         }
1827     }
1828
1829     if (port->peer) {
1830         port->peer->peer = NULL;
1831         port->peer = NULL;
1832     }
1833
1834     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1835         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1836         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1837         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1838     }
1839
1840     if (port->is_tunnel) {
1841         atomic_count_dec(&ofproto->backer->tnl_count);
1842     }
1843
1844     if (port->is_tunnel && ofproto->ipfix) {
1845        dpif_ipfix_del_tunnel_port(ofproto->ipfix, port->odp_port);
1846     }
1847
1848     tnl_port_del(port);
1849     sset_find_and_delete(&ofproto->ports, devname);
1850     sset_find_and_delete(&ofproto->ghost_ports, devname);
1851     bundle_remove(port_);
1852     set_cfm(port_, NULL);
1853     set_bfd(port_, NULL);
1854     set_lldp(port_, NULL);
1855     if (port->stp_port) {
1856         stp_port_disable(port->stp_port);
1857     }
1858     set_rstp_port(port_, NULL);
1859     if (ofproto->sflow) {
1860         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1861     }
1862
1863     free(port->qdscp);
1864 }
1865
1866 static void
1867 port_modified(struct ofport *port_)
1868 {
1869     struct ofport_dpif *port = ofport_dpif_cast(port_);
1870     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1871     const char *dp_port_name;
1872     struct netdev *netdev = port->up.netdev;
1873
1874     if (port->bundle && port->bundle->bond) {
1875         bond_slave_set_netdev(port->bundle->bond, port, netdev);
1876     }
1877
1878     if (port->cfm) {
1879         cfm_set_netdev(port->cfm, netdev);
1880     }
1881
1882     if (port->bfd) {
1883         bfd_set_netdev(port->bfd, netdev);
1884     }
1885
1886     ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1887                                      port->lldp, &port->up.pp.hw_addr);
1888
1889     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1890
1891     if (port->is_tunnel) {
1892         struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1893
1894         if (tnl_port_reconfigure(port, netdev, port->odp_port,
1895                                  ovs_native_tunneling_is_on(ofproto),
1896                                  dp_port_name)) {
1897             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1898         }
1899     }
1900
1901     ofport_update_peer(port);
1902 }
1903
1904 static void
1905 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1906 {
1907     struct ofport_dpif *port = ofport_dpif_cast(port_);
1908     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1909     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1910
1911     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1912                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1913                    OFPUTIL_PC_NO_PACKET_IN)) {
1914         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1915
1916         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1917             bundle_update(port->bundle);
1918         }
1919     }
1920 }
1921
1922 static int
1923 set_sflow(struct ofproto *ofproto_,
1924           const struct ofproto_sflow_options *sflow_options)
1925 {
1926     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1927     struct dpif_sflow *ds = ofproto->sflow;
1928
1929     if (sflow_options) {
1930         uint32_t old_probability = ds ? dpif_sflow_get_probability(ds) : 0;
1931         if (!ds) {
1932             struct ofport_dpif *ofport;
1933
1934             ds = ofproto->sflow = dpif_sflow_create();
1935             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1936                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1937             }
1938         }
1939         dpif_sflow_set_options(ds, sflow_options);
1940         if (dpif_sflow_get_probability(ds) != old_probability) {
1941             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1942         }
1943     } else {
1944         if (ds) {
1945             dpif_sflow_unref(ds);
1946             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1947             ofproto->sflow = NULL;
1948         }
1949     }
1950     return 0;
1951 }
1952
1953 static int
1954 set_ipfix(
1955     struct ofproto *ofproto_,
1956     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1957     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1958     size_t n_flow_exporters_options)
1959 {
1960     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1961     struct dpif_ipfix *di = ofproto->ipfix;
1962     bool has_options = bridge_exporter_options || flow_exporters_options;
1963     bool new_di = false;
1964
1965     if (has_options && !di) {
1966         di = ofproto->ipfix = dpif_ipfix_create();
1967         new_di = true;
1968     }
1969
1970     if (di) {
1971         /* Call set_options in any case to cleanly flush the flow
1972          * caches in the last exporters that are to be destroyed. */
1973         dpif_ipfix_set_options(
1974             di, bridge_exporter_options, flow_exporters_options,
1975             n_flow_exporters_options);
1976
1977         /* Add tunnel ports only when a new ipfix created */
1978         if (new_di == true) {
1979             struct ofport_dpif *ofport;
1980             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1981                 if (ofport->is_tunnel == true) {
1982                     dpif_ipfix_add_tunnel_port(di, &ofport->up, ofport->odp_port);
1983                 }
1984             }
1985         }
1986
1987         if (!has_options) {
1988             dpif_ipfix_unref(di);
1989             ofproto->ipfix = NULL;
1990         }
1991     }
1992
1993     return 0;
1994 }
1995
1996 static int
1997 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1998 {
1999     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2000     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2001     struct cfm *old = ofport->cfm;
2002     int error = 0;
2003
2004     if (s) {
2005         if (!ofport->cfm) {
2006             ofport->cfm = cfm_create(ofport->up.netdev);
2007         }
2008
2009         if (cfm_configure(ofport->cfm, s)) {
2010             error = 0;
2011             goto out;
2012         }
2013
2014         error = EINVAL;
2015     }
2016     cfm_unref(ofport->cfm);
2017     ofport->cfm = NULL;
2018 out:
2019     if (ofport->cfm != old) {
2020         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2021     }
2022     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
2023                                      ofport->lldp, &ofport->up.pp.hw_addr);
2024     return error;
2025 }
2026
2027 static bool
2028 cfm_status_changed(struct ofport *ofport_)
2029 {
2030     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2031
2032     return ofport->cfm ? cfm_check_status_change(ofport->cfm) : true;
2033 }
2034
2035 static int
2036 get_cfm_status(const struct ofport *ofport_,
2037                struct cfm_status *status)
2038 {
2039     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2040     int ret = 0;
2041
2042     if (ofport->cfm) {
2043         cfm_get_status(ofport->cfm, status);
2044     } else {
2045         ret = ENOENT;
2046     }
2047
2048     return ret;
2049 }
2050
2051 static int
2052 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2053 {
2054     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2055     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2056     struct bfd *old;
2057
2058     old = ofport->bfd;
2059     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2060                                 cfg, ofport->up.netdev);
2061     if (ofport->bfd != old) {
2062         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2063     }
2064     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
2065                                      ofport->lldp, &ofport->up.pp.hw_addr);
2066     return 0;
2067 }
2068
2069 static bool
2070 bfd_status_changed(struct ofport *ofport_)
2071 {
2072     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2073
2074     return ofport->bfd ? bfd_check_status_change(ofport->bfd) : true;
2075 }
2076
2077 static int
2078 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2079 {
2080     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2081     int ret = 0;
2082
2083     if (ofport->bfd) {
2084         bfd_get_status(ofport->bfd, smap);
2085     } else {
2086         ret = ENOENT;
2087     }
2088
2089     return ret;
2090 }
2091
2092 static int
2093 set_lldp(struct ofport *ofport_,
2094          const struct smap *cfg)
2095 {
2096     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2097     int error = 0;
2098
2099     if (cfg) {
2100         if (!ofport->lldp) {
2101             struct ofproto_dpif *ofproto;
2102
2103             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2104             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2105             ofport->lldp = lldp_create(ofport->up.netdev, ofport_->mtu, cfg);
2106         }
2107
2108         if (!lldp_configure(ofport->lldp, cfg)) {
2109             error = EINVAL;
2110         }
2111     }
2112     if (error) {
2113         lldp_unref(ofport->lldp);
2114         ofport->lldp = NULL;
2115     }
2116
2117     ofproto_dpif_monitor_port_update(ofport,
2118                                      ofport->bfd,
2119                                      ofport->cfm,
2120                                      ofport->lldp,
2121                                      &ofport->up.pp.hw_addr);
2122     return error;
2123 }
2124
2125 static bool
2126 get_lldp_status(const struct ofport *ofport_,
2127                struct lldp_status *status OVS_UNUSED)
2128 {
2129     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2130
2131     return ofport->lldp ? true : false;
2132 }
2133
2134 static int
2135 set_aa(struct ofproto *ofproto OVS_UNUSED,
2136        const struct aa_settings *s)
2137 {
2138     return aa_configure(s);
2139 }
2140
2141 static int
2142 aa_mapping_set(struct ofproto *ofproto_ OVS_UNUSED, void *aux,
2143                const struct aa_mapping_settings *s)
2144 {
2145     return aa_mapping_register(aux, s);
2146 }
2147
2148 static int
2149 aa_mapping_unset(struct ofproto *ofproto OVS_UNUSED, void *aux)
2150 {
2151     return aa_mapping_unregister(aux);
2152 }
2153
2154 static int
2155 aa_vlan_get_queued(struct ofproto *ofproto OVS_UNUSED, struct ovs_list *list)
2156 {
2157     return aa_get_vlan_queued(list);
2158 }
2159
2160 static unsigned int
2161 aa_vlan_get_queue_size(struct ofproto *ofproto OVS_UNUSED)
2162 {
2163     return aa_get_vlan_queue_size();
2164 }
2165
2166 \f
2167 /* Spanning Tree. */
2168
2169 /* Called while rstp_mutex is held. */
2170 static void
2171 rstp_send_bpdu_cb(struct dp_packet *pkt, void *ofport_, void *ofproto_)
2172 {
2173     struct ofproto_dpif *ofproto = ofproto_;
2174     struct ofport_dpif *ofport = ofport_;
2175     struct eth_header *eth = dp_packet_l2(pkt);
2176
2177     netdev_get_etheraddr(ofport->up.netdev, &eth->eth_src);
2178     if (eth_addr_is_zero(eth->eth_src)) {
2179         VLOG_WARN_RL(&rl, "%s port %d: cannot send RSTP BPDU on a port which "
2180                      "does not have a configured source MAC address.",
2181                      ofproto->up.name, ofp_to_u16(ofport->up.ofp_port));
2182     } else {
2183         ofproto_dpif_send_packet(ofport, pkt);
2184     }
2185     dp_packet_delete(pkt);
2186 }
2187
2188 static void
2189 send_bpdu_cb(struct dp_packet *pkt, int port_num, void *ofproto_)
2190 {
2191     struct ofproto_dpif *ofproto = ofproto_;
2192     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2193     struct ofport_dpif *ofport;
2194
2195     ofport = stp_port_get_aux(sp);
2196     if (!ofport) {
2197         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2198                      ofproto->up.name, port_num);
2199     } else {
2200         struct eth_header *eth = dp_packet_l2(pkt);
2201
2202         netdev_get_etheraddr(ofport->up.netdev, &eth->eth_src);
2203         if (eth_addr_is_zero(eth->eth_src)) {
2204             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2205                          "with unknown MAC", ofproto->up.name, port_num);
2206         } else {
2207             ofproto_dpif_send_packet(ofport, pkt);
2208         }
2209     }
2210     dp_packet_delete(pkt);
2211 }
2212
2213 /* Configure RSTP on 'ofproto_' using the settings defined in 's'. */
2214 static void
2215 set_rstp(struct ofproto *ofproto_, const struct ofproto_rstp_settings *s)
2216 {
2217     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2218
2219     /* Only revalidate flows if the configuration changed. */
2220     if (!s != !ofproto->rstp) {
2221         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2222     }
2223
2224     if (s) {
2225         if (!ofproto->rstp) {
2226             ofproto->rstp = rstp_create(ofproto_->name, s->address,
2227                                         rstp_send_bpdu_cb, ofproto);
2228             ofproto->rstp_last_tick = time_msec();
2229         }
2230         rstp_set_bridge_address(ofproto->rstp, s->address);
2231         rstp_set_bridge_priority(ofproto->rstp, s->priority);
2232         rstp_set_bridge_ageing_time(ofproto->rstp, s->ageing_time);
2233         rstp_set_bridge_force_protocol_version(ofproto->rstp,
2234                                                s->force_protocol_version);
2235         rstp_set_bridge_max_age(ofproto->rstp, s->bridge_max_age);
2236         rstp_set_bridge_forward_delay(ofproto->rstp, s->bridge_forward_delay);
2237         rstp_set_bridge_transmit_hold_count(ofproto->rstp,
2238                                             s->transmit_hold_count);
2239     } else {
2240         struct ofport *ofport;
2241         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2242             set_rstp_port(ofport, NULL);
2243         }
2244         rstp_unref(ofproto->rstp);
2245         ofproto->rstp = NULL;
2246     }
2247 }
2248
2249 static void
2250 get_rstp_status(struct ofproto *ofproto_, struct ofproto_rstp_status *s)
2251 {
2252     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2253
2254     if (ofproto->rstp) {
2255         s->enabled = true;
2256         s->root_id = rstp_get_root_id(ofproto->rstp);
2257         s->bridge_id = rstp_get_bridge_id(ofproto->rstp);
2258         s->designated_id = rstp_get_designated_id(ofproto->rstp);
2259         s->root_path_cost = rstp_get_root_path_cost(ofproto->rstp);
2260         s->designated_port_id = rstp_get_designated_port_id(ofproto->rstp);
2261         s->bridge_port_id = rstp_get_bridge_port_id(ofproto->rstp);
2262     } else {
2263         s->enabled = false;
2264     }
2265 }
2266
2267 static void
2268 update_rstp_port_state(struct ofport_dpif *ofport)
2269 {
2270     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2271     enum rstp_state state;
2272
2273     /* Figure out new state. */
2274     state = ofport->rstp_port ? rstp_port_get_state(ofport->rstp_port)
2275         : RSTP_DISABLED;
2276
2277     /* Update state. */
2278     if (ofport->rstp_state != state) {
2279         enum ofputil_port_state of_state;
2280         bool fwd_change;
2281
2282         VLOG_DBG("port %s: RSTP state changed from %s to %s",
2283                  netdev_get_name(ofport->up.netdev),
2284                  rstp_state_name(ofport->rstp_state),
2285                  rstp_state_name(state));
2286
2287         if (rstp_learn_in_state(ofport->rstp_state)
2288             != rstp_learn_in_state(state)) {
2289             /* XXX: Learning action flows should also be flushed. */
2290             if (ofport->bundle) {
2291                 if (!rstp_shift_root_learned_address(ofproto->rstp)
2292                     || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2293                     bundle_flush_macs(ofport->bundle, false);
2294                 }
2295             }
2296         }
2297         fwd_change = rstp_forward_in_state(ofport->rstp_state)
2298             != rstp_forward_in_state(state);
2299
2300         ofproto->backer->need_revalidate = REV_RSTP;
2301         ofport->rstp_state = state;
2302
2303         if (fwd_change && ofport->bundle) {
2304             bundle_update(ofport->bundle);
2305         }
2306
2307         /* Update the RSTP state bits in the OpenFlow port description. */
2308         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2309         of_state |= (state == RSTP_LEARNING ? OFPUTIL_PS_STP_LEARN
2310                 : state == RSTP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2311                 : state == RSTP_DISCARDING ?  OFPUTIL_PS_STP_LISTEN
2312                 : 0);
2313         ofproto_port_set_state(&ofport->up, of_state);
2314     }
2315 }
2316
2317 static void
2318 rstp_run(struct ofproto_dpif *ofproto)
2319 {
2320     if (ofproto->rstp) {
2321         long long int now = time_msec();
2322         long long int elapsed = now - ofproto->rstp_last_tick;
2323         struct rstp_port *rp;
2324         struct ofport_dpif *ofport;
2325
2326         /* Every second, decrease the values of the timers. */
2327         if (elapsed >= 1000) {
2328             rstp_tick_timers(ofproto->rstp);
2329             ofproto->rstp_last_tick = now;
2330         }
2331         rp = NULL;
2332         while ((ofport = rstp_get_next_changed_port_aux(ofproto->rstp, &rp))) {
2333             update_rstp_port_state(ofport);
2334         }
2335         rp = NULL;
2336         ofport = NULL;
2337         /* FIXME: This check should be done on-event (i.e., when setting
2338          * p->fdb_flush) and not periodically.
