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 static void rstp_run(struct ofproto_dpif *ofproto);
 149 static void set_rstp_port(struct ofport *,
 150                          const struct ofproto_port_rstp_settings *);
 151
 152 struct ofport_dpif {
 153     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 154     struct ofport up;
 155
 156     odp_port_t odp_port;
 157     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 158     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
 159     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 160     struct bfd *bfd;            /* BFD, if any. */
 161     bool may_enable;            /* May be enabled in bonds. */
 162     bool is_tunnel;             /* This port is a tunnel. */
 163     bool is_layer3;             /* This is a layer 3 port. */
 164     long long int carrier_seq;  /* Carrier status changes. */
 165     struct ofport_dpif *peer;   /* Peer if patch port. */
 166
 167     /* Spanning tree. */
 168     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 169     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 170     long long int stp_state_entered;
 171
 172     /* Rapid Spanning Tree. */
 173     struct rstp_port *rstp_port; /* Rapid Spanning Tree Protocol, if any. */
 174     enum rstp_state rstp_state; /* Always RSTP_DISABLED if RSTP not in use. */
 175
 176     /* Queue to DSCP mapping. */
 177     struct ofproto_port_queue *qdscp;
 178     size_t n_qdscp;
 179
 180     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 181      *
 182      * This is deprecated.  It is only for compatibility with broken device
 183      * drivers in old versions of Linux that do not properly support VLANs when
 184      * VLAN devices are not used.  When broken device drivers are no longer in
 185      * widespread use, we will delete these interfaces. */
 186     ofp_port_t realdev_ofp_port;
 187     int vlandev_vid;
 188 };
 189
 190 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 191  *
 192  * This is deprecated.  It is only for compatibility with broken device drivers
 193  * in old versions of Linux that do not properly support VLANs when VLAN
 194  * devices are not used.  When broken device drivers are no longer in
 195  * widespread use, we will delete these interfaces. */
 196 struct vlan_splinter {
 197     struct hmap_node realdev_vid_node;
 198     struct hmap_node vlandev_node;
 199     ofp_port_t realdev_ofp_port;
 200     ofp_port_t vlandev_ofp_port;
 201     int vid;
 202 };
 203
 204 static void vsp_remove(struct ofport_dpif *);
 205 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 206
 207 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 208                                        ofp_port_t);
 209
 210 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 211                                        odp_port_t);
 212
 213 static struct ofport_dpif *
 214 ofport_dpif_cast(const struct ofport *ofport)
 215 {
 216     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 217 }
 218
 219 static void port_run(struct ofport_dpif *);
 220 static int set_bfd(struct ofport *, const struct smap *);
 221 static int set_cfm(struct ofport *, const struct cfm_settings *);
 222 static void ofport_update_peer(struct ofport_dpif *);
 223
 224 /* Reasons that we might need to revalidate every datapath flow, and
 225  * corresponding coverage counters.
 226  *
 227  * A value of 0 means that there is no need to revalidate.
 228  *
 229  * It would be nice to have some cleaner way to integrate with coverage
 230  * counters, but with only a few reasons I guess this is good enough for
 231  * now. */
 232 enum revalidate_reason {
 233     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 234     REV_STP,                   /* Spanning tree protocol port status change. */
 235     REV_RSTP,                  /* RSTP port status change. */
 236     REV_BOND,                  /* Bonding changed. */
 237     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 238     REV_FLOW_TABLE,            /* Flow table changed. */
 239     REV_MAC_LEARNING,          /* Mac learning changed. */
 240     REV_MCAST_SNOOPING,        /* Multicast snooping changed. */
 241 };
 242 COVERAGE_DEFINE(rev_reconfigure);
 243 COVERAGE_DEFINE(rev_stp);
 244 COVERAGE_DEFINE(rev_rstp);
 245 COVERAGE_DEFINE(rev_bond);
 246 COVERAGE_DEFINE(rev_port_toggled);
 247 COVERAGE_DEFINE(rev_flow_table);
 248 COVERAGE_DEFINE(rev_mac_learning);
 249 COVERAGE_DEFINE(rev_mcast_snooping);
 250
 251 /* All datapaths of a given type share a single dpif backer instance. */
 252 struct dpif_backer {
 253     char *type;
 254     int refcount;
 255     struct dpif *dpif;
 256     struct udpif *udpif;
 257
 258     struct ovs_rwlock odp_to_ofport_lock;
 259     struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
 260
 261     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 262
 263     enum revalidate_reason need_revalidate; /* Revalidate all flows. */
 264
 265     bool recv_set_enable; /* Enables or disables receiving packets. */
 266
 267     /* Recirculation. */
 268     struct recirc_id_pool *rid_pool;       /* Recirculation ID pool. */
 269     bool enable_recirc;   /* True if the datapath supports recirculation */
 270
 271     /* True if the datapath supports variable-length
 272      * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
 273      * False if the datapath supports only 8-byte (or shorter) userdata. */
 274     bool variable_length_userdata;
 275
 276     /* Maximum number of MPLS label stack entries that the datapath supports
 277      * in a match */
 278     size_t max_mpls_depth;
 279 };
 280
 281 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 282 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 283
 284 struct ofproto_dpif {
 285     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 286     struct ofproto up;
 287     struct dpif_backer *backer;
 288
 289     uint64_t dump_seq; /* Last read of udpif_dump_seq(). */
 290
 291     /* Special OpenFlow rules. */
 292     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 293     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 294     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 295
 296     /* Bridging. */
 297     struct netflow *netflow;
 298     struct dpif_sflow *sflow;
 299     struct dpif_ipfix *ipfix;
 300     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 301     struct mac_learning *ml;
 302     struct mcast_snooping *ms;
 303     bool has_bonded_bundles;
 304     bool lacp_enabled;
 305     struct mbridge *mbridge;
 306
 307     struct ovs_mutex stats_mutex;
 308     struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
 309                                             * consumed in userspace. */
 310
 311     /* Spanning tree. */
 312     struct stp *stp;
 313     long long int stp_last_tick;
 314
 315     /* Rapid Spanning Tree. */
 316     struct rstp *rstp;
 317     long long int rstp_last_tick;
 318
 319     /* VLAN splinters. */
 320     struct ovs_mutex vsp_mutex;
 321     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 322     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 323
 324     /* Ports. */
 325     struct sset ports;             /* Set of standard port names. */
 326     struct sset ghost_ports;       /* Ports with no datapath port. */
 327     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 328     int port_poll_errno;           /* Last errno for port_poll() reply. */
 329     uint64_t change_seq;           /* Connectivity status changes. */
 330
 331     /* Work queues. */
 332     struct guarded_list pins;      /* Contains "struct ofputil_packet_in"s. */
 333     struct seq *pins_seq;          /* For notifying 'pins' reception. */
 334     uint64_t pins_seqno;
 335 };
 336
 337 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 338 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 339
 340 static void ofproto_dpif_unixctl_init(void);
 341
 342 static inline struct ofproto_dpif *
 343 ofproto_dpif_cast(const struct ofproto *ofproto)
 344 {
 345     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 346     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 347 }
 348
 349 size_t
 350 ofproto_dpif_get_max_mpls_depth(const struct ofproto_dpif *ofproto)
 351 {
 352     return ofproto->backer->max_mpls_depth;
 353 }
 354
 355 bool
 356 ofproto_dpif_get_enable_recirc(const struct ofproto_dpif *ofproto)
 357 {
 358     return ofproto->backer->enable_recirc;
 359 }
 360
 361 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
 362                                         ofp_port_t ofp_port);
 363 static void ofproto_trace(struct ofproto_dpif *, struct flow *,
 364                           const struct ofpbuf *packet,
 365                           const struct ofpact[], size_t ofpacts_len,
 366                           struct ds *);
 367
 368 /* Global variables. */
 369 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 370
 371 /* Initial mappings of port to bridge mappings. */
 372 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 373
 374 /* Executes 'fm'.  The caller retains ownership of 'fm' and everything in
 375  * it. */
 376 void
 377 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 378                       struct ofputil_flow_mod *fm)
 379 {
 380     ofproto_flow_mod(&ofproto->up, fm);
 381 }
 382
 383 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 384  * Takes ownership of 'pin' and pin->packet. */
 385 void
 386 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 387                             struct ofproto_packet_in *pin)
 388 {
 389     if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
 390         COVERAGE_INC(packet_in_overflow);
 391         free(CONST_CAST(void *, pin->up.packet));
 392         free(pin);
 393     }
 394
 395     /* Wakes up main thread for packet-in I/O. */
 396     seq_change(ofproto->pins_seq);
 397 }
 398
 399 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
 400  * packet rather than to send the packet to the controller.
 401  *
 402  * This function returns false to indicate that a packet_in message
 403  * for a "table-miss" should be sent to at least one controller.
 404  * False otherwise. */
 405 bool
 406 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif *ofproto)
 407 {
 408     return connmgr_wants_packet_in_on_miss(ofproto->up.connmgr);
 409 }
 410 \f
 411 /* Factory functions. */
 412
 413 static void
 414 init(const struct shash *iface_hints)
 415 {
 416     struct shash_node *node;
 417
 418     /* Make a local copy, since we don't own 'iface_hints' elements. */
 419     SHASH_FOR_EACH(node, iface_hints) {
 420         const struct iface_hint *orig_hint = node->data;
 421         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 422
 423         new_hint->br_name = xstrdup(orig_hint->br_name);
 424         new_hint->br_type = xstrdup(orig_hint->br_type);
 425         new_hint->ofp_port = orig_hint->ofp_port;
 426
 427         shash_add(&init_ofp_ports, node->name, new_hint);
 428     }
 429 }
 430
 431 static void
 432 enumerate_types(struct sset *types)
 433 {
 434     dp_enumerate_types(types);
 435 }
 436
 437 static int
 438 enumerate_names(const char *type, struct sset *names)
 439 {
 440     struct ofproto_dpif *ofproto;
 441
 442     sset_clear(names);
 443     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 444         if (strcmp(type, ofproto->up.type)) {
 445             continue;
 446         }
 447         sset_add(names, ofproto->up.name);
 448     }
 449
 450     return 0;
 451 }
 452
 453 static int
 454 del(const char *type, const char *name)
 455 {
 456     struct dpif *dpif;
 457     int error;
 458
 459     error = dpif_open(name, type, &dpif);
 460     if (!error) {
 461         error = dpif_delete(dpif);
 462         dpif_close(dpif);
 463     }
 464     return error;
 465 }
 466 \f
 467 static const char *
 468 port_open_type(const char *datapath_type, const char *port_type)
 469 {
 470     return dpif_port_open_type(datapath_type, port_type);
 471 }
 472
 473 /* Type functions. */
 474
 475 static void process_dpif_port_changes(struct dpif_backer *);
 476 static void process_dpif_all_ports_changed(struct dpif_backer *);
 477 static void process_dpif_port_change(struct dpif_backer *,
 478                                      const char *devname);
 479 static void process_dpif_port_error(struct dpif_backer *, int error);
 480
 481 static struct ofproto_dpif *
 482 lookup_ofproto_dpif_by_port_name(const char *name)
 483 {
 484     struct ofproto_dpif *ofproto;
 485
 486     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 487         if (sset_contains(&ofproto->ports, name)) {
 488             return ofproto;
 489         }
 490     }
 491
 492     return NULL;
 493 }
 494
 495 static int
 496 type_run(const char *type)
 497 {
 498     struct dpif_backer *backer;
 499
 500     backer = shash_find_data(&all_dpif_backers, type);
 501     if (!backer) {
 502         /* This is not necessarily a problem, since backers are only
 503          * created on demand. */
 504         return 0;
 505     }
 506
 507     dpif_run(backer->dpif);
 508     udpif_run(backer->udpif);
 509
 510     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 511      * and the configuration has now changed to "false", enable receiving
 512      * packets from the datapath. */
 513     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 514         int error;
 515
 516         backer->recv_set_enable = true;
 517
 518         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 519         if (error) {
 520             VLOG_ERR("Failed to enable receiving packets in dpif.");
 521             return error;
 522         }
 523         dpif_flow_flush(backer->dpif);
 524         backer->need_revalidate = REV_RECONFIGURE;
 525     }
 526
 527     if (backer->recv_set_enable) {
 528         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 529     }
 530
 531     if (backer->need_revalidate) {
 532         struct ofproto_dpif *ofproto;
 533         struct simap_node *node;
 534         struct simap tmp_backers;
 535
 536         /* Handle tunnel garbage collection. */
 537         simap_init(&tmp_backers);
 538         simap_swap(&backer->tnl_backers, &tmp_backers);
 539
 540         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 541             struct ofport_dpif *iter;
 542
 543             if (backer != ofproto->backer) {
 544                 continue;
 545             }
 546
 547             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 548                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 549                 const char *dp_port;
 550
 551                 if (!iter->is_tunnel) {
 552                     continue;
 553                 }
 554
 555                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 556                                                      namebuf, sizeof namebuf);
 557                 node = simap_find(&tmp_backers, dp_port);
 558                 if (node) {
 559                     simap_put(&backer->tnl_backers, dp_port, node->data);
 560                     simap_delete(&tmp_backers, node);
 561                     node = simap_find(&backer->tnl_backers, dp_port);
 562                 } else {
 563                     node = simap_find(&backer->tnl_backers, dp_port);
 564                     if (!node) {
 565                         odp_port_t odp_port = ODPP_NONE;
 566
 567                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 568                                            &odp_port)) {
 569                             simap_put(&backer->tnl_backers, dp_port,
 570                                       odp_to_u32(odp_port));
 571                             node = simap_find(&backer->tnl_backers, dp_port);
 572                         }
 573                     }
 574                 }
 575
 576                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 577                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 578                                          iter->odp_port)) {
 579                     backer->need_revalidate = REV_RECONFIGURE;
 580                 }
 581             }
 582         }
 583
 584         SIMAP_FOR_EACH (node, &tmp_backers) {
 585             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 586         }
 587         simap_destroy(&tmp_backers);
 588
 589         switch (backer->need_revalidate) {
 590         case REV_RECONFIGURE:    COVERAGE_INC(rev_reconfigure);    break;
 591         case REV_STP:            COVERAGE_INC(rev_stp);            break;
 592         case REV_RSTP:           COVERAGE_INC(rev_rstp);           break;
 593         case REV_BOND:           COVERAGE_INC(rev_bond);           break;
 594         case REV_PORT_TOGGLED:   COVERAGE_INC(rev_port_toggled);   break;
 595         case REV_FLOW_TABLE:     COVERAGE_INC(rev_flow_table);     break;
 596         case REV_MAC_LEARNING:   COVERAGE_INC(rev_mac_learning);   break;
 597         case REV_MCAST_SNOOPING: COVERAGE_INC(rev_mcast_snooping); break;
 598         }
 599         backer->need_revalidate = 0;
 600
 601         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 602             struct ofport_dpif *ofport;
 603             struct ofbundle *bundle;
 604
 605             if (ofproto->backer != backer) {
 606                 continue;
 607             }
 608
 609             xlate_txn_start();
 610             xlate_ofproto_set(ofproto, ofproto->up.name,
 611                               ofproto->backer->dpif, ofproto->miss_rule,
 612                               ofproto->no_packet_in_rule, ofproto->ml,
 613                               ofproto->stp, ofproto->rstp, ofproto->ms,
 614                               ofproto->mbridge, ofproto->sflow, ofproto->ipfix,
 615                               ofproto->netflow, ofproto->up.frag_handling,
 616                               ofproto->up.forward_bpdu,
 617                               connmgr_has_in_band(ofproto->up.connmgr),
 618                               ofproto->backer->enable_recirc,
 619                               ofproto->backer->variable_length_userdata,
 620                               ofproto->backer->max_mpls_depth);
 621
 622             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 623                 xlate_bundle_set(ofproto, bundle, bundle->name,
 624                                  bundle->vlan_mode, bundle->vlan,
 625                                  bundle->trunks, bundle->use_priority_tags,
 626                                  bundle->bond, bundle->lacp,
 627                                  bundle->floodable);
 628             }
 629
 630             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 631                 int stp_port = ofport->stp_port
 632                     ? stp_port_no(ofport->stp_port)
 633                     : -1;
 634                 int rstp_port = ofport->rstp_port
 635                     ? rstp_port_number(ofport->rstp_port)
 636                     : -1;
 637                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 638                                  ofport->up.ofp_port, ofport->odp_port,
 639                                  ofport->up.netdev, ofport->cfm,
 640                                  ofport->bfd, ofport->peer, stp_port,
 641                                  rstp_port, ofport->qdscp, ofport->n_qdscp,
 642                                  ofport->up.pp.config, ofport->up.pp.state,
 643                                  ofport->is_tunnel, ofport->may_enable);
 644             }
 645             xlate_txn_commit();
 646         }
 647
 648         udpif_revalidate(backer->udpif);
 649     }
 650
 651     process_dpif_port_changes(backer);
 652
 653     return 0;
 654 }
 655
 656 /* Check for and handle port changes in 'backer''s dpif. */
 657 static void
 658 process_dpif_port_changes(struct dpif_backer *backer)
 659 {
 660     for (;;) {
 661         char *devname;
 662         int error;
 663
 664         error = dpif_port_poll(backer->dpif, &devname);
 665         switch (error) {
 666         case EAGAIN:
 667             return;
 668
 669         case ENOBUFS:
 670             process_dpif_all_ports_changed(backer);
 671             break;
 672
 673         case 0:
 674             process_dpif_port_change(backer, devname);
 675             free(devname);
 676             break;
 677
 678         default:
 679             process_dpif_port_error(backer, error);
 680             break;
 681         }
 682     }
 683 }
 684
 685 static void
 686 process_dpif_all_ports_changed(struct dpif_backer *backer)
 687 {
 688     struct ofproto_dpif *ofproto;
 689     struct dpif_port dpif_port;
 690     struct dpif_port_dump dump;
 691     struct sset devnames;
 692     const char *devname;
 693
 694     sset_init(&devnames);
 695     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 696         if (ofproto->backer == backer) {
 697             struct ofport *ofport;
 698
 699             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 700                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 701             }
 702         }
 703     }
 704     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 705         sset_add(&devnames, dpif_port.name);
 706     }
 707
 708     SSET_FOR_EACH (devname, &devnames) {
 709         process_dpif_port_change(backer, devname);
 710     }
 711     sset_destroy(&devnames);
 712 }
 713
 714 static void
 715 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 716 {
 717     struct ofproto_dpif *ofproto;
 718     struct dpif_port port;
 719
 720     /* Don't report on the datapath's device. */
 721     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 722         return;
 723     }
 724
 725     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 726                    &all_ofproto_dpifs) {
 727         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 728             return;
 729         }
 730     }
 731
 732     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 733     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 734         /* The port was removed.  If we know the datapath,
 735          * report it through poll_set().  If we don't, it may be
 736          * notifying us of a removal we initiated, so ignore it.
