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 size_t
 363 ofproto_dpif_get_max_mpls_depth(const struct ofproto_dpif *ofproto)
 364 {
 365     return ofproto->backer->support.max_mpls_depth;
 366 }
 367
 368 bool
 369 ofproto_dpif_get_enable_recirc(const struct ofproto_dpif *ofproto)
 370 {
 371     return ofproto->backer->support.recirc;
 372 }
 373
 374 bool
 375 ofproto_dpif_get_enable_ufid(struct dpif_backer *backer)
 376 {
 377     return backer->support.ufid;
 378 }
 379
 380 static void ofproto_trace(struct ofproto_dpif *, struct flow *,
 381                           const struct dp_packet *packet,
 382                           const struct ofpact[], size_t ofpacts_len,
 383                           struct ds *);
 384
 385 /* Global variables. */
 386 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 387
 388 /* Initial mappings of port to bridge mappings. */
 389 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 390
 391 /* Executes 'fm'.  The caller retains ownership of 'fm' and everything in
 392  * it. */
 393 void
 394 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 395                       struct ofputil_flow_mod *fm)
 396 {
 397     ofproto_flow_mod(&ofproto->up, fm);
 398 }
 399
 400 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 401  * Takes ownership of 'pin' and pin->packet. */
 402 void
 403 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 404                             struct ofproto_packet_in *pin)
 405 {
 406     if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
 407         COVERAGE_INC(packet_in_overflow);
 408         free(CONST_CAST(void *, pin->up.packet));
 409         free(pin);
 410     }
 411
 412     /* Wakes up main thread for packet-in I/O. */
 413     seq_change(ofproto->pins_seq);
 414 }
 415
 416 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
 417  * packet rather than to send the packet to the controller.
 418  *
 419  * This function returns false to indicate that a packet_in message
 420  * for a "table-miss" should be sent to at least one controller.
 421  * False otherwise. */
 422 bool
 423 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif *ofproto)
 424 {
 425     return connmgr_wants_packet_in_on_miss(ofproto->up.connmgr);
 426 }
 427 \f
 428 /* Factory functions. */
 429
 430 static void
 431 init(const struct shash *iface_hints)
 432 {
 433     struct shash_node *node;
 434
 435     /* Make a local copy, since we don't own 'iface_hints' elements. */
 436     SHASH_FOR_EACH(node, iface_hints) {
 437         const struct iface_hint *orig_hint = node->data;
 438         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 439
 440         new_hint->br_name = xstrdup(orig_hint->br_name);
 441         new_hint->br_type = xstrdup(orig_hint->br_type);
 442         new_hint->ofp_port = orig_hint->ofp_port;
 443
 444         shash_add(&init_ofp_ports, node->name, new_hint);
 445     }
 446 }
 447
 448 static void
 449 enumerate_types(struct sset *types)
 450 {
 451     dp_enumerate_types(types);
 452 }
 453
 454 static int
 455 enumerate_names(const char *type, struct sset *names)
 456 {
 457     struct ofproto_dpif *ofproto;
 458
 459     sset_clear(names);
 460     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 461         if (strcmp(type, ofproto->up.type)) {
 462             continue;
 463         }
 464         sset_add(names, ofproto->up.name);
 465     }
 466
 467     return 0;
 468 }
 469
 470 static int
 471 del(const char *type, const char *name)
 472 {
 473     struct dpif *dpif;
 474     int error;
 475
 476     error = dpif_open(name, type, &dpif);
 477     if (!error) {
 478         error = dpif_delete(dpif);
 479         dpif_close(dpif);
 480     }
 481     return error;
 482 }
 483 \f
 484 static const char *
 485 port_open_type(const char *datapath_type, const char *port_type)
 486 {
 487     return dpif_port_open_type(datapath_type, port_type);
 488 }
 489
 490 /* Type functions. */
 491
 492 static void process_dpif_port_changes(struct dpif_backer *);
 493 static void process_dpif_all_ports_changed(struct dpif_backer *);
 494 static void process_dpif_port_change(struct dpif_backer *,
 495                                      const char *devname);
 496 static void process_dpif_port_error(struct dpif_backer *, int error);
 497
 498 static struct ofproto_dpif *
 499 lookup_ofproto_dpif_by_port_name(const char *name)
 500 {
 501     struct ofproto_dpif *ofproto;
 502
 503     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 504         if (sset_contains(&ofproto->ports, name)) {
 505             return ofproto;
 506         }
 507     }
 508
 509     return NULL;
 510 }
 511
 512 static int
 513 type_run(const char *type)
 514 {
 515     struct dpif_backer *backer;
 516
 517     backer = shash_find_data(&all_dpif_backers, type);
 518     if (!backer) {
 519         /* This is not necessarily a problem, since backers are only
 520          * created on demand. */
 521         return 0;
 522     }
 523
 524
 525     if (dpif_run(backer->dpif)) {
 526         backer->need_revalidate = REV_RECONFIGURE;
 527     }
 528
 529     udpif_run(backer->udpif);
 530
 531     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 532      * and the configuration has now changed to "false", enable receiving
 533      * packets from the datapath. */
 534     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 535         int error;
 536
 537         backer->recv_set_enable = true;
 538
 539         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 540         if (error) {
 541             VLOG_ERR("Failed to enable receiving packets in dpif.");
 542             return error;
 543         }
 544         dpif_flow_flush(backer->dpif);
 545         backer->need_revalidate = REV_RECONFIGURE;
 546     }
 547
 548     if (backer->recv_set_enable) {
 549         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 550     }
 551
 552     dpif_poll_threads_set(backer->dpif, n_dpdk_rxqs, pmd_cpu_mask);
 553
 554     if (backer->need_revalidate) {
 555         struct ofproto_dpif *ofproto;
 556         struct simap_node *node;
 557         struct simap tmp_backers;
 558
 559         /* Handle tunnel garbage collection. */
 560         simap_init(&tmp_backers);
 561         simap_swap(&backer->tnl_backers, &tmp_backers);
 562
 563         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 564             struct ofport_dpif *iter;
 565
 566             if (backer != ofproto->backer) {
 567                 continue;
 568             }
 569
 570             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 571                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 572                 const char *dp_port;
 573
 574                 if (!iter->is_tunnel) {
 575                     continue;
 576                 }
 577
 578                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 579                                                      namebuf, sizeof namebuf);
 580                 node = simap_find(&tmp_backers, dp_port);
 581                 if (node) {
 582                     simap_put(&backer->tnl_backers, dp_port, node->data);
 583                     simap_delete(&tmp_backers, node);
 584                     node = simap_find(&backer->tnl_backers, dp_port);
 585                 } else {
 586                     node = simap_find(&backer->tnl_backers, dp_port);
 587                     if (!node) {
 588                         odp_port_t odp_port = ODPP_NONE;
 589
 590                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 591                                            &odp_port)) {
 592                             simap_put(&backer->tnl_backers, dp_port,
 593                                       odp_to_u32(odp_port));
 594                             node = simap_find(&backer->tnl_backers, dp_port);
 595                         }
 596                     }
 597                 }
 598
 599                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 600                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 601                                          iter->odp_port,
 602                                          ovs_native_tunneling_is_on(ofproto), dp_port)) {
 603                     backer->need_revalidate = REV_RECONFIGURE;
 604                 }
 605             }
 606         }
 607
 608         SIMAP_FOR_EACH (node, &tmp_backers) {
 609             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 610         }
 611         simap_destroy(&tmp_backers);
 612
 613         switch (backer->need_revalidate) {
 614         case REV_RECONFIGURE:    COVERAGE_INC(rev_reconfigure);    break;
 615         case REV_STP:            COVERAGE_INC(rev_stp);            break;
 616         case REV_RSTP:           COVERAGE_INC(rev_rstp);           break;
 617         case REV_BOND:           COVERAGE_INC(rev_bond);           break;
 618         case REV_PORT_TOGGLED:   COVERAGE_INC(rev_port_toggled);   break;
 619         case REV_FLOW_TABLE:     COVERAGE_INC(rev_flow_table);     break;
 620         case REV_MAC_LEARNING:   COVERAGE_INC(rev_mac_learning);   break;
 621         case REV_MCAST_SNOOPING: COVERAGE_INC(rev_mcast_snooping); break;
 622         }
 623         backer->need_revalidate = 0;
 624
 625         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 626             struct ofport_dpif *ofport;
 627             struct ofbundle *bundle;
 628
 629             if (ofproto->backer != backer) {
 630                 continue;
 631             }
 632
 633             xlate_txn_start();
 634             xlate_ofproto_set(ofproto, ofproto->up.name,
 635                               ofproto->backer->dpif, ofproto->ml,
 636                               ofproto->stp, ofproto->rstp, ofproto->ms,
 637                               ofproto->mbridge, ofproto->sflow, ofproto->ipfix,
 638                               ofproto->netflow,
 639                               ofproto->up.forward_bpdu,
 640                               connmgr_has_in_band(ofproto->up.connmgr),
 641                               &ofproto->backer->support);
 642
 643             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 644                 xlate_bundle_set(ofproto, bundle, bundle->name,
 645                                  bundle->vlan_mode, bundle->vlan,
 646                                  bundle->trunks, bundle->use_priority_tags,
 647                                  bundle->bond, bundle->lacp,
 648                                  bundle->floodable);
 649             }
 650
 651             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 652                 int stp_port = ofport->stp_port
 653                     ? stp_port_no(ofport->stp_port)
 654                     : -1;
 655                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 656                                  ofport->up.ofp_port, ofport->odp_port,
 657                                  ofport->up.netdev, ofport->cfm, ofport->bfd,
 658                                  ofport->lldp, ofport->peer, stp_port,
 659                                  ofport->rstp_port, ofport->qdscp,
 660                                  ofport->n_qdscp, ofport->up.pp.config,
 661                                  ofport->up.pp.state, ofport->is_tunnel,
 662                                  ofport->may_enable);
 663             }
 664             xlate_txn_commit();
 665         }
 666
 667         udpif_revalidate(backer->udpif);
 668     }
 669
 670     process_dpif_port_changes(backer);
 671
 672     return 0;
 673 }
 674
 675 /* Check for and handle port changes in 'backer''s dpif. */
 676 static void
 677 process_dpif_port_changes(struct dpif_backer *backer)
 678 {
 679     for (;;) {
 680         char *devname;
 681         int error;
 682
 683         error = dpif_port_poll(backer->dpif, &devname);
 684         switch (error) {
 685         case EAGAIN:
 686             return;
 687
 688         case ENOBUFS:
 689             process_dpif_all_ports_changed(backer);
 690             break;
 691
 692         case 0:
 693             process_dpif_port_change(backer, devname);
 694             free(devname);
 695             break;
 696
 697         default:
 698             process_dpif_port_error(backer, error);
 699             break;
 700         }
 701     }
 702 }
 703
 704 static void
 705 process_dpif_all_ports_changed(struct dpif_backer *backer)
 706 {
 707     struct ofproto_dpif *ofproto;
 708     struct dpif_port dpif_port;
 709     struct dpif_port_dump dump;
 710     struct sset devnames;
 711     const char *devname;
 712
 713     sset_init(&devnames);
 714     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 715         if (ofproto->backer == backer) {
 716             struct ofport *ofport;
 717
 718             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 719                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 720             }
 721         }
 722     }
 723     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 724         sset_add(&devnames, dpif_port.name);
 725     }
 726
 727     SSET_FOR_EACH (devname, &devnames) {
 728         process_dpif_port_change(backer, devname);
 729     }
 730     sset_destroy(&devnames);
 731 }
 732
 733 static void
 734 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 735 {
 736     struct ofproto_dpif *ofproto;
 737     struct dpif_port port;
 738
 739     /* Don't report on the datapath's device. */
 740     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 741         return;
 742     }
 743
 744     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 745                    &all_ofproto_dpifs) {
 746         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 747             return;
 748         }
 749     }
 750
 751     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 752     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 753         /* The port was removed.  If we know the datapath,
 754          * report it through poll_set().  If we don't, it may be
 755          * notifying us of a removal we initiated, so ignore it.
 756          * If there's a pending ENOBUFS, let it stand, since
 757          * everything will be reevaluated. */
 758         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 759             sset_add(&ofproto->port_poll_set, devname);
 760             ofproto->port_poll_errno = 0;
 761         }
 762     } else if (!ofproto) {
 763         /* The port was added, but we don't know with which
 764          * ofproto we should associate it.  Delete it. */
 765         dpif_port_del(backer->dpif, port.port_no);
 766     } else {
 767         struct ofport_dpif *ofport;
 768
 769         ofport = ofport_dpif_cast(shash_find_data(
 770                                       &ofproto->up.port_by_name, devname));
 771         if (ofport
 772             && ofport->odp_port != port.port_no
 773             && !odp_port_to_ofport(backer, port.port_no))
 774         {
 775             /* 'ofport''s datapath port number has changed from
 776              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 777              * structures to match. */
 778             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 779             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 780             ofport->odp_port = port.port_no;
 781             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 782                         hash_odp_port(port.port_no));
 783             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 784             backer->need_revalidate = REV_RECONFIGURE;
 785         }
 786     }
 787     dpif_port_destroy(&port);
 788 }
 789
 790 /* Propagate 'error' to all ofprotos based on 'backer'. */
 791 static void
 792 process_dpif_port_error(struct dpif_backer *backer, int error)
 793 {
 794     struct ofproto_dpif *ofproto;
 795
 796     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 797         if (ofproto->backer == backer) {
 798             sset_clear(&ofproto->port_poll_set);
 799             ofproto->port_poll_errno = error;
 800         }
 801     }
 802 }
 803
 804 static void
 805 type_wait(const char *type)
 806 {
 807     struct dpif_backer *backer;
 808
 809     backer = shash_find_data(&all_dpif_backers, type);
 810     if (!backer) {
 811         /* This is not necessarily a problem, since backers are only
 812          * created on demand. */
 813         return;
 814     }
 815
 816     dpif_wait(backer->dpif);
 817 }
 818 \f
 819 /* Basic life-cycle. */
 820
 821 static int add_internal_flows(struct ofproto_dpif *);
 822
 823 static struct ofproto *
 824 alloc(void)
 825 {
 826     struct ofproto_dpif *ofproto = xzalloc(sizeof *ofproto);
 827     return &ofproto->up;
 828 }
 829
 830 static void
 831 dealloc(struct ofproto *ofproto_)
 832 {
 833     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
 834     free(ofproto);
 835 }
 836
 837 static void
 838 close_dpif_backer(struct dpif_backer *backer)
 839 {
 840     ovs_assert(backer->refcount > 0);
 841
 842     if (--backer->refcount) {
 843         return;
 844     }
 845
 846     udpif_destroy(backer->udpif);
 847
 848     simap_destroy(&backer->tnl_backers);
 849     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
 850     hmap_destroy(&backer->odp_to_ofport_map);
 851     shash_find_and_delete(&all_dpif_backers, backer->type);
 852     free(backer->type);
 853     free(backer->dp_version_string);
 854     dpif_close(backer->dpif);
 855     free(backer);
 856 }
 857
 858 /* Datapath port slated for removal from datapath. */
 859 struct odp_garbage {
 860     struct ovs_list list_node;
 861     odp_port_t odp_port;
 862 };
 863
 864 static bool check_variable_length_userdata(struct dpif_backer *backer);
 865 static void check_support(struct dpif_backer *backer);
 866
 867 static int
 868 open_dpif_backer(const char *type, struct dpif_backer **backerp)
 869 {
 870     struct dpif_backer *backer;
 871     struct dpif_port_dump port_dump;
 872     struct dpif_port port;
 873     struct shash_node *node;
 874     struct ovs_list garbage_list;
 875     struct odp_garbage *garbage;
 876
 877     struct sset names;
 878     char *backer_name;
 879     const char *name;
 880     int error;
 881
 882     recirc_init();
 883
 884     backer = shash_find_data(&all_dpif_backers, type);
 885     if (backer) {
 886         backer->refcount++;
 887         *backerp = backer;
 888         return 0;
 889     }
 890
 891     backer_name = xasprintf("ovs-%s", type);
 892
 893     /* Remove any existing datapaths, since we assume we're the only
 894      * userspace controlling the datapath. */
 895     sset_init(&names);
 896     dp_enumerate_names(type, &names);
 897     SSET_FOR_EACH(name, &names) {
 898         struct dpif *old_dpif;
 899
 900         /* Don't remove our backer if it exists. */
 901         if (!strcmp(name, backer_name)) {
 902             continue;
 903         }
 904
 905         if (dpif_open(name, type, &old_dpif)) {
 906             VLOG_WARN("couldn't open old datapath %s to remove it", name);
 907         } else {
 908             dpif_delete(old_dpif);
 909             dpif_close(old_dpif);
 910         }
 911     }
 912     sset_destroy(&names);
 913
 914     backer = xmalloc(sizeof *backer);
 915
 916     error = dpif_create_and_open(backer_name, type, &backer->dpif);
 917     free(backer_name);
 918     if (error) {
 919         VLOG_ERR("failed to open datapath of type %s: %s", type,
 920                  ovs_strerror(error));
 921         free(backer);
 922         return error;
 923     }
 924     backer->udpif = udpif_create(backer, backer->dpif);
 925
 926     backer->type = xstrdup(type);
 927     backer->refcount = 1;
 928     hmap_init(&backer->odp_to_ofport_map);
 929     ovs_rwlock_init(&backer->odp_to_ofport_lock);
 930     backer->need_revalidate = 0;
 931     simap_init(&backer->tnl_backers);
 932     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
 933     *backerp = backer;
 934
 935     if (backer->recv_set_enable) {
 936         dpif_flow_flush(backer->dpif);
 937     }
 938
 939     /* Loop through the ports already on the datapath and remove any
 940      * that we don't need anymore. */
 941     list_init(&garbage_list);
 942     dpif_port_dump_start(&port_dump, backer->dpif);
 943     while (dpif_port_dump_next(&port_dump, &port)) {
 944         node = shash_find(&init_ofp_ports, port.name);
 945         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
 946             garbage = xmalloc(sizeof *garbage);
 947             garbage->odp_port = port.port_no;
 948             list_push_front(&garbage_list, &garbage->list_node);
 949         }
 950     }
 951     dpif_port_dump_done(&port_dump);
 952
 953     LIST_FOR_EACH_POP (garbage, list_node, &garbage_list) {
 954         dpif_port_del(backer->dpif, garbage->odp_port);
 955         free(garbage);
 956     }
 957
 958     shash_add(&all_dpif_backers, type, backer);
 959
 960     check_support(backer);
 961     atomic_count_init(&backer->tnl_count, 0);
 962
 963     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 964     if (error) {
 965         VLOG_ERR("failed to listen on datapath of type %s: %s",
 966                  type, ovs_strerror(error));
 967         close_dpif_backer(backer);
 968         return error;
 969     }
 970
 971     if (backer->recv_set_enable) {
 972         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 973     }
 974
 975     /* This check fails if performed before udpif threads have been set,
 976      * as the kernel module checks that the 'pid' in userspace action
 977      * is non-zero. */
 978     backer->support.variable_length_userdata
 979         = check_variable_length_userdata(backer);
 980     backer->dp_version_string = dpif_get_dp_version(backer->dpif);
 981
 982     return error;
 983 }
 984
 985 bool
 986 ovs_native_tunneling_is_on(struct ofproto_dpif *ofproto)
 987 {
 988     return ofproto_use_tnl_push_pop && ofproto->backer->support.tnl_push_pop &&
 989            atomic_count_get(&ofproto->backer->tnl_count);
 990 }
 991
 992 /* Tests whether 'backer''s datapath supports recirculation.  Only newer
 993  * datapaths support OVS_KEY_ATTR_RECIRC_ID in keys.  We need to disable some
 994  * features on older datapaths that don't support this feature.
 995  *
 996  * Returns false if 'backer' definitely does not support recirculation, true if
 997  * it seems to support recirculation or if at least the error we get is
 998  * ambiguous. */
 999 static bool
1000 check_recirc(struct dpif_backer *backer)
1001 {
1002     struct flow flow;
1003     struct odputil_keybuf keybuf;
1004     struct ofpbuf key;
1005     bool enable_recirc;
1006     struct odp_flow_key_parms odp_parms = {
1007         .flow = &flow,
1008         .recirc = true,
1009     };
1010
1011     memset(&flow, 0, sizeof flow);
1012     flow.recirc_id = 1;
1013     flow.dp_hash = 1;
1014
1015     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1016     odp_flow_key_from_flow(&odp_parms, &key);
1017     enable_recirc = dpif_probe_feature(backer->dpif, "recirculation", &key,
1018                                        NULL);
1019
1020     if (enable_recirc) {
1021         VLOG_INFO("%s: Datapath supports recirculation",
1022                   dpif_name(backer->dpif));
1023     } else {
1024         VLOG_INFO("%s: Datapath does not support recirculation",
1025                   dpif_name(backer->dpif));
1026     }
1027
1028     return enable_recirc;
1029 }
1030
1031 /* Tests whether 'dpif' supports unique flow ids. We can skip serializing
1032  * some flow attributes for datapaths that support this feature.
