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->in_port = OFPP_IN_PORT;
1250     resubmit->table_id = 0;
1251
1252     match_init_catchall(&match);
1253     match_set_recirc_id(&match, 0);
1254
1255     error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, 0, &ofpacts,
1256                                            &unused_rulep);
1257     if (error) {
1258         return error;
1259     }
1260
1261     /* Drop any run away recirc rule lookups. Recirc_id has to be
1262      * non-zero when reaching this rule.
1263      *
1264      * (priority=1), *, actions=drop
1265      */
1266     ofpbuf_clear(&ofpacts);
1267     match_init_catchall(&match);
1268     error = ofproto_dpif_add_internal_flow(ofproto, &match, 1, 0, &ofpacts,
1269                                            &unused_rulep);
1270
1271     return error;
1272 }
1273
1274 static void
1275 destruct(struct ofproto *ofproto_)
1276 {
1277     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1278     struct ofproto_packet_in *pin, *next_pin;
1279     struct rule_dpif *rule;
1280     struct oftable *table;
1281     struct list pins;
1282
1283     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1284     xlate_txn_start();
1285     xlate_remove_ofproto(ofproto);
1286     xlate_txn_commit();
1287
1288     /* Ensure that the upcall processing threads have no remaining references
1289      * to the ofproto or anything in it. */
1290     udpif_synchronize(ofproto->backer->udpif);
1291
1292     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1293
1294     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1295         CLS_FOR_EACH_SAFE (rule, up.cr, &table->cls) {
1296             ofproto_rule_delete(&ofproto->up, &rule->up);
1297         }
1298     }
1299
1300     guarded_list_pop_all(&ofproto->pins, &pins);
1301     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1302         list_remove(&pin->list_node);
1303         free(CONST_CAST(void *, pin->up.packet));
1304         free(pin);
1305     }
1306     guarded_list_destroy(&ofproto->pins);
1307
1308     mbridge_unref(ofproto->mbridge);
1309
1310     netflow_unref(ofproto->netflow);
1311     dpif_sflow_unref(ofproto->sflow);
1312     dpif_ipfix_unref(ofproto->ipfix);
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         if (ofproto->ipfix) {
1587            dpif_ipfix_add_tunnel_port(ofproto->ipfix, port_, port->odp_port);
1588         }
1589     } else {
1590         /* Sanity-check that a mapping doesn't already exist.  This
1591          * shouldn't happen for non-tunnel ports. */
1592         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1593             VLOG_ERR("port %s already has an OpenFlow port number",
1594                      dpif_port.name);
1595             dpif_port_destroy(&dpif_port);
1596             return EBUSY;
1597         }
1598
1599         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1600         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1601                     hash_odp_port(port->odp_port));
1602         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1603     }
1604     dpif_port_destroy(&dpif_port);
1605
1606     if (ofproto->sflow) {
1607         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1608     }
1609
1610     return 0;
1611 }
1612
1613 static void
1614 port_destruct(struct ofport *port_)
1615 {
1616     struct ofport_dpif *port = ofport_dpif_cast(port_);
1617     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1618     const char *devname = netdev_get_name(port->up.netdev);
1619     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1620     const char *dp_port_name;
1621
1622     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1623     xlate_txn_start();
1624     xlate_ofport_remove(port);
1625     xlate_txn_commit();
1626
1627     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1628                                               sizeof namebuf);
1629     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1630         /* The underlying device is still there, so delete it.  This
1631          * happens when the ofproto is being destroyed, since the caller
1632          * assumes that removal of attached ports will happen as part of
1633          * destruction. */
1634         if (!port->is_tunnel) {
1635             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1636         }
1637     }
1638
1639     if (port->peer) {
1640         port->peer->peer = NULL;
1641         port->peer = NULL;
1642     }
1643
1644     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1645         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1646         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1647         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1648     }
1649
1650     if (port->is_tunnel && ofproto->ipfix) {
1651        dpif_ipfix_del_tunnel_port(ofproto->ipfix, port->odp_port);
1652     }
1653
1654     tnl_port_del(port);
1655     sset_find_and_delete(&ofproto->ports, devname);
1656     sset_find_and_delete(&ofproto->ghost_ports, devname);
1657     bundle_remove(port_);
1658     set_cfm(port_, NULL);
1659     set_bfd(port_, NULL);
1660     if (port->stp_port) {
1661         stp_port_disable(port->stp_port);
1662     }
1663     if (ofproto->sflow) {
1664         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1665     }
1666
1667     free(port->qdscp);
1668 }
1669
1670 static void
1671 port_modified(struct ofport *port_)
1672 {
1673     struct ofport_dpif *port = ofport_dpif_cast(port_);
1674
1675     if (port->bundle && port->bundle->bond) {
1676         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1677     }
1678
1679     if (port->cfm) {
1680         cfm_set_netdev(port->cfm, port->up.netdev);
1681     }
1682
1683     if (port->bfd) {
1684         bfd_set_netdev(port->bfd, port->up.netdev);
1685     }
1686
1687     ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1688                                      port->up.pp.hw_addr);
1689
1690     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1691                                                 port->odp_port)) {
1692         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1693             REV_RECONFIGURE;
1694     }
1695
1696     ofport_update_peer(port);
1697 }
1698
1699 static void
1700 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1701 {
1702     struct ofport_dpif *port = ofport_dpif_cast(port_);
1703     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1704     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1705
1706     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1707                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1708                    OFPUTIL_PC_NO_PACKET_IN)) {
1709         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1710
1711         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1712             bundle_update(port->bundle);
1713         }
1714     }
1715 }
1716
1717 static int
1718 set_sflow(struct ofproto *ofproto_,
1719           const struct ofproto_sflow_options *sflow_options)
1720 {
1721     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1722     struct dpif_sflow *ds = ofproto->sflow;
1723
1724     if (sflow_options) {
1725         if (!ds) {
1726             struct ofport_dpif *ofport;
1727
1728             ds = ofproto->sflow = dpif_sflow_create();
1729             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1730                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1731             }
1732             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1733         }
1734         dpif_sflow_set_options(ds, sflow_options);
1735     } else {
1736         if (ds) {
1737             dpif_sflow_unref(ds);
1738             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1739             ofproto->sflow = NULL;
1740         }
1741     }
1742     return 0;
1743 }
1744
1745 static int
1746 set_ipfix(
1747     struct ofproto *ofproto_,
1748     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1749     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1750     size_t n_flow_exporters_options)
1751 {
1752     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1753     struct dpif_ipfix *di = ofproto->ipfix;
1754     bool has_options = bridge_exporter_options || flow_exporters_options;
1755     bool new_di = false;
1756
1757     if (has_options && !di) {
1758         di = ofproto->ipfix = dpif_ipfix_create();
1759         new_di = true;
1760     }
1761
1762     if (di) {
1763         /* Call set_options in any case to cleanly flush the flow
1764          * caches in the last exporters that are to be destroyed. */
1765         dpif_ipfix_set_options(
1766             di, bridge_exporter_options, flow_exporters_options,
1767             n_flow_exporters_options);
1768
1769         /* Add tunnel ports only when a new ipfix created */
1770         if (new_di == true) {
1771             struct ofport_dpif *ofport;
1772             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1773                 if (ofport->is_tunnel == true) {
1774                     dpif_ipfix_add_tunnel_port(di, &ofport->up, ofport->odp_port);
1775                 }
1776             }
1777         }
1778
1779         if (!has_options) {
1780             dpif_ipfix_unref(di);
1781             ofproto->ipfix = NULL;
1782         }
1783     }
1784
1785     return 0;
1786 }
1787
1788 static int
1789 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1790 {
1791     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1792     int error = 0;
1793
1794     if (s) {
1795         if (!ofport->cfm) {
1796             struct ofproto_dpif *ofproto;
1797
1798             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1799             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1800             ofport->cfm = cfm_create(ofport->up.netdev);
1801         }
1802
1803         if (cfm_configure(ofport->cfm, s)) {
1804             error = 0;
1805             goto out;
1806         }
1807
1808         error = EINVAL;
1809     }
1810     cfm_unref(ofport->cfm);
1811     ofport->cfm = NULL;
1812 out:
1813     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1814                                      ofport->up.pp.hw_addr);
1815     return error;
1816 }
1817
1818 static bool
1819 cfm_status_changed(struct ofport *ofport_)
1820 {
1821     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1822
1823     return ofport->cfm ? cfm_check_status_change(ofport->cfm) : true;
1824 }
1825
1826 static int
1827 get_cfm_status(const struct ofport *ofport_,
1828                struct cfm_status *status)
1829 {
1830     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1831     int ret = 0;
1832
1833     if (ofport->cfm) {
1834         cfm_get_status(ofport->cfm, status);
1835     } else {
1836         ret = ENOENT;
1837     }
1838
1839     return ret;
1840 }
1841
1842 static int
1843 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1844 {
1845     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1846     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1847     struct bfd *old;
1848
1849     old = ofport->bfd;
1850     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1851                                 cfg, ofport->up.netdev);
1852     if (ofport->bfd != old) {
1853         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1854     }
1855     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1856                                      ofport->up.pp.hw_addr);
1857     return 0;
1858 }
1859
1860 static bool
1861 bfd_status_changed(struct ofport *ofport_)
1862 {
1863     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1864
1865     return ofport->bfd ? bfd_check_status_change(ofport->bfd) : true;
1866 }
1867
1868 static int
1869 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1870 {
1871     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1872     int ret = 0;
1873
1874     if (ofport->bfd) {
1875         bfd_get_status(ofport->bfd, smap);
1876     } else {
1877         ret = ENOENT;
1878     }
1879
1880     return ret;
1881 }
1882 \f
1883 /* Spanning Tree. */
1884
1885 static void
1886 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1887 {
1888     struct ofproto_dpif *ofproto = ofproto_;
1889     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1890     struct ofport_dpif *ofport;
1891
1892     ofport = stp_port_get_aux(sp);
1893     if (!ofport) {
1894         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1895                      ofproto->up.name, port_num);
1896     } else {
1897         struct eth_header *eth = ofpbuf_l2(pkt);
1898
1899         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1900         if (eth_addr_is_zero(eth->eth_src)) {
1901             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1902                          "with unknown MAC", ofproto->up.name, port_num);
1903         } else {
1904             ofproto_dpif_send_packet(ofport, pkt);
1905         }
1906     }
1907     ofpbuf_delete(pkt);
1908 }
1909
1910 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1911 static int
1912 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1913 {
1914     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1915
1916     /* Only revalidate flows if the configuration changed. */
1917     if (!s != !ofproto->stp) {
1918         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1919     }
1920
1921     if (s) {
1922         if (!ofproto->stp) {
1923             ofproto->stp = stp_create(ofproto_->name, s->system_id,
1924                                       send_bpdu_cb, ofproto);
1925             ofproto->stp_last_tick = time_msec();
1926         }
1927
1928         stp_set_bridge_id(ofproto->stp, s->system_id);
1929         stp_set_bridge_priority(ofproto->stp, s->priority);
1930         stp_set_hello_time(ofproto->stp, s->hello_time);
1931         stp_set_max_age(ofproto->stp, s->max_age);
1932         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1933     }  else {
1934         struct ofport *ofport;
1935
1936         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1937             set_stp_port(ofport, NULL);
1938         }
1939
1940         stp_unref(ofproto->stp);
1941         ofproto->stp = NULL;
1942     }
1943
1944     return 0;
1945 }
1946
1947 static int
1948 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1949 {
1950     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1951
1952     if (ofproto->stp) {
1953         s->enabled = true;
1954         s->bridge_id = stp_get_bridge_id(ofproto->stp);
1955         s->designated_root = stp_get_designated_root(ofproto->stp);
1956         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1957     } else {
1958         s->enabled = false;
1959     }
1960
1961     return 0;
1962 }
1963
1964 static void
1965 update_stp_port_state(struct ofport_dpif *ofport)
1966 {
1967     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1968     enum stp_state state;
1969
1970     /* Figure out new state. */
1971     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1972                              : STP_DISABLED;
1973
1974     /* Update state. */
1975     if (ofport->stp_state != state) {
1976         enum ofputil_port_state of_state;
1977         bool fwd_change;
1978
1979         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1980                     netdev_get_name(ofport->up.netdev),
1981                     stp_state_name(ofport->stp_state),
1982                     stp_state_name(state));
1983         if (stp_learn_in_state(ofport->stp_state)
1984                 != stp_learn_in_state(state)) {
1985             /* xxx Learning action flows should also be flushed. */
1986             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1987             mac_learning_flush(ofproto->ml);
1988             ovs_rwlock_unlock(&ofproto->ml->rwlock);
1989             mcast_snooping_mdb_flush(ofproto->ms);
1990         }
1991         fwd_change = stp_forward_in_state(ofport->stp_state)
1992                         != stp_forward_in_state(state);
1993
1994         ofproto->backer->need_revalidate = REV_STP;
1995         ofport->stp_state = state;
1996         ofport->stp_state_entered = time_msec();
1997
1998         if (fwd_change && ofport->bundle) {
1999             bundle_update(ofport->bundle);
2000         }
2001
2002         /* Update the STP state bits in the OpenFlow port description. */
2003         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2004         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2005                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2006                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2007                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2008                      : 0);
2009         ofproto_port_set_state(&ofport->up, of_state);
2010     }
2011 }
2012
2013 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2014  * caller is responsible for assigning STP port numbers and ensuring
2015  * there are no duplicates. */
2016 static int
2017 set_stp_port(struct ofport *ofport_,
2018              const struct ofproto_port_stp_settings *s)
2019 {
2020     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2021     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2022     struct stp_port *sp = ofport->stp_port;
2023
2024     if (!s || !s->enable) {
2025         if (sp) {
2026             ofport->stp_port = NULL;
2027             stp_port_disable(sp);
2028             update_stp_port_state(ofport);
2029         }
2030         return 0;
2031     } else if (sp && stp_port_no(sp) != s->port_num
2032             && ofport == stp_port_get_aux(sp)) {
2033         /* The port-id changed, so disable the old one if it's not
2034          * already in use by another port. */
2035         stp_port_disable(sp);
2036     }
2037
2038     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2039     stp_port_enable(sp);
2040
2041     stp_port_set_name(sp, netdev_get_name(ofport->up.netdev));
2042     stp_port_set_aux(sp, ofport);
2043     stp_port_set_priority(sp, s->priority);
2044     stp_port_set_path_cost(sp, s->path_cost);
2045
2046     update_stp_port_state(ofport);
2047
2048     return 0;
2049 }
2050
2051 static int
2052 get_stp_port_status(struct ofport *ofport_,
2053                     struct ofproto_port_stp_status *s)
2054 {
2055     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2056     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2057     struct stp_port *sp = ofport->stp_port;
2058
2059     if (!ofproto->stp || !sp) {
2060         s->enabled = false;
2061         return 0;
2062     }
2063
2064     s->enabled = true;
2065     s->port_id = stp_port_get_id(sp);
2066     s->state = stp_port_get_state(sp);
2067     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2068     s->role = stp_port_get_role(sp);
2069
2070     return 0;
2071 }
2072
2073 static int
2074 get_stp_port_stats(struct ofport *ofport_,
2075                    struct ofproto_port_stp_stats *s)
2076 {
2077     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2078     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2079     struct stp_port *sp = ofport->stp_port;
2080
2081     if (!ofproto->stp || !sp) {
2082         s->enabled = false;
2083         return 0;
2084     }
2085
2086     s->enabled = true;
2087     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2088
2089     return 0;
2090 }
2091
2092 static void
2093 stp_run(struct ofproto_dpif *ofproto)
2094 {
2095     if (ofproto->stp) {
2096         long long int now = time_msec();
2097         long long int elapsed = now - ofproto->stp_last_tick;
2098         struct stp_port *sp;
2099
2100         if (elapsed > 0) {
2101             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2102             ofproto->stp_last_tick = now;
2103         }
2104         while (stp_get_changed_port(ofproto->stp, &sp)) {
2105             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2106
2107             if (ofport) {
2108                 update_stp_port_state(ofport);
2109             }
2110         }
2111
2112         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2113             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2114             mac_learning_flush(ofproto->ml);
2115             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2116             mcast_snooping_mdb_flush(ofproto->ms);
2117         }
2118     }
2119 }
2120
2121 static void
2122 stp_wait(struct ofproto_dpif *ofproto)
2123 {
2124     if (ofproto->stp) {
2125         poll_timer_wait(1000);
2126     }
2127 }
2128 \f
2129 static int
2130 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2131            size_t n_qdscp)
2132 {
2133     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2134     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2135
2136     if (ofport->n_qdscp != n_qdscp
2137         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2138                               n_qdscp * sizeof *qdscp))) {
2139         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2140         free(ofport->qdscp);
2141         ofport->qdscp = n_qdscp
2142             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2143             : NULL;
2144         ofport->n_qdscp = n_qdscp;
2145     }
2146
2147     return 0;
2148 }
2149 \f
2150 /* Bundles. */
2151
2152 /* Expires all MAC learning entries associated with 'bundle' and forces its
2153  * ofproto to revalidate every flow.
