ofproto/ofproto-dpif.c

   1 /*
   2  * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015 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 "ovs-lldp.h"
  33 #include "dpif.h"
  34 #include "dynamic-string.h"
  35 #include "fail-open.h"
  36 #include "guarded-list.h"
  37 #include "hmapx.h"
  38 #include "lacp.h"
  39 #include "learn.h"
  40 #include "mac-learning.h"
  41 #include "mcast-snooping.h"
  42 #include "meta-flow.h"
  43 #include "multipath.h"
  44 #include "netdev-vport.h"
  45 #include "netdev.h"
  46 #include "netlink.h"
  47 #include "nx-match.h"
  48 #include "odp-util.h"
  49 #include "odp-execute.h"
  50 #include "ofp-util.h"
  51 #include "ofpbuf.h"
  52 #include "ofp-actions.h"
  53 #include "ofp-parse.h"
  54 #include "ofp-print.h"
  55 #include "ofproto-dpif-ipfix.h"
  56 #include "ofproto-dpif-mirror.h"
  57 #include "ofproto-dpif-monitor.h"
  58 #include "ofproto-dpif-rid.h"
  59 #include "ofproto-dpif-sflow.h"
  60 #include "ofproto-dpif-upcall.h"
  61 #include "ofproto-dpif-xlate.h"
  62 #include "poll-loop.h"
  63 #include "ovs-rcu.h"
  64 #include "ovs-router.h"
  65 #include "seq.h"
  66 #include "simap.h"
  67 #include "smap.h"
  68 #include "timer.h"
  69 #include "tunnel.h"
  70 #include "unaligned.h"
  71 #include "unixctl.h"
  72 #include "vlan-bitmap.h"
  73 #include "openvswitch/vlog.h"
  74
  75 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  76
  77 COVERAGE_DEFINE(ofproto_dpif_expired);
  78 COVERAGE_DEFINE(packet_in_overflow);
  79
  80 struct flow_miss;
  81
  82 struct rule_dpif {
  83     struct rule up;
  84
  85     /* These statistics:
  86      *
  87      *   - Do include packets and bytes from datapath flows which have not
  88      *   recently been processed by a revalidator. */
  89     struct ovs_mutex stats_mutex;
  90     struct dpif_flow_stats stats OVS_GUARDED;
  91
  92    /* In non-NULL, will point to a new rule (for which a reference is held) to
  93     * which all the stats updates should be forwarded. This exists only
  94     * transitionally when flows are replaced.
  95     *
  96     * Protected by stats_mutex.  If both 'rule->stats_mutex' and
  97     * 'rule->new_rule->stats_mutex' must be held together, acquire them in that
  98     * order, */
  99     struct rule_dpif *new_rule OVS_GUARDED;
 100
 101     /* If non-zero then the recirculation id that has
 102      * been allocated for use with this rule.
 103      * The recirculation id and associated internal flow should
 104      * be freed when the rule is freed */
 105     uint32_t recirc_id;
 106 };
 107
 108 /* RULE_CAST() depends on this. */
 109 BUILD_ASSERT_DECL(offsetof(struct rule_dpif, up) == 0);
 110
 111 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes,
 112                            long long int *used);
 113 static struct rule_dpif *rule_dpif_cast(const struct rule *);
 114 static void rule_expire(struct rule_dpif *);
 115
 116 struct group_dpif {
 117     struct ofgroup up;
 118
 119     /* These statistics:
 120      *
 121      *   - Do include packets and bytes from datapath flows which have not
 122      *   recently been processed by a revalidator. */
 123     struct ovs_mutex stats_mutex;
 124     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 125     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 126 };
 127
 128 struct ofbundle {
 129     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
 130     struct ofproto_dpif *ofproto; /* Owning ofproto. */
 131     void *aux;                  /* Key supplied by ofproto's client. */
 132     char *name;                 /* Identifier for log messages. */
 133
 134     /* Configuration. */
 135     struct ovs_list ports;      /* Contains "struct ofport"s. */
 136     enum port_vlan_mode vlan_mode; /* VLAN mode */
 137     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
 138     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
 139                                  * NULL if all VLANs are trunked. */
 140     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 141     struct bond *bond;          /* Nonnull iff more than one port. */
 142     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 143
 144     /* Status. */
 145     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 146 };
 147
 148 static void bundle_remove(struct ofport *);
 149 static void bundle_update(struct ofbundle *);
 150 static void bundle_destroy(struct ofbundle *);
 151 static void bundle_del_port(struct ofport_dpif *);
 152 static void bundle_run(struct ofbundle *);
 153 static void bundle_wait(struct ofbundle *);
 154 static void bundle_flush_macs(struct ofbundle *, bool);
 155 static void bundle_move(struct ofbundle *, struct ofbundle *);
 156
 157 static void stp_run(struct ofproto_dpif *ofproto);
 158 static void stp_wait(struct ofproto_dpif *ofproto);
 159 static int set_stp_port(struct ofport *,
 160                         const struct ofproto_port_stp_settings *);
 161
 162 static void rstp_run(struct ofproto_dpif *ofproto);
 163 static void set_rstp_port(struct ofport *,
 164                          const struct ofproto_port_rstp_settings *);
 165
 166 struct ofport_dpif {
 167     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 168     struct ofport up;
 169
 170     odp_port_t odp_port;
 171     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 172     struct ovs_list bundle_node;/* In struct ofbundle's "ports" list. */
 173     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 174     struct bfd *bfd;            /* BFD, if any. */
 175     struct lldp *lldp;          /* lldp, if any. */
 176     bool may_enable;            /* May be enabled in bonds. */
 177     bool is_tunnel;             /* This port is a tunnel. */
 178     bool is_layer3;             /* This is a layer 3 port. */
 179     long long int carrier_seq;  /* Carrier status changes. */
 180     struct ofport_dpif *peer;   /* Peer if patch port. */
 181
 182     /* Spanning tree. */
 183     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 184     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 185     long long int stp_state_entered;
 186
 187     /* Rapid Spanning Tree. */
 188     struct rstp_port *rstp_port; /* Rapid Spanning Tree Protocol, if any. */
 189     enum rstp_state rstp_state; /* Always RSTP_DISABLED if RSTP not in use. */
 190
 191     /* Queue to DSCP mapping. */
 192     struct ofproto_port_queue *qdscp;
 193     size_t n_qdscp;
 194
 195     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 196      *
 197      * This is deprecated.  It is only for compatibility with broken device
 198      * drivers in old versions of Linux that do not properly support VLANs when
 199      * VLAN devices are not used.  When broken device drivers are no longer in
 200      * widespread use, we will delete these interfaces. */
 201     ofp_port_t realdev_ofp_port;
 202     int vlandev_vid;
 203 };
 204
 205 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 206  *
 207  * This is deprecated.  It is only for compatibility with broken device drivers
 208  * in old versions of Linux that do not properly support VLANs when VLAN
 209  * devices are not used.  When broken device drivers are no longer in
 210  * widespread use, we will delete these interfaces. */
 211 struct vlan_splinter {
 212     struct hmap_node realdev_vid_node;
 213     struct hmap_node vlandev_node;
 214     ofp_port_t realdev_ofp_port;
 215     ofp_port_t vlandev_ofp_port;
 216     int vid;
 217 };
 218
 219 static void vsp_remove(struct ofport_dpif *);
 220 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 221
 222 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 223                                        ofp_port_t);
 224
 225 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 226                                        odp_port_t);
 227
 228 static struct ofport_dpif *
 229 ofport_dpif_cast(const struct ofport *ofport)
 230 {
 231     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 232 }
 233
 234 static void port_run(struct ofport_dpif *);
 235 static int set_bfd(struct ofport *, const struct smap *);
 236 static int set_cfm(struct ofport *, const struct cfm_settings *);
 237 static int set_lldp(struct ofport *ofport_, const struct smap *cfg);
 238 static void ofport_update_peer(struct ofport_dpif *);
 239
 240 /* Reasons that we might need to revalidate every datapath flow, and
 241  * corresponding coverage counters.
 242  *
 243  * A value of 0 means that there is no need to revalidate.
 244  *
 245  * It would be nice to have some cleaner way to integrate with coverage
 246  * counters, but with only a few reasons I guess this is good enough for
 247  * now. */
 248 enum revalidate_reason {
 249     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 250     REV_STP,                   /* Spanning tree protocol port status change. */
 251     REV_RSTP,                  /* RSTP port status change. */
 252     REV_BOND,                  /* Bonding changed. */
 253     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 254     REV_FLOW_TABLE,            /* Flow table changed. */
 255     REV_MAC_LEARNING,          /* Mac learning changed. */
 256     REV_MCAST_SNOOPING,        /* Multicast snooping changed. */
 257 };
 258 COVERAGE_DEFINE(rev_reconfigure);
 259 COVERAGE_DEFINE(rev_stp);
 260 COVERAGE_DEFINE(rev_rstp);
 261 COVERAGE_DEFINE(rev_bond);
 262 COVERAGE_DEFINE(rev_port_toggled);
 263 COVERAGE_DEFINE(rev_flow_table);
 264 COVERAGE_DEFINE(rev_mac_learning);
 265 COVERAGE_DEFINE(rev_mcast_snooping);
 266
 267 /* All datapaths of a given type share a single dpif backer instance. */
 268 struct dpif_backer {
 269     char *type;
 270     int refcount;
 271     struct dpif *dpif;
 272     struct udpif *udpif;
 273
 274     struct ovs_rwlock odp_to_ofport_lock;
 275     struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
 276
 277     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 278
 279     enum revalidate_reason need_revalidate; /* Revalidate all flows. */
 280
 281     bool recv_set_enable; /* Enables or disables receiving packets. */
 282
 283     /* Version string of the datapath stored in OVSDB. */
 284     char *dp_version_string;
 285
 286     /* Datapath feature support. */
 287     struct dpif_backer_support support;
 288     struct atomic_count tnl_count;
 289 };
 290
 291 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 292 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 293
 294 struct ofproto_dpif {
 295     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 296     struct ofproto up;
 297     struct dpif_backer *backer;
 298
 299     ATOMIC(cls_version_t) tables_version;  /* For classifier lookups. */
 300
 301     uint64_t dump_seq; /* Last read of udpif_dump_seq(). */
 302
 303     /* Special OpenFlow rules. */
 304     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 305     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 306     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 307
 308     /* Bridging. */
 309     struct netflow *netflow;
 310     struct dpif_sflow *sflow;
 311     struct dpif_ipfix *ipfix;
 312     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 313     struct mac_learning *ml;
 314     struct mcast_snooping *ms;
 315     bool has_bonded_bundles;
 316     bool lacp_enabled;
 317     struct mbridge *mbridge;
 318
 319     struct ovs_mutex stats_mutex;
 320     struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
 321                                             * consumed in userspace. */
 322
 323     /* Spanning tree. */
 324     struct stp *stp;
 325     long long int stp_last_tick;
 326
 327     /* Rapid Spanning Tree. */
 328     struct rstp *rstp;
 329     long long int rstp_last_tick;
 330
 331     /* VLAN splinters. */
 332     struct ovs_mutex vsp_mutex;
 333     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 334     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 335
 336     /* Ports. */
 337     struct sset ports;             /* Set of standard port names. */
 338     struct sset ghost_ports;       /* Ports with no datapath port. */
 339     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 340     int port_poll_errno;           /* Last errno for port_poll() reply. */
 341     uint64_t change_seq;           /* Connectivity status changes. */
 342
 343     /* Work queues. */
 344     struct guarded_list pins;      /* Contains "struct ofputil_packet_in"s. */
 345     struct seq *pins_seq;          /* For notifying 'pins' reception. */
 346     uint64_t pins_seqno;
 347 };
 348
 349 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 350 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 351
 352 static bool ofproto_use_tnl_push_pop = true;
 353 static void ofproto_unixctl_init(void);
 354
 355 static inline struct ofproto_dpif *
 356 ofproto_dpif_cast(const struct ofproto *ofproto)
 357 {
 358     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 359     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 360 }
 361
 362 bool
 363 ofproto_dpif_get_enable_ufid(const struct dpif_backer *backer)
 364 {
 365     return backer->support.ufid;
 366 }
 367
 368 struct dpif_backer_support *
 369 ofproto_dpif_get_support(const struct ofproto_dpif *ofproto)
 370 {
 371     return &ofproto->backer->support;
 372 }
 373
 374 static void ofproto_trace(struct ofproto_dpif *, struct flow *,
 375                           const struct dp_packet *packet,
 376                           const struct ofpact[], size_t ofpacts_len,
 377                           struct ds *);
 378
 379 /* Global variables. */
 380 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 381
 382 /* Initial mappings of port to bridge mappings. */
 383 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 384
 385 /* Executes 'fm'.  The caller retains ownership of 'fm' and everything in
 386  * it. */
 387 void
 388 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 389                       const struct ofputil_flow_mod *fm)
 390 {
 391     struct ofproto_flow_mod ofm;
 392
 393     /* Multiple threads may do this for the same 'fm' at the same time.
 394      * Allocate ofproto_flow_mod with execution context from stack.
 395      *
 396      * Note: This copy could be avoided by making ofproto_flow_mod more
 397      * complex, but that may not be desireable, and a learn action is not that
 398      * fast to begin with. */
 399     ofm.fm = *fm;
 400     ofproto_flow_mod(&ofproto->up, &ofm);
 401 }
 402
 403 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 404  * Takes ownership of 'pin' and pin->packet. */
 405 void
 406 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 407                             struct ofproto_packet_in *pin)
 408 {
 409     if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
 410         COVERAGE_INC(packet_in_overflow);
 411         free(CONST_CAST(void *, pin->up.packet));
 412         free(pin);
 413     }
 414
 415     /* Wakes up main thread for packet-in I/O. */
 416     seq_change(ofproto->pins_seq);
 417 }
 418
 419 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
 420  * packet rather than to send the packet to the controller.
 421  *
 422  * This function returns false to indicate that a packet_in message
 423  * for a "table-miss" should be sent to at least one controller.
 424  * False otherwise. */
 425 bool
 426 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif *ofproto)
 427 {
 428     return connmgr_wants_packet_in_on_miss(ofproto->up.connmgr);
 429 }
 430 \f
 431 /* Factory functions. */
 432
 433 static void
 434 init(const struct shash *iface_hints)
 435 {
 436     struct shash_node *node;
 437
 438     /* Make a local copy, since we don't own 'iface_hints' elements. */
 439     SHASH_FOR_EACH(node, iface_hints) {
 440         const struct iface_hint *orig_hint = node->data;
 441         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 442
 443         new_hint->br_name = xstrdup(orig_hint->br_name);
 444         new_hint->br_type = xstrdup(orig_hint->br_type);
 445         new_hint->ofp_port = orig_hint->ofp_port;
 446
 447         shash_add(&init_ofp_ports, node->name, new_hint);
 448     }
 449
 450     ofproto_unixctl_init();
 451     udpif_init();
 452 }
 453
 454 static void
 455 enumerate_types(struct sset *types)
 456 {
 457     dp_enumerate_types(types);
 458 }
 459
 460 static int
 461 enumerate_names(const char *type, struct sset *names)
 462 {
 463     struct ofproto_dpif *ofproto;
 464
 465     sset_clear(names);
 466     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 467         if (strcmp(type, ofproto->up.type)) {
 468             continue;
 469         }
 470         sset_add(names, ofproto->up.name);
 471     }
 472
 473     return 0;
 474 }
 475
 476 static int
 477 del(const char *type, const char *name)
 478 {
 479     struct dpif *dpif;
 480     int error;
 481
 482     error = dpif_open(name, type, &dpif);
 483     if (!error) {
 484         error = dpif_delete(dpif);
 485         dpif_close(dpif);
 486     }
 487     return error;
 488 }
 489 \f
 490 static const char *
 491 port_open_type(const char *datapath_type, const char *port_type)
 492 {
 493     return dpif_port_open_type(datapath_type, port_type);
 494 }
 495
 496 /* Type functions. */
 497
 498 static void process_dpif_port_changes(struct dpif_backer *);
 499 static void process_dpif_all_ports_changed(struct dpif_backer *);
 500 static void process_dpif_port_change(struct dpif_backer *,
 501                                      const char *devname);
 502 static void process_dpif_port_error(struct dpif_backer *, int error);
 503
 504 static struct ofproto_dpif *
 505 lookup_ofproto_dpif_by_port_name(const char *name)
 506 {
 507     struct ofproto_dpif *ofproto;
 508
 509     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 510         if (sset_contains(&ofproto->ports, name)) {
 511             return ofproto;
 512         }
 513     }
 514
 515     return NULL;
 516 }
 517
 518 static int
 519 type_run(const char *type)
 520 {
 521     struct dpif_backer *backer;
 522
 523     backer = shash_find_data(&all_dpif_backers, type);
 524     if (!backer) {
 525         /* This is not necessarily a problem, since backers are only
 526          * created on demand. */
 527         return 0;
 528     }
 529
 530
 531     if (dpif_run(backer->dpif)) {
 532         backer->need_revalidate = REV_RECONFIGURE;
 533     }
 534
 535     udpif_run(backer->udpif);
 536
 537     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 538      * and the configuration has now changed to "false", enable receiving
 539      * packets from the datapath. */
 540     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 541         int error;
 542
 543         backer->recv_set_enable = true;
 544
 545         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 546         if (error) {
 547             VLOG_ERR("Failed to enable receiving packets in dpif.");
 548             return error;
 549         }
 550         dpif_flow_flush(backer->dpif);
 551         backer->need_revalidate = REV_RECONFIGURE;
 552     }
 553
 554     if (backer->recv_set_enable) {
 555         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 556     }
 557
 558     dpif_poll_threads_set(backer->dpif, n_dpdk_rxqs, pmd_cpu_mask);
 559
 560     if (backer->need_revalidate) {
 561         struct ofproto_dpif *ofproto;
 562         struct simap_node *node;
 563         struct simap tmp_backers;
 564
 565         /* Handle tunnel garbage collection. */
 566         simap_init(&tmp_backers);
 567         simap_swap(&backer->tnl_backers, &tmp_backers);
 568
 569         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 570             struct ofport_dpif *iter;
 571
 572             if (backer != ofproto->backer) {
 573                 continue;
 574             }
 575
 576             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 577                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 578                 const char *dp_port;
 579
 580                 if (!iter->is_tunnel) {
 581                     continue;
 582                 }
 583
 584                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 585                                                      namebuf, sizeof namebuf);
 586                 node = simap_find(&tmp_backers, dp_port);
 587                 if (node) {
 588                     simap_put(&backer->tnl_backers, dp_port, node->data);
 589                     simap_delete(&tmp_backers, node);
 590                     node = simap_find(&backer->tnl_backers, dp_port);
 591                 } else {
 592                     node = simap_find(&backer->tnl_backers, dp_port);
 593                     if (!node) {
 594                         odp_port_t odp_port = ODPP_NONE;
 595
 596                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 597                                            &odp_port)) {
 598                             simap_put(&backer->tnl_backers, dp_port,
 599                                       odp_to_u32(odp_port));
 600                             node = simap_find(&backer->tnl_backers, dp_port);
 601                         }
 602                     }
 603                 }
 604
 605                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 606                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 607                                          iter->odp_port,
 608                                          ovs_native_tunneling_is_on(ofproto), dp_port)) {
 609                     backer->need_revalidate = REV_RECONFIGURE;
 610                 }
 611             }
 612         }
 613
 614         SIMAP_FOR_EACH (node, &tmp_backers) {
 615             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 616         }
 617         simap_destroy(&tmp_backers);
 618
 619         switch (backer->need_revalidate) {
 620         case REV_RECONFIGURE:    COVERAGE_INC(rev_reconfigure);    break;
 621         case REV_STP:            COVERAGE_INC(rev_stp);            break;
 622         case REV_RSTP:           COVERAGE_INC(rev_rstp);           break;
 623         case REV_BOND:           COVERAGE_INC(rev_bond);           break;
 624         case REV_PORT_TOGGLED:   COVERAGE_INC(rev_port_toggled);   break;
 625         case REV_FLOW_TABLE:     COVERAGE_INC(rev_flow_table);     break;
 626         case REV_MAC_LEARNING:   COVERAGE_INC(rev_mac_learning);   break;
 627         case REV_MCAST_SNOOPING: COVERAGE_INC(rev_mcast_snooping); break;
 628         }
 629         backer->need_revalidate = 0;
 630
 631         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 632             struct ofport_dpif *ofport;
 633             struct ofbundle *bundle;
 634
 635             if (ofproto->backer != backer) {
 636                 continue;
 637             }
 638
 639             xlate_txn_start();
 640             xlate_ofproto_set(ofproto, ofproto->up.name,
 641                               ofproto->backer->dpif, ofproto->ml,
 642                               ofproto->stp, ofproto->rstp, ofproto->ms,
 643                               ofproto->mbridge, ofproto->sflow, ofproto->ipfix,
 644                               ofproto->netflow,
 645                               ofproto->up.forward_bpdu,
 646                               connmgr_has_in_band(ofproto->up.connmgr),
 647                               &ofproto->backer->support);
 648
 649             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 650                 xlate_bundle_set(ofproto, bundle, bundle->name,
 651                                  bundle->vlan_mode, bundle->vlan,
 652                                  bundle->trunks, bundle->use_priority_tags,
 653                                  bundle->bond, bundle->lacp,
 654                                  bundle->floodable);
 655             }
 656
 657             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 658                 int stp_port = ofport->stp_port
 659                     ? stp_port_no(ofport->stp_port)
 660                     : -1;
 661                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 662                                  ofport->up.ofp_port, ofport->odp_port,
 663                                  ofport->up.netdev, ofport->cfm, ofport->bfd,
 664                                  ofport->lldp, ofport->peer, stp_port,
 665                                  ofport->rstp_port, ofport->qdscp,
 666                                  ofport->n_qdscp, ofport->up.pp.config,
 667                                  ofport->up.pp.state, ofport->is_tunnel,
 668                                  ofport->may_enable);
 669             }
 670             xlate_txn_commit();
 671         }
 672
 673         udpif_revalidate(backer->udpif);
 674     }
 675
 676     process_dpif_port_changes(backer);
 677
 678     return 0;
 679 }
 680
 681 /* Check for and handle port changes in 'backer''s dpif. */
 682 static void
 683 process_dpif_port_changes(struct dpif_backer *backer)
 684 {
 685     for (;;) {
 686         char *devname;
 687         int error;
 688
 689         error = dpif_port_poll(backer->dpif, &devname);
 690         switch (error) {
 691         case EAGAIN:
 692             return;
 693
 694         case ENOBUFS:
 695             process_dpif_all_ports_changed(backer);
 696             break;
 697
 698         case 0:
 699             process_dpif_port_change(backer, devname);
 700             free(devname);
 701             break;
 702
 703         default:
 704             process_dpif_port_error(backer, error);
 705             break;
 706         }
 707     }
 708 }
 709
 710 static void
 711 process_dpif_all_ports_changed(struct dpif_backer *backer)
 712 {
 713     struct ofproto_dpif *ofproto;
 714     struct dpif_port dpif_port;
 715     struct dpif_port_dump dump;
 716     struct sset devnames;
 717     const char *devname;
 718
 719     sset_init(&devnames);
 720     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 721         if (ofproto->backer == backer) {
 722             struct ofport *ofport;
 723
 724             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 725                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 726             }
 727         }
 728     }
 729     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 730         sset_add(&devnames, dpif_port.name);
 731     }
 732
 733     SSET_FOR_EACH (devname, &devnames) {
 734         process_dpif_port_change(backer, devname);
 735     }
 736     sset_destroy(&devnames);
 737 }
 738
 739 static void
 740 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 741 {
 742     struct ofproto_dpif *ofproto;
 743     struct dpif_port port;
 744
 745     /* Don't report on the datapath's device. */
 746     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 747         return;
 748     }
 749
 750     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 751                    &all_ofproto_dpifs) {
 752         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 753             return;
 754         }
 755     }
 756
 757     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 758     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 759         /* The port was removed.  If we know the datapath,
 760          * report it through poll_set().  If we don't, it may be
 761          * notifying us of a removal we initiated, so ignore it.
