ofproto/ofproto-dpif.c

   1 /*
   2  * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18
  19 #include "ofproto/ofproto-dpif.h"
  20 #include "ofproto/ofproto-provider.h"
  21
  22 #include <errno.h>
  23
  24 #include "bfd.h"
  25 #include "bond.h"
  26 #include "bundle.h"
  27 #include "byte-order.h"
  28 #include "connectivity.h"
  29 #include "connmgr.h"
  30 #include "coverage.h"
  31 #include "cfm.h"
  32 #include "dpif.h"
  33 #include "dynamic-string.h"
  34 #include "fail-open.h"
  35 #include "guarded-list.h"
  36 #include "hmapx.h"
  37 #include "lacp.h"
  38 #include "learn.h"
  39 #include "mac-learning.h"
  40 #include "mcast-snooping.h"
  41 #include "meta-flow.h"
  42 #include "multipath.h"
  43 #include "netdev-vport.h"
  44 #include "netdev.h"
  45 #include "netlink.h"
  46 #include "nx-match.h"
  47 #include "odp-util.h"
  48 #include "odp-execute.h"
  49 #include "ofp-util.h"
  50 #include "ofpbuf.h"
  51 #include "ofp-actions.h"
  52 #include "ofp-parse.h"
  53 #include "ofp-print.h"
  54 #include "ofproto-dpif-ipfix.h"
  55 #include "ofproto-dpif-mirror.h"
  56 #include "ofproto-dpif-monitor.h"
  57 #include "ofproto-dpif-rid.h"
  58 #include "ofproto-dpif-sflow.h"
  59 #include "ofproto-dpif-upcall.h"
  60 #include "ofproto-dpif-xlate.h"
  61 #include "poll-loop.h"
  62 #include "seq.h"
  63 #include "simap.h"
  64 #include "smap.h"
  65 #include "timer.h"
  66 #include "tunnel.h"
  67 #include "unaligned.h"
  68 #include "unixctl.h"
  69 #include "vlan-bitmap.h"
  70 #include "vlog.h"
  71
  72 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  73
  74 COVERAGE_DEFINE(ofproto_dpif_expired);
  75 COVERAGE_DEFINE(packet_in_overflow);
  76
  77 struct flow_miss;
  78
  79 struct rule_dpif {
  80     struct rule up;
  81
  82     /* These statistics:
  83      *
  84      *   - Do include packets and bytes from datapath flows which have not
  85      *   recently been processed by a revalidator. */
  86     struct ovs_mutex stats_mutex;
  87     struct dpif_flow_stats stats OVS_GUARDED;
  88
  89     /* If non-zero then the recirculation id that has
  90      * been allocated for use with this rule.
  91      * The recirculation id and associated internal flow should
  92      * be freed when the rule is freed */
  93     uint32_t recirc_id;
  94 };
  95
  96 /* RULE_CAST() depends on this. */
  97 BUILD_ASSERT_DECL(offsetof(struct rule_dpif, up) == 0);
  98
  99 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
 100                            long long int *used);
 101 static struct rule_dpif *rule_dpif_cast(const struct rule *);
 102 static void rule_expire(struct rule_dpif *);
 103
 104 struct group_dpif {
 105     struct ofgroup up;
 106
 107     /* These statistics:
 108      *
 109      *   - Do include packets and bytes from datapath flows which have not
 110      *   recently been processed by a revalidator. */
 111     struct ovs_mutex stats_mutex;
 112     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 113     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 114 };
 115
 116 struct ofbundle {
 117     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
 118     struct ofproto_dpif *ofproto; /* Owning ofproto. */
 119     void *aux;                  /* Key supplied by ofproto's client. */
 120     char *name;                 /* Identifier for log messages. */
 121
 122     /* Configuration. */
 123     struct list ports;          /* Contains "struct ofport"s. */
 124     enum port_vlan_mode vlan_mode; /* VLAN mode */
 125     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
 126     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
 127                                  * NULL if all VLANs are trunked. */
 128     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 129     struct bond *bond;          /* Nonnull iff more than one port. */
 130     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 131
 132     /* Status. */
 133     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 134 };
 135
 136 static void bundle_remove(struct ofport *);
 137 static void bundle_update(struct ofbundle *);
 138 static void bundle_destroy(struct ofbundle *);
 139 static void bundle_del_port(struct ofport_dpif *);
 140 static void bundle_run(struct ofbundle *);
 141 static void bundle_wait(struct ofbundle *);
 142
 143 static void stp_run(struct ofproto_dpif *ofproto);
 144 static void stp_wait(struct ofproto_dpif *ofproto);
 145 static int set_stp_port(struct ofport *,
 146                         const struct ofproto_port_stp_settings *);
 147
 148 struct ofport_dpif {
 149     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 150     struct ofport up;
 151
 152     odp_port_t odp_port;
 153     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 154     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
 155     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 156     struct bfd *bfd;            /* BFD, if any. */
 157     bool may_enable;            /* May be enabled in bonds. */
 158     bool is_tunnel;             /* This port is a tunnel. */
 159     bool is_layer3;             /* This is a layer 3 port. */
 160     long long int carrier_seq;  /* Carrier status changes. */
 161     struct ofport_dpif *peer;   /* Peer if patch port. */
 162
 163     /* Spanning tree. */
 164     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 165     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 166     long long int stp_state_entered;
 167
 168     /* Queue to DSCP mapping. */
 169     struct ofproto_port_queue *qdscp;
 170     size_t n_qdscp;
 171
 172     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 173      *
 174      * This is deprecated.  It is only for compatibility with broken device
 175      * drivers in old versions of Linux that do not properly support VLANs when
 176      * VLAN devices are not used.  When broken device drivers are no longer in
 177      * widespread use, we will delete these interfaces. */
 178     ofp_port_t realdev_ofp_port;
 179     int vlandev_vid;
 180 };
 181
 182 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 183  *
 184  * This is deprecated.  It is only for compatibility with broken device drivers
 185  * in old versions of Linux that do not properly support VLANs when VLAN
 186  * devices are not used.  When broken device drivers are no longer in
 187  * widespread use, we will delete these interfaces. */
 188 struct vlan_splinter {
 189     struct hmap_node realdev_vid_node;
 190     struct hmap_node vlandev_node;
 191     ofp_port_t realdev_ofp_port;
 192     ofp_port_t vlandev_ofp_port;
 193     int vid;
 194 };
 195
 196 static void vsp_remove(struct ofport_dpif *);
 197 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 198
 199 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 200                                        ofp_port_t);
 201
 202 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 203                                        odp_port_t);
 204
 205 static struct ofport_dpif *
 206 ofport_dpif_cast(const struct ofport *ofport)
 207 {
 208     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 209 }
 210
 211 static void port_run(struct ofport_dpif *);
 212 static int set_bfd(struct ofport *, const struct smap *);
 213 static int set_cfm(struct ofport *, const struct cfm_settings *);
 214 static void ofport_update_peer(struct ofport_dpif *);
 215
 216 /* Reasons that we might need to revalidate every datapath flow, and
 217  * corresponding coverage counters.
 218  *
 219  * A value of 0 means that there is no need to revalidate.
 220  *
 221  * It would be nice to have some cleaner way to integrate with coverage
 222  * counters, but with only a few reasons I guess this is good enough for
 223  * now. */
 224 enum revalidate_reason {
 225     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 226     REV_STP,                   /* Spanning tree protocol port status change. */
 227     REV_BOND,                  /* Bonding changed. */
 228     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 229     REV_FLOW_TABLE,            /* Flow table changed. */
 230     REV_MAC_LEARNING,          /* Mac learning changed. */
 231     REV_MCAST_SNOOPING,        /* Multicast snooping changed. */
 232 };
 233 COVERAGE_DEFINE(rev_reconfigure);
 234 COVERAGE_DEFINE(rev_stp);
 235 COVERAGE_DEFINE(rev_bond);
 236 COVERAGE_DEFINE(rev_port_toggled);
 237 COVERAGE_DEFINE(rev_flow_table);
 238 COVERAGE_DEFINE(rev_mac_learning);
 239 COVERAGE_DEFINE(rev_mcast_snooping);
 240
 241 /* All datapaths of a given type share a single dpif backer instance. */
 242 struct dpif_backer {
 243     char *type;
 244     int refcount;
 245     struct dpif *dpif;
 246     struct udpif *udpif;
 247
 248     struct ovs_rwlock odp_to_ofport_lock;
 249     struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
 250
 251     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 252
 253     enum revalidate_reason need_revalidate; /* Revalidate all flows. */
 254
 255     bool recv_set_enable; /* Enables or disables receiving packets. */
 256
 257     /* Recirculation. */
 258     struct recirc_id_pool *rid_pool;       /* Recirculation ID pool. */
 259     bool enable_recirc;   /* True if the datapath supports recirculation */
 260
 261     /* True if the datapath supports variable-length
 262      * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
 263      * False if the datapath supports only 8-byte (or shorter) userdata. */
 264     bool variable_length_userdata;
 265
 266     /* Maximum number of MPLS label stack entries that the datapath supports
 267      * in a match */
 268     size_t max_mpls_depth;
 269 };
 270
 271 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 272 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 273
 274 struct ofproto_dpif {
 275     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 276     struct ofproto up;
 277     struct dpif_backer *backer;
 278
 279     uint64_t dump_seq; /* Last read of udpif_dump_seq(). */
 280
 281     /* Special OpenFlow rules. */
 282     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 283     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 284     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 285
 286     /* Bridging. */
 287     struct netflow *netflow;
 288     struct dpif_sflow *sflow;
 289     struct dpif_ipfix *ipfix;
 290     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 291     struct mac_learning *ml;
 292     struct mcast_snooping *ms;
 293     bool has_bonded_bundles;
 294     bool lacp_enabled;
 295     struct mbridge *mbridge;
 296
 297     struct ovs_mutex stats_mutex;
 298     struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
 299                                             * consumed in userspace. */
 300
 301     /* Spanning tree. */
 302     struct stp *stp;
 303     long long int stp_last_tick;
 304
 305     /* VLAN splinters. */
 306     struct ovs_mutex vsp_mutex;
 307     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 308     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 309
 310     /* Ports. */
 311     struct sset ports;             /* Set of standard port names. */
 312     struct sset ghost_ports;       /* Ports with no datapath port. */
 313     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 314     int port_poll_errno;           /* Last errno for port_poll() reply. */
 315     uint64_t change_seq;           /* Connectivity status changes. */
 316
 317     /* Work queues. */
 318     struct guarded_list pins;      /* Contains "struct ofputil_packet_in"s. */
 319     struct seq *pins_seq;          /* For notifying 'pins' reception. */
 320     uint64_t pins_seqno;
 321 };
 322
 323 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 324 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 325
 326 static void ofproto_dpif_unixctl_init(void);
 327
 328 static inline struct ofproto_dpif *
 329 ofproto_dpif_cast(const struct ofproto *ofproto)
 330 {
 331     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 332     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 333 }
 334
 335 size_t
 336 ofproto_dpif_get_max_mpls_depth(const struct ofproto_dpif *ofproto)
 337 {
 338     return ofproto->backer->max_mpls_depth;
 339 }
 340
 341 bool
 342 ofproto_dpif_get_enable_recirc(const struct ofproto_dpif *ofproto)
 343 {
 344     return ofproto->backer->enable_recirc;
 345 }
 346
 347 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
 348                                         ofp_port_t ofp_port);
 349 static void ofproto_trace(struct ofproto_dpif *, struct flow *,
 350                           const struct ofpbuf *packet,
 351                           const struct ofpact[], size_t ofpacts_len,
 352                           struct ds *);
 353
 354 /* Global variables. */
 355 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 356
 357 /* Initial mappings of port to bridge mappings. */
 358 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 359
 360 /* Executes 'fm'.  The caller retains ownership of 'fm' and everything in
 361  * it. */
 362 void
 363 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 364                       struct ofputil_flow_mod *fm)
 365 {
 366     ofproto_flow_mod(&ofproto->up, fm);
 367 }
 368
 369 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 370  * Takes ownership of 'pin' and pin->packet. */
 371 void
 372 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 373                             struct ofproto_packet_in *pin)
 374 {
 375     if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
 376         COVERAGE_INC(packet_in_overflow);
 377         free(CONST_CAST(void *, pin->up.packet));
 378         free(pin);
 379     }
 380
 381     /* Wakes up main thread for packet-in I/O. */
 382     seq_change(ofproto->pins_seq);
 383 }
 384
 385 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
 386  * packet rather than to send the packet to the controller.
 387  *
 388  * This function returns false to indicate that a packet_in message
 389  * for a "table-miss" should be sent to at least one controller.
 390  * False otherwise. */
 391 bool
 392 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif *ofproto)
 393 {
 394     return connmgr_wants_packet_in_on_miss(ofproto->up.connmgr);
 395 }
 396 \f
 397 /* Factory functions. */
 398
 399 static void
 400 init(const struct shash *iface_hints)
 401 {
 402     struct shash_node *node;
 403
 404     /* Make a local copy, since we don't own 'iface_hints' elements. */
 405     SHASH_FOR_EACH(node, iface_hints) {
 406         const struct iface_hint *orig_hint = node->data;
 407         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 408
 409         new_hint->br_name = xstrdup(orig_hint->br_name);
 410         new_hint->br_type = xstrdup(orig_hint->br_type);
 411         new_hint->ofp_port = orig_hint->ofp_port;
 412
 413         shash_add(&init_ofp_ports, node->name, new_hint);
 414     }
 415 }
 416
 417 static void
 418 enumerate_types(struct sset *types)
 419 {
 420     dp_enumerate_types(types);
 421 }
 422
 423 static int
 424 enumerate_names(const char *type, struct sset *names)
 425 {
 426     struct ofproto_dpif *ofproto;
 427
 428     sset_clear(names);
 429     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 430         if (strcmp(type, ofproto->up.type)) {
 431             continue;
 432         }
 433         sset_add(names, ofproto->up.name);
 434     }
 435
 436     return 0;
 437 }
 438
 439 static int
 440 del(const char *type, const char *name)
 441 {
 442     struct dpif *dpif;
 443     int error;
 444
 445     error = dpif_open(name, type, &dpif);
 446     if (!error) {
 447         error = dpif_delete(dpif);
 448         dpif_close(dpif);
 449     }
 450     return error;
 451 }
 452 \f
 453 static const char *
 454 port_open_type(const char *datapath_type, const char *port_type)
 455 {
 456     return dpif_port_open_type(datapath_type, port_type);
 457 }
 458
 459 /* Type functions. */
 460
 461 static void process_dpif_port_changes(struct dpif_backer *);
 462 static void process_dpif_all_ports_changed(struct dpif_backer *);
 463 static void process_dpif_port_change(struct dpif_backer *,
 464                                      const char *devname);
 465 static void process_dpif_port_error(struct dpif_backer *, int error);
 466
 467 static struct ofproto_dpif *
 468 lookup_ofproto_dpif_by_port_name(const char *name)
 469 {
 470     struct ofproto_dpif *ofproto;
 471
 472     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 473         if (sset_contains(&ofproto->ports, name)) {
 474             return ofproto;
 475         }
 476     }
 477
 478     return NULL;
 479 }
 480
 481 static int
 482 type_run(const char *type)
 483 {
 484     struct dpif_backer *backer;
 485
 486     backer = shash_find_data(&all_dpif_backers, type);
 487     if (!backer) {
 488         /* This is not necessarily a problem, since backers are only
 489          * created on demand. */
 490         return 0;
 491     }
 492
 493     dpif_run(backer->dpif);
 494     udpif_run(backer->udpif);
 495
 496     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 497      * and the configuration has now changed to "false", enable receiving
 498      * packets from the datapath. */
 499     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 500         int error;
 501
 502         backer->recv_set_enable = true;
 503
 504         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 505         if (error) {
 506             VLOG_ERR("Failed to enable receiving packets in dpif.");
 507             return error;
 508         }
 509         dpif_flow_flush(backer->dpif);
 510         backer->need_revalidate = REV_RECONFIGURE;
 511     }
 512
 513     if (backer->recv_set_enable) {
 514         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 515     }
 516
 517     if (backer->need_revalidate) {
 518         struct ofproto_dpif *ofproto;
 519         struct simap_node *node;
 520         struct simap tmp_backers;
 521
 522         /* Handle tunnel garbage collection. */
 523         simap_init(&tmp_backers);
 524         simap_swap(&backer->tnl_backers, &tmp_backers);
 525
 526         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 527             struct ofport_dpif *iter;
 528
 529             if (backer != ofproto->backer) {
 530                 continue;
 531             }
 532
 533             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 534                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 535                 const char *dp_port;
 536
 537                 if (!iter->is_tunnel) {
 538                     continue;
 539                 }
 540
 541                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 542                                                      namebuf, sizeof namebuf);
 543                 node = simap_find(&tmp_backers, dp_port);
 544                 if (node) {
 545                     simap_put(&backer->tnl_backers, dp_port, node->data);
 546                     simap_delete(&tmp_backers, node);
 547                     node = simap_find(&backer->tnl_backers, dp_port);
 548                 } else {
 549                     node = simap_find(&backer->tnl_backers, dp_port);
 550                     if (!node) {
 551                         odp_port_t odp_port = ODPP_NONE;
 552
 553                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 554                                            &odp_port)) {
 555                             simap_put(&backer->tnl_backers, dp_port,
 556                                       odp_to_u32(odp_port));
 557                             node = simap_find(&backer->tnl_backers, dp_port);
 558                         }
 559                     }
 560                 }
 561
 562                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 563                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 564                                          iter->odp_port)) {
 565                     backer->need_revalidate = REV_RECONFIGURE;
 566                 }
 567             }
 568         }
 569
 570         SIMAP_FOR_EACH (node, &tmp_backers) {
 571             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 572         }
 573         simap_destroy(&tmp_backers);
 574
 575         switch (backer->need_revalidate) {
 576         case REV_RECONFIGURE:    COVERAGE_INC(rev_reconfigure);    break;
 577         case REV_STP:            COVERAGE_INC(rev_stp);            break;
 578         case REV_BOND:           COVERAGE_INC(rev_bond);           break;
 579         case REV_PORT_TOGGLED:   COVERAGE_INC(rev_port_toggled);   break;
 580         case REV_FLOW_TABLE:     COVERAGE_INC(rev_flow_table);     break;
 581         case REV_MAC_LEARNING:   COVERAGE_INC(rev_mac_learning);   break;
 582         case REV_MCAST_SNOOPING: COVERAGE_INC(rev_mcast_snooping); break;
 583         }
 584         backer->need_revalidate = 0;
 585
 586         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 587             struct ofport_dpif *ofport;
 588             struct ofbundle *bundle;
 589
 590             if (ofproto->backer != backer) {
 591                 continue;
 592             }
 593
 594             xlate_txn_start();
 595             xlate_ofproto_set(ofproto, ofproto->up.name,
 596                               ofproto->backer->dpif, ofproto->miss_rule,
 597                               ofproto->no_packet_in_rule, ofproto->ml,
 598                               ofproto->stp, ofproto->ms, ofproto->mbridge,
 599                               ofproto->sflow, ofproto->ipfix,
 600                               ofproto->netflow, ofproto->up.frag_handling,
 601                               ofproto->up.forward_bpdu,
 602                               connmgr_has_in_band(ofproto->up.connmgr),
 603                               ofproto->backer->enable_recirc,
 604                               ofproto->backer->variable_length_userdata,
 605                               ofproto->backer->max_mpls_depth);
 606
 607             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 608                 xlate_bundle_set(ofproto, bundle, bundle->name,
 609                                  bundle->vlan_mode, bundle->vlan,
 610                                  bundle->trunks, bundle->use_priority_tags,
 611                                  bundle->bond, bundle->lacp,
 612                                  bundle->floodable);
 613             }
 614
 615             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 616                 int stp_port = ofport->stp_port
 617                     ? stp_port_no(ofport->stp_port)
 618                     : -1;
 619                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 620                                  ofport->up.ofp_port, ofport->odp_port,
 621                                  ofport->up.netdev, ofport->cfm,
 622                                  ofport->bfd, ofport->peer, stp_port,
 623                                  ofport->qdscp, ofport->n_qdscp,
 624                                  ofport->up.pp.config, ofport->up.pp.state,
 625                                  ofport->is_tunnel, ofport->may_enable);
 626             }
 627             xlate_txn_commit();
 628         }
 629
 630         udpif_revalidate(backer->udpif);
 631     }
 632
 633     process_dpif_port_changes(backer);
 634
 635     return 0;
 636 }
 637
 638 /* Check for and handle port changes in 'backer''s dpif. */
 639 static void
 640 process_dpif_port_changes(struct dpif_backer *backer)
 641 {
 642     for (;;) {
 643         char *devname;
 644         int error;
 645
 646         error = dpif_port_poll(backer->dpif, &devname);
 647         switch (error) {
 648         case EAGAIN:
 649             return;
 650
 651         case ENOBUFS:
 652             process_dpif_all_ports_changed(backer);
 653             break;
 654
 655         case 0:
 656             process_dpif_port_change(backer, devname);
 657             free(devname);
 658             break;
 659
 660         default:
 661             process_dpif_port_error(backer, error);
 662             break;
 663         }
 664     }
 665 }
 666
 667 static void
 668 process_dpif_all_ports_changed(struct dpif_backer *backer)
 669 {
 670     struct ofproto_dpif *ofproto;
 671     struct dpif_port dpif_port;
 672     struct dpif_port_dump dump;
 673     struct sset devnames;
 674     const char *devname;
 675
 676     sset_init(&devnames);
 677     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 678         if (ofproto->backer == backer) {
 679             struct ofport *ofport;
 680
 681             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 682                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 683             }
 684         }
 685     }
 686     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 687         sset_add(&devnames, dpif_port.name);
 688     }
 689
 690     SSET_FOR_EACH (devname, &devnames) {
 691         process_dpif_port_change(backer, devname);
 692     }
 693     sset_destroy(&devnames);
 694 }
 695
 696 static void
 697 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 698 {
 699     struct ofproto_dpif *ofproto;
 700     struct dpif_port port;
 701
 702     /* Don't report on the datapath's device. */
 703     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 704         return;
 705     }
 706
 707     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 708                    &all_ofproto_dpifs) {
 709         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 710             return;
 711         }
 712     }
 713
 714     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 715     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 716         /* The port was removed.  If we know the datapath,
 717          * report it through poll_set().  If we don't, it may be
 718          * notifying us of a removal we initiated, so ignore it.