2339          */
2340         while ((ofport = rstp_check_and_reset_fdb_flush(ofproto->rstp, &rp))) {
2341             if (!rstp_shift_root_learned_address(ofproto->rstp)
2342                 || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2343                 bundle_flush_macs(ofport->bundle, false);
2344             }
2345         }
2346
2347         if (rstp_shift_root_learned_address(ofproto->rstp)) {
2348             struct ofport_dpif *old_root_aux =
2349                 (struct ofport_dpif *)rstp_get_old_root_aux(ofproto->rstp);
2350             struct ofport_dpif *new_root_aux =
2351                 (struct ofport_dpif *)rstp_get_new_root_aux(ofproto->rstp);
2352             if (old_root_aux != NULL && new_root_aux != NULL) {
2353                 bundle_move(old_root_aux->bundle, new_root_aux->bundle);
2354                 rstp_reset_root_changed(ofproto->rstp);
2355             }
2356         }
2357     }
2358 }
2359
2360 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2361 static int
2362 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2363 {
2364     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2365
2366     /* Only revalidate flows if the configuration changed. */
2367     if (!s != !ofproto->stp) {
2368         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2369     }
2370
2371     if (s) {
2372         if (!ofproto->stp) {
2373             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2374                                       send_bpdu_cb, ofproto);
2375             ofproto->stp_last_tick = time_msec();
2376         }
2377
2378         stp_set_bridge_id(ofproto->stp, s->system_id);
2379         stp_set_bridge_priority(ofproto->stp, s->priority);
2380         stp_set_hello_time(ofproto->stp, s->hello_time);
2381         stp_set_max_age(ofproto->stp, s->max_age);
2382         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2383     }  else {
2384         struct ofport *ofport;
2385
2386         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2387             set_stp_port(ofport, NULL);
2388         }
2389
2390         stp_unref(ofproto->stp);
2391         ofproto->stp = NULL;
2392     }
2393
2394     return 0;
2395 }
2396
2397 static int
2398 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2399 {
2400     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2401
2402     if (ofproto->stp) {
2403         s->enabled = true;
2404         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2405         s->designated_root = stp_get_designated_root(ofproto->stp);
2406         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2407     } else {
2408         s->enabled = false;
2409     }
2410
2411     return 0;
2412 }
2413
2414 static void
2415 update_stp_port_state(struct ofport_dpif *ofport)
2416 {
2417     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2418     enum stp_state state;
2419
2420     /* Figure out new state. */
2421     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2422                              : STP_DISABLED;
2423
2424     /* Update state. */
2425     if (ofport->stp_state != state) {
2426         enum ofputil_port_state of_state;
2427         bool fwd_change;
2428
2429         VLOG_DBG("port %s: STP state changed from %s to %s",
2430                  netdev_get_name(ofport->up.netdev),
2431                  stp_state_name(ofport->stp_state),
2432                  stp_state_name(state));
2433         if (stp_learn_in_state(ofport->stp_state)
2434                 != stp_learn_in_state(state)) {
2435             /* xxx Learning action flows should also be flushed. */
2436             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2437             mac_learning_flush(ofproto->ml);
2438             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2439             mcast_snooping_mdb_flush(ofproto->ms);
2440         }
2441         fwd_change = stp_forward_in_state(ofport->stp_state)
2442                         != stp_forward_in_state(state);
2443
2444         ofproto->backer->need_revalidate = REV_STP;
2445         ofport->stp_state = state;
2446         ofport->stp_state_entered = time_msec();
2447
2448         if (fwd_change && ofport->bundle) {
2449             bundle_update(ofport->bundle);
2450         }
2451
2452         /* Update the STP state bits in the OpenFlow port description. */
2453         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2454         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2455                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2456                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2457                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2458                      : 0);
2459         ofproto_port_set_state(&ofport->up, of_state);
2460     }
2461 }
2462
2463 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2464  * caller is responsible for assigning STP port numbers and ensuring
2465  * there are no duplicates. */
2466 static int
2467 set_stp_port(struct ofport *ofport_,
2468              const struct ofproto_port_stp_settings *s)
2469 {
2470     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2471     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2472     struct stp_port *sp = ofport->stp_port;
2473
2474     if (!s || !s->enable) {
2475         if (sp) {
2476             ofport->stp_port = NULL;
2477             stp_port_disable(sp);
2478             update_stp_port_state(ofport);
2479         }
2480         return 0;
2481     } else if (sp && stp_port_no(sp) != s->port_num
2482                && ofport == stp_port_get_aux(sp)) {
2483         /* The port-id changed, so disable the old one if it's not
2484          * already in use by another port. */
2485         stp_port_disable(sp);
2486     }
2487
2488     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2489
2490     /* Set name before enabling the port so that debugging messages can print
2491      * the name. */
2492     stp_port_set_name(sp, netdev_get_name(ofport->up.netdev));
2493     stp_port_enable(sp);
2494
2495     stp_port_set_aux(sp, ofport);
2496     stp_port_set_priority(sp, s->priority);
2497     stp_port_set_path_cost(sp, s->path_cost);
2498
2499     update_stp_port_state(ofport);
2500
2501     return 0;
2502 }
2503
2504 static int
2505 get_stp_port_status(struct ofport *ofport_,
2506                     struct ofproto_port_stp_status *s)
2507 {
2508     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2509     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2510     struct stp_port *sp = ofport->stp_port;
2511
2512     if (!ofproto->stp || !sp) {
2513         s->enabled = false;
2514         return 0;
2515     }
2516
2517     s->enabled = true;
2518     s->port_id = stp_port_get_id(sp);
2519     s->state = stp_port_get_state(sp);
2520     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2521     s->role = stp_port_get_role(sp);
2522
2523     return 0;
2524 }
2525
2526 static int
2527 get_stp_port_stats(struct ofport *ofport_,
2528                    struct ofproto_port_stp_stats *s)
2529 {
2530     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2531     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2532     struct stp_port *sp = ofport->stp_port;
2533
2534     if (!ofproto->stp || !sp) {
2535         s->enabled = false;
2536         return 0;
2537     }
2538
2539     s->enabled = true;
2540     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2541
2542     return 0;
2543 }
2544
2545 static void
2546 stp_run(struct ofproto_dpif *ofproto)
2547 {
2548     if (ofproto->stp) {
2549         long long int now = time_msec();
2550         long long int elapsed = now - ofproto->stp_last_tick;
2551         struct stp_port *sp;
2552
2553         if (elapsed > 0) {
2554             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2555             ofproto->stp_last_tick = now;
2556         }
2557         while (stp_get_changed_port(ofproto->stp, &sp)) {
2558             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2559
2560             if (ofport) {
2561                 update_stp_port_state(ofport);
2562             }
2563         }
2564
2565         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2566             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2567             mac_learning_flush(ofproto->ml);
2568             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2569             mcast_snooping_mdb_flush(ofproto->ms);
2570         }
2571     }
2572 }
2573
2574 static void
2575 stp_wait(struct ofproto_dpif *ofproto)
2576 {
2577     if (ofproto->stp) {
2578         poll_timer_wait(1000);
2579     }
2580 }
2581
2582 /* Configures RSTP on 'ofport_' using the settings defined in 's'.  The
2583  * caller is responsible for assigning RSTP port numbers and ensuring
2584  * there are no duplicates. */
2585 static void
2586 set_rstp_port(struct ofport *ofport_,
2587               const struct ofproto_port_rstp_settings *s)
2588 {
2589     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2590     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2591     struct rstp_port *rp = ofport->rstp_port;
2592
2593     if (!s || !s->enable) {
2594         if (rp) {
2595             rstp_port_set_aux(rp, NULL);
2596             rstp_port_set_state(rp, RSTP_DISABLED);
2597             rstp_port_set_mac_operational(rp, false);
2598             ofport->rstp_port = NULL;
2599             rstp_port_unref(rp);
2600             update_rstp_port_state(ofport);
2601         }
2602         return;
2603     }
2604
2605     /* Check if need to add a new port. */
2606     if (!rp) {
2607         rp = ofport->rstp_port = rstp_add_port(ofproto->rstp);
2608     }
2609
2610     rstp_port_set(rp, s->port_num, s->priority, s->path_cost,
2611                   s->admin_edge_port, s->auto_edge,
2612                   s->admin_p2p_mac_state, s->admin_port_state, s->mcheck,
2613                   ofport);
2614     update_rstp_port_state(ofport);
2615     /* Synchronize operational status. */
2616     rstp_port_set_mac_operational(rp, ofport->may_enable);
2617 }
2618
2619 static void
2620 get_rstp_port_status(struct ofport *ofport_,
2621         struct ofproto_port_rstp_status *s)
2622 {
2623     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2624     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2625     struct rstp_port *rp = ofport->rstp_port;
2626
2627     if (!ofproto->rstp || !rp) {
2628         s->enabled = false;
2629         return;
2630     }
2631
2632     s->enabled = true;
2633     rstp_port_get_status(rp, &s->port_id, &s->state, &s->role,
2634                          &s->designated_bridge_id, &s->designated_port_id,
2635                          &s->designated_path_cost, &s->tx_count,
2636                          &s->rx_count, &s->error_count, &s->uptime);
2637 }
2638
2639 \f
2640 static int
2641 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2642            size_t n_qdscp)
2643 {
2644     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2645     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2646
2647     if (ofport->n_qdscp != n_qdscp
2648         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2649                               n_qdscp * sizeof *qdscp))) {
2650         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2651         free(ofport->qdscp);
2652         ofport->qdscp = n_qdscp
2653             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2654             : NULL;
2655         ofport->n_qdscp = n_qdscp;
2656     }
2657
2658     return 0;
2659 }
2660 \f
2661 /* Bundles. */
2662
2663 /* Expires all MAC learning entries associated with 'bundle' and forces its
2664  * ofproto to revalidate every flow.
2665  *
2666  * Normally MAC learning entries are removed only from the ofproto associated
2667  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2668  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2669  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2670  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2671  * with the host from which it migrated. */
2672 static void
2673 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2674 {
2675     struct ofproto_dpif *ofproto = bundle->ofproto;
2676     struct mac_learning *ml = ofproto->ml;
2677     struct mac_entry *mac, *next_mac;
2678
2679     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2680     ovs_rwlock_wrlock(&ml->rwlock);
2681     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2682         if (mac_entry_get_port(ml, mac) == bundle) {
2683             if (all_ofprotos) {
2684                 struct ofproto_dpif *o;
2685
2686                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2687                     if (o != ofproto) {
2688                         struct mac_entry *e;
2689
2690                         ovs_rwlock_wrlock(&o->ml->rwlock);
2691                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2692                         if (e) {
2693                             mac_learning_expire(o->ml, e);
2694                         }
2695                         ovs_rwlock_unlock(&o->ml->rwlock);
2696                     }
2697                 }
2698             }
2699
2700             mac_learning_expire(ml, mac);
2701         }
2702     }
2703     ovs_rwlock_unlock(&ml->rwlock);
2704 }
2705
2706 static void
2707 bundle_move(struct ofbundle *old, struct ofbundle *new)
2708 {
2709     struct ofproto_dpif *ofproto = old->ofproto;
2710     struct mac_learning *ml = ofproto->ml;
2711     struct mac_entry *mac, *next_mac;
2712
2713     ovs_assert(new->ofproto == old->ofproto);
2714
2715     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2716     ovs_rwlock_wrlock(&ml->rwlock);
2717     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2718         if (mac_entry_get_port(ml, mac) == old) {
2719             mac_entry_set_port(ml, mac, new);
2720         }
2721     }
2722     ovs_rwlock_unlock(&ml->rwlock);
2723 }
2724
2725 static struct ofbundle *
2726 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2727 {
2728     struct ofbundle *bundle;
2729
2730     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2731                              &ofproto->bundles) {
2732         if (bundle->aux == aux) {
2733             return bundle;
2734         }
2735     }
2736     return NULL;
2737 }
2738
2739 static void
2740 bundle_update(struct ofbundle *bundle)
2741 {
2742     struct ofport_dpif *port;
2743
2744     bundle->floodable = true;
2745     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2746         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2747             || port->is_layer3
2748             || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2749             || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2750             bundle->floodable = false;
2751             break;
2752         }
2753     }
2754 }
2755
2756 static void
2757 bundle_del_port(struct ofport_dpif *port)
2758 {
2759     struct ofbundle *bundle = port->bundle;
2760
2761     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2762
2763     list_remove(&port->bundle_node);
2764     port->bundle = NULL;
2765
2766     if (bundle->lacp) {
2767         lacp_slave_unregister(bundle->lacp, port);
2768     }
2769     if (bundle->bond) {
2770         bond_slave_unregister(bundle->bond, port);
2771     }
2772
2773     bundle_update(bundle);
2774 }
2775
2776 static bool
2777 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2778                 struct lacp_slave_settings *lacp)
2779 {
2780     struct ofport_dpif *port;
2781
2782     port = ofp_port_to_ofport(bundle->ofproto, ofp_port);
2783     if (!port) {
2784         return false;
2785     }
2786
2787     if (port->bundle != bundle) {
2788         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2789         if (port->bundle) {
2790             bundle_remove(&port->up);
2791         }
2792
2793         port->bundle = bundle;
2794         list_push_back(&bundle->ports, &port->bundle_node);
2795         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2796             || port->is_layer3
2797             || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2798             || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2799             bundle->floodable = false;
2800         }
2801     }
2802     if (lacp) {
2803         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2804         lacp_slave_register(bundle->lacp, port, lacp);
2805     }
2806
2807     return true;
2808 }
2809
2810 static void
2811 bundle_destroy(struct ofbundle *bundle)
2812 {
2813     struct ofproto_dpif *ofproto;
2814     struct ofport_dpif *port, *next_port;
2815
2816     if (!bundle) {
2817         return;
2818     }
2819
2820     ofproto = bundle->ofproto;
2821     mbridge_unregister_bundle(ofproto->mbridge, bundle);
2822
2823     xlate_txn_start();
2824     xlate_bundle_remove(bundle);
2825     xlate_txn_commit();
2826
2827     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2828         bundle_del_port(port);
2829     }
2830
2831     bundle_flush_macs(bundle, true);
2832     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2833     free(bundle->name);
2834     free(bundle->trunks);
2835     lacp_unref(bundle->lacp);
2836     bond_unref(bundle->bond);
2837     free(bundle);
2838 }
2839
2840 static int
2841 bundle_set(struct ofproto *ofproto_, void *aux,
2842            const struct ofproto_bundle_settings *s)
2843 {
2844     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2845     bool need_flush = false;
2846     struct ofport_dpif *port;
2847     struct ofbundle *bundle;
2848     unsigned long *trunks;
2849     int vlan;
2850     size_t i;
2851     bool ok;
2852
2853     if (!s) {
2854         bundle_destroy(bundle_lookup(ofproto, aux));
2855         return 0;
2856     }
2857
2858     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2859     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2860
2861     bundle = bundle_lookup(ofproto, aux);
2862     if (!bundle) {
2863         bundle = xmalloc(sizeof *bundle);
2864
2865         bundle->ofproto = ofproto;
2866         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2867                     hash_pointer(aux, 0));
2868         bundle->aux = aux;
2869         bundle->name = NULL;
2870
2871         list_init(&bundle->ports);
2872         bundle->vlan_mode = PORT_VLAN_TRUNK;
2873         bundle->vlan = -1;
2874         bundle->trunks = NULL;
2875         bundle->use_priority_tags = s->use_priority_tags;
2876         bundle->lacp = NULL;
2877         bundle->bond = NULL;
2878
2879         bundle->floodable = true;
2880         mbridge_register_bundle(ofproto->mbridge, bundle);
2881     }
2882
2883     if (!bundle->name || strcmp(s->name, bundle->name)) {
2884         free(bundle->name);
2885         bundle->name = xstrdup(s->name);
2886     }
2887
2888     /* LACP. */
2889     if (s->lacp) {
2890         ofproto->lacp_enabled = true;
2891         if (!bundle->lacp) {
2892             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2893             bundle->lacp = lacp_create();
2894         }
2895         lacp_configure(bundle->lacp, s->lacp);
2896     } else {
2897         lacp_unref(bundle->lacp);
2898         bundle->lacp = NULL;
2899     }
2900
2901     /* Update set of ports. */
2902     ok = true;
2903     for (i = 0; i < s->n_slaves; i++) {
2904         if (!bundle_add_port(bundle, s->slaves[i],
2905                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2906             ok = false;
2907         }
2908     }
2909     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2910         struct ofport_dpif *next_port;
2911
2912         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2913             for (i = 0; i < s->n_slaves; i++) {
2914                 if (s->slaves[i] == port->up.ofp_port) {
2915                     goto found;
2916                 }
2917             }
2918
2919             bundle_del_port(port);
2920         found: ;
2921         }
2922     }
2923     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2924
2925     if (list_is_empty(&bundle->ports)) {
2926         bundle_destroy(bundle);
2927         return EINVAL;
2928     }
2929
2930     /* Set VLAN tagging mode */
2931     if (s->vlan_mode != bundle->vlan_mode
2932         || s->use_priority_tags != bundle->use_priority_tags) {