 737          * If there's a pending ENOBUFS, let it stand, since
 738          * everything will be reevaluated. */
 739         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 740             sset_add(&ofproto->port_poll_set, devname);
 741             ofproto->port_poll_errno = 0;
 742         }
 743     } else if (!ofproto) {
 744         /* The port was added, but we don't know with which
 745          * ofproto we should associate it.  Delete it. */
 746         dpif_port_del(backer->dpif, port.port_no);
 747     } else {
 748         struct ofport_dpif *ofport;
 749
 750         ofport = ofport_dpif_cast(shash_find_data(
 751                                       &ofproto->up.port_by_name, devname));
 752         if (ofport
 753             && ofport->odp_port != port.port_no
 754             && !odp_port_to_ofport(backer, port.port_no))
 755         {
 756             /* 'ofport''s datapath port number has changed from
 757              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 758              * structures to match. */
 759             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 760             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 761             ofport->odp_port = port.port_no;
 762             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 763                         hash_odp_port(port.port_no));
 764             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 765             backer->need_revalidate = REV_RECONFIGURE;
 766         }
 767     }
 768     dpif_port_destroy(&port);
 769 }
 770
 771 /* Propagate 'error' to all ofprotos based on 'backer'. */
 772 static void
 773 process_dpif_port_error(struct dpif_backer *backer, int error)
 774 {
 775     struct ofproto_dpif *ofproto;
 776
 777     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 778         if (ofproto->backer == backer) {
 779             sset_clear(&ofproto->port_poll_set);
 780             ofproto->port_poll_errno = error;
 781         }
 782     }
 783 }
 784
 785 static void
 786 type_wait(const char *type)
 787 {
 788     struct dpif_backer *backer;
 789
 790     backer = shash_find_data(&all_dpif_backers, type);
 791     if (!backer) {
 792         /* This is not necessarily a problem, since backers are only
 793          * created on demand. */
 794         return;
 795     }
 796
 797     dpif_wait(backer->dpif);
 798 }
 799 \f
 800 /* Basic life-cycle. */
 801
 802 static int add_internal_flows(struct ofproto_dpif *);
 803
 804 static struct ofproto *
 805 alloc(void)
 806 {
 807     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
 808     return &ofproto->up;
 809 }
 810
 811 static void
 812 dealloc(struct ofproto *ofproto_)
 813 {
 814     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
 815     free(ofproto);
 816 }
 817
 818 static void
 819 close_dpif_backer(struct dpif_backer *backer)
 820 {
 821     ovs_assert(backer->refcount > 0);
 822
 823     if (--backer->refcount) {
 824         return;
 825     }
 826
 827     udpif_destroy(backer->udpif);
 828
 829     simap_destroy(&backer->tnl_backers);
 830     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
 831     hmap_destroy(&backer->odp_to_ofport_map);
 832     shash_find_and_delete(&all_dpif_backers, backer->type);
 833     recirc_id_pool_destroy(backer->rid_pool);
 834     free(backer->type);
 835     dpif_close(backer->dpif);
 836     free(backer);
 837 }
 838
 839 /* Datapath port slated for removal from datapath. */
 840 struct odp_garbage {
 841     struct list list_node;
 842     odp_port_t odp_port;
 843 };
 844
 845 static bool check_variable_length_userdata(struct dpif_backer *backer);
 846 static size_t check_max_mpls_depth(struct dpif_backer *backer);
 847 static bool check_recirc(struct dpif_backer *backer);
 848
 849 static int
 850 open_dpif_backer(const char *type, struct dpif_backer **backerp)
 851 {
 852     struct dpif_backer *backer;
 853     struct dpif_port_dump port_dump;
 854     struct dpif_port port;
 855     struct shash_node *node;
 856     struct list garbage_list;
 857     struct odp_garbage *garbage, *next;
 858
 859     struct sset names;
 860     char *backer_name;
 861     const char *name;
 862     int error;
 863
 864     backer = shash_find_data(&all_dpif_backers, type);
 865     if (backer) {
 866         backer->refcount++;
 867         *backerp = backer;
 868         return 0;
 869     }
 870
 871     backer_name = xasprintf("ovs-%s", type);
 872
 873     /* Remove any existing datapaths, since we assume we're the only
 874      * userspace controlling the datapath. */
 875     sset_init(&names);
 876     dp_enumerate_names(type, &names);
 877     SSET_FOR_EACH(name, &names) {
 878         struct dpif *old_dpif;
 879
 880         /* Don't remove our backer if it exists. */
 881         if (!strcmp(name, backer_name)) {
 882             continue;
 883         }
 884
 885         if (dpif_open(name, type, &old_dpif)) {
 886             VLOG_WARN("couldn't open old datapath %s to remove it", name);
 887         } else {
 888             dpif_delete(old_dpif);
 889             dpif_close(old_dpif);
 890         }
 891     }
 892     sset_destroy(&names);
 893
 894     backer = xmalloc(sizeof *backer);
 895
 896     error = dpif_create_and_open(backer_name, type, &backer->dpif);
 897     free(backer_name);
 898     if (error) {
 899         VLOG_ERR("failed to open datapath of type %s: %s", type,
 900                  ovs_strerror(error));
 901         free(backer);
 902         return error;
 903     }
 904     backer->udpif = udpif_create(backer, backer->dpif);
 905
 906     backer->type = xstrdup(type);
 907     backer->refcount = 1;
 908     hmap_init(&backer->odp_to_ofport_map);
 909     ovs_rwlock_init(&backer->odp_to_ofport_lock);
 910     backer->need_revalidate = 0;
 911     simap_init(&backer->tnl_backers);
 912     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
 913     *backerp = backer;
 914
 915     if (backer->recv_set_enable) {
 916         dpif_flow_flush(backer->dpif);
 917     }
 918
 919     /* Loop through the ports already on the datapath and remove any
 920      * that we don't need anymore. */
 921     list_init(&garbage_list);
 922     dpif_port_dump_start(&port_dump, backer->dpif);
 923     while (dpif_port_dump_next(&port_dump, &port)) {
 924         node = shash_find(&init_ofp_ports, port.name);
 925         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
 926             garbage = xmalloc(sizeof *garbage);
 927             garbage->odp_port = port.port_no;
 928             list_push_front(&garbage_list, &garbage->list_node);
 929         }
 930     }
 931     dpif_port_dump_done(&port_dump);
 932
 933     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
 934         dpif_port_del(backer->dpif, garbage->odp_port);
 935         list_remove(&garbage->list_node);
 936         free(garbage);
 937     }
 938
 939     shash_add(&all_dpif_backers, type, backer);
 940
 941     backer->enable_recirc = check_recirc(backer);
 942     backer->variable_length_userdata = check_variable_length_userdata(backer);
 943     backer->max_mpls_depth = check_max_mpls_depth(backer);
 944     backer->rid_pool = recirc_id_pool_create();
 945
 946     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 947     if (error) {
 948         VLOG_ERR("failed to listen on datapath of type %s: %s",
 949                  type, ovs_strerror(error));
 950         close_dpif_backer(backer);
 951         return error;
 952     }
 953
 954     if (backer->recv_set_enable) {
 955         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 956     }
 957
 958     return error;
 959 }
 960
 961 /* Tests whether 'backer''s datapath supports recirculation.  Only newer
 962  * datapaths support OVS_KEY_ATTR_RECIRC_ID in keys.  We need to disable some
 963  * features on older datapaths that don't support this feature.
 964  *
 965  * Returns false if 'backer' definitely does not support recirculation, true if
 966  * it seems to support recirculation or if at least the error we get is
 967  * ambiguous. */
 968 static bool
 969 check_recirc(struct dpif_backer *backer)
 970 {
 971     struct flow flow;
 972     struct odputil_keybuf keybuf;
 973     struct ofpbuf key;
 974     int error;
 975     bool enable_recirc = false;
 976
 977     memset(&flow, 0, sizeof flow);
 978     flow.recirc_id = 1;
 979     flow.dp_hash = 1;
 980
 981     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
 982     odp_flow_key_from_flow(&key, &flow, NULL, 0, true);
 983
 984     error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE,
 985                           ofpbuf_data(&key), ofpbuf_size(&key), NULL, 0, NULL,
 986                           0, NULL);
 987     if (error && error != EEXIST) {
 988         if (error != EINVAL) {
 989             VLOG_WARN("%s: Reciculation flow probe failed (%s)",
 990                       dpif_name(backer->dpif), ovs_strerror(error));
 991         }
 992         goto done;
 993     }
 994
 995     error = dpif_flow_del(backer->dpif, ofpbuf_data(&key), ofpbuf_size(&key),
 996                           NULL);
 997     if (error) {
 998         VLOG_WARN("%s: failed to delete recirculation feature probe flow",
 999                   dpif_name(backer->dpif));
1000     }
1001
1002     enable_recirc = true;
1003
1004 done:
1005     if (enable_recirc) {
1006         VLOG_INFO("%s: Datapath supports recirculation",
1007                   dpif_name(backer->dpif));
1008     } else {
1009         VLOG_INFO("%s: Datapath does not support recirculation",
1010                   dpif_name(backer->dpif));
1011     }
1012
1013     return enable_recirc;
1014 }
1015
1016 /* Tests whether 'backer''s datapath supports variable-length
1017  * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.  We need
1018  * to disable some features on older datapaths that don't support this
1019  * feature.
1020  *
1021  * Returns false if 'backer' definitely does not support variable-length
1022  * userdata, true if it seems to support them or if at least the error we get
1023  * is ambiguous. */
1024 static bool
1025 check_variable_length_userdata(struct dpif_backer *backer)
1026 {
1027     struct eth_header *eth;
1028     struct ofpbuf actions;
1029     struct dpif_execute execute;
1030     struct ofpbuf packet;
1031     size_t start;
1032     int error;
1033
1034     /* Compose a userspace action that will cause an ERANGE error on older
1035      * datapaths that don't support variable-length userdata.
1036      *
1037      * We really test for using userdata longer than 8 bytes, but older
1038      * datapaths accepted these, silently truncating the userdata to 8 bytes.
1039      * The same older datapaths rejected userdata shorter than 8 bytes, so we
1040      * test for that instead as a proxy for longer userdata support. */
1041     ofpbuf_init(&actions, 64);
1042     start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
1043     nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
1044                    dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
1045     nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
1046     nl_msg_end_nested(&actions, start);
1047
1048     /* Compose a dummy ethernet packet. */
1049     ofpbuf_init(&packet, ETH_HEADER_LEN);
1050     eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
1051     eth->eth_type = htons(0x1234);
1052
1053     /* Execute the actions.  On older datapaths this fails with ERANGE, on
1054      * newer datapaths it succeeds. */
1055     execute.actions = ofpbuf_data(&actions);
1056     execute.actions_len = ofpbuf_size(&actions);
1057     execute.packet = &packet;
1058     execute.md = PKT_METADATA_INITIALIZER(0);
1059     execute.needs_help = false;
1060
1061     error = dpif_execute(backer->dpif, &execute);
1062
1063     ofpbuf_uninit(&packet);
1064     ofpbuf_uninit(&actions);
1065
1066     switch (error) {
1067     case 0:
1068         return true;
1069
1070     case ERANGE:
1071         /* Variable-length userdata is not supported. */
1072         VLOG_WARN("%s: datapath does not support variable-length userdata "
1073                   "feature (needs Linux 3.10+ or kernel module from OVS "
1074                   "1..11+).  The NXAST_SAMPLE action will be ignored.",
1075                   dpif_name(backer->dpif));
1076         return false;
1077
1078     default:
1079         /* Something odd happened.  We're not sure whether variable-length
1080          * userdata is supported.  Default to "yes". */
1081         VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1082                   dpif_name(backer->dpif), ovs_strerror(error));
1083         return true;
1084     }
1085 }
1086
1087 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1088  *
1089  * Returns the number of elements in a struct flow's mpls_lse field
1090  * if the datapath supports at least that many entries in an
1091  * MPLS label stack.
1092  * Otherwise returns the number of MPLS push actions supported by
1093  * the datapath. */
1094 static size_t
1095 check_max_mpls_depth(struct dpif_backer *backer)
1096 {
1097     struct flow flow;
1098     int n;
1099
1100     for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
1101         struct odputil_keybuf keybuf;
1102         struct ofpbuf key;
1103         int error;
1104
1105         memset(&flow, 0, sizeof flow);
1106         flow.dl_type = htons(ETH_TYPE_MPLS);
1107         flow_set_mpls_bos(&flow, n, 1);
1108
1109         ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1110         odp_flow_key_from_flow(&key, &flow, NULL, 0, false);
1111
1112         error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE,
1113                               ofpbuf_data(&key), ofpbuf_size(&key), NULL, 0, NULL, 0, NULL);
1114         if (error && error != EEXIST) {
1115             if (error != EINVAL) {
1116                 VLOG_WARN("%s: MPLS stack length feature probe failed (%s)",
1117                           dpif_name(backer->dpif), ovs_strerror(error));
1118             }
1119             break;
1120         }
1121
1122         error = dpif_flow_del(backer->dpif, ofpbuf_data(&key), ofpbuf_size(&key), NULL);
1123         if (error) {
1124             VLOG_WARN("%s: failed to delete MPLS feature probe flow",
1125                       dpif_name(backer->dpif));
1126         }
1127     }
1128
1129     VLOG_INFO("%s: MPLS label stack length probed as %d",
1130               dpif_name(backer->dpif), n);
1131     return n;
1132 }
1133
1134 static int
1135 construct(struct ofproto *ofproto_)
1136 {
1137     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1138     struct shash_node *node, *next;
1139     int error;
1140
1141     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1142     if (error) {
1143         return error;
1144     }
1145
1146     ofproto->netflow = NULL;
1147     ofproto->sflow = NULL;
1148     ofproto->ipfix = NULL;
1149     ofproto->stp = NULL;
1150     ofproto->rstp = NULL;
1151     ofproto->dump_seq = 0;
1152     hmap_init(&ofproto->bundles);
1153     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1154     ofproto->ms = NULL;
1155     ofproto->mbridge = mbridge_create();
1156     ofproto->has_bonded_bundles = false;
1157     ofproto->lacp_enabled = false;
1158     ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1159     ovs_mutex_init(&ofproto->vsp_mutex);
1160
1161     guarded_list_init(&ofproto->pins);
1162
1163     ofproto_dpif_unixctl_init();
1164
1165     hmap_init(&ofproto->vlandev_map);
1166     hmap_init(&ofproto->realdev_vid_map);
1167
1168     sset_init(&ofproto->ports);
1169     sset_init(&ofproto->ghost_ports);
1170     sset_init(&ofproto->port_poll_set);
1171     ofproto->port_poll_errno = 0;
1172     ofproto->change_seq = 0;
1173     ofproto->pins_seq = seq_create();
1174     ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1175
1176
1177     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1178         struct iface_hint *iface_hint = node->data;
1179
1180         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1181             /* Check if the datapath already has this port. */
1182             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1183                 sset_add(&ofproto->ports, node->name);
1184             }
1185
1186             free(iface_hint->br_name);
1187             free(iface_hint->br_type);
1188             free(iface_hint);
1189             shash_delete(&init_ofp_ports, node);
1190         }
1191     }
1192
1193     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1194                 hash_string(ofproto->up.name, 0));
1195     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1196
1197     ofproto_init_tables(ofproto_, N_TABLES);
1198     error = add_internal_flows(ofproto);
1199
1200     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1201
1202     return error;
1203 }
1204
1205 static int
1206 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1207                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1208 {
1209     struct match match;
1210     int error;
1211     struct rule *rule;
1212
1213     match_init_catchall(&match);
1214     match_set_reg(&match, 0, id);
1215
1216     error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, 0, ofpacts,
1217                                            &rule);
1218     *rulep = error ? NULL : rule_dpif_cast(rule);
1219
1220     return error;
1221 }
1222
1223 static int
1224 add_internal_flows(struct ofproto_dpif *ofproto)
1225 {
1226     struct ofpact_controller *controller;
1227     uint64_t ofpacts_stub[128 / 8];
1228     struct ofpbuf ofpacts;
1229     struct rule *unused_rulep OVS_UNUSED;
1230     struct ofpact_resubmit *resubmit;
1231     struct match match;
1232     int error;
1233     int id;
1234
1235     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1236     id = 1;
1237
1238     controller = ofpact_put_CONTROLLER(&ofpacts);
1239     controller->max_len = UINT16_MAX;
1240     controller->controller_id = 0;
1241     controller->reason = OFPR_NO_MATCH;
1242     ofpact_pad(&ofpacts);
1243
1244     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1245                                    &ofproto->miss_rule);
1246     if (error) {
1247         return error;
1248     }
1249
1250     ofpbuf_clear(&ofpacts);
1251     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1252                                    &ofproto->no_packet_in_rule);
1253     if (error) {
1254         return error;
1255     }
1256
1257     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1258                                    &ofproto->drop_frags_rule);
1259     if (error) {
1260         return error;
1261     }
1262
1263     /* Continue non-recirculation rule lookups from table 0.
1264      *
1265      * (priority=2), recirc=0, actions=resubmit(, 0)
1266      */
1267     resubmit = ofpact_put_RESUBMIT(&ofpacts);
1268     resubmit->in_port = OFPP_IN_PORT;
1269     resubmit->table_id = 0;
1270
1271     match_init_catchall(&match);
1272     match_set_recirc_id(&match, 0);
1273
1274     error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, 0, &ofpacts,
1275                                            &unused_rulep);
1276     if (error) {
1277         return error;
1278     }
1279
1280     /* Drop any run away recirc rule lookups. Recirc_id has to be
1281      * non-zero when reaching this rule.