1033  *
1034  * Returns true if 'dpif' supports UFID for flow operations.
1035  * Returns false if  'dpif' does not support UFID. */
1036 static bool
1037 check_ufid(struct dpif_backer *backer)
1038 {
1039     struct flow flow;
1040     struct odputil_keybuf keybuf;
1041     struct ofpbuf key;
1042     ovs_u128 ufid;
1043     bool enable_ufid;
1044     struct odp_flow_key_parms odp_parms = {
1045         .flow = &flow,
1046     };
1047
1048     memset(&flow, 0, sizeof flow);
1049     flow.dl_type = htons(0x1234);
1050
1051     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1052     odp_flow_key_from_flow(&odp_parms, &key);
1053     dpif_flow_hash(backer->dpif, key.data, key.size, &ufid);
1054
1055     enable_ufid = dpif_probe_feature(backer->dpif, "UFID", &key, &ufid);
1056
1057     if (enable_ufid) {
1058         VLOG_INFO("%s: Datapath supports unique flow ids",
1059                   dpif_name(backer->dpif));
1060     } else {
1061         VLOG_INFO("%s: Datapath does not support unique flow ids",
1062                   dpif_name(backer->dpif));
1063     }
1064     return enable_ufid;
1065 }
1066
1067 /* Tests whether 'backer''s datapath supports variable-length
1068  * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.  We need
1069  * to disable some features on older datapaths that don't support this
1070  * feature.
1071  *
1072  * Returns false if 'backer' definitely does not support variable-length
1073  * userdata, true if it seems to support them or if at least the error we get
1074  * is ambiguous. */
1075 static bool
1076 check_variable_length_userdata(struct dpif_backer *backer)
1077 {
1078     struct eth_header *eth;
1079     struct ofpbuf actions;
1080     struct dpif_execute execute;
1081     struct dp_packet packet;
1082     size_t start;
1083     int error;
1084
1085     /* Compose a userspace action that will cause an ERANGE error on older
1086      * datapaths that don't support variable-length userdata.
1087      *
1088      * We really test for using userdata longer than 8 bytes, but older
1089      * datapaths accepted these, silently truncating the userdata to 8 bytes.
1090      * The same older datapaths rejected userdata shorter than 8 bytes, so we
1091      * test for that instead as a proxy for longer userdata support. */
1092     ofpbuf_init(&actions, 64);
1093     start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
1094     nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
1095                    dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
1096     nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
1097     nl_msg_end_nested(&actions, start);
1098
1099     /* Compose a dummy ethernet packet. */
1100     dp_packet_init(&packet, ETH_HEADER_LEN);
1101     eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1102     eth->eth_type = htons(0x1234);
1103
1104     /* Execute the actions.  On older datapaths this fails with ERANGE, on
1105      * newer datapaths it succeeds. */
1106     execute.actions = actions.data;
1107     execute.actions_len = actions.size;
1108     execute.packet = &packet;
1109     execute.needs_help = false;
1110     execute.probe = true;
1111
1112     error = dpif_execute(backer->dpif, &execute);
1113
1114     dp_packet_uninit(&packet);
1115     ofpbuf_uninit(&actions);
1116
1117     switch (error) {
1118     case 0:
1119         return true;
1120
1121     case ERANGE:
1122         /* Variable-length userdata is not supported. */
1123         VLOG_WARN("%s: datapath does not support variable-length userdata "
1124                   "feature (needs Linux 3.10+ or kernel module from OVS "
1125                   "1..11+).  The NXAST_SAMPLE action will be ignored.",
1126                   dpif_name(backer->dpif));
1127         return false;
1128
1129     default:
1130         /* Something odd happened.  We're not sure whether variable-length
1131          * userdata is supported.  Default to "yes". */
1132         VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1133                   dpif_name(backer->dpif), ovs_strerror(error));
1134         return true;
1135     }
1136 }
1137
1138 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1139  *
1140  * Returns the number of elements in a struct flow's mpls_lse field
1141  * if the datapath supports at least that many entries in an
1142  * MPLS label stack.
1143  * Otherwise returns the number of MPLS push actions supported by
1144  * the datapath. */
1145 static size_t
1146 check_max_mpls_depth(struct dpif_backer *backer)
1147 {
1148     struct flow flow;
1149     int n;
1150
1151     for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
1152         struct odputil_keybuf keybuf;
1153         struct ofpbuf key;
1154         struct odp_flow_key_parms odp_parms = {
1155             .flow = &flow,
1156         };
1157
1158         memset(&flow, 0, sizeof flow);
1159         flow.dl_type = htons(ETH_TYPE_MPLS);
1160         flow_set_mpls_bos(&flow, n, 1);
1161
1162         ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1163         odp_flow_key_from_flow(&odp_parms, &key);
1164         if (!dpif_probe_feature(backer->dpif, "MPLS", &key, NULL)) {
1165             break;
1166         }
1167     }
1168
1169     VLOG_INFO("%s: MPLS label stack length probed as %d",
1170               dpif_name(backer->dpif), n);
1171     return n;
1172 }
1173
1174 /* Tests whether 'backer''s datapath supports masked data in
1175  * OVS_ACTION_ATTR_SET actions.  We need to disable some features on older
1176  * datapaths that don't support this feature. */
1177 static bool
1178 check_masked_set_action(struct dpif_backer *backer)
1179 {
1180     struct eth_header *eth;
1181     struct ofpbuf actions;
1182     struct dpif_execute execute;
1183     struct dp_packet packet;
1184     int error;
1185     struct ovs_key_ethernet key, mask;
1186
1187     /* Compose a set action that will cause an EINVAL error on older
1188      * datapaths that don't support masked set actions.
1189      * Avoid using a full mask, as it could be translated to a non-masked
1190      * set action instead. */
1191     ofpbuf_init(&actions, 64);
1192     memset(&key, 0x53, sizeof key);
1193     memset(&mask, 0x7f, sizeof mask);
1194     commit_masked_set_action(&actions, OVS_KEY_ATTR_ETHERNET, &key, &mask,
1195                              sizeof key);
1196
1197     /* Compose a dummy ethernet packet. */
1198     dp_packet_init(&packet, ETH_HEADER_LEN);
1199     eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1200     eth->eth_type = htons(0x1234);
1201
1202     /* Execute the actions.  On older datapaths this fails with EINVAL, on
1203      * newer datapaths it succeeds. */
1204     execute.actions = actions.data;
1205     execute.actions_len = actions.size;
1206     execute.packet = &packet;
1207     execute.needs_help = false;
1208     execute.probe = true;
1209
1210     error = dpif_execute(backer->dpif, &execute);
1211
1212     dp_packet_uninit(&packet);
1213     ofpbuf_uninit(&actions);
1214
1215     if (error) {
1216         /* Masked set action is not supported. */
1217         VLOG_INFO("%s: datapath does not support masked set action feature.",
1218                   dpif_name(backer->dpif));
1219     }
1220     return !error;
1221 }
1222
1223 static void
1224 check_support(struct dpif_backer *backer)
1225 {
1226     /* This feature needs to be tested after udpif threads are set. */
1227     backer->support.variable_length_userdata = false;
1228
1229     backer->support.recirc = check_recirc(backer);
1230     backer->support.max_mpls_depth = check_max_mpls_depth(backer);
1231     backer->support.masked_set_action = check_masked_set_action(backer);
1232     backer->support.ufid = check_ufid(backer);
1233     backer->support.tnl_push_pop = dpif_supports_tnl_push_pop(backer->dpif);
1234 }
1235
1236 static int
1237 construct(struct ofproto *ofproto_)
1238 {
1239     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1240     struct shash_node *node, *next;
1241     int error;
1242
1243     /* Tunnel module can get used right after the udpif threads are running. */
1244     ofproto_tunnel_init();
1245
1246     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1247     if (error) {
1248         return error;
1249     }
1250
1251     atomic_init(&ofproto->tables_version, CLS_MIN_VERSION);
1252     ofproto->netflow = NULL;
1253     ofproto->sflow = NULL;
1254     ofproto->ipfix = NULL;
1255     ofproto->stp = NULL;
1256     ofproto->rstp = NULL;
1257     ofproto->dump_seq = 0;
1258     hmap_init(&ofproto->bundles);
1259     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1260     ofproto->ms = NULL;
1261     ofproto->mbridge = mbridge_create();
1262     ofproto->has_bonded_bundles = false;
1263     ofproto->lacp_enabled = false;
1264     ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1265     ovs_mutex_init(&ofproto->vsp_mutex);
1266
1267     guarded_list_init(&ofproto->pins);
1268
1269     ofproto_unixctl_init();
1270
1271     hmap_init(&ofproto->vlandev_map);
1272     hmap_init(&ofproto->realdev_vid_map);
1273
1274     sset_init(&ofproto->ports);
1275     sset_init(&ofproto->ghost_ports);
1276     sset_init(&ofproto->port_poll_set);
1277     ofproto->port_poll_errno = 0;
1278     ofproto->change_seq = 0;
1279     ofproto->pins_seq = seq_create();
1280     ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1281
1282
1283     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1284         struct iface_hint *iface_hint = node->data;
1285
1286         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1287             /* Check if the datapath already has this port. */
1288             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1289                 sset_add(&ofproto->ports, node->name);
1290             }
1291
1292             free(iface_hint->br_name);
1293             free(iface_hint->br_type);
1294             free(iface_hint);
1295             shash_delete(&init_ofp_ports, node);
1296         }
1297     }
1298
1299     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1300                 hash_string(ofproto->up.name, 0));
1301     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1302
1303     ofproto_init_tables(ofproto_, N_TABLES);
1304     error = add_internal_flows(ofproto);
1305
1306     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1307
1308     return error;
1309 }
1310
1311 static int
1312 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1313                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1314 {
1315     struct match match;
1316     int error;
1317     struct rule *rule;
1318
1319     match_init_catchall(&match);
1320     match_set_reg(&match, 0, id);
1321
1322     error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, 0, ofpacts,
1323                                            &rule);
1324     *rulep = error ? NULL : rule_dpif_cast(rule);
1325
1326     return error;
1327 }
1328
1329 static int
1330 add_internal_flows(struct ofproto_dpif *ofproto)
1331 {
1332     struct ofpact_controller *controller;
1333     uint64_t ofpacts_stub[128 / 8];
1334     struct ofpbuf ofpacts;
1335     struct rule *unused_rulep OVS_UNUSED;
1336     struct match match;
1337     int error;
1338     int id;
1339
1340     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1341     id = 1;
1342
1343     controller = ofpact_put_CONTROLLER(&ofpacts);
1344     controller->max_len = UINT16_MAX;
1345     controller->controller_id = 0;
1346     controller->reason = OFPR_NO_MATCH;
1347     ofpact_pad(&ofpacts);
1348
1349     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1350                                    &ofproto->miss_rule);
1351     if (error) {
1352         return error;
1353     }
1354
1355     ofpbuf_clear(&ofpacts);
1356     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1357                                    &ofproto->no_packet_in_rule);
1358     if (error) {
1359         return error;
1360     }
1361
1362     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1363                                    &ofproto->drop_frags_rule);
1364     if (error) {
1365         return error;
1366     }
1367
1368     /* Drop any run away non-recirc rule lookups. Recirc_id has to be
1369      * zero when reaching this rule.
1370      *
1371      * (priority=2), recirc_id=0, actions=drop
1372      */
1373     ofpbuf_clear(&ofpacts);
1374     match_init_catchall(&match);
1375     match_set_recirc_id(&match, 0);
1376     error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, 0, &ofpacts,
1377                                            &unused_rulep);
1378     return error;
1379 }
1380
1381 static void
1382 destruct(struct ofproto *ofproto_)
1383 {
1384     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1385     struct ofproto_packet_in *pin;
1386     struct rule_dpif *rule;
1387     struct oftable *table;
1388     struct ovs_list pins;
1389
1390     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1391     xlate_txn_start();
1392     xlate_remove_ofproto(ofproto);
1393     xlate_txn_commit();
1394
1395     /* Ensure that the upcall processing threads have no remaining references
1396      * to the ofproto or anything in it. */
1397     udpif_synchronize(ofproto->backer->udpif);
1398
1399     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1400
1401     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1402         CLS_FOR_EACH (rule, up.cr, &table->cls) {
1403             ofproto_rule_delete(&ofproto->up, &rule->up);
1404         }
1405     }
1406
1407     guarded_list_pop_all(&ofproto->pins, &pins);
1408     LIST_FOR_EACH_POP (pin, list_node, &pins) {
1409         free(CONST_CAST(void *, pin->up.packet));
1410         free(pin);
1411     }
1412     guarded_list_destroy(&ofproto->pins);
1413
1414     recirc_free_ofproto(ofproto, ofproto->up.name);
1415
1416     mbridge_unref(ofproto->mbridge);
1417
1418     netflow_unref(ofproto->netflow);
1419     dpif_sflow_unref(ofproto->sflow);
1420     dpif_ipfix_unref(ofproto->ipfix);
1421     hmap_destroy(&ofproto->bundles);
1422     mac_learning_unref(ofproto->ml);
1423     mcast_snooping_unref(ofproto->ms);
1424
1425     hmap_destroy(&ofproto->vlandev_map);
1426     hmap_destroy(&ofproto->realdev_vid_map);
1427
1428     sset_destroy(&ofproto->ports);
1429     sset_destroy(&ofproto->ghost_ports);
1430     sset_destroy(&ofproto->port_poll_set);
1431
1432     ovs_mutex_destroy(&ofproto->stats_mutex);
1433     ovs_mutex_destroy(&ofproto->vsp_mutex);
1434
1435     seq_destroy(ofproto->pins_seq);
1436
1437     close_dpif_backer(ofproto->backer);
1438 }
1439
1440 static int
1441 run(struct ofproto *ofproto_)
1442 {
1443     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1444     uint64_t new_seq, new_dump_seq;
1445
1446     if (mbridge_need_revalidate(ofproto->mbridge)) {
1447         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1448         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1449         mac_learning_flush(ofproto->ml);
1450         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1451         mcast_snooping_mdb_flush(ofproto->ms);
1452     }
1453
1454     /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
1455      * flow restore.  Even though nothing is processed during flow restore,
1456      * all queued 'pins' will be handled immediately when flow restore
1457      * completes. */
1458     ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1459
1460     /* Do not perform any periodic activity required by 'ofproto' while
1461      * waiting for flow restore to complete. */
1462     if (!ofproto_get_flow_restore_wait()) {
1463         struct ofproto_packet_in *pin;
1464         struct ovs_list pins;
1465
1466         guarded_list_pop_all(&ofproto->pins, &pins);
1467         LIST_FOR_EACH_POP (pin, list_node, &pins) {
1468             connmgr_send_packet_in(ofproto->up.connmgr, pin);
1469             free(CONST_CAST(void *, pin->up.packet));
1470             free(pin);
1471         }
1472     }
1473
1474     if (ofproto->netflow) {
1475         netflow_run(ofproto->netflow);
1476     }
1477     if (ofproto->sflow) {
1478         dpif_sflow_run(ofproto->sflow);
1479     }
1480     if (ofproto->ipfix) {
1481         dpif_ipfix_run(ofproto->ipfix);
1482     }
1483
1484     new_seq = seq_read(connectivity_seq_get());
1485     if (ofproto->change_seq != new_seq) {
1486         struct ofport_dpif *ofport;
1487
1488         HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1489             port_run(ofport);
1490         }
1491
1492         ofproto->change_seq = new_seq;
1493     }
1494     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1495         struct ofbundle *bundle;
1496
1497         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1498             bundle_run(bundle);
1499         }
1500     }
1501
1502     stp_run(ofproto);
1503     rstp_run(ofproto);
1504     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1505     if (mac_learning_run(ofproto->ml)) {
1506         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1507     }
1508     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1509
1510     if (mcast_snooping_run(ofproto->ms)) {
1511         ofproto->backer->need_revalidate = REV_MCAST_SNOOPING;
1512     }
1513
1514     new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1515     if (ofproto->dump_seq != new_dump_seq) {
1516         struct rule *rule, *next_rule;
1517
1518         /* We know stats are relatively fresh, so now is a good time to do some
1519          * periodic work. */
1520         ofproto->dump_seq = new_dump_seq;
1521
1522         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1523          * has passed. */
1524         ovs_mutex_lock(&ofproto_mutex);
1525         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1526                             &ofproto->up.expirable) {
1527             rule_expire(rule_dpif_cast(rule));
1528         }
1529         ovs_mutex_unlock(&ofproto_mutex);
1530
1531         /* All outstanding data in existing flows has been accounted, so it's a
1532          * good time to do bond rebalancing. */
1533         if (ofproto->has_bonded_bundles) {
1534             struct ofbundle *bundle;
1535
1536             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1537                 if (bundle->bond) {
1538                     bond_rebalance(bundle->bond);
1539                 }
1540             }
1541         }
1542     }
1543     return 0;
1544 }
1545
1546 static void
1547 wait(struct ofproto *ofproto_)
1548 {
1549     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1550
1551     if (ofproto_get_flow_restore_wait()) {
1552         return;
1553     }
1554
1555     if (ofproto->sflow) {
1556         dpif_sflow_wait(ofproto->sflow);
1557     }
1558     if (ofproto->ipfix) {
1559         dpif_ipfix_wait(ofproto->ipfix);
1560     }
1561     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1562         struct ofbundle *bundle;
1563
1564         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1565             bundle_wait(bundle);
1566         }
1567     }
1568     if (ofproto->netflow) {
1569         netflow_wait(ofproto->netflow);
1570     }
1571     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1572     mac_learning_wait(ofproto->ml);
1573     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1574     mcast_snooping_wait(ofproto->ms);
1575     stp_wait(ofproto);
1576     if (ofproto->backer->need_revalidate) {
1577         /* Shouldn't happen, but if it does just go around again. */
1578         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1579         poll_immediate_wake();
1580     }
1581
1582     seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1583     seq_wait(ofproto->pins_seq, ofproto->pins_seqno);
1584 }
1585
1586 static void
1587 type_get_memory_usage(const char *type, struct simap *usage)
1588 {
1589     struct dpif_backer *backer;
1590
1591     backer = shash_find_data(&all_dpif_backers, type);
1592     if (backer) {
1593         udpif_get_memory_usage(backer->udpif, usage);
1594     }
1595 }
1596
1597 static void
1598 flush(struct ofproto *ofproto_)
1599 {
1600     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1601     struct dpif_backer *backer = ofproto->backer;
1602
1603     if (backer) {
1604         udpif_flush(backer->udpif);
1605     }
1606 }
1607
1608 static void
1609 query_tables(struct ofproto *ofproto,
1610              struct ofputil_table_features *features,
1611              struct ofputil_table_stats *stats)
1612 {
1613     strcpy(features->name, "classifier");
1614
1615     if (stats) {
1616         int i;
1617
1618         for (i = 0; i < ofproto->n_tables; i++) {
1619             unsigned long missed, matched;
1620
1621             atomic_read_relaxed(&ofproto->tables[i].n_matched, &matched);
1622             atomic_read_relaxed(&ofproto->tables[i].n_missed, &missed);
1623
1624             stats[i].matched_count = matched;
1625             stats[i].lookup_count = matched + missed;
1626         }
1627     }
1628 }
1629
1630 static void
1631 set_tables_version(struct ofproto *ofproto_, cls_version_t version)
1632 {
1633     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1634
1635     atomic_store_relaxed(&ofproto->tables_version, version);
1636 }
1637
1638
1639 static struct ofport *
1640 port_alloc(void)
1641 {
1642     struct ofport_dpif *port = xzalloc(sizeof *port);
1643     return &port->up;
1644 }
1645
1646 static void
1647 port_dealloc(struct ofport *port_)
1648 {
1649     struct ofport_dpif *port = ofport_dpif_cast(port_);
1650     free(port);
1651 }
1652
1653 static int
1654 port_construct(struct ofport *port_)
1655 {
1656     struct ofport_dpif *port = ofport_dpif_cast(port_);
1657     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1658     const struct netdev *netdev = port->up.netdev;