 719          * If there's a pending ENOBUFS, let it stand, since
 720          * everything will be reevaluated. */
 721         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 722             sset_add(&ofproto->port_poll_set, devname);
 723             ofproto->port_poll_errno = 0;
 724         }
 725     } else if (!ofproto) {
 726         /* The port was added, but we don't know with which
 727          * ofproto we should associate it.  Delete it. */
 728         dpif_port_del(backer->dpif, port.port_no);
 729     } else {
 730         struct ofport_dpif *ofport;
 731
 732         ofport = ofport_dpif_cast(shash_find_data(
 733                                       &ofproto->up.port_by_name, devname));
 734         if (ofport
 735             && ofport->odp_port != port.port_no
 736             && !odp_port_to_ofport(backer, port.port_no))
 737         {
 738             /* 'ofport''s datapath port number has changed from
 739              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 740              * structures to match. */
 741             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 742             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 743             ofport->odp_port = port.port_no;
 744             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 745                         hash_odp_port(port.port_no));
 746             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 747             backer->need_revalidate = REV_RECONFIGURE;
 748         }
 749     }
 750     dpif_port_destroy(&port);
 751 }
 752
 753 /* Propagate 'error' to all ofprotos based on 'backer'. */
 754 static void
 755 process_dpif_port_error(struct dpif_backer *backer, int error)
 756 {
 757     struct ofproto_dpif *ofproto;
 758
 759     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 760         if (ofproto->backer == backer) {
 761             sset_clear(&ofproto->port_poll_set);
 762             ofproto->port_poll_errno = error;
 763         }
 764     }
 765 }
 766
 767 static void
 768 type_wait(const char *type)
 769 {
 770     struct dpif_backer *backer;
 771
 772     backer = shash_find_data(&all_dpif_backers, type);
 773     if (!backer) {
 774         /* This is not necessarily a problem, since backers are only
 775          * created on demand. */
 776         return;
 777     }
 778
 779     dpif_wait(backer->dpif);
 780 }
 781 \f
 782 /* Basic life-cycle. */
 783
 784 static int add_internal_flows(struct ofproto_dpif *);
 785
 786 static struct ofproto *
 787 alloc(void)
 788 {
 789     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
 790     return &ofproto->up;
 791 }
 792
 793 static void
 794 dealloc(struct ofproto *ofproto_)
 795 {
 796     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
 797     free(ofproto);
 798 }
 799
 800 static void
 801 close_dpif_backer(struct dpif_backer *backer)
 802 {
 803     ovs_assert(backer->refcount > 0);
 804
 805     if (--backer->refcount) {
 806         return;
 807     }
 808
 809     udpif_destroy(backer->udpif);
 810
 811     simap_destroy(&backer->tnl_backers);
 812     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
 813     hmap_destroy(&backer->odp_to_ofport_map);
 814     shash_find_and_delete(&all_dpif_backers, backer->type);
 815     recirc_id_pool_destroy(backer->rid_pool);
 816     free(backer->type);
 817     dpif_close(backer->dpif);
 818     free(backer);
 819 }
 820
 821 /* Datapath port slated for removal from datapath. */
 822 struct odp_garbage {
 823     struct list list_node;
 824     odp_port_t odp_port;
 825 };
 826
 827 static bool check_variable_length_userdata(struct dpif_backer *backer);
 828 static size_t check_max_mpls_depth(struct dpif_backer *backer);
 829 static bool check_recirc(struct dpif_backer *backer);
 830
 831 static int
 832 open_dpif_backer(const char *type, struct dpif_backer **backerp)
 833 {
 834     struct dpif_backer *backer;
 835     struct dpif_port_dump port_dump;
 836     struct dpif_port port;
 837     struct shash_node *node;
 838     struct list garbage_list;
 839     struct odp_garbage *garbage, *next;
 840
 841     struct sset names;
 842     char *backer_name;
 843     const char *name;
 844     int error;
 845
 846     backer = shash_find_data(&all_dpif_backers, type);
 847     if (backer) {
 848         backer->refcount++;
 849         *backerp = backer;
 850         return 0;
 851     }
 852
 853     backer_name = xasprintf("ovs-%s", type);
 854
 855     /* Remove any existing datapaths, since we assume we're the only
 856      * userspace controlling the datapath. */
 857     sset_init(&names);
 858     dp_enumerate_names(type, &names);
 859     SSET_FOR_EACH(name, &names) {
 860         struct dpif *old_dpif;
 861
 862         /* Don't remove our backer if it exists. */
 863         if (!strcmp(name, backer_name)) {
 864             continue;
 865         }
 866
 867         if (dpif_open(name, type, &old_dpif)) {
 868             VLOG_WARN("couldn't open old datapath %s to remove it", name);
 869         } else {
 870             dpif_delete(old_dpif);
 871             dpif_close(old_dpif);
 872         }
 873     }
 874     sset_destroy(&names);
 875
 876     backer = xmalloc(sizeof *backer);
 877
 878     error = dpif_create_and_open(backer_name, type, &backer->dpif);
 879     free(backer_name);
 880     if (error) {
 881         VLOG_ERR("failed to open datapath of type %s: %s", type,
 882                  ovs_strerror(error));
 883         free(backer);
 884         return error;
 885     }
 886     backer->udpif = udpif_create(backer, backer->dpif);
 887
 888     backer->type = xstrdup(type);
 889     backer->refcount = 1;
 890     hmap_init(&backer->odp_to_ofport_map);
 891     ovs_rwlock_init(&backer->odp_to_ofport_lock);
 892     backer->need_revalidate = 0;
 893     simap_init(&backer->tnl_backers);
 894     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
 895     *backerp = backer;
 896
 897     if (backer->recv_set_enable) {
 898         dpif_flow_flush(backer->dpif);
 899     }
 900
 901     /* Loop through the ports already on the datapath and remove any
 902      * that we don't need anymore. */
 903     list_init(&garbage_list);
 904     dpif_port_dump_start(&port_dump, backer->dpif);
 905     while (dpif_port_dump_next(&port_dump, &port)) {
 906         node = shash_find(&init_ofp_ports, port.name);
 907         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
 908             garbage = xmalloc(sizeof *garbage);
 909             garbage->odp_port = port.port_no;
 910             list_push_front(&garbage_list, &garbage->list_node);
 911         }
 912     }
 913     dpif_port_dump_done(&port_dump);
 914
 915     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
 916         dpif_port_del(backer->dpif, garbage->odp_port);
 917         list_remove(&garbage->list_node);
 918         free(garbage);
 919     }
 920
 921     shash_add(&all_dpif_backers, type, backer);
 922
 923     backer->enable_recirc = check_recirc(backer);
 924     backer->variable_length_userdata = check_variable_length_userdata(backer);
 925     backer->max_mpls_depth = check_max_mpls_depth(backer);
 926     backer->rid_pool = recirc_id_pool_create();
 927
 928     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 929     if (error) {
 930         VLOG_ERR("failed to listen on datapath of type %s: %s",
 931                  type, ovs_strerror(error));
 932         close_dpif_backer(backer);
 933         return error;
 934     }
 935
 936     if (backer->recv_set_enable) {
 937         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 938     }
 939
 940     return error;
 941 }
 942
 943 /* Tests whether 'backer''s datapath supports recirculation.  Only newer
 944  * datapaths support OVS_KEY_ATTR_RECIRC_ID in keys.  We need to disable some
 945  * features on older datapaths that don't support this feature.
 946  *
 947  * Returns false if 'backer' definitely does not support recirculation, true if
 948  * it seems to support recirculation or if at least the error we get is
 949  * ambiguous. */
 950 static bool
 951 check_recirc(struct dpif_backer *backer)
 952 {
 953     struct flow flow;
 954     struct odputil_keybuf keybuf;
 955     struct ofpbuf key;
 956     int error;
 957     bool enable_recirc = false;
 958
 959     memset(&flow, 0, sizeof flow);
 960     flow.recirc_id = 1;
 961     flow.dp_hash = 1;
 962
 963     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
 964     odp_flow_key_from_flow(&key, &flow, NULL, 0, true);
 965
 966     error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE,
 967                           ofpbuf_data(&key), ofpbuf_size(&key), NULL, 0, NULL,
 968                           0, NULL);
 969     if (error && error != EEXIST) {
 970         if (error != EINVAL) {
 971             VLOG_WARN("%s: Reciculation flow probe failed (%s)",
 972                       dpif_name(backer->dpif), ovs_strerror(error));
 973         }
 974         goto done;
 975     }
 976
 977     error = dpif_flow_del(backer->dpif, ofpbuf_data(&key), ofpbuf_size(&key),
 978                           NULL);
 979     if (error) {
 980         VLOG_WARN("%s: failed to delete recirculation feature probe flow",
 981                   dpif_name(backer->dpif));
 982     }
 983
 984     enable_recirc = true;
 985
 986 done:
 987     if (enable_recirc) {
 988         VLOG_INFO("%s: Datapath supports recirculation",
 989                   dpif_name(backer->dpif));
 990     } else {
 991         VLOG_INFO("%s: Datapath does not support recirculation",
 992                   dpif_name(backer->dpif));
 993     }
 994
 995     return enable_recirc;
 996 }
 997
 998 /* Tests whether 'backer''s datapath supports variable-length
 999  * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.  We need
1000  * to disable some features on older datapaths that don't support this
1001  * feature.
1002  *
1003  * Returns false if 'backer' definitely does not support variable-length
1004  * userdata, true if it seems to support them or if at least the error we get
1005  * is ambiguous. */
1006 static bool
1007 check_variable_length_userdata(struct dpif_backer *backer)
1008 {
1009     struct eth_header *eth;
1010     struct ofpbuf actions;
1011     struct dpif_execute execute;
1012     struct ofpbuf packet;
1013     size_t start;
1014     int error;
1015
1016     /* Compose a userspace action that will cause an ERANGE error on older
1017      * datapaths that don't support variable-length userdata.
1018      *
1019      * We really test for using userdata longer than 8 bytes, but older
1020      * datapaths accepted these, silently truncating the userdata to 8 bytes.
1021      * The same older datapaths rejected userdata shorter than 8 bytes, so we
1022      * test for that instead as a proxy for longer userdata support. */
1023     ofpbuf_init(&actions, 64);
1024     start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
1025     nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
1026                    dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
1027     nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
1028     nl_msg_end_nested(&actions, start);
1029
1030     /* Compose a dummy ethernet packet. */
1031     ofpbuf_init(&packet, ETH_HEADER_LEN);
1032     eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
1033     eth->eth_type = htons(0x1234);
1034
1035     /* Execute the actions.  On older datapaths this fails with ERANGE, on
1036      * newer datapaths it succeeds. */
1037     execute.actions = ofpbuf_data(&actions);
1038     execute.actions_len = ofpbuf_size(&actions);
1039     execute.packet = &packet;
1040     execute.md = PKT_METADATA_INITIALIZER(0);
1041     execute.needs_help = false;
1042
1043     error = dpif_execute(backer->dpif, &execute);
1044
1045     ofpbuf_uninit(&packet);
1046     ofpbuf_uninit(&actions);
1047
1048     switch (error) {
1049     case 0:
1050         return true;
1051
1052     case ERANGE:
1053         /* Variable-length userdata is not supported. */
1054         VLOG_WARN("%s: datapath does not support variable-length userdata "
1055                   "feature (needs Linux 3.10+ or kernel module from OVS "
1056                   "1..11+).  The NXAST_SAMPLE action will be ignored.",
1057                   dpif_name(backer->dpif));
1058         return false;
1059
1060     default:
1061         /* Something odd happened.  We're not sure whether variable-length
1062          * userdata is supported.  Default to "yes". */
1063         VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1064                   dpif_name(backer->dpif), ovs_strerror(error));
1065         return true;
1066     }
1067 }
1068
1069 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1070  *
1071  * Returns the number of elements in a struct flow's mpls_lse field
1072  * if the datapath supports at least that many entries in an
1073  * MPLS label stack.
1074  * Otherwise returns the number of MPLS push actions supported by
1075  * the datapath. */
1076 static size_t
1077 check_max_mpls_depth(struct dpif_backer *backer)
1078 {
1079     struct flow flow;
1080     int n;
1081
1082     for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
1083         struct odputil_keybuf keybuf;
1084         struct ofpbuf key;
1085         int error;
1086
1087         memset(&flow, 0, sizeof flow);
1088         flow.dl_type = htons(ETH_TYPE_MPLS);
1089         flow_set_mpls_bos(&flow, n, 1);
1090
1091         ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1092         odp_flow_key_from_flow(&key, &flow, NULL, 0, false);
1093
1094         error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE,
1095                               ofpbuf_data(&key), ofpbuf_size(&key), NULL, 0, NULL, 0, NULL);
1096         if (error && error != EEXIST) {
1097             if (error != EINVAL) {
1098                 VLOG_WARN("%s: MPLS stack length feature probe failed (%s)",
1099                           dpif_name(backer->dpif), ovs_strerror(error));
1100             }
1101             break;
1102         }
1103
1104         error = dpif_flow_del(backer->dpif, ofpbuf_data(&key), ofpbuf_size(&key), NULL);
1105         if (error) {
1106             VLOG_WARN("%s: failed to delete MPLS feature probe flow",
1107                       dpif_name(backer->dpif));
1108         }
1109     }
1110
1111     VLOG_INFO("%s: MPLS label stack length probed as %d",
1112               dpif_name(backer->dpif), n);
1113     return n;
1114 }
1115
1116 static int
1117 construct(struct ofproto *ofproto_)
1118 {
1119     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1120     struct shash_node *node, *next;
1121     int error;
1122
1123     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1124     if (error) {
1125         return error;
1126     }
1127
1128     ofproto->netflow = NULL;
1129     ofproto->sflow = NULL;
1130     ofproto->ipfix = NULL;
1131     ofproto->stp = NULL;
1132     ofproto->dump_seq = 0;
1133     hmap_init(&ofproto->bundles);
1134     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1135     ofproto->ms = NULL;
1136     ofproto->mbridge = mbridge_create();
1137     ofproto->has_bonded_bundles = false;
1138     ofproto->lacp_enabled = false;
1139     ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1140     ovs_mutex_init(&ofproto->vsp_mutex);
1141
1142     guarded_list_init(&ofproto->pins);
1143
1144     ofproto_dpif_unixctl_init();
1145
1146     hmap_init(&ofproto->vlandev_map);
1147     hmap_init(&ofproto->realdev_vid_map);
1148
1149     sset_init(&ofproto->ports);
1150     sset_init(&ofproto->ghost_ports);
1151     sset_init(&ofproto->port_poll_set);
1152     ofproto->port_poll_errno = 0;
1153     ofproto->change_seq = 0;
1154     ofproto->pins_seq = seq_create();
1155     ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1156
1157
1158     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1159         struct iface_hint *iface_hint = node->data;
1160
1161         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1162             /* Check if the datapath already has this port. */
1163             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1164                 sset_add(&ofproto->ports, node->name);
1165             }
1166
1167             free(iface_hint->br_name);
1168             free(iface_hint->br_type);
1169             free(iface_hint);
1170             shash_delete(&init_ofp_ports, node);
1171         }
1172     }
1173
1174     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1175                 hash_string(ofproto->up.name, 0));
1176     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1177
1178     ofproto_init_tables(ofproto_, N_TABLES);
1179     error = add_internal_flows(ofproto);
1180
1181     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1182
1183     return error;
1184 }
1185
1186 static int
1187 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1188                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1189 {
1190     struct match match;
1191     int error;
1192     struct rule *rule;
1193
1194     match_init_catchall(&match);
1195     match_set_reg(&match, 0, id);
1196
1197     error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, 0, ofpacts,
1198                                            &rule);
1199     *rulep = error ? NULL : rule_dpif_cast(rule);
1200
1201     return error;
1202 }
1203
1204 static int
1205 add_internal_flows(struct ofproto_dpif *ofproto)
1206 {
1207     struct ofpact_controller *controller;
1208     uint64_t ofpacts_stub[128 / 8];
1209     struct ofpbuf ofpacts;
1210     struct rule *unused_rulep OVS_UNUSED;
1211     struct ofpact_resubmit *resubmit;
1212     struct match match;
1213     int error;
1214     int id;
1215
1216     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1217     id = 1;
1218
1219     controller = ofpact_put_CONTROLLER(&ofpacts);
1220     controller->max_len = UINT16_MAX;
1221     controller->controller_id = 0;
1222     controller->reason = OFPR_NO_MATCH;
1223     ofpact_pad(&ofpacts);
1224
1225     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1226                                    &ofproto->miss_rule);
1227     if (error) {
1228         return error;
1229     }
1230
1231     ofpbuf_clear(&ofpacts);
1232     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1233                                    &ofproto->no_packet_in_rule);
1234     if (error) {
1235         return error;
1236     }
1237
1238     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1239                                    &ofproto->drop_frags_rule);
1240     if (error) {
1241         return error;
1242     }
1243
1244     /* Continue non-recirculation rule lookups from table 0.