1282      *
1283      * (priority=1), *, actions=drop
1284      */
1285     ofpbuf_clear(&ofpacts);
1286     match_init_catchall(&match);
1287     error = ofproto_dpif_add_internal_flow(ofproto, &match, 1, 0, &ofpacts,
1288                                            &unused_rulep);
1289
1290     return error;
1291 }
1292
1293 static void
1294 destruct(struct ofproto *ofproto_)
1295 {
1296     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1297     struct ofproto_packet_in *pin, *next_pin;
1298     struct rule_dpif *rule;
1299     struct oftable *table;
1300     struct list pins;
1301
1302     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1303     xlate_txn_start();
1304     xlate_remove_ofproto(ofproto);
1305     xlate_txn_commit();
1306
1307     /* Ensure that the upcall processing threads have no remaining references
1308      * to the ofproto or anything in it. */
1309     udpif_synchronize(ofproto->backer->udpif);
1310
1311     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1312
1313     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1314         CLS_FOR_EACH_SAFE (rule, up.cr, &table->cls) {
1315             ofproto_rule_delete(&ofproto->up, &rule->up);
1316         }
1317     }
1318
1319     guarded_list_pop_all(&ofproto->pins, &pins);
1320     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1321         list_remove(&pin->list_node);
1322         free(CONST_CAST(void *, pin->up.packet));
1323         free(pin);
1324     }
1325     guarded_list_destroy(&ofproto->pins);
1326
1327     mbridge_unref(ofproto->mbridge);
1328
1329     netflow_unref(ofproto->netflow);
1330     dpif_sflow_unref(ofproto->sflow);
1331     dpif_ipfix_unref(ofproto->ipfix);
1332     hmap_destroy(&ofproto->bundles);
1333     mac_learning_unref(ofproto->ml);
1334     mcast_snooping_unref(ofproto->ms);
1335
1336     hmap_destroy(&ofproto->vlandev_map);
1337     hmap_destroy(&ofproto->realdev_vid_map);
1338
1339     sset_destroy(&ofproto->ports);
1340     sset_destroy(&ofproto->ghost_ports);
1341     sset_destroy(&ofproto->port_poll_set);
1342
1343     ovs_mutex_destroy(&ofproto->stats_mutex);
1344     ovs_mutex_destroy(&ofproto->vsp_mutex);
1345
1346     seq_destroy(ofproto->pins_seq);
1347
1348     close_dpif_backer(ofproto->backer);
1349 }
1350
1351 static int
1352 run(struct ofproto *ofproto_)
1353 {
1354     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1355     uint64_t new_seq, new_dump_seq;
1356
1357     if (mbridge_need_revalidate(ofproto->mbridge)) {
1358         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1359         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1360         mac_learning_flush(ofproto->ml);
1361         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1362         mcast_snooping_mdb_flush(ofproto->ms);
1363     }
1364
1365     /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
1366      * flow restore.  Even though nothing is processed during flow restore,
1367      * all queued 'pins' will be handled immediately when flow restore
1368      * completes. */
1369     ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1370
1371     /* Do not perform any periodic activity required by 'ofproto' while
1372      * waiting for flow restore to complete. */
1373     if (!ofproto_get_flow_restore_wait()) {
1374         struct ofproto_packet_in *pin, *next_pin;
1375         struct list pins;
1376
1377         guarded_list_pop_all(&ofproto->pins, &pins);
1378         LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1379             connmgr_send_packet_in(ofproto->up.connmgr, pin);
1380             list_remove(&pin->list_node);
1381             free(CONST_CAST(void *, pin->up.packet));
1382             free(pin);
1383         }
1384     }
1385
1386     if (ofproto->netflow) {
1387         netflow_run(ofproto->netflow);
1388     }
1389     if (ofproto->sflow) {
1390         dpif_sflow_run(ofproto->sflow);
1391     }
1392     if (ofproto->ipfix) {
1393         dpif_ipfix_run(ofproto->ipfix);
1394     }
1395
1396     new_seq = seq_read(connectivity_seq_get());
1397     if (ofproto->change_seq != new_seq) {
1398         struct ofport_dpif *ofport;
1399
1400         HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1401             port_run(ofport);
1402         }
1403
1404         ofproto->change_seq = new_seq;
1405     }
1406     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1407         struct ofbundle *bundle;
1408
1409         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1410             bundle_run(bundle);
1411         }
1412     }
1413
1414     stp_run(ofproto);
1415     rstp_run(ofproto);
1416     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1417     if (mac_learning_run(ofproto->ml)) {
1418         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1419     }
1420     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1421
1422     if (mcast_snooping_run(ofproto->ms)) {
1423         ofproto->backer->need_revalidate = REV_MCAST_SNOOPING;
1424     }
1425
1426     new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1427     if (ofproto->dump_seq != new_dump_seq) {
1428         struct rule *rule, *next_rule;
1429
1430         /* We know stats are relatively fresh, so now is a good time to do some
1431          * periodic work. */
1432         ofproto->dump_seq = new_dump_seq;
1433
1434         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1435          * has passed. */
1436         ovs_mutex_lock(&ofproto_mutex);
1437         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1438                             &ofproto->up.expirable) {
1439             rule_expire(rule_dpif_cast(rule));
1440         }
1441         ovs_mutex_unlock(&ofproto_mutex);
1442
1443         /* All outstanding data in existing flows has been accounted, so it's a
1444          * good time to do bond rebalancing. */
1445         if (ofproto->has_bonded_bundles) {
1446             struct ofbundle *bundle;
1447
1448             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1449                 if (bundle->bond) {
1450                     bond_rebalance(bundle->bond);
1451                 }
1452             }
1453         }
1454     }
1455
1456     return 0;
1457 }
1458
1459 static void
1460 wait(struct ofproto *ofproto_)
1461 {
1462     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1463
1464     if (ofproto_get_flow_restore_wait()) {
1465         return;
1466     }
1467
1468     if (ofproto->sflow) {
1469         dpif_sflow_wait(ofproto->sflow);
1470     }
1471     if (ofproto->ipfix) {
1472         dpif_ipfix_wait(ofproto->ipfix);
1473     }
1474     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1475         struct ofbundle *bundle;
1476
1477         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1478             bundle_wait(bundle);
1479         }
1480     }
1481     if (ofproto->netflow) {
1482         netflow_wait(ofproto->netflow);
1483     }
1484     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1485     mac_learning_wait(ofproto->ml);
1486     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1487     mcast_snooping_wait(ofproto->ms);
1488     stp_wait(ofproto);
1489     if (ofproto->backer->need_revalidate) {
1490         /* Shouldn't happen, but if it does just go around again. */
1491         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1492         poll_immediate_wake();
1493     }
1494
1495     seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1496     seq_wait(ofproto->pins_seq, ofproto->pins_seqno);
1497 }
1498
1499 static void
1500 type_get_memory_usage(const char *type, struct simap *usage)
1501 {
1502     struct dpif_backer *backer;
1503
1504     backer = shash_find_data(&all_dpif_backers, type);
1505     if (backer) {
1506         udpif_get_memory_usage(backer->udpif, usage);
1507     }
1508 }
1509
1510 static void
1511 flush(struct ofproto *ofproto_)
1512 {
1513     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1514     struct dpif_backer *backer = ofproto->backer;
1515
1516     if (backer) {
1517         udpif_flush(backer->udpif);
1518     }
1519 }
1520
1521 static void
1522 query_tables(struct ofproto *ofproto,
1523              struct ofputil_table_features *features,
1524              struct ofputil_table_stats *stats)
1525 {
1526     strcpy(features->name, "classifier");
1527
1528     if (stats) {
1529         int i;
1530
1531         for (i = 0; i < ofproto->n_tables; i++) {
1532             unsigned long missed, matched;
1533
1534             atomic_read_relaxed(&ofproto->tables[i].n_matched, &matched);
1535             atomic_read_relaxed(&ofproto->tables[i].n_missed, &missed);
1536
1537             stats[i].matched_count = matched;
1538             stats[i].lookup_count = matched + missed;
1539         }
1540     }
1541 }
1542
1543 static struct ofport *
1544 port_alloc(void)
1545 {
1546     struct ofport_dpif *port = xmalloc(sizeof *port);
1547     return &port->up;
1548 }
1549
1550 static void
1551 port_dealloc(struct ofport *port_)
1552 {
1553     struct ofport_dpif *port = ofport_dpif_cast(port_);
1554     free(port);
1555 }
1556
1557 static int
1558 port_construct(struct ofport *port_)
1559 {
1560     struct ofport_dpif *port = ofport_dpif_cast(port_);
1561     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1562     const struct netdev *netdev = port->up.netdev;
1563     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1564     struct dpif_port dpif_port;
1565     int error;
1566
1567     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1568     port->bundle = NULL;
1569     port->cfm = NULL;
1570     port->bfd = NULL;
1571     port->may_enable = true;
1572     port->stp_port = NULL;
1573     port->stp_state = STP_DISABLED;
1574     port->rstp_port = NULL;
1575     port->rstp_state = RSTP_DISABLED;
1576     port->is_tunnel = false;
1577     port->peer = NULL;
1578     port->qdscp = NULL;
1579     port->n_qdscp = 0;
1580     port->realdev_ofp_port = 0;
1581     port->vlandev_vid = 0;
1582     port->carrier_seq = netdev_get_carrier_resets(netdev);
1583     port->is_layer3 = netdev_vport_is_layer3(netdev);
1584
1585     if (netdev_vport_is_patch(netdev)) {
1586         /* By bailing out here, we don't submit the port to the sFlow module
1587          * to be considered for counter polling export.  This is correct
1588          * because the patch port represents an interface that sFlow considers
1589          * to be "internal" to the switch as a whole, and therefore not an
1590          * candidate for counter polling. */
1591         port->odp_port = ODPP_NONE;
1592         ofport_update_peer(port);
1593         return 0;
1594     }
1595
1596     error = dpif_port_query_by_name(ofproto->backer->dpif,
1597                                     netdev_vport_get_dpif_port(netdev, namebuf,
1598                                                                sizeof namebuf),
1599                                     &dpif_port);
1600     if (error) {
1601         return error;
1602     }
1603
1604     port->odp_port = dpif_port.port_no;
1605
1606     if (netdev_get_tunnel_config(netdev)) {
1607         tnl_port_add(port, port->up.netdev, port->odp_port);
1608         port->is_tunnel = true;
1609         if (ofproto->ipfix) {
1610            dpif_ipfix_add_tunnel_port(ofproto->ipfix, port_, port->odp_port);
1611         }
1612     } else {
1613         /* Sanity-check that a mapping doesn't already exist.  This
1614          * shouldn't happen for non-tunnel ports. */
1615         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1616             VLOG_ERR("port %s already has an OpenFlow port number",
1617                      dpif_port.name);
1618             dpif_port_destroy(&dpif_port);
1619             return EBUSY;
1620         }
1621
1622         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1623         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1624                     hash_odp_port(port->odp_port));
1625         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1626     }
1627     dpif_port_destroy(&dpif_port);
1628
1629     if (ofproto->sflow) {
1630         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1631     }
1632
1633     return 0;
1634 }
1635
1636 static void
1637 port_destruct(struct ofport *port_)
1638 {
1639     struct ofport_dpif *port = ofport_dpif_cast(port_);
1640     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1641     const char *devname = netdev_get_name(port->up.netdev);
1642     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1643     const char *dp_port_name;
1644
1645     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1646     xlate_txn_start();
1647     xlate_ofport_remove(port);
1648     xlate_txn_commit();
1649
1650     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1651                                               sizeof namebuf);
1652     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1653         /* The underlying device is still there, so delete it.  This
1654          * happens when the ofproto is being destroyed, since the caller
1655          * assumes that removal of attached ports will happen as part of
1656          * destruction. */
1657         if (!port->is_tunnel) {
1658             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1659         }
1660     }
1661
1662     if (port->peer) {
1663         port->peer->peer = NULL;
1664         port->peer = NULL;
1665     }
1666
1667     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1668         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1669         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1670         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1671     }
1672
1673     if (port->is_tunnel && ofproto->ipfix) {
1674        dpif_ipfix_del_tunnel_port(ofproto->ipfix, port->odp_port);
1675     }
1676
1677     tnl_port_del(port);
1678     sset_find_and_delete(&ofproto->ports, devname);
1679     sset_find_and_delete(&ofproto->ghost_ports, devname);
1680     bundle_remove(port_);
1681     set_cfm(port_, NULL);
1682     set_bfd(port_, NULL);
1683     if (port->stp_port) {
1684         stp_port_disable(port->stp_port);
1685     }
1686     if (port->rstp_port) {
1687         rstp_delete_port(port->rstp_port);
1688     }
1689     if (ofproto->sflow) {
1690         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1691     }
1692
1693     free(port->qdscp);
1694 }
1695
1696 static void
1697 port_modified(struct ofport *port_)
1698 {
1699     struct ofport_dpif *port = ofport_dpif_cast(port_);
1700
1701     if (port->bundle && port->bundle->bond) {
1702         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1703     }
1704
1705     if (port->cfm) {
1706         cfm_set_netdev(port->cfm, port->up.netdev);
1707     }
1708
1709     if (port->bfd) {
1710         bfd_set_netdev(port->bfd, port->up.netdev);
1711     }
1712
1713     ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1714                                      port->up.pp.hw_addr);
1715
1716     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1717                                                 port->odp_port)) {
1718         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1719             REV_RECONFIGURE;
1720     }
1721
1722     ofport_update_peer(port);
1723 }
1724
1725 static void
1726 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1727 {
1728     struct ofport_dpif *port = ofport_dpif_cast(port_);
1729     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1730     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1731
1732     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1733                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1734                    OFPUTIL_PC_NO_PACKET_IN)) {
1735         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1736
1737         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1738             bundle_update(port->bundle);
1739         }
1740     }
1741 }
1742
1743 static int
1744 set_sflow(struct ofproto *ofproto_,
1745           const struct ofproto_sflow_options *sflow_options)
1746 {
1747     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1748     struct dpif_sflow *ds = ofproto->sflow;
1749
1750     if (sflow_options) {
1751         if (!ds) {
1752             struct ofport_dpif *ofport;
1753
1754             ds = ofproto->sflow = dpif_sflow_create();
1755             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1756                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1757             }
1758             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1759         }
1760         dpif_sflow_set_options(ds, sflow_options);
1761     } else {
1762         if (ds) {
1763             dpif_sflow_unref(ds);
1764             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1765             ofproto->sflow = NULL;
1766         }
1767     }
1768     return 0;
1769 }
1770
1771 static int
1772 set_ipfix(
1773     struct ofproto *ofproto_,
1774     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1775     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1776     size_t n_flow_exporters_options)
1777 {
1778     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1779     struct dpif_ipfix *di = ofproto->ipfix;
1780     bool has_options = bridge_exporter_options || flow_exporters_options;
1781     bool new_di = false;
1782
1783     if (has_options && !di) {
1784         di = ofproto->ipfix = dpif_ipfix_create();
1785         new_di = true;
1786     }
1787
1788     if (di) {
1789         /* Call set_options in any case to cleanly flush the flow
1790          * caches in the last exporters that are to be destroyed. */
1791         dpif_ipfix_set_options(
1792             di, bridge_exporter_options, flow_exporters_options,
1793             n_flow_exporters_options);
1794
1795         /* Add tunnel ports only when a new ipfix created */
1796         if (new_di == true) {
1797             struct ofport_dpif *ofport;
1798             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1799                 if (ofport->is_tunnel == true) {
1800                     dpif_ipfix_add_tunnel_port(di, &ofport->up, ofport->odp_port);
1801                 }
1802             }
1803         }
1804
1805         if (!has_options) {
1806             dpif_ipfix_unref(di);
1807             ofproto->ipfix = NULL;
1808         }
1809     }
1810
1811     return 0;
1812 }
1813
1814 static int
1815 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1816 {
1817     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1818     int error = 0;
1819
1820     if (s) {
1821         if (!ofport->cfm) {
1822             struct ofproto_dpif *ofproto;
1823
1824             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1825             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1826             ofport->cfm = cfm_create(ofport->up.netdev);
1827         }
1828
1829         if (cfm_configure(ofport->cfm, s)) {
1830             error = 0;
1831             goto out;
1832         }
1833
1834         error = EINVAL;
1835     }
1836     cfm_unref(ofport->cfm);
1837     ofport->cfm = NULL;
1838 out:
1839     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1840                                      ofport->up.pp.hw_addr);
1841     return error;
1842 }
1843
1844 static bool
1845 cfm_status_changed(struct ofport *ofport_)
1846 {
1847     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1848
1849     return ofport->cfm ? cfm_check_status_change(ofport->cfm) : true;
1850 }
1851
1852 static int
1853 get_cfm_status(const struct ofport *ofport_,
1854                struct cfm_status *status)
1855 {
1856     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1857     int ret = 0;
1858
1859     if (ofport->cfm) {
1860         cfm_get_status(ofport->cfm, status);
1861     } else {
1862         ret = ENOENT;
1863     }
1864
1865     return ret;
1866 }
1867
1868 static int
1869 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1870 {
1871     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1872     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1873     struct bfd *old;
1874
1875     old = ofport->bfd;
1876     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1877                                 cfg, ofport->up.netdev);
1878     if (ofport->bfd != old) {
1879         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1880     }
1881     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1882                                      ofport->up.pp.hw_addr);
1883     return 0;
1884 }
1885
1886 static bool
1887 bfd_status_changed(struct ofport *ofport_)
1888 {
1889     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1890
1891     return ofport->bfd ? bfd_check_status_change(ofport->bfd) : true;
1892 }
1893
1894 static int
1895 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1896 {
1897     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1898     int ret = 0;
1899
1900     if (ofport->bfd) {
1901         bfd_get_status(ofport->bfd, smap);
1902     } else {
1903         ret = ENOENT;
1904     }
1905
1906     return ret;
1907 }
1908 \f
1909 /* Spanning Tree. */
1910
1911 static void
1912 rstp_send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1913 {
1914     struct ofproto_dpif *ofproto = ofproto_;
1915     struct rstp_port *rp = rstp_get_port(ofproto->rstp, port_num);
1916     struct ofport_dpif *ofport;
1917
1918     ofport = rstp_port_get_aux(rp);
1919     if (!ofport) {
1920         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1921                 ofproto->up.name, port_num);
1922     } else {
1923         struct eth_header *eth = ofpbuf_l2(pkt);
1924
1925         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1926         if (eth_addr_is_zero(eth->eth_src)) {
1927             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1928                     "with unknown MAC", ofproto->up.name, port_num);
1929         } else {
1930             ofproto_dpif_send_packet(ofport, pkt);
1931         }
1932     }
1933     ofpbuf_delete(pkt);
1934 }
1935
1936 static void
1937 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1938 {
1939     struct ofproto_dpif *ofproto = ofproto_;
1940     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1941     struct ofport_dpif *ofport;
1942
1943     ofport = stp_port_get_aux(sp);
1944     if (!ofport) {
1945         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1946                      ofproto->up.name, port_num);
1947     } else {
1948         struct eth_header *eth = ofpbuf_l2(pkt);
1949
1950         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1951         if (eth_addr_is_zero(eth->eth_src)) {
1952             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1953                          "with unknown MAC", ofproto->up.name, port_num);
1954         } else {
1955             ofproto_dpif_send_packet(ofport, pkt);
1956         }
1957     }
1958     ofpbuf_delete(pkt);
1959 }
1960
1961 /* Configure RSTP on 'ofproto_' using the settings defined in 's'. */
1962 static void
1963 set_rstp(struct ofproto *ofproto_, const struct ofproto_rstp_settings *s)
1964 {
1965     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1966
1967     /* Only revalidate flows if the configuration changed. */
1968     if (!s != !ofproto->rstp) {
1969         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1970     }
1971
1972     if (s) {
1973         if (!ofproto->rstp) {
1974             ofproto->rstp = rstp_create(ofproto_->name, s->address,
1975               rstp_send_bpdu_cb, ofproto);
1976             ofproto->rstp_last_tick = time_msec();
1977         }
1978         rstp_set_bridge_address(ofproto->rstp, s->address);
1979         rstp_set_bridge_priority(ofproto->rstp, s->priority);
1980         rstp_set_bridge_ageing_time(ofproto->rstp, s->ageing_time);
1981         rstp_set_bridge_force_protocol_version(ofproto->rstp,
1982                                                s->force_protocol_version);
1983         rstp_set_bridge_max_age(ofproto->rstp, s->bridge_max_age);
1984         rstp_set_bridge_forward_delay(ofproto->rstp, s->bridge_forward_delay);
1985         rstp_set_bridge_transmit_hold_count(ofproto->rstp,
1986                                             s->transmit_hold_count);
1987     } else {
1988         struct ofport *ofport;
1989         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1990             set_rstp_port(ofport, NULL);
1991         }
1992         rstp_unref(ofproto->rstp);
1993         ofproto->rstp = NULL;
1994     }
1995 }
1996
1997 static void
1998 get_rstp_status(struct ofproto *ofproto_, struct ofproto_rstp_status *s)
1999 {
2000     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2001
2002     if (ofproto->rstp) {
2003         s->enabled = true;
2004         s->root_id = rstp_get_root_id(ofproto->rstp);
2005         s->bridge_id = rstp_get_bridge_id(ofproto->rstp);
2006         s->designated_id = rstp_get_designated_id(ofproto->rstp);
2007         s->root_path_cost = rstp_get_root_path_cost(ofproto->rstp);
2008         s->designated_port_id = rstp_get_designated_port_id(ofproto->rstp);
2009         s->bridge_port_id = rstp_get_bridge_port_id(ofproto->rstp);
2010     } else {
2011         s->enabled = false;
2012     }
2013 }
2014
2015 static void
2016 update_rstp_port_state(struct ofport_dpif *ofport)
2017 {
2018     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2019     enum rstp_state state;
2020
2021     /* Figure out new state. */
2022     state = ofport->rstp_port ? rstp_port_get_state(ofport->rstp_port)
2023         : RSTP_DISABLED;
2024
2025     /* Update state. */
2026     if (ofport->rstp_state != state) {
2027         enum ofputil_port_state of_state;
2028         bool fwd_change;
2029
2030         VLOG_DBG("port %s: RSTP state changed from %s to %s",
2031                  netdev_get_name(ofport->up.netdev),
2032                  rstp_state_name(ofport->rstp_state),
2033                  rstp_state_name(state));
2034         if (rstp_learn_in_state(ofport->rstp_state)
2035                 != rstp_learn_in_state(state)) {
2036             /* xxx Learning action flows should also be flushed. */
2037             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2038             mac_learning_flush(ofproto->ml);
2039             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2040         }
2041         fwd_change = rstp_forward_in_state(ofport->rstp_state)
2042             != rstp_forward_in_state(state);
2043
2044         ofproto->backer->need_revalidate = REV_RSTP;
2045         ofport->rstp_state = state;
2046
2047         if (fwd_change && ofport->bundle) {
2048             bundle_update(ofport->bundle);
2049         }
2050
2051         /* Update the RSTP state bits in the OpenFlow port description. */
2052         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2053         of_state |= (state == RSTP_LEARNING ? OFPUTIL_PS_STP_LEARN
2054                 : state == RSTP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2055                 : state == RSTP_DISCARDING ?  OFPUTIL_PS_STP_LISTEN
2056                 : 0);
2057         ofproto_port_set_state(&ofport->up, of_state);
2058     }
2059 }
2060
2061 static void
2062 rstp_run(struct ofproto_dpif *ofproto)
2063 {
2064     if (ofproto->rstp) {
2065         long long int now = time_msec();
2066         long long int elapsed = now - ofproto->rstp_last_tick;
2067         struct rstp_port *rp;
2068         /* Every second, decrease the values of the timers. */
2069         if (elapsed >= 1000) {
2070             rstp_tick_timers(ofproto->rstp);
2071             ofproto->rstp_last_tick = now;
2072         }
2073         while (rstp_get_changed_port(ofproto->rstp, &rp)) {
2074             struct ofport_dpif *ofport = rstp_port_get_aux(rp);
2075             if (ofport) {
2076                 update_rstp_port_state(ofport);
2077             }
2078         }
2079         /* FIXME: This check should be done on-event (i.e., when setting
2080          * p->fdb_flush) and not periodically.
2081          */
2082         if (rstp_check_and_reset_fdb_flush(ofproto->rstp)) {
2083             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2084             /* FIXME: RSTP should be able to flush the entries pertaining to a
2085              * single port, not the whole table.
