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 "meta-flow.h"
  41 #include "multipath.h"
  42 #include "netdev-vport.h"
  43 #include "netdev.h"
  44 #include "netlink.h"
  45 #include "nx-match.h"
  46 #include "odp-util.h"
  47 #include "odp-execute.h"
  48 #include "ofp-util.h"
  49 #include "ofpbuf.h"
  50 #include "ofp-actions.h"
  51 #include "ofp-parse.h"
  52 #include "ofp-print.h"
  53 #include "ofproto-dpif-ipfix.h"
  54 #include "ofproto-dpif-mirror.h"
  55 #include "ofproto-dpif-monitor.h"
  56 #include "ofproto-dpif-sflow.h"
  57 #include "ofproto-dpif-upcall.h"
  58 #include "ofproto-dpif-xlate.h"
  59 #include "poll-loop.h"
  60 #include "seq.h"
  61 #include "simap.h"
  62 #include "smap.h"
  63 #include "timer.h"
  64 #include "tunnel.h"
  65 #include "unaligned.h"
  66 #include "unixctl.h"
  67 #include "vlan-bitmap.h"
  68 #include "vlog.h"
  69
  70 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  71
  72 COVERAGE_DEFINE(ofproto_dpif_expired);
  73 COVERAGE_DEFINE(packet_in_overflow);
  74
  75 /* Number of implemented OpenFlow tables. */
  76 enum { N_TABLES = 255 };
  77 enum { TBL_INTERNAL = N_TABLES - 1 };    /* Used for internal hidden rules. */
  78 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
  79
  80 struct flow_miss;
  81
  82 struct rule_dpif {
  83     struct rule up;
  84
  85     /* These statistics:
  86      *
  87      *   - Do include packets and bytes from datapath flows which have not
  88      *   recently been processed by a revalidator. */
  89     struct ovs_mutex stats_mutex;
  90     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
  91     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
  92 };
  93
  94 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
  95 static struct rule_dpif *rule_dpif_cast(const struct rule *);
  96 static void rule_expire(struct rule_dpif *);
  97
  98 struct group_dpif {
  99     struct ofgroup up;
 100
 101     /* These statistics:
 102      *
 103      *   - Do include packets and bytes from datapath flows which have not
 104      *   recently been processed by a revalidator. */
 105     struct ovs_mutex stats_mutex;
 106     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 107     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 108     struct bucket_counter *bucket_stats OVS_GUARDED;  /* Bucket statistics. */
 109 };
 110
 111 struct ofbundle {
 112     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
 113     struct ofproto_dpif *ofproto; /* Owning ofproto. */
 114     void *aux;                  /* Key supplied by ofproto's client. */
 115     char *name;                 /* Identifier for log messages. */
 116
 117     /* Configuration. */
 118     struct list ports;          /* Contains "struct ofport"s. */
 119     enum port_vlan_mode vlan_mode; /* VLAN mode */
 120     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
 121     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
 122                                  * NULL if all VLANs are trunked. */
 123     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 124     struct bond *bond;          /* Nonnull iff more than one port. */
 125     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 126
 127     /* Status. */
 128     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 129 };
 130
 131 static void bundle_remove(struct ofport *);
 132 static void bundle_update(struct ofbundle *);
 133 static void bundle_destroy(struct ofbundle *);
 134 static void bundle_del_port(struct ofport_dpif *);
 135 static void bundle_run(struct ofbundle *);
 136 static void bundle_wait(struct ofbundle *);
 137
 138 static void stp_run(struct ofproto_dpif *ofproto);
 139 static void stp_wait(struct ofproto_dpif *ofproto);
 140 static int set_stp_port(struct ofport *,
 141                         const struct ofproto_port_stp_settings *);
 142
 143 struct ofport_dpif {
 144     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 145     struct ofport up;
 146
 147     odp_port_t odp_port;
 148     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 149     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
 150     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 151     struct bfd *bfd;            /* BFD, if any. */
 152     bool may_enable;            /* May be enabled in bonds. */
 153     bool is_tunnel;             /* This port is a tunnel. */
 154     bool is_layer3;             /* This is a layer 3 port. */
 155     long long int carrier_seq;  /* Carrier status changes. */
 156     struct ofport_dpif *peer;   /* Peer if patch port. */
 157
 158     /* Spanning tree. */
 159     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 160     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 161     long long int stp_state_entered;
 162
 163     /* Queue to DSCP mapping. */
 164     struct ofproto_port_queue *qdscp;
 165     size_t n_qdscp;
 166
 167     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 168      *
 169      * This is deprecated.  It is only for compatibility with broken device
 170      * drivers in old versions of Linux that do not properly support VLANs when
 171      * VLAN devices are not used.  When broken device drivers are no longer in
 172      * widespread use, we will delete these interfaces. */
 173     ofp_port_t realdev_ofp_port;
 174     int vlandev_vid;
 175 };
 176
 177 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 178  *
 179  * This is deprecated.  It is only for compatibility with broken device drivers
 180  * in old versions of Linux that do not properly support VLANs when VLAN
 181  * devices are not used.  When broken device drivers are no longer in
 182  * widespread use, we will delete these interfaces. */
 183 struct vlan_splinter {
 184     struct hmap_node realdev_vid_node;
 185     struct hmap_node vlandev_node;
 186     ofp_port_t realdev_ofp_port;
 187     ofp_port_t vlandev_ofp_port;
 188     int vid;
 189 };
 190
 191 static void vsp_remove(struct ofport_dpif *);
 192 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 193
 194 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 195                                        ofp_port_t);
 196
 197 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 198                                        odp_port_t);
 199
 200 static struct ofport_dpif *
 201 ofport_dpif_cast(const struct ofport *ofport)
 202 {
 203     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 204 }
 205
 206 static void port_run(struct ofport_dpif *);
 207 static int set_bfd(struct ofport *, const struct smap *);
 208 static int set_cfm(struct ofport *, const struct cfm_settings *);
 209 static void ofport_update_peer(struct ofport_dpif *);
 210
 211 struct dpif_completion {
 212     struct list list_node;
 213     struct ofoperation *op;
 214 };
 215
 216 /* Reasons that we might need to revalidate every datapath flow, and
 217  * corresponding coverage counters.
 218  *
 219  * A value of 0 means that there is no need to revalidate.
 220  *
 221  * It would be nice to have some cleaner way to integrate with coverage
 222  * counters, but with only a few reasons I guess this is good enough for
 223  * now. */
 224 enum revalidate_reason {
 225     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 226     REV_STP,                   /* Spanning tree protocol port status change. */
 227     REV_BOND,                  /* Bonding changed. */
 228     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 229     REV_FLOW_TABLE,            /* Flow table changed. */
 230     REV_MAC_LEARNING,          /* Mac learning changed. */
 231 };
 232 COVERAGE_DEFINE(rev_reconfigure);
 233 COVERAGE_DEFINE(rev_stp);
 234 COVERAGE_DEFINE(rev_bond);
 235 COVERAGE_DEFINE(rev_port_toggled);
 236 COVERAGE_DEFINE(rev_flow_table);
 237 COVERAGE_DEFINE(rev_mac_learning);
 238
 239 /* All datapaths of a given type share a single dpif backer instance. */
 240 struct dpif_backer {
 241     char *type;
 242     int refcount;
 243     struct dpif *dpif;
 244     struct udpif *udpif;
 245
 246     struct ovs_rwlock odp_to_ofport_lock;
 247     struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
 248
 249     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 250
 251     enum revalidate_reason need_revalidate; /* Revalidate all flows. */
 252
 253     bool recv_set_enable; /* Enables or disables receiving packets. */
 254 };
 255
 256 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 257 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 258
 259 struct ofproto_dpif {
 260     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 261     struct ofproto up;
 262     struct dpif_backer *backer;
 263
 264     uint64_t dump_seq; /* Last read of udpif_dump_seq(). */
 265
 266     /* Special OpenFlow rules. */
 267     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 268     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 269     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 270
 271     /* Bridging. */
 272     struct netflow *netflow;
 273     struct dpif_sflow *sflow;
 274     struct dpif_ipfix *ipfix;
 275     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 276     struct mac_learning *ml;
 277     bool has_bonded_bundles;
 278     bool lacp_enabled;
 279     struct mbridge *mbridge;
 280
 281     struct ovs_mutex stats_mutex;
 282     struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
 283                                             * consumed in userspace. */
 284
 285     /* Spanning tree. */
 286     struct stp *stp;
 287     long long int stp_last_tick;
 288
 289     /* VLAN splinters. */
 290     struct ovs_mutex vsp_mutex;
 291     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 292     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 293
 294     /* Ports. */
 295     struct sset ports;             /* Set of standard port names. */
 296     struct sset ghost_ports;       /* Ports with no datapath port. */
 297     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 298     int port_poll_errno;           /* Last errno for port_poll() reply. */
 299     uint64_t change_seq;           /* Connectivity status changes. */
 300
 301     /* Work queues. */
 302     struct guarded_list pins;      /* Contains "struct ofputil_packet_in"s. */
 303 };
 304
 305 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 306 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 307
 308 static void ofproto_dpif_unixctl_init(void);
 309
 310 static inline struct ofproto_dpif *
 311 ofproto_dpif_cast(const struct ofproto *ofproto)
 312 {
 313     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 314     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 315 }
 316
 317 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
 318                                         ofp_port_t ofp_port);
 319 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
 320                           const struct ofpbuf *packet,
 321                           const struct ofpact[], size_t ofpacts_len,
 322                           struct ds *);
 323
 324 /* Global variables. */
 325 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 326
 327 /* Initial mappings of port to bridge mappings. */
 328 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 329
 330 /* Executes 'fm'.  The caller retains ownership of 'fm' and everything in
 331  * it. */
 332 void
 333 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 334                       struct ofputil_flow_mod *fm)
 335 {
 336     ofproto_flow_mod(&ofproto->up, fm);
 337 }
 338
 339 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 340  * Takes ownership of 'pin' and pin->packet. */
 341 void
 342 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 343                             struct ofproto_packet_in *pin)
 344 {
 345     if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
 346         COVERAGE_INC(packet_in_overflow);
 347         free(CONST_CAST(void *, pin->up.packet));
 348         free(pin);
 349     }
 350 }
 351 \f
 352 /* Factory functions. */
 353
 354 static void
 355 init(const struct shash *iface_hints)
 356 {
 357     struct shash_node *node;
 358
 359     /* Make a local copy, since we don't own 'iface_hints' elements. */
 360     SHASH_FOR_EACH(node, iface_hints) {
 361         const struct iface_hint *orig_hint = node->data;
 362         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 363
 364         new_hint->br_name = xstrdup(orig_hint->br_name);
 365         new_hint->br_type = xstrdup(orig_hint->br_type);
 366         new_hint->ofp_port = orig_hint->ofp_port;
 367
 368         shash_add(&init_ofp_ports, node->name, new_hint);
 369     }
 370 }
 371
 372 static void
 373 enumerate_types(struct sset *types)
 374 {
 375     dp_enumerate_types(types);
 376 }
 377
 378 static int
 379 enumerate_names(const char *type, struct sset *names)
 380 {
 381     struct ofproto_dpif *ofproto;
 382
 383     sset_clear(names);
 384     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 385         if (strcmp(type, ofproto->up.type)) {
 386             continue;
 387         }
 388         sset_add(names, ofproto->up.name);
 389     }
 390
 391     return 0;
 392 }
 393
 394 static int
 395 del(const char *type, const char *name)
 396 {
 397     struct dpif *dpif;
 398     int error;
 399
 400     error = dpif_open(name, type, &dpif);
 401     if (!error) {
 402         error = dpif_delete(dpif);
 403         dpif_close(dpif);
 404     }
 405     return error;
 406 }
 407 \f
 408 static const char *
 409 port_open_type(const char *datapath_type, const char *port_type)
 410 {
 411     return dpif_port_open_type(datapath_type, port_type);
 412 }
 413
 414 /* Type functions. */
 415
 416 static void process_dpif_port_changes(struct dpif_backer *);
 417 static void process_dpif_all_ports_changed(struct dpif_backer *);
 418 static void process_dpif_port_change(struct dpif_backer *,
 419                                      const char *devname);
 420 static void process_dpif_port_error(struct dpif_backer *, int error);
 421
 422 static struct ofproto_dpif *
 423 lookup_ofproto_dpif_by_port_name(const char *name)
 424 {
 425     struct ofproto_dpif *ofproto;
 426
 427     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 428         if (sset_contains(&ofproto->ports, name)) {
 429             return ofproto;
 430         }
 431     }
 432
 433     return NULL;
 434 }
 435
 436 static int
 437 type_run(const char *type)
 438 {
 439     struct dpif_backer *backer;
 440
 441     backer = shash_find_data(&all_dpif_backers, type);
 442     if (!backer) {
 443         /* This is not necessarily a problem, since backers are only
 444          * created on demand. */
 445         return 0;
 446     }
 447
 448     dpif_run(backer->dpif);
 449
 450     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 451      * and the configuration has now changed to "false", enable receiving
 452      * packets from the datapath. */
 453     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 454         int error;
 455
 456         backer->recv_set_enable = true;
 457
 458         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 459         if (error) {
 460             VLOG_ERR("Failed to enable receiving packets in dpif.");
 461             return error;
 462         }
 463         dpif_flow_flush(backer->dpif);
 464         backer->need_revalidate = REV_RECONFIGURE;
 465     }
 466
 467     if (backer->recv_set_enable) {
 468         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 469     }
 470
 471     if (backer->need_revalidate) {
 472         struct ofproto_dpif *ofproto;
 473         struct simap_node *node;
 474         struct simap tmp_backers;
 475
 476         /* Handle tunnel garbage collection. */
 477         simap_init(&tmp_backers);
 478         simap_swap(&backer->tnl_backers, &tmp_backers);
 479
 480         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 481             struct ofport_dpif *iter;
 482
 483             if (backer != ofproto->backer) {
 484                 continue;
 485             }
 486
 487             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 488                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 489                 const char *dp_port;
 490
 491                 if (!iter->is_tunnel) {
 492                     continue;
 493                 }
 494
 495                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 496                                                      namebuf, sizeof namebuf);
 497                 node = simap_find(&tmp_backers, dp_port);
 498                 if (node) {
 499                     simap_put(&backer->tnl_backers, dp_port, node->data);
 500                     simap_delete(&tmp_backers, node);
 501                     node = simap_find(&backer->tnl_backers, dp_port);
 502                 } else {
 503                     node = simap_find(&backer->tnl_backers, dp_port);
 504                     if (!node) {
 505                         odp_port_t odp_port = ODPP_NONE;
 506
 507                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 508                                            &odp_port)) {
 509                             simap_put(&backer->tnl_backers, dp_port,
 510                                       odp_to_u32(odp_port));
 511                             node = simap_find(&backer->tnl_backers, dp_port);
 512                         }
 513                     }
 514                 }
 515
 516                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 517                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 518                                          iter->odp_port)) {
 519                     backer->need_revalidate = REV_RECONFIGURE;
 520                 }
 521             }
 522         }
 523
 524         SIMAP_FOR_EACH (node, &tmp_backers) {
 525             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 526         }
 527         simap_destroy(&tmp_backers);
 528
 529         switch (backer->need_revalidate) {
 530         case REV_RECONFIGURE:   COVERAGE_INC(rev_reconfigure);   break;
 531         case REV_STP:           COVERAGE_INC(rev_stp);           break;
 532         case REV_BOND:          COVERAGE_INC(rev_bond);          break;
 533         case REV_PORT_TOGGLED:  COVERAGE_INC(rev_port_toggled);  break;
 534         case REV_FLOW_TABLE:    COVERAGE_INC(rev_flow_table);    break;
 535         case REV_MAC_LEARNING:  COVERAGE_INC(rev_mac_learning);  break;
 536         }
 537         backer->need_revalidate = 0;
 538
 539         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 540             struct ofport_dpif *ofport;
 541             struct ofbundle *bundle;
 542
 543             if (ofproto->backer != backer) {
 544                 continue;
 545             }
 546
 547             ovs_rwlock_wrlock(&xlate_rwlock);
 548             xlate_ofproto_set(ofproto, ofproto->up.name,
 549                               ofproto->backer->dpif, ofproto->miss_rule,
 550                               ofproto->no_packet_in_rule, ofproto->ml,
 551                               ofproto->stp, ofproto->mbridge,
 552                               ofproto->sflow, ofproto->ipfix,
 553                               ofproto->netflow, ofproto->up.frag_handling,
 554                               ofproto->up.forward_bpdu,
 555                               connmgr_has_in_band(ofproto->up.connmgr));
 556
 557             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 558                 xlate_bundle_set(ofproto, bundle, bundle->name,
 559                                  bundle->vlan_mode, bundle->vlan,
 560                                  bundle->trunks, bundle->use_priority_tags,
 561                                  bundle->bond, bundle->lacp,
 562                                  bundle->floodable);
 563             }
 564
 565             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 566                 int stp_port = ofport->stp_port
 567                     ? stp_port_no(ofport->stp_port)
 568                     : -1;
 569                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 570                                  ofport->up.ofp_port, ofport->odp_port,
 571                                  ofport->up.netdev, ofport->cfm,
 572                                  ofport->bfd, ofport->peer, stp_port,
 573                                  ofport->qdscp, ofport->n_qdscp,
 574                                  ofport->up.pp.config, ofport->up.pp.state,
 575                                  ofport->is_tunnel, ofport->may_enable);
 576             }
 577             ovs_rwlock_unlock(&xlate_rwlock);
 578         }
 579
 580         udpif_revalidate(backer->udpif);
 581     }
 582
 583     process_dpif_port_changes(backer);
 584
 585     return 0;
 586 }
 587
 588 /* Check for and handle port changes in 'backer''s dpif. */
 589 static void
 590 process_dpif_port_changes(struct dpif_backer *backer)
 591 {
 592     for (;;) {
 593         char *devname;
 594         int error;
 595
 596         error = dpif_port_poll(backer->dpif, &devname);
 597         switch (error) {
 598         case EAGAIN:
 599             return;
 600
 601         case ENOBUFS:
 602             process_dpif_all_ports_changed(backer);
 603             break;
 604
 605         case 0:
 606             process_dpif_port_change(backer, devname);
 607             free(devname);
 608             break;
 609
 610         default:
 611             process_dpif_port_error(backer, error);
 612             break;
 613         }
 614     }
 615 }
 616
 617 static void
 618 process_dpif_all_ports_changed(struct dpif_backer *backer)
 619 {
 620     struct ofproto_dpif *ofproto;
 621     struct dpif_port dpif_port;
 622     struct dpif_port_dump dump;
 623     struct sset devnames;
 624     const char *devname;
 625
 626     sset_init(&devnames);
 627     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 628         if (ofproto->backer == backer) {
 629             struct ofport *ofport;
 630
 631             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 632                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 633             }
 634         }
 635     }
 636     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 637         sset_add(&devnames, dpif_port.name);
 638     }
 639
 640     SSET_FOR_EACH (devname, &devnames) {
 641         process_dpif_port_change(backer, devname);
 642     }
 643     sset_destroy(&devnames);
 644 }
 645
 646 static void
 647 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 648 {
 649     struct ofproto_dpif *ofproto;
 650     struct dpif_port port;
 651
 652     /* Don't report on the datapath's device. */
 653     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 654         return;
 655     }
 656
 657     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 658                    &all_ofproto_dpifs) {
 659         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 660             return;
 661         }
 662     }
 663
 664     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 665     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 666         /* The port was removed.  If we know the datapath,
 667          * report it through poll_set().  If we don't, it may be
 668          * notifying us of a removal we initiated, so ignore it.