2154  *
2155  * Normally MAC learning entries are removed only from the ofproto associated
2156  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2157  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2158  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2159  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2160  * with the host from which it migrated. */
2161 static void
2162 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2163 {
2164     struct ofproto_dpif *ofproto = bundle->ofproto;
2165     struct mac_learning *ml = ofproto->ml;
2166     struct mac_entry *mac, *next_mac;
2167
2168     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2169     ovs_rwlock_wrlock(&ml->rwlock);
2170     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2171         if (mac->port.p == bundle) {
2172             if (all_ofprotos) {
2173                 struct ofproto_dpif *o;
2174
2175                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2176                     if (o != ofproto) {
2177                         struct mac_entry *e;
2178
2179                         ovs_rwlock_wrlock(&o->ml->rwlock);
2180                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2181                         if (e) {
2182                             mac_learning_expire(o->ml, e);
2183                         }
2184                         ovs_rwlock_unlock(&o->ml->rwlock);
2185                     }
2186                 }
2187             }
2188
2189             mac_learning_expire(ml, mac);
2190         }
2191     }
2192     ovs_rwlock_unlock(&ml->rwlock);
2193 }
2194
2195 static struct ofbundle *
2196 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2197 {
2198     struct ofbundle *bundle;
2199
2200     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2201                              &ofproto->bundles) {
2202         if (bundle->aux == aux) {
2203             return bundle;
2204         }
2205     }
2206     return NULL;
2207 }
2208
2209 static void
2210 bundle_update(struct ofbundle *bundle)
2211 {
2212     struct ofport_dpif *port;
2213
2214     bundle->floodable = true;
2215     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2216         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2217             || port->is_layer3
2218             || !stp_forward_in_state(port->stp_state)) {
2219             bundle->floodable = false;
2220             break;
2221         }
2222     }
2223 }
2224
2225 static void
2226 bundle_del_port(struct ofport_dpif *port)
2227 {
2228     struct ofbundle *bundle = port->bundle;
2229
2230     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2231
2232     list_remove(&port->bundle_node);
2233     port->bundle = NULL;
2234
2235     if (bundle->lacp) {
2236         lacp_slave_unregister(bundle->lacp, port);
2237     }
2238     if (bundle->bond) {
2239         bond_slave_unregister(bundle->bond, port);
2240     }
2241
2242     bundle_update(bundle);
2243 }
2244
2245 static bool
2246 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2247                 struct lacp_slave_settings *lacp)
2248 {
2249     struct ofport_dpif *port;
2250
2251     port = get_ofp_port(bundle->ofproto, ofp_port);
2252     if (!port) {
2253         return false;
2254     }
2255
2256     if (port->bundle != bundle) {
2257         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2258         if (port->bundle) {
2259             bundle_remove(&port->up);
2260         }
2261
2262         port->bundle = bundle;
2263         list_push_back(&bundle->ports, &port->bundle_node);
2264         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2265             || port->is_layer3
2266             || !stp_forward_in_state(port->stp_state)) {
2267             bundle->floodable = false;
2268         }
2269     }
2270     if (lacp) {
2271         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2272         lacp_slave_register(bundle->lacp, port, lacp);
2273     }
2274
2275     return true;
2276 }
2277
2278 static void
2279 bundle_destroy(struct ofbundle *bundle)
2280 {
2281     struct ofproto_dpif *ofproto;
2282     struct ofport_dpif *port, *next_port;
2283
2284     if (!bundle) {
2285         return;
2286     }
2287
2288     ofproto = bundle->ofproto;
2289     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2290
2291     xlate_txn_start();
2292     xlate_bundle_remove(bundle);
2293     xlate_txn_commit();
2294
2295     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2296         bundle_del_port(port);
2297     }
2298
2299     bundle_flush_macs(bundle, true);
2300     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2301     free(bundle->name);
2302     free(bundle->trunks);
2303     lacp_unref(bundle->lacp);
2304     bond_unref(bundle->bond);
2305     free(bundle);
2306 }
2307
2308 static int
2309 bundle_set(struct ofproto *ofproto_, void *aux,
2310            const struct ofproto_bundle_settings *s)
2311 {
2312     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2313     bool need_flush = false;
2314     struct ofport_dpif *port;
2315     struct ofbundle *bundle;
2316     unsigned long *trunks;
2317     int vlan;
2318     size_t i;
2319     bool ok;
2320
2321     if (!s) {
2322         bundle_destroy(bundle_lookup(ofproto, aux));
2323         return 0;
2324     }
2325
2326     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2327     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2328
2329     bundle = bundle_lookup(ofproto, aux);
2330     if (!bundle) {
2331         bundle = xmalloc(sizeof *bundle);
2332
2333         bundle->ofproto = ofproto;
2334         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2335                     hash_pointer(aux, 0));
2336         bundle->aux = aux;
2337         bundle->name = NULL;
2338
2339         list_init(&bundle->ports);
2340         bundle->vlan_mode = PORT_VLAN_TRUNK;
2341         bundle->vlan = -1;
2342         bundle->trunks = NULL;
2343         bundle->use_priority_tags = s->use_priority_tags;
2344         bundle->lacp = NULL;
2345         bundle->bond = NULL;
2346
2347         bundle->floodable = true;
2348         mbridge_register_bundle(ofproto->mbridge, bundle);
2349     }
2350
2351     if (!bundle->name || strcmp(s->name, bundle->name)) {
2352         free(bundle->name);
2353         bundle->name = xstrdup(s->name);
2354     }
2355
2356     /* LACP. */
2357     if (s->lacp) {
2358         ofproto->lacp_enabled = true;
2359         if (!bundle->lacp) {
2360             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2361             bundle->lacp = lacp_create();
2362         }
2363         lacp_configure(bundle->lacp, s->lacp);
2364     } else {
2365         lacp_unref(bundle->lacp);
2366         bundle->lacp = NULL;
2367     }
2368
2369     /* Update set of ports. */
2370     ok = true;
2371     for (i = 0; i < s->n_slaves; i++) {
2372         if (!bundle_add_port(bundle, s->slaves[i],
2373                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2374             ok = false;
2375         }
2376     }
2377     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2378         struct ofport_dpif *next_port;
2379
2380         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2381             for (i = 0; i < s->n_slaves; i++) {
2382                 if (s->slaves[i] == port->up.ofp_port) {
2383                     goto found;
2384                 }
2385             }
2386
2387             bundle_del_port(port);
2388         found: ;
2389         }
2390     }
2391     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2392
2393     if (list_is_empty(&bundle->ports)) {
2394         bundle_destroy(bundle);
2395         return EINVAL;
2396     }
2397
2398     /* Set VLAN tagging mode */
2399     if (s->vlan_mode != bundle->vlan_mode
2400         || s->use_priority_tags != bundle->use_priority_tags) {
2401         bundle->vlan_mode = s->vlan_mode;
2402         bundle->use_priority_tags = s->use_priority_tags;
2403         need_flush = true;
2404     }
2405
2406     /* Set VLAN tag. */
2407     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2408             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2409             : 0);
2410     if (vlan != bundle->vlan) {
2411         bundle->vlan = vlan;
2412         need_flush = true;
2413     }
2414
2415     /* Get trunked VLANs. */
2416     switch (s->vlan_mode) {
2417     case PORT_VLAN_ACCESS:
2418         trunks = NULL;
2419         break;
2420
2421     case PORT_VLAN_TRUNK:
2422         trunks = CONST_CAST(unsigned long *, s->trunks);
2423         break;
2424
2425     case PORT_VLAN_NATIVE_UNTAGGED:
2426     case PORT_VLAN_NATIVE_TAGGED:
2427         if (vlan != 0 && (!s->trunks
2428                           || !bitmap_is_set(s->trunks, vlan)
2429                           || bitmap_is_set(s->trunks, 0))) {
2430             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2431             if (s->trunks) {
2432                 trunks = bitmap_clone(s->trunks, 4096);
2433             } else {
2434                 trunks = bitmap_allocate1(4096);
2435             }
2436             bitmap_set1(trunks, vlan);
2437             bitmap_set0(trunks, 0);
2438         } else {
2439             trunks = CONST_CAST(unsigned long *, s->trunks);
2440         }
2441         break;
2442
2443     default:
2444         OVS_NOT_REACHED();
2445     }
2446     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2447         free(bundle->trunks);
2448         if (trunks == s->trunks) {
2449             bundle->trunks = vlan_bitmap_clone(trunks);
2450         } else {
2451             bundle->trunks = trunks;
2452             trunks = NULL;
2453         }
2454         need_flush = true;
2455     }
2456     if (trunks != s->trunks) {
2457         free(trunks);
2458     }
2459
2460     /* Bonding. */
2461     if (!list_is_short(&bundle->ports)) {
2462         bundle->ofproto->has_bonded_bundles = true;
2463         if (bundle->bond) {
2464             if (bond_reconfigure(bundle->bond, s->bond)) {
2465                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2466             }
2467         } else {
2468             bundle->bond = bond_create(s->bond, ofproto);
2469             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2470         }
2471
2472         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2473             bond_slave_register(bundle->bond, port,
2474                                 port->up.ofp_port, port->up.netdev);
2475         }
2476     } else {
2477         bond_unref(bundle->bond);
2478         bundle->bond = NULL;
2479     }
2480
2481     /* If we changed something that would affect MAC learning, un-learn
2482      * everything on this port and force flow revalidation. */
2483     if (need_flush) {
2484         bundle_flush_macs(bundle, false);
2485     }
2486
2487     return 0;
2488 }
2489
2490 static void
2491 bundle_remove(struct ofport *port_)
2492 {
2493     struct ofport_dpif *port = ofport_dpif_cast(port_);
2494     struct ofbundle *bundle = port->bundle;
2495
2496     if (bundle) {
2497         bundle_del_port(port);
2498         if (list_is_empty(&bundle->ports)) {
2499             bundle_destroy(bundle);
2500         } else if (list_is_short(&bundle->ports)) {
2501             bond_unref(bundle->bond);
2502             bundle->bond = NULL;
2503         }
2504     }
2505 }
2506
2507 static void
2508 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2509 {
2510     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2511     struct ofport_dpif *port = port_;
2512     uint8_t ea[ETH_ADDR_LEN];
2513     int error;
2514
2515     error = netdev_get_etheraddr(port->up.netdev, ea);
2516     if (!error) {
2517         struct ofpbuf packet;
2518         void *packet_pdu;
2519
2520         ofpbuf_init(&packet, 0);
2521         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2522                                  pdu_size);
2523         memcpy(packet_pdu, pdu, pdu_size);
2524
2525         ofproto_dpif_send_packet(port, &packet);
2526         ofpbuf_uninit(&packet);
2527     } else {
2528         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2529                     "%s (%s)", port->bundle->name,
2530                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2531     }
2532 }
2533
2534 static void
2535 bundle_send_learning_packets(struct ofbundle *bundle)
2536 {
2537     struct ofproto_dpif *ofproto = bundle->ofproto;
2538     struct ofpbuf *learning_packet;
2539     int error, n_packets, n_errors;
2540     struct mac_entry *e;
2541     struct list packets;
2542
2543     list_init(&packets);
2544     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2545     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2546         if (e->port.p != bundle) {
2547             void *port_void;
2548
2549             learning_packet = bond_compose_learning_packet(bundle->bond,
2550                                                            e->mac, e->vlan,
2551                                                            &port_void);
2552             /* Temporarily use 'frame' as a private pointer (see below). */
2553             ovs_assert(learning_packet->frame == ofpbuf_data(learning_packet));
2554             learning_packet->frame = port_void;
2555             list_push_back(&packets, &learning_packet->list_node);
2556         }
2557     }
2558     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2559
2560     error = n_packets = n_errors = 0;
2561     LIST_FOR_EACH (learning_packet, list_node, &packets) {
2562         int ret;
2563         void *port_void = learning_packet->frame;
2564
2565         /* Restore 'frame'. */
2566         learning_packet->frame = ofpbuf_data(learning_packet);
2567         ret = ofproto_dpif_send_packet(port_void, learning_packet);
2568         if (ret) {
2569             error = ret;
2570             n_errors++;
2571         }
2572         n_packets++;
2573     }
2574     ofpbuf_list_delete(&packets);
2575
2576     if (n_errors) {