2933         bundle->vlan_mode = s->vlan_mode;
2934         bundle->use_priority_tags = s->use_priority_tags;
2935         need_flush = true;
2936     }
2937
2938     /* Set VLAN tag. */
2939     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2940             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2941             : 0);
2942     if (vlan != bundle->vlan) {
2943         bundle->vlan = vlan;
2944         need_flush = true;
2945     }
2946
2947     /* Get trunked VLANs. */
2948     switch (s->vlan_mode) {
2949     case PORT_VLAN_ACCESS:
2950         trunks = NULL;
2951         break;
2952
2953     case PORT_VLAN_TRUNK:
2954         trunks = CONST_CAST(unsigned long *, s->trunks);
2955         break;
2956
2957     case PORT_VLAN_NATIVE_UNTAGGED:
2958     case PORT_VLAN_NATIVE_TAGGED:
2959         if (vlan != 0 && (!s->trunks
2960                           || !bitmap_is_set(s->trunks, vlan)
2961                           || bitmap_is_set(s->trunks, 0))) {
2962             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2963             if (s->trunks) {
2964                 trunks = bitmap_clone(s->trunks, 4096);
2965             } else {
2966                 trunks = bitmap_allocate1(4096);
2967             }
2968             bitmap_set1(trunks, vlan);
2969             bitmap_set0(trunks, 0);
2970         } else {
2971             trunks = CONST_CAST(unsigned long *, s->trunks);
2972         }
2973         break;
2974
2975     default:
2976         OVS_NOT_REACHED();
2977     }
2978     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2979         free(bundle->trunks);
2980         if (trunks == s->trunks) {
2981             bundle->trunks = vlan_bitmap_clone(trunks);
2982         } else {
2983             bundle->trunks = trunks;
2984             trunks = NULL;
2985         }
2986         need_flush = true;
2987     }
2988     if (trunks != s->trunks) {
2989         free(trunks);
2990     }
2991
2992     /* Bonding. */
2993     if (!list_is_short(&bundle->ports)) {
2994         bundle->ofproto->has_bonded_bundles = true;
2995         if (bundle->bond) {
2996             if (bond_reconfigure(bundle->bond, s->bond)) {
2997                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2998             }
2999         } else {
3000             bundle->bond = bond_create(s->bond, ofproto);
3001             ofproto->backer->need_revalidate = REV_RECONFIGURE;
3002         }
3003
3004         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
3005             bond_slave_register(bundle->bond, port,
3006                                 port->up.ofp_port, port->up.netdev);
3007         }
3008     } else {
3009         bond_unref(bundle->bond);
3010         bundle->bond = NULL;
3011     }
3012
3013     /* If we changed something that would affect MAC learning, un-learn
3014      * everything on this port and force flow revalidation. */
3015     if (need_flush) {
3016         bundle_flush_macs(bundle, false);
3017     }
3018
3019     return 0;
3020 }
3021
3022 static void
3023 bundle_remove(struct ofport *port_)
3024 {
3025     struct ofport_dpif *port = ofport_dpif_cast(port_);
3026     struct ofbundle *bundle = port->bundle;
3027
3028     if (bundle) {
3029         bundle_del_port(port);
3030         if (list_is_empty(&bundle->ports)) {
3031             bundle_destroy(bundle);
3032         } else if (list_is_short(&bundle->ports)) {
3033             bond_unref(bundle->bond);
3034             bundle->bond = NULL;
3035         }
3036     }
3037 }
3038
3039 static void
3040 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
3041 {
3042     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
3043     struct ofport_dpif *port = port_;
3044     struct eth_addr ea;
3045     int error;
3046
3047     error = netdev_get_etheraddr(port->up.netdev, &ea);
3048     if (!error) {
3049         struct dp_packet packet;
3050         void *packet_pdu;
3051
3052         dp_packet_init(&packet, 0);
3053         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
3054                                  pdu_size);
3055         memcpy(packet_pdu, pdu, pdu_size);
3056
3057         ofproto_dpif_send_packet(port, &packet);
3058         dp_packet_uninit(&packet);
3059     } else {
3060         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
3061                     "%s (%s)", port->bundle->name,
3062                     netdev_get_name(port->up.netdev), ovs_strerror(error));
3063     }
3064 }
3065
3066 static void
3067 bundle_send_learning_packets(struct ofbundle *bundle)
3068 {
3069     struct ofproto_dpif *ofproto = bundle->ofproto;
3070     int error, n_packets, n_errors;
3071     struct mac_entry *e;
3072     struct pkt_list {
3073         struct ovs_list list_node;
3074         struct ofport_dpif *port;
3075         struct dp_packet *pkt;
3076     } *pkt_node;
3077     struct ovs_list packets;
3078
3079     list_init(&packets);
3080     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3081     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3082         if (mac_entry_get_port(ofproto->ml, e) != bundle) {
3083             pkt_node = xmalloc(sizeof *pkt_node);
3084             pkt_node->pkt = bond_compose_learning_packet(bundle->bond,
3085                                                          e->mac, e->vlan,
3086                                                          (void **)&pkt_node->port);
3087             list_push_back(&packets, &pkt_node->list_node);
3088         }
3089     }
3090     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3091
3092     error = n_packets = n_errors = 0;
3093     LIST_FOR_EACH_POP (pkt_node, list_node, &packets) {
3094         int ret;
3095
3096         ret = ofproto_dpif_send_packet(pkt_node->port, pkt_node->pkt);
3097         dp_packet_delete(pkt_node->pkt);
3098         free(pkt_node);
3099         if (ret) {
3100             error = ret;
3101             n_errors++;
3102         }
3103         n_packets++;
3104     }
3105
3106     if (n_errors) {
3107         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3108         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3109                      "packets, last error was: %s",
3110                      bundle->name, n_errors, n_packets, ovs_strerror(error));
3111     } else {
3112         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3113                  bundle->name, n_packets);
3114     }
3115 }
3116
3117 static void
3118 bundle_run(struct ofbundle *bundle)
3119 {
3120     if (bundle->lacp) {
3121         lacp_run(bundle->lacp, send_pdu_cb);
3122     }
3123     if (bundle->bond) {
3124         struct ofport_dpif *port;
3125
3126         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
3127             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
3128         }
3129
3130         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
3131             bundle->ofproto->backer->need_revalidate = REV_BOND;
3132         }
3133
3134         if (bond_should_send_learning_packets(bundle->bond)) {
3135             bundle_send_learning_packets(bundle);
3136         }
3137     }
3138 }
3139
3140 static void
3141 bundle_wait(struct ofbundle *bundle)
3142 {
3143     if (bundle->lacp) {
3144         lacp_wait(bundle->lacp);
3145     }
3146     if (bundle->bond) {
3147         bond_wait(bundle->bond);
3148     }
3149 }
3150 \f
3151 /* Mirrors. */
3152
3153 static int
3154 mirror_set__(struct ofproto *ofproto_, void *aux,
3155              const struct ofproto_mirror_settings *s)
3156 {
3157     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3158     struct ofbundle **srcs, **dsts;
3159     int error;
3160     size_t i;
3161
3162     if (!s) {
3163         mirror_destroy(ofproto->mbridge, aux);
3164         return 0;
3165     }
3166
3167     srcs = xmalloc(s->n_srcs * sizeof *srcs);
3168     dsts = xmalloc(s->n_dsts * sizeof *dsts);
3169
3170     for (i = 0; i < s->n_srcs; i++) {
3171         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
3172     }
3173
3174     for (i = 0; i < s->n_dsts; i++) {
3175         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
3176     }
3177
3178     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
3179                        s->n_dsts, s->src_vlans,
3180                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
3181     free(srcs);
3182     free(dsts);
3183     return error;
3184 }
3185
3186 static int
3187 mirror_get_stats__(struct ofproto *ofproto, void *aux,
3188                    uint64_t *packets, uint64_t *bytes)
3189 {
3190     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
3191                             bytes);
3192 }
3193
3194 static int
3195 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3196 {
3197     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3198     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3199     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3200         mac_learning_flush(ofproto->ml);
3201     }
3202     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3203     return 0;
3204 }
3205
3206 static bool
3207 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3208 {
3209     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3210     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3211     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
3212 }
3213
3214 static void
3215 forward_bpdu_changed(struct ofproto *ofproto_)
3216 {
3217     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3218     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3219 }
3220
3221 static void
3222 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3223                      size_t max_entries)
3224 {
3225     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3226     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3227     mac_learning_set_idle_time(ofproto->ml, idle_time);
3228     mac_learning_set_max_entries(ofproto->ml, max_entries);
3229     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3230 }
3231
3232 /* Configures multicast snooping on 'ofport' using the settings
3233  * defined in 's'. */
3234 static int
3235 set_mcast_snooping(struct ofproto *ofproto_,
3236                    const struct ofproto_mcast_snooping_settings *s)
3237 {
3238     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3239
3240     /* Only revalidate flows if the configuration changed. */
3241     if (!s != !ofproto->ms) {
3242         ofproto->backer->need_revalidate = REV_RECONFIGURE;
3243     }
3244
3245     if (s) {
3246         if (!ofproto->ms) {
3247             ofproto->ms = mcast_snooping_create();
3248         }
3249
3250         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3251         mcast_snooping_set_idle_time(ofproto->ms, s->idle_time);
3252         mcast_snooping_set_max_entries(ofproto->ms, s->max_entries);
3253         if (mcast_snooping_set_flood_unreg(ofproto->ms, s->flood_unreg)) {
3254             ofproto->backer->need_revalidate = REV_RECONFIGURE;
3255         }
3256         ovs_rwlock_unlock(&ofproto->ms->rwlock);
3257     } else {
3258         mcast_snooping_unref(ofproto->ms);
3259         ofproto->ms = NULL;
3260     }
3261
3262     return 0;
3263 }
3264
3265 /* Configures multicast snooping port's flood settings on 'ofproto'. */
3266 static int
3267 set_mcast_snooping_port(struct ofproto *ofproto_, void *aux,
3268                         const struct ofproto_mcast_snooping_port_settings *s)
3269 {
3270     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3271     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3272
3273     if (ofproto->ms && s) {
3274         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3275         mcast_snooping_set_port_flood(ofproto->ms, bundle, s->flood);
3276         mcast_snooping_set_port_flood_reports(ofproto->ms, bundle,
3277                                               s->flood_reports);
3278         ovs_rwlock_unlock(&ofproto->ms->rwlock);
3279     }
3280     return 0;
3281 }
3282
3283 \f
3284 /* Ports. */
3285
3286 struct ofport_dpif *
3287 ofp_port_to_ofport(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
3288 {
3289     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3290     return ofport ? ofport_dpif_cast(ofport) : NULL;
3291 }
3292
3293 static void
3294 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3295                             struct ofproto_port *ofproto_port,
3296                             struct dpif_port *dpif_port)
3297 {
3298     ofproto_port->name = dpif_port->name;
3299     ofproto_port->type = dpif_port->type;
3300     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3301 }
3302
3303 static void
3304 ofport_update_peer(struct ofport_dpif *ofport)
3305 {
3306     const struct ofproto_dpif *ofproto;
3307     struct dpif_backer *backer;
3308     char *peer_name;
3309
3310     if (!netdev_vport_is_patch(ofport->up.netdev)) {
3311         return;
3312     }
3313
3314     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
3315     backer->need_revalidate = REV_RECONFIGURE;
3316
3317     if (ofport->peer) {
3318         ofport->peer->peer = NULL;
3319         ofport->peer = NULL;
3320     }
3321
3322     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
3323     if (!peer_name) {
3324         return;
3325     }
3326
3327     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3328         struct ofport *peer_ofport;
3329         struct ofport_dpif *peer;
3330         char *peer_peer;
3331
3332         if (ofproto->backer != backer) {
3333             continue;
3334         }
3335
3336         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
3337         if (!peer_ofport) {
3338             continue;
3339         }
3340
3341         peer = ofport_dpif_cast(peer_ofport);
3342         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
3343         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
3344                                  peer_peer)) {
3345             ofport->peer = peer;
3346             ofport->peer->peer = ofport;
3347         }
3348         free(peer_peer);
3349
3350         break;
3351     }
3352     free(peer_name);
3353 }
3354
3355 static void
3356 port_run(struct ofport_dpif *ofport)
3357 {
3358     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3359     bool carrier_changed = carrier_seq != ofport->carrier_seq;
3360     bool enable = netdev_get_carrier(ofport->up.netdev);
3361     bool cfm_enable = false;
3362     bool bfd_enable = false;
3363
3364     ofport->carrier_seq = carrier_seq;
3365
3366     if (ofport->cfm) {
3367         int cfm_opup = cfm_get_opup(ofport->cfm);
3368
3369         cfm_enable = !cfm_get_fault(ofport->cfm);
3370
3371         if (cfm_opup >= 0) {
3372             cfm_enable = cfm_enable && cfm_opup;
3373         }
3374     }
3375
3376     if (ofport->bfd) {
3377         bfd_enable = bfd_forwarding(ofport->bfd);
3378     }
3379
3380     if (ofport->bfd || ofport->cfm) {
3381         enable = enable && (cfm_enable || bfd_enable);
3382     }
3383
3384     if (ofport->bundle) {
3385         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3386         if (carrier_changed) {
3387             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3388         }
3389     }
3390
3391     if (ofport->may_enable != enable) {
3392         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3393
3394         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3395
3396         if (ofport->rstp_port) {
3397             rstp_port_set_mac_operational(ofport->rstp_port, enable);
3398         }
3399     }
3400
3401     ofport->may_enable = enable;
3402 }
3403
3404 static int
3405 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3406                    struct ofproto_port *ofproto_port)
3407 {
3408     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3409     struct dpif_port dpif_port;
3410     int error;
3411
3412     if (sset_contains(&ofproto->ghost_ports, devname)) {
3413         const char *type = netdev_get_type_from_name(devname);
3414
3415         /* We may be called before ofproto->up.port_by_name is populated with
3416          * the appropriate ofport.  For this reason, we must get the name and
3417          * type from the netdev layer directly. */
3418         if (type) {
3419             const struct ofport *ofport;
3420
3421             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3422             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3423             ofproto_port->name = xstrdup(devname);
3424             ofproto_port->type = xstrdup(type);
3425             return 0;
3426         }
3427         return ENODEV;
3428     }
3429
3430     if (!sset_contains(&ofproto->ports, devname)) {
3431         return ENODEV;
3432     }
3433     error = dpif_port_query_by_name(ofproto->backer->dpif,
3434                                     devname, &dpif_port);
3435     if (!error) {
3436         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3437     }
3438     return error;
3439 }
3440
3441 static int
3442 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3443 {
3444     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3445     const char *devname = netdev_get_name(netdev);
3446     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3447     const char *dp_port_name;
3448
3449     if (netdev_vport_is_patch(netdev)) {
3450         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3451         return 0;
3452     }
3453
3454     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3455     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3456         odp_port_t port_no = ODPP_NONE;
3457         int error;
3458
3459         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3460         if (error) {
3461             return error;
3462         }
3463         if (netdev_get_tunnel_config(netdev)) {
3464             simap_put(&ofproto->backer->tnl_backers,
3465                       dp_port_name, odp_to_u32(port_no));
3466         }
3467     }
3468
3469     if (netdev_get_tunnel_config(netdev)) {
3470         sset_add(&ofproto->ghost_ports, devname);
3471     } else {
3472         sset_add(&ofproto->ports, devname);
3473     }
3474     return 0;
3475 }
3476
3477 static int
3478 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3479 {
3480     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3481     struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
3482     int error = 0;
3483
3484     if (!ofport) {
3485         return 0;
3486     }
3487
3488     sset_find_and_delete(&ofproto->ghost_ports,
3489                          netdev_get_name(ofport->up.netdev));
3490     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3491     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3492         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3493         if (!error) {
3494             /* The caller is going to close ofport->up.netdev.  If this is a
3495              * bonded port, then the bond is using that netdev, so remove it
3496              * from the bond.  The client will need to reconfigure everything
3497              * after deleting ports, so then the slave will get re-added. */
3498             bundle_remove(&ofport->up);
3499         }
3500     }
3501     return error;
3502 }
3503
3504 static int
3505 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3506 {
3507     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3508     int error;
3509
3510     error = netdev_get_stats(ofport->up.netdev, stats);
3511
3512     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3513         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3514
3515         ovs_mutex_lock(&ofproto->stats_mutex);
3516         /* ofproto->stats.tx_packets represents packets that we created
3517          * internally and sent to some port (e.g. packets sent with
3518          * ofproto_dpif_send_packet()).  Account for them as if they had
3519          * come from OFPP_LOCAL and got forwarded. */
3520
3521         if (stats->rx_packets != UINT64_MAX) {
3522             stats->rx_packets += ofproto->stats.tx_packets;
3523         }
3524
3525         if (stats->rx_bytes != UINT64_MAX) {
3526             stats->rx_bytes += ofproto->stats.tx_bytes;
3527         }
3528
3529         /* ofproto->stats.rx_packets represents packets that were received on
3530          * some port and we processed internally and dropped (e.g. STP).