 669          * If there's a pending ENOBUFS, let it stand, since
 670          * everything will be reevaluated. */
 671         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 672             sset_add(&ofproto->port_poll_set, devname);
 673             ofproto->port_poll_errno = 0;
 674         }
 675     } else if (!ofproto) {
 676         /* The port was added, but we don't know with which
 677          * ofproto we should associate it.  Delete it. */
 678         dpif_port_del(backer->dpif, port.port_no);
 679     } else {
 680         struct ofport_dpif *ofport;
 681
 682         ofport = ofport_dpif_cast(shash_find_data(
 683                                       &ofproto->up.port_by_name, devname));
 684         if (ofport
 685             && ofport->odp_port != port.port_no
 686             && !odp_port_to_ofport(backer, port.port_no))
 687         {
 688             /* 'ofport''s datapath port number has changed from
 689              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 690              * structures to match. */
 691             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 692             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 693             ofport->odp_port = port.port_no;
 694             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 695                         hash_odp_port(port.port_no));
 696             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 697             backer->need_revalidate = REV_RECONFIGURE;
 698         }
 699     }
 700     dpif_port_destroy(&port);
 701 }
 702
 703 /* Propagate 'error' to all ofprotos based on 'backer'. */
 704 static void
 705 process_dpif_port_error(struct dpif_backer *backer, int error)
 706 {
 707     struct ofproto_dpif *ofproto;
 708
 709     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 710         if (ofproto->backer == backer) {
 711             sset_clear(&ofproto->port_poll_set);
 712             ofproto->port_poll_errno = error;
 713         }
 714     }
 715 }
 716
 717 static void
 718 type_wait(const char *type)
 719 {
 720     struct dpif_backer *backer;
 721
 722     backer = shash_find_data(&all_dpif_backers, type);
 723     if (!backer) {
 724         /* This is not necessarily a problem, since backers are only
 725          * created on demand. */
 726         return;
 727     }
 728
 729     dpif_wait(backer->dpif);
 730 }
 731 \f
 732 /* Basic life-cycle. */
 733
 734 static int add_internal_flows(struct ofproto_dpif *);
 735
 736 static struct ofproto *
 737 alloc(void)
 738 {
 739     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
 740     return &ofproto->up;
 741 }
 742
 743 static void
 744 dealloc(struct ofproto *ofproto_)
 745 {
 746     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
 747     free(ofproto);
 748 }
 749
 750 static void
 751 close_dpif_backer(struct dpif_backer *backer)
 752 {
 753     ovs_assert(backer->refcount > 0);
 754
 755     if (--backer->refcount) {
 756         return;
 757     }
 758
 759     udpif_destroy(backer->udpif);
 760
 761     simap_destroy(&backer->tnl_backers);
 762     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
 763     hmap_destroy(&backer->odp_to_ofport_map);
 764     shash_find_and_delete(&all_dpif_backers, backer->type);
 765     free(backer->type);
 766     dpif_close(backer->dpif);
 767
 768     free(backer);
 769 }
 770
 771 /* Datapath port slated for removal from datapath. */
 772 struct odp_garbage {
 773     struct list list_node;
 774     odp_port_t odp_port;
 775 };
 776
 777 static int
 778 open_dpif_backer(const char *type, struct dpif_backer **backerp)
 779 {
 780     struct dpif_backer *backer;
 781     struct dpif_port_dump port_dump;
 782     struct dpif_port port;
 783     struct shash_node *node;
 784     struct list garbage_list;
 785     struct odp_garbage *garbage, *next;
 786     struct sset names;
 787     char *backer_name;
 788     const char *name;
 789     int error;
 790
 791     backer = shash_find_data(&all_dpif_backers, type);
 792     if (backer) {
 793         backer->refcount++;
 794         *backerp = backer;
 795         return 0;
 796     }
 797
 798     backer_name = xasprintf("ovs-%s", type);
 799
 800     /* Remove any existing datapaths, since we assume we're the only
 801      * userspace controlling the datapath. */
 802     sset_init(&names);
 803     dp_enumerate_names(type, &names);
 804     SSET_FOR_EACH(name, &names) {
 805         struct dpif *old_dpif;
 806
 807         /* Don't remove our backer if it exists. */
 808         if (!strcmp(name, backer_name)) {
 809             continue;
 810         }
 811
 812         if (dpif_open(name, type, &old_dpif)) {
 813             VLOG_WARN("couldn't open old datapath %s to remove it", name);
 814         } else {
 815             dpif_delete(old_dpif);
 816             dpif_close(old_dpif);
 817         }
 818     }
 819     sset_destroy(&names);
 820
 821     backer = xmalloc(sizeof *backer);
 822
 823     error = dpif_create_and_open(backer_name, type, &backer->dpif);
 824     free(backer_name);
 825     if (error) {
 826         VLOG_ERR("failed to open datapath of type %s: %s", type,
 827                  ovs_strerror(error));
 828         free(backer);
 829         return error;
 830     }
 831     backer->udpif = udpif_create(backer, backer->dpif);
 832
 833     backer->type = xstrdup(type);
 834     backer->refcount = 1;
 835     hmap_init(&backer->odp_to_ofport_map);
 836     ovs_rwlock_init(&backer->odp_to_ofport_lock);
 837     backer->need_revalidate = 0;
 838     simap_init(&backer->tnl_backers);
 839     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
 840     *backerp = backer;
 841
 842     if (backer->recv_set_enable) {
 843         dpif_flow_flush(backer->dpif);
 844     }
 845
 846     /* Loop through the ports already on the datapath and remove any
 847      * that we don't need anymore. */
 848     list_init(&garbage_list);
 849     dpif_port_dump_start(&port_dump, backer->dpif);
 850     while (dpif_port_dump_next(&port_dump, &port)) {
 851         node = shash_find(&init_ofp_ports, port.name);
 852         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
 853             garbage = xmalloc(sizeof *garbage);
 854             garbage->odp_port = port.port_no;
 855             list_push_front(&garbage_list, &garbage->list_node);
 856         }
 857     }
 858     dpif_port_dump_done(&port_dump);
 859
 860     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
 861         dpif_port_del(backer->dpif, garbage->odp_port);
 862         list_remove(&garbage->list_node);
 863         free(garbage);
 864     }
 865
 866     shash_add(&all_dpif_backers, type, backer);
 867
 868     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 869     if (error) {
 870         VLOG_ERR("failed to listen on datapath of type %s: %s",
 871                  type, ovs_strerror(error));
 872         close_dpif_backer(backer);
 873         return error;
 874     }
 875
 876     if (backer->recv_set_enable) {
 877         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 878     }
 879
 880     return error;
 881 }
 882
 883 static int
 884 construct(struct ofproto *ofproto_)
 885 {
 886     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
 887     struct shash_node *node, *next;
 888     int error;
 889
 890     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
 891     if (error) {
 892         return error;
 893     }
 894
 895     ofproto->netflow = NULL;
 896     ofproto->sflow = NULL;
 897     ofproto->ipfix = NULL;
 898     ofproto->stp = NULL;
 899     ofproto->dump_seq = 0;
 900     hmap_init(&ofproto->bundles);
 901     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
 902     ofproto->mbridge = mbridge_create();
 903     ofproto->has_bonded_bundles = false;
 904     ofproto->lacp_enabled = false;
 905     ovs_mutex_init(&ofproto->stats_mutex);
 906     ovs_mutex_init(&ofproto->vsp_mutex);
 907
 908     guarded_list_init(&ofproto->pins);
 909
 910     ofproto_dpif_unixctl_init();
 911
 912     hmap_init(&ofproto->vlandev_map);
 913     hmap_init(&ofproto->realdev_vid_map);
 914
 915     sset_init(&ofproto->ports);
 916     sset_init(&ofproto->ghost_ports);
 917     sset_init(&ofproto->port_poll_set);
 918     ofproto->port_poll_errno = 0;
 919     ofproto->change_seq = 0;
 920
 921     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
 922         struct iface_hint *iface_hint = node->data;
 923
 924         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
 925             /* Check if the datapath already has this port. */
 926             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
 927                 sset_add(&ofproto->ports, node->name);
 928             }
 929
 930             free(iface_hint->br_name);
 931             free(iface_hint->br_type);
 932             free(iface_hint);
 933             shash_delete(&init_ofp_ports, node);
 934         }
 935     }
 936
 937     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
 938                 hash_string(ofproto->up.name, 0));
 939     memset(&ofproto->stats, 0, sizeof ofproto->stats);
 940
 941     ofproto_init_tables(ofproto_, N_TABLES);
 942     error = add_internal_flows(ofproto);
 943     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
 944
 945     return error;
 946 }
 947
 948 static int
 949 add_internal_flow(struct ofproto_dpif *ofproto, int id,
 950                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
 951 {
 952     struct ofputil_flow_mod fm;
 953     int error;
 954
 955     match_init_catchall(&fm.match);
 956     fm.priority = 0;
 957     match_set_reg(&fm.match, 0, id);
 958     fm.new_cookie = htonll(0);
 959     fm.cookie = htonll(0);
 960     fm.cookie_mask = htonll(0);
 961     fm.modify_cookie = false;
 962     fm.table_id = TBL_INTERNAL;
 963     fm.command = OFPFC_ADD;
 964     fm.idle_timeout = 0;
 965     fm.hard_timeout = 0;
 966     fm.buffer_id = 0;
 967     fm.out_port = 0;
 968     fm.flags = 0;
 969     fm.ofpacts = ofpacts->data;
 970     fm.ofpacts_len = ofpacts->size;
 971
 972     error = ofproto_flow_mod(&ofproto->up, &fm);
 973     if (error) {
 974         VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
 975                     id, ofperr_to_string(error));
 976         return error;
 977     }
 978
 979     if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
 980                                   rulep)) {
 981         rule_dpif_unref(*rulep);
 982     } else {
 983         OVS_NOT_REACHED();
 984     }
 985
 986     return 0;
 987 }
 988
 989 static int
 990 add_internal_flows(struct ofproto_dpif *ofproto)
 991 {
 992     struct ofpact_controller *controller;
 993     uint64_t ofpacts_stub[128 / 8];
 994     struct ofpbuf ofpacts;
 995     int error;
 996     int id;
 997
 998     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
 999     id = 1;
1000
1001     controller = ofpact_put_CONTROLLER(&ofpacts);
1002     controller->max_len = UINT16_MAX;
1003     controller->controller_id = 0;
1004     controller->reason = OFPR_NO_MATCH;
1005     ofpact_pad(&ofpacts);
1006
1007     error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1008     if (error) {
1009         return error;
1010     }
1011
1012     ofpbuf_clear(&ofpacts);
1013     error = add_internal_flow(ofproto, id++, &ofpacts,
1014                               &ofproto->no_packet_in_rule);
1015     if (error) {
1016         return error;
1017     }
1018
1019     error = add_internal_flow(ofproto, id++, &ofpacts,
1020                               &ofproto->drop_frags_rule);
1021     return error;
1022 }
1023
1024 static void
1025 destruct(struct ofproto *ofproto_)
1026 {
1027     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1028     struct rule_dpif *rule, *next_rule;
1029     struct ofproto_packet_in *pin, *next_pin;
1030     struct oftable *table;
1031     struct list pins;
1032
1033     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1034     ovs_rwlock_wrlock(&xlate_rwlock);
1035     xlate_remove_ofproto(ofproto);
1036     ovs_rwlock_unlock(&xlate_rwlock);
1037
1038     /* Ensure that the upcall processing threads have no remaining references
1039      * to the ofproto or anything in it. */
1040     udpif_synchronize(ofproto->backer->udpif);
1041
1042     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1043
1044     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1045         struct cls_cursor cursor;
1046
1047         fat_rwlock_rdlock(&table->cls.rwlock);
1048         cls_cursor_init(&cursor, &table->cls, NULL);
1049         fat_rwlock_unlock(&table->cls.rwlock);
1050         CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1051             ofproto_rule_delete(&ofproto->up, &rule->up);
1052         }
1053     }
1054
1055     guarded_list_pop_all(&ofproto->pins, &pins);
1056     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1057         list_remove(&pin->list_node);
1058         free(CONST_CAST(void *, pin->up.packet));
1059         free(pin);
1060     }
1061     guarded_list_destroy(&ofproto->pins);
1062
1063     mbridge_unref(ofproto->mbridge);
1064
1065     netflow_unref(ofproto->netflow);
1066     dpif_sflow_unref(ofproto->sflow);
1067     hmap_destroy(&ofproto->bundles);
1068     mac_learning_unref(ofproto->ml);
1069
1070     hmap_destroy(&ofproto->vlandev_map);
1071     hmap_destroy(&ofproto->realdev_vid_map);
1072
1073     sset_destroy(&ofproto->ports);
1074     sset_destroy(&ofproto->ghost_ports);
1075     sset_destroy(&ofproto->port_poll_set);
1076
1077     ovs_mutex_destroy(&ofproto->stats_mutex);
1078     ovs_mutex_destroy(&ofproto->vsp_mutex);
1079
1080     close_dpif_backer(ofproto->backer);
1081 }
1082
1083 static int
1084 run(struct ofproto *ofproto_)
1085 {
1086     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1087     uint64_t new_seq, new_dump_seq;
1088
1089     if (mbridge_need_revalidate(ofproto->mbridge)) {
1090         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1091         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1092         mac_learning_flush(ofproto->ml);
1093         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1094     }
1095
1096     /* Do not perform any periodic activity required by 'ofproto' while
1097      * waiting for flow restore to complete. */
1098     if (!ofproto_get_flow_restore_wait()) {
1099         struct ofproto_packet_in *pin, *next_pin;
1100         struct list pins;
1101
1102         guarded_list_pop_all(&ofproto->pins, &pins);
1103         LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1104             connmgr_send_packet_in(ofproto->up.connmgr, pin);
1105             list_remove(&pin->list_node);
1106             free(CONST_CAST(void *, pin->up.packet));
1107             free(pin);
1108         }
1109     }
1110
1111     if (ofproto->netflow) {
1112         netflow_run(ofproto->netflow);
1113     }
1114     if (ofproto->sflow) {
1115         dpif_sflow_run(ofproto->sflow);
1116     }
1117     if (ofproto->ipfix) {
1118         dpif_ipfix_run(ofproto->ipfix);
1119     }
1120
1121     new_seq = seq_read(connectivity_seq_get());
1122     if (ofproto->change_seq != new_seq) {
1123         struct ofport_dpif *ofport;
1124
1125         HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1126             port_run(ofport);
1127         }
1128
1129         ofproto->change_seq = new_seq;
1130     }
1131     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1132         struct ofbundle *bundle;
1133
1134         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1135             bundle_run(bundle);
1136         }
1137     }
1138
1139     stp_run(ofproto);
1140     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1141     if (mac_learning_run(ofproto->ml)) {
1142         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1143     }
1144     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1145
1146     new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1147     if (ofproto->dump_seq != new_dump_seq) {
1148         struct rule *rule, *next_rule;
1149
1150         /* We know stats are relatively fresh, so now is a good time to do some
1151          * periodic work. */
1152         ofproto->dump_seq = new_dump_seq;
1153
1154         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1155          * has passed. */
1156         ovs_mutex_lock(&ofproto_mutex);
1157         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1158                             &ofproto->up.expirable) {
1159             rule_expire(rule_dpif_cast(rule));
1160         }
1161         ovs_mutex_unlock(&ofproto_mutex);
1162
1163         /* All outstanding data in existing flows has been accounted, so it's a
1164          * good time to do bond rebalancing. */
1165         if (ofproto->has_bonded_bundles) {
