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-rid.h"
  57 #include "ofproto-dpif-sflow.h"
  58 #include "ofproto-dpif-upcall.h"
  59 #include "ofproto-dpif-xlate.h"
  60 #include "poll-loop.h"
  61 #include "seq.h"
  62 #include "simap.h"
  63 #include "smap.h"
  64 #include "timer.h"
  65 #include "tunnel.h"
  66 #include "unaligned.h"
  67 #include "unixctl.h"
  68 #include "vlan-bitmap.h"
  69 #include "vlog.h"
  70
  71 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  72
  73 COVERAGE_DEFINE(ofproto_dpif_expired);
  74 COVERAGE_DEFINE(packet_in_overflow);
  75
  76 /* Number of implemented OpenFlow tables. */
  77 enum { N_TABLES = 255 };
  78 enum { TBL_INTERNAL = N_TABLES - 1 };    /* Used for internal hidden rules. */
  79 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
  80
  81 struct flow_miss;
  82
  83 struct rule_dpif {
  84     struct rule up;
  85
  86     /* These statistics:
  87      *
  88      *   - Do include packets and bytes from datapath flows which have not
  89      *   recently been processed by a revalidator. */
  90     struct ovs_mutex stats_mutex;
  91     struct dpif_flow_stats stats OVS_GUARDED;
  92 };
  93
  94 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
  95                            long long int *used);
  96 static struct rule_dpif *rule_dpif_cast(const struct rule *);
  97 static void rule_expire(struct rule_dpif *);
  98
  99 struct group_dpif {
 100     struct ofgroup up;
 101
 102     /* These statistics:
 103      *
 104      *   - Do include packets and bytes from datapath flows which have not
 105      *   recently been processed by a revalidator. */
 106     struct ovs_mutex stats_mutex;
 107     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 108     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 109     struct bucket_counter *bucket_stats OVS_GUARDED;  /* Bucket statistics. */
 110 };
 111
 112 struct ofbundle {
 113     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
 114     struct ofproto_dpif *ofproto; /* Owning ofproto. */
 115     void *aux;                  /* Key supplied by ofproto's client. */
 116     char *name;                 /* Identifier for log messages. */
 117
 118     /* Configuration. */
 119     struct list ports;          /* Contains "struct ofport"s. */
 120     enum port_vlan_mode vlan_mode; /* VLAN mode */
 121     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
 122     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
 123                                  * NULL if all VLANs are trunked. */
 124     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 125     struct bond *bond;          /* Nonnull iff more than one port. */
 126     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 127
 128     /* Status. */
 129     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 130 };
 131
 132 static void bundle_remove(struct ofport *);
 133 static void bundle_update(struct ofbundle *);
 134 static void bundle_destroy(struct ofbundle *);
 135 static void bundle_del_port(struct ofport_dpif *);
 136 static void bundle_run(struct ofbundle *);
 137 static void bundle_wait(struct ofbundle *);
 138
 139 static void stp_run(struct ofproto_dpif *ofproto);
 140 static void stp_wait(struct ofproto_dpif *ofproto);
 141 static int set_stp_port(struct ofport *,
 142                         const struct ofproto_port_stp_settings *);
 143
 144 struct ofport_dpif {
 145     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 146     struct ofport up;
 147
 148     odp_port_t odp_port;
 149     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 150     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
 151     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 152     struct bfd *bfd;            /* BFD, if any. */
 153     bool may_enable;            /* May be enabled in bonds. */
 154     bool is_tunnel;             /* This port is a tunnel. */
 155     bool is_layer3;             /* This is a layer 3 port. */
 156     long long int carrier_seq;  /* Carrier status changes. */
 157     struct ofport_dpif *peer;   /* Peer if patch port. */
 158
 159     /* Spanning tree. */
 160     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 161     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 162     long long int stp_state_entered;
 163
 164     /* Queue to DSCP mapping. */
 165     struct ofproto_port_queue *qdscp;
 166     size_t n_qdscp;
 167
 168     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 169      *
 170      * This is deprecated.  It is only for compatibility with broken device
 171      * drivers in old versions of Linux that do not properly support VLANs when
 172      * VLAN devices are not used.  When broken device drivers are no longer in
 173      * widespread use, we will delete these interfaces. */
 174     ofp_port_t realdev_ofp_port;
 175     int vlandev_vid;
 176 };
 177
 178 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 179  *
 180  * This is deprecated.  It is only for compatibility with broken device drivers
 181  * in old versions of Linux that do not properly support VLANs when VLAN
 182  * devices are not used.  When broken device drivers are no longer in
 183  * widespread use, we will delete these interfaces. */
 184 struct vlan_splinter {
 185     struct hmap_node realdev_vid_node;
 186     struct hmap_node vlandev_node;
 187     ofp_port_t realdev_ofp_port;
 188     ofp_port_t vlandev_ofp_port;
 189     int vid;
 190 };
 191
 192 static void vsp_remove(struct ofport_dpif *);
 193 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 194
 195 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 196                                        ofp_port_t);
 197
 198 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 199                                        odp_port_t);
 200
 201 static struct ofport_dpif *
 202 ofport_dpif_cast(const struct ofport *ofport)
 203 {
 204     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 205 }
 206
 207 static void port_run(struct ofport_dpif *);
 208 static int set_bfd(struct ofport *, const struct smap *);
 209 static int set_cfm(struct ofport *, const struct cfm_settings *);
 210 static void ofport_update_peer(struct ofport_dpif *);
 211
 212 struct dpif_completion {
 213     struct list list_node;
 214     struct ofoperation *op;
 215 };
 216
 217 /* Reasons that we might need to revalidate every datapath flow, and
 218  * corresponding coverage counters.
 219  *
 220  * A value of 0 means that there is no need to revalidate.
 221  *
 222  * It would be nice to have some cleaner way to integrate with coverage
 223  * counters, but with only a few reasons I guess this is good enough for
 224  * now. */
 225 enum revalidate_reason {
 226     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 227     REV_STP,                   /* Spanning tree protocol port status change. */
 228     REV_BOND,                  /* Bonding changed. */
 229     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 230     REV_FLOW_TABLE,            /* Flow table changed. */
 231     REV_MAC_LEARNING,          /* Mac learning changed. */
 232 };
 233 COVERAGE_DEFINE(rev_reconfigure);
 234 COVERAGE_DEFINE(rev_stp);
 235 COVERAGE_DEFINE(rev_bond);
 236 COVERAGE_DEFINE(rev_port_toggled);
 237 COVERAGE_DEFINE(rev_flow_table);
 238 COVERAGE_DEFINE(rev_mac_learning);
 239
 240 /* All datapaths of a given type share a single dpif backer instance. */
 241 struct dpif_backer {
 242     char *type;
 243     int refcount;
 244     struct dpif *dpif;
 245     struct udpif *udpif;
 246
 247     struct ovs_rwlock odp_to_ofport_lock;
 248     struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
 249
 250     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 251
 252     enum revalidate_reason need_revalidate; /* Revalidate all flows. */
 253
 254     bool recv_set_enable; /* Enables or disables receiving packets. */
 255
 256     /* Recirculation. */
 257     struct recirc_id_pool *rid_pool;       /* Recirculation ID pool. */
 258     bool enable_recirc;   /* True if the datapath supports recirculation */
 259
 260     /* True if the datapath supports variable-length
 261      * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
 262      * False if the datapath supports only 8-byte (or shorter) userdata. */
 263     bool variable_length_userdata;
 264
 265     /* Maximum number of MPLS label stack entries that the datapath supports
 266      * in a match */
 267     size_t max_mpls_depth;
 268 };
 269
 270 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 271 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 272
 273 struct ofproto_dpif {
 274     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 275     struct ofproto up;
 276     struct dpif_backer *backer;
 277
 278     uint64_t dump_seq; /* Last read of udpif_dump_seq(). */
 279
 280     /* Special OpenFlow rules. */
 281     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 282     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 283     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 284
 285     /* Bridging. */
 286     struct netflow *netflow;
 287     struct dpif_sflow *sflow;
 288     struct dpif_ipfix *ipfix;
 289     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 290     struct mac_learning *ml;
 291     bool has_bonded_bundles;
 292     bool lacp_enabled;
 293     struct mbridge *mbridge;
 294
 295     struct ovs_mutex stats_mutex;
 296     struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
 297                                             * consumed in userspace. */
 298
 299     /* Spanning tree. */
 300     struct stp *stp;
 301     long long int stp_last_tick;
 302
 303     /* VLAN splinters. */
 304     struct ovs_mutex vsp_mutex;
 305     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 306     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 307
 308     /* Ports. */
 309     struct sset ports;             /* Set of standard port names. */
 310     struct sset ghost_ports;       /* Ports with no datapath port. */
 311     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 312     int port_poll_errno;           /* Last errno for port_poll() reply. */
 313     uint64_t change_seq;           /* Connectivity status changes. */
 314
 315     /* Work queues. */
 316     struct guarded_list pins;      /* Contains "struct ofputil_packet_in"s. */
 317 };
 318
 319 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 320 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 321
 322 static void ofproto_dpif_unixctl_init(void);
 323
 324 static inline struct ofproto_dpif *
 325 ofproto_dpif_cast(const struct ofproto *ofproto)
 326 {
 327     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 328     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 329 }
 330
 331 size_t
 332 ofproto_dpif_get_max_mpls_depth(const struct ofproto_dpif *ofproto)
 333 {
 334     return ofproto->backer->max_mpls_depth;
 335 }
 336
 337 bool
 338 ofproto_dpif_get_enable_recirc(const struct ofproto_dpif *ofproto)
 339 {
 340     return ofproto->backer->enable_recirc;
 341 }
 342
 343 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
 344                                         ofp_port_t ofp_port);
 345 static void ofproto_trace(struct ofproto_dpif *, struct flow *,
 346                           const struct ofpbuf *packet,
 347                           const struct ofpact[], size_t ofpacts_len,
 348                           struct ds *);
 349
 350 /* Global variables. */
 351 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 352
 353 /* Initial mappings of port to bridge mappings. */
 354 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 355
 356 /* Executes 'fm'.  The caller retains ownership of 'fm' and everything in
 357  * it. */
 358 void
 359 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 360                       struct ofputil_flow_mod *fm)
 361 {
 362     ofproto_flow_mod(&ofproto->up, fm);
 363 }
 364
 365 /* Resets the modified time for 'rule' or an equivalent rule. If 'rule' is not
 366  * in the classifier, but an equivalent rule is, unref 'rule' and ref the new
 367  * rule. Otherwise if 'rule' is no longer installed in the classifier,
 368  * reinstall it.
 369  *
 370  * Returns the rule whose modified time has been reset. */
 371 struct rule_dpif *
 372 ofproto_dpif_refresh_rule(struct rule_dpif *rule)
 373 {
 374     return rule_dpif_cast(ofproto_refresh_rule(&rule->up));
 375 }
 376
 377 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 378  * Takes ownership of 'pin' and pin->packet. */
 379 void
 380 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 381                             struct ofproto_packet_in *pin)
 382 {
 383     if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
 384         COVERAGE_INC(packet_in_overflow);
 385         free(CONST_CAST(void *, pin->up.packet));
 386         free(pin);
 387     }
 388 }
 389
 390 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
 391  * packet rather than to send the packet to the controller.
 392  *
 393  * This function returns false to indicate that a packet_in message
 394  * for a "table-miss" should be sent to at least one controller.
 395  * False otherwise. */
 396 bool
 397 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif *ofproto)
 398 {
 399     return connmgr_wants_packet_in_on_miss(ofproto->up.connmgr);
 400 }
 401 \f
 402 /* Factory functions. */
 403
 404 static void
 405 init(const struct shash *iface_hints)
 406 {
 407     struct shash_node *node;
 408
 409     /* Make a local copy, since we don't own 'iface_hints' elements. */
 410     SHASH_FOR_EACH(node, iface_hints) {
 411         const struct iface_hint *orig_hint = node->data;
 412         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 413
 414         new_hint->br_name = xstrdup(orig_hint->br_name);
 415         new_hint->br_type = xstrdup(orig_hint->br_type);
 416         new_hint->ofp_port = orig_hint->ofp_port;
 417
 418         shash_add(&init_ofp_ports, node->name, new_hint);
 419     }
 420 }
 421
 422 static void
 423 enumerate_types(struct sset *types)
 424 {
 425     dp_enumerate_types(types);
 426 }
 427
 428 static int
 429 enumerate_names(const char *type, struct sset *names)
 430 {
 431     struct ofproto_dpif *ofproto;
 432
 433     sset_clear(names);
 434     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 435         if (strcmp(type, ofproto->up.type)) {
 436             continue;
 437         }
 438         sset_add(names, ofproto->up.name);
 439     }
 440
 441     return 0;
 442 }
 443
 444 static int
 445 del(const char *type, const char *name)
 446 {
 447     struct dpif *dpif;
 448     int error;
 449
 450     error = dpif_open(name, type, &dpif);
 451     if (!error) {
 452         error = dpif_delete(dpif);
 453         dpif_close(dpif);
 454     }
 455     return error;
 456 }
 457 \f
 458 static const char *
 459 port_open_type(const char *datapath_type, const char *port_type)
 460 {
 461     return dpif_port_open_type(datapath_type, port_type);
 462 }
 463
 464 /* Type functions. */
 465
 466 static void process_dpif_port_changes(struct dpif_backer *);
 467 static void process_dpif_all_ports_changed(struct dpif_backer *);
 468 static void process_dpif_port_change(struct dpif_backer *,
 469                                      const char *devname);
 470 static void process_dpif_port_error(struct dpif_backer *, int error);
 471
 472 static struct ofproto_dpif *
 473 lookup_ofproto_dpif_by_port_name(const char *name)
 474 {
 475     struct ofproto_dpif *ofproto;
 476
 477     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 478         if (sset_contains(&ofproto->ports, name)) {
 479             return ofproto;
 480         }
 481     }
 482
 483     return NULL;
 484 }
 485
 486 static int
 487 type_run(const char *type)
 488 {
 489     struct dpif_backer *backer;
 490
 491     backer = shash_find_data(&all_dpif_backers, type);
 492     if (!backer) {
 493         /* This is not necessarily a problem, since backers are only
 494          * created on demand. */
 495         return 0;
 496     }
 497
 498     dpif_run(backer->dpif);
 499
 500     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 501      * and the configuration has now changed to "false", enable receiving
 502      * packets from the datapath. */
 503     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 504         int error;
 505
 506         backer->recv_set_enable = true;
 507
 508         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 509         if (error) {
 510             VLOG_ERR("Failed to enable receiving packets in dpif.");
 511             return error;
 512         }
 513         dpif_flow_flush(backer->dpif);
 514         backer->need_revalidate = REV_RECONFIGURE;
 515     }
 516
 517     if (backer->recv_set_enable) {
 518         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 519     }
 520
 521     if (backer->need_revalidate) {
 522         struct ofproto_dpif *ofproto;
 523         struct simap_node *node;
 524         struct simap tmp_backers;
 525
 526         /* Handle tunnel garbage collection. */
 527         simap_init(&tmp_backers);
 528         simap_swap(&backer->tnl_backers, &tmp_backers);
 529
 530         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 531             struct ofport_dpif *iter;
 532
 533             if (backer != ofproto->backer) {
 534                 continue;
 535             }
 536
 537             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 538                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 539                 const char *dp_port;
 540
 541                 if (!iter->is_tunnel) {
 542                     continue;
 543                 }
 544
 545                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 546                                                      namebuf, sizeof namebuf);
 547                 node = simap_find(&tmp_backers, dp_port);
 548                 if (node) {
 549                     simap_put(&backer->tnl_backers, dp_port, node->data);
 550                     simap_delete(&tmp_backers, node);
 551                     node = simap_find(&backer->tnl_backers, dp_port);
 552                 } else {
 553                     node = simap_find(&backer->tnl_backers, dp_port);
 554                     if (!node) {
 555                         odp_port_t odp_port = ODPP_NONE;
 556
 557                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 558                                            &odp_port)) {
 559                             simap_put(&backer->tnl_backers, dp_port,
 560                                       odp_to_u32(odp_port));
 561                             node = simap_find(&backer->tnl_backers, dp_port);
 562                         }
 563                     }
 564                 }
 565
 566                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 567                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 568                                          iter->odp_port)) {
 569                     backer->need_revalidate = REV_RECONFIGURE;
 570                 }
 571             }
 572         }
 573
 574         SIMAP_FOR_EACH (node, &tmp_backers) {
 575             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 576         }
 577         simap_destroy(&tmp_backers);
 578
 579         switch (backer->need_revalidate) {
 580         case REV_RECONFIGURE:   COVERAGE_INC(rev_reconfigure);   break;
 581         case REV_STP:           COVERAGE_INC(rev_stp);           break;
 582         case REV_BOND:          COVERAGE_INC(rev_bond);          break;
 583         case REV_PORT_TOGGLED:  COVERAGE_INC(rev_port_toggled);  break;
 584         case REV_FLOW_TABLE:    COVERAGE_INC(rev_flow_table);    break;
 585         case REV_MAC_LEARNING:  COVERAGE_INC(rev_mac_learning);  break;
 586         }
 587         backer->need_revalidate = 0;
 588
 589         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 590             struct ofport_dpif *ofport;
 591             struct ofbundle *bundle;
 592
 593             if (ofproto->backer != backer) {
 594                 continue;
 595             }
 596
 597             ovs_rwlock_wrlock(&xlate_rwlock);
 598             xlate_ofproto_set(ofproto, ofproto->up.name,
 599                               ofproto->backer->dpif, ofproto->miss_rule,
 600                               ofproto->no_packet_in_rule, ofproto->ml,
 601                               ofproto->stp, ofproto->mbridge,
 602                               ofproto->sflow, ofproto->ipfix,
 603                               ofproto->netflow, ofproto->up.frag_handling,
 604                               ofproto->up.forward_bpdu,
 605                               connmgr_has_in_band(ofproto->up.connmgr),
 606                               ofproto->backer->enable_recirc,
 607                               ofproto->backer->variable_length_userdata,
 608                               ofproto->backer->max_mpls_depth);
 609
 610             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 611                 xlate_bundle_set(ofproto, bundle, bundle->name,
 612                                  bundle->vlan_mode, bundle->vlan,
 613                                  bundle->trunks, bundle->use_priority_tags,
 614                                  bundle->bond, bundle->lacp,
 615                                  bundle->floodable);
 616             }
 617
 618             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 619                 int stp_port = ofport->stp_port
 620                     ? stp_port_no(ofport->stp_port)
 621                     : -1;
 622                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 623                                  ofport->up.ofp_port, ofport->odp_port,
 624                                  ofport->up.netdev, ofport->cfm,
 625                                  ofport->bfd, ofport->peer, stp_port,
 626                                  ofport->qdscp, ofport->n_qdscp,
 627                                  ofport->up.pp.config, ofport->up.pp.state,
 628                                  ofport->is_tunnel, ofport->may_enable);
 629             }
 630             ovs_rwlock_unlock(&xlate_rwlock);
 631         }
 632
 633         udpif_revalidate(backer->udpif);
 634     }
 635
 636     process_dpif_port_changes(backer);
 637
 638     return 0;
 639 }
 640
 641 /* Check for and handle port changes in 'backer''s dpif. */
 642 static void
 643 process_dpif_port_changes(struct dpif_backer *backer)
 644 {
 645     for (;;) {
 646         char *devname;
 647         int error;
 648
 649         error = dpif_port_poll(backer->dpif, &devname);
 650         switch (error) {
 651         case EAGAIN:
 652             return;
 653
 654         case ENOBUFS:
 655             process_dpif_all_ports_changed(backer);
 656             break;
 657
 658         case 0:
 659             process_dpif_port_change(backer, devname);
 660             free(devname);
 661             break;
 662
 663         default:
 664             process_dpif_port_error(backer, error);
 665             break;
 666         }
 667     }
 668 }
 669
 670 static void
 671 process_dpif_all_ports_changed(struct dpif_backer *backer)
 672 {
 673     struct ofproto_dpif *ofproto;
 674     struct dpif_port dpif_port;
 675     struct dpif_port_dump dump;
 676     struct sset devnames;
 677     const char *devname;
 678
 679     sset_init(&devnames);
 680     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 681         if (ofproto->backer == backer) {
 682             struct ofport *ofport;
 683
 684             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 685                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 686             }
 687         }
 688     }
 689     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 690         sset_add(&devnames, dpif_port.name);
 691     }
 692
 693     SSET_FOR_EACH (devname, &devnames) {
 694         process_dpif_port_change(backer, devname);
 695     }
 696     sset_destroy(&devnames);
 697 }
 698
 699 static void
 700 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 701 {
 702     struct ofproto_dpif *ofproto;
 703     struct dpif_port port;
 704
 705     /* Don't report on the datapath's device. */
 706     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 707         return;
 708     }
 709
 710     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 711                    &all_ofproto_dpifs) {
 712         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 713             return;
 714         }
 715     }
 716
 717     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 718     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 719         /* The port was removed.  If we know the datapath,
 720          * report it through poll_set().  If we don't, it may be
 721          * notifying us of a removal we initiated, so ignore it.