2086              */
2087             mac_learning_flush(ofproto->ml);
2088             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2089         }
2090     }
2091 }
2092
2093 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2094 static int
2095 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2096 {
2097     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2098
2099     /* Only revalidate flows if the configuration changed. */
2100     if (!s != !ofproto->stp) {
2101         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2102     }
2103
2104     if (s) {
2105         if (!ofproto->stp) {
2106             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2107                                       send_bpdu_cb, ofproto);
2108             ofproto->stp_last_tick = time_msec();
2109         }
2110
2111         stp_set_bridge_id(ofproto->stp, s->system_id);
2112         stp_set_bridge_priority(ofproto->stp, s->priority);
2113         stp_set_hello_time(ofproto->stp, s->hello_time);
2114         stp_set_max_age(ofproto->stp, s->max_age);
2115         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2116     }  else {
2117         struct ofport *ofport;
2118
2119         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2120             set_stp_port(ofport, NULL);
2121         }
2122
2123         stp_unref(ofproto->stp);
2124         ofproto->stp = NULL;
2125     }
2126
2127     return 0;
2128 }
2129
2130 static int
2131 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2132 {
2133     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2134
2135     if (ofproto->stp) {
2136         s->enabled = true;
2137         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2138         s->designated_root = stp_get_designated_root(ofproto->stp);
2139         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2140     } else {
2141         s->enabled = false;
2142     }
2143
2144     return 0;
2145 }
2146
2147 static void
2148 update_stp_port_state(struct ofport_dpif *ofport)
2149 {
2150     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2151     enum stp_state state;
2152
2153     /* Figure out new state. */
2154     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2155                              : STP_DISABLED;
2156
2157     /* Update state. */
2158     if (ofport->stp_state != state) {
2159         enum ofputil_port_state of_state;
2160         bool fwd_change;
2161
2162         VLOG_DBG("port %s: STP state changed from %s to %s",
2163                  netdev_get_name(ofport->up.netdev),
2164                  stp_state_name(ofport->stp_state),
2165                  stp_state_name(state));
2166         if (stp_learn_in_state(ofport->stp_state)
2167                 != stp_learn_in_state(state)) {
2168             /* xxx Learning action flows should also be flushed. */
2169             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2170             mac_learning_flush(ofproto->ml);
2171             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2172             mcast_snooping_mdb_flush(ofproto->ms);
2173         }
2174         fwd_change = stp_forward_in_state(ofport->stp_state)
2175                         != stp_forward_in_state(state);
2176
2177         ofproto->backer->need_revalidate = REV_STP;
2178         ofport->stp_state = state;
2179         ofport->stp_state_entered = time_msec();
2180
2181         if (fwd_change && ofport->bundle) {
2182             bundle_update(ofport->bundle);
2183         }
2184
2185         /* Update the STP state bits in the OpenFlow port description. */
2186         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2187         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2188                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2189                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2190                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2191                      : 0);
2192         ofproto_port_set_state(&ofport->up, of_state);
2193     }
2194 }
2195
2196 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2197  * caller is responsible for assigning STP port numbers and ensuring
2198  * there are no duplicates. */
2199 static int
2200 set_stp_port(struct ofport *ofport_,
2201              const struct ofproto_port_stp_settings *s)
2202 {
2203     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2204     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2205     struct stp_port *sp = ofport->stp_port;
2206
2207     if (!s || !s->enable) {
2208         if (sp) {
2209             ofport->stp_port = NULL;
2210             stp_port_disable(sp);
2211             update_stp_port_state(ofport);
2212         }
2213         return 0;
2214     } else if (sp && stp_port_no(sp) != s->port_num
2215             && ofport == stp_port_get_aux(sp)) {
2216         /* The port-id changed, so disable the old one if it's not
2217          * already in use by another port. */
2218         stp_port_disable(sp);
2219     }
2220
2221     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2222     stp_port_enable(sp);
2223
2224     stp_port_set_name(sp, netdev_get_name(ofport->up.netdev));
2225     stp_port_set_aux(sp, ofport);
2226     stp_port_set_priority(sp, s->priority);
2227     stp_port_set_path_cost(sp, s->path_cost);
2228
2229     update_stp_port_state(ofport);
2230
2231     return 0;
2232 }
2233
2234 static int
2235 get_stp_port_status(struct ofport *ofport_,
2236                     struct ofproto_port_stp_status *s)
2237 {
2238     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2239     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2240     struct stp_port *sp = ofport->stp_port;
2241
2242     if (!ofproto->stp || !sp) {
2243         s->enabled = false;
2244         return 0;
2245     }
2246
2247     s->enabled = true;
2248     s->port_id = stp_port_get_id(sp);
2249     s->state = stp_port_get_state(sp);
2250     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2251     s->role = stp_port_get_role(sp);
2252
2253     return 0;
2254 }
2255
2256 static int
2257 get_stp_port_stats(struct ofport *ofport_,
2258                    struct ofproto_port_stp_stats *s)
2259 {
2260     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2261     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2262     struct stp_port *sp = ofport->stp_port;
2263
2264     if (!ofproto->stp || !sp) {
2265         s->enabled = false;
2266         return 0;
2267     }
2268
2269     s->enabled = true;
2270     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2271
2272     return 0;
2273 }
2274
2275 static void
2276 stp_run(struct ofproto_dpif *ofproto)
2277 {
2278     if (ofproto->stp) {
2279         long long int now = time_msec();
2280         long long int elapsed = now - ofproto->stp_last_tick;
2281         struct stp_port *sp;
2282
2283         if (elapsed > 0) {
2284             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2285             ofproto->stp_last_tick = now;
2286         }
2287         while (stp_get_changed_port(ofproto->stp, &sp)) {
2288             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2289
2290             if (ofport) {
2291                 update_stp_port_state(ofport);
2292             }
2293         }
2294
2295         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2296             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2297             mac_learning_flush(ofproto->ml);
2298             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2299             mcast_snooping_mdb_flush(ofproto->ms);
2300         }
2301     }
2302 }
2303
2304 static void
2305 stp_wait(struct ofproto_dpif *ofproto)
2306 {
2307     if (ofproto->stp) {
2308         poll_timer_wait(1000);
2309     }
2310 }
2311
2312 /* Configures RSTP on 'ofport_' using the settings defined in 's'.  The
2313  * caller is responsible for assigning RSTP port numbers and ensuring
2314  * there are no duplicates. */
2315 static void
2316 set_rstp_port(struct ofport *ofport_,
2317         const struct ofproto_port_rstp_settings *s)
2318 {
2319     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2320     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2321     struct rstp_port *rp = ofport->rstp_port;
2322     int stp_port;
2323
2324     if (!s || !s->enable) {
2325         if (rp) {
2326             ofport->rstp_port = NULL;
2327             rstp_delete_port(rp);
2328             update_rstp_port_state(ofport);
2329         }
2330         return;
2331     } else if (rp && rstp_port_number(rp) != s->port_num
2332                   && ofport == rstp_port_get_aux(rp)) {
2333         /* The port-id changed, so disable the old one if it's not
2334          * already in use by another port. */
2335         if (s->port_num != 0) {
2336             xlate_txn_start();
2337             stp_port = ofport->stp_port ? stp_port_no(ofport->stp_port) : -1;
2338             xlate_ofport_set(ofproto, ofport->bundle, ofport,
2339                     ofport->up.ofp_port, ofport->odp_port,
2340                     ofport->up.netdev, ofport->cfm,
2341                     ofport->bfd, ofport->peer, stp_port,
2342                     s->port_num,
2343                     ofport->qdscp, ofport->n_qdscp,
2344                     ofport->up.pp.config, ofport->up.pp.state,
2345                     ofport->is_tunnel, ofport->may_enable);
2346             xlate_txn_commit();
2347         }
2348
2349         rstp_port_set_aux(rp, ofport);
2350         rstp_port_set_priority(rp, s->priority);
2351         rstp_port_set_port_number(rp, s->port_num);
2352         rstp_port_set_path_cost(rp, s->path_cost);
2353         rstp_port_set_admin_edge(rp, s->admin_edge_port);
2354         rstp_port_set_auto_edge(rp, s->auto_edge);
2355         rstp_port_set_mcheck(rp, s->mcheck);
2356
2357         update_rstp_port_state(ofport);
2358
2359         return;
2360     }
2361     rp = ofport->rstp_port = rstp_get_port(ofproto->rstp, s->port_num);
2362     /* Enable RSTP on port */
2363     if (!rp) {
2364         rp = ofport->rstp_port = rstp_add_port(ofproto->rstp);
2365     }
2366     /* Setters */
2367     rstp_port_set_aux(rp, ofport);
2368     rstp_port_set_priority(rp, s->priority);
2369     rstp_port_set_port_number(rp, s->port_num);
2370     rstp_port_set_path_cost(rp, s->path_cost);
2371     rstp_port_set_admin_edge(rp, s->admin_edge_port);
2372     rstp_port_set_auto_edge(rp, s->auto_edge);
2373     rstp_port_set_mcheck(rp, s->mcheck);
2374
2375     update_rstp_port_state(ofport);
2376 }
2377
2378 static void
2379 get_rstp_port_status(struct ofport *ofport_,
2380         struct ofproto_port_rstp_status *s)
2381 {
2382     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2383     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2384     struct rstp_port *rp = ofport->rstp_port;
2385
2386     if (!ofproto->rstp || !rp) {
2387         s->enabled = false;
2388         return;
2389     }
2390
2391     s->enabled = true;
2392     s->port_id = rstp_port_get_id(rp);
2393     s->state = rstp_port_get_state(rp);
2394     s->role = rstp_port_get_role(rp);
2395     rstp_port_get_counts(rp, &s->tx_count, &s->rx_count,
2396                          &s->error_count, &s->uptime);
2397 }
2398
2399 \f
2400 static int
2401 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2402            size_t n_qdscp)
2403 {
2404     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2405     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2406
2407     if (ofport->n_qdscp != n_qdscp
2408         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2409                               n_qdscp * sizeof *qdscp))) {
2410         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2411         free(ofport->qdscp);
2412         ofport->qdscp = n_qdscp
2413             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2414             : NULL;
2415         ofport->n_qdscp = n_qdscp;
2416     }
2417
2418     return 0;
2419 }
2420 \f
2421 /* Bundles. */
2422
2423 /* Expires all MAC learning entries associated with 'bundle' and forces its
2424  * ofproto to revalidate every flow.
2425  *
2426  * Normally MAC learning entries are removed only from the ofproto associated
2427  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2428  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2429  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2430  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2431  * with the host from which it migrated. */
2432 static void
2433 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2434 {
2435     struct ofproto_dpif *ofproto = bundle->ofproto;
2436     struct mac_learning *ml = ofproto->ml;
2437     struct mac_entry *mac, *next_mac;
2438
2439     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2440     ovs_rwlock_wrlock(&ml->rwlock);
2441     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2442         if (mac->port.p == bundle) {
2443             if (all_ofprotos) {
2444                 struct ofproto_dpif *o;
2445
2446                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2447                     if (o != ofproto) {
2448                         struct mac_entry *e;
2449
2450                         ovs_rwlock_wrlock(&o->ml->rwlock);
2451                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2452                         if (e) {
2453                             mac_learning_expire(o->ml, e);
2454                         }
2455                         ovs_rwlock_unlock(&o->ml->rwlock);
2456                     }
2457                 }
2458             }
2459
2460             mac_learning_expire(ml, mac);
2461         }
2462     }
2463     ovs_rwlock_unlock(&ml->rwlock);
2464 }
2465
2466 static struct ofbundle *
2467 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2468 {
2469     struct ofbundle *bundle;
2470
2471     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2472                              &ofproto->bundles) {
2473         if (bundle->aux == aux) {
2474             return bundle;
2475         }
2476     }
2477     return NULL;
2478 }
2479
2480 static void
2481 bundle_update(struct ofbundle *bundle)
2482 {
2483     struct ofport_dpif *port;
2484
2485     bundle->floodable = true;
2486     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2487         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2488             || port->is_layer3
2489             || !stp_forward_in_state(port->stp_state)) {
2490             bundle->floodable = false;
2491             break;
2492         }
2493     }
2494 }
2495
2496 static void
2497 bundle_del_port(struct ofport_dpif *port)
2498 {
2499     struct ofbundle *bundle = port->bundle;
2500
2501     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2502
2503     list_remove(&port->bundle_node);
2504     port->bundle = NULL;
2505
2506     if (bundle->lacp) {
2507         lacp_slave_unregister(bundle->lacp, port);
2508     }
2509     if (bundle->bond) {
2510         bond_slave_unregister(bundle->bond, port);
2511     }
2512
2513     bundle_update(bundle);
2514 }
2515
2516 static bool
2517 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2518                 struct lacp_slave_settings *lacp)
2519 {
2520     struct ofport_dpif *port;
2521
2522     port = get_ofp_port(bundle->ofproto, ofp_port);
2523     if (!port) {
2524         return false;
2525     }
2526
2527     if (port->bundle != bundle) {
2528         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2529         if (port->bundle) {
2530             bundle_remove(&port->up);
2531         }
2532
2533         port->bundle = bundle;
2534         list_push_back(&bundle->ports, &port->bundle_node);
2535         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2536             || port->is_layer3
2537             || !stp_forward_in_state(port->stp_state)) {
2538             bundle->floodable = false;
2539         }
2540     }
2541     if (lacp) {
2542         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2543         lacp_slave_register(bundle->lacp, port, lacp);
2544     }
2545
2546     return true;
2547 }
2548
2549 static void
2550 bundle_destroy(struct ofbundle *bundle)
2551 {
2552     struct ofproto_dpif *ofproto;
2553     struct ofport_dpif *port, *next_port;
2554
2555     if (!bundle) {
2556         return;
2557     }
2558
2559     ofproto = bundle->ofproto;
2560     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2561
2562     xlate_txn_start();
2563     xlate_bundle_remove(bundle);
2564     xlate_txn_commit();
2565
2566     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2567         bundle_del_port(port);
2568     }
2569
2570     bundle_flush_macs(bundle, true);
2571     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2572     free(bundle->name);
2573     free(bundle->trunks);
2574     lacp_unref(bundle->lacp);
2575     bond_unref(bundle->bond);
2576     free(bundle);
2577 }
2578
2579 static int
2580 bundle_set(struct ofproto *ofproto_, void *aux,
2581            const struct ofproto_bundle_settings *s)
2582 {
2583     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2584     bool need_flush = false;
2585     struct ofport_dpif *port;
2586     struct ofbundle *bundle;
2587     unsigned long *trunks;
2588     int vlan;
2589     size_t i;
2590     bool ok;
2591
2592     if (!s) {
2593         bundle_destroy(bundle_lookup(ofproto, aux));
2594         return 0;
2595     }
2596
2597     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2598     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2599
2600     bundle = bundle_lookup(ofproto, aux);
2601     if (!bundle) {
2602         bundle = xmalloc(sizeof *bundle);
2603
2604         bundle->ofproto = ofproto;
2605         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2606                     hash_pointer(aux, 0));
2607         bundle->aux = aux;
2608         bundle->name = NULL;
2609
2610         list_init(&bundle->ports);
2611         bundle->vlan_mode = PORT_VLAN_TRUNK;
2612         bundle->vlan = -1;
2613         bundle->trunks = NULL;
2614         bundle->use_priority_tags = s->use_priority_tags;
2615         bundle->lacp = NULL;
2616         bundle->bond = NULL;
2617
2618         bundle->floodable = true;
2619         mbridge_register_bundle(ofproto->mbridge, bundle);
2620     }
2621
2622     if (!bundle->name || strcmp(s->name, bundle->name)) {
2623         free(bundle->name);
2624         bundle->name = xstrdup(s->name);
2625     }
2626
2627     /* LACP. */
2628     if (s->lacp) {
2629         ofproto->lacp_enabled = true;
2630         if (!bundle->lacp) {
2631             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2632             bundle->lacp = lacp_create();
2633         }
2634         lacp_configure(bundle->lacp, s->lacp);
2635     } else {
2636         lacp_unref(bundle->lacp);
2637         bundle->lacp = NULL;
2638     }
2639
2640     /* Update set of ports. */
2641     ok = true;
2642     for (i = 0; i < s->n_slaves; i++) {
2643         if (!bundle_add_port(bundle, s->slaves[i],
2644                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2645             ok = false;
2646         }
2647     }
2648     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2649         struct ofport_dpif *next_port;
2650
2651         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2652             for (i = 0; i < s->n_slaves; i++) {
2653                 if (s->slaves[i] == port->up.ofp_port) {
2654                     goto found;
2655                 }
2656             }
2657
2658             bundle_del_port(port);
2659         found: ;
2660         }
2661     }
2662     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2663
2664     if (list_is_empty(&bundle->ports)) {
2665         bundle_destroy(bundle);
2666         return EINVAL;
2667     }
2668
2669     /* Set VLAN tagging mode */
2670     if (s->vlan_mode != bundle->vlan_mode
2671         || s->use_priority_tags != bundle->use_priority_tags) {
2672         bundle->vlan_mode = s->vlan_mode;
2673         bundle->use_priority_tags = s->use_priority_tags;
2674         need_flush = true;
2675     }
2676
2677     /* Set VLAN tag. */
2678     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2679             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2680             : 0);
2681     if (vlan != bundle->vlan) {
2682         bundle->vlan = vlan;
2683         need_flush = true;
2684     }
2685
2686     /* Get trunked VLANs. */
2687     switch (s->vlan_mode) {
2688     case PORT_VLAN_ACCESS:
2689         trunks = NULL;
2690         break;
2691
2692     case PORT_VLAN_TRUNK:
2693         trunks = CONST_CAST(unsigned long *, s->trunks);
2694         break;
2695
2696     case PORT_VLAN_NATIVE_UNTAGGED:
2697     case PORT_VLAN_NATIVE_TAGGED:
2698         if (vlan != 0 && (!s->trunks
2699                           || !bitmap_is_set(s->trunks, vlan)
2700                           || bitmap_is_set(s->trunks, 0))) {
2701             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2702             if (s->trunks) {
2703                 trunks = bitmap_clone(s->trunks, 4096);
2704             } else {
2705                 trunks = bitmap_allocate1(4096);
2706             }
2707             bitmap_set1(trunks, vlan);
2708             bitmap_set0(trunks, 0);
2709         } else {
2710             trunks = CONST_CAST(unsigned long *, s->trunks);
2711         }
2712         break;
2713
2714     default:
2715         OVS_NOT_REACHED();
2716     }
2717     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2718         free(bundle->trunks);
2719         if (trunks == s->trunks) {
2720             bundle->trunks = vlan_bitmap_clone(trunks);
2721         } else {
2722             bundle->trunks = trunks;
2723             trunks = NULL;
2724         }
2725         need_flush = true;
2726     }
2727     if (trunks != s->trunks) {
2728         free(trunks);
2729     }
2730
2731     /* Bonding. */
2732     if (!list_is_short(&bundle->ports)) {
2733         bundle->ofproto->has_bonded_bundles = true;
2734         if (bundle->bond) {
2735             if (bond_reconfigure(bundle->bond, s->bond)) {
2736                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2737             }
2738         } else {
2739             bundle->bond = bond_create(s->bond, ofproto);
2740             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2741         }
2742
2743         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2744             bond_slave_register(bundle->bond, port,
2745                                 port->up.ofp_port, port->up.netdev);
2746         }
2747     } else {
2748         bond_unref(bundle->bond);
2749         bundle->bond = NULL;
2750     }
2751
2752     /* If we changed something that would affect MAC learning, un-learn
2753      * everything on this port and force flow revalidation. */
2754     if (need_flush) {
2755         bundle_flush_macs(bundle, false);
2756     }
2757
2758     return 0;
2759 }
2760
2761 static void
2762 bundle_remove(struct ofport *port_)
2763 {
2764     struct ofport_dpif *port = ofport_dpif_cast(port_);
2765     struct ofbundle *bundle = port->bundle;
2766
2767     if (bundle) {
2768         bundle_del_port(port);
2769         if (list_is_empty(&bundle->ports)) {
2770             bundle_destroy(bundle);
2771         } else if (list_is_short(&bundle->ports)) {
2772             bond_unref(bundle->bond);
2773             bundle->bond = NULL;
2774         }
2775     }
2776 }
2777
2778 static void
2779 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2780 {
2781     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2782     struct ofport_dpif *port = port_;
2783     uint8_t ea[ETH_ADDR_LEN];
2784     int error;
2785
2786     error = netdev_get_etheraddr(port->up.netdev, ea);
2787     if (!error) {
2788         struct ofpbuf packet;
2789         void *packet_pdu;
2790
2791         ofpbuf_init(&packet, 0);
2792         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2793                                  pdu_size);
2794         memcpy(packet_pdu, pdu, pdu_size);
2795
2796         ofproto_dpif_send_packet(port, &packet);
2797         ofpbuf_uninit(&packet);
2798     } else {
2799         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2800                     "%s (%s)", port->bundle->name,
2801                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2802     }
2803 }
2804
2805 static void
2806 bundle_send_learning_packets(struct ofbundle *bundle)
2807 {
2808     struct ofproto_dpif *ofproto = bundle->ofproto;
2809     struct ofpbuf *learning_packet;