1659     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1660     const char *dp_port_name;
1661     struct dpif_port dpif_port;
1662     int error;
1663
1664     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1665     port->bundle = NULL;
1666     port->cfm = NULL;
1667     port->bfd = NULL;
1668     port->lldp = NULL;
1669     port->may_enable = false;
1670     port->stp_port = NULL;
1671     port->stp_state = STP_DISABLED;
1672     port->rstp_port = NULL;
1673     port->rstp_state = RSTP_DISABLED;
1674     port->is_tunnel = false;
1675     port->peer = NULL;
1676     port->qdscp = NULL;
1677     port->n_qdscp = 0;
1678     port->realdev_ofp_port = 0;
1679     port->vlandev_vid = 0;
1680     port->carrier_seq = netdev_get_carrier_resets(netdev);
1681     port->is_layer3 = netdev_vport_is_layer3(netdev);
1682
1683     if (netdev_vport_is_patch(netdev)) {
1684         /* By bailing out here, we don't submit the port to the sFlow module
1685          * to be considered for counter polling export.  This is correct
1686          * because the patch port represents an interface that sFlow considers
1687          * to be "internal" to the switch as a whole, and therefore not a
1688          * candidate for counter polling. */
1689         port->odp_port = ODPP_NONE;
1690         ofport_update_peer(port);
1691         return 0;
1692     }
1693
1694     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1695     error = dpif_port_query_by_name(ofproto->backer->dpif, dp_port_name,
1696                                     &dpif_port);
1697     if (error) {
1698         return error;
1699     }
1700
1701     port->odp_port = dpif_port.port_no;
1702
1703     if (netdev_get_tunnel_config(netdev)) {
1704         atomic_count_inc(&ofproto->backer->tnl_count);
1705         error = tnl_port_add(port, port->up.netdev, port->odp_port,
1706                              ovs_native_tunneling_is_on(ofproto), dp_port_name);
1707         if (error) {
1708             atomic_count_dec(&ofproto->backer->tnl_count);
1709             dpif_port_destroy(&dpif_port);
1710             return error;
1711         }
1712
1713         port->is_tunnel = true;
1714         if (ofproto->ipfix) {
1715            dpif_ipfix_add_tunnel_port(ofproto->ipfix, port_, port->odp_port);
1716         }
1717     } else {
1718         /* Sanity-check that a mapping doesn't already exist.  This
1719          * shouldn't happen for non-tunnel ports. */
1720         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1721             VLOG_ERR("port %s already has an OpenFlow port number",
1722                      dpif_port.name);
1723             dpif_port_destroy(&dpif_port);
1724             return EBUSY;
1725         }
1726
1727         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1728         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1729                     hash_odp_port(port->odp_port));
1730         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1731     }
1732     dpif_port_destroy(&dpif_port);
1733
1734     if (ofproto->sflow) {
1735         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1736     }
1737
1738     return 0;
1739 }
1740
1741 static void
1742 port_destruct(struct ofport *port_)
1743 {
1744     struct ofport_dpif *port = ofport_dpif_cast(port_);
1745     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1746     const char *devname = netdev_get_name(port->up.netdev);
1747     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1748     const char *dp_port_name;
1749
1750     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1751     xlate_txn_start();
1752     xlate_ofport_remove(port);
1753     xlate_txn_commit();
1754
1755     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1756                                               sizeof namebuf);
1757     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1758         /* The underlying device is still there, so delete it.  This
1759          * happens when the ofproto is being destroyed, since the caller
1760          * assumes that removal of attached ports will happen as part of
1761          * destruction. */
1762         if (!port->is_tunnel) {
1763             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1764         }
1765     }
1766
1767     if (port->peer) {
1768         port->peer->peer = NULL;
1769         port->peer = NULL;
1770     }
1771
1772     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1773         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1774         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1775         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1776     }
1777
1778     if (port->is_tunnel) {
1779         atomic_count_dec(&ofproto->backer->tnl_count);
1780     }
1781
1782     if (port->is_tunnel && ofproto->ipfix) {
1783        dpif_ipfix_del_tunnel_port(ofproto->ipfix, port->odp_port);
1784     }
1785
1786     tnl_port_del(port);
1787     sset_find_and_delete(&ofproto->ports, devname);
1788     sset_find_and_delete(&ofproto->ghost_ports, devname);
1789     bundle_remove(port_);
1790     set_cfm(port_, NULL);
1791     set_bfd(port_, NULL);
1792     set_lldp(port_, NULL);
1793     if (port->stp_port) {
1794         stp_port_disable(port->stp_port);
1795     }
1796     set_rstp_port(port_, NULL);
1797     if (ofproto->sflow) {
1798         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1799     }
1800
1801     free(port->qdscp);
1802 }
1803
1804 static void
1805 port_modified(struct ofport *port_)
1806 {
1807     struct ofport_dpif *port = ofport_dpif_cast(port_);
1808     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1809     const char *dp_port_name;
1810     struct netdev *netdev = port->up.netdev;
1811
1812     if (port->bundle && port->bundle->bond) {
1813         bond_slave_set_netdev(port->bundle->bond, port, netdev);
1814     }
1815
1816     if (port->cfm) {
1817         cfm_set_netdev(port->cfm, netdev);
1818     }
1819
1820     if (port->bfd) {
1821         bfd_set_netdev(port->bfd, netdev);
1822     }
1823
1824     ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1825                                      port->lldp, port->up.pp.hw_addr);
1826
1827     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1828
1829     if (port->is_tunnel) {
1830         struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1831
1832         if (tnl_port_reconfigure(port, netdev, port->odp_port,
1833                                  ovs_native_tunneling_is_on(ofproto),
1834                                  dp_port_name)) {
1835             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1836         }
1837     }
1838
1839     ofport_update_peer(port);
1840 }
1841
1842 static void
1843 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1844 {
1845     struct ofport_dpif *port = ofport_dpif_cast(port_);
1846     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1847     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1848
1849     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1850                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1851                    OFPUTIL_PC_NO_PACKET_IN)) {
1852         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1853
1854         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1855             bundle_update(port->bundle);
1856         }
1857     }
1858 }
1859
1860 static int
1861 set_sflow(struct ofproto *ofproto_,
1862           const struct ofproto_sflow_options *sflow_options)
1863 {
1864     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1865     struct dpif_sflow *ds = ofproto->sflow;
1866
1867     if (sflow_options) {
1868         uint32_t old_probability = ds ? dpif_sflow_get_probability(ds) : 0;
1869         if (!ds) {
1870             struct ofport_dpif *ofport;
1871
1872             ds = ofproto->sflow = dpif_sflow_create();
1873             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1874                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1875             }
1876         }
1877         dpif_sflow_set_options(ds, sflow_options);
1878         if (dpif_sflow_get_probability(ds) != old_probability) {
1879             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1880         }
1881     } else {
1882         if (ds) {
1883             dpif_sflow_unref(ds);
1884             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1885             ofproto->sflow = NULL;
1886         }
1887     }
1888     return 0;
1889 }
1890
1891 static int
1892 set_ipfix(
1893     struct ofproto *ofproto_,
1894     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1895     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1896     size_t n_flow_exporters_options)
1897 {
1898     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1899     struct dpif_ipfix *di = ofproto->ipfix;
1900     bool has_options = bridge_exporter_options || flow_exporters_options;
1901     bool new_di = false;
1902
1903     if (has_options && !di) {
1904         di = ofproto->ipfix = dpif_ipfix_create();
1905         new_di = true;
1906     }
1907
1908     if (di) {
1909         /* Call set_options in any case to cleanly flush the flow
1910          * caches in the last exporters that are to be destroyed. */
1911         dpif_ipfix_set_options(
1912             di, bridge_exporter_options, flow_exporters_options,
1913             n_flow_exporters_options);
1914
1915         /* Add tunnel ports only when a new ipfix created */
1916         if (new_di == true) {
1917             struct ofport_dpif *ofport;
1918             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1919                 if (ofport->is_tunnel == true) {
1920                     dpif_ipfix_add_tunnel_port(di, &ofport->up, ofport->odp_port);
1921                 }
1922             }
1923         }
1924
1925         if (!has_options) {
1926             dpif_ipfix_unref(di);
1927             ofproto->ipfix = NULL;
1928         }
1929     }
1930
1931     return 0;
1932 }
1933
1934 static int
1935 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1936 {
1937     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1938     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1939     struct cfm *old = ofport->cfm;
1940     int error = 0;
1941
1942     if (s) {
1943         if (!ofport->cfm) {
1944             ofport->cfm = cfm_create(ofport->up.netdev);
1945         }
1946
1947         if (cfm_configure(ofport->cfm, s)) {
1948             error = 0;
1949             goto out;
1950         }
1951
1952         error = EINVAL;
1953     }
1954     cfm_unref(ofport->cfm);
1955     ofport->cfm = NULL;
1956 out:
1957     if (ofport->cfm != old) {
1958         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1959     }
1960     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1961                                      ofport->lldp, ofport->up.pp.hw_addr);
1962     return error;
1963 }
1964
1965 static bool
1966 cfm_status_changed(struct ofport *ofport_)
1967 {
1968     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1969
1970     return ofport->cfm ? cfm_check_status_change(ofport->cfm) : true;
1971 }
1972
1973 static int
1974 get_cfm_status(const struct ofport *ofport_,
1975                struct cfm_status *status)
1976 {
1977     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1978     int ret = 0;
1979
1980     if (ofport->cfm) {
1981         cfm_get_status(ofport->cfm, status);
1982     } else {
1983         ret = ENOENT;
1984     }
1985
1986     return ret;
1987 }
1988
1989 static int
1990 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1991 {
1992     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1993     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1994     struct bfd *old;
1995
1996     old = ofport->bfd;
1997     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1998                                 cfg, ofport->up.netdev);
1999     if (ofport->bfd != old) {
2000         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2001     }
2002     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
2003                                      ofport->lldp, ofport->up.pp.hw_addr);
2004     return 0;
2005 }
2006
2007 static bool
2008 bfd_status_changed(struct ofport *ofport_)
2009 {
2010     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2011
2012     return ofport->bfd ? bfd_check_status_change(ofport->bfd) : true;
2013 }
2014
2015 static int
2016 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2017 {
2018     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2019     int ret = 0;
2020
2021     if (ofport->bfd) {
2022         bfd_get_status(ofport->bfd, smap);
2023     } else {
2024         ret = ENOENT;
2025     }
2026
2027     return ret;
2028 }
2029
2030 static int
2031 set_lldp(struct ofport *ofport_,
2032          const struct smap *cfg)
2033 {
2034     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2035     int error = 0;
2036
2037     if (cfg) {
2038         if (!ofport->lldp) {
2039             struct ofproto_dpif *ofproto;
2040
2041             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2042             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2043             ofport->lldp = lldp_create(ofport->up.netdev, ofport_->mtu, cfg);
2044         }
2045
2046         if (!lldp_configure(ofport->lldp, cfg)) {
2047             error = EINVAL;
2048         }
2049     }
2050     if (error) {
2051         lldp_unref(ofport->lldp);
2052         ofport->lldp = NULL;
2053     }
2054
2055     ofproto_dpif_monitor_port_update(ofport,
2056                                      ofport->bfd,
2057                                      ofport->cfm,
2058                                      ofport->lldp,
2059                                      ofport->up.pp.hw_addr);
2060     return error;
2061 }
2062
2063 static bool
2064 get_lldp_status(const struct ofport *ofport_,
2065                struct lldp_status *status OVS_UNUSED)
2066 {
2067     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2068
2069     return ofport->lldp ? true : false;
2070 }
2071
2072 static int
2073 set_aa(struct ofproto *ofproto OVS_UNUSED,
2074        const struct aa_settings *s)
2075 {
2076     return aa_configure(s);
2077 }
2078
2079 static int
2080 aa_mapping_set(struct ofproto *ofproto_ OVS_UNUSED, void *aux,
2081                const struct aa_mapping_settings *s)
2082 {
2083     return aa_mapping_register(aux, s);
2084 }
2085
2086 static int
2087 aa_mapping_unset(struct ofproto *ofproto OVS_UNUSED, void *aux)
2088 {
2089     return aa_mapping_unregister(aux);
2090 }
2091
2092 static int
2093 aa_vlan_get_queued(struct ofproto *ofproto OVS_UNUSED, struct ovs_list *list)
2094 {
2095     return aa_get_vlan_queued(list);
2096 }
2097
2098 static unsigned int
2099 aa_vlan_get_queue_size(struct ofproto *ofproto OVS_UNUSED)
2100 {
2101     return aa_get_vlan_queue_size();
2102 }
2103
2104 \f
2105 /* Spanning Tree. */
2106
2107 /* Called while rstp_mutex is held. */
2108 static void
2109 rstp_send_bpdu_cb(struct dp_packet *pkt, void *ofport_, void *ofproto_)
2110 {
2111     struct ofproto_dpif *ofproto = ofproto_;
2112     struct ofport_dpif *ofport = ofport_;
2113     struct eth_header *eth = dp_packet_l2(pkt);
2114
2115     netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2116     if (eth_addr_is_zero(eth->eth_src)) {
2117         VLOG_WARN_RL(&rl, "%s port %d: cannot send RSTP BPDU on a port which "
2118                      "does not have a configured source MAC address.",
2119                      ofproto->up.name, ofp_to_u16(ofport->up.ofp_port));
2120     } else {
2121         ofproto_dpif_send_packet(ofport, pkt);
2122     }
2123     dp_packet_delete(pkt);
2124 }
2125
2126 static void
2127 send_bpdu_cb(struct dp_packet *pkt, int port_num, void *ofproto_)
2128 {
2129     struct ofproto_dpif *ofproto = ofproto_;
2130     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2131     struct ofport_dpif *ofport;
2132
2133     ofport = stp_port_get_aux(sp);
2134     if (!ofport) {
2135         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2136                      ofproto->up.name, port_num);
2137     } else {
2138         struct eth_header *eth = dp_packet_l2(pkt);
2139
2140         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2141         if (eth_addr_is_zero(eth->eth_src)) {
2142             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2143                          "with unknown MAC", ofproto->up.name, port_num);
2144         } else {
2145             ofproto_dpif_send_packet(ofport, pkt);
2146         }
2147     }
2148     dp_packet_delete(pkt);
2149 }
2150
2151 /* Configure RSTP on 'ofproto_' using the settings defined in 's'. */
2152 static void
2153 set_rstp(struct ofproto *ofproto_, const struct ofproto_rstp_settings *s)
2154 {
2155     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2156
2157     /* Only revalidate flows if the configuration changed. */
2158     if (!s != !ofproto->rstp) {
2159         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2160     }
2161
2162     if (s) {
2163         if (!ofproto->rstp) {
2164             ofproto->rstp = rstp_create(ofproto_->name, s->address,
2165                                         rstp_send_bpdu_cb, ofproto);
2166             ofproto->rstp_last_tick = time_msec();
2167         }
2168         rstp_set_bridge_address(ofproto->rstp, s->address);
2169         rstp_set_bridge_priority(ofproto->rstp, s->priority);
2170         rstp_set_bridge_ageing_time(ofproto->rstp, s->ageing_time);
2171         rstp_set_bridge_force_protocol_version(ofproto->rstp,
2172                                                s->force_protocol_version);
2173         rstp_set_bridge_max_age(ofproto->rstp, s->bridge_max_age);
2174         rstp_set_bridge_forward_delay(ofproto->rstp, s->bridge_forward_delay);
2175         rstp_set_bridge_transmit_hold_count(ofproto->rstp,
2176                                             s->transmit_hold_count);
2177     } else {
2178         struct ofport *ofport;
2179         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2180             set_rstp_port(ofport, NULL);
2181         }
2182         rstp_unref(ofproto->rstp);
2183         ofproto->rstp = NULL;
2184     }
2185 }
2186
2187 static void
2188 get_rstp_status(struct ofproto *ofproto_, struct ofproto_rstp_status *s)
2189 {
2190     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2191
2192     if (ofproto->rstp) {
2193         s->enabled = true;
2194         s->root_id = rstp_get_root_id(ofproto->rstp);
2195         s->bridge_id = rstp_get_bridge_id(ofproto->rstp);
2196         s->designated_id = rstp_get_designated_id(ofproto->rstp);
2197         s->root_path_cost = rstp_get_root_path_cost(ofproto->rstp);
2198         s->designated_port_id = rstp_get_designated_port_id(ofproto->rstp);
2199         s->bridge_port_id = rstp_get_bridge_port_id(ofproto->rstp);
2200     } else {
2201         s->enabled = false;
2202     }
2203 }
2204
2205 static void
2206 update_rstp_port_state(struct ofport_dpif *ofport)
2207 {
2208     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2209     enum rstp_state state;
2210
2211     /* Figure out new state. */
2212     state = ofport->rstp_port ? rstp_port_get_state(ofport->rstp_port)
2213         : RSTP_DISABLED;
2214
2215     /* Update state. */
2216     if (ofport->rstp_state != state) {
2217         enum ofputil_port_state of_state;
2218         bool fwd_change;
2219
2220         VLOG_DBG("port %s: RSTP state changed from %s to %s",
2221                  netdev_get_name(ofport->up.netdev),
2222                  rstp_state_name(ofport->rstp_state),
2223                  rstp_state_name(state));
2224
2225         if (rstp_learn_in_state(ofport->rstp_state)
2226             != rstp_learn_in_state(state)) {
2227             /* XXX: Learning action flows should also be flushed. */
2228             if (ofport->bundle) {
2229                 if (!rstp_shift_root_learned_address(ofproto->rstp)
2230                     || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2231                     bundle_flush_macs(ofport->bundle, false);
2232                 }
2233             }
2234         }
2235         fwd_change = rstp_forward_in_state(ofport->rstp_state)
2236             != rstp_forward_in_state(state);
2237
2238         ofproto->backer->need_revalidate = REV_RSTP;
2239         ofport->rstp_state = state;
2240
2241         if (fwd_change && ofport->bundle) {
2242             bundle_update(ofport->bundle);
2243         }
2244
2245         /* Update the RSTP state bits in the OpenFlow port description. */
2246         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2247         of_state |= (state == RSTP_LEARNING ? OFPUTIL_PS_STP_LEARN
2248                 : state == RSTP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2249                 : state == RSTP_DISCARDING ?  OFPUTIL_PS_STP_LISTEN
2250                 : 0);
2251         ofproto_port_set_state(&ofport->up, of_state);
2252     }
2253 }
2254
2255 static void
2256 rstp_run(struct ofproto_dpif *ofproto)
2257 {
2258     if (ofproto->rstp) {
2259         long long int now = time_msec();
2260         long long int elapsed = now - ofproto->rstp_last_tick;
2261         struct rstp_port *rp;
2262         struct ofport_dpif *ofport;
2263
2264         /* Every second, decrease the values of the timers. */
2265         if (elapsed >= 1000) {
2266             rstp_tick_timers(ofproto->rstp);
2267             ofproto->rstp_last_tick = now;
2268         }
2269         rp = NULL;
2270         while ((ofport = rstp_get_next_changed_port_aux(ofproto->rstp, &rp))) {
2271             update_rstp_port_state(ofport);
2272         }
2273         rp = NULL;
2274         ofport = NULL;
2275         /* FIXME: This check should be done on-event (i.e., when setting
2276          * p->fdb_flush) and not periodically.