 722          * If there's a pending ENOBUFS, let it stand, since
 723          * everything will be reevaluated. */
 724         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 725             sset_add(&ofproto->port_poll_set, devname);
 726             ofproto->port_poll_errno = 0;
 727         }
 728     } else if (!ofproto) {
 729         /* The port was added, but we don't know with which
 730          * ofproto we should associate it.  Delete it. */
 731         dpif_port_del(backer->dpif, port.port_no);
 732     } else {
 733         struct ofport_dpif *ofport;
 734
 735         ofport = ofport_dpif_cast(shash_find_data(
 736                                       &ofproto->up.port_by_name, devname));
 737         if (ofport
 738             && ofport->odp_port != port.port_no
 739             && !odp_port_to_ofport(backer, port.port_no))
 740         {
 741             /* 'ofport''s datapath port number has changed from
 742              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 743              * structures to match. */
 744             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 745             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 746             ofport->odp_port = port.port_no;
 747             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 748                         hash_odp_port(port.port_no));
 749             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 750             backer->need_revalidate = REV_RECONFIGURE;
 751         }
 752     }
 753     dpif_port_destroy(&port);
 754 }
 755
 756 /* Propagate 'error' to all ofprotos based on 'backer'. */
 757 static void
 758 process_dpif_port_error(struct dpif_backer *backer, int error)
 759 {
 760     struct ofproto_dpif *ofproto;
 761
 762     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 763         if (ofproto->backer == backer) {
 764             sset_clear(&ofproto->port_poll_set);
 765             ofproto->port_poll_errno = error;
 766         }
 767     }
 768 }
 769
 770 static void
 771 type_wait(const char *type)
 772 {
 773     struct dpif_backer *backer;
 774
 775     backer = shash_find_data(&all_dpif_backers, type);
 776     if (!backer) {
 777         /* This is not necessarily a problem, since backers are only
 778          * created on demand. */
 779         return;
 780     }
 781
 782     dpif_wait(backer->dpif);
 783 }
 784 \f
 785 /* Basic life-cycle. */
 786
 787 static int add_internal_flows(struct ofproto_dpif *);
 788
 789 static struct ofproto *
 790 alloc(void)
 791 {
 792     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
 793     return &ofproto->up;
 794 }
 795
 796 static void
 797 dealloc(struct ofproto *ofproto_)
 798 {
 799     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
 800     free(ofproto);
 801 }
 802
 803 static void
 804 close_dpif_backer(struct dpif_backer *backer)
 805 {
 806     ovs_assert(backer->refcount > 0);
 807
 808     if (--backer->refcount) {
 809         return;
 810     }
 811
 812     udpif_destroy(backer->udpif);
 813
 814     simap_destroy(&backer->tnl_backers);
 815     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
 816     hmap_destroy(&backer->odp_to_ofport_map);
 817     shash_find_and_delete(&all_dpif_backers, backer->type);
 818     recirc_id_pool_destroy(backer->rid_pool);
 819     free(backer->type);
 820     dpif_close(backer->dpif);
 821     free(backer);
 822 }
 823
 824 /* Datapath port slated for removal from datapath. */
 825 struct odp_garbage {
 826     struct list list_node;
 827     odp_port_t odp_port;
 828 };
 829
 830 static bool check_variable_length_userdata(struct dpif_backer *backer);
 831 static size_t check_max_mpls_depth(struct dpif_backer *backer);
 832 static bool check_recirc(struct dpif_backer *backer);
 833
 834 static int
 835 open_dpif_backer(const char *type, struct dpif_backer **backerp)
 836 {
 837     struct dpif_backer *backer;
 838     struct dpif_port_dump port_dump;
 839     struct dpif_port port;
 840     struct shash_node *node;
 841     struct list garbage_list;
 842     struct odp_garbage *garbage, *next;
 843
 844     struct sset names;
 845     char *backer_name;
 846     const char *name;
 847     int error;
 848
 849     backer = shash_find_data(&all_dpif_backers, type);
 850     if (backer) {
 851         backer->refcount++;
 852         *backerp = backer;
 853         return 0;
 854     }
 855
 856     backer_name = xasprintf("ovs-%s", type);
 857
 858     /* Remove any existing datapaths, since we assume we're the only
 859      * userspace controlling the datapath. */
 860     sset_init(&names);
 861     dp_enumerate_names(type, &names);
 862     SSET_FOR_EACH(name, &names) {
 863         struct dpif *old_dpif;
 864
 865         /* Don't remove our backer if it exists. */
 866         if (!strcmp(name, backer_name)) {
 867             continue;
 868         }
 869
 870         if (dpif_open(name, type, &old_dpif)) {
 871             VLOG_WARN("couldn't open old datapath %s to remove it", name);
 872         } else {
 873             dpif_delete(old_dpif);
 874             dpif_close(old_dpif);
 875         }
 876     }
 877     sset_destroy(&names);
 878
 879     backer = xmalloc(sizeof *backer);
 880
 881     error = dpif_create_and_open(backer_name, type, &backer->dpif);
 882     free(backer_name);
 883     if (error) {
 884         VLOG_ERR("failed to open datapath of type %s: %s", type,
 885                  ovs_strerror(error));
 886         free(backer);
 887         return error;
 888     }
 889     backer->udpif = udpif_create(backer, backer->dpif);
 890
 891     backer->type = xstrdup(type);
 892     backer->refcount = 1;
 893     hmap_init(&backer->odp_to_ofport_map);
 894     ovs_rwlock_init(&backer->odp_to_ofport_lock);
 895     backer->need_revalidate = 0;
 896     simap_init(&backer->tnl_backers);
 897     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
 898     *backerp = backer;
 899
 900     if (backer->recv_set_enable) {
 901         dpif_flow_flush(backer->dpif);
 902     }
 903
 904     /* Loop through the ports already on the datapath and remove any
 905      * that we don't need anymore. */
 906     list_init(&garbage_list);
 907     dpif_port_dump_start(&port_dump, backer->dpif);
 908     while (dpif_port_dump_next(&port_dump, &port)) {
 909         node = shash_find(&init_ofp_ports, port.name);
 910         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
 911             garbage = xmalloc(sizeof *garbage);
 912             garbage->odp_port = port.port_no;
 913             list_push_front(&garbage_list, &garbage->list_node);
 914         }
 915     }
 916     dpif_port_dump_done(&port_dump);
 917
 918     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
 919         dpif_port_del(backer->dpif, garbage->odp_port);
 920         list_remove(&garbage->list_node);
 921         free(garbage);
 922     }
 923
 924     shash_add(&all_dpif_backers, type, backer);
 925
 926     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 927     if (error) {
 928         VLOG_ERR("failed to listen on datapath of type %s: %s",
 929                  type, ovs_strerror(error));
 930         close_dpif_backer(backer);
 931         return error;
 932     }
 933     backer->enable_recirc = check_recirc(backer);
 934     backer->variable_length_userdata = check_variable_length_userdata(backer);
 935     backer->max_mpls_depth = check_max_mpls_depth(backer);
 936     backer->rid_pool = recirc_id_pool_create();
 937
 938     if (backer->recv_set_enable) {
 939         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 940     }
 941
 942     return error;
 943 }
 944
 945 /* Tests whether 'backer''s datapath supports recirculation Only newer datapath
 946  * supports OVS_KEY_ATTR in OVS_ACTION_ATTR_USERSPACE actions.  We need to
 947  * disable some features on older datapaths that don't support this feature.
 948  *
 949  * Returns false if 'backer' definitely does not support recirculation, true if
 950  * it seems to support recirculation or if at least the error we get is
 951  * ambiguous. */
 952 static bool
 953 check_recirc(struct dpif_backer *backer)
 954 {
 955     struct flow flow;
 956     struct odputil_keybuf keybuf;
 957     struct ofpbuf key;
 958     int error;
 959     bool enable_recirc = false;
 960
 961     memset(&flow, 0, sizeof flow);
 962     flow.recirc_id = 1;
 963     flow.dp_hash = 1;
 964
 965     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
 966     odp_flow_key_from_flow(&key, &flow, NULL, 0);
 967
 968     error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
 969                           ofpbuf_data(&key), ofpbuf_size(&key), NULL, 0, NULL,
 970                           0, NULL);
 971     if (error && error != EEXIST) {
 972         if (error != EINVAL) {
 973             VLOG_WARN("%s: Reciculation flow probe failed (%s)",
 974                       dpif_name(backer->dpif), ovs_strerror(error));
 975         }
 976         goto done;
 977     }
 978
 979     error = dpif_flow_del(backer->dpif, ofpbuf_data(&key), ofpbuf_size(&key),
 980                           NULL);
 981     if (error) {
 982         VLOG_WARN("%s: failed to delete recirculation feature probe flow",
 983                   dpif_name(backer->dpif));
 984     }
 985
 986     enable_recirc = true;
 987
 988 done:
 989     if (enable_recirc) {
 990         VLOG_INFO("%s: Datapath supports recirculation",
 991                   dpif_name(backer->dpif));
 992     } else {
 993         VLOG_INFO("%s: Datapath does not support recirculation",
 994                   dpif_name(backer->dpif));
 995     }
 996
 997     return enable_recirc;
 998 }
 999
1000 /* Tests whether 'backer''s datapath supports variable-length
1001  * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.  We need
1002  * to disable some features on older datapaths that don't support this
1003  * feature.
1004  *
1005  * Returns false if 'backer' definitely does not support variable-length
1006  * userdata, true if it seems to support them or if at least the error we get
1007  * is ambiguous. */
1008 static bool
1009 check_variable_length_userdata(struct dpif_backer *backer)
1010 {
1011     struct eth_header *eth;
1012     struct ofpbuf actions;
1013     struct dpif_execute execute;
1014     struct ofpbuf packet;
1015     size_t start;
1016     int error;
1017
1018     /* Compose a userspace action that will cause an ERANGE error on older
1019      * datapaths that don't support variable-length userdata.
1020      *
1021      * We really test for using userdata longer than 8 bytes, but older
1022      * datapaths accepted these, silently truncating the userdata to 8 bytes.
1023      * The same older datapaths rejected userdata shorter than 8 bytes, so we
1024      * test for that instead as a proxy for longer userdata support. */
1025     ofpbuf_init(&actions, 64);
1026     start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
1027     nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
1028                    dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
1029     nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
1030     nl_msg_end_nested(&actions, start);
1031
1032     /* Compose a dummy ethernet packet. */
1033     ofpbuf_init(&packet, ETH_HEADER_LEN);
1034     eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
1035     eth->eth_type = htons(0x1234);
1036
1037     /* Execute the actions.  On older datapaths this fails with ERANGE, on
1038      * newer datapaths it succeeds. */
1039     execute.actions = ofpbuf_data(&actions);
1040     execute.actions_len = ofpbuf_size(&actions);
1041     execute.packet = &packet;
1042     execute.md = PKT_METADATA_INITIALIZER(0);
1043     execute.needs_help = false;
1044
1045     error = dpif_execute(backer->dpif, &execute);
1046
1047     ofpbuf_uninit(&packet);
1048     ofpbuf_uninit(&actions);
1049
1050     switch (error) {
1051     case 0:
1052         /* Variable-length userdata is supported.
1053          *
1054          * Purge received packets to avoid processing the nonsense packet we
1055          * sent to userspace, then report success. */
1056         dpif_recv_purge(backer->dpif);
1057         return true;
1058
1059     case ERANGE:
1060         /* Variable-length userdata is not supported. */
1061         VLOG_WARN("%s: datapath does not support variable-length userdata "
1062                   "feature (needs Linux 3.10+ or kernel module from OVS "
1063                   "1..11+).  The NXAST_SAMPLE action will be ignored.",
1064                   dpif_name(backer->dpif));
1065         return false;
1066
1067     default:
1068         /* Something odd happened.  We're not sure whether variable-length
1069          * userdata is supported.  Default to "yes". */
1070         VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1071                   dpif_name(backer->dpif), ovs_strerror(error));
1072         return true;
1073     }
1074 }
1075
1076 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1077  *
1078  * Returns the number of elements in a struct flow's mpls_lse field
1079  * if the datapath supports at least that many entries in an
1080  * MPLS label stack.
1081  * Otherwise returns the number of MPLS push actions supported by
1082  * the datapath. */
1083 static size_t
1084 check_max_mpls_depth(struct dpif_backer *backer)
1085 {
1086     struct flow flow;
1087     int n;
1088
1089     for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
1090         struct odputil_keybuf keybuf;
1091         struct ofpbuf key;
1092         int error;
1093
1094         memset(&flow, 0, sizeof flow);
1095         flow.dl_type = htons(ETH_TYPE_MPLS);
1096         flow_set_mpls_bos(&flow, n, 1);
1097
1098         ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1099         odp_flow_key_from_flow(&key, &flow, NULL, 0);
1100
1101         error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
1102                               ofpbuf_data(&key), ofpbuf_size(&key), NULL, 0, NULL, 0, NULL);
1103         if (error && error != EEXIST) {
1104             if (error != EINVAL) {
1105                 VLOG_WARN("%s: MPLS stack length feature probe failed (%s)",
1106                           dpif_name(backer->dpif), ovs_strerror(error));
1107             }
1108             break;
1109         }
1110
1111         error = dpif_flow_del(backer->dpif, ofpbuf_data(&key), ofpbuf_size(&key), NULL);
1112         if (error) {
1113             VLOG_WARN("%s: failed to delete MPLS feature probe flow",
1114                       dpif_name(backer->dpif));
1115         }
1116     }
1117
1118     VLOG_INFO("%s: MPLS label stack length probed as %d",
1119               dpif_name(backer->dpif), n);
1120     return n;
1121 }
1122
1123 static int
1124 construct(struct ofproto *ofproto_)
1125 {
1126     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1127     struct shash_node *node, *next;
1128     int error;
1129
1130     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1131     if (error) {
1132         return error;
1133     }
1134
1135     ofproto->netflow = NULL;
1136     ofproto->sflow = NULL;
1137     ofproto->ipfix = NULL;
1138     ofproto->stp = NULL;
1139     ofproto->dump_seq = 0;
1140     hmap_init(&ofproto->bundles);
1141     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1142     ofproto->mbridge = mbridge_create();
1143     ofproto->has_bonded_bundles = false;
1144     ofproto->lacp_enabled = false;
1145     ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1146     ovs_mutex_init(&ofproto->vsp_mutex);
1147
1148     guarded_list_init(&ofproto->pins);
1149
1150     ofproto_dpif_unixctl_init();
1151
1152     hmap_init(&ofproto->vlandev_map);
1153     hmap_init(&ofproto->realdev_vid_map);
1154
1155     sset_init(&ofproto->ports);
1156     sset_init(&ofproto->ghost_ports);
1157     sset_init(&ofproto->port_poll_set);
1158     ofproto->port_poll_errno = 0;
1159     ofproto->change_seq = 0;
1160
1161     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1162         struct iface_hint *iface_hint = node->data;
1163
1164         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1165             /* Check if the datapath already has this port. */
1166             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1167                 sset_add(&ofproto->ports, node->name);
1168             }
1169
1170             free(iface_hint->br_name);
1171             free(iface_hint->br_type);
1172             free(iface_hint);
1173             shash_delete(&init_ofp_ports, node);
1174         }
1175     }
1176
1177     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1178                 hash_string(ofproto->up.name, 0));
1179     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1180
1181     ofproto_init_tables(ofproto_, N_TABLES);
1182     error = add_internal_flows(ofproto);
1183
1184     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1185
1186     return error;
1187 }
1188
1189 static int
1190 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1191                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1192 {
1193     struct match match;
1194     int error;
1195     struct rule *rule;
1196
1197     match_init_catchall(&match);
1198     match_set_reg(&match, 0, id);
1199
1200     error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, ofpacts, &rule);
1201     *rulep = error ? NULL : rule_dpif_cast(rule);
1202
1203     return error;
1204 }
1205
1206 static int
1207 add_internal_flows(struct ofproto_dpif *ofproto)
1208 {
1209     struct ofpact_controller *controller;
1210     uint64_t ofpacts_stub[128 / 8];
1211     struct ofpbuf ofpacts;
1212     struct rule *unused_rulep OVS_UNUSED;
1213     struct ofpact_resubmit *resubmit;
1214     struct match match;
1215     int error;
1216     int id;
1217
1218     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1219     id = 1;
1220
1221     controller = ofpact_put_CONTROLLER(&ofpacts);
1222     controller->max_len = UINT16_MAX;
1223     controller->controller_id = 0;
1224     controller->reason = OFPR_NO_MATCH;
1225     ofpact_pad(&ofpacts);
1226
1227     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1228                                    &ofproto->miss_rule);
1229     if (error) {
1230         return error;
1231     }
1232
1233     ofpbuf_clear(&ofpacts);
1234     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1235                               &ofproto->no_packet_in_rule);
1236     if (error) {
1237         return error;
1238     }
1239
1240     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1241                               &ofproto->drop_frags_rule);
1242     if (error) {
1243         return error;
1244     }
1245
1246     /* Continue non-recirculation rule lookups from table 0.