 762          * If there's a pending ENOBUFS, let it stand, since
 763          * everything will be reevaluated. */
 764         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 765             sset_add(&ofproto->port_poll_set, devname);
 766             ofproto->port_poll_errno = 0;
 767         }
 768     } else if (!ofproto) {
 769         /* The port was added, but we don't know with which
 770          * ofproto we should associate it.  Delete it. */
 771         dpif_port_del(backer->dpif, port.port_no);
 772     } else {
 773         struct ofport_dpif *ofport;
 774
 775         ofport = ofport_dpif_cast(shash_find_data(
 776                                       &ofproto->up.port_by_name, devname));
 777         if (ofport
 778             && ofport->odp_port != port.port_no
 779             && !odp_port_to_ofport(backer, port.port_no))
 780         {
 781             /* 'ofport''s datapath port number has changed from
 782              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 783              * structures to match. */
 784             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 785             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 786             ofport->odp_port = port.port_no;
 787             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 788                         hash_odp_port(port.port_no));
 789             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 790             backer->need_revalidate = REV_RECONFIGURE;
 791         }
 792     }
 793     dpif_port_destroy(&port);
 794 }
 795
 796 /* Propagate 'error' to all ofprotos based on 'backer'. */
 797 static void
 798 process_dpif_port_error(struct dpif_backer *backer, int error)
 799 {
 800     struct ofproto_dpif *ofproto;
 801
 802     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 803         if (ofproto->backer == backer) {
 804             sset_clear(&ofproto->port_poll_set);
 805             ofproto->port_poll_errno = error;
 806         }
 807     }
 808 }
 809
 810 static void
 811 type_wait(const char *type)
 812 {
 813     struct dpif_backer *backer;
 814
 815     backer = shash_find_data(&all_dpif_backers, type);
 816     if (!backer) {
 817         /* This is not necessarily a problem, since backers are only
 818          * created on demand. */
 819         return;
 820     }
 821
 822     dpif_wait(backer->dpif);
 823 }
 824 \f
 825 /* Basic life-cycle. */
 826
 827 static int add_internal_flows(struct ofproto_dpif *);
 828
 829 static struct ofproto *
 830 alloc(void)
 831 {
 832     struct ofproto_dpif *ofproto = xzalloc(sizeof *ofproto);
 833     return &ofproto->up;
 834 }
 835
 836 static void
 837 dealloc(struct ofproto *ofproto_)
 838 {
 839     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
 840     free(ofproto);
 841 }
 842
 843 static void
 844 close_dpif_backer(struct dpif_backer *backer)
 845 {
 846     ovs_assert(backer->refcount > 0);
 847
 848     if (--backer->refcount) {
 849         return;
 850     }
 851
 852     udpif_destroy(backer->udpif);
 853
 854     simap_destroy(&backer->tnl_backers);
 855     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
 856     hmap_destroy(&backer->odp_to_ofport_map);
 857     shash_find_and_delete(&all_dpif_backers, backer->type);
 858     free(backer->type);
 859     free(backer->dp_version_string);
 860     dpif_close(backer->dpif);
 861     free(backer);
 862 }
 863
 864 /* Datapath port slated for removal from datapath. */
 865 struct odp_garbage {
 866     struct ovs_list list_node;
 867     odp_port_t odp_port;
 868 };
 869
 870 static bool check_variable_length_userdata(struct dpif_backer *backer);
 871 static void check_support(struct dpif_backer *backer);
 872
 873 static int
 874 open_dpif_backer(const char *type, struct dpif_backer **backerp)
 875 {
 876     struct dpif_backer *backer;
 877     struct dpif_port_dump port_dump;
 878     struct dpif_port port;
 879     struct shash_node *node;
 880     struct ovs_list garbage_list;
 881     struct odp_garbage *garbage;
 882
 883     struct sset names;
 884     char *backer_name;
 885     const char *name;
 886     int error;
 887
 888     recirc_init();
 889
 890     backer = shash_find_data(&all_dpif_backers, type);
 891     if (backer) {
 892         backer->refcount++;
 893         *backerp = backer;
 894         return 0;
 895     }
 896
 897     backer_name = xasprintf("ovs-%s", type);
 898
 899     /* Remove any existing datapaths, since we assume we're the only
 900      * userspace controlling the datapath. */
 901     sset_init(&names);
 902     dp_enumerate_names(type, &names);
 903     SSET_FOR_EACH(name, &names) {
 904         struct dpif *old_dpif;
 905
 906         /* Don't remove our backer if it exists. */
 907         if (!strcmp(name, backer_name)) {
 908             continue;
 909         }
 910
 911         if (dpif_open(name, type, &old_dpif)) {
 912             VLOG_WARN("couldn't open old datapath %s to remove it", name);
 913         } else {
 914             dpif_delete(old_dpif);
 915             dpif_close(old_dpif);
 916         }
 917     }
 918     sset_destroy(&names);
 919
 920     backer = xmalloc(sizeof *backer);
 921
 922     error = dpif_create_and_open(backer_name, type, &backer->dpif);
 923     free(backer_name);
 924     if (error) {
 925         VLOG_ERR("failed to open datapath of type %s: %s", type,
 926                  ovs_strerror(error));
 927         free(backer);
 928         return error;
 929     }
 930     backer->udpif = udpif_create(backer, backer->dpif);
 931
 932     backer->type = xstrdup(type);
 933     backer->refcount = 1;
 934     hmap_init(&backer->odp_to_ofport_map);
 935     ovs_rwlock_init(&backer->odp_to_ofport_lock);
 936     backer->need_revalidate = 0;
 937     simap_init(&backer->tnl_backers);
 938     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
 939     *backerp = backer;
 940
 941     if (backer->recv_set_enable) {
 942         dpif_flow_flush(backer->dpif);
 943     }
 944
 945     /* Loop through the ports already on the datapath and remove any
 946      * that we don't need anymore. */
 947     list_init(&garbage_list);
 948     dpif_port_dump_start(&port_dump, backer->dpif);
 949     while (dpif_port_dump_next(&port_dump, &port)) {
 950         node = shash_find(&init_ofp_ports, port.name);
 951         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
 952             garbage = xmalloc(sizeof *garbage);
 953             garbage->odp_port = port.port_no;
 954             list_push_front(&garbage_list, &garbage->list_node);
 955         }
 956     }
 957     dpif_port_dump_done(&port_dump);
 958
 959     LIST_FOR_EACH_POP (garbage, list_node, &garbage_list) {
 960         dpif_port_del(backer->dpif, garbage->odp_port);
 961         free(garbage);
 962     }
 963
 964     shash_add(&all_dpif_backers, type, backer);
 965
 966     check_support(backer);
 967     atomic_count_init(&backer->tnl_count, 0);
 968
 969     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 970     if (error) {
 971         VLOG_ERR("failed to listen on datapath of type %s: %s",
 972                  type, ovs_strerror(error));
 973         close_dpif_backer(backer);
 974         return error;
 975     }
 976
 977     if (backer->recv_set_enable) {
 978         udpif_set_threads(backer->udpif, n_handlers, n_revalidators);
 979     }
 980
 981     /* This check fails if performed before udpif threads have been set,
 982      * as the kernel module checks that the 'pid' in userspace action
 983      * is non-zero. */
 984     backer->support.variable_length_userdata
 985         = check_variable_length_userdata(backer);
 986     backer->dp_version_string = dpif_get_dp_version(backer->dpif);
 987
 988     return error;
 989 }
 990
 991 bool
 992 ovs_native_tunneling_is_on(struct ofproto_dpif *ofproto)
 993 {
 994     return ofproto_use_tnl_push_pop && ofproto->backer->support.tnl_push_pop &&
 995            atomic_count_get(&ofproto->backer->tnl_count);
 996 }
 997
 998 /* Tests whether 'backer''s datapath supports recirculation.  Only newer
 999  * datapaths support OVS_KEY_ATTR_RECIRC_ID in keys.  We need to disable some
1000  * features on older datapaths that don't support this feature.
1001  *
1002  * Returns false if 'backer' definitely does not support recirculation, true if
1003  * it seems to support recirculation or if at least the error we get is
1004  * ambiguous. */
1005 static bool
1006 check_recirc(struct dpif_backer *backer)
1007 {
1008     struct flow flow;
1009     struct odputil_keybuf keybuf;
1010     struct ofpbuf key;
1011     bool enable_recirc;
1012     struct odp_flow_key_parms odp_parms = {
1013         .flow = &flow,
1014         .support = {
1015             .recirc = true,
1016         },
1017     };
1018
1019     memset(&flow, 0, sizeof flow);
1020     flow.recirc_id = 1;
1021     flow.dp_hash = 1;
1022
1023     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1024     odp_flow_key_from_flow(&odp_parms, &key);
1025     enable_recirc = dpif_probe_feature(backer->dpif, "recirculation", &key,
1026                                        NULL);
1027
1028     if (enable_recirc) {
1029         VLOG_INFO("%s: Datapath supports recirculation",
1030                   dpif_name(backer->dpif));
1031     } else {
1032         VLOG_INFO("%s: Datapath does not support recirculation",
1033                   dpif_name(backer->dpif));
1034     }
1035
1036     return enable_recirc;
1037 }
1038
1039 /* Tests whether 'dpif' supports unique flow ids. We can skip serializing
1040  * some flow attributes for datapaths that support this feature.
1041  *
1042  * Returns true if 'dpif' supports UFID for flow operations.
1043  * Returns false if  'dpif' does not support UFID. */
1044 static bool
1045 check_ufid(struct dpif_backer *backer)
1046 {
1047     struct flow flow;
1048     struct odputil_keybuf keybuf;
1049     struct ofpbuf key;
1050     ovs_u128 ufid;
1051     bool enable_ufid;
1052     struct odp_flow_key_parms odp_parms = {
1053         .flow = &flow,
1054     };
1055
1056     memset(&flow, 0, sizeof flow);
1057     flow.dl_type = htons(0x1234);
1058
1059     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1060     odp_flow_key_from_flow(&odp_parms, &key);
1061     dpif_flow_hash(backer->dpif, key.data, key.size, &ufid);
1062
1063     enable_ufid = dpif_probe_feature(backer->dpif, "UFID", &key, &ufid);
1064
1065     if (enable_ufid) {
1066         VLOG_INFO("%s: Datapath supports unique flow ids",
1067                   dpif_name(backer->dpif));
1068     } else {
1069         VLOG_INFO("%s: Datapath does not support unique flow ids",
1070                   dpif_name(backer->dpif));
1071     }
1072     return enable_ufid;
1073 }
1074
1075 /* Tests whether 'backer''s datapath supports variable-length
1076  * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.  We need
1077  * to disable some features on older datapaths that don't support this
1078  * feature.
1079  *
1080  * Returns false if 'backer' definitely does not support variable-length
1081  * userdata, true if it seems to support them or if at least the error we get
1082  * is ambiguous. */
1083 static bool
1084 check_variable_length_userdata(struct dpif_backer *backer)
1085 {
1086     struct eth_header *eth;
1087     struct ofpbuf actions;
1088     struct dpif_execute execute;
1089     struct dp_packet packet;
1090     size_t start;
1091     int error;
1092
1093     /* Compose a userspace action that will cause an ERANGE error on older
1094      * datapaths that don't support variable-length userdata.
1095      *
1096      * We really test for using userdata longer than 8 bytes, but older
1097      * datapaths accepted these, silently truncating the userdata to 8 bytes.
1098      * The same older datapaths rejected userdata shorter than 8 bytes, so we
1099      * test for that instead as a proxy for longer userdata support. */
1100     ofpbuf_init(&actions, 64);
1101     start = nl_msg_start_nested(&actions, OVS_ACTION_ATTR_USERSPACE);
1102     nl_msg_put_u32(&actions, OVS_USERSPACE_ATTR_PID,
1103                    dpif_port_get_pid(backer->dpif, ODPP_NONE, 0));
1104     nl_msg_put_unspec_zero(&actions, OVS_USERSPACE_ATTR_USERDATA, 4);
1105     nl_msg_end_nested(&actions, start);
1106
1107     /* Compose a dummy ethernet packet. */
1108     dp_packet_init(&packet, ETH_HEADER_LEN);
1109     eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1110     eth->eth_type = htons(0x1234);
1111
1112     /* Execute the actions.  On older datapaths this fails with ERANGE, on
1113      * newer datapaths it succeeds. */
1114     execute.actions = actions.data;
1115     execute.actions_len = actions.size;
1116     execute.packet = &packet;
1117     execute.needs_help = false;
1118     execute.probe = true;
1119
1120     error = dpif_execute(backer->dpif, &execute);
1121
1122     dp_packet_uninit(&packet);
1123     ofpbuf_uninit(&actions);
1124
1125     switch (error) {
1126     case 0:
1127         return true;
1128
1129     case ERANGE:
1130         /* Variable-length userdata is not supported. */
1131         VLOG_WARN("%s: datapath does not support variable-length userdata "
1132                   "feature (needs Linux 3.10+ or kernel module from OVS "
1133                   "1..11+).  The NXAST_SAMPLE action will be ignored.",
1134                   dpif_name(backer->dpif));
1135         return false;
1136
1137     default:
1138         /* Something odd happened.  We're not sure whether variable-length
1139          * userdata is supported.  Default to "yes". */
1140         VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1141                   dpif_name(backer->dpif), ovs_strerror(error));
1142         return true;
1143     }
1144 }
1145
1146 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1147  *
1148  * Returns the number of elements in a struct flow's mpls_lse field
1149  * if the datapath supports at least that many entries in an
1150  * MPLS label stack.
1151  * Otherwise returns the number of MPLS push actions supported by
1152  * the datapath. */
1153 static size_t
1154 check_max_mpls_depth(struct dpif_backer *backer)
1155 {
1156     struct flow flow;
1157     int n;
1158
1159     for (n = 0; n < FLOW_MAX_MPLS_LABELS; n++) {
1160         struct odputil_keybuf keybuf;
1161         struct ofpbuf key;
1162         struct odp_flow_key_parms odp_parms = {
1163             .flow = &flow,
1164         };
1165
1166         memset(&flow, 0, sizeof flow);
1167         flow.dl_type = htons(ETH_TYPE_MPLS);
1168         flow_set_mpls_bos(&flow, n, 1);
1169
1170         ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
1171         odp_flow_key_from_flow(&odp_parms, &key);
1172         if (!dpif_probe_feature(backer->dpif, "MPLS", &key, NULL)) {
1173             break;
1174         }
1175     }
1176
1177     VLOG_INFO("%s: MPLS label stack length probed as %d",
1178               dpif_name(backer->dpif), n);
1179     return n;
1180 }
1181
1182 /* Tests whether 'backer''s datapath supports masked data in
1183  * OVS_ACTION_ATTR_SET actions.  We need to disable some features on older
1184  * datapaths that don't support this feature. */
1185 static bool
1186 check_masked_set_action(struct dpif_backer *backer)
1187 {
1188     struct eth_header *eth;
1189     struct ofpbuf actions;
1190     struct dpif_execute execute;
1191     struct dp_packet packet;
1192     int error;
1193     struct ovs_key_ethernet key, mask;
1194
1195     /* Compose a set action that will cause an EINVAL error on older
1196      * datapaths that don't support masked set actions.
1197      * Avoid using a full mask, as it could be translated to a non-masked
1198      * set action instead. */
1199     ofpbuf_init(&actions, 64);
1200     memset(&key, 0x53, sizeof key);
1201     memset(&mask, 0x7f, sizeof mask);
1202     commit_masked_set_action(&actions, OVS_KEY_ATTR_ETHERNET, &key, &mask,
1203                              sizeof key);
1204
1205     /* Compose a dummy ethernet packet. */
1206     dp_packet_init(&packet, ETH_HEADER_LEN);
1207     eth = dp_packet_put_zeros(&packet, ETH_HEADER_LEN);
1208     eth->eth_type = htons(0x1234);
1209
1210     /* Execute the actions.  On older datapaths this fails with EINVAL, on
1211      * newer datapaths it succeeds. */
1212     execute.actions = actions.data;
1213     execute.actions_len = actions.size;
1214     execute.packet = &packet;
1215     execute.needs_help = false;
1216     execute.probe = true;
1217
1218     error = dpif_execute(backer->dpif, &execute);
1219
1220     dp_packet_uninit(&packet);
1221     ofpbuf_uninit(&actions);
1222
1223     if (error) {
1224         /* Masked set action is not supported. */
1225         VLOG_INFO("%s: datapath does not support masked set action feature.",
1226                   dpif_name(backer->dpif));
1227     }
1228     return !error;
1229 }
1230
1231 static void
1232 check_support(struct dpif_backer *backer)
1233 {
1234     /* This feature needs to be tested after udpif threads are set. */
1235     backer->support.variable_length_userdata = false;
1236
1237     backer->support.odp.recirc = check_recirc(backer);
1238     backer->support.odp.max_mpls_depth = check_max_mpls_depth(backer);
1239     backer->support.masked_set_action = check_masked_set_action(backer);
1240     backer->support.ufid = check_ufid(backer);
1241     backer->support.tnl_push_pop = dpif_supports_tnl_push_pop(backer->dpif);
1242 }
1243
1244 static int
1245 construct(struct ofproto *ofproto_)
1246 {
1247     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1248     struct shash_node *node, *next;
1249     int error;
1250
1251     /* Tunnel module can get used right after the udpif threads are running. */
1252     ofproto_tunnel_init();
1253
1254     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1255     if (error) {
1256         return error;
1257     }
1258
1259     atomic_init(&ofproto->tables_version, CLS_MIN_VERSION);
1260     ofproto->netflow = NULL;
1261     ofproto->sflow = NULL;
1262     ofproto->ipfix = NULL;
1263     ofproto->stp = NULL;
1264     ofproto->rstp = NULL;
1265     ofproto->dump_seq = 0;
1266     hmap_init(&ofproto->bundles);
1267     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1268     ofproto->ms = NULL;
1269     ofproto->mbridge = mbridge_create();
1270     ofproto->has_bonded_bundles = false;
1271     ofproto->lacp_enabled = false;
1272     ovs_mutex_init_adaptive(&ofproto->stats_mutex);
1273     ovs_mutex_init(&ofproto->vsp_mutex);
1274
1275     guarded_list_init(&ofproto->pins);
1276
1277     hmap_init(&ofproto->vlandev_map);
1278     hmap_init(&ofproto->realdev_vid_map);
1279
1280     sset_init(&ofproto->ports);
1281     sset_init(&ofproto->ghost_ports);
1282     sset_init(&ofproto->port_poll_set);
1283     ofproto->port_poll_errno = 0;
1284     ofproto->change_seq = 0;
1285     ofproto->pins_seq = seq_create();
1286     ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1287
1288
1289     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1290         struct iface_hint *iface_hint = node->data;
1291
1292         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1293             /* Check if the datapath already has this port. */
1294             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1295                 sset_add(&ofproto->ports, node->name);
1296             }
1297
1298             free(iface_hint->br_name);
1299             free(iface_hint->br_type);
1300             free(iface_hint);
1301             shash_delete(&init_ofp_ports, node);
1302         }
1303     }
1304
1305     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1306                 hash_string(ofproto->up.name, 0));
1307     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1308
1309     ofproto_init_tables(ofproto_, N_TABLES);
1310     error = add_internal_flows(ofproto);
1311
1312     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1313
1314     return error;
1315 }
1316
1317 static int
1318 add_internal_miss_flow(struct ofproto_dpif *ofproto, int id,
1319                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1320 {
1321     struct match match;
1322     int error;
1323     struct rule *rule;
1324
1325     match_init_catchall(&match);
1326     match_set_reg(&match, 0, id);
1327
1328     error = ofproto_dpif_add_internal_flow(ofproto, &match, 0, 0, ofpacts,
1329                                            &rule);
1330     *rulep = error ? NULL : rule_dpif_cast(rule);
1331
1332     return error;
1333 }
1334
1335 static int
1336 add_internal_flows(struct ofproto_dpif *ofproto)
1337 {
1338     struct ofpact_controller *controller;
1339     uint64_t ofpacts_stub[128 / 8];
1340     struct ofpbuf ofpacts;
1341     struct rule *unused_rulep OVS_UNUSED;
1342     struct match match;
1343     int error;
1344     int id;
1345
1346     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1347     id = 1;
1348
1349     controller = ofpact_put_CONTROLLER(&ofpacts);
1350     controller->max_len = UINT16_MAX;
1351     controller->controller_id = 0;
1352     controller->reason = OFPR_NO_MATCH;
1353     ofpact_pad(&ofpacts);
1354
1355     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1356                                    &ofproto->miss_rule);
1357     if (error) {
1358         return error;
1359     }
1360
1361     ofpbuf_clear(&ofpacts);
1362     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1363                                    &ofproto->no_packet_in_rule);
1364     if (error) {
1365         return error;
1366     }
1367
1368     error = add_internal_miss_flow(ofproto, id++, &ofpacts,
1369                                    &ofproto->drop_frags_rule);
1370     if (error) {
1371         return error;
1372     }
1373
1374     /* Drop any run away non-recirc rule lookups. Recirc_id has to be
1375      * zero when reaching this rule.