1245      *
1246      * (priority=2), recirc=0, actions=resubmit(, 0)
1247      */
1248     resubmit = ofpact_put_RESUBMIT(&ofpacts);
1249     resubmit->ofpact.compat = 0;
1250     resubmit->in_port = OFPP_IN_PORT;
1251     resubmit->table_id = 0;
1252
1253     match_init_catchall(&match);
1254     match_set_recirc_id(&match, 0);
1255
1256     error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, 0, &ofpacts,
1257                                            &unused_rulep);
1258     if (error) {
1259         return error;
1260     }
1261
1262     /* Drop any run away recirc rule lookups. Recirc_id has to be
1263      * non-zero when reaching this rule.
1264      *
1265      * (priority=1), *, actions=drop
1266      */
1267     ofpbuf_clear(&ofpacts);
1268     match_init_catchall(&match);
1269     error = ofproto_dpif_add_internal_flow(ofproto, &match, 1, 0, &ofpacts,
1270                                            &unused_rulep);
1271
1272     return error;
1273 }
1274
1275 static void
1276 destruct(struct ofproto *ofproto_)
1277 {
1278     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1279     struct ofproto_packet_in *pin, *next_pin;
1280     struct rule_dpif *rule;
1281     struct oftable *table;
1282     struct list pins;
1283
1284     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1285     xlate_txn_start();
1286     xlate_remove_ofproto(ofproto);
1287     xlate_txn_commit();
1288
1289     /* Ensure that the upcall processing threads have no remaining references
1290      * to the ofproto or anything in it. */
1291     udpif_synchronize(ofproto->backer->udpif);
1292
1293     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1294
1295     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1296         CLS_FOR_EACH_SAFE (rule, up.cr, &table->cls) {
1297             ofproto_rule_delete(&ofproto->up, &rule->up);
1298         }
1299     }
1300
1301     guarded_list_pop_all(&ofproto->pins, &pins);
1302     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1303         list_remove(&pin->list_node);
1304         free(CONST_CAST(void *, pin->up.packet));
1305         free(pin);
1306     }
1307     guarded_list_destroy(&ofproto->pins);
1308
1309     mbridge_unref(ofproto->mbridge);
1310
1311     netflow_unref(ofproto->netflow);
1312     dpif_sflow_unref(ofproto->sflow);
1313     hmap_destroy(&ofproto->bundles);
1314     mac_learning_unref(ofproto->ml);
1315     mcast_snooping_unref(ofproto->ms);
1316
1317     hmap_destroy(&ofproto->vlandev_map);
1318     hmap_destroy(&ofproto->realdev_vid_map);
1319
1320     sset_destroy(&ofproto->ports);
1321     sset_destroy(&ofproto->ghost_ports);
1322     sset_destroy(&ofproto->port_poll_set);
1323
1324     ovs_mutex_destroy(&ofproto->stats_mutex);
1325     ovs_mutex_destroy(&ofproto->vsp_mutex);
1326
1327     seq_destroy(ofproto->pins_seq);
1328
1329     close_dpif_backer(ofproto->backer);
1330 }
1331
1332 static int
1333 run(struct ofproto *ofproto_)
1334 {
1335     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1336     uint64_t new_seq, new_dump_seq;
1337
1338     if (mbridge_need_revalidate(ofproto->mbridge)) {
1339         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1340         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1341         mac_learning_flush(ofproto->ml);
1342         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1343         mcast_snooping_mdb_flush(ofproto->ms);
1344     }
1345
1346     /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
1347      * flow restore.  Even though nothing is processed during flow restore,
1348      * all queued 'pins' will be handled immediately when flow restore
1349      * completes. */
1350     ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1351
1352     /* Do not perform any periodic activity required by 'ofproto' while
1353      * waiting for flow restore to complete. */
1354     if (!ofproto_get_flow_restore_wait()) {
1355         struct ofproto_packet_in *pin, *next_pin;
1356         struct list pins;
1357
1358         guarded_list_pop_all(&ofproto->pins, &pins);
1359         LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1360             connmgr_send_packet_in(ofproto->up.connmgr, pin);
1361             list_remove(&pin->list_node);
1362             free(CONST_CAST(void *, pin->up.packet));
1363             free(pin);
1364         }
1365     }
1366
1367     if (ofproto->netflow) {
1368         netflow_run(ofproto->netflow);
1369     }
1370     if (ofproto->sflow) {
1371         dpif_sflow_run(ofproto->sflow);
1372     }
1373     if (ofproto->ipfix) {
1374         dpif_ipfix_run(ofproto->ipfix);
1375     }
1376
1377     new_seq = seq_read(connectivity_seq_get());
1378     if (ofproto->change_seq != new_seq) {
1379         struct ofport_dpif *ofport;
1380
1381         HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1382             port_run(ofport);
1383         }
1384
1385         ofproto->change_seq = new_seq;
1386     }
1387     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1388         struct ofbundle *bundle;
1389
1390         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1391             bundle_run(bundle);
1392         }
1393     }
1394
1395     stp_run(ofproto);
1396     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1397     if (mac_learning_run(ofproto->ml)) {
1398         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1399     }
1400     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1401
1402     if (mcast_snooping_run(ofproto->ms)) {
1403         ofproto->backer->need_revalidate = REV_MCAST_SNOOPING;
1404     }
1405
1406     new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1407     if (ofproto->dump_seq != new_dump_seq) {
1408         struct rule *rule, *next_rule;
1409
1410         /* We know stats are relatively fresh, so now is a good time to do some
1411          * periodic work. */
1412         ofproto->dump_seq = new_dump_seq;
1413
1414         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1415          * has passed. */
1416         ovs_mutex_lock(&ofproto_mutex);
1417         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1418                             &ofproto->up.expirable) {
1419             rule_expire(rule_dpif_cast(rule));
1420         }
1421         ovs_mutex_unlock(&ofproto_mutex);
1422
1423         /* All outstanding data in existing flows has been accounted, so it's a
1424          * good time to do bond rebalancing. */
1425         if (ofproto->has_bonded_bundles) {
1426             struct ofbundle *bundle;
1427
1428             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1429                 if (bundle->bond) {
1430                     bond_rebalance(bundle->bond);
1431                 }
1432             }
1433         }
1434     }
1435
1436     return 0;
1437 }
1438
1439 static void
1440 wait(struct ofproto *ofproto_)
1441 {
1442     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1443
1444     if (ofproto_get_flow_restore_wait()) {
1445         return;
1446     }
1447
1448     if (ofproto->sflow) {
1449         dpif_sflow_wait(ofproto->sflow);
1450     }
1451     if (ofproto->ipfix) {
1452         dpif_ipfix_wait(ofproto->ipfix);
1453     }
1454     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1455         struct ofbundle *bundle;
1456
1457         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1458             bundle_wait(bundle);
1459         }
1460     }
1461     if (ofproto->netflow) {
1462         netflow_wait(ofproto->netflow);
1463     }
1464     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1465     mac_learning_wait(ofproto->ml);
1466     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1467     mcast_snooping_wait(ofproto->ms);
1468     stp_wait(ofproto);
1469     if (ofproto->backer->need_revalidate) {
1470         /* Shouldn't happen, but if it does just go around again. */
1471         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1472         poll_immediate_wake();
1473     }
1474
1475     seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1476     seq_wait(ofproto->pins_seq, ofproto->pins_seqno);
1477 }
1478
1479 static void
1480 type_get_memory_usage(const char *type, struct simap *usage)
1481 {
1482     struct dpif_backer *backer;
1483
1484     backer = shash_find_data(&all_dpif_backers, type);
1485     if (backer) {
1486         udpif_get_memory_usage(backer->udpif, usage);
1487     }
1488 }
1489
1490 static void
1491 flush(struct ofproto *ofproto_)
1492 {
1493     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1494     struct dpif_backer *backer = ofproto->backer;
1495
1496     if (backer) {
1497         udpif_flush(backer->udpif);
1498     }
1499 }
1500
1501 static void
1502 query_tables(struct ofproto *ofproto,
1503              struct ofputil_table_features *features,
1504              struct ofputil_table_stats *stats)
1505 {
1506     strcpy(features->name, "classifier");
1507
1508     if (stats) {
1509         int i;
1510
1511         for (i = 0; i < ofproto->n_tables; i++) {
1512             unsigned long missed, matched;
1513
1514             atomic_read(&ofproto->tables[i].n_matched, &matched);
1515             stats[i].matched_count = matched;
1516             atomic_read(&ofproto->tables[i].n_missed, &missed);
1517             stats[i].lookup_count = matched + missed;
1518         }
1519     }
1520 }
1521
1522 static struct ofport *
1523 port_alloc(void)
1524 {
1525     struct ofport_dpif *port = xmalloc(sizeof *port);
1526     return &port->up;
1527 }
1528
1529 static void
1530 port_dealloc(struct ofport *port_)
1531 {
1532     struct ofport_dpif *port = ofport_dpif_cast(port_);
1533     free(port);
1534 }
1535
1536 static int
1537 port_construct(struct ofport *port_)
1538 {
1539     struct ofport_dpif *port = ofport_dpif_cast(port_);
1540     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1541     const struct netdev *netdev = port->up.netdev;
1542     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1543     struct dpif_port dpif_port;
1544     int error;
1545
1546     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1547     port->bundle = NULL;
1548     port->cfm = NULL;
1549     port->bfd = NULL;
1550     port->may_enable = true;
1551     port->stp_port = NULL;
1552     port->stp_state = STP_DISABLED;
1553     port->is_tunnel = false;
1554     port->peer = NULL;
1555     port->qdscp = NULL;
1556     port->n_qdscp = 0;
1557     port->realdev_ofp_port = 0;
1558     port->vlandev_vid = 0;
1559     port->carrier_seq = netdev_get_carrier_resets(netdev);
1560     port->is_layer3 = netdev_vport_is_layer3(netdev);
1561
1562     if (netdev_vport_is_patch(netdev)) {
1563         /* By bailing out here, we don't submit the port to the sFlow module
1564          * to be considered for counter polling export.  This is correct
1565          * because the patch port represents an interface that sFlow considers
1566          * to be "internal" to the switch as a whole, and therefore not an
1567          * candidate for counter polling. */
1568         port->odp_port = ODPP_NONE;
1569         ofport_update_peer(port);
1570         return 0;
1571     }
1572
1573     error = dpif_port_query_by_name(ofproto->backer->dpif,
1574                                     netdev_vport_get_dpif_port(netdev, namebuf,
1575                                                                sizeof namebuf),
1576                                     &dpif_port);
1577     if (error) {
1578         return error;
1579     }
1580
1581     port->odp_port = dpif_port.port_no;
1582
1583     if (netdev_get_tunnel_config(netdev)) {
1584         tnl_port_add(port, port->up.netdev, port->odp_port);
1585         port->is_tunnel = true;
1586     } else {
1587         /* Sanity-check that a mapping doesn't already exist.  This
1588          * shouldn't happen for non-tunnel ports. */
1589         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1590             VLOG_ERR("port %s already has an OpenFlow port number",
1591                      dpif_port.name);
1592             dpif_port_destroy(&dpif_port);
1593             return EBUSY;
1594         }
1595
1596         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1597         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1598                     hash_odp_port(port->odp_port));
1599         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1600     }
1601     dpif_port_destroy(&dpif_port);
1602
1603     if (ofproto->sflow) {
1604         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1605     }
1606
1607     return 0;
1608 }
1609
1610 static void
1611 port_destruct(struct ofport *port_)
1612 {
1613     struct ofport_dpif *port = ofport_dpif_cast(port_);
1614     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1615     const char *devname = netdev_get_name(port->up.netdev);
1616     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1617     const char *dp_port_name;
1618
1619     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1620     xlate_txn_start();
1621     xlate_ofport_remove(port);
1622     xlate_txn_commit();
1623
1624     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1625                                               sizeof namebuf);
1626     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1627         /* The underlying device is still there, so delete it.  This
1628          * happens when the ofproto is being destroyed, since the caller
1629          * assumes that removal of attached ports will happen as part of
1630          * destruction. */
1631         if (!port->is_tunnel) {
1632             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1633         }
1634     }
1635
1636     if (port->peer) {
1637         port->peer->peer = NULL;
1638         port->peer = NULL;
1639     }
1640
1641     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1642         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1643         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1644         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1645     }
1646
1647     tnl_port_del(port);
1648     sset_find_and_delete(&ofproto->ports, devname);
1649     sset_find_and_delete(&ofproto->ghost_ports, devname);
1650     bundle_remove(port_);
1651     set_cfm(port_, NULL);
1652     set_bfd(port_, NULL);
1653     if (port->stp_port) {
1654         stp_port_disable(port->stp_port);
1655     }
1656     if (ofproto->sflow) {
1657         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1658     }
1659
1660     free(port->qdscp);
1661 }
1662
1663 static void
1664 port_modified(struct ofport *port_)
1665 {
1666     struct ofport_dpif *port = ofport_dpif_cast(port_);
1667
1668     if (port->bundle && port->bundle->bond) {
1669         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1670     }
1671
1672     if (port->cfm) {
1673         cfm_set_netdev(port->cfm, port->up.netdev);
1674     }
1675
1676     if (port->bfd) {
1677         bfd_set_netdev(port->bfd, port->up.netdev);
1678     }
1679
1680     ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1681                                      port->up.pp.hw_addr);
1682
1683     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1684                                                 port->odp_port)) {
1685         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1686             REV_RECONFIGURE;
1687     }
1688
1689     ofport_update_peer(port);
1690 }
1691
1692 static void
1693 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1694 {
1695     struct ofport_dpif *port = ofport_dpif_cast(port_);
1696     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1697     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1698
1699     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1700                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1701                    OFPUTIL_PC_NO_PACKET_IN)) {
1702         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1703
1704         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1705             bundle_update(port->bundle);
1706         }
1707     }
1708 }
1709
1710 static int
1711 set_sflow(struct ofproto *ofproto_,
1712           const struct ofproto_sflow_options *sflow_options)
1713 {
1714     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1715     struct dpif_sflow *ds = ofproto->sflow;
1716
1717     if (sflow_options) {
1718         if (!ds) {
1719             struct ofport_dpif *ofport;
1720
1721             ds = ofproto->sflow = dpif_sflow_create();
1722             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1723                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1724             }
1725             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1726         }
1727         dpif_sflow_set_options(ds, sflow_options);
1728     } else {
1729         if (ds) {
1730             dpif_sflow_unref(ds);
1731             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1732             ofproto->sflow = NULL;
1733         }
1734     }
1735     return 0;
1736 }
1737
1738 static int
1739 set_ipfix(
1740     struct ofproto *ofproto_,
1741     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1742     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1743     size_t n_flow_exporters_options)
1744 {
1745     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1746     struct dpif_ipfix *di = ofproto->ipfix;
1747     bool has_options = bridge_exporter_options || flow_exporters_options;
1748
1749     if (has_options && !di) {
1750         di = ofproto->ipfix = dpif_ipfix_create();
1751     }
1752
1753     if (di) {
1754         /* Call set_options in any case to cleanly flush the flow
1755          * caches in the last exporters that are to be destroyed. */
1756         dpif_ipfix_set_options(
1757             di, bridge_exporter_options, flow_exporters_options,
1758             n_flow_exporters_options);
1759
1760         if (!has_options) {
1761             dpif_ipfix_unref(di);
1762             ofproto->ipfix = NULL;
1763         }
1764     }
1765
1766     return 0;
1767 }
1768
1769 static int
1770 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1771 {
1772     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1773     int error = 0;
1774
1775     if (s) {
1776         if (!ofport->cfm) {
1777             struct ofproto_dpif *ofproto;
1778
1779             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1780             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1781             ofport->cfm = cfm_create(ofport->up.netdev);
1782         }
1783
1784         if (cfm_configure(ofport->cfm, s)) {
1785             error = 0;
1786             goto out;
1787         }
1788
1789         error = EINVAL;
1790     }
1791     cfm_unref(ofport->cfm);
1792     ofport->cfm = NULL;
1793 out:
1794     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1795                                      ofport->up.pp.hw_addr);
1796     return error;
1797 }
1798
1799 static bool
1800 cfm_status_changed(struct ofport *ofport_)
1801 {
1802     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1803
1804     return ofport->cfm ? cfm_check_status_change(ofport->cfm) : true;
1805 }
1806
1807 static int
1808 get_cfm_status(const struct ofport *ofport_,
1809                struct cfm_status *status)
1810 {
1811     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1812     int ret = 0;
1813
1814     if (ofport->cfm) {
1815         cfm_get_status(ofport->cfm, status);
1816     } else {
1817         ret = ENOENT;
1818     }
1819
1820     return ret;
1821 }
1822
1823 static int
1824 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1825 {
1826     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1827     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1828     struct bfd *old;
1829
1830     old = ofport->bfd;
1831     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1832                                 cfg, ofport->up.netdev);
1833     if (ofport->bfd != old) {
1834         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1835     }
1836     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1837                                      ofport->up.pp.hw_addr);
1838     return 0;
1839 }
1840
1841 static bool
1842 bfd_status_changed(struct ofport *ofport_)
1843 {
1844     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1845
1846     return ofport->bfd ? bfd_check_status_change(ofport->bfd) : true;
1847 }
1848
1849 static int
1850 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1851 {
1852     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1853     int ret = 0;
1854
1855     if (ofport->bfd) {
1856         bfd_get_status(ofport->bfd, smap);
1857     } else {
1858         ret = ENOENT;
1859     }
1860
1861     return ret;
1862 }
1863 \f
1864 /* Spanning Tree. */
1865
1866 static void
1867 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1868 {
1869     struct ofproto_dpif *ofproto = ofproto_;
1870     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1871     struct ofport_dpif *ofport;
1872
1873     ofport = stp_port_get_aux(sp);
1874     if (!ofport) {
1875         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1876                      ofproto->up.name, port_num);
1877     } else {
1878         struct eth_header *eth = ofpbuf_l2(pkt);
1879
1880         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1881         if (eth_addr_is_zero(eth->eth_src)) {
1882             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1883                          "with unknown MAC", ofproto->up.name, port_num);
1884         } else {
1885             ofproto_dpif_send_packet(ofport, pkt);
1886         }
1887     }
1888     ofpbuf_delete(pkt);
1889 }
1890
1891 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1892 static int
1893 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1894 {
1895     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1896
1897     /* Only revalidate flows if the configuration changed. */
1898     if (!s != !ofproto->stp) {
1899         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1900     }
1901
1902     if (s) {
1903         if (!ofproto->stp) {
1904             ofproto->stp = stp_create(ofproto_->name, s->system_id,
1905                                       send_bpdu_cb, ofproto);
1906             ofproto->stp_last_tick = time_msec();
1907         }
1908
1909         stp_set_bridge_id(ofproto->stp, s->system_id);
1910         stp_set_bridge_priority(ofproto->stp, s->priority);
1911         stp_set_hello_time(ofproto->stp, s->hello_time);
1912         stp_set_max_age(ofproto->stp, s->max_age);
1913         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1914     }  else {
1915         struct ofport *ofport;
1916
1917         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1918             set_stp_port(ofport, NULL);
1919         }
1920
1921         stp_unref(ofproto->stp);
1922         ofproto->stp = NULL;
1923     }
1924
1925     return 0;
1926 }
1927
1928 static int
1929 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1930 {
1931     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1932
1933     if (ofproto->stp) {
1934         s->enabled = true;
1935         s->bridge_id = stp_get_bridge_id(ofproto->stp);
1936         s->designated_root = stp_get_designated_root(ofproto->stp);
1937         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1938     } else {
1939         s->enabled = false;
1940     }
1941
1942     return 0;
1943 }
1944
1945 static void
1946 update_stp_port_state(struct ofport_dpif *ofport)
1947 {
1948     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1949     enum stp_state state;
1950
1951     /* Figure out new state. */
1952     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1953                              : STP_DISABLED;
1954
1955     /* Update state. */
1956     if (ofport->stp_state != state) {
1957         enum ofputil_port_state of_state;
1958         bool fwd_change;
1959
1960         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1961                     netdev_get_name(ofport->up.netdev),
1962                     stp_state_name(ofport->stp_state),
1963                     stp_state_name(state));
1964         if (stp_learn_in_state(ofport->stp_state)
1965                 != stp_learn_in_state(state)) {
1966             /* xxx Learning action flows should also be flushed. */
1967             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1968             mac_learning_flush(ofproto->ml);
1969             ovs_rwlock_unlock(&ofproto->ml->rwlock);
1970             mcast_snooping_mdb_flush(ofproto->ms);
1971         }
1972         fwd_change = stp_forward_in_state(ofport->stp_state)
1973                         != stp_forward_in_state(state);
1974
1975         ofproto->backer->need_revalidate = REV_STP;
1976         ofport->stp_state = state;
1977         ofport->stp_state_entered = time_msec();
1978
1979         if (fwd_change && ofport->bundle) {
1980             bundle_update(ofport->bundle);
1981         }
1982
1983         /* Update the STP state bits in the OpenFlow port description. */
1984         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1985         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1986                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1987                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1988                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
1989                      : 0);
1990         ofproto_port_set_state(&ofport->up, of_state);
1991     }
1992 }
1993
1994 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
1995  * caller is responsible for assigning STP port numbers and ensuring
1996  * there are no duplicates. */
1997 static int
1998 set_stp_port(struct ofport *ofport_,
1999              const struct ofproto_port_stp_settings *s)
2000 {
2001     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2002     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2003     struct stp_port *sp = ofport->stp_port;
2004
2005     if (!s || !s->enable) {
2006         if (sp) {
2007             ofport->stp_port = NULL;
2008             stp_port_disable(sp);
2009             update_stp_port_state(ofport);
2010         }
2011         return 0;
2012     } else if (sp && stp_port_no(sp) != s->port_num
2013             && ofport == stp_port_get_aux(sp)) {
2014         /* The port-id changed, so disable the old one if it's not
2015          * already in use by another port. */
2016         stp_port_disable(sp);
2017     }
2018
2019     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2020     stp_port_enable(sp);
2021
2022     stp_port_set_name(sp, netdev_get_name(ofport->up.netdev));
2023     stp_port_set_aux(sp, ofport);
2024     stp_port_set_priority(sp, s->priority);
2025     stp_port_set_path_cost(sp, s->path_cost);
2026
2027     update_stp_port_state(ofport);
2028
2029     return 0;
2030 }
2031
2032 static int
2033 get_stp_port_status(struct ofport *ofport_,
2034                     struct ofproto_port_stp_status *s)
2035 {
2036     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2037     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2038     struct stp_port *sp = ofport->stp_port;
2039
2040     if (!ofproto->stp || !sp) {
2041         s->enabled = false;
2042         return 0;
2043     }
2044
2045     s->enabled = true;
2046     s->port_id = stp_port_get_id(sp);
2047     s->state = stp_port_get_state(sp);
2048     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2049     s->role = stp_port_get_role(sp);
2050
2051     return 0;
2052 }
2053
2054 static int
2055 get_stp_port_stats(struct ofport *ofport_,
2056                    struct ofproto_port_stp_stats *s)
2057 {
2058     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2059     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2060     struct stp_port *sp = ofport->stp_port;
2061
2062     if (!ofproto->stp || !sp) {
2063         s->enabled = false;
2064         return 0;
2065     }
2066
2067     s->enabled = true;
2068     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2069
2070     return 0;
2071 }
2072
2073 static void
2074 stp_run(struct ofproto_dpif *ofproto)
2075 {
2076     if (ofproto->stp) {
2077         long long int now = time_msec();
2078         long long int elapsed = now - ofproto->stp_last_tick;
2079         struct stp_port *sp;
2080
2081         if (elapsed > 0) {
2082             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2083             ofproto->stp_last_tick = now;
2084         }
2085         while (stp_get_changed_port(ofproto->stp, &sp)) {
2086             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2087
2088             if (ofport) {
2089                 update_stp_port_state(ofport);
2090             }
2091         }
2092
2093         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2094             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2095             mac_learning_flush(ofproto->ml);
2096             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2097             mcast_snooping_mdb_flush(ofproto->ms);
2098         }
2099     }
2100 }
2101
2102 static void
2103 stp_wait(struct ofproto_dpif *ofproto)
2104 {
2105     if (ofproto->stp) {
2106         poll_timer_wait(1000);
2107     }
2108 }
2109 \f
2110 static int
2111 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2112            size_t n_qdscp)
2113 {
2114     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2115     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2116
2117     if (ofport->n_qdscp != n_qdscp
2118         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2119                               n_qdscp * sizeof *qdscp))) {
2120         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2121         free(ofport->qdscp);
2122         ofport->qdscp = n_qdscp
2123             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2124             : NULL;
2125         ofport->n_qdscp = n_qdscp;
2126     }
2127
2128     return 0;
2129 }
2130 \f
2131 /* Bundles. */
2132
2133 /* Expires all MAC learning entries associated with 'bundle' and forces its
2134  * ofproto to revalidate every flow.