3531          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3532
3533         if (stats->tx_packets != UINT64_MAX) {
3534             stats->tx_packets += ofproto->stats.rx_packets;
3535         }
3536
3537         if (stats->tx_bytes != UINT64_MAX) {
3538             stats->tx_bytes += ofproto->stats.rx_bytes;
3539         }
3540         ovs_mutex_unlock(&ofproto->stats_mutex);
3541     }
3542
3543     return error;
3544 }
3545
3546 static int
3547 port_get_lacp_stats(const struct ofport *ofport_, struct lacp_slave_stats *stats)
3548 {
3549     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3550     if (ofport->bundle && ofport->bundle->lacp) {
3551         if (lacp_get_slave_stats(ofport->bundle->lacp, ofport, stats)) {
3552             return 0;
3553         }
3554     }
3555     return -1;
3556 }
3557
3558 struct port_dump_state {
3559     uint32_t bucket;
3560     uint32_t offset;
3561     bool ghost;
3562
3563     struct ofproto_port port;
3564     bool has_port;
3565 };
3566
3567 static int
3568 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3569 {
3570     *statep = xzalloc(sizeof(struct port_dump_state));
3571     return 0;
3572 }
3573
3574 static int
3575 port_dump_next(const struct ofproto *ofproto_, void *state_,
3576                struct ofproto_port *port)
3577 {
3578     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3579     struct port_dump_state *state = state_;
3580     const struct sset *sset;
3581     struct sset_node *node;
3582
3583     if (state->has_port) {
3584         ofproto_port_destroy(&state->port);
3585         state->has_port = false;
3586     }
3587     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3588     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3589         int error;
3590
3591         error = port_query_by_name(ofproto_, node->name, &state->port);
3592         if (!error) {
3593             *port = state->port;
3594             state->has_port = true;
3595             return 0;
3596         } else if (error != ENODEV) {
3597             return error;
3598         }
3599     }
3600
3601     if (!state->ghost) {
3602         state->ghost = true;
3603         state->bucket = 0;
3604         state->offset = 0;
3605         return port_dump_next(ofproto_, state_, port);
3606     }
3607
3608     return EOF;
3609 }
3610
3611 static int
3612 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3613 {
3614     struct port_dump_state *state = state_;
3615
3616     if (state->has_port) {
3617         ofproto_port_destroy(&state->port);
3618     }
3619     free(state);
3620     return 0;
3621 }
3622
3623 static int
3624 port_poll(const struct ofproto *ofproto_, char **devnamep)
3625 {
3626     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3627
3628     if (ofproto->port_poll_errno) {
3629         int error = ofproto->port_poll_errno;
3630         ofproto->port_poll_errno = 0;
3631         return error;
3632     }
3633
3634     if (sset_is_empty(&ofproto->port_poll_set)) {
3635         return EAGAIN;
3636     }
3637
3638     *devnamep = sset_pop(&ofproto->port_poll_set);
3639     return 0;
3640 }
3641
3642 static void
3643 port_poll_wait(const struct ofproto *ofproto_)
3644 {
3645     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3646     dpif_port_poll_wait(ofproto->backer->dpif);
3647 }
3648
3649 static int
3650 port_is_lacp_current(const struct ofport *ofport_)
3651 {
3652     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3653     return (ofport->bundle && ofport->bundle->lacp
3654             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3655             : -1);
3656 }
3657 \f
3658 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3659  * then delete it entirely. */
3660 static void
3661 rule_expire(struct rule_dpif *rule)
3662     OVS_REQUIRES(ofproto_mutex)
3663 {
3664     uint16_t hard_timeout, idle_timeout;
3665     long long int now = time_msec();
3666     int reason = -1;
3667
3668     hard_timeout = rule->up.hard_timeout;
3669     idle_timeout = rule->up.idle_timeout;
3670
3671     /* Has 'rule' expired? */
3672     if (hard_timeout) {
3673         long long int modified;
3674
3675         ovs_mutex_lock(&rule->up.mutex);
3676         modified = rule->up.modified;
3677         ovs_mutex_unlock(&rule->up.mutex);
3678
3679         if (now > modified + hard_timeout * 1000) {
3680             reason = OFPRR_HARD_TIMEOUT;
3681         }
3682     }
3683
3684     if (reason < 0 && idle_timeout) {
3685         long long int used;
3686
3687         ovs_mutex_lock(&rule->stats_mutex);
3688         used = rule->stats.used;
3689         ovs_mutex_unlock(&rule->stats_mutex);
3690
3691         if (now > used + idle_timeout * 1000) {
3692             reason = OFPRR_IDLE_TIMEOUT;
3693         }
3694     }
3695
3696     if (reason >= 0) {
3697         COVERAGE_INC(ofproto_dpif_expired);
3698         ofproto_rule_expire(&rule->up, reason);
3699     }
3700 }
3701
3702 int
3703 ofproto_dpif_execute_actions__(struct ofproto_dpif *ofproto,
3704                                const struct flow *flow,
3705                                struct rule_dpif *rule,
3706                                const struct ofpact *ofpacts, size_t ofpacts_len,
3707                                int recurse, int resubmits,
3708                                struct dp_packet *packet)
3709 {
3710     struct dpif_flow_stats stats;
3711     struct xlate_out xout;
3712     struct xlate_in xin;
3713     ofp_port_t in_port;
3714     struct dpif_execute execute;
3715     int error;
3716
3717     ovs_assert((rule != NULL) != (ofpacts != NULL));
3718
3719     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3720
3721     if (rule) {
3722         rule_dpif_credit_stats(rule, &stats);
3723     }
3724
3725     uint64_t odp_actions_stub[1024 / 8];
3726     struct ofpbuf odp_actions = OFPBUF_STUB_INITIALIZER(odp_actions_stub);
3727     xlate_in_init(&xin, ofproto, flow, flow->in_port.ofp_port, rule,
3728                   stats.tcp_flags, packet, NULL, &odp_actions);
3729     xin.ofpacts = ofpacts;
3730     xin.ofpacts_len = ofpacts_len;
3731     xin.resubmit_stats = &stats;
3732     xin.recurse = recurse;
3733     xin.resubmits = resubmits;
3734     if (xlate_actions(&xin, &xout) != XLATE_OK) {
3735         error = EINVAL;
3736         goto out;
3737     }
3738
3739     execute.actions = odp_actions.data;
3740     execute.actions_len = odp_actions.size;
3741
3742     pkt_metadata_from_flow(&packet->md, flow);
3743     execute.packet = packet;
3744     execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3745     execute.probe = false;
3746     execute.mtu = 0;
3747
3748     /* Fix up in_port. */
3749     in_port = flow->in_port.ofp_port;
3750     if (in_port == OFPP_NONE) {
3751         in_port = OFPP_LOCAL;
3752     }
3753     execute.packet->md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3754
3755     error = dpif_execute(ofproto->backer->dpif, &execute);
3756 out:
3757     xlate_out_uninit(&xout);
3758     ofpbuf_uninit(&odp_actions);
3759
3760     return error;
3761 }
3762
3763 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3764  * 'flow' must reflect the data in 'packet'. */
3765 int
3766 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3767                              const struct flow *flow,
3768                              struct rule_dpif *rule,
3769                              const struct ofpact *ofpacts, size_t ofpacts_len,
3770                              struct dp_packet *packet)
3771 {
3772     return ofproto_dpif_execute_actions__(ofproto, flow, rule, ofpacts,
3773                                           ofpacts_len, 0, 0, packet);
3774 }
3775
3776 void
3777 rule_dpif_credit_stats(struct rule_dpif *rule,
3778                        const struct dpif_flow_stats *stats)
3779 {
3780     ovs_mutex_lock(&rule->stats_mutex);
3781     if (OVS_UNLIKELY(rule->new_rule)) {
3782         rule_dpif_credit_stats(rule->new_rule, stats);
3783     } else {
3784         rule->stats.n_packets += stats->n_packets;
3785         rule->stats.n_bytes += stats->n_bytes;
3786         rule->stats.used = MAX(rule->stats.used, stats->used);
3787     }
3788     ovs_mutex_unlock(&rule->stats_mutex);
3789 }
3790
3791 ovs_be64
3792 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3793     OVS_REQUIRES(rule->up.mutex)
3794 {
3795     return rule->up.flow_cookie;
3796 }
3797
3798 void
3799 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3800                      uint16_t hard_timeout)
3801 {
3802     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3803 }
3804
3805 /* Returns 'rule''s actions.  The returned actions are RCU-protected, and can
3806  * be read until the calling thread quiesces. */
3807 const struct rule_actions *
3808 rule_dpif_get_actions(const struct rule_dpif *rule)
3809 {
3810     return rule_get_actions(&rule->up);
3811 }
3812
3813 /* Sets 'rule''s recirculation id. */
3814 static void
3815 rule_dpif_set_recirc_id(struct rule_dpif *rule, uint32_t id)
3816     OVS_REQUIRES(rule->up.mutex)
3817 {
3818     ovs_assert(!rule->recirc_id || rule->recirc_id == id);
3819     if (rule->recirc_id == id) {
3820         /* Release the new reference to the same id. */
3821         recirc_free_id(id);
3822     } else {
3823         rule->recirc_id = id;
3824     }
3825 }
3826
3827 /* Sets 'rule''s recirculation id. */
3828 void
3829 rule_set_recirc_id(struct rule *rule_, uint32_t id)
3830 {
3831     struct rule_dpif *rule = rule_dpif_cast(rule_);
3832
3833     ovs_mutex_lock(&rule->up.mutex);
3834     rule_dpif_set_recirc_id(rule, id);
3835     ovs_mutex_unlock(&rule->up.mutex);
3836 }
3837
3838 cls_version_t
3839 ofproto_dpif_get_tables_version(struct ofproto_dpif *ofproto OVS_UNUSED)
3840 {
3841     cls_version_t version;
3842
3843     atomic_read_relaxed(&ofproto->tables_version, &version);
3844
3845     return version;
3846 }
3847
3848 /* The returned rule (if any) is valid at least until the next RCU quiescent
3849  * period.  If the rule needs to stay around longer, the caller should take
3850  * a reference.
3851  *
3852  * 'flow' is non-const to allow for temporary modifications during the lookup.
3853  * Any changes are restored before returning. */
3854 static struct rule_dpif *
3855 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, cls_version_t version,
3856                           uint8_t table_id, struct flow *flow,
3857                           struct flow_wildcards *wc)
3858 {
3859     struct classifier *cls = &ofproto->up.tables[table_id].cls;
3860     return rule_dpif_cast(rule_from_cls_rule(classifier_lookup(cls, version,
3861                                                                flow, wc)));
3862 }
3863
3864 /* Look up 'flow' in 'ofproto''s classifier version 'version', starting from
3865  * table '*table_id'.  Returns the rule that was found, which may be one of the
3866  * special rules according to packet miss hadling.  If 'may_packet_in' is
3867  * false, returning of the miss_rule (which issues packet ins for the
3868  * controller) is avoided.  Updates 'wc', if nonnull, to reflect the fields
3869  * that were used during the lookup.
3870  *
3871  * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3872  * if none is found then the table miss configuration for that table is
3873  * honored, which can result in additional lookups in other OpenFlow tables.
3874  * In this case the function updates '*table_id' to reflect the final OpenFlow
3875  * table that was searched.
3876  *
3877  * If 'honor_table_miss' is false, then only one table lookup occurs, in
3878  * '*table_id'.
3879  *
3880  * The rule is returned in '*rule', which is valid at least until the next
3881  * RCU quiescent period.  If the '*rule' needs to stay around longer, the
3882  * caller must take a reference.
3883  *
3884  * 'in_port' allows the lookup to take place as if the in port had the value
3885  * 'in_port'.  This is needed for resubmit action support.
3886  *
3887  * 'flow' is non-const to allow for temporary modifications during the lookup.
3888  * Any changes are restored before returning. */
3889 struct rule_dpif *
3890 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3891                             cls_version_t version, struct flow *flow,
3892                             struct flow_wildcards *wc,
3893                             const struct dpif_flow_stats *stats,
3894                             uint8_t *table_id, ofp_port_t in_port,
3895                             bool may_packet_in, bool honor_table_miss)
3896 {
3897     ovs_be16 old_tp_src = flow->tp_src, old_tp_dst = flow->tp_dst;
3898     ofp_port_t old_in_port = flow->in_port.ofp_port;
3899     enum ofputil_table_miss miss_config;
3900     struct rule_dpif *rule;
3901     uint8_t next_id;
3902
3903     /* We always unwildcard nw_frag (for IP), so they
3904      * need not be unwildcarded here. */
3905     if (flow->nw_frag & FLOW_NW_FRAG_ANY
3906         && ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3907         if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3908             /* We must pretend that transport ports are unavailable. */
3909             flow->tp_src = htons(0);
3910             flow->tp_dst = htons(0);
3911         } else {
3912             /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM).