1376      *
1377      * (priority=2), recirc_id=0, actions=drop
1378      */
1379     ofpbuf_clear(&ofpacts);
1380     match_init_catchall(&match);
1381     match_set_recirc_id(&match, 0);
1382     error = ofproto_dpif_add_internal_flow(ofproto, &match, 2, 0, &ofpacts,
1383                                            &unused_rulep);
1384     return error;
1385 }
1386
1387 static void
1388 destruct(struct ofproto *ofproto_)
1389 {
1390     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1391     struct ofproto_packet_in *pin;
1392     struct rule_dpif *rule;
1393     struct oftable *table;
1394     struct ovs_list pins;
1395
1396     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1397     xlate_txn_start();
1398     xlate_remove_ofproto(ofproto);
1399     xlate_txn_commit();
1400
1401     /* Ensure that the upcall processing threads have no remaining references
1402      * to the ofproto or anything in it. */
1403     udpif_synchronize(ofproto->backer->udpif);
1404
1405     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1406
1407     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1408         CLS_FOR_EACH (rule, up.cr, &table->cls) {
1409             ofproto_rule_delete(&ofproto->up, &rule->up);
1410         }
1411     }
1412     ofproto_group_delete_all(&ofproto->up);
1413
1414     guarded_list_pop_all(&ofproto->pins, &pins);
1415     LIST_FOR_EACH_POP (pin, list_node, &pins) {
1416         free(CONST_CAST(void *, pin->up.packet));
1417         free(pin);
1418     }
1419     guarded_list_destroy(&ofproto->pins);
1420
1421     recirc_free_ofproto(ofproto, ofproto->up.name);
1422
1423     mbridge_unref(ofproto->mbridge);
1424
1425     netflow_unref(ofproto->netflow);
1426     dpif_sflow_unref(ofproto->sflow);
1427     dpif_ipfix_unref(ofproto->ipfix);
1428     hmap_destroy(&ofproto->bundles);
1429     mac_learning_unref(ofproto->ml);
1430     mcast_snooping_unref(ofproto->ms);
1431
1432     hmap_destroy(&ofproto->vlandev_map);
1433     hmap_destroy(&ofproto->realdev_vid_map);
1434
1435     sset_destroy(&ofproto->ports);
1436     sset_destroy(&ofproto->ghost_ports);
1437     sset_destroy(&ofproto->port_poll_set);
1438
1439     ovs_mutex_destroy(&ofproto->stats_mutex);
1440     ovs_mutex_destroy(&ofproto->vsp_mutex);
1441
1442     seq_destroy(ofproto->pins_seq);
1443
1444     close_dpif_backer(ofproto->backer);
1445 }
1446
1447 static int
1448 run(struct ofproto *ofproto_)
1449 {
1450     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1451     uint64_t new_seq, new_dump_seq;
1452
1453     if (mbridge_need_revalidate(ofproto->mbridge)) {
1454         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1455         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1456         mac_learning_flush(ofproto->ml);
1457         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1458         mcast_snooping_mdb_flush(ofproto->ms);
1459     }
1460
1461     /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
1462      * flow restore.  Even though nothing is processed during flow restore,
1463      * all queued 'pins' will be handled immediately when flow restore
1464      * completes. */
1465     ofproto->pins_seqno = seq_read(ofproto->pins_seq);
1466
1467     /* Do not perform any periodic activity required by 'ofproto' while
1468      * waiting for flow restore to complete. */
1469     if (!ofproto_get_flow_restore_wait()) {
1470         struct ofproto_packet_in *pin;
1471         struct ovs_list pins;
1472
1473         guarded_list_pop_all(&ofproto->pins, &pins);
1474         LIST_FOR_EACH_POP (pin, list_node, &pins) {
1475             connmgr_send_packet_in(ofproto->up.connmgr, pin);
1476             free(CONST_CAST(void *, pin->up.packet));
1477             free(pin);
1478         }
1479     }
1480
1481     if (ofproto->netflow) {
1482         netflow_run(ofproto->netflow);
1483     }
1484     if (ofproto->sflow) {
1485         dpif_sflow_run(ofproto->sflow);
1486     }
1487     if (ofproto->ipfix) {
1488         dpif_ipfix_run(ofproto->ipfix);
1489     }
1490
1491     new_seq = seq_read(connectivity_seq_get());
1492     if (ofproto->change_seq != new_seq) {
1493         struct ofport_dpif *ofport;
1494
1495         HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1496             port_run(ofport);
1497         }
1498
1499         ofproto->change_seq = new_seq;
1500     }
1501     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1502         struct ofbundle *bundle;
1503
1504         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1505             bundle_run(bundle);
1506         }
1507     }
1508
1509     stp_run(ofproto);
1510     rstp_run(ofproto);
1511     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1512     if (mac_learning_run(ofproto->ml)) {
1513         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1514     }
1515     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1516
1517     if (mcast_snooping_run(ofproto->ms)) {
1518         ofproto->backer->need_revalidate = REV_MCAST_SNOOPING;
1519     }
1520
1521     new_dump_seq = seq_read(udpif_dump_seq(ofproto->backer->udpif));
1522     if (ofproto->dump_seq != new_dump_seq) {
1523         struct rule *rule, *next_rule;
1524
1525         /* We know stats are relatively fresh, so now is a good time to do some
1526          * periodic work. */
1527         ofproto->dump_seq = new_dump_seq;
1528
1529         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1530          * has passed. */
1531         ovs_mutex_lock(&ofproto_mutex);
1532         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
1533                             &ofproto->up.expirable) {
1534             rule_expire(rule_dpif_cast(rule));
1535         }
1536         ovs_mutex_unlock(&ofproto_mutex);
1537
1538         /* All outstanding data in existing flows has been accounted, so it's a
1539          * good time to do bond rebalancing. */
1540         if (ofproto->has_bonded_bundles) {
1541             struct ofbundle *bundle;
1542
1543             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1544                 if (bundle->bond) {
1545                     bond_rebalance(bundle->bond);
1546                 }
1547             }
1548         }
1549     }
1550     return 0;
1551 }
1552
1553 static void
1554 wait(struct ofproto *ofproto_)
1555 {
1556     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1557
1558     if (ofproto_get_flow_restore_wait()) {
1559         return;
1560     }
1561
1562     if (ofproto->sflow) {
1563         dpif_sflow_wait(ofproto->sflow);
1564     }
1565     if (ofproto->ipfix) {
1566         dpif_ipfix_wait(ofproto->ipfix);
1567     }
1568     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1569         struct ofbundle *bundle;
1570
1571         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1572             bundle_wait(bundle);
1573         }
1574     }
1575     if (ofproto->netflow) {
1576         netflow_wait(ofproto->netflow);
1577     }
1578     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1579     mac_learning_wait(ofproto->ml);
1580     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1581     mcast_snooping_wait(ofproto->ms);
1582     stp_wait(ofproto);
1583     if (ofproto->backer->need_revalidate) {
1584         /* Shouldn't happen, but if it does just go around again. */
1585         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1586         poll_immediate_wake();
1587     }
1588
1589     seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq);
1590     seq_wait(ofproto->pins_seq, ofproto->pins_seqno);
1591 }
1592
1593 static void
1594 type_get_memory_usage(const char *type, struct simap *usage)
1595 {
1596     struct dpif_backer *backer;
1597
1598     backer = shash_find_data(&all_dpif_backers, type);
1599     if (backer) {
1600         udpif_get_memory_usage(backer->udpif, usage);
1601     }
1602 }
1603
1604 static void
1605 flush(struct ofproto *ofproto_)
1606 {
1607     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1608     struct dpif_backer *backer = ofproto->backer;
1609
1610     if (backer) {
1611         udpif_flush(backer->udpif);
1612     }
1613 }
1614
1615 static void
1616 query_tables(struct ofproto *ofproto,
1617              struct ofputil_table_features *features,
1618              struct ofputil_table_stats *stats)
1619 {
1620     strcpy(features->name, "classifier");
1621
1622     if (stats) {
1623         int i;
1624
1625         for (i = 0; i < ofproto->n_tables; i++) {
1626             unsigned long missed, matched;
1627
1628             atomic_read_relaxed(&ofproto->tables[i].n_matched, &matched);
1629             atomic_read_relaxed(&ofproto->tables[i].n_missed, &missed);
1630
1631             stats[i].matched_count = matched;
1632             stats[i].lookup_count = matched + missed;
1633         }
1634     }
1635 }
1636
1637 static void
1638 set_tables_version(struct ofproto *ofproto_, cls_version_t version)
1639 {
1640     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1641
1642     atomic_store_relaxed(&ofproto->tables_version, version);
1643 }
1644
1645
1646 static struct ofport *
1647 port_alloc(void)
1648 {
1649     struct ofport_dpif *port = xzalloc(sizeof *port);
1650     return &port->up;
1651 }
1652
1653 static void
1654 port_dealloc(struct ofport *port_)
1655 {
1656     struct ofport_dpif *port = ofport_dpif_cast(port_);
1657     free(port);
1658 }
1659
1660 static int
1661 port_construct(struct ofport *port_)
1662 {
1663     struct ofport_dpif *port = ofport_dpif_cast(port_);
1664     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1665     const struct netdev *netdev = port->up.netdev;
1666     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1667     const char *dp_port_name;
1668     struct dpif_port dpif_port;
1669     int error;
1670
1671     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1672     port->bundle = NULL;
1673     port->cfm = NULL;
1674     port->bfd = NULL;
1675     port->lldp = NULL;
1676     port->may_enable = false;
1677     port->stp_port = NULL;
1678     port->stp_state = STP_DISABLED;
1679     port->rstp_port = NULL;
1680     port->rstp_state = RSTP_DISABLED;
1681     port->is_tunnel = false;
1682     port->peer = NULL;
1683     port->qdscp = NULL;
1684     port->n_qdscp = 0;
1685     port->realdev_ofp_port = 0;
1686     port->vlandev_vid = 0;
1687     port->carrier_seq = netdev_get_carrier_resets(netdev);
1688     port->is_layer3 = netdev_vport_is_layer3(netdev);
1689
1690     if (netdev_vport_is_patch(netdev)) {
1691         /* By bailing out here, we don't submit the port to the sFlow module
1692          * to be considered for counter polling export.  This is correct
1693          * because the patch port represents an interface that sFlow considers
1694          * to be "internal" to the switch as a whole, and therefore not a
1695          * candidate for counter polling. */
1696         port->odp_port = ODPP_NONE;
1697         ofport_update_peer(port);
1698         return 0;
1699     }
1700
1701     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1702     error = dpif_port_query_by_name(ofproto->backer->dpif, dp_port_name,
1703                                     &dpif_port);
1704     if (error) {
1705         return error;
1706     }
1707
1708     port->odp_port = dpif_port.port_no;
1709
1710     if (netdev_get_tunnel_config(netdev)) {
1711         atomic_count_inc(&ofproto->backer->tnl_count);
1712         error = tnl_port_add(port, port->up.netdev, port->odp_port,
1713                              ovs_native_tunneling_is_on(ofproto), dp_port_name);
1714         if (error) {
1715             atomic_count_dec(&ofproto->backer->tnl_count);
1716             dpif_port_destroy(&dpif_port);
1717             return error;
1718         }
1719
1720         port->is_tunnel = true;
1721         if (ofproto->ipfix) {
1722            dpif_ipfix_add_tunnel_port(ofproto->ipfix, port_, port->odp_port);
1723         }
1724     } else {
1725         /* Sanity-check that a mapping doesn't already exist.  This
1726          * shouldn't happen for non-tunnel ports. */
1727         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1728             VLOG_ERR("port %s already has an OpenFlow port number",
1729                      dpif_port.name);
1730             dpif_port_destroy(&dpif_port);
1731             return EBUSY;
1732         }
1733
1734         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1735         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1736                     hash_odp_port(port->odp_port));
1737         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1738     }
1739     dpif_port_destroy(&dpif_port);
1740
1741     if (ofproto->sflow) {
1742         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1743     }
1744
1745     return 0;
1746 }
1747
1748 static void
1749 port_destruct(struct ofport *port_)
1750 {
1751     struct ofport_dpif *port = ofport_dpif_cast(port_);
1752     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1753     const char *devname = netdev_get_name(port->up.netdev);
1754     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1755     const char *dp_port_name;
1756
1757     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1758     xlate_txn_start();
1759     xlate_ofport_remove(port);
1760     xlate_txn_commit();
1761
1762     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1763                                               sizeof namebuf);
1764     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1765         /* The underlying device is still there, so delete it.  This
1766          * happens when the ofproto is being destroyed, since the caller
1767          * assumes that removal of attached ports will happen as part of
1768          * destruction. */
1769         if (!port->is_tunnel) {
1770             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1771         }
1772     }
1773
1774     if (port->peer) {
1775         port->peer->peer = NULL;
1776         port->peer = NULL;
1777     }
1778
1779     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1780         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1781         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1782         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1783     }
1784
1785     if (port->is_tunnel) {
1786         atomic_count_dec(&ofproto->backer->tnl_count);
1787     }
1788
1789     if (port->is_tunnel && ofproto->ipfix) {
1790        dpif_ipfix_del_tunnel_port(ofproto->ipfix, port->odp_port);
1791     }
1792
1793     tnl_port_del(port);
1794     sset_find_and_delete(&ofproto->ports, devname);
1795     sset_find_and_delete(&ofproto->ghost_ports, devname);
1796     bundle_remove(port_);
1797     set_cfm(port_, NULL);
1798     set_bfd(port_, NULL);
1799     set_lldp(port_, NULL);
1800     if (port->stp_port) {
1801         stp_port_disable(port->stp_port);
1802     }
1803     set_rstp_port(port_, NULL);
1804     if (ofproto->sflow) {
1805         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1806     }
1807
1808     free(port->qdscp);
1809 }
1810
1811 static void
1812 port_modified(struct ofport *port_)
1813 {
1814     struct ofport_dpif *port = ofport_dpif_cast(port_);
1815     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1816     const char *dp_port_name;
1817     struct netdev *netdev = port->up.netdev;
1818
1819     if (port->bundle && port->bundle->bond) {
1820         bond_slave_set_netdev(port->bundle->bond, port, netdev);
1821     }
1822
1823     if (port->cfm) {
1824         cfm_set_netdev(port->cfm, netdev);
1825     }
1826
1827     if (port->bfd) {
1828         bfd_set_netdev(port->bfd, netdev);
1829     }
1830
1831     ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1832                                      port->lldp, port->up.pp.hw_addr);
1833
1834     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
1835
1836     if (port->is_tunnel) {
1837         struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1838
1839         if (tnl_port_reconfigure(port, netdev, port->odp_port,
1840                                  ovs_native_tunneling_is_on(ofproto),
1841                                  dp_port_name)) {
1842             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1843         }
1844     }
1845
1846     ofport_update_peer(port);
1847 }
1848
1849 static void
1850 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1851 {
1852     struct ofport_dpif *port = ofport_dpif_cast(port_);
1853     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1854     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1855
1856     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1857                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1858                    OFPUTIL_PC_NO_PACKET_IN)) {
1859         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1860
1861         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1862             bundle_update(port->bundle);
1863         }
1864     }
1865 }
1866
1867 static int
1868 set_sflow(struct ofproto *ofproto_,
1869           const struct ofproto_sflow_options *sflow_options)
1870 {
1871     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1872     struct dpif_sflow *ds = ofproto->sflow;
1873
1874     if (sflow_options) {
1875         uint32_t old_probability = ds ? dpif_sflow_get_probability(ds) : 0;
1876         if (!ds) {
1877             struct ofport_dpif *ofport;
1878
1879             ds = ofproto->sflow = dpif_sflow_create();
1880             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1881                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1882             }
1883         }
1884         dpif_sflow_set_options(ds, sflow_options);
1885         if (dpif_sflow_get_probability(ds) != old_probability) {
1886             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1887         }
1888     } else {
1889         if (ds) {
1890             dpif_sflow_unref(ds);
1891             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1892             ofproto->sflow = NULL;
1893         }
1894     }
1895     return 0;
1896 }
1897
1898 static int
1899 set_ipfix(
1900     struct ofproto *ofproto_,
1901     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1902     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1903     size_t n_flow_exporters_options)
1904 {
1905     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1906     struct dpif_ipfix *di = ofproto->ipfix;
1907     bool has_options = bridge_exporter_options || flow_exporters_options;
1908     bool new_di = false;
1909
1910     if (has_options && !di) {
1911         di = ofproto->ipfix = dpif_ipfix_create();
1912         new_di = true;
1913     }
1914
1915     if (di) {
1916         /* Call set_options in any case to cleanly flush the flow
1917          * caches in the last exporters that are to be destroyed. */
1918         dpif_ipfix_set_options(
1919             di, bridge_exporter_options, flow_exporters_options,
1920             n_flow_exporters_options);
1921
1922         /* Add tunnel ports only when a new ipfix created */
1923         if (new_di == true) {
1924             struct ofport_dpif *ofport;
1925             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1926                 if (ofport->is_tunnel == true) {
1927                     dpif_ipfix_add_tunnel_port(di, &ofport->up, ofport->odp_port);
1928                 }
1929             }
1930         }
1931
1932         if (!has_options) {
1933             dpif_ipfix_unref(di);
1934             ofproto->ipfix = NULL;
1935         }
1936     }
1937
1938     return 0;
1939 }
1940
1941 static int
1942 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1943 {
1944     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1945     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1946     struct cfm *old = ofport->cfm;
1947     int error = 0;
1948
1949     if (s) {
1950         if (!ofport->cfm) {
1951             ofport->cfm = cfm_create(ofport->up.netdev);
1952         }
1953
1954         if (cfm_configure(ofport->cfm, s)) {
1955             error = 0;
1956             goto out;
1957         }
1958
1959         error = EINVAL;
1960     }
1961     cfm_unref(ofport->cfm);
1962     ofport->cfm = NULL;
1963 out:
1964     if (ofport->cfm != old) {
1965         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1966     }
1967     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1968                                      ofport->lldp, ofport->up.pp.hw_addr);
1969     return error;
1970 }
1971
1972 static bool
1973 cfm_status_changed(struct ofport *ofport_)
1974 {
1975     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1976
1977     return ofport->cfm ? cfm_check_status_change(ofport->cfm) : true;
1978 }
1979
1980 static int
1981 get_cfm_status(const struct ofport *ofport_,
1982                struct cfm_status *status)
1983 {
1984     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1985     int ret = 0;
1986
1987     if (ofport->cfm) {
1988         cfm_get_status(ofport->cfm, status);
1989     } else {
1990         ret = ENOENT;
1991     }
1992
1993     return ret;
1994 }
1995
1996 static int
1997 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1998 {
1999     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2000     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2001     struct bfd *old;
2002
2003     old = ofport->bfd;
2004     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2005                                 cfg, ofport->up.netdev);
2006     if (ofport->bfd != old) {
2007         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2008     }
2009     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
2010                                      ofport->lldp, ofport->up.pp.hw_addr);
2011     return 0;
2012 }
2013
2014 static bool
2015 bfd_status_changed(struct ofport *ofport_)
2016 {
2017     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2018
2019     return ofport->bfd ? bfd_check_status_change(ofport->bfd) : true;
2020 }
2021
2022 static int
2023 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2024 {
2025     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2026     int ret = 0;
2027
2028     if (ofport->bfd) {
2029         bfd_get_status(ofport->bfd, smap);
2030     } else {
2031         ret = ENOENT;
2032     }
2033
2034     return ret;
2035 }
2036
2037 static int
2038 set_lldp(struct ofport *ofport_,
2039          const struct smap *cfg)
2040 {
2041     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2042     int error = 0;
2043
2044     if (cfg) {
2045         if (!ofport->lldp) {
2046             struct ofproto_dpif *ofproto;
2047
2048             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2049             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2050             ofport->lldp = lldp_create(ofport->up.netdev, ofport_->mtu, cfg);
2051         }
2052
2053         if (!lldp_configure(ofport->lldp, cfg)) {
2054             error = EINVAL;
2055         }
2056     }
2057     if (error) {
2058         lldp_unref(ofport->lldp);
2059         ofport->lldp = NULL;
2060     }
2061
2062     ofproto_dpif_monitor_port_update(ofport,
2063                                      ofport->bfd,
2064                                      ofport->cfm,
2065                                      ofport->lldp,
2066                                      ofport->up.pp.hw_addr);
2067     return error;
2068 }
2069
2070 static bool
2071 get_lldp_status(const struct ofport *ofport_,
2072                struct lldp_status *status OVS_UNUSED)
2073 {
2074     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2075
2076     return ofport->lldp ? true : false;
2077 }
2078
2079 static int
2080 set_aa(struct ofproto *ofproto OVS_UNUSED,
2081        const struct aa_settings *s)
2082 {
2083     return aa_configure(s);
2084 }
2085
2086 static int
2087 aa_mapping_set(struct ofproto *ofproto_ OVS_UNUSED, void *aux,
2088                const struct aa_mapping_settings *s)
2089 {
2090     return aa_mapping_register(aux, s);
2091 }
2092
2093 static int
2094 aa_mapping_unset(struct ofproto *ofproto OVS_UNUSED, void *aux)
2095 {
2096     return aa_mapping_unregister(aux);
2097 }
2098
2099 static int
2100 aa_vlan_get_queued(struct ofproto *ofproto OVS_UNUSED, struct ovs_list *list)
2101 {
2102     return aa_get_vlan_queued(list);
2103 }
2104
2105 static unsigned int
2106 aa_vlan_get_queue_size(struct ofproto *ofproto OVS_UNUSED)
2107 {
2108     return aa_get_vlan_queue_size();
2109 }
2110
2111 \f
2112 /* Spanning Tree. */
2113
2114 /* Called while rstp_mutex is held. */
2115 static void
2116 rstp_send_bpdu_cb(struct dp_packet *pkt, void *ofport_, void *ofproto_)
2117 {
2118     struct ofproto_dpif *ofproto = ofproto_;
2119     struct ofport_dpif *ofport = ofport_;
2120     struct eth_header *eth = dp_packet_l2(pkt);
2121
2122     netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2123     if (eth_addr_is_zero(eth->eth_src)) {
2124         VLOG_WARN_RL(&rl, "%s port %d: cannot send RSTP BPDU on a port which "
2125                      "does not have a configured source MAC address.",
2126                      ofproto->up.name, ofp_to_u16(ofport->up.ofp_port));
2127     } else {
2128         ofproto_dpif_send_packet(ofport, pkt);
2129     }
2130     dp_packet_delete(pkt);
2131 }
2132
2133 static void
2134 send_bpdu_cb(struct dp_packet *pkt, int port_num, void *ofproto_)
2135 {
2136     struct ofproto_dpif *ofproto = ofproto_;
2137     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2138     struct ofport_dpif *ofport;
2139
2140     ofport = stp_port_get_aux(sp);
2141     if (!ofport) {
2142         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2143                      ofproto->up.name, port_num);
2144     } else {
2145         struct eth_header *eth = dp_packet_l2(pkt);
2146
2147         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2148         if (eth_addr_is_zero(eth->eth_src)) {
2149             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2150                          "with unknown MAC", ofproto->up.name, port_num);
2151         } else {
2152             ofproto_dpif_send_packet(ofport, pkt);
2153         }
2154     }
2155     dp_packet_delete(pkt);
2156 }
2157
2158 /* Configure RSTP on 'ofproto_' using the settings defined in 's'. */
2159 static void
2160 set_rstp(struct ofproto *ofproto_, const struct ofproto_rstp_settings *s)
2161 {
2162     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2163
2164     /* Only revalidate flows if the configuration changed. */
2165     if (!s != !ofproto->rstp) {
2166         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2167     }
2168
2169     if (s) {
2170         if (!ofproto->rstp) {
2171             ofproto->rstp = rstp_create(ofproto_->name, s->address,
2172                                         rstp_send_bpdu_cb, ofproto);
2173             ofproto->rstp_last_tick = time_msec();
2174         }
2175         rstp_set_bridge_address(ofproto->rstp, s->address);
2176         rstp_set_bridge_priority(ofproto->rstp, s->priority);
2177         rstp_set_bridge_ageing_time(ofproto->rstp, s->ageing_time);
2178         rstp_set_bridge_force_protocol_version(ofproto->rstp,
2179                                                s->force_protocol_version);
2180         rstp_set_bridge_max_age(ofproto->rstp, s->bridge_max_age);
2181         rstp_set_bridge_forward_delay(ofproto->rstp, s->bridge_forward_delay);
2182         rstp_set_bridge_transmit_hold_count(ofproto->rstp,
2183                                             s->transmit_hold_count);
2184     } else {
2185         struct ofport *ofport;
2186         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2187             set_rstp_port(ofport, NULL);
2188         }
2189         rstp_unref(ofproto->rstp);
2190         ofproto->rstp = NULL;
2191     }
2192 }
2193
2194 static void
2195 get_rstp_status(struct ofproto *ofproto_, struct ofproto_rstp_status *s)
2196 {
2197     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2198
2199     if (ofproto->rstp) {
2200         s->enabled = true;
2201         s->root_id = rstp_get_root_id(ofproto->rstp);
2202         s->bridge_id = rstp_get_bridge_id(ofproto->rstp);
2203         s->designated_id = rstp_get_designated_id(ofproto->rstp);
2204         s->root_path_cost = rstp_get_root_path_cost(ofproto->rstp);
2205         s->designated_port_id = rstp_get_designated_port_id(ofproto->rstp);
2206         s->bridge_port_id = rstp_get_bridge_port_id(ofproto->rstp);
2207     } else {
2208         s->enabled = false;
2209     }
2210 }
2211
2212 static void
2213 update_rstp_port_state(struct ofport_dpif *ofport)
2214 {
2215     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2216     enum rstp_state state;
2217
2218     /* Figure out new state. */
2219     state = ofport->rstp_port ? rstp_port_get_state(ofport->rstp_port)
2220         : RSTP_DISABLED;
2221
2222     /* Update state. */
2223     if (ofport->rstp_state != state) {
2224         enum ofputil_port_state of_state;
2225         bool fwd_change;
2226
2227         VLOG_DBG("port %s: RSTP state changed from %s to %s",
2228                  netdev_get_name(ofport->up.netdev),
2229                  rstp_state_name(ofport->rstp_state),
2230                  rstp_state_name(state));
2231
2232         if (rstp_learn_in_state(ofport->rstp_state)
2233             != rstp_learn_in_state(state)) {
2234             /* XXX: Learning action flows should also be flushed. */
2235             if (ofport->bundle) {
2236                 if (!rstp_shift_root_learned_address(ofproto->rstp)
2237                     || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2238                     bundle_flush_macs(ofport->bundle, false);
2239                 }
2240             }
2241         }
2242         fwd_change = rstp_forward_in_state(ofport->rstp_state)
2243             != rstp_forward_in_state(state);
2244
2245         ofproto->backer->need_revalidate = REV_RSTP;
2246         ofport->rstp_state = state;
2247
2248         if (fwd_change && ofport->bundle) {
2249             bundle_update(ofport->bundle);
2250         }
2251
2252         /* Update the RSTP state bits in the OpenFlow port description. */
2253         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2254         of_state |= (state == RSTP_LEARNING ? OFPUTIL_PS_STP_LEARN
2255                 : state == RSTP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2256                 : state == RSTP_DISCARDING ?  OFPUTIL_PS_STP_LISTEN
2257                 : 0);
2258         ofproto_port_set_state(&ofport->up, of_state);
2259     }
2260 }
2261
2262 static void
2263 rstp_run(struct ofproto_dpif *ofproto)
2264 {
2265     if (ofproto->rstp) {
2266         long long int now = time_msec();
2267         long long int elapsed = now - ofproto->rstp_last_tick;
2268         struct rstp_port *rp;
2269         struct ofport_dpif *ofport;
2270
2271         /* Every second, decrease the values of the timers. */
2272         if (elapsed >= 1000) {
2273             rstp_tick_timers(ofproto->rstp);
2274             ofproto->rstp_last_tick = now;
2275         }
2276         rp = NULL;
2277         while ((ofport = rstp_get_next_changed_port_aux(ofproto->rstp, &rp))) {
2278             update_rstp_port_state(ofport);
2279         }
2280         rp = NULL;
2281         ofport = NULL;
2282         /* FIXME: This check should be done on-event (i.e., when setting
2283          * p->fdb_flush) and not periodically.