2577         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2578         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2579                      "packets, last error was: %s",
2580                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2581     } else {
2582         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2583                  bundle->name, n_packets);
2584     }
2585 }
2586
2587 static void
2588 bundle_run(struct ofbundle *bundle)
2589 {
2590     if (bundle->lacp) {
2591         lacp_run(bundle->lacp, send_pdu_cb);
2592     }
2593     if (bundle->bond) {
2594         struct ofport_dpif *port;
2595
2596         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2597             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2598         }
2599
2600         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2601             bundle->ofproto->backer->need_revalidate = REV_BOND;
2602         }
2603
2604         if (bond_should_send_learning_packets(bundle->bond)) {
2605             bundle_send_learning_packets(bundle);
2606         }
2607     }
2608 }
2609
2610 static void
2611 bundle_wait(struct ofbundle *bundle)
2612 {
2613     if (bundle->lacp) {
2614         lacp_wait(bundle->lacp);
2615     }
2616     if (bundle->bond) {
2617         bond_wait(bundle->bond);
2618     }
2619 }
2620 \f
2621 /* Mirrors. */
2622
2623 static int
2624 mirror_set__(struct ofproto *ofproto_, void *aux,
2625              const struct ofproto_mirror_settings *s)
2626 {
2627     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2628     struct ofbundle **srcs, **dsts;
2629     int error;
2630     size_t i;
2631
2632     if (!s) {
2633         mirror_destroy(ofproto->mbridge, aux);
2634         return 0;
2635     }
2636
2637     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2638     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2639
2640     for (i = 0; i < s->n_srcs; i++) {
2641         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2642     }
2643
2644     for (i = 0; i < s->n_dsts; i++) {
2645         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2646     }
2647
2648     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2649                        s->n_dsts, s->src_vlans,
2650                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2651     free(srcs);
2652     free(dsts);
2653     return error;
2654 }
2655
2656 static int
2657 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2658                    uint64_t *packets, uint64_t *bytes)
2659 {
2660     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2661                             bytes);
2662 }
2663
2664 static int
2665 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2666 {
2667     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2668     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2669     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2670         mac_learning_flush(ofproto->ml);
2671     }
2672     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2673     return 0;
2674 }
2675
2676 static bool
2677 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2678 {
2679     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2680     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2681     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2682 }
2683
2684 static void
2685 forward_bpdu_changed(struct ofproto *ofproto_)
2686 {
2687     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2688     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2689 }
2690
2691 static void
2692 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2693                      size_t max_entries)
2694 {
2695     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2696     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2697     mac_learning_set_idle_time(ofproto->ml, idle_time);
2698     mac_learning_set_max_entries(ofproto->ml, max_entries);
2699     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2700 }
2701
2702 /* Configures multicast snooping on 'ofport' using the settings
2703  * defined in 's'. */
2704 static int
2705 set_mcast_snooping(struct ofproto *ofproto_,
2706                    const struct ofproto_mcast_snooping_settings *s)
2707 {
2708     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2709
2710     /* Only revalidate flows if the configuration changed. */
2711     if (!s != !ofproto->ms) {
2712         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2713     }
2714
2715     if (s) {
2716         if (!ofproto->ms) {
2717             ofproto->ms = mcast_snooping_create();
2718         }
2719
2720         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
2721         mcast_snooping_set_idle_time(ofproto->ms, s->idle_time);
2722         mcast_snooping_set_max_entries(ofproto->ms, s->max_entries);
2723         if (mcast_snooping_set_flood_unreg(ofproto->ms, s->flood_unreg)) {
2724             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2725         }
2726         ovs_rwlock_unlock(&ofproto->ms->rwlock);
2727     } else {
2728         mcast_snooping_unref(ofproto->ms);
2729         ofproto->ms = NULL;
2730     }
2731
2732     return 0;
2733 }
2734
2735 /* Configures multicast snooping port's flood setting on 'ofproto'. */
2736 static int
2737 set_mcast_snooping_port(struct ofproto *ofproto_, void *aux, bool flood)
2738 {
2739     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2740     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2741
2742     if (ofproto->ms) {
2743         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
2744         mcast_snooping_set_port_flood(ofproto->ms, bundle->vlan, bundle,
2745                                       flood);
2746         ovs_rwlock_unlock(&ofproto->ms->rwlock);
2747     }
2748     return 0;
2749 }
2750
2751 \f
2752 /* Ports. */
2753
2754 static struct ofport_dpif *
2755 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2756 {
2757     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2758     return ofport ? ofport_dpif_cast(ofport) : NULL;
2759 }
2760
2761 static void
2762 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2763                             struct ofproto_port *ofproto_port,
2764                             struct dpif_port *dpif_port)
2765 {
2766     ofproto_port->name = dpif_port->name;
2767     ofproto_port->type = dpif_port->type;
2768     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2769 }
2770
2771 static void
2772 ofport_update_peer(struct ofport_dpif *ofport)
2773 {
2774     const struct ofproto_dpif *ofproto;
2775     struct dpif_backer *backer;
2776     char *peer_name;
2777
2778     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2779         return;
2780     }
2781
2782     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2783     backer->need_revalidate = REV_RECONFIGURE;
2784
2785     if (ofport->peer) {
2786         ofport->peer->peer = NULL;
2787         ofport->peer = NULL;
2788     }
2789
2790     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2791     if (!peer_name) {
2792         return;
2793     }
2794
2795     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2796         struct ofport *peer_ofport;
2797         struct ofport_dpif *peer;
2798         char *peer_peer;
2799
2800         if (ofproto->backer != backer) {
2801             continue;
2802         }
2803
2804         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2805         if (!peer_ofport) {
2806             continue;
2807         }
2808
2809         peer = ofport_dpif_cast(peer_ofport);
2810         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2811         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2812                                  peer_peer)) {
2813             ofport->peer = peer;
2814             ofport->peer->peer = ofport;
2815         }
2816         free(peer_peer);
2817
2818         break;
2819     }
2820     free(peer_name);
2821 }
2822
2823 static void
2824 port_run(struct ofport_dpif *ofport)
2825 {
2826     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2827     bool carrier_changed = carrier_seq != ofport->carrier_seq;
2828     bool enable = netdev_get_carrier(ofport->up.netdev);
2829     bool cfm_enable = false;
2830     bool bfd_enable = false;
2831
2832     ofport->carrier_seq = carrier_seq;
2833
2834     if (ofport->cfm) {
2835         int cfm_opup = cfm_get_opup(ofport->cfm);
2836
2837         cfm_enable = !cfm_get_fault(ofport->cfm);
2838
2839         if (cfm_opup >= 0) {
2840             cfm_enable = cfm_enable && cfm_opup;
2841         }
2842     }
2843
2844     if (ofport->bfd) {
2845         bfd_enable = bfd_forwarding(ofport->bfd);
2846     }
2847
2848     if (ofport->bfd || ofport->cfm) {
2849         enable = enable && (cfm_enable || bfd_enable);
2850     }
2851
2852     if (ofport->bundle) {
2853         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2854         if (carrier_changed) {
2855             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2856         }
2857     }
2858
2859     if (ofport->may_enable != enable) {
2860         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2861         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2862     }
2863
2864     ofport->may_enable = enable;
2865 }
2866
2867 static int
2868 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2869                    struct ofproto_port *ofproto_port)
2870 {
2871     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2872     struct dpif_port dpif_port;
2873     int error;
2874
2875     if (sset_contains(&ofproto->ghost_ports, devname)) {
2876         const char *type = netdev_get_type_from_name(devname);
2877
2878         /* We may be called before ofproto->up.port_by_name is populated with
2879          * the appropriate ofport.  For this reason, we must get the name and
2880          * type from the netdev layer directly. */
2881         if (type) {
2882             const struct ofport *ofport;
2883
2884             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2885             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2886             ofproto_port->name = xstrdup(devname);
2887             ofproto_port->type = xstrdup(type);
2888             return 0;
2889         }
2890         return ENODEV;
2891     }
2892
2893     if (!sset_contains(&ofproto->ports, devname)) {
2894         return ENODEV;
2895     }
2896     error = dpif_port_query_by_name(ofproto->backer->dpif,
2897                                     devname, &dpif_port);
2898     if (!error) {
2899         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2900     }
2901     return error;
2902 }
2903
2904 static int
2905 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2906 {
2907     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2908     const char *devname = netdev_get_name(netdev);
2909     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2910     const char *dp_port_name;
2911
2912     if (netdev_vport_is_patch(netdev)) {
2913         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2914         return 0;
2915     }
2916
2917     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2918     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2919         odp_port_t port_no = ODPP_NONE;
2920         int error;
2921
2922         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2923         if (error) {
2924             return error;
2925         }
2926         if (netdev_get_tunnel_config(netdev)) {
2927             simap_put(&ofproto->backer->tnl_backers,
2928                       dp_port_name, odp_to_u32(port_no));
2929         }
2930     }
2931
2932     if (netdev_get_tunnel_config(netdev)) {
2933         sset_add(&ofproto->ghost_ports, devname);
2934     } else {
2935         sset_add(&ofproto->ports, devname);
2936     }
2937     return 0;
2938 }
2939
2940 static int
2941 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
2942 {
2943     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2944     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2945     int error = 0;
2946
2947     if (!ofport) {
2948         return 0;
2949     }
2950
2951     sset_find_and_delete(&ofproto->ghost_ports,
2952                          netdev_get_name(ofport->up.netdev));
2953     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2954     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
2955         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
2956         if (!error) {
2957             /* The caller is going to close ofport->up.netdev.  If this is a
2958              * bonded port, then the bond is using that netdev, so remove it
2959              * from the bond.  The client will need to reconfigure everything
2960              * after deleting ports, so then the slave will get re-added. */
2961             bundle_remove(&ofport->up);
2962         }
2963     }
2964     return error;
2965 }
2966
2967 static int
2968 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2969 {
2970     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2971     int error;
2972
2973     error = netdev_get_stats(ofport->up.netdev, stats);
2974
2975     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
2976         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2977
2978         ovs_mutex_lock(&ofproto->stats_mutex);
2979         /* ofproto->stats.tx_packets represents packets that we created
2980          * internally and sent to some port (e.g. packets sent with
2981          * ofproto_dpif_send_packet()).  Account for them as if they had
2982          * come from OFPP_LOCAL and got forwarded. */
2983
2984         if (stats->rx_packets != UINT64_MAX) {
2985             stats->rx_packets += ofproto->stats.tx_packets;
2986         }
2987
2988         if (stats->rx_bytes != UINT64_MAX) {
2989             stats->rx_bytes += ofproto->stats.tx_bytes;
2990         }
2991
2992         /* ofproto->stats.rx_packets represents packets that were received on
2993          * some port and we processed internally and dropped (e.g. STP).