3913              * Use the drop_frags_rule (which cannot disappear). */
3914             rule = ofproto->drop_frags_rule;
3915             if (stats) {
3916                 struct oftable *tbl = &ofproto->up.tables[*table_id];
3917                 unsigned long orig;
3918
3919                 atomic_add_relaxed(&tbl->n_matched, stats->n_packets, &orig);
3920             }
3921             return rule;
3922         }
3923     }
3924
3925     /* Look up a flow with 'in_port' as the input port.  Then restore the
3926      * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
3927      * have surprising behavior). */
3928     flow->in_port.ofp_port = in_port;
3929
3930     /* Our current implementation depends on n_tables == N_TABLES, and
3931      * TBL_INTERNAL being the last table. */
3932     BUILD_ASSERT_DECL(N_TABLES == TBL_INTERNAL + 1);
3933
3934     miss_config = OFPUTIL_TABLE_MISS_CONTINUE;
3935
3936     for (next_id = *table_id;
3937          next_id < ofproto->up.n_tables;
3938          next_id++, next_id += (next_id == TBL_INTERNAL))
3939     {
3940         *table_id = next_id;
3941         rule = rule_dpif_lookup_in_table(ofproto, version, next_id, flow, wc);
3942         if (stats) {
3943             struct oftable *tbl = &ofproto->up.tables[next_id];
3944             unsigned long orig;
3945
3946             atomic_add_relaxed(rule ? &tbl->n_matched : &tbl->n_missed,
3947                                stats->n_packets, &orig);
3948         }
3949         if (rule) {
3950             goto out;   /* Match. */
3951         }
3952         if (honor_table_miss) {
3953             miss_config = ofproto_table_get_miss_config(&ofproto->up,
3954                                                         *table_id);
3955             if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE) {
3956                 continue;
3957             }
3958         }
3959         break;
3960     }
3961     /* Miss. */
3962     rule = ofproto->no_packet_in_rule;
3963     if (may_packet_in) {
3964         if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE
3965             || miss_config == OFPUTIL_TABLE_MISS_CONTROLLER) {
3966             struct ofport_dpif *port;
3967
3968             port = ofp_port_to_ofport(ofproto, old_in_port);
3969             if (!port) {
3970                 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3971                              old_in_port);
3972             } else if (!(port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN)) {
3973                 rule = ofproto->miss_rule;
3974             }
3975         } else if (miss_config == OFPUTIL_TABLE_MISS_DEFAULT &&
3976                    connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3977             rule = ofproto->miss_rule;
3978         }
3979     }
3980 out:
3981     /* Restore port numbers, as they may have been modified above. */
3982     flow->tp_src = old_tp_src;
3983     flow->tp_dst = old_tp_dst;
3984     /* Restore the old in port. */
3985     flow->in_port.ofp_port = old_in_port;
3986
3987     return rule;
3988 }
3989
3990 static void
3991 complete_operation(struct rule_dpif *rule)
3992     OVS_REQUIRES(ofproto_mutex)
3993 {
3994     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3995
3996     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3997 }
3998
3999 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4000 {
4001     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4002 }
4003
4004 static struct rule *
4005 rule_alloc(void)
4006 {
4007     struct rule_dpif *rule = xzalloc(sizeof *rule);
4008     return &rule->up;
4009 }
4010
4011 static void
4012 rule_dealloc(struct rule *rule_)
4013 {
4014     struct rule_dpif *rule = rule_dpif_cast(rule_);
4015     free(rule);
4016 }
4017
4018 static enum ofperr
4019 check_mask(struct ofproto_dpif *ofproto, const struct miniflow *flow)
4020 {
4021     const struct odp_support *support;
4022     uint16_t ct_state, ct_zone;
4023     ovs_u128 ct_label;
4024     uint32_t ct_mark;
4025
4026     support = &ofproto_dpif_get_support(ofproto)->odp;
4027     ct_state = MINIFLOW_GET_U16(flow, ct_state);
4028     if (support->ct_state && support->ct_zone && support->ct_mark
4029         && support->ct_label && support->ct_state_nat) {
4030         return ct_state & CS_UNSUPPORTED_MASK ? OFPERR_OFPBMC_BAD_MASK : 0;
4031     }
4032
4033     ct_zone = MINIFLOW_GET_U16(flow, ct_zone);
4034     ct_mark = MINIFLOW_GET_U32(flow, ct_mark);
4035     ct_label = MINIFLOW_GET_U128(flow, ct_label);
4036
4037     if ((ct_state && !support->ct_state)
4038         || (ct_state & CS_UNSUPPORTED_MASK)
4039         || ((ct_state & (CS_SRC_NAT | CS_DST_NAT)) && !support->ct_state_nat)
4040         || (ct_zone && !support->ct_zone)
4041         || (ct_mark && !support->ct_mark)
4042         || (!ovs_u128_is_zero(&ct_label) && !support->ct_label)) {
4043         return OFPERR_OFPBMC_BAD_MASK;
4044     }
4045
4046     return 0;
4047 }
4048
4049 static enum ofperr
4050 check_actions(const struct ofproto_dpif *ofproto,
4051               const struct rule_actions *const actions)
4052 {
4053     const struct ofpact *ofpact;
4054
4055     OFPACT_FOR_EACH (ofpact, actions->ofpacts, actions->ofpacts_len) {
4056         const struct odp_support *support;
4057         const struct ofpact_conntrack *ct;
4058         const struct ofpact *a;
4059
4060         if (ofpact->type != OFPACT_CT) {
4061             continue;
4062         }
4063
4064         ct = CONTAINER_OF(ofpact, struct ofpact_conntrack, ofpact);
4065         support = &ofproto_dpif_get_support(ofproto)->odp;
4066
4067         if (!support->ct_state) {
4068             return OFPERR_OFPBAC_BAD_TYPE;
4069         }
4070         if ((ct->zone_imm || ct->zone_src.field) && !support->ct_zone) {
4071             return OFPERR_OFPBAC_BAD_ARGUMENT;
4072         }
4073
4074         OFPACT_FOR_EACH(a, ct->actions, ofpact_ct_get_action_len(ct)) {
4075             const struct mf_field *dst = ofpact_get_mf_dst(a);
4076
4077             if (a->type == OFPACT_NAT && !support->ct_state_nat) {
4078                 /* The backer doesn't seem to support the NAT bits in
4079                  * 'ct_state': assume that it doesn't support the NAT
4080                  * action. */
4081                 return OFPERR_OFPBAC_BAD_TYPE;
4082             }
4083             if (dst && ((dst->id == MFF_CT_MARK && !support->ct_mark)
4084                         || (dst->id == MFF_CT_LABEL && !support->ct_label))) {
4085                 return OFPERR_OFPBAC_BAD_SET_ARGUMENT;
4086             }
4087         }
4088     }
4089
4090     return 0;
4091 }
4092
4093 static enum ofperr
4094 rule_check(struct rule *rule)
4095 {
4096     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->ofproto);
4097     enum ofperr err;
4098
4099     err = check_mask(ofproto, &rule->cr.match.mask->masks);
4100     if (err) {
4101         return err;
4102     }
4103     return check_actions(ofproto, rule->actions);
4104 }
4105
4106 static enum ofperr
4107 rule_construct(struct rule *rule_)
4108     OVS_NO_THREAD_SAFETY_ANALYSIS
4109 {
4110     struct rule_dpif *rule = rule_dpif_cast(rule_);
4111     int error;
4112
4113     error = rule_check(rule_);
4114     if (error) {
4115         return error;
4116     }
4117
4118     ovs_mutex_init_adaptive(&rule->stats_mutex);
4119     rule->stats.n_packets = 0;
4120     rule->stats.n_bytes = 0;
4121     rule->stats.used = rule->up.modified;
4122     rule->recirc_id = 0;
4123     rule->new_rule = NULL;
4124
4125     return 0;
4126 }
4127
4128 static void
4129 rule_insert(struct rule *rule_, struct rule *old_rule_, bool forward_stats)
4130     OVS_REQUIRES(ofproto_mutex)
4131 {
4132     struct rule_dpif *rule = rule_dpif_cast(rule_);
4133
4134     if (old_rule_ && forward_stats) {
4135         struct rule_dpif *old_rule = rule_dpif_cast(old_rule_);
4136
4137         ovs_assert(!old_rule->new_rule);
4138
4139         /* Take a reference to the new rule, and refer all stats updates from
4140          * the old rule to the new rule. */
4141         rule_dpif_ref(rule);
4142
4143         ovs_mutex_lock(&old_rule->stats_mutex);
4144         ovs_mutex_lock(&rule->stats_mutex);
4145         old_rule->new_rule = rule;       /* Forward future stats. */
4146         rule->stats = old_rule->stats;   /* Transfer stats to the new rule. */
4147         ovs_mutex_unlock(&rule->stats_mutex);
4148         ovs_mutex_unlock(&old_rule->stats_mutex);
4149     }
4150
4151     complete_operation(rule);
4152 }
4153
4154 static void
4155 rule_delete(struct rule *rule_)
4156     OVS_REQUIRES(ofproto_mutex)
4157 {
4158     struct rule_dpif *rule = rule_dpif_cast(rule_);
4159     complete_operation(rule);
4160 }
4161
4162 static void
4163 rule_destruct(struct rule *rule_)
4164     OVS_NO_THREAD_SAFETY_ANALYSIS
4165 {
4166     struct rule_dpif *rule = rule_dpif_cast(rule_);
4167
4168     ovs_mutex_destroy(&rule->stats_mutex);
4169     /* Release reference to the new rule, if any. */
4170     if (rule->new_rule) {
4171         rule_dpif_unref(rule->new_rule);
4172     }
4173     if (rule->recirc_id) {
4174         recirc_free_id(rule->recirc_id);
4175     }
4176 }
4177
4178 static void
4179 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
4180                long long int *used)
4181 {
4182     struct rule_dpif *rule = rule_dpif_cast(rule_);
4183
4184     ovs_mutex_lock(&rule->stats_mutex);
4185     if (OVS_UNLIKELY(rule->new_rule)) {
4186         rule_get_stats(&rule->new_rule->up, packets, bytes, used);
4187     } else {
4188         *packets = rule->stats.n_packets;
4189         *bytes = rule->stats.n_bytes;
4190         *used = rule->stats.used;
4191     }
4192     ovs_mutex_unlock(&rule->stats_mutex);
4193 }
4194
4195 static void
4196 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4197                   struct dp_packet *packet)
4198 {
4199     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4200
4201     ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
4202 }
4203
4204 static enum ofperr
4205 rule_execute(struct rule *rule, const struct flow *flow,
4206              struct dp_packet *packet)
4207 {
4208     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4209     dp_packet_delete(packet);
4210     return 0;
4211 }
4212
4213 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4214 {
4215     return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4216 }
4217
4218 static struct ofgroup *
4219 group_alloc(void)
4220 {
4221     struct group_dpif *group = xzalloc(sizeof *group);
4222     return &group->up;
4223 }
4224
4225 static void
4226 group_dealloc(struct ofgroup *group_)
4227 {
4228     struct group_dpif *group = group_dpif_cast(group_);
4229     free(group);
4230 }
4231
4232 static void
4233 group_construct_stats(struct group_dpif *group)
4234     OVS_REQUIRES(group->stats_mutex)
4235 {
4236     struct ofputil_bucket *bucket;
4237     const struct ovs_list *buckets;
4238
4239     group->packet_count = 0;
4240     group->byte_count = 0;
4241
4242     group_dpif_get_buckets(group, &buckets);
4243     LIST_FOR_EACH (bucket, list_node, buckets) {
4244         bucket->stats.packet_count = 0;
4245         bucket->stats.byte_count = 0;
4246     }
4247 }
4248
4249 void
4250 group_dpif_credit_stats(struct group_dpif *group,
4251                         struct ofputil_bucket *bucket,
4252                         const struct dpif_flow_stats *stats)
4253 {
4254     ovs_mutex_lock(&group->stats_mutex);
4255     group->packet_count += stats->n_packets;
4256     group->byte_count += stats->n_bytes;
4257     if (bucket) {
4258         bucket->stats.packet_count += stats->n_packets;
4259         bucket->stats.byte_count += stats->n_bytes;
4260     } else { /* Credit to all buckets */
4261         const struct ovs_list *buckets;
4262
4263         group_dpif_get_buckets(group, &buckets);
4264         LIST_FOR_EACH (bucket, list_node, buckets) {
4265             bucket->stats.packet_count += stats->n_packets;
4266             bucket->stats.byte_count += stats->n_bytes;
4267         }
4268     }
4269     ovs_mutex_unlock(&group->stats_mutex);
4270 }
4271
4272 static enum ofperr
4273 group_construct(struct ofgroup *group_)
4274 {
4275     struct group_dpif *group = group_dpif_cast(group_);
4276
4277     ovs_mutex_init_adaptive(&group->stats_mutex);
4278     ovs_mutex_lock(&group->stats_mutex);
4279     group_construct_stats(group);
4280     ovs_mutex_unlock(&group->stats_mutex);
4281     return 0;
4282 }
4283
4284 static void
4285 group_destruct(struct ofgroup *group_)
4286 {
4287     struct group_dpif *group = group_dpif_cast(group_);
4288     ovs_mutex_destroy(&group->stats_mutex);
4289 }
4290
4291 static enum ofperr
4292 group_modify(struct ofgroup *group_)
4293 {
4294     struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
4295
4296     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4297
4298     return 0;
4299 }
4300
4301 static enum ofperr
4302 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4303 {
4304     struct group_dpif *group = group_dpif_cast(group_);
4305     struct ofputil_bucket *bucket;
4306     const struct ovs_list *buckets;
4307     struct bucket_counter *bucket_stats;
4308
4309     ovs_mutex_lock(&group->stats_mutex);
4310     ogs->packet_count = group->packet_count;
4311     ogs->byte_count = group->byte_count;
4312
4313     group_dpif_get_buckets(group, &buckets);
4314     bucket_stats = ogs->bucket_stats;
4315     LIST_FOR_EACH (bucket, list_node, buckets) {
4316         bucket_stats->packet_count = bucket->stats.packet_count;
4317         bucket_stats->byte_count = bucket->stats.byte_count;
4318         bucket_stats++;
4319     }
4320     ovs_mutex_unlock(&group->stats_mutex);
4321
4322     return 0;
4323 }
4324
4325 /* If the group exists, this function increments the groups's reference count.
4326  *
4327  * Make sure to call group_dpif_unref() after no longer needing to maintain
4328  * a reference to the group. */
4329 bool
4330 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4331                   struct group_dpif **group)
4332 {
4333     struct ofgroup *ofgroup;
4334     bool found;
4335
4336     found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4337     *group = found ?  group_dpif_cast(ofgroup) : NULL;
4338
4339     return found;
4340 }
4341
4342 void
4343 group_dpif_get_buckets(const struct group_dpif *group,
4344                        const struct ovs_list **buckets)
4345 {
4346     *buckets = &group->up.buckets;
4347 }
4348
4349 enum ofp11_group_type
4350 group_dpif_get_type(const struct group_dpif *group)
4351 {
4352     return group->up.type;
4353 }
4354
4355 const char *
4356 group_dpif_get_selection_method(const struct group_dpif *group)
4357 {
4358     return group->up.props.selection_method;
4359 }
4360 \f
4361 /* Sends 'packet' out 'ofport'.
4362  * May modify 'packet'.
4363  * Returns 0 if successful, otherwise a positive errno value. */
4364 int
4365 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct dp_packet *packet)
4366 {
4367     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4368     int error;
4369
4370     error = xlate_send_packet(ofport, packet);
4371
4372     ovs_mutex_lock(&ofproto->stats_mutex);
4373     ofproto->stats.tx_packets++;
4374     ofproto->stats.tx_bytes += dp_packet_size(packet);
4375     ovs_mutex_unlock(&ofproto->stats_mutex);
4376     return error;
4377 }
4378
4379 uint64_t
4380 group_dpif_get_selection_method_param(const struct group_dpif *group)
4381 {
4382     return group->up.props.selection_method_param;
4383 }
4384
4385 const struct field_array *
4386 group_dpif_get_fields(const struct group_dpif *group)
4387 {
4388     return &group->up.props.fields;
4389 }
4390 \f
4391 /* Return the version string of the datapath that backs up
4392  * this 'ofproto'.