1247      *
1248      * (priority=2), recirc=0, actions=resubmit(, 0)
1249      */
1250     resubmit = ofpact_put_RESUBMIT(&ofpacts);
1251     resubmit->ofpact.compat = 0;
1252     resubmit->in_port = OFPP_IN_PORT;
1253     resubmit->table_id = 0;
1254
1255     match_init_catchall(&match);
1256     match_set_recirc_id(&match, 0);
1257
1258     error = ofproto_dpif_add_internal_flow(ofproto, &match, 2,  &ofpacts,
1259                                            &unused_rulep);
1260     if (error) {
1261         return error;
1262     }
1263
1264     /* Drop any run away recirc rule lookups. Recirc_id has to be
1265      * non-zero when reaching this rule.
1266      *
1267      * (priority=1), *, actions=drop
1268      */
1269     ofpbuf_clear(&ofpacts);
1270     match_init_catchall(&match);
1271     error = ofproto_dpif_add_internal_flow(ofproto, &match, 1,  &ofpacts,
1272                                            &unused_rulep);
1273
1274     return error;
1275 }
1276
1277 static void
1278 destruct(struct ofproto *ofproto_)
1279 {
1280     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1281     struct rule_dpif *rule, *next_rule;
1282     struct ofproto_packet_in *pin, *next_pin;
1283     struct oftable *table;
1284     struct list pins;
1285
1286     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1287     ovs_rwlock_wrlock(&xlate_rwlock);
1288     xlate_remove_ofproto(ofproto);
1289     ovs_rwlock_unlock(&xlate_rwlock);
1290
1291     /* Ensure that the upcall processing threads have no remaining references
1292      * to the ofproto or anything in it. */
1293     udpif_synchronize(ofproto->backer->udpif);
1294
1295     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1296
1297     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1298         struct cls_cursor cursor;
1299
1300         fat_rwlock_rdlock(&table->cls.rwlock);
1301         cls_cursor_init(&cursor, &table->cls, NULL);
1302         fat_rwlock_unlock(&table->cls.rwlock);
1303         CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1304             ofproto_rule_delete(&ofproto->up, &rule->up);
1305         }
1306     }
1307
1308     guarded_list_pop_all(&ofproto->pins, &pins);
1309     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1310         list_remove(&pin->list_node);
1311         free(CONST_CAST(void *, pin->up.packet));
1312         free(pin);
1313     }
1314     guarded_list_destroy(&ofproto->pins);
1315
1316     mbridge_unref(ofproto->mbridge);
1317
1318     netflow_unref(ofproto->netflow);
1319     dpif_sflow_unref(ofproto->sflow);
1320     hmap_destroy(&ofproto->bundles);
1321     mac_learning_unref(ofproto->ml);
1322
1323     hmap_destroy(&ofproto->vlandev_map);
1324     hmap_destroy(&ofproto->realdev_vid_map);
1325
1326     sset_destroy(&ofproto->ports);
1327     sset_destroy(&ofproto->ghost_ports);
1328     sset_destroy(&ofproto->port_poll_set);
1329
1330     ovs_mutex_destroy(&ofproto->stats_mutex);
1331     ovs_mutex_destroy(&ofproto->vsp_mutex);
1332
1333     close_dpif_backer(ofproto->backer);
1334 }
1335
1336 static int
1337 run(struct ofproto *ofproto_)
1338 {
1339     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1340     uint64_t new_seq, new_dump_seq;
1341     const bool enable_recirc = ofproto_dpif_get_enable_recirc(ofproto);
1342
1343     if (mbridge_need_revalidate(ofproto->mbridge)) {
1344         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1345         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1346         mac_learning_flush(ofproto->ml);
1347         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1348     }
1349
1350     /* Do not perform any periodic activity required by 'ofproto' while
1351      * waiting for flow restore to complete. */
1352     if (!ofproto_get_flow_restore_wait()) {
1353         struct ofproto_packet_in *pin, *next_pin;
1354         struct list pins;
1355
1356         guarded_list_pop_all(&ofproto->pins, &pins);
1357         LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1358             connmgr_send_packet_in(ofproto->up.connmgr, pin);
1359             list_remove(&pin->list_node);
1360             free(CONST_CAST(void *, pin->up.packet));
1361             free(pin);
1362         }
1363     }
1364
1365     if (ofproto->netflow) {
1366         netflow_run(ofproto->netflow);
1367     }
1368     if (ofproto->sflow) {
1369         dpif_sflow_run(ofproto->sflow);
1370     }
1371     if (ofproto->ipfix) {
1372         dpif_ipfix_run(ofproto->ipfix);
1373     }
1374
1375     new_seq = seq_read(connectivity_seq_get());
1376     if (ofproto->change_seq != new_seq) {
1377         struct ofport_dpif *ofport;
1378
1379         HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1380             port_run(ofport);
1381         }
1382
1383         ofproto->change_seq = new_seq;
1384     }
1385     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1386         struct ofbundle *bundle;
1387
1388         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1389             bundle_run(bundle);
1390         }
1391     }
1392
1393     stp_run(ofproto);
1394     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1395     if (mac_learning_run(ofproto->ml)) {
1396         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1397     }
1398     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1399
1400     new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1401     if (ofproto->dump_seq != new_dump_seq) {
1402         struct rule *rule, *next_rule;
1403
1404         /* We know stats are relatively fresh, so now is a good time to do some
1405          * periodic work. */
1406         ofproto->dump_seq = new_dump_seq;
1407
1408         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1409          * has passed. */
1410         ovs_mutex_lock(&ofproto_mutex);
1411         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1412                             &ofproto->up.expirable) {
1413             rule_expire(rule_dpif_cast(rule));
1414         }
1415         ovs_mutex_unlock(&ofproto_mutex);
1416
1417         /* All outstanding data in existing flows has been accounted, so it's a
1418          * good time to do bond rebalancing. */
1419         if (enable_recirc && ofproto->has_bonded_bundles) {
1420             struct ofbundle *bundle;
1421
1422             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1423                 struct bond *bond = bundle->bond;
1424
1425                 if (bond && bond_may_recirc(bond, NULL, NULL)) {
1426                     bond_recirculation_account(bond);
1427                     if (bond_rebalance(bundle->bond)) {
1428                         bond_update_post_recirc_rules(bond, true);
1429                     }
1430                 }
1431             }
1432         }
1433     }
1434
1435     return 0;
1436 }
1437
1438 static void
1439 wait(struct ofproto *ofproto_)
1440 {
1441     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1442
1443     if (ofproto_get_flow_restore_wait()) {
1444         return;
1445     }
1446
1447     if (ofproto->sflow) {
1448         dpif_sflow_wait(ofproto->sflow);
1449     }
1450     if (ofproto->ipfix) {
1451         dpif_ipfix_wait(ofproto->ipfix);
1452     }
1453     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1454         struct ofbundle *bundle;
1455
1456         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1457             bundle_wait(bundle);
1458         }
1459     }
1460     if (ofproto->netflow) {
1461         netflow_wait(ofproto->netflow);
1462     }
1463     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1464     mac_learning_wait(ofproto->ml);
1465     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1466     stp_wait(ofproto);
1467     if (ofproto->backer->need_revalidate) {
1468         /* Shouldn't happen, but if it does just go around again. */
1469         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1470         poll_immediate_wake();
1471     }
1472
1473     seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1474 }
1475
1476 static void
1477 type_get_memory_usage(const char *type, struct simap *usage)
1478 {
1479     struct dpif_backer *backer;
1480
1481     backer = shash_find_data(&all_dpif_backers, type);
1482     if (backer) {
1483         udpif_get_memory_usage(backer->udpif, usage);
1484     }
1485 }
1486
1487 static void
1488 flush(struct ofproto *ofproto_)
1489 {
1490     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1491     struct dpif_backer *backer = ofproto->backer;
1492
1493     if (backer) {
1494         udpif_flush(backer->udpif);
1495     }
1496 }
1497
1498 static void
1499 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1500              bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1501 {
1502     *arp_match_ip = true;
1503     *actions = (OFPUTIL_A_OUTPUT |
1504                 OFPUTIL_A_SET_VLAN_VID |
1505                 OFPUTIL_A_SET_VLAN_PCP |
1506                 OFPUTIL_A_STRIP_VLAN |
1507                 OFPUTIL_A_SET_DL_SRC |
1508                 OFPUTIL_A_SET_DL_DST |
1509                 OFPUTIL_A_SET_NW_SRC |
1510                 OFPUTIL_A_SET_NW_DST |
1511                 OFPUTIL_A_SET_NW_TOS |
1512                 OFPUTIL_A_SET_TP_SRC |
1513                 OFPUTIL_A_SET_TP_DST |
1514                 OFPUTIL_A_ENQUEUE);
1515 }
1516
1517 static void
1518 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1519 {
1520     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1521     struct dpif_dp_stats s;
1522     uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1523     uint64_t n_lookup;
1524     long long int used;
1525
1526     strcpy(ots->name, "classifier");
1527
1528     dpif_get_dp_stats(ofproto->backer->dpif, &s);
1529     rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes, &used);
1530     rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes,
1531                    &used);
1532     rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes,
1533                    &used);
1534     n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1535     ots->lookup_count = htonll(n_lookup);
1536     ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1537 }
1538
1539 static struct ofport *
1540 port_alloc(void)
1541 {
1542     struct ofport_dpif *port = xmalloc(sizeof *port);
1543     return &port->up;
1544 }
1545
1546 static void
1547 port_dealloc(struct ofport *port_)
1548 {
1549     struct ofport_dpif *port = ofport_dpif_cast(port_);
1550     free(port);
1551 }
1552
1553 static int
1554 port_construct(struct ofport *port_)
1555 {
1556     struct ofport_dpif *port = ofport_dpif_cast(port_);
1557     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1558     const struct netdev *netdev = port->up.netdev;
1559     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1560     struct dpif_port dpif_port;
1561     int error;
1562
1563     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1564     port->bundle = NULL;
1565     port->cfm = NULL;
1566     port->bfd = NULL;
1567     port->may_enable = true;
1568     port->stp_port = NULL;
1569     port->stp_state = STP_DISABLED;
1570     port->is_tunnel = false;
1571     port->peer = NULL;
1572     port->qdscp = NULL;
1573     port->n_qdscp = 0;
1574     port->realdev_ofp_port = 0;
1575     port->vlandev_vid = 0;
1576     port->carrier_seq = netdev_get_carrier_resets(netdev);
1577     port->is_layer3 = netdev_vport_is_layer3(netdev);
1578
1579     if (netdev_vport_is_patch(netdev)) {
1580         /* By bailing out here, we don't submit the port to the sFlow module
1581          * to be considered for counter polling export.  This is correct
1582          * because the patch port represents an interface that sFlow considers
1583          * to be "internal" to the switch as a whole, and therefore not an
1584          * candidate for counter polling. */
1585         port->odp_port = ODPP_NONE;
1586         ofport_update_peer(port);
1587         return 0;
1588     }
1589
1590     error = dpif_port_query_by_name(ofproto->backer->dpif,
1591                                     netdev_vport_get_dpif_port(netdev, namebuf,
1592                                                                sizeof namebuf),
1593                                     &dpif_port);
1594     if (error) {
1595         return error;
1596     }
1597
1598     port->odp_port = dpif_port.port_no;
1599
1600     if (netdev_get_tunnel_config(netdev)) {
1601         tnl_port_add(port, port->up.netdev, port->odp_port);
1602         port->is_tunnel = true;
1603     } else {
1604         /* Sanity-check that a mapping doesn't already exist.  This
1605          * shouldn't happen for non-tunnel ports. */
1606         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1607             VLOG_ERR("port %s already has an OpenFlow port number",
1608                      dpif_port.name);
1609             dpif_port_destroy(&dpif_port);
1610             return EBUSY;
1611         }
1612
1613         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1614         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1615                     hash_odp_port(port->odp_port));
1616         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1617     }
1618     dpif_port_destroy(&dpif_port);
1619
1620     if (ofproto->sflow) {
1621         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1622     }
1623
1624     return 0;
1625 }
1626
1627 static void
1628 port_destruct(struct ofport *port_)
1629 {
1630     struct ofport_dpif *port = ofport_dpif_cast(port_);
1631     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1632     const char *devname = netdev_get_name(port->up.netdev);
1633     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1634     const char *dp_port_name;
1635
1636     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1637     ovs_rwlock_wrlock(&xlate_rwlock);
1638     xlate_ofport_remove(port);
1639     ovs_rwlock_unlock(&xlate_rwlock);
1640
1641     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1642                                               sizeof namebuf);
1643     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1644         /* The underlying device is still there, so delete it.  This
1645          * happens when the ofproto is being destroyed, since the caller
1646          * assumes that removal of attached ports will happen as part of
1647          * destruction. */
1648         if (!port->is_tunnel) {
1649             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1650         }
1651     }
1652
1653     if (port->peer) {
1654         port->peer->peer = NULL;
1655         port->peer = NULL;
1656     }
1657
1658     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1659         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1660         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1661         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1662     }
1663
1664     tnl_port_del(port);
1665     sset_find_and_delete(&ofproto->ports, devname);
1666     sset_find_and_delete(&ofproto->ghost_ports, devname);
1667     bundle_remove(port_);
1668     set_cfm(port_, NULL);
1669     set_bfd(port_, NULL);
1670     if (port->stp_port) {
1671         stp_port_disable(port->stp_port);
1672     }
1673     if (ofproto->sflow) {
1674         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1675     }
1676
1677     free(port->qdscp);
1678 }
1679
1680 static void
1681 port_modified(struct ofport *port_)
1682 {
1683     struct ofport_dpif *port = ofport_dpif_cast(port_);
1684
1685     if (port->bundle && port->bundle->bond) {
1686         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1687     }
1688
1689     if (port->cfm) {
1690         cfm_set_netdev(port->cfm, port->up.netdev);
1691     }
1692
1693     if (port->bfd) {
1694         bfd_set_netdev(port->bfd, port->up.netdev);
1695     }
1696
1697     ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1698                                      port->up.pp.hw_addr);
1699
1700     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1701                                                 port->odp_port)) {
1702         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1703             REV_RECONFIGURE;
1704     }
1705
1706     ofport_update_peer(port);
1707 }
1708
1709 static void
1710 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1711 {
1712     struct ofport_dpif *port = ofport_dpif_cast(port_);
1713     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1714     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1715
1716     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1717                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1718                    OFPUTIL_PC_NO_PACKET_IN)) {
1719         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1720
1721         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1722             bundle_update(port->bundle);
1723         }
1724     }
1725 }
1726
1727 static int
1728 set_sflow(struct ofproto *ofproto_,
1729           const struct ofproto_sflow_options *sflow_options)
1730 {
1731     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1732     struct dpif_sflow *ds = ofproto->sflow;
1733
1734     if (sflow_options) {
1735         if (!ds) {
1736             struct ofport_dpif *ofport;
1737
1738             ds = ofproto->sflow = dpif_sflow_create();
1739             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1740                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1741             }
1742             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1743         }
1744         dpif_sflow_set_options(ds, sflow_options);
1745     } else {
1746         if (ds) {
1747             dpif_sflow_unref(ds);
1748             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1749             ofproto->sflow = NULL;
1750         }
1751     }
1752     return 0;
1753 }
1754
1755 static int
1756 set_ipfix(
1757     struct ofproto *ofproto_,
1758     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1759     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1760     size_t n_flow_exporters_options)
1761 {
1762     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1763     struct dpif_ipfix *di = ofproto->ipfix;
1764     bool has_options = bridge_exporter_options || flow_exporters_options;
1765
1766     if (has_options && !di) {
1767         di = ofproto->ipfix = dpif_ipfix_create();
1768     }
1769
1770     if (di) {
1771         /* Call set_options in any case to cleanly flush the flow
1772          * caches in the last exporters that are to be destroyed. */
1773         dpif_ipfix_set_options(
1774             di, bridge_exporter_options, flow_exporters_options,
1775             n_flow_exporters_options);
1776
1777         if (!has_options) {
1778             dpif_ipfix_unref(di);
1779             ofproto->ipfix = NULL;
1780         }
1781     }
1782
1783     return 0;
1784 }
1785
1786 static int
1787 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1788 {
1789     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1790     int error = 0;
1791
1792     if (s) {
1793         if (!ofport->cfm) {
1794             struct ofproto_dpif *ofproto;
1795
1796             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1797             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1798             ofport->cfm = cfm_create(ofport->up.netdev);
1799         }
1800
1801         if (cfm_configure(ofport->cfm, s)) {
1802             error = 0;
1803             goto out;
1804         }
1805
1806         error = EINVAL;
1807     }
1808     cfm_unref(ofport->cfm);
1809     ofport->cfm = NULL;
1810 out:
1811     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1812                                      ofport->up.pp.hw_addr);
1813     return error;
1814 }
1815
1816 static bool
1817 get_cfm_status(const struct ofport *ofport_,
1818                struct ofproto_cfm_status *status)
1819 {
1820     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1821
1822     if (ofport->cfm) {
1823         status->faults = cfm_get_fault(ofport->cfm);
1824         status->flap_count = cfm_get_flap_count(ofport->cfm);
1825         status->remote_opstate = cfm_get_opup(ofport->cfm);
1826         status->health = cfm_get_health(ofport->cfm);
1827         cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1828         return true;
1829     } else {
1830         return false;
1831     }
1832 }
1833
1834 static int
1835 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1836 {
1837     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1838     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1839     struct bfd *old;
1840
1841     old = ofport->bfd;
1842     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1843                                 cfg, ofport->up.netdev);
1844     if (ofport->bfd != old) {
1845         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1846     }
1847     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1848                                      ofport->up.pp.hw_addr);
1849     return 0;
1850 }
1851
1852 static int
1853 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1854 {
1855     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1856
1857     if (ofport->bfd) {
1858         bfd_get_status(ofport->bfd, smap);
1859         return 0;
1860     } else {
1861         return ENOENT;
1862     }
1863 }
1864 \f
1865 /* Spanning Tree. */
1866
1867 static void
1868 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1869 {
1870     struct ofproto_dpif *ofproto = ofproto_;
1871     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1872     struct ofport_dpif *ofport;