2135  *
2136  * Normally MAC learning entries are removed only from the ofproto associated
2137  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2138  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2139  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2140  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2141  * with the host from which it migrated. */
2142 static void
2143 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2144 {
2145     struct ofproto_dpif *ofproto = bundle->ofproto;
2146     struct mac_learning *ml = ofproto->ml;
2147     struct mac_entry *mac, *next_mac;
2148
2149     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2150     ovs_rwlock_wrlock(&ml->rwlock);
2151     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2152         if (mac->port.p == bundle) {
2153             if (all_ofprotos) {
2154                 struct ofproto_dpif *o;
2155
2156                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2157                     if (o != ofproto) {
2158                         struct mac_entry *e;
2159
2160                         ovs_rwlock_wrlock(&o->ml->rwlock);
2161                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2162                         if (e) {
2163                             mac_learning_expire(o->ml, e);
2164                         }
2165                         ovs_rwlock_unlock(&o->ml->rwlock);
2166                     }
2167                 }
2168             }
2169
2170             mac_learning_expire(ml, mac);
2171         }
2172     }
2173     ovs_rwlock_unlock(&ml->rwlock);
2174 }
2175
2176 static struct ofbundle *
2177 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2178 {
2179     struct ofbundle *bundle;
2180
2181     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2182                              &ofproto->bundles) {
2183         if (bundle->aux == aux) {
2184             return bundle;
2185         }
2186     }
2187     return NULL;
2188 }
2189
2190 static void
2191 bundle_update(struct ofbundle *bundle)
2192 {
2193     struct ofport_dpif *port;
2194
2195     bundle->floodable = true;
2196     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2197         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2198             || port->is_layer3
2199             || !stp_forward_in_state(port->stp_state)) {
2200             bundle->floodable = false;
2201             break;
2202         }
2203     }
2204 }
2205
2206 static void
2207 bundle_del_port(struct ofport_dpif *port)
2208 {
2209     struct ofbundle *bundle = port->bundle;
2210
2211     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2212
2213     list_remove(&port->bundle_node);
2214     port->bundle = NULL;
2215
2216     if (bundle->lacp) {
2217         lacp_slave_unregister(bundle->lacp, port);
2218     }
2219     if (bundle->bond) {
2220         bond_slave_unregister(bundle->bond, port);
2221     }
2222
2223     bundle_update(bundle);
2224 }
2225
2226 static bool
2227 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2228                 struct lacp_slave_settings *lacp)
2229 {
2230     struct ofport_dpif *port;
2231
2232     port = get_ofp_port(bundle->ofproto, ofp_port);
2233     if (!port) {
2234         return false;
2235     }
2236
2237     if (port->bundle != bundle) {
2238         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2239         if (port->bundle) {
2240             bundle_remove(&port->up);
2241         }
2242
2243         port->bundle = bundle;
2244         list_push_back(&bundle->ports, &port->bundle_node);
2245         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2246             || port->is_layer3
2247             || !stp_forward_in_state(port->stp_state)) {
2248             bundle->floodable = false;
2249         }
2250     }
2251     if (lacp) {
2252         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2253         lacp_slave_register(bundle->lacp, port, lacp);
2254     }
2255
2256     return true;
2257 }
2258
2259 static void
2260 bundle_destroy(struct ofbundle *bundle)
2261 {
2262     struct ofproto_dpif *ofproto;
2263     struct ofport_dpif *port, *next_port;
2264
2265     if (!bundle) {
2266         return;
2267     }
2268
2269     ofproto = bundle->ofproto;
2270     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2271
2272     xlate_txn_start();
2273     xlate_bundle_remove(bundle);
2274     xlate_txn_commit();
2275
2276     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2277         bundle_del_port(port);
2278     }
2279
2280     bundle_flush_macs(bundle, true);
2281     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2282     free(bundle->name);
2283     free(bundle->trunks);
2284     lacp_unref(bundle->lacp);
2285     bond_unref(bundle->bond);
2286     free(bundle);
2287 }
2288
2289 static int
2290 bundle_set(struct ofproto *ofproto_, void *aux,
2291            const struct ofproto_bundle_settings *s)
2292 {
2293     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2294     bool need_flush = false;
2295     struct ofport_dpif *port;
2296     struct ofbundle *bundle;
2297     unsigned long *trunks;
2298     int vlan;
2299     size_t i;
2300     bool ok;
2301
2302     if (!s) {
2303         bundle_destroy(bundle_lookup(ofproto, aux));
2304         return 0;
2305     }
2306
2307     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2308     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2309
2310     bundle = bundle_lookup(ofproto, aux);
2311     if (!bundle) {
2312         bundle = xmalloc(sizeof *bundle);
2313
2314         bundle->ofproto = ofproto;
2315         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2316                     hash_pointer(aux, 0));
2317         bundle->aux = aux;
2318         bundle->name = NULL;
2319
2320         list_init(&bundle->ports);
2321         bundle->vlan_mode = PORT_VLAN_TRUNK;
2322         bundle->vlan = -1;
2323         bundle->trunks = NULL;
2324         bundle->use_priority_tags = s->use_priority_tags;
2325         bundle->lacp = NULL;
2326         bundle->bond = NULL;
2327
2328         bundle->floodable = true;
2329         mbridge_register_bundle(ofproto->mbridge, bundle);
2330     }
2331
2332     if (!bundle->name || strcmp(s->name, bundle->name)) {
2333         free(bundle->name);
2334         bundle->name = xstrdup(s->name);
2335     }
2336
2337     /* LACP. */
2338     if (s->lacp) {
2339         ofproto->lacp_enabled = true;
2340         if (!bundle->lacp) {
2341             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2342             bundle->lacp = lacp_create();
2343         }
2344         lacp_configure(bundle->lacp, s->lacp);
2345     } else {
2346         lacp_unref(bundle->lacp);
2347         bundle->lacp = NULL;
2348     }
2349
2350     /* Update set of ports. */
2351     ok = true;
2352     for (i = 0; i < s->n_slaves; i++) {
2353         if (!bundle_add_port(bundle, s->slaves[i],
2354                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2355             ok = false;
2356         }
2357     }
2358     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2359         struct ofport_dpif *next_port;
2360
2361         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2362             for (i = 0; i < s->n_slaves; i++) {
2363                 if (s->slaves[i] == port->up.ofp_port) {
2364                     goto found;
2365                 }
2366             }
2367
2368             bundle_del_port(port);
2369         found: ;
2370         }
2371     }
2372     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2373
2374     if (list_is_empty(&bundle->ports)) {
2375         bundle_destroy(bundle);
2376         return EINVAL;
2377     }
2378
2379     /* Set VLAN tagging mode */
2380     if (s->vlan_mode != bundle->vlan_mode
2381         || s->use_priority_tags != bundle->use_priority_tags) {
2382         bundle->vlan_mode = s->vlan_mode;
2383         bundle->use_priority_tags = s->use_priority_tags;
2384         need_flush = true;
2385     }
2386
2387     /* Set VLAN tag. */
2388     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2389             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2390             : 0);
2391     if (vlan != bundle->vlan) {
2392         bundle->vlan = vlan;
2393         need_flush = true;
2394     }
2395
2396     /* Get trunked VLANs. */
2397     switch (s->vlan_mode) {
2398     case PORT_VLAN_ACCESS:
2399         trunks = NULL;
2400         break;
2401
2402     case PORT_VLAN_TRUNK:
2403         trunks = CONST_CAST(unsigned long *, s->trunks);
2404         break;
2405
2406     case PORT_VLAN_NATIVE_UNTAGGED:
2407     case PORT_VLAN_NATIVE_TAGGED:
2408         if (vlan != 0 && (!s->trunks
2409                           || !bitmap_is_set(s->trunks, vlan)
2410                           || bitmap_is_set(s->trunks, 0))) {
2411             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2412             if (s->trunks) {
2413                 trunks = bitmap_clone(s->trunks, 4096);
2414             } else {
2415                 trunks = bitmap_allocate1(4096);
2416             }
2417             bitmap_set1(trunks, vlan);
2418             bitmap_set0(trunks, 0);
2419         } else {
2420             trunks = CONST_CAST(unsigned long *, s->trunks);
2421         }
2422         break;
2423
2424     default:
2425         OVS_NOT_REACHED();
2426     }
2427     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2428         free(bundle->trunks);
2429         if (trunks == s->trunks) {
2430             bundle->trunks = vlan_bitmap_clone(trunks);
2431         } else {
2432             bundle->trunks = trunks;
2433             trunks = NULL;
2434         }
2435         need_flush = true;
2436     }
2437     if (trunks != s->trunks) {
2438         free(trunks);
2439     }
2440
2441     /* Bonding. */
2442     if (!list_is_short(&bundle->ports)) {
2443         bundle->ofproto->has_bonded_bundles = true;
2444         if (bundle->bond) {
2445             if (bond_reconfigure(bundle->bond, s->bond)) {
2446                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2447             }
2448         } else {
2449             bundle->bond = bond_create(s->bond, ofproto);
2450             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2451         }
2452
2453         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2454             bond_slave_register(bundle->bond, port,
2455                                 port->up.ofp_port, port->up.netdev);
2456         }
2457     } else {
2458         bond_unref(bundle->bond);
2459         bundle->bond = NULL;
2460     }
2461
2462     /* If we changed something that would affect MAC learning, un-learn
2463      * everything on this port and force flow revalidation. */
2464     if (need_flush) {
2465         bundle_flush_macs(bundle, false);
2466     }
2467
2468     return 0;
2469 }
2470
2471 static void
2472 bundle_remove(struct ofport *port_)
2473 {
2474     struct ofport_dpif *port = ofport_dpif_cast(port_);
2475     struct ofbundle *bundle = port->bundle;
2476
2477     if (bundle) {
2478         bundle_del_port(port);
2479         if (list_is_empty(&bundle->ports)) {
2480             bundle_destroy(bundle);
2481         } else if (list_is_short(&bundle->ports)) {
2482             bond_unref(bundle->bond);
2483             bundle->bond = NULL;
2484         }
2485     }
2486 }
2487
2488 static void
2489 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2490 {
2491     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2492     struct ofport_dpif *port = port_;
2493     uint8_t ea[ETH_ADDR_LEN];
2494     int error;
2495
2496     error = netdev_get_etheraddr(port->up.netdev, ea);
2497     if (!error) {
2498         struct ofpbuf packet;
2499         void *packet_pdu;
2500
2501         ofpbuf_init(&packet, 0);
2502         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2503                                  pdu_size);
2504         memcpy(packet_pdu, pdu, pdu_size);
2505
2506         ofproto_dpif_send_packet(port, &packet);
2507         ofpbuf_uninit(&packet);
2508     } else {
2509         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2510                     "%s (%s)", port->bundle->name,
2511                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2512     }
2513 }
2514
2515 static void
2516 bundle_send_learning_packets(struct ofbundle *bundle)
2517 {
2518     struct ofproto_dpif *ofproto = bundle->ofproto;
2519     struct ofpbuf *learning_packet;
2520     int error, n_packets, n_errors;
2521     struct mac_entry *e;
2522     struct list packets;
2523
2524     list_init(&packets);
2525     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2526     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2527         if (e->port.p != bundle) {
2528             void *port_void;
2529
2530             learning_packet = bond_compose_learning_packet(bundle->bond,
2531                                                            e->mac, e->vlan,
2532                                                            &port_void);
2533             /* Temporarily use 'frame' as a private pointer (see below). */
2534             ovs_assert(learning_packet->frame == ofpbuf_data(learning_packet));
2535             learning_packet->frame = port_void;
2536             list_push_back(&packets, &learning_packet->list_node);
2537         }
2538     }
2539     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2540
2541     error = n_packets = n_errors = 0;
2542     LIST_FOR_EACH (learning_packet, list_node, &packets) {
2543         int ret;
2544         void *port_void = learning_packet->frame;
2545
2546         /* Restore 'frame'. */
2547         learning_packet->frame = ofpbuf_data(learning_packet);
2548         ret = ofproto_dpif_send_packet(port_void, learning_packet);
2549         if (ret) {
2550             error = ret;
2551             n_errors++;
2552         }
2553         n_packets++;
2554     }
2555     ofpbuf_list_delete(&packets);
2556
2557     if (n_errors) {
2558         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2559         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2560                      "packets, last error was: %s",
2561                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2562     } else {
2563         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2564                  bundle->name, n_packets);
2565     }
2566 }
2567
2568 static void
2569 bundle_run(struct ofbundle *bundle)
2570 {
2571     if (bundle->lacp) {
2572         lacp_run(bundle->lacp, send_pdu_cb);
2573     }
2574     if (bundle->bond) {
2575         struct ofport_dpif *port;
2576
2577         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2578             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2579         }
2580
2581         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2582             bundle->ofproto->backer->need_revalidate = REV_BOND;
2583         }
2584
2585         if (bond_should_send_learning_packets(bundle->bond)) {
2586             bundle_send_learning_packets(bundle);
2587         }
2588     }
2589 }
2590
2591 static void
2592 bundle_wait(struct ofbundle *bundle)
2593 {
2594     if (bundle->lacp) {
2595         lacp_wait(bundle->lacp);
2596     }
2597     if (bundle->bond) {
2598         bond_wait(bundle->bond);
2599     }
2600 }
2601 \f
2602 /* Mirrors. */
2603
2604 static int
2605 mirror_set__(struct ofproto *ofproto_, void *aux,
2606              const struct ofproto_mirror_settings *s)
2607 {
2608     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2609     struct ofbundle **srcs, **dsts;
2610     int error;
2611     size_t i;
2612
2613     if (!s) {
2614         mirror_destroy(ofproto->mbridge, aux);
2615         return 0;
2616     }
2617
2618     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2619     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2620
2621     for (i = 0; i < s->n_srcs; i++) {
2622         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2623     }
2624
2625     for (i = 0; i < s->n_dsts; i++) {
2626         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2627     }
2628
2629     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2630                        s->n_dsts, s->src_vlans,
2631                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2632     free(srcs);
2633     free(dsts);
2634     return error;
2635 }
2636
2637 static int
2638 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2639                    uint64_t *packets, uint64_t *bytes)
2640 {
2641     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2642                             bytes);
2643 }
2644
2645 static int
2646 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2647 {
2648     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2649     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2650     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2651         mac_learning_flush(ofproto->ml);
2652     }
2653     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2654     return 0;
2655 }
2656
2657 static bool
2658 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2659 {
2660     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2661     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2662     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2663 }
2664
2665 static void
2666 forward_bpdu_changed(struct ofproto *ofproto_)
2667 {
2668     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2669     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2670 }
2671
2672 static void
2673 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2674                      size_t max_entries)
2675 {
2676     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2677     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2678     mac_learning_set_idle_time(ofproto->ml, idle_time);
2679     mac_learning_set_max_entries(ofproto->ml, max_entries);
2680     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2681 }
2682
2683 /* Configures multicast snooping on 'ofport' using the settings
2684  * defined in 's'. */
2685 static int
2686 set_mcast_snooping(struct ofproto *ofproto_,
2687                    const struct ofproto_mcast_snooping_settings *s)
2688 {
2689     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2690
2691     /* Only revalidate flows if the configuration changed. */
2692     if (!s != !ofproto->ms) {
2693         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2694     }
2695
2696     if (s) {
2697         if (!ofproto->ms) {
2698             ofproto->ms = mcast_snooping_create();
2699         }
2700
2701         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
2702         mcast_snooping_set_idle_time(ofproto->ms, s->idle_time);
2703         mcast_snooping_set_max_entries(ofproto->ms, s->max_entries);
2704         if (mcast_snooping_set_flood_unreg(ofproto->ms, s->flood_unreg)) {
2705             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2706         }
2707         ovs_rwlock_unlock(&ofproto->ms->rwlock);
2708     } else {
2709         mcast_snooping_unref(ofproto->ms);
2710         ofproto->ms = NULL;
2711     }
2712
2713     return 0;
2714 }
2715
2716 /* Configures multicast snooping port's flood setting on 'ofproto'. */
2717 static int
2718 set_mcast_snooping_port(struct ofproto *ofproto_, void *aux, bool flood)
2719 {
2720     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2721     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2722
2723     if (ofproto->ms) {
2724         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
2725         mcast_snooping_set_port_flood(ofproto->ms, bundle->vlan, bundle,
2726                                       flood);
2727         ovs_rwlock_unlock(&ofproto->ms->rwlock);
2728     }
2729     return 0;
