ofproto/ofproto-dpif.c

   1 /*
   2  * Copyright (c) 2009, 2010, 2011, 2012, 2013 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 "connmgr.h"
  29 #include "coverage.h"
  30 #include "cfm.h"
  31 #include "dpif.h"
  32 #include "dynamic-string.h"
  33 #include "fail-open.h"
  34 #include "guarded-list.h"
  35 #include "hmapx.h"
  36 #include "lacp.h"
  37 #include "learn.h"
  38 #include "mac-learning.h"
  39 #include "meta-flow.h"
  40 #include "multipath.h"
  41 #include "netdev-vport.h"
  42 #include "netdev.h"
  43 #include "netlink.h"
  44 #include "nx-match.h"
  45 #include "odp-util.h"
  46 #include "odp-execute.h"
  47 #include "ofp-util.h"
  48 #include "ofpbuf.h"
  49 #include "ofp-actions.h"
  50 #include "ofp-parse.h"
  51 #include "ofp-print.h"
  52 #include "ofproto-dpif-governor.h"
  53 #include "ofproto-dpif-ipfix.h"
  54 #include "ofproto-dpif-mirror.h"
  55 #include "ofproto-dpif-monitor.h"
  56 #include "ofproto-dpif-sflow.h"
  57 #include "ofproto-dpif-upcall.h"
  58 #include "ofproto-dpif-xlate.h"
  59 #include "poll-loop.h"
  60 #include "simap.h"
  61 #include "smap.h"
  62 #include "timer.h"
  63 #include "tunnel.h"
  64 #include "unaligned.h"
  65 #include "unixctl.h"
  66 #include "vlan-bitmap.h"
  67 #include "vlog.h"
  68
  69 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  70
  71 COVERAGE_DEFINE(ofproto_dpif_expired);
  72 COVERAGE_DEFINE(facet_revalidate);
  73 COVERAGE_DEFINE(facet_unexpected);
  74 COVERAGE_DEFINE(facet_create);
  75 COVERAGE_DEFINE(facet_remove);
  76 COVERAGE_DEFINE(subfacet_create);
  77 COVERAGE_DEFINE(subfacet_destroy);
  78 COVERAGE_DEFINE(subfacet_install_fail);
  79 COVERAGE_DEFINE(packet_in_overflow);
  80
  81 /* Number of implemented OpenFlow tables. */
  82 enum { N_TABLES = 255 };
  83 enum { TBL_INTERNAL = N_TABLES - 1 };    /* Used for internal hidden rules. */
  84 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
  85
  86 struct flow_miss;
  87 struct facet;
  88
  89 struct rule_dpif {
  90     struct rule up;
  91
  92     /* These statistics:
  93      *
  94      *   - Do include packets and bytes from facets that have been deleted or
  95      *     whose own statistics have been folded into the rule.
  96      *
  97      *   - Do include packets and bytes sent "by hand" that were accounted to
  98      *     the rule without any facet being involved (this is a rare corner
  99      *     case in rule_execute()).
 100      *
 101      *   - Do not include packet or bytes that can be obtained from any facet's
 102      *     packet_count or byte_count member or that can be obtained from the
 103      *     datapath by, e.g., dpif_flow_get() for any subfacet.
 104      */
 105     struct ovs_mutex stats_mutex;
 106     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 107     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 108 };
 109
 110 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
 111 static struct rule_dpif *rule_dpif_cast(const struct rule *);
 112
 113 struct group_dpif {
 114     struct ofgroup up;
 115
 116     /* These statistics:
 117      *
 118      *   - Do include packets and bytes from facets that have been deleted or
 119      *     whose own statistics have been folded into the rule.
 120      *
 121      *   - Do include packets and bytes sent "by hand" that were accounted to
 122      *     the rule without any facet being involved (this is a rare corner
 123      *     case in rule_execute()).
 124      *
 125      *   - Do not include packet or bytes that can be obtained from any facet's
 126      *     packet_count or byte_count member or that can be obtained from the
 127      *     datapath by, e.g., dpif_flow_get() for any subfacet.
 128      */
 129     struct ovs_mutex stats_mutex;
 130     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 131     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 132     struct bucket_counter *bucket_stats OVS_GUARDED;  /* Bucket statistics. */
 133 };
 134
 135 struct ofbundle {
 136     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
 137     struct ofproto_dpif *ofproto; /* Owning ofproto. */
 138     void *aux;                  /* Key supplied by ofproto's client. */
 139     char *name;                 /* Identifier for log messages. */
 140
 141     /* Configuration. */
 142     struct list ports;          /* Contains "struct ofport"s. */
 143     enum port_vlan_mode vlan_mode; /* VLAN mode */
 144     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
 145     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
 146                                  * NULL if all VLANs are trunked. */
 147     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 148     struct bond *bond;          /* Nonnull iff more than one port. */
 149     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 150
 151     /* Status. */
 152     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 153 };
 154
 155 static void bundle_remove(struct ofport *);
 156 static void bundle_update(struct ofbundle *);
 157 static void bundle_destroy(struct ofbundle *);
 158 static void bundle_del_port(struct ofport_dpif *);
 159 static void bundle_run(struct ofbundle *);
 160 static void bundle_wait(struct ofbundle *);
 161
 162 static void stp_run(struct ofproto_dpif *ofproto);
 163 static void stp_wait(struct ofproto_dpif *ofproto);
 164 static int set_stp_port(struct ofport *,
 165                         const struct ofproto_port_stp_settings *);
 166
 167 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
 168                               enum slow_path_reason,
 169                               uint64_t *stub, size_t stub_size,
 170                               const struct nlattr **actionsp,
 171                               size_t *actions_lenp);
 172
 173 /* A subfacet (see "struct subfacet" below) has three possible installation
 174  * states:
 175  *
 176  *   - SF_NOT_INSTALLED: Not installed in the datapath.  This will only be the
 177  *     case just after the subfacet is created, just before the subfacet is
 178  *     destroyed, or if the datapath returns an error when we try to install a
 179  *     subfacet.
 180  *
 181  *   - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
 182  *
 183  *   - SF_SLOW_PATH: An action that sends every packet for the subfacet through
 184  *     ofproto_dpif is installed in the datapath.
 185  */
 186 enum subfacet_path {
 187     SF_NOT_INSTALLED,           /* No datapath flow for this subfacet. */
 188     SF_FAST_PATH,               /* Full actions are installed. */
 189     SF_SLOW_PATH,               /* Send-to-userspace action is installed. */
 190 };
 191
 192 /* A dpif flow and actions associated with a facet.
 193  *
 194  * See also the large comment on struct facet. */
 195 struct subfacet {
 196     /* Owners. */
 197     struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
 198     struct list list_node;      /* In struct facet's 'facets' list. */
 199     struct facet *facet;        /* Owning facet. */
 200     struct dpif_backer *backer; /* Owning backer. */
 201
 202     struct nlattr *key;
 203     int key_len;
 204
 205     long long int used;         /* Time last used; time created if not used. */
 206     long long int created;      /* Time created. */
 207
 208     uint64_t dp_packet_count;   /* Last known packet count in the datapath. */
 209     uint64_t dp_byte_count;     /* Last known byte count in the datapath. */
 210
 211     enum subfacet_path path;    /* Installed in datapath? */
 212 };
 213
 214 #define SUBFACET_DESTROY_MAX_BATCH 50
 215
 216 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *);
 217 static struct subfacet *subfacet_find(struct dpif_backer *,
 218                                       const struct nlattr *key, size_t key_len,
 219                                       uint32_t key_hash);
 220 static void subfacet_destroy(struct subfacet *);
 221 static void subfacet_destroy__(struct subfacet *);
 222 static void subfacet_destroy_batch(struct dpif_backer *,
 223                                    struct subfacet **, int n);
 224 static void subfacet_reset_dp_stats(struct subfacet *,
 225                                     struct dpif_flow_stats *);
 226 static void subfacet_update_stats(struct subfacet *,
 227                                   const struct dpif_flow_stats *);
 228 static int subfacet_install(struct subfacet *,
 229                             const struct ofpbuf *odp_actions,
 230                             struct dpif_flow_stats *);
 231 static void subfacet_uninstall(struct subfacet *);
 232
 233 /* A unique, non-overlapping instantiation of an OpenFlow flow.
 234  *
 235  * A facet associates a "struct flow", which represents the Open vSwitch
 236  * userspace idea of an exact-match flow, with one or more subfacets.
 237  * While the facet is created based on an exact-match flow, it is stored
 238  * within the ofproto based on the wildcards that could be expressed
 239  * based on the flow table and other configuration.  (See the 'wc'
 240  * description in "struct xlate_out" for more details.)
 241  *
 242  * Each subfacet tracks the datapath's idea of the flow equivalent to
 243  * the facet.  When the kernel module (or other dpif implementation) and
 244  * Open vSwitch userspace agree on the definition of a flow key, there
 245  * is exactly one subfacet per facet.  If the dpif implementation
 246  * supports more-specific flow matching than userspace, however, a facet
 247  * can have more than one subfacet.  Examples include the dpif
 248  * implementation not supporting the same wildcards as userspace or some
 249  * distinction in flow that userspace simply doesn't understand.
 250  *
 251  * Flow expiration works in terms of subfacets, so a facet must have at
 252  * least one subfacet or it will never expire, leaking memory. */
 253 struct facet {
 254     /* Owner. */
 255     struct ofproto_dpif *ofproto;
 256
 257     /* Owned data. */
 258     struct list subfacets;
 259     long long int used;         /* Time last used; time created if not used. */
 260
 261     /* Key. */
 262     struct flow flow;           /* Flow of the creating subfacet. */
 263     struct cls_rule cr;         /* In 'ofproto_dpif's facets classifier. */
 264
 265     /* These statistics:
 266      *
 267      *   - Do include packets and bytes sent "by hand", e.g. with
 268      *     dpif_execute().
 269      *
 270      *   - Do include packets and bytes that were obtained from the datapath
 271      *     when a subfacet's statistics were reset (e.g. dpif_flow_put() with
 272      *     DPIF_FP_ZERO_STATS).
 273      *
 274      *   - Do not include packets or bytes that can be obtained from the
 275      *     datapath for any existing subfacet.
 276      */
 277     uint64_t packet_count;       /* Number of packets received. */
 278     uint64_t byte_count;         /* Number of bytes received. */
 279
 280     /* Resubmit statistics. */
 281     uint64_t prev_packet_count;  /* Number of packets from last stats push. */
 282     uint64_t prev_byte_count;    /* Number of bytes from last stats push. */
 283     long long int prev_used;     /* Used time from last stats push. */
 284
 285     /* Accounting. */
 286     uint64_t accounted_bytes;    /* Bytes processed by facet_account(). */
 287     struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
 288     uint16_t tcp_flags;          /* TCP flags seen for this 'rule'. */
 289
 290     struct xlate_out xout;
 291
 292     /* Storage for a single subfacet, to reduce malloc() time and space
 293      * overhead.  (A facet always has at least one subfacet and in the common
 294      * case has exactly one subfacet.  However, 'one_subfacet' may not
 295      * always be valid, since it could have been removed after newer
 296      * subfacets were pushed onto the 'subfacets' list.) */
 297     struct subfacet one_subfacet;
 298
 299     long long int learn_rl;      /* Rate limiter for facet_learn(). */
 300 };
 301
 302 static struct facet *facet_create(const struct flow_miss *);
 303 static void facet_remove(struct facet *);
 304 static void facet_free(struct facet *);
 305
 306 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
 307 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
 308                                         const struct flow *);
 309 static bool facet_revalidate(struct facet *);
 310 static bool facet_check_consistency(struct facet *);
 311
 312 static void facet_flush_stats(struct facet *);
 313
 314 static void facet_reset_counters(struct facet *);
 315 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
 316                             struct dpif_flow_stats *, bool may_learn);
 317 static void facet_push_stats(struct facet *, bool may_learn);
 318 static void facet_learn(struct facet *);
 319 static void facet_account(struct facet *);
 320 static void push_all_stats(void);
 321
 322 static bool facet_is_controller_flow(struct facet *);
 323
 324 struct ofport_dpif {
 325     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 326     struct ofport up;
 327
 328     odp_port_t odp_port;
 329     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 330     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
 331     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 332     struct bfd *bfd;            /* BFD, if any. */
 333     bool may_enable;            /* May be enabled in bonds. */
 334     bool is_tunnel;             /* This port is a tunnel. */
 335     long long int carrier_seq;  /* Carrier status changes. */
 336     struct ofport_dpif *peer;   /* Peer if patch port. */
 337
 338     /* Spanning tree. */
 339     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 340     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 341     long long int stp_state_entered;
 342
 343     /* Queue to DSCP mapping. */
 344     struct ofproto_port_queue *qdscp;
 345     size_t n_qdscp;
 346
 347     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 348      *
 349      * This is deprecated.  It is only for compatibility with broken device
 350      * drivers in old versions of Linux that do not properly support VLANs when
 351      * VLAN devices are not used.  When broken device drivers are no longer in
 352      * widespread use, we will delete these interfaces. */
 353     ofp_port_t realdev_ofp_port;
 354     int vlandev_vid;
 355 };
 356
 357 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 358  *
 359  * This is deprecated.  It is only for compatibility with broken device drivers
 360  * in old versions of Linux that do not properly support VLANs when VLAN
 361  * devices are not used.  When broken device drivers are no longer in
 362  * widespread use, we will delete these interfaces. */
 363 struct vlan_splinter {
 364     struct hmap_node realdev_vid_node;
 365     struct hmap_node vlandev_node;
 366     ofp_port_t realdev_ofp_port;
 367     ofp_port_t vlandev_ofp_port;
 368     int vid;
 369 };
 370
 371 static void vsp_remove(struct ofport_dpif *);
 372 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 373
 374 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 375                                        ofp_port_t);
 376
 377 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 378                                        odp_port_t);
 379
 380 static struct ofport_dpif *
 381 ofport_dpif_cast(const struct ofport *ofport)
 382 {
 383     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 384 }
 385
 386 static void port_run(struct ofport_dpif *);
 387 static int set_bfd(struct ofport *, const struct smap *);
 388 static int set_cfm(struct ofport *, const struct cfm_settings *);
 389 static void ofport_update_peer(struct ofport_dpif *);
 390 static void run_fast_rl(void);
 391 static int run_fast(struct ofproto *);
 392
 393 struct dpif_completion {
 394     struct list list_node;
 395     struct ofoperation *op;
 396 };
 397
 398 /* Reasons that we might need to revalidate every facet, and corresponding
 399  * coverage counters.