2994          * Account for them as if they had been forwarded to OFPP_LOCAL. */
2995
2996         if (stats->tx_packets != UINT64_MAX) {
2997             stats->tx_packets += ofproto->stats.rx_packets;
2998         }
2999
3000         if (stats->tx_bytes != UINT64_MAX) {
3001             stats->tx_bytes += ofproto->stats.rx_bytes;
3002         }
3003         ovs_mutex_unlock(&ofproto->stats_mutex);
3004     }
3005
3006     return error;
3007 }
3008
3009 struct port_dump_state {
3010     uint32_t bucket;
3011     uint32_t offset;
3012     bool ghost;
3013
3014     struct ofproto_port port;
3015     bool has_port;
3016 };
3017
3018 static int
3019 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3020 {
3021     *statep = xzalloc(sizeof(struct port_dump_state));
3022     return 0;
3023 }
3024
3025 static int
3026 port_dump_next(const struct ofproto *ofproto_, void *state_,
3027                struct ofproto_port *port)
3028 {
3029     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3030     struct port_dump_state *state = state_;
3031     const struct sset *sset;
3032     struct sset_node *node;
3033
3034     if (state->has_port) {
3035         ofproto_port_destroy(&state->port);
3036         state->has_port = false;
3037     }
3038     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3039     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3040         int error;
3041
3042         error = port_query_by_name(ofproto_, node->name, &state->port);
3043         if (!error) {
3044             *port = state->port;
3045             state->has_port = true;
3046             return 0;
3047         } else if (error != ENODEV) {
3048             return error;
3049         }
3050     }
3051
3052     if (!state->ghost) {
3053         state->ghost = true;
3054         state->bucket = 0;
3055         state->offset = 0;
3056         return port_dump_next(ofproto_, state_, port);
3057     }
3058
3059     return EOF;
3060 }
3061
3062 static int
3063 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3064 {
3065     struct port_dump_state *state = state_;
3066
3067     if (state->has_port) {
3068         ofproto_port_destroy(&state->port);
3069     }
3070     free(state);
3071     return 0;
3072 }
3073
3074 static int
3075 port_poll(const struct ofproto *ofproto_, char **devnamep)
3076 {
3077     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3078
3079     if (ofproto->port_poll_errno) {
3080         int error = ofproto->port_poll_errno;
3081         ofproto->port_poll_errno = 0;
3082         return error;
3083     }
3084
3085     if (sset_is_empty(&ofproto->port_poll_set)) {
3086         return EAGAIN;
3087     }
3088
3089     *devnamep = sset_pop(&ofproto->port_poll_set);
3090     return 0;
3091 }
3092
3093 static void
3094 port_poll_wait(const struct ofproto *ofproto_)
3095 {
3096     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3097     dpif_port_poll_wait(ofproto->backer->dpif);
3098 }
3099
3100 static int
3101 port_is_lacp_current(const struct ofport *ofport_)
3102 {
3103     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3104     return (ofport->bundle && ofport->bundle->lacp
3105             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3106             : -1);
3107 }
3108 \f
3109 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3110  * then delete it entirely. */
3111 static void
3112 rule_expire(struct rule_dpif *rule)
3113     OVS_REQUIRES(ofproto_mutex)
3114 {
3115     uint16_t hard_timeout, idle_timeout;
3116     long long int now = time_msec();
3117     int reason = -1;
3118
3119     hard_timeout = rule->up.hard_timeout;
3120     idle_timeout = rule->up.idle_timeout;
3121
3122     /* Has 'rule' expired? */
3123     if (hard_timeout) {
3124         long long int modified;
3125
3126         ovs_mutex_lock(&rule->up.mutex);
3127         modified = rule->up.modified;
3128         ovs_mutex_unlock(&rule->up.mutex);
3129
3130         if (now > modified + hard_timeout * 1000) {
3131             reason = OFPRR_HARD_TIMEOUT;
3132         }
3133     }
3134
3135     if (reason < 0 && idle_timeout) {
3136         long long int used;
3137
3138         ovs_mutex_lock(&rule->stats_mutex);
3139         used = rule->stats.used;
3140         ovs_mutex_unlock(&rule->stats_mutex);
3141
3142         if (now > used + idle_timeout * 1000) {
3143             reason = OFPRR_IDLE_TIMEOUT;
3144         }
3145     }
3146
3147     if (reason >= 0) {
3148         COVERAGE_INC(ofproto_dpif_expired);
3149         ofproto_rule_expire(&rule->up, reason);
3150     }
3151 }
3152
3153 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3154  * 'flow' must reflect the data in 'packet'. */
3155 int
3156 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3157                              const struct flow *flow,
3158                              struct rule_dpif *rule,
3159                              const struct ofpact *ofpacts, size_t ofpacts_len,
3160                              struct ofpbuf *packet)
3161 {
3162     struct dpif_flow_stats stats;
3163     struct xlate_out xout;
3164     struct xlate_in xin;
3165     ofp_port_t in_port;
3166     struct dpif_execute execute;
3167     int error;
3168
3169     ovs_assert((rule != NULL) != (ofpacts != NULL));
3170
3171     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3172
3173     if (rule) {
3174         rule_dpif_credit_stats(rule, &stats);
3175     }
3176
3177     xlate_in_init(&xin, ofproto, flow, flow->in_port.ofp_port, rule,
3178                   stats.tcp_flags, packet);
3179     xin.ofpacts = ofpacts;
3180     xin.ofpacts_len = ofpacts_len;
3181     xin.resubmit_stats = &stats;
3182     xlate_actions(&xin, &xout);
3183
3184     execute.actions = ofpbuf_data(xout.odp_actions);
3185     execute.actions_len = ofpbuf_size(xout.odp_actions);
3186     execute.packet = packet;
3187     execute.md = pkt_metadata_from_flow(flow);
3188     execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3189
3190     /* Fix up in_port. */
3191     in_port = flow->in_port.ofp_port;
3192     if (in_port == OFPP_NONE) {
3193         in_port = OFPP_LOCAL;
3194     }
3195     execute.md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3196
3197     error = dpif_execute(ofproto->backer->dpif, &execute);
3198
3199     xlate_out_uninit(&xout);
3200
3201     return error;
3202 }
3203
3204 void
3205 rule_dpif_credit_stats(struct rule_dpif *rule,
3206                        const struct dpif_flow_stats *stats)
3207 {
3208     ovs_mutex_lock(&rule->stats_mutex);
3209     rule->stats.n_packets += stats->n_packets;
3210     rule->stats.n_bytes += stats->n_bytes;
3211     rule->stats.used = MAX(rule->stats.used, stats->used);
3212     ovs_mutex_unlock(&rule->stats_mutex);
3213 }
3214
3215 ovs_be64
3216 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3217     OVS_REQUIRES(rule->up.mutex)
3218 {
3219     return rule->up.flow_cookie;
3220 }
3221
3222 void
3223 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3224                      uint16_t hard_timeout)
3225 {
3226     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3227 }
3228
3229 /* Returns 'rule''s actions.  The caller owns a reference on the returned
3230  * actions and must eventually release it (with rule_actions_unref()) to avoid
3231  * a memory leak. */
3232 const struct rule_actions *
3233 rule_dpif_get_actions(const struct rule_dpif *rule)
3234 {
3235     return rule_get_actions(&rule->up);
3236 }
3237
3238 /* Sets 'rule''s recirculation id. */
3239 static void
3240 rule_dpif_set_recirc_id(struct rule_dpif *rule, uint32_t id)
3241     OVS_REQUIRES(rule->up.mutex)
3242 {
3243     ovs_assert(!rule->recirc_id);
3244     rule->recirc_id = id;
3245 }
3246
3247 /* Returns 'rule''s recirculation id. */
3248 uint32_t
3249 rule_dpif_get_recirc_id(struct rule_dpif *rule)
3250     OVS_REQUIRES(rule->up.mutex)
3251 {
3252     if (!rule->recirc_id) {
3253         struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3254
3255         rule_dpif_set_recirc_id(rule, ofproto_dpif_alloc_recirc_id(ofproto));
3256     }
3257     return rule->recirc_id;
3258 }
3259
3260 /* Sets 'rule''s recirculation id. */
3261 void
3262 rule_set_recirc_id(struct rule *rule_, uint32_t id)
3263 {
3264     struct rule_dpif *rule = rule_dpif_cast(rule_);
3265
3266     ovs_mutex_lock(&rule->up.mutex);
3267     rule_dpif_set_recirc_id(rule, id);
3268     ovs_mutex_unlock(&rule->up.mutex);
3269 }
3270
3271 /* Lookup 'flow' in table 0 of 'ofproto''s classifier.
3272  * If 'wc' is non-null, sets the fields that were relevant as part of
3273  * the lookup. Returns the table_id where a match or miss occurred.
3274  *
3275  * The return value will be zero unless there was a miss and
3276  * OFPTC11_TABLE_MISS_CONTINUE is in effect for the sequence of tables
3277  * where misses occur.
3278  *
3279  * The rule is returned in '*rule', which is valid at least until the next
3280  * RCU quiescent period.  If the '*rule' needs to stay around longer,
3281  * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3282  * on it before this returns. */
3283 uint8_t
3284 rule_dpif_lookup(struct ofproto_dpif *ofproto, struct flow *flow,
3285                  struct flow_wildcards *wc, struct rule_dpif **rule,
3286                  bool take_ref, const struct dpif_flow_stats *stats)
3287 {
3288     enum rule_dpif_lookup_verdict verdict;
3289     enum ofputil_port_config config = 0;
3290     uint8_t table_id;
3291
3292     if (ofproto_dpif_get_enable_recirc(ofproto)) {
3293         /* Always exactly match recirc_id since datapath supports
3294          * recirculation.  */
3295         if (wc) {
3296             wc->masks.recirc_id = UINT32_MAX;
3297         }
3298
3299         /* Start looking up from internal table for post recirculation flows
3300          * or packets. We can also simply send all, including normal flows
3301          * or packets to the internal table. They will not match any post
3302          * recirculation rules except the 'catch all' rule that resubmit
3303          * them to table 0.
3304          *
3305          * As an optimization, we send normal flows and packets to table 0
3306          * directly, saving one table lookup.  */
3307         table_id = flow->recirc_id ? TBL_INTERNAL : 0;
3308     } else {
3309         table_id = 0;
3310     }
3311
3312     verdict = rule_dpif_lookup_from_table(ofproto, flow, wc, true,
3313                                           &table_id, rule, take_ref, stats);
3314
3315     switch (verdict) {
3316     case RULE_DPIF_LOOKUP_VERDICT_MATCH:
3317         return table_id;
3318     case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER: {
3319         struct ofport_dpif *port;
3320
3321         port = get_ofp_port(ofproto, flow->in_port.ofp_port);
3322         if (!port) {
3323             VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3324                          flow->in_port.ofp_port);
3325         }
3326         config = port ? port->up.pp.config : 0;
3327         break;
3328     }
3329     case RULE_DPIF_LOOKUP_VERDICT_DROP:
3330         config = OFPUTIL_PC_NO_PACKET_IN;
3331         break;
3332     case RULE_DPIF_LOOKUP_VERDICT_DEFAULT:
3333         if (!connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3334             config = OFPUTIL_PC_NO_PACKET_IN;
3335         }
3336         break;
3337     default:
3338         OVS_NOT_REACHED();
3339     }
3340
3341     choose_miss_rule(config, ofproto->miss_rule,
3342                      ofproto->no_packet_in_rule, rule, take_ref);
3343     return table_id;
3344 }
3345
3346 /* The returned rule is valid at least until the next RCU quiescent period.
3347  * If the '*rule' needs to stay around longer, a non-zero 'take_ref' must be
3348  * passed in to cause a reference to be taken on it before this returns. */
3349 static struct rule_dpif *
3350 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, uint8_t table_id,
3351                           const struct flow *flow, struct flow_wildcards *wc,
3352                           bool take_ref)
3353 {
3354     struct classifier *cls = &ofproto->up.tables[table_id].cls;
3355     const struct cls_rule *cls_rule;
3356     struct rule_dpif *rule;
3357     struct flow ofpc_normal_flow;
3358
3359     if (ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3360         /* We always unwildcard dl_type and nw_frag (for IP), so they
3361          * need not be unwildcarded here. */
3362
3363         if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
3364             if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3365                 /* We must pretend that transport ports are unavailable. */
3366                 ofpc_normal_flow = *flow;
3367                 ofpc_normal_flow.tp_src = htons(0);
3368                 ofpc_normal_flow.tp_dst = htons(0);
3369                 flow = &ofpc_normal_flow;
3370             } else {
3371                 /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM).
3372                  * Use the drop_frags_rule (which cannot disappear). */
3373                 cls_rule = &ofproto->drop_frags_rule->up.cr;
3374                 rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3375                 if (take_ref) {
3376                     rule_dpif_ref(rule);
3377                 }
3378                 return rule;
3379             }
3380         }
3381     }
3382
3383     do {
3384         cls_rule = classifier_lookup(cls, flow, wc);
3385
3386         rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3387
3388         /* Try again if the rule was released before we get the reference. */
3389     } while (rule && take_ref && !rule_dpif_try_ref(rule));
3390
3391     return rule;
3392 }
3393
3394 /* Look up 'flow' in 'ofproto''s classifier starting from table '*table_id'.
3395  * Stores the rule that was found in '*rule', or NULL if none was found.
3396  * Updates 'wc', if nonnull, to reflect the fields that were used during the
3397  * lookup.
3398  *
3399  * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3400  * if none is found then the table miss configuration for that table is
3401  * honored, which can result in additional lookups in other OpenFlow tables.
3402  * In this case the function updates '*table_id' to reflect the final OpenFlow
3403  * table that was searched.
3404  *
3405  * If 'honor_table_miss' is false, then only one table lookup occurs, in
3406  * '*table_id'.
3407  *
3408  * Returns:
3409  *
3410  *    - RULE_DPIF_LOOKUP_VERDICT_MATCH if a rule (in '*rule') was found.
3411  *
3412  *    - RULE_OFPTC_TABLE_MISS_CONTROLLER if no rule was found and either:
3413  *      + 'honor_table_miss' is false
3414  *      + a table miss configuration specified that the packet should be
3415  *        sent to the controller in this case.
3416  *
3417  *    - RULE_DPIF_LOOKUP_VERDICT_DROP if no rule was found, 'honor_table_miss'
3418  *      is true and a table miss configuration specified that the packet
3419  *      should be dropped in this case.
3420  *
3421  *    - RULE_DPIF_LOOKUP_VERDICT_DEFAULT if no rule was found,
3422  *      'honor_table_miss' is true and a table miss configuration has
3423  *      not been specified in this case.