4393  */
4394 static const char *
4395 get_datapath_version(const struct ofproto *ofproto_)
4396 {
4397     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4398
4399     return ofproto->backer->dp_version_string;
4400 }
4401
4402 static bool
4403 set_frag_handling(struct ofproto *ofproto_,
4404                   enum ofp_config_flags frag_handling)
4405 {
4406     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4407     if (frag_handling != OFPC_FRAG_REASM) {
4408         ofproto->backer->need_revalidate = REV_RECONFIGURE;
4409         return true;
4410     } else {
4411         return false;
4412     }
4413 }
4414
4415 static enum ofperr
4416 packet_out(struct ofproto *ofproto_, struct dp_packet *packet,
4417            const struct flow *flow,
4418            const struct ofpact *ofpacts, size_t ofpacts_len)
4419 {
4420     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4421
4422     ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
4423                                  ofpacts_len, packet);
4424     return 0;
4425 }
4426 \f
4427 /* NetFlow. */
4428
4429 static int
4430 set_netflow(struct ofproto *ofproto_,
4431             const struct netflow_options *netflow_options)
4432 {
4433     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4434
4435     if (netflow_options) {
4436         if (!ofproto->netflow) {
4437             ofproto->netflow = netflow_create();
4438             ofproto->backer->need_revalidate = REV_RECONFIGURE;
4439         }
4440         return netflow_set_options(ofproto->netflow, netflow_options);
4441     } else if (ofproto->netflow) {
4442         ofproto->backer->need_revalidate = REV_RECONFIGURE;
4443         netflow_unref(ofproto->netflow);
4444         ofproto->netflow = NULL;
4445     }
4446
4447     return 0;
4448 }
4449
4450 static void
4451 get_netflow_ids(const struct ofproto *ofproto_,
4452                 uint8_t *engine_type, uint8_t *engine_id)
4453 {
4454     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4455
4456     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
4457 }
4458 \f
4459 static struct ofproto_dpif *
4460 ofproto_dpif_lookup(const char *name)
4461 {
4462     struct ofproto_dpif *ofproto;
4463
4464     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
4465                              hash_string(name, 0), &all_ofproto_dpifs) {
4466         if (!strcmp(ofproto->up.name, name)) {
4467             return ofproto;
4468         }
4469     }
4470     return NULL;
4471 }
4472
4473 static void
4474 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
4475                           const char *argv[], void *aux OVS_UNUSED)
4476 {
4477     struct ofproto_dpif *ofproto;
4478
4479     if (argc > 1) {
4480         ofproto = ofproto_dpif_lookup(argv[1]);
4481         if (!ofproto) {
4482             unixctl_command_reply_error(conn, "no such bridge");
4483             return;
4484         }
4485         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4486         mac_learning_flush(ofproto->ml);
4487         ovs_rwlock_unlock(&ofproto->ml->rwlock);
4488     } else {
4489         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4490             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4491             mac_learning_flush(ofproto->ml);
4492             ovs_rwlock_unlock(&ofproto->ml->rwlock);
4493         }
4494     }
4495
4496     unixctl_command_reply(conn, "table successfully flushed");
4497 }
4498
4499 static void
4500 ofproto_unixctl_mcast_snooping_flush(struct unixctl_conn *conn, int argc,
4501                                      const char *argv[], void *aux OVS_UNUSED)
4502 {
4503     struct ofproto_dpif *ofproto;
4504
4505     if (argc > 1) {
4506         ofproto = ofproto_dpif_lookup(argv[1]);
4507         if (!ofproto) {
4508             unixctl_command_reply_error(conn, "no such bridge");
4509             return;
4510         }
4511
4512         if (!mcast_snooping_enabled(ofproto->ms)) {
4513             unixctl_command_reply_error(conn, "multicast snooping is disabled");
4514             return;
4515         }
4516         mcast_snooping_mdb_flush(ofproto->ms);
4517     } else {
4518         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4519             if (!mcast_snooping_enabled(ofproto->ms)) {
4520                 continue;
4521             }
4522             mcast_snooping_mdb_flush(ofproto->ms);
4523         }
4524     }
4525
4526     unixctl_command_reply(conn, "table successfully flushed");
4527 }
4528
4529 static struct ofport_dpif *
4530 ofbundle_get_a_port(const struct ofbundle *bundle)
4531 {
4532     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
4533                         bundle_node);
4534 }
4535
4536 static void
4537 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4538                          const char *argv[], void *aux OVS_UNUSED)
4539 {
4540     struct ds ds = DS_EMPTY_INITIALIZER;
4541     const struct ofproto_dpif *ofproto;
4542     const struct mac_entry *e;
4543
4544     ofproto = ofproto_dpif_lookup(argv[1]);
4545     if (!ofproto) {
4546         unixctl_command_reply_error(conn, "no such bridge");
4547         return;
4548     }
4549
4550     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
4551     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
4552     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
4553         struct ofbundle *bundle = mac_entry_get_port(ofproto->ml, e);
4554         char name[OFP_MAX_PORT_NAME_LEN];
4555
4556         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4557                                name, sizeof name);
4558         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
4559                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
4560                       mac_entry_age(ofproto->ml, e));
4561     }
4562     ovs_rwlock_unlock(&ofproto->ml->rwlock);
4563     unixctl_command_reply(conn, ds_cstr(&ds));
4564     ds_destroy(&ds);
4565 }
4566
4567 static void
4568 ofproto_unixctl_mcast_snooping_show(struct unixctl_conn *conn,
4569                                     int argc OVS_UNUSED,
4570                                     const char *argv[],
4571                                     void *aux OVS_UNUSED)
4572 {
4573     struct ds ds = DS_EMPTY_INITIALIZER;
4574     const struct ofproto_dpif *ofproto;
4575     const struct ofbundle *bundle;
4576     const struct mcast_group *grp;
4577     struct mcast_group_bundle *b;
4578     struct mcast_mrouter_bundle *mrouter;
4579
4580     ofproto = ofproto_dpif_lookup(argv[1]);
4581     if (!ofproto) {
4582         unixctl_command_reply_error(conn, "no such bridge");
4583         return;
4584     }
4585
4586     if (!mcast_snooping_enabled(ofproto->ms)) {
4587         unixctl_command_reply_error(conn, "multicast snooping is disabled");
4588         return;
4589     }
4590
4591     ds_put_cstr(&ds, " port  VLAN  GROUP                Age\n");
4592     ovs_rwlock_rdlock(&ofproto->ms->rwlock);
4593     LIST_FOR_EACH (grp, group_node, &ofproto->ms->group_lru) {
4594         LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
4595             char name[OFP_MAX_PORT_NAME_LEN];
4596
4597             bundle = b->port;
4598             ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4599                                    name, sizeof name);
4600             ds_put_format(&ds, "%5s  %4d  ", name, grp->vlan);
4601             ipv6_format_mapped(&grp->addr, &ds);
4602             ds_put_format(&ds, "         %3d\n",
4603                           mcast_bundle_age(ofproto->ms, b));
4604         }
4605     }
4606
4607     /* ports connected to multicast routers */
4608     LIST_FOR_EACH(mrouter, mrouter_node, &ofproto->ms->mrouter_lru) {
4609         char name[OFP_MAX_PORT_NAME_LEN];
4610
4611         bundle = mrouter->port;
4612         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4613                                name, sizeof name);
4614         ds_put_format(&ds, "%5s  %4d  querier             %3d\n",
4615                       name, mrouter->vlan,
4616                       mcast_mrouter_age(ofproto->ms, mrouter));
4617     }
4618     ovs_rwlock_unlock(&ofproto->ms->rwlock);
4619     unixctl_command_reply(conn, ds_cstr(&ds));
4620     ds_destroy(&ds);
4621 }
4622
4623 struct trace_ctx {
4624     struct xlate_out xout;
4625     struct xlate_in xin;
4626     const struct flow *key;
4627     struct flow flow;
4628     struct ds *result;
4629     struct flow_wildcards wc;
4630     struct ofpbuf odp_actions;
4631 };
4632
4633 static void
4634 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
4635 {
4636     const struct rule_actions *actions;
4637     ovs_be64 cookie;
4638
4639     ds_put_char_multiple(result, '\t', level);
4640     if (!rule) {
4641         ds_put_cstr(result, "No match\n");
4642         return;
4643     }
4644
4645     ovs_mutex_lock(&rule->up.mutex);
4646     cookie = rule->up.flow_cookie;
4647     ovs_mutex_unlock(&rule->up.mutex);
4648
4649     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
4650                   rule ? rule->up.table_id : 0, ntohll(cookie));
4651     cls_rule_format(&rule->up.cr, result);
4652     ds_put_char(result, '\n');
4653
4654     actions = rule_dpif_get_actions(rule);
4655
4656     ds_put_char_multiple(result, '\t', level);
4657     ds_put_cstr(result, "OpenFlow actions=");
4658     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
4659     ds_put_char(result, '\n');
4660 }
4661
4662 static void
4663 trace_format_flow(struct ds *result, int level, const char *title,
4664                   struct trace_ctx *trace)
4665 {
4666     ds_put_char_multiple(result, '\t', level);
4667     ds_put_format(result, "%s: ", title);
4668     /* Do not report unchanged flows for resubmits. */
4669     if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
4670         || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
4671         ds_put_cstr(result, "unchanged");
4672     } else {
4673         flow_format(result, &trace->xin.flow);
4674         trace->flow = trace->xin.flow;
4675     }
4676     ds_put_char(result, '\n');
4677 }
4678
4679 static void
4680 trace_format_regs(struct ds *result, int level, const char *title,
4681                   struct trace_ctx *trace)
4682 {
4683     size_t i;
4684
4685     ds_put_char_multiple(result, '\t', level);
4686     ds_put_format(result, "%s:", title);
4687     for (i = 0; i < FLOW_N_REGS; i++) {
4688         ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
4689     }
4690     ds_put_char(result, '\n');
4691 }
4692
4693 static void
4694 trace_format_odp(struct ds *result, int level, const char *title,
4695                  struct trace_ctx *trace)
4696 {
4697     struct ofpbuf *odp_actions = &trace->odp_actions;
4698
4699     ds_put_char_multiple(result, '\t', level);
4700     ds_put_format(result, "%s: ", title);
4701     format_odp_actions(result, odp_actions->data, odp_actions->size);
4702     ds_put_char(result, '\n');
4703 }
4704
4705 static void
4706 trace_format_megaflow(struct ds *result, int level, const char *title,
4707                       struct trace_ctx *trace)
4708 {
4709     struct match match;
4710
4711     ds_put_char_multiple(result, '\t', level);
4712     ds_put_format(result, "%s: ", title);
4713     match_init(&match, trace->key, &trace->wc);
4714     match_format(&match, result, OFP_DEFAULT_PRIORITY);
4715     ds_put_char(result, '\n');
4716 }
4717
4718 static void trace_report(struct xlate_in *, int recurse,
4719                          const char *format, ...)
4720     OVS_PRINTF_FORMAT(3, 4);
4721 static void trace_report_valist(struct xlate_in *, int recurse,
4722                                 const char *format, va_list args)
4723     OVS_PRINTF_FORMAT(3, 0);
4724
4725 static void
4726 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
4727 {
4728     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4729     struct ds *result = trace->result;
4730
4731     if (!recurse) {
4732         if (rule == xin->ofproto->miss_rule) {
4733             trace_report(xin, recurse,
4734                          "No match, flow generates \"packet in\"s.");
4735         } else if (rule == xin->ofproto->no_packet_in_rule) {
4736             trace_report(xin, recurse, "No match, packets dropped because "
4737                          "OFPPC_NO_PACKET_IN is set on in_port.");
4738         } else if (rule == xin->ofproto->drop_frags_rule) {
4739             trace_report(xin, recurse, "Packets dropped because they are IP "
4740                          "fragments and the fragment handling mode is "
4741                          "\"drop\".");
4742         }
4743     }
4744
4745     ds_put_char(result, '\n');
4746     if (recurse) {
4747         trace_format_flow(result, recurse, "Resubmitted flow", trace);
4748         trace_format_regs(result, recurse, "Resubmitted regs", trace);
4749         trace_format_odp(result,  recurse, "Resubmitted  odp", trace);
4750         trace_format_megaflow(result, recurse, "Resubmitted megaflow", trace);
4751     }
4752     trace_format_rule(result, recurse, rule);
4753 }
4754
4755 static void
4756 trace_report_valist(struct xlate_in *xin, int recurse,
4757                     const char *format, va_list args)
4758 {
4759     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4760     struct ds *result = trace->result;
4761
4762     ds_put_char_multiple(result, '\t', recurse);
4763     ds_put_format_valist(result, format, args);
4764     ds_put_char(result, '\n');
4765 }
4766
4767 static void
4768 trace_report(struct xlate_in *xin, int recurse, const char *format, ...)
4769 {
4770     va_list args;
4771
4772     va_start(args, format);
4773     trace_report_valist(xin, recurse, format, args);
4774     va_end(args);
4775 }
4776
4777 /* Parses the 'argc' elements of 'argv', ignoring argv[0].  The following
4778  * forms are supported:
4779  *
4780  *     - [dpname] odp_flow [-generate | packet]
4781  *     - bridge br_flow [-generate | packet]
4782  *
4783  * On success, initializes '*ofprotop' and 'flow' and returns NULL.  On failure
4784  * returns a nonnull malloced error message. */
4785 static char * OVS_WARN_UNUSED_RESULT
4786 parse_flow_and_packet(int argc, const char *argv[],
4787                       struct ofproto_dpif **ofprotop, struct flow *flow,
4788                       struct dp_packet **packetp)
4789 {
4790     const struct dpif_backer *backer = NULL;
4791     const char *error = NULL;
4792     char *m_err = NULL;
4793     struct simap port_names = SIMAP_INITIALIZER(&port_names);
4794     struct dp_packet *packet;
4795     struct ofpbuf odp_key;
4796     struct ofpbuf odp_mask;
4797
4798     ofpbuf_init(&odp_key, 0);
4799     ofpbuf_init(&odp_mask, 0);
4800
4801     /* Handle "-generate" or a hex string as the last argument. */
4802     if (!strcmp(argv[argc - 1], "-generate")) {
4803         packet = dp_packet_new(0);
4804         argc--;
4805     } else {
4806         error = eth_from_hex(argv[argc - 1], &packet);
4807         if (!error) {
4808             argc--;
4809         } else if (argc == 4) {
4810             /* The 3-argument form must end in "-generate' or a hex string. */
4811             goto exit;
4812         }
4813         error = NULL;
4814     }
4815
4816     /* odp_flow can have its in_port specified as a name instead of port no.
4817      * We do not yet know whether a given flow is a odp_flow or a br_flow.