 400  *
 401  * A value of 0 means that there is no need to revalidate.
 402  *
 403  * It would be nice to have some cleaner way to integrate with coverage
 404  * counters, but with only a few reasons I guess this is good enough for
 405  * now. */
 406 enum revalidate_reason {
 407     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 408     REV_STP,                   /* Spanning tree protocol port status change. */
 409     REV_BOND,                  /* Bonding changed. */
 410     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 411     REV_FLOW_TABLE,            /* Flow table changed. */
 412     REV_MAC_LEARNING,          /* Mac learning changed. */
 413     REV_INCONSISTENCY          /* Facet self-check failed. */
 414 };
 415 COVERAGE_DEFINE(rev_reconfigure);
 416 COVERAGE_DEFINE(rev_stp);
 417 COVERAGE_DEFINE(rev_bond);
 418 COVERAGE_DEFINE(rev_port_toggled);
 419 COVERAGE_DEFINE(rev_flow_table);
 420 COVERAGE_DEFINE(rev_mac_learning);
 421 COVERAGE_DEFINE(rev_inconsistency);
 422
 423 struct avg_subfacet_rates {
 424     double add_rate;   /* Moving average of new flows created per minute. */
 425     double del_rate;   /* Moving average of flows deleted per minute. */
 426 };
 427
 428 /* All datapaths of a given type share a single dpif backer instance. */
 429 struct dpif_backer {
 430     char *type;
 431     int refcount;
 432     struct dpif *dpif;
 433     struct udpif *udpif;
 434     struct timer next_expiration;
 435
 436     struct ovs_rwlock odp_to_ofport_lock;
 437     struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
 438
 439     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 440
 441     /* Facet revalidation flags applying to facets which use this backer. */
 442     enum revalidate_reason need_revalidate; /* Revalidate every facet. */
 443
 444     struct hmap drop_keys; /* Set of dropped odp keys. */
 445     bool recv_set_enable; /* Enables or disables receiving packets. */
 446
 447     struct hmap subfacets;
 448     struct governor *governor;
 449
 450     /* Subfacet statistics.
 451      *
 452      * These keep track of the total number of subfacets added and deleted and
 453      * flow life span.  They are useful for computing the flow rates stats
 454      * exposed via "ovs-appctl dpif/show".  The goal is to learn about
 455      * traffic patterns in ways that we can use later to improve Open vSwitch
 456      * performance in new situations.  */
 457     long long int created;           /* Time when it is created. */
 458     unsigned max_n_subfacet;         /* Maximum number of flows */
 459     unsigned avg_n_subfacet;         /* Average number of flows. */
 460     long long int avg_subfacet_life; /* Average life span of subfacets. */
 461
 462     /* Number of upcall handling threads. */
 463     unsigned int n_handler_threads;
 464 };
 465
 466 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 467 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 468
 469 static void drop_key_clear(struct dpif_backer *);
 470
 471 struct ofproto_dpif {
 472     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 473     struct ofproto up;
 474     struct dpif_backer *backer;
 475
 476     /* Special OpenFlow rules. */
 477     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 478     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 479     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 480
 481     /* Bridging. */
 482     struct netflow *netflow;
 483     struct dpif_sflow *sflow;
 484     struct dpif_ipfix *ipfix;
 485     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 486     struct mac_learning *ml;
 487     bool has_bonded_bundles;
 488     struct mbridge *mbridge;
 489
 490     /* Facets. */
 491     struct classifier facets;     /* Contains 'struct facet's. */
 492     long long int consistency_rl;
 493
 494     struct ovs_mutex stats_mutex;
 495     struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
 496                                             * consumed in userspace. */
 497
 498     /* Spanning tree. */
 499     struct stp *stp;
 500     long long int stp_last_tick;
 501
 502     /* VLAN splinters. */
 503     struct ovs_mutex vsp_mutex;
 504     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 505     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 506
 507     /* Ports. */
 508     struct sset ports;             /* Set of standard port names. */
 509     struct sset ghost_ports;       /* Ports with no datapath port. */
 510     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 511     int port_poll_errno;           /* Last errno for port_poll() reply. */
 512
 513     /* Per ofproto's dpif stats. */
 514     uint64_t n_hit;
 515     uint64_t n_missed;
 516
 517     /* Work queues. */
 518     struct guarded_list pins;      /* Contains "struct ofputil_packet_in"s. */
 519 };
 520
 521 /* By default, flows in the datapath are wildcarded (megaflows).  They
 522  * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
 523 static bool enable_megaflows = true;
 524
 525 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 526 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 527
 528 static void ofproto_dpif_unixctl_init(void);
 529
 530 static inline struct ofproto_dpif *
 531 ofproto_dpif_cast(const struct ofproto *ofproto)
 532 {
 533     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 534     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 535 }
 536
 537 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
 538                                         ofp_port_t ofp_port);
 539 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
 540                           const struct ofpbuf *packet, struct ds *);
 541
 542 /* Upcalls. */
 543 static void handle_upcalls(struct dpif_backer *);
 544
 545 /* Flow expiration. */
 546 static int expire(struct dpif_backer *);
 547
 548 /* NetFlow. */
 549 static void send_netflow_active_timeouts(struct ofproto_dpif *);
 550
 551 /* Global variables. */
 552 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 553
 554 /* Initial mappings of port to bridge mappings. */
 555 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 556
 557 /* Executes 'fm'.  The caller retains ownership of 'fm' and everything in
 558  * it. */
 559 void
 560 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 561                       struct ofputil_flow_mod *fm)
 562 {
 563     ofproto_flow_mod(&ofproto->up, fm);
 564 }
 565
 566 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 567  * Takes ownership of 'pin' and pin->packet. */
 568 void
 569 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 570                             struct ofproto_packet_in *pin)
 571 {
 572     if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
 573         COVERAGE_INC(packet_in_overflow);
 574         free(CONST_CAST(void *, pin->up.packet));
 575         free(pin);
 576     }
 577 }
 578 \f
 579 /* Factory functions. */
 580
 581 static void
 582 init(const struct shash *iface_hints)
 583 {
 584     struct shash_node *node;
 585
 586     /* Make a local copy, since we don't own 'iface_hints' elements. */
 587     SHASH_FOR_EACH(node, iface_hints) {
 588         const struct iface_hint *orig_hint = node->data;
 589         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 590
 591         new_hint->br_name = xstrdup(orig_hint->br_name);
 592         new_hint->br_type = xstrdup(orig_hint->br_type);
 593         new_hint->ofp_port = orig_hint->ofp_port;
 594
 595         shash_add(&init_ofp_ports, node->name, new_hint);
 596     }
 597 }
 598
 599 static void
 600 enumerate_types(struct sset *types)
 601 {
 602     dp_enumerate_types(types);
 603 }
 604
 605 static int
 606 enumerate_names(const char *type, struct sset *names)
 607 {
 608     struct ofproto_dpif *ofproto;
 609
 610     sset_clear(names);
 611     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 612         if (strcmp(type, ofproto->up.type)) {
 613             continue;
 614         }
 615         sset_add(names, ofproto->up.name);
 616     }
 617
 618     return 0;
 619 }
 620
 621 static int
 622 del(const char *type, const char *name)
 623 {
 624     struct dpif *dpif;
 625     int error;
 626
 627     error = dpif_open(name, type, &dpif);
 628     if (!error) {
 629         error = dpif_delete(dpif);
 630         dpif_close(dpif);
 631     }
 632     return error;
 633 }
 634 \f
 635 static const char *
 636 port_open_type(const char *datapath_type, const char *port_type)
 637 {
 638     return dpif_port_open_type(datapath_type, port_type);
 639 }
 640
 641 /* Type functions. */
 642
 643 static void process_dpif_port_changes(struct dpif_backer *);
 644 static void process_dpif_all_ports_changed(struct dpif_backer *);
 645 static void process_dpif_port_change(struct dpif_backer *,
 646                                      const char *devname);
 647 static void process_dpif_port_error(struct dpif_backer *, int error);
 648
 649 static struct ofproto_dpif *
 650 lookup_ofproto_dpif_by_port_name(const char *name)
 651 {
 652     struct ofproto_dpif *ofproto;
 653
 654     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 655         if (sset_contains(&ofproto->ports, name)) {
 656             return ofproto;
 657         }
 658     }
 659
 660     return NULL;
 661 }
 662
 663 static int
 664 type_run(const char *type)
 665 {
 666     static long long int push_timer = LLONG_MIN;
 667     struct dpif_backer *backer;
 668
 669     backer = shash_find_data(&all_dpif_backers, type);
 670     if (!backer) {
 671         /* This is not necessarily a problem, since backers are only
 672          * created on demand. */
 673         return 0;
 674     }
 675
 676     dpif_run(backer->dpif);
 677
 678     /* The most natural place to push facet statistics is when they're pulled
 679      * from the datapath.  However, when there are many flows in the datapath,
 680      * this expensive operation can occur so frequently, that it reduces our
 681      * ability to quickly set up flows.  To reduce the cost, we push statistics
 682      * here instead. */
 683     if (time_msec() > push_timer) {
 684         push_timer = time_msec() + 2000;
 685         push_all_stats();
 686     }
 687
 688     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 689      * and the configuration has now changed to "false", enable receiving
 690      * packets from the datapath. */
 691     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 692         int error;
 693
 694         backer->recv_set_enable = true;
 695
 696         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 697         if (error) {
 698             udpif_recv_set(backer->udpif, 0, false);
 699             VLOG_ERR("Failed to enable receiving packets in dpif.");
 700             return error;
 701         }
 702         udpif_recv_set(backer->udpif, n_handler_threads,
 703                        backer->recv_set_enable);
 704         dpif_flow_flush(backer->dpif);
 705         backer->need_revalidate = REV_RECONFIGURE;
 706     }
 707
 708     /* If the n_handler_threads is reconfigured, call udpif_recv_set()
 709      * to reset the handler threads. */
 710     if (backer->n_handler_threads != n_handler_threads) {
 711         udpif_recv_set(backer->udpif, n_handler_threads,
 712                        backer->recv_set_enable);
 713         backer->n_handler_threads = n_handler_threads;
 714     }
 715
 716     if (backer->need_revalidate) {
 717         struct ofproto_dpif *ofproto;
 718         struct simap_node *node;
 719         struct simap tmp_backers;
 720
 721         /* Handle tunnel garbage collection. */
 722         simap_init(&tmp_backers);
 723         simap_swap(&backer->tnl_backers, &tmp_backers);
 724
 725         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 726             struct ofport_dpif *iter;
 727
 728             if (backer != ofproto->backer) {
 729                 continue;
 730             }
 731
 732             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 733                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 734                 const char *dp_port;
 735
 736                 if (!iter->is_tunnel) {
 737                     continue;
 738                 }
 739
 740                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 741                                                      namebuf, sizeof namebuf);
 742                 node = simap_find(&tmp_backers, dp_port);
 743                 if (node) {
 744                     simap_put(&backer->tnl_backers, dp_port, node->data);
 745                     simap_delete(&tmp_backers, node);
 746                     node = simap_find(&backer->tnl_backers, dp_port);
 747                 } else {
 748                     node = simap_find(&backer->tnl_backers, dp_port);
 749                     if (!node) {
 750                         odp_port_t odp_port = ODPP_NONE;
 751
 752                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 753                                            &odp_port)) {
 754                             simap_put(&backer->tnl_backers, dp_port,
 755                                       odp_to_u32(odp_port));
 756                             node = simap_find(&backer->tnl_backers, dp_port);
 757                         }
 758                     }
 759                 }
 760
 761                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 762                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 763                                          iter->odp_port)) {
 764                     backer->need_revalidate = REV_RECONFIGURE;
 765                 }
 766             }
 767         }
 768
 769         SIMAP_FOR_EACH (node, &tmp_backers) {
 770             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 771         }
 772         simap_destroy(&tmp_backers);
 773
 774         switch (backer->need_revalidate) {
 775         case REV_RECONFIGURE:   COVERAGE_INC(rev_reconfigure);   break;
 776         case REV_STP:           COVERAGE_INC(rev_stp);           break;
 777         case REV_BOND:          COVERAGE_INC(rev_bond);          break;
 778         case REV_PORT_TOGGLED:  COVERAGE_INC(rev_port_toggled);  break;
 779         case REV_FLOW_TABLE:    COVERAGE_INC(rev_flow_table);    break;
 780         case REV_MAC_LEARNING:  COVERAGE_INC(rev_mac_learning);  break;
 781         case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
 782         }
 783         backer->need_revalidate = 0;
 784
 785         /* Clear the drop_keys in case we should now be accepting some
 786          * formerly dropped flows. */
 787         drop_key_clear(backer);
 788
 789         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 790             struct facet *facet, *next;
 791             struct ofport_dpif *ofport;
 792             struct cls_cursor cursor;
 793             struct ofbundle *bundle;
 794
 795             if (ofproto->backer != backer) {
 796                 continue;
 797             }
 798
 799             ovs_rwlock_wrlock(&xlate_rwlock);
 800             xlate_ofproto_set(ofproto, ofproto->up.name,