1166             struct ofbundle *bundle;
1167
1168             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1169                 if (bundle->bond) {
1170                     bond_rebalance(bundle->bond);
1171                 }
1172             }
1173         }
1174     }
1175
1176     return 0;
1177 }
1178
1179 static void
1180 wait(struct ofproto *ofproto_)
1181 {
1182     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1183
1184     if (ofproto_get_flow_restore_wait()) {
1185         return;
1186     }
1187
1188     if (ofproto->sflow) {
1189         dpif_sflow_wait(ofproto->sflow);
1190     }
1191     if (ofproto->ipfix) {
1192         dpif_ipfix_wait(ofproto->ipfix);
1193     }
1194     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1195         struct ofbundle *bundle;
1196
1197         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1198             bundle_wait(bundle);
1199         }
1200     }
1201     if (ofproto->netflow) {
1202         netflow_wait(ofproto->netflow);
1203     }
1204     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1205     mac_learning_wait(ofproto->ml);
1206     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1207     stp_wait(ofproto);
1208     if (ofproto->backer->need_revalidate) {
1209         /* Shouldn't happen, but if it does just go around again. */
1210         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1211         poll_immediate_wake();
1212     }
1213
1214     seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1215 }
1216
1217 static void
1218 type_get_memory_usage(const char *type, struct simap *usage)
1219 {
1220     struct dpif_backer *backer;
1221
1222     backer = shash_find_data(&all_dpif_backers, type);
1223     if (backer) {
1224         udpif_get_memory_usage(backer->udpif, usage);
1225     }
1226 }
1227
1228 static void
1229 flush(struct ofproto *ofproto OVS_UNUSED)
1230 {
1231     udpif_flush();
1232 }
1233
1234 static void
1235 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1236              bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1237 {
1238     *arp_match_ip = true;
1239     *actions = (OFPUTIL_A_OUTPUT |
1240                 OFPUTIL_A_SET_VLAN_VID |
1241                 OFPUTIL_A_SET_VLAN_PCP |
1242                 OFPUTIL_A_STRIP_VLAN |
1243                 OFPUTIL_A_SET_DL_SRC |
1244                 OFPUTIL_A_SET_DL_DST |
1245                 OFPUTIL_A_SET_NW_SRC |
1246                 OFPUTIL_A_SET_NW_DST |
1247                 OFPUTIL_A_SET_NW_TOS |
1248                 OFPUTIL_A_SET_TP_SRC |
1249                 OFPUTIL_A_SET_TP_DST |
1250                 OFPUTIL_A_ENQUEUE);
1251 }
1252
1253 static void
1254 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1255 {
1256     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1257     struct dpif_dp_stats s;
1258     uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1259     uint64_t n_lookup;
1260
1261     strcpy(ots->name, "classifier");
1262
1263     dpif_get_dp_stats(ofproto->backer->dpif, &s);
1264     rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1265     rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1266     rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1267
1268     n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1269     ots->lookup_count = htonll(n_lookup);
1270     ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1271 }
1272
1273 static struct ofport *
1274 port_alloc(void)
1275 {
1276     struct ofport_dpif *port = xmalloc(sizeof *port);
1277     return &port->up;
1278 }
1279
1280 static void
1281 port_dealloc(struct ofport *port_)
1282 {
1283     struct ofport_dpif *port = ofport_dpif_cast(port_);
1284     free(port);
1285 }
1286
1287 static int
1288 port_construct(struct ofport *port_)
1289 {
1290     struct ofport_dpif *port = ofport_dpif_cast(port_);
1291     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1292     const struct netdev *netdev = port->up.netdev;
1293     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1294     struct dpif_port dpif_port;
1295     int error;
1296
1297     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1298     port->bundle = NULL;
1299     port->cfm = NULL;
1300     port->bfd = NULL;
1301     port->may_enable = true;
1302     port->stp_port = NULL;
1303     port->stp_state = STP_DISABLED;
1304     port->is_tunnel = false;
1305     port->peer = NULL;
1306     port->qdscp = NULL;
1307     port->n_qdscp = 0;
1308     port->realdev_ofp_port = 0;
1309     port->vlandev_vid = 0;
1310     port->carrier_seq = netdev_get_carrier_resets(netdev);
1311     port->is_layer3 = netdev_vport_is_layer3(netdev);
1312
1313     if (netdev_vport_is_patch(netdev)) {
1314         /* By bailing out here, we don't submit the port to the sFlow module
1315          * to be considered for counter polling export.  This is correct
1316          * because the patch port represents an interface that sFlow considers
1317          * to be "internal" to the switch as a whole, and therefore not an
1318          * candidate for counter polling. */
1319         port->odp_port = ODPP_NONE;
1320         ofport_update_peer(port);
1321         return 0;
1322     }
1323
1324     error = dpif_port_query_by_name(ofproto->backer->dpif,
1325                                     netdev_vport_get_dpif_port(netdev, namebuf,
1326                                                                sizeof namebuf),
1327                                     &dpif_port);
1328     if (error) {
1329         return error;
1330     }
1331
1332     port->odp_port = dpif_port.port_no;
1333
1334     if (netdev_get_tunnel_config(netdev)) {
1335         tnl_port_add(port, port->up.netdev, port->odp_port);
1336         port->is_tunnel = true;
1337     } else {
1338         /* Sanity-check that a mapping doesn't already exist.  This
1339          * shouldn't happen for non-tunnel ports. */
1340         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1341             VLOG_ERR("port %s already has an OpenFlow port number",
1342                      dpif_port.name);
1343             dpif_port_destroy(&dpif_port);
1344             return EBUSY;
1345         }
1346
1347         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1348         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1349                     hash_odp_port(port->odp_port));
1350         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1351     }
1352     dpif_port_destroy(&dpif_port);
1353
1354     if (ofproto->sflow) {
1355         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1356     }
1357
1358     return 0;
1359 }
1360
1361 static void
1362 port_destruct(struct ofport *port_)
1363 {
1364     struct ofport_dpif *port = ofport_dpif_cast(port_);
1365     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1366     const char *devname = netdev_get_name(port->up.netdev);
1367     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1368     const char *dp_port_name;
1369
1370     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1371     ovs_rwlock_wrlock(&xlate_rwlock);
1372     xlate_ofport_remove(port);
1373     ovs_rwlock_unlock(&xlate_rwlock);
1374
1375     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1376                                               sizeof namebuf);
1377     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1378         /* The underlying device is still there, so delete it.  This
1379          * happens when the ofproto is being destroyed, since the caller
1380          * assumes that removal of attached ports will happen as part of
1381          * destruction. */
1382         if (!port->is_tunnel) {
1383             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1384         }
1385     }
1386
1387     if (port->peer) {
1388         port->peer->peer = NULL;
1389         port->peer = NULL;
1390     }
1391
1392     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1393         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1394         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1395         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1396     }
1397
1398     tnl_port_del(port);
1399     sset_find_and_delete(&ofproto->ports, devname);
1400     sset_find_and_delete(&ofproto->ghost_ports, devname);
1401     bundle_remove(port_);
1402     set_cfm(port_, NULL);
1403     set_bfd(port_, NULL);
1404     if (port->stp_port) {
1405         stp_port_disable(port->stp_port);
1406     }
1407     if (ofproto->sflow) {
1408         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1409     }
1410
1411     free(port->qdscp);
1412 }
1413
1414 static void
1415 port_modified(struct ofport *port_)
1416 {
1417     struct ofport_dpif *port = ofport_dpif_cast(port_);
1418
1419     if (port->bundle && port->bundle->bond) {
1420         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1421     }
1422
1423     if (port->cfm) {
1424         cfm_set_netdev(port->cfm, port->up.netdev);
1425     }
1426
1427     if (port->bfd) {
1428         bfd_set_netdev(port->bfd, port->up.netdev);
1429     }
1430
1431     ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1432                                      port->up.pp.hw_addr);
1433
1434     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1435                                                 port->odp_port)) {
1436         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1437             REV_RECONFIGURE;
1438     }
1439
1440     ofport_update_peer(port);
1441 }
1442
1443 static void
1444 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1445 {
1446     struct ofport_dpif *port = ofport_dpif_cast(port_);
1447     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1448     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1449
1450     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1451                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1452                    OFPUTIL_PC_NO_PACKET_IN)) {
1453         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1454
1455         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1456             bundle_update(port->bundle);
1457         }
1458     }
1459 }
1460
1461 static int
1462 set_sflow(struct ofproto *ofproto_,
1463           const struct ofproto_sflow_options *sflow_options)
1464 {
1465     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1466     struct dpif_sflow *ds = ofproto->sflow;
1467
1468     if (sflow_options) {
1469         if (!ds) {
1470             struct ofport_dpif *ofport;
1471
1472             ds = ofproto->sflow = dpif_sflow_create();
1473             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1474                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1475             }
1476             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1477         }
1478         dpif_sflow_set_options(ds, sflow_options);
1479     } else {
1480         if (ds) {
1481             dpif_sflow_unref(ds);
1482             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1483             ofproto->sflow = NULL;
1484         }
1485     }
1486     return 0;
1487 }
1488
1489 static int
1490 set_ipfix(
1491     struct ofproto *ofproto_,
1492     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1493     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1494     size_t n_flow_exporters_options)
1495 {
1496     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1497     struct dpif_ipfix *di = ofproto->ipfix;
1498     bool has_options = bridge_exporter_options || flow_exporters_options;
1499
1500     if (has_options && !di) {
1501         di = ofproto->ipfix = dpif_ipfix_create();
1502     }
1503
1504     if (di) {
1505         /* Call set_options in any case to cleanly flush the flow
1506          * caches in the last exporters that are to be destroyed. */
1507         dpif_ipfix_set_options(
1508             di, bridge_exporter_options, flow_exporters_options,
1509             n_flow_exporters_options);
1510
1511         if (!has_options) {
1512             dpif_ipfix_unref(di);
1513             ofproto->ipfix = NULL;
1514         }
1515     }
1516
1517     return 0;
1518 }
1519
1520 static int
1521 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1522 {
1523     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1524     int error = 0;
1525
1526     if (s) {
1527         if (!ofport->cfm) {
1528             struct ofproto_dpif *ofproto;
1529
1530             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1531             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1532             ofport->cfm = cfm_create(ofport->up.netdev);
1533         }
1534
1535         if (cfm_configure(ofport->cfm, s)) {
1536             error = 0;
1537             goto out;
1538         }
1539
1540         error = EINVAL;
1541     }
1542     cfm_unref(ofport->cfm);
1543     ofport->cfm = NULL;
1544 out:
1545     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1546                                      ofport->up.pp.hw_addr);
1547     return error;
1548 }
1549
1550 static bool
1551 get_cfm_status(const struct ofport *ofport_,
1552                struct ofproto_cfm_status *status)
1553 {
1554     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1555
1556     if (ofport->cfm) {
1557         status->faults = cfm_get_fault(ofport->cfm);
1558         status->flap_count = cfm_get_flap_count(ofport->cfm);
1559         status->remote_opstate = cfm_get_opup(ofport->cfm);
1560         status->health = cfm_get_health(ofport->cfm);
1561         cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1562         return true;
1563     } else {
1564         return false;
1565     }
1566 }
1567
1568 static int
1569 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1570 {
1571     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1572     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1573     struct bfd *old;
1574
1575     old = ofport->bfd;
1576     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1577                                 cfg, ofport->up.netdev);
1578     if (ofport->bfd != old) {
1579         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1580     }
1581     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1582                                      ofport->up.pp.hw_addr);
1583     return 0;
1584 }
1585
1586 static int
1587 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1588 {
1589     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1590
1591     if (ofport->bfd) {
1592         bfd_get_status(ofport->bfd, smap);
1593         return 0;
1594     } else {
1595         return ENOENT;
1596     }
1597 }
1598 \f
1599 /* Spanning Tree. */
1600
1601 static void
1602 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1603 {
1604     struct ofproto_dpif *ofproto = ofproto_;
1605     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1606     struct ofport_dpif *ofport;
1607
1608     ofport = stp_port_get_aux(sp);
1609     if (!ofport) {
1610         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1611                      ofproto->up.name, port_num);
1612     } else {
1613         struct eth_header *eth = pkt->l2;
1614
1615         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1616         if (eth_addr_is_zero(eth->eth_src)) {
1617             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1618                          "with unknown MAC", ofproto->up.name, port_num);
1619         } else {
1620             ofproto_dpif_send_packet(ofport, pkt);
1621         }
1622     }
1623     ofpbuf_delete(pkt);
1624 }
1625
1626 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1627 static int
1628 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1629 {
1630     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1631
1632     /* Only revalidate flows if the configuration changed. */
1633     if (!s != !ofproto->stp) {
1634         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1635     }
1636
1637     if (s) {
1638         if (!ofproto->stp) {
1639             ofproto->stp = stp_create(ofproto_->name, s->system_id,
1640                                       send_bpdu_cb, ofproto);
1641             ofproto->stp_last_tick = time_msec();
1642         }
1643
1644         stp_set_bridge_id(ofproto->stp, s->system_id);
1645         stp_set_bridge_priority(ofproto->stp, s->priority);
1646         stp_set_hello_time(ofproto->stp, s->hello_time);
1647         stp_set_max_age(ofproto->stp, s->max_age);
1648         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1649     }  else {
1650         struct ofport *ofport;
1651
1652         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1653             set_stp_port(ofport, NULL);
1654         }
1655
1656         stp_unref(ofproto->stp);