1873
1874     ofport = stp_port_get_aux(sp);
1875     if (!ofport) {
1876         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1877                      ofproto->up.name, port_num);
1878     } else {
1879         struct eth_header *eth = ofpbuf_l2(pkt);
1880
1881         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1882         if (eth_addr_is_zero(eth->eth_src)) {
1883             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1884                          "with unknown MAC", ofproto->up.name, port_num);
1885         } else {
1886             ofproto_dpif_send_packet(ofport, pkt);
1887         }
1888     }
1889     ofpbuf_delete(pkt);
1890 }
1891
1892 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1893 static int
1894 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1895 {
1896     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1897
1898     /* Only revalidate flows if the configuration changed. */
1899     if (!s != !ofproto->stp) {
1900         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1901     }
1902
1903     if (s) {
1904         if (!ofproto->stp) {
1905             ofproto->stp = stp_create(ofproto_->name, s->system_id,
1906                                       send_bpdu_cb, ofproto);
1907             ofproto->stp_last_tick = time_msec();
1908         }
1909
1910         stp_set_bridge_id(ofproto->stp, s->system_id);
1911         stp_set_bridge_priority(ofproto->stp, s->priority);
1912         stp_set_hello_time(ofproto->stp, s->hello_time);
1913         stp_set_max_age(ofproto->stp, s->max_age);
1914         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1915     }  else {
1916         struct ofport *ofport;
1917
1918         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1919             set_stp_port(ofport, NULL);
1920         }
1921
1922         stp_unref(ofproto->stp);
1923         ofproto->stp = NULL;
1924     }
1925
1926     return 0;
1927 }
1928
1929 static int
1930 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1931 {
1932     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1933
1934     if (ofproto->stp) {
1935         s->enabled = true;
1936         s->bridge_id = stp_get_bridge_id(ofproto->stp);
1937         s->designated_root = stp_get_designated_root(ofproto->stp);
1938         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1939     } else {
1940         s->enabled = false;
1941     }
1942
1943     return 0;
1944 }
1945
1946 static void
1947 update_stp_port_state(struct ofport_dpif *ofport)
1948 {
1949     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1950     enum stp_state state;
1951
1952     /* Figure out new state. */
1953     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1954                              : STP_DISABLED;
1955
1956     /* Update state. */
1957     if (ofport->stp_state != state) {
1958         enum ofputil_port_state of_state;
1959         bool fwd_change;
1960
1961         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1962                     netdev_get_name(ofport->up.netdev),
1963                     stp_state_name(ofport->stp_state),
1964                     stp_state_name(state));
1965         if (stp_learn_in_state(ofport->stp_state)
1966                 != stp_learn_in_state(state)) {
1967             /* xxx Learning action flows should also be flushed. */
1968             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1969             mac_learning_flush(ofproto->ml);
1970             ovs_rwlock_unlock(&ofproto->ml->rwlock);
1971         }
1972         fwd_change = stp_forward_in_state(ofport->stp_state)
1973                         != stp_forward_in_state(state);
1974
1975         ofproto->backer->need_revalidate = REV_STP;
1976         ofport->stp_state = state;
1977         ofport->stp_state_entered = time_msec();
1978
1979         if (fwd_change && ofport->bundle) {
1980             bundle_update(ofport->bundle);
1981         }
1982
1983         /* Update the STP state bits in the OpenFlow port description. */
1984         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1985         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1986                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1987                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1988                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
1989                      : 0);
1990         ofproto_port_set_state(&ofport->up, of_state);
1991     }
1992 }
1993
1994 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
1995  * caller is responsible for assigning STP port numbers and ensuring
1996  * there are no duplicates. */
1997 static int
1998 set_stp_port(struct ofport *ofport_,
1999              const struct ofproto_port_stp_settings *s)
2000 {
2001     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2002     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2003     struct stp_port *sp = ofport->stp_port;
2004
2005     if (!s || !s->enable) {
2006         if (sp) {
2007             ofport->stp_port = NULL;
2008             stp_port_disable(sp);
2009             update_stp_port_state(ofport);
2010         }
2011         return 0;
2012     } else if (sp && stp_port_no(sp) != s->port_num
2013             && ofport == stp_port_get_aux(sp)) {
2014         /* The port-id changed, so disable the old one if it's not
2015          * already in use by another port. */
2016         stp_port_disable(sp);
2017     }
2018
2019     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2020     stp_port_enable(sp);
2021
2022     stp_port_set_aux(sp, ofport);
2023     stp_port_set_priority(sp, s->priority);
2024     stp_port_set_path_cost(sp, s->path_cost);
2025
2026     update_stp_port_state(ofport);
2027
2028     return 0;
2029 }
2030
2031 static int
2032 get_stp_port_status(struct ofport *ofport_,
2033                     struct ofproto_port_stp_status *s)
2034 {
2035     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2036     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2037     struct stp_port *sp = ofport->stp_port;
2038
2039     if (!ofproto->stp || !sp) {
2040         s->enabled = false;
2041         return 0;
2042     }
2043
2044     s->enabled = true;
2045     s->port_id = stp_port_get_id(sp);
2046     s->state = stp_port_get_state(sp);
2047     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2048     s->role = stp_port_get_role(sp);
2049
2050     return 0;
2051 }
2052
2053 static int
2054 get_stp_port_stats(struct ofport *ofport_,
2055                    struct ofproto_port_stp_stats *s)
2056 {
2057     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2058     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2059     struct stp_port *sp = ofport->stp_port;
2060
2061     if (!ofproto->stp || !sp) {
2062         s->enabled = false;
2063         return 0;
2064     }
2065
2066     s->enabled = true;
2067     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2068
2069     return 0;
2070 }
2071
2072 static void
2073 stp_run(struct ofproto_dpif *ofproto)
2074 {
2075     if (ofproto->stp) {
2076         long long int now = time_msec();
2077         long long int elapsed = now - ofproto->stp_last_tick;
2078         struct stp_port *sp;
2079
2080         if (elapsed > 0) {
2081             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2082             ofproto->stp_last_tick = now;
2083         }
2084         while (stp_get_changed_port(ofproto->stp, &sp)) {
2085             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2086
2087             if (ofport) {
2088                 update_stp_port_state(ofport);
2089             }
2090         }
2091
2092         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2093             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2094             mac_learning_flush(ofproto->ml);
2095             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2096         }
2097     }
2098 }
2099
2100 static void
2101 stp_wait(struct ofproto_dpif *ofproto)
2102 {
2103     if (ofproto->stp) {
2104         poll_timer_wait(1000);
2105     }
2106 }
2107 \f
2108 static int
2109 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2110            size_t n_qdscp)
2111 {
2112     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2113     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2114
2115     if (ofport->n_qdscp != n_qdscp
2116         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2117                               n_qdscp * sizeof *qdscp))) {
2118         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2119         free(ofport->qdscp);
2120         ofport->qdscp = n_qdscp
2121             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2122             : NULL;
2123         ofport->n_qdscp = n_qdscp;
2124     }
2125
2126     return 0;
2127 }
2128 \f
2129 /* Bundles. */
2130
2131 /* Expires all MAC learning entries associated with 'bundle' and forces its
2132  * ofproto to revalidate every flow.
2133  *
2134  * Normally MAC learning entries are removed only from the ofproto associated
2135  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2136  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2137  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2138  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2139  * with the host from which it migrated. */
2140 static void
2141 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2142 {
2143     struct ofproto_dpif *ofproto = bundle->ofproto;
2144     struct mac_learning *ml = ofproto->ml;
2145     struct mac_entry *mac, *next_mac;
2146
2147     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2148     ovs_rwlock_wrlock(&ml->rwlock);
2149     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2150         if (mac->port.p == bundle) {
2151             if (all_ofprotos) {
2152                 struct ofproto_dpif *o;
2153
2154                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2155                     if (o != ofproto) {
2156                         struct mac_entry *e;
2157
2158                         ovs_rwlock_wrlock(&o->ml->rwlock);
2159                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2160                         if (e) {
2161                             mac_learning_expire(o->ml, e);
2162                         }
2163                         ovs_rwlock_unlock(&o->ml->rwlock);
2164                     }
2165                 }
2166             }
2167
2168             mac_learning_expire(ml, mac);
2169         }
2170     }
2171     ovs_rwlock_unlock(&ml->rwlock);
2172 }
2173
2174 static struct ofbundle *
2175 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2176 {
2177     struct ofbundle *bundle;
2178
2179     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2180                              &ofproto->bundles) {
2181         if (bundle->aux == aux) {
2182             return bundle;
2183         }
2184     }
2185     return NULL;
2186 }
2187
2188 static void
2189 bundle_update(struct ofbundle *bundle)
2190 {
2191     struct ofport_dpif *port;
2192
2193     bundle->floodable = true;
2194     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2195         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2196             || port->is_layer3
2197             || !stp_forward_in_state(port->stp_state)) {
2198             bundle->floodable = false;
2199             break;
2200         }
2201     }
2202 }
2203
2204 static void
2205 bundle_del_port(struct ofport_dpif *port)
2206 {
2207     struct ofbundle *bundle = port->bundle;
2208
2209     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2210
2211     list_remove(&port->bundle_node);
2212     port->bundle = NULL;
2213
2214     if (bundle->lacp) {
2215         lacp_slave_unregister(bundle->lacp, port);
2216     }
2217     if (bundle->bond) {
2218         bond_slave_unregister(bundle->bond, port);
2219     }
2220
2221     bundle_update(bundle);
2222 }
2223
2224 static bool
2225 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2226                 struct lacp_slave_settings *lacp)
2227 {
2228     struct ofport_dpif *port;
2229
2230     port = get_ofp_port(bundle->ofproto, ofp_port);
2231     if (!port) {
2232         return false;
2233     }
2234
2235     if (port->bundle != bundle) {
2236         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2237         if (port->bundle) {
2238             bundle_remove(&port->up);
2239         }
2240
2241         port->bundle = bundle;
2242         list_push_back(&bundle->ports, &port->bundle_node);
2243         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2244             || port->is_layer3
2245             || !stp_forward_in_state(port->stp_state)) {
2246             bundle->floodable = false;
2247         }
2248     }
2249     if (lacp) {
2250         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2251         lacp_slave_register(bundle->lacp, port, lacp);
2252     }
2253
2254     return true;
2255 }
2256
2257 static void
2258 bundle_destroy(struct ofbundle *bundle)
2259 {
2260     struct ofproto_dpif *ofproto;
2261     struct ofport_dpif *port, *next_port;
2262
2263     if (!bundle) {
2264         return;
2265     }
2266
2267     ofproto = bundle->ofproto;
2268     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2269
2270     ovs_rwlock_wrlock(&xlate_rwlock);
2271     xlate_bundle_remove(bundle);
2272     ovs_rwlock_unlock(&xlate_rwlock);
2273
2274     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2275         bundle_del_port(port);
2276     }
2277
2278     bundle_flush_macs(bundle, true);
2279     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2280     free(bundle->name);
2281     free(bundle->trunks);
2282     lacp_unref(bundle->lacp);
2283     bond_unref(bundle->bond);
2284     free(bundle);
2285 }
2286
2287 static int
2288 bundle_set(struct ofproto *ofproto_, void *aux,
2289            const struct ofproto_bundle_settings *s)
2290 {
2291     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2292     bool need_flush = false;
2293     struct ofport_dpif *port;
2294     struct ofbundle *bundle;
2295     unsigned long *trunks;
2296     int vlan;
2297     size_t i;
2298     bool ok;
2299
2300     if (!s) {
2301         bundle_destroy(bundle_lookup(ofproto, aux));
2302         return 0;
2303     }
2304
2305     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2306     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2307
2308     bundle = bundle_lookup(ofproto, aux);
2309     if (!bundle) {
2310         bundle = xmalloc(sizeof *bundle);
2311
2312         bundle->ofproto = ofproto;
2313         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2314                     hash_pointer(aux, 0));
2315         bundle->aux = aux;
2316         bundle->name = NULL;
2317
2318         list_init(&bundle->ports);
2319         bundle->vlan_mode = PORT_VLAN_TRUNK;
2320         bundle->vlan = -1;
2321         bundle->trunks = NULL;
2322         bundle->use_priority_tags = s->use_priority_tags;
2323         bundle->lacp = NULL;
2324         bundle->bond = NULL;
2325
2326         bundle->floodable = true;
2327         mbridge_register_bundle(ofproto->mbridge, bundle);
2328     }
2329
2330     if (!bundle->name || strcmp(s->name, bundle->name)) {
2331         free(bundle->name);
2332         bundle->name = xstrdup(s->name);
2333     }
2334
2335     /* LACP. */
2336     if (s->lacp) {
2337         ofproto->lacp_enabled = true;
2338         if (!bundle->lacp) {
2339             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2340             bundle->lacp = lacp_create();
2341         }
2342         lacp_configure(bundle->lacp, s->lacp);
2343     } else {
2344         lacp_unref(bundle->lacp);
2345         bundle->lacp = NULL;
2346     }
2347
2348     /* Update set of ports. */
2349     ok = true;
2350     for (i = 0; i < s->n_slaves; i++) {
2351         if (!bundle_add_port(bundle, s->slaves[i],
2352                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2353             ok = false;
2354         }
2355     }
2356     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2357         struct ofport_dpif *next_port;
2358
2359         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2360             for (i = 0; i < s->n_slaves; i++) {
2361                 if (s->slaves[i] == port->up.ofp_port) {
2362                     goto found;
2363                 }
2364             }
2365
2366             bundle_del_port(port);
2367         found: ;
2368         }
2369     }
2370     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2371
2372     if (list_is_empty(&bundle->ports)) {
2373         bundle_destroy(bundle);
2374         return EINVAL;
2375     }
2376
2377     /* Set VLAN tagging mode */
2378     if (s->vlan_mode != bundle->vlan_mode
2379         || s->use_priority_tags != bundle->use_priority_tags) {
2380         bundle->vlan_mode = s->vlan_mode;
2381         bundle->use_priority_tags = s->use_priority_tags;
2382         need_flush = true;
2383     }
2384
2385     /* Set VLAN tag. */
2386     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2387             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2388             : 0);
2389     if (vlan != bundle->vlan) {
2390         bundle->vlan = vlan;
2391         need_flush = true;
2392     }
2393
2394     /* Get trunked VLANs. */
2395     switch (s->vlan_mode) {
2396     case PORT_VLAN_ACCESS:
2397         trunks = NULL;
2398         break;
2399
2400     case PORT_VLAN_TRUNK:
2401         trunks = CONST_CAST(unsigned long *, s->trunks);
2402         break;
2403
2404     case PORT_VLAN_NATIVE_UNTAGGED:
2405     case PORT_VLAN_NATIVE_TAGGED:
2406         if (vlan != 0 && (!s->trunks
2407                           || !bitmap_is_set(s->trunks, vlan)
2408                           || bitmap_is_set(s->trunks, 0))) {
2409             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2410             if (s->trunks) {
2411                 trunks = bitmap_clone(s->trunks, 4096);
2412             } else {
2413                 trunks = bitmap_allocate1(4096);
2414             }
2415             bitmap_set1(trunks, vlan);
2416             bitmap_set0(trunks, 0);
2417         } else {
2418             trunks = CONST_CAST(unsigned long *, s->trunks);
2419         }
2420         break;
2421
2422     default:
2423         OVS_NOT_REACHED();
2424     }
2425     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2426         free(bundle->trunks);
2427         if (trunks == s->trunks) {
2428             bundle->trunks = vlan_bitmap_clone(trunks);
2429         } else {
2430             bundle->trunks = trunks;
2431             trunks = NULL;
2432         }
2433         need_flush = true;
2434     }
2435     if (trunks != s->trunks) {
2436         free(trunks);
2437     }
2438
2439     /* Bonding. */
2440     if (!list_is_short(&bundle->ports)) {
2441         bundle->ofproto->has_bonded_bundles = true;
2442         if (bundle->bond) {
2443             if (bond_reconfigure(bundle->bond, s->bond)) {
2444                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2445             }
2446         } else {
2447             bundle->bond = bond_create(s->bond, ofproto);
2448             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2449         }
2450
2451         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2452             bond_slave_register(bundle->bond, port,
2453                                 port->up.ofp_port, port->up.netdev);
2454         }
2455     } else {
2456         bond_unref(bundle->bond);
2457         bundle->bond = NULL;
2458     }
2459
2460     /* If we changed something that would affect MAC learning, un-learn
2461      * everything on this port and force flow revalidation. */
2462     if (need_flush) {
2463         bundle_flush_macs(bundle, false);
2464     }
2465
2466     return 0;
2467 }
2468
2469 static void
2470 bundle_remove(struct ofport *port_)
2471 {
2472     struct ofport_dpif *port = ofport_dpif_cast(port_);
2473     struct ofbundle *bundle = port->bundle;
2474
2475     if (bundle) {
2476         bundle_del_port(port);
2477         if (list_is_empty(&bundle->ports)) {
2478             bundle_destroy(bundle);
2479         } else if (list_is_short(&bundle->ports)) {
2480             bond_unref(bundle->bond);
2481             bundle->bond = NULL;
2482         }
2483     }
2484 }
2485
2486 static void
2487 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2488 {
2489     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2490     struct ofport_dpif *port = port_;
2491     uint8_t ea[ETH_ADDR_LEN];
2492     int error;
2493
2494     error = netdev_get_etheraddr(port->up.netdev, ea);
2495     if (!error) {
2496         struct ofpbuf packet;
2497         void *packet_pdu;
2498
2499         ofpbuf_init(&packet, 0);
2500         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2501                                  pdu_size);
2502         memcpy(packet_pdu, pdu, pdu_size);
2503
2504         ofproto_dpif_send_packet(port, &packet);
2505         ofpbuf_uninit(&packet);
2506     } else {
2507         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2508                     "%s (%s)", port->bundle->name,
2509                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2510     }
2511 }
2512
2513 static void
2514 bundle_send_learning_packets(struct ofbundle *bundle)
2515 {
2516     struct ofproto_dpif *ofproto = bundle->ofproto;
2517     struct ofpbuf *learning_packet;
2518     int error, n_packets, n_errors;
2519     struct mac_entry *e;
2520     struct list packets;
2521
2522     list_init(&packets);