 801                               ofproto->backer->dpif, ofproto->miss_rule,
 802                               ofproto->no_packet_in_rule, ofproto->ml,
 803                               ofproto->stp, ofproto->mbridge,
 804                               ofproto->sflow, ofproto->ipfix,
 805                               ofproto->up.frag_handling,
 806                               ofproto->up.forward_bpdu,
 807                               connmgr_has_in_band(ofproto->up.connmgr),
 808                               ofproto->netflow != NULL);
 809
 810             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 811                 xlate_bundle_set(ofproto, bundle, bundle->name,
 812                                  bundle->vlan_mode, bundle->vlan,
 813                                  bundle->trunks, bundle->use_priority_tags,
 814                                  bundle->bond, bundle->lacp,
 815                                  bundle->floodable);
 816             }
 817
 818             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 819                 int stp_port = ofport->stp_port
 820                     ? stp_port_no(ofport->stp_port)
 821                     : -1;
 822                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 823                                  ofport->up.ofp_port, ofport->odp_port,
 824                                  ofport->up.netdev, ofport->cfm,
 825                                  ofport->bfd, ofport->peer, stp_port,
 826                                  ofport->qdscp, ofport->n_qdscp,
 827                                  ofport->up.pp.config, ofport->is_tunnel,
 828                                  ofport->may_enable);
 829             }
 830             ovs_rwlock_unlock(&xlate_rwlock);
 831
 832             /* Only ofproto-dpif cares about the facet classifier so we just
 833              * lock cls_cursor_init() to appease the thread safety analysis. */
 834             ovs_rwlock_rdlock(&ofproto->facets.rwlock);
 835             cls_cursor_init(&cursor, &ofproto->facets, NULL);
 836             ovs_rwlock_unlock(&ofproto->facets.rwlock);
 837             CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
 838                 facet_revalidate(facet);
 839                 run_fast_rl();
 840             }
 841         }
 842
 843         udpif_revalidate(backer->udpif);
 844     }
 845
 846     if (!backer->recv_set_enable) {
 847         /* Wake up before a max of 1000ms. */
 848         timer_set_duration(&backer->next_expiration, 1000);
 849     } else if (timer_expired(&backer->next_expiration)) {
 850         int delay = expire(backer);
 851         timer_set_duration(&backer->next_expiration, delay);
 852     }
 853
 854     process_dpif_port_changes(backer);
 855
 856     if (backer->governor) {
 857         size_t n_subfacets;
 858
 859         governor_run(backer->governor);
 860
 861         /* If the governor has shrunk to its minimum size and the number of
 862          * subfacets has dwindled, then drop the governor entirely.
 863          *
 864          * For hysteresis, the number of subfacets to drop the governor is
 865          * smaller than the number needed to trigger its creation. */
 866         n_subfacets = hmap_count(&backer->subfacets);
 867         if (n_subfacets * 4 < flow_eviction_threshold
 868             && governor_is_idle(backer->governor)) {
 869             governor_destroy(backer->governor);
 870             backer->governor = NULL;
 871         }
 872     }
 873
 874     return 0;
 875 }
 876
 877 /* Check for and handle port changes in 'backer''s dpif. */
 878 static void
 879 process_dpif_port_changes(struct dpif_backer *backer)
 880 {
 881     for (;;) {
 882         char *devname;
 883         int error;
 884
 885         error = dpif_port_poll(backer->dpif, &devname);
 886         switch (error) {
 887         case EAGAIN:
 888             return;
 889
 890         case ENOBUFS:
 891             process_dpif_all_ports_changed(backer);
 892             break;
 893
 894         case 0:
 895             process_dpif_port_change(backer, devname);
 896             free(devname);
 897             break;
 898
 899         default:
 900             process_dpif_port_error(backer, error);
 901             break;
 902         }
 903     }
 904 }
 905
 906 static void
 907 process_dpif_all_ports_changed(struct dpif_backer *backer)
 908 {
 909     struct ofproto_dpif *ofproto;
 910     struct dpif_port dpif_port;
 911     struct dpif_port_dump dump;
 912     struct sset devnames;
 913     const char *devname;
 914
 915     sset_init(&devnames);
 916     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 917         if (ofproto->backer == backer) {
 918             struct ofport *ofport;
 919
 920             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 921                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 922             }
 923         }
 924     }
 925     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 926         sset_add(&devnames, dpif_port.name);
 927     }
 928
 929     SSET_FOR_EACH (devname, &devnames) {
 930         process_dpif_port_change(backer, devname);
 931     }
 932     sset_destroy(&devnames);
 933 }
 934
 935 static void
 936 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 937 {
 938     struct ofproto_dpif *ofproto;
 939     struct dpif_port port;
 940
 941     /* Don't report on the datapath's device. */
 942     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 943         return;
 944     }
 945
 946     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 947                    &all_ofproto_dpifs) {
 948         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 949             return;
 950         }
 951     }
 952
 953     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 954     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 955         /* The port was removed.  If we know the datapath,
 956          * report it through poll_set().  If we don't, it may be
 957          * notifying us of a removal we initiated, so ignore it.
 958          * If there's a pending ENOBUFS, let it stand, since
 959          * everything will be reevaluated. */
 960         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 961             sset_add(&ofproto->port_poll_set, devname);
 962             ofproto->port_poll_errno = 0;
 963         }
 964     } else if (!ofproto) {
 965         /* The port was added, but we don't know with which
 966          * ofproto we should associate it.  Delete it. */
 967         dpif_port_del(backer->dpif, port.port_no);
 968     } else {
 969         struct ofport_dpif *ofport;
 970
 971         ofport = ofport_dpif_cast(shash_find_data(
 972                                       &ofproto->up.port_by_name, devname));
 973         if (ofport
 974             && ofport->odp_port != port.port_no
 975             && !odp_port_to_ofport(backer, port.port_no))
 976         {
 977             /* 'ofport''s datapath port number has changed from
 978              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 979              * structures to match. */
 980             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 981             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 982             ofport->odp_port = port.port_no;
 983             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 984                         hash_odp_port(port.port_no));
 985             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 986             backer->need_revalidate = REV_RECONFIGURE;
 987         }
 988     }
 989     dpif_port_destroy(&port);
 990 }
 991
 992 /* Propagate 'error' to all ofprotos based on 'backer'. */
 993 static void
 994 process_dpif_port_error(struct dpif_backer *backer, int error)
 995 {
 996     struct ofproto_dpif *ofproto;
 997
 998     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 999         if (ofproto->backer == backer) {
1000             sset_clear(&ofproto->port_poll_set);
1001             ofproto->port_poll_errno = error;
1002         }
1003     }
1004 }
1005
1006 static int
1007 dpif_backer_run_fast(struct dpif_backer *backer)
1008 {
1009     handle_upcalls(backer);
1010
1011     return 0;
1012 }
1013
1014 static int
1015 type_run_fast(const char *type)
1016 {
1017     struct dpif_backer *backer;
1018
1019     backer = shash_find_data(&all_dpif_backers, type);
1020     if (!backer) {
1021         /* This is not necessarily a problem, since backers are only
1022          * created on demand. */
1023         return 0;
1024     }
1025
1026     return dpif_backer_run_fast(backer);
1027 }
1028
1029 static void
1030 run_fast_rl(void)
1031 {
1032     static long long int port_rl = LLONG_MIN;
1033
1034     if (time_msec() >= port_rl) {
1035         struct ofproto_dpif *ofproto;
1036
1037         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1038             run_fast(&ofproto->up);
1039         }
1040         port_rl = time_msec() + 200;
1041     }
1042 }
1043
1044 static void
1045 type_wait(const char *type)
1046 {
1047     struct dpif_backer *backer;
1048
1049     backer = shash_find_data(&all_dpif_backers, type);
1050     if (!backer) {
1051         /* This is not necessarily a problem, since backers are only
1052          * created on demand. */
1053         return;
1054     }
1055
1056     if (backer->governor) {
1057         governor_wait(backer->governor);
1058     }
1059
1060     timer_wait(&backer->next_expiration);
1061     dpif_wait(backer->dpif);
1062     udpif_wait(backer->udpif);
1063 }
1064 \f
1065 /* Basic life-cycle. */
1066
1067 static int add_internal_flows(struct ofproto_dpif *);
1068
1069 static struct ofproto *
1070 alloc(void)
1071 {
1072     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1073     return &ofproto->up;
1074 }
1075
1076 static void
1077 dealloc(struct ofproto *ofproto_)
1078 {
1079     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1080     free(ofproto);
1081 }
1082
1083 static void
1084 close_dpif_backer(struct dpif_backer *backer)
1085 {
1086     ovs_assert(backer->refcount > 0);
1087
1088     if (--backer->refcount) {
1089         return;
1090     }
1091
1092     drop_key_clear(backer);
1093     hmap_destroy(&backer->drop_keys);
1094
1095     udpif_destroy(backer->udpif);
1096
1097     simap_destroy(&backer->tnl_backers);
1098     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1099     hmap_destroy(&backer->odp_to_ofport_map);
1100     shash_find_and_delete(&all_dpif_backers, backer->type);
1101     free(backer->type);
1102     dpif_close(backer->dpif);
1103
1104     ovs_assert(hmap_is_empty(&backer->subfacets));
1105     hmap_destroy(&backer->subfacets);
1106     governor_destroy(backer->governor);
1107
1108     free(backer);
1109 }
1110
1111 /* Datapath port slated for removal from datapath. */
1112 struct odp_garbage {
1113     struct list list_node;
1114     odp_port_t odp_port;
1115 };
1116
1117 static int
1118 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1119 {
1120     struct dpif_backer *backer;
1121     struct dpif_port_dump port_dump;
1122     struct dpif_port port;
1123     struct shash_node *node;
1124     struct list garbage_list;
1125     struct odp_garbage *garbage, *next;
1126     struct sset names;
1127     char *backer_name;
1128     const char *name;
1129     int error;
1130
1131     backer = shash_find_data(&all_dpif_backers, type);
1132     if (backer) {
1133         backer->refcount++;
1134         *backerp = backer;
1135         return 0;
1136     }
1137
1138     backer_name = xasprintf("ovs-%s", type);
1139
1140     /* Remove any existing datapaths, since we assume we're the only
1141      * userspace controlling the datapath. */
1142     sset_init(&names);
1143     dp_enumerate_names(type, &names);
1144     SSET_FOR_EACH(name, &names) {
1145         struct dpif *old_dpif;
1146
1147         /* Don't remove our backer if it exists. */
1148         if (!strcmp(name, backer_name)) {
1149             continue;
1150         }
1151
1152         if (dpif_open(name, type, &old_dpif)) {
1153             VLOG_WARN("couldn't open old datapath %s to remove it", name);
1154         } else {
1155             dpif_delete(old_dpif);
1156             dpif_close(old_dpif);
1157         }
1158     }
1159     sset_destroy(&names);
1160
1161     backer = xmalloc(sizeof *backer);
1162
1163     error = dpif_create_and_open(backer_name, type, &backer->dpif);
1164     free(backer_name);
1165     if (error) {
1166         VLOG_ERR("failed to open datapath of type %s: %s", type,
1167                  ovs_strerror(error));
1168         free(backer);
1169         return error;
1170     }
1171     backer->udpif = udpif_create(backer, backer->dpif);
1172
1173     backer->type = xstrdup(type);
1174     backer->governor = NULL;
1175     backer->refcount = 1;
1176     hmap_init(&backer->odp_to_ofport_map);
1177     ovs_rwlock_init(&backer->odp_to_ofport_lock);
1178     hmap_init(&backer->drop_keys);
1179     hmap_init(&backer->subfacets);
1180     timer_set_duration(&backer->next_expiration, 1000);
1181     backer->need_revalidate = 0;
1182     simap_init(&backer->tnl_backers);
1183     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1184     *backerp = backer;
1185
1186     if (backer->recv_set_enable) {
1187         dpif_flow_flush(backer->dpif);
1188     }
1189
1190     /* Loop through the ports already on the datapath and remove any
1191      * that we don't need anymore. */
1192     list_init(&garbage_list);
1193     dpif_port_dump_start(&port_dump, backer->dpif);
1194     while (dpif_port_dump_next(&port_dump, &port)) {
1195         node = shash_find(&init_ofp_ports, port.name);
1196         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1197             garbage = xmalloc(sizeof *garbage);
1198             garbage->odp_port = port.port_no;
1199             list_push_front(&garbage_list, &garbage->list_node);
1200         }
1201     }
1202     dpif_port_dump_done(&port_dump);
1203
1204     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1205         dpif_port_del(backer->dpif, garbage->odp_port);
1206         list_remove(&garbage->list_node);
1207         free(garbage);
1208     }
1209
1210     shash_add(&all_dpif_backers, type, backer);
1211
1212     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1213     if (error) {
1214         VLOG_ERR("failed to listen on datapath of type %s: %s",
1215                  type, ovs_strerror(error));
1216         close_dpif_backer(backer);
1217         return error;
1218     }
1219     udpif_recv_set(backer->udpif, n_handler_threads,
1220                    backer->recv_set_enable);
1221     backer->n_handler_threads = n_handler_threads;
1222
1223     backer->max_n_subfacet = 0;
1224     backer->created = time_msec();
1225     backer->avg_n_subfacet = 0;
1226     backer->avg_subfacet_life = 0;
1227
1228     return error;
1229 }
1230
1231 static int
1232 construct(struct ofproto *ofproto_)
1233 {
1234     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1235     struct shash_node *node, *next;
1236     uint32_t max_ports;
1237     int error;
1238