2277          */
2278         while ((ofport = rstp_check_and_reset_fdb_flush(ofproto->rstp, &rp))) {
2279             if (!rstp_shift_root_learned_address(ofproto->rstp)
2280                 || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2281                 bundle_flush_macs(ofport->bundle, false);
2282             }
2283         }
2284
2285         if (rstp_shift_root_learned_address(ofproto->rstp)) {
2286             bundle_move(((struct ofport_dpif *)rstp_get_old_root_aux(ofproto->rstp))->bundle,
2287                         ((struct ofport_dpif *)rstp_get_new_root_aux(ofproto->rstp))->bundle);
2288             rstp_reset_root_changed(ofproto->rstp);
2289         }
2290     }
2291 }
2292
2293 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2294 static int
2295 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2296 {
2297     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2298
2299     /* Only revalidate flows if the configuration changed. */
2300     if (!s != !ofproto->stp) {
2301         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2302     }
2303
2304     if (s) {
2305         if (!ofproto->stp) {
2306             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2307                                       send_bpdu_cb, ofproto);
2308             ofproto->stp_last_tick = time_msec();
2309         }
2310
2311         stp_set_bridge_id(ofproto->stp, s->system_id);
2312         stp_set_bridge_priority(ofproto->stp, s->priority);
2313         stp_set_hello_time(ofproto->stp, s->hello_time);
2314         stp_set_max_age(ofproto->stp, s->max_age);
2315         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2316     }  else {
2317         struct ofport *ofport;
2318
2319         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2320             set_stp_port(ofport, NULL);
2321         }
2322
2323         stp_unref(ofproto->stp);
2324         ofproto->stp = NULL;
2325     }
2326
2327     return 0;
2328 }
2329
2330 static int
2331 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2332 {
2333     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2334
2335     if (ofproto->stp) {
2336         s->enabled = true;
2337         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2338         s->designated_root = stp_get_designated_root(ofproto->stp);
2339         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2340     } else {
2341         s->enabled = false;
2342     }
2343
2344     return 0;
2345 }
2346
2347 static void
2348 update_stp_port_state(struct ofport_dpif *ofport)
2349 {
2350     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2351     enum stp_state state;
2352
2353     /* Figure out new state. */
2354     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2355                              : STP_DISABLED;
2356
2357     /* Update state. */
2358     if (ofport->stp_state != state) {
2359         enum ofputil_port_state of_state;
2360         bool fwd_change;
2361
2362         VLOG_DBG("port %s: STP state changed from %s to %s",
2363                  netdev_get_name(ofport->up.netdev),
2364                  stp_state_name(ofport->stp_state),
2365                  stp_state_name(state));
2366         if (stp_learn_in_state(ofport->stp_state)
2367                 != stp_learn_in_state(state)) {
2368             /* xxx Learning action flows should also be flushed. */
2369             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2370             mac_learning_flush(ofproto->ml);
2371             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2372             mcast_snooping_mdb_flush(ofproto->ms);
2373         }
2374         fwd_change = stp_forward_in_state(ofport->stp_state)
2375                         != stp_forward_in_state(state);
2376
2377         ofproto->backer->need_revalidate = REV_STP;
2378         ofport->stp_state = state;
2379         ofport->stp_state_entered = time_msec();
2380
2381         if (fwd_change && ofport->bundle) {
2382             bundle_update(ofport->bundle);
2383         }
2384
2385         /* Update the STP state bits in the OpenFlow port description. */
2386         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2387         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2388                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2389                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2390                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2391                      : 0);
2392         ofproto_port_set_state(&ofport->up, of_state);
2393     }
2394 }
2395
2396 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2397  * caller is responsible for assigning STP port numbers and ensuring
2398  * there are no duplicates. */
2399 static int
2400 set_stp_port(struct ofport *ofport_,
2401              const struct ofproto_port_stp_settings *s)
2402 {
2403     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2404     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2405     struct stp_port *sp = ofport->stp_port;
2406
2407     if (!s || !s->enable) {
2408         if (sp) {
2409             ofport->stp_port = NULL;
2410             stp_port_disable(sp);
2411             update_stp_port_state(ofport);
2412         }
2413         return 0;
2414     } else if (sp && stp_port_no(sp) != s->port_num
2415                && ofport == stp_port_get_aux(sp)) {
2416         /* The port-id changed, so disable the old one if it's not
2417          * already in use by another port. */
2418         stp_port_disable(sp);
2419     }
2420
2421     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2422
2423     /* Set name before enabling the port so that debugging messages can print
2424      * the name. */
2425     stp_port_set_name(sp, netdev_get_name(ofport->up.netdev));
2426     stp_port_enable(sp);
2427
2428     stp_port_set_aux(sp, ofport);
2429     stp_port_set_priority(sp, s->priority);
2430     stp_port_set_path_cost(sp, s->path_cost);
2431
2432     update_stp_port_state(ofport);
2433
2434     return 0;
2435 }
2436
2437 static int
2438 get_stp_port_status(struct ofport *ofport_,
2439                     struct ofproto_port_stp_status *s)
2440 {
2441     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2442     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2443     struct stp_port *sp = ofport->stp_port;
2444
2445     if (!ofproto->stp || !sp) {
2446         s->enabled = false;
2447         return 0;
2448     }
2449
2450     s->enabled = true;
2451     s->port_id = stp_port_get_id(sp);
2452     s->state = stp_port_get_state(sp);
2453     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2454     s->role = stp_port_get_role(sp);
2455
2456     return 0;
2457 }
2458
2459 static int
2460 get_stp_port_stats(struct ofport *ofport_,
2461                    struct ofproto_port_stp_stats *s)
2462 {
2463     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2464     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2465     struct stp_port *sp = ofport->stp_port;
2466
2467     if (!ofproto->stp || !sp) {
2468         s->enabled = false;
2469         return 0;
2470     }
2471
2472     s->enabled = true;
2473     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2474
2475     return 0;
2476 }
2477
2478 static void
2479 stp_run(struct ofproto_dpif *ofproto)
2480 {
2481     if (ofproto->stp) {
2482         long long int now = time_msec();
2483         long long int elapsed = now - ofproto->stp_last_tick;
2484         struct stp_port *sp;
2485
2486         if (elapsed > 0) {
2487             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2488             ofproto->stp_last_tick = now;
2489         }
2490         while (stp_get_changed_port(ofproto->stp, &sp)) {
2491             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2492
2493             if (ofport) {
2494                 update_stp_port_state(ofport);
2495             }
2496         }
2497
2498         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2499             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2500             mac_learning_flush(ofproto->ml);
2501             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2502             mcast_snooping_mdb_flush(ofproto->ms);
2503         }
2504     }
2505 }
2506
2507 static void
2508 stp_wait(struct ofproto_dpif *ofproto)
2509 {
2510     if (ofproto->stp) {
2511         poll_timer_wait(1000);
2512     }
2513 }
2514
2515 /* Configures RSTP on 'ofport_' using the settings defined in 's'.  The
2516  * caller is responsible for assigning RSTP port numbers and ensuring
2517  * there are no duplicates. */
2518 static void
2519 set_rstp_port(struct ofport *ofport_,
2520               const struct ofproto_port_rstp_settings *s)
2521 {
2522     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2523     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2524     struct rstp_port *rp = ofport->rstp_port;
2525
2526     if (!s || !s->enable) {
2527         if (rp) {
2528             rstp_port_unref(rp);
2529             ofport->rstp_port = NULL;
2530             update_rstp_port_state(ofport);
2531         }
2532         return;
2533     }
2534
2535     /* Check if need to add a new port. */
2536     if (!rp) {
2537         rp = ofport->rstp_port = rstp_add_port(ofproto->rstp);
2538     }
2539
2540     rstp_port_set(rp, s->port_num, s->priority, s->path_cost,
2541                   s->admin_edge_port, s->auto_edge,
2542                   s->admin_p2p_mac_state, s->admin_port_state, s->mcheck,
2543                   ofport);
2544     update_rstp_port_state(ofport);
2545     /* Synchronize operational status. */
2546     rstp_port_set_mac_operational(rp, ofport->may_enable);
2547 }
2548
2549 static void
2550 get_rstp_port_status(struct ofport *ofport_,
2551         struct ofproto_port_rstp_status *s)
2552 {
2553     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2554     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2555     struct rstp_port *rp = ofport->rstp_port;
2556
2557     if (!ofproto->rstp || !rp) {
2558         s->enabled = false;
2559         return;
2560     }
2561
2562     s->enabled = true;
2563     rstp_port_get_status(rp, &s->port_id, &s->state, &s->role,
2564                          &s->designated_bridge_id, &s->designated_port_id,
2565                          &s->designated_path_cost, &s->tx_count,
2566                          &s->rx_count, &s->error_count, &s->uptime);
2567 }
2568
2569 \f
2570 static int
2571 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2572            size_t n_qdscp)
2573 {
2574     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2575     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2576
2577     if (ofport->n_qdscp != n_qdscp
2578         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2579                               n_qdscp * sizeof *qdscp))) {
2580         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2581         free(ofport->qdscp);
2582         ofport->qdscp = n_qdscp
2583             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2584             : NULL;
2585         ofport->n_qdscp = n_qdscp;
2586     }
2587
2588     return 0;
2589 }
2590 \f
2591 /* Bundles. */
2592
2593 /* Expires all MAC learning entries associated with 'bundle' and forces its
2594  * ofproto to revalidate every flow.
2595  *
2596  * Normally MAC learning entries are removed only from the ofproto associated
2597  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2598  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2599  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2600  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2601  * with the host from which it migrated. */
2602 static void
2603 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2604 {
2605     struct ofproto_dpif *ofproto = bundle->ofproto;
2606     struct mac_learning *ml = ofproto->ml;
2607     struct mac_entry *mac, *next_mac;
2608
2609     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2610     ovs_rwlock_wrlock(&ml->rwlock);
2611     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2612         if (mac_entry_get_port(ml, mac) == bundle) {
2613             if (all_ofprotos) {
2614                 struct ofproto_dpif *o;
2615
2616                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2617                     if (o != ofproto) {
2618                         struct mac_entry *e;
2619
2620                         ovs_rwlock_wrlock(&o->ml->rwlock);
2621                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2622                         if (e) {
2623                             mac_learning_expire(o->ml, e);
2624                         }
2625                         ovs_rwlock_unlock(&o->ml->rwlock);
2626                     }
2627                 }
2628             }
2629
2630             mac_learning_expire(ml, mac);
2631         }
2632     }
2633     ovs_rwlock_unlock(&ml->rwlock);
2634 }
2635
2636 static void
2637 bundle_move(struct ofbundle *old, struct ofbundle *new)
2638 {
2639     struct ofproto_dpif *ofproto = old->ofproto;
2640     struct mac_learning *ml = ofproto->ml;
2641     struct mac_entry *mac, *next_mac;
2642
2643     ovs_assert(new->ofproto == old->ofproto);
2644
2645     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2646     ovs_rwlock_wrlock(&ml->rwlock);
2647     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2648         if (mac_entry_get_port(ml, mac) == old) {
2649             mac_entry_set_port(ml, mac, new);
2650         }
2651     }
2652     ovs_rwlock_unlock(&ml->rwlock);
2653 }
2654
2655 static struct ofbundle *
2656 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2657 {
2658     struct ofbundle *bundle;
2659
2660     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2661                              &ofproto->bundles) {
2662         if (bundle->aux == aux) {
2663             return bundle;
2664         }
2665     }
2666     return NULL;
2667 }
2668
2669 static void
2670 bundle_update(struct ofbundle *bundle)
2671 {
2672     struct ofport_dpif *port;
2673
2674     bundle->floodable = true;
2675     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2676         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2677             || port->is_layer3
2678             || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2679             || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2680             bundle->floodable = false;
2681             break;
2682         }
2683     }
2684 }
2685
2686 static void
2687 bundle_del_port(struct ofport_dpif *port)
2688 {
2689     struct ofbundle *bundle = port->bundle;
2690
2691     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2692
2693     list_remove(&port->bundle_node);
2694     port->bundle = NULL;
2695
2696     if (bundle->lacp) {
2697         lacp_slave_unregister(bundle->lacp, port);
2698     }
2699     if (bundle->bond) {
2700         bond_slave_unregister(bundle->bond, port);
2701     }
2702
2703     bundle_update(bundle);
2704 }
2705
2706 static bool
2707 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2708                 struct lacp_slave_settings *lacp)
2709 {
2710     struct ofport_dpif *port;
2711
2712     port = ofp_port_to_ofport(bundle->ofproto, ofp_port);
2713     if (!port) {
2714         return false;
2715     }
2716
2717     if (port->bundle != bundle) {
2718         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2719         if (port->bundle) {
2720             bundle_remove(&port->up);
2721         }
2722
2723         port->bundle = bundle;
2724         list_push_back(&bundle->ports, &port->bundle_node);
2725         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2726             || port->is_layer3
2727             || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2728             || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2729             bundle->floodable = false;
2730         }
2731     }
2732     if (lacp) {
2733         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2734         lacp_slave_register(bundle->lacp, port, lacp);
2735     }
2736
2737     return true;
2738 }
2739
2740 static void
2741 bundle_destroy(struct ofbundle *bundle)
2742 {
2743     struct ofproto_dpif *ofproto;
2744     struct ofport_dpif *port, *next_port;
2745
2746     if (!bundle) {
2747         return;
2748     }
2749
2750     ofproto = bundle->ofproto;
2751     mbridge_unregister_bundle(ofproto->mbridge, bundle);
2752
2753     xlate_txn_start();
2754     xlate_bundle_remove(bundle);
2755     xlate_txn_commit();
2756
2757     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2758         bundle_del_port(port);
2759     }
2760
2761     bundle_flush_macs(bundle, true);
2762     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2763     free(bundle->name);
2764     free(bundle->trunks);
2765     lacp_unref(bundle->lacp);
2766     bond_unref(bundle->bond);
2767     free(bundle);
2768 }
2769
2770 static int
2771 bundle_set(struct ofproto *ofproto_, void *aux,
2772            const struct ofproto_bundle_settings *s)
2773 {
2774     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2775     bool need_flush = false;
2776     struct ofport_dpif *port;
2777     struct ofbundle *bundle;
2778     unsigned long *trunks;
2779     int vlan;
2780     size_t i;
2781     bool ok;
2782
2783     if (!s) {
2784         bundle_destroy(bundle_lookup(ofproto, aux));
2785         return 0;
2786     }
2787
2788     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2789     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2790
2791     bundle = bundle_lookup(ofproto, aux);
2792     if (!bundle) {
2793         bundle = xmalloc(sizeof *bundle);
2794
2795         bundle->ofproto = ofproto;
2796         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2797                     hash_pointer(aux, 0));
2798         bundle->aux = aux;
2799         bundle->name = NULL;
2800
2801         list_init(&bundle->ports);
2802         bundle->vlan_mode = PORT_VLAN_TRUNK;
2803         bundle->vlan = -1;
2804         bundle->trunks = NULL;
2805         bundle->use_priority_tags = s->use_priority_tags;
2806         bundle->lacp = NULL;
2807         bundle->bond = NULL;
2808
2809         bundle->floodable = true;
2810         mbridge_register_bundle(ofproto->mbridge, bundle);
2811     }
2812
2813     if (!bundle->name || strcmp(s->name, bundle->name)) {
2814         free(bundle->name);
2815         bundle->name = xstrdup(s->name);
2816     }
2817
2818     /* LACP. */
2819     if (s->lacp) {
2820         ofproto->lacp_enabled = true;
2821         if (!bundle->lacp) {
2822             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2823             bundle->lacp = lacp_create();
2824         }
2825         lacp_configure(bundle->lacp, s->lacp);
2826     } else {
2827         lacp_unref(bundle->lacp);
2828         bundle->lacp = NULL;
2829     }
2830
2831     /* Update set of ports. */
2832     ok = true;
2833     for (i = 0; i < s->n_slaves; i++) {
2834         if (!bundle_add_port(bundle, s->slaves[i],
2835                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2836             ok = false;
2837         }
2838     }
2839     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2840         struct ofport_dpif *next_port;
2841
2842         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2843             for (i = 0; i < s->n_slaves; i++) {
2844                 if (s->slaves[i] == port->up.ofp_port) {
2845                     goto found;
2846                 }
2847             }
2848
2849             bundle_del_port(port);
2850         found: ;
2851         }
2852     }
2853     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2854
2855     if (list_is_empty(&bundle->ports)) {
2856         bundle_destroy(bundle);
2857         return EINVAL;
2858     }
2859
2860     /* Set VLAN tagging mode */
2861     if (s->vlan_mode != bundle->vlan_mode
2862         || s->use_priority_tags != bundle->use_priority_tags) {
2863         bundle->vlan_mode = s->vlan_mode;
2864         bundle->use_priority_tags = s->use_priority_tags;
2865         need_flush = true;
2866     }
2867
2868     /* Set VLAN tag. */
2869     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2870             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2871             : 0);
2872     if (vlan != bundle->vlan) {
2873         bundle->vlan = vlan;
2874         need_flush = true;
2875     }
2876
2877     /* Get trunked VLANs. */
2878     switch (s->vlan_mode) {
2879     case PORT_VLAN_ACCESS:
2880         trunks = NULL;
2881         break;
2882
2883     case PORT_VLAN_TRUNK:
2884         trunks = CONST_CAST(unsigned long *, s->trunks);
2885         break;
2886
2887     case PORT_VLAN_NATIVE_UNTAGGED:
2888     case PORT_VLAN_NATIVE_TAGGED:
2889         if (vlan != 0 && (!s->trunks
2890                           || !bitmap_is_set(s->trunks, vlan)
2891                           || bitmap_is_set(s->trunks, 0))) {
2892             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2893             if (s->trunks) {
2894                 trunks = bitmap_clone(s->trunks, 4096);
2895             } else {
2896                 trunks = bitmap_allocate1(4096);
2897             }
2898             bitmap_set1(trunks, vlan);
2899             bitmap_set0(trunks, 0);
2900         } else {
2901             trunks = CONST_CAST(unsigned long *, s->trunks);
2902         }
2903         break;
2904
2905     default:
2906         OVS_NOT_REACHED();
2907     }
2908     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2909         free(bundle->trunks);
2910         if (trunks == s->trunks) {
2911             bundle->trunks = vlan_bitmap_clone(trunks);
2912         } else {
2913             bundle->trunks = trunks;
2914             trunks = NULL;
2915         }
2916         need_flush = true;
2917     }
2918     if (trunks != s->trunks) {
2919         free(trunks);
2920     }
2921
2922     /* Bonding. */
2923     if (!list_is_short(&bundle->ports)) {
2924         bundle->ofproto->has_bonded_bundles = true;
2925         if (bundle->bond) {
2926             if (bond_reconfigure(bundle->bond, s->bond)) {
2927                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2928             }
2929         } else {
2930             bundle->bond = bond_create(s->bond, ofproto);
2931             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2932         }
2933
2934         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2935             bond_slave_register(bundle->bond, port,
2936                                 port->up.ofp_port, port->up.netdev);
2937         }
2938     } else {
2939         bond_unref(bundle->bond);
2940         bundle->bond = NULL;
2941     }
2942
2943     /* If we changed something that would affect MAC learning, un-learn
2944      * everything on this port and force flow revalidation. */
2945     if (need_flush) {
2946         bundle_flush_macs(bundle, false);
2947     }
2948
2949     return 0;
2950 }
2951
2952 static void
2953 bundle_remove(struct ofport *port_)
2954 {
2955     struct ofport_dpif *port = ofport_dpif_cast(port_);
2956     struct ofbundle *bundle = port->bundle;
2957
2958     if (bundle) {
2959         bundle_del_port(port);
2960         if (list_is_empty(&bundle->ports)) {
2961             bundle_destroy(bundle);
2962         } else if (list_is_short(&bundle->ports)) {
2963             bond_unref(bundle->bond);
2964             bundle->bond = NULL;
2965         }
2966     }
2967 }
2968
2969 static void
2970 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2971 {
2972     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2973     struct ofport_dpif *port = port_;
2974     uint8_t ea[ETH_ADDR_LEN];
2975     int error;
2976
2977     error = netdev_get_etheraddr(port->up.netdev, ea);
2978     if (!error) {
2979         struct dp_packet packet;
2980         void *packet_pdu;
2981
2982         dp_packet_init(&packet, 0);
2983         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2984                                  pdu_size);
2985         memcpy(packet_pdu, pdu, pdu_size);
2986
2987         ofproto_dpif_send_packet(port, &packet);
2988         dp_packet_uninit(&packet);
2989     } else {
2990         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2991                     "%s (%s)", port->bundle->name,
2992                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2993     }
2994 }
2995
2996 static void
2997 bundle_send_learning_packets(struct ofbundle *bundle)
2998 {
2999     struct ofproto_dpif *ofproto = bundle->ofproto;
3000     int error, n_packets, n_errors;
3001     struct mac_entry *e;
3002     struct pkt_list {
3003         struct ovs_list list_node;
3004         struct ofport_dpif *port;
3005         struct dp_packet *pkt;