2810     int error, n_packets, n_errors;
2811     struct mac_entry *e;
2812     struct list packets;
2813
2814     list_init(&packets);
2815     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2816     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2817         if (e->port.p != bundle) {
2818             void *port_void;
2819
2820             learning_packet = bond_compose_learning_packet(bundle->bond,
2821                                                            e->mac, e->vlan,
2822                                                            &port_void);
2823             /* Temporarily use 'frame' as a private pointer (see below). */
2824             ovs_assert(learning_packet->frame == ofpbuf_data(learning_packet));
2825             learning_packet->frame = port_void;
2826             list_push_back(&packets, &learning_packet->list_node);
2827         }
2828     }
2829     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2830
2831     error = n_packets = n_errors = 0;
2832     LIST_FOR_EACH (learning_packet, list_node, &packets) {
2833         int ret;
2834         void *port_void = learning_packet->frame;
2835
2836         /* Restore 'frame'. */
2837         learning_packet->frame = ofpbuf_data(learning_packet);
2838         ret = ofproto_dpif_send_packet(port_void, learning_packet);
2839         if (ret) {
2840             error = ret;
2841             n_errors++;
2842         }
2843         n_packets++;
2844     }
2845     ofpbuf_list_delete(&packets);
2846
2847     if (n_errors) {
2848         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2849         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2850                      "packets, last error was: %s",
2851                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2852     } else {
2853         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2854                  bundle->name, n_packets);
2855     }
2856 }
2857
2858 static void
2859 bundle_run(struct ofbundle *bundle)
2860 {
2861     if (bundle->lacp) {
2862         lacp_run(bundle->lacp, send_pdu_cb);
2863     }
2864     if (bundle->bond) {
2865         struct ofport_dpif *port;
2866
2867         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2868             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2869         }
2870
2871         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2872             bundle->ofproto->backer->need_revalidate = REV_BOND;
2873         }
2874
2875         if (bond_should_send_learning_packets(bundle->bond)) {
2876             bundle_send_learning_packets(bundle);
2877         }
2878     }
2879 }
2880
2881 static void
2882 bundle_wait(struct ofbundle *bundle)
2883 {
2884     if (bundle->lacp) {
2885         lacp_wait(bundle->lacp);
2886     }
2887     if (bundle->bond) {
2888         bond_wait(bundle->bond);
2889     }
2890 }
2891 \f
2892 /* Mirrors. */
2893
2894 static int
2895 mirror_set__(struct ofproto *ofproto_, void *aux,
2896              const struct ofproto_mirror_settings *s)
2897 {
2898     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2899     struct ofbundle **srcs, **dsts;
2900     int error;
2901     size_t i;
2902
2903     if (!s) {
2904         mirror_destroy(ofproto->mbridge, aux);
2905         return 0;
2906     }
2907
2908     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2909     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2910
2911     for (i = 0; i < s->n_srcs; i++) {
2912         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2913     }
2914
2915     for (i = 0; i < s->n_dsts; i++) {
2916         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2917     }
2918
2919     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2920                        s->n_dsts, s->src_vlans,
2921                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2922     free(srcs);
2923     free(dsts);
2924     return error;
2925 }
2926
2927 static int
2928 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2929                    uint64_t *packets, uint64_t *bytes)
2930 {
2931     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2932                             bytes);
2933 }
2934
2935 static int
2936 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2937 {
2938     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2939     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2940     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2941         mac_learning_flush(ofproto->ml);
2942     }
2943     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2944     return 0;
2945 }
2946
2947 static bool
2948 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2949 {
2950     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2951     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2952     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2953 }
2954
2955 static void
2956 forward_bpdu_changed(struct ofproto *ofproto_)
2957 {
2958     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2959     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2960 }
2961
2962 static void
2963 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2964                      size_t max_entries)
2965 {
2966     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2967     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2968     mac_learning_set_idle_time(ofproto->ml, idle_time);
2969     mac_learning_set_max_entries(ofproto->ml, max_entries);
2970     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2971 }
2972
2973 /* Configures multicast snooping on 'ofport' using the settings
2974  * defined in 's'. */
2975 static int
2976 set_mcast_snooping(struct ofproto *ofproto_,
2977                    const struct ofproto_mcast_snooping_settings *s)
2978 {
2979     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2980
2981     /* Only revalidate flows if the configuration changed. */
2982     if (!s != !ofproto->ms) {
2983         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2984     }
2985
2986     if (s) {
2987         if (!ofproto->ms) {
2988             ofproto->ms = mcast_snooping_create();
2989         }
2990
2991         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
2992         mcast_snooping_set_idle_time(ofproto->ms, s->idle_time);
2993         mcast_snooping_set_max_entries(ofproto->ms, s->max_entries);
2994         if (mcast_snooping_set_flood_unreg(ofproto->ms, s->flood_unreg)) {
2995             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2996         }
2997         ovs_rwlock_unlock(&ofproto->ms->rwlock);
2998     } else {
2999         mcast_snooping_unref(ofproto->ms);
3000         ofproto->ms = NULL;
3001     }
3002
3003     return 0;
3004 }
3005
3006 /* Configures multicast snooping port's flood setting on 'ofproto'. */
3007 static int
3008 set_mcast_snooping_port(struct ofproto *ofproto_, void *aux, bool flood)
3009 {
3010     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3011     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3012
3013     if (ofproto->ms) {
3014         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3015         mcast_snooping_set_port_flood(ofproto->ms, bundle->vlan, bundle,
3016                                       flood);
3017         ovs_rwlock_unlock(&ofproto->ms->rwlock);
3018     }
3019     return 0;
3020 }
3021
3022 \f
3023 /* Ports. */
3024
3025 static struct ofport_dpif *
3026 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
3027 {
3028     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3029     return ofport ? ofport_dpif_cast(ofport) : NULL;
3030 }
3031
3032 static void
3033 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3034                             struct ofproto_port *ofproto_port,
3035                             struct dpif_port *dpif_port)
3036 {
3037     ofproto_port->name = dpif_port->name;
3038     ofproto_port->type = dpif_port->type;
3039     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3040 }
3041
3042 static void
3043 ofport_update_peer(struct ofport_dpif *ofport)
3044 {
3045     const struct ofproto_dpif *ofproto;
3046     struct dpif_backer *backer;
3047     char *peer_name;
3048
3049     if (!netdev_vport_is_patch(ofport->up.netdev)) {
3050         return;
3051     }
3052
3053     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
3054     backer->need_revalidate = REV_RECONFIGURE;
3055
3056     if (ofport->peer) {
3057         ofport->peer->peer = NULL;
3058         ofport->peer = NULL;
3059     }
3060
3061     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
3062     if (!peer_name) {
3063         return;
3064     }
3065
3066     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3067         struct ofport *peer_ofport;
3068         struct ofport_dpif *peer;
3069         char *peer_peer;
3070
3071         if (ofproto->backer != backer) {
3072             continue;
3073         }
3074
3075         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
3076         if (!peer_ofport) {
3077             continue;
3078         }
3079
3080         peer = ofport_dpif_cast(peer_ofport);
3081         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
3082         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
3083                                  peer_peer)) {
3084             ofport->peer = peer;
3085             ofport->peer->peer = ofport;
3086         }
3087         free(peer_peer);
3088
3089         break;
3090     }
3091     free(peer_name);
3092 }
3093
3094 static void
3095 port_run(struct ofport_dpif *ofport)
3096 {
3097     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3098     bool carrier_changed = carrier_seq != ofport->carrier_seq;
3099     bool enable = netdev_get_carrier(ofport->up.netdev);
3100     bool cfm_enable = false;
3101     bool bfd_enable = false;
3102
3103     ofport->carrier_seq = carrier_seq;
3104
3105     if (ofport->cfm) {
3106         int cfm_opup = cfm_get_opup(ofport->cfm);
3107
3108         cfm_enable = !cfm_get_fault(ofport->cfm);
3109
3110         if (cfm_opup >= 0) {
3111             cfm_enable = cfm_enable && cfm_opup;
3112         }
3113     }
3114
3115     if (ofport->bfd) {
3116         bfd_enable = bfd_forwarding(ofport->bfd);
3117     }
3118
3119     if (ofport->bfd || ofport->cfm) {
3120         enable = enable && (cfm_enable || bfd_enable);
3121     }
3122
3123     if (ofport->bundle) {
3124         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3125         if (carrier_changed) {
3126             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3127         }
3128     }
3129
3130     if (ofport->may_enable != enable) {
3131         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3132         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3133     }
3134
3135     ofport->may_enable = enable;
3136
3137     if (ofport->rstp_port) {
3138         if (rstp_port_get_mac_operational(ofport->rstp_port) != enable) {
3139             rstp_port_set_mac_operational(ofport->rstp_port, enable);
3140         }
3141     }
3142 }
3143
3144 static int
3145 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3146                    struct ofproto_port *ofproto_port)
3147 {
3148     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3149     struct dpif_port dpif_port;
3150     int error;
3151
3152     if (sset_contains(&ofproto->ghost_ports, devname)) {
3153         const char *type = netdev_get_type_from_name(devname);
3154
3155         /* We may be called before ofproto->up.port_by_name is populated with
3156          * the appropriate ofport.  For this reason, we must get the name and
3157          * type from the netdev layer directly. */
3158         if (type) {
3159             const struct ofport *ofport;
3160
3161             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3162             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3163             ofproto_port->name = xstrdup(devname);
3164             ofproto_port->type = xstrdup(type);
3165             return 0;
3166         }
3167         return ENODEV;
3168     }
3169
3170     if (!sset_contains(&ofproto->ports, devname)) {
3171         return ENODEV;
3172     }
3173     error = dpif_port_query_by_name(ofproto->backer->dpif,
3174                                     devname, &dpif_port);
3175     if (!error) {
3176         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3177     }
3178     return error;
3179 }
3180
3181 static int
3182 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3183 {
3184     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3185     const char *devname = netdev_get_name(netdev);
3186     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3187     const char *dp_port_name;
3188
3189     if (netdev_vport_is_patch(netdev)) {
3190         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3191         return 0;
3192     }
3193
3194     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3195     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3196         odp_port_t port_no = ODPP_NONE;
3197         int error;
3198
3199         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3200         if (error) {
3201             return error;
3202         }
3203         if (netdev_get_tunnel_config(netdev)) {
3204             simap_put(&ofproto->backer->tnl_backers,
3205                       dp_port_name, odp_to_u32(port_no));
3206         }
3207     }
3208
3209     if (netdev_get_tunnel_config(netdev)) {
3210         sset_add(&ofproto->ghost_ports, devname);
3211     } else {
3212         sset_add(&ofproto->ports, devname);
3213     }
3214     return 0;
3215 }
3216
3217 static int
3218 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3219 {
3220     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3221     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3222     int error = 0;
3223
3224     if (!ofport) {
3225         return 0;
3226     }
3227
3228     sset_find_and_delete(&ofproto->ghost_ports,
3229                          netdev_get_name(ofport->up.netdev));
3230     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3231     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3232         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3233         if (!error) {
3234             /* The caller is going to close ofport->up.netdev.  If this is a
3235              * bonded port, then the bond is using that netdev, so remove it
3236              * from the bond.  The client will need to reconfigure everything
3237              * after deleting ports, so then the slave will get re-added. */
3238             bundle_remove(&ofport->up);
3239         }
3240     }
3241     return error;
3242 }
3243
3244 static int
3245 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3246 {
3247     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3248     int error;
3249
3250     error = netdev_get_stats(ofport->up.netdev, stats);
3251
3252     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3253         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3254
3255         ovs_mutex_lock(&ofproto->stats_mutex);
3256         /* ofproto->stats.tx_packets represents packets that we created
3257          * internally and sent to some port (e.g. packets sent with
3258          * ofproto_dpif_send_packet()).  Account for them as if they had
3259          * come from OFPP_LOCAL and got forwarded. */
3260
3261         if (stats->rx_packets != UINT64_MAX) {
3262             stats->rx_packets += ofproto->stats.tx_packets;
3263         }
3264
3265         if (stats->rx_bytes != UINT64_MAX) {
3266             stats->rx_bytes += ofproto->stats.tx_bytes;
3267         }
3268
3269         /* ofproto->stats.rx_packets represents packets that were received on
3270          * some port and we processed internally and dropped (e.g. STP).
3271          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3272
3273         if (stats->tx_packets != UINT64_MAX) {
3274             stats->tx_packets += ofproto->stats.rx_packets;
3275         }
3276
3277         if (stats->tx_bytes != UINT64_MAX) {
3278             stats->tx_bytes += ofproto->stats.rx_bytes;
3279         }
3280         ovs_mutex_unlock(&ofproto->stats_mutex);
3281     }
3282
3283     return error;
3284 }
3285
3286 struct port_dump_state {
3287     uint32_t bucket;
3288     uint32_t offset;
3289     bool ghost;
3290
3291     struct ofproto_port port;
3292     bool has_port;
3293 };
3294
3295 static int
3296 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3297 {
3298     *statep = xzalloc(sizeof(struct port_dump_state));
3299     return 0;
3300 }
3301
3302 static int
3303 port_dump_next(const struct ofproto *ofproto_, void *state_,
3304                struct ofproto_port *port)
3305 {
3306     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3307     struct port_dump_state *state = state_;
3308     const struct sset *sset;
3309     struct sset_node *node;
3310
3311     if (state->has_port) {
3312         ofproto_port_destroy(&state->port);
3313         state->has_port = false;
3314     }
3315     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3316     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3317         int error;
3318
3319         error = port_query_by_name(ofproto_, node->name, &state->port);
3320         if (!error) {
3321             *port = state->port;
3322             state->has_port = true;
3323             return 0;
3324         } else if (error != ENODEV) {
3325             return error;
3326         }
3327     }
3328
3329     if (!state->ghost) {
3330         state->ghost = true;
3331         state->bucket = 0;
3332         state->offset = 0;
3333         return port_dump_next(ofproto_, state_, port);
3334     }
3335
3336     return EOF;
3337 }
3338
3339 static int
3340 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3341 {
3342     struct port_dump_state *state = state_;
3343
3344     if (state->has_port) {
3345         ofproto_port_destroy(&state->port);
3346     }
3347     free(state);
3348     return 0;
3349 }
3350
3351 static int
3352 port_poll(const struct ofproto *ofproto_, char **devnamep)
3353 {
3354     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3355
3356     if (ofproto->port_poll_errno) {
3357         int error = ofproto->port_poll_errno;
3358         ofproto->port_poll_errno = 0;
3359         return error;
3360     }
3361
3362     if (sset_is_empty(&ofproto->port_poll_set)) {
3363         return EAGAIN;
3364     }
3365
3366     *devnamep = sset_pop(&ofproto->port_poll_set);
3367     return 0;
3368 }
3369
3370 static void
3371 port_poll_wait(const struct ofproto *ofproto_)
3372 {
3373     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3374     dpif_port_poll_wait(ofproto->backer->dpif);
3375 }
3376
3377 static int
3378 port_is_lacp_current(const struct ofport *ofport_)
3379 {
3380     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3381     return (ofport->bundle && ofport->bundle->lacp
3382             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3383             : -1);
3384 }
3385 \f
3386 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3387  * then delete it entirely. */
3388 static void
3389 rule_expire(struct rule_dpif *rule)
3390     OVS_REQUIRES(ofproto_mutex)
3391 {
3392     uint16_t hard_timeout, idle_timeout;
3393     long long int now = time_msec();
3394     int reason = -1;
3395
3396     hard_timeout = rule->up.hard_timeout;
3397     idle_timeout = rule->up.idle_timeout;
3398
3399     /* Has 'rule' expired? */
3400     if (hard_timeout) {
3401         long long int modified;
3402
3403         ovs_mutex_lock(&rule->up.mutex);
3404         modified = rule->up.modified;
3405         ovs_mutex_unlock(&rule->up.mutex);
3406
3407         if (now > modified + hard_timeout * 1000) {
3408             reason = OFPRR_HARD_TIMEOUT;
3409         }
3410     }
3411
3412     if (reason < 0 && idle_timeout) {
3413         long long int used;
3414
3415         ovs_mutex_lock(&rule->stats_mutex);
3416         used = rule->stats.used;
3417         ovs_mutex_unlock(&rule->stats_mutex);
3418
3419         if (now > used + idle_timeout * 1000) {
3420             reason = OFPRR_IDLE_TIMEOUT;
3421         }
3422     }
3423
3424     if (reason >= 0) {
3425         COVERAGE_INC(ofproto_dpif_expired);
3426         ofproto_rule_expire(&rule->up, reason);
3427     }
3428 }
3429
3430 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3431  * 'flow' must reflect the data in 'packet'. */
3432 int
3433 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3434                              const struct flow *flow,
3435                              struct rule_dpif *rule,
3436                              const struct ofpact *ofpacts, size_t ofpacts_len,
3437                              struct ofpbuf *packet)
3438 {
3439     struct dpif_flow_stats stats;
3440     struct xlate_out xout;
3441     struct xlate_in xin;
3442     ofp_port_t in_port;
3443     struct dpif_execute execute;
3444     int error;
3445
3446     ovs_assert((rule != NULL) != (ofpacts != NULL));
3447
3448     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3449
3450     if (rule) {
3451         rule_dpif_credit_stats(rule, &stats);
3452     }
3453
3454     xlate_in_init(&xin, ofproto, flow, flow->in_port.ofp_port, rule,
3455                   stats.tcp_flags, packet);
3456     xin.ofpacts = ofpacts;
3457     xin.ofpacts_len = ofpacts_len;
3458     xin.resubmit_stats = &stats;
3459     xlate_actions(&xin, &xout);
3460
3461     execute.actions = ofpbuf_data(xout.odp_actions);
3462     execute.actions_len = ofpbuf_size(xout.odp_actions);
3463     execute.packet = packet;
3464     execute.md = pkt_metadata_from_flow(flow);
3465     execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3466
3467     /* Fix up in_port. */
3468     in_port = flow->in_port.ofp_port;
3469     if (in_port == OFPP_NONE) {
3470         in_port = OFPP_LOCAL;
3471     }
3472     execute.md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3473
3474     error = dpif_execute(ofproto->backer->dpif, &execute);
3475
3476     xlate_out_uninit(&xout);
3477
3478     return error;
3479 }
3480
3481 void
3482 rule_dpif_credit_stats(struct rule_dpif *rule,
3483                        const struct dpif_flow_stats *stats)
3484 {
3485     ovs_mutex_lock(&rule->stats_mutex);
3486     rule->stats.n_packets += stats->n_packets;
3487     rule->stats.n_bytes += stats->n_bytes;
3488     rule->stats.used = MAX(rule->stats.used, stats->used);
3489     ovs_mutex_unlock(&rule->stats_mutex);
3490 }
3491
3492 ovs_be64
3493 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3494     OVS_REQUIRES(rule->up.mutex)
3495 {
3496     return rule->up.flow_cookie;
3497 }
3498
3499 void
3500 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3501                      uint16_t hard_timeout)
3502 {
3503     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3504 }
3505
3506 /* Returns 'rule''s actions.  The caller owns a reference on the returned
3507  * actions and must eventually release it (with rule_actions_unref()) to avoid
3508  * a memory leak. */
3509 const struct rule_actions *
3510 rule_dpif_get_actions(const struct rule_dpif *rule)
3511 {
3512     return rule_get_actions(&rule->up);
3513 }
3514
3515 /* Sets 'rule''s recirculation id. */
3516 static void
3517 rule_dpif_set_recirc_id(struct rule_dpif *rule, uint32_t id)
3518     OVS_REQUIRES(rule->up.mutex)
3519 {
3520     ovs_assert(!rule->recirc_id);
3521     rule->recirc_id = id;
3522 }
3523
3524 /* Returns 'rule''s recirculation id. */
3525 uint32_t
3526 rule_dpif_get_recirc_id(struct rule_dpif *rule)
3527     OVS_REQUIRES(rule->up.mutex)
3528 {
3529     if (!rule->recirc_id) {
3530         struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3531
3532         rule_dpif_set_recirc_id(rule, ofproto_dpif_alloc_recirc_id(ofproto));
3533     }
3534     return rule->recirc_id;
3535 }
3536
3537 /* Sets 'rule''s recirculation id. */
3538 void
3539 rule_set_recirc_id(struct rule *rule_, uint32_t id)
3540 {
3541     struct rule_dpif *rule = rule_dpif_cast(rule_);
3542
3543     ovs_mutex_lock(&rule->up.mutex);
3544     rule_dpif_set_recirc_id(rule, id);
3545     ovs_mutex_unlock(&rule->up.mutex);
3546 }
3547
3548 /* Lookup 'flow' in table 0 of 'ofproto''s classifier.
3549  * If 'wc' is non-null, sets the fields that were relevant as part of
3550  * the lookup. Returns the table_id where a match or miss occurred.
3551  *
3552  * The return value will be zero unless there was a miss and
3553  * OFPTC11_TABLE_MISS_CONTINUE is in effect for the sequence of tables
3554  * where misses occur.