2730 }
2731
2732 \f
2733 /* Ports. */
2734
2735 static struct ofport_dpif *
2736 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2737 {
2738     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2739     return ofport ? ofport_dpif_cast(ofport) : NULL;
2740 }
2741
2742 static void
2743 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2744                             struct ofproto_port *ofproto_port,
2745                             struct dpif_port *dpif_port)
2746 {
2747     ofproto_port->name = dpif_port->name;
2748     ofproto_port->type = dpif_port->type;
2749     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2750 }
2751
2752 static void
2753 ofport_update_peer(struct ofport_dpif *ofport)
2754 {
2755     const struct ofproto_dpif *ofproto;
2756     struct dpif_backer *backer;
2757     char *peer_name;
2758
2759     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2760         return;
2761     }
2762
2763     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2764     backer->need_revalidate = REV_RECONFIGURE;
2765
2766     if (ofport->peer) {
2767         ofport->peer->peer = NULL;
2768         ofport->peer = NULL;
2769     }
2770
2771     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2772     if (!peer_name) {
2773         return;
2774     }
2775
2776     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2777         struct ofport *peer_ofport;
2778         struct ofport_dpif *peer;
2779         char *peer_peer;
2780
2781         if (ofproto->backer != backer) {
2782             continue;
2783         }
2784
2785         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2786         if (!peer_ofport) {
2787             continue;
2788         }
2789
2790         peer = ofport_dpif_cast(peer_ofport);
2791         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2792         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2793                                  peer_peer)) {
2794             ofport->peer = peer;
2795             ofport->peer->peer = ofport;
2796         }
2797         free(peer_peer);
2798
2799         break;
2800     }
2801     free(peer_name);
2802 }
2803
2804 static void
2805 port_run(struct ofport_dpif *ofport)
2806 {
2807     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2808     bool carrier_changed = carrier_seq != ofport->carrier_seq;
2809     bool enable = netdev_get_carrier(ofport->up.netdev);
2810     bool cfm_enable = false;
2811     bool bfd_enable = false;
2812
2813     ofport->carrier_seq = carrier_seq;
2814
2815     if (ofport->cfm) {
2816         int cfm_opup = cfm_get_opup(ofport->cfm);
2817
2818         cfm_enable = !cfm_get_fault(ofport->cfm);
2819
2820         if (cfm_opup >= 0) {
2821             cfm_enable = cfm_enable && cfm_opup;
2822         }
2823     }
2824
2825     if (ofport->bfd) {
2826         bfd_enable = bfd_forwarding(ofport->bfd);
2827     }
2828
2829     if (ofport->bfd || ofport->cfm) {
2830         enable = enable && (cfm_enable || bfd_enable);
2831     }
2832
2833     if (ofport->bundle) {
2834         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2835         if (carrier_changed) {
2836             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2837         }
2838     }
2839
2840     if (ofport->may_enable != enable) {
2841         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2842         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2843     }
2844
2845     ofport->may_enable = enable;
2846 }
2847
2848 static int
2849 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2850                    struct ofproto_port *ofproto_port)
2851 {
2852     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2853     struct dpif_port dpif_port;
2854     int error;
2855
2856     if (sset_contains(&ofproto->ghost_ports, devname)) {
2857         const char *type = netdev_get_type_from_name(devname);
2858
2859         /* We may be called before ofproto->up.port_by_name is populated with
2860          * the appropriate ofport.  For this reason, we must get the name and
2861          * type from the netdev layer directly. */
2862         if (type) {
2863             const struct ofport *ofport;
2864
2865             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2866             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2867             ofproto_port->name = xstrdup(devname);
2868             ofproto_port->type = xstrdup(type);
2869             return 0;
2870         }
2871         return ENODEV;
2872     }
2873
2874     if (!sset_contains(&ofproto->ports, devname)) {
2875         return ENODEV;
2876     }
2877     error = dpif_port_query_by_name(ofproto->backer->dpif,
2878                                     devname, &dpif_port);
2879     if (!error) {
2880         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2881     }
2882     return error;
2883 }
2884
2885 static int
2886 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2887 {
2888     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2889     const char *devname = netdev_get_name(netdev);
2890     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2891     const char *dp_port_name;
2892
2893     if (netdev_vport_is_patch(netdev)) {
2894         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2895         return 0;
2896     }
2897
2898     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2899     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2900         odp_port_t port_no = ODPP_NONE;
2901         int error;
2902
2903         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2904         if (error) {
2905             return error;
2906         }
2907         if (netdev_get_tunnel_config(netdev)) {
2908             simap_put(&ofproto->backer->tnl_backers,
2909                       dp_port_name, odp_to_u32(port_no));
2910         }
2911     }
2912
2913     if (netdev_get_tunnel_config(netdev)) {
2914         sset_add(&ofproto->ghost_ports, devname);
2915     } else {
2916         sset_add(&ofproto->ports, devname);
2917     }
2918     return 0;
2919 }
2920
2921 static int
2922 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
2923 {
2924     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2925     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2926     int error = 0;
2927
2928     if (!ofport) {
2929         return 0;
2930     }
2931
2932     sset_find_and_delete(&ofproto->ghost_ports,
2933                          netdev_get_name(ofport->up.netdev));
2934     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2935     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
2936         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
2937         if (!error) {
2938             /* The caller is going to close ofport->up.netdev.  If this is a
2939              * bonded port, then the bond is using that netdev, so remove it
2940              * from the bond.  The client will need to reconfigure everything
2941              * after deleting ports, so then the slave will get re-added. */
2942             bundle_remove(&ofport->up);
2943         }
2944     }
2945     return error;
2946 }
2947
2948 static int
2949 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2950 {
2951     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2952     int error;
2953
2954     error = netdev_get_stats(ofport->up.netdev, stats);
2955
2956     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
2957         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2958
2959         ovs_mutex_lock(&ofproto->stats_mutex);
2960         /* ofproto->stats.tx_packets represents packets that we created
2961          * internally and sent to some port (e.g. packets sent with
2962          * ofproto_dpif_send_packet()).  Account for them as if they had
2963          * come from OFPP_LOCAL and got forwarded. */
2964
2965         if (stats->rx_packets != UINT64_MAX) {
2966             stats->rx_packets += ofproto->stats.tx_packets;
2967         }
2968
2969         if (stats->rx_bytes != UINT64_MAX) {
2970             stats->rx_bytes += ofproto->stats.tx_bytes;
2971         }
2972
2973         /* ofproto->stats.rx_packets represents packets that were received on
2974          * some port and we processed internally and dropped (e.g. STP).
2975          * Account for them as if they had been forwarded to OFPP_LOCAL. */
2976
2977         if (stats->tx_packets != UINT64_MAX) {
2978             stats->tx_packets += ofproto->stats.rx_packets;
2979         }
2980
2981         if (stats->tx_bytes != UINT64_MAX) {
2982             stats->tx_bytes += ofproto->stats.rx_bytes;
2983         }
2984         ovs_mutex_unlock(&ofproto->stats_mutex);
2985     }
2986
2987     return error;
2988 }
2989
2990 struct port_dump_state {
2991     uint32_t bucket;
2992     uint32_t offset;
2993     bool ghost;
2994
2995     struct ofproto_port port;
2996     bool has_port;
2997 };
2998
2999 static int
3000 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3001 {
3002     *statep = xzalloc(sizeof(struct port_dump_state));
3003     return 0;
3004 }
3005
3006 static int
3007 port_dump_next(const struct ofproto *ofproto_, void *state_,
3008                struct ofproto_port *port)
3009 {
3010     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3011     struct port_dump_state *state = state_;
3012     const struct sset *sset;
3013     struct sset_node *node;
3014
3015     if (state->has_port) {
3016         ofproto_port_destroy(&state->port);
3017         state->has_port = false;
3018     }
3019     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3020     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3021         int error;
3022
3023         error = port_query_by_name(ofproto_, node->name, &state->port);
3024         if (!error) {
3025             *port = state->port;
3026             state->has_port = true;
3027             return 0;
3028         } else if (error != ENODEV) {
3029             return error;
3030         }
3031     }
3032
3033     if (!state->ghost) {
3034         state->ghost = true;
3035         state->bucket = 0;
3036         state->offset = 0;
3037         return port_dump_next(ofproto_, state_, port);
3038     }
3039
3040     return EOF;
3041 }
3042
3043 static int
3044 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3045 {
3046     struct port_dump_state *state = state_;
3047
3048     if (state->has_port) {
3049         ofproto_port_destroy(&state->port);
3050     }
3051     free(state);
3052     return 0;
3053 }
3054
3055 static int
3056 port_poll(const struct ofproto *ofproto_, char **devnamep)
3057 {
3058     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3059
3060     if (ofproto->port_poll_errno) {
3061         int error = ofproto->port_poll_errno;
3062         ofproto->port_poll_errno = 0;
3063         return error;
3064     }
3065
3066     if (sset_is_empty(&ofproto->port_poll_set)) {
3067         return EAGAIN;
3068     }
3069
3070     *devnamep = sset_pop(&ofproto->port_poll_set);
3071     return 0;
3072 }
3073
3074 static void
3075 port_poll_wait(const struct ofproto *ofproto_)
3076 {
3077     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3078     dpif_port_poll_wait(ofproto->backer->dpif);
3079 }
3080
3081 static int
3082 port_is_lacp_current(const struct ofport *ofport_)
3083 {
3084     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3085     return (ofport->bundle && ofport->bundle->lacp
3086             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3087             : -1);
3088 }
3089 \f
3090 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3091  * then delete it entirely. */
3092 static void
3093 rule_expire(struct rule_dpif *rule)
3094     OVS_REQUIRES(ofproto_mutex)
3095 {
3096     uint16_t hard_timeout, idle_timeout;
3097     long long int now = time_msec();
3098     int reason = -1;
3099
3100     hard_timeout = rule->up.hard_timeout;
3101     idle_timeout = rule->up.idle_timeout;
3102
3103     /* Has 'rule' expired? */
3104     if (hard_timeout) {
3105         long long int modified;
3106
3107         ovs_mutex_lock(&rule->up.mutex);
3108         modified = rule->up.modified;
3109         ovs_mutex_unlock(&rule->up.mutex);
3110
3111         if (now > modified + hard_timeout * 1000) {
3112             reason = OFPRR_HARD_TIMEOUT;
3113         }
3114     }
3115
3116     if (reason < 0 && idle_timeout) {
3117         long long int used;
3118
3119         ovs_mutex_lock(&rule->stats_mutex);
3120         used = rule->stats.used;
3121         ovs_mutex_unlock(&rule->stats_mutex);
3122
3123         if (now > used + idle_timeout * 1000) {
3124             reason = OFPRR_IDLE_TIMEOUT;
3125         }
3126     }
3127
3128     if (reason >= 0) {
3129         COVERAGE_INC(ofproto_dpif_expired);
3130         ofproto_rule_expire(&rule->up, reason);
3131     }
3132 }
3133
3134 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3135  * 'flow' must reflect the data in 'packet'. */
3136 int
3137 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3138                              const struct flow *flow,
3139                              struct rule_dpif *rule,
3140                              const struct ofpact *ofpacts, size_t ofpacts_len,
3141                              struct ofpbuf *packet)
3142 {
3143     struct dpif_flow_stats stats;
3144     struct xlate_out xout;
3145     struct xlate_in xin;
3146     ofp_port_t in_port;
3147     struct dpif_execute execute;
3148     int error;
3149
3150     ovs_assert((rule != NULL) != (ofpacts != NULL));
3151
3152     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3153
3154     if (rule) {
3155         rule_dpif_credit_stats(rule, &stats);
3156     }
3157
3158     xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
3159     xin.ofpacts = ofpacts;
3160     xin.ofpacts_len = ofpacts_len;
3161     xin.resubmit_stats = &stats;
3162     xlate_actions(&xin, &xout);
3163
3164     execute.actions = ofpbuf_data(&xout.odp_actions);
3165     execute.actions_len = ofpbuf_size(&xout.odp_actions);
3166     execute.packet = packet;
3167     execute.md = pkt_metadata_from_flow(flow);
3168     execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3169
3170     /* Fix up in_port. */
3171     in_port = flow->in_port.ofp_port;
3172     if (in_port == OFPP_NONE) {
3173         in_port = OFPP_LOCAL;
3174     }
3175     execute.md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3176
3177     error = dpif_execute(ofproto->backer->dpif, &execute);
3178
3179     xlate_out_uninit(&xout);
3180
3181     return error;
3182 }
3183
3184 void
3185 rule_dpif_credit_stats(struct rule_dpif *rule,
3186                        const struct dpif_flow_stats *stats)
3187 {
3188     ovs_mutex_lock(&rule->stats_mutex);
3189     rule->stats.n_packets += stats->n_packets;
3190     rule->stats.n_bytes += stats->n_bytes;
3191     rule->stats.used = MAX(rule->stats.used, stats->used);
3192     ovs_mutex_unlock(&rule->stats_mutex);
3193 }
3194
3195 ovs_be64
3196 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3197     OVS_REQUIRES(rule->up.mutex)
3198 {
3199     return rule->up.flow_cookie;
3200 }
3201
3202 void
3203 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3204                      uint16_t hard_timeout)
3205 {
3206     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3207 }
3208
3209 /* Returns 'rule''s actions.  The caller owns a reference on the returned
3210  * actions and must eventually release it (with rule_actions_unref()) to avoid
3211  * a memory leak. */
3212 const struct rule_actions *
3213 rule_dpif_get_actions(const struct rule_dpif *rule)
3214 {
3215     return rule_get_actions(&rule->up);
3216 }
3217
3218 /* Sets 'rule''s recirculation id. */
3219 static void
3220 rule_dpif_set_recirc_id(struct rule_dpif *rule, uint32_t id)
3221     OVS_REQUIRES(rule->up.mutex)
3222 {
3223     ovs_assert(!rule->recirc_id);
3224     rule->recirc_id = id;
3225 }
3226
3227 /* Returns 'rule''s recirculation id. */
3228 uint32_t
3229 rule_dpif_get_recirc_id(struct rule_dpif *rule)
3230     OVS_REQUIRES(rule->up.mutex)
3231 {
3232     if (!rule->recirc_id) {
3233         struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3234
3235         rule_dpif_set_recirc_id(rule, ofproto_dpif_alloc_recirc_id(ofproto));
3236     }
3237     return rule->recirc_id;
3238 }
3239
3240 /* Sets 'rule''s recirculation id. */
3241 void
3242 rule_set_recirc_id(struct rule *rule_, uint32_t id)
3243 {
3244     struct rule_dpif *rule = rule_dpif_cast(rule_);
3245
3246     ovs_mutex_lock(&rule->up.mutex);
3247     rule_dpif_set_recirc_id(rule, id);
3248     ovs_mutex_unlock(&rule->up.mutex);
3249 }
3250
3251 /* Lookup 'flow' in table 0 of 'ofproto''s classifier.
3252  * If 'wc' is non-null, sets the fields that were relevant as part of
3253  * the lookup. Returns the table_id where a match or miss occurred.
3254  *
3255  * The return value will be zero unless there was a miss and
3256  * OFPTC11_TABLE_MISS_CONTINUE is in effect for the sequence of tables
3257  * where misses occur.
3258  *
3259  * The rule is returned in '*rule', which is valid at least until the next
3260  * RCU quiescent period.  If the '*rule' needs to stay around longer,
3261  * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3262  * on it before this returns. */
3263 uint8_t
3264 rule_dpif_lookup(struct ofproto_dpif *ofproto, struct flow *flow,
3265                  struct flow_wildcards *wc, struct rule_dpif **rule,
3266                  bool take_ref, const struct dpif_flow_stats *stats)
3267 {
3268     enum rule_dpif_lookup_verdict verdict;
3269     enum ofputil_port_config config = 0;
3270     uint8_t table_id;
3271
3272     if (ofproto_dpif_get_enable_recirc(ofproto)) {
3273         /* Always exactly match recirc_id since datapath supports
3274          * recirculation.  */
3275         if (wc) {
3276             wc->masks.recirc_id = UINT32_MAX;
3277         }
3278
3279         /* Start looking up from internal table for post recirculation flows
3280          * or packets. We can also simply send all, including normal flows
3281          * or packets to the internal table. They will not match any post
3282          * recirculation rules except the 'catch all' rule that resubmit
3283          * them to table 0.
3284          *
3285          * As an optimization, we send normal flows and packets to table 0
3286          * directly, saving one table lookup.  */
3287         table_id = flow->recirc_id ? TBL_INTERNAL : 0;
3288     } else {
3289         table_id = 0;
3290     }
3291
3292     verdict = rule_dpif_lookup_from_table(ofproto, flow, wc, true,
3293                                           &table_id, rule, take_ref, stats);
3294
3295     switch (verdict) {
3296     case RULE_DPIF_LOOKUP_VERDICT_MATCH:
3297         return table_id;
3298     case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER: {
3299         struct ofport_dpif *port;
3300
3301         port = get_ofp_port(ofproto, flow->in_port.ofp_port);
3302         if (!port) {
3303             VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3304                          flow->in_port.ofp_port);
3305         }
3306         config = port ? port->up.pp.config : 0;
3307         break;
3308     }
3309     case RULE_DPIF_LOOKUP_VERDICT_DROP:
3310         config = OFPUTIL_PC_NO_PACKET_IN;
3311         break;
3312     case RULE_DPIF_LOOKUP_VERDICT_DEFAULT:
3313         if (!connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3314             config = OFPUTIL_PC_NO_PACKET_IN;
3315         }
3316         break;
3317     default:
3318         OVS_NOT_REACHED();
3319     }
3320
3321     choose_miss_rule(config, ofproto->miss_rule,
3322                      ofproto->no_packet_in_rule, rule, take_ref);
3323     return table_id;
3324 }
3325
3326 /* The returned rule is valid at least until the next RCU quiescent period.