2284          */
2285         while ((ofport = rstp_check_and_reset_fdb_flush(ofproto->rstp, &rp))) {
2286             if (!rstp_shift_root_learned_address(ofproto->rstp)
2287                 || rstp_get_old_root_aux(ofproto->rstp) != ofport) {
2288                 bundle_flush_macs(ofport->bundle, false);
2289             }
2290         }
2291
2292         if (rstp_shift_root_learned_address(ofproto->rstp)) {
2293             bundle_move(((struct ofport_dpif *)rstp_get_old_root_aux(ofproto->rstp))->bundle,
2294                         ((struct ofport_dpif *)rstp_get_new_root_aux(ofproto->rstp))->bundle);
2295             rstp_reset_root_changed(ofproto->rstp);
2296         }
2297     }
2298 }
2299
2300 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2301 static int
2302 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2303 {
2304     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2305
2306     /* Only revalidate flows if the configuration changed. */
2307     if (!s != !ofproto->stp) {
2308         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2309     }
2310
2311     if (s) {
2312         if (!ofproto->stp) {
2313             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2314                                       send_bpdu_cb, ofproto);
2315             ofproto->stp_last_tick = time_msec();
2316         }
2317
2318         stp_set_bridge_id(ofproto->stp, s->system_id);
2319         stp_set_bridge_priority(ofproto->stp, s->priority);
2320         stp_set_hello_time(ofproto->stp, s->hello_time);
2321         stp_set_max_age(ofproto->stp, s->max_age);
2322         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2323     }  else {
2324         struct ofport *ofport;
2325
2326         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2327             set_stp_port(ofport, NULL);
2328         }
2329
2330         stp_unref(ofproto->stp);
2331         ofproto->stp = NULL;
2332     }
2333
2334     return 0;
2335 }
2336
2337 static int
2338 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2339 {
2340     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2341
2342     if (ofproto->stp) {
2343         s->enabled = true;
2344         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2345         s->designated_root = stp_get_designated_root(ofproto->stp);
2346         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2347     } else {
2348         s->enabled = false;
2349     }
2350
2351     return 0;
2352 }
2353
2354 static void
2355 update_stp_port_state(struct ofport_dpif *ofport)
2356 {
2357     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2358     enum stp_state state;
2359
2360     /* Figure out new state. */
2361     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2362                              : STP_DISABLED;
2363
2364     /* Update state. */
2365     if (ofport->stp_state != state) {
2366         enum ofputil_port_state of_state;
2367         bool fwd_change;
2368
2369         VLOG_DBG("port %s: STP state changed from %s to %s",
2370                  netdev_get_name(ofport->up.netdev),
2371                  stp_state_name(ofport->stp_state),
2372                  stp_state_name(state));
2373         if (stp_learn_in_state(ofport->stp_state)
2374                 != stp_learn_in_state(state)) {
2375             /* xxx Learning action flows should also be flushed. */
2376             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2377             mac_learning_flush(ofproto->ml);
2378             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2379             mcast_snooping_mdb_flush(ofproto->ms);
2380         }
2381         fwd_change = stp_forward_in_state(ofport->stp_state)
2382                         != stp_forward_in_state(state);
2383
2384         ofproto->backer->need_revalidate = REV_STP;
2385         ofport->stp_state = state;
2386         ofport->stp_state_entered = time_msec();
2387
2388         if (fwd_change && ofport->bundle) {
2389             bundle_update(ofport->bundle);
2390         }
2391
2392         /* Update the STP state bits in the OpenFlow port description. */
2393         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2394         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2395                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2396                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2397                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2398                      : 0);
2399         ofproto_port_set_state(&ofport->up, of_state);
2400     }
2401 }
2402
2403 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2404  * caller is responsible for assigning STP port numbers and ensuring
2405  * there are no duplicates. */
2406 static int
2407 set_stp_port(struct ofport *ofport_,
2408              const struct ofproto_port_stp_settings *s)
2409 {
2410     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2411     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2412     struct stp_port *sp = ofport->stp_port;
2413
2414     if (!s || !s->enable) {
2415         if (sp) {
2416             ofport->stp_port = NULL;
2417             stp_port_disable(sp);
2418             update_stp_port_state(ofport);
2419         }
2420         return 0;
2421     } else if (sp && stp_port_no(sp) != s->port_num
2422                && ofport == stp_port_get_aux(sp)) {
2423         /* The port-id changed, so disable the old one if it's not
2424          * already in use by another port. */
2425         stp_port_disable(sp);
2426     }
2427
2428     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2429
2430     /* Set name before enabling the port so that debugging messages can print
2431      * the name. */
2432     stp_port_set_name(sp, netdev_get_name(ofport->up.netdev));
2433     stp_port_enable(sp);
2434
2435     stp_port_set_aux(sp, ofport);
2436     stp_port_set_priority(sp, s->priority);
2437     stp_port_set_path_cost(sp, s->path_cost);
2438
2439     update_stp_port_state(ofport);
2440
2441     return 0;
2442 }
2443
2444 static int
2445 get_stp_port_status(struct ofport *ofport_,
2446                     struct ofproto_port_stp_status *s)
2447 {
2448     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2449     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2450     struct stp_port *sp = ofport->stp_port;
2451
2452     if (!ofproto->stp || !sp) {
2453         s->enabled = false;
2454         return 0;
2455     }
2456
2457     s->enabled = true;
2458     s->port_id = stp_port_get_id(sp);
2459     s->state = stp_port_get_state(sp);
2460     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2461     s->role = stp_port_get_role(sp);
2462
2463     return 0;
2464 }
2465
2466 static int
2467 get_stp_port_stats(struct ofport *ofport_,
2468                    struct ofproto_port_stp_stats *s)
2469 {
2470     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2471     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2472     struct stp_port *sp = ofport->stp_port;
2473
2474     if (!ofproto->stp || !sp) {
2475         s->enabled = false;
2476         return 0;
2477     }
2478
2479     s->enabled = true;
2480     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2481
2482     return 0;
2483 }
2484
2485 static void
2486 stp_run(struct ofproto_dpif *ofproto)
2487 {
2488     if (ofproto->stp) {
2489         long long int now = time_msec();
2490         long long int elapsed = now - ofproto->stp_last_tick;
2491         struct stp_port *sp;
2492
2493         if (elapsed > 0) {
2494             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2495             ofproto->stp_last_tick = now;
2496         }
2497         while (stp_get_changed_port(ofproto->stp, &sp)) {
2498             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2499
2500             if (ofport) {
2501                 update_stp_port_state(ofport);
2502             }
2503         }
2504
2505         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2506             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2507             mac_learning_flush(ofproto->ml);
2508             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2509             mcast_snooping_mdb_flush(ofproto->ms);
2510         }
2511     }
2512 }
2513
2514 static void
2515 stp_wait(struct ofproto_dpif *ofproto)
2516 {
2517     if (ofproto->stp) {
2518         poll_timer_wait(1000);
2519     }
2520 }
2521
2522 /* Configures RSTP on 'ofport_' using the settings defined in 's'.  The
2523  * caller is responsible for assigning RSTP port numbers and ensuring
2524  * there are no duplicates. */
2525 static void
2526 set_rstp_port(struct ofport *ofport_,
2527               const struct ofproto_port_rstp_settings *s)
2528 {
2529     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2530     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2531     struct rstp_port *rp = ofport->rstp_port;
2532
2533     if (!s || !s->enable) {
2534         if (rp) {
2535             rstp_port_unref(rp);
2536             ofport->rstp_port = NULL;
2537             update_rstp_port_state(ofport);
2538         }
2539         return;
2540     }
2541
2542     /* Check if need to add a new port. */
2543     if (!rp) {
2544         rp = ofport->rstp_port = rstp_add_port(ofproto->rstp);
2545     }
2546
2547     rstp_port_set(rp, s->port_num, s->priority, s->path_cost,
2548                   s->admin_edge_port, s->auto_edge,
2549                   s->admin_p2p_mac_state, s->admin_port_state, s->mcheck,
2550                   ofport);
2551     update_rstp_port_state(ofport);
2552     /* Synchronize operational status. */
2553     rstp_port_set_mac_operational(rp, ofport->may_enable);
2554 }
2555
2556 static void
2557 get_rstp_port_status(struct ofport *ofport_,
2558         struct ofproto_port_rstp_status *s)
2559 {
2560     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2561     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2562     struct rstp_port *rp = ofport->rstp_port;
2563
2564     if (!ofproto->rstp || !rp) {
2565         s->enabled = false;
2566         return;
2567     }
2568
2569     s->enabled = true;
2570     rstp_port_get_status(rp, &s->port_id, &s->state, &s->role,
2571                          &s->designated_bridge_id, &s->designated_port_id,
2572                          &s->designated_path_cost, &s->tx_count,
2573                          &s->rx_count, &s->error_count, &s->uptime);
2574 }
2575
2576 \f
2577 static int
2578 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2579            size_t n_qdscp)
2580 {
2581     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2582     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2583
2584     if (ofport->n_qdscp != n_qdscp
2585         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2586                               n_qdscp * sizeof *qdscp))) {
2587         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2588         free(ofport->qdscp);
2589         ofport->qdscp = n_qdscp
2590             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2591             : NULL;
2592         ofport->n_qdscp = n_qdscp;
2593     }
2594
2595     return 0;
2596 }
2597 \f
2598 /* Bundles. */
2599
2600 /* Expires all MAC learning entries associated with 'bundle' and forces its
2601  * ofproto to revalidate every flow.
2602  *
2603  * Normally MAC learning entries are removed only from the ofproto associated
2604  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2605  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2606  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2607  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2608  * with the host from which it migrated. */
2609 static void
2610 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2611 {
2612     struct ofproto_dpif *ofproto = bundle->ofproto;
2613     struct mac_learning *ml = ofproto->ml;
2614     struct mac_entry *mac, *next_mac;
2615
2616     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2617     ovs_rwlock_wrlock(&ml->rwlock);
2618     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2619         if (mac_entry_get_port(ml, mac) == bundle) {
2620             if (all_ofprotos) {
2621                 struct ofproto_dpif *o;
2622
2623                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2624                     if (o != ofproto) {
2625                         struct mac_entry *e;
2626
2627                         ovs_rwlock_wrlock(&o->ml->rwlock);
2628                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2629                         if (e) {
2630                             mac_learning_expire(o->ml, e);
2631                         }
2632                         ovs_rwlock_unlock(&o->ml->rwlock);
2633                     }
2634                 }
2635             }
2636
2637             mac_learning_expire(ml, mac);
2638         }
2639     }
2640     ovs_rwlock_unlock(&ml->rwlock);
2641 }
2642
2643 static void
2644 bundle_move(struct ofbundle *old, struct ofbundle *new)
2645 {
2646     struct ofproto_dpif *ofproto = old->ofproto;
2647     struct mac_learning *ml = ofproto->ml;
2648     struct mac_entry *mac, *next_mac;
2649
2650     ovs_assert(new->ofproto == old->ofproto);
2651
2652     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2653     ovs_rwlock_wrlock(&ml->rwlock);
2654     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2655         if (mac_entry_get_port(ml, mac) == old) {
2656             mac_entry_set_port(ml, mac, new);
2657         }
2658     }
2659     ovs_rwlock_unlock(&ml->rwlock);
2660 }
2661
2662 static struct ofbundle *
2663 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2664 {
2665     struct ofbundle *bundle;
2666
2667     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2668                              &ofproto->bundles) {
2669         if (bundle->aux == aux) {
2670             return bundle;
2671         }
2672     }
2673     return NULL;
2674 }
2675
2676 static void
2677 bundle_update(struct ofbundle *bundle)
2678 {
2679     struct ofport_dpif *port;
2680
2681     bundle->floodable = true;
2682     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2683         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2684             || port->is_layer3
2685             || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2686             || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2687             bundle->floodable = false;
2688             break;
2689         }
2690     }
2691 }
2692
2693 static void
2694 bundle_del_port(struct ofport_dpif *port)
2695 {
2696     struct ofbundle *bundle = port->bundle;
2697
2698     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2699
2700     list_remove(&port->bundle_node);
2701     port->bundle = NULL;
2702
2703     if (bundle->lacp) {
2704         lacp_slave_unregister(bundle->lacp, port);
2705     }
2706     if (bundle->bond) {
2707         bond_slave_unregister(bundle->bond, port);
2708     }
2709
2710     bundle_update(bundle);
2711 }
2712
2713 static bool
2714 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2715                 struct lacp_slave_settings *lacp)
2716 {
2717     struct ofport_dpif *port;
2718
2719     port = ofp_port_to_ofport(bundle->ofproto, ofp_port);
2720     if (!port) {
2721         return false;
2722     }
2723
2724     if (port->bundle != bundle) {
2725         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2726         if (port->bundle) {
2727             bundle_remove(&port->up);
2728         }
2729
2730         port->bundle = bundle;
2731         list_push_back(&bundle->ports, &port->bundle_node);
2732         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2733             || port->is_layer3
2734             || (bundle->ofproto->stp && !stp_forward_in_state(port->stp_state))
2735             || (bundle->ofproto->rstp && !rstp_forward_in_state(port->rstp_state))) {
2736             bundle->floodable = false;
2737         }
2738     }
2739     if (lacp) {
2740         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2741         lacp_slave_register(bundle->lacp, port, lacp);
2742     }
2743
2744     return true;
2745 }
2746
2747 static void
2748 bundle_destroy(struct ofbundle *bundle)
2749 {
2750     struct ofproto_dpif *ofproto;
2751     struct ofport_dpif *port, *next_port;
2752
2753     if (!bundle) {
2754         return;
2755     }
2756
2757     ofproto = bundle->ofproto;
2758     mbridge_unregister_bundle(ofproto->mbridge, bundle);
2759
2760     xlate_txn_start();
2761     xlate_bundle_remove(bundle);
2762     xlate_txn_commit();
2763
2764     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2765         bundle_del_port(port);
2766     }
2767
2768     bundle_flush_macs(bundle, true);
2769     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2770     free(bundle->name);
2771     free(bundle->trunks);
2772     lacp_unref(bundle->lacp);
2773     bond_unref(bundle->bond);
2774     free(bundle);
2775 }
2776
2777 static int
2778 bundle_set(struct ofproto *ofproto_, void *aux,
2779            const struct ofproto_bundle_settings *s)
2780 {
2781     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2782     bool need_flush = false;
2783     struct ofport_dpif *port;
2784     struct ofbundle *bundle;
2785     unsigned long *trunks;
2786     int vlan;
2787     size_t i;
2788     bool ok;
2789
2790     if (!s) {
2791         bundle_destroy(bundle_lookup(ofproto, aux));
2792         return 0;
2793     }
2794
2795     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2796     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2797
2798     bundle = bundle_lookup(ofproto, aux);
2799     if (!bundle) {
2800         bundle = xmalloc(sizeof *bundle);
2801
2802         bundle->ofproto = ofproto;
2803         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2804                     hash_pointer(aux, 0));
2805         bundle->aux = aux;
2806         bundle->name = NULL;
2807
2808         list_init(&bundle->ports);
2809         bundle->vlan_mode = PORT_VLAN_TRUNK;
2810         bundle->vlan = -1;
2811         bundle->trunks = NULL;
2812         bundle->use_priority_tags = s->use_priority_tags;
2813         bundle->lacp = NULL;
2814         bundle->bond = NULL;
2815
2816         bundle->floodable = true;
2817         mbridge_register_bundle(ofproto->mbridge, bundle);
2818     }
2819
2820     if (!bundle->name || strcmp(s->name, bundle->name)) {
2821         free(bundle->name);
2822         bundle->name = xstrdup(s->name);
2823     }
2824
2825     /* LACP. */
2826     if (s->lacp) {
2827         ofproto->lacp_enabled = true;
2828         if (!bundle->lacp) {
2829             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2830             bundle->lacp = lacp_create();
2831         }
2832         lacp_configure(bundle->lacp, s->lacp);
2833     } else {
2834         lacp_unref(bundle->lacp);
2835         bundle->lacp = NULL;
2836     }
2837
2838     /* Update set of ports. */
2839     ok = true;
2840     for (i = 0; i < s->n_slaves; i++) {
2841         if (!bundle_add_port(bundle, s->slaves[i],
2842                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2843             ok = false;
2844         }
2845     }
2846     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2847         struct ofport_dpif *next_port;
2848
2849         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2850             for (i = 0; i < s->n_slaves; i++) {
2851                 if (s->slaves[i] == port->up.ofp_port) {
2852                     goto found;
2853                 }
2854             }
2855
2856             bundle_del_port(port);
2857         found: ;
2858         }
2859     }
2860     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2861
2862     if (list_is_empty(&bundle->ports)) {
2863         bundle_destroy(bundle);
2864         return EINVAL;
2865     }
2866
2867     /* Set VLAN tagging mode */
2868     if (s->vlan_mode != bundle->vlan_mode
2869         || s->use_priority_tags != bundle->use_priority_tags) {
2870         bundle->vlan_mode = s->vlan_mode;
2871         bundle->use_priority_tags = s->use_priority_tags;
2872         need_flush = true;
2873     }
2874
2875     /* Set VLAN tag. */
2876     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2877             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2878             : 0);
2879     if (vlan != bundle->vlan) {
2880         bundle->vlan = vlan;
2881         need_flush = true;
2882     }
2883
2884     /* Get trunked VLANs. */
2885     switch (s->vlan_mode) {
2886     case PORT_VLAN_ACCESS:
2887         trunks = NULL;
2888         break;
2889
2890     case PORT_VLAN_TRUNK:
2891         trunks = CONST_CAST(unsigned long *, s->trunks);
2892         break;
2893
2894     case PORT_VLAN_NATIVE_UNTAGGED:
2895     case PORT_VLAN_NATIVE_TAGGED:
2896         if (vlan != 0 && (!s->trunks
2897                           || !bitmap_is_set(s->trunks, vlan)
2898                           || bitmap_is_set(s->trunks, 0))) {
2899             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2900             if (s->trunks) {
2901                 trunks = bitmap_clone(s->trunks, 4096);
2902             } else {
2903                 trunks = bitmap_allocate1(4096);
2904             }
2905             bitmap_set1(trunks, vlan);
2906             bitmap_set0(trunks, 0);
2907         } else {
2908             trunks = CONST_CAST(unsigned long *, s->trunks);
2909         }
2910         break;
2911
2912     default:
2913         OVS_NOT_REACHED();
2914     }
2915     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2916         free(bundle->trunks);
2917         if (trunks == s->trunks) {
2918             bundle->trunks = vlan_bitmap_clone(trunks);
2919         } else {
2920             bundle->trunks = trunks;
2921             trunks = NULL;
2922         }
2923         need_flush = true;
2924     }
2925     if (trunks != s->trunks) {
2926         free(trunks);
2927     }
2928
2929     /* Bonding. */
2930     if (!list_is_short(&bundle->ports)) {
2931         bundle->ofproto->has_bonded_bundles = true;
2932         if (bundle->bond) {
2933             if (bond_reconfigure(bundle->bond, s->bond)) {
2934                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2935             }
2936         } else {
2937             bundle->bond = bond_create(s->bond, ofproto);
2938             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2939         }
2940
2941         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2942             bond_slave_register(bundle->bond, port,
2943                                 port->up.ofp_port, port->up.netdev);
2944         }
2945     } else {
2946         bond_unref(bundle->bond);
2947         bundle->bond = NULL;
2948     }
2949
2950     /* If we changed something that would affect MAC learning, un-learn
2951      * everything on this port and force flow revalidation. */
2952     if (need_flush) {
2953         bundle_flush_macs(bundle, false);
2954     }
2955
2956     return 0;
2957 }
2958
2959 static void
2960 bundle_remove(struct ofport *port_)
2961 {
2962     struct ofport_dpif *port = ofport_dpif_cast(port_);
2963     struct ofbundle *bundle = port->bundle;
2964
2965     if (bundle) {
2966         bundle_del_port(port);
2967         if (list_is_empty(&bundle->ports)) {
2968             bundle_destroy(bundle);
2969         } else if (list_is_short(&bundle->ports)) {
2970             bond_unref(bundle->bond);
2971             bundle->bond = NULL;
2972         }
2973     }
2974 }
2975
2976 static void
2977 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2978 {
2979     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2980     struct ofport_dpif *port = port_;
2981     uint8_t ea[ETH_ADDR_LEN];
2982     int error;
2983
2984     error = netdev_get_etheraddr(port->up.netdev, ea);
2985     if (!error) {
2986         struct dp_packet packet;
2987         void *packet_pdu;
2988
2989         dp_packet_init(&packet, 0);
2990         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2991                                  pdu_size);
2992         memcpy(packet_pdu, pdu, pdu_size);
2993
2994         ofproto_dpif_send_packet(port, &packet);
2995         dp_packet_uninit(&packet);
2996     } else {
2997         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2998                     "%s (%s)", port->bundle->name,
2999                     netdev_get_name(port->up.netdev), ovs_strerror(error));
3000     }
3001 }
3002
3003 static void
3004 bundle_send_learning_packets(struct ofbundle *bundle)
3005 {
3006     struct ofproto_dpif *ofproto = bundle->ofproto;
3007     int error, n_packets, n_errors;
3008     struct mac_entry *e;
3009     struct pkt_list {
3010         struct ovs_list list_node;
3011         struct ofport_dpif *port;
3012         struct dp_packet *pkt;
3013     } *pkt_node;