4818      * But, to know whether a flow is odp_flow through odp_flow_from_string(),
4819      * we need to create a simap of name to port no. */
4820     if (argc == 3) {
4821         const char *dp_type;
4822         if (!strncmp(argv[1], "ovs-", 4)) {
4823             dp_type = argv[1] + 4;
4824         } else {
4825             dp_type = argv[1];
4826         }
4827         backer = shash_find_data(&all_dpif_backers, dp_type);
4828     } else if (argc == 2) {
4829         struct shash_node *node;
4830         if (shash_count(&all_dpif_backers) == 1) {
4831             node = shash_first(&all_dpif_backers);
4832             backer = node->data;
4833         }
4834     } else {
4835         error = "Syntax error";
4836         goto exit;
4837     }
4838     if (backer && backer->dpif) {
4839         struct dpif_port dpif_port;
4840         struct dpif_port_dump port_dump;
4841         DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4842             simap_put(&port_names, dpif_port.name,
4843                       odp_to_u32(dpif_port.port_no));
4844         }
4845     }
4846
4847     /* Parse the flow and determine whether a datapath or
4848      * bridge is specified. If function odp_flow_key_from_string()
4849      * returns 0, the flow is a odp_flow. If function
4850      * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4851     if (!odp_flow_from_string(argv[argc - 1], &port_names,
4852                               &odp_key, &odp_mask)) {
4853         if (!backer) {
4854             error = "Cannot find the datapath";
4855             goto exit;
4856         }
4857
4858         if (odp_flow_key_to_flow(odp_key.data, odp_key.size, flow) == ODP_FIT_ERROR) {
4859             error = "Failed to parse datapath flow key";
4860             goto exit;
4861         }
4862
4863         *ofprotop = xlate_lookup_ofproto(backer, flow,
4864                                          &flow->in_port.ofp_port);
4865         if (*ofprotop == NULL) {
4866             error = "Invalid datapath flow";
4867             goto exit;
4868         }
4869
4870         vsp_adjust_flow(*ofprotop, flow, NULL);
4871
4872     } else {
4873         char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4874
4875         if (err) {
4876             m_err = xasprintf("Bad openflow flow syntax: %s", err);
4877             free(err);
4878             goto exit;
4879         } else {
4880             if (argc != 3) {
4881                 error = "Must specify bridge name";
4882                 goto exit;
4883             }
4884
4885             *ofprotop = ofproto_dpif_lookup(argv[1]);
4886             if (!*ofprotop) {
4887                 error = "Unknown bridge name";
4888                 goto exit;
4889             }
4890         }
4891     }
4892
4893     /* Generate a packet, if requested. */
4894     if (packet) {
4895         if (!dp_packet_size(packet)) {
4896             flow_compose(packet, flow);
4897         } else {
4898             /* Use the metadata from the flow and the packet argument
4899              * to reconstruct the flow. */
4900             pkt_metadata_from_flow(&packet->md, flow);
4901             flow_extract(packet, flow);
4902         }
4903     }
4904
4905 exit:
4906     if (error && !m_err) {
4907         m_err = xstrdup(error);
4908     }
4909     if (m_err) {
4910         dp_packet_delete(packet);
4911         packet = NULL;
4912     }
4913     *packetp = packet;
4914     ofpbuf_uninit(&odp_key);
4915     ofpbuf_uninit(&odp_mask);
4916     simap_destroy(&port_names);
4917     return m_err;
4918 }
4919
4920 static void
4921 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4922                       void *aux OVS_UNUSED)
4923 {
4924     struct ofproto_dpif *ofproto;
4925     struct dp_packet *packet;
4926     char *error;
4927     struct flow flow;
4928
4929     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4930     if (!error) {
4931         struct ds result;
4932
4933         ds_init(&result);
4934         ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4935         unixctl_command_reply(conn, ds_cstr(&result));
4936         ds_destroy(&result);
4937         dp_packet_delete(packet);
4938     } else {
4939         unixctl_command_reply_error(conn, error);
4940         free(error);
4941     }
4942 }
4943
4944 static void
4945 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4946                               const char *argv[], void *aux OVS_UNUSED)
4947 {
4948     enum ofputil_protocol usable_protocols;
4949     struct ofproto_dpif *ofproto;
4950     bool enforce_consistency;
4951     struct ofpbuf ofpacts;
4952     struct dp_packet *packet;
4953     struct ds result;
4954     struct flow flow;
4955     uint16_t in_port;
4956
4957     /* Three kinds of error return values! */
4958     enum ofperr retval;
4959     char *error;
4960
4961     packet = NULL;
4962     ds_init(&result);
4963     ofpbuf_init(&ofpacts, 0);
4964
4965     /* Parse actions. */
4966     error = ofpacts_parse_actions(argv[--argc], &ofpacts, &usable_protocols);
4967     if (error) {
4968         unixctl_command_reply_error(conn, error);
4969         free(error);
4970         goto exit;
4971     }
4972
4973     /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4974      * consistency between the flow and the actions.  With -consistent, we
4975      * enforce consistency even for a flow supported in OpenFlow 1.0. */
4976     if (!strcmp(argv[1], "-consistent")) {
4977         enforce_consistency = true;
4978         argv++;
4979         argc--;
4980     } else {
4981         enforce_consistency = false;
4982     }
4983
4984     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4985     if (error) {
4986         unixctl_command_reply_error(conn, error);
4987         free(error);
4988         goto exit;
4989     }
4990
4991     /* Do the same checks as handle_packet_out() in ofproto.c.
4992      *
4993      * We pass a 'table_id' of 0 to ofpacts_check(), which isn't
4994      * strictly correct because these actions aren't in any table, but it's OK
4995      * because it 'table_id' is used only to check goto_table instructions, but
4996      * packet-outs take a list of actions and therefore it can't include
4997      * instructions.
4998      *
4999      * We skip the "meter" check here because meter is an instruction, not an
5000      * action, and thus cannot appear in ofpacts. */
5001     in_port = ofp_to_u16(flow.in_port.ofp_port);
5002     if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
5003         unixctl_command_reply_error(conn, "invalid in_port");
5004         goto exit;
5005     }
5006     if (enforce_consistency) {
5007         retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size, &flow,
5008                                            u16_to_ofp(ofproto->up.max_ports),
5009                                            0, ofproto->up.n_tables,
5010                                            usable_protocols);
5011     } else {
5012         retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow,
5013                                u16_to_ofp(ofproto->up.max_ports), 0,
5014                                ofproto->up.n_tables, &usable_protocols);
5015     }
5016     if (!retval) {
5017         retval = ofproto_check_ofpacts(&ofproto->up, ofpacts.data,
5018                                        ofpacts.size);
5019     }
5020
5021     if (retval) {
5022         ds_clear(&result);
5023         ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
5024         unixctl_command_reply_error(conn, ds_cstr(&result));
5025         goto exit;
5026     }
5027
5028     ofproto_trace(ofproto, &flow, packet,
5029                   ofpacts.data, ofpacts.size, &result);
5030     unixctl_command_reply(conn, ds_cstr(&result));
5031
5032 exit:
5033     ds_destroy(&result);
5034     dp_packet_delete(packet);
5035     ofpbuf_uninit(&ofpacts);
5036 }
5037
5038 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
5039  * description of the results to 'ds'.
5040  *
5041  * The trace follows a packet with the specified 'flow' through the flow
5042  * table.  'packet' may be nonnull to trace an actual packet, with consequent
5043  * side effects (if it is nonnull then its flow must be 'flow').
5044  *
5045  * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
5046  * trace, otherwise the actions are determined by a flow table lookup. */
5047 static void
5048 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
5049               const struct dp_packet *packet,
5050               const struct ofpact ofpacts[], size_t ofpacts_len,
5051               struct ds *ds)
5052 {
5053     struct trace_ctx trace;
5054     enum xlate_error error;
5055
5056     ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
5057     ds_put_cstr(ds, "Flow: ");
5058     flow_format(ds, flow);
5059     ds_put_char(ds, '\n');
5060
5061     ofpbuf_init(&trace.odp_actions, 0);
5062
5063     trace.result = ds;
5064     trace.key = flow; /* Original flow key, used for megaflow. */
5065     trace.flow = *flow; /* May be modified by actions. */
5066     xlate_in_init(&trace.xin, ofproto, flow, flow->in_port.ofp_port, NULL,
5067                   ntohs(flow->tcp_flags), packet, &trace.wc,
5068                   &trace.odp_actions);
5069     trace.xin.ofpacts = ofpacts;
5070     trace.xin.ofpacts_len = ofpacts_len;
5071     trace.xin.resubmit_hook = trace_resubmit;
5072     trace.xin.report_hook = trace_report_valist;
5073
5074     error = xlate_actions(&trace.xin, &trace.xout);
5075     ds_put_char(ds, '\n');
5076     trace_format_flow(ds, 0, "Final flow", &trace);
5077     trace_format_megaflow(ds, 0, "Megaflow", &trace);
5078
5079     ds_put_cstr(ds, "Datapath actions: ");
5080     format_odp_actions(ds, trace.odp_actions.data, trace.odp_actions.size);
5081
5082     if (error != XLATE_OK) {
5083         ds_put_format(ds, "\nTranslation failed (%s), packet is dropped.\n",
5084                       xlate_strerror(error));
5085     } else if (trace.xout.slow) {
5086         enum slow_path_reason slow;
5087
5088         ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5089                     "slow path because it:");
5090
5091         slow = trace.xout.slow;
5092         while (slow) {
5093             enum slow_path_reason bit = rightmost_1bit(slow);
5094
5095             ds_put_format(ds, "\n\t- %s.",
5096                           slow_path_reason_to_explanation(bit));
5097
5098             slow &= ~bit;
5099         }
5100     }
5101
5102     xlate_out_uninit(&trace.xout);
5103     ofpbuf_uninit(&trace.odp_actions);
5104 }
5105
5106 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
5107  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
5108  * to destroy 'ofproto_shash' and free the returned value. */
5109 static const struct shash_node **
5110 get_ofprotos(struct shash *ofproto_shash)
5111 {
5112     const struct ofproto_dpif *ofproto;
5113
5114     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5115         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5116         shash_add_nocopy(ofproto_shash, name, ofproto);
5117     }
5118
5119     return shash_sort(ofproto_shash);
5120 }
5121
5122 static void
5123 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5124                               const char *argv[] OVS_UNUSED,
5125                               void *aux OVS_UNUSED)
5126 {
5127     struct ds ds = DS_EMPTY_INITIALIZER;
5128     struct shash ofproto_shash;
5129     const struct shash_node **sorted_ofprotos;
5130     int i;
5131
5132     shash_init(&ofproto_shash);
5133     sorted_ofprotos = get_ofprotos(&ofproto_shash);
5134     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5135         const struct shash_node *node = sorted_ofprotos[i];
5136         ds_put_format(&ds, "%s\n", node->name);
5137     }
5138
5139     shash_destroy(&ofproto_shash);
5140     free(sorted_ofprotos);
5141
5142     unixctl_command_reply(conn, ds_cstr(&ds));
5143     ds_destroy(&ds);
5144 }
5145
5146 static void
5147 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5148 {
5149     const struct shash_node **ofprotos;
5150     struct dpif_dp_stats dp_stats;
5151     struct shash ofproto_shash;
5152     size_t i;
5153
5154     dpif_get_dp_stats(backer->dpif, &dp_stats);
5155
5156     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5157                   dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
5158
5159     shash_init(&ofproto_shash);
5160     ofprotos = get_ofprotos(&ofproto_shash);
5161     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5162         struct ofproto_dpif *ofproto = ofprotos[i]->data;
5163         const struct shash_node **ports;
5164         size_t j;
5165
5166         if (ofproto->backer != backer) {
5167             continue;
5168         }
5169
5170         ds_put_format(ds, "\t%s:\n", ofproto->up.name);
5171
5172         ports = shash_sort(&ofproto->up.port_by_name);
5173         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5174             const struct shash_node *node = ports[j];
5175             struct ofport *ofport = node->data;
5176             struct smap config;
5177             odp_port_t odp_port;
5178
5179             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5180                           ofport->ofp_port);
5181
5182             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5183             if (odp_port != ODPP_NONE) {
5184                 ds_put_format(ds, "%"PRIu32":", odp_port);
5185             } else {
5186                 ds_put_cstr(ds, "none:");
5187             }
5188
5189             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5190
5191             smap_init(&config);
5192             if (!netdev_get_config(ofport->netdev, &config)) {
5193                 const struct smap_node **nodes;
5194                 size_t i;
5195
5196                 nodes = smap_sort(&config);
5197                 for (i = 0; i < smap_count(&config); i++) {
5198                     const struct smap_node *node = nodes[i];
5199                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5200                                   node->key, node->value);
5201                 }
5202                 free(nodes);
5203             }
5204             smap_destroy(&config);
5205
5206             ds_put_char(ds, ')');
5207             ds_put_char(ds, '\n');
5208         }
5209         free(ports);
5210     }
5211     shash_destroy(&ofproto_shash);
5212     free(ofprotos);
5213 }
5214
5215 static void
5216 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5217                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5218 {
5219     struct ds ds = DS_EMPTY_INITIALIZER;
5220     const struct shash_node **backers;
5221     int i;
5222
5223     backers = shash_sort(&all_dpif_backers);
5224     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5225         dpif_show_backer(backers[i]->data, &ds);
5226     }
5227     free(backers);
5228
5229     unixctl_command_reply(conn, ds_cstr(&ds));
5230     ds_destroy(&ds);
5231 }
5232
5233 static void
5234 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5235                                 int argc OVS_UNUSED, const char *argv[],
5236                                 void *aux OVS_UNUSED)
5237 {
5238     const struct ofproto_dpif *ofproto;
5239
5240     struct ds ds = DS_EMPTY_INITIALIZER;
5241     bool verbosity = false;
5242
5243     struct dpif_port dpif_port;
5244     struct dpif_port_dump port_dump;
5245     struct hmap portno_names;
5246
5247     struct dpif_flow_dump *flow_dump;
5248     struct dpif_flow_dump_thread *flow_dump_thread;
5249     struct dpif_flow f;
5250     int error;
5251
5252     ofproto = ofproto_dpif_lookup(argv[argc - 1]);
5253     if (!ofproto) {
5254         unixctl_command_reply_error(conn, "no such bridge");
5255         return;
5256     }
5257
5258     if (argc > 2 && !strcmp(argv[1], "-m")) {
5259         verbosity = true;
5260     }
5261
5262     hmap_init(&portno_names);
5263     DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
5264         odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
5265     }
5266
5267     ds_init(&ds);
5268     flow_dump = dpif_flow_dump_create(ofproto->backer->dpif, false);
5269     flow_dump_thread = dpif_flow_dump_thread_create(flow_dump);
5270     while (dpif_flow_dump_next(flow_dump_thread, &f, 1)) {
5271         struct flow flow;
5272
5273         if (odp_flow_key_to_flow(f.key, f.key_len, &flow) == ODP_FIT_ERROR
5274             || xlate_lookup_ofproto(ofproto->backer, &flow, NULL) != ofproto) {
5275             continue;
5276         }
5277
5278         if (verbosity) {
5279             odp_format_ufid(&f.ufid, &ds);
5280             ds_put_cstr(&ds, " ");
5281         }
5282         odp_flow_format(f.key, f.key_len, f.mask, f.mask_len,
5283                         &portno_names, &ds, verbosity);
5284         ds_put_cstr(&ds, ", ");
5285         dpif_flow_stats_format(&f.stats, &ds);
5286         ds_put_cstr(&ds, ", actions:");
5287         format_odp_actions(&ds, f.actions, f.actions_len);
5288         ds_put_char(&ds, '\n');
5289     }
5290     dpif_flow_dump_thread_destroy(flow_dump_thread);
5291     error = dpif_flow_dump_destroy(flow_dump);
5292
5293     if (error) {
5294         ds_clear(&ds);
5295         ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
5296         unixctl_command_reply_error(conn, ds_cstr(&ds));
5297     } else {
5298         unixctl_command_reply(conn, ds_cstr(&ds));
5299     }
5300     odp_portno_names_destroy(&portno_names);
5301     hmap_destroy(&portno_names);
5302     ds_destroy(&ds);
5303 }
5304
5305 static void
5306 ofproto_revalidate_all_backers(void)
5307 {
5308     const struct shash_node **backers;
5309     int i;
5310
5311     backers = shash_sort(&all_dpif_backers);
5312     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5313         struct dpif_backer *backer = backers[i]->data;
5314         backer->need_revalidate = REV_RECONFIGURE;
5315     }
5316     free(backers);
5317 }
5318
5319 static void
5320 disable_tnl_push_pop(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5321                      const char *argv[], void *aux OVS_UNUSED)
5322 {
5323     if (!strcasecmp(argv[1], "off")) {
5324         ofproto_use_tnl_push_pop = false;
5325         unixctl_command_reply(conn, "Tunnel push-pop off");
5326         ofproto_revalidate_all_backers();
5327     } else if (!strcasecmp(argv[1], "on")) {
5328         ofproto_use_tnl_push_pop = true;
5329         unixctl_command_reply(conn, "Tunnel push-pop on");
5330         ofproto_revalidate_all_backers();
5331     } else {
5332         unixctl_command_reply_error(conn, "Invalid argument");
5333     }
5334 }
5335
5336 static void
5337 ofproto_unixctl_init(void)
5338 {
5339     static bool registered;
5340     if (registered) {
5341         return;
5342     }
5343     registered = true;
5344
5345     unixctl_command_register(
5346         "ofproto/trace",
5347         "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
5348         1, 3, ofproto_unixctl_trace, NULL);
5349     unixctl_command_register(
5350         "ofproto/trace-packet-out",
5351         "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
5352         2, 6, ofproto_unixctl_trace_actions, NULL);
5353     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5354                              ofproto_unixctl_fdb_flush, NULL);
5355     unixctl_command_register("fdb/show", "bridge", 1, 1,
5356                              ofproto_unixctl_fdb_show, NULL);
5357     unixctl_command_register("mdb/flush", "[bridge]", 0, 1,
5358                              ofproto_unixctl_mcast_snooping_flush, NULL);
5359     unixctl_command_register("mdb/show", "bridge", 1, 1,
5360                              ofproto_unixctl_mcast_snooping_show, NULL);
5361     unixctl_command_register("dpif/dump-dps", "", 0, 0,
5362                              ofproto_unixctl_dpif_dump_dps, NULL);
5363     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5364                              NULL);
5365     unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
5366                              ofproto_unixctl_dpif_dump_flows, NULL);
5367
5368     unixctl_command_register("ofproto/tnl-push-pop", "[on]|[off]", 1, 1,
5369                              disable_tnl_push_pop, NULL);
5370 }
5371
5372 /* Returns true if 'table' is the table used for internal rules,
5373  * false otherwise. */
5374 bool
5375 table_is_internal(uint8_t table_id)
5376 {
5377     return table_id == TBL_INTERNAL;
5378 }
5379 \f
5380 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5381  *
5382  * This is deprecated.  It is only for compatibility with broken device drivers
5383  * in old versions of Linux that do not properly support VLANs when VLAN
5384  * devices are not used.  When broken device drivers are no longer in
5385  * widespread use, we will delete these interfaces. */
5386
5387 static int
5388 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5389 {
5390     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5391     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5392
5393     if (realdev_ofp_port == ofport->realdev_ofp_port
5394         && vid == ofport->vlandev_vid) {
5395         return 0;
5396     }
5397
5398     ofproto->backer->need_revalidate = REV_RECONFIGURE;
5399
5400     if (ofport->realdev_ofp_port) {
5401         vsp_remove(ofport);
5402     }
5403     if (realdev_ofp_port && ofport->bundle) {
5404         /* vlandevs are enslaved to their realdevs, so they are not allowed to
5405          * themselves be part of a bundle. */
5406         bundle_set(ofport_->ofproto, ofport->bundle, NULL);
5407     }
5408
5409     ofport->realdev_ofp_port = realdev_ofp_port;
5410     ofport->vlandev_vid = vid;
5411
5412     if (realdev_ofp_port) {
5413         vsp_add(ofport, realdev_ofp_port, vid);
5414     }
5415
5416     return 0;
5417 }
5418
5419 static uint32_t
5420 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
5421 {
5422     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
5423 }
5424
5425 bool
5426 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
5427     OVS_EXCLUDED(ofproto->vsp_mutex)
5428 {
5429     /* hmap_is_empty is thread safe. */
5430     return !hmap_is_empty(&ofproto->realdev_vid_map);
5431 }
5432
5433
5434 static ofp_port_t
5435 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
5436                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5437     OVS_REQUIRES(ofproto->vsp_mutex)
5438 {
5439     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5440         int vid = vlan_tci_to_vid(vlan_tci);
5441         const struct vlan_splinter *vsp;
5442
5443         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5444                                  hash_realdev_vid(realdev_ofp_port, vid),
5445                                  &ofproto->realdev_vid_map) {
5446             if (vsp->realdev_ofp_port == realdev_ofp_port
5447                 && vsp->vid == vid) {
5448                 return vsp->vlandev_ofp_port;
5449             }
5450         }
5451     }
5452     return realdev_ofp_port;
5453 }
5454
5455 /* Returns the OFP port number of the Linux VLAN device that corresponds to
5456  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
5457  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
5458  * 'vlan_tci' 9, it would return the port number of eth0.9.