3327  * If the '*rule' needs to stay around longer, a non-zero 'take_ref' must be
3328  * passed in to cause a reference to be taken on it before this returns. */
3329 static struct rule_dpif *
3330 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, uint8_t table_id,
3331                           const struct flow *flow, struct flow_wildcards *wc,
3332                           bool take_ref)
3333 {
3334     struct classifier *cls = &ofproto->up.tables[table_id].cls;
3335     const struct cls_rule *cls_rule;
3336     struct rule_dpif *rule;
3337     struct flow ofpc_normal_flow;
3338
3339     if (ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3340         /* We always unwildcard dl_type and nw_frag (for IP), so they
3341          * need not be unwildcarded here. */
3342
3343         if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
3344             if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3345                 /* We must pretend that transport ports are unavailable. */
3346                 ofpc_normal_flow = *flow;
3347                 ofpc_normal_flow.tp_src = htons(0);
3348                 ofpc_normal_flow.tp_dst = htons(0);
3349                 flow = &ofpc_normal_flow;
3350             } else {
3351                 /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM).
3352                  * Use the drop_frags_rule (which cannot disappear). */
3353                 cls_rule = &ofproto->drop_frags_rule->up.cr;
3354                 rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3355                 if (take_ref) {
3356                     rule_dpif_ref(rule);
3357                 }
3358                 return rule;
3359             }
3360         }
3361     }
3362
3363     do {
3364         cls_rule = classifier_lookup(cls, flow, wc);
3365
3366         rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3367
3368         /* Try again if the rule was released before we get the reference. */
3369     } while (rule && take_ref && !rule_dpif_try_ref(rule));
3370
3371     return rule;
3372 }
3373
3374 /* Look up 'flow' in 'ofproto''s classifier starting from table '*table_id'.
3375  * Stores the rule that was found in '*rule', or NULL if none was found.
3376  * Updates 'wc', if nonnull, to reflect the fields that were used during the
3377  * lookup.
3378  *
3379  * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3380  * if none is found then the table miss configuration for that table is
3381  * honored, which can result in additional lookups in other OpenFlow tables.
3382  * In this case the function updates '*table_id' to reflect the final OpenFlow
3383  * table that was searched.
3384  *
3385  * If 'honor_table_miss' is false, then only one table lookup occurs, in
3386  * '*table_id'.
3387  *
3388  * Returns:
3389  *
3390  *    - RULE_DPIF_LOOKUP_VERDICT_MATCH if a rule (in '*rule') was found.
3391  *
3392  *    - RULE_OFPTC_TABLE_MISS_CONTROLLER if no rule was found and either:
3393  *      + 'honor_table_miss' is false
3394  *      + a table miss configuration specified that the packet should be
3395  *        sent to the controller in this case.
3396  *
3397  *    - RULE_DPIF_LOOKUP_VERDICT_DROP if no rule was found, 'honor_table_miss'
3398  *      is true and a table miss configuration specified that the packet
3399  *      should be dropped in this case.
3400  *
3401  *    - RULE_DPIF_LOOKUP_VERDICT_DEFAULT if no rule was found,
3402  *      'honor_table_miss' is true and a table miss configuration has
3403  *      not been specified in this case.
3404  *
3405  * The rule is returned in '*rule', which is valid at least until the next
3406  * RCU quiescent period.  If the '*rule' needs to stay around longer,
3407  * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3408  * on it before this returns. */
3409 enum rule_dpif_lookup_verdict
3410 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3411                             const struct flow *flow,
3412                             struct flow_wildcards *wc,
3413                             bool honor_table_miss,
3414                             uint8_t *table_id, struct rule_dpif **rule,
3415                             bool take_ref, const struct dpif_flow_stats *stats)
3416 {
3417     uint8_t next_id;
3418
3419     for (next_id = *table_id;
3420          next_id < ofproto->up.n_tables;
3421          next_id++, next_id += (next_id == TBL_INTERNAL))
3422     {
3423         *table_id = next_id;
3424         *rule = rule_dpif_lookup_in_table(ofproto, *table_id, flow, wc,
3425                                           take_ref);
3426         if (stats) {
3427             struct oftable *tbl = &ofproto->up.tables[next_id];
3428             atomic_ulong *stat = *rule ? &tbl->n_matched : &tbl->n_missed;
3429             unsigned long orig;
3430             atomic_add(stat, stats->n_packets, &orig);
3431         }
3432         if (*rule) {
3433             return RULE_DPIF_LOOKUP_VERDICT_MATCH;
3434         } else if (!honor_table_miss) {
3435             return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3436         } else {
3437             switch (ofproto_table_get_miss_config(&ofproto->up, *table_id)) {
3438             case OFPUTIL_TABLE_MISS_CONTINUE:
3439                 break;
3440
3441             case OFPUTIL_TABLE_MISS_CONTROLLER:
3442                 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3443
3444             case OFPUTIL_TABLE_MISS_DROP:
3445                 return RULE_DPIF_LOOKUP_VERDICT_DROP;
3446
3447             case OFPUTIL_TABLE_MISS_DEFAULT:
3448                 return RULE_DPIF_LOOKUP_VERDICT_DEFAULT;
3449             }
3450         }
3451     }
3452
3453     return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3454 }
3455
3456 /* Given a port configuration (specified as zero if there's no port), chooses
3457  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
3458  * flow table miss.
3459  *
3460  * The rule is returned in '*rule', which is valid at least until the next
3461  * RCU quiescent period.  If the '*rule' needs to stay around longer,
3462  * a reference must be taken on it (rule_dpif_ref()).
3463  */
3464 void
3465 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
3466                  struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule,
3467                  bool take_ref)
3468 {
3469     *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
3470     if (take_ref) {
3471         rule_dpif_ref(*rule);
3472     }
3473 }
3474
3475 static void
3476 complete_operation(struct rule_dpif *rule)
3477     OVS_REQUIRES(ofproto_mutex)
3478 {
3479     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3480
3481     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3482 }
3483
3484 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3485 {
3486     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3487 }
3488
3489 static struct rule *
3490 rule_alloc(void)
3491 {
3492     struct rule_dpif *rule = xmalloc(sizeof *rule);
3493     return &rule->up;
3494 }
3495
3496 static void
3497 rule_dealloc(struct rule *rule_)
3498 {
3499     struct rule_dpif *rule = rule_dpif_cast(rule_);
3500     free(rule);
3501 }
3502
3503 static enum ofperr
3504 rule_construct(struct rule *rule_)
3505     OVS_NO_THREAD_SAFETY_ANALYSIS
3506 {
3507     struct rule_dpif *rule = rule_dpif_cast(rule_);
3508     ovs_mutex_init_adaptive(&rule->stats_mutex);
3509     rule->stats.n_packets = 0;
3510     rule->stats.n_bytes = 0;
3511     rule->stats.used = rule->up.modified;
3512     rule->recirc_id = 0;
3513
3514     return 0;
3515 }
3516
3517 static enum ofperr
3518 rule_insert(struct rule *rule_)
3519     OVS_REQUIRES(ofproto_mutex)
3520 {
3521     struct rule_dpif *rule = rule_dpif_cast(rule_);
3522     complete_operation(rule);
3523     return 0;
3524 }
3525
3526 static void
3527 rule_delete(struct rule *rule_)
3528     OVS_REQUIRES(ofproto_mutex)
3529 {
3530     struct rule_dpif *rule = rule_dpif_cast(rule_);
3531     complete_operation(rule);
3532 }
3533
3534 static void
3535 rule_destruct(struct rule *rule_)
3536 {
3537     struct rule_dpif *rule = rule_dpif_cast(rule_);
3538
3539     ovs_mutex_destroy(&rule->stats_mutex);
3540     if (rule->recirc_id) {
3541         struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3542
3543         ofproto_dpif_free_recirc_id(ofproto, rule->recirc_id);
3544     }
3545 }
3546
3547 static void
3548 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
3549                long long int *used)
3550 {
3551     struct rule_dpif *rule = rule_dpif_cast(rule_);
3552
3553     ovs_mutex_lock(&rule->stats_mutex);
3554     *packets = rule->stats.n_packets;
3555     *bytes = rule->stats.n_bytes;
3556     *used = rule->stats.used;
3557     ovs_mutex_unlock(&rule->stats_mutex);
3558 }
3559
3560 static void
3561 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
3562                   struct ofpbuf *packet)
3563 {
3564     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3565
3566     ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
3567 }
3568
3569 static enum ofperr
3570 rule_execute(struct rule *rule, const struct flow *flow,
3571              struct ofpbuf *packet)
3572 {
3573     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
3574     ofpbuf_delete(packet);
3575     return 0;
3576 }
3577
3578 static void
3579 rule_modify_actions(struct rule *rule_, bool reset_counters)
3580     OVS_REQUIRES(ofproto_mutex)
3581 {
3582     struct rule_dpif *rule = rule_dpif_cast(rule_);
3583
3584     if (reset_counters) {
3585         ovs_mutex_lock(&rule->stats_mutex);
3586         rule->stats.n_packets = 0;
3587         rule->stats.n_bytes = 0;
3588         ovs_mutex_unlock(&rule->stats_mutex);
3589     }
3590
3591     complete_operation(rule);
3592 }
3593
3594 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
3595 {
3596     return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
3597 }
3598
3599 static struct ofgroup *
3600 group_alloc(void)
3601 {
3602     struct group_dpif *group = xzalloc(sizeof *group);
3603     return &group->up;
3604 }
3605
3606 static void
3607 group_dealloc(struct ofgroup *group_)
3608 {
3609     struct group_dpif *group = group_dpif_cast(group_);
3610     free(group);
3611 }
3612
3613 static void
3614 group_construct_stats(struct group_dpif *group)
3615     OVS_REQUIRES(group->stats_mutex)
3616 {
3617     struct ofputil_bucket *bucket;
3618     const struct list *buckets;
3619
3620     group->packet_count = 0;
3621     group->byte_count = 0;
3622
3623     group_dpif_get_buckets(group, &buckets);
3624     LIST_FOR_EACH (bucket, list_node, buckets) {
3625         bucket->stats.packet_count = 0;
3626         bucket->stats.byte_count = 0;
3627     }
3628 }
3629
3630 void
3631 group_dpif_credit_stats(struct group_dpif *group,
3632                         struct ofputil_bucket *bucket,
3633                         const struct dpif_flow_stats *stats)
3634 {
3635     ovs_mutex_lock(&group->stats_mutex);
3636     group->packet_count += stats->n_packets;
3637     group->byte_count += stats->n_bytes;
3638     if (bucket) {
3639         bucket->stats.packet_count += stats->n_packets;
3640         bucket->stats.byte_count += stats->n_bytes;
3641     } else { /* Credit to all buckets */
3642         const struct list *buckets;
3643
3644         group_dpif_get_buckets(group, &buckets);
3645         LIST_FOR_EACH (bucket, list_node, buckets) {
3646             bucket->stats.packet_count += stats->n_packets;
3647             bucket->stats.byte_count += stats->n_bytes;
3648         }
3649     }
3650     ovs_mutex_unlock(&group->stats_mutex);
3651 }
3652
3653 static enum ofperr
3654 group_construct(struct ofgroup *group_)
3655 {
3656     struct group_dpif *group = group_dpif_cast(group_);
3657     const struct ofputil_bucket *bucket;
3658
3659     /* Prevent group chaining because our locking structure makes it hard to
3660      * implement deadlock-free.  (See xlate_group_resource_check().) */
3661     LIST_FOR_EACH (bucket, list_node, &group->up.buckets) {
3662         const struct ofpact *a;
3663
3664         OFPACT_FOR_EACH (a, bucket->ofpacts, bucket->ofpacts_len) {
3665             if (a->type == OFPACT_GROUP) {
3666                 return OFPERR_OFPGMFC_CHAINING_UNSUPPORTED;
3667             }
3668         }
3669     }
3670
3671     ovs_mutex_init_adaptive(&group->stats_mutex);
3672     ovs_mutex_lock(&group->stats_mutex);
3673     group_construct_stats(group);
3674     ovs_mutex_unlock(&group->stats_mutex);
3675     return 0;
3676 }
3677
3678 static void
3679 group_destruct(struct ofgroup *group_)
3680 {
3681     struct group_dpif *group = group_dpif_cast(group_);
3682     ovs_mutex_destroy(&group->stats_mutex);
3683 }
3684
3685 static enum ofperr
3686 group_modify(struct ofgroup *group_)
3687 {
3688     struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
3689
3690     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3691
3692     return 0;
3693 }
3694
3695 static enum ofperr
3696 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
3697 {
3698     struct group_dpif *group = group_dpif_cast(group_);
3699     struct ofputil_bucket *bucket;
3700     const struct list *buckets;
3701     struct bucket_counter *bucket_stats;
3702
3703     ovs_mutex_lock(&group->stats_mutex);
3704     ogs->packet_count = group->packet_count;
3705     ogs->byte_count = group->byte_count;
3706
3707     group_dpif_get_buckets(group, &buckets);
3708     bucket_stats = ogs->bucket_stats;
3709     LIST_FOR_EACH (bucket, list_node, buckets) {
3710         bucket_stats->packet_count = bucket->stats.packet_count;
3711         bucket_stats->byte_count = bucket->stats.byte_count;
3712         bucket_stats++;
3713     }
3714     ovs_mutex_unlock(&group->stats_mutex);
3715
3716     return 0;
3717 }
3718
3719 /* If the group exists, this function increments the groups's reference count.
3720  *
3721  * Make sure to call group_dpif_unref() after no longer needing to maintain
3722  * a reference to the group. */
3723 bool
3724 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
3725                   struct group_dpif **group)
3726 {
3727     struct ofgroup *ofgroup;
3728     bool found;
3729
3730     found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
3731     *group = found ?  group_dpif_cast(ofgroup) : NULL;
3732
3733     return found;
3734 }
3735
3736 void
3737 group_dpif_get_buckets(const struct group_dpif *group,
3738                        const struct list **buckets)
3739 {
3740     *buckets = &group->up.buckets;
3741 }
3742
3743 enum ofp11_group_type
3744 group_dpif_get_type(const struct group_dpif *group)
3745 {
3746     return group->up.type;
3747 }
3748 \f
3749 /* Sends 'packet' out 'ofport'.
3750  * May modify 'packet'.