2523     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2524     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2525         if (e->port.p != bundle) {
2526             void *port_void;
2527
2528             learning_packet = bond_compose_learning_packet(bundle->bond,
2529                                                            e->mac, e->vlan,
2530                                                            &port_void);
2531             /* Temporarily use 'frame' as a private pointer (see below). */
2532             ovs_assert(learning_packet->frame == ofpbuf_data(learning_packet));
2533             learning_packet->frame = port_void;
2534             list_push_back(&packets, &learning_packet->list_node);
2535         }
2536     }
2537     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2538
2539     error = n_packets = n_errors = 0;
2540     LIST_FOR_EACH (learning_packet, list_node, &packets) {
2541         int ret;
2542         void *port_void = learning_packet->frame;
2543
2544         /* Restore 'frame'. */
2545         learning_packet->frame = ofpbuf_data(learning_packet);
2546         ret = ofproto_dpif_send_packet(port_void, learning_packet);
2547         if (ret) {
2548             error = ret;
2549             n_errors++;
2550         }
2551         n_packets++;
2552     }
2553     ofpbuf_list_delete(&packets);
2554
2555     if (n_errors) {
2556         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2557         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2558                      "packets, last error was: %s",
2559                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2560     } else {
2561         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2562                  bundle->name, n_packets);
2563     }
2564 }
2565
2566 static void
2567 bundle_run(struct ofbundle *bundle)
2568 {
2569     if (bundle->lacp) {
2570         lacp_run(bundle->lacp, send_pdu_cb);
2571     }
2572     if (bundle->bond) {
2573         struct ofport_dpif *port;
2574
2575         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2576             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2577         }
2578
2579         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2580             bundle->ofproto->backer->need_revalidate = REV_BOND;
2581         }
2582
2583         if (bond_should_send_learning_packets(bundle->bond)) {
2584             bundle_send_learning_packets(bundle);
2585         }
2586     }
2587 }
2588
2589 static void
2590 bundle_wait(struct ofbundle *bundle)
2591 {
2592     if (bundle->lacp) {
2593         lacp_wait(bundle->lacp);
2594     }
2595     if (bundle->bond) {
2596         bond_wait(bundle->bond);
2597     }
2598 }
2599 \f
2600 /* Mirrors. */
2601
2602 static int
2603 mirror_set__(struct ofproto *ofproto_, void *aux,
2604              const struct ofproto_mirror_settings *s)
2605 {
2606     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2607     struct ofbundle **srcs, **dsts;
2608     int error;
2609     size_t i;
2610
2611     if (!s) {
2612         mirror_destroy(ofproto->mbridge, aux);
2613         return 0;
2614     }
2615
2616     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2617     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2618
2619     for (i = 0; i < s->n_srcs; i++) {
2620         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2621     }
2622
2623     for (i = 0; i < s->n_dsts; i++) {
2624         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2625     }
2626
2627     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2628                        s->n_dsts, s->src_vlans,
2629                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2630     free(srcs);
2631     free(dsts);
2632     return error;
2633 }
2634
2635 static int
2636 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2637                    uint64_t *packets, uint64_t *bytes)
2638 {
2639     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2640                             bytes);
2641 }
2642
2643 static int
2644 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2645 {
2646     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2647     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2648     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2649         mac_learning_flush(ofproto->ml);
2650     }
2651     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2652     return 0;
2653 }
2654
2655 static bool
2656 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2657 {
2658     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2659     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2660     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2661 }
2662
2663 static void
2664 forward_bpdu_changed(struct ofproto *ofproto_)
2665 {
2666     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2667     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2668 }
2669
2670 static void
2671 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2672                      size_t max_entries)
2673 {
2674     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2675     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2676     mac_learning_set_idle_time(ofproto->ml, idle_time);
2677     mac_learning_set_max_entries(ofproto->ml, max_entries);
2678     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2679 }
2680 \f
2681 /* Ports. */
2682
2683 static struct ofport_dpif *
2684 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2685 {
2686     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2687     return ofport ? ofport_dpif_cast(ofport) : NULL;
2688 }
2689
2690 static void
2691 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2692                             struct ofproto_port *ofproto_port,
2693                             struct dpif_port *dpif_port)
2694 {
2695     ofproto_port->name = dpif_port->name;
2696     ofproto_port->type = dpif_port->type;
2697     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2698 }
2699
2700 static void
2701 ofport_update_peer(struct ofport_dpif *ofport)
2702 {
2703     const struct ofproto_dpif *ofproto;
2704     struct dpif_backer *backer;
2705     char *peer_name;
2706
2707     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2708         return;
2709     }
2710
2711     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2712     backer->need_revalidate = REV_RECONFIGURE;
2713
2714     if (ofport->peer) {
2715         ofport->peer->peer = NULL;
2716         ofport->peer = NULL;
2717     }
2718
2719     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2720     if (!peer_name) {
2721         return;
2722     }
2723
2724     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2725         struct ofport *peer_ofport;
2726         struct ofport_dpif *peer;
2727         char *peer_peer;
2728
2729         if (ofproto->backer != backer) {
2730             continue;
2731         }
2732
2733         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2734         if (!peer_ofport) {
2735             continue;
2736         }
2737
2738         peer = ofport_dpif_cast(peer_ofport);
2739         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2740         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2741                                  peer_peer)) {
2742             ofport->peer = peer;
2743             ofport->peer->peer = ofport;
2744         }
2745         free(peer_peer);
2746
2747         break;
2748     }
2749     free(peer_name);
2750 }
2751
2752 static void
2753 port_run(struct ofport_dpif *ofport)
2754 {
2755     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2756     bool carrier_changed = carrier_seq != ofport->carrier_seq;
2757     bool enable = netdev_get_carrier(ofport->up.netdev);
2758     bool cfm_enable = false;
2759     bool bfd_enable = false;
2760
2761     ofport->carrier_seq = carrier_seq;
2762
2763     if (ofport->cfm) {
2764         int cfm_opup = cfm_get_opup(ofport->cfm);
2765
2766         cfm_enable = !cfm_get_fault(ofport->cfm);
2767
2768         if (cfm_opup >= 0) {
2769             cfm_enable = cfm_enable && cfm_opup;
2770         }
2771     }
2772
2773     if (ofport->bfd) {
2774         bfd_enable = bfd_forwarding(ofport->bfd);
2775     }
2776
2777     if (ofport->bfd || ofport->cfm) {
2778         enable = enable && (cfm_enable || bfd_enable);
2779     }
2780
2781     if (ofport->bundle) {
2782         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2783         if (carrier_changed) {
2784             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2785         }
2786     }
2787
2788     if (ofport->may_enable != enable) {
2789         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2790         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2791     }
2792
2793     ofport->may_enable = enable;
2794 }
2795
2796 static int
2797 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2798                    struct ofproto_port *ofproto_port)
2799 {
2800     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2801     struct dpif_port dpif_port;
2802     int error;
2803
2804     if (sset_contains(&ofproto->ghost_ports, devname)) {
2805         const char *type = netdev_get_type_from_name(devname);
2806
2807         /* We may be called before ofproto->up.port_by_name is populated with
2808          * the appropriate ofport.  For this reason, we must get the name and
2809          * type from the netdev layer directly. */
2810         if (type) {
2811             const struct ofport *ofport;
2812
2813             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2814             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2815             ofproto_port->name = xstrdup(devname);
2816             ofproto_port->type = xstrdup(type);
2817             return 0;
2818         }
2819         return ENODEV;
2820     }
2821
2822     if (!sset_contains(&ofproto->ports, devname)) {
2823         return ENODEV;
2824     }
2825     error = dpif_port_query_by_name(ofproto->backer->dpif,
2826                                     devname, &dpif_port);
2827     if (!error) {
2828         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2829     }
2830     return error;
2831 }
2832
2833 static int
2834 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2835 {
2836     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2837     const char *devname = netdev_get_name(netdev);
2838     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2839     const char *dp_port_name;
2840
2841     if (netdev_vport_is_patch(netdev)) {
2842         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2843         return 0;
2844     }
2845
2846     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2847     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2848         odp_port_t port_no = ODPP_NONE;
2849         int error;
2850
2851         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2852         if (error) {
2853             return error;
2854         }
2855         if (netdev_get_tunnel_config(netdev)) {
2856             simap_put(&ofproto->backer->tnl_backers,
2857                       dp_port_name, odp_to_u32(port_no));
2858         }
2859     }
2860
2861     if (netdev_get_tunnel_config(netdev)) {
2862         sset_add(&ofproto->ghost_ports, devname);
2863     } else {
2864         sset_add(&ofproto->ports, devname);
2865     }
2866     return 0;
2867 }
2868
2869 static int
2870 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
2871 {
2872     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2873     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2874     int error = 0;
2875
2876     if (!ofport) {
2877         return 0;
2878     }
2879
2880     sset_find_and_delete(&ofproto->ghost_ports,
2881                          netdev_get_name(ofport->up.netdev));
2882     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2883     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
2884         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
2885         if (!error) {
2886             /* The caller is going to close ofport->up.netdev.  If this is a
2887              * bonded port, then the bond is using that netdev, so remove it
2888              * from the bond.  The client will need to reconfigure everything
2889              * after deleting ports, so then the slave will get re-added. */
2890             bundle_remove(&ofport->up);
2891         }
2892     }
2893     return error;
2894 }
2895
2896 static int
2897 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
2898 {
2899     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2900     int error;
2901
2902     error = netdev_get_stats(ofport->up.netdev, stats);
2903
2904     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
2905         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2906
2907         ovs_mutex_lock(&ofproto->stats_mutex);
2908         /* ofproto->stats.tx_packets represents packets that we created
2909          * internally and sent to some port (e.g. packets sent with
2910          * ofproto_dpif_send_packet()).  Account for them as if they had
2911          * come from OFPP_LOCAL and got forwarded. */
2912
2913         if (stats->rx_packets != UINT64_MAX) {
2914             stats->rx_packets += ofproto->stats.tx_packets;
2915         }
2916
2917         if (stats->rx_bytes != UINT64_MAX) {
2918             stats->rx_bytes += ofproto->stats.tx_bytes;
2919         }
2920
2921         /* ofproto->stats.rx_packets represents packets that were received on
2922          * some port and we processed internally and dropped (e.g. STP).
2923          * Account for them as if they had been forwarded to OFPP_LOCAL. */
2924
2925         if (stats->tx_packets != UINT64_MAX) {
2926             stats->tx_packets += ofproto->stats.rx_packets;
2927         }
2928
2929         if (stats->tx_bytes != UINT64_MAX) {
2930             stats->tx_bytes += ofproto->stats.rx_bytes;
2931         }
2932         ovs_mutex_unlock(&ofproto->stats_mutex);
2933     }
2934
2935     return error;
2936 }
2937
2938 struct port_dump_state {
2939     uint32_t bucket;
2940     uint32_t offset;
2941     bool ghost;
2942
2943     struct ofproto_port port;
2944     bool has_port;
2945 };
2946
2947 static int
2948 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
2949 {
2950     *statep = xzalloc(sizeof(struct port_dump_state));
2951     return 0;
2952 }
2953
2954 static int
2955 port_dump_next(const struct ofproto *ofproto_, void *state_,
2956                struct ofproto_port *port)
2957 {
2958     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2959     struct port_dump_state *state = state_;
2960     const struct sset *sset;
2961     struct sset_node *node;
2962
2963     if (state->has_port) {
2964         ofproto_port_destroy(&state->port);
2965         state->has_port = false;
2966     }
2967     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
2968     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
2969         int error;
2970
2971         error = port_query_by_name(ofproto_, node->name, &state->port);
2972         if (!error) {
2973             *port = state->port;
2974             state->has_port = true;
2975             return 0;
2976         } else if (error != ENODEV) {
2977             return error;
2978         }
2979     }
2980
2981     if (!state->ghost) {
2982         state->ghost = true;
2983         state->bucket = 0;
2984         state->offset = 0;
2985         return port_dump_next(ofproto_, state_, port);
2986     }
2987
2988     return EOF;
2989 }
2990
2991 static int
2992 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
2993 {
2994     struct port_dump_state *state = state_;
2995
2996     if (state->has_port) {
2997         ofproto_port_destroy(&state->port);
2998     }
2999     free(state);
3000     return 0;
3001 }
3002
3003 static int
3004 port_poll(const struct ofproto *ofproto_, char **devnamep)
3005 {
3006     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3007
3008     if (ofproto->port_poll_errno) {
3009         int error = ofproto->port_poll_errno;
3010         ofproto->port_poll_errno = 0;
3011         return error;
3012     }
3013
3014     if (sset_is_empty(&ofproto->port_poll_set)) {
3015         return EAGAIN;
3016     }
3017
3018     *devnamep = sset_pop(&ofproto->port_poll_set);
3019     return 0;
3020 }
3021
3022 static void
3023 port_poll_wait(const struct ofproto *ofproto_)
3024 {
3025     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3026     dpif_port_poll_wait(ofproto->backer->dpif);
3027 }
3028
3029 static int
3030 port_is_lacp_current(const struct ofport *ofport_)
3031 {
3032     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3033     return (ofport->bundle && ofport->bundle->lacp
3034             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3035             : -1);
3036 }
3037 \f
3038 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3039  * then delete it entirely. */
3040 static void
3041 rule_expire(struct rule_dpif *rule)
3042     OVS_REQUIRES(ofproto_mutex)
3043 {
3044     uint16_t hard_timeout, idle_timeout;
3045     long long int now = time_msec();
3046     int reason = -1;
3047
3048     ovs_assert(!rule->up.pending);
3049
3050     hard_timeout = rule->up.hard_timeout;
3051     idle_timeout = rule->up.idle_timeout;
3052
3053     /* Has 'rule' expired? */
3054     if (hard_timeout) {
3055         long long int modified;
3056
3057         ovs_mutex_lock(&rule->up.mutex);
3058         modified = rule->up.modified;
3059         ovs_mutex_unlock(&rule->up.mutex);
3060
3061         if (now > modified + hard_timeout * 1000) {
3062             reason = OFPRR_HARD_TIMEOUT;
3063         }
3064     }
3065
3066     if (reason < 0 && idle_timeout) {
3067         long long int used;
3068
3069         ovs_mutex_lock(&rule->stats_mutex);
3070         used = rule->stats.used;
3071         ovs_mutex_unlock(&rule->stats_mutex);
3072
3073         if (now > used + idle_timeout * 1000) {
3074             reason = OFPRR_IDLE_TIMEOUT;
3075         }
3076     }
3077
3078     if (reason >= 0) {
3079         COVERAGE_INC(ofproto_dpif_expired);
3080         ofproto_rule_expire(&rule->up, reason);
3081     }
3082 }
3083
3084 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3085  * 'flow' must reflect the data in 'packet'. */
3086 int
3087 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3088                              const struct flow *flow,
3089                              struct rule_dpif *rule,
3090                              const struct ofpact *ofpacts, size_t ofpacts_len,
3091                              struct ofpbuf *packet)
3092 {
3093     struct dpif_flow_stats stats;
3094     struct xlate_out xout;
3095     struct xlate_in xin;
3096     ofp_port_t in_port;
3097     struct dpif_execute execute;
3098     int error;
3099
3100     ovs_assert((rule != NULL) != (ofpacts != NULL));
3101
3102     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3103
3104     if (rule) {
3105         rule_dpif_credit_stats(rule, &stats);
3106     }
3107
3108     xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
3109     xin.ofpacts = ofpacts;
3110     xin.ofpacts_len = ofpacts_len;
3111     xin.resubmit_stats = &stats;
3112     xlate_actions(&xin, &xout);
3113
3114     in_port = flow->in_port.ofp_port;
3115     if (in_port == OFPP_NONE) {
3116         in_port = OFPP_LOCAL;
3117     }
3118     execute.actions = ofpbuf_data(&xout.odp_actions);
3119     execute.actions_len = ofpbuf_size(&xout.odp_actions);
3120     execute.packet = packet;
3121     execute.md.tunnel = flow->tunnel;
3122     execute.md.skb_priority = flow->skb_priority;
3123     execute.md.pkt_mark = flow->pkt_mark;
3124     execute.md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3125     execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3126
3127     error = dpif_execute(ofproto->backer->dpif, &execute);
3128
3129     xlate_out_uninit(&xout);
3130
3131     return error;
3132 }
3133
3134 void
3135 rule_dpif_credit_stats(struct rule_dpif *rule,
3136                        const struct dpif_flow_stats *stats)
3137 {
3138     ovs_mutex_lock(&rule->stats_mutex);
3139     rule->stats.n_packets += stats->n_packets;
3140     rule->stats.n_bytes += stats->n_bytes;
3141     rule->stats.used = MAX(rule->stats.used, stats->used);
3142     ovs_mutex_unlock(&rule->stats_mutex);
3143 }
3144
3145 bool
3146 rule_dpif_is_fail_open(const struct rule_dpif *rule)
3147 {
3148     return is_fail_open_rule(&rule->up);
3149 }
3150
3151 bool
3152 rule_dpif_is_table_miss(const struct rule_dpif *rule)
3153 {
3154     return rule_is_table_miss(&rule->up);
3155 }
3156
3157 bool
3158 rule_dpif_is_internal(const struct rule_dpif *rule)
3159 {
3160     return rule_is_internal(&rule->up);
3161 }
3162
3163 ovs_be64
3164 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3165     OVS_REQUIRES(rule->up.mutex)
3166 {
3167     return rule->up.flow_cookie;
3168 }
3169
3170 void
3171 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3172                      uint16_t hard_timeout)
3173 {
3174     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3175 }
3176
3177 /* Returns 'rule''s actions.  The caller owns a reference on the returned
3178  * actions and must eventually release it (with rule_actions_unref()) to avoid
3179  * a memory leak. */
3180 struct rule_actions *
3181 rule_dpif_get_actions(const struct rule_dpif *rule)
3182 {
3183     return rule_get_actions(&rule->up);
3184 }
3185
3186 /* Lookup 'flow' in table 0 of 'ofproto''s classifier.
3187  * If 'wc' is non-null, sets the fields that were relevant as part of
3188  * the lookup. Returns the table_id where a match or miss occurred.