3424  *
3425  * The rule is returned in '*rule', which is valid at least until the next
3426  * RCU quiescent period.  If the '*rule' needs to stay around longer,
3427  * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3428  * on it before this returns. */
3429 enum rule_dpif_lookup_verdict
3430 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3431                             const struct flow *flow,
3432                             struct flow_wildcards *wc,
3433                             bool honor_table_miss,
3434                             uint8_t *table_id, struct rule_dpif **rule,
3435                             bool take_ref, const struct dpif_flow_stats *stats)
3436 {
3437     uint8_t next_id;
3438
3439     for (next_id = *table_id;
3440          next_id < ofproto->up.n_tables;
3441          next_id++, next_id += (next_id == TBL_INTERNAL))
3442     {
3443         *table_id = next_id;
3444         *rule = rule_dpif_lookup_in_table(ofproto, *table_id, flow, wc,
3445                                           take_ref);
3446         if (stats) {
3447             struct oftable *tbl = &ofproto->up.tables[next_id];
3448             atomic_ulong *stat = *rule ? &tbl->n_matched : &tbl->n_missed;
3449             unsigned long orig;
3450             atomic_add(stat, stats->n_packets, &orig);
3451         }
3452         if (*rule) {
3453             return RULE_DPIF_LOOKUP_VERDICT_MATCH;
3454         } else if (!honor_table_miss) {
3455             return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3456         } else {
3457             switch (ofproto_table_get_miss_config(&ofproto->up, *table_id)) {
3458             case OFPUTIL_TABLE_MISS_CONTINUE:
3459                 break;
3460
3461             case OFPUTIL_TABLE_MISS_CONTROLLER:
3462                 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3463
3464             case OFPUTIL_TABLE_MISS_DROP:
3465                 return RULE_DPIF_LOOKUP_VERDICT_DROP;
3466
3467             case OFPUTIL_TABLE_MISS_DEFAULT:
3468                 return RULE_DPIF_LOOKUP_VERDICT_DEFAULT;
3469             }
3470         }
3471     }
3472
3473     return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3474 }
3475
3476 /* Given a port configuration (specified as zero if there's no port), chooses
3477  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
3478  * flow table miss.
3479  *
3480  * The rule is returned in '*rule', which is valid at least until the next
3481  * RCU quiescent period.  If the '*rule' needs to stay around longer,
3482  * a reference must be taken on it (rule_dpif_ref()).
3483  */
3484 void
3485 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
3486                  struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule,
3487                  bool take_ref)
3488 {
3489     *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
3490     if (take_ref) {
3491         rule_dpif_ref(*rule);
3492     }
3493 }
3494
3495 static void
3496 complete_operation(struct rule_dpif *rule)
3497     OVS_REQUIRES(ofproto_mutex)
3498 {
3499     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3500
3501     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3502 }
3503
3504 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3505 {
3506     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3507 }
3508
3509 static struct rule *
3510 rule_alloc(void)
3511 {
3512     struct rule_dpif *rule = xmalloc(sizeof *rule);
3513     return &rule->up;
3514 }
3515
3516 static void
3517 rule_dealloc(struct rule *rule_)
3518 {
3519     struct rule_dpif *rule = rule_dpif_cast(rule_);
3520     free(rule);
3521 }
3522
3523 static enum ofperr
3524 rule_construct(struct rule *rule_)
3525     OVS_NO_THREAD_SAFETY_ANALYSIS
3526 {
3527     struct rule_dpif *rule = rule_dpif_cast(rule_);
3528     ovs_mutex_init_adaptive(&rule->stats_mutex);
3529     rule->stats.n_packets = 0;
3530     rule->stats.n_bytes = 0;
3531     rule->stats.used = rule->up.modified;
3532     rule->recirc_id = 0;
3533
3534     return 0;
3535 }
3536
3537 static enum ofperr
3538 rule_insert(struct rule *rule_)
3539     OVS_REQUIRES(ofproto_mutex)
3540 {
3541     struct rule_dpif *rule = rule_dpif_cast(rule_);
3542     complete_operation(rule);
3543     return 0;
3544 }
3545
3546 static void
3547 rule_delete(struct rule *rule_)
3548     OVS_REQUIRES(ofproto_mutex)
3549 {
3550     struct rule_dpif *rule = rule_dpif_cast(rule_);
3551     complete_operation(rule);
3552 }
3553
3554 static void
3555 rule_destruct(struct rule *rule_)
3556 {
3557     struct rule_dpif *rule = rule_dpif_cast(rule_);
3558
3559     ovs_mutex_destroy(&rule->stats_mutex);
3560     if (rule->recirc_id) {
3561         struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3562
3563         ofproto_dpif_free_recirc_id(ofproto, rule->recirc_id);
3564     }
3565 }
3566
3567 static void
3568 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
3569                long long int *used)
3570 {
3571     struct rule_dpif *rule = rule_dpif_cast(rule_);
3572
3573     ovs_mutex_lock(&rule->stats_mutex);
3574     *packets = rule->stats.n_packets;
3575     *bytes = rule->stats.n_bytes;
3576     *used = rule->stats.used;
3577     ovs_mutex_unlock(&rule->stats_mutex);
3578 }
3579
3580 static void
3581 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
3582                   struct ofpbuf *packet)
3583 {
3584     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3585
3586     ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
3587 }
3588
3589 static enum ofperr
3590 rule_execute(struct rule *rule, const struct flow *flow,
3591              struct ofpbuf *packet)
3592 {
3593     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
3594     ofpbuf_delete(packet);
3595     return 0;
3596 }
3597
3598 static void
3599 rule_modify_actions(struct rule *rule_, bool reset_counters)
3600     OVS_REQUIRES(ofproto_mutex)
3601 {
3602     struct rule_dpif *rule = rule_dpif_cast(rule_);
3603
3604     if (reset_counters) {
3605         ovs_mutex_lock(&rule->stats_mutex);
3606         rule->stats.n_packets = 0;
3607         rule->stats.n_bytes = 0;
3608         ovs_mutex_unlock(&rule->stats_mutex);
3609     }
3610
3611     complete_operation(rule);
3612 }
3613
3614 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
3615 {
3616     return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
3617 }
3618
3619 static struct ofgroup *
3620 group_alloc(void)
3621 {
3622     struct group_dpif *group = xzalloc(sizeof *group);
3623     return &group->up;
3624 }
3625
3626 static void
3627 group_dealloc(struct ofgroup *group_)
3628 {
3629     struct group_dpif *group = group_dpif_cast(group_);
3630     free(group);
3631 }
3632
3633 static void
3634 group_construct_stats(struct group_dpif *group)
3635     OVS_REQUIRES(group->stats_mutex)
3636 {
3637     struct ofputil_bucket *bucket;
3638     const struct list *buckets;
3639
3640     group->packet_count = 0;
3641     group->byte_count = 0;
3642
3643     group_dpif_get_buckets(group, &buckets);
3644     LIST_FOR_EACH (bucket, list_node, buckets) {
3645         bucket->stats.packet_count = 0;
3646         bucket->stats.byte_count = 0;
3647     }
3648 }
3649
3650 void
3651 group_dpif_credit_stats(struct group_dpif *group,
3652                         struct ofputil_bucket *bucket,
3653                         const struct dpif_flow_stats *stats)
3654 {
3655     ovs_mutex_lock(&group->stats_mutex);
3656     group->packet_count += stats->n_packets;
3657     group->byte_count += stats->n_bytes;
3658     if (bucket) {
3659         bucket->stats.packet_count += stats->n_packets;
3660         bucket->stats.byte_count += stats->n_bytes;
3661     } else { /* Credit to all buckets */
3662         const struct list *buckets;
3663
3664         group_dpif_get_buckets(group, &buckets);
3665         LIST_FOR_EACH (bucket, list_node, buckets) {
3666             bucket->stats.packet_count += stats->n_packets;
3667             bucket->stats.byte_count += stats->n_bytes;
3668         }
3669     }
3670     ovs_mutex_unlock(&group->stats_mutex);
3671 }
3672
3673 static enum ofperr
3674 group_construct(struct ofgroup *group_)
3675 {
3676     struct group_dpif *group = group_dpif_cast(group_);
3677     const struct ofputil_bucket *bucket;
3678
3679     /* Prevent group chaining because our locking structure makes it hard to
3680      * implement deadlock-free.  (See xlate_group_resource_check().) */
3681     LIST_FOR_EACH (bucket, list_node, &group->up.buckets) {
3682         const struct ofpact *a;
3683
3684         OFPACT_FOR_EACH (a, bucket->ofpacts, bucket->ofpacts_len) {
3685             if (a->type == OFPACT_GROUP) {
3686                 return OFPERR_OFPGMFC_CHAINING_UNSUPPORTED;
3687             }
3688         }
3689     }
3690
3691     ovs_mutex_init_adaptive(&group->stats_mutex);
3692     ovs_mutex_lock(&group->stats_mutex);
3693     group_construct_stats(group);
3694     ovs_mutex_unlock(&group->stats_mutex);
3695     return 0;
3696 }
3697
3698 static void
3699 group_destruct(struct ofgroup *group_)
3700 {
3701     struct group_dpif *group = group_dpif_cast(group_);
3702     ovs_mutex_destroy(&group->stats_mutex);
3703 }
3704
3705 static enum ofperr
3706 group_modify(struct ofgroup *group_)
3707 {
3708     struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
3709
3710     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3711
3712     return 0;
3713 }
3714
3715 static enum ofperr
3716 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
3717 {
3718     struct group_dpif *group = group_dpif_cast(group_);
3719     struct ofputil_bucket *bucket;
3720     const struct list *buckets;
3721     struct bucket_counter *bucket_stats;
3722
3723     ovs_mutex_lock(&group->stats_mutex);
3724     ogs->packet_count = group->packet_count;
3725     ogs->byte_count = group->byte_count;
3726
3727     group_dpif_get_buckets(group, &buckets);
3728     bucket_stats = ogs->bucket_stats;
3729     LIST_FOR_EACH (bucket, list_node, buckets) {
3730         bucket_stats->packet_count = bucket->stats.packet_count;
3731         bucket_stats->byte_count = bucket->stats.byte_count;
3732         bucket_stats++;
3733     }
3734     ovs_mutex_unlock(&group->stats_mutex);
3735
3736     return 0;
3737 }
3738
3739 /* If the group exists, this function increments the groups's reference count.
3740  *
3741  * Make sure to call group_dpif_unref() after no longer needing to maintain
3742  * a reference to the group. */
3743 bool
3744 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
3745                   struct group_dpif **group)
3746 {
3747     struct ofgroup *ofgroup;
3748     bool found;
3749
3750     found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
3751     *group = found ?  group_dpif_cast(ofgroup) : NULL;
3752
3753     return found;
3754 }
3755
3756 void
3757 group_dpif_get_buckets(const struct group_dpif *group,
3758                        const struct list **buckets)
3759 {
3760     *buckets = &group->up.buckets;
3761 }
3762
3763 enum ofp11_group_type
3764 group_dpif_get_type(const struct group_dpif *group)
3765 {
3766     return group->up.type;
3767 }
3768 \f
3769 /* Sends 'packet' out 'ofport'.
3770  * May modify 'packet'.