3555  *
3556  * The rule is returned in '*rule', which is valid at least until the next
3557  * RCU quiescent period.  If the '*rule' needs to stay around longer,
3558  * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3559  * on it before this returns. */
3560 uint8_t
3561 rule_dpif_lookup(struct ofproto_dpif *ofproto, struct flow *flow,
3562                  struct flow_wildcards *wc, struct rule_dpif **rule,
3563                  bool take_ref, const struct dpif_flow_stats *stats)
3564 {
3565     enum rule_dpif_lookup_verdict verdict;
3566     enum ofputil_port_config config = 0;
3567     uint8_t table_id;
3568
3569     if (ofproto_dpif_get_enable_recirc(ofproto)) {
3570         /* Always exactly match recirc_id since datapath supports
3571          * recirculation.  */
3572         if (wc) {
3573             wc->masks.recirc_id = UINT32_MAX;
3574         }
3575
3576         /* Start looking up from internal table for post recirculation flows
3577          * or packets. We can also simply send all, including normal flows
3578          * or packets to the internal table. They will not match any post
3579          * recirculation rules except the 'catch all' rule that resubmit
3580          * them to table 0.
3581          *
3582          * As an optimization, we send normal flows and packets to table 0
3583          * directly, saving one table lookup.  */
3584         table_id = flow->recirc_id ? TBL_INTERNAL : 0;
3585     } else {
3586         table_id = 0;
3587     }
3588
3589     verdict = rule_dpif_lookup_from_table(ofproto, flow, wc, true,
3590                                           &table_id, rule, take_ref, stats);
3591
3592     switch (verdict) {
3593     case RULE_DPIF_LOOKUP_VERDICT_MATCH:
3594         return table_id;
3595     case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER: {
3596         struct ofport_dpif *port;
3597
3598         port = get_ofp_port(ofproto, flow->in_port.ofp_port);
3599         if (!port) {
3600             VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3601                          flow->in_port.ofp_port);
3602         }
3603         config = port ? port->up.pp.config : 0;
3604         break;
3605     }
3606     case RULE_DPIF_LOOKUP_VERDICT_DROP:
3607         config = OFPUTIL_PC_NO_PACKET_IN;
3608         break;
3609     case RULE_DPIF_LOOKUP_VERDICT_DEFAULT:
3610         if (!connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3611             config = OFPUTIL_PC_NO_PACKET_IN;
3612         }
3613         break;
3614     default:
3615         OVS_NOT_REACHED();
3616     }
3617
3618     choose_miss_rule(config, ofproto->miss_rule,
3619                      ofproto->no_packet_in_rule, rule, take_ref);
3620     return table_id;
3621 }
3622
3623 /* The returned rule is valid at least until the next RCU quiescent period.
3624  * If the '*rule' needs to stay around longer, a non-zero 'take_ref' must be
3625  * passed in to cause a reference to be taken on it before this returns. */
3626 static struct rule_dpif *
3627 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, uint8_t table_id,
3628                           const struct flow *flow, struct flow_wildcards *wc,
3629                           bool take_ref)
3630 {
3631     struct classifier *cls = &ofproto->up.tables[table_id].cls;
3632     const struct cls_rule *cls_rule;
3633     struct rule_dpif *rule;
3634     struct flow ofpc_normal_flow;
3635
3636     if (ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3637         /* We always unwildcard dl_type and nw_frag (for IP), so they
3638          * need not be unwildcarded here. */
3639
3640         if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
3641             if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3642                 /* We must pretend that transport ports are unavailable. */
3643                 ofpc_normal_flow = *flow;
3644                 ofpc_normal_flow.tp_src = htons(0);
3645                 ofpc_normal_flow.tp_dst = htons(0);
3646                 flow = &ofpc_normal_flow;
3647             } else {
3648                 /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM).
3649                  * Use the drop_frags_rule (which cannot disappear). */
3650                 cls_rule = &ofproto->drop_frags_rule->up.cr;
3651                 rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3652                 if (take_ref) {
3653                     rule_dpif_ref(rule);
3654                 }
3655                 return rule;
3656             }
3657         }
3658     }
3659
3660     do {
3661         cls_rule = classifier_lookup(cls, flow, wc);
3662
3663         rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3664
3665         /* Try again if the rule was released before we get the reference. */
3666     } while (rule && take_ref && !rule_dpif_try_ref(rule));
3667
3668     return rule;
3669 }
3670
3671 /* Look up 'flow' in 'ofproto''s classifier starting from table '*table_id'.
3672  * Stores the rule that was found in '*rule', or NULL if none was found.
3673  * Updates 'wc', if nonnull, to reflect the fields that were used during the
3674  * lookup.
3675  *
3676  * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3677  * if none is found then the table miss configuration for that table is
3678  * honored, which can result in additional lookups in other OpenFlow tables.
3679  * In this case the function updates '*table_id' to reflect the final OpenFlow
3680  * table that was searched.
3681  *
3682  * If 'honor_table_miss' is false, then only one table lookup occurs, in
3683  * '*table_id'.
3684  *
3685  * Returns:
3686  *
3687  *    - RULE_DPIF_LOOKUP_VERDICT_MATCH if a rule (in '*rule') was found.
3688  *
3689  *    - RULE_OFPTC_TABLE_MISS_CONTROLLER if no rule was found and either:
3690  *      + 'honor_table_miss' is false
3691  *      + a table miss configuration specified that the packet should be
3692  *        sent to the controller in this case.
3693  *
3694  *    - RULE_DPIF_LOOKUP_VERDICT_DROP if no rule was found, 'honor_table_miss'
3695  *      is true and a table miss configuration specified that the packet
3696  *      should be dropped in this case.
3697  *
3698  *    - RULE_DPIF_LOOKUP_VERDICT_DEFAULT if no rule was found,
3699  *      'honor_table_miss' is true and a table miss configuration has
3700  *      not been specified in this case.
3701  *
3702  * The rule is returned in '*rule', which is valid at least until the next
3703  * RCU quiescent period.  If the '*rule' needs to stay around longer,
3704  * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3705  * on it before this returns. */
3706 enum rule_dpif_lookup_verdict
3707 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3708                             const struct flow *flow,
3709                             struct flow_wildcards *wc,
3710                             bool honor_table_miss,
3711                             uint8_t *table_id, struct rule_dpif **rule,
3712                             bool take_ref, const struct dpif_flow_stats *stats)
3713 {
3714     uint8_t next_id;
3715
3716     for (next_id = *table_id;
3717          next_id < ofproto->up.n_tables;
3718          next_id++, next_id += (next_id == TBL_INTERNAL))
3719     {
3720         *table_id = next_id;
3721         *rule = rule_dpif_lookup_in_table(ofproto, *table_id, flow, wc,
3722                                           take_ref);
3723         if (stats) {
3724             struct oftable *tbl = &ofproto->up.tables[next_id];
3725             unsigned long orig;
3726
3727             atomic_add_relaxed(*rule ? &tbl->n_matched : &tbl->n_missed,
3728                                stats->n_packets, &orig);
3729         }
3730         if (*rule) {
3731             return RULE_DPIF_LOOKUP_VERDICT_MATCH;
3732         } else if (!honor_table_miss) {
3733             return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3734         } else {
3735             switch (ofproto_table_get_miss_config(&ofproto->up, *table_id)) {
3736             case OFPUTIL_TABLE_MISS_CONTINUE:
3737                 break;
3738
3739             case OFPUTIL_TABLE_MISS_CONTROLLER:
3740                 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3741
3742             case OFPUTIL_TABLE_MISS_DROP:
3743                 return RULE_DPIF_LOOKUP_VERDICT_DROP;
3744
3745             case OFPUTIL_TABLE_MISS_DEFAULT:
3746                 return RULE_DPIF_LOOKUP_VERDICT_DEFAULT;
3747             }
3748         }
3749     }
3750
3751     return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3752 }
3753
3754 /* Given a port configuration (specified as zero if there's no port), chooses
3755  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
3756  * flow table miss.
3757  *
3758  * The rule is returned in '*rule', which is valid at least until the next
3759  * RCU quiescent period.  If the '*rule' needs to stay around longer,
3760  * a reference must be taken on it (rule_dpif_ref()).
3761  */
3762 void
3763 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
3764                  struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule,
3765                  bool take_ref)
3766 {
3767     *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
3768     if (take_ref) {
3769         rule_dpif_ref(*rule);
3770     }
3771 }
3772
3773 static void
3774 complete_operation(struct rule_dpif *rule)
3775     OVS_REQUIRES(ofproto_mutex)
3776 {
3777     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3778
3779     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3780 }
3781
3782 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3783 {
3784     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3785 }
3786
3787 static struct rule *
3788 rule_alloc(void)
3789 {
3790     struct rule_dpif *rule = xmalloc(sizeof *rule);
3791     return &rule->up;
3792 }
3793
3794 static void
3795 rule_dealloc(struct rule *rule_)
3796 {
3797     struct rule_dpif *rule = rule_dpif_cast(rule_);
3798     free(rule);
3799 }
3800
3801 static enum ofperr
3802 rule_construct(struct rule *rule_)
3803     OVS_NO_THREAD_SAFETY_ANALYSIS
3804 {
3805     struct rule_dpif *rule = rule_dpif_cast(rule_);
3806     ovs_mutex_init_adaptive(&rule->stats_mutex);
3807     rule->stats.n_packets = 0;
3808     rule->stats.n_bytes = 0;
3809     rule->stats.used = rule->up.modified;
3810     rule->recirc_id = 0;
3811
3812     return 0;
3813 }
3814
3815 static enum ofperr
3816 rule_insert(struct rule *rule_)
3817     OVS_REQUIRES(ofproto_mutex)
3818 {
3819     struct rule_dpif *rule = rule_dpif_cast(rule_);
3820     complete_operation(rule);
3821     return 0;
3822 }
3823
3824 static void
3825 rule_delete(struct rule *rule_)
3826     OVS_REQUIRES(ofproto_mutex)
3827 {
3828     struct rule_dpif *rule = rule_dpif_cast(rule_);
3829     complete_operation(rule);
3830 }
3831
3832 static void
3833 rule_destruct(struct rule *rule_)
3834 {
3835     struct rule_dpif *rule = rule_dpif_cast(rule_);
3836
3837     ovs_mutex_destroy(&rule->stats_mutex);
3838     if (rule->recirc_id) {
3839         struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3840
3841         ofproto_dpif_free_recirc_id(ofproto, rule->recirc_id);
3842     }
3843 }
3844
3845 static void
3846 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
3847                long long int *used)
3848 {
3849     struct rule_dpif *rule = rule_dpif_cast(rule_);
3850
3851     ovs_mutex_lock(&rule->stats_mutex);
3852     *packets = rule->stats.n_packets;
3853     *bytes = rule->stats.n_bytes;
3854     *used = rule->stats.used;
3855     ovs_mutex_unlock(&rule->stats_mutex);
3856 }
3857
3858 static void
3859 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
3860                   struct ofpbuf *packet)
3861 {
3862     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3863
3864     ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
3865 }
3866
3867 static enum ofperr
3868 rule_execute(struct rule *rule, const struct flow *flow,
3869              struct ofpbuf *packet)
3870 {
3871     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
3872     ofpbuf_delete(packet);
3873     return 0;
3874 }
3875
3876 static void
3877 rule_modify_actions(struct rule *rule_, bool reset_counters)
3878     OVS_REQUIRES(ofproto_mutex)
3879 {
3880     struct rule_dpif *rule = rule_dpif_cast(rule_);
3881
3882     if (reset_counters) {
3883         ovs_mutex_lock(&rule->stats_mutex);
3884         rule->stats.n_packets = 0;
3885         rule->stats.n_bytes = 0;
3886         ovs_mutex_unlock(&rule->stats_mutex);
3887     }
3888
3889     complete_operation(rule);
3890 }
3891
3892 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
3893 {
3894     return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
3895 }
3896
3897 static struct ofgroup *
3898 group_alloc(void)
3899 {
3900     struct group_dpif *group = xzalloc(sizeof *group);
3901     return &group->up;
3902 }
3903
3904 static void
3905 group_dealloc(struct ofgroup *group_)
3906 {
3907     struct group_dpif *group = group_dpif_cast(group_);
3908     free(group);
3909 }
3910
3911 static void
3912 group_construct_stats(struct group_dpif *group)
3913     OVS_REQUIRES(group->stats_mutex)
3914 {
3915     struct ofputil_bucket *bucket;
3916     const struct list *buckets;
3917
3918     group->packet_count = 0;
3919     group->byte_count = 0;
3920
3921     group_dpif_get_buckets(group, &buckets);
3922     LIST_FOR_EACH (bucket, list_node, buckets) {
3923         bucket->stats.packet_count = 0;
3924         bucket->stats.byte_count = 0;
3925     }
3926 }
3927
3928 void
3929 group_dpif_credit_stats(struct group_dpif *group,
3930                         struct ofputil_bucket *bucket,
3931                         const struct dpif_flow_stats *stats)
3932 {
3933     ovs_mutex_lock(&group->stats_mutex);
3934     group->packet_count += stats->n_packets;
3935     group->byte_count += stats->n_bytes;
3936     if (bucket) {
3937         bucket->stats.packet_count += stats->n_packets;
3938         bucket->stats.byte_count += stats->n_bytes;
3939     } else { /* Credit to all buckets */
3940         const struct list *buckets;
3941
3942         group_dpif_get_buckets(group, &buckets);
3943         LIST_FOR_EACH (bucket, list_node, buckets) {
3944             bucket->stats.packet_count += stats->n_packets;
3945             bucket->stats.byte_count += stats->n_bytes;
3946         }
3947     }
3948     ovs_mutex_unlock(&group->stats_mutex);
3949 }
3950
3951 static enum ofperr
3952 group_construct(struct ofgroup *group_)
3953 {
3954     struct group_dpif *group = group_dpif_cast(group_);
3955     const struct ofputil_bucket *bucket;
3956
3957     /* Prevent group chaining because our locking structure makes it hard to
3958      * implement deadlock-free.  (See xlate_group_resource_check().) */
3959     LIST_FOR_EACH (bucket, list_node, &group->up.buckets) {
3960         const struct ofpact *a;
3961
3962         OFPACT_FOR_EACH (a, bucket->ofpacts, bucket->ofpacts_len) {
3963             if (a->type == OFPACT_GROUP) {
3964                 return OFPERR_OFPGMFC_CHAINING_UNSUPPORTED;
3965             }
3966         }
3967     }
3968
3969     ovs_mutex_init_adaptive(&group->stats_mutex);
3970     ovs_mutex_lock(&group->stats_mutex);
3971     group_construct_stats(group);
3972     ovs_mutex_unlock(&group->stats_mutex);
3973     return 0;
3974 }
3975
3976 static void
3977 group_destruct(struct ofgroup *group_)
3978 {
3979     struct group_dpif *group = group_dpif_cast(group_);
3980     ovs_mutex_destroy(&group->stats_mutex);
3981 }
3982
3983 static enum ofperr
3984 group_modify(struct ofgroup *group_)
3985 {
3986     struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
3987
3988     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3989
3990     return 0;
3991 }
3992
3993 static enum ofperr
3994 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
3995 {
3996     struct group_dpif *group = group_dpif_cast(group_);
3997     struct ofputil_bucket *bucket;
3998     const struct list *buckets;
3999     struct bucket_counter *bucket_stats;
4000
4001     ovs_mutex_lock(&group->stats_mutex);
4002     ogs->packet_count = group->packet_count;
4003     ogs->byte_count = group->byte_count;
4004
4005     group_dpif_get_buckets(group, &buckets);
4006     bucket_stats = ogs->bucket_stats;
4007     LIST_FOR_EACH (bucket, list_node, buckets) {
4008         bucket_stats->packet_count = bucket->stats.packet_count;
4009         bucket_stats->byte_count = bucket->stats.byte_count;
4010         bucket_stats++;
4011     }
4012     ovs_mutex_unlock(&group->stats_mutex);
4013
4014     return 0;
4015 }
4016
4017 /* If the group exists, this function increments the groups's reference count.
4018  *
4019  * Make sure to call group_dpif_unref() after no longer needing to maintain
4020  * a reference to the group. */
4021 bool
4022 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4023                   struct group_dpif **group)
4024 {
4025     struct ofgroup *ofgroup;
4026     bool found;
4027
4028     found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4029     *group = found ?  group_dpif_cast(ofgroup) : NULL;
4030
4031     return found;
4032 }
4033
4034 void
4035 group_dpif_get_buckets(const struct group_dpif *group,
4036                        const struct list **buckets)
4037 {
4038     *buckets = &group->up.buckets;
4039 }
4040
4041 enum ofp11_group_type
4042 group_dpif_get_type(const struct group_dpif *group)
4043 {
4044     return group->up.type;
4045 }
4046 \f
4047 /* Sends 'packet' out 'ofport'.
4048  * May modify 'packet'.