3189  *
3190  * The return value will be zero unless there was a miss and
3191  * OFPTC11_TABLE_MISS_CONTINUE is in effect for the sequence of tables
3192  * where misses occur. */
3193 uint8_t
3194 rule_dpif_lookup(struct ofproto_dpif *ofproto, struct flow *flow,
3195                  struct flow_wildcards *wc, struct rule_dpif **rule)
3196 {
3197     enum rule_dpif_lookup_verdict verdict;
3198     enum ofputil_port_config config = 0;
3199     uint8_t table_id;
3200
3201     if (ofproto_dpif_get_enable_recirc(ofproto)) {
3202         if (flow->recirc_id == 0) {
3203             if (wc) {
3204                 wc->masks.recirc_id = UINT32_MAX;
3205             }
3206             table_id = 0;
3207         } else {
3208             table_id = TBL_INTERNAL;
3209         }
3210     } else {
3211         table_id = 0;
3212     }
3213
3214     verdict = rule_dpif_lookup_from_table(ofproto, flow, wc, true,
3215                                           &table_id, rule);
3216
3217     switch (verdict) {
3218     case RULE_DPIF_LOOKUP_VERDICT_MATCH:
3219         return table_id;
3220     case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER: {
3221         struct ofport_dpif *port;
3222
3223         port = get_ofp_port(ofproto, flow->in_port.ofp_port);
3224         if (!port) {
3225             VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3226                          flow->in_port.ofp_port);
3227         }
3228         config = port ? port->up.pp.config : 0;
3229         break;
3230     }
3231     case RULE_DPIF_LOOKUP_VERDICT_DROP:
3232         config = OFPUTIL_PC_NO_PACKET_IN;
3233         break;
3234     case RULE_DPIF_LOOKUP_VERDICT_DEFAULT:
3235         if (!connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3236             config = OFPUTIL_PC_NO_PACKET_IN;
3237         }
3238         break;
3239     default:
3240         OVS_NOT_REACHED();
3241     }
3242
3243     choose_miss_rule(config, ofproto->miss_rule,
3244                      ofproto->no_packet_in_rule, rule);
3245     return table_id;
3246 }
3247
3248 static struct rule_dpif *
3249 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, uint8_t table_id,
3250                           const struct flow *flow, struct flow_wildcards *wc)
3251 {
3252     struct classifier *cls = &ofproto->up.tables[table_id].cls;
3253     const struct cls_rule *cls_rule;
3254     struct rule_dpif *rule;
3255
3256     fat_rwlock_rdlock(&cls->rwlock);
3257     if (ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3258         if (wc) {
3259             memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
3260             if (is_ip_any(flow)) {
3261                 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
3262             }
3263         }
3264
3265         if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
3266             if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3267                 /* We must pretend that transport ports are unavailable. */
3268                 struct flow ofpc_normal_flow = *flow;
3269                 ofpc_normal_flow.tp_src = htons(0);
3270                 ofpc_normal_flow.tp_dst = htons(0);
3271                 cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
3272             } else {
3273                 /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM). */
3274                 cls_rule = &ofproto->drop_frags_rule->up.cr;
3275             }
3276         } else {
3277             cls_rule = classifier_lookup(cls, flow, wc);
3278         }
3279     } else {
3280         cls_rule = classifier_lookup(cls, flow, wc);
3281     }
3282
3283     rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3284     rule_dpif_ref(rule);
3285     fat_rwlock_unlock(&cls->rwlock);
3286
3287     return rule;
3288 }
3289
3290 /* Look up 'flow' in 'ofproto''s classifier starting from table '*table_id'.
3291  * Stores the rule that was found in '*rule', or NULL if none was found.
3292  * Updates 'wc', if nonnull, to reflect the fields that were used during the
3293  * lookup.
3294  *
3295  * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3296  * if none is found then the table miss configuration for that table is
3297  * honored, which can result in additional lookups in other OpenFlow tables.
3298  * In this case the function updates '*table_id' to reflect the final OpenFlow
3299  * table that was searched.
3300  *
3301  * If 'honor_table_miss' is false, then only one table lookup occurs, in
3302  * '*table_id'.
3303  *
3304  * Returns:
3305  *
3306  *    - RULE_DPIF_LOOKUP_VERDICT_MATCH if a rule (in '*rule') was found.
3307  *
3308  *    - RULE_OFPTC_TABLE_MISS_CONTROLLER if no rule was found and either:
3309  *      + 'honor_table_miss' is false
3310  *      + a table miss configuration specified that the packet should be
3311  *        sent to the controller in this case.
3312  *
3313  *    - RULE_DPIF_LOOKUP_VERDICT_DROP if no rule was found, 'honor_table_miss'
3314  *      is true and a table miss configuration specified that the packet
3315  *      should be dropped in this case.
3316  *
3317  *    - RULE_DPIF_LOOKUP_VERDICT_DEFAULT if no rule was found,
3318  *      'honor_table_miss' is true and a table miss configuration has
3319  *      not been specified in this case. */
3320 enum rule_dpif_lookup_verdict
3321 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3322                             const struct flow *flow,
3323                             struct flow_wildcards *wc,
3324                             bool honor_table_miss,
3325                             uint8_t *table_id, struct rule_dpif **rule)
3326 {
3327     uint8_t next_id;
3328
3329     for (next_id = *table_id;
3330          next_id < ofproto->up.n_tables;
3331          next_id++, next_id += (next_id == TBL_INTERNAL))
3332     {
3333         *table_id = next_id;
3334         *rule = rule_dpif_lookup_in_table(ofproto, *table_id, flow, wc);
3335         if (*rule) {
3336             return RULE_DPIF_LOOKUP_VERDICT_MATCH;
3337         } else if (!honor_table_miss) {
3338             return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3339         } else {
3340             switch (ofproto_table_get_config(&ofproto->up, *table_id)) {
3341             case OFPROTO_TABLE_MISS_CONTINUE:
3342                 break;
3343
3344             case OFPROTO_TABLE_MISS_CONTROLLER:
3345                 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3346
3347             case OFPROTO_TABLE_MISS_DROP:
3348                 return RULE_DPIF_LOOKUP_VERDICT_DROP;
3349
3350             case OFPROTO_TABLE_MISS_DEFAULT:
3351                 return RULE_DPIF_LOOKUP_VERDICT_DEFAULT;
3352             }
3353         }
3354     }
3355
3356     return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER;
3357 }
3358
3359 /* Given a port configuration (specified as zero if there's no port), chooses
3360  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
3361  * flow table miss. */
3362 void
3363 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
3364                  struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
3365 {
3366     *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
3367     rule_dpif_ref(*rule);
3368 }
3369
3370 void
3371 rule_dpif_ref(struct rule_dpif *rule)
3372 {
3373     if (rule) {
3374         ofproto_rule_ref(&rule->up);
3375     }
3376 }
3377
3378 void
3379 rule_dpif_unref(struct rule_dpif *rule)
3380 {
3381     if (rule) {
3382         ofproto_rule_unref(&rule->up);
3383     }
3384 }
3385
3386 static void
3387 complete_operation(struct rule_dpif *rule)
3388     OVS_REQUIRES(ofproto_mutex)
3389 {
3390     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3391
3392     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3393     ofoperation_complete(rule->up.pending, 0);
3394 }
3395
3396 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3397 {
3398     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3399 }
3400
3401 static struct rule *
3402 rule_alloc(void)
3403 {
3404     struct rule_dpif *rule = xmalloc(sizeof *rule);
3405     return &rule->up;
3406 }
3407
3408 static void
3409 rule_dealloc(struct rule *rule_)
3410 {
3411     struct rule_dpif *rule = rule_dpif_cast(rule_);
3412     free(rule);
3413 }
3414
3415 static enum ofperr
3416 rule_construct(struct rule *rule_)
3417     OVS_NO_THREAD_SAFETY_ANALYSIS
3418 {
3419     struct rule_dpif *rule = rule_dpif_cast(rule_);
3420     ovs_mutex_init_adaptive(&rule->stats_mutex);
3421     rule->stats.n_packets = 0;
3422     rule->stats.n_bytes = 0;
3423     rule->stats.used = rule->up.modified;
3424     return 0;
3425 }
3426
3427 static void
3428 rule_insert(struct rule *rule_)
3429     OVS_REQUIRES(ofproto_mutex)
3430 {
3431     struct rule_dpif *rule = rule_dpif_cast(rule_);
3432     complete_operation(rule);
3433 }
3434
3435 static void
3436 rule_delete(struct rule *rule_)
3437     OVS_REQUIRES(ofproto_mutex)
3438 {
3439     struct rule_dpif *rule = rule_dpif_cast(rule_);
3440     complete_operation(rule);
3441 }
3442
3443 static void
3444 rule_destruct(struct rule *rule_)
3445 {
3446     struct rule_dpif *rule = rule_dpif_cast(rule_);
3447     ovs_mutex_destroy(&rule->stats_mutex);
3448 }
3449
3450 static void
3451 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
3452                long long int *used)
3453 {
3454     struct rule_dpif *rule = rule_dpif_cast(rule_);
3455
3456     ovs_mutex_lock(&rule->stats_mutex);
3457     *packets = rule->stats.n_packets;
3458     *bytes = rule->stats.n_bytes;
3459     *used = rule->stats.used;
3460     ovs_mutex_unlock(&rule->stats_mutex);
3461 }
3462
3463 static void
3464 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
3465                   struct ofpbuf *packet)
3466 {
3467     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3468
3469     ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
3470 }
3471
3472 static enum ofperr
3473 rule_execute(struct rule *rule, const struct flow *flow,
3474              struct ofpbuf *packet)
3475 {
3476     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
3477     ofpbuf_delete(packet);
3478     return 0;
3479 }
3480
3481 static void
3482 rule_modify_actions(struct rule *rule_, bool reset_counters)
3483     OVS_REQUIRES(ofproto_mutex)
3484 {
3485     struct rule_dpif *rule = rule_dpif_cast(rule_);
3486
3487     if (reset_counters) {
3488         ovs_mutex_lock(&rule->stats_mutex);
3489         rule->stats.n_packets = 0;
3490         rule->stats.n_bytes = 0;
3491         ovs_mutex_unlock(&rule->stats_mutex);
3492     }
3493
3494     complete_operation(rule);
3495 }
3496
3497 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
3498 {
3499     return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
3500 }
3501
3502 static struct ofgroup *
3503 group_alloc(void)
3504 {
3505     struct group_dpif *group = xzalloc(sizeof *group);
3506     return &group->up;
3507 }
3508
3509 static void
3510 group_dealloc(struct ofgroup *group_)
3511 {
3512     struct group_dpif *group = group_dpif_cast(group_);
3513     free(group);
3514 }
3515
3516 static void
3517 group_construct_stats(struct group_dpif *group)
3518     OVS_REQUIRES(group->stats_mutex)
3519 {
3520     group->packet_count = 0;
3521     group->byte_count = 0;
3522     if (!group->bucket_stats) {
3523         group->bucket_stats = xcalloc(group->up.n_buckets,
3524                                       sizeof *group->bucket_stats);
3525     } else {
3526         memset(group->bucket_stats, 0, group->up.n_buckets *
3527                sizeof *group->bucket_stats);
3528     }
3529 }
3530
3531 static enum ofperr
3532 group_construct(struct ofgroup *group_)
3533 {
3534     struct group_dpif *group = group_dpif_cast(group_);
3535     const struct ofputil_bucket *bucket;
3536
3537     /* Prevent group chaining because our locking structure makes it hard to
3538      * implement deadlock-free.  (See xlate_group_resource_check().) */
3539     LIST_FOR_EACH (bucket, list_node, &group->up.buckets) {
3540         const struct ofpact *a;
3541
3542         OFPACT_FOR_EACH (a, bucket->ofpacts, bucket->ofpacts_len) {
3543             if (a->type == OFPACT_GROUP) {
3544                 return OFPERR_OFPGMFC_CHAINING_UNSUPPORTED;
3545             }
3546         }
3547     }
3548
3549     ovs_mutex_init_adaptive(&group->stats_mutex);
3550     ovs_mutex_lock(&group->stats_mutex);
3551     group_construct_stats(group);
3552     ovs_mutex_unlock(&group->stats_mutex);
3553     return 0;
3554 }
3555
3556 static void
3557 group_destruct__(struct group_dpif *group)
3558     OVS_REQUIRES(group->stats_mutex)
3559 {
3560     free(group->bucket_stats);
3561     group->bucket_stats = NULL;
3562 }
3563
3564 static void
3565 group_destruct(struct ofgroup *group_)
3566 {
3567     struct group_dpif *group = group_dpif_cast(group_);
3568     ovs_mutex_lock(&group->stats_mutex);
3569     group_destruct__(group);
3570     ovs_mutex_unlock(&group->stats_mutex);
3571     ovs_mutex_destroy(&group->stats_mutex);
3572 }
3573
3574 static enum ofperr
3575 group_modify(struct ofgroup *group_, struct ofgroup *victim_)
3576 {
3577     struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
3578     struct group_dpif *group = group_dpif_cast(group_);
3579     struct group_dpif *victim = group_dpif_cast(victim_);
3580
3581     ovs_mutex_lock(&group->stats_mutex);
3582     if (victim->up.n_buckets < group->up.n_buckets) {
3583         group_destruct__(group);
3584     }
3585     group_construct_stats(group);
3586     ovs_mutex_unlock(&group->stats_mutex);
3587
3588     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3589
3590     return 0;
3591 }
3592
3593 static enum ofperr
3594 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
3595 {
3596     struct group_dpif *group = group_dpif_cast(group_);
3597
3598     ovs_mutex_lock(&group->stats_mutex);
3599     ogs->packet_count = group->packet_count;
3600     ogs->byte_count = group->byte_count;
3601     memcpy(ogs->bucket_stats, group->bucket_stats,
3602            group->up.n_buckets * sizeof *group->bucket_stats);
3603     ovs_mutex_unlock(&group->stats_mutex);
3604
3605     return 0;
3606 }
3607
3608 bool
3609 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
3610                   struct group_dpif **group)
3611     OVS_TRY_RDLOCK(true, (*group)->up.rwlock)
3612 {
3613     struct ofgroup *ofgroup;
3614     bool found;
3615
3616     *group = NULL;
3617     found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
3618     *group = found ?  group_dpif_cast(ofgroup) : NULL;
3619
3620     return found;
3621 }
3622
3623 void
3624 group_dpif_release(struct group_dpif *group)
3625     OVS_RELEASES(group->up.rwlock)
3626 {
3627     ofproto_group_release(&group->up);
3628 }
3629
3630 void
3631 group_dpif_get_buckets(const struct group_dpif *group,
3632                        const struct list **buckets)
3633 {
3634     *buckets = &group->up.buckets;
3635 }
3636
3637 enum ofp11_group_type
3638 group_dpif_get_type(const struct group_dpif *group)
3639 {
3640     return group->up.type;
3641 }
3642 \f
3643 /* Sends 'packet' out 'ofport'.
3644  * May modify 'packet'.