3006     } *pkt_node;
3007     struct ovs_list packets;
3008
3009     list_init(&packets);
3010     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3011     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3012         if (mac_entry_get_port(ofproto->ml, e) != bundle) {
3013             pkt_node = xmalloc(sizeof *pkt_node);
3014             pkt_node->pkt = bond_compose_learning_packet(bundle->bond,
3015                                                          e->mac, e->vlan,
3016                                                          (void **)&pkt_node->port);
3017             list_push_back(&packets, &pkt_node->list_node);
3018         }
3019     }
3020     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3021
3022     error = n_packets = n_errors = 0;
3023     LIST_FOR_EACH_POP (pkt_node, list_node, &packets) {
3024         int ret;
3025
3026         ret = ofproto_dpif_send_packet(pkt_node->port, pkt_node->pkt);
3027         dp_packet_delete(pkt_node->pkt);
3028         free(pkt_node);
3029         if (ret) {
3030             error = ret;
3031             n_errors++;
3032         }
3033         n_packets++;
3034     }
3035
3036     if (n_errors) {
3037         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3038         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3039                      "packets, last error was: %s",
3040                      bundle->name, n_errors, n_packets, ovs_strerror(error));
3041     } else {
3042         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3043                  bundle->name, n_packets);
3044     }
3045 }
3046
3047 static void
3048 bundle_run(struct ofbundle *bundle)
3049 {
3050     if (bundle->lacp) {
3051         lacp_run(bundle->lacp, send_pdu_cb);
3052     }
3053     if (bundle->bond) {
3054         struct ofport_dpif *port;
3055
3056         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
3057             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
3058         }
3059
3060         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
3061             bundle->ofproto->backer->need_revalidate = REV_BOND;
3062         }
3063
3064         if (bond_should_send_learning_packets(bundle->bond)) {
3065             bundle_send_learning_packets(bundle);
3066         }
3067     }
3068 }
3069
3070 static void
3071 bundle_wait(struct ofbundle *bundle)
3072 {
3073     if (bundle->lacp) {
3074         lacp_wait(bundle->lacp);
3075     }
3076     if (bundle->bond) {
3077         bond_wait(bundle->bond);
3078     }
3079 }
3080 \f
3081 /* Mirrors. */
3082
3083 static int
3084 mirror_set__(struct ofproto *ofproto_, void *aux,
3085              const struct ofproto_mirror_settings *s)
3086 {
3087     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3088     struct ofbundle **srcs, **dsts;
3089     int error;
3090     size_t i;
3091
3092     if (!s) {
3093         mirror_destroy(ofproto->mbridge, aux);
3094         return 0;
3095     }
3096
3097     srcs = xmalloc(s->n_srcs * sizeof *srcs);
3098     dsts = xmalloc(s->n_dsts * sizeof *dsts);
3099
3100     for (i = 0; i < s->n_srcs; i++) {
3101         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
3102     }
3103
3104     for (i = 0; i < s->n_dsts; i++) {
3105         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
3106     }
3107
3108     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
3109                        s->n_dsts, s->src_vlans,
3110                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
3111     free(srcs);
3112     free(dsts);
3113     return error;
3114 }
3115
3116 static int
3117 mirror_get_stats__(struct ofproto *ofproto, void *aux,
3118                    uint64_t *packets, uint64_t *bytes)
3119 {
3120     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
3121                             bytes);
3122 }
3123
3124 static int
3125 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3126 {
3127     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3128     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3129     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3130         mac_learning_flush(ofproto->ml);
3131     }
3132     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3133     return 0;
3134 }
3135
3136 static bool
3137 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3138 {
3139     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3140     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3141     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
3142 }
3143
3144 static void
3145 forward_bpdu_changed(struct ofproto *ofproto_)
3146 {
3147     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3148     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3149 }
3150
3151 static void
3152 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3153                      size_t max_entries)
3154 {
3155     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3156     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3157     mac_learning_set_idle_time(ofproto->ml, idle_time);
3158     mac_learning_set_max_entries(ofproto->ml, max_entries);
3159     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3160 }
3161
3162 /* Configures multicast snooping on 'ofport' using the settings
3163  * defined in 's'. */
3164 static int
3165 set_mcast_snooping(struct ofproto *ofproto_,
3166                    const struct ofproto_mcast_snooping_settings *s)
3167 {
3168     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3169
3170     /* Only revalidate flows if the configuration changed. */
3171     if (!s != !ofproto->ms) {
3172         ofproto->backer->need_revalidate = REV_RECONFIGURE;
3173     }
3174
3175     if (s) {
3176         if (!ofproto->ms) {
3177             ofproto->ms = mcast_snooping_create();
3178         }
3179
3180         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3181         mcast_snooping_set_idle_time(ofproto->ms, s->idle_time);
3182         mcast_snooping_set_max_entries(ofproto->ms, s->max_entries);
3183         if (mcast_snooping_set_flood_unreg(ofproto->ms, s->flood_unreg)) {
3184             ofproto->backer->need_revalidate = REV_RECONFIGURE;
3185         }
3186         ovs_rwlock_unlock(&ofproto->ms->rwlock);
3187     } else {
3188         mcast_snooping_unref(ofproto->ms);
3189         ofproto->ms = NULL;
3190     }
3191
3192     return 0;
3193 }
3194
3195 /* Configures multicast snooping port's flood settings on 'ofproto'. */
3196 static int
3197 set_mcast_snooping_port(struct ofproto *ofproto_, void *aux,
3198                         const struct ofproto_mcast_snooping_port_settings *s)
3199 {
3200     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3201     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3202
3203     if (ofproto->ms && s) {
3204         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3205         mcast_snooping_set_port_flood(ofproto->ms, bundle, s->flood);
3206         mcast_snooping_set_port_flood_reports(ofproto->ms, bundle,
3207                                               s->flood_reports);
3208         ovs_rwlock_unlock(&ofproto->ms->rwlock);
3209     }
3210     return 0;
3211 }
3212
3213 \f
3214 /* Ports. */
3215
3216 struct ofport_dpif *
3217 ofp_port_to_ofport(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
3218 {
3219     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3220     return ofport ? ofport_dpif_cast(ofport) : NULL;
3221 }
3222
3223 static void
3224 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3225                             struct ofproto_port *ofproto_port,
3226                             struct dpif_port *dpif_port)
3227 {
3228     ofproto_port->name = dpif_port->name;
3229     ofproto_port->type = dpif_port->type;
3230     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3231 }
3232
3233 static void
3234 ofport_update_peer(struct ofport_dpif *ofport)
3235 {
3236     const struct ofproto_dpif *ofproto;
3237     struct dpif_backer *backer;
3238     char *peer_name;
3239
3240     if (!netdev_vport_is_patch(ofport->up.netdev)) {
3241         return;
3242     }
3243
3244     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
3245     backer->need_revalidate = REV_RECONFIGURE;
3246
3247     if (ofport->peer) {
3248         ofport->peer->peer = NULL;
3249         ofport->peer = NULL;
3250     }
3251
3252     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
3253     if (!peer_name) {
3254         return;
3255     }
3256
3257     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3258         struct ofport *peer_ofport;
3259         struct ofport_dpif *peer;
3260         char *peer_peer;
3261
3262         if (ofproto->backer != backer) {
3263             continue;
3264         }
3265
3266         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
3267         if (!peer_ofport) {
3268             continue;
3269         }
3270
3271         peer = ofport_dpif_cast(peer_ofport);
3272         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
3273         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
3274                                  peer_peer)) {
3275             ofport->peer = peer;
3276             ofport->peer->peer = ofport;
3277         }
3278         free(peer_peer);
3279
3280         break;
3281     }
3282     free(peer_name);
3283 }
3284
3285 static void
3286 port_run(struct ofport_dpif *ofport)
3287 {
3288     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3289     bool carrier_changed = carrier_seq != ofport->carrier_seq;
3290     bool enable = netdev_get_carrier(ofport->up.netdev);
3291     bool cfm_enable = false;
3292     bool bfd_enable = false;
3293
3294     ofport->carrier_seq = carrier_seq;
3295
3296     if (ofport->cfm) {
3297         int cfm_opup = cfm_get_opup(ofport->cfm);
3298
3299         cfm_enable = !cfm_get_fault(ofport->cfm);
3300
3301         if (cfm_opup >= 0) {
3302             cfm_enable = cfm_enable && cfm_opup;
3303         }
3304     }
3305
3306     if (ofport->bfd) {
3307         bfd_enable = bfd_forwarding(ofport->bfd);
3308     }
3309
3310     if (ofport->bfd || ofport->cfm) {
3311         enable = enable && (cfm_enable || bfd_enable);
3312     }
3313
3314     if (ofport->bundle) {
3315         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3316         if (carrier_changed) {
3317             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3318         }
3319     }
3320
3321     if (ofport->may_enable != enable) {
3322         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3323
3324         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3325
3326         if (ofport->rstp_port) {
3327             rstp_port_set_mac_operational(ofport->rstp_port, enable);
3328         }
3329     }
3330
3331     ofport->may_enable = enable;
3332 }
3333
3334 static int
3335 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3336                    struct ofproto_port *ofproto_port)
3337 {
3338     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3339     struct dpif_port dpif_port;
3340     int error;
3341
3342     if (sset_contains(&ofproto->ghost_ports, devname)) {
3343         const char *type = netdev_get_type_from_name(devname);
3344
3345         /* We may be called before ofproto->up.port_by_name is populated with
3346          * the appropriate ofport.  For this reason, we must get the name and
3347          * type from the netdev layer directly. */
3348         if (type) {
3349             const struct ofport *ofport;
3350
3351             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3352             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3353             ofproto_port->name = xstrdup(devname);
3354             ofproto_port->type = xstrdup(type);
3355             return 0;
3356         }
3357         return ENODEV;
3358     }
3359
3360     if (!sset_contains(&ofproto->ports, devname)) {
3361         return ENODEV;
3362     }
3363     error = dpif_port_query_by_name(ofproto->backer->dpif,
3364                                     devname, &dpif_port);
3365     if (!error) {
3366         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3367     }
3368     return error;
3369 }
3370
3371 static int
3372 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3373 {
3374     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3375     const char *devname = netdev_get_name(netdev);
3376     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3377     const char *dp_port_name;
3378
3379     if (netdev_vport_is_patch(netdev)) {
3380         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3381         return 0;
3382     }
3383
3384     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3385     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3386         odp_port_t port_no = ODPP_NONE;
3387         int error;
3388
3389         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3390         if (error) {
3391             return error;
3392         }
3393         if (netdev_get_tunnel_config(netdev)) {
3394             simap_put(&ofproto->backer->tnl_backers,
3395                       dp_port_name, odp_to_u32(port_no));
3396         }
3397     }
3398
3399     if (netdev_get_tunnel_config(netdev)) {
3400         sset_add(&ofproto->ghost_ports, devname);
3401     } else {
3402         sset_add(&ofproto->ports, devname);
3403     }
3404     return 0;
3405 }
3406
3407 static int
3408 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3409 {
3410     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3411     struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
3412     int error = 0;
3413
3414     if (!ofport) {
3415         return 0;
3416     }
3417
3418     sset_find_and_delete(&ofproto->ghost_ports,
3419                          netdev_get_name(ofport->up.netdev));
3420     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3421     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3422         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3423         if (!error) {
3424             /* The caller is going to close ofport->up.netdev.  If this is a
3425              * bonded port, then the bond is using that netdev, so remove it
3426              * from the bond.  The client will need to reconfigure everything
3427              * after deleting ports, so then the slave will get re-added. */
3428             bundle_remove(&ofport->up);
3429         }
3430     }
3431     return error;
3432 }
3433
3434 static int
3435 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3436 {
3437     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3438     int error;
3439
3440     error = netdev_get_stats(ofport->up.netdev, stats);
3441
3442     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3443         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3444
3445         ovs_mutex_lock(&ofproto->stats_mutex);
3446         /* ofproto->stats.tx_packets represents packets that we created
3447          * internally and sent to some port (e.g. packets sent with
3448          * ofproto_dpif_send_packet()).  Account for them as if they had
3449          * come from OFPP_LOCAL and got forwarded. */
3450
3451         if (stats->rx_packets != UINT64_MAX) {
3452             stats->rx_packets += ofproto->stats.tx_packets;
3453         }
3454
3455         if (stats->rx_bytes != UINT64_MAX) {
3456             stats->rx_bytes += ofproto->stats.tx_bytes;
3457         }
3458
3459         /* ofproto->stats.rx_packets represents packets that were received on
3460          * some port and we processed internally and dropped (e.g. STP).
3461          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3462
3463         if (stats->tx_packets != UINT64_MAX) {
3464             stats->tx_packets += ofproto->stats.rx_packets;
3465         }
3466
3467         if (stats->tx_bytes != UINT64_MAX) {
3468             stats->tx_bytes += ofproto->stats.rx_bytes;
3469         }
3470         ovs_mutex_unlock(&ofproto->stats_mutex);
3471     }
3472
3473     return error;
3474 }
3475
3476 static int
3477 port_get_lacp_stats(const struct ofport *ofport_, struct lacp_slave_stats *stats)
3478 {
3479     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3480     if (ofport->bundle && ofport->bundle->lacp) {
3481         if (lacp_get_slave_stats(ofport->bundle->lacp, ofport, stats)) {
3482             return 0;
3483         }
3484     }
3485     return -1;
3486 }
3487
3488 struct port_dump_state {
3489     uint32_t bucket;
3490     uint32_t offset;
3491     bool ghost;
3492
3493     struct ofproto_port port;
3494     bool has_port;
3495 };
3496
3497 static int
3498 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3499 {
3500     *statep = xzalloc(sizeof(struct port_dump_state));
3501     return 0;
3502 }
3503
3504 static int
3505 port_dump_next(const struct ofproto *ofproto_, void *state_,
3506                struct ofproto_port *port)
3507 {
3508     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3509     struct port_dump_state *state = state_;
3510     const struct sset *sset;
3511     struct sset_node *node;
3512
3513     if (state->has_port) {
3514         ofproto_port_destroy(&state->port);
3515         state->has_port = false;
3516     }
3517     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3518     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3519         int error;
3520
3521         error = port_query_by_name(ofproto_, node->name, &state->port);
3522         if (!error) {
3523             *port = state->port;
3524             state->has_port = true;
3525             return 0;
3526         } else if (error != ENODEV) {
3527             return error;
3528         }
3529     }
3530
3531     if (!state->ghost) {
3532         state->ghost = true;
3533         state->bucket = 0;
3534         state->offset = 0;
3535         return port_dump_next(ofproto_, state_, port);
3536     }
3537
3538     return EOF;
3539 }
3540
3541 static int
3542 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3543 {
3544     struct port_dump_state *state = state_;
3545
3546     if (state->has_port) {
3547         ofproto_port_destroy(&state->port);
3548     }
3549     free(state);
3550     return 0;
3551 }
3552
3553 static int
3554 port_poll(const struct ofproto *ofproto_, char **devnamep)
3555 {
3556     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3557
3558     if (ofproto->port_poll_errno) {
3559         int error = ofproto->port_poll_errno;
3560         ofproto->port_poll_errno = 0;
3561         return error;
3562     }
3563
3564     if (sset_is_empty(&ofproto->port_poll_set)) {
3565         return EAGAIN;
3566     }
3567
3568     *devnamep = sset_pop(&ofproto->port_poll_set);
3569     return 0;
3570 }
3571
3572 static void
3573 port_poll_wait(const struct ofproto *ofproto_)
3574 {
3575     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3576     dpif_port_poll_wait(ofproto->backer->dpif);
3577 }
3578
3579 static int
3580 port_is_lacp_current(const struct ofport *ofport_)
3581 {
3582     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3583     return (ofport->bundle && ofport->bundle->lacp
3584             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3585             : -1);
3586 }
3587 \f
3588 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3589  * then delete it entirely. */
3590 static void
3591 rule_expire(struct rule_dpif *rule)
3592     OVS_REQUIRES(ofproto_mutex)
3593 {
3594     uint16_t hard_timeout, idle_timeout;
3595     long long int now = time_msec();
3596     int reason = -1;
3597
3598     hard_timeout = rule->up.hard_timeout;
3599     idle_timeout = rule->up.idle_timeout;
3600
3601     /* Has 'rule' expired? */
3602     if (hard_timeout) {
3603         long long int modified;
3604
3605         ovs_mutex_lock(&rule->up.mutex);
3606         modified = rule->up.modified;
3607         ovs_mutex_unlock(&rule->up.mutex);
3608
3609         if (now > modified + hard_timeout * 1000) {
3610             reason = OFPRR_HARD_TIMEOUT;
3611         }
3612     }
3613
3614     if (reason < 0 && idle_timeout) {
3615         long long int used;
3616
3617         ovs_mutex_lock(&rule->stats_mutex);
3618         used = rule->stats.used;
3619         ovs_mutex_unlock(&rule->stats_mutex);
3620
3621         if (now > used + idle_timeout * 1000) {
3622             reason = OFPRR_IDLE_TIMEOUT;
3623         }
3624     }
3625
3626     if (reason >= 0) {
3627         COVERAGE_INC(ofproto_dpif_expired);
3628         ofproto_rule_expire(&rule->up, reason);
3629     }
3630 }
3631
3632 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3633  * 'flow' must reflect the data in 'packet'. */
3634 int
3635 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3636                              const struct flow *flow,
3637                              struct rule_dpif *rule,
3638                              const struct ofpact *ofpacts, size_t ofpacts_len,
3639                              struct dp_packet *packet)
3640 {
3641     struct dpif_flow_stats stats;
3642     struct xlate_out xout;
3643     struct xlate_in xin;
3644     ofp_port_t in_port;
3645     struct dpif_execute execute;
3646     int error;
3647
3648     ovs_assert((rule != NULL) != (ofpacts != NULL));
3649
3650     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3651
3652     if (rule) {
3653         rule_dpif_credit_stats(rule, &stats);
3654     }
3655
3656     xlate_in_init(&xin, ofproto, flow, flow->in_port.ofp_port, rule,
3657                   stats.tcp_flags, packet);
3658     xin.ofpacts = ofpacts;
3659     xin.ofpacts_len = ofpacts_len;
3660     xin.resubmit_stats = &stats;
3661     xlate_actions(&xin, &xout);
3662
3663     execute.actions = xout.odp_actions->data;
3664     execute.actions_len = xout.odp_actions->size;
3665
3666     pkt_metadata_from_flow(&packet->md, flow);
3667     execute.packet = packet;
3668     execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3669     execute.probe = false;
3670
3671     /* Fix up in_port. */
3672     in_port = flow->in_port.ofp_port;
3673     if (in_port == OFPP_NONE) {
3674         in_port = OFPP_LOCAL;
3675     }
3676     execute.packet->md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3677
3678     error = dpif_execute(ofproto->backer->dpif, &execute);
3679
3680     xlate_out_uninit(&xout);
3681
3682     return error;
3683 }
3684
3685 void
3686 rule_dpif_credit_stats(struct rule_dpif *rule,
3687                        const struct dpif_flow_stats *stats)
3688 {
3689     ovs_mutex_lock(&rule->stats_mutex);
3690     if (OVS_UNLIKELY(rule->new_rule)) {
3691         rule_dpif_credit_stats(rule->new_rule, stats);
3692     } else {
3693         rule->stats.n_packets += stats->n_packets;
3694         rule->stats.n_bytes += stats->n_bytes;
3695         rule->stats.used = MAX(rule->stats.used, stats->used);
3696     }
3697     ovs_mutex_unlock(&rule->stats_mutex);
3698 }
3699
3700 ovs_be64
3701 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3702     OVS_REQUIRES(rule->up.mutex)
3703 {
3704     return rule->up.flow_cookie;
3705 }
3706
3707 void
3708 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3709                      uint16_t hard_timeout)
3710 {
3711     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3712 }
3713
3714 /* Returns 'rule''s actions.  The returned actions are RCU-protected, and can
3715  * be read until the calling thread quiesces. */
3716 const struct rule_actions *
3717 rule_dpif_get_actions(const struct rule_dpif *rule)
3718 {
3719     return rule_get_actions(&rule->up);
3720 }
3721
3722 /* Sets 'rule''s recirculation id. */
3723 static void
3724 rule_dpif_set_recirc_id(struct rule_dpif *rule, uint32_t id)
3725     OVS_REQUIRES(rule->up.mutex)
3726 {
3727     ovs_assert(!rule->recirc_id || rule->recirc_id == id);
3728     if (rule->recirc_id == id) {
3729         /* Release the new reference to the same id. */
3730         recirc_free_id(id);
3731     } else {
3732         rule->recirc_id = id;
3733     }
3734 }
3735
3736 /* Sets 'rule''s recirculation id. */
3737 void
3738 rule_set_recirc_id(struct rule *rule_, uint32_t id)
3739 {
3740     struct rule_dpif *rule = rule_dpif_cast(rule_);
3741
3742     ovs_mutex_lock(&rule->up.mutex);
3743     rule_dpif_set_recirc_id(rule, id);
3744     ovs_mutex_unlock(&rule->up.mutex);
3745 }
3746
3747 cls_version_t
3748 ofproto_dpif_get_tables_version(struct ofproto_dpif *ofproto OVS_UNUSED)
3749 {
3750     cls_version_t version;
3751
3752     atomic_read_relaxed(&ofproto->tables_version, &version);
3753
3754     return version;
3755 }
3756
3757 /* The returned rule (if any) is valid at least until the next RCU quiescent
3758  * period.  If the rule needs to stay around longer, a non-zero 'take_ref'
3759  * must be passed in to cause a reference to be taken on it.
3760  *
3761  * 'flow' is non-const to allow for temporary modifications during the lookup.