3014     struct ovs_list packets;
3015
3016     list_init(&packets);
3017     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
3018     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
3019         if (mac_entry_get_port(ofproto->ml, e) != bundle) {
3020             pkt_node = xmalloc(sizeof *pkt_node);
3021             pkt_node->pkt = bond_compose_learning_packet(bundle->bond,
3022                                                          e->mac, e->vlan,
3023                                                          (void **)&pkt_node->port);
3024             list_push_back(&packets, &pkt_node->list_node);
3025         }
3026     }
3027     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3028
3029     error = n_packets = n_errors = 0;
3030     LIST_FOR_EACH_POP (pkt_node, list_node, &packets) {
3031         int ret;
3032
3033         ret = ofproto_dpif_send_packet(pkt_node->port, pkt_node->pkt);
3034         dp_packet_delete(pkt_node->pkt);
3035         free(pkt_node);
3036         if (ret) {
3037             error = ret;
3038             n_errors++;
3039         }
3040         n_packets++;
3041     }
3042
3043     if (n_errors) {
3044         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3045         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
3046                      "packets, last error was: %s",
3047                      bundle->name, n_errors, n_packets, ovs_strerror(error));
3048     } else {
3049         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3050                  bundle->name, n_packets);
3051     }
3052 }
3053
3054 static void
3055 bundle_run(struct ofbundle *bundle)
3056 {
3057     if (bundle->lacp) {
3058         lacp_run(bundle->lacp, send_pdu_cb);
3059     }
3060     if (bundle->bond) {
3061         struct ofport_dpif *port;
3062
3063         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
3064             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
3065         }
3066
3067         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
3068             bundle->ofproto->backer->need_revalidate = REV_BOND;
3069         }
3070
3071         if (bond_should_send_learning_packets(bundle->bond)) {
3072             bundle_send_learning_packets(bundle);
3073         }
3074     }
3075 }
3076
3077 static void
3078 bundle_wait(struct ofbundle *bundle)
3079 {
3080     if (bundle->lacp) {
3081         lacp_wait(bundle->lacp);
3082     }
3083     if (bundle->bond) {
3084         bond_wait(bundle->bond);
3085     }
3086 }
3087 \f
3088 /* Mirrors. */
3089
3090 static int
3091 mirror_set__(struct ofproto *ofproto_, void *aux,
3092              const struct ofproto_mirror_settings *s)
3093 {
3094     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3095     struct ofbundle **srcs, **dsts;
3096     int error;
3097     size_t i;
3098
3099     if (!s) {
3100         mirror_destroy(ofproto->mbridge, aux);
3101         return 0;
3102     }
3103
3104     srcs = xmalloc(s->n_srcs * sizeof *srcs);
3105     dsts = xmalloc(s->n_dsts * sizeof *dsts);
3106
3107     for (i = 0; i < s->n_srcs; i++) {
3108         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
3109     }
3110
3111     for (i = 0; i < s->n_dsts; i++) {
3112         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
3113     }
3114
3115     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
3116                        s->n_dsts, s->src_vlans,
3117                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
3118     free(srcs);
3119     free(dsts);
3120     return error;
3121 }
3122
3123 static int
3124 mirror_get_stats__(struct ofproto *ofproto, void *aux,
3125                    uint64_t *packets, uint64_t *bytes)
3126 {
3127     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
3128                             bytes);
3129 }
3130
3131 static int
3132 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3133 {
3134     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3135     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3136     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3137         mac_learning_flush(ofproto->ml);
3138     }
3139     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3140     return 0;
3141 }
3142
3143 static bool
3144 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3145 {
3146     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3147     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3148     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
3149 }
3150
3151 static void
3152 forward_bpdu_changed(struct ofproto *ofproto_)
3153 {
3154     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3155     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3156 }
3157
3158 static void
3159 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3160                      size_t max_entries)
3161 {
3162     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3163     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
3164     mac_learning_set_idle_time(ofproto->ml, idle_time);
3165     mac_learning_set_max_entries(ofproto->ml, max_entries);
3166     ovs_rwlock_unlock(&ofproto->ml->rwlock);
3167 }
3168
3169 /* Configures multicast snooping on 'ofport' using the settings
3170  * defined in 's'. */
3171 static int
3172 set_mcast_snooping(struct ofproto *ofproto_,
3173                    const struct ofproto_mcast_snooping_settings *s)
3174 {
3175     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3176
3177     /* Only revalidate flows if the configuration changed. */
3178     if (!s != !ofproto->ms) {
3179         ofproto->backer->need_revalidate = REV_RECONFIGURE;
3180     }
3181
3182     if (s) {
3183         if (!ofproto->ms) {
3184             ofproto->ms = mcast_snooping_create();
3185         }
3186
3187         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3188         mcast_snooping_set_idle_time(ofproto->ms, s->idle_time);
3189         mcast_snooping_set_max_entries(ofproto->ms, s->max_entries);
3190         if (mcast_snooping_set_flood_unreg(ofproto->ms, s->flood_unreg)) {
3191             ofproto->backer->need_revalidate = REV_RECONFIGURE;
3192         }
3193         ovs_rwlock_unlock(&ofproto->ms->rwlock);
3194     } else {
3195         mcast_snooping_unref(ofproto->ms);
3196         ofproto->ms = NULL;
3197     }
3198
3199     return 0;
3200 }
3201
3202 /* Configures multicast snooping port's flood settings on 'ofproto'. */
3203 static int
3204 set_mcast_snooping_port(struct ofproto *ofproto_, void *aux,
3205                         const struct ofproto_mcast_snooping_port_settings *s)
3206 {
3207     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3208     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3209
3210     if (ofproto->ms && s) {
3211         ovs_rwlock_wrlock(&ofproto->ms->rwlock);
3212         mcast_snooping_set_port_flood(ofproto->ms, bundle, s->flood);
3213         mcast_snooping_set_port_flood_reports(ofproto->ms, bundle,
3214                                               s->flood_reports);
3215         ovs_rwlock_unlock(&ofproto->ms->rwlock);
3216     }
3217     return 0;
3218 }
3219
3220 \f
3221 /* Ports. */
3222
3223 struct ofport_dpif *
3224 ofp_port_to_ofport(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
3225 {
3226     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3227     return ofport ? ofport_dpif_cast(ofport) : NULL;
3228 }
3229
3230 static void
3231 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3232                             struct ofproto_port *ofproto_port,
3233                             struct dpif_port *dpif_port)
3234 {
3235     ofproto_port->name = dpif_port->name;
3236     ofproto_port->type = dpif_port->type;
3237     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3238 }
3239
3240 static void
3241 ofport_update_peer(struct ofport_dpif *ofport)
3242 {
3243     const struct ofproto_dpif *ofproto;
3244     struct dpif_backer *backer;
3245     char *peer_name;
3246
3247     if (!netdev_vport_is_patch(ofport->up.netdev)) {
3248         return;
3249     }
3250
3251     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
3252     backer->need_revalidate = REV_RECONFIGURE;
3253
3254     if (ofport->peer) {
3255         ofport->peer->peer = NULL;
3256         ofport->peer = NULL;
3257     }
3258
3259     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
3260     if (!peer_name) {
3261         return;
3262     }
3263
3264     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3265         struct ofport *peer_ofport;
3266         struct ofport_dpif *peer;
3267         char *peer_peer;
3268
3269         if (ofproto->backer != backer) {
3270             continue;
3271         }
3272
3273         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
3274         if (!peer_ofport) {
3275             continue;
3276         }
3277
3278         peer = ofport_dpif_cast(peer_ofport);
3279         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
3280         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
3281                                  peer_peer)) {
3282             ofport->peer = peer;
3283             ofport->peer->peer = ofport;
3284         }
3285         free(peer_peer);
3286
3287         break;
3288     }
3289     free(peer_name);
3290 }
3291
3292 static void
3293 port_run(struct ofport_dpif *ofport)
3294 {
3295     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3296     bool carrier_changed = carrier_seq != ofport->carrier_seq;
3297     bool enable = netdev_get_carrier(ofport->up.netdev);
3298     bool cfm_enable = false;
3299     bool bfd_enable = false;
3300
3301     ofport->carrier_seq = carrier_seq;
3302
3303     if (ofport->cfm) {
3304         int cfm_opup = cfm_get_opup(ofport->cfm);
3305
3306         cfm_enable = !cfm_get_fault(ofport->cfm);
3307
3308         if (cfm_opup >= 0) {
3309             cfm_enable = cfm_enable && cfm_opup;
3310         }
3311     }
3312
3313     if (ofport->bfd) {
3314         bfd_enable = bfd_forwarding(ofport->bfd);
3315     }
3316
3317     if (ofport->bfd || ofport->cfm) {
3318         enable = enable && (cfm_enable || bfd_enable);
3319     }
3320
3321     if (ofport->bundle) {
3322         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3323         if (carrier_changed) {
3324             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3325         }
3326     }
3327
3328     if (ofport->may_enable != enable) {
3329         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3330
3331         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3332
3333         if (ofport->rstp_port) {
3334             rstp_port_set_mac_operational(ofport->rstp_port, enable);
3335         }
3336     }
3337
3338     ofport->may_enable = enable;
3339 }
3340
3341 static int
3342 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3343                    struct ofproto_port *ofproto_port)
3344 {
3345     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3346     struct dpif_port dpif_port;
3347     int error;
3348
3349     if (sset_contains(&ofproto->ghost_ports, devname)) {
3350         const char *type = netdev_get_type_from_name(devname);
3351
3352         /* We may be called before ofproto->up.port_by_name is populated with
3353          * the appropriate ofport.  For this reason, we must get the name and
3354          * type from the netdev layer directly. */
3355         if (type) {
3356             const struct ofport *ofport;
3357
3358             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3359             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3360             ofproto_port->name = xstrdup(devname);
3361             ofproto_port->type = xstrdup(type);
3362             return 0;
3363         }
3364         return ENODEV;
3365     }
3366
3367     if (!sset_contains(&ofproto->ports, devname)) {
3368         return ENODEV;
3369     }
3370     error = dpif_port_query_by_name(ofproto->backer->dpif,
3371                                     devname, &dpif_port);
3372     if (!error) {
3373         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3374     }
3375     return error;
3376 }
3377
3378 static int
3379 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3380 {
3381     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3382     const char *devname = netdev_get_name(netdev);
3383     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3384     const char *dp_port_name;
3385
3386     if (netdev_vport_is_patch(netdev)) {
3387         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3388         return 0;
3389     }
3390
3391     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3392     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3393         odp_port_t port_no = ODPP_NONE;
3394         int error;
3395
3396         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3397         if (error) {
3398             return error;
3399         }
3400         if (netdev_get_tunnel_config(netdev)) {
3401             simap_put(&ofproto->backer->tnl_backers,
3402                       dp_port_name, odp_to_u32(port_no));
3403         }
3404     }
3405
3406     if (netdev_get_tunnel_config(netdev)) {
3407         sset_add(&ofproto->ghost_ports, devname);
3408     } else {
3409         sset_add(&ofproto->ports, devname);
3410     }
3411     return 0;
3412 }
3413
3414 static int
3415 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3416 {
3417     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3418     struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
3419     int error = 0;
3420
3421     if (!ofport) {
3422         return 0;
3423     }
3424
3425     sset_find_and_delete(&ofproto->ghost_ports,
3426                          netdev_get_name(ofport->up.netdev));
3427     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3428     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3429         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3430         if (!error) {
3431             /* The caller is going to close ofport->up.netdev.  If this is a
3432              * bonded port, then the bond is using that netdev, so remove it
3433              * from the bond.  The client will need to reconfigure everything
3434              * after deleting ports, so then the slave will get re-added. */
3435             bundle_remove(&ofport->up);
3436         }
3437     }
3438     return error;
3439 }
3440
3441 static int
3442 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3443 {
3444     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3445     int error;
3446
3447     error = netdev_get_stats(ofport->up.netdev, stats);
3448
3449     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3450         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3451
3452         ovs_mutex_lock(&ofproto->stats_mutex);
3453         /* ofproto->stats.tx_packets represents packets that we created
3454          * internally and sent to some port (e.g. packets sent with
3455          * ofproto_dpif_send_packet()).  Account for them as if they had
3456          * come from OFPP_LOCAL and got forwarded. */
3457
3458         if (stats->rx_packets != UINT64_MAX) {
3459             stats->rx_packets += ofproto->stats.tx_packets;
3460         }
3461
3462         if (stats->rx_bytes != UINT64_MAX) {
3463             stats->rx_bytes += ofproto->stats.tx_bytes;
3464         }
3465
3466         /* ofproto->stats.rx_packets represents packets that were received on
3467          * some port and we processed internally and dropped (e.g. STP).
3468          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3469
3470         if (stats->tx_packets != UINT64_MAX) {
3471             stats->tx_packets += ofproto->stats.rx_packets;
3472         }
3473
3474         if (stats->tx_bytes != UINT64_MAX) {
3475             stats->tx_bytes += ofproto->stats.rx_bytes;
3476         }
3477         ovs_mutex_unlock(&ofproto->stats_mutex);
3478     }
3479
3480     return error;
3481 }
3482
3483 static int
3484 port_get_lacp_stats(const struct ofport *ofport_, struct lacp_slave_stats *stats)
3485 {
3486     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3487     if (ofport->bundle && ofport->bundle->lacp) {
3488         if (lacp_get_slave_stats(ofport->bundle->lacp, ofport, stats)) {
3489             return 0;
3490         }
3491     }
3492     return -1;
3493 }
3494
3495 struct port_dump_state {
3496     uint32_t bucket;
3497     uint32_t offset;
3498     bool ghost;
3499
3500     struct ofproto_port port;
3501     bool has_port;
3502 };
3503
3504 static int
3505 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3506 {
3507     *statep = xzalloc(sizeof(struct port_dump_state));
3508     return 0;
3509 }
3510
3511 static int
3512 port_dump_next(const struct ofproto *ofproto_, void *state_,
3513                struct ofproto_port *port)
3514 {
3515     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3516     struct port_dump_state *state = state_;
3517     const struct sset *sset;
3518     struct sset_node *node;
3519
3520     if (state->has_port) {
3521         ofproto_port_destroy(&state->port);
3522         state->has_port = false;
3523     }
3524     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3525     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3526         int error;
3527
3528         error = port_query_by_name(ofproto_, node->name, &state->port);
3529         if (!error) {
3530             *port = state->port;
3531             state->has_port = true;
3532             return 0;
3533         } else if (error != ENODEV) {
3534             return error;
3535         }
3536     }
3537
3538     if (!state->ghost) {
3539         state->ghost = true;
3540         state->bucket = 0;
3541         state->offset = 0;
3542         return port_dump_next(ofproto_, state_, port);
3543     }
3544
3545     return EOF;
3546 }
3547
3548 static int
3549 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3550 {
3551     struct port_dump_state *state = state_;
3552
3553     if (state->has_port) {
3554         ofproto_port_destroy(&state->port);
3555     }
3556     free(state);
3557     return 0;
3558 }
3559
3560 static int
3561 port_poll(const struct ofproto *ofproto_, char **devnamep)
3562 {
3563     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3564
3565     if (ofproto->port_poll_errno) {
3566         int error = ofproto->port_poll_errno;
3567         ofproto->port_poll_errno = 0;
3568         return error;
3569     }
3570
3571     if (sset_is_empty(&ofproto->port_poll_set)) {
3572         return EAGAIN;
3573     }
3574
3575     *devnamep = sset_pop(&ofproto->port_poll_set);
3576     return 0;
3577 }
3578
3579 static void
3580 port_poll_wait(const struct ofproto *ofproto_)
3581 {
3582     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3583     dpif_port_poll_wait(ofproto->backer->dpif);
3584 }
3585
3586 static int
3587 port_is_lacp_current(const struct ofport *ofport_)
3588 {
3589     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3590     return (ofport->bundle && ofport->bundle->lacp
3591             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3592             : -1);
3593 }
3594 \f
3595 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3596  * then delete it entirely. */
3597 static void
3598 rule_expire(struct rule_dpif *rule)
3599     OVS_REQUIRES(ofproto_mutex)
3600 {
3601     uint16_t hard_timeout, idle_timeout;
3602     long long int now = time_msec();
3603     int reason = -1;
3604
3605     hard_timeout = rule->up.hard_timeout;
3606     idle_timeout = rule->up.idle_timeout;
3607
3608     /* Has 'rule' expired? */
3609     if (hard_timeout) {
3610         long long int modified;
3611
3612         ovs_mutex_lock(&rule->up.mutex);
3613         modified = rule->up.modified;
3614         ovs_mutex_unlock(&rule->up.mutex);
3615
3616         if (now > modified + hard_timeout * 1000) {
3617             reason = OFPRR_HARD_TIMEOUT;
3618         }
3619     }
3620
3621     if (reason < 0 && idle_timeout) {
3622         long long int used;
3623
3624         ovs_mutex_lock(&rule->stats_mutex);
3625         used = rule->stats.used;
3626         ovs_mutex_unlock(&rule->stats_mutex);
3627
3628         if (now > used + idle_timeout * 1000) {
3629             reason = OFPRR_IDLE_TIMEOUT;
3630         }
3631     }
3632
3633     if (reason >= 0) {
3634         COVERAGE_INC(ofproto_dpif_expired);
3635         ofproto_rule_expire(&rule->up, reason);
3636     }
3637 }
3638
3639 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3640  * 'flow' must reflect the data in 'packet'. */
3641 int
3642 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3643                              const struct flow *flow,
3644                              struct rule_dpif *rule,
3645                              const struct ofpact *ofpacts, size_t ofpacts_len,
3646                              struct dp_packet *packet)
3647 {
3648     struct dpif_flow_stats stats;
3649     struct xlate_out xout;
3650     struct xlate_in xin;
3651     ofp_port_t in_port;
3652     struct dpif_execute execute;
3653     int error;
3654
3655     ovs_assert((rule != NULL) != (ofpacts != NULL));
3656
3657     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3658
3659     if (rule) {
3660         rule_dpif_credit_stats(rule, &stats);
3661     }
3662
3663     uint64_t odp_actions_stub[1024 / 8];
3664     struct ofpbuf odp_actions = OFPBUF_STUB_INITIALIZER(odp_actions_stub);
3665     xlate_in_init(&xin, ofproto, flow, flow->in_port.ofp_port, rule,
3666                   stats.tcp_flags, packet, NULL, &odp_actions);
3667     xin.ofpacts = ofpacts;
3668     xin.ofpacts_len = ofpacts_len;
3669     xin.resubmit_stats = &stats;
3670     xlate_actions(&xin, &xout);
3671
3672     execute.actions = odp_actions.data;
3673     execute.actions_len = odp_actions.size;
3674
3675     pkt_metadata_from_flow(&packet->md, flow);
3676     execute.packet = packet;
3677     execute.needs_help = (xout.slow & SLOW_ACTION) != 0;
3678     execute.probe = false;
3679
3680     /* Fix up in_port. */
3681     in_port = flow->in_port.ofp_port;
3682     if (in_port == OFPP_NONE) {
3683         in_port = OFPP_LOCAL;
3684     }
3685     execute.packet->md.in_port.odp_port = ofp_port_to_odp_port(ofproto, in_port);
3686
3687     error = dpif_execute(ofproto->backer->dpif, &execute);
3688
3689     xlate_out_uninit(&xout);
3690     ofpbuf_uninit(&odp_actions);
3691
3692     return error;
3693 }
3694
3695 void
3696 rule_dpif_credit_stats(struct rule_dpif *rule,
3697                        const struct dpif_flow_stats *stats)
3698 {
3699     ovs_mutex_lock(&rule->stats_mutex);
3700     if (OVS_UNLIKELY(rule->new_rule)) {
3701         rule_dpif_credit_stats(rule->new_rule, stats);
3702     } else {
3703         rule->stats.n_packets += stats->n_packets;
3704         rule->stats.n_bytes += stats->n_bytes;
3705         rule->stats.used = MAX(rule->stats.used, stats->used);
3706     }
3707     ovs_mutex_unlock(&rule->stats_mutex);
3708 }
3709
3710 ovs_be64
3711 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
3712     OVS_REQUIRES(rule->up.mutex)
3713 {
3714     return rule->up.flow_cookie;
3715 }
3716
3717 void
3718 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
3719                      uint16_t hard_timeout)
3720 {
3721     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
3722 }
3723
3724 /* Returns 'rule''s actions.  The returned actions are RCU-protected, and can
3725  * be read until the calling thread quiesces. */
3726 const struct rule_actions *
3727 rule_dpif_get_actions(const struct rule_dpif *rule)
3728 {
3729     return rule_get_actions(&rule->up);
3730 }
3731
3732 /* Sets 'rule''s recirculation id. */
3733 static void
3734 rule_dpif_set_recirc_id(struct rule_dpif *rule, uint32_t id)
3735     OVS_REQUIRES(rule->up.mutex)
3736 {
3737     ovs_assert(!rule->recirc_id || rule->recirc_id == id);
3738     if (rule->recirc_id == id) {
3739         /* Release the new reference to the same id. */
3740         recirc_free_id(id);
3741     } else {
3742         rule->recirc_id = id;
3743     }
3744 }
3745
3746 /* Sets 'rule''s recirculation id. */
3747 void
3748 rule_set_recirc_id(struct rule *rule_, uint32_t id)
3749 {
3750     struct rule_dpif *rule = rule_dpif_cast(rule_);
3751
3752     ovs_mutex_lock(&rule->up.mutex);
3753     rule_dpif_set_recirc_id(rule, id);
3754     ovs_mutex_unlock(&rule->up.mutex);
3755 }
3756
3757 cls_version_t
3758 ofproto_dpif_get_tables_version(struct ofproto_dpif *ofproto OVS_UNUSED)
3759 {
3760     cls_version_t version;
3761
3762     atomic_read_relaxed(&ofproto->tables_version, &version);
3763
3764     return version;
3765 }
3766
3767 /* The returned rule (if any) is valid at least until the next RCU quiescent
3768  * period.  If the rule needs to stay around longer, a non-zero 'take_ref'
3769  * must be passed in to cause a reference to be taken on it.