3645  * Returns 0 if successful, otherwise a positive errno value. */
3646 int
3647 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3648 {
3649     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3650     int error;
3651
3652     error = xlate_send_packet(ofport, packet);
3653
3654     ovs_mutex_lock(&ofproto->stats_mutex);
3655     ofproto->stats.tx_packets++;
3656     ofproto->stats.tx_bytes += ofpbuf_size(packet);
3657     ovs_mutex_unlock(&ofproto->stats_mutex);
3658     return error;
3659 }
3660 \f
3661 static bool
3662 set_frag_handling(struct ofproto *ofproto_,
3663                   enum ofp_config_flags frag_handling)
3664 {
3665     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3666     if (frag_handling != OFPC_FRAG_REASM) {
3667         ofproto->backer->need_revalidate = REV_RECONFIGURE;
3668         return true;
3669     } else {
3670         return false;
3671     }
3672 }
3673
3674 static enum ofperr
3675 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
3676            const struct flow *flow,
3677            const struct ofpact *ofpacts, size_t ofpacts_len)
3678 {
3679     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3680
3681     ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
3682                                  ofpacts_len, packet);
3683     return 0;
3684 }
3685 \f
3686 /* NetFlow. */
3687
3688 static int
3689 set_netflow(struct ofproto *ofproto_,
3690             const struct netflow_options *netflow_options)
3691 {
3692     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3693
3694     if (netflow_options) {
3695         if (!ofproto->netflow) {
3696             ofproto->netflow = netflow_create();
3697             ofproto->backer->need_revalidate = REV_RECONFIGURE;
3698         }
3699         return netflow_set_options(ofproto->netflow, netflow_options);
3700     } else if (ofproto->netflow) {
3701         ofproto->backer->need_revalidate = REV_RECONFIGURE;
3702         netflow_unref(ofproto->netflow);
3703         ofproto->netflow = NULL;
3704     }
3705
3706     return 0;
3707 }
3708
3709 static void
3710 get_netflow_ids(const struct ofproto *ofproto_,
3711                 uint8_t *engine_type, uint8_t *engine_id)
3712 {
3713     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3714
3715     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
3716 }
3717 \f
3718 static struct ofproto_dpif *
3719 ofproto_dpif_lookup(const char *name)
3720 {
3721     struct ofproto_dpif *ofproto;
3722
3723     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
3724                              hash_string(name, 0), &all_ofproto_dpifs) {
3725         if (!strcmp(ofproto->up.name, name)) {
3726             return ofproto;
3727         }
3728     }
3729     return NULL;
3730 }
3731
3732 static void
3733 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
3734                           const char *argv[], void *aux OVS_UNUSED)
3735 {
3736     struct ofproto_dpif *ofproto;
3737
3738     if (argc > 1) {
3739         ofproto = ofproto_dpif_lookup(argv[1]);
3740         if (!ofproto) {
3741             unixctl_command_reply_error(conn, "no such bridge");
3742             return;
3743         }
3744         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3745         mac_learning_flush(ofproto->ml);
3746         ovs_rwlock_unlock(&ofproto->ml->rwlock);
3747     } else {
3748         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3749             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3750             mac_learning_flush(ofproto->ml);
3751             ovs_rwlock_unlock(&ofproto->ml->rwlock);
3752         }
3753     }
3754
3755     unixctl_command_reply(conn, "table successfully flushed");
3756 }
3757
3758 static struct ofport_dpif *
3759 ofbundle_get_a_port(const struct ofbundle *bundle)
3760 {
3761     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
3762                         bundle_node);
3763 }
3764
3765 static void
3766 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
3767                          const char *argv[], void *aux OVS_UNUSED)
3768 {
3769     struct ds ds = DS_EMPTY_INITIALIZER;
3770     const struct ofproto_dpif *ofproto;
3771     const struct mac_entry *e;
3772
3773     ofproto = ofproto_dpif_lookup(argv[1]);
3774     if (!ofproto) {
3775         unixctl_command_reply_error(conn, "no such bridge");
3776         return;
3777     }
3778
3779     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
3780     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3781     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3782         struct ofbundle *bundle = e->port.p;
3783         char name[OFP_MAX_PORT_NAME_LEN];
3784
3785         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
3786                                name, sizeof name);
3787         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
3788                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
3789                       mac_entry_age(ofproto->ml, e));
3790     }
3791     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3792     unixctl_command_reply(conn, ds_cstr(&ds));
3793     ds_destroy(&ds);
3794 }
3795
3796 struct trace_ctx {
3797     struct xlate_out xout;
3798     struct xlate_in xin;
3799     const struct flow *key;
3800     struct flow flow;
3801     struct flow_wildcards wc;
3802     struct ds *result;
3803 };
3804
3805 static void
3806 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
3807 {
3808     struct rule_actions *actions;
3809     ovs_be64 cookie;
3810
3811     ds_put_char_multiple(result, '\t', level);
3812     if (!rule) {
3813         ds_put_cstr(result, "No match\n");
3814         return;
3815     }
3816
3817     ovs_mutex_lock(&rule->up.mutex);
3818     cookie = rule->up.flow_cookie;
3819     ovs_mutex_unlock(&rule->up.mutex);
3820
3821     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
3822                   rule ? rule->up.table_id : 0, ntohll(cookie));
3823     cls_rule_format(&rule->up.cr, result);
3824     ds_put_char(result, '\n');
3825
3826     actions = rule_dpif_get_actions(rule);
3827
3828     ds_put_char_multiple(result, '\t', level);
3829     ds_put_cstr(result, "OpenFlow actions=");
3830     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
3831     ds_put_char(result, '\n');
3832 }
3833
3834 static void
3835 trace_format_flow(struct ds *result, int level, const char *title,
3836                   struct trace_ctx *trace)
3837 {
3838     ds_put_char_multiple(result, '\t', level);
3839     ds_put_format(result, "%s: ", title);
3840     /* Do not report unchanged flows for resubmits. */
3841     if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
3842         || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
3843         ds_put_cstr(result, "unchanged");
3844     } else {
3845         flow_format(result, &trace->xin.flow);
3846         trace->flow = trace->xin.flow;
3847     }
3848     ds_put_char(result, '\n');
3849 }
3850
3851 static void
3852 trace_format_regs(struct ds *result, int level, const char *title,
3853                   struct trace_ctx *trace)
3854 {
3855     size_t i;
3856
3857     ds_put_char_multiple(result, '\t', level);
3858     ds_put_format(result, "%s:", title);
3859     for (i = 0; i < FLOW_N_REGS; i++) {
3860         ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
3861     }
3862     ds_put_char(result, '\n');
3863 }
3864
3865 static void
3866 trace_format_odp(struct ds *result, int level, const char *title,
3867                  struct trace_ctx *trace)
3868 {
3869     struct ofpbuf *odp_actions = &trace->xout.odp_actions;
3870
3871     ds_put_char_multiple(result, '\t', level);
3872     ds_put_format(result, "%s: ", title);
3873     format_odp_actions(result, ofpbuf_data(odp_actions),
3874                                ofpbuf_size(odp_actions));
3875     ds_put_char(result, '\n');
3876 }
3877
3878 static void
3879 trace_format_megaflow(struct ds *result, int level, const char *title,
3880                       struct trace_ctx *trace)
3881 {
3882     struct match match;
3883
3884     ds_put_char_multiple(result, '\t', level);
3885     ds_put_format(result, "%s: ", title);
3886     flow_wildcards_or(&trace->wc, &trace->xout.wc, &trace->wc);
3887     match_init(&match, trace->key, &trace->wc);
3888     match_format(&match, result, OFP_DEFAULT_PRIORITY);
3889     ds_put_char(result, '\n');
3890 }
3891
3892 static void
3893 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
3894 {
3895     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
3896     struct ds *result = trace->result;
3897
3898     ds_put_char(result, '\n');
3899     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
3900     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
3901     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
3902     trace_format_megaflow(result, recurse + 1, "Resubmitted megaflow", trace);
3903     trace_format_rule(result, recurse + 1, rule);
3904 }
3905
3906 static void
3907 trace_report(struct xlate_in *xin, const char *s, int recurse)
3908 {
3909     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
3910     struct ds *result = trace->result;
3911
3912     ds_put_char_multiple(result, '\t', recurse);
3913     ds_put_cstr(result, s);
3914     ds_put_char(result, '\n');
3915 }
3916
3917 /* Parses the 'argc' elements of 'argv', ignoring argv[0].  The following
3918  * forms are supported:
3919  *
3920  *     - [dpname] odp_flow [-generate | packet]
3921  *     - bridge br_flow [-generate | packet]
3922  *
3923  * On success, initializes '*ofprotop' and 'flow' and returns NULL.  On failure
3924  * returns a nonnull malloced error message. */
3925 static char * WARN_UNUSED_RESULT
3926 parse_flow_and_packet(int argc, const char *argv[],
3927                       struct ofproto_dpif **ofprotop, struct flow *flow,
3928                       struct ofpbuf **packetp)
3929 {
3930     const struct dpif_backer *backer = NULL;
3931     const char *error = NULL;
3932     char *m_err = NULL;
3933     struct simap port_names = SIMAP_INITIALIZER(&port_names);
3934     struct ofpbuf *packet;
3935     struct ofpbuf odp_key;
3936     struct ofpbuf odp_mask;
3937
3938     ofpbuf_init(&odp_key, 0);
3939     ofpbuf_init(&odp_mask, 0);
3940
3941     /* Handle "-generate" or a hex string as the last argument. */
3942     if (!strcmp(argv[argc - 1], "-generate")) {
3943         packet = ofpbuf_new(0);
3944         argc--;
3945     } else {
3946         error = eth_from_hex(argv[argc - 1], &packet);
3947         if (!error) {
3948             argc--;
3949         } else if (argc == 4) {
3950             /* The 3-argument form must end in "-generate' or a hex string. */
3951             goto exit;
3952         }
3953         error = NULL;
3954     }
3955
3956     /* odp_flow can have its in_port specified as a name instead of port no.
3957      * We do not yet know whether a given flow is a odp_flow or a br_flow.
3958      * But, to know whether a flow is odp_flow through odp_flow_from_string(),
3959      * we need to create a simap of name to port no. */
3960     if (argc == 3) {
3961         const char *dp_type;
3962         if (!strncmp(argv[1], "ovs-", 4)) {
3963             dp_type = argv[1] + 4;
3964         } else {
3965             dp_type = argv[1];
3966         }
3967         backer = shash_find_data(&all_dpif_backers, dp_type);
3968     } else if (argc == 2) {
3969         struct shash_node *node;
3970         if (shash_count(&all_dpif_backers) == 1) {
3971             node = shash_first(&all_dpif_backers);
3972             backer = node->data;
3973         }
3974     } else {
3975         error = "Syntax error";
3976         goto exit;
3977     }
3978     if (backer && backer->dpif) {
3979         struct dpif_port dpif_port;
3980         struct dpif_port_dump port_dump;
3981         DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
3982             simap_put(&port_names, dpif_port.name,
3983                       odp_to_u32(dpif_port.port_no));
3984         }
3985     }
3986
3987     /* Parse the flow and determine whether a datapath or
3988      * bridge is specified. If function odp_flow_key_from_string()
3989      * returns 0, the flow is a odp_flow. If function
3990      * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
3991     if (!odp_flow_from_string(argv[argc - 1], &port_names,
3992                               &odp_key, &odp_mask)) {
3993         if (!backer) {
3994             error = "Cannot find the datapath";
3995             goto exit;
3996         }
3997
3998         if (xlate_receive(backer, NULL, ofpbuf_data(&odp_key),
3999                           ofpbuf_size(&odp_key), flow,
4000                           ofprotop, NULL, NULL, NULL, NULL)) {
4001             error = "Invalid datapath flow";
4002             goto exit;
4003         }
4004     } else {
4005         char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4006
4007         if (err) {
4008             m_err = xasprintf("Bad flow syntax: %s", err);
4009             free(err);
4010             goto exit;
4011         } else {
4012             if (argc != 3) {
4013                 error = "Must specify bridge name";
4014                 goto exit;
4015             }
4016
4017             *ofprotop = ofproto_dpif_lookup(argv[1]);
4018             if (!*ofprotop) {
4019                 error = "Unknown bridge name";
4020                 goto exit;
4021             }
4022         }
4023     }
4024
4025     /* Generate a packet, if requested. */
4026     if (packet) {
4027         if (!ofpbuf_size(packet)) {
4028             flow_compose(packet, flow);
4029         } else {
4030             struct pkt_metadata md = pkt_metadata_from_flow(flow);
4031
4032             /* Use the metadata from the flow and the packet argument
4033              * to reconstruct the flow. */
4034             flow_extract(packet, &md, flow);
4035         }
4036     }
4037
4038 exit:
4039     if (error && !m_err) {
4040         m_err = xstrdup(error);
4041     }
4042     if (m_err) {
4043         ofpbuf_delete(packet);
4044         packet = NULL;
4045     }
4046     *packetp = packet;
4047     ofpbuf_uninit(&odp_key);
4048     ofpbuf_uninit(&odp_mask);
4049     simap_destroy(&port_names);
4050     return m_err;
4051 }
4052
4053 static void
4054 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4055                       void *aux OVS_UNUSED)
4056 {
4057     struct ofproto_dpif *ofproto;
4058     struct ofpbuf *packet;
4059     char *error;
4060     struct flow flow;
4061
4062     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4063     if (!error) {
4064         struct ds result;
4065
4066         ds_init(&result);
4067         ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4068         unixctl_command_reply(conn, ds_cstr(&result));
4069         ds_destroy(&result);
4070         ofpbuf_delete(packet);
4071     } else {
4072         unixctl_command_reply_error(conn, error);
4073         free(error);
4074     }
4075 }
4076
4077 static void
4078 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4079                               const char *argv[], void *aux OVS_UNUSED)
4080 {
4081     enum ofputil_protocol usable_protocols;
4082     struct ofproto_dpif *ofproto;
4083     bool enforce_consistency;
4084     struct ofpbuf ofpacts;
4085     struct ofpbuf *packet;
4086     struct ds result;
4087     struct flow flow;
4088     uint16_t in_port;
4089
4090     /* Three kinds of error return values! */
4091     enum ofperr retval;
4092     char *error;
4093
4094     packet = NULL;
4095     ds_init(&result);
4096     ofpbuf_init(&ofpacts, 0);
4097
4098     /* Parse actions. */
4099     error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols);
4100     if (error) {
4101         unixctl_command_reply_error(conn, error);
4102         free(error);
4103         goto exit;
4104     }
4105
4106     /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4107      * consistency between the flow and the actions.  With -consistent, we
4108      * enforce consistency even for a flow supported in OpenFlow 1.0. */
4109     if (!strcmp(argv[1], "-consistent")) {
4110         enforce_consistency = true;
4111         argv++;
4112         argc--;
4113     } else {
4114         enforce_consistency = false;
4115     }
4116
4117     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4118     if (error) {
4119         unixctl_command_reply_error(conn, error);
4120         free(error);
4121         goto exit;
4122     }
4123
4124     /* Do the same checks as handle_packet_out() in ofproto.c.
4125      *
4126      * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
4127      * strictly correct because these actions aren't in any table, but it's OK
4128      * because it 'table_id' is used only to check goto_table instructions, but
4129      * packet-outs take a list of actions and therefore it can't include
4130      * instructions.
4131      *
4132      * We skip the "meter" check here because meter is an instruction, not an
4133      * action, and thus cannot appear in ofpacts. */
4134     in_port = ofp_to_u16(flow.in_port.ofp_port);
4135     if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
4136         unixctl_command_reply_error(conn, "invalid in_port");
4137         goto exit;
4138     }
4139     if (enforce_consistency) {
4140         retval = ofpacts_check_consistency(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts),
4141                                            &flow, u16_to_ofp(ofproto->up.max_ports),
4142                                            0, 0, usable_protocols);
4143     } else {
4144         retval = ofpacts_check(ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &flow,
4145                                u16_to_ofp(ofproto->up.max_ports), 0, 0,
4146                                &usable_protocols);
4147     }
4148
4149     if (retval) {
4150         ds_clear(&result);
4151         ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
4152         unixctl_command_reply_error(conn, ds_cstr(&result));
4153         goto exit;
4154     }
4155
4156     ofproto_trace(ofproto, &flow, packet,
4157                   ofpbuf_data(&ofpacts), ofpbuf_size(&ofpacts), &result);
4158     unixctl_command_reply(conn, ds_cstr(&result));
4159
4160 exit:
4161     ds_destroy(&result);
4162     ofpbuf_delete(packet);
4163     ofpbuf_uninit(&ofpacts);
4164 }
4165
4166 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4167  * description of the results to 'ds'.
4168  *
4169  * The trace follows a packet with the specified 'flow' through the flow
4170  * table.  'packet' may be nonnull to trace an actual packet, with consequent
4171  * side effects (if it is nonnull then its flow must be 'flow').