3762  * Any changes are restored before returning. */
3763 static struct rule_dpif *
3764 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, cls_version_t version,
3765                           uint8_t table_id, struct flow *flow,
3766                           struct flow_wildcards *wc, bool take_ref)
3767 {
3768     struct classifier *cls = &ofproto->up.tables[table_id].cls;
3769     const struct cls_rule *cls_rule;
3770     struct rule_dpif *rule;
3771
3772     do {
3773         cls_rule = classifier_lookup(cls, version, flow, wc);
3774
3775         rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3776
3777         /* Try again if the rule was released before we get the reference. */
3778     } while (rule && take_ref && !rule_dpif_try_ref(rule));
3779
3780     return rule;
3781 }
3782
3783 /* Look up 'flow' in 'ofproto''s classifier version 'version', starting from
3784  * table '*table_id'.  Returns the rule that was found, which may be one of the
3785  * special rules according to packet miss hadling.  If 'may_packet_in' is
3786  * false, returning of the miss_rule (which issues packet ins for the
3787  * controller) is avoided.  Updates 'wc', if nonnull, to reflect the fields
3788  * that were used during the lookup.
3789  *
3790  * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3791  * if none is found then the table miss configuration for that table is
3792  * honored, which can result in additional lookups in other OpenFlow tables.
3793  * In this case the function updates '*table_id' to reflect the final OpenFlow
3794  * table that was searched.
3795  *
3796  * If 'honor_table_miss' is false, then only one table lookup occurs, in
3797  * '*table_id'.
3798  *
3799  * The rule is returned in '*rule', which is valid at least until the next
3800  * RCU quiescent period.  If the '*rule' needs to stay around longer,
3801  * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3802  * on it before this returns.
3803  *
3804  * 'in_port' allows the lookup to take place as if the in port had the value
3805  * 'in_port'.  This is needed for resubmit action support.
3806  *
3807  * 'flow' is non-const to allow for temporary modifications during the lookup.
3808  * Any changes are restored before returning. */
3809 struct rule_dpif *
3810 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3811                             cls_version_t version, struct flow *flow,
3812                             struct flow_wildcards *wc, bool take_ref,
3813                             const struct dpif_flow_stats *stats,
3814                             uint8_t *table_id, ofp_port_t in_port,
3815                             bool may_packet_in, bool honor_table_miss)
3816 {
3817     ovs_be16 old_tp_src = flow->tp_src, old_tp_dst = flow->tp_dst;
3818     ofp_port_t old_in_port = flow->in_port.ofp_port;
3819     enum ofputil_table_miss miss_config;
3820     struct rule_dpif *rule;
3821     uint8_t next_id;
3822
3823     /* We always unwildcard nw_frag (for IP), so they
3824      * need not be unwildcarded here. */
3825     if (flow->nw_frag & FLOW_NW_FRAG_ANY
3826         && ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3827         if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3828             /* We must pretend that transport ports are unavailable. */
3829             flow->tp_src = htons(0);
3830             flow->tp_dst = htons(0);
3831         } else {
3832             /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM).
3833              * Use the drop_frags_rule (which cannot disappear). */
3834             rule = ofproto->drop_frags_rule;
3835             if (take_ref) {
3836                 rule_dpif_ref(rule);
3837             }
3838             if (stats) {
3839                 struct oftable *tbl = &ofproto->up.tables[*table_id];
3840                 unsigned long orig;
3841
3842                 atomic_add_relaxed(&tbl->n_matched, stats->n_packets, &orig);
3843             }
3844             return rule;
3845         }
3846     }
3847
3848     /* Look up a flow with 'in_port' as the input port.  Then restore the
3849      * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
3850      * have surprising behavior). */
3851     flow->in_port.ofp_port = in_port;
3852
3853     /* Our current implementation depends on n_tables == N_TABLES, and
3854      * TBL_INTERNAL being the last table. */
3855     BUILD_ASSERT_DECL(N_TABLES == TBL_INTERNAL + 1);
3856
3857     miss_config = OFPUTIL_TABLE_MISS_CONTINUE;
3858
3859     for (next_id = *table_id;
3860          next_id < ofproto->up.n_tables;
3861          next_id++, next_id += (next_id == TBL_INTERNAL))
3862     {
3863         *table_id = next_id;
3864         rule = rule_dpif_lookup_in_table(ofproto, version, next_id, flow, wc,
3865                                          take_ref);
3866         if (stats) {
3867             struct oftable *tbl = &ofproto->up.tables[next_id];
3868             unsigned long orig;
3869
3870             atomic_add_relaxed(rule ? &tbl->n_matched : &tbl->n_missed,
3871                                stats->n_packets, &orig);
3872         }
3873         if (rule) {
3874             goto out;   /* Match. */
3875         }
3876         if (honor_table_miss) {
3877             miss_config = ofproto_table_get_miss_config(&ofproto->up,
3878                                                         *table_id);
3879             if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE) {
3880                 continue;
3881             }
3882         }
3883         break;
3884     }
3885     /* Miss. */
3886     rule = ofproto->no_packet_in_rule;
3887     if (may_packet_in) {
3888         if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE
3889             || miss_config == OFPUTIL_TABLE_MISS_CONTROLLER) {
3890             struct ofport_dpif *port;
3891
3892             port = ofp_port_to_ofport(ofproto, old_in_port);
3893             if (!port) {
3894                 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3895                              old_in_port);
3896             } else if (!(port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN)) {
3897                 rule = ofproto->miss_rule;
3898             }
3899         } else if (miss_config == OFPUTIL_TABLE_MISS_DEFAULT &&
3900                    connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3901             rule = ofproto->miss_rule;
3902         }
3903     }
3904     if (take_ref) {
3905         rule_dpif_ref(rule);
3906     }
3907 out:
3908     /* Restore port numbers, as they may have been modified above. */
3909     flow->tp_src = old_tp_src;
3910     flow->tp_dst = old_tp_dst;
3911     /* Restore the old in port. */
3912     flow->in_port.ofp_port = old_in_port;
3913
3914     return rule;
3915 }
3916
3917 static void
3918 complete_operation(struct rule_dpif *rule)
3919     OVS_REQUIRES(ofproto_mutex)
3920 {
3921     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3922
3923     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3924 }
3925
3926 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3927 {
3928     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3929 }
3930
3931 static struct rule *
3932 rule_alloc(void)
3933 {
3934     struct rule_dpif *rule = xzalloc(sizeof *rule);
3935     return &rule->up;
3936 }
3937
3938 static void
3939 rule_dealloc(struct rule *rule_)
3940 {
3941     struct rule_dpif *rule = rule_dpif_cast(rule_);
3942     free(rule);
3943 }
3944
3945 static enum ofperr
3946 rule_construct(struct rule *rule_)
3947     OVS_NO_THREAD_SAFETY_ANALYSIS
3948 {
3949     struct rule_dpif *rule = rule_dpif_cast(rule_);
3950     ovs_mutex_init_adaptive(&rule->stats_mutex);
3951     rule->stats.n_packets = 0;
3952     rule->stats.n_bytes = 0;
3953     rule->stats.used = rule->up.modified;
3954     rule->recirc_id = 0;
3955     rule->new_rule = NULL;
3956
3957     return 0;
3958 }
3959
3960 static void
3961 rule_insert(struct rule *rule_, struct rule *old_rule_, bool forward_stats)
3962     OVS_REQUIRES(ofproto_mutex)
3963 {
3964     struct rule_dpif *rule = rule_dpif_cast(rule_);
3965
3966     if (old_rule_ && forward_stats) {
3967         struct rule_dpif *old_rule = rule_dpif_cast(old_rule_);
3968
3969         ovs_assert(!old_rule->new_rule);
3970
3971         /* Take a reference to the new rule, and refer all stats updates from
3972          * the old rule to the new rule. */
3973         rule_dpif_ref(rule);
3974
3975         ovs_mutex_lock(&old_rule->stats_mutex);
3976         ovs_mutex_lock(&rule->stats_mutex);
3977         old_rule->new_rule = rule;       /* Forward future stats. */
3978         rule->stats = old_rule->stats;   /* Transfer stats to the new rule. */
3979         ovs_mutex_unlock(&rule->stats_mutex);
3980         ovs_mutex_unlock(&old_rule->stats_mutex);
3981     }
3982
3983     complete_operation(rule);
3984 }
3985
3986 static void
3987 rule_delete(struct rule *rule_)
3988     OVS_REQUIRES(ofproto_mutex)
3989 {
3990     struct rule_dpif *rule = rule_dpif_cast(rule_);
3991     complete_operation(rule);
3992 }
3993
3994 static void
3995 rule_destruct(struct rule *rule_)
3996     OVS_NO_THREAD_SAFETY_ANALYSIS
3997 {
3998     struct rule_dpif *rule = rule_dpif_cast(rule_);
3999
4000     ovs_mutex_destroy(&rule->stats_mutex);
4001     /* Release reference to the new rule, if any. */
4002     if (rule->new_rule) {
4003         rule_dpif_unref(rule->new_rule);
4004     }
4005     if (rule->recirc_id) {
4006         recirc_free_id(rule->recirc_id);
4007     }
4008 }
4009
4010 static void
4011 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
4012                long long int *used)
4013 {
4014     struct rule_dpif *rule = rule_dpif_cast(rule_);
4015
4016     ovs_mutex_lock(&rule->stats_mutex);
4017     if (OVS_UNLIKELY(rule->new_rule)) {
4018         rule_get_stats(&rule->new_rule->up, packets, bytes, used);
4019     } else {
4020         *packets = rule->stats.n_packets;
4021         *bytes = rule->stats.n_bytes;
4022         *used = rule->stats.used;
4023     }
4024     ovs_mutex_unlock(&rule->stats_mutex);
4025 }
4026
4027 static void
4028 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4029                   struct dp_packet *packet)
4030 {
4031     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4032
4033     ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
4034 }
4035
4036 static enum ofperr
4037 rule_execute(struct rule *rule, const struct flow *flow,
4038              struct dp_packet *packet)
4039 {
4040     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4041     dp_packet_delete(packet);
4042     return 0;
4043 }
4044
4045 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4046 {
4047     return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4048 }
4049
4050 static struct ofgroup *
4051 group_alloc(void)
4052 {
4053     struct group_dpif *group = xzalloc(sizeof *group);
4054     return &group->up;
4055 }
4056
4057 static void
4058 group_dealloc(struct ofgroup *group_)
4059 {
4060     struct group_dpif *group = group_dpif_cast(group_);
4061     free(group);
4062 }
4063
4064 static void
4065 group_construct_stats(struct group_dpif *group)
4066     OVS_REQUIRES(group->stats_mutex)
4067 {
4068     struct ofputil_bucket *bucket;
4069     const struct ovs_list *buckets;
4070
4071     group->packet_count = 0;
4072     group->byte_count = 0;
4073
4074     group_dpif_get_buckets(group, &buckets);
4075     LIST_FOR_EACH (bucket, list_node, buckets) {
4076         bucket->stats.packet_count = 0;
4077         bucket->stats.byte_count = 0;
4078     }
4079 }
4080
4081 void
4082 group_dpif_credit_stats(struct group_dpif *group,
4083                         struct ofputil_bucket *bucket,
4084                         const struct dpif_flow_stats *stats)
4085 {
4086     ovs_mutex_lock(&group->stats_mutex);
4087     group->packet_count += stats->n_packets;
4088     group->byte_count += stats->n_bytes;
4089     if (bucket) {
4090         bucket->stats.packet_count += stats->n_packets;
4091         bucket->stats.byte_count += stats->n_bytes;
4092     } else { /* Credit to all buckets */
4093         const struct ovs_list *buckets;
4094
4095         group_dpif_get_buckets(group, &buckets);
4096         LIST_FOR_EACH (bucket, list_node, buckets) {
4097             bucket->stats.packet_count += stats->n_packets;
4098             bucket->stats.byte_count += stats->n_bytes;
4099         }
4100     }
4101     ovs_mutex_unlock(&group->stats_mutex);
4102 }
4103
4104 static enum ofperr
4105 group_construct(struct ofgroup *group_)
4106 {
4107     struct group_dpif *group = group_dpif_cast(group_);
4108
4109     ovs_mutex_init_adaptive(&group->stats_mutex);
4110     ovs_mutex_lock(&group->stats_mutex);
4111     group_construct_stats(group);
4112     ovs_mutex_unlock(&group->stats_mutex);
4113     return 0;
4114 }
4115
4116 static void
4117 group_destruct(struct ofgroup *group_)
4118 {
4119     struct group_dpif *group = group_dpif_cast(group_);
4120     ovs_mutex_destroy(&group->stats_mutex);
4121 }
4122
4123 static enum ofperr
4124 group_modify(struct ofgroup *group_)
4125 {
4126     struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
4127
4128     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4129
4130     return 0;
4131 }
4132
4133 static enum ofperr
4134 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4135 {
4136     struct group_dpif *group = group_dpif_cast(group_);
4137     struct ofputil_bucket *bucket;
4138     const struct ovs_list *buckets;
4139     struct bucket_counter *bucket_stats;
4140
4141     ovs_mutex_lock(&group->stats_mutex);
4142     ogs->packet_count = group->packet_count;
4143     ogs->byte_count = group->byte_count;
4144
4145     group_dpif_get_buckets(group, &buckets);
4146     bucket_stats = ogs->bucket_stats;
4147     LIST_FOR_EACH (bucket, list_node, buckets) {
4148         bucket_stats->packet_count = bucket->stats.packet_count;
4149         bucket_stats->byte_count = bucket->stats.byte_count;
4150         bucket_stats++;
4151     }
4152     ovs_mutex_unlock(&group->stats_mutex);
4153
4154     return 0;
4155 }
4156
4157 /* If the group exists, this function increments the groups's reference count.
4158  *
4159  * Make sure to call group_dpif_unref() after no longer needing to maintain
4160  * a reference to the group. */
4161 bool
4162 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4163                   struct group_dpif **group)
4164 {
4165     struct ofgroup *ofgroup;
4166     bool found;
4167
4168     found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4169     *group = found ?  group_dpif_cast(ofgroup) : NULL;
4170
4171     return found;
4172 }
4173
4174 void
4175 group_dpif_get_buckets(const struct group_dpif *group,
4176                        const struct ovs_list **buckets)
4177 {
4178     *buckets = &group->up.buckets;
4179 }
4180
4181 enum ofp11_group_type
4182 group_dpif_get_type(const struct group_dpif *group)
4183 {
4184     return group->up.type;
4185 }
4186
4187 const char *
4188 group_dpif_get_selection_method(const struct group_dpif *group)
4189 {
4190     return group->up.props.selection_method;
4191 }
4192 \f
4193 /* Sends 'packet' out 'ofport'.
4194  * May modify 'packet'.
4195  * Returns 0 if successful, otherwise a positive errno value. */
4196 int
4197 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct dp_packet *packet)
4198 {
4199     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4200     int error;
4201
4202     error = xlate_send_packet(ofport, packet);
4203
4204     ovs_mutex_lock(&ofproto->stats_mutex);
4205     ofproto->stats.tx_packets++;
4206     ofproto->stats.tx_bytes += dp_packet_size(packet);
4207     ovs_mutex_unlock(&ofproto->stats_mutex);
4208     return error;
4209 }
4210
4211 uint64_t
4212 group_dpif_get_selection_method_param(const struct group_dpif *group)
4213 {
4214     return group->up.props.selection_method_param;
4215 }
4216
4217 const struct field_array *
4218 group_dpif_get_fields(const struct group_dpif *group)
4219 {
4220     return &group->up.props.fields;
4221 }
4222 \f
4223 /* Return the version string of the datapath that backs up
4224  * this 'ofproto'.