1239     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1240     if (error) {
1241         return error;
1242     }
1243
1244     max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1245     ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX)));
1246
1247     ofproto->netflow = NULL;
1248     ofproto->sflow = NULL;
1249     ofproto->ipfix = NULL;
1250     ofproto->stp = NULL;
1251     hmap_init(&ofproto->bundles);
1252     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1253     ofproto->mbridge = mbridge_create();
1254     ofproto->has_bonded_bundles = false;
1255     ovs_mutex_init(&ofproto->stats_mutex);
1256     ovs_mutex_init(&ofproto->vsp_mutex);
1257
1258     classifier_init(&ofproto->facets);
1259     ofproto->consistency_rl = LLONG_MIN;
1260
1261     guarded_list_init(&ofproto->pins);
1262
1263     ofproto_dpif_unixctl_init();
1264
1265     hmap_init(&ofproto->vlandev_map);
1266     hmap_init(&ofproto->realdev_vid_map);
1267
1268     sset_init(&ofproto->ports);
1269     sset_init(&ofproto->ghost_ports);
1270     sset_init(&ofproto->port_poll_set);
1271     ofproto->port_poll_errno = 0;
1272
1273     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1274         struct iface_hint *iface_hint = node->data;
1275
1276         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1277             /* Check if the datapath already has this port. */
1278             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1279                 sset_add(&ofproto->ports, node->name);
1280             }
1281
1282             free(iface_hint->br_name);
1283             free(iface_hint->br_type);
1284             free(iface_hint);
1285             shash_delete(&init_ofp_ports, node);
1286         }
1287     }
1288
1289     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1290                 hash_string(ofproto->up.name, 0));
1291     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1292
1293     ofproto_init_tables(ofproto_, N_TABLES);
1294     error = add_internal_flows(ofproto);
1295     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1296
1297     ofproto->n_hit = 0;
1298     ofproto->n_missed = 0;
1299
1300     return error;
1301 }
1302
1303 static int
1304 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1305                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1306 {
1307     struct ofputil_flow_mod fm;
1308     int error;
1309
1310     match_init_catchall(&fm.match);
1311     fm.priority = 0;
1312     match_set_reg(&fm.match, 0, id);
1313     fm.new_cookie = htonll(0);
1314     fm.cookie = htonll(0);
1315     fm.cookie_mask = htonll(0);
1316     fm.modify_cookie = false;
1317     fm.table_id = TBL_INTERNAL;
1318     fm.command = OFPFC_ADD;
1319     fm.idle_timeout = 0;
1320     fm.hard_timeout = 0;
1321     fm.buffer_id = 0;
1322     fm.out_port = 0;
1323     fm.flags = 0;
1324     fm.ofpacts = ofpacts->data;
1325     fm.ofpacts_len = ofpacts->size;
1326
1327     error = ofproto_flow_mod(&ofproto->up, &fm);
1328     if (error) {
1329         VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1330                     id, ofperr_to_string(error));
1331         return error;
1332     }
1333
1334     if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1335                                   rulep)) {
1336         rule_dpif_unref(*rulep);
1337     } else {
1338         NOT_REACHED();
1339     }
1340
1341     return 0;
1342 }
1343
1344 static int
1345 add_internal_flows(struct ofproto_dpif *ofproto)
1346 {
1347     struct ofpact_controller *controller;
1348     uint64_t ofpacts_stub[128 / 8];
1349     struct ofpbuf ofpacts;
1350     int error;
1351     int id;
1352
1353     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1354     id = 1;
1355
1356     controller = ofpact_put_CONTROLLER(&ofpacts);
1357     controller->max_len = UINT16_MAX;
1358     controller->controller_id = 0;
1359     controller->reason = OFPR_NO_MATCH;
1360     ofpact_pad(&ofpacts);
1361
1362     error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1363     if (error) {
1364         return error;
1365     }
1366
1367     ofpbuf_clear(&ofpacts);
1368     error = add_internal_flow(ofproto, id++, &ofpacts,
1369                               &ofproto->no_packet_in_rule);
1370     if (error) {
1371         return error;
1372     }
1373
1374     error = add_internal_flow(ofproto, id++, &ofpacts,
1375                               &ofproto->drop_frags_rule);
1376     return error;
1377 }
1378
1379 static void
1380 destruct(struct ofproto *ofproto_)
1381 {
1382     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1383     struct rule_dpif *rule, *next_rule;
1384     struct ofproto_packet_in *pin, *next_pin;
1385     struct facet *facet, *next_facet;
1386     struct cls_cursor cursor;
1387     struct oftable *table;
1388     struct list pins;
1389
1390     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1391     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1392     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1393     CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1394         facet_remove(facet);
1395     }
1396
1397     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1398     ovs_rwlock_wrlock(&xlate_rwlock);
1399     xlate_remove_ofproto(ofproto);
1400     ovs_rwlock_unlock(&xlate_rwlock);
1401
1402     /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a
1403      * use-after-free error. */
1404     udpif_revalidate(ofproto->backer->udpif);
1405
1406     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1407
1408     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1409         struct cls_cursor cursor;
1410
1411         ovs_rwlock_rdlock(&table->cls.rwlock);
1412         cls_cursor_init(&cursor, &table->cls, NULL);
1413         ovs_rwlock_unlock(&table->cls.rwlock);
1414         CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1415             ofproto_rule_delete(&ofproto->up, &rule->up);
1416         }
1417     }
1418
1419     guarded_list_pop_all(&ofproto->pins, &pins);
1420     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1421         list_remove(&pin->list_node);
1422         free(CONST_CAST(void *, pin->up.packet));
1423         free(pin);
1424     }
1425     guarded_list_destroy(&ofproto->pins);
1426
1427     mbridge_unref(ofproto->mbridge);
1428
1429     netflow_destroy(ofproto->netflow);
1430     dpif_sflow_unref(ofproto->sflow);
1431     hmap_destroy(&ofproto->bundles);
1432     mac_learning_unref(ofproto->ml);
1433
1434     classifier_destroy(&ofproto->facets);
1435
1436     hmap_destroy(&ofproto->vlandev_map);
1437     hmap_destroy(&ofproto->realdev_vid_map);
1438
1439     sset_destroy(&ofproto->ports);
1440     sset_destroy(&ofproto->ghost_ports);
1441     sset_destroy(&ofproto->port_poll_set);
1442
1443     ovs_mutex_destroy(&ofproto->stats_mutex);
1444     ovs_mutex_destroy(&ofproto->vsp_mutex);
1445
1446     close_dpif_backer(ofproto->backer);
1447 }
1448
1449 static int
1450 run_fast(struct ofproto *ofproto_)
1451 {
1452     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1453     struct ofproto_packet_in *pin, *next_pin;
1454     struct list pins;
1455
1456     /* Do not perform any periodic activity required by 'ofproto' while
1457      * waiting for flow restore to complete. */
1458     if (ofproto_get_flow_restore_wait()) {
1459         return 0;
1460     }
1461
1462     guarded_list_pop_all(&ofproto->pins, &pins);
1463     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1464         connmgr_send_packet_in(ofproto->up.connmgr, pin);
1465         list_remove(&pin->list_node);
1466         free(CONST_CAST(void *, pin->up.packet));
1467         free(pin);
1468     }
1469
1470     return 0;
1471 }
1472
1473 static int
1474 run(struct ofproto *ofproto_)
1475 {
1476     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1477     struct ofport_dpif *ofport;
1478     struct ofbundle *bundle;
1479     int error;
1480
1481     if (mbridge_need_revalidate(ofproto->mbridge)) {
1482         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1483         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1484         mac_learning_flush(ofproto->ml);
1485         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1486     }
1487
1488     /* Do not perform any periodic activity below required by 'ofproto' while
1489      * waiting for flow restore to complete. */
1490     if (ofproto_get_flow_restore_wait()) {
1491         return 0;
1492     }
1493
1494     error = run_fast(ofproto_);
1495     if (error) {
1496         return error;
1497     }
1498
1499     if (ofproto->netflow) {
1500         if (netflow_run(ofproto->netflow)) {
1501             send_netflow_active_timeouts(ofproto);
1502         }
1503     }
1504     if (ofproto->sflow) {
1505         dpif_sflow_run(ofproto->sflow);
1506     }
1507     if (ofproto->ipfix) {
1508         dpif_ipfix_run(ofproto->ipfix);
1509     }
1510
1511     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1512         port_run(ofport);
1513     }
1514     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1515         bundle_run(bundle);
1516     }
1517
1518     stp_run(ofproto);
1519     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1520     if (mac_learning_run(ofproto->ml)) {
1521         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1522     }
1523     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1524
1525     /* Check the consistency of a random facet, to aid debugging. */
1526     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1527     if (time_msec() >= ofproto->consistency_rl
1528         && !classifier_is_empty(&ofproto->facets)
1529         && !ofproto->backer->need_revalidate) {
1530         struct cls_subtable *table;
1531         struct cls_rule *cr;
1532         struct facet *facet;
1533
1534         ofproto->consistency_rl = time_msec() + 250;
1535
1536         table = CONTAINER_OF(hmap_random_node(&ofproto->facets.subtables),
1537                              struct cls_subtable, hmap_node);
1538         cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1539                           hmap_node);
1540         facet = CONTAINER_OF(cr, struct facet, cr);
1541
1542         if (!facet_check_consistency(facet)) {
1543             ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1544         }
1545     }
1546     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1547
1548     return 0;
1549 }
1550
1551 static void
1552 wait(struct ofproto *ofproto_)
1553 {
1554     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1555     struct ofbundle *bundle;
1556
1557     if (ofproto_get_flow_restore_wait()) {
1558         return;
1559     }
1560
1561     if (ofproto->sflow) {
1562         dpif_sflow_wait(ofproto->sflow);
1563     }
1564     if (ofproto->ipfix) {
1565         dpif_ipfix_wait(ofproto->ipfix);
1566     }
1567     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1568         bundle_wait(bundle);
1569     }
1570     if (ofproto->netflow) {
1571         netflow_wait(ofproto->netflow);
1572     }
1573     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1574     mac_learning_wait(ofproto->ml);
1575     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1576     stp_wait(ofproto);
1577     if (ofproto->backer->need_revalidate) {
1578         /* Shouldn't happen, but if it does just go around again. */
1579         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1580         poll_immediate_wake();
1581     }
1582 }
1583
1584 static void
1585 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1586 {
1587     const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1588     struct cls_cursor cursor;
1589     size_t n_subfacets = 0;
1590     struct facet *facet;
1591
1592     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1593     simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1594     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1595
1596     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1597     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1598     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1599         n_subfacets += list_size(&facet->subfacets);
1600     }
1601     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1602     simap_increase(usage, "subfacets", n_subfacets);
1603 }
1604
1605 static void
1606 flush(struct ofproto *ofproto_)
1607 {
1608     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1609     struct subfacet *subfacet, *next_subfacet;
1610     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1611     int n_batch;
1612
1613     n_batch = 0;
1614     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1615                         &ofproto->backer->subfacets) {
1616         if (subfacet->facet->ofproto != ofproto) {
1617             continue;
1618         }
1619
1620         if (subfacet->path != SF_NOT_INSTALLED) {
1621             batch[n_batch++] = subfacet;
1622             if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1623                 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1624                 n_batch = 0;
1625             }
1626         } else {
1627             subfacet_destroy(subfacet);
1628         }
1629     }
1630
1631     if (n_batch > 0) {
1632         subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1633     }
1634 }
1635
1636 static void
1637 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1638              bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1639 {
1640     *arp_match_ip = true;
1641     *actions = (OFPUTIL_A_OUTPUT |
1642                 OFPUTIL_A_SET_VLAN_VID |
1643                 OFPUTIL_A_SET_VLAN_PCP |
1644                 OFPUTIL_A_STRIP_VLAN |
1645                 OFPUTIL_A_SET_DL_SRC |
1646                 OFPUTIL_A_SET_DL_DST |
1647                 OFPUTIL_A_SET_NW_SRC |
1648                 OFPUTIL_A_SET_NW_DST |
1649                 OFPUTIL_A_SET_NW_TOS |
1650                 OFPUTIL_A_SET_TP_SRC |
1651                 OFPUTIL_A_SET_TP_DST |
1652                 OFPUTIL_A_ENQUEUE);
1653 }
1654
1655 static void
1656 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1657 {
1658     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1659     struct dpif_dp_stats s;
1660     uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1661     uint64_t n_lookup;
1662
1663     strcpy(ots->name, "classifier");
1664
1665     dpif_get_dp_stats(ofproto->backer->dpif, &s);
1666     rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1667     rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1668     rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1669
1670     n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1671     ots->lookup_count = htonll(n_lookup);
1672     ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1673 }
1674
1675 static struct ofport *
1676 port_alloc(void)
1677 {
1678     struct ofport_dpif *port = xmalloc(sizeof *port);
1679     return &port->up;
1680 }
1681
1682 static void
1683 port_dealloc(struct ofport *port_)
1684 {
1685     struct ofport_dpif *port = ofport_dpif_cast(port_);
1686     free(port);
1687 }
1688
1689 static int
1690 port_construct(struct ofport *port_)
1691 {