3770  *
3771  * 'flow' is non-const to allow for temporary modifications during the lookup.
3772  * Any changes are restored before returning. */
3773 static struct rule_dpif *
3774 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, cls_version_t version,
3775                           uint8_t table_id, struct flow *flow,
3776                           struct flow_wildcards *wc, bool take_ref)
3777 {
3778     struct classifier *cls = &ofproto->up.tables[table_id].cls;
3779     const struct cls_rule *cls_rule;
3780     struct rule_dpif *rule;
3781
3782     do {
3783         cls_rule = classifier_lookup(cls, version, flow, wc);
3784
3785         rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
3786
3787         /* Try again if the rule was released before we get the reference. */
3788     } while (rule && take_ref && !rule_dpif_try_ref(rule));
3789
3790     return rule;
3791 }
3792
3793 /* Look up 'flow' in 'ofproto''s classifier version 'version', starting from
3794  * table '*table_id'.  Returns the rule that was found, which may be one of the
3795  * special rules according to packet miss hadling.  If 'may_packet_in' is
3796  * false, returning of the miss_rule (which issues packet ins for the
3797  * controller) is avoided.  Updates 'wc', if nonnull, to reflect the fields
3798  * that were used during the lookup.
3799  *
3800  * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3801  * if none is found then the table miss configuration for that table is
3802  * honored, which can result in additional lookups in other OpenFlow tables.
3803  * In this case the function updates '*table_id' to reflect the final OpenFlow
3804  * table that was searched.
3805  *
3806  * If 'honor_table_miss' is false, then only one table lookup occurs, in
3807  * '*table_id'.
3808  *
3809  * The rule is returned in '*rule', which is valid at least until the next
3810  * RCU quiescent period.  If the '*rule' needs to stay around longer,
3811  * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3812  * on it before this returns.
3813  *
3814  * 'in_port' allows the lookup to take place as if the in port had the value
3815  * 'in_port'.  This is needed for resubmit action support.
3816  *
3817  * 'flow' is non-const to allow for temporary modifications during the lookup.
3818  * Any changes are restored before returning. */
3819 struct rule_dpif *
3820 rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto,
3821                             cls_version_t version, struct flow *flow,
3822                             struct flow_wildcards *wc, bool take_ref,
3823                             const struct dpif_flow_stats *stats,
3824                             uint8_t *table_id, ofp_port_t in_port,
3825                             bool may_packet_in, bool honor_table_miss)
3826 {
3827     ovs_be16 old_tp_src = flow->tp_src, old_tp_dst = flow->tp_dst;
3828     ofp_port_t old_in_port = flow->in_port.ofp_port;
3829     enum ofputil_table_miss miss_config;
3830     struct rule_dpif *rule;
3831     uint8_t next_id;
3832
3833     /* We always unwildcard nw_frag (for IP), so they
3834      * need not be unwildcarded here. */
3835     if (flow->nw_frag & FLOW_NW_FRAG_ANY
3836         && ofproto->up.frag_handling != OFPC_FRAG_NX_MATCH) {
3837         if (ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
3838             /* We must pretend that transport ports are unavailable. */
3839             flow->tp_src = htons(0);
3840             flow->tp_dst = htons(0);
3841         } else {
3842             /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM).
3843              * Use the drop_frags_rule (which cannot disappear). */
3844             rule = ofproto->drop_frags_rule;
3845             if (take_ref) {
3846                 rule_dpif_ref(rule);
3847             }
3848             if (stats) {
3849                 struct oftable *tbl = &ofproto->up.tables[*table_id];
3850                 unsigned long orig;
3851
3852                 atomic_add_relaxed(&tbl->n_matched, stats->n_packets, &orig);
3853             }
3854             return rule;
3855         }
3856     }
3857
3858     /* Look up a flow with 'in_port' as the input port.  Then restore the
3859      * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
3860      * have surprising behavior). */
3861     flow->in_port.ofp_port = in_port;
3862
3863     /* Our current implementation depends on n_tables == N_TABLES, and
3864      * TBL_INTERNAL being the last table. */
3865     BUILD_ASSERT_DECL(N_TABLES == TBL_INTERNAL + 1);
3866
3867     miss_config = OFPUTIL_TABLE_MISS_CONTINUE;
3868
3869     for (next_id = *table_id;
3870          next_id < ofproto->up.n_tables;
3871          next_id++, next_id += (next_id == TBL_INTERNAL))
3872     {
3873         *table_id = next_id;
3874         rule = rule_dpif_lookup_in_table(ofproto, version, next_id, flow, wc,
3875                                          take_ref);
3876         if (stats) {
3877             struct oftable *tbl = &ofproto->up.tables[next_id];
3878             unsigned long orig;
3879
3880             atomic_add_relaxed(rule ? &tbl->n_matched : &tbl->n_missed,
3881                                stats->n_packets, &orig);
3882         }
3883         if (rule) {
3884             goto out;   /* Match. */
3885         }
3886         if (honor_table_miss) {
3887             miss_config = ofproto_table_get_miss_config(&ofproto->up,
3888                                                         *table_id);
3889             if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE) {
3890                 continue;
3891             }
3892         }
3893         break;
3894     }
3895     /* Miss. */
3896     rule = ofproto->no_packet_in_rule;
3897     if (may_packet_in) {
3898         if (miss_config == OFPUTIL_TABLE_MISS_CONTINUE
3899             || miss_config == OFPUTIL_TABLE_MISS_CONTROLLER) {
3900             struct ofport_dpif *port;
3901
3902             port = ofp_port_to_ofport(ofproto, old_in_port);
3903             if (!port) {
3904                 VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
3905                              old_in_port);
3906             } else if (!(port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN)) {
3907                 rule = ofproto->miss_rule;
3908             }
3909         } else if (miss_config == OFPUTIL_TABLE_MISS_DEFAULT &&
3910                    connmgr_wants_packet_in_on_miss(ofproto->up.connmgr)) {
3911             rule = ofproto->miss_rule;
3912         }
3913     }
3914     if (take_ref) {
3915         rule_dpif_ref(rule);
3916     }
3917 out:
3918     /* Restore port numbers, as they may have been modified above. */
3919     flow->tp_src = old_tp_src;
3920     flow->tp_dst = old_tp_dst;
3921     /* Restore the old in port. */
3922     flow->in_port.ofp_port = old_in_port;
3923
3924     return rule;
3925 }
3926
3927 static void
3928 complete_operation(struct rule_dpif *rule)
3929     OVS_REQUIRES(ofproto_mutex)
3930 {
3931     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3932
3933     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
3934 }
3935
3936 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
3937 {
3938     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
3939 }
3940
3941 static struct rule *
3942 rule_alloc(void)
3943 {
3944     struct rule_dpif *rule = xzalloc(sizeof *rule);
3945     return &rule->up;
3946 }
3947
3948 static void
3949 rule_dealloc(struct rule *rule_)
3950 {
3951     struct rule_dpif *rule = rule_dpif_cast(rule_);
3952     free(rule);
3953 }
3954
3955 static enum ofperr
3956 rule_construct(struct rule *rule_)
3957     OVS_NO_THREAD_SAFETY_ANALYSIS
3958 {
3959     struct rule_dpif *rule = rule_dpif_cast(rule_);
3960     ovs_mutex_init_adaptive(&rule->stats_mutex);
3961     rule->stats.n_packets = 0;
3962     rule->stats.n_bytes = 0;
3963     rule->stats.used = rule->up.modified;
3964     rule->recirc_id = 0;
3965     rule->new_rule = NULL;
3966
3967     return 0;
3968 }
3969
3970 static void
3971 rule_insert(struct rule *rule_, struct rule *old_rule_, bool forward_stats)
3972     OVS_REQUIRES(ofproto_mutex)
3973 {
3974     struct rule_dpif *rule = rule_dpif_cast(rule_);
3975
3976     if (old_rule_ && forward_stats) {
3977         struct rule_dpif *old_rule = rule_dpif_cast(old_rule_);
3978
3979         ovs_assert(!old_rule->new_rule);
3980
3981         /* Take a reference to the new rule, and refer all stats updates from
3982          * the old rule to the new rule. */
3983         rule_dpif_ref(rule);
3984
3985         ovs_mutex_lock(&old_rule->stats_mutex);
3986         ovs_mutex_lock(&rule->stats_mutex);
3987         old_rule->new_rule = rule;       /* Forward future stats. */
3988         rule->stats = old_rule->stats;   /* Transfer stats to the new rule. */
3989         ovs_mutex_unlock(&rule->stats_mutex);
3990         ovs_mutex_unlock(&old_rule->stats_mutex);
3991     }
3992
3993     complete_operation(rule);
3994 }
3995
3996 static void
3997 rule_delete(struct rule *rule_)
3998     OVS_REQUIRES(ofproto_mutex)
3999 {
4000     struct rule_dpif *rule = rule_dpif_cast(rule_);
4001     complete_operation(rule);
4002 }
4003
4004 static void
4005 rule_destruct(struct rule *rule_)
4006     OVS_NO_THREAD_SAFETY_ANALYSIS
4007 {
4008     struct rule_dpif *rule = rule_dpif_cast(rule_);
4009
4010     ovs_mutex_destroy(&rule->stats_mutex);
4011     /* Release reference to the new rule, if any. */
4012     if (rule->new_rule) {
4013         rule_dpif_unref(rule->new_rule);
4014     }
4015     if (rule->recirc_id) {
4016         recirc_free_id(rule->recirc_id);
4017     }
4018 }
4019
4020 static void
4021 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes,
4022                long long int *used)
4023 {
4024     struct rule_dpif *rule = rule_dpif_cast(rule_);
4025
4026     ovs_mutex_lock(&rule->stats_mutex);
4027     if (OVS_UNLIKELY(rule->new_rule)) {
4028         rule_get_stats(&rule->new_rule->up, packets, bytes, used);
4029     } else {
4030         *packets = rule->stats.n_packets;
4031         *bytes = rule->stats.n_bytes;
4032         *used = rule->stats.used;
4033     }
4034     ovs_mutex_unlock(&rule->stats_mutex);
4035 }
4036
4037 static void
4038 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4039                   struct dp_packet *packet)
4040 {
4041     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4042
4043     ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
4044 }
4045
4046 static enum ofperr
4047 rule_execute(struct rule *rule, const struct flow *flow,
4048              struct dp_packet *packet)
4049 {
4050     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4051     dp_packet_delete(packet);
4052     return 0;
4053 }
4054
4055 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4056 {
4057     return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4058 }
4059
4060 static struct ofgroup *
4061 group_alloc(void)
4062 {
4063     struct group_dpif *group = xzalloc(sizeof *group);
4064     return &group->up;
4065 }
4066
4067 static void
4068 group_dealloc(struct ofgroup *group_)
4069 {
4070     struct group_dpif *group = group_dpif_cast(group_);
4071     free(group);
4072 }
4073
4074 static void
4075 group_construct_stats(struct group_dpif *group)
4076     OVS_REQUIRES(group->stats_mutex)
4077 {
4078     struct ofputil_bucket *bucket;
4079     const struct ovs_list *buckets;
4080
4081     group->packet_count = 0;
4082     group->byte_count = 0;
4083
4084     group_dpif_get_buckets(group, &buckets);
4085     LIST_FOR_EACH (bucket, list_node, buckets) {
4086         bucket->stats.packet_count = 0;
4087         bucket->stats.byte_count = 0;
4088     }
4089 }
4090
4091 void
4092 group_dpif_credit_stats(struct group_dpif *group,
4093                         struct ofputil_bucket *bucket,
4094                         const struct dpif_flow_stats *stats)
4095 {
4096     ovs_mutex_lock(&group->stats_mutex);
4097     group->packet_count += stats->n_packets;
4098     group->byte_count += stats->n_bytes;
4099     if (bucket) {
4100         bucket->stats.packet_count += stats->n_packets;
4101         bucket->stats.byte_count += stats->n_bytes;
4102     } else { /* Credit to all buckets */
4103         const struct ovs_list *buckets;
4104
4105         group_dpif_get_buckets(group, &buckets);
4106         LIST_FOR_EACH (bucket, list_node, buckets) {
4107             bucket->stats.packet_count += stats->n_packets;
4108             bucket->stats.byte_count += stats->n_bytes;
4109         }
4110     }
4111     ovs_mutex_unlock(&group->stats_mutex);
4112 }
4113
4114 static enum ofperr
4115 group_construct(struct ofgroup *group_)
4116 {
4117     struct group_dpif *group = group_dpif_cast(group_);
4118
4119     ovs_mutex_init_adaptive(&group->stats_mutex);
4120     ovs_mutex_lock(&group->stats_mutex);
4121     group_construct_stats(group);
4122     ovs_mutex_unlock(&group->stats_mutex);
4123     return 0;
4124 }
4125
4126 static void
4127 group_destruct(struct ofgroup *group_)
4128 {
4129     struct group_dpif *group = group_dpif_cast(group_);
4130     ovs_mutex_destroy(&group->stats_mutex);
4131 }
4132
4133 static enum ofperr
4134 group_modify(struct ofgroup *group_)
4135 {
4136     struct ofproto_dpif *ofproto = ofproto_dpif_cast(group_->ofproto);
4137
4138     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4139
4140     return 0;
4141 }
4142
4143 static enum ofperr
4144 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4145 {
4146     struct group_dpif *group = group_dpif_cast(group_);
4147     struct ofputil_bucket *bucket;
4148     const struct ovs_list *buckets;
4149     struct bucket_counter *bucket_stats;
4150
4151     ovs_mutex_lock(&group->stats_mutex);
4152     ogs->packet_count = group->packet_count;
4153     ogs->byte_count = group->byte_count;
4154
4155     group_dpif_get_buckets(group, &buckets);
4156     bucket_stats = ogs->bucket_stats;
4157     LIST_FOR_EACH (bucket, list_node, buckets) {
4158         bucket_stats->packet_count = bucket->stats.packet_count;
4159         bucket_stats->byte_count = bucket->stats.byte_count;
4160         bucket_stats++;
4161     }
4162     ovs_mutex_unlock(&group->stats_mutex);
4163
4164     return 0;
4165 }
4166
4167 /* If the group exists, this function increments the groups's reference count.
4168  *
4169  * Make sure to call group_dpif_unref() after no longer needing to maintain
4170  * a reference to the group. */
4171 bool
4172 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4173                   struct group_dpif **group)
4174 {
4175     struct ofgroup *ofgroup;
4176     bool found;
4177
4178     found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4179     *group = found ?  group_dpif_cast(ofgroup) : NULL;
4180
4181     return found;
4182 }
4183
4184 void
4185 group_dpif_get_buckets(const struct group_dpif *group,
4186                        const struct ovs_list **buckets)
4187 {
4188     *buckets = &group->up.buckets;
4189 }
4190
4191 enum ofp11_group_type
4192 group_dpif_get_type(const struct group_dpif *group)
4193 {
4194     return group->up.type;
4195 }
4196
4197 const char *
4198 group_dpif_get_selection_method(const struct group_dpif *group)
4199 {
4200     return group->up.props.selection_method;
4201 }
4202 \f
4203 /* Sends 'packet' out 'ofport'.
4204  * May modify 'packet'.
4205  * Returns 0 if successful, otherwise a positive errno value. */
4206 int
4207 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct dp_packet *packet)
4208 {
4209     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4210     int error;
4211
4212     error = xlate_send_packet(ofport, packet);
4213
4214     ovs_mutex_lock(&ofproto->stats_mutex);
4215     ofproto->stats.tx_packets++;
4216     ofproto->stats.tx_bytes += dp_packet_size(packet);
4217     ovs_mutex_unlock(&ofproto->stats_mutex);
4218     return error;
4219 }
4220
4221 uint64_t
4222 group_dpif_get_selection_method_param(const struct group_dpif *group)
4223 {
4224     return group->up.props.selection_method_param;
4225 }
4226
4227 const struct field_array *
4228 group_dpif_get_fields(const struct group_dpif *group)
4229 {
4230     return &group->up.props.fields;
4231 }
4232 \f
4233 /* Return the version string of the datapath that backs up
4234  * this 'ofproto'.