1657         ofproto->stp = NULL;
1658     }
1659
1660     return 0;
1661 }
1662
1663 static int
1664 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1665 {
1666     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1667
1668     if (ofproto->stp) {
1669         s->enabled = true;
1670         s->bridge_id = stp_get_bridge_id(ofproto->stp);
1671         s->designated_root = stp_get_designated_root(ofproto->stp);
1672         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1673     } else {
1674         s->enabled = false;
1675     }
1676
1677     return 0;
1678 }
1679
1680 static void
1681 update_stp_port_state(struct ofport_dpif *ofport)
1682 {
1683     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1684     enum stp_state state;
1685
1686     /* Figure out new state. */
1687     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1688                              : STP_DISABLED;
1689
1690     /* Update state. */
1691     if (ofport->stp_state != state) {
1692         enum ofputil_port_state of_state;
1693         bool fwd_change;
1694
1695         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1696                     netdev_get_name(ofport->up.netdev),
1697                     stp_state_name(ofport->stp_state),
1698                     stp_state_name(state));
1699         if (stp_learn_in_state(ofport->stp_state)
1700                 != stp_learn_in_state(state)) {
1701             /* xxx Learning action flows should also be flushed. */
1702             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1703             mac_learning_flush(ofproto->ml);
1704             ovs_rwlock_unlock(&ofproto->ml->rwlock);
1705         }
1706         fwd_change = stp_forward_in_state(ofport->stp_state)
1707                         != stp_forward_in_state(state);
1708
1709         ofproto->backer->need_revalidate = REV_STP;
1710         ofport->stp_state = state;
1711         ofport->stp_state_entered = time_msec();
1712
1713         if (fwd_change && ofport->bundle) {
1714             bundle_update(ofport->bundle);
1715         }
1716
1717         /* Update the STP state bits in the OpenFlow port description. */
1718         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1719         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1720                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1721                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1722                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
1723                      : 0);
1724         ofproto_port_set_state(&ofport->up, of_state);
1725     }
1726 }
1727
1728 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
1729  * caller is responsible for assigning STP port numbers and ensuring
1730  * there are no duplicates. */
1731 static int
1732 set_stp_port(struct ofport *ofport_,
1733              const struct ofproto_port_stp_settings *s)
1734 {
1735     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1736     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1737     struct stp_port *sp = ofport->stp_port;
1738
1739     if (!s || !s->enable) {
1740         if (sp) {
1741             ofport->stp_port = NULL;
1742             stp_port_disable(sp);
1743             update_stp_port_state(ofport);
1744         }
1745         return 0;
1746     } else if (sp && stp_port_no(sp) != s->port_num
1747             && ofport == stp_port_get_aux(sp)) {
1748         /* The port-id changed, so disable the old one if it's not
1749          * already in use by another port. */
1750         stp_port_disable(sp);
1751     }
1752
1753     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1754     stp_port_enable(sp);
1755
1756     stp_port_set_aux(sp, ofport);
1757     stp_port_set_priority(sp, s->priority);
1758     stp_port_set_path_cost(sp, s->path_cost);
1759
1760     update_stp_port_state(ofport);
1761
1762     return 0;
1763 }
1764
1765 static int
1766 get_stp_port_status(struct ofport *ofport_,
1767                     struct ofproto_port_stp_status *s)
1768 {
1769     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1770     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1771     struct stp_port *sp = ofport->stp_port;
1772
1773     if (!ofproto->stp || !sp) {
1774         s->enabled = false;
1775         return 0;
1776     }
1777
1778     s->enabled = true;
1779     s->port_id = stp_port_get_id(sp);
1780     s->state = stp_port_get_state(sp);
1781     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
1782     s->role = stp_port_get_role(sp);
1783
1784     return 0;
1785 }
1786
1787 static int
1788 get_stp_port_stats(struct ofport *ofport_,
1789                    struct ofproto_port_stp_stats *s)
1790 {
1791     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1792     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1793     struct stp_port *sp = ofport->stp_port;
1794
1795     if (!ofproto->stp || !sp) {
1796         s->enabled = false;
1797         return 0;
1798     }
1799
1800     s->enabled = true;
1801     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
1802
1803     return 0;
1804 }
1805
1806 static void
1807 stp_run(struct ofproto_dpif *ofproto)
1808 {
1809     if (ofproto->stp) {
1810         long long int now = time_msec();
1811         long long int elapsed = now - ofproto->stp_last_tick;
1812         struct stp_port *sp;
1813
1814         if (elapsed > 0) {
1815             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
1816             ofproto->stp_last_tick = now;
1817         }
1818         while (stp_get_changed_port(ofproto->stp, &sp)) {
1819             struct ofport_dpif *ofport = stp_port_get_aux(sp);
1820
1821             if (ofport) {
1822                 update_stp_port_state(ofport);
1823             }
1824         }
1825
1826         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
1827             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1828             mac_learning_flush(ofproto->ml);
1829             ovs_rwlock_unlock(&ofproto->ml->rwlock);
1830         }
1831     }
1832 }
1833
1834 static void
1835 stp_wait(struct ofproto_dpif *ofproto)
1836 {
1837     if (ofproto->stp) {
1838         poll_timer_wait(1000);
1839     }
1840 }
1841 \f
1842 static int
1843 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
1844            size_t n_qdscp)
1845 {
1846     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1847     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1848
1849     if (ofport->n_qdscp != n_qdscp
1850         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
1851                               n_qdscp * sizeof *qdscp))) {
1852         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1853         free(ofport->qdscp);
1854         ofport->qdscp = n_qdscp
1855             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
1856             : NULL;
1857         ofport->n_qdscp = n_qdscp;
1858     }
1859
1860     return 0;
1861 }
1862 \f
1863 /* Bundles. */
1864
1865 /* Expires all MAC learning entries associated with 'bundle' and forces its
1866  * ofproto to revalidate every flow.
1867  *
1868  * Normally MAC learning entries are removed only from the ofproto associated
1869  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
1870  * are removed from every ofproto.  When patch ports and SLB bonds are in use
1871  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
1872  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
1873  * with the host from which it migrated. */
1874 static void
1875 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
1876 {
1877     struct ofproto_dpif *ofproto = bundle->ofproto;
1878     struct mac_learning *ml = ofproto->ml;
1879     struct mac_entry *mac, *next_mac;
1880
1881     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1882     ovs_rwlock_wrlock(&ml->rwlock);
1883     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
1884         if (mac->port.p == bundle) {
1885             if (all_ofprotos) {
1886                 struct ofproto_dpif *o;
1887
1888                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1889                     if (o != ofproto) {
1890                         struct mac_entry *e;
1891
1892                         ovs_rwlock_wrlock(&o->ml->rwlock);
1893                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
1894                         if (e) {
1895                             mac_learning_expire(o->ml, e);
1896                         }
1897                         ovs_rwlock_unlock(&o->ml->rwlock);
1898                     }
1899                 }
1900             }
1901
1902             mac_learning_expire(ml, mac);
1903         }
1904     }
1905     ovs_rwlock_unlock(&ml->rwlock);
1906 }
1907
1908 static struct ofbundle *
1909 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
1910 {
1911     struct ofbundle *bundle;
1912
1913     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
1914                              &ofproto->bundles) {
1915         if (bundle->aux == aux) {
1916             return bundle;
1917         }
1918     }
1919     return NULL;
1920 }
1921
1922 static void
1923 bundle_update(struct ofbundle *bundle)
1924 {
1925     struct ofport_dpif *port;
1926
1927     bundle->floodable = true;
1928     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
1929         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1930             || port->is_layer3
1931             || !stp_forward_in_state(port->stp_state)) {
1932             bundle->floodable = false;
1933             break;
1934         }
1935     }
1936 }
1937
1938 static void
1939 bundle_del_port(struct ofport_dpif *port)
1940 {
1941     struct ofbundle *bundle = port->bundle;
1942
1943     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
1944
1945     list_remove(&port->bundle_node);
1946     port->bundle = NULL;
1947
1948     if (bundle->lacp) {
1949         lacp_slave_unregister(bundle->lacp, port);
1950     }
1951     if (bundle->bond) {
1952         bond_slave_unregister(bundle->bond, port);
1953     }
1954
1955     bundle_update(bundle);
1956 }
1957
1958 static bool
1959 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
1960                 struct lacp_slave_settings *lacp)
1961 {
1962     struct ofport_dpif *port;
1963
1964     port = get_ofp_port(bundle->ofproto, ofp_port);
1965     if (!port) {
1966         return false;
1967     }
1968
1969     if (port->bundle != bundle) {
1970         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
1971         if (port->bundle) {
1972             bundle_remove(&port->up);
1973         }
1974
1975         port->bundle = bundle;
1976         list_push_back(&bundle->ports, &port->bundle_node);
1977         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
1978             || port->is_layer3
1979             || !stp_forward_in_state(port->stp_state)) {
1980             bundle->floodable = false;
1981         }
1982     }
1983     if (lacp) {
1984         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
1985         lacp_slave_register(bundle->lacp, port, lacp);
1986     }
1987
1988     return true;
1989 }
1990
1991 static void
1992 bundle_destroy(struct ofbundle *bundle)
1993 {
1994     struct ofproto_dpif *ofproto;
1995     struct ofport_dpif *port, *next_port;
1996
1997     if (!bundle) {
1998         return;
1999     }
2000
2001     ofproto = bundle->ofproto;
2002     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2003
2004     ovs_rwlock_wrlock(&xlate_rwlock);
2005     xlate_bundle_remove(bundle);
2006     ovs_rwlock_unlock(&xlate_rwlock);
2007
2008     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2009         bundle_del_port(port);
2010     }
2011
2012     bundle_flush_macs(bundle, true);
2013     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2014     free(bundle->name);
2015     free(bundle->trunks);
2016     lacp_unref(bundle->lacp);
2017     bond_unref(bundle->bond);
2018     free(bundle);
2019 }
2020
2021 static int
2022 bundle_set(struct ofproto *ofproto_, void *aux,
2023            const struct ofproto_bundle_settings *s)
2024 {
2025     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2026     bool need_flush = false;
2027     struct ofport_dpif *port;
2028     struct ofbundle *bundle;
2029     unsigned long *trunks;
2030     int vlan;
2031     size_t i;
2032     bool ok;
2033
2034     if (!s) {
2035         bundle_destroy(bundle_lookup(ofproto, aux));
2036         return 0;
2037     }
2038
2039     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2040     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2041
2042     bundle = bundle_lookup(ofproto, aux);
2043     if (!bundle) {
2044         bundle = xmalloc(sizeof *bundle);
2045
2046         bundle->ofproto = ofproto;
2047         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2048                     hash_pointer(aux, 0));
2049         bundle->aux = aux;
2050         bundle->name = NULL;
2051
2052         list_init(&bundle->ports);
2053         bundle->vlan_mode = PORT_VLAN_TRUNK;
2054         bundle->vlan = -1;
2055         bundle->trunks = NULL;
2056         bundle->use_priority_tags = s->use_priority_tags;
2057         bundle->lacp = NULL;
2058         bundle->bond = NULL;
2059
2060         bundle->floodable = true;
2061         mbridge_register_bundle(ofproto->mbridge, bundle);
2062     }
2063
2064     if (!bundle->name || strcmp(s->name, bundle->name)) {
2065         free(bundle->name);
2066         bundle->name = xstrdup(s->name);
2067     }
2068
2069     /* LACP. */
2070     if (s->lacp) {
2071         ofproto->lacp_enabled = true;
2072         if (!bundle->lacp) {
2073             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2074             bundle->lacp = lacp_create();
2075         }
2076         lacp_configure(bundle->lacp, s->lacp);
2077     } else {
2078         lacp_unref(bundle->lacp);
2079         bundle->lacp = NULL;
2080     }
2081
2082     /* Update set of ports. */
2083     ok = true;
2084     for (i = 0; i < s->n_slaves; i++) {
2085         if (!bundle_add_port(bundle, s->slaves[i],
2086                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2087             ok = false;
2088         }
2089     }
2090     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2091         struct ofport_dpif *next_port;
2092
2093         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2094             for (i = 0; i < s->n_slaves; i++) {
2095                 if (s->slaves[i] == port->up.ofp_port) {
2096                     goto found;
2097                 }
2098             }
2099
2100             bundle_del_port(port);
2101         found: ;
2102         }
2103     }
2104     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2105
2106     if (list_is_empty(&bundle->ports)) {
2107         bundle_destroy(bundle);
2108         return EINVAL;
2109     }
2110
2111     /* Set VLAN tagging mode */
2112     if (s->vlan_mode != bundle->vlan_mode
2113         || s->use_priority_tags != bundle->use_priority_tags) {
2114         bundle->vlan_mode = s->vlan_mode;
2115         bundle->use_priority_tags = s->use_priority_tags;
2116         need_flush = true;
2117     }
2118
2119     /* Set VLAN tag. */
2120     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2121             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2122             : 0);
2123     if (vlan != bundle->vlan) {
2124         bundle->vlan = vlan;
2125         need_flush = true;
2126     }
2127
2128     /* Get trunked VLANs. */
2129     switch (s->vlan_mode) {
2130     case PORT_VLAN_ACCESS:
2131         trunks = NULL;
2132         break;
2133
2134     case PORT_VLAN_TRUNK:
2135         trunks = CONST_CAST(unsigned long *, s->trunks);
2136         break;
2137
2138     case PORT_VLAN_NATIVE_UNTAGGED:
2139     case PORT_VLAN_NATIVE_TAGGED:
2140         if (vlan != 0 && (!s->trunks
2141                           || !bitmap_is_set(s->trunks, vlan)
2142                           || bitmap_is_set(s->trunks, 0))) {
2143             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2144             if (s->trunks) {
2145                 trunks = bitmap_clone(s->trunks, 4096);
2146             } else {
2147                 trunks = bitmap_allocate1(4096);
2148             }
2149             bitmap_set1(trunks, vlan);
2150             bitmap_set0(trunks, 0);
2151         } else {
2152             trunks = CONST_CAST(unsigned long *, s->trunks);
2153         }
2154         break;
2155
2156     default:
2157         OVS_NOT_REACHED();
2158     }
2159     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2160         free(bundle->trunks);
2161         if (trunks == s->trunks) {
2162             bundle->trunks = vlan_bitmap_clone(trunks);
2163         } else {
2164             bundle->trunks = trunks;