5459  *
5460  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
5461  * function just returns its 'realdev_ofp_port' argument. */
5462 ofp_port_t
5463 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5464                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5465     OVS_EXCLUDED(ofproto->vsp_mutex)
5466 {
5467     ofp_port_t ret;
5468
5469     /* hmap_is_empty is thread safe, see if we can return immediately. */
5470     if (hmap_is_empty(&ofproto->realdev_vid_map)) {
5471         return realdev_ofp_port;
5472     }
5473     ovs_mutex_lock(&ofproto->vsp_mutex);
5474     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
5475     ovs_mutex_unlock(&ofproto->vsp_mutex);
5476     return ret;
5477 }
5478
5479 static struct vlan_splinter *
5480 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
5481 {
5482     struct vlan_splinter *vsp;
5483
5484     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
5485                              hash_ofp_port(vlandev_ofp_port),
5486                              &ofproto->vlandev_map) {
5487         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5488             return vsp;
5489         }
5490     }
5491
5492     return NULL;
5493 }
5494
5495 /* Returns the OpenFlow port number of the "real" device underlying the Linux
5496  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
5497  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
5498  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
5499  * eth0 and store 9 in '*vid'.
5500  *
5501  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
5502  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
5503  * always does.*/
5504 static ofp_port_t
5505 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5506                        ofp_port_t vlandev_ofp_port, int *vid)
5507     OVS_REQUIRES(ofproto->vsp_mutex)
5508 {
5509     if (!hmap_is_empty(&ofproto->vlandev_map)) {
5510         const struct vlan_splinter *vsp;
5511
5512         vsp = vlandev_find(ofproto, vlandev_ofp_port);
5513         if (vsp) {
5514             if (vid) {
5515                 *vid = vsp->vid;
5516             }
5517             return vsp->realdev_ofp_port;
5518         }
5519     }
5520     return 0;
5521 }
5522
5523 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
5524  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
5525  * 'flow->in_port' to the "real" device backing the VLAN device, sets
5526  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Optionally pushes the
5527  * appropriate VLAN on 'packet' if provided.  Otherwise (which is always the
5528  * case unless VLAN splinters are enabled), returns false without making any
5529  * changes. */
5530 bool
5531 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow,
5532                 struct dp_packet *packet)
5533     OVS_EXCLUDED(ofproto->vsp_mutex)
5534 {
5535     ofp_port_t realdev;
5536     int vid;
5537
5538     /* hmap_is_empty is thread safe. */
5539     if (hmap_is_empty(&ofproto->vlandev_map)) {
5540         return false;
5541     }
5542
5543     ovs_mutex_lock(&ofproto->vsp_mutex);
5544     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
5545     ovs_mutex_unlock(&ofproto->vsp_mutex);
5546     if (!realdev) {
5547         return false;
5548     }
5549
5550     /* Cause the flow to be processed as if it came in on the real device with
5551      * the VLAN device's VLAN ID. */
5552     flow->in_port.ofp_port = realdev;
5553     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
5554
5555     if (packet) {
5556         /* Make the packet resemble the flow, so that it gets sent to an
5557          * OpenFlow controller properly, so that it looks correct for sFlow,
5558          * and so that flow_extract() will get the correct vlan_tci if it is
5559          * called on 'packet'. */
5560         eth_push_vlan(packet, htons(ETH_TYPE_VLAN), flow->vlan_tci);
5561     }
5562
5563     return true;
5564 }
5565
5566 static void
5567 vsp_remove(struct ofport_dpif *port)
5568 {
5569     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5570     struct vlan_splinter *vsp;
5571
5572     ovs_mutex_lock(&ofproto->vsp_mutex);
5573     vsp = vlandev_find(ofproto, port->up.ofp_port);
5574     if (vsp) {
5575         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5576         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
5577         free(vsp);
5578
5579         port->realdev_ofp_port = 0;
5580     } else {
5581         VLOG_ERR("missing vlan device record");
5582     }
5583     ovs_mutex_unlock(&ofproto->vsp_mutex);
5584 }
5585
5586 static void
5587 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
5588 {
5589     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5590
5591     ovs_mutex_lock(&ofproto->vsp_mutex);
5592     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
5593         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
5594             == realdev_ofp_port)) {
5595         struct vlan_splinter *vsp;
5596
5597         vsp = xmalloc(sizeof *vsp);
5598         vsp->realdev_ofp_port = realdev_ofp_port;
5599         vsp->vlandev_ofp_port = port->up.ofp_port;
5600         vsp->vid = vid;
5601
5602         port->realdev_ofp_port = realdev_ofp_port;
5603
5604         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
5605                     hash_ofp_port(port->up.ofp_port));
5606         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
5607                     hash_realdev_vid(realdev_ofp_port, vid));
5608     } else {
5609         VLOG_ERR("duplicate vlan device record");
5610     }
5611     ovs_mutex_unlock(&ofproto->vsp_mutex);
5612 }
5613
5614 static odp_port_t
5615 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
5616 {
5617     const struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
5618     return ofport ? ofport->odp_port : ODPP_NONE;
5619 }
5620
5621 struct ofport_dpif *
5622 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
5623 {
5624     struct ofport_dpif *port;
5625
5626     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
5627     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
5628                              &backer->odp_to_ofport_map) {
5629         if (port->odp_port == odp_port) {
5630             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5631             return port;
5632         }
5633     }
5634
5635     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5636     return NULL;
5637 }
5638
5639 static ofp_port_t
5640 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
5641 {
5642     struct ofport_dpif *port;
5643
5644     port = odp_port_to_ofport(ofproto->backer, odp_port);
5645     if (port && &ofproto->up == port->up.ofproto) {
5646         return port->up.ofp_port;
5647     } else {
5648         return OFPP_NONE;
5649     }
5650 }
5651
5652 int
5653 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
5654                                const struct match *match, int priority,
5655                                uint16_t idle_timeout,
5656                                const struct ofpbuf *ofpacts,
5657                                struct rule **rulep)
5658 {
5659     struct ofproto_flow_mod ofm;
5660     struct rule_dpif *rule;
5661     int error;
5662
5663     ofm.fm.match = *match;
5664     ofm.fm.priority = priority;
5665     ofm.fm.new_cookie = htonll(0);
5666     ofm.fm.cookie = htonll(0);
5667     ofm.fm.cookie_mask = htonll(0);
5668     ofm.fm.modify_cookie = false;
5669     ofm.fm.table_id = TBL_INTERNAL;
5670     ofm.fm.command = OFPFC_ADD;
5671     ofm.fm.idle_timeout = idle_timeout;
5672     ofm.fm.hard_timeout = 0;
5673     ofm.fm.importance = 0;
5674     ofm.fm.buffer_id = 0;
5675     ofm.fm.out_port = 0;
5676     ofm.fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5677     ofm.fm.ofpacts = ofpacts->data;
5678     ofm.fm.ofpacts_len = ofpacts->size;
5679
5680     error = ofproto_flow_mod(&ofproto->up, &ofm);
5681     if (error) {
5682         VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
5683                     ofperr_to_string(error));
5684         *rulep = NULL;
5685         return error;
5686     }
5687
5688     rule = rule_dpif_lookup_in_table(ofproto,
5689                                      ofproto_dpif_get_tables_version(ofproto),
5690                                      TBL_INTERNAL, &ofm.fm.match.flow,
5691                                      &ofm.fm.match.wc);
5692     if (rule) {
5693         *rulep = &rule->up;
5694     } else {
5695         OVS_NOT_REACHED();
5696     }
5697     return 0;
5698 }
5699
5700 int
5701 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
5702                                   struct match *match, int priority)
5703 {
5704     struct ofproto_flow_mod ofm;
5705     int error;
5706
5707     ofm.fm.match = *match;
5708     ofm.fm.priority = priority;
5709     ofm.fm.new_cookie = htonll(0);
5710     ofm.fm.cookie = htonll(0);
5711     ofm.fm.cookie_mask = htonll(0);
5712     ofm.fm.modify_cookie = false;
5713     ofm.fm.table_id = TBL_INTERNAL;
5714     ofm.fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5715     ofm.fm.command = OFPFC_DELETE_STRICT;
5716
5717     error = ofproto_flow_mod(&ofproto->up, &ofm);
5718     if (error) {
5719         VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
5720                     ofperr_to_string(error));
5721         return error;
5722     }
5723
5724     return 0;
5725 }
5726
5727 const struct ofproto_class ofproto_dpif_class = {
5728     init,
5729     enumerate_types,
5730     enumerate_names,
5731     del,
5732     port_open_type,
5733     type_run,
5734     type_wait,
5735     alloc,
5736     construct,
5737     destruct,
5738     dealloc,
5739     run,
5740     wait,
5741     NULL,                       /* get_memory_usage. */
5742     type_get_memory_usage,
5743     flush,
5744     query_tables,
5745     set_tables_version,
5746     port_alloc,
5747     port_construct,
5748     port_destruct,
5749     port_dealloc,
5750     port_modified,
5751     port_reconfigured,
5752     port_query_by_name,
5753     port_add,
5754     port_del,
5755     port_get_stats,
5756     port_dump_start,
5757     port_dump_next,
5758     port_dump_done,
5759     port_poll,
5760     port_poll_wait,
5761     port_is_lacp_current,
5762     port_get_lacp_stats,
5763     NULL,                       /* rule_choose_table */
5764     rule_alloc,
5765     rule_construct,
5766     rule_insert,
5767     rule_delete,
5768     rule_destruct,
5769     rule_dealloc,
5770     rule_get_stats,
5771     rule_execute,
5772     set_frag_handling,
5773     packet_out,
5774     set_netflow,
5775     get_netflow_ids,
5776     set_sflow,
5777     set_ipfix,
5778     set_cfm,
5779     cfm_status_changed,
5780     get_cfm_status,
5781     set_lldp,
5782     get_lldp_status,
5783     set_aa,
5784     aa_mapping_set,
5785     aa_mapping_unset,
5786     aa_vlan_get_queued,
5787     aa_vlan_get_queue_size,
5788     set_bfd,
5789     bfd_status_changed,
5790     get_bfd_status,
5791     set_stp,
5792     get_stp_status,
5793     set_stp_port,
5794     get_stp_port_status,
5795     get_stp_port_stats,
5796     set_rstp,
5797     get_rstp_status,
5798     set_rstp_port,
5799     get_rstp_port_status,
5800     set_queues,
5801     bundle_set,
5802     bundle_remove,
5803     mirror_set__,
5804     mirror_get_stats__,
5805     set_flood_vlans,
5806     is_mirror_output_bundle,
5807     forward_bpdu_changed,
5808     set_mac_table_config,
5809     set_mcast_snooping,
5810     set_mcast_snooping_port,
5811     set_realdev,
5812     NULL,                       /* meter_get_features */
5813     NULL,                       /* meter_set */
5814     NULL,                       /* meter_get */
5815     NULL,                       /* meter_del */
5816     group_alloc,                /* group_alloc */
5817     group_construct,            /* group_construct */
5818     group_destruct,             /* group_destruct */
5819     group_dealloc,              /* group_dealloc */
5820     group_modify,               /* group_modify */
5821     group_get_stats,            /* group_get_stats */
5822     get_datapath_version,       /* get_datapath_version */
5823 };