3751  * Returns 0 if successful, otherwise a positive errno value. */
3752 int
3753 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3754 {
3755     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3756     int error;
3757
3758     error = xlate_send_packet(ofport, packet);
3759
3760     ovs_mutex_lock(&ofproto->stats_mutex);
3761     ofproto->stats.tx_packets++;
3762     ofproto->stats.tx_bytes += ofpbuf_size(packet);
3763     ovs_mutex_unlock(&ofproto->stats_mutex);
3764     return error;
3765 }
3766 \f
3767 static bool
3768 set_frag_handling(struct ofproto *ofproto_,
3769                   enum ofp_config_flags frag_handling)
3770 {
3771     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3772     if (frag_handling != OFPC_FRAG_REASM) {
3773         ofproto->backer->need_revalidate = REV_RECONFIGURE;
3774         return true;
3775     } else {
3776         return false;
3777     }
3778 }
3779
3780 static enum ofperr
3781 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
3782            const struct flow *flow,
3783            const struct ofpact *ofpacts, size_t ofpacts_len)
3784 {
3785     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3786
3787     ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
3788                                  ofpacts_len, packet);
3789     return 0;
3790 }
3791 \f
3792 /* NetFlow. */
3793
3794 static int
3795 set_netflow(struct ofproto *ofproto_,
3796             const struct netflow_options *netflow_options)
3797 {
3798     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3799
3800     if (netflow_options) {
3801         if (!ofproto->netflow) {
3802             ofproto->netflow = netflow_create();
3803             ofproto->backer->need_revalidate = REV_RECONFIGURE;
3804         }
3805         return netflow_set_options(ofproto->netflow, netflow_options);
3806     } else if (ofproto->netflow) {
3807         ofproto->backer->need_revalidate = REV_RECONFIGURE;
3808         netflow_unref(ofproto->netflow);
3809         ofproto->netflow = NULL;
3810     }
3811
3812     return 0;
3813 }
3814
3815 static void
3816 get_netflow_ids(const struct ofproto *ofproto_,
3817                 uint8_t *engine_type, uint8_t *engine_id)
3818 {
3819     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3820
3821     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
3822 }
3823 \f
3824 static struct ofproto_dpif *
3825 ofproto_dpif_lookup(const char *name)
3826 {
3827     struct ofproto_dpif *ofproto;
3828
3829     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
3830                              hash_string(name, 0), &all_ofproto_dpifs) {
3831         if (!strcmp(ofproto->up.name, name)) {
3832             return ofproto;
3833         }
3834     }
3835     return NULL;
3836 }
3837
3838 static void
3839 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
3840                           const char *argv[], void *aux OVS_UNUSED)
3841 {
3842     struct ofproto_dpif *ofproto;
3843
3844     if (argc > 1) {
3845         ofproto = ofproto_dpif_lookup(argv[1]);
3846         if (!ofproto) {
3847             unixctl_command_reply_error(conn, "no such bridge");
3848             return;
3849         }
3850         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3851         mac_learning_flush(ofproto->ml);
3852         ovs_rwlock_unlock(&ofproto->ml->rwlock);
3853     } else {
3854         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3855             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3856             mac_learning_flush(ofproto->ml);
3857             ovs_rwlock_unlock(&ofproto->ml->rwlock);
3858         }
3859     }
3860
3861     unixctl_command_reply(conn, "table successfully flushed");
3862 }
3863
3864 static void
3865 ofproto_unixctl_mcast_snooping_flush(struct unixctl_conn *conn, int argc,
3866                                      const char *argv[], void *aux OVS_UNUSED)
3867 {
3868     struct ofproto_dpif *ofproto;
3869
3870     if (argc > 1) {
3871         ofproto = ofproto_dpif_lookup(argv[1]);
3872         if (!ofproto) {
3873             unixctl_command_reply_error(conn, "no such bridge");
3874             return;
3875         }
3876
3877         if (!mcast_snooping_enabled(ofproto->ms)) {
3878             unixctl_command_reply_error(conn, "multicast snooping is disabled");
3879             return;
3880         }
3881         mcast_snooping_mdb_flush(ofproto->ms);
3882     } else {
3883         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3884             if (!mcast_snooping_enabled(ofproto->ms)) {
3885                 continue;
3886             }
3887             mcast_snooping_mdb_flush(ofproto->ms);
3888         }
3889     }
3890
3891     unixctl_command_reply(conn, "table successfully flushed");
3892 }
3893
3894 static struct ofport_dpif *
3895 ofbundle_get_a_port(const struct ofbundle *bundle)
3896 {
3897     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
3898                         bundle_node);
3899 }
3900
3901 static void
3902 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
3903                          const char *argv[], void *aux OVS_UNUSED)
3904 {
3905     struct ds ds = DS_EMPTY_INITIALIZER;
3906     const struct ofproto_dpif *ofproto;
3907     const struct mac_entry *e;
3908
3909     ofproto = ofproto_dpif_lookup(argv[1]);
3910     if (!ofproto) {
3911         unixctl_command_reply_error(conn, "no such bridge");
3912         return;
3913     }
3914
3915     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
3916     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3917     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3918         struct ofbundle *bundle = e->port.p;
3919         char name[OFP_MAX_PORT_NAME_LEN];
3920
3921         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
3922                                name, sizeof name);
3923         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
3924                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
3925                       mac_entry_age(ofproto->ml, e));
3926     }
3927     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3928     unixctl_command_reply(conn, ds_cstr(&ds));
3929     ds_destroy(&ds);
3930 }
3931
3932 static void
3933 ofproto_unixctl_mcast_snooping_show(struct unixctl_conn *conn,
3934                                     int argc OVS_UNUSED,
3935                                     const char *argv[],
3936                                     void *aux OVS_UNUSED)
3937 {
3938     struct ds ds = DS_EMPTY_INITIALIZER;
3939     const struct ofproto_dpif *ofproto;
3940     const struct ofbundle *bundle;
3941     const struct mcast_group *grp;
3942     struct mcast_group_bundle *b;
3943     struct mcast_mrouter_bundle *mrouter;
3944
3945     ofproto = ofproto_dpif_lookup(argv[1]);
3946     if (!ofproto) {
3947         unixctl_command_reply_error(conn, "no such bridge");
3948         return;
3949     }
3950
3951     if (!mcast_snooping_enabled(ofproto->ms)) {
3952         unixctl_command_reply_error(conn, "multicast snooping is disabled");
3953         return;
3954     }
3955
3956     ds_put_cstr(&ds, " port  VLAN  GROUP                Age\n");
3957     ovs_rwlock_rdlock(&ofproto->ms->rwlock);
3958     LIST_FOR_EACH (grp, group_node, &ofproto->ms->group_lru) {
3959         LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
3960             char name[OFP_MAX_PORT_NAME_LEN];
3961
3962             bundle = b->port;
3963             ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
3964                                    name, sizeof name);
3965             ds_put_format(&ds, "%5s  %4d  "IP_FMT"         %3d\n",
3966                           name, grp->vlan, IP_ARGS(grp->ip4),
3967                           mcast_bundle_age(ofproto->ms, b));
3968         }
3969     }
3970
3971     /* ports connected to multicast routers */
3972     LIST_FOR_EACH(mrouter, mrouter_node, &ofproto->ms->mrouter_lru) {
3973         char name[OFP_MAX_PORT_NAME_LEN];
3974
3975         bundle = mrouter->port;
3976         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
3977                                name, sizeof name);
3978             ds_put_format(&ds, "%5s  %4d  querier             %3d\n",
3979                       name, mrouter->vlan,
3980                       mcast_mrouter_age(ofproto->ms, mrouter));
3981     }
3982     ovs_rwlock_unlock(&ofproto->ms->rwlock);
3983     unixctl_command_reply(conn, ds_cstr(&ds));
3984     ds_destroy(&ds);
3985 }
3986
3987 struct trace_ctx {
3988     struct xlate_out xout;
3989     struct xlate_in xin;
3990     const struct flow *key;
3991     struct flow flow;
3992     struct flow_wildcards wc;
3993     struct ds *result;
3994 };
3995
3996 static void
3997 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
3998 {
3999     const struct rule_actions *actions;
4000     ovs_be64 cookie;
4001
4002     ds_put_char_multiple(result, '\t', level);
4003     if (!rule) {
4004         ds_put_cstr(result, "No match\n");
4005         return;
4006     }
4007
4008     ovs_mutex_lock(&rule->up.mutex);
4009     cookie = rule->up.flow_cookie;
4010     ovs_mutex_unlock(&rule->up.mutex);
4011
4012     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
4013                   rule ? rule->up.table_id : 0, ntohll(cookie));
4014     cls_rule_format(&rule->up.cr, result);
4015     ds_put_char(result, '\n');
4016
4017     actions = rule_dpif_get_actions(rule);
4018
4019     ds_put_char_multiple(result, '\t', level);
4020     ds_put_cstr(result, "OpenFlow actions=");
4021     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
4022     ds_put_char(result, '\n');
4023 }
4024
4025 static void
4026 trace_format_flow(struct ds *result, int level, const char *title,
4027                   struct trace_ctx *trace)
4028 {
4029     ds_put_char_multiple(result, '\t', level);
4030     ds_put_format(result, "%s: ", title);
4031     /* Do not report unchanged flows for resubmits. */
4032     if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
4033         || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
4034         ds_put_cstr(result, "unchanged");
4035     } else {
4036         flow_format(result, &trace->xin.flow);
4037         trace->flow = trace->xin.flow;
4038     }
4039     ds_put_char(result, '\n');
4040 }
4041
4042 static void
4043 trace_format_regs(struct ds *result, int level, const char *title,
4044                   struct trace_ctx *trace)
4045 {
4046     size_t i;
4047
4048     ds_put_char_multiple(result, '\t', level);
4049     ds_put_format(result, "%s:", title);
4050     for (i = 0; i < FLOW_N_REGS; i++) {
4051         ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
4052     }
4053     ds_put_char(result, '\n');
4054 }
4055
4056 static void
4057 trace_format_odp(struct ds *result, int level, const char *title,
4058                  struct trace_ctx *trace)
4059 {
4060     struct ofpbuf *odp_actions = &trace->xout.odp_actions;
4061
4062     ds_put_char_multiple(result, '\t', level);
4063     ds_put_format(result, "%s: ", title);
4064     format_odp_actions(result, ofpbuf_data(odp_actions),
4065                                ofpbuf_size(odp_actions));
4066     ds_put_char(result, '\n');
4067 }
4068
4069 static void
4070 trace_format_megaflow(struct ds *result, int level, const char *title,
4071                       struct trace_ctx *trace)
4072 {
4073     struct match match;
4074
4075     ds_put_char_multiple(result, '\t', level);
4076     ds_put_format(result, "%s: ", title);
4077     flow_wildcards_or(&trace->wc, &trace->xout.wc, &trace->wc);
4078     match_init(&match, trace->key, &trace->wc);
4079     match_format(&match, result, OFP_DEFAULT_PRIORITY);
4080     ds_put_char(result, '\n');
4081 }
4082
4083 static void
4084 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
4085 {
4086     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4087     struct ds *result = trace->result;
4088
4089     ds_put_char(result, '\n');
4090     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
4091     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
4092     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
4093     trace_format_megaflow(result, recurse + 1, "Resubmitted megaflow", trace);
4094     trace_format_rule(result, recurse + 1, rule);
4095 }
4096
4097 static void
4098 trace_report(struct xlate_in *xin, const char *s, int recurse)
4099 {
4100     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4101     struct ds *result = trace->result;
4102
4103     ds_put_char_multiple(result, '\t', recurse);
4104     ds_put_cstr(result, s);
4105     ds_put_char(result, '\n');
4106 }
4107
4108 /* Parses the 'argc' elements of 'argv', ignoring argv[0].  The following
4109  * forms are supported:
4110  *
4111  *     - [dpname] odp_flow [-generate | packet]
4112  *     - bridge br_flow [-generate | packet]
4113  *
4114  * On success, initializes '*ofprotop' and 'flow' and returns NULL.  On failure
4115  * returns a nonnull malloced error message. */
4116 static char * WARN_UNUSED_RESULT
4117 parse_flow_and_packet(int argc, const char *argv[],
4118                       struct ofproto_dpif **ofprotop, struct flow *flow,
4119                       struct ofpbuf **packetp)
4120 {
4121     const struct dpif_backer *backer = NULL;
4122     const char *error = NULL;
4123     char *m_err = NULL;
4124     struct simap port_names = SIMAP_INITIALIZER(&port_names);
4125     struct ofpbuf *packet;
4126     struct ofpbuf odp_key;
4127     struct ofpbuf odp_mask;
4128
4129     ofpbuf_init(&odp_key, 0);
4130     ofpbuf_init(&odp_mask, 0);
4131
4132     /* Handle "-generate" or a hex string as the last argument. */
4133     if (!strcmp(argv[argc - 1], "-generate")) {
4134         packet = ofpbuf_new(0);
4135         argc--;
4136     } else {
4137         error = eth_from_hex(argv[argc - 1], &packet);
4138         if (!error) {
4139             argc--;
4140         } else if (argc == 4) {
4141             /* The 3-argument form must end in "-generate' or a hex string. */
4142             goto exit;
4143         }
4144         error = NULL;
4145     }
4146
4147     /* odp_flow can have its in_port specified as a name instead of port no.
4148      * We do not yet know whether a given flow is a odp_flow or a br_flow.
4149      * But, to know whether a flow is odp_flow through odp_flow_from_string(),
4150      * we need to create a simap of name to port no. */
4151     if (argc == 3) {
4152         const char *dp_type;
4153         if (!strncmp(argv[1], "ovs-", 4)) {
4154             dp_type = argv[1] + 4;
4155         } else {
4156             dp_type = argv[1];
4157         }
4158         backer = shash_find_data(&all_dpif_backers, dp_type);
4159     } else if (argc == 2) {
4160         struct shash_node *node;
4161         if (shash_count(&all_dpif_backers) == 1) {
4162             node = shash_first(&all_dpif_backers);
4163             backer = node->data;
4164         }
4165     } else {
4166         error = "Syntax error";
4167         goto exit;
4168     }
4169     if (backer && backer->dpif) {
4170         struct dpif_port dpif_port;
4171         struct dpif_port_dump port_dump;
4172         DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4173             simap_put(&port_names, dpif_port.name,
4174                       odp_to_u32(dpif_port.port_no));
4175         }
4176     }
4177
4178     /* Parse the flow and determine whether a datapath or
4179      * bridge is specified. If function odp_flow_key_from_string()
4180      * returns 0, the flow is a odp_flow. If function
4181      * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4182     if (!odp_flow_from_string(argv[argc - 1], &port_names,
4183                               &odp_key, &odp_mask)) {
4184         if (!backer) {
4185             error = "Cannot find the datapath";
4186             goto exit;
4187         }
4188
4189         if (xlate_receive(backer, NULL, ofpbuf_data(&odp_key),
4190                           ofpbuf_size(&odp_key), flow,
4191                           ofprotop, NULL, NULL, NULL, NULL)) {
4192             error = "Invalid datapath flow";
4193             goto exit;
4194         }
4195     } else {
4196         char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4197
4198         if (err) {
4199             m_err = xasprintf("Bad flow syntax: %s", err);
4200             free(err);
4201             goto exit;
4202         } else {
4203             if (argc != 3) {
4204                 error = "Must specify bridge name";
4205                 goto exit;
4206             }
4207
4208             *ofprotop = ofproto_dpif_lookup(argv[1]);
4209             if (!*ofprotop) {
4210                 error = "Unknown bridge name";
4211                 goto exit;
4212             }
4213         }
4214     }
4215
4216     /* Generate a packet, if requested. */
4217     if (packet) {
4218         if (!ofpbuf_size(packet)) {
4219             flow_compose(packet, flow);
4220         } else {
4221             struct pkt_metadata md = pkt_metadata_from_flow(flow);
4222
4223             /* Use the metadata from the flow and the packet argument
4224              * to reconstruct the flow. */
4225             flow_extract(packet, &md, flow);
4226         }
4227     }
4228
4229 exit:
4230     if (error && !m_err) {
4231         m_err = xstrdup(error);
4232     }
4233     if (m_err) {
4234         ofpbuf_delete(packet);
4235         packet = NULL;
4236     }
4237     *packetp = packet;
4238     ofpbuf_uninit(&odp_key);
4239     ofpbuf_uninit(&odp_mask);
4240     simap_destroy(&port_names);
4241     return m_err;
4242 }
4243
4244 static void
4245 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4246                       void *aux OVS_UNUSED)
4247 {
4248     struct ofproto_dpif *ofproto;
4249     struct ofpbuf *packet;
4250     char *error;
4251     struct flow flow;
4252
4253     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4254     if (!error) {
4255         struct ds result;
4256
4257         ds_init(&result);
4258         ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4259         unixctl_command_reply(conn, ds_cstr(&result));
4260         ds_destroy(&result);
4261         ofpbuf_delete(packet);
4262     } else {
4263         unixctl_command_reply_error(conn, error);
4264         free(error);
4265     }
4266 }
4267
4268 static void
4269 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4270                               const char *argv[], void *aux OVS_UNUSED)
4271 {
4272     enum ofputil_protocol usable_protocols;
4273     struct ofproto_dpif *ofproto;
4274     bool enforce_consistency;
4275     struct ofpbuf ofpacts;
4276     struct ofpbuf *packet;
4277     struct ds result;
4278     struct flow flow;
4279     uint16_t in_port;
4280
4281     /* Three kinds of error return values! */
4282     enum ofperr retval;
4283     char *error;
4284
4285     packet = NULL;
4286     ds_init(&result);
4287     ofpbuf_init(&ofpacts, 0);
4288
4289     /* Parse actions. */
4290     error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols);
4291     if (error) {
4292         unixctl_command_reply_error(conn, error);
4293         free(error);
4294         goto exit;
4295     }
4296
4297     /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4298      * consistency between the flow and the actions.  With -consistent, we
4299      * enforce consistency even for a flow supported in OpenFlow 1.0. */
4300     if (!strcmp(argv[1], "-consistent")) {
4301         enforce_consistency = true;
4302         argv++;
4303         argc--;
4304     } else {
4305         enforce_consistency = false;
4306     }
4307
4308     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4309     if (error) {
4310         unixctl_command_reply_error(conn, error);
4311         free(error);
4312         goto exit;
4313     }
4314
4315     /* Do the same checks as handle_packet_out() in ofproto.c.
4316      *
4317      * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
4318      * strictly correct because these actions aren't in any table, but it's OK
4319      * because it 'table_id' is used only to check goto_table instructions, but
4320      * packet-outs take a list of actions and therefore it can't include
4321      * instructions.
4322      *
4323      * We skip the "meter" check here because meter is an instruction, not an
4324      * action, and thus cannot appear in ofpacts. */
4325     in_port = ofp_to_u16(flow.in_port.ofp_port);
4326     if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
4327         unixctl_command_reply_error(conn, "invalid in_port");
4328         goto exit;
4329     }
4330     if (enforce_consistency) {
4331         retval = ofpacts_check_consistency(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts),
4332                                            &flow, u16_to_ofp(ofproto->up.max_ports),
4333                                            0, 0, usable_protocols);
4334     } else {
4335         retval = ofpacts_check(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &flow,
4336                                u16_to_ofp(ofproto->up.max_ports), 0, 0,
4337                                &usable_protocols);
4338     }
4339
4340     if (retval) {
4341         ds_clear(&result);
4342         ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
4343         unixctl_command_reply_error(conn, ds_cstr(&result));
4344         goto exit;
4345     }
4346
4347     ofproto_trace(ofproto, &flow, packet,
4348                   ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &result);
4349     unixctl_command_reply(conn, ds_cstr(&result));
4350
4351 exit:
4352     ds_destroy(&result);
4353     ofpbuf_delete(packet);
4354     ofpbuf_uninit(&ofpacts);
4355 }
4356
4357 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4358  * description of the results to 'ds'.
4359  *
4360  * The trace follows a packet with the specified 'flow' through the flow
4361  * table.  'packet' may be nonnull to trace an actual packet, with consequent
4362  * side effects (if it is nonnull then its flow must be 'flow').