4049  * Returns 0 if successful, otherwise a positive errno value. */
4050 int
4051 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4052 {
4053     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4054     int error;
4055
4056     error = xlate_send_packet(ofport, packet);
4057
4058     ovs_mutex_lock(&ofproto->stats_mutex);
4059     ofproto->stats.tx_packets++;
4060     ofproto->stats.tx_bytes += ofpbuf_size(packet);
4061     ovs_mutex_unlock(&ofproto->stats_mutex);
4062     return error;
4063 }
4064 \f
4065 static bool
4066 set_frag_handling(struct ofproto *ofproto_,
4067                   enum ofp_config_flags frag_handling)
4068 {
4069     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4070     if (frag_handling != OFPC_FRAG_REASM) {
4071         ofproto->backer->need_revalidate = REV_RECONFIGURE;
4072         return true;
4073     } else {
4074         return false;
4075     }
4076 }
4077
4078 static enum ofperr
4079 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
4080            const struct flow *flow,
4081            const struct ofpact *ofpacts, size_t ofpacts_len)
4082 {
4083     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4084
4085     ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
4086                                  ofpacts_len, packet);
4087     return 0;
4088 }
4089 \f
4090 /* NetFlow. */
4091
4092 static int
4093 set_netflow(struct ofproto *ofproto_,
4094             const struct netflow_options *netflow_options)
4095 {
4096     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4097
4098     if (netflow_options) {
4099         if (!ofproto->netflow) {
4100             ofproto->netflow = netflow_create();
4101             ofproto->backer->need_revalidate = REV_RECONFIGURE;
4102         }
4103         return netflow_set_options(ofproto->netflow, netflow_options);
4104     } else if (ofproto->netflow) {
4105         ofproto->backer->need_revalidate = REV_RECONFIGURE;
4106         netflow_unref(ofproto->netflow);
4107         ofproto->netflow = NULL;
4108     }
4109
4110     return 0;
4111 }
4112
4113 static void
4114 get_netflow_ids(const struct ofproto *ofproto_,
4115                 uint8_t *engine_type, uint8_t *engine_id)
4116 {
4117     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4118
4119     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
4120 }
4121 \f
4122 static struct ofproto_dpif *
4123 ofproto_dpif_lookup(const char *name)
4124 {
4125     struct ofproto_dpif *ofproto;
4126
4127     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
4128                              hash_string(name, 0), &all_ofproto_dpifs) {
4129         if (!strcmp(ofproto->up.name, name)) {
4130             return ofproto;
4131         }
4132     }
4133     return NULL;
4134 }
4135
4136 static void
4137 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
4138                           const char *argv[], void *aux OVS_UNUSED)
4139 {
4140     struct ofproto_dpif *ofproto;
4141
4142     if (argc > 1) {
4143         ofproto = ofproto_dpif_lookup(argv[1]);
4144         if (!ofproto) {
4145             unixctl_command_reply_error(conn, "no such bridge");
4146             return;
4147         }
4148         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4149         mac_learning_flush(ofproto->ml);
4150         ovs_rwlock_unlock(&ofproto->ml->rwlock);
4151     } else {
4152         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4153             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4154             mac_learning_flush(ofproto->ml);
4155             ovs_rwlock_unlock(&ofproto->ml->rwlock);
4156         }
4157     }
4158
4159     unixctl_command_reply(conn, "table successfully flushed");
4160 }
4161
4162 static void
4163 ofproto_unixctl_mcast_snooping_flush(struct unixctl_conn *conn, int argc,
4164                                      const char *argv[], void *aux OVS_UNUSED)
4165 {
4166     struct ofproto_dpif *ofproto;
4167
4168     if (argc > 1) {
4169         ofproto = ofproto_dpif_lookup(argv[1]);
4170         if (!ofproto) {
4171             unixctl_command_reply_error(conn, "no such bridge");
4172             return;
4173         }
4174
4175         if (!mcast_snooping_enabled(ofproto->ms)) {
4176             unixctl_command_reply_error(conn, "multicast snooping is disabled");
4177             return;
4178         }
4179         mcast_snooping_mdb_flush(ofproto->ms);
4180     } else {
4181         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4182             if (!mcast_snooping_enabled(ofproto->ms)) {
4183                 continue;
4184             }
4185             mcast_snooping_mdb_flush(ofproto->ms);
4186         }
4187     }
4188
4189     unixctl_command_reply(conn, "table successfully flushed");
4190 }
4191
4192 static struct ofport_dpif *
4193 ofbundle_get_a_port(const struct ofbundle *bundle)
4194 {
4195     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
4196                         bundle_node);
4197 }
4198
4199 static void
4200 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4201                          const char *argv[], void *aux OVS_UNUSED)
4202 {
4203     struct ds ds = DS_EMPTY_INITIALIZER;
4204     const struct ofproto_dpif *ofproto;
4205     const struct mac_entry *e;
4206
4207     ofproto = ofproto_dpif_lookup(argv[1]);
4208     if (!ofproto) {
4209         unixctl_command_reply_error(conn, "no such bridge");
4210         return;
4211     }
4212
4213     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
4214     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
4215     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
4216         struct ofbundle *bundle = e->port.p;
4217         char name[OFP_MAX_PORT_NAME_LEN];
4218
4219         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4220                                name, sizeof name);
4221         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
4222                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
4223                       mac_entry_age(ofproto->ml, e));
4224     }
4225     ovs_rwlock_unlock(&ofproto->ml->rwlock);
4226     unixctl_command_reply(conn, ds_cstr(&ds));
4227     ds_destroy(&ds);
4228 }
4229
4230 static void
4231 ofproto_unixctl_mcast_snooping_show(struct unixctl_conn *conn,
4232                                     int argc OVS_UNUSED,
4233                                     const char *argv[],
4234                                     void *aux OVS_UNUSED)
4235 {
4236     struct ds ds = DS_EMPTY_INITIALIZER;
4237     const struct ofproto_dpif *ofproto;
4238     const struct ofbundle *bundle;
4239     const struct mcast_group *grp;
4240     struct mcast_group_bundle *b;
4241     struct mcast_mrouter_bundle *mrouter;
4242
4243     ofproto = ofproto_dpif_lookup(argv[1]);
4244     if (!ofproto) {
4245         unixctl_command_reply_error(conn, "no such bridge");
4246         return;
4247     }
4248
4249     if (!mcast_snooping_enabled(ofproto->ms)) {
4250         unixctl_command_reply_error(conn, "multicast snooping is disabled");
4251         return;
4252     }
4253
4254     ds_put_cstr(&ds, " port  VLAN  GROUP                Age\n");
4255     ovs_rwlock_rdlock(&ofproto->ms->rwlock);
4256     LIST_FOR_EACH (grp, group_node, &ofproto->ms->group_lru) {
4257         LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
4258             char name[OFP_MAX_PORT_NAME_LEN];
4259
4260             bundle = b->port;
4261             ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4262                                    name, sizeof name);
4263             ds_put_format(&ds, "%5s  %4d  "IP_FMT"         %3d\n",
4264                           name, grp->vlan, IP_ARGS(grp->ip4),
4265                           mcast_bundle_age(ofproto->ms, b));
4266         }
4267     }
4268
4269     /* ports connected to multicast routers */
4270     LIST_FOR_EACH(mrouter, mrouter_node, &ofproto->ms->mrouter_lru) {
4271         char name[OFP_MAX_PORT_NAME_LEN];
4272
4273         bundle = mrouter->port;
4274         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4275                                name, sizeof name);
4276             ds_put_format(&ds, "%5s  %4d  querier             %3d\n",
4277                       name, mrouter->vlan,
4278                       mcast_mrouter_age(ofproto->ms, mrouter));
4279     }
4280     ovs_rwlock_unlock(&ofproto->ms->rwlock);
4281     unixctl_command_reply(conn, ds_cstr(&ds));
4282     ds_destroy(&ds);
4283 }
4284
4285 struct trace_ctx {
4286     struct xlate_out xout;
4287     struct xlate_in xin;
4288     const struct flow *key;
4289     struct flow flow;
4290     struct flow_wildcards wc;
4291     struct ds *result;
4292 };
4293
4294 static void
4295 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
4296 {
4297     const struct rule_actions *actions;
4298     ovs_be64 cookie;
4299
4300     ds_put_char_multiple(result, '\t', level);
4301     if (!rule) {
4302         ds_put_cstr(result, "No match\n");
4303         return;
4304     }
4305
4306     ovs_mutex_lock(&rule->up.mutex);
4307     cookie = rule->up.flow_cookie;
4308     ovs_mutex_unlock(&rule->up.mutex);
4309
4310     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
4311                   rule ? rule->up.table_id : 0, ntohll(cookie));
4312     cls_rule_format(&rule->up.cr, result);
4313     ds_put_char(result, '\n');
4314
4315     actions = rule_dpif_get_actions(rule);
4316
4317     ds_put_char_multiple(result, '\t', level);
4318     ds_put_cstr(result, "OpenFlow actions=");
4319     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
4320     ds_put_char(result, '\n');
4321 }
4322
4323 static void
4324 trace_format_flow(struct ds *result, int level, const char *title,
4325                   struct trace_ctx *trace)
4326 {
4327     ds_put_char_multiple(result, '\t', level);
4328     ds_put_format(result, "%s: ", title);
4329     /* Do not report unchanged flows for resubmits. */
4330     if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
4331         || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
4332         ds_put_cstr(result, "unchanged");
4333     } else {
4334         flow_format(result, &trace->xin.flow);
4335         trace->flow = trace->xin.flow;
4336     }
4337     ds_put_char(result, '\n');
4338 }
4339
4340 static void
4341 trace_format_regs(struct ds *result, int level, const char *title,
4342                   struct trace_ctx *trace)
4343 {
4344     size_t i;
4345
4346     ds_put_char_multiple(result, '\t', level);
4347     ds_put_format(result, "%s:", title);
4348     for (i = 0; i < FLOW_N_REGS; i++) {
4349         ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
4350     }
4351     ds_put_char(result, '\n');
4352 }
4353
4354 static void
4355 trace_format_odp(struct ds *result, int level, const char *title,
4356                  struct trace_ctx *trace)
4357 {
4358     struct ofpbuf *odp_actions = trace->xout.odp_actions;
4359
4360     ds_put_char_multiple(result, '\t', level);
4361     ds_put_format(result, "%s: ", title);
4362     format_odp_actions(result, ofpbuf_data(odp_actions),
4363                                ofpbuf_size(odp_actions));
4364     ds_put_char(result, '\n');
4365 }
4366
4367 static void
4368 trace_format_megaflow(struct ds *result, int level, const char *title,
4369                       struct trace_ctx *trace)
4370 {
4371     struct match match;
4372
4373     ds_put_char_multiple(result, '\t', level);
4374     ds_put_format(result, "%s: ", title);
4375     flow_wildcards_or(&trace->wc, &trace->xout.wc, &trace->wc);
4376     match_init(&match, trace->key, &trace->wc);
4377     match_format(&match, result, OFP_DEFAULT_PRIORITY);
4378     ds_put_char(result, '\n');
4379 }
4380
4381 static void
4382 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
4383 {
4384     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4385     struct ds *result = trace->result;
4386
4387     ds_put_char(result, '\n');
4388     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
4389     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
4390     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
4391     trace_format_megaflow(result, recurse + 1, "Resubmitted megaflow", trace);
4392     trace_format_rule(result, recurse + 1, rule);
4393 }
4394
4395 static void
4396 trace_report(struct xlate_in *xin, const char *s, int recurse)
4397 {
4398     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4399     struct ds *result = trace->result;
4400
4401     ds_put_char_multiple(result, '\t', recurse);
4402     ds_put_cstr(result, s);
4403     ds_put_char(result, '\n');
4404 }
4405
4406 /* Parses the 'argc' elements of 'argv', ignoring argv[0].  The following
4407  * forms are supported:
4408  *
4409  *     - [dpname] odp_flow [-generate | packet]
4410  *     - bridge br_flow [-generate | packet]
4411  *
4412  * On success, initializes '*ofprotop' and 'flow' and returns NULL.  On failure
4413  * returns a nonnull malloced error message. */
4414 static char * WARN_UNUSED_RESULT
4415 parse_flow_and_packet(int argc, const char *argv[],
4416                       struct ofproto_dpif **ofprotop, struct flow *flow,
4417                       struct ofpbuf **packetp)
4418 {
4419     const struct dpif_backer *backer = NULL;
4420     const char *error = NULL;
4421     char *m_err = NULL;
4422     struct simap port_names = SIMAP_INITIALIZER(&port_names);
4423     struct ofpbuf *packet;
4424     struct ofpbuf odp_key;
4425     struct ofpbuf odp_mask;
4426
4427     ofpbuf_init(&odp_key, 0);
4428     ofpbuf_init(&odp_mask, 0);
4429
4430     /* Handle "-generate" or a hex string as the last argument. */
4431     if (!strcmp(argv[argc - 1], "-generate")) {
4432         packet = ofpbuf_new(0);
4433         argc--;
4434     } else {
4435         error = eth_from_hex(argv[argc - 1], &packet);
4436         if (!error) {
4437             argc--;
4438         } else if (argc == 4) {
4439             /* The 3-argument form must end in "-generate' or a hex string. */
4440             goto exit;
4441         }
4442         error = NULL;
4443     }
4444
4445     /* odp_flow can have its in_port specified as a name instead of port no.
4446      * We do not yet know whether a given flow is a odp_flow or a br_flow.
4447      * But, to know whether a flow is odp_flow through odp_flow_from_string(),
4448      * we need to create a simap of name to port no. */
4449     if (argc == 3) {
4450         const char *dp_type;
4451         if (!strncmp(argv[1], "ovs-", 4)) {
4452             dp_type = argv[1] + 4;
4453         } else {
4454             dp_type = argv[1];
4455         }
4456         backer = shash_find_data(&all_dpif_backers, dp_type);
4457     } else if (argc == 2) {
4458         struct shash_node *node;
4459         if (shash_count(&all_dpif_backers) == 1) {
4460             node = shash_first(&all_dpif_backers);
4461             backer = node->data;
4462         }
4463     } else {
4464         error = "Syntax error";
4465         goto exit;
4466     }
4467     if (backer && backer->dpif) {
4468         struct dpif_port dpif_port;
4469         struct dpif_port_dump port_dump;
4470         DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4471             simap_put(&port_names, dpif_port.name,
4472                       odp_to_u32(dpif_port.port_no));
4473         }
4474     }
4475
4476     /* Parse the flow and determine whether a datapath or
4477      * bridge is specified. If function odp_flow_key_from_string()
4478      * returns 0, the flow is a odp_flow. If function
4479      * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4480     if (!odp_flow_from_string(argv[argc - 1], &port_names,
4481                               &odp_key, &odp_mask)) {
4482         if (!backer) {
4483             error = "Cannot find the datapath";
4484             goto exit;
4485         }
4486
4487         if (odp_flow_key_to_flow(ofpbuf_data(&odp_key), ofpbuf_size(&odp_key),
4488                                  flow) == ODP_FIT_ERROR) {
4489             error = "Failed to parse flow key";
4490             goto exit;
4491         }
4492
4493         *ofprotop = xlate_lookup_ofproto(backer, flow,
4494                                          &flow->in_port.ofp_port);
4495         if (*ofprotop == NULL) {
4496             error = "Invalid datapath flow";
4497             goto exit;
4498         }
4499
4500         vsp_adjust_flow(*ofprotop, flow, NULL);
4501
4502     } else {
4503         char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4504
4505         if (err) {
4506             m_err = xasprintf("Bad flow syntax: %s", err);
4507             free(err);
4508             goto exit;
4509         } else {
4510             if (argc != 3) {
4511                 error = "Must specify bridge name";
4512                 goto exit;
4513             }
4514
4515             *ofprotop = ofproto_dpif_lookup(argv[1]);
4516             if (!*ofprotop) {
4517                 error = "Unknown bridge name";
4518                 goto exit;
4519             }
4520         }
4521     }
4522
4523     /* Generate a packet, if requested. */
4524     if (packet) {
4525         if (!ofpbuf_size(packet)) {
4526             flow_compose(packet, flow);
4527         } else {
4528             struct pkt_metadata md = pkt_metadata_from_flow(flow);
4529
4530             /* Use the metadata from the flow and the packet argument
4531              * to reconstruct the flow. */
4532             flow_extract(packet, &md, flow);
4533         }
4534     }
4535
4536 exit:
4537     if (error && !m_err) {
4538         m_err = xstrdup(error);
4539     }
4540     if (m_err) {
4541         ofpbuf_delete(packet);
4542         packet = NULL;
4543     }
4544     *packetp = packet;
4545     ofpbuf_uninit(&odp_key);
4546     ofpbuf_uninit(&odp_mask);
4547     simap_destroy(&port_names);
4548     return m_err;
4549 }
4550
4551 static void
4552 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4553                       void *aux OVS_UNUSED)
4554 {
4555     struct ofproto_dpif *ofproto;
4556     struct ofpbuf *packet;
4557     char *error;
4558     struct flow flow;
4559
4560     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4561     if (!error) {
4562         struct ds result;
4563
4564         ds_init(&result);
4565         ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4566         unixctl_command_reply(conn, ds_cstr(&result));
4567         ds_destroy(&result);
4568         ofpbuf_delete(packet);
4569     } else {
4570         unixctl_command_reply_error(conn, error);
4571         free(error);
4572     }
4573 }
4574
4575 static void
4576 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4577                               const char *argv[], void *aux OVS_UNUSED)
4578 {
4579     enum ofputil_protocol usable_protocols;
4580     struct ofproto_dpif *ofproto;
4581     bool enforce_consistency;
4582     struct ofpbuf ofpacts;
4583     struct ofpbuf *packet;
4584     struct ds result;
4585     struct flow flow;
4586     uint16_t in_port;
4587
4588     /* Three kinds of error return values! */
4589     enum ofperr retval;
4590     char *error;
4591
4592     packet = NULL;
4593     ds_init(&result);
4594     ofpbuf_init(&ofpacts, 0);
4595
4596     /* Parse actions. */
4597     error = ofpacts_parse_actions(argv[--argc], &ofpacts, &usable_protocols);
4598     if (error) {
4599         unixctl_command_reply_error(conn, error);
4600         free(error);
4601         goto exit;
4602     }
4603
4604     /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4605      * consistency between the flow and the actions.  With -consistent, we
4606      * enforce consistency even for a flow supported in OpenFlow 1.0. */
4607     if (!strcmp(argv[1], "-consistent")) {
4608         enforce_consistency = true;
4609         argv++;
4610         argc--;
4611     } else {
4612         enforce_consistency = false;
4613     }
4614
4615     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4616     if (error) {
4617         unixctl_command_reply_error(conn, error);
4618         free(error);
4619         goto exit;
4620     }
4621
4622     /* Do the same checks as handle_packet_out() in ofproto.c.
4623      *
4624      * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
4625      * strictly correct because these actions aren't in any table, but it's OK
4626      * because it 'table_id' is used only to check goto_table instructions, but
4627      * packet-outs take a list of actions and therefore it can't include
4628      * instructions.
4629      *
4630      * We skip the "meter" check here because meter is an instruction, not an
4631      * action, and thus cannot appear in ofpacts. */
4632     in_port = ofp_to_u16(flow.in_port.ofp_port);
4633     if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
4634         unixctl_command_reply_error(conn, "invalid in_port");
4635         goto exit;
4636     }
4637     if (enforce_consistency) {
4638         retval = ofpacts_check_consistency(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts),
4639                                            &flow, u16_to_ofp(ofproto->up.max_ports),
4640                                            0, 0, usable_protocols);
4641     } else {
4642         retval = ofpacts_check(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &flow,
4643                                u16_to_ofp(ofproto->up.max_ports), 0, 0,
4644                                &usable_protocols);
4645     }
4646
4647     if (retval) {
4648         ds_clear(&result);
4649         ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
4650         unixctl_command_reply_error(conn, ds_cstr(&result));
4651         goto exit;
4652     }
4653
4654     ofproto_trace(ofproto, &flow, packet,
4655                   ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &result);
4656     unixctl_command_reply(conn, ds_cstr(&result));
4657
4658 exit:
4659     ds_destroy(&result);
4660     ofpbuf_delete(packet);
4661     ofpbuf_uninit(&ofpacts);
4662 }
4663
4664 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4665  * description of the results to 'ds'.
4666  *
4667  * The trace follows a packet with the specified 'flow' through the flow
4668  * table.  'packet' may be nonnull to trace an actual packet, with consequent
4669  * side effects (if it is nonnull then its flow must be 'flow').