1692     struct ofport_dpif *port = ofport_dpif_cast(port_);
1693     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1694     const struct netdev *netdev = port->up.netdev;
1695     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1696     struct dpif_port dpif_port;
1697     int error;
1698
1699     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1700     port->bundle = NULL;
1701     port->cfm = NULL;
1702     port->bfd = NULL;
1703     port->may_enable = true;
1704     port->stp_port = NULL;
1705     port->stp_state = STP_DISABLED;
1706     port->is_tunnel = false;
1707     port->peer = NULL;
1708     port->qdscp = NULL;
1709     port->n_qdscp = 0;
1710     port->realdev_ofp_port = 0;
1711     port->vlandev_vid = 0;
1712     port->carrier_seq = netdev_get_carrier_resets(netdev);
1713
1714     if (netdev_vport_is_patch(netdev)) {
1715         /* By bailing out here, we don't submit the port to the sFlow module
1716          * to be considered for counter polling export.  This is correct
1717          * because the patch port represents an interface that sFlow considers
1718          * to be "internal" to the switch as a whole, and therefore not an
1719          * candidate for counter polling. */
1720         port->odp_port = ODPP_NONE;
1721         ofport_update_peer(port);
1722         return 0;
1723     }
1724
1725     error = dpif_port_query_by_name(ofproto->backer->dpif,
1726                                     netdev_vport_get_dpif_port(netdev, namebuf,
1727                                                                sizeof namebuf),
1728                                     &dpif_port);
1729     if (error) {
1730         return error;
1731     }
1732
1733     port->odp_port = dpif_port.port_no;
1734
1735     if (netdev_get_tunnel_config(netdev)) {
1736         tnl_port_add(port, port->up.netdev, port->odp_port);
1737         port->is_tunnel = true;
1738     } else {
1739         /* Sanity-check that a mapping doesn't already exist.  This
1740          * shouldn't happen for non-tunnel ports. */
1741         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1742             VLOG_ERR("port %s already has an OpenFlow port number",
1743                      dpif_port.name);
1744             dpif_port_destroy(&dpif_port);
1745             return EBUSY;
1746         }
1747
1748         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1749         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1750                     hash_odp_port(port->odp_port));
1751         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1752     }
1753     dpif_port_destroy(&dpif_port);
1754
1755     if (ofproto->sflow) {
1756         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1757     }
1758
1759     return 0;
1760 }
1761
1762 static void
1763 port_destruct(struct ofport *port_)
1764 {
1765     struct ofport_dpif *port = ofport_dpif_cast(port_);
1766     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1767     const char *devname = netdev_get_name(port->up.netdev);
1768     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1769     const char *dp_port_name;
1770
1771     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1772     ovs_rwlock_wrlock(&xlate_rwlock);
1773     xlate_ofport_remove(port);
1774     ovs_rwlock_unlock(&xlate_rwlock);
1775
1776     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1777                                               sizeof namebuf);
1778     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1779         /* The underlying device is still there, so delete it.  This
1780          * happens when the ofproto is being destroyed, since the caller
1781          * assumes that removal of attached ports will happen as part of
1782          * destruction. */
1783         if (!port->is_tunnel) {
1784             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1785         }
1786     }
1787
1788     if (port->peer) {
1789         port->peer->peer = NULL;
1790         port->peer = NULL;
1791     }
1792
1793     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1794         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1795         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1796         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1797     }
1798
1799     tnl_port_del(port);
1800     sset_find_and_delete(&ofproto->ports, devname);
1801     sset_find_and_delete(&ofproto->ghost_ports, devname);
1802     bundle_remove(port_);
1803     set_cfm(port_, NULL);
1804     set_bfd(port_, NULL);
1805     if (ofproto->sflow) {
1806         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1807     }
1808
1809     free(port->qdscp);
1810 }
1811
1812 static void
1813 port_modified(struct ofport *port_)
1814 {
1815     struct ofport_dpif *port = ofport_dpif_cast(port_);
1816
1817     if (port->bundle && port->bundle->bond) {
1818         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1819     }
1820
1821     if (port->cfm) {
1822         cfm_set_netdev(port->cfm, port->up.netdev);
1823     }
1824
1825     if (port->bfd) {
1826         bfd_set_netdev(port->bfd, port->up.netdev);
1827     }
1828
1829     ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1830                                      port->up.pp.hw_addr);
1831
1832     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1833                                                 port->odp_port)) {
1834         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1835             REV_RECONFIGURE;
1836     }
1837
1838     ofport_update_peer(port);
1839 }
1840
1841 static void
1842 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1843 {
1844     struct ofport_dpif *port = ofport_dpif_cast(port_);
1845     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1846     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1847
1848     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1849                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1850                    OFPUTIL_PC_NO_PACKET_IN)) {
1851         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1852
1853         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1854             bundle_update(port->bundle);
1855         }
1856     }
1857 }
1858
1859 static int
1860 set_sflow(struct ofproto *ofproto_,
1861           const struct ofproto_sflow_options *sflow_options)
1862 {
1863     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1864     struct dpif_sflow *ds = ofproto->sflow;
1865
1866     if (sflow_options) {
1867         if (!ds) {
1868             struct ofport_dpif *ofport;
1869
1870             ds = ofproto->sflow = dpif_sflow_create();
1871             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1872                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1873             }
1874             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1875         }
1876         dpif_sflow_set_options(ds, sflow_options);
1877     } else {
1878         if (ds) {
1879             dpif_sflow_unref(ds);
1880             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1881             ofproto->sflow = NULL;
1882         }
1883     }
1884     return 0;
1885 }
1886
1887 static int
1888 set_ipfix(
1889     struct ofproto *ofproto_,
1890     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1891     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1892     size_t n_flow_exporters_options)
1893 {
1894     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1895     struct dpif_ipfix *di = ofproto->ipfix;
1896     bool has_options = bridge_exporter_options || flow_exporters_options;
1897
1898     if (has_options && !di) {
1899         di = ofproto->ipfix = dpif_ipfix_create();
1900     }
1901
1902     if (di) {
1903         /* Call set_options in any case to cleanly flush the flow
1904          * caches in the last exporters that are to be destroyed. */
1905         dpif_ipfix_set_options(
1906             di, bridge_exporter_options, flow_exporters_options,
1907             n_flow_exporters_options);
1908
1909         if (!has_options) {
1910             dpif_ipfix_unref(di);
1911             ofproto->ipfix = NULL;
1912         }
1913     }
1914
1915     return 0;
1916 }
1917
1918 static int
1919 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1920 {
1921     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1922     int error = 0;
1923
1924     if (s) {
1925         if (!ofport->cfm) {
1926             struct ofproto_dpif *ofproto;
1927
1928             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1929             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1930             ofport->cfm = cfm_create(ofport->up.netdev);
1931         }
1932
1933         if (cfm_configure(ofport->cfm, s)) {
1934             error = 0;
1935             goto out;
1936         }
1937
1938         error = EINVAL;
1939     }
1940     cfm_unref(ofport->cfm);
1941     ofport->cfm = NULL;
1942 out:
1943     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1944                                      ofport->up.pp.hw_addr);
1945     return error;
1946 }
1947
1948 static bool
1949 get_cfm_status(const struct ofport *ofport_,
1950                struct ofproto_cfm_status *status)
1951 {
1952     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1953
1954     if (ofport->cfm) {
1955         status->faults = cfm_get_fault(ofport->cfm);
1956         status->remote_opstate = cfm_get_opup(ofport->cfm);
1957         status->health = cfm_get_health(ofport->cfm);
1958         cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1959         return true;
1960     } else {
1961         return false;
1962     }
1963 }
1964
1965 static int
1966 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1967 {
1968     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1969     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1970     struct bfd *old;
1971
1972     old = ofport->bfd;
1973     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1974                                 cfg, ofport->up.netdev);
1975     if (ofport->bfd != old) {
1976         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1977     }
1978     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1979                                      ofport->up.pp.hw_addr);
1980     return 0;
1981 }
1982
1983 static int
1984 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1985 {
1986     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1987
1988     if (ofport->bfd) {
1989         bfd_get_status(ofport->bfd, smap);
1990         return 0;
1991     } else {
1992         return ENOENT;
1993     }
1994 }
1995 \f
1996 /* Spanning Tree. */
1997
1998 static void
1999 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2000 {
2001     struct ofproto_dpif *ofproto = ofproto_;
2002     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2003     struct ofport_dpif *ofport;
2004
2005     ofport = stp_port_get_aux(sp);
2006     if (!ofport) {
2007         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2008                      ofproto->up.name, port_num);
2009     } else {
2010         struct eth_header *eth = pkt->l2;
2011
2012         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2013         if (eth_addr_is_zero(eth->eth_src)) {
2014             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2015                          "with unknown MAC", ofproto->up.name, port_num);
2016         } else {
2017             ofproto_dpif_send_packet(ofport, pkt);
2018         }
2019     }
2020     ofpbuf_delete(pkt);
2021 }
2022
2023 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2024 static int
2025 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2026 {
2027     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2028
2029     /* Only revalidate flows if the configuration changed. */
2030     if (!s != !ofproto->stp) {
2031         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2032     }
2033
2034     if (s) {
2035         if (!ofproto->stp) {
2036             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2037                                       send_bpdu_cb, ofproto);
2038             ofproto->stp_last_tick = time_msec();
2039         }
2040
2041         stp_set_bridge_id(ofproto->stp, s->system_id);
2042         stp_set_bridge_priority(ofproto->stp, s->priority);
2043         stp_set_hello_time(ofproto->stp, s->hello_time);
2044         stp_set_max_age(ofproto->stp, s->max_age);
2045         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2046     }  else {
2047         struct ofport *ofport;
2048
2049         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2050             set_stp_port(ofport, NULL);
2051         }
2052
2053         stp_unref(ofproto->stp);
2054         ofproto->stp = NULL;
2055     }
2056
2057     return 0;
2058 }
2059
2060 static int
2061 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2062 {
2063     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2064
2065     if (ofproto->stp) {
2066         s->enabled = true;
2067         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2068         s->designated_root = stp_get_designated_root(ofproto->stp);
2069         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2070     } else {
2071         s->enabled = false;
2072     }
2073
2074     return 0;
2075 }
2076
2077 static void
2078 update_stp_port_state(struct ofport_dpif *ofport)
2079 {
2080     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2081     enum stp_state state;
2082
2083     /* Figure out new state. */
2084     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2085                              : STP_DISABLED;
2086
2087     /* Update state. */
2088     if (ofport->stp_state != state) {
2089         enum ofputil_port_state of_state;
2090         bool fwd_change;
2091
2092         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2093                     netdev_get_name(ofport->up.netdev),
2094                     stp_state_name(ofport->stp_state),
2095                     stp_state_name(state));
2096         if (stp_learn_in_state(ofport->stp_state)
2097                 != stp_learn_in_state(state)) {
2098             /* xxx Learning action flows should also be flushed. */
2099             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2100             mac_learning_flush(ofproto->ml);
2101             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2102         }
2103         fwd_change = stp_forward_in_state(ofport->stp_state)
2104                         != stp_forward_in_state(state);
2105
2106         ofproto->backer->need_revalidate = REV_STP;
2107         ofport->stp_state = state;
2108         ofport->stp_state_entered = time_msec();
2109
2110         if (fwd_change && ofport->bundle) {
2111             bundle_update(ofport->bundle);
2112         }
2113
2114         /* Update the STP state bits in the OpenFlow port description. */
2115         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2116         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2117                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2118                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2119                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2120                      : 0);