3771  * Returns 0 if successful, otherwise a positive errno value. */
3772 int
3773 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3774 {
3775     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3776     int error;
3777
3778     error = xlate_send_packet(ofport, packet);
3779
3780     ovs_mutex_lock(&ofproto->stats_mutex);
3781     ofproto->stats.tx_packets++;
3782     ofproto->stats.tx_bytes += ofpbuf_size(packet);
3783     ovs_mutex_unlock(&ofproto->stats_mutex);
3784     return error;
3785 }
3786 \f
3787 static bool
3788 set_frag_handling(struct ofproto *ofproto_,
3789                   enum ofp_config_flags frag_handling)
3790 {
3791     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3792     if (frag_handling != OFPC_FRAG_REASM) {
3793         ofproto->backer->need_revalidate = REV_RECONFIGURE;
3794         return true;
3795     } else {
3796         return false;
3797     }
3798 }
3799
3800 static enum ofperr
3801 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
3802            const struct flow *flow,
3803            const struct ofpact *ofpacts, size_t ofpacts_len)
3804 {
3805     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3806
3807     ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
3808                                  ofpacts_len, packet);
3809     return 0;
3810 }
3811 \f
3812 /* NetFlow. */
3813
3814 static int
3815 set_netflow(struct ofproto *ofproto_,
3816             const struct netflow_options *netflow_options)
3817 {
3818     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3819
3820     if (netflow_options) {
3821         if (!ofproto->netflow) {
3822             ofproto->netflow = netflow_create();
3823             ofproto->backer->need_revalidate = REV_RECONFIGURE;
3824         }
3825         return netflow_set_options(ofproto->netflow, netflow_options);
3826     } else if (ofproto->netflow) {
3827         ofproto->backer->need_revalidate = REV_RECONFIGURE;
3828         netflow_unref(ofproto->netflow);
3829         ofproto->netflow = NULL;
3830     }
3831
3832     return 0;
3833 }
3834
3835 static void
3836 get_netflow_ids(const struct ofproto *ofproto_,
3837                 uint8_t *engine_type, uint8_t *engine_id)
3838 {
3839     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3840
3841     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
3842 }
3843 \f
3844 static struct ofproto_dpif *
3845 ofproto_dpif_lookup(const char *name)
3846 {
3847     struct ofproto_dpif *ofproto;
3848
3849     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
3850                              hash_string(name, 0), &all_ofproto_dpifs) {
3851         if (!strcmp(ofproto->up.name, name)) {
3852             return ofproto;
3853         }
3854     }
3855     return NULL;
3856 }
3857
3858 static void
3859 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
3860                           const char *argv[], void *aux OVS_UNUSED)
3861 {
3862     struct ofproto_dpif *ofproto;
3863
3864     if (argc > 1) {
3865         ofproto = ofproto_dpif_lookup(argv[1]);
3866         if (!ofproto) {
3867             unixctl_command_reply_error(conn, "no such bridge");
3868             return;
3869         }
3870         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3871         mac_learning_flush(ofproto->ml);
3872         ovs_rwlock_unlock(&ofproto->ml->rwlock);
3873     } else {
3874         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3875             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3876             mac_learning_flush(ofproto->ml);
3877             ovs_rwlock_unlock(&ofproto->ml->rwlock);
3878         }
3879     }
3880
3881     unixctl_command_reply(conn, "table successfully flushed");
3882 }
3883
3884 static void
3885 ofproto_unixctl_mcast_snooping_flush(struct unixctl_conn *conn, int argc,
3886                                      const char *argv[], void *aux OVS_UNUSED)
3887 {
3888     struct ofproto_dpif *ofproto;
3889
3890     if (argc > 1) {
3891         ofproto = ofproto_dpif_lookup(argv[1]);
3892         if (!ofproto) {
3893             unixctl_command_reply_error(conn, "no such bridge");
3894             return;
3895         }
3896
3897         if (!mcast_snooping_enabled(ofproto->ms)) {
3898             unixctl_command_reply_error(conn, "multicast snooping is disabled");
3899             return;
3900         }
3901         mcast_snooping_mdb_flush(ofproto->ms);
3902     } else {
3903         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3904             if (!mcast_snooping_enabled(ofproto->ms)) {
3905                 continue;
3906             }
3907             mcast_snooping_mdb_flush(ofproto->ms);
3908         }
3909     }
3910
3911     unixctl_command_reply(conn, "table successfully flushed");
3912 }
3913
3914 static struct ofport_dpif *
3915 ofbundle_get_a_port(const struct ofbundle *bundle)
3916 {
3917     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
3918                         bundle_node);
3919 }
3920
3921 static void
3922 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
3923                          const char *argv[], void *aux OVS_UNUSED)
3924 {
3925     struct ds ds = DS_EMPTY_INITIALIZER;
3926     const struct ofproto_dpif *ofproto;
3927     const struct mac_entry *e;
3928
3929     ofproto = ofproto_dpif_lookup(argv[1]);
3930     if (!ofproto) {
3931         unixctl_command_reply_error(conn, "no such bridge");
3932         return;
3933     }
3934
3935     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
3936     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3937     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3938         struct ofbundle *bundle = e->port.p;
3939         char name[OFP_MAX_PORT_NAME_LEN];
3940
3941         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
3942                                name, sizeof name);
3943         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
3944                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
3945                       mac_entry_age(ofproto->ml, e));
3946     }
3947     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3948     unixctl_command_reply(conn, ds_cstr(&ds));
3949     ds_destroy(&ds);
3950 }
3951
3952 static void
3953 ofproto_unixctl_mcast_snooping_show(struct unixctl_conn *conn,
3954                                     int argc OVS_UNUSED,
3955                                     const char *argv[],
3956                                     void *aux OVS_UNUSED)
3957 {
3958     struct ds ds = DS_EMPTY_INITIALIZER;
3959     const struct ofproto_dpif *ofproto;
3960     const struct ofbundle *bundle;
3961     const struct mcast_group *grp;
3962     struct mcast_group_bundle *b;
3963     struct mcast_mrouter_bundle *mrouter;
3964
3965     ofproto = ofproto_dpif_lookup(argv[1]);
3966     if (!ofproto) {
3967         unixctl_command_reply_error(conn, "no such bridge");
3968         return;
3969     }
3970
3971     if (!mcast_snooping_enabled(ofproto->ms)) {
3972         unixctl_command_reply_error(conn, "multicast snooping is disabled");
3973         return;
3974     }
3975
3976     ds_put_cstr(&ds, " port  VLAN  GROUP                Age\n");
3977     ovs_rwlock_rdlock(&ofproto->ms->rwlock);
3978     LIST_FOR_EACH (grp, group_node, &ofproto->ms->group_lru) {
3979         LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
3980             char name[OFP_MAX_PORT_NAME_LEN];
3981
3982             bundle = b->port;
3983             ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
3984                                    name, sizeof name);
3985             ds_put_format(&ds, "%5s  %4d  "IP_FMT"         %3d\n",
3986                           name, grp->vlan, IP_ARGS(grp->ip4),
3987                           mcast_bundle_age(ofproto->ms, b));
3988         }
3989     }
3990
3991     /* ports connected to multicast routers */
3992     LIST_FOR_EACH(mrouter, mrouter_node, &ofproto->ms->mrouter_lru) {
3993         char name[OFP_MAX_PORT_NAME_LEN];
3994
3995         bundle = mrouter->port;
3996         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
3997                                name, sizeof name);
3998             ds_put_format(&ds, "%5s  %4d  querier             %3d\n",
3999                       name, mrouter->vlan,
4000                       mcast_mrouter_age(ofproto->ms, mrouter));
4001     }
4002     ovs_rwlock_unlock(&ofproto->ms->rwlock);
4003     unixctl_command_reply(conn, ds_cstr(&ds));
4004     ds_destroy(&ds);
4005 }
4006
4007 struct trace_ctx {
4008     struct xlate_out xout;
4009     struct xlate_in xin;
4010     const struct flow *key;
4011     struct flow flow;
4012     struct flow_wildcards wc;
4013     struct ds *result;
4014 };
4015
4016 static void
4017 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
4018 {
4019     const struct rule_actions *actions;
4020     ovs_be64 cookie;
4021
4022     ds_put_char_multiple(result, '\t', level);
4023     if (!rule) {
4024         ds_put_cstr(result, "No match\n");
4025         return;
4026     }
4027
4028     ovs_mutex_lock(&rule->up.mutex);
4029     cookie = rule->up.flow_cookie;
4030     ovs_mutex_unlock(&rule->up.mutex);
4031
4032     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
4033                   rule ? rule->up.table_id : 0, ntohll(cookie));
4034     cls_rule_format(&rule->up.cr, result);
4035     ds_put_char(result, '\n');
4036
4037     actions = rule_dpif_get_actions(rule);
4038
4039     ds_put_char_multiple(result, '\t', level);
4040     ds_put_cstr(result, "OpenFlow actions=");
4041     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
4042     ds_put_char(result, '\n');
4043 }
4044
4045 static void
4046 trace_format_flow(struct ds *result, int level, const char *title,
4047                   struct trace_ctx *trace)
4048 {
4049     ds_put_char_multiple(result, '\t', level);
4050     ds_put_format(result, "%s: ", title);
4051     /* Do not report unchanged flows for resubmits. */
4052     if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
4053         || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
4054         ds_put_cstr(result, "unchanged");
4055     } else {
4056         flow_format(result, &trace->xin.flow);
4057         trace->flow = trace->xin.flow;
4058     }
4059     ds_put_char(result, '\n');
4060 }
4061
4062 static void
4063 trace_format_regs(struct ds *result, int level, const char *title,
4064                   struct trace_ctx *trace)
4065 {
4066     size_t i;
4067
4068     ds_put_char_multiple(result, '\t', level);
4069     ds_put_format(result, "%s:", title);
4070     for (i = 0; i < FLOW_N_REGS; i++) {
4071         ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
4072     }
4073     ds_put_char(result, '\n');
4074 }
4075
4076 static void
4077 trace_format_odp(struct ds *result, int level, const char *title,
4078                  struct trace_ctx *trace)
4079 {
4080     struct ofpbuf *odp_actions = trace->xout.odp_actions;
4081
4082     ds_put_char_multiple(result, '\t', level);
4083     ds_put_format(result, "%s: ", title);
4084     format_odp_actions(result, ofpbuf_data(odp_actions),
4085                                ofpbuf_size(odp_actions));
4086     ds_put_char(result, '\n');
4087 }
4088
4089 static void
4090 trace_format_megaflow(struct ds *result, int level, const char *title,
4091                       struct trace_ctx *trace)
4092 {
4093     struct match match;
4094
4095     ds_put_char_multiple(result, '\t', level);
4096     ds_put_format(result, "%s: ", title);
4097     flow_wildcards_or(&trace->wc, &trace->xout.wc, &trace->wc);
4098     match_init(&match, trace->key, &trace->wc);
4099     match_format(&match, result, OFP_DEFAULT_PRIORITY);
4100     ds_put_char(result, '\n');
4101 }
4102
4103 static void
4104 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
4105 {
4106     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4107     struct ds *result = trace->result;
4108
4109     ds_put_char(result, '\n');
4110     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
4111     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
4112     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
4113     trace_format_megaflow(result, recurse + 1, "Resubmitted megaflow", trace);
4114     trace_format_rule(result, recurse + 1, rule);
4115 }
4116
4117 static void
4118 trace_report(struct xlate_in *xin, const char *s, int recurse)
4119 {
4120     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4121     struct ds *result = trace->result;
4122
4123     ds_put_char_multiple(result, '\t', recurse);
4124     ds_put_cstr(result, s);
4125     ds_put_char(result, '\n');
4126 }
4127
4128 /* Parses the 'argc' elements of 'argv', ignoring argv[0].  The following
4129  * forms are supported:
4130  *
4131  *     - [dpname] odp_flow [-generate | packet]
4132  *     - bridge br_flow [-generate | packet]
4133  *
4134  * On success, initializes '*ofprotop' and 'flow' and returns NULL.  On failure
4135  * returns a nonnull malloced error message. */
4136 static char * WARN_UNUSED_RESULT
4137 parse_flow_and_packet(int argc, const char *argv[],
4138                       struct ofproto_dpif **ofprotop, struct flow *flow,
4139                       struct ofpbuf **packetp)
4140 {
4141     const struct dpif_backer *backer = NULL;
4142     const char *error = NULL;
4143     char *m_err = NULL;
4144     struct simap port_names = SIMAP_INITIALIZER(&port_names);
4145     struct ofpbuf *packet;
4146     struct ofpbuf odp_key;
4147     struct ofpbuf odp_mask;
4148
4149     ofpbuf_init(&odp_key, 0);
4150     ofpbuf_init(&odp_mask, 0);
4151
4152     /* Handle "-generate" or a hex string as the last argument. */
4153     if (!strcmp(argv[argc - 1], "-generate")) {
4154         packet = ofpbuf_new(0);
4155         argc--;
4156     } else {
4157         error = eth_from_hex(argv[argc - 1], &packet);
4158         if (!error) {
4159             argc--;
4160         } else if (argc == 4) {
4161             /* The 3-argument form must end in "-generate' or a hex string. */
4162             goto exit;
4163         }
4164         error = NULL;
4165     }
4166
4167     /* odp_flow can have its in_port specified as a name instead of port no.
4168      * We do not yet know whether a given flow is a odp_flow or a br_flow.
4169      * But, to know whether a flow is odp_flow through odp_flow_from_string(),
4170      * we need to create a simap of name to port no. */
4171     if (argc == 3) {
4172         const char *dp_type;
4173         if (!strncmp(argv[1], "ovs-", 4)) {
4174             dp_type = argv[1] + 4;
4175         } else {
4176             dp_type = argv[1];
4177         }
4178         backer = shash_find_data(&all_dpif_backers, dp_type);
4179     } else if (argc == 2) {
4180         struct shash_node *node;
4181         if (shash_count(&all_dpif_backers) == 1) {
4182             node = shash_first(&all_dpif_backers);
4183             backer = node->data;
4184         }
4185     } else {
4186         error = "Syntax error";
4187         goto exit;
4188     }
4189     if (backer && backer->dpif) {
4190         struct dpif_port dpif_port;
4191         struct dpif_port_dump port_dump;
4192         DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4193             simap_put(&port_names, dpif_port.name,
4194                       odp_to_u32(dpif_port.port_no));
4195         }
4196     }
4197
4198     /* Parse the flow and determine whether a datapath or
4199      * bridge is specified. If function odp_flow_key_from_string()
4200      * returns 0, the flow is a odp_flow. If function
4201      * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4202     if (!odp_flow_from_string(argv[argc - 1], &port_names,
4203                               &odp_key, &odp_mask)) {
4204         if (!backer) {
4205             error = "Cannot find the datapath";
4206             goto exit;
4207         }
4208
4209         if (odp_flow_key_to_flow(ofpbuf_data(&odp_key), ofpbuf_size(&odp_key),
4210                                  flow) == ODP_FIT_ERROR) {
4211             error = "Failed to parse flow key";
4212             goto exit;
4213         }
4214
4215         if (xlate_receive(backer, flow, ofprotop, NULL, NULL, NULL,
4216                           &flow->in_port.ofp_port)) {
4217             error = "Invalid datapath flow";
4218             goto exit;
4219         }
4220
4221         vsp_adjust_flow(*ofprotop, flow, NULL);
4222
4223     } else {
4224         char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4225
4226         if (err) {
4227             m_err = xasprintf("Bad flow syntax: %s", err);
4228             free(err);
4229             goto exit;
4230         } else {
4231             if (argc != 3) {
4232                 error = "Must specify bridge name";
4233                 goto exit;
4234             }
4235
4236             *ofprotop = ofproto_dpif_lookup(argv[1]);
4237             if (!*ofprotop) {
4238                 error = "Unknown bridge name";
4239                 goto exit;
4240             }
4241         }
4242     }
4243
4244     /* Generate a packet, if requested. */
4245     if (packet) {
4246         if (!ofpbuf_size(packet)) {
4247             flow_compose(packet, flow);
4248         } else {
4249             struct pkt_metadata md = pkt_metadata_from_flow(flow);
4250
4251             /* Use the metadata from the flow and the packet argument
4252              * to reconstruct the flow. */
4253             flow_extract(packet, &md, flow);
4254         }
4255     }
4256
4257 exit:
4258     if (error && !m_err) {
4259         m_err = xstrdup(error);
4260     }
4261     if (m_err) {
4262         ofpbuf_delete(packet);
4263         packet = NULL;
4264     }
4265     *packetp = packet;
4266     ofpbuf_uninit(&odp_key);
4267     ofpbuf_uninit(&odp_mask);
4268     simap_destroy(&port_names);
4269     return m_err;
4270 }
4271
4272 static void
4273 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4274                       void *aux OVS_UNUSED)
4275 {
4276     struct ofproto_dpif *ofproto;
4277     struct ofpbuf *packet;
4278     char *error;
4279     struct flow flow;
4280
4281     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4282     if (!error) {
4283         struct ds result;
4284
4285         ds_init(&result);
4286         ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4287         unixctl_command_reply(conn, ds_cstr(&result));
4288         ds_destroy(&result);
4289         ofpbuf_delete(packet);
4290     } else {
4291         unixctl_command_reply_error(conn, error);
4292         free(error);
4293     }
4294 }
4295
4296 static void
4297 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4298                               const char *argv[], void *aux OVS_UNUSED)
4299 {
4300     enum ofputil_protocol usable_protocols;
4301     struct ofproto_dpif *ofproto;
4302     bool enforce_consistency;
4303     struct ofpbuf ofpacts;
4304     struct ofpbuf *packet;
4305     struct ds result;
4306     struct flow flow;
4307     uint16_t in_port;
4308
4309     /* Three kinds of error return values! */
4310     enum ofperr retval;
4311     char *error;
4312
4313     packet = NULL;
4314     ds_init(&result);
4315     ofpbuf_init(&ofpacts, 0);
4316
4317     /* Parse actions. */
4318     error = ofpacts_parse_actions(argv[--argc], &ofpacts, &usable_protocols);
4319     if (error) {
4320         unixctl_command_reply_error(conn, error);
4321         free(error);
4322         goto exit;
4323     }
4324
4325     /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4326      * consistency between the flow and the actions.  With -consistent, we
4327      * enforce consistency even for a flow supported in OpenFlow 1.0. */
4328     if (!strcmp(argv[1], "-consistent")) {
4329         enforce_consistency = true;
4330         argv++;
4331         argc--;
4332     } else {
4333         enforce_consistency = false;
4334     }
4335
4336     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4337     if (error) {
4338         unixctl_command_reply_error(conn, error);
4339         free(error);
4340         goto exit;
4341     }
4342
4343     /* Do the same checks as handle_packet_out() in ofproto.c.