2165             trunks = NULL;
2166         }
2167         need_flush = true;
2168     }
2169     if (trunks != s->trunks) {
2170         free(trunks);
2171     }
2172
2173     /* Bonding. */
2174     if (!list_is_short(&bundle->ports)) {
2175         bundle->ofproto->has_bonded_bundles = true;
2176         if (bundle->bond) {
2177             if (bond_reconfigure(bundle->bond, s->bond)) {
2178                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2179             }
2180         } else {
2181             bundle->bond = bond_create(s->bond);
2182             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2183         }
2184
2185         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2186             bond_slave_register(bundle->bond, port, port->up.netdev);
2187         }
2188     } else {
2189         bond_unref(bundle->bond);
2190         bundle->bond = NULL;
2191     }
2192
2193     /* If we changed something that would affect MAC learning, un-learn
2194      * everything on this port and force flow revalidation. */
2195     if (need_flush) {
2196         bundle_flush_macs(bundle, false);
2197     }
2198
2199     return 0;
2200 }
2201
2202 static void
2203 bundle_remove(struct ofport *port_)
2204 {
2205     struct ofport_dpif *port = ofport_dpif_cast(port_);
2206     struct ofbundle *bundle = port->bundle;
2207
2208     if (bundle) {
2209         bundle_del_port(port);
2210         if (list_is_empty(&bundle->ports)) {
2211             bundle_destroy(bundle);
2212         } else if (list_is_short(&bundle->ports)) {
2213             bond_unref(bundle->bond);
2214             bundle->bond = NULL;
2215         }
2216     }
2217 }
2218
2219 static void
2220 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2221 {
2222     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2223     struct ofport_dpif *port = port_;
2224     uint8_t ea[ETH_ADDR_LEN];
2225     int error;
2226
2227     error = netdev_get_etheraddr(port->up.netdev, ea);
2228     if (!error) {
2229         struct ofpbuf packet;
2230         void *packet_pdu;
2231
2232         ofpbuf_init(&packet, 0);
2233         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2234                                  pdu_size);
2235         memcpy(packet_pdu, pdu, pdu_size);
2236
2237         ofproto_dpif_send_packet(port, &packet);
2238         ofpbuf_uninit(&packet);
2239     } else {
2240         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2241                     "%s (%s)", port->bundle->name,
2242                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2243     }
2244 }
2245
2246 static void
2247 bundle_send_learning_packets(struct ofbundle *bundle)
2248 {
2249     struct ofproto_dpif *ofproto = bundle->ofproto;
2250     struct ofpbuf *learning_packet;
2251     int error, n_packets, n_errors;
2252     struct mac_entry *e;
2253     struct list packets;
2254
2255     list_init(&packets);
2256     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2257     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2258         if (e->port.p != bundle) {
2259             void *port_void;
2260
2261             learning_packet = bond_compose_learning_packet(bundle->bond,
2262                                                            e->mac, e->vlan,
2263                                                            &port_void);
2264             learning_packet->private_p = port_void;
2265             list_push_back(&packets, &learning_packet->list_node);
2266         }
2267     }
2268     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2269
2270     error = n_packets = n_errors = 0;
2271     LIST_FOR_EACH (learning_packet, list_node, &packets) {
2272         int ret;
2273
2274         ret = ofproto_dpif_send_packet(learning_packet->private_p, learning_packet);
2275         if (ret) {
2276             error = ret;
2277             n_errors++;
2278         }
2279         n_packets++;
2280     }
2281     ofpbuf_list_delete(&packets);
2282
2283     if (n_errors) {
2284         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2285         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2286                      "packets, last error was: %s",
2287                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2288     } else {
2289         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2290                  bundle->name, n_packets);
2291     }
2292 }
2293
2294 static void
2295 bundle_run(struct ofbundle *bundle)
2296 {
2297     if (bundle->lacp) {
2298         lacp_run(bundle->lacp, send_pdu_cb);
2299     }
2300     if (bundle->bond) {
2301         struct ofport_dpif *port;
2302
2303         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2304             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2305         }
2306
2307         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2308             bundle->ofproto->backer->need_revalidate = REV_BOND;
2309         }
2310
2311         if (bond_should_send_learning_packets(bundle->bond)) {
2312             bundle_send_learning_packets(bundle);
2313         }
2314     }
2315 }
2316
2317 static void
2318 bundle_wait(struct ofbundle *bundle)
2319 {
2320     if (bundle->lacp) {
2321         lacp_wait(bundle->lacp);
2322     }
2323     if (bundle->bond) {
2324         bond_wait(bundle->bond);
2325     }
2326 }
2327 \f
2328 /* Mirrors. */
2329
2330 static int
2331 mirror_set__(struct ofproto *ofproto_, void *aux,
2332              const struct ofproto_mirror_settings *s)
2333 {
2334     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2335     struct ofbundle **srcs, **dsts;
2336     int error;
2337     size_t i;
2338
2339     if (!s) {
2340         mirror_destroy(ofproto->mbridge, aux);
2341         return 0;
2342     }
2343
2344     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2345     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2346
2347     for (i = 0; i < s->n_srcs; i++) {
2348         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2349     }
2350
2351     for (i = 0; i < s->n_dsts; i++) {
2352         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2353     }
2354
2355     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2356                        s->n_dsts, s->src_vlans,
2357                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2358     free(srcs);
2359     free(dsts);
2360     return error;
2361 }
2362
2363 static int
2364 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2365                    uint64_t *packets, uint64_t *bytes)
2366 {
2367     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2368                             bytes);
2369 }
2370
2371 static int
2372 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2373 {
2374     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2375     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2376     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2377         mac_learning_flush(ofproto->ml);
2378     }
2379     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2380     return 0;
2381 }
2382
2383 static bool
2384 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2385 {
2386     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2387     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2388     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2389 }
2390
2391 static void
2392 forward_bpdu_changed(struct ofproto *ofproto_)
2393 {
2394     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2395     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2396 }
2397
2398 static void
2399 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2400                      size_t max_entries)
2401 {
2402     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2403     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2404     mac_learning_set_idle_time(ofproto->ml, idle_time);
2405     mac_learning_set_max_entries(ofproto->ml, max_entries);
2406     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2407 }
2408 \f
2409 /* Ports. */
2410
2411 static struct ofport_dpif *
2412 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2413 {
2414     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2415     return ofport ? ofport_dpif_cast(ofport) : NULL;
2416 }
2417
2418 static void
2419 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2420                             struct ofproto_port *ofproto_port,
2421                             struct dpif_port *dpif_port)
2422 {
2423     ofproto_port->name = dpif_port->name;
2424     ofproto_port->type = dpif_port->type;
2425     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2426 }
2427
2428 static void
2429 ofport_update_peer(struct ofport_dpif *ofport)
2430 {
2431     const struct ofproto_dpif *ofproto;
2432     struct dpif_backer *backer;
2433     char *peer_name;
2434
2435     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2436         return;
2437     }
2438
2439     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2440     backer->need_revalidate = REV_RECONFIGURE;
2441
2442     if (ofport->peer) {
2443         ofport->peer->peer = NULL;
2444         ofport->peer = NULL;
2445     }
2446
2447     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2448     if (!peer_name) {
2449         return;
2450     }
2451
2452     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2453         struct ofport *peer_ofport;
2454         struct ofport_dpif *peer;
2455         char *peer_peer;
2456
2457         if (ofproto->backer != backer) {
2458             continue;
2459         }
2460
2461         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2462         if (!peer_ofport) {
2463             continue;
2464         }
2465
2466         peer = ofport_dpif_cast(peer_ofport);
2467         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2468         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2469                                  peer_peer)) {
2470             ofport->peer = peer;
2471             ofport->peer->peer = ofport;
2472         }
2473         free(peer_peer);
2474
2475         break;
2476     }
2477     free(peer_name);
2478 }
2479
2480 static void
2481 port_run(struct ofport_dpif *ofport)
2482 {
2483     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2484     bool carrier_changed = carrier_seq != ofport->carrier_seq;
2485     bool enable = netdev_get_carrier(ofport->up.netdev);
2486     bool cfm_enable = false;
2487     bool bfd_enable = false;
2488
2489     ofport->carrier_seq = carrier_seq;
2490
2491     if (ofport->cfm) {
2492         int cfm_opup = cfm_get_opup(ofport->cfm);
2493
2494         cfm_enable = !cfm_get_fault(ofport->cfm);
2495
2496         if (cfm_opup >= 0) {
2497             cfm_enable = cfm_enable && cfm_opup;
2498         }
2499     }
2500
2501     if (ofport->bfd) {
2502         bfd_enable = bfd_forwarding(ofport->bfd);
2503     }
2504
2505     if (ofport->bfd || ofport->cfm) {
2506         enable = enable && (cfm_enable || bfd_enable);
2507     }
2508
2509     if (ofport->bundle) {
2510         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2511         if (carrier_changed) {
2512             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2513         }
2514     }
2515
2516     if (ofport->may_enable != enable) {
2517         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2518         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2519     }
2520
2521     ofport->may_enable = enable;
2522 }
2523
2524 static int
2525 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2526                    struct ofproto_port *ofproto_port)
2527 {
2528     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2529     struct dpif_port dpif_port;
2530     int error;
2531
2532     if (sset_contains(&ofproto->ghost_ports, devname)) {
2533         const char *type = netdev_get_type_from_name(devname);
2534
2535         /* We may be called before ofproto->up.port_by_name is populated with
2536          * the appropriate ofport.  For this reason, we must get the name and
2537          * type from the netdev layer directly. */
2538         if (type) {
2539             const struct ofport *ofport;
2540
2541             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2542             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2543             ofproto_port->name = xstrdup(devname);
2544             ofproto_port->type = xstrdup(type);
2545             return 0;
2546         }
2547         return ENODEV;
2548     }
2549
2550     if (!sset_contains(&ofproto->ports, devname)) {
2551         return ENODEV;
2552     }
2553     error = dpif_port_query_by_name(ofproto->backer->dpif,
2554                                     devname, &dpif_port);
2555     if (!error) {
2556         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2557     }
2558     return error;
2559 }
2560
2561 static int
2562 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2563 {
2564     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2565     const char *devname = netdev_get_name(netdev);
2566     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2567     const char *dp_port_name;
2568
2569     if (netdev_vport_is_patch(netdev)) {
2570         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2571         return 0;
2572     }
2573
2574     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2575     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2576         odp_port_t port_no = ODPP_NONE;
2577         int error;
2578
2579         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2580         if (error) {
2581             return error;
2582         }
2583         if (netdev_get_tunnel_config(netdev)) {
2584             simap_put(&ofproto->backer->tnl_backers,
2585                       dp_port_name, odp_to_u32(port_no));
2586         }
2587     }
2588
2589     if (netdev_get_tunnel_config(netdev)) {
2590         sset_add(&ofproto->ghost_ports, devname);
2591     } else {
2592         sset_add(&ofproto->ports, devname);
2593     }
2594     return 0;
2595 }
2596
2597 static int
2598 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
2599 {
2600     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2601     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2602     int error = 0;
2603
2604     if (!ofport) {
2605         return 0;
2606     }
2607
2608     sset_find_and_delete(&ofproto->ghost_ports,
2609                          netdev_get_name(ofport->up.netdev));
2610     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2611     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
2612         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
2613         if (!error) {
2614             /* The caller is going to close ofport->up.netdev.  If this is a
2615              * bonded port, then the bond is using that netdev, so remove it
2616              * from the bond.  The client will need to reconfigure everything
2617              * after deleting ports, so then the slave will get re-added. */
2618             bundle_remove(&ofport->up);
2619         }
2620     }
2621     return error;
2622 }
2623
2624 static int
2625 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2626 {
2627     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2628     int error;
2629
2630     error = netdev_get_stats(ofport->up.netdev, stats);
2631
2632     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
2633         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2634
2635         ovs_mutex_lock(&ofproto->stats_mutex);
2636         /* ofproto->stats.tx_packets represents packets that we created
2637          * internally and sent to some port (e.g. packets sent with
2638          * ofproto_dpif_send_packet()).  Account for them as if they had
2639          * come from OFPP_LOCAL and got forwarded. */
2640
2641         if (stats->rx_packets != UINT64_MAX) {
2642             stats->rx_packets += ofproto->stats.tx_packets;
2643         }
2644
2645         if (stats->rx_bytes != UINT64_MAX) {
2646             stats->rx_bytes += ofproto->stats.tx_bytes;
2647         }
2648
2649         /* ofproto->stats.rx_packets represents packets that were received on
2650          * some port and we processed internally and dropped (e.g. STP).
2651          * Account for them as if they had been forwarded to OFPP_LOCAL. */
2652
2653         if (stats->tx_packets != UINT64_MAX) {
2654             stats->tx_packets += ofproto->stats.rx_packets;
2655         }
2656
2657         if (stats->tx_bytes != UINT64_MAX) {
2658             stats->tx_bytes += ofproto->stats.rx_bytes;
2659         }
2660         ovs_mutex_unlock(&ofproto->stats_mutex);
2661     }
2662
2663     return error;
2664 }
2665
2666 struct port_dump_state {
2667     uint32_t bucket;
2668     uint32_t offset;