4235  */
4236 static const char *
4237 get_datapath_version(const struct ofproto *ofproto_)
4238 {
4239     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4240
4241     return ofproto->backer->dp_version_string;
4242 }
4243
4244 static bool
4245 set_frag_handling(struct ofproto *ofproto_,
4246                   enum ofp_config_flags frag_handling)
4247 {
4248     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4249     if (frag_handling != OFPC_FRAG_REASM) {
4250         ofproto->backer->need_revalidate = REV_RECONFIGURE;
4251         return true;
4252     } else {
4253         return false;
4254     }
4255 }
4256
4257 static enum ofperr
4258 packet_out(struct ofproto *ofproto_, struct dp_packet *packet,
4259            const struct flow *flow,
4260            const struct ofpact *ofpacts, size_t ofpacts_len)
4261 {
4262     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4263
4264     ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
4265                                  ofpacts_len, packet);
4266     return 0;
4267 }
4268 \f
4269 /* NetFlow. */
4270
4271 static int
4272 set_netflow(struct ofproto *ofproto_,
4273             const struct netflow_options *netflow_options)
4274 {
4275     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4276
4277     if (netflow_options) {
4278         if (!ofproto->netflow) {
4279             ofproto->netflow = netflow_create();
4280             ofproto->backer->need_revalidate = REV_RECONFIGURE;
4281         }
4282         return netflow_set_options(ofproto->netflow, netflow_options);
4283     } else if (ofproto->netflow) {
4284         ofproto->backer->need_revalidate = REV_RECONFIGURE;
4285         netflow_unref(ofproto->netflow);
4286         ofproto->netflow = NULL;
4287     }
4288
4289     return 0;
4290 }
4291
4292 static void
4293 get_netflow_ids(const struct ofproto *ofproto_,
4294                 uint8_t *engine_type, uint8_t *engine_id)
4295 {
4296     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4297
4298     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
4299 }
4300 \f
4301 static struct ofproto_dpif *
4302 ofproto_dpif_lookup(const char *name)
4303 {
4304     struct ofproto_dpif *ofproto;
4305
4306     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
4307                              hash_string(name, 0), &all_ofproto_dpifs) {
4308         if (!strcmp(ofproto->up.name, name)) {
4309             return ofproto;
4310         }
4311     }
4312     return NULL;
4313 }
4314
4315 static void
4316 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
4317                           const char *argv[], void *aux OVS_UNUSED)
4318 {
4319     struct ofproto_dpif *ofproto;
4320
4321     if (argc > 1) {
4322         ofproto = ofproto_dpif_lookup(argv[1]);
4323         if (!ofproto) {
4324             unixctl_command_reply_error(conn, "no such bridge");
4325             return;
4326         }
4327         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4328         mac_learning_flush(ofproto->ml);
4329         ovs_rwlock_unlock(&ofproto->ml->rwlock);
4330     } else {
4331         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4332             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
4333             mac_learning_flush(ofproto->ml);
4334             ovs_rwlock_unlock(&ofproto->ml->rwlock);
4335         }
4336     }
4337
4338     unixctl_command_reply(conn, "table successfully flushed");
4339 }
4340
4341 static void
4342 ofproto_unixctl_mcast_snooping_flush(struct unixctl_conn *conn, int argc,
4343                                      const char *argv[], void *aux OVS_UNUSED)
4344 {
4345     struct ofproto_dpif *ofproto;
4346
4347     if (argc > 1) {
4348         ofproto = ofproto_dpif_lookup(argv[1]);
4349         if (!ofproto) {
4350             unixctl_command_reply_error(conn, "no such bridge");
4351             return;
4352         }
4353
4354         if (!mcast_snooping_enabled(ofproto->ms)) {
4355             unixctl_command_reply_error(conn, "multicast snooping is disabled");
4356             return;
4357         }
4358         mcast_snooping_mdb_flush(ofproto->ms);
4359     } else {
4360         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4361             if (!mcast_snooping_enabled(ofproto->ms)) {
4362                 continue;
4363             }
4364             mcast_snooping_mdb_flush(ofproto->ms);
4365         }
4366     }
4367
4368     unixctl_command_reply(conn, "table successfully flushed");
4369 }
4370
4371 static struct ofport_dpif *
4372 ofbundle_get_a_port(const struct ofbundle *bundle)
4373 {
4374     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
4375                         bundle_node);
4376 }
4377
4378 static void
4379 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
4380                          const char *argv[], void *aux OVS_UNUSED)
4381 {
4382     struct ds ds = DS_EMPTY_INITIALIZER;
4383     const struct ofproto_dpif *ofproto;
4384     const struct mac_entry *e;
4385
4386     ofproto = ofproto_dpif_lookup(argv[1]);
4387     if (!ofproto) {
4388         unixctl_command_reply_error(conn, "no such bridge");
4389         return;
4390     }
4391
4392     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
4393     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
4394     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
4395         struct ofbundle *bundle = mac_entry_get_port(ofproto->ml, e);
4396         char name[OFP_MAX_PORT_NAME_LEN];
4397
4398         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4399                                name, sizeof name);
4400         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
4401                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
4402                       mac_entry_age(ofproto->ml, e));
4403     }
4404     ovs_rwlock_unlock(&ofproto->ml->rwlock);
4405     unixctl_command_reply(conn, ds_cstr(&ds));
4406     ds_destroy(&ds);
4407 }
4408
4409 static void
4410 ofproto_unixctl_mcast_snooping_show(struct unixctl_conn *conn,
4411                                     int argc OVS_UNUSED,
4412                                     const char *argv[],
4413                                     void *aux OVS_UNUSED)
4414 {
4415     struct ds ds = DS_EMPTY_INITIALIZER;
4416     const struct ofproto_dpif *ofproto;
4417     const struct ofbundle *bundle;
4418     const struct mcast_group *grp;
4419     struct mcast_group_bundle *b;
4420     struct mcast_mrouter_bundle *mrouter;
4421
4422     ofproto = ofproto_dpif_lookup(argv[1]);
4423     if (!ofproto) {
4424         unixctl_command_reply_error(conn, "no such bridge");
4425         return;
4426     }
4427
4428     if (!mcast_snooping_enabled(ofproto->ms)) {
4429         unixctl_command_reply_error(conn, "multicast snooping is disabled");
4430         return;
4431     }
4432
4433     ds_put_cstr(&ds, " port  VLAN  GROUP                Age\n");
4434     ovs_rwlock_rdlock(&ofproto->ms->rwlock);
4435     LIST_FOR_EACH (grp, group_node, &ofproto->ms->group_lru) {
4436         LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
4437             char name[OFP_MAX_PORT_NAME_LEN];
4438
4439             bundle = b->port;
4440             ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4441                                    name, sizeof name);
4442             ds_put_format(&ds, "%5s  %4d  ", name, grp->vlan);
4443             print_ipv6_mapped(&ds, &grp->addr);
4444             ds_put_format(&ds, "         %3d\n",
4445                           mcast_bundle_age(ofproto->ms, b));
4446         }
4447     }
4448
4449     /* ports connected to multicast routers */
4450     LIST_FOR_EACH(mrouter, mrouter_node, &ofproto->ms->mrouter_lru) {
4451         char name[OFP_MAX_PORT_NAME_LEN];
4452
4453         bundle = mrouter->port;
4454         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
4455                                name, sizeof name);
4456         ds_put_format(&ds, "%5s  %4d  querier             %3d\n",
4457                       name, mrouter->vlan,
4458                       mcast_mrouter_age(ofproto->ms, mrouter));
4459     }
4460     ovs_rwlock_unlock(&ofproto->ms->rwlock);
4461     unixctl_command_reply(conn, ds_cstr(&ds));
4462     ds_destroy(&ds);
4463 }
4464
4465 struct trace_ctx {
4466     struct xlate_out xout;
4467     struct xlate_in xin;
4468     const struct flow *key;
4469     struct flow flow;
4470     struct ds *result;
4471     struct flow_wildcards wc;
4472     struct ofpbuf odp_actions;
4473 };
4474
4475 static void
4476 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
4477 {
4478     const struct rule_actions *actions;
4479     ovs_be64 cookie;
4480
4481     ds_put_char_multiple(result, '\t', level);
4482     if (!rule) {
4483         ds_put_cstr(result, "No match\n");
4484         return;
4485     }
4486
4487     ovs_mutex_lock(&rule->up.mutex);
4488     cookie = rule->up.flow_cookie;
4489     ovs_mutex_unlock(&rule->up.mutex);
4490
4491     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
4492                   rule ? rule->up.table_id : 0, ntohll(cookie));
4493     cls_rule_format(&rule->up.cr, result);
4494     ds_put_char(result, '\n');
4495
4496     actions = rule_dpif_get_actions(rule);
4497
4498     ds_put_char_multiple(result, '\t', level);
4499     ds_put_cstr(result, "OpenFlow actions=");
4500     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
4501     ds_put_char(result, '\n');
4502 }
4503
4504 static void
4505 trace_format_flow(struct ds *result, int level, const char *title,
4506                   struct trace_ctx *trace)
4507 {
4508     ds_put_char_multiple(result, '\t', level);
4509     ds_put_format(result, "%s: ", title);
4510     /* Do not report unchanged flows for resubmits. */
4511     if ((level > 0 && flow_equal(&trace->xin.flow, &trace->flow))
4512         || (level == 0 && flow_equal(&trace->xin.flow, trace->key))) {
4513         ds_put_cstr(result, "unchanged");
4514     } else {
4515         flow_format(result, &trace->xin.flow);
4516         trace->flow = trace->xin.flow;
4517     }
4518     ds_put_char(result, '\n');
4519 }
4520
4521 static void
4522 trace_format_regs(struct ds *result, int level, const char *title,
4523                   struct trace_ctx *trace)
4524 {
4525     size_t i;
4526
4527     ds_put_char_multiple(result, '\t', level);
4528     ds_put_format(result, "%s:", title);
4529     for (i = 0; i < FLOW_N_REGS; i++) {
4530         ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
4531     }
4532     ds_put_char(result, '\n');
4533 }
4534
4535 static void
4536 trace_format_odp(struct ds *result, int level, const char *title,
4537                  struct trace_ctx *trace)
4538 {
4539     struct ofpbuf *odp_actions = &trace->odp_actions;
4540
4541     ds_put_char_multiple(result, '\t', level);
4542     ds_put_format(result, "%s: ", title);
4543     format_odp_actions(result, odp_actions->data, odp_actions->size);
4544     ds_put_char(result, '\n');
4545 }
4546
4547 static void
4548 trace_format_megaflow(struct ds *result, int level, const char *title,
4549                       struct trace_ctx *trace)
4550 {
4551     struct match match;
4552
4553     ds_put_char_multiple(result, '\t', level);
4554     ds_put_format(result, "%s: ", title);
4555     match_init(&match, trace->key, &trace->wc);
4556     match_format(&match, result, OFP_DEFAULT_PRIORITY);
4557     ds_put_char(result, '\n');
4558 }
4559
4560 static void trace_report(struct xlate_in *, int recurse,
4561                          const char *format, ...)
4562     OVS_PRINTF_FORMAT(3, 4);
4563 static void trace_report_valist(struct xlate_in *, int recurse,
4564                                 const char *format, va_list args)
4565     OVS_PRINTF_FORMAT(3, 0);
4566
4567 static void
4568 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
4569 {
4570     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4571     struct ds *result = trace->result;
4572
4573     if (!recurse) {
4574         if (rule == xin->ofproto->miss_rule) {
4575             trace_report(xin, recurse,
4576                          "No match, flow generates \"packet in\"s.");
4577         } else if (rule == xin->ofproto->no_packet_in_rule) {
4578             trace_report(xin, recurse, "No match, packets dropped because "
4579                          "OFPPC_NO_PACKET_IN is set on in_port.");
4580         } else if (rule == xin->ofproto->drop_frags_rule) {
4581             trace_report(xin, recurse, "Packets dropped because they are IP "
4582                          "fragments and the fragment handling mode is "
4583                          "\"drop\".");
4584         }
4585     }
4586
4587     ds_put_char(result, '\n');
4588     if (recurse) {
4589         trace_format_flow(result, recurse, "Resubmitted flow", trace);
4590         trace_format_regs(result, recurse, "Resubmitted regs", trace);
4591         trace_format_odp(result,  recurse, "Resubmitted  odp", trace);
4592         trace_format_megaflow(result, recurse, "Resubmitted megaflow", trace);
4593     }
4594     trace_format_rule(result, recurse, rule);
4595 }
4596
4597 static void
4598 trace_report_valist(struct xlate_in *xin, int recurse,
4599                     const char *format, va_list args)
4600 {
4601     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
4602     struct ds *result = trace->result;
4603
4604     ds_put_char_multiple(result, '\t', recurse);
4605     ds_put_format_valist(result, format, args);
4606     ds_put_char(result, '\n');
4607 }
4608
4609 static void
4610 trace_report(struct xlate_in *xin, int recurse, const char *format, ...)
4611 {
4612     va_list args;
4613
4614     va_start(args, format);
4615     trace_report_valist(xin, recurse, format, args);
4616     va_end(args);
4617 }
4618
4619 /* Parses the 'argc' elements of 'argv', ignoring argv[0].  The following
4620  * forms are supported:
4621  *
4622  *     - [dpname] odp_flow [-generate | packet]
4623  *     - bridge br_flow [-generate | packet]
4624  *
4625  * On success, initializes '*ofprotop' and 'flow' and returns NULL.  On failure
4626  * returns a nonnull malloced error message. */
4627 static char * OVS_WARN_UNUSED_RESULT
4628 parse_flow_and_packet(int argc, const char *argv[],
4629                       struct ofproto_dpif **ofprotop, struct flow *flow,
4630                       struct dp_packet **packetp)
4631 {
4632     const struct dpif_backer *backer = NULL;
4633     const char *error = NULL;
4634     char *m_err = NULL;
4635     struct simap port_names = SIMAP_INITIALIZER(&port_names);
4636     struct dp_packet *packet;
4637     struct ofpbuf odp_key;
4638     struct ofpbuf odp_mask;
4639
4640     ofpbuf_init(&odp_key, 0);
4641     ofpbuf_init(&odp_mask, 0);
4642
4643     /* Handle "-generate" or a hex string as the last argument. */
4644     if (!strcmp(argv[argc - 1], "-generate")) {
4645         packet = dp_packet_new(0);
4646         argc--;
4647     } else {
4648         error = eth_from_hex(argv[argc - 1], &packet);
4649         if (!error) {
4650             argc--;
4651         } else if (argc == 4) {
4652             /* The 3-argument form must end in "-generate' or a hex string. */
4653             goto exit;
4654         }
4655         error = NULL;
4656     }
4657
4658     /* odp_flow can have its in_port specified as a name instead of port no.
4659      * We do not yet know whether a given flow is a odp_flow or a br_flow.
4660      * But, to know whether a flow is odp_flow through odp_flow_from_string(),
4661      * we need to create a simap of name to port no. */
4662     if (argc == 3) {
4663         const char *dp_type;
4664         if (!strncmp(argv[1], "ovs-", 4)) {
4665             dp_type = argv[1] + 4;
4666         } else {
4667             dp_type = argv[1];
4668         }
4669         backer = shash_find_data(&all_dpif_backers, dp_type);
4670     } else if (argc == 2) {
4671         struct shash_node *node;
4672         if (shash_count(&all_dpif_backers) == 1) {
4673             node = shash_first(&all_dpif_backers);
4674             backer = node->data;
4675         }
4676     } else {
4677         error = "Syntax error";
4678         goto exit;
4679     }
4680     if (backer && backer->dpif) {
4681         struct dpif_port dpif_port;
4682         struct dpif_port_dump port_dump;
4683         DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
4684             simap_put(&port_names, dpif_port.name,
4685                       odp_to_u32(dpif_port.port_no));
4686         }
4687     }
4688
4689     /* Parse the flow and determine whether a datapath or
4690      * bridge is specified. If function odp_flow_key_from_string()
4691      * returns 0, the flow is a odp_flow. If function
4692      * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4693     if (!odp_flow_from_string(argv[argc - 1], &port_names,
4694                               &odp_key, &odp_mask)) {
4695         if (!backer) {
4696             error = "Cannot find the datapath";
4697             goto exit;
4698         }
4699
4700         if (odp_flow_key_to_flow(odp_key.data, odp_key.size, flow) == ODP_FIT_ERROR) {
4701             error = "Failed to parse datapath flow key";
4702             goto exit;
4703         }
4704
4705         *ofprotop = xlate_lookup_ofproto(backer, flow,
4706                                          &flow->in_port.ofp_port);
4707         if (*ofprotop == NULL) {
4708             error = "Invalid datapath flow";
4709             goto exit;
4710         }
4711
4712         vsp_adjust_flow(*ofprotop, flow, NULL);
4713
4714     } else {
4715         char *err = parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL);
4716
4717         if (err) {
4718             m_err = xasprintf("Bad openflow flow syntax: %s", err);
4719             free(err);
4720             goto exit;
4721         } else {
4722             if (argc != 3) {
4723                 error = "Must specify bridge name";
4724                 goto exit;
4725             }
4726
4727             *ofprotop = ofproto_dpif_lookup(argv[1]);
4728             if (!*ofprotop) {
4729                 error = "Unknown bridge name";
4730                 goto exit;
4731             }
4732         }
4733     }
4734
4735     /* Generate a packet, if requested. */
4736     if (packet) {
4737         if (!dp_packet_size(packet)) {
4738             flow_compose(packet, flow);
4739         } else {
4740             /* Use the metadata from the flow and the packet argument
4741              * to reconstruct the flow. */
4742             pkt_metadata_from_flow(&packet->md, flow);
4743             flow_extract(packet, flow);
4744         }
4745     }
4746
4747 exit:
4748     if (error && !m_err) {
4749         m_err = xstrdup(error);
4750     }
4751     if (m_err) {
4752         dp_packet_delete(packet);
4753         packet = NULL;
4754     }
4755     *packetp = packet;
4756     ofpbuf_uninit(&odp_key);
4757     ofpbuf_uninit(&odp_mask);
4758     simap_destroy(&port_names);
4759     return m_err;
4760 }
4761
4762 static void
4763 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
4764                       void *aux OVS_UNUSED)
4765 {
4766     struct ofproto_dpif *ofproto;
4767     struct dp_packet *packet;
4768     char *error;
4769     struct flow flow;
4770
4771     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4772     if (!error) {
4773         struct ds result;
4774
4775         ds_init(&result);
4776         ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
4777         unixctl_command_reply(conn, ds_cstr(&result));
4778         ds_destroy(&result);
4779         dp_packet_delete(packet);
4780     } else {
4781         unixctl_command_reply_error(conn, error);
4782         free(error);
4783     }
4784 }
4785
4786 static void
4787 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
4788                               const char *argv[], void *aux OVS_UNUSED)
4789 {
4790     enum ofputil_protocol usable_protocols;
4791     struct ofproto_dpif *ofproto;
4792     bool enforce_consistency;
4793     struct ofpbuf ofpacts;
4794     struct dp_packet *packet;
4795     struct ds result;
4796     struct flow flow;
4797     uint16_t in_port;
4798
4799     /* Three kinds of error return values! */
4800     enum ofperr retval;
4801     char *error;
4802
4803     packet = NULL;
4804     ds_init(&result);
4805     ofpbuf_init(&ofpacts, 0);
4806
4807     /* Parse actions. */
4808     error = ofpacts_parse_actions(argv[--argc], &ofpacts, &usable_protocols);
4809     if (error) {
4810         unixctl_command_reply_error(conn, error);
4811         free(error);
4812         goto exit;
4813     }
4814
4815     /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4816      * consistency between the flow and the actions.  With -consistent, we
4817      * enforce consistency even for a flow supported in OpenFlow 1.0. */
4818     if (!strcmp(argv[1], "-consistent")) {
4819         enforce_consistency = true;
4820         argv++;
4821         argc--;
4822     } else {
4823         enforce_consistency = false;
4824     }
4825
4826     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
4827     if (error) {
4828         unixctl_command_reply_error(conn, error);
4829         free(error);
4830         goto exit;
4831     }
4832
4833     /* Do the same checks as handle_packet_out() in ofproto.c.
4834      *
4835      * We pass a 'table_id' of 0 to ofpacts_check(), which isn't
4836      * strictly correct because these actions aren't in any table, but it's OK
4837      * because it 'table_id' is used only to check goto_table instructions, but
4838      * packet-outs take a list of actions and therefore it can't include
4839      * instructions.
4840      *
4841      * We skip the "meter" check here because meter is an instruction, not an
4842      * action, and thus cannot appear in ofpacts. */
4843     in_port = ofp_to_u16(flow.in_port.ofp_port);
4844     if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
4845         unixctl_command_reply_error(conn, "invalid in_port");
4846         goto exit;
4847     }
4848     if (enforce_consistency) {
4849         retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size,
4850                                            &flow, u16_to_ofp(ofproto->up.max_ports),
4851                                            0, 0, usable_protocols);
4852     } else {
4853         retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow,
4854                                u16_to_ofp(ofproto->up.max_ports), 0, 0,
4855                                &usable_protocols);
4856     }
4857
4858     if (retval) {
4859         ds_clear(&result);
4860         ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
4861         unixctl_command_reply_error(conn, ds_cstr(&result));
4862         goto exit;
4863     }
4864
4865     ofproto_trace(ofproto, &flow, packet,
4866                   ofpacts.data, ofpacts.size, &result);
4867     unixctl_command_reply(conn, ds_cstr(&result));
4868
4869 exit:
4870     ds_destroy(&result);
4871     dp_packet_delete(packet);
4872     ofpbuf_uninit(&ofpacts);
4873 }
4874
4875 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4876  * description of the results to 'ds'.
4877  *
4878  * The trace follows a packet with the specified 'flow' through the flow
4879  * table.  'packet' may be nonnull to trace an actual packet, with consequent
4880  * side effects (if it is nonnull then its flow must be 'flow').