2121         ofproto_port_set_state(&ofport->up, of_state);
2122     }
2123 }
2124
2125 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2126  * caller is responsible for assigning STP port numbers and ensuring
2127  * there are no duplicates. */
2128 static int
2129 set_stp_port(struct ofport *ofport_,
2130              const struct ofproto_port_stp_settings *s)
2131 {
2132     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2133     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2134     struct stp_port *sp = ofport->stp_port;
2135
2136     if (!s || !s->enable) {
2137         if (sp) {
2138             ofport->stp_port = NULL;
2139             stp_port_disable(sp);
2140             update_stp_port_state(ofport);
2141         }
2142         return 0;
2143     } else if (sp && stp_port_no(sp) != s->port_num
2144             && ofport == stp_port_get_aux(sp)) {
2145         /* The port-id changed, so disable the old one if it's not
2146          * already in use by another port. */
2147         stp_port_disable(sp);
2148     }
2149
2150     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2151     stp_port_enable(sp);
2152
2153     stp_port_set_aux(sp, ofport);
2154     stp_port_set_priority(sp, s->priority);
2155     stp_port_set_path_cost(sp, s->path_cost);
2156
2157     update_stp_port_state(ofport);
2158
2159     return 0;
2160 }
2161
2162 static int
2163 get_stp_port_status(struct ofport *ofport_,
2164                     struct ofproto_port_stp_status *s)
2165 {
2166     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2167     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2168     struct stp_port *sp = ofport->stp_port;
2169
2170     if (!ofproto->stp || !sp) {
2171         s->enabled = false;
2172         return 0;
2173     }
2174
2175     s->enabled = true;
2176     s->port_id = stp_port_get_id(sp);
2177     s->state = stp_port_get_state(sp);
2178     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2179     s->role = stp_port_get_role(sp);
2180     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2181
2182     return 0;
2183 }
2184
2185 static void
2186 stp_run(struct ofproto_dpif *ofproto)
2187 {
2188     if (ofproto->stp) {
2189         long long int now = time_msec();
2190         long long int elapsed = now - ofproto->stp_last_tick;
2191         struct stp_port *sp;
2192
2193         if (elapsed > 0) {
2194             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2195             ofproto->stp_last_tick = now;
2196         }
2197         while (stp_get_changed_port(ofproto->stp, &sp)) {
2198             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2199
2200             if (ofport) {
2201                 update_stp_port_state(ofport);
2202             }
2203         }
2204
2205         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2206             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2207             mac_learning_flush(ofproto->ml);
2208             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2209         }
2210     }
2211 }
2212
2213 static void
2214 stp_wait(struct ofproto_dpif *ofproto)
2215 {
2216     if (ofproto->stp) {
2217         poll_timer_wait(1000);
2218     }
2219 }
2220 \f
2221 static int
2222 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2223            size_t n_qdscp)
2224 {
2225     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2226     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2227
2228     if (ofport->n_qdscp != n_qdscp
2229         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2230                               n_qdscp * sizeof *qdscp))) {
2231         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2232         free(ofport->qdscp);
2233         ofport->qdscp = n_qdscp
2234             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2235             : NULL;
2236         ofport->n_qdscp = n_qdscp;
2237     }
2238
2239     return 0;
2240 }
2241 \f
2242 /* Bundles. */
2243
2244 /* Expires all MAC learning entries associated with 'bundle' and forces its
2245  * ofproto to revalidate every flow.
2246  *
2247  * Normally MAC learning entries are removed only from the ofproto associated
2248  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2249  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2250  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2251  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2252  * with the host from which it migrated. */
2253 static void
2254 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2255 {
2256     struct ofproto_dpif *ofproto = bundle->ofproto;
2257     struct mac_learning *ml = ofproto->ml;
2258     struct mac_entry *mac, *next_mac;
2259
2260     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2261     ovs_rwlock_wrlock(&ml->rwlock);
2262     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2263         if (mac->port.p == bundle) {
2264             if (all_ofprotos) {
2265                 struct ofproto_dpif *o;
2266
2267                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2268                     if (o != ofproto) {
2269                         struct mac_entry *e;
2270
2271                         ovs_rwlock_wrlock(&o->ml->rwlock);
2272                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2273                         if (e) {
2274                             mac_learning_expire(o->ml, e);
2275                         }
2276                         ovs_rwlock_unlock(&o->ml->rwlock);
2277                     }
2278                 }
2279             }
2280
2281             mac_learning_expire(ml, mac);
2282         }
2283     }
2284     ovs_rwlock_unlock(&ml->rwlock);
2285 }
2286
2287 static struct ofbundle *
2288 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2289 {
2290     struct ofbundle *bundle;
2291
2292     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2293                              &ofproto->bundles) {
2294         if (bundle->aux == aux) {
2295             return bundle;
2296         }
2297     }
2298     return NULL;
2299 }
2300
2301 static void
2302 bundle_update(struct ofbundle *bundle)
2303 {
2304     struct ofport_dpif *port;
2305
2306     bundle->floodable = true;
2307     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2308         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2309             || !stp_forward_in_state(port->stp_state)) {
2310             bundle->floodable = false;
2311             break;
2312         }
2313     }
2314 }
2315
2316 static void
2317 bundle_del_port(struct ofport_dpif *port)
2318 {
2319     struct ofbundle *bundle = port->bundle;
2320
2321     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2322
2323     list_remove(&port->bundle_node);
2324     port->bundle = NULL;
2325
2326     if (bundle->lacp) {
2327         lacp_slave_unregister(bundle->lacp, port);
2328     }
2329     if (bundle->bond) {
2330         bond_slave_unregister(bundle->bond, port);
2331     }
2332
2333     bundle_update(bundle);
2334 }
2335
2336 static bool
2337 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2338                 struct lacp_slave_settings *lacp)
2339 {
2340     struct ofport_dpif *port;
2341
2342     port = get_ofp_port(bundle->ofproto, ofp_port);
2343     if (!port) {
2344         return false;
2345     }
2346
2347     if (port->bundle != bundle) {
2348         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2349         if (port->bundle) {
2350             bundle_remove(&port->up);
2351         }
2352
2353         port->bundle = bundle;
2354         list_push_back(&bundle->ports, &port->bundle_node);
2355         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2356             || !stp_forward_in_state(port->stp_state)) {
2357             bundle->floodable = false;
2358         }
2359     }
2360     if (lacp) {
2361         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2362         lacp_slave_register(bundle->lacp, port, lacp);
2363     }
2364
2365     return true;
2366 }
2367
2368 static void
2369 bundle_destroy(struct ofbundle *bundle)
2370 {
2371     struct ofproto_dpif *ofproto;
2372     struct ofport_dpif *port, *next_port;
2373
2374     if (!bundle) {
2375         return;
2376     }
2377
2378     ofproto = bundle->ofproto;
2379     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2380
2381     ovs_rwlock_wrlock(&xlate_rwlock);
2382     xlate_bundle_remove(bundle);
2383     ovs_rwlock_unlock(&xlate_rwlock);
2384
2385     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2386         bundle_del_port(port);
2387     }
2388
2389     bundle_flush_macs(bundle, true);
2390     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2391     free(bundle->name);
2392     free(bundle->trunks);
2393     lacp_unref(bundle->lacp);
2394     bond_unref(bundle->bond);
2395     free(bundle);
2396 }
2397
2398 static int
2399 bundle_set(struct ofproto *ofproto_, void *aux,
2400            const struct ofproto_bundle_settings *s)
2401 {
2402     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2403     bool need_flush = false;
2404     struct ofport_dpif *port;
2405     struct ofbundle *bundle;
2406     unsigned long *trunks;
2407     int vlan;
2408     size_t i;
2409     bool ok;
2410
2411     if (!s) {
2412         bundle_destroy(bundle_lookup(ofproto, aux));
2413         return 0;
2414     }
2415
2416     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2417     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2418
2419     bundle = bundle_lookup(ofproto, aux);
2420     if (!bundle) {
2421         bundle = xmalloc(sizeof *bundle);
2422
2423         bundle->ofproto = ofproto;
2424         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2425                     hash_pointer(aux, 0));
2426         bundle->aux = aux;
2427         bundle->name = NULL;
2428
2429         list_init(&bundle->ports);
2430         bundle->vlan_mode = PORT_VLAN_TRUNK;
2431         bundle->vlan = -1;
2432         bundle->trunks = NULL;
2433         bundle->use_priority_tags = s->use_priority_tags;
2434         bundle->lacp = NULL;
2435         bundle->bond = NULL;
2436
2437         bundle->floodable = true;
2438         mbridge_register_bundle(ofproto->mbridge, bundle);
2439     }
2440
2441     if (!bundle->name || strcmp(s->name, bundle->name)) {
2442         free(bundle->name);
2443         bundle->name = xstrdup(s->name);
2444     }
2445
2446     /* LACP. */
2447     if (s->lacp) {
2448         if (!bundle->lacp) {
2449             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2450             bundle->lacp = lacp_create();
2451         }
2452         lacp_configure(bundle->lacp, s->lacp);
2453     } else {
2454         lacp_unref(bundle->lacp);
2455         bundle->lacp = NULL;
2456     }
2457
2458     /* Update set of ports. */
2459     ok = true;
2460     for (i = 0; i < s->n_slaves; i++) {
2461         if (!bundle_add_port(bundle, s->slaves[i],
2462                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2463             ok = false;
2464         }
2465     }
2466     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2467         struct ofport_dpif *next_port;
2468
2469         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2470             for (i = 0; i < s->n_slaves; i++) {
2471                 if (s->slaves[i] == port->up.ofp_port) {
2472                     goto found;
2473                 }
2474             }
2475
2476             bundle_del_port(port);
2477         found: ;
2478         }
2479     }
2480     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2481
2482     if (list_is_empty(&bundle->ports)) {
2483         bundle_destroy(bundle);
2484         return EINVAL;
2485     }
2486
2487     /* Set VLAN tagging mode */
2488     if (s->vlan_mode != bundle->vlan_mode
2489         || s->use_priority_tags != bundle->use_priority_tags) {
2490         bundle->vlan_mode = s->vlan_mode;
2491         bundle->use_priority_tags = s->use_priority_tags;
2492         need_flush = true;
2493     }
2494
2495     /* Set VLAN tag. */
2496     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2497             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2498             : 0);
2499     if (vlan != bundle->vlan) {
2500         bundle->vlan = vlan;
2501         need_flush = true;
2502     }
2503
2504     /* Get trunked VLANs. */
2505     switch (s->vlan_mode) {
2506     case PORT_VLAN_ACCESS:
2507         trunks = NULL;
2508         break;
2509
2510     case PORT_VLAN_TRUNK:
2511         trunks = CONST_CAST(unsigned long *, s->trunks);
2512         break;
2513
2514     case PORT_VLAN_NATIVE_UNTAGGED:
2515     case PORT_VLAN_NATIVE_TAGGED:
2516         if (vlan != 0 && (!s->trunks
2517                           || !bitmap_is_set(s->trunks, vlan)
2518                           || bitmap_is_set(s->trunks, 0))) {
2519             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2520             if (s->trunks) {
2521                 trunks = bitmap_clone(s->trunks, 4096);
2522             } else {
2523                 trunks = bitmap_allocate1(4096);
2524             }
2525             bitmap_set1(trunks, vlan);
2526             bitmap_set0(trunks, 0);
2527         } else {
2528             trunks = CONST_CAST(unsigned long *, s->trunks);
2529         }
2530         break;
2531
2532     default:
2533         NOT_REACHED();
2534     }
2535     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2536         free(bundle->trunks);
2537         if (trunks == s->trunks) {
2538             bundle->trunks = vlan_bitmap_clone(trunks);
2539         } else {
2540             bundle->trunks = trunks;
2541             trunks = NULL;
2542         }
2543         need_flush = true;
2544     }
2545     if (trunks != s->trunks) {
2546         free(trunks);
2547     }
2548
2549     /* Bonding. */
2550     if (!list_is_short(&bundle->ports)) {
2551         bundle->ofproto->has_bonded_bundles = true;
2552         if (bundle->bond) {
2553             if (bond_reconfigure(bundle->bond, s->bond)) {
2554                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2555             }
2556         } else {
2557             bundle->bond = bond_create(s->bond);
2558             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2559         }
2560
2561         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2562             bond_slave_register(bundle->bond, port, port->up.netdev);
2563         }
2564     } else {
2565         bond_unref(bundle->bond);
2566         bundle->bond = NULL;
2567     }
2568
2569     /* If we changed something that would affect MAC learning, un-learn
2570      * everything on this port and force flow revalidation. */
2571     if (need_flush) {
2572         bundle_flush_macs(bundle, false);
2573     }
2574
2575     return 0;
2576 }
2577
2578 static void
2579 bundle_remove(struct ofport *port_)
2580 {
2581     struct ofport_dpif *port = ofport_dpif_cast(port_);
2582     struct ofbundle *bundle = port->bundle;
2583
2584     if (bundle) {
2585         bundle_del_port(port);