4225  */
4226 static const char *
4227 get_datapath_version(const struct ofproto *ofproto_)
4228 {
4229     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4230
4231     return ofproto->backer->dp_version_string;
4232 }
4233
4234 static bool
4235 set_frag_handling(struct ofproto *ofproto_,
4236                   enum ofp_config_flags frag_handling)
4237 {
4238     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4239     if (frag_handling != OFPC_FRAG_REASM) {
4240         ofproto->backer->need_revalidate = REV_RECONFIGURE;
4241         return true;
4242     } else {
4243         return false;
4244     }
4245 }
4246
4247 static enum ofperr
4248 packet_out(struct ofproto *ofproto_, struct dp_packet *packet,
4249            const struct flow *flow,
4250            const struct ofpact *ofpacts, size_t ofpacts_len)
4251 {
4252     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4253
4254     ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
4255                                  ofpacts_len, packet);
4256     return 0;
4257 }
4258 \f
4259 /* NetFlow. */
4260
4261 static int
4262 set_netflow(struct ofproto *ofproto_,
4263             const struct netflow_options *netflow_options)
4264 {
4265     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4266
4267     if (netflow_options) {
4268         if (!ofproto->netflow) {
4269             ofproto->netflow = netflow_create();
4270             ofproto->backer->need_revalidate = REV_RECONFIGURE;
4271         }
4272         return netflow_set_options(ofproto->netflow, netflow_options);
4273     } else if (ofproto->netflow) {
4274         ofproto->backer->need_revalidate = REV_RECONFIGURE;
4275         netflow_unref(ofproto->netflow);
4276         ofproto->netflow = NULL;
4277     }
4278
4279     return 0;
4280 }
4281
4282 static void
4283 get_netflow_ids(const struct ofproto *ofproto_,
4284                 uint8_t *engine_type, uint8_t *engine_id)
4285 {
4286     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4287
4288     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
4289 }
4290 \f
4291 static struct ofproto_dpif *
4292 ofproto_dpif_lookup(const char *name)
4293 {
4294     struct ofproto_dpif *ofproto;
4295
4296     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
4297                              hash_string(name, 0), &all_ofproto_dpifs) {
4298         if (!strcmp(ofproto->up.name, name)) {
4299             return ofproto;
4300         }
4301     }
4302     return NULL;
4303 }
4304
4305 static void
4306 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
4307                           const char *argv[], void *aux OVS_UNUSED)
4308 {
4309     struct ofproto_dpif *ofproto;
4310
4311     if (argc > 1) {
4312         ofproto = ofproto_dpif_lookup(argv[1]);
4313         if (!ofproto) {
4314             unixctl_command_reply_error(conn, "no such bridge");
4315             return;
4316         }
4317         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4318         mac_learning_flush(ofproto->ml);
4319         ovs_rwlock_unlock(&ofproto->ml->rwlock);
4320     } else {
4321         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4322             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4323             mac_learning_flush(ofproto->ml);
4324             ovs_rwlock_unlock(&ofproto->ml->rwlock);
4325         }
4326     }
4327
4328     unixctl_command_reply(conn, "table successfully flushed");
4329 }
4330
4331 static void
4332 ofproto_unixctl_mcast_snooping_flush(struct unixctl_conn *conn, int argc,
4333                                      const char *argv[], void *aux OVS_UNUSED)
4334 {
4335     struct ofproto_dpif *ofproto;
4336
4337     if (argc > 1) {
4338         ofproto = ofproto_dpif_lookup(argv[1]);
4339         if (!ofproto) {
4340             unixctl_command_reply_error(conn, "no such bridge");
4341             return;
4342         }
4343
4344         if (!mcast_snooping_enabled(ofproto->ms)) {
4345             unixctl_command_reply_error(conn, "multicast snooping is disabled");
4346             return;
4347         }
4348         mcast_snooping_mdb_flush(ofproto->ms);
4349     } else {
4350         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4351             if (!mcast_snooping_enabled(ofproto->ms)) {
4352                 continue;
4353             }
4354             mcast_snooping_mdb_flush(ofproto->ms);
4355         }
4356     }
4357
4358     unixctl_command_reply(conn, "table successfully flushed");
4359 }
4360
4361 static struct ofport_dpif *
4362 ofbundle_get_a_port(const struct ofbundle *bundle)
4363 {
4364     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
4365                         bundle_node);
4366 }
4367
4368 static void
4369 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4370                          const char *argv[], void *aux OVS_UNUSED)
4371 {
4372     struct ds ds = DS_EMPTY_INITIALIZER;
4373     const struct ofproto_dpif *ofproto;
4374     const struct mac_entry *e;
4375
4376     ofproto = ofproto_dpif_lookup(argv[1]);
4377     if (!ofproto) {
4378         unixctl_command_reply_error(conn, "no such bridge");
4379         return;
4380     }
4381
4382     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
4383     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
4384     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
4385         struct ofbundle *bundle = mac_entry_get_port(ofproto->ml, e);
4386         char name[OFP_MAX_PORT_NAME_LEN];
4387
4388         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4389                                name, sizeof name);
4390         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
4391                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
4392                       mac_entry_age(ofproto->ml, e));
4393     }
4394     ovs_rwlock_unlock(&ofproto->ml->rwlock);
4395     unixctl_command_reply(conn, ds_cstr(&ds));
4396     ds_destroy(&ds);
4397 }
4398
4399 static void
4400 ofproto_unixctl_mcast_snooping_show(struct unixctl_conn *conn,
4401                                     int argc OVS_UNUSED,
4402                                     const char *argv[],
4403                                     void *aux OVS_UNUSED)
4404 {
4405     struct ds ds = DS_EMPTY_INITIALIZER;
4406     const struct ofproto_dpif *ofproto;
4407     const struct ofbundle *bundle;
4408     const struct mcast_group *grp;
4409     struct mcast_group_bundle *b;
4410     struct mcast_mrouter_bundle *mrouter;
4411
4412     ofproto = ofproto_dpif_lookup(argv[1]);
4413     if (!ofproto) {
4414         unixctl_command_reply_error(conn, "no such bridge");
4415         return;
4416     }
4417
4418     if (!mcast_snooping_enabled(ofproto->ms)) {
4419         unixctl_command_reply_error(conn, "multicast snooping is disabled");
4420         return;
4421     }
4422
4423     ds_put_cstr(&ds, " port  VLAN  GROUP                Age\n");
4424     ovs_rwlock_rdlock(&ofproto->ms->rwlock);
4425     LIST_FOR_EACH (grp, group_node, &ofproto->ms->group_lru) {
4426         LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
4427             char name[OFP_MAX_PORT_NAME_LEN];
4428
4429             bundle = b->port;
4430             ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4431                                    name, sizeof name);
4432             ds_put_format(&ds, "%5s  %4d  "IP_FMT"         %3d\n",
4433                           name, grp->vlan, IP_ARGS(grp->ip4),
4434                           mcast_bundle_age(ofproto->ms, b));
4435         }
4436     }
4437
4438     /* ports connected to multicast routers */
4439     LIST_FOR_EACH(mrouter, mrouter_node, &ofproto->ms->mrouter_lru) {
4440         char name[OFP_MAX_PORT_NAME_LEN];
4441
4442         bundle = mrouter->port;
4443         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4444                                name, sizeof name);
4445             ds_put_format(&ds, "%5s  %4d  querier             %3d\n",
4446                       name, mrouter->vlan,
4447                       mcast_mrouter_age(ofproto->ms, mrouter));
4448     }
4449     ovs_rwlock_unlock(&ofproto->ms->rwlock);
4450     unixctl_command_reply(conn, ds_cstr(&ds));
4451     ds_destroy(&ds);
4452 }
4453
4454 struct trace_ctx {
4455     struct xlate_out xout;
4456     struct xlate_in xin;
4457     const struct flow *key;
4458     struct flow flow;
4459     struct flow_wildcards wc;
4460     struct ds *result;
4461 };
4462
4463 static void
4464 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
4465 {
4466     const struct rule_actions *actions;
4467     ovs_be64 cookie;
4468
4469     ds_put_char_multiple(result, '\t', level);
4470     if (!rule) {
4471         ds_put_cstr(result, "No match\n");
4472         return;
4473     }
4474
4475     ovs_mutex_lock(&rule->up.mutex);
4476     cookie = rule->up.flow_cookie;
4477     ovs_mutex_unlock(&rule->up.mutex);
4478
4479     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
4480                   rule ? rule->up.table_id : 0, ntohll(cookie));
4481     cls_rule_format(&rule->up.cr, result);
4482     ds_put_char(result, '\n');
4483
4484     actions = rule_dpif_get_actions(rule);
4485
4486     ds_put_char_multiple(result, '\t', level);
4487     ds_put_cstr(result, "OpenFlow actions=");
4488     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
4489     ds_put_char(result, '\n');
4490 }
4491
4492 static void
4493 trace_format_flow(struct ds *result, int level, const char *title,
4494                   struct trace_ctx *trace)
4495 {
4496     ds_put_char_multiple(result, '\t', level);
4497     ds_put_format(result, "%s: ", title);
4498     /* Do not report unchanged flows for resubmits. */
4499     if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
4500         || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
4501         ds_put_cstr(result, "unchanged");
4502     } else {
4503         flow_format(result, &trace->xin.flow);
4504         trace->flow = trace->xin.flow;
4505     }
4506     ds_put_char(result, '\n');
4507 }
4508
4509 static void
4510 trace_format_regs(struct ds *result, int level, const char *title,
4511                   struct trace_ctx *trace)
4512 {
4513     size_t i;
4514
4515     ds_put_char_multiple(result, '\t', level);
4516     ds_put_format(result, "%s:", title);
4517     for (i = 0; i < FLOW_N_REGS; i++) {
4518         ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
4519     }
4520     ds_put_char(result, '\n');
4521 }
4522
4523 static void
4524 trace_format_odp(struct ds *result, int level, const char *title,
4525                  struct trace_ctx *trace)
4526 {
4527     struct ofpbuf *odp_actions = trace->xout.odp_actions;
4528
4529     ds_put_char_multiple(result, '\t', level);
4530     ds_put_format(result, "%s: ", title);
4531     format_odp_actions(result, odp_actions->data, odp_actions->size);
4532     ds_put_char(result, '\n');
4533 }
4534
4535 static void
4536 trace_format_megaflow(struct ds *result, int level, const char *title,
4537                       struct trace_ctx *trace)
4538 {
4539     struct match match;
4540
4541     ds_put_char_multiple(result, '\t', level);
4542     ds_put_format(result, "%s: ", title);
4543     flow_wildcards_or(&trace->wc, &trace->xout.wc, &trace->wc);
4544     match_init(&match, trace->key, &trace->wc);
4545     match_format(&match, result, OFP_DEFAULT_PRIORITY);
4546     ds_put_char(result, '\n');
4547 }
4548
4549 static void trace_report(struct xlate_in *, int recurse,
4550                          const char *format, ...)
4551     OVS_PRINTF_FORMAT(3, 4);
4552 static void trace_report_valist(struct xlate_in *, int recurse,
4553                                 const char *format, va_list args)
4554     OVS_PRINTF_FORMAT(3, 0);
4555
4556 static void
4557 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
4558 {
4559     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4560     struct ds *result = trace->result;
4561
4562     if (!recurse) {
4563         if (rule == xin->ofproto->miss_rule) {
4564             trace_report(xin, recurse,
4565                          "No match, flow generates \"packet in\"s.");
4566         } else if (rule == xin->ofproto->no_packet_in_rule) {
4567             trace_report(xin, recurse, "No match, packets dropped because "
4568                          "OFPPC_NO_PACKET_IN is set on in_port.");
4569         } else if (rule == xin->ofproto->drop_frags_rule) {
4570             trace_report(xin, recurse, "Packets dropped because they are IP "
4571                          "fragments and the fragment handling mode is "
4572                          "\"drop\".");
4573         }
4574     }
4575
4576     ds_put_char(result, '\n');
4577     if (recurse) {
4578         trace_format_flow(result, recurse, "Resubmitted flow", trace);
4579         trace_format_regs(result, recurse, "Resubmitted regs", trace);
4580         trace_format_odp(result,  recurse, "Resubmitted  odp", trace);
4581         trace_format_megaflow(result, recurse, "Resubmitted megaflow", trace);
4582     }
4583     trace_format_rule(result, recurse, rule);
4584 }
4585
4586 static void
4587 trace_report_valist(struct xlate_in *xin, int recurse,
4588                     const char *format, va_list args)
4589 {
4590     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4591     struct ds *result = trace->result;
4592
4593     ds_put_char_multiple(result, '\t', recurse);
4594     ds_put_format_valist(result, format, args);
4595     ds_put_char(result, '\n');
4596 }
4597
4598 static void
4599 trace_report(struct xlate_in *xin, int recurse, const char *format, ...)
4600 {
4601     va_list args;
4602
4603     va_start(args, format);
4604     trace_report_valist(xin, recurse, format, args);
4605     va_end(args);
4606 }
4607
4608 /* Parses the 'argc' elements of 'argv', ignoring argv[0].  The following
4609  * forms are supported:
4610  *
4611  *     - [dpname] odp_flow [-generate | packet]
4612  *     - bridge br_flow [-generate | packet]
4613  *
4614  * On success, initializes '*ofprotop' and 'flow' and returns NULL.  On failure
4615  * returns a nonnull malloced error message. */
4616 static char * OVS_WARN_UNUSED_RESULT
4617 parse_flow_and_packet(int argc, const char *argv[],
4618                       struct ofproto_dpif **ofprotop, struct flow *flow,
4619                       struct dp_packet **packetp)
4620 {
4621     const struct dpif_backer *backer = NULL;
4622     const char *error = NULL;
4623     char *m_err = NULL;
4624     struct simap port_names = SIMAP_INITIALIZER(&port_names);
4625     struct dp_packet *packet;
4626     struct ofpbuf odp_key;
4627     struct ofpbuf odp_mask;
4628
4629     ofpbuf_init(&odp_key, 0);
4630     ofpbuf_init(&odp_mask, 0);
4631
4632     /* Handle "-generate" or a hex string as the last argument. */
4633     if (!strcmp(argv[argc - 1], "-generate")) {
4634         packet = dp_packet_new(0);
4635         argc--;
4636     } else {
4637         error = eth_from_hex(argv[argc - 1], &packet);
4638         if (!error) {
4639             argc--;
4640         } else if (argc == 4) {
4641             /* The 3-argument form must end in "-generate' or a hex string. */
4642             goto exit;
4643         }
4644         error = NULL;
4645     }
4646
4647     /* odp_flow can have its in_port specified as a name instead of port no.
4648      * We do not yet know whether a given flow is a odp_flow or a br_flow.
4649      * But, to know whether a flow is odp_flow through odp_flow_from_string(),
4650      * we need to create a simap of name to port no. */
4651     if (argc == 3) {
4652         const char *dp_type;
4653         if (!strncmp(argv[1], "ovs-", 4)) {
4654             dp_type = argv[1] + 4;
4655         } else {
4656             dp_type = argv[1];
4657         }
4658         backer = shash_find_data(&all_dpif_backers, dp_type);
4659     } else if (argc == 2) {
4660         struct shash_node *node;
4661         if (shash_count(&all_dpif_backers) == 1) {
4662             node = shash_first(&all_dpif_backers);
4663             backer = node->data;
4664         }
4665     } else {
4666         error = "Syntax error";
4667         goto exit;
4668     }
4669     if (backer && backer->dpif) {
4670         struct dpif_port dpif_port;
4671         struct dpif_port_dump port_dump;
4672         DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4673             simap_put(&port_names, dpif_port.name,
4674                       odp_to_u32(dpif_port.port_no));
4675         }
4676     }
4677
4678     /* Parse the flow and determine whether a datapath or
4679      * bridge is specified. If function odp_flow_key_from_string()
4680      * returns 0, the flow is a odp_flow. If function
4681      * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4682     if (!odp_flow_from_string(argv[argc - 1], &port_names,
4683                               &odp_key, &odp_mask)) {
4684         if (!backer) {
4685             error = "Cannot find the datapath";
4686             goto exit;
4687         }
4688
4689         if (odp_flow_key_to_flow(odp_key.data, odp_key.size, flow) == ODP_FIT_ERROR) {
4690             error = "Failed to parse datapath flow key";
4691             goto exit;
4692         }
4693
4694         *ofprotop = xlate_lookup_ofproto(backer, flow,
4695                                          &flow->in_port.ofp_port);
4696         if (*ofprotop == NULL) {
4697             error = "Invalid datapath flow";
4698             goto exit;
4699         }
4700
4701         vsp_adjust_flow(*ofprotop, flow, NULL);
4702
4703     } else {
4704         char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4705
4706         if (err) {
4707             m_err = xasprintf("Bad openflow flow syntax: %s", err);
4708             free(err);
4709             goto exit;
4710         } else {
4711             if (argc != 3) {
4712                 error = "Must specify bridge name";
4713                 goto exit;
4714             }
4715
4716             *ofprotop = ofproto_dpif_lookup(argv[1]);
4717             if (!*ofprotop) {
4718                 error = "Unknown bridge name";
4719                 goto exit;
4720             }
4721         }
4722     }
4723
4724     /* Generate a packet, if requested. */
4725     if (packet) {
4726         if (!dp_packet_size(packet)) {
4727             flow_compose(packet, flow);
4728         } else {
4729             /* Use the metadata from the flow and the packet argument
4730              * to reconstruct the flow. */
4731             pkt_metadata_from_flow(&packet->md, flow);
4732             flow_extract(packet, flow);
4733         }
4734     }
4735
4736 exit:
4737     if (error && !m_err) {
4738         m_err = xstrdup(error);
4739     }
4740     if (m_err) {
4741         dp_packet_delete(packet);
4742         packet = NULL;
4743     }
4744     *packetp = packet;
4745     ofpbuf_uninit(&odp_key);
4746     ofpbuf_uninit(&odp_mask);
4747     simap_destroy(&port_names);
4748     return m_err;
4749 }
4750
4751 static void
4752 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4753                       void *aux OVS_UNUSED)
4754 {
4755     struct ofproto_dpif *ofproto;
4756     struct dp_packet *packet;
4757     char *error;
4758     struct flow flow;
4759
4760     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4761     if (!error) {
4762         struct ds result;
4763
4764         ds_init(&result);
4765         ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4766         unixctl_command_reply(conn, ds_cstr(&result));
4767         ds_destroy(&result);
4768         dp_packet_delete(packet);
4769     } else {
4770         unixctl_command_reply_error(conn, error);
4771         free(error);
4772     }
4773 }
4774
4775 static void
4776 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4777                               const char *argv[], void *aux OVS_UNUSED)
4778 {
4779     enum ofputil_protocol usable_protocols;
4780     struct ofproto_dpif *ofproto;
4781     bool enforce_consistency;
4782     struct ofpbuf ofpacts;
4783     struct dp_packet *packet;
4784     struct ds result;
4785     struct flow flow;
4786     uint16_t in_port;
4787
4788     /* Three kinds of error return values! */
4789     enum ofperr retval;
4790     char *error;
4791
4792     packet = NULL;
4793     ds_init(&result);
4794     ofpbuf_init(&ofpacts, 0);
4795
4796     /* Parse actions. */
4797     error = ofpacts_parse_actions(argv[--argc], &ofpacts, &usable_protocols);
4798     if (error) {
4799         unixctl_command_reply_error(conn, error);
4800         free(error);
4801         goto exit;
4802     }
4803
4804     /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4805      * consistency between the flow and the actions.  With -consistent, we
4806      * enforce consistency even for a flow supported in OpenFlow 1.0. */
4807     if (!strcmp(argv[1], "-consistent")) {
4808         enforce_consistency = true;
4809         argv++;
4810         argc--;
4811     } else {
4812         enforce_consistency = false;
4813     }
4814
4815     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4816     if (error) {
4817         unixctl_command_reply_error(conn, error);
4818         free(error);
4819         goto exit;
4820     }
4821
4822     /* Do the same checks as handle_packet_out() in ofproto.c.
4823      *
4824      * We pass a 'table_id' of 0 to ofpacts_check(), which isn't
4825      * strictly correct because these actions aren't in any table, but it's OK
4826      * because it 'table_id' is used only to check goto_table instructions, but
4827      * packet-outs take a list of actions and therefore it can't include
4828      * instructions.
4829      *
4830      * We skip the "meter" check here because meter is an instruction, not an
4831      * action, and thus cannot appear in ofpacts. */
4832     in_port = ofp_to_u16(flow.in_port.ofp_port);
4833     if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
4834         unixctl_command_reply_error(conn, "invalid in_port");
4835         goto exit;
4836     }
4837     if (enforce_consistency) {
4838         retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size,
4839                                            &flow, u16_to_ofp(ofproto->up.max_ports),
4840                                            0, 0, usable_protocols);
4841     } else {
4842         retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow,
4843                                u16_to_ofp(ofproto->up.max_ports), 0, 0,
4844                                &usable_protocols);
4845     }
4846
4847     if (retval) {
4848         ds_clear(&result);
4849         ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
4850         unixctl_command_reply_error(conn, ds_cstr(&result));
4851         goto exit;
4852     }
4853
4854     ofproto_trace(ofproto, &flow, packet,
4855                   ofpacts.data, ofpacts.size, &result);
4856     unixctl_command_reply(conn, ds_cstr(&result));
4857
4858 exit:
4859     ds_destroy(&result);
4860     dp_packet_delete(packet);
4861     ofpbuf_uninit(&ofpacts);
4862 }
4863
4864 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4865  * description of the results to 'ds'.
4866  *
4867  * The trace follows a packet with the specified 'flow' through the flow
4868  * table.  'packet' may be nonnull to trace an actual packet, with consequent
4869  * side effects (if it is nonnull then its flow must be 'flow').