4344      *
4345      * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
4346      * strictly correct because these actions aren't in any table, but it's OK
4347      * because it 'table_id' is used only to check goto_table instructions, but
4348      * packet-outs take a list of actions and therefore it can't include
4349      * instructions.
4350      *
4351      * We skip the "meter" check here because meter is an instruction, not an
4352      * action, and thus cannot appear in ofpacts. */
4353     in_port = ofp_to_u16(flow.in_port.ofp_port);
4354     if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
4355         unixctl_command_reply_error(conn, "invalid in_port");
4356         goto exit;
4357     }
4358     if (enforce_consistency) {
4359         retval = ofpacts_check_consistency(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts),
4360                                            &flow, u16_to_ofp(ofproto->up.max_ports),
4361                                            0, 0, usable_protocols);
4362     } else {
4363         retval = ofpacts_check(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &flow,
4364                                u16_to_ofp(ofproto->up.max_ports), 0, 0,
4365                                &usable_protocols);
4366     }
4367
4368     if (retval) {
4369         ds_clear(&result);
4370         ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
4371         unixctl_command_reply_error(conn, ds_cstr(&result));
4372         goto exit;
4373     }
4374
4375     ofproto_trace(ofproto, &flow, packet,
4376                   ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &result);
4377     unixctl_command_reply(conn, ds_cstr(&result));
4378
4379 exit:
4380     ds_destroy(&result);
4381     ofpbuf_delete(packet);
4382     ofpbuf_uninit(&ofpacts);
4383 }
4384
4385 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4386  * description of the results to 'ds'.
4387  *
4388  * The trace follows a packet with the specified 'flow' through the flow
4389  * table.  'packet' may be nonnull to trace an actual packet, with consequent
4390  * side effects (if it is nonnull then its flow must be 'flow').
4391  *
4392  * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4393  * trace, otherwise the actions are determined by a flow table lookup. */
4394 static void
4395 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
4396               const struct ofpbuf *packet,
4397               const struct ofpact ofpacts[], size_t ofpacts_len,
4398               struct ds *ds)
4399 {
4400     struct rule_dpif *rule;
4401     struct trace_ctx trace;
4402
4403     ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
4404     ds_put_cstr(ds, "Flow: ");
4405     flow_format(ds, flow);
4406     ds_put_char(ds, '\n');
4407
4408     flow_wildcards_init_catchall(&trace.wc);
4409     if (ofpacts) {
4410         rule = NULL;
4411     } else {
4412         rule_dpif_lookup(ofproto, flow, &trace.wc, &rule, false, NULL);
4413
4414         trace_format_rule(ds, 0, rule);
4415         if (rule == ofproto->miss_rule) {
4416             ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
4417         } else if (rule == ofproto->no_packet_in_rule) {
4418             ds_put_cstr(ds, "\nNo match, packets dropped because "
4419                         "OFPPC_NO_PACKET_IN is set on in_port.\n");
4420         } else if (rule == ofproto->drop_frags_rule) {
4421             ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
4422                         "and the fragment handling mode is \"drop\".\n");
4423         }
4424     }
4425
4426     if (rule || ofpacts) {
4427         trace.result = ds;
4428         trace.key = flow; /* Original flow key, used for megaflow. */
4429         trace.flow = *flow; /* May be modified by actions. */
4430         xlate_in_init(&trace.xin, ofproto, flow, flow->in_port.ofp_port, rule,
4431                       ntohs(flow->tcp_flags), packet);
4432         if (ofpacts) {
4433             trace.xin.ofpacts = ofpacts;
4434             trace.xin.ofpacts_len = ofpacts_len;
4435         }
4436         trace.xin.resubmit_hook = trace_resubmit;
4437         trace.xin.report_hook = trace_report;
4438
4439         xlate_actions(&trace.xin, &trace.xout);
4440
4441         ds_put_char(ds, '\n');
4442         trace_format_flow(ds, 0, "Final flow", &trace);
4443         trace_format_megaflow(ds, 0, "Megaflow", &trace);
4444
4445         ds_put_cstr(ds, "Datapath actions: ");
4446         format_odp_actions(ds, ofpbuf_data(trace.xout.odp_actions),
4447                            ofpbuf_size(trace.xout.odp_actions));
4448
4449         if (trace.xout.slow) {
4450             enum slow_path_reason slow;
4451
4452             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
4453                         "slow path because it:");
4454
4455             slow = trace.xout.slow;
4456             while (slow) {
4457                 enum slow_path_reason bit = rightmost_1bit(slow);
4458
4459                 ds_put_format(ds, "\n\t- %s.",
4460                               slow_path_reason_to_explanation(bit));
4461
4462                 slow &= ~bit;
4463             }
4464         }
4465
4466         xlate_out_uninit(&trace.xout);
4467     }
4468 }
4469
4470 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
4471  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
4472  * to destroy 'ofproto_shash' and free the returned value. */
4473 static const struct shash_node **
4474 get_ofprotos(struct shash *ofproto_shash)
4475 {
4476     const struct ofproto_dpif *ofproto;
4477
4478     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4479         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
4480         shash_add_nocopy(ofproto_shash, name, ofproto);
4481     }
4482
4483     return shash_sort(ofproto_shash);
4484 }
4485
4486 static void
4487 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
4488                               const char *argv[] OVS_UNUSED,
4489                               void *aux OVS_UNUSED)
4490 {
4491     struct ds ds = DS_EMPTY_INITIALIZER;
4492     struct shash ofproto_shash;
4493     const struct shash_node **sorted_ofprotos;
4494     int i;
4495
4496     shash_init(&ofproto_shash);
4497     sorted_ofprotos = get_ofprotos(&ofproto_shash);
4498     for (i = 0; i < shash_count(&ofproto_shash); i++) {
4499         const struct shash_node *node = sorted_ofprotos[i];
4500         ds_put_format(&ds, "%s\n", node->name);
4501     }
4502
4503     shash_destroy(&ofproto_shash);
4504     free(sorted_ofprotos);
4505
4506     unixctl_command_reply(conn, ds_cstr(&ds));
4507     ds_destroy(&ds);
4508 }
4509
4510 static void
4511 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
4512 {
4513     const struct shash_node **ofprotos;
4514     struct dpif_dp_stats dp_stats;
4515     struct shash ofproto_shash;
4516     size_t i;
4517
4518     dpif_get_dp_stats(backer->dpif, &dp_stats);
4519
4520     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
4521                   dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
4522
4523     shash_init(&ofproto_shash);
4524     ofprotos = get_ofprotos(&ofproto_shash);
4525     for (i = 0; i < shash_count(&ofproto_shash); i++) {
4526         struct ofproto_dpif *ofproto = ofprotos[i]->data;
4527         const struct shash_node **ports;
4528         size_t j;
4529
4530         if (ofproto->backer != backer) {
4531             continue;
4532         }
4533
4534         ds_put_format(ds, "\t%s:\n", ofproto->up.name);
4535
4536         ports = shash_sort(&ofproto->up.port_by_name);
4537         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
4538             const struct shash_node *node = ports[j];
4539             struct ofport *ofport = node->data;
4540             struct smap config;
4541             odp_port_t odp_port;
4542
4543             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
4544                           ofport->ofp_port);
4545
4546             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
4547             if (odp_port != ODPP_NONE) {
4548                 ds_put_format(ds, "%"PRIu32":", odp_port);
4549             } else {
4550                 ds_put_cstr(ds, "none:");
4551             }
4552
4553             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
4554
4555             smap_init(&config);
4556             if (!netdev_get_config(ofport->netdev, &config)) {
4557                 const struct smap_node **nodes;
4558                 size_t i;
4559
4560                 nodes = smap_sort(&config);
4561                 for (i = 0; i < smap_count(&config); i++) {
4562                     const struct smap_node *node = nodes[i];
4563                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
4564                                   node->key, node->value);
4565                 }
4566                 free(nodes);
4567             }
4568             smap_destroy(&config);
4569
4570             ds_put_char(ds, ')');
4571             ds_put_char(ds, '\n');
4572         }
4573         free(ports);
4574     }
4575     shash_destroy(&ofproto_shash);
4576     free(ofprotos);
4577 }
4578
4579 static void
4580 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4581                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4582 {
4583     struct ds ds = DS_EMPTY_INITIALIZER;
4584     const struct shash_node **backers;
4585     int i;
4586
4587     backers = shash_sort(&all_dpif_backers);
4588     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
4589         dpif_show_backer(backers[i]->data, &ds);
4590     }
4591     free(backers);
4592
4593     unixctl_command_reply(conn, ds_cstr(&ds));
4594     ds_destroy(&ds);
4595 }
4596
4597 static bool
4598 ofproto_dpif_contains_flow(const struct ofproto_dpif *ofproto,
4599                            const struct nlattr *key, size_t key_len)
4600 {
4601     struct ofproto_dpif *ofp;
4602     struct flow flow;
4603
4604     if (odp_flow_key_to_flow(key, key_len, &flow) == ODP_FIT_ERROR) {
4605         return false;
4606     }
4607
4608     if (xlate_receive(ofproto->backer, &flow, &ofp, NULL, NULL, NULL, NULL)) {
4609         return false;
4610     }
4611
4612     return ofp == ofproto;
4613 }
4614
4615 static void
4616 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
4617                                 int argc OVS_UNUSED, const char *argv[],
4618                                 void *aux OVS_UNUSED)
4619 {
4620     const struct ofproto_dpif *ofproto;
4621
4622     struct ds ds = DS_EMPTY_INITIALIZER;
4623     bool verbosity = false;
4624
4625     struct dpif_port dpif_port;
4626     struct dpif_port_dump port_dump;
4627     struct hmap portno_names;
4628
4629     struct dpif_flow_dump *flow_dump;
4630     struct dpif_flow_dump_thread *flow_dump_thread;
4631     struct dpif_flow f;
4632     int error;
4633
4634     ofproto = ofproto_dpif_lookup(argv[argc - 1]);
4635     if (!ofproto) {
4636         unixctl_command_reply_error(conn, "no such bridge");
4637         return;
4638     }
4639
4640     if (argc > 2 && !strcmp(argv[1], "-m")) {
4641         verbosity = true;
4642     }
4643
4644     hmap_init(&portno_names);
4645     DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
4646         odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
4647     }
4648
4649     ds_init(&ds);
4650     flow_dump = dpif_flow_dump_create(ofproto->backer->dpif);
4651     flow_dump_thread = dpif_flow_dump_thread_create(flow_dump);
4652     while (dpif_flow_dump_next(flow_dump_thread, &f, 1)) {
4653         if (!ofproto_dpif_contains_flow(ofproto, f.key, f.key_len)) {
4654             continue;
4655         }
4656
4657         odp_flow_format(f.key, f.key_len, f.mask, f.mask_len,
4658                         &portno_names, &ds, verbosity);
4659         ds_put_cstr(&ds, ", ");
4660         dpif_flow_stats_format(&f.stats, &ds);
4661         ds_put_cstr(&ds, ", actions:");
4662         format_odp_actions(&ds, f.actions, f.actions_len);
4663         ds_put_char(&ds, '\n');
4664     }
4665     dpif_flow_dump_thread_destroy(flow_dump_thread);
4666     error = dpif_flow_dump_destroy(flow_dump);
4667
4668     if (error) {
4669         ds_clear(&ds);
4670         ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
4671         unixctl_command_reply_error(conn, ds_cstr(&ds));
4672     } else {
4673         unixctl_command_reply(conn, ds_cstr(&ds));
4674     }
4675     odp_portno_names_destroy(&portno_names);
4676     hmap_destroy(&portno_names);
4677     ds_destroy(&ds);
4678 }
4679
4680 static void
4681 ofproto_dpif_unixctl_init(void)
4682 {
4683     static bool registered;
4684     if (registered) {
4685         return;
4686     }
4687     registered = true;
4688
4689     unixctl_command_register(
4690         "ofproto/trace",
4691         "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
4692         1, 3, ofproto_unixctl_trace, NULL);
4693     unixctl_command_register(
4694         "ofproto/trace-packet-out",
4695         "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
4696         2, 6, ofproto_unixctl_trace_actions, NULL);
4697     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
4698                              ofproto_unixctl_fdb_flush, NULL);
4699     unixctl_command_register("fdb/show", "bridge", 1, 1,
4700                              ofproto_unixctl_fdb_show, NULL);
4701     unixctl_command_register("mdb/flush", "[bridge]", 0, 1,
4702                              ofproto_unixctl_mcast_snooping_flush, NULL);
4703     unixctl_command_register("mdb/show", "bridge", 1, 1,
4704                              ofproto_unixctl_mcast_snooping_show, NULL);
4705     unixctl_command_register("dpif/dump-dps", "", 0, 0,
4706                              ofproto_unixctl_dpif_dump_dps, NULL);
4707     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
4708                              NULL);
4709     unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
4710                              ofproto_unixctl_dpif_dump_flows, NULL);
4711 }
4712
4713 /* Returns true if 'table' is the table used for internal rules,
4714  * false otherwise. */
4715 bool
4716 table_is_internal(uint8_t table_id)
4717 {
4718     return table_id == TBL_INTERNAL;
4719 }
4720 \f
4721 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4722  *
4723  * This is deprecated.  It is only for compatibility with broken device drivers
4724  * in old versions of Linux that do not properly support VLANs when VLAN
4725  * devices are not used.  When broken device drivers are no longer in
4726  * widespread use, we will delete these interfaces. */
4727
4728 static int
4729 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
4730 {
4731     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
4732     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
4733
4734     if (realdev_ofp_port == ofport->realdev_ofp_port
4735         && vid == ofport->vlandev_vid) {
4736         return 0;
4737     }
4738
4739     ofproto->backer->need_revalidate = REV_RECONFIGURE;
4740
4741     if (ofport->realdev_ofp_port) {
4742         vsp_remove(ofport);
4743     }
4744     if (realdev_ofp_port && ofport->bundle) {
4745         /* vlandevs are enslaved to their realdevs, so they are not allowed to
4746          * themselves be part of a bundle. */
4747         bundle_set(ofport_->ofproto, ofport->bundle, NULL);
4748     }
4749
4750     ofport->realdev_ofp_port = realdev_ofp_port;
4751     ofport->vlandev_vid = vid;
4752
4753     if (realdev_ofp_port) {
4754         vsp_add(ofport, realdev_ofp_port, vid);
4755     }
4756
4757     return 0;
4758 }
4759
4760 static uint32_t
4761 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
4762 {
4763     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
4764 }
4765
4766 bool
4767 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
4768     OVS_EXCLUDED(ofproto->vsp_mutex)
4769 {
4770     /* hmap_is_empty is thread safe. */
4771     return !hmap_is_empty(&ofproto->realdev_vid_map);
4772 }
4773
4774
4775 static ofp_port_t
4776 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
4777                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4778     OVS_REQUIRES(ofproto->vsp_mutex)
4779 {
4780     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
4781         int vid = vlan_tci_to_vid(vlan_tci);
4782         const struct vlan_splinter *vsp;
4783
4784         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
4785                                  hash_realdev_vid(realdev_ofp_port, vid),
4786                                  &ofproto->realdev_vid_map) {
4787             if (vsp->realdev_ofp_port == realdev_ofp_port
4788                 && vsp->vid == vid) {
4789                 return vsp->vlandev_ofp_port;
4790             }
4791         }
4792     }
4793     return realdev_ofp_port;
4794 }
4795
4796 /* Returns the OFP port number of the Linux VLAN device that corresponds to
4797  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
4798  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
4799  * 'vlan_tci' 9, it would return the port number of eth0.9.