2586         if (list_is_empty(&bundle->ports)) {
2587             bundle_destroy(bundle);
2588         } else if (list_is_short(&bundle->ports)) {
2589             bond_unref(bundle->bond);
2590             bundle->bond = NULL;
2591         }
2592     }
2593 }
2594
2595 static void
2596 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2597 {
2598     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2599     struct ofport_dpif *port = port_;
2600     uint8_t ea[ETH_ADDR_LEN];
2601     int error;
2602
2603     error = netdev_get_etheraddr(port->up.netdev, ea);
2604     if (!error) {
2605         struct ofpbuf packet;
2606         void *packet_pdu;
2607
2608         ofpbuf_init(&packet, 0);
2609         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2610                                  pdu_size);
2611         memcpy(packet_pdu, pdu, pdu_size);
2612
2613         ofproto_dpif_send_packet(port, &packet);
2614         ofpbuf_uninit(&packet);
2615     } else {
2616         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2617                     "%s (%s)", port->bundle->name,
2618                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2619     }
2620 }
2621
2622 static void
2623 bundle_send_learning_packets(struct ofbundle *bundle)
2624 {
2625     struct ofproto_dpif *ofproto = bundle->ofproto;
2626     struct ofpbuf *learning_packet;
2627     int error, n_packets, n_errors;
2628     struct mac_entry *e;
2629     struct list packets;
2630
2631     list_init(&packets);
2632     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2633     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2634         if (e->port.p != bundle) {
2635             void *port_void;
2636
2637             learning_packet = bond_compose_learning_packet(bundle->bond,
2638                                                            e->mac, e->vlan,
2639                                                            &port_void);
2640             learning_packet->private_p = port_void;
2641             list_push_back(&packets, &learning_packet->list_node);
2642         }
2643     }
2644     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2645
2646     error = n_packets = n_errors = 0;
2647     LIST_FOR_EACH (learning_packet, list_node, &packets) {
2648         int ret;
2649
2650         ret = ofproto_dpif_send_packet(learning_packet->private_p, learning_packet);
2651         if (ret) {
2652             error = ret;
2653             n_errors++;
2654         }
2655         n_packets++;
2656     }
2657     ofpbuf_list_delete(&packets);
2658
2659     if (n_errors) {
2660         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2661         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2662                      "packets, last error was: %s",
2663                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2664     } else {
2665         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2666                  bundle->name, n_packets);
2667     }
2668 }
2669
2670 static void
2671 bundle_run(struct ofbundle *bundle)
2672 {
2673     if (bundle->lacp) {
2674         lacp_run(bundle->lacp, send_pdu_cb);
2675     }
2676     if (bundle->bond) {
2677         struct ofport_dpif *port;
2678
2679         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2680             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2681         }
2682
2683         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2684             bundle->ofproto->backer->need_revalidate = REV_BOND;
2685         }
2686
2687         if (bond_should_send_learning_packets(bundle->bond)) {
2688             bundle_send_learning_packets(bundle);
2689         }
2690     }
2691 }
2692
2693 static void
2694 bundle_wait(struct ofbundle *bundle)
2695 {
2696     if (bundle->lacp) {
2697         lacp_wait(bundle->lacp);
2698     }
2699     if (bundle->bond) {
2700         bond_wait(bundle->bond);
2701     }
2702 }
2703 \f
2704 /* Mirrors. */
2705
2706 static int
2707 mirror_set__(struct ofproto *ofproto_, void *aux,
2708              const struct ofproto_mirror_settings *s)
2709 {
2710     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2711     struct ofbundle **srcs, **dsts;
2712     int error;
2713     size_t i;
2714
2715     if (!s) {
2716         mirror_destroy(ofproto->mbridge, aux);
2717         return 0;
2718     }
2719
2720     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2721     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2722
2723     for (i = 0; i < s->n_srcs; i++) {
2724         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2725     }
2726
2727     for (i = 0; i < s->n_dsts; i++) {
2728         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2729     }
2730
2731     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2732                        s->n_dsts, s->src_vlans,
2733                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2734     free(srcs);
2735     free(dsts);
2736     return error;
2737 }
2738
2739 static int
2740 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2741                    uint64_t *packets, uint64_t *bytes)
2742 {
2743     push_all_stats();
2744     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2745                             bytes);
2746 }
2747
2748 static int
2749 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2750 {
2751     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2752     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2753     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2754         mac_learning_flush(ofproto->ml);
2755     }
2756     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2757     return 0;
2758 }
2759
2760 static bool
2761 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2762 {
2763     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2764     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2765     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2766 }
2767
2768 static void
2769 forward_bpdu_changed(struct ofproto *ofproto_)
2770 {
2771     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2772     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2773 }
2774
2775 static void
2776 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2777                      size_t max_entries)
2778 {
2779     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2780     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2781     mac_learning_set_idle_time(ofproto->ml, idle_time);
2782     mac_learning_set_max_entries(ofproto->ml, max_entries);
2783     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2784 }
2785 \f
2786 /* Ports. */
2787
2788 static struct ofport_dpif *
2789 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2790 {
2791     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2792     return ofport ? ofport_dpif_cast(ofport) : NULL;
2793 }
2794
2795 static struct ofport_dpif *
2796 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2797 {
2798     struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2799     return port && &ofproto->up == port->up.ofproto ? port : NULL;
2800 }
2801
2802 static void
2803 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2804                             struct ofproto_port *ofproto_port,
2805                             struct dpif_port *dpif_port)
2806 {
2807     ofproto_port->name = dpif_port->name;
2808     ofproto_port->type = dpif_port->type;
2809     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2810 }
2811
2812 static void
2813 ofport_update_peer(struct ofport_dpif *ofport)
2814 {
2815     const struct ofproto_dpif *ofproto;
2816     struct dpif_backer *backer;
2817     char *peer_name;
2818
2819     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2820         return;
2821     }
2822
2823     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2824     backer->need_revalidate = REV_RECONFIGURE;
2825
2826     if (ofport->peer) {
2827         ofport->peer->peer = NULL;
2828         ofport->peer = NULL;
2829     }
2830
2831     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2832     if (!peer_name) {
2833         return;
2834     }
2835
2836     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2837         struct ofport *peer_ofport;
2838         struct ofport_dpif *peer;
2839         char *peer_peer;
2840
2841         if (ofproto->backer != backer) {
2842             continue;
2843         }
2844
2845         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2846         if (!peer_ofport) {
2847             continue;
2848         }
2849
2850         peer = ofport_dpif_cast(peer_ofport);
2851         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2852         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2853                                  peer_peer)) {
2854             ofport->peer = peer;
2855             ofport->peer->peer = ofport;
2856         }
2857         free(peer_peer);
2858
2859         break;
2860     }
2861     free(peer_name);
2862 }
2863
2864 static void
2865 port_run(struct ofport_dpif *ofport)
2866 {
2867     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2868     bool carrier_changed = carrier_seq != ofport->carrier_seq;
2869     bool enable = netdev_get_carrier(ofport->up.netdev);
2870     bool cfm_enable = false;
2871     bool bfd_enable = false;
2872
2873     ofport->carrier_seq = carrier_seq;
2874
2875     if (ofport->cfm) {
2876         int cfm_opup = cfm_get_opup(ofport->cfm);
2877
2878         cfm_enable = !cfm_get_fault(ofport->cfm);
2879
2880         if (cfm_opup >= 0) {
2881             cfm_enable = cfm_enable && cfm_opup;
2882         }
2883     }
2884
2885     if (ofport->bfd) {
2886         bfd_enable = bfd_forwarding(ofport->bfd);
2887     }
2888
2889     if (ofport->bfd || ofport->cfm) {
2890         enable = enable && (cfm_enable || bfd_enable);
2891     }
2892
2893     if (ofport->bundle) {
2894         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2895         if (carrier_changed) {
2896             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2897         }
2898     }
2899
2900     if (ofport->may_enable != enable) {
2901         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2902         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2903     }
2904
2905     ofport->may_enable = enable;
2906 }
2907
2908 static int
2909 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2910                    struct ofproto_port *ofproto_port)
2911 {
2912     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2913     struct dpif_port dpif_port;
2914     int error;
2915
2916     if (sset_contains(&ofproto->ghost_ports, devname)) {
2917         const char *type = netdev_get_type_from_name(devname);
2918
2919         /* We may be called before ofproto->up.port_by_name is populated with
2920          * the appropriate ofport.  For this reason, we must get the name and
2921          * type from the netdev layer directly. */
2922         if (type) {
2923             const struct ofport *ofport;
2924
2925             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2926             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2927             ofproto_port->name = xstrdup(devname);
2928             ofproto_port->type = xstrdup(type);
2929             return 0;
2930         }
2931         return ENODEV;
2932     }
2933
2934     if (!sset_contains(&ofproto->ports, devname)) {
2935         return ENODEV;
2936     }
2937     error = dpif_port_query_by_name(ofproto->backer->dpif,
2938                                     devname, &dpif_port);
2939     if (!error) {
2940         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2941     }
2942     return error;
2943 }
2944
2945 static int
2946 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2947 {
2948     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2949     const char *devname = netdev_get_name(netdev);
2950     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2951     const char *dp_port_name;
2952
2953     if (netdev_vport_is_patch(netdev)) {
2954         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2955         return 0;
2956     }
2957
2958     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2959     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2960         odp_port_t port_no = ODPP_NONE;
2961         int error;
2962
2963         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2964         if (error) {
2965             return error;
2966         }
2967         if (netdev_get_tunnel_config(netdev)) {
2968             simap_put(&ofproto->backer->tnl_backers,
2969                       dp_port_name, odp_to_u32(port_no));
2970         }
2971     }
2972
2973     if (netdev_get_tunnel_config(netdev)) {
2974         sset_add(&ofproto->ghost_ports, devname);
2975     } else {
2976         sset_add(&ofproto->ports, devname);
2977     }
2978     return 0;
2979 }
2980
2981 static int
2982 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
2983 {
2984     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2985     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2986     int error = 0;
2987
2988     if (!ofport) {
2989         return 0;
2990     }
2991
2992     sset_find_and_delete(&ofproto->ghost_ports,
2993                          netdev_get_name(ofport->up.netdev));
2994     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2995     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
2996         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
2997         if (!error) {
2998             /* The caller is going to close ofport->up.netdev.  If this is a
2999              * bonded port, then the bond is using that netdev, so remove it
3000              * from the bond.  The client will need to reconfigure everything
3001              * after deleting ports, so then the slave will get re-added. */
3002             bundle_remove(&ofport->up);
3003         }
3004     }
3005     return error;
3006 }
3007
3008 static int
3009 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3010 {
3011     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3012     int error;
3013
3014     push_all_stats();
3015
3016     error = netdev_get_stats(ofport->up.netdev, stats);
3017
3018     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3019         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3020
3021         ovs_mutex_lock(&ofproto->stats_mutex);
3022         /* ofproto->stats.tx_packets represents packets that we created
3023          * internally and sent to some port (e.g. packets sent with
3024          * ofproto_dpif_send_packet()).  Account for them as if they had
3025          * come from OFPP_LOCAL and got forwarded. */
3026
3027         if (stats->rx_packets != UINT64_MAX) {
3028             stats->rx_packets += ofproto->stats.tx_packets;
3029         }
3030
3031         if (stats->rx_bytes != UINT64_MAX) {
3032             stats->rx_bytes += ofproto->stats.tx_bytes;
3033         }
3034
3035         /* ofproto->stats.rx_packets represents packets that were received on
3036          * some port and we processed internally and dropped (e.g. STP).