2669     bool ghost;
2670
2671     struct ofproto_port port;
2672     bool has_port;
2673 };
2674
2675 static int
2676 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
2677 {
2678     *statep = xzalloc(sizeof(struct port_dump_state));
2679     return 0;
2680 }
2681
2682 static int
2683 port_dump_next(const struct ofproto *ofproto_, void *state_,
2684                struct ofproto_port *port)
2685 {
2686     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2687     struct port_dump_state *state = state_;
2688     const struct sset *sset;
2689     struct sset_node *node;
2690
2691     if (state->has_port) {
2692         ofproto_port_destroy(&state->port);
2693         state->has_port = false;
2694     }
2695     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
2696     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
2697         int error;
2698
2699         error = port_query_by_name(ofproto_, node->name, &state->port);
2700         if (!error) {
2701             *port = state->port;
2702             state->has_port = true;
2703             return 0;
2704         } else if (error != ENODEV) {
2705             return error;
2706         }
2707     }
2708
2709     if (!state->ghost) {
2710         state->ghost = true;
2711         state->bucket = 0;
2712         state->offset = 0;
2713         return port_dump_next(ofproto_, state_, port);
2714     }
2715
2716     return EOF;
2717 }
2718
2719 static int
2720 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2721 {
2722     struct port_dump_state *state = state_;
2723
2724     if (state->has_port) {
2725         ofproto_port_destroy(&state->port);
2726     }
2727     free(state);
2728     return 0;
2729 }
2730
2731 static int
2732 port_poll(const struct ofproto *ofproto_, char **devnamep)
2733 {
2734     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2735
2736     if (ofproto->port_poll_errno) {
2737         int error = ofproto->port_poll_errno;
2738         ofproto->port_poll_errno = 0;
2739         return error;
2740     }
2741
2742     if (sset_is_empty(&ofproto->port_poll_set)) {
2743         return EAGAIN;
2744     }
2745
2746     *devnamep = sset_pop(&ofproto->port_poll_set);
2747     return 0;
2748 }
2749
2750 static void
2751 port_poll_wait(const struct ofproto *ofproto_)
2752 {
2753     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2754     dpif_port_poll_wait(ofproto->backer->dpif);
2755 }
2756
2757 static int
2758 port_is_lacp_current(const struct ofport *ofport_)
2759 {
2760     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2761     return (ofport->bundle && ofport->bundle->lacp
2762             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
2763             : -1);
2764 }
2765 \f
2766 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
2767  * then delete it entirely. */
2768 static void
2769 rule_expire(struct rule_dpif *rule)
2770     OVS_REQUIRES(ofproto_mutex)
2771 {
2772     uint16_t idle_timeout, hard_timeout;
2773     long long int now = time_msec();
2774     int reason;
2775
2776     ovs_assert(!rule->up.pending);
2777
2778     /* Has 'rule' expired? */
2779     ovs_mutex_lock(&rule->up.mutex);
2780     hard_timeout = rule->up.hard_timeout;
2781     idle_timeout = rule->up.idle_timeout;
2782     if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
2783         reason = OFPRR_HARD_TIMEOUT;
2784     } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
2785         reason = OFPRR_IDLE_TIMEOUT;
2786     } else {
2787         reason = -1;
2788     }
2789     ovs_mutex_unlock(&rule->up.mutex);
2790
2791     if (reason >= 0) {
2792         COVERAGE_INC(ofproto_dpif_expired);
2793         ofproto_rule_expire(&rule->up, reason);
2794     }
2795 }
2796
2797 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
2798  * 'flow' must reflect the data in 'packet'. */
2799 int
2800 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
2801                              const struct flow *flow,
2802                              struct rule_dpif *rule,
2803                              const struct ofpact *ofpacts, size_t ofpacts_len,
2804                              struct ofpbuf *packet)
2805 {
2806     struct odputil_keybuf keybuf;
2807     struct dpif_flow_stats stats;
2808     struct xlate_out xout;
2809     struct xlate_in xin;
2810     ofp_port_t in_port;
2811     struct ofpbuf key;
2812     int error;
2813
2814     ovs_assert((rule != NULL) != (ofpacts != NULL));
2815
2816     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
2817     if (rule) {
2818         rule_dpif_credit_stats(rule, &stats);
2819     }
2820
2821     xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
2822     xin.ofpacts = ofpacts;
2823     xin.ofpacts_len = ofpacts_len;
2824     xin.resubmit_stats = &stats;
2825     xlate_actions(&xin, &xout);
2826
2827     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
2828     in_port = flow->in_port.ofp_port;
2829     if (in_port == OFPP_NONE) {
2830         in_port = OFPP_LOCAL;
2831     }
2832     odp_flow_key_from_flow(&key, flow, ofp_port_to_odp_port(ofproto, in_port));
2833
2834     error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
2835                          xout.odp_actions.data, xout.odp_actions.size, packet,
2836                          (xout.slow & SLOW_ACTION) != 0);
2837     xlate_out_uninit(&xout);
2838
2839     return error;
2840 }
2841
2842 void
2843 rule_dpif_credit_stats(struct rule_dpif *rule,
2844                        const struct dpif_flow_stats *stats)
2845 {
2846     ovs_mutex_lock(&rule->stats_mutex);
2847     rule->packet_count += stats->n_packets;
2848     rule->byte_count += stats->n_bytes;
2849     rule->up.used = MAX(rule->up.used, stats->used);
2850     ovs_mutex_unlock(&rule->stats_mutex);
2851 }
2852
2853 bool
2854 rule_dpif_is_fail_open(const struct rule_dpif *rule)
2855 {
2856     return is_fail_open_rule(&rule->up);
2857 }
2858
2859 bool
2860 rule_dpif_is_table_miss(const struct rule_dpif *rule)
2861 {
2862     return rule_is_table_miss(&rule->up);
2863 }
2864
2865 ovs_be64
2866 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
2867     OVS_REQUIRES(rule->up.mutex)
2868 {
2869     return rule->up.flow_cookie;
2870 }
2871
2872 void
2873 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
2874                      uint16_t hard_timeout)
2875 {
2876     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
2877 }
2878
2879 /* Returns 'rule''s actions.  The caller owns a reference on the returned
2880  * actions and must eventually release it (with rule_actions_unref()) to avoid
2881  * a memory leak. */
2882 struct rule_actions *
2883 rule_dpif_get_actions(const struct rule_dpif *rule)
2884 {
2885     return rule_get_actions(&rule->up);
2886 }
2887
2888 /* Lookup 'flow' in 'ofproto''s classifier.  If 'wc' is non-null, sets
2889  * the fields that were relevant as part of the lookup. */
2890 void
2891 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
2892                  struct flow_wildcards *wc, struct rule_dpif **rule)
2893 {
2894     struct ofport_dpif *port;
2895
2896     if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
2897         return;
2898     }
2899     port = get_ofp_port(ofproto, flow->in_port.ofp_port);
2900     if (!port) {
2901         VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
2902                      flow->in_port.ofp_port);
2903     }
2904
2905     choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
2906                      ofproto->no_packet_in_rule, rule);
2907 }
2908
2909 bool
2910 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
2911                           const struct flow *flow, struct flow_wildcards *wc,
2912                           uint8_t table_id, struct rule_dpif **rule)
2913 {
2914     const struct cls_rule *cls_rule;
2915     struct classifier *cls;
2916     bool frag;
2917
2918     *rule = NULL;
2919     if (table_id >= N_TABLES) {
2920         return false;
2921     }
2922
2923     if (wc) {
2924         memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
2925         if (is_ip_any(flow)) {
2926             wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
2927         }
2928     }
2929
2930     cls = &ofproto->up.tables[table_id].cls;
2931     fat_rwlock_rdlock(&cls->rwlock);
2932     frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
2933     if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
2934         /* We must pretend that transport ports are unavailable. */
2935         struct flow ofpc_normal_flow = *flow;
2936         ofpc_normal_flow.tp_src = htons(0);
2937         ofpc_normal_flow.tp_dst = htons(0);
2938         cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
2939     } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
2940         cls_rule = &ofproto->drop_frags_rule->up.cr;
2941         /* Frag mask in wc already set above. */
2942     } else {
2943         cls_rule = classifier_lookup(cls, flow, wc);
2944     }
2945
2946     *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
2947     rule_dpif_ref(*rule);
2948     fat_rwlock_unlock(&cls->rwlock);
2949
2950     return *rule != NULL;
2951 }
2952
2953 /* Given a port configuration (specified as zero if there's no port), chooses
2954  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
2955  * flow table miss. */
2956 void
2957 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
2958                  struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
2959 {
2960     *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
2961     rule_dpif_ref(*rule);
2962 }
2963
2964 void
2965 rule_dpif_ref(struct rule_dpif *rule)
2966 {
2967     if (rule) {
2968         ofproto_rule_ref(&rule->up);
2969     }
2970 }
2971
2972 void
2973 rule_dpif_unref(struct rule_dpif *rule)
2974 {
2975     if (rule) {
2976         ofproto_rule_unref(&rule->up);
2977     }
2978 }
2979
2980 static void
2981 complete_operation(struct rule_dpif *rule)
2982     OVS_REQUIRES(ofproto_mutex)
2983 {
2984     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
2985
2986     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
2987     ofoperation_complete(rule->up.pending, 0);
2988 }
2989
2990 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
2991 {
2992     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
2993 }
2994
2995 static struct rule *
2996 rule_alloc(void)
2997 {
2998     struct rule_dpif *rule = xmalloc(sizeof *rule);
2999     return &rule->up;
3000 }
3001
3002 static void
3003 rule_dealloc(struct rule *rule_)
3004 {
3005     struct rule_dpif *rule = rule_dpif_cast(rule_);
3006     free(rule);
3007 }
3008
3009 static enum ofperr
3010 rule_construct(struct rule *rule_)
3011 {
3012     struct rule_dpif *rule = rule_dpif_cast(rule_);
3013     ovs_mutex_init(&rule->stats_mutex);
3014     ovs_mutex_lock(&rule->stats_mutex);
3015     rule->packet_count = 0;
3016     rule->byte_count = 0;
3017     ovs_mutex_unlock(&rule->stats_mutex);
3018     return 0;
3019 }
3020
3021 static void
3022 rule_insert(struct rule *rule_)
3023     OVS_REQUIRES(ofproto_mutex)
3024 {
3025     struct rule_dpif *rule = rule_dpif_cast(rule_);
3026     complete_operation(rule);
3027 }
3028
3029 static void
3030 rule_delete(struct rule *rule_)
3031     OVS_REQUIRES(ofproto_mutex)
3032 {
3033     struct rule_dpif *rule = rule_dpif_cast(rule_);
3034     complete_operation(rule);
3035 }
3036
3037 static void
3038 rule_destruct(struct rule *rule_)
3039 {
3040     struct rule_dpif *rule = rule_dpif_cast(rule_);
3041     ovs_mutex_destroy(&rule->stats_mutex);
3042 }
3043
3044 static void
3045 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
3046 {
3047     struct rule_dpif *rule = rule_dpif_cast(rule_);
3048
3049     ovs_mutex_lock(&rule->stats_mutex);
3050     *packets = rule->packet_count;
3051     *bytes = rule->byte_count;
3052     ovs_mutex_unlock(&rule->stats_mutex);
3053 }
3054
3055 static void
3056 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
3057                   struct ofpbuf *packet)
3058 {
3059     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3060
3061     ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
3062 }
3063
3064 static enum ofperr
3065 rule_execute(struct rule *rule, const struct flow *flow,
3066              struct ofpbuf *packet)
3067 {
3068     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
3069     ofpbuf_delete(packet);
3070     return 0;
3071 }
3072
3073 static void
3074 rule_modify_actions(struct rule *rule_, bool reset_counters)
3075     OVS_REQUIRES(ofproto_mutex)
3076 {
3077     struct rule_dpif *rule = rule_dpif_cast(rule_);
3078
3079     if (reset_counters) {
3080         ovs_mutex_lock(&rule->stats_mutex);
3081         rule->packet_count = 0;
3082         rule->byte_count = 0;
3083         ovs_mutex_unlock(&rule->stats_mutex);
3084     }
3085
3086     complete_operation(rule);
3087 }
3088
3089 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
3090 {
3091     return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
3092 }
3093
3094 static struct ofgroup *
3095 group_alloc(void)
3096 {
3097     struct group_dpif *group = xzalloc(sizeof *group);
3098     return &group->up;
3099 }
3100
3101 static void
3102 group_dealloc(struct ofgroup *group_)
3103 {
3104     struct group_dpif *group = group_dpif_cast(group_);
3105     free(group);
3106 }
3107
3108 static void
3109 group_construct_stats(struct group_dpif *group)
3110     OVS_REQUIRES(group->stats_mutex)
3111 {
3112     group->packet_count = 0;
3113     group->byte_count = 0;
3114     if (!group->bucket_stats) {
3115         group->bucket_stats = xcalloc(group->up.n_buckets,
3116                                       sizeof *group->bucket_stats);
3117     } else {
3118         memset(group->bucket_stats, 0, group->up.n_buckets *
3119                sizeof *group->bucket_stats);
3120     }
3121 }
3122
3123 static enum ofperr
3124 group_construct(struct ofgroup *group_)
3125 {
3126     struct group_dpif *group = group_dpif_cast(group_);
3127     ovs_mutex_init(&group->stats_mutex);
3128     ovs_mutex_lock(&group->stats_mutex);
3129     group_construct_stats(group);
3130     ovs_mutex_unlock(&group->stats_mutex);
3131     return 0;
3132 }
3133
3134 static void
3135 group_destruct__(struct group_dpif *group)
3136     OVS_REQUIRES(group->stats_mutex)
3137 {
3138     free(group->bucket_stats);
3139     group->bucket_stats = NULL;
3140 }
3141
3142 static void
3143 group_destruct(struct ofgroup *group_)
3144 {
3145     struct group_dpif *group = group_dpif_cast(group_);
3146     ovs_mutex_lock(&group->stats_mutex);
3147     group_destruct__(group);
3148     ovs_mutex_unlock(&group->stats_mutex);
3149     ovs_mutex_destroy(&group->stats_mutex);
3150 }
3151
3152 static enum ofperr
3153 group_modify(struct ofgroup *group_, struct ofgroup *victim_)
3154 {
3155     struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
3156     struct group_dpif *group = group_dpif_cast(group_);
3157     struct group_dpif *victim = group_dpif_cast(victim_);
3158
3159     ovs_mutex_lock(&group->stats_mutex);
3160     if (victim->up.n_buckets < group->up.n_buckets) {
3161         group_destruct__(group);
3162     }
3163     group_construct_stats(group);
3164     ovs_mutex_unlock(&group->stats_mutex);
3165
3166     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3167
3168     return 0;
3169 }
3170
3171 static enum ofperr
3172 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
3173 {
3174     struct group_dpif *group = group_dpif_cast(group_);
3175
3176     ovs_mutex_lock(&group->stats_mutex);
3177     ogs->packet_count = group->packet_count;
3178     ogs->byte_count = group->byte_count;
3179     memcpy(ogs->bucket_stats, group->bucket_stats,
3180            group->up.n_buckets * sizeof *group->bucket_stats);
3181     ovs_mutex_unlock(&group->stats_mutex);
3182
3183     return 0;
3184 }
3185
3186 bool
3187 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
3188                   struct group_dpif **group)
3189     OVS_TRY_RDLOCK(true, (*group)->up.rwlock)
3190 {
3191     struct ofgroup *ofgroup;
3192     bool found;
3193
3194     *group = NULL;
3195     found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
3196     *group = found ?  group_dpif_cast(ofgroup) : NULL;
3197
3198     return found;
3199 }
3200
3201 void
3202 group_dpif_release(struct group_dpif *group)
3203     OVS_RELEASES(group->up.rwlock)
3204 {
3205     ofproto_group_release(&group->up);
3206 }
3207
3208 void
3209 group_dpif_get_buckets(const struct group_dpif *group,
3210                        const struct list **buckets)
3211 {
3212     *buckets = &group->up.buckets;
3213 }
3214
3215 enum ofp11_group_type
3216 group_dpif_get_type(const struct group_dpif *group)
3217 {
3218     return group->up.type;
3219 }
3220 \f
3221 /* Sends 'packet' out 'ofport'.
3222  * May modify 'packet'.