4800  *
4801  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
4802  * function just returns its 'realdev_ofp_port' argument. */
4803 ofp_port_t
4804 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
4805                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4806     OVS_EXCLUDED(ofproto->vsp_mutex)
4807 {
4808     ofp_port_t ret;
4809
4810     /* hmap_is_empty is thread safe, see if we can return immediately. */
4811     if (hmap_is_empty(&ofproto->realdev_vid_map)) {
4812         return realdev_ofp_port;
4813     }
4814     ovs_mutex_lock(&ofproto->vsp_mutex);
4815     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
4816     ovs_mutex_unlock(&ofproto->vsp_mutex);
4817     return ret;
4818 }
4819
4820 static struct vlan_splinter *
4821 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
4822 {
4823     struct vlan_splinter *vsp;
4824
4825     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
4826                              hash_ofp_port(vlandev_ofp_port),
4827                              &ofproto->vlandev_map) {
4828         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
4829             return vsp;
4830         }
4831     }
4832
4833     return NULL;
4834 }
4835
4836 /* Returns the OpenFlow port number of the "real" device underlying the Linux
4837  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
4838  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
4839  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
4840  * eth0 and store 9 in '*vid'.
4841  *
4842  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
4843  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
4844  * always does.*/
4845 static ofp_port_t
4846 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
4847                        ofp_port_t vlandev_ofp_port, int *vid)
4848     OVS_REQUIRES(ofproto->vsp_mutex)
4849 {
4850     if (!hmap_is_empty(&ofproto->vlandev_map)) {
4851         const struct vlan_splinter *vsp;
4852
4853         vsp = vlandev_find(ofproto, vlandev_ofp_port);
4854         if (vsp) {
4855             if (vid) {
4856                 *vid = vsp->vid;
4857             }
4858             return vsp->realdev_ofp_port;
4859         }
4860     }
4861     return 0;
4862 }
4863
4864 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
4865  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
4866  * 'flow->in_port' to the "real" device backing the VLAN device, sets
4867  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Optionally pushes the
4868  * appropriate VLAN on 'packet' if provided.  Otherwise (which is always the
4869  * case unless VLAN splinters are enabled), returns false without making any
4870  * changes. */
4871 bool
4872 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow,
4873                 struct ofpbuf *packet)
4874     OVS_EXCLUDED(ofproto->vsp_mutex)
4875 {
4876     ofp_port_t realdev;
4877     int vid;
4878
4879     /* hmap_is_empty is thread safe. */
4880     if (hmap_is_empty(&ofproto->vlandev_map)) {
4881         return false;
4882     }
4883
4884     ovs_mutex_lock(&ofproto->vsp_mutex);
4885     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
4886     ovs_mutex_unlock(&ofproto->vsp_mutex);
4887     if (!realdev) {
4888         return false;
4889     }
4890
4891     /* Cause the flow to be processed as if it came in on the real device with
4892      * the VLAN device's VLAN ID. */
4893     flow->in_port.ofp_port = realdev;
4894     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
4895
4896     if (packet) {
4897         /* Make the packet resemble the flow, so that it gets sent to an
4898          * OpenFlow controller properly, so that it looks correct for sFlow,
4899          * and so that flow_extract() will get the correct vlan_tci if it is
4900          * called on 'packet'. */
4901         eth_push_vlan(packet, htons(ETH_TYPE_VLAN), flow->vlan_tci);
4902     }
4903
4904     return true;
4905 }
4906
4907 static void
4908 vsp_remove(struct ofport_dpif *port)
4909 {
4910     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4911     struct vlan_splinter *vsp;
4912
4913     ovs_mutex_lock(&ofproto->vsp_mutex);
4914     vsp = vlandev_find(ofproto, port->up.ofp_port);
4915     if (vsp) {
4916         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
4917         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
4918         free(vsp);
4919
4920         port->realdev_ofp_port = 0;
4921     } else {
4922         VLOG_ERR("missing vlan device record");
4923     }
4924     ovs_mutex_unlock(&ofproto->vsp_mutex);
4925 }
4926
4927 static void
4928 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
4929 {
4930     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4931
4932     ovs_mutex_lock(&ofproto->vsp_mutex);
4933     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
4934         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
4935             == realdev_ofp_port)) {
4936         struct vlan_splinter *vsp;
4937
4938         vsp = xmalloc(sizeof *vsp);
4939         vsp->realdev_ofp_port = realdev_ofp_port;
4940         vsp->vlandev_ofp_port = port->up.ofp_port;
4941         vsp->vid = vid;
4942
4943         port->realdev_ofp_port = realdev_ofp_port;
4944
4945         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
4946                     hash_ofp_port(port->up.ofp_port));
4947         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
4948                     hash_realdev_vid(realdev_ofp_port, vid));
4949     } else {
4950         VLOG_ERR("duplicate vlan device record");
4951     }
4952     ovs_mutex_unlock(&ofproto->vsp_mutex);
4953 }
4954
4955 static odp_port_t
4956 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
4957 {
4958     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
4959     return ofport ? ofport->odp_port : ODPP_NONE;
4960 }
4961
4962 struct ofport_dpif *
4963 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
4964 {
4965     struct ofport_dpif *port;
4966
4967     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
4968     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
4969                              &backer->odp_to_ofport_map) {
4970         if (port->odp_port == odp_port) {
4971             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4972             return port;
4973         }
4974     }
4975
4976     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4977     return NULL;
4978 }
4979
4980 static ofp_port_t
4981 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
4982 {
4983     struct ofport_dpif *port;
4984
4985     port = odp_port_to_ofport(ofproto->backer, odp_port);
4986     if (port && &ofproto->up == port->up.ofproto) {
4987         return port->up.ofp_port;
4988     } else {
4989         return OFPP_NONE;
4990     }
4991 }
4992
4993 uint32_t
4994 ofproto_dpif_alloc_recirc_id(struct ofproto_dpif *ofproto)
4995 {
4996     struct dpif_backer *backer = ofproto->backer;
4997
4998     return  recirc_id_alloc(backer->rid_pool);
4999 }
5000
5001 void
5002 ofproto_dpif_free_recirc_id(struct ofproto_dpif *ofproto, uint32_t recirc_id)
5003 {
5004     struct dpif_backer *backer = ofproto->backer;
5005
5006     recirc_id_free(backer->rid_pool, recirc_id);
5007 }
5008
5009 int
5010 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
5011                                const struct match *match, int priority,
5012                                uint16_t idle_timeout,
5013                                const struct ofpbuf *ofpacts,
5014                                struct rule **rulep)
5015 {
5016     struct ofputil_flow_mod fm;
5017     struct rule_dpif *rule;
5018     int error;
5019
5020     fm.match = *match;
5021     fm.priority = priority;
5022     fm.new_cookie = htonll(0);
5023     fm.cookie = htonll(0);
5024     fm.cookie_mask = htonll(0);
5025     fm.modify_cookie = false;
5026     fm.table_id = TBL_INTERNAL;
5027     fm.command = OFPFC_ADD;
5028     fm.idle_timeout = idle_timeout;
5029     fm.hard_timeout = 0;
5030     fm.buffer_id = 0;
5031     fm.out_port = 0;
5032     fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5033     fm.ofpacts = ofpbuf_data(ofpacts);
5034     fm.ofpacts_len = ofpbuf_size(ofpacts);
5035
5036     error = ofproto_flow_mod(&ofproto->up, &fm);
5037     if (error) {
5038         VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
5039                     ofperr_to_string(error));
5040         *rulep = NULL;
5041         return error;
5042     }
5043
5044     rule = rule_dpif_lookup_in_table(ofproto, TBL_INTERNAL, &fm.match.flow,
5045                                      &fm.match.wc, false);
5046     if (rule) {
5047         *rulep = &rule->up;
5048     } else {
5049         OVS_NOT_REACHED();
5050     }
5051     return 0;
5052 }
5053
5054 int
5055 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
5056                                   struct match *match, int priority)
5057 {
5058     struct ofputil_flow_mod fm;
5059     int error;
5060
5061     fm.match = *match;
5062     fm.priority = priority;
5063     fm.new_cookie = htonll(0);
5064     fm.cookie = htonll(0);
5065     fm.cookie_mask = htonll(0);
5066     fm.modify_cookie = false;
5067     fm.table_id = TBL_INTERNAL;
5068     fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5069     fm.command = OFPFC_DELETE_STRICT;
5070
5071     error = ofproto_flow_mod(&ofproto->up, &fm);
5072     if (error) {
5073         VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
5074                     ofperr_to_string(error));
5075         return error;
5076     }
5077
5078     return 0;
5079 }
5080
5081 const struct ofproto_class ofproto_dpif_class = {
5082     init,
5083     enumerate_types,
5084     enumerate_names,
5085     del,
5086     port_open_type,
5087     type_run,
5088     type_wait,
5089     alloc,
5090     construct,
5091     destruct,
5092     dealloc,
5093     run,
5094     wait,
5095     NULL,                       /* get_memory_usage. */
5096     type_get_memory_usage,
5097     flush,
5098     query_tables,
5099     port_alloc,
5100     port_construct,
5101     port_destruct,
5102     port_dealloc,
5103     port_modified,
5104     port_reconfigured,
5105     port_query_by_name,
5106     port_add,
5107     port_del,
5108     port_get_stats,
5109     port_dump_start,
5110     port_dump_next,
5111     port_dump_done,
5112     port_poll,
5113     port_poll_wait,
5114     port_is_lacp_current,
5115     NULL,                       /* rule_choose_table */
5116     rule_alloc,
5117     rule_construct,
5118     rule_insert,
5119     rule_delete,
5120     rule_destruct,
5121     rule_dealloc,
5122     rule_get_stats,
5123     rule_execute,
5124     NULL,                       /* rule_premodify_actions */
5125     rule_modify_actions,
5126     set_frag_handling,
5127     packet_out,
5128     set_netflow,
5129     get_netflow_ids,
5130     set_sflow,
5131     set_ipfix,
5132     set_cfm,
5133     cfm_status_changed,
5134     get_cfm_status,
5135     set_bfd,
5136     bfd_status_changed,
5137     get_bfd_status,
5138     set_stp,
5139     get_stp_status,
5140     set_stp_port,
5141     get_stp_port_status,
5142     get_stp_port_stats,
5143     set_queues,
5144     bundle_set,
5145     bundle_remove,
5146     mirror_set__,
5147     mirror_get_stats__,
5148     set_flood_vlans,
5149     is_mirror_output_bundle,
5150     forward_bpdu_changed,
5151     set_mac_table_config,
5152     set_mcast_snooping,
5153     set_mcast_snooping_port,
5154     set_realdev,
5155     NULL,                       /* meter_get_features */
5156     NULL,                       /* meter_set */
5157     NULL,                       /* meter_get */
5158     NULL,                       /* meter_del */
5159     group_alloc,                /* group_alloc */
5160     group_construct,            /* group_construct */
5161     group_destruct,             /* group_destruct */
5162     group_dealloc,              /* group_dealloc */
5163     group_modify,               /* group_modify */
5164     group_get_stats,            /* group_get_stats */
5165 };