4870  *
4871  * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4872  * trace, otherwise the actions are determined by a flow table lookup. */
4873 static void
4874 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
4875               const struct dp_packet *packet,
4876               const struct ofpact ofpacts[], size_t ofpacts_len,
4877               struct ds *ds)
4878 {
4879     struct trace_ctx trace;
4880
4881     ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
4882     ds_put_cstr(ds, "Flow: ");
4883     flow_format(ds, flow);
4884     ds_put_char(ds, '\n');
4885
4886     flow_wildcards_init_catchall(&trace.wc);
4887
4888     trace.result = ds;
4889     trace.key = flow; /* Original flow key, used for megaflow. */
4890     trace.flow = *flow; /* May be modified by actions. */
4891     xlate_in_init(&trace.xin, ofproto, flow, flow->in_port.ofp_port, NULL,
4892                   ntohs(flow->tcp_flags), packet);
4893     trace.xin.ofpacts = ofpacts;
4894     trace.xin.ofpacts_len = ofpacts_len;
4895     trace.xin.resubmit_hook = trace_resubmit;
4896     trace.xin.report_hook = trace_report_valist;
4897
4898     xlate_actions(&trace.xin, &trace.xout);
4899
4900     ds_put_char(ds, '\n');
4901     trace_format_flow(ds, 0, "Final flow", &trace);
4902     trace_format_megaflow(ds, 0, "Megaflow", &trace);
4903
4904     ds_put_cstr(ds, "Datapath actions: ");
4905     format_odp_actions(ds, trace.xout.odp_actions->data,
4906                        trace.xout.odp_actions->size);
4907
4908     if (trace.xout.slow) {
4909         enum slow_path_reason slow;
4910
4911         ds_put_cstr(ds, "\nThis flow is handled by the userspace "
4912                     "slow path because it:");
4913
4914         slow = trace.xout.slow;
4915         while (slow) {
4916             enum slow_path_reason bit = rightmost_1bit(slow);
4917
4918             ds_put_format(ds, "\n\t- %s.",
4919                           slow_path_reason_to_explanation(bit));
4920
4921             slow &= ~bit;
4922         }
4923     }
4924
4925     xlate_out_uninit(&trace.xout);
4926 }
4927
4928 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
4929  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
4930  * to destroy 'ofproto_shash' and free the returned value. */
4931 static const struct shash_node **
4932 get_ofprotos(struct shash *ofproto_shash)
4933 {
4934     const struct ofproto_dpif *ofproto;
4935
4936     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4937         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
4938         shash_add_nocopy(ofproto_shash, name, ofproto);
4939     }
4940
4941     return shash_sort(ofproto_shash);
4942 }
4943
4944 static void
4945 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
4946                               const char *argv[] OVS_UNUSED,
4947                               void *aux OVS_UNUSED)
4948 {
4949     struct ds ds = DS_EMPTY_INITIALIZER;
4950     struct shash ofproto_shash;
4951     const struct shash_node **sorted_ofprotos;
4952     int i;
4953
4954     shash_init(&ofproto_shash);
4955     sorted_ofprotos = get_ofprotos(&ofproto_shash);
4956     for (i = 0; i < shash_count(&ofproto_shash); i++) {
4957         const struct shash_node *node = sorted_ofprotos[i];
4958         ds_put_format(&ds, "%s\n", node->name);
4959     }
4960
4961     shash_destroy(&ofproto_shash);
4962     free(sorted_ofprotos);
4963
4964     unixctl_command_reply(conn, ds_cstr(&ds));
4965     ds_destroy(&ds);
4966 }
4967
4968 static void
4969 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
4970 {
4971     const struct shash_node **ofprotos;
4972     struct dpif_dp_stats dp_stats;
4973     struct shash ofproto_shash;
4974     size_t i;
4975
4976     dpif_get_dp_stats(backer->dpif, &dp_stats);
4977
4978     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
4979                   dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
4980
4981     shash_init(&ofproto_shash);
4982     ofprotos = get_ofprotos(&ofproto_shash);
4983     for (i = 0; i < shash_count(&ofproto_shash); i++) {
4984         struct ofproto_dpif *ofproto = ofprotos[i]->data;
4985         const struct shash_node **ports;
4986         size_t j;
4987
4988         if (ofproto->backer != backer) {
4989             continue;
4990         }
4991
4992         ds_put_format(ds, "\t%s:\n", ofproto->up.name);
4993
4994         ports = shash_sort(&ofproto->up.port_by_name);
4995         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
4996             const struct shash_node *node = ports[j];
4997             struct ofport *ofport = node->data;
4998             struct smap config;
4999             odp_port_t odp_port;
5000
5001             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5002                           ofport->ofp_port);
5003
5004             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5005             if (odp_port != ODPP_NONE) {
5006                 ds_put_format(ds, "%"PRIu32":", odp_port);
5007             } else {
5008                 ds_put_cstr(ds, "none:");
5009             }
5010
5011             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5012
5013             smap_init(&config);
5014             if (!netdev_get_config(ofport->netdev, &config)) {
5015                 const struct smap_node **nodes;
5016                 size_t i;
5017
5018                 nodes = smap_sort(&config);
5019                 for (i = 0; i < smap_count(&config); i++) {
5020                     const struct smap_node *node = nodes[i];
5021                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5022                                   node->key, node->value);
5023                 }
5024                 free(nodes);
5025             }
5026             smap_destroy(&config);
5027
5028             ds_put_char(ds, ')');
5029             ds_put_char(ds, '\n');
5030         }
5031         free(ports);
5032     }
5033     shash_destroy(&ofproto_shash);
5034     free(ofprotos);
5035 }
5036
5037 static void
5038 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5039                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5040 {
5041     struct ds ds = DS_EMPTY_INITIALIZER;
5042     const struct shash_node **backers;
5043     int i;
5044
5045     backers = shash_sort(&all_dpif_backers);
5046     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5047         dpif_show_backer(backers[i]->data, &ds);
5048     }
5049     free(backers);
5050
5051     unixctl_command_reply(conn, ds_cstr(&ds));
5052     ds_destroy(&ds);
5053 }
5054
5055 static void
5056 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5057                                 int argc OVS_UNUSED, const char *argv[],
5058                                 void *aux OVS_UNUSED)
5059 {
5060     const struct ofproto_dpif *ofproto;
5061
5062     struct ds ds = DS_EMPTY_INITIALIZER;
5063     bool verbosity = false;
5064
5065     struct dpif_port dpif_port;
5066     struct dpif_port_dump port_dump;
5067     struct hmap portno_names;
5068
5069     struct dpif_flow_dump *flow_dump;
5070     struct dpif_flow_dump_thread *flow_dump_thread;
5071     struct dpif_flow f;
5072     int error;
5073
5074     ofproto = ofproto_dpif_lookup(argv[argc - 1]);
5075     if (!ofproto) {
5076         unixctl_command_reply_error(conn, "no such bridge");
5077         return;
5078     }
5079
5080     if (argc > 2 && !strcmp(argv[1], "-m")) {
5081         verbosity = true;
5082     }
5083
5084     hmap_init(&portno_names);
5085     DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
5086         odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
5087     }
5088
5089     ds_init(&ds);
5090     flow_dump = dpif_flow_dump_create(ofproto->backer->dpif, false);
5091     flow_dump_thread = dpif_flow_dump_thread_create(flow_dump);
5092     while (dpif_flow_dump_next(flow_dump_thread, &f, 1)) {
5093         struct flow flow;
5094
5095         if (odp_flow_key_to_flow(f.key, f.key_len, &flow) == ODP_FIT_ERROR
5096             || xlate_lookup_ofproto(ofproto->backer, &flow, NULL) != ofproto) {
5097             continue;
5098         }
5099
5100         if (verbosity) {
5101             odp_format_ufid(&f.ufid, &ds);
5102             ds_put_cstr(&ds, " ");
5103         }
5104         odp_flow_format(f.key, f.key_len, f.mask, f.mask_len,
5105                         &portno_names, &ds, verbosity);
5106         ds_put_cstr(&ds, ", ");
5107         dpif_flow_stats_format(&f.stats, &ds);
5108         ds_put_cstr(&ds, ", actions:");
5109         format_odp_actions(&ds, f.actions, f.actions_len);
5110         ds_put_char(&ds, '\n');
5111     }
5112     dpif_flow_dump_thread_destroy(flow_dump_thread);
5113     error = dpif_flow_dump_destroy(flow_dump);
5114
5115     if (error) {
5116         ds_clear(&ds);
5117         ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
5118         unixctl_command_reply_error(conn, ds_cstr(&ds));
5119     } else {
5120         unixctl_command_reply(conn, ds_cstr(&ds));
5121     }
5122     odp_portno_names_destroy(&portno_names);
5123     hmap_destroy(&portno_names);
5124     ds_destroy(&ds);
5125 }
5126
5127 static void
5128 ofproto_revalidate_all_backers(void)
5129 {
5130     const struct shash_node **backers;
5131     int i;
5132
5133     backers = shash_sort(&all_dpif_backers);
5134     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5135         struct dpif_backer *backer = backers[i]->data;
5136         backer->need_revalidate = REV_RECONFIGURE;
5137     }
5138     free(backers);
5139 }
5140
5141 static void
5142 disable_tnl_push_pop(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5143                      const char *argv[], void *aux OVS_UNUSED)
5144 {
5145     if (!strcasecmp(argv[1], "off")) {
5146         ofproto_use_tnl_push_pop = false;
5147         unixctl_command_reply(conn, "Tunnel push-pop off");
5148         ofproto_revalidate_all_backers();
5149     } else if (!strcasecmp(argv[1], "on")) {
5150         ofproto_use_tnl_push_pop = true;
5151         unixctl_command_reply(conn, "Tunnel push-pop on");
5152         ofproto_revalidate_all_backers();
5153     }
5154 }
5155
5156 static void
5157 ofproto_unixctl_init(void)
5158 {
5159     static bool registered;
5160     if (registered) {
5161         return;
5162     }
5163     registered = true;
5164
5165     unixctl_command_register(
5166         "ofproto/trace",
5167         "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
5168         1, 3, ofproto_unixctl_trace, NULL);
5169     unixctl_command_register(
5170         "ofproto/trace-packet-out",
5171         "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
5172         2, 6, ofproto_unixctl_trace_actions, NULL);
5173     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5174                              ofproto_unixctl_fdb_flush, NULL);
5175     unixctl_command_register("fdb/show", "bridge", 1, 1,
5176                              ofproto_unixctl_fdb_show, NULL);
5177     unixctl_command_register("mdb/flush", "[bridge]", 0, 1,
5178                              ofproto_unixctl_mcast_snooping_flush, NULL);
5179     unixctl_command_register("mdb/show", "bridge", 1, 1,
5180                              ofproto_unixctl_mcast_snooping_show, NULL);
5181     unixctl_command_register("dpif/dump-dps", "", 0, 0,
5182                              ofproto_unixctl_dpif_dump_dps, NULL);
5183     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5184                              NULL);
5185     unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
5186                              ofproto_unixctl_dpif_dump_flows, NULL);
5187
5188     unixctl_command_register("ofproto/tnl-push-pop", "[on]|[off]", 1, 1,
5189                              disable_tnl_push_pop, NULL);
5190 }
5191
5192 /* Returns true if 'table' is the table used for internal rules,
5193  * false otherwise. */
5194 bool
5195 table_is_internal(uint8_t table_id)
5196 {
5197     return table_id == TBL_INTERNAL;
5198 }
5199 \f
5200 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5201  *
5202  * This is deprecated.  It is only for compatibility with broken device drivers
5203  * in old versions of Linux that do not properly support VLANs when VLAN
5204  * devices are not used.  When broken device drivers are no longer in
5205  * widespread use, we will delete these interfaces. */
5206
5207 static int
5208 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5209 {
5210     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5211     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5212
5213     if (realdev_ofp_port == ofport->realdev_ofp_port
5214         && vid == ofport->vlandev_vid) {
5215         return 0;
5216     }
5217
5218     ofproto->backer->need_revalidate = REV_RECONFIGURE;
5219
5220     if (ofport->realdev_ofp_port) {
5221         vsp_remove(ofport);
5222     }
5223     if (realdev_ofp_port && ofport->bundle) {
5224         /* vlandevs are enslaved to their realdevs, so they are not allowed to
5225          * themselves be part of a bundle. */
5226         bundle_set(ofport_->ofproto, ofport->bundle, NULL);
5227     }
5228
5229     ofport->realdev_ofp_port = realdev_ofp_port;
5230     ofport->vlandev_vid = vid;
5231
5232     if (realdev_ofp_port) {
5233         vsp_add(ofport, realdev_ofp_port, vid);
5234     }
5235
5236     return 0;
5237 }
5238
5239 static uint32_t
5240 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
5241 {
5242     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
5243 }
5244
5245 bool
5246 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
5247     OVS_EXCLUDED(ofproto->vsp_mutex)
5248 {
5249     /* hmap_is_empty is thread safe. */
5250     return !hmap_is_empty(&ofproto->realdev_vid_map);
5251 }
5252
5253
5254 static ofp_port_t
5255 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
5256                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5257     OVS_REQUIRES(ofproto->vsp_mutex)
5258 {
5259     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5260         int vid = vlan_tci_to_vid(vlan_tci);
5261         const struct vlan_splinter *vsp;
5262
5263         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5264                                  hash_realdev_vid(realdev_ofp_port, vid),
5265                                  &ofproto->realdev_vid_map) {
5266             if (vsp->realdev_ofp_port == realdev_ofp_port
5267                 && vsp->vid == vid) {
5268                 return vsp->vlandev_ofp_port;
5269             }
5270         }
5271     }
5272     return realdev_ofp_port;
5273 }
5274
5275 /* Returns the OFP port number of the Linux VLAN device that corresponds to
5276  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
5277  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
5278  * 'vlan_tci' 9, it would return the port number of eth0.9.
5279  *
5280  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
5281  * function just returns its 'realdev_ofp_port' argument. */
5282 ofp_port_t
5283 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5284                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5285     OVS_EXCLUDED(ofproto->vsp_mutex)
5286 {
5287     ofp_port_t ret;
5288
5289     /* hmap_is_empty is thread safe, see if we can return immediately. */
5290     if (hmap_is_empty(&ofproto->realdev_vid_map)) {
5291         return realdev_ofp_port;
5292     }
5293     ovs_mutex_lock(&ofproto->vsp_mutex);
5294     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
5295     ovs_mutex_unlock(&ofproto->vsp_mutex);
5296     return ret;
5297 }
5298
5299 static struct vlan_splinter *
5300 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
5301 {
5302     struct vlan_splinter *vsp;
5303
5304     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
5305                              hash_ofp_port(vlandev_ofp_port),
5306                              &ofproto->vlandev_map) {
5307         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5308             return vsp;
5309         }
5310     }
5311
5312     return NULL;
5313 }
5314
5315 /* Returns the OpenFlow port number of the "real" device underlying the Linux
5316  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
5317  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
5318  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
5319  * eth0 and store 9 in '*vid'.
5320  *
5321  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
5322  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
5323  * always does.*/
5324 static ofp_port_t
5325 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5326                        ofp_port_t vlandev_ofp_port, int *vid)
5327     OVS_REQUIRES(ofproto->vsp_mutex)
5328 {
5329     if (!hmap_is_empty(&ofproto->vlandev_map)) {
5330         const struct vlan_splinter *vsp;
5331
5332         vsp = vlandev_find(ofproto, vlandev_ofp_port);
5333         if (vsp) {
5334             if (vid) {
5335                 *vid = vsp->vid;
5336             }
5337             return vsp->realdev_ofp_port;
5338         }
5339     }
5340     return 0;
5341 }
5342
5343 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
5344  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
5345  * 'flow->in_port' to the "real" device backing the VLAN device, sets
5346  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Optionally pushes the
5347  * appropriate VLAN on 'packet' if provided.  Otherwise (which is always the
5348  * case unless VLAN splinters are enabled), returns false without making any
5349  * changes. */
5350 bool
5351 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow,
5352                 struct dp_packet *packet)
5353     OVS_EXCLUDED(ofproto->vsp_mutex)
5354 {
5355     ofp_port_t realdev;
5356     int vid;
5357
5358     /* hmap_is_empty is thread safe. */
5359     if (hmap_is_empty(&ofproto->vlandev_map)) {
5360         return false;
5361     }
5362
5363     ovs_mutex_lock(&ofproto->vsp_mutex);
5364     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
5365     ovs_mutex_unlock(&ofproto->vsp_mutex);
5366     if (!realdev) {
5367         return false;
5368     }
5369
5370     /* Cause the flow to be processed as if it came in on the real device with
5371      * the VLAN device's VLAN ID. */
5372     flow->in_port.ofp_port = realdev;
5373     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
5374
5375     if (packet) {
5376         /* Make the packet resemble the flow, so that it gets sent to an
5377          * OpenFlow controller properly, so that it looks correct for sFlow,
5378          * and so that flow_extract() will get the correct vlan_tci if it is
5379          * called on 'packet'. */
5380         eth_push_vlan(packet, htons(ETH_TYPE_VLAN), flow->vlan_tci);
5381     }
5382
5383     return true;
5384 }
5385
5386 static void
5387 vsp_remove(struct ofport_dpif *port)
5388 {
5389     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5390     struct vlan_splinter *vsp;
5391
5392     ovs_mutex_lock(&ofproto->vsp_mutex);
5393     vsp = vlandev_find(ofproto, port->up.ofp_port);
5394     if (vsp) {
5395         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5396         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
5397         free(vsp);
5398
5399         port->realdev_ofp_port = 0;
5400     } else {
5401         VLOG_ERR("missing vlan device record");
5402     }
5403     ovs_mutex_unlock(&ofproto->vsp_mutex);
5404 }
5405
5406 static void
5407 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
5408 {
5409     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5410
5411     ovs_mutex_lock(&ofproto->vsp_mutex);
5412     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
5413         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
5414             == realdev_ofp_port)) {
5415         struct vlan_splinter *vsp;
5416
5417         vsp = xmalloc(sizeof *vsp);
5418         vsp->realdev_ofp_port = realdev_ofp_port;
5419         vsp->vlandev_ofp_port = port->up.ofp_port;
5420         vsp->vid = vid;
5421
5422         port->realdev_ofp_port = realdev_ofp_port;
5423
5424         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
5425                     hash_ofp_port(port->up.ofp_port));
5426         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
5427                     hash_realdev_vid(realdev_ofp_port, vid));
5428     } else {
5429         VLOG_ERR("duplicate vlan device record");
5430     }
5431     ovs_mutex_unlock(&ofproto->vsp_mutex);
5432 }
5433
5434 static odp_port_t
5435 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
5436 {
5437     const struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
5438     return ofport ? ofport->odp_port : ODPP_NONE;
5439 }
5440
5441 struct ofport_dpif *
5442 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
5443 {
5444     struct ofport_dpif *port;
5445
5446     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
5447     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
5448                              &backer->odp_to_ofport_map) {
5449         if (port->odp_port == odp_port) {
5450             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5451             return port;
5452         }
5453     }
5454
5455     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5456     return NULL;
5457 }
5458
5459 static ofp_port_t
5460 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
5461 {
5462     struct ofport_dpif *port;
5463
5464     port = odp_port_to_ofport(ofproto->backer, odp_port);
5465     if (port && &ofproto->up == port->up.ofproto) {
5466         return port->up.ofp_port;
5467     } else {
5468         return OFPP_NONE;
5469     }
5470 }
5471
5472 int
5473 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
5474                                const struct match *match, int priority,
5475                                uint16_t idle_timeout,
5476                                const struct ofpbuf *ofpacts,
5477                                struct rule **rulep)
5478 {
5479     struct ofputil_flow_mod fm;
5480     struct rule_dpif *rule;
5481     int error;
5482
5483     fm.match = *match;
5484     fm.priority = priority;
5485     fm.new_cookie = htonll(0);
5486     fm.cookie = htonll(0);
5487     fm.cookie_mask = htonll(0);
5488     fm.modify_cookie = false;
5489     fm.table_id = TBL_INTERNAL;
5490     fm.command = OFPFC_ADD;
5491     fm.idle_timeout = idle_timeout;
5492     fm.hard_timeout = 0;
5493     fm.importance = 0;
5494     fm.buffer_id = 0;
5495     fm.out_port = 0;
5496     fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5497     fm.ofpacts = ofpacts->data;
5498     fm.ofpacts_len = ofpacts->size;
5499
5500     error = ofproto_flow_mod(&ofproto->up, &fm);
5501     if (error) {
5502         VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
5503                     ofperr_to_string(error));
5504         *rulep = NULL;
5505         return error;
5506     }
5507
5508     rule = rule_dpif_lookup_in_table(ofproto,
5509                                      ofproto_dpif_get_tables_version(ofproto),
5510                                      TBL_INTERNAL, &fm.match.flow,
5511                                      &fm.match.wc, false);
5512     if (rule) {
5513         *rulep = &rule->up;
5514     } else {
5515         OVS_NOT_REACHED();
5516     }
5517     return 0;
5518 }
5519
5520 int
5521 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
5522                                   struct match *match, int priority)
5523 {
5524     struct ofputil_flow_mod fm;
5525     int error;
5526
5527     fm.match = *match;
5528     fm.priority = priority;
5529     fm.new_cookie = htonll(0);
5530     fm.cookie = htonll(0);
5531     fm.cookie_mask = htonll(0);
5532     fm.modify_cookie = false;
5533     fm.table_id = TBL_INTERNAL;
5534     fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5535     fm.command = OFPFC_DELETE_STRICT;
5536
5537     error = ofproto_flow_mod(&ofproto->up, &fm);
5538     if (error) {
5539         VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
5540                     ofperr_to_string(error));
5541         return error;
5542     }
5543
5544     return 0;
5545 }
5546
5547 const struct ofproto_class ofproto_dpif_class = {
5548     init,
5549     enumerate_types,
5550     enumerate_names,
5551     del,
5552     port_open_type,
5553     type_run,
5554     type_wait,
5555     alloc,
5556     construct,
5557     destruct,
5558     dealloc,
5559     run,
5560     wait,
5561     NULL,                       /* get_memory_usage. */
5562     type_get_memory_usage,
5563     flush,
5564     query_tables,
5565     set_tables_version,
5566     port_alloc,
5567     port_construct,
5568     port_destruct,
5569     port_dealloc,
5570     port_modified,
5571     port_reconfigured,
5572     port_query_by_name,
5573     port_add,
5574     port_del,
5575     port_get_stats,
5576     port_dump_start,
5577     port_dump_next,
5578     port_dump_done,
5579     port_poll,
5580     port_poll_wait,
5581     port_is_lacp_current,
5582     port_get_lacp_stats,
5583     NULL,                       /* rule_choose_table */
5584     rule_alloc,
5585     rule_construct,
5586     rule_insert,
5587     rule_delete,
5588     rule_destruct,
5589     rule_dealloc,
5590     rule_get_stats,
5591     rule_execute,
5592     set_frag_handling,
5593     packet_out,
5594     set_netflow,
5595     get_netflow_ids,
5596     set_sflow,
5597     set_ipfix,
5598     set_cfm,
5599     cfm_status_changed,
5600     get_cfm_status,
5601     set_lldp,
5602     get_lldp_status,
5603     set_aa,
5604     aa_mapping_set,
5605     aa_mapping_unset,
5606     aa_vlan_get_queued,
5607     aa_vlan_get_queue_size,
5608     set_bfd,
5609     bfd_status_changed,
5610     get_bfd_status,
5611     set_stp,
5612     get_stp_status,
5613     set_stp_port,
5614     get_stp_port_status,
5615     get_stp_port_stats,
5616     set_rstp,
5617     get_rstp_status,
5618     set_rstp_port,
5619     get_rstp_port_status,
5620     set_queues,
5621     bundle_set,
5622     bundle_remove,
5623     mirror_set__,
5624     mirror_get_stats__,
5625     set_flood_vlans,
5626     is_mirror_output_bundle,
5627     forward_bpdu_changed,
5628     set_mac_table_config,
5629     set_mcast_snooping,
5630     set_mcast_snooping_port,
5631     set_realdev,
5632     NULL,                       /* meter_get_features */
5633     NULL,                       /* meter_set */
5634     NULL,                       /* meter_get */
5635     NULL,                       /* meter_del */
5636     group_alloc,                /* group_alloc */
5637     group_construct,            /* group_construct */
5638     group_destruct,             /* group_destruct */
5639     group_dealloc,              /* group_dealloc */
5640     group_modify,               /* group_modify */
5641     group_get_stats,            /* group_get_stats */
5642     get_datapath_version,       /* get_datapath_version */
5643 };