4172  *
4173  * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4174  * trace, otherwise the actions are determined by a flow table lookup. */
4175 static void
4176 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
4177               const struct ofpbuf *packet,
4178               const struct ofpact ofpacts[], size_t ofpacts_len,
4179               struct ds *ds)
4180 {
4181     struct rule_dpif *rule;
4182     struct trace_ctx trace;
4183
4184     ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
4185     ds_put_cstr(ds, "Flow: ");
4186     flow_format(ds, flow);
4187     ds_put_char(ds, '\n');
4188
4189     flow_wildcards_init_catchall(&trace.wc);
4190     if (ofpacts) {
4191         rule = NULL;
4192     } else {
4193         rule_dpif_lookup(ofproto, flow, &trace.wc, &rule);
4194
4195         trace_format_rule(ds, 0, rule);
4196         if (rule == ofproto->miss_rule) {
4197             ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
4198         } else if (rule == ofproto->no_packet_in_rule) {
4199             ds_put_cstr(ds, "\nNo match, packets dropped because "
4200                         "OFPPC_NO_PACKET_IN is set on in_port.\n");
4201         } else if (rule == ofproto->drop_frags_rule) {
4202             ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
4203                         "and the fragment handling mode is \"drop\".\n");
4204         }
4205     }
4206
4207     if (rule || ofpacts) {
4208         trace.result = ds;
4209         trace.key = flow; /* Original flow key, used for megaflow. */
4210         trace.flow = *flow; /* May be modified by actions. */
4211         xlate_in_init(&trace.xin, ofproto, flow, rule, ntohs(flow->tcp_flags),
4212                       packet);
4213         if (ofpacts) {
4214             trace.xin.ofpacts = ofpacts;
4215             trace.xin.ofpacts_len = ofpacts_len;
4216         }
4217         trace.xin.resubmit_hook = trace_resubmit;
4218         trace.xin.report_hook = trace_report;
4219
4220         xlate_actions(&trace.xin, &trace.xout);
4221
4222         ds_put_char(ds, '\n');
4223         trace_format_flow(ds, 0, "Final flow", &trace);
4224         trace_format_megaflow(ds, 0, "Megaflow", &trace);
4225
4226         ds_put_cstr(ds, "Datapath actions: ");
4227         format_odp_actions(ds, ofpbuf_data(&trace.xout.odp_actions),
4228                            ofpbuf_size(&trace.xout.odp_actions));
4229
4230         if (trace.xout.slow) {
4231             enum slow_path_reason slow;
4232
4233             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
4234                         "slow path because it:");
4235
4236             slow = trace.xout.slow;
4237             while (slow) {
4238                 enum slow_path_reason bit = rightmost_1bit(slow);
4239
4240                 ds_put_format(ds, "\n\t- %s.",
4241                               slow_path_reason_to_explanation(bit));
4242
4243                 slow &= ~bit;
4244             }
4245         }
4246
4247         xlate_out_uninit(&trace.xout);
4248     }
4249
4250     rule_dpif_unref(rule);
4251 }
4252
4253 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
4254  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
4255  * to destroy 'ofproto_shash' and free the returned value. */
4256 static const struct shash_node **
4257 get_ofprotos(struct shash *ofproto_shash)
4258 {
4259     const struct ofproto_dpif *ofproto;
4260
4261     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4262         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
4263         shash_add_nocopy(ofproto_shash, name, ofproto);
4264     }
4265
4266     return shash_sort(ofproto_shash);
4267 }
4268
4269 static void
4270 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
4271                               const char *argv[] OVS_UNUSED,
4272                               void *aux OVS_UNUSED)
4273 {
4274     struct ds ds = DS_EMPTY_INITIALIZER;
4275     struct shash ofproto_shash;
4276     const struct shash_node **sorted_ofprotos;
4277     int i;
4278
4279     shash_init(&ofproto_shash);
4280     sorted_ofprotos = get_ofprotos(&ofproto_shash);
4281     for (i = 0; i < shash_count(&ofproto_shash); i++) {
4282         const struct shash_node *node = sorted_ofprotos[i];
4283         ds_put_format(&ds, "%s\n", node->name);
4284     }
4285
4286     shash_destroy(&ofproto_shash);
4287     free(sorted_ofprotos);
4288
4289     unixctl_command_reply(conn, ds_cstr(&ds));
4290     ds_destroy(&ds);
4291 }
4292
4293 static void
4294 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
4295 {
4296     const struct shash_node **ofprotos;
4297     struct dpif_dp_stats dp_stats;
4298     struct shash ofproto_shash;
4299     size_t i;
4300
4301     dpif_get_dp_stats(backer->dpif, &dp_stats);
4302
4303     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
4304                   dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
4305
4306     shash_init(&ofproto_shash);
4307     ofprotos = get_ofprotos(&ofproto_shash);
4308     for (i = 0; i < shash_count(&ofproto_shash); i++) {
4309         struct ofproto_dpif *ofproto = ofprotos[i]->data;
4310         const struct shash_node **ports;
4311         size_t j;
4312
4313         if (ofproto->backer != backer) {
4314             continue;
4315         }
4316
4317         ds_put_format(ds, "\t%s:\n", ofproto->up.name);
4318
4319         ports = shash_sort(&ofproto->up.port_by_name);
4320         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
4321             const struct shash_node *node = ports[j];
4322             struct ofport *ofport = node->data;
4323             struct smap config;
4324             odp_port_t odp_port;
4325
4326             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
4327                           ofport->ofp_port);
4328
4329             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
4330             if (odp_port != ODPP_NONE) {
4331                 ds_put_format(ds, "%"PRIu32":", odp_port);
4332             } else {
4333                 ds_put_cstr(ds, "none:");
4334             }
4335
4336             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
4337
4338             smap_init(&config);
4339             if (!netdev_get_config(ofport->netdev, &config)) {
4340                 const struct smap_node **nodes;
4341                 size_t i;
4342
4343                 nodes = smap_sort(&config);
4344                 for (i = 0; i < smap_count(&config); i++) {
4345                     const struct smap_node *node = nodes[i];
4346                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
4347                                   node->key, node->value);
4348                 }
4349                 free(nodes);
4350             }
4351             smap_destroy(&config);
4352
4353             ds_put_char(ds, ')');
4354             ds_put_char(ds, '\n');
4355         }
4356         free(ports);
4357     }
4358     shash_destroy(&ofproto_shash);
4359     free(ofprotos);
4360 }
4361
4362 static void
4363 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4364                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4365 {
4366     struct ds ds = DS_EMPTY_INITIALIZER;
4367     const struct shash_node **backers;
4368     int i;
4369
4370     backers = shash_sort(&all_dpif_backers);
4371     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
4372         dpif_show_backer(backers[i]->data, &ds);
4373     }
4374     free(backers);
4375
4376     unixctl_command_reply(conn, ds_cstr(&ds));
4377     ds_destroy(&ds);
4378 }
4379
4380 static bool
4381 ofproto_dpif_contains_flow(const struct ofproto_dpif *ofproto,
4382                            const struct nlattr *key, size_t key_len)
4383 {
4384     struct ofproto_dpif *ofp;
4385     struct flow flow;
4386
4387     xlate_receive(ofproto->backer, NULL, key, key_len, &flow, &ofp,
4388                   NULL, NULL, NULL, NULL);
4389     return ofp == ofproto;
4390 }
4391
4392 static void
4393 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
4394                                 int argc OVS_UNUSED, const char *argv[],
4395                                 void *aux OVS_UNUSED)
4396 {
4397     struct ds ds = DS_EMPTY_INITIALIZER;
4398     const struct dpif_flow_stats *stats;
4399     const struct ofproto_dpif *ofproto;
4400     struct dpif_flow_dump flow_dump;
4401     const struct nlattr *actions;
4402     const struct nlattr *mask;
4403     const struct nlattr *key;
4404     size_t actions_len;
4405     size_t mask_len;
4406     size_t key_len;
4407     bool verbosity = false;
4408     struct dpif_port dpif_port;
4409     struct dpif_port_dump port_dump;
4410     struct hmap portno_names;
4411     void *state = NULL;
4412     int error;
4413
4414     ofproto = ofproto_dpif_lookup(argv[argc - 1]);
4415     if (!ofproto) {
4416         unixctl_command_reply_error(conn, "no such bridge");
4417         return;
4418     }
4419
4420     if (argc > 2 && !strcmp(argv[1], "-m")) {
4421         verbosity = true;
4422     }
4423
4424     hmap_init(&portno_names);
4425     DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
4426         odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
4427     }
4428
4429     ds_init(&ds);
4430     error = dpif_flow_dump_start(&flow_dump, ofproto->backer->dpif);
4431     if (error) {
4432         goto exit;
4433     }
4434     dpif_flow_dump_state_init(ofproto->backer->dpif, &state);
4435     while (dpif_flow_dump_next(&flow_dump, state, &key, &key_len,
4436                                &mask, &mask_len, &actions, &actions_len,
4437                                &stats)) {
4438         if (!ofproto_dpif_contains_flow(ofproto, key, key_len)) {
4439             continue;
4440         }
4441
4442         odp_flow_format(key, key_len, mask, mask_len, &portno_names, &ds,
4443                         verbosity);
4444         ds_put_cstr(&ds, ", ");
4445         dpif_flow_stats_format(stats, &ds);
4446         ds_put_cstr(&ds, ", actions:");
4447         format_odp_actions(&ds, actions, actions_len);
4448         ds_put_char(&ds, '\n');
4449     }
4450     dpif_flow_dump_state_uninit(ofproto->backer->dpif, state);
4451     error = dpif_flow_dump_done(&flow_dump);
4452
4453 exit:
4454     if (error) {
4455         ds_clear(&ds);
4456         ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
4457         unixctl_command_reply_error(conn, ds_cstr(&ds));
4458     } else {
4459         unixctl_command_reply(conn, ds_cstr(&ds));
4460     }
4461     odp_portno_names_destroy(&portno_names);
4462     hmap_destroy(&portno_names);
4463     ds_destroy(&ds);
4464 }
4465
4466 static void
4467 ofproto_dpif_unixctl_init(void)
4468 {
4469     static bool registered;
4470     if (registered) {
4471         return;
4472     }
4473     registered = true;
4474
4475     unixctl_command_register(
4476         "ofproto/trace",
4477         "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
4478         1, 3, ofproto_unixctl_trace, NULL);
4479     unixctl_command_register(
4480         "ofproto/trace-packet-out",
4481         "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
4482         2, 6, ofproto_unixctl_trace_actions, NULL);
4483     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
4484                              ofproto_unixctl_fdb_flush, NULL);
4485     unixctl_command_register("fdb/show", "bridge", 1, 1,
4486                              ofproto_unixctl_fdb_show, NULL);
4487     unixctl_command_register("dpif/dump-dps", "", 0, 0,
4488                              ofproto_unixctl_dpif_dump_dps, NULL);
4489     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
4490                              NULL);
4491     unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
4492                              ofproto_unixctl_dpif_dump_flows, NULL);
4493 }
4494
4495
4496 /* Returns true if 'rule' is an internal rule, false otherwise. */
4497 bool
4498 rule_is_internal(const struct rule *rule)
4499 {
4500     return rule->table_id == TBL_INTERNAL;
4501 }
4502 \f
4503 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4504  *
4505  * This is deprecated.  It is only for compatibility with broken device drivers
4506  * in old versions of Linux that do not properly support VLANs when VLAN
4507  * devices are not used.  When broken device drivers are no longer in
4508  * widespread use, we will delete these interfaces. */
4509
4510 static int
4511 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
4512 {
4513     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
4514     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
4515
4516     if (realdev_ofp_port == ofport->realdev_ofp_port
4517         && vid == ofport->vlandev_vid) {
4518         return 0;
4519     }
4520
4521     ofproto->backer->need_revalidate = REV_RECONFIGURE;
4522
4523     if (ofport->realdev_ofp_port) {
4524         vsp_remove(ofport);
4525     }
4526     if (realdev_ofp_port && ofport->bundle) {
4527         /* vlandevs are enslaved to their realdevs, so they are not allowed to
4528          * themselves be part of a bundle. */
4529         bundle_set(ofport_->ofproto, ofport->bundle, NULL);
4530     }
4531
4532     ofport->realdev_ofp_port = realdev_ofp_port;
4533     ofport->vlandev_vid = vid;
4534
4535     if (realdev_ofp_port) {
4536         vsp_add(ofport, realdev_ofp_port, vid);
4537     }
4538
4539     return 0;
4540 }
4541
4542 static uint32_t
4543 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
4544 {
4545     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
4546 }
4547
4548 bool
4549 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
4550     OVS_EXCLUDED(ofproto->vsp_mutex)
4551 {
4552     /* hmap_is_empty is thread safe. */
4553     return !hmap_is_empty(&ofproto->realdev_vid_map);
4554 }
4555
4556
4557 static ofp_port_t
4558 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
4559                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4560     OVS_REQUIRES(ofproto->vsp_mutex)
4561 {
4562     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
4563         int vid = vlan_tci_to_vid(vlan_tci);
4564         const struct vlan_splinter *vsp;
4565
4566         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
4567                                  hash_realdev_vid(realdev_ofp_port, vid),
4568                                  &ofproto->realdev_vid_map) {
4569             if (vsp->realdev_ofp_port == realdev_ofp_port
4570                 && vsp->vid == vid) {
4571                 return vsp->vlandev_ofp_port;
4572             }
4573         }
4574     }
4575     return realdev_ofp_port;
4576 }
4577
4578 /* Returns the OFP port number of the Linux VLAN device that corresponds to
4579  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
4580  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
4581  * 'vlan_tci' 9, it would return the port number of eth0.9.
4582  *
4583  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
4584  * function just returns its 'realdev_ofp_port' argument. */
4585 ofp_port_t
4586 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
4587                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
4588     OVS_EXCLUDED(ofproto->vsp_mutex)
4589 {
4590     ofp_port_t ret;
4591
4592     /* hmap_is_empty is thread safe, see if we can return immediately. */
4593     if (hmap_is_empty(&ofproto->realdev_vid_map)) {
4594         return realdev_ofp_port;
4595     }
4596     ovs_mutex_lock(&ofproto->vsp_mutex);
4597     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
4598     ovs_mutex_unlock(&ofproto->vsp_mutex);
4599     return ret;
4600 }
4601
4602 static struct vlan_splinter *
4603 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
4604 {
4605     struct vlan_splinter *vsp;
4606
4607     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
4608                              hash_ofp_port(vlandev_ofp_port),
4609                              &ofproto->vlandev_map) {
4610         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
4611             return vsp;
4612         }
4613     }
4614
4615     return NULL;
4616 }
4617
4618 /* Returns the OpenFlow port number of the "real" device underlying the Linux
4619  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
4620  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
4621  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
4622  * eth0 and store 9 in '*vid'.
4623  *
4624  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
4625  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
4626  * always does.*/
4627 static ofp_port_t
4628 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
4629                        ofp_port_t vlandev_ofp_port, int *vid)
4630     OVS_REQUIRES(ofproto->vsp_mutex)
4631 {
4632     if (!hmap_is_empty(&ofproto->vlandev_map)) {
4633         const struct vlan_splinter *vsp;
4634
4635         vsp = vlandev_find(ofproto, vlandev_ofp_port);
4636         if (vsp) {
4637             if (vid) {
4638                 *vid = vsp->vid;
4639             }
4640             return vsp->realdev_ofp_port;
4641         }
4642     }
4643     return 0;
4644 }
4645
4646 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
4647  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
4648  * 'flow->in_port' to the "real" device backing the VLAN device, sets
4649  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Otherwise (which is
4650  * always the case unless VLAN splinters are enabled), returns false without
4651  * making any changes. */
4652 bool
4653 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
4654     OVS_EXCLUDED(ofproto->vsp_mutex)
4655 {
4656     ofp_port_t realdev;
4657     int vid;
4658
4659     /* hmap_is_empty is thread safe. */
4660     if (hmap_is_empty(&ofproto->vlandev_map)) {
4661         return false;
4662     }
4663
4664     ovs_mutex_lock(&ofproto->vsp_mutex);
4665     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
4666     ovs_mutex_unlock(&ofproto->vsp_mutex);
4667     if (!realdev) {
4668         return false;
4669     }
4670
4671     /* Cause the flow to be processed as if it came in on the real device with
4672      * the VLAN device's VLAN ID. */
4673     flow->in_port.ofp_port = realdev;
4674     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
4675     return true;
4676 }
4677
4678 static void
4679 vsp_remove(struct ofport_dpif *port)
4680 {
4681     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4682     struct vlan_splinter *vsp;
4683
4684     ovs_mutex_lock(&ofproto->vsp_mutex);
4685     vsp = vlandev_find(ofproto, port->up.ofp_port);
4686     if (vsp) {
4687         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
4688         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
4689         free(vsp);
4690
4691         port->realdev_ofp_port = 0;
4692     } else {
4693         VLOG_ERR("missing vlan device record");
4694     }
4695     ovs_mutex_unlock(&ofproto->vsp_mutex);
4696 }
4697
4698 static void
4699 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
4700 {
4701     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
4702
4703     ovs_mutex_lock(&ofproto->vsp_mutex);
4704     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
4705         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
4706             == realdev_ofp_port)) {
4707         struct vlan_splinter *vsp;
4708
4709         vsp = xmalloc(sizeof *vsp);
4710         vsp->realdev_ofp_port = realdev_ofp_port;
4711         vsp->vlandev_ofp_port = port->up.ofp_port;
4712         vsp->vid = vid;
4713
4714         port->realdev_ofp_port = realdev_ofp_port;
4715
4716         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
4717                     hash_ofp_port(port->up.ofp_port));
4718         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
4719                     hash_realdev_vid(realdev_ofp_port, vid));
4720     } else {
4721         VLOG_ERR("duplicate vlan device record");
4722     }
4723     ovs_mutex_unlock(&ofproto->vsp_mutex);
4724 }
4725
4726 static odp_port_t
4727 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
4728 {
4729     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
4730     return ofport ? ofport->odp_port : ODPP_NONE;
4731 }
4732
4733 struct ofport_dpif *
4734 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
4735 {
4736     struct ofport_dpif *port;
4737
4738     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
4739     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
4740                              &backer->odp_to_ofport_map) {
4741         if (port->odp_port == odp_port) {
4742             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4743             return port;
4744         }
4745     }
4746
4747     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
4748     return NULL;
4749 }
4750
4751 static ofp_port_t
4752 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
4753 {
4754     struct ofport_dpif *port;
4755
4756     port = odp_port_to_ofport(ofproto->backer, odp_port);
4757     if (port && &ofproto->up == port->up.ofproto) {
4758         return port->up.ofp_port;
4759     } else {
4760         return OFPP_NONE;
4761     }
4762 }
4763
4764 uint32_t
4765 ofproto_dpif_alloc_recirc_id(struct ofproto_dpif *ofproto)
4766 {
4767     struct dpif_backer *backer = ofproto->backer;
4768
4769     return  recirc_id_alloc(backer->rid_pool);
4770 }
4771
4772 void
4773 ofproto_dpif_free_recirc_id(struct ofproto_dpif *ofproto, uint32_t recirc_id)
4774 {
4775     struct dpif_backer *backer = ofproto->backer;
4776
4777     recirc_id_free(backer->rid_pool, recirc_id);
4778 }
4779
4780 int
4781 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
4782                                struct match *match, int priority,
4783                                const struct ofpbuf *ofpacts,
4784                                struct rule **rulep)
4785 {
4786     struct ofputil_flow_mod fm;
4787     struct rule_dpif *rule;
4788     int error;
4789
4790     fm.match = *match;
4791     fm.priority = priority;
4792     fm.new_cookie = htonll(0);
4793     fm.cookie = htonll(0);
4794     fm.cookie_mask = htonll(0);
4795     fm.modify_cookie = false;
4796     fm.table_id = TBL_INTERNAL;
4797     fm.command = OFPFC_ADD;
4798     fm.idle_timeout = 0;
4799     fm.hard_timeout = 0;
4800     fm.buffer_id = 0;
4801     fm.out_port = 0;
4802     fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
4803     fm.ofpacts = ofpbuf_data(ofpacts);
4804     fm.ofpacts_len = ofpbuf_size(ofpacts);
4805
4806     error = ofproto_flow_mod(&ofproto->up, &fm);
4807     if (error) {
4808         VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
4809                     ofperr_to_string(error));
4810         *rulep = NULL;
4811         return error;
4812     }
4813
4814     rule = rule_dpif_lookup_in_table(ofproto, TBL_INTERNAL, &match->flow,
4815                                      &match->wc);
4816     if (rule) {
4817         rule_dpif_unref(rule);
4818         *rulep = &rule->up;
4819     } else {
4820         OVS_NOT_REACHED();
4821     }
4822     return 0;
4823 }
4824
4825 int
4826 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
4827                                   struct match *match, int priority)
4828 {
4829     struct ofputil_flow_mod fm;
4830     int error;
4831
4832     fm.match = *match;
4833     fm.priority = priority;
4834     fm.new_cookie = htonll(0);
4835     fm.cookie = htonll(0);
4836     fm.cookie_mask = htonll(0);
4837     fm.modify_cookie = false;
4838     fm.table_id = TBL_INTERNAL;
4839     fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
4840     fm.command = OFPFC_DELETE_STRICT;
4841
4842     error = ofproto_flow_mod(&ofproto->up, &fm);
4843     if (error) {
4844         VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
4845                     ofperr_to_string(error));
4846         return error;
4847     }
4848
4849     return 0;
4850 }
4851
4852 const struct ofproto_class ofproto_dpif_class = {
4853     init,
4854     enumerate_types,
4855     enumerate_names,
4856     del,
4857     port_open_type,
4858     type_run,
4859     type_wait,
4860     alloc,
4861     construct,
4862     destruct,
4863     dealloc,
4864     run,
4865     wait,
4866     NULL,                       /* get_memory_usage. */
4867     type_get_memory_usage,
4868     flush,
4869     get_features,
4870     get_tables,
4871     port_alloc,
4872     port_construct,
4873     port_destruct,
4874     port_dealloc,
4875     port_modified,
4876     port_reconfigured,
4877     port_query_by_name,
4878     port_add,
4879     port_del,
4880     port_get_stats,
4881     port_dump_start,
4882     port_dump_next,
4883     port_dump_done,
4884     port_poll,
4885     port_poll_wait,
4886     port_is_lacp_current,
4887     NULL,                       /* rule_choose_table */
4888     rule_alloc,
4889     rule_construct,
4890     rule_insert,
4891     rule_delete,
4892     rule_destruct,
4893     rule_dealloc,
4894     rule_get_stats,
4895     rule_execute,
4896     rule_modify_actions,
4897     set_frag_handling,
4898     packet_out,
4899     set_netflow,
4900     get_netflow_ids,
4901     set_sflow,
4902     set_ipfix,
4903     set_cfm,
4904     get_cfm_status,
4905     set_bfd,
4906     get_bfd_status,
4907     set_stp,
4908     get_stp_status,
4909     set_stp_port,
4910     get_stp_port_status,
4911     get_stp_port_stats,
4912     set_queues,
4913     bundle_set,
4914     bundle_remove,
4915     mirror_set__,
4916     mirror_get_stats__,
4917     set_flood_vlans,
4918     is_mirror_output_bundle,
4919     forward_bpdu_changed,
4920     set_mac_table_config,
4921     set_realdev,
4922     NULL,                       /* meter_get_features */
4923     NULL,                       /* meter_set */
4924     NULL,                       /* meter_get */
4925     NULL,                       /* meter_del */
4926     group_alloc,                /* group_alloc */
4927     group_construct,            /* group_construct */
4928     group_destruct,             /* group_destruct */
4929     group_dealloc,              /* group_dealloc */
4930     group_modify,               /* group_modify */
4931     group_get_stats,            /* group_get_stats */
4932 };