4881  *
4882  * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4883  * trace, otherwise the actions are determined by a flow table lookup. */
4884 static void
4885 ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow,
4886               const struct dp_packet *packet,
4887               const struct ofpact ofpacts[], size_t ofpacts_len,
4888               struct ds *ds)
4889 {
4890     struct trace_ctx trace;
4891
4892     ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
4893     ds_put_cstr(ds, "Flow: ");
4894     flow_format(ds, flow);
4895     ds_put_char(ds, '\n');
4896
4897     ofpbuf_init(&trace.odp_actions, 0);
4898
4899     trace.result = ds;
4900     trace.key = flow; /* Original flow key, used for megaflow. */
4901     trace.flow = *flow; /* May be modified by actions. */
4902     xlate_in_init(&trace.xin, ofproto, flow, flow->in_port.ofp_port, NULL,
4903                   ntohs(flow->tcp_flags), packet, &trace.wc,
4904                   &trace.odp_actions);
4905     trace.xin.ofpacts = ofpacts;
4906     trace.xin.ofpacts_len = ofpacts_len;
4907     trace.xin.resubmit_hook = trace_resubmit;
4908     trace.xin.report_hook = trace_report_valist;
4909
4910     xlate_actions(&trace.xin, &trace.xout);
4911
4912     ds_put_char(ds, '\n');
4913     trace_format_flow(ds, 0, "Final flow", &trace);
4914     trace_format_megaflow(ds, 0, "Megaflow", &trace);
4915
4916     ds_put_cstr(ds, "Datapath actions: ");
4917     format_odp_actions(ds, trace.odp_actions.data, trace.odp_actions.size);
4918
4919     if (trace.xout.slow) {
4920         enum slow_path_reason slow;
4921
4922         ds_put_cstr(ds, "\nThis flow is handled by the userspace "
4923                     "slow path because it:");
4924
4925         slow = trace.xout.slow;
4926         while (slow) {
4927             enum slow_path_reason bit = rightmost_1bit(slow);
4928
4929             ds_put_format(ds, "\n\t- %s.",
4930                           slow_path_reason_to_explanation(bit));
4931
4932             slow &= ~bit;
4933         }
4934     }
4935
4936     xlate_out_uninit(&trace.xout);
4937     ofpbuf_uninit(&trace.odp_actions);
4938 }
4939
4940 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
4941  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
4942  * to destroy 'ofproto_shash' and free the returned value. */
4943 static const struct shash_node **
4944 get_ofprotos(struct shash *ofproto_shash)
4945 {
4946     const struct ofproto_dpif *ofproto;
4947
4948     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4949         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
4950         shash_add_nocopy(ofproto_shash, name, ofproto);
4951     }
4952
4953     return shash_sort(ofproto_shash);
4954 }
4955
4956 static void
4957 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
4958                               const char *argv[] OVS_UNUSED,
4959                               void *aux OVS_UNUSED)
4960 {
4961     struct ds ds = DS_EMPTY_INITIALIZER;
4962     struct shash ofproto_shash;
4963     const struct shash_node **sorted_ofprotos;
4964     int i;
4965
4966     shash_init(&ofproto_shash);
4967     sorted_ofprotos = get_ofprotos(&ofproto_shash);
4968     for (i = 0; i < shash_count(&ofproto_shash); i++) {
4969         const struct shash_node *node = sorted_ofprotos[i];
4970         ds_put_format(&ds, "%s\n", node->name);
4971     }
4972
4973     shash_destroy(&ofproto_shash);
4974     free(sorted_ofprotos);
4975
4976     unixctl_command_reply(conn, ds_cstr(&ds));
4977     ds_destroy(&ds);
4978 }
4979
4980 static void
4981 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
4982 {
4983     const struct shash_node **ofprotos;
4984     struct dpif_dp_stats dp_stats;
4985     struct shash ofproto_shash;
4986     size_t i;
4987
4988     dpif_get_dp_stats(backer->dpif, &dp_stats);
4989
4990     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
4991                   dpif_name(backer->dpif), dp_stats.n_hit, dp_stats.n_missed);
4992
4993     shash_init(&ofproto_shash);
4994     ofprotos = get_ofprotos(&ofproto_shash);
4995     for (i = 0; i < shash_count(&ofproto_shash); i++) {
4996         struct ofproto_dpif *ofproto = ofprotos[i]->data;
4997         const struct shash_node **ports;
4998         size_t j;
4999
5000         if (ofproto->backer != backer) {
5001             continue;
5002         }
5003
5004         ds_put_format(ds, "\t%s:\n", ofproto->up.name);
5005
5006         ports = shash_sort(&ofproto->up.port_by_name);
5007         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5008             const struct shash_node *node = ports[j];
5009             struct ofport *ofport = node->data;
5010             struct smap config;
5011             odp_port_t odp_port;
5012
5013             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5014                           ofport->ofp_port);
5015
5016             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5017             if (odp_port != ODPP_NONE) {
5018                 ds_put_format(ds, "%"PRIu32":", odp_port);
5019             } else {
5020                 ds_put_cstr(ds, "none:");
5021             }
5022
5023             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5024
5025             smap_init(&config);
5026             if (!netdev_get_config(ofport->netdev, &config)) {
5027                 const struct smap_node **nodes;
5028                 size_t i;
5029
5030                 nodes = smap_sort(&config);
5031                 for (i = 0; i < smap_count(&config); i++) {
5032                     const struct smap_node *node = nodes[i];
5033                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5034                                   node->key, node->value);
5035                 }
5036                 free(nodes);
5037             }
5038             smap_destroy(&config);
5039
5040             ds_put_char(ds, ')');
5041             ds_put_char(ds, '\n');
5042         }
5043         free(ports);
5044     }
5045     shash_destroy(&ofproto_shash);
5046     free(ofprotos);
5047 }
5048
5049 static void
5050 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5051                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5052 {
5053     struct ds ds = DS_EMPTY_INITIALIZER;
5054     const struct shash_node **backers;
5055     int i;
5056
5057     backers = shash_sort(&all_dpif_backers);
5058     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5059         dpif_show_backer(backers[i]->data, &ds);
5060     }
5061     free(backers);
5062
5063     unixctl_command_reply(conn, ds_cstr(&ds));
5064     ds_destroy(&ds);
5065 }
5066
5067 static void
5068 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5069                                 int argc OVS_UNUSED, const char *argv[],
5070                                 void *aux OVS_UNUSED)
5071 {
5072     const struct ofproto_dpif *ofproto;
5073
5074     struct ds ds = DS_EMPTY_INITIALIZER;
5075     bool verbosity = false;
5076
5077     struct dpif_port dpif_port;
5078     struct dpif_port_dump port_dump;
5079     struct hmap portno_names;
5080
5081     struct dpif_flow_dump *flow_dump;
5082     struct dpif_flow_dump_thread *flow_dump_thread;
5083     struct dpif_flow f;
5084     int error;
5085
5086     ofproto = ofproto_dpif_lookup(argv[argc - 1]);
5087     if (!ofproto) {
5088         unixctl_command_reply_error(conn, "no such bridge");
5089         return;
5090     }
5091
5092     if (argc > 2 && !strcmp(argv[1], "-m")) {
5093         verbosity = true;
5094     }
5095
5096     hmap_init(&portno_names);
5097     DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, ofproto->backer->dpif) {
5098         odp_portno_names_set(&portno_names, dpif_port.port_no, dpif_port.name);
5099     }
5100
5101     ds_init(&ds);
5102     flow_dump = dpif_flow_dump_create(ofproto->backer->dpif, false);
5103     flow_dump_thread = dpif_flow_dump_thread_create(flow_dump);
5104     while (dpif_flow_dump_next(flow_dump_thread, &f, 1)) {
5105         struct flow flow;
5106
5107         if (odp_flow_key_to_flow(f.key, f.key_len, &flow) == ODP_FIT_ERROR
5108             || xlate_lookup_ofproto(ofproto->backer, &flow, NULL) != ofproto) {
5109             continue;
5110         }
5111
5112         if (verbosity) {
5113             odp_format_ufid(&f.ufid, &ds);
5114             ds_put_cstr(&ds, " ");
5115         }
5116         odp_flow_format(f.key, f.key_len, f.mask, f.mask_len,
5117                         &portno_names, &ds, verbosity);
5118         ds_put_cstr(&ds, ", ");
5119         dpif_flow_stats_format(&f.stats, &ds);
5120         ds_put_cstr(&ds, ", actions:");
5121         format_odp_actions(&ds, f.actions, f.actions_len);
5122         ds_put_char(&ds, '\n');
5123     }
5124     dpif_flow_dump_thread_destroy(flow_dump_thread);
5125     error = dpif_flow_dump_destroy(flow_dump);
5126
5127     if (error) {
5128         ds_clear(&ds);
5129         ds_put_format(&ds, "dpif/dump_flows failed: %s", ovs_strerror(errno));
5130         unixctl_command_reply_error(conn, ds_cstr(&ds));
5131     } else {
5132         unixctl_command_reply(conn, ds_cstr(&ds));
5133     }
5134     odp_portno_names_destroy(&portno_names);
5135     hmap_destroy(&portno_names);
5136     ds_destroy(&ds);
5137 }
5138
5139 static void
5140 ofproto_revalidate_all_backers(void)
5141 {
5142     const struct shash_node **backers;
5143     int i;
5144
5145     backers = shash_sort(&all_dpif_backers);
5146     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5147         struct dpif_backer *backer = backers[i]->data;
5148         backer->need_revalidate = REV_RECONFIGURE;
5149     }
5150     free(backers);
5151 }
5152
5153 static void
5154 disable_tnl_push_pop(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5155                      const char *argv[], void *aux OVS_UNUSED)
5156 {
5157     if (!strcasecmp(argv[1], "off")) {
5158         ofproto_use_tnl_push_pop = false;
5159         unixctl_command_reply(conn, "Tunnel push-pop off");
5160         ofproto_revalidate_all_backers();
5161     } else if (!strcasecmp(argv[1], "on")) {
5162         ofproto_use_tnl_push_pop = true;
5163         unixctl_command_reply(conn, "Tunnel push-pop on");
5164         ofproto_revalidate_all_backers();
5165     }
5166 }
5167
5168 static void
5169 ofproto_unixctl_init(void)
5170 {
5171     static bool registered;
5172     if (registered) {
5173         return;
5174     }
5175     registered = true;
5176
5177     unixctl_command_register(
5178         "ofproto/trace",
5179         "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
5180         1, 3, ofproto_unixctl_trace, NULL);
5181     unixctl_command_register(
5182         "ofproto/trace-packet-out",
5183         "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
5184         2, 6, ofproto_unixctl_trace_actions, NULL);
5185     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5186                              ofproto_unixctl_fdb_flush, NULL);
5187     unixctl_command_register("fdb/show", "bridge", 1, 1,
5188                              ofproto_unixctl_fdb_show, NULL);
5189     unixctl_command_register("mdb/flush", "[bridge]", 0, 1,
5190                              ofproto_unixctl_mcast_snooping_flush, NULL);
5191     unixctl_command_register("mdb/show", "bridge", 1, 1,
5192                              ofproto_unixctl_mcast_snooping_show, NULL);
5193     unixctl_command_register("dpif/dump-dps", "", 0, 0,
5194                              ofproto_unixctl_dpif_dump_dps, NULL);
5195     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5196                              NULL);
5197     unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
5198                              ofproto_unixctl_dpif_dump_flows, NULL);
5199
5200     unixctl_command_register("ofproto/tnl-push-pop", "[on]|[off]", 1, 1,
5201                              disable_tnl_push_pop, NULL);
5202 }
5203
5204 /* Returns true if 'table' is the table used for internal rules,
5205  * false otherwise. */
5206 bool
5207 table_is_internal(uint8_t table_id)
5208 {
5209     return table_id == TBL_INTERNAL;
5210 }
5211 \f
5212 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5213  *
5214  * This is deprecated.  It is only for compatibility with broken device drivers
5215  * in old versions of Linux that do not properly support VLANs when VLAN
5216  * devices are not used.  When broken device drivers are no longer in
5217  * widespread use, we will delete these interfaces. */
5218
5219 static int
5220 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5221 {
5222     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5223     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5224
5225     if (realdev_ofp_port == ofport->realdev_ofp_port
5226         && vid == ofport->vlandev_vid) {
5227         return 0;
5228     }
5229
5230     ofproto->backer->need_revalidate = REV_RECONFIGURE;
5231
5232     if (ofport->realdev_ofp_port) {
5233         vsp_remove(ofport);
5234     }
5235     if (realdev_ofp_port && ofport->bundle) {
5236         /* vlandevs are enslaved to their realdevs, so they are not allowed to
5237          * themselves be part of a bundle. */
5238         bundle_set(ofport_->ofproto, ofport->bundle, NULL);
5239     }
5240
5241     ofport->realdev_ofp_port = realdev_ofp_port;
5242     ofport->vlandev_vid = vid;
5243
5244     if (realdev_ofp_port) {
5245         vsp_add(ofport, realdev_ofp_port, vid);
5246     }
5247
5248     return 0;
5249 }
5250
5251 static uint32_t
5252 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
5253 {
5254     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
5255 }
5256
5257 bool
5258 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
5259     OVS_EXCLUDED(ofproto->vsp_mutex)
5260 {
5261     /* hmap_is_empty is thread safe. */
5262     return !hmap_is_empty(&ofproto->realdev_vid_map);
5263 }
5264
5265
5266 static ofp_port_t
5267 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
5268                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5269     OVS_REQUIRES(ofproto->vsp_mutex)
5270 {
5271     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5272         int vid = vlan_tci_to_vid(vlan_tci);
5273         const struct vlan_splinter *vsp;
5274
5275         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5276                                  hash_realdev_vid(realdev_ofp_port, vid),
5277                                  &ofproto->realdev_vid_map) {
5278             if (vsp->realdev_ofp_port == realdev_ofp_port
5279                 && vsp->vid == vid) {
5280                 return vsp->vlandev_ofp_port;
5281             }
5282         }
5283     }
5284     return realdev_ofp_port;
5285 }
5286
5287 /* Returns the OFP port number of the Linux VLAN device that corresponds to
5288  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
5289  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
5290  * 'vlan_tci' 9, it would return the port number of eth0.9.
5291  *
5292  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
5293  * function just returns its 'realdev_ofp_port' argument. */
5294 ofp_port_t
5295 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5296                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5297     OVS_EXCLUDED(ofproto->vsp_mutex)
5298 {
5299     ofp_port_t ret;
5300
5301     /* hmap_is_empty is thread safe, see if we can return immediately. */
5302     if (hmap_is_empty(&ofproto->realdev_vid_map)) {
5303         return realdev_ofp_port;
5304     }
5305     ovs_mutex_lock(&ofproto->vsp_mutex);
5306     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
5307     ovs_mutex_unlock(&ofproto->vsp_mutex);
5308     return ret;
5309 }
5310
5311 static struct vlan_splinter *
5312 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
5313 {
5314     struct vlan_splinter *vsp;
5315
5316     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
5317                              hash_ofp_port(vlandev_ofp_port),
5318                              &ofproto->vlandev_map) {
5319         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5320             return vsp;
5321         }
5322     }
5323
5324     return NULL;
5325 }
5326
5327 /* Returns the OpenFlow port number of the "real" device underlying the Linux
5328  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
5329  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
5330  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
5331  * eth0 and store 9 in '*vid'.
5332  *
5333  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
5334  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
5335  * always does.*/
5336 static ofp_port_t
5337 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5338                        ofp_port_t vlandev_ofp_port, int *vid)
5339     OVS_REQUIRES(ofproto->vsp_mutex)
5340 {
5341     if (!hmap_is_empty(&ofproto->vlandev_map)) {
5342         const struct vlan_splinter *vsp;
5343
5344         vsp = vlandev_find(ofproto, vlandev_ofp_port);
5345         if (vsp) {
5346             if (vid) {
5347                 *vid = vsp->vid;
5348             }
5349             return vsp->realdev_ofp_port;
5350         }
5351     }
5352     return 0;
5353 }
5354
5355 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
5356  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
5357  * 'flow->in_port' to the "real" device backing the VLAN device, sets
5358  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Optionally pushes the
5359  * appropriate VLAN on 'packet' if provided.  Otherwise (which is always the
5360  * case unless VLAN splinters are enabled), returns false without making any
5361  * changes. */
5362 bool
5363 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow,
5364                 struct dp_packet *packet)
5365     OVS_EXCLUDED(ofproto->vsp_mutex)
5366 {
5367     ofp_port_t realdev;
5368     int vid;
5369
5370     /* hmap_is_empty is thread safe. */
5371     if (hmap_is_empty(&ofproto->vlandev_map)) {
5372         return false;
5373     }
5374
5375     ovs_mutex_lock(&ofproto->vsp_mutex);
5376     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
5377     ovs_mutex_unlock(&ofproto->vsp_mutex);
5378     if (!realdev) {
5379         return false;
5380     }
5381
5382     /* Cause the flow to be processed as if it came in on the real device with
5383      * the VLAN device's VLAN ID. */
5384     flow->in_port.ofp_port = realdev;
5385     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
5386
5387     if (packet) {
5388         /* Make the packet resemble the flow, so that it gets sent to an
5389          * OpenFlow controller properly, so that it looks correct for sFlow,
5390          * and so that flow_extract() will get the correct vlan_tci if it is
5391          * called on 'packet'. */
5392         eth_push_vlan(packet, htons(ETH_TYPE_VLAN), flow->vlan_tci);
5393     }
5394
5395     return true;
5396 }
5397
5398 static void
5399 vsp_remove(struct ofport_dpif *port)
5400 {
5401     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5402     struct vlan_splinter *vsp;
5403
5404     ovs_mutex_lock(&ofproto->vsp_mutex);
5405     vsp = vlandev_find(ofproto, port->up.ofp_port);
5406     if (vsp) {
5407         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5408         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
5409         free(vsp);
5410
5411         port->realdev_ofp_port = 0;
5412     } else {
5413         VLOG_ERR("missing vlan device record");
5414     }
5415     ovs_mutex_unlock(&ofproto->vsp_mutex);
5416 }
5417
5418 static void
5419 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
5420 {
5421     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5422
5423     ovs_mutex_lock(&ofproto->vsp_mutex);
5424     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
5425         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
5426             == realdev_ofp_port)) {
5427         struct vlan_splinter *vsp;
5428
5429         vsp = xmalloc(sizeof *vsp);
5430         vsp->realdev_ofp_port = realdev_ofp_port;
5431         vsp->vlandev_ofp_port = port->up.ofp_port;
5432         vsp->vid = vid;
5433
5434         port->realdev_ofp_port = realdev_ofp_port;
5435
5436         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
5437                     hash_ofp_port(port->up.ofp_port));
5438         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
5439                     hash_realdev_vid(realdev_ofp_port, vid));
5440     } else {
5441         VLOG_ERR("duplicate vlan device record");
5442     }
5443     ovs_mutex_unlock(&ofproto->vsp_mutex);
5444 }
5445
5446 static odp_port_t
5447 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
5448 {
5449     const struct ofport_dpif *ofport = ofp_port_to_ofport(ofproto, ofp_port);
5450     return ofport ? ofport->odp_port : ODPP_NONE;
5451 }
5452
5453 struct ofport_dpif *
5454 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
5455 {
5456     struct ofport_dpif *port;
5457
5458     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
5459     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
5460                              &backer->odp_to_ofport_map) {
5461         if (port->odp_port == odp_port) {
5462             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5463             return port;
5464         }
5465     }
5466
5467     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
5468     return NULL;
5469 }
5470
5471 static ofp_port_t
5472 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
5473 {
5474     struct ofport_dpif *port;
5475
5476     port = odp_port_to_ofport(ofproto->backer, odp_port);
5477     if (port && &ofproto->up == port->up.ofproto) {
5478         return port->up.ofp_port;
5479     } else {
5480         return OFPP_NONE;
5481     }
5482 }
5483
5484 int
5485 ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto,
5486                                const struct match *match, int priority,
5487                                uint16_t idle_timeout,
5488                                const struct ofpbuf *ofpacts,
5489                                struct rule **rulep)
5490 {
5491     struct ofproto_flow_mod ofm;
5492     struct rule_dpif *rule;
5493     int error;
5494
5495     ofm.fm.match = *match;
5496     ofm.fm.priority = priority;
5497     ofm.fm.new_cookie = htonll(0);
5498     ofm.fm.cookie = htonll(0);
5499     ofm.fm.cookie_mask = htonll(0);
5500     ofm.fm.modify_cookie = false;
5501     ofm.fm.table_id = TBL_INTERNAL;
5502     ofm.fm.command = OFPFC_ADD;
5503     ofm.fm.idle_timeout = idle_timeout;
5504     ofm.fm.hard_timeout = 0;
5505     ofm.fm.importance = 0;
5506     ofm.fm.buffer_id = 0;
5507     ofm.fm.out_port = 0;
5508     ofm.fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5509     ofm.fm.ofpacts = ofpacts->data;
5510     ofm.fm.ofpacts_len = ofpacts->size;
5511
5512     error = ofproto_flow_mod(&ofproto->up, &ofm);
5513     if (error) {
5514         VLOG_ERR_RL(&rl, "failed to add internal flow (%s)",
5515                     ofperr_to_string(error));
5516         *rulep = NULL;
5517         return error;
5518     }
5519
5520     rule = rule_dpif_lookup_in_table(ofproto,
5521                                      ofproto_dpif_get_tables_version(ofproto),
5522                                      TBL_INTERNAL, &ofm.fm.match.flow,
5523                                      &ofm.fm.match.wc, false);
5524     if (rule) {
5525         *rulep = &rule->up;
5526     } else {
5527         OVS_NOT_REACHED();
5528     }
5529     return 0;
5530 }
5531
5532 int
5533 ofproto_dpif_delete_internal_flow(struct ofproto_dpif *ofproto,
5534                                   struct match *match, int priority)
5535 {
5536     struct ofproto_flow_mod ofm;
5537     int error;
5538
5539     ofm.fm.match = *match;
5540     ofm.fm.priority = priority;
5541     ofm.fm.new_cookie = htonll(0);
5542     ofm.fm.cookie = htonll(0);
5543     ofm.fm.cookie_mask = htonll(0);
5544     ofm.fm.modify_cookie = false;
5545     ofm.fm.table_id = TBL_INTERNAL;
5546     ofm.fm.flags = OFPUTIL_FF_HIDDEN_FIELDS | OFPUTIL_FF_NO_READONLY;
5547     ofm.fm.command = OFPFC_DELETE_STRICT;
5548
5549     error = ofproto_flow_mod(&ofproto->up, &ofm);
5550     if (error) {
5551         VLOG_ERR_RL(&rl, "failed to delete internal flow (%s)",
5552                     ofperr_to_string(error));
5553         return error;
5554     }
5555
5556     return 0;
5557 }
5558
5559 const struct ofproto_class ofproto_dpif_class = {
5560     init,
5561     enumerate_types,
5562     enumerate_names,
5563     del,
5564     port_open_type,
5565     type_run,
5566     type_wait,
5567     alloc,
5568     construct,
5569     destruct,
5570     dealloc,
5571     run,
5572     wait,
5573     NULL,                       /* get_memory_usage. */
5574     type_get_memory_usage,
5575     flush,
5576     query_tables,
5577     set_tables_version,
5578     port_alloc,
5579     port_construct,
5580     port_destruct,
5581     port_dealloc,
5582     port_modified,
5583     port_reconfigured,
5584     port_query_by_name,
5585     port_add,
5586     port_del,
5587     port_get_stats,
5588     port_dump_start,
5589     port_dump_next,
5590     port_dump_done,
5591     port_poll,
5592     port_poll_wait,
5593     port_is_lacp_current,
5594     port_get_lacp_stats,
5595     NULL,                       /* rule_choose_table */
5596     rule_alloc,
5597     rule_construct,
5598     rule_insert,
5599     rule_delete,
5600     rule_destruct,
5601     rule_dealloc,
5602     rule_get_stats,
5603     rule_execute,
5604     set_frag_handling,
5605     packet_out,
5606     set_netflow,
5607     get_netflow_ids,
5608     set_sflow,
5609     set_ipfix,
5610     set_cfm,
5611     cfm_status_changed,
5612     get_cfm_status,
5613     set_lldp,
5614     get_lldp_status,
5615     set_aa,
5616     aa_mapping_set,
5617     aa_mapping_unset,
5618     aa_vlan_get_queued,
5619     aa_vlan_get_queue_size,
5620     set_bfd,
5621     bfd_status_changed,
5622     get_bfd_status,
5623     set_stp,
5624     get_stp_status,
5625     set_stp_port,
5626     get_stp_port_status,
5627     get_stp_port_stats,
5628     set_rstp,
5629     get_rstp_status,
5630     set_rstp_port,
5631     get_rstp_port_status,
5632     set_queues,
5633     bundle_set,
5634     bundle_remove,
5635     mirror_set__,
5636     mirror_get_stats__,
5637     set_flood_vlans,
5638     is_mirror_output_bundle,
5639     forward_bpdu_changed,
5640     set_mac_table_config,
5641     set_mcast_snooping,
5642     set_mcast_snooping_port,
5643     set_realdev,
5644     NULL,                       /* meter_get_features */
5645     NULL,                       /* meter_set */
5646     NULL,                       /* meter_get */
5647     NULL,                       /* meter_del */
5648     group_alloc,                /* group_alloc */
5649     group_construct,            /* group_construct */
5650     group_destruct,             /* group_destruct */
5651     group_dealloc,              /* group_dealloc */
5652     group_modify,               /* group_modify */
5653     group_get_stats,            /* group_get_stats */
5654     get_datapath_version,       /* get_datapath_version */
5655 };