3223  * Returns 0 if successful, otherwise a positive errno value. */
3224 int
3225 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3226 {
3227     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3228     int error;
3229
3230     error = xlate_send_packet(ofport, packet);
3231
3232     ovs_mutex_lock(&ofproto->stats_mutex);
3233     ofproto->stats.tx_packets++;
3234     ofproto->stats.tx_bytes += packet->size;
3235     ovs_mutex_unlock(&ofproto->stats_mutex);
3236     return error;
3237 }
3238 \f
3239 static bool
3240 set_frag_handling(struct ofproto *ofproto_,
3241                   enum ofp_config_flags frag_handling)
3242 {
3243     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3244     if (frag_handling != OFPC_FRAG_REASM) {
3245         ofproto->backer->need_revalidate = REV_RECONFIGURE;
3246         return true;
3247     } else {
3248         return false;
3249     }
3250 }
3251
3252 static enum ofperr
3253 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
3254            const struct flow *flow,
3255            const struct ofpact *ofpacts, size_t ofpacts_len)
3256 {
3257     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3258
3259     ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
3260                                  ofpacts_len, packet);
3261     return 0;
3262 }
3263 \f
3264 /* NetFlow. */
3265
3266 static int
3267 set_netflow(struct ofproto *ofproto_,
3268             const struct netflow_options *netflow_options)
3269 {
3270     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3271
3272     if (netflow_options) {
3273         if (!ofproto->netflow) {
3274             ofproto->netflow = netflow_create();
3275             ofproto->backer->need_revalidate = REV_RECONFIGURE;
3276         }
3277         return netflow_set_options(ofproto->netflow, netflow_options);
3278     } else if (ofproto->netflow) {
3279         ofproto->backer->need_revalidate = REV_RECONFIGURE;
3280         netflow_unref(ofproto->netflow);
3281         ofproto->netflow = NULL;
3282     }
3283
3284     return 0;
3285 }
3286
3287 static void
3288 get_netflow_ids(const struct ofproto *ofproto_,
3289                 uint8_t *engine_type, uint8_t *engine_id)
3290 {
3291     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3292
3293     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
3294 }
3295 \f
3296 static struct ofproto_dpif *
3297 ofproto_dpif_lookup(const char *name)
3298 {
3299     struct ofproto_dpif *ofproto;
3300
3301     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
3302                              hash_string(name, 0), &all_ofproto_dpifs) {
3303         if (!strcmp(ofproto->up.name, name)) {
3304             return ofproto;
3305         }
3306     }
3307     return NULL;
3308 }
3309
3310 static void
3311 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
3312                           const char *argv[], void *aux OVS_UNUSED)
3313 {
3314     struct ofproto_dpif *ofproto;
3315
3316     if (argc > 1) {
3317         ofproto = ofproto_dpif_lookup(argv[1]);
3318         if (!ofproto) {
3319             unixctl_command_reply_error(conn, "no such bridge");
3320             return;
3321         }
3322         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3323         mac_learning_flush(ofproto->ml);
3324         ovs_rwlock_unlock(&ofproto->ml->rwlock);
3325     } else {
3326         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3327             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3328             mac_learning_flush(ofproto->ml);
3329             ovs_rwlock_unlock(&ofproto->ml->rwlock);
3330         }
3331     }
3332
3333     unixctl_command_reply(conn, "table successfully flushed");
3334 }
3335
3336 static struct ofport_dpif *
3337 ofbundle_get_a_port(const struct ofbundle *bundle)
3338 {
3339     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
3340                         bundle_node);
3341 }
3342
3343 static void
3344 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
3345                          const char *argv[], void *aux OVS_UNUSED)
3346 {
3347     struct ds ds = DS_EMPTY_INITIALIZER;
3348     const struct ofproto_dpif *ofproto;
3349     const struct mac_entry *e;
3350
3351     ofproto = ofproto_dpif_lookup(argv[1]);
3352     if (!ofproto) {
3353         unixctl_command_reply_error(conn, "no such bridge");
3354         return;
3355     }
3356
3357     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
3358     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3359     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3360         struct ofbundle *bundle = e->port.p;
3361         char name[OFP_MAX_PORT_NAME_LEN];
3362
3363         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
3364                                name, sizeof name);
3365         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
3366                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
3367                       mac_entry_age(ofproto->ml, e));
3368     }
3369     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3370     unixctl_command_reply(conn, ds_cstr(&ds));
3371     ds_destroy(&ds);
3372 }
3373
3374 struct trace_ctx {
3375     struct xlate_out xout;
3376     struct xlate_in xin;
3377     struct flow flow;
3378     struct ds *result;
3379 };
3380
3381 static void
3382 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
3383 {
3384     struct rule_actions *actions;
3385     ovs_be64 cookie;
3386
3387     ds_put_char_multiple(result, '\t', level);
3388     if (!rule) {
3389         ds_put_cstr(result, "No match\n");
3390         return;
3391     }
3392
3393     ovs_mutex_lock(&rule->up.mutex);
3394     cookie = rule->up.flow_cookie;
3395     ovs_mutex_unlock(&rule->up.mutex);
3396
3397     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
3398                   rule ? rule->up.table_id : 0, ntohll(cookie));
3399     cls_rule_format(&rule->up.cr, result);
3400     ds_put_char(result, '\n');
3401
3402     actions = rule_dpif_get_actions(rule);
3403
3404     ds_put_char_multiple(result, '\t', level);
3405     ds_put_cstr(result, "OpenFlow actions=");
3406     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
3407     ds_put_char(result, '\n');
3408
3409     rule_actions_unref(actions);
3410 }
3411
3412 static void
3413 trace_format_flow(struct ds *result, int level, const char *title,
3414                   struct trace_ctx *trace)
3415 {
3416     ds_put_char_multiple(result, '\t', level);
3417     ds_put_format(result, "%s: ", title);
3418     if (flow_equal(&trace->xin.flow, &trace->flow)) {
3419         ds_put_cstr(result, "unchanged");
3420     } else {
3421         flow_format(result, &trace->xin.flow);
3422         trace->flow = trace->xin.flow;
3423     }
3424     ds_put_char(result, '\n');
3425 }
3426
3427 static void
3428 trace_format_regs(struct ds *result, int level, const char *title,
3429                   struct trace_ctx *trace)
3430 {
3431     size_t i;
3432
3433     ds_put_char_multiple(result, '\t', level);
3434     ds_put_format(result, "%s:", title);
3435     for (i = 0; i < FLOW_N_REGS; i++) {
3436         ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
3437     }
3438     ds_put_char(result, '\n');
3439 }
3440
3441 static void
3442 trace_format_odp(struct ds *result, int level, const char *title,
3443                  struct trace_ctx *trace)
3444 {
3445     struct ofpbuf *odp_actions = &trace->xout.odp_actions;
3446
3447     ds_put_char_multiple(result, '\t', level);
3448     ds_put_format(result, "%s: ", title);
3449     format_odp_actions(result, odp_actions->data, odp_actions->size);
3450     ds_put_char(result, '\n');
3451 }
3452
3453 static void
3454 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
3455 {
3456     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
3457     struct ds *result = trace->result;
3458
3459     ds_put_char(result, '\n');
3460     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
3461     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
3462     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
3463     trace_format_rule(result, recurse + 1, rule);
3464 }
3465
3466 static void
3467 trace_report(struct xlate_in *xin, const char *s, int recurse)
3468 {
3469     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
3470     struct ds *result = trace->result;
3471
3472     ds_put_char_multiple(result, '\t', recurse);
3473     ds_put_cstr(result, s);
3474     ds_put_char(result, '\n');
3475 }
3476
3477 /* Parses the 'argc' elements of 'argv', ignoring argv[0].  The following
3478  * forms are supported:
3479  *
3480  *     - [dpname] odp_flow [-generate | packet]
3481  *     - bridge br_flow [-generate | packet]
3482  *
3483  * On success, initializes '*ofprotop' and 'flow' and returns NULL.  On failure
3484  * returns a nonnull malloced error message. */
3485 static char * WARN_UNUSED_RESULT
3486 parse_flow_and_packet(int argc, const char *argv[],
3487                       struct ofproto_dpif **ofprotop, struct flow *flow,
3488                       struct ofpbuf **packetp)
3489 {
3490     const struct dpif_backer *backer = NULL;
3491     const char *error = NULL;
3492     char *m_err = NULL;
3493     struct simap port_names = SIMAP_INITIALIZER(&port_names);
3494     struct ofpbuf *packet;
3495     struct ofpbuf odp_key;
3496     struct ofpbuf odp_mask;
3497
3498     ofpbuf_init(&odp_key, 0);
3499     ofpbuf_init(&odp_mask, 0);
3500
3501     /* Handle "-generate" or a hex string as the last argument. */
3502     if (!strcmp(argv[argc - 1], "-generate")) {
3503         packet = ofpbuf_new(0);
3504         argc--;
3505     } else {
3506         error = eth_from_hex(argv[argc - 1], &packet);
3507         if (!error) {
3508             argc--;
3509         } else if (argc == 4) {
3510             /* The 3-argument form must end in "-generate' or a hex string. */
3511             goto exit;
3512         }
3513         error = NULL;
3514     }
3515
3516     /* odp_flow can have its in_port specified as a name instead of port no.
3517      * We do not yet know whether a given flow is a odp_flow or a br_flow.
3518      * But, to know whether a flow is odp_flow through odp_flow_from_string(),
3519      * we need to create a simap of name to port no. */
3520     if (argc == 3) {
3521         const char *dp_type;
3522         if (!strncmp(argv[1], "ovs-", 4)) {
3523             dp_type = argv[1] + 4;
3524         } else {
3525             dp_type = argv[1];
3526         }
3527         backer = shash_find_data(&all_dpif_backers, dp_type);
3528     } else if (argc == 2) {
3529         struct shash_node *node;
3530         if (shash_count(&all_dpif_backers) == 1) {
3531             node = shash_first(&all_dpif_backers);
3532             backer = node->data;
3533         }
3534     } else {
3535         error = "Syntax error";
3536         goto exit;
3537     }
3538     if (backer && backer->dpif) {
3539         struct dpif_port dpif_port;
3540         struct dpif_port_dump port_dump;
3541         DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
3542             simap_put(&port_names, dpif_port.name,
3543                       odp_to_u32(dpif_port.port_no));
3544         }
3545     }
3546
3547     /* Parse the flow and determine whether a datapath or
3548      * bridge is specified. If function odp_flow_key_from_string()
3549      * returns 0, the flow is a odp_flow. If function
3550      * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
3551     if (!odp_flow_from_string(argv[argc - 1], &port_names,
3552                               &odp_key, &odp_mask)) {
3553         if (!backer) {
3554             error = "Cannot find the datapath";
3555             goto exit;
3556         }
3557
3558         if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, flow,
3559                           NULL, ofprotop, NULL, NULL, NULL, NULL)) {
3560             error = "Invalid datapath flow";
3561             goto exit;
3562         }
3563     } else {
3564         char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
3565
3566         if (err) {
3567             m_err = xasprintf("Bad flow syntax: %s", err);
3568             free(err);
3569             goto exit;
3570         } else {
3571             if (argc != 3) {
3572                 error = "Must specify bridge name";
3573                 goto exit;
3574             }
3575
3576             *ofprotop = ofproto_dpif_lookup(argv[1]);
3577             if (!*ofprotop) {
3578                 error = "Unknown bridge name";
3579                 goto exit;
3580             }
3581         }
3582     }
3583
3584     /* Generate a packet, if requested. */
3585     if (packet) {
3586         if (!packet->size) {
3587             flow_compose(packet, flow);
3588         } else {
3589             union flow_in_port in_port = flow->in_port;
3590
3591             /* Use the metadata from the flow and the packet argument
3592              * to reconstruct the flow. */
3593             flow_extract(packet, flow->skb_priority, flow->pkt_mark, NULL,
3594                          &in_port, flow);
3595         }
3596     }
3597
3598 exit:
3599     if (error && !m_err) {
3600         m_err = xstrdup(error);
3601     }
3602     if (m_err) {
3603         ofpbuf_delete(packet);
3604         packet = NULL;
3605     }
3606     *packetp = packet;
3607     ofpbuf_uninit(&odp_key);
3608     ofpbuf_uninit(&odp_mask);
3609     simap_destroy(&port_names);
3610     return m_err;
3611 }
3612
3613 static void
3614 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
3615                       void *aux OVS_UNUSED)
3616 {
3617     struct ofproto_dpif *ofproto;
3618     struct ofpbuf *packet;
3619     char *error;
3620     struct flow flow;
3621
3622     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
3623     if (!error) {
3624         struct ds result;
3625
3626         ds_init(&result);
3627         ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
3628         unixctl_command_reply(conn, ds_cstr(&result));
3629         ds_destroy(&result);
3630         ofpbuf_delete(packet);
3631     } else {
3632         unixctl_command_reply_error(conn, error);
3633         free(error);
3634     }
3635 }
3636
3637 static void
3638 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
3639                               const char *argv[], void *aux OVS_UNUSED)
3640 {
3641     enum ofputil_protocol usable_protocols;
3642     struct ofproto_dpif *ofproto;
3643     bool enforce_consistency;
3644     struct ofpbuf ofpacts;
3645     struct ofpbuf *packet;
3646     struct ds result;
3647     struct flow flow;
3648     uint16_t in_port;
3649
3650     /* Three kinds of error return values! */
3651     enum ofperr retval;
3652     char *error;
3653
3654     packet = NULL;
3655     ds_init(&result);
3656     ofpbuf_init(&ofpacts, 0);
3657
3658     /* Parse actions. */
3659     error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols);
3660     if (error) {
3661         unixctl_command_reply_error(conn, error);
3662         free(error);
3663         goto exit;
3664     }
3665
3666     /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
3667      * consistency between the flow and the actions.  With -consistent, we
3668      * enforce consistency even for a flow supported in OpenFlow 1.0. */
3669     if (!strcmp(argv[1], "-consistent")) {
3670         enforce_consistency = true;
3671         argv++;
3672         argc--;
3673     } else {
3674         enforce_consistency = false;
3675     }
3676
3677     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
3678     if (error) {
3679         unixctl_command_reply_error(conn, error);
3680         free(error);
3681         goto exit;
3682     }
3683
3684     /* Do the same checks as handle_packet_out() in ofproto.c.
3685      *
3686      * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
3687      * strictly correct because these actions aren't in any table, but it's OK
3688      * because it 'table_id' is used only to check goto_table instructions, but
3689      * packet-outs take a list of actions and therefore it can't include
3690      * instructions.
3691      *
3692      * We skip the "meter" check here because meter is an instruction, not an
3693      * action, and thus cannot appear in ofpacts. */
3694     in_port = ofp_to_u16(flow.in_port.ofp_port);
3695     if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
3696         unixctl_command_reply_error(conn, "invalid in_port");
3697         goto exit;
3698     }
3699     if (enforce_consistency) {
3700         retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size, &flow,
3701                                            u16_to_ofp(ofproto->up.max_ports),
3702                                            0, 0, usable_protocols);
3703     } else {
3704         retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow,
3705                                u16_to_ofp(ofproto->up.max_ports), 0, 0,
3706                                &usable_protocols);
3707     }
3708
3709     if (retval) {
3710         ds_clear(&result);
3711         ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
3712         unixctl_command_reply_error(conn, ds_cstr(&result));
3713         goto exit;
3714     }
3715
3716     ofproto_trace(ofproto, &flow, packet, ofpacts.data, ofpacts.size, &result);
3717     unixctl_command_reply(conn, ds_cstr(&result));
3718
3719 exit:
3720     ds_destroy(&result);
3721     ofpbuf_delete(packet);
3722     ofpbuf_uninit(&ofpacts);
3723 }
3724
3725 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
3726  * description of the results to 'ds'.
3727  *
3728  * The trace follows a packet with the specified 'flow' through the flow
3729  * table.  'packet' may be nonnull to trace an actual packet, with consequent
3730  * side effects (if it is nonnull then its flow must be 'flow').