4670  *
4671  * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4672  * trace, otherwise the actions are determined by a flow table lookup. */
4673 static void
4674 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
4675               const struct ofpbuf *packet,
4676               const struct ofpact ofpacts[], size_t ofpacts_len,
4677               struct ds *ds)
4678 {
4679     struct rule_dpif *rule;
4680     struct trace_ctx trace;
4681
4682     ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
4683     ds_put_cstr(ds, "Flow: ");
4684     flow_format(ds, flow);
4685     ds_put_char(ds, '\n');
4686
4687     flow_wildcards_init_catchall(&trace.wc);
4688     if (ofpacts) {
4689         rule = NULL;
4690     } else {
4691         rule_dpif_lookup(ofproto, flow, &trace.wc, &rule, false, NULL);
4692
4693         trace_format_rule(ds, 0, rule);
4694         if (rule == ofproto->miss_rule) {
4695             ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
4696         } else if (rule == ofproto->no_packet_in_rule) {
4697             ds_put_cstr(ds, "\nNo match, packets dropped because "
4698                         "OFPPC_NO_PACKET_IN is set on in_port.\n");
4699         } else if (rule == ofproto->drop_frags_rule) {
4700             ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
4701                         "and the fragment handling mode is \"drop\".\n");
4702         }
4703     }
4704
4705     if (rule || ofpacts) {
4706         trace.result = ds;
4707         trace.key = flow; /* Original flow key, used for megaflow. */
4708         trace.flow = *flow; /* May be modified by actions. */
4709         xlate_in_init(&trace.xin, ofproto, flow, flow->in_port.ofp_port, rule,
4710                       ntohs(flow->tcp_flags), packet);
4711         if (ofpacts) {
4712             trace.xin.ofpacts = ofpacts;
4713             trace.xin.ofpacts_len = ofpacts_len;
4714         }
4715         trace.xin.resubmit_hook = trace_resubmit;
4716         trace.xin.report_hook = trace_report;
4717
4718         xlate_actions(&trace.xin, &trace.xout);
4719
4720         ds_put_char(ds, '\n');
4721         trace_format_flow(ds, 0, "Final flow", &trace);
4722         trace_format_megaflow(ds, 0, "Megaflow", &trace);
4723
4724         ds_put_cstr(ds, "Datapath actions: ");
4725         format_odp_actions(ds, ofpbuf_data(trace.xout.odp_actions),
4726                            ofpbuf_size(trace.xout.odp_actions));
4727
4728         if (trace.xout.slow) {
4729             enum slow_path_reason slow;
4730
4731             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
4732                         "slow path because it:");
4733
4734             slow = trace.xout.slow;
4735             while (slow) {
4736                 enum slow_path_reason bit = rightmost_1bit(slow);
4737
4738                 ds_put_format(ds, "\n\t- %s.",
4739                               slow_path_reason_to_explanation(bit));
4740
4741                 slow &= ~bit;
4742             }
4743         }
4744
4745         xlate_out_uninit(&trace.xout);
4746     }
4747 }
4748
4749 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
4750  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
4751  * to destroy 'ofproto_shash' and free the returned value. */
4752 static const struct shash_node **
4753 get_ofprotos(struct shash *ofproto_shash)
4754 {
4755     const struct ofproto_dpif *ofproto;
4756
4757     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4758         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
4759         shash_add_nocopy(ofproto_shash, name, ofproto);
4760     }
4761
4762     return shash_sort(ofproto_shash);
4763 }
4764
4765 static void
4766 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
4767                               const char *argv[] OVS_UNUSED,
4768                               void *aux OVS_UNUSED)
4769 {
4770     struct ds ds = DS_EMPTY_INITIALIZER;
4771     struct shash ofproto_shash;
4772     const struct shash_node **sorted_ofprotos;
4773     int i;
4774
4775     shash_init(&ofproto_shash);
4776     sorted_ofprotos = get_ofprotos(&ofproto_shash);
4777     for (i = 0; i < shash_count(&ofproto_shash); i++) {
4778         const struct shash_node *node = sorted_ofprotos[i];
4779         ds_put_format(&ds, "%s\n", node->name);
4780     }
4781
4782     shash_destroy(&ofproto_shash);
4783     free(sorted_ofprotos);
4784
4785     unixctl_command_reply(conn, ds_cstr(&ds));
4786     ds_destroy(&ds);
4787 }
4788
4789 static void
4790 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
4791 {
4792     const struct shash_node **ofprotos;
4793     struct dpif_dp_stats dp_stats;
4794     struct shash ofproto_shash;
4795     size_t i;
4796
4797     dpif_get_dp_stats(backer->dpif, &dp_stats);
4798
4799     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
4800                   dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
4801
4802     shash_init(&ofproto_shash);
4803     ofprotos = get_ofprotos(&ofproto_shash);
4804     for (i = 0; i < shash_count(&ofproto_shash); i++) {
4805         struct ofproto_dpif *ofproto = ofprotos[i]->data;
4806         const struct shash_node **ports;
4807         size_t j;
4808
4809         if (ofproto->backer != backer) {
4810             continue;
4811         }
4812
4813         ds_put_format(ds, "\t%s:\n", ofproto->up.name);
4814
4815         ports = shash_sort(&ofproto->up.port_by_name);
4816         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
4817             const struct shash_node *node = ports[j];
4818             struct ofport *ofport = node->data;
4819             struct smap config;
4820             odp_port_t odp_port;
4821
4822             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
4823                           ofport->ofp_port);
4824
4825             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
4826             if (odp_port != ODPP_NONE) {
4827                 ds_put_format(ds, "%"PRIu32":", odp_port);
4828             } else {
4829                 ds_put_cstr(ds, "none:");
4830             }
4831
4832             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
4833
4834             smap_init(&config);
4835             if (!netdev_get_config(ofport->netdev, &config)) {
4836                 const struct smap_node **nodes;
4837                 size_t i;
4838
4839                 nodes = smap_sort(&config);
4840                 for (i = 0; i < smap_count(&config); i++) {
4841                     const struct smap_node *node = nodes[i];
4842                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
4843                                   node->key, node->value);
4844                 }
4845                 free(nodes);
4846             }
4847             smap_destroy(&config);
4848
4849             ds_put_char(ds, ')');
4850             ds_put_char(ds, '\n');
4851         }
4852         free(ports);
4853     }
4854     shash_destroy(&ofproto_shash);
4855     free(ofprotos);
4856 }
4857
4858 static void
4859 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4860                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4861 {
4862     struct ds ds = DS_EMPTY_INITIALIZER;
4863     const struct shash_node **backers;
4864     int i;
4865
4866     backers = shash_sort(&all_dpif_backers);
4867     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
4868         dpif_show_backer(backers[i]->data, &ds);
4869     }
4870     free(backers);
4871
4872     unixctl_command_reply(conn, ds_cstr(&ds));
4873     ds_destroy(&ds);
4874 }
4875
4876 static void
4877 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
4878                                 int argc OVS_UNUSED, const char *argv[],
4879                                 void *aux OVS_UNUSED)
4880 {
4881     const struct ofproto_dpif *ofproto;
4882
4883     struct ds ds = DS_EMPTY_INITIALIZER;
4884     bool verbosity = false;
4885
4886     struct dpif_port dpif_port;
4887     struct dpif_port_dump port_dump;
4888     struct hmap portno_names;
4889
4890     struct dpif_flow_dump *flow_dump;
4891     struct dpif_flow_dump_thread *flow_dump_thread;
4892     struct dpif_flow f;
4893     int error;
4894
4895     ofproto = ofproto_dpif_lookup(argv[argc - 1]);
4896     if (!ofproto) {
4897         unixctl_command_reply_error(conn, "no such bridge");
4898         return;
4899     }
4900
4901     if (argc > 2 && !strcmp(argv[1], "-m")) {
4902         verbosity = true;
4903     }
4904
4905     hmap_init(&portno_names);
4906     DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
4907         odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
4908     }
4909
4910     ds_init(&ds);
4911     flow_dump = dpif_flow_dump_create(ofproto->backer->dpif);
4912     flow_dump_thread = dpif_flow_dump_thread_create(flow_dump);
4913     while (dpif_flow_dump_next(flow_dump_thread, &f, 1)) {
4914         struct flow flow;
4915
4916         if (odp_flow_key_to_flow(f.key, f.key_len, &flow) == ODP_FIT_ERROR
4917             || xlate_lookup_ofproto(ofproto->backer, &flow, NULL) != ofproto) {
4918             continue;
4919         }
4920
4921         odp_flow_format(f.key, f.key_len, f.mask, f.mask_len,
4922                         &portno_names, &ds, verbosity);
4923         ds_put_cstr(&ds, ", ");
4924         dpif_flow_stats_format(&f.stats, &ds);
4925         ds_put_cstr(&ds, ", actions:");
4926         format_odp_actions(&ds, f.actions, f.actions_len);
4927         ds_put_char(&ds, '\n');
4928     }
4929     dpif_flow_dump_thread_destroy(flow_dump_thread);
4930     error = dpif_flow_dump_destroy(flow_dump);
4931
4932     if (error) {
4933         ds_clear(&ds);
4934         ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
4935         unixctl_command_reply_error(conn, ds_cstr(&ds));
4936     } else {
4937         unixctl_command_reply(conn, ds_cstr(&ds));
4938     }
4939     odp_portno_names_destroy(&portno_names);
4940     hmap_destroy(&portno_names);
4941     ds_destroy(&ds);
4942 }
4943
4944 static void
4945 ofproto_dpif_unixctl_init(void)
4946 {
4947     static bool registered;
4948     if (registered) {
4949         return;
4950     }
4951     registered = true;
4952
4953     unixctl_command_register(
4954         "ofproto/trace",
4955         "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
4956         1, 3, ofproto_unixctl_trace, NULL);
4957     unixctl_command_register(
4958         "ofproto/trace-packet-out",
4959         "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
4960         2, 6, ofproto_unixctl_trace_actions, NULL);
4961     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
4962                              ofproto_unixctl_fdb_flush, NULL);
4963     unixctl_command_register("fdb/show", "bridge", 1, 1,
4964                              ofproto_unixctl_fdb_show, NULL);
4965     unixctl_command_register("mdb/flush", "[bridge]", 0, 1,
4966                              ofproto_unixctl_mcast_snooping_flush, NULL);
4967     unixctl_command_register("mdb/show", "bridge", 1, 1,
4968                              ofproto_unixctl_mcast_snooping_show, NULL);
4969     unixctl_command_register("dpif/dump-dps", "", 0, 0,
4970                              ofproto_unixctl_dpif_dump_dps, NULL);
4971     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
4972                              NULL);
4973     unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
4974                              ofproto_unixctl_dpif_dump_flows, NULL);
4975 }
4976
4977 /* Returns true if 'table' is the table used for internal rules,
4978  * false otherwise. */
4979 bool
4980 table_is_internal(uint8_t table_id)
4981 {
4982     return table_id == TBL_INTERNAL;
4983 }
4984 \f
4985 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4986  *
4987  * This is deprecated.  It is only for compatibility with broken device drivers
4988  * in old versions of Linux that do not properly support VLANs when VLAN
4989  * devices are not used.  When broken device drivers are no longer in
4990  * widespread use, we will delete these interfaces. */
4991
4992 static int
4993 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
4994 {
4995     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
4996     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
4997
4998     if (realdev_ofp_port == ofport->realdev_ofp_port
4999         && vid == ofport->vlandev_vid) {
5000         return 0;
5001     }
5002
5003     ofproto->backer->need_revalidate = REV_RECONFIGURE;
5004
5005     if (ofport->realdev_ofp_port) {
5006         vsp_remove(ofport);
5007     }
5008     if (realdev_ofp_port && ofport->bundle) {
5009         /* vlandevs are enslaved to their realdevs, so they are not allowed to
5010          * themselves be part of a bundle. */
5011         bundle_set(ofport_->ofproto, ofport->bundle, NULL);
5012     }
5013
5014     ofport->realdev_ofp_port = realdev_ofp_port;
5015     ofport->vlandev_vid = vid;
5016
5017     if (realdev_ofp_port) {
5018         vsp_add(ofport, realdev_ofp_port, vid);
5019     }
5020
5021     return 0;
5022 }
5023
5024 static uint32_t
5025 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
5026 {
5027     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
5028 }
5029
5030 bool
5031 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
5032     OVS_EXCLUDED(ofproto->vsp_mutex)
5033 {
5034     /* hmap_is_empty is thread safe. */
5035     return !hmap_is_empty(&ofproto->realdev_vid_map);
5036 }
5037
5038
5039 static ofp_port_t
5040 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
5041                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5042     OVS_REQUIRES(ofproto->vsp_mutex)
5043 {
5044     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5045         int vid = vlan_tci_to_vid(vlan_tci);
5046         const struct vlan_splinter *vsp;
5047
5048         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5049                                  hash_realdev_vid(realdev_ofp_port, vid),
5050                                  &ofproto->realdev_vid_map) {
5051             if (vsp->realdev_ofp_port == realdev_ofp_port
5052                 && vsp->vid == vid) {
5053                 return vsp->vlandev_ofp_port;
5054             }
5055         }
5056     }
5057     return realdev_ofp_port;
5058 }
5059
5060 /* Returns the OFP port number of the Linux VLAN device that corresponds to
5061  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
5062  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
5063  * 'vlan_tci' 9, it would return the port number of eth0.9.
5064  *
5065  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
5066  * function just returns its 'realdev_ofp_port' argument. */
5067 ofp_port_t
5068 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5069                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5070     OVS_EXCLUDED(ofproto->vsp_mutex)
5071 {
5072     ofp_port_t ret;
5073
5074     /* hmap_is_empty is thread safe, see if we can return immediately. */
5075     if (hmap_is_empty(&ofproto->realdev_vid_map)) {
5076         return realdev_ofp_port;
5077     }
5078     ovs_mutex_lock(&ofproto->vsp_mutex);
5079     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
5080     ovs_mutex_unlock(&ofproto->vsp_mutex);
5081     return ret;
5082 }
5083
5084 static struct vlan_splinter *
5085 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
5086 {
5087     struct vlan_splinter *vsp;
5088
5089     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
5090                              hash_ofp_port(vlandev_ofp_port),
5091                              &ofproto->vlandev_map) {
5092         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5093             return vsp;
5094         }
5095     }
5096
5097     return NULL;
5098 }
5099
5100 /* Returns the OpenFlow port number of the "real" device underlying the Linux
5101  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
5102  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
5103  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
5104  * eth0 and store 9 in '*vid'.
5105  *
5106  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
5107  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
5108  * always does.*/
5109 static ofp_port_t
5110 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5111                        ofp_port_t vlandev_ofp_port, int *vid)
5112     OVS_REQUIRES(ofproto->vsp_mutex)
5113 {
5114     if (!hmap_is_empty(&ofproto->vlandev_map)) {
5115         const struct vlan_splinter *vsp;
5116
5117         vsp = vlandev_find(ofproto, vlandev_ofp_port);
5118         if (vsp) {
5119             if (vid) {
5120                 *vid = vsp->vid;
5121             }
5122             return vsp->realdev_ofp_port;
5123         }
5124     }
5125     return 0;
5126 }
5127
5128 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
5129  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
5130  * 'flow->in_port' to the "real" device backing the VLAN device, sets
5131  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Optionally pushes the
5132  * appropriate VLAN on 'packet' if provided.  Otherwise (which is always the
5133  * case unless VLAN splinters are enabled), returns false without making any
5134  * changes. */
5135 bool
5136 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow,
5137                 struct ofpbuf *packet)
5138     OVS_EXCLUDED(ofproto->vsp_mutex)
5139 {
5140     ofp_port_t realdev;
5141     int vid;
5142
5143     /* hmap_is_empty is thread safe. */
5144     if (hmap_is_empty(&ofproto->vlandev_map)) {
5145         return false;
5146     }
5147
5148     ovs_mutex_lock(&ofproto->vsp_mutex);
5149     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
5150     ovs_mutex_unlock(&ofproto->vsp_mutex);
5151     if (!realdev) {
5152         return false;
5153     }
5154
5155     /* Cause the flow to be processed as if it came in on the real device with
5156      * the VLAN device's VLAN ID. */
5157     flow->in_port.ofp_port = realdev;
5158     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
5159
5160     if (packet) {
5161         /* Make the packet resemble the flow, so that it gets sent to an
5162          * OpenFlow controller properly, so that it looks correct for sFlow,
5163          * and so that flow_extract() will get the correct vlan_tci if it is
5164          * called on 'packet'. */
5165         eth_push_vlan(packet, htons(ETH_TYPE_VLAN), flow->vlan_tci);
5166     }
5167
5168     return true;
5169 }
5170
5171 static void
5172 vsp_remove(struct ofport_dpif *port)
5173 {
5174     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5175     struct vlan_splinter *vsp;
5176
5177     ovs_mutex_lock(&ofproto->vsp_mutex);
5178     vsp = vlandev_find(ofproto, port->up.ofp_port);
5179     if (vsp) {
5180         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5181         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
5182         free(vsp);
5183
5184         port->realdev_ofp_port = 0;
5185     } else {
5186         VLOG_ERR("missing vlan device record");
5187     }
5188     ovs_mutex_unlock(&ofproto->vsp_mutex);
5189 }
5190
5191 static void
5192 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
5193 {
5194     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5195
5196     ovs_mutex_lock(&ofproto->vsp_mutex);
5197     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
5198         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
5199             == realdev_ofp_port)) {
5200         struct vlan_splinter *vsp;
5201
5202         vsp = xmalloc(sizeof *vsp);
5203         vsp->realdev_ofp_port = realdev_ofp_port;
5204         vsp->vlandev_ofp_port = port->up.ofp_port;
5205         vsp->vid = vid;
5206
5207         port->realdev_ofp_port = realdev_ofp_port;
5208
5209         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
5210                     hash_ofp_port(port->up.ofp_port));
5211         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
5212                     hash_realdev_vid(realdev_ofp_port, vid));
5213     } else {
5214         VLOG_ERR("duplicate vlan device record");
5215     }
5216     ovs_mutex_unlock(&ofproto->vsp_mutex);
5217 }
5218
5219 static odp_port_t
5220 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
5221 {
5222     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
5223     return ofport ? ofport->odp_port : ODPP_NONE;
5224 }
5225
5226 struct ofport_dpif *
5227 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
5228 {
5229     struct ofport_dpif *port;
5230
5231     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
5232     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
5233                              &backer->odp_to_ofport_map) {
5234         if (port->odp_port == odp_port) {
5235             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5236             return port;
5237         }
5238     }
5239
5240     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5241     return NULL;
5242 }
5243
5244 static ofp_port_t
5245 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
5246 {
5247     struct ofport_dpif *port;
5248
5249     port = odp_port_to_ofport(ofproto->backer, odp_port);
5250     if (port && &ofproto->up == port->up.ofproto) {
5251         return port->up.ofp_port;
5252     } else {
5253         return OFPP_NONE;
5254     }
5255 }
5256
5257 uint32_t
5258 ofproto_dpif_alloc_recirc_id(struct ofproto_dpif *ofproto)
5259 {
5260     struct dpif_backer *backer = ofproto->backer;
5261
5262     return  recirc_id_alloc(backer->rid_pool);
5263 }
5264
5265 void
5266 ofproto_dpif_free_recirc_id(struct ofproto_dpif *ofproto, uint32_t recirc_id)
5267 {
5268     struct dpif_backer *backer = ofproto->backer;
5269
5270     recirc_id_free(backer->rid_pool, recirc_id);
5271 }
5272
5273 int
5274 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
5275                                const struct match *match, int priority,
5276                                uint16_t idle_timeout,
5277                                const struct ofpbuf *ofpacts,
5278                                struct rule **rulep)
5279 {
5280     struct ofputil_flow_mod fm;
5281     struct rule_dpif *rule;
5282     int error;
5283
5284     fm.match = *match;
5285     fm.priority = priority;
5286     fm.new_cookie = htonll(0);
5287     fm.cookie = htonll(0);
5288     fm.cookie_mask = htonll(0);
5289     fm.modify_cookie = false;
5290     fm.table_id = TBL_INTERNAL;
5291     fm.command = OFPFC_ADD;
5292     fm.idle_timeout = idle_timeout;
5293     fm.hard_timeout = 0;
5294     fm.buffer_id = 0;
5295     fm.out_port = 0;
5296     fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5297     fm.ofpacts = ofpbuf_data(ofpacts);
5298     fm.ofpacts_len = ofpbuf_size(ofpacts);
5299
5300     error = ofproto_flow_mod(&ofproto->up, &fm);
5301     if (error) {
5302         VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
5303                     ofperr_to_string(error));
5304         *rulep = NULL;
5305         return error;
5306     }
5307
5308     rule = rule_dpif_lookup_in_table(ofproto, TBL_INTERNAL, &fm.match.flow,
5309                                      &fm.match.wc, false);
5310     if (rule) {
5311         *rulep = &rule->up;
5312     } else {
5313         OVS_NOT_REACHED();
5314     }
5315     return 0;
5316 }
5317
5318 int
5319 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
5320                                   struct match *match, int priority)
5321 {
5322     struct ofputil_flow_mod fm;
5323     int error;
5324
5325     fm.match = *match;
5326     fm.priority = priority;
5327     fm.new_cookie = htonll(0);
5328     fm.cookie = htonll(0);
5329     fm.cookie_mask = htonll(0);
5330     fm.modify_cookie = false;
5331     fm.table_id = TBL_INTERNAL;
5332     fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5333     fm.command = OFPFC_DELETE_STRICT;
5334
5335     error = ofproto_flow_mod(&ofproto->up, &fm);
5336     if (error) {
5337         VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
5338                     ofperr_to_string(error));
5339         return error;
5340     }
5341
5342     return 0;
5343 }
5344
5345 const struct ofproto_class ofproto_dpif_class = {
5346     init,
5347     enumerate_types,
5348     enumerate_names,
5349     del,
5350     port_open_type,
5351     type_run,
5352     type_wait,
5353     alloc,
5354     construct,
5355     destruct,
5356     dealloc,
5357     run,
5358     wait,
5359     NULL,                       /* get_memory_usage. */
5360     type_get_memory_usage,
5361     flush,
5362     query_tables,
5363     port_alloc,
5364     port_construct,
5365     port_destruct,
5366     port_dealloc,
5367     port_modified,
5368     port_reconfigured,
5369     port_query_by_name,
5370     port_add,
5371     port_del,
5372     port_get_stats,
5373     port_dump_start,
5374     port_dump_next,
5375     port_dump_done,
5376     port_poll,
5377     port_poll_wait,
5378     port_is_lacp_current,
5379     NULL,                       /* rule_choose_table */
5380     rule_alloc,
5381     rule_construct,
5382     rule_insert,
5383     rule_delete,
5384     rule_destruct,
5385     rule_dealloc,
5386     rule_get_stats,
5387     rule_execute,
5388     NULL,                       /* rule_premodify_actions */
5389     rule_modify_actions,
5390     set_frag_handling,
5391     packet_out,
5392     set_netflow,
5393     get_netflow_ids,
5394     set_sflow,
5395     set_ipfix,
5396     set_cfm,
5397     cfm_status_changed,
5398     get_cfm_status,
5399     set_bfd,
5400     bfd_status_changed,
5401     get_bfd_status,
5402     set_stp,
5403     get_stp_status,
5404     set_stp_port,
5405     get_stp_port_status,
5406     get_stp_port_stats,
5407     set_rstp,
5408     get_rstp_status,
5409     set_rstp_port,
5410     get_rstp_port_status,
5411     set_queues,
5412     bundle_set,
5413     bundle_remove,
5414     mirror_set__,
5415     mirror_get_stats__,
5416     set_flood_vlans,
5417     is_mirror_output_bundle,
5418     forward_bpdu_changed,
5419     set_mac_table_config,
5420     set_mcast_snooping,
5421     set_mcast_snooping_port,
5422     set_realdev,
5423     NULL,                       /* meter_get_features */
5424     NULL,                       /* meter_set */
5425     NULL,                       /* meter_get */
5426     NULL,                       /* meter_del */
5427     group_alloc,                /* group_alloc */
5428     group_construct,            /* group_construct */
5429     group_destruct,             /* group_destruct */
5430     group_dealloc,              /* group_dealloc */
5431     group_modify,               /* group_modify */
5432     group_get_stats,            /* group_get_stats */
5433 };