4363  *
4364  * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4365  * trace, otherwise the actions are determined by a flow table lookup. */
4366 static void
4367 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
4368               const struct ofpbuf *packet,
4369               const struct ofpact ofpacts[], size_t ofpacts_len,
4370               struct ds *ds)
4371 {
4372     struct rule_dpif *rule;
4373     struct trace_ctx trace;
4374
4375     ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
4376     ds_put_cstr(ds, "Flow: ");
4377     flow_format(ds, flow);
4378     ds_put_char(ds, '\n');
4379
4380     flow_wildcards_init_catchall(&trace.wc);
4381     if (ofpacts) {
4382         rule = NULL;
4383     } else {
4384         rule_dpif_lookup(ofproto, flow, &trace.wc, &rule, false, NULL);
4385
4386         trace_format_rule(ds, 0, rule);
4387         if (rule == ofproto->miss_rule) {
4388             ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
4389         } else if (rule == ofproto->no_packet_in_rule) {
4390             ds_put_cstr(ds, "\nNo match, packets dropped because "
4391                         "OFPPC_NO_PACKET_IN is set on in_port.\n");
4392         } else if (rule == ofproto->drop_frags_rule) {
4393             ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
4394                         "and the fragment handling mode is \"drop\".\n");
4395         }
4396     }
4397
4398     if (rule || ofpacts) {
4399         trace.result = ds;
4400         trace.key = flow; /* Original flow key, used for megaflow. */
4401         trace.flow = *flow; /* May be modified by actions. */
4402         xlate_in_init(&trace.xin, ofproto, flow, rule, ntohs(flow->tcp_flags),
4403                       packet);
4404         if (ofpacts) {
4405             trace.xin.ofpacts = ofpacts;
4406             trace.xin.ofpacts_len = ofpacts_len;
4407         }
4408         trace.xin.resubmit_hook = trace_resubmit;
4409         trace.xin.report_hook = trace_report;
4410
4411         xlate_actions(&trace.xin, &trace.xout);
4412
4413         ds_put_char(ds, '\n');
4414         trace_format_flow(ds, 0, "Final flow", &trace);
4415         trace_format_megaflow(ds, 0, "Megaflow", &trace);
4416
4417         ds_put_cstr(ds, "Datapath actions: ");
4418         format_odp_actions(ds, ofpbuf_data(&trace.xout.odp_actions),
4419                            ofpbuf_size(&trace.xout.odp_actions));
4420
4421         if (trace.xout.slow) {
4422             enum slow_path_reason slow;
4423
4424             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
4425                         "slow path because it:");
4426
4427             slow = trace.xout.slow;
4428             while (slow) {
4429                 enum slow_path_reason bit = rightmost_1bit(slow);
4430
4431                 ds_put_format(ds, "\n\t- %s.",
4432                               slow_path_reason_to_explanation(bit));
4433
4434                 slow &= ~bit;
4435             }
4436         }
4437
4438         xlate_out_uninit(&trace.xout);
4439     }
4440 }
4441
4442 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
4443  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
4444  * to destroy 'ofproto_shash' and free the returned value. */
4445 static const struct shash_node **
4446 get_ofprotos(struct shash *ofproto_shash)
4447 {
4448     const struct ofproto_dpif *ofproto;
4449
4450     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4451         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
4452         shash_add_nocopy(ofproto_shash, name, ofproto);
4453     }
4454
4455     return shash_sort(ofproto_shash);
4456 }
4457
4458 static void
4459 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
4460                               const char *argv[] OVS_UNUSED,
4461                               void *aux OVS_UNUSED)
4462 {
4463     struct ds ds = DS_EMPTY_INITIALIZER;
4464     struct shash ofproto_shash;
4465     const struct shash_node **sorted_ofprotos;
4466     int i;
4467
4468     shash_init(&ofproto_shash);
4469     sorted_ofprotos = get_ofprotos(&ofproto_shash);
4470     for (i = 0; i < shash_count(&ofproto_shash); i++) {
4471         const struct shash_node *node = sorted_ofprotos[i];
4472         ds_put_format(&ds, "%s\n", node->name);
4473     }
4474
4475     shash_destroy(&ofproto_shash);
4476     free(sorted_ofprotos);
4477
4478     unixctl_command_reply(conn, ds_cstr(&ds));
4479     ds_destroy(&ds);
4480 }
4481
4482 static void
4483 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
4484 {
4485     const struct shash_node **ofprotos;
4486     struct dpif_dp_stats dp_stats;
4487     struct shash ofproto_shash;
4488     size_t i;
4489
4490     dpif_get_dp_stats(backer->dpif, &dp_stats);
4491
4492     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
4493                   dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
4494
4495     shash_init(&ofproto_shash);
4496     ofprotos = get_ofprotos(&ofproto_shash);
4497     for (i = 0; i < shash_count(&ofproto_shash); i++) {
4498         struct ofproto_dpif *ofproto = ofprotos[i]->data;
4499         const struct shash_node **ports;
4500         size_t j;
4501
4502         if (ofproto->backer != backer) {
4503             continue;
4504         }
4505
4506         ds_put_format(ds, "\t%s:\n", ofproto->up.name);
4507
4508         ports = shash_sort(&ofproto->up.port_by_name);
4509         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
4510             const struct shash_node *node = ports[j];
4511             struct ofport *ofport = node->data;
4512             struct smap config;
4513             odp_port_t odp_port;
4514
4515             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
4516                           ofport->ofp_port);
4517
4518             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
4519             if (odp_port != ODPP_NONE) {
4520                 ds_put_format(ds, "%"PRIu32":", odp_port);
4521             } else {
4522                 ds_put_cstr(ds, "none:");
4523             }
4524
4525             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
4526
4527             smap_init(&config);
4528             if (!netdev_get_config(ofport->netdev, &config)) {
4529                 const struct smap_node **nodes;
4530                 size_t i;
4531
4532                 nodes = smap_sort(&config);
4533                 for (i = 0; i < smap_count(&config); i++) {
4534                     const struct smap_node *node = nodes[i];
4535                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
4536                                   node->key, node->value);
4537                 }
4538                 free(nodes);
4539             }
4540             smap_destroy(&config);
4541
4542             ds_put_char(ds, ')');
4543             ds_put_char(ds, '\n');
4544         }
4545         free(ports);
4546     }
4547     shash_destroy(&ofproto_shash);
4548     free(ofprotos);
4549 }
4550
4551 static void
4552 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4553                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4554 {
4555     struct ds ds = DS_EMPTY_INITIALIZER;
4556     const struct shash_node **backers;
4557     int i;
4558
4559     backers = shash_sort(&all_dpif_backers);
4560     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
4561         dpif_show_backer(backers[i]->data, &ds);
4562     }
4563     free(backers);
4564
4565     unixctl_command_reply(conn, ds_cstr(&ds));
4566     ds_destroy(&ds);
4567 }
4568
4569 static bool
4570 ofproto_dpif_contains_flow(const struct ofproto_dpif *ofproto,
4571                            const struct nlattr *key, size_t key_len)
4572 {
4573     struct ofproto_dpif *ofp;
4574     struct flow flow;
4575
4576     xlate_receive(ofproto->backer, NULL, key, key_len, &flow, &ofp,
4577                   NULL, NULL, NULL, NULL);
4578     return ofp == ofproto;
4579 }
4580
4581 static void
4582 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
4583                                 int argc OVS_UNUSED, const char *argv[],
4584                                 void *aux OVS_UNUSED)
4585 {
4586     const struct ofproto_dpif *ofproto;
4587
4588     struct ds ds = DS_EMPTY_INITIALIZER;
4589     bool verbosity = false;
4590
4591     struct dpif_port dpif_port;
4592     struct dpif_port_dump port_dump;
4593     struct hmap portno_names;
4594
4595     struct dpif_flow_dump *flow_dump;
4596     struct dpif_flow_dump_thread *flow_dump_thread;
4597     struct dpif_flow f;
4598     int error;
4599
4600     ofproto = ofproto_dpif_lookup(argv[argc - 1]);
4601     if (!ofproto) {
4602         unixctl_command_reply_error(conn, "no such bridge");
4603         return;
4604     }
4605
4606     if (argc > 2 && !strcmp(argv[1], "-m")) {
4607         verbosity = true;
4608     }
4609
4610     hmap_init(&portno_names);
4611     DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
4612         odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
4613     }
4614
4615     ds_init(&ds);
4616     flow_dump = dpif_flow_dump_create(ofproto->backer->dpif);
4617     flow_dump_thread = dpif_flow_dump_thread_create(flow_dump);
4618     while (dpif_flow_dump_next(flow_dump_thread, &f, 1)) {
4619         if (!ofproto_dpif_contains_flow(ofproto, f.key, f.key_len)) {
4620             continue;
4621         }
4622
4623         odp_flow_format(f.key, f.key_len, f.mask, f.mask_len,
4624                         &portno_names, &ds, verbosity);
4625         ds_put_cstr(&ds, ", ");
4626         dpif_flow_stats_format(&f.stats, &ds);
4627         ds_put_cstr(&ds, ", actions:");
4628         format_odp_actions(&ds, f.actions, f.actions_len);
4629         ds_put_char(&ds, '\n');
4630     }
4631     dpif_flow_dump_thread_destroy(flow_dump_thread);
4632     error = dpif_flow_dump_destroy(flow_dump);
4633
4634     if (error) {
4635         ds_clear(&ds);
4636         ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
4637         unixctl_command_reply_error(conn, ds_cstr(&ds));
4638     } else {
4639         unixctl_command_reply(conn, ds_cstr(&ds));
4640     }
4641     odp_portno_names_destroy(&portno_names);
4642     hmap_destroy(&portno_names);
4643     ds_destroy(&ds);
4644 }
4645
4646 static void
4647 ofproto_dpif_unixctl_init(void)
4648 {
4649     static bool registered;
4650     if (registered) {
4651         return;
4652     }
4653     registered = true;
4654
4655     unixctl_command_register(
4656         "ofproto/trace",
4657         "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
4658         1, 3, ofproto_unixctl_trace, NULL);
4659     unixctl_command_register(
4660         "ofproto/trace-packet-out",
4661         "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
4662         2, 6, ofproto_unixctl_trace_actions, NULL);
4663     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
4664                              ofproto_unixctl_fdb_flush, NULL);
4665     unixctl_command_register("fdb/show", "bridge", 1, 1,
4666                              ofproto_unixctl_fdb_show, NULL);
4667     unixctl_command_register("mdb/flush", "[bridge]", 0, 1,
4668                              ofproto_unixctl_mcast_snooping_flush, NULL);
4669     unixctl_command_register("mdb/show", "bridge", 1, 1,
4670                              ofproto_unixctl_mcast_snooping_show, NULL);
4671     unixctl_command_register("dpif/dump-dps", "", 0, 0,
4672                              ofproto_unixctl_dpif_dump_dps, NULL);
4673     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
4674                              NULL);
4675     unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
4676                              ofproto_unixctl_dpif_dump_flows, NULL);
4677 }
4678
4679 /* Returns true if 'table' is the table used for internal rules,
4680  * false otherwise. */
4681 bool
4682 table_is_internal(uint8_t table_id)
4683 {
4684     return table_id == TBL_INTERNAL;
4685 }
4686 \f
4687 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4688  *
4689  * This is deprecated.  It is only for compatibility with broken device drivers
4690  * in old versions of Linux that do not properly support VLANs when VLAN
4691  * devices are not used.  When broken device drivers are no longer in
4692  * widespread use, we will delete these interfaces. */
4693
4694 static int
4695 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
4696 {
4697     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
4698     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
4699
4700     if (realdev_ofp_port == ofport->realdev_ofp_port
4701         && vid == ofport->vlandev_vid) {
4702         return 0;
4703     }
4704
4705     ofproto->backer->need_revalidate = REV_RECONFIGURE;
4706
4707     if (ofport->realdev_ofp_port) {
4708         vsp_remove(ofport);
4709     }
4710     if (realdev_ofp_port && ofport->bundle) {
4711         /* vlandevs are enslaved to their realdevs, so they are not allowed to
4712          * themselves be part of a bundle. */
4713         bundle_set(ofport_->ofproto, ofport->bundle, NULL);
4714     }
4715
4716     ofport->realdev_ofp_port = realdev_ofp_port;
4717     ofport->vlandev_vid = vid;
4718
4719     if (realdev_ofp_port) {
4720         vsp_add(ofport, realdev_ofp_port, vid);
4721     }
4722
4723     return 0;
4724 }
4725
4726 static uint32_t
4727 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
4728 {
4729     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
4730 }
4731
4732 bool
4733 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
4734     OVS_EXCLUDED(ofproto->vsp_mutex)
4735 {
4736     /* hmap_is_empty is thread safe. */
4737     return !hmap_is_empty(&ofproto->realdev_vid_map);
4738 }
4739
4740
4741 static ofp_port_t
4742 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
4743                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4744     OVS_REQUIRES(ofproto->vsp_mutex)
4745 {
4746     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
4747         int vid = vlan_tci_to_vid(vlan_tci);
4748         const struct vlan_splinter *vsp;
4749
4750         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
4751                                  hash_realdev_vid(realdev_ofp_port, vid),
4752                                  &ofproto->realdev_vid_map) {
4753             if (vsp->realdev_ofp_port == realdev_ofp_port
4754                 && vsp->vid == vid) {
4755                 return vsp->vlandev_ofp_port;
4756             }
4757         }
4758     }
4759     return realdev_ofp_port;
4760 }
4761
4762 /* Returns the OFP port number of the Linux VLAN device that corresponds to
4763  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
4764  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
4765  * 'vlan_tci' 9, it would return the port number of eth0.9.
4766  *
4767  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
4768  * function just returns its 'realdev_ofp_port' argument. */
4769 ofp_port_t
4770 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
4771                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4772     OVS_EXCLUDED(ofproto->vsp_mutex)
4773 {
4774     ofp_port_t ret;
4775
4776     /* hmap_is_empty is thread safe, see if we can return immediately. */
4777     if (hmap_is_empty(&ofproto->realdev_vid_map)) {
4778         return realdev_ofp_port;
4779     }
4780     ovs_mutex_lock(&ofproto->vsp_mutex);
4781     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
4782     ovs_mutex_unlock(&ofproto->vsp_mutex);
4783     return ret;
4784 }
4785
4786 static struct vlan_splinter *
4787 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
4788 {
4789     struct vlan_splinter *vsp;
4790
4791     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
4792                              hash_ofp_port(vlandev_ofp_port),
4793                              &ofproto->vlandev_map) {
4794         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
4795             return vsp;
4796         }
4797     }
4798
4799     return NULL;
4800 }
4801
4802 /* Returns the OpenFlow port number of the "real" device underlying the Linux
4803  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
4804  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
4805  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
4806  * eth0 and store 9 in '*vid'.
4807  *
4808  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
4809  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
4810  * always does.*/
4811 static ofp_port_t
4812 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
4813                        ofp_port_t vlandev_ofp_port, int *vid)
4814     OVS_REQUIRES(ofproto->vsp_mutex)
4815 {
4816     if (!hmap_is_empty(&ofproto->vlandev_map)) {
4817         const struct vlan_splinter *vsp;
4818
4819         vsp = vlandev_find(ofproto, vlandev_ofp_port);
4820         if (vsp) {
4821             if (vid) {
4822                 *vid = vsp->vid;
4823             }
4824             return vsp->realdev_ofp_port;
4825         }
4826     }
4827     return 0;
4828 }
4829
4830 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
4831  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
4832  * 'flow->in_port' to the "real" device backing the VLAN device, sets
4833  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Otherwise (which is
4834  * always the case unless VLAN splinters are enabled), returns false without
4835  * making any changes. */
4836 bool
4837 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
4838     OVS_EXCLUDED(ofproto->vsp_mutex)
4839 {
4840     ofp_port_t realdev;
4841     int vid;
4842
4843     /* hmap_is_empty is thread safe. */
4844     if (hmap_is_empty(&ofproto->vlandev_map)) {
4845         return false;
4846     }
4847
4848     ovs_mutex_lock(&ofproto->vsp_mutex);
4849     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
4850     ovs_mutex_unlock(&ofproto->vsp_mutex);
4851     if (!realdev) {
4852         return false;
4853     }
4854
4855     /* Cause the flow to be processed as if it came in on the real device with
4856      * the VLAN device's VLAN ID. */
4857     flow->in_port.ofp_port = realdev;
4858     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
4859     return true;
4860 }
4861
4862 static void
4863 vsp_remove(struct ofport_dpif *port)
4864 {
4865     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4866     struct vlan_splinter *vsp;
4867
4868     ovs_mutex_lock(&ofproto->vsp_mutex);
4869     vsp = vlandev_find(ofproto, port->up.ofp_port);
4870     if (vsp) {
4871         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
4872         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
4873         free(vsp);
4874
4875         port->realdev_ofp_port = 0;
4876     } else {
4877         VLOG_ERR("missing vlan device record");
4878     }
4879     ovs_mutex_unlock(&ofproto->vsp_mutex);
4880 }
4881
4882 static void
4883 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
4884 {
4885     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4886
4887     ovs_mutex_lock(&ofproto->vsp_mutex);
4888     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
4889         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
4890             == realdev_ofp_port)) {
4891         struct vlan_splinter *vsp;
4892
4893         vsp = xmalloc(sizeof *vsp);
4894         vsp->realdev_ofp_port = realdev_ofp_port;
4895         vsp->vlandev_ofp_port = port->up.ofp_port;
4896         vsp->vid = vid;
4897
4898         port->realdev_ofp_port = realdev_ofp_port;
4899
4900         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
4901                     hash_ofp_port(port->up.ofp_port));
4902         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
4903                     hash_realdev_vid(realdev_ofp_port, vid));
4904     } else {
4905         VLOG_ERR("duplicate vlan device record");
4906     }
4907     ovs_mutex_unlock(&ofproto->vsp_mutex);
4908 }
4909
4910 static odp_port_t
4911 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
4912 {
4913     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
4914     return ofport ? ofport->odp_port : ODPP_NONE;
4915 }
4916
4917 struct ofport_dpif *
4918 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
4919 {
4920     struct ofport_dpif *port;
4921
4922     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
4923     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
4924                              &backer->odp_to_ofport_map) {
4925         if (port->odp_port == odp_port) {
4926             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4927             return port;
4928         }
4929     }
4930
4931     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4932     return NULL;
4933 }
4934
4935 static ofp_port_t
4936 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
4937 {
4938     struct ofport_dpif *port;
4939
4940     port = odp_port_to_ofport(ofproto->backer, odp_port);
4941     if (port && &ofproto->up == port->up.ofproto) {
4942         return port->up.ofp_port;
4943     } else {
4944         return OFPP_NONE;
4945     }
4946 }
4947
4948 uint32_t
4949 ofproto_dpif_alloc_recirc_id(struct ofproto_dpif *ofproto)
4950 {
4951     struct dpif_backer *backer = ofproto->backer;
4952
4953     return  recirc_id_alloc(backer->rid_pool);
4954 }
4955
4956 void
4957 ofproto_dpif_free_recirc_id(struct ofproto_dpif *ofproto, uint32_t recirc_id)
4958 {
4959     struct dpif_backer *backer = ofproto->backer;
4960
4961     recirc_id_free(backer->rid_pool, recirc_id);
4962 }
4963
4964 int
4965 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
4966                                const struct match *match, int priority,
4967                                uint16_t idle_timeout,
4968                                const struct ofpbuf *ofpacts,
4969                                struct rule **rulep)
4970 {
4971     struct ofputil_flow_mod fm;
4972     struct rule_dpif *rule;
4973     int error;
4974
4975     fm.match = *match;
4976     fm.priority = priority;
4977     fm.new_cookie = htonll(0);
4978     fm.cookie = htonll(0);
4979     fm.cookie_mask = htonll(0);
4980     fm.modify_cookie = false;
4981     fm.table_id = TBL_INTERNAL;
4982     fm.command = OFPFC_ADD;
4983     fm.idle_timeout = idle_timeout;
4984     fm.hard_timeout = 0;
4985     fm.buffer_id = 0;
4986     fm.out_port = 0;
4987     fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
4988     fm.ofpacts = ofpbuf_data(ofpacts);
4989     fm.ofpacts_len = ofpbuf_size(ofpacts);
4990
4991     error = ofproto_flow_mod(&ofproto->up, &fm);
4992     if (error) {
4993         VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
4994                     ofperr_to_string(error));
4995         *rulep = NULL;
4996         return error;
4997     }
4998
4999     rule = rule_dpif_lookup_in_table(ofproto, TBL_INTERNAL, &fm.match.flow,
5000                                      &fm.match.wc, false);
5001     if (rule) {
5002         *rulep = &rule->up;
5003     } else {
5004         OVS_NOT_REACHED();
5005     }
5006     return 0;
5007 }
5008
5009 int
5010 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
5011                                   struct match *match, int priority)
5012 {
5013     struct ofputil_flow_mod fm;
5014     int error;
5015
5016     fm.match = *match;
5017     fm.priority = priority;
5018     fm.new_cookie = htonll(0);
5019     fm.cookie = htonll(0);
5020     fm.cookie_mask = htonll(0);
5021     fm.modify_cookie = false;
5022     fm.table_id = TBL_INTERNAL;
5023     fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5024     fm.command = OFPFC_DELETE_STRICT;
5025
5026     error = ofproto_flow_mod(&ofproto->up, &fm);
5027     if (error) {
5028         VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
5029                     ofperr_to_string(error));
5030         return error;
5031     }
5032
5033     return 0;
5034 }
5035
5036 const struct ofproto_class ofproto_dpif_class = {
5037     init,
5038     enumerate_types,
5039     enumerate_names,
5040     del,
5041     port_open_type,
5042     type_run,
5043     type_wait,
5044     alloc,
5045     construct,
5046     destruct,
5047     dealloc,
5048     run,
5049     wait,
5050     NULL,                       /* get_memory_usage. */
5051     type_get_memory_usage,
5052     flush,
5053     query_tables,
5054     port_alloc,
5055     port_construct,
5056     port_destruct,
5057     port_dealloc,
5058     port_modified,
5059     port_reconfigured,
5060     port_query_by_name,
5061     port_add,
5062     port_del,
5063     port_get_stats,
5064     port_dump_start,
5065     port_dump_next,
5066     port_dump_done,
5067     port_poll,
5068     port_poll_wait,
5069     port_is_lacp_current,
5070     NULL,                       /* rule_choose_table */
5071     rule_alloc,
5072     rule_construct,
5073     rule_insert,
5074     rule_delete,
5075     rule_destruct,
5076     rule_dealloc,
5077     rule_get_stats,
5078     rule_execute,
5079     NULL,                       /* rule_premodify_actions */
5080     rule_modify_actions,
5081     set_frag_handling,
5082     packet_out,
5083     set_netflow,
5084     get_netflow_ids,
5085     set_sflow,
5086     set_ipfix,
5087     set_cfm,
5088     cfm_status_changed,
5089     get_cfm_status,
5090     set_bfd,
5091     bfd_status_changed,
5092     get_bfd_status,
5093     set_stp,
5094     get_stp_status,
5095     set_stp_port,
5096     get_stp_port_status,
5097     get_stp_port_stats,
5098     set_queues,
5099     bundle_set,
5100     bundle_remove,
5101     mirror_set__,
5102     mirror_get_stats__,
5103     set_flood_vlans,
5104     is_mirror_output_bundle,
5105     forward_bpdu_changed,
5106     set_mac_table_config,
5107     set_mcast_snooping,
5108     set_mcast_snooping_port,
5109     set_realdev,
5110     NULL,                       /* meter_get_features */
5111     NULL,                       /* meter_set */
5112     NULL,                       /* meter_get */
5113     NULL,                       /* meter_del */
5114     group_alloc,                /* group_alloc */
5115     group_construct,            /* group_construct */
5116     group_destruct,             /* group_destruct */
5117     group_dealloc,              /* group_dealloc */
5118     group_modify,               /* group_modify */
5119     group_get_stats,            /* group_get_stats */
5120 };