3731  *
3732  * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
3733  * trace, otherwise the actions are determined by a flow table lookup. */
3734 static void
3735 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
3736               const struct ofpbuf *packet,
3737               const struct ofpact ofpacts[], size_t ofpacts_len,
3738               struct ds *ds)
3739 {
3740     struct rule_dpif *rule;
3741     struct flow_wildcards wc;
3742
3743     ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
3744     ds_put_cstr(ds, "Flow: ");
3745     flow_format(ds, flow);
3746     ds_put_char(ds, '\n');
3747
3748     flow_wildcards_init_catchall(&wc);
3749     if (ofpacts) {
3750         rule = NULL;
3751     } else {
3752         rule_dpif_lookup(ofproto, flow, &wc, &rule);
3753
3754         trace_format_rule(ds, 0, rule);
3755         if (rule == ofproto->miss_rule) {
3756             ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
3757         } else if (rule == ofproto->no_packet_in_rule) {
3758             ds_put_cstr(ds, "\nNo match, packets dropped because "
3759                         "OFPPC_NO_PACKET_IN is set on in_port.\n");
3760         } else if (rule == ofproto->drop_frags_rule) {
3761             ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
3762                         "and the fragment handling mode is \"drop\".\n");
3763         }
3764     }
3765
3766     if (rule || ofpacts) {
3767         uint64_t odp_actions_stub[1024 / 8];
3768         struct ofpbuf odp_actions;
3769         struct trace_ctx trace;
3770         struct match match;
3771         uint16_t tcp_flags;
3772
3773         tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
3774         trace.result = ds;
3775         trace.flow = *flow;
3776         ofpbuf_use_stub(&odp_actions,
3777                         odp_actions_stub, sizeof odp_actions_stub);
3778         xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
3779         if (ofpacts) {
3780             trace.xin.ofpacts = ofpacts;
3781             trace.xin.ofpacts_len = ofpacts_len;
3782         }
3783         trace.xin.resubmit_hook = trace_resubmit;
3784         trace.xin.report_hook = trace_report;
3785
3786         xlate_actions(&trace.xin, &trace.xout);
3787         flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
3788
3789         ds_put_char(ds, '\n');
3790         trace_format_flow(ds, 0, "Final flow", &trace);
3791
3792         match_init(&match, flow, &trace.xout.wc);
3793         ds_put_cstr(ds, "Relevant fields: ");
3794         match_format(&match, ds, OFP_DEFAULT_PRIORITY);
3795         ds_put_char(ds, '\n');
3796
3797         ds_put_cstr(ds, "Datapath actions: ");
3798         format_odp_actions(ds, trace.xout.odp_actions.data,
3799                            trace.xout.odp_actions.size);
3800
3801         if (trace.xout.slow) {
3802             enum slow_path_reason slow;
3803
3804             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
3805                         "slow path because it:");
3806
3807             slow = trace.xout.slow;
3808             while (slow) {
3809                 enum slow_path_reason bit = rightmost_1bit(slow);
3810
3811                 ds_put_format(ds, "\n\t- %s.",
3812                               slow_path_reason_to_explanation(bit));
3813
3814                 slow &= ~bit;
3815             }
3816         }
3817
3818         xlate_out_uninit(&trace.xout);
3819     }
3820
3821     rule_dpif_unref(rule);
3822 }
3823
3824 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
3825  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
3826  * to destroy 'ofproto_shash' and free the returned value. */
3827 static const struct shash_node **
3828 get_ofprotos(struct shash *ofproto_shash)
3829 {
3830     const struct ofproto_dpif *ofproto;
3831
3832     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3833         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
3834         shash_add_nocopy(ofproto_shash, name, ofproto);
3835     }
3836
3837     return shash_sort(ofproto_shash);
3838 }
3839
3840 static void
3841 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
3842                               const char *argv[] OVS_UNUSED,
3843                               void *aux OVS_UNUSED)
3844 {
3845     struct ds ds = DS_EMPTY_INITIALIZER;
3846     struct shash ofproto_shash;
3847     const struct shash_node **sorted_ofprotos;
3848     int i;
3849
3850     shash_init(&ofproto_shash);
3851     sorted_ofprotos = get_ofprotos(&ofproto_shash);
3852     for (i = 0; i < shash_count(&ofproto_shash); i++) {
3853         const struct shash_node *node = sorted_ofprotos[i];
3854         ds_put_format(&ds, "%s\n", node->name);
3855     }
3856
3857     shash_destroy(&ofproto_shash);
3858     free(sorted_ofprotos);
3859
3860     unixctl_command_reply(conn, ds_cstr(&ds));
3861     ds_destroy(&ds);
3862 }
3863
3864 static void
3865 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
3866 {
3867     const struct shash_node **ofprotos;
3868     struct dpif_dp_stats dp_stats;
3869     struct shash ofproto_shash;
3870     size_t i;
3871
3872     dpif_get_dp_stats(backer->dpif, &dp_stats);
3873
3874     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
3875                   dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
3876
3877     shash_init(&ofproto_shash);
3878     ofprotos = get_ofprotos(&ofproto_shash);
3879     for (i = 0; i < shash_count(&ofproto_shash); i++) {
3880         struct ofproto_dpif *ofproto = ofprotos[i]->data;
3881         const struct shash_node **ports;
3882         size_t j;
3883
3884         if (ofproto->backer != backer) {
3885             continue;
3886         }
3887
3888         ds_put_format(ds, "\t%s:\n", ofproto->up.name);
3889
3890         ports = shash_sort(&ofproto->up.port_by_name);
3891         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
3892             const struct shash_node *node = ports[j];
3893             struct ofport *ofport = node->data;
3894             struct smap config;
3895             odp_port_t odp_port;
3896
3897             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
3898                           ofport->ofp_port);
3899
3900             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
3901             if (odp_port != ODPP_NONE) {
3902                 ds_put_format(ds, "%"PRIu32":", odp_port);
3903             } else {
3904                 ds_put_cstr(ds, "none:");
3905             }
3906
3907             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
3908
3909             smap_init(&config);
3910             if (!netdev_get_config(ofport->netdev, &config)) {
3911                 const struct smap_node **nodes;
3912                 size_t i;
3913
3914                 nodes = smap_sort(&config);
3915                 for (i = 0; i < smap_count(&config); i++) {
3916                     const struct smap_node *node = nodes[i];
3917                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
3918                                   node->key, node->value);
3919                 }
3920                 free(nodes);
3921             }
3922             smap_destroy(&config);
3923
3924             ds_put_char(ds, ')');
3925             ds_put_char(ds, '\n');
3926         }
3927         free(ports);
3928     }
3929     shash_destroy(&ofproto_shash);
3930     free(ofprotos);
3931 }
3932
3933 static void
3934 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
3935                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
3936 {
3937     struct ds ds = DS_EMPTY_INITIALIZER;
3938     const struct shash_node **backers;
3939     int i;
3940
3941     backers = shash_sort(&all_dpif_backers);
3942     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
3943         dpif_show_backer(backers[i]->data, &ds);
3944     }
3945     free(backers);
3946
3947     unixctl_command_reply(conn, ds_cstr(&ds));
3948     ds_destroy(&ds);
3949 }
3950
3951 static bool
3952 ofproto_dpif_contains_flow(const struct ofproto_dpif *ofproto,
3953                            const struct nlattr *key, size_t key_len)
3954 {
3955     enum odp_key_fitness fitness;
3956     struct ofproto_dpif *ofp;
3957     struct flow flow;
3958
3959     xlate_receive(ofproto->backer, NULL, key, key_len, &flow, &fitness, &ofp,
3960                   NULL, NULL, NULL, NULL);
3961     return ofp == ofproto;
3962 }
3963
3964 static void
3965 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
3966                                 int argc OVS_UNUSED, const char *argv[],
3967                                 void *aux OVS_UNUSED)
3968 {
3969     struct ds ds = DS_EMPTY_INITIALIZER;
3970     const struct dpif_flow_stats *stats;
3971     const struct ofproto_dpif *ofproto;
3972     struct dpif_flow_dump flow_dump;
3973     const struct nlattr *actions;
3974     const struct nlattr *mask;
3975     const struct nlattr *key;
3976     size_t actions_len;
3977     size_t mask_len;
3978     size_t key_len;
3979
3980     ofproto = ofproto_dpif_lookup(argv[1]);
3981     if (!ofproto) {
3982         unixctl_command_reply_error(conn, "no such bridge");
3983         return;
3984     }
3985
3986     ds_init(&ds);
3987     dpif_flow_dump_start(&flow_dump, ofproto->backer->dpif);
3988     while (dpif_flow_dump_next(&flow_dump, &key, &key_len, &mask, &mask_len,
3989                                &actions, &actions_len, &stats)) {
3990         if (!ofproto_dpif_contains_flow(ofproto, key, key_len)) {
3991             continue;
3992         }
3993
3994         odp_flow_format(key, key_len, mask, mask_len, NULL, &ds, false);
3995         ds_put_cstr(&ds, ", ");
3996         dpif_flow_stats_format(stats, &ds);
3997         ds_put_cstr(&ds, ", actions:");
3998         format_odp_actions(&ds, actions, actions_len);
3999         ds_put_char(&ds, '\n');
4000     }
4001
4002     if (dpif_flow_dump_done(&flow_dump)) {
4003         ds_clear(&ds);
4004         ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
4005         unixctl_command_reply_error(conn, ds_cstr(&ds));
4006     } else {
4007         unixctl_command_reply(conn, ds_cstr(&ds));
4008     }
4009     ds_destroy(&ds);
4010 }
4011
4012 static void
4013 ofproto_dpif_unixctl_init(void)
4014 {
4015     static bool registered;
4016     if (registered) {
4017         return;
4018     }
4019     registered = true;
4020
4021     unixctl_command_register(
4022         "ofproto/trace",
4023         "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
4024         1, 3, ofproto_unixctl_trace, NULL);
4025     unixctl_command_register(
4026         "ofproto/trace-packet-out",
4027         "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
4028         2, 6, ofproto_unixctl_trace_actions, NULL);
4029     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
4030                              ofproto_unixctl_fdb_flush, NULL);
4031     unixctl_command_register("fdb/show", "bridge", 1, 1,
4032                              ofproto_unixctl_fdb_show, NULL);
4033     unixctl_command_register("dpif/dump-dps", "", 0, 0,
4034                              ofproto_unixctl_dpif_dump_dps, NULL);
4035     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
4036                              NULL);
4037     unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
4038                              ofproto_unixctl_dpif_dump_flows, NULL);
4039 }
4040 \f
4041 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4042  *
4043  * This is deprecated.  It is only for compatibility with broken device drivers
4044  * in old versions of Linux that do not properly support VLANs when VLAN
4045  * devices are not used.  When broken device drivers are no longer in
4046  * widespread use, we will delete these interfaces. */
4047
4048 static int
4049 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
4050 {
4051     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
4052     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
4053
4054     if (realdev_ofp_port == ofport->realdev_ofp_port
4055         && vid == ofport->vlandev_vid) {
4056         return 0;
4057     }
4058
4059     ofproto->backer->need_revalidate = REV_RECONFIGURE;
4060
4061     if (ofport->realdev_ofp_port) {
4062         vsp_remove(ofport);
4063     }
4064     if (realdev_ofp_port && ofport->bundle) {
4065         /* vlandevs are enslaved to their realdevs, so they are not allowed to
4066          * themselves be part of a bundle. */
4067         bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
4068     }
4069
4070     ofport->realdev_ofp_port = realdev_ofp_port;
4071     ofport->vlandev_vid = vid;
4072
4073     if (realdev_ofp_port) {
4074         vsp_add(ofport, realdev_ofp_port, vid);
4075     }
4076
4077     return 0;
4078 }
4079
4080 static uint32_t
4081 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
4082 {
4083     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
4084 }
4085
4086 bool
4087 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
4088     OVS_EXCLUDED(ofproto->vsp_mutex)
4089 {
4090     bool ret;
4091
4092     ovs_mutex_lock(&ofproto->vsp_mutex);
4093     ret = !hmap_is_empty(&ofproto->realdev_vid_map);
4094     ovs_mutex_unlock(&ofproto->vsp_mutex);
4095     return ret;
4096 }
4097
4098 static ofp_port_t
4099 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
4100                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4101     OVS_REQUIRES(ofproto->vsp_mutex)
4102 {
4103     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
4104         int vid = vlan_tci_to_vid(vlan_tci);
4105         const struct vlan_splinter *vsp;
4106
4107         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
4108                                  hash_realdev_vid(realdev_ofp_port, vid),
4109                                  &ofproto->realdev_vid_map) {
4110             if (vsp->realdev_ofp_port == realdev_ofp_port
4111                 && vsp->vid == vid) {
4112                 return vsp->vlandev_ofp_port;
4113             }
4114         }
4115     }
4116     return realdev_ofp_port;
4117 }
4118
4119 /* Returns the OFP port number of the Linux VLAN device that corresponds to
4120  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
4121  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
4122  * 'vlan_tci' 9, it would return the port number of eth0.9.
4123  *
4124  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
4125  * function just returns its 'realdev_ofp_port' argument. */
4126 ofp_port_t
4127 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
4128                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4129     OVS_EXCLUDED(ofproto->vsp_mutex)
4130 {
4131     ofp_port_t ret;
4132
4133     ovs_mutex_lock(&ofproto->vsp_mutex);
4134     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
4135     ovs_mutex_unlock(&ofproto->vsp_mutex);
4136     return ret;
4137 }
4138
4139 static struct vlan_splinter *
4140 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
4141 {
4142     struct vlan_splinter *vsp;
4143
4144     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
4145                              hash_ofp_port(vlandev_ofp_port),
4146                              &ofproto->vlandev_map) {
4147         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
4148             return vsp;
4149         }
4150     }
4151
4152     return NULL;
4153 }
4154
4155 /* Returns the OpenFlow port number of the "real" device underlying the Linux
4156  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
4157  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
4158  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
4159  * eth0 and store 9 in '*vid'.
4160  *
4161  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
4162  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
4163  * always does.*/
4164 static ofp_port_t
4165 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
4166                        ofp_port_t vlandev_ofp_port, int *vid)
4167     OVS_REQUIRES(ofproto->vsp_mutex)
4168 {
4169     if (!hmap_is_empty(&ofproto->vlandev_map)) {
4170         const struct vlan_splinter *vsp;
4171
4172         vsp = vlandev_find(ofproto, vlandev_ofp_port);
4173         if (vsp) {
4174             if (vid) {
4175                 *vid = vsp->vid;
4176             }
4177             return vsp->realdev_ofp_port;
4178         }
4179     }
4180     return 0;
4181 }
4182
4183 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
4184  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
4185  * 'flow->in_port' to the "real" device backing the VLAN device, sets
4186  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Otherwise (which is
4187  * always the case unless VLAN splinters are enabled), returns false without
4188  * making any changes. */
4189 bool
4190 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
4191     OVS_EXCLUDED(ofproto->vsp_mutex)
4192 {
4193     ofp_port_t realdev;
4194     int vid;
4195
4196     ovs_mutex_lock(&ofproto->vsp_mutex);
4197     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
4198     ovs_mutex_unlock(&ofproto->vsp_mutex);
4199     if (!realdev) {
4200         return false;
4201     }
4202
4203     /* Cause the flow to be processed as if it came in on the real device with
4204      * the VLAN device's VLAN ID. */
4205     flow->in_port.ofp_port = realdev;
4206     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
4207     return true;
4208 }
4209
4210 static void
4211 vsp_remove(struct ofport_dpif *port)
4212 {
4213     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4214     struct vlan_splinter *vsp;
4215
4216     ovs_mutex_lock(&ofproto->vsp_mutex);
4217     vsp = vlandev_find(ofproto, port->up.ofp_port);
4218     if (vsp) {
4219         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
4220         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
4221         free(vsp);
4222
4223         port->realdev_ofp_port = 0;
4224     } else {
4225         VLOG_ERR("missing vlan device record");
4226     }
4227     ovs_mutex_unlock(&ofproto->vsp_mutex);
4228 }
4229
4230 static void
4231 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
4232 {
4233     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4234
4235     ovs_mutex_lock(&ofproto->vsp_mutex);
4236     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
4237         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
4238             == realdev_ofp_port)) {
4239         struct vlan_splinter *vsp;
4240
4241         vsp = xmalloc(sizeof *vsp);
4242         vsp->realdev_ofp_port = realdev_ofp_port;
4243         vsp->vlandev_ofp_port = port->up.ofp_port;
4244         vsp->vid = vid;
4245
4246         port->realdev_ofp_port = realdev_ofp_port;
4247
4248         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
4249                     hash_ofp_port(port->up.ofp_port));
4250         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
4251                     hash_realdev_vid(realdev_ofp_port, vid));
4252     } else {
4253         VLOG_ERR("duplicate vlan device record");
4254     }
4255     ovs_mutex_unlock(&ofproto->vsp_mutex);
4256 }
4257
4258 static odp_port_t
4259 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
4260 {
4261     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
4262     return ofport ? ofport->odp_port : ODPP_NONE;
4263 }
4264
4265 struct ofport_dpif *
4266 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
4267 {
4268     struct ofport_dpif *port;
4269
4270     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
4271     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
4272                              &backer->odp_to_ofport_map) {
4273         if (port->odp_port == odp_port) {
4274             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4275             return port;
4276         }
4277     }
4278
4279     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4280     return NULL;
4281 }
4282
4283 static ofp_port_t
4284 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
4285 {
4286     struct ofport_dpif *port;
4287
4288     port = odp_port_to_ofport(ofproto->backer, odp_port);
4289     if (port && &ofproto->up == port->up.ofproto) {
4290         return port->up.ofp_port;
4291     } else {
4292         return OFPP_NONE;
4293     }
4294 }
4295
4296 const struct ofproto_class ofproto_dpif_class = {
4297     init,
4298     enumerate_types,
4299     enumerate_names,
4300     del,
4301     port_open_type,
4302     type_run,
4303     type_wait,
4304     alloc,
4305     construct,
4306     destruct,
4307     dealloc,
4308     run,
4309     wait,
4310     NULL,                       /* get_memory_usage. */
4311     type_get_memory_usage,
4312     flush,
4313     get_features,
4314     get_tables,
4315     port_alloc,
4316     port_construct,
4317     port_destruct,
4318     port_dealloc,
4319     port_modified,
4320     port_reconfigured,
4321     port_query_by_name,
4322     port_add,
4323     port_del,
4324     port_get_stats,
4325     port_dump_start,
4326     port_dump_next,
4327     port_dump_done,
4328     port_poll,
4329     port_poll_wait,
4330     port_is_lacp_current,
4331     NULL,                       /* rule_choose_table */
4332     rule_alloc,
4333     rule_construct,
4334     rule_insert,
4335     rule_delete,
4336     rule_destruct,
4337     rule_dealloc,
4338     rule_get_stats,
4339     rule_execute,
4340     rule_modify_actions,
4341     set_frag_handling,
4342     packet_out,
4343     set_netflow,
4344     get_netflow_ids,
4345     set_sflow,
4346     set_ipfix,
4347     set_cfm,
4348     get_cfm_status,
4349     set_bfd,
4350     get_bfd_status,
4351     set_stp,
4352     get_stp_status,
4353     set_stp_port,
4354     get_stp_port_status,
4355     get_stp_port_stats,
4356     set_queues,
4357     bundle_set,
4358     bundle_remove,
4359     mirror_set__,
4360     mirror_get_stats__,
4361     set_flood_vlans,
4362     is_mirror_output_bundle,
4363     forward_bpdu_changed,
4364     set_mac_table_config,
4365     set_realdev,
4366     NULL,                       /* meter_get_features */
4367     NULL,                       /* meter_set */
4368     NULL,                       /* meter_get */
4369     NULL,                       /* meter_del */
4370     group_alloc,                /* group_alloc */
4371     group_construct,            /* group_construct */
4372     group_destruct,             /* group_destruct */
4373     group_dealloc,              /* group_dealloc */
4374     group_modify,               /* group_modify */
4375     group_get_stats,            /* group_get_stats */
4376 };