3037          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3038
3039         if (stats->tx_packets != UINT64_MAX) {
3040             stats->tx_packets += ofproto->stats.rx_packets;
3041         }
3042
3043         if (stats->tx_bytes != UINT64_MAX) {
3044             stats->tx_bytes += ofproto->stats.rx_bytes;
3045         }
3046         ovs_mutex_unlock(&ofproto->stats_mutex);
3047     }
3048
3049     return error;
3050 }
3051
3052 struct port_dump_state {
3053     uint32_t bucket;
3054     uint32_t offset;
3055     bool ghost;
3056
3057     struct ofproto_port port;
3058     bool has_port;
3059 };
3060
3061 static int
3062 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3063 {
3064     *statep = xzalloc(sizeof(struct port_dump_state));
3065     return 0;
3066 }
3067
3068 static int
3069 port_dump_next(const struct ofproto *ofproto_, void *state_,
3070                struct ofproto_port *port)
3071 {
3072     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3073     struct port_dump_state *state = state_;
3074     const struct sset *sset;
3075     struct sset_node *node;
3076
3077     if (state->has_port) {
3078         ofproto_port_destroy(&state->port);
3079         state->has_port = false;
3080     }
3081     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3082     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3083         int error;
3084
3085         error = port_query_by_name(ofproto_, node->name, &state->port);
3086         if (!error) {
3087             *port = state->port;
3088             state->has_port = true;
3089             return 0;
3090         } else if (error != ENODEV) {
3091             return error;
3092         }
3093     }
3094
3095     if (!state->ghost) {
3096         state->ghost = true;
3097         state->bucket = 0;
3098         state->offset = 0;
3099         return port_dump_next(ofproto_, state_, port);
3100     }
3101
3102     return EOF;
3103 }
3104
3105 static int
3106 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3107 {
3108     struct port_dump_state *state = state_;
3109
3110     if (state->has_port) {
3111         ofproto_port_destroy(&state->port);
3112     }
3113     free(state);
3114     return 0;
3115 }
3116
3117 static int
3118 port_poll(const struct ofproto *ofproto_, char **devnamep)
3119 {
3120     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3121
3122     if (ofproto->port_poll_errno) {
3123         int error = ofproto->port_poll_errno;
3124         ofproto->port_poll_errno = 0;
3125         return error;
3126     }
3127
3128     if (sset_is_empty(&ofproto->port_poll_set)) {
3129         return EAGAIN;
3130     }
3131
3132     *devnamep = sset_pop(&ofproto->port_poll_set);
3133     return 0;
3134 }
3135
3136 static void
3137 port_poll_wait(const struct ofproto *ofproto_)
3138 {
3139     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3140     dpif_port_poll_wait(ofproto->backer->dpif);
3141 }
3142
3143 static int
3144 port_is_lacp_current(const struct ofport *ofport_)
3145 {
3146     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3147     return (ofport->bundle && ofport->bundle->lacp
3148             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3149             : -1);
3150 }
3151 \f
3152 /* Upcall handling. */
3153
3154 struct flow_miss_op {
3155     struct dpif_op dpif_op;
3156
3157     uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3158     struct xlate_out xout;
3159     bool xout_garbage;           /* 'xout' needs to be uninitialized? */
3160
3161     struct ofpbuf mask;          /* Flow mask for "put" ops. */
3162     struct odputil_keybuf maskbuf;
3163
3164     /* If this is a "put" op, then a pointer to the subfacet that should
3165      * be marked as uninstalled if the operation fails. */
3166     struct subfacet *subfacet;
3167 };
3168
3169 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3170  * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3171  * and (usually) installing a datapath flow.  The answer is usually "yes" (a
3172  * return value of true).  However, for short flows the cost of bookkeeping is
3173  * much higher than the benefits, so when the datapath holds a large number of
3174  * flows we impose some heuristics to decide which flows are likely to be worth
3175  * tracking. */
3176 static bool
3177 flow_miss_should_make_facet(struct flow_miss *miss)
3178 {
3179     struct dpif_backer *backer = miss->ofproto->backer;
3180     uint32_t hash;
3181
3182     switch (flow_miss_model) {
3183     case OFPROTO_HANDLE_MISS_AUTO:
3184         break;
3185     case OFPROTO_HANDLE_MISS_WITH_FACETS:
3186         return true;
3187     case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3188         return false;
3189     }
3190
3191     if (!backer->governor) {
3192         size_t n_subfacets;
3193
3194         n_subfacets = hmap_count(&backer->subfacets);
3195         if (n_subfacets * 2 <= flow_eviction_threshold) {
3196             return true;
3197         }
3198
3199         backer->governor = governor_create();
3200     }
3201
3202     hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3203     return governor_should_install_flow(backer->governor, hash,
3204                                         miss->stats.n_packets);
3205 }
3206
3207 /* Handles 'miss', which matches 'facet'.  May add any required datapath
3208  * operations to 'ops', incrementing '*n_ops' for each new op.
3209  *
3210  * All of the packets in 'miss' are considered to have arrived at time
3211  * 'miss->stats.used'.  This is really important only for new facets: if we
3212  * just called time_msec() here, then the new subfacet or its packets could
3213  * look (occasionally) as though it was used some time after the facet was
3214  * used.  That can make a one-packet flow look like it has a nonzero duration,
3215  * which looks odd in e.g. NetFlow statistics. */
3216 static void
3217 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3218                             struct flow_miss_op *ops, size_t *n_ops)
3219 {
3220     enum subfacet_path want_path;
3221     struct subfacet *subfacet;
3222
3223     facet->packet_count += miss->stats.n_packets;
3224     facet->prev_packet_count += miss->stats.n_packets;
3225     facet->byte_count += miss->stats.n_bytes;
3226     facet->prev_byte_count += miss->stats.n_bytes;
3227
3228     want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3229
3230     /* Don't install the flow if it's the result of the "userspace"
3231      * action for an already installed facet.  This can occur when a
3232      * datapath flow with wildcards has a "userspace" action and flows
3233      * sent to userspace result in a different subfacet, which will then
3234      * be rejected as overlapping by the datapath. */
3235     if (miss->upcall_type == DPIF_UC_ACTION
3236         && !list_is_empty(&facet->subfacets)) {
3237         return;
3238     }
3239
3240     subfacet = subfacet_create(facet, miss);
3241     if (subfacet->path != want_path) {
3242         struct flow_miss_op *op = &ops[(*n_ops)++];
3243         struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3244
3245         subfacet->path = want_path;
3246
3247         ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3248         if (enable_megaflows) {
3249             odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3250                                    &miss->flow, UINT32_MAX);
3251         }
3252
3253         op->xout_garbage = false;
3254         op->dpif_op.type = DPIF_OP_FLOW_PUT;
3255         op->subfacet = subfacet;
3256         put->flags = DPIF_FP_CREATE;
3257         put->key = miss->key;
3258         put->key_len = miss->key_len;
3259         put->mask = op->mask.data;
3260         put->mask_len = op->mask.size;
3261
3262         if (want_path == SF_FAST_PATH) {
3263             put->actions = facet->xout.odp_actions.data;
3264             put->actions_len = facet->xout.odp_actions.size;
3265         } else {
3266             compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3267                               op->slow_stub, sizeof op->slow_stub,
3268                               &put->actions, &put->actions_len);
3269         }
3270         put->stats = NULL;
3271     }
3272 }
3273
3274 /* Handles flow miss 'miss'.  May add any required datapath operations
3275  * to 'ops', incrementing '*n_ops' for each new op. */
3276 static void
3277 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3278                  size_t *n_ops)
3279 {
3280     struct facet *facet;
3281
3282     miss->ofproto->n_missed += miss->stats.n_packets;
3283
3284     facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3285     if (!facet) {
3286         /* There does not exist a bijection between 'struct flow' and datapath
3287          * flow keys with fitness ODP_FIT_TO_LITTLE.  This breaks a fundamental
3288          * assumption used throughout the facet and subfacet handling code.
3289          * Since we have to handle these misses in userspace anyway, we simply
3290          * skip facet creation, avoiding the problem altogether. */
3291         if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3292             || !flow_miss_should_make_facet(miss)) {
3293             return;
3294         }
3295
3296         facet = facet_create(miss);
3297     }
3298     handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3299 }
3300
3301 static struct drop_key *
3302 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3303                 size_t key_len)
3304 {
3305     struct drop_key *drop_key;
3306
3307     HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3308                              &backer->drop_keys) {
3309         if (drop_key->key_len == key_len
3310             && !memcmp(drop_key->key, key, key_len)) {
3311             return drop_key;
3312         }
3313     }
3314     return NULL;
3315 }
3316
3317 static void
3318 drop_key_clear(struct dpif_backer *backer)
3319 {
3320     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3321     struct drop_key *drop_key, *next;
3322
3323     HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3324         int error;
3325
3326         error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3327                               NULL);
3328         if (error && !VLOG_DROP_WARN(&rl)) {
3329             struct ds ds = DS_EMPTY_INITIALIZER;
3330             odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3331             VLOG_WARN("Failed to delete drop key (%s) (%s)",
3332                       ovs_strerror(error), ds_cstr(&ds));
3333             ds_destroy(&ds);
3334         }
3335
3336         hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3337         drop_key_destroy(drop_key);
3338     }
3339
3340     udpif_drop_key_clear(backer->udpif);
3341 }
3342
3343 static void
3344 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3345 {
3346     struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3347     struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3348     struct flow_miss *miss;
3349     size_t n_ops, i;
3350
3351     /* Process each element in the to-do list, constructing the set of
3352      * operations to batch. */
3353     n_ops = 0;
3354     HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3355         handle_flow_miss(miss, flow_miss_ops, &n_ops);
3356     }
3357     ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3358
3359     /* Execute batch. */
3360     for (i = 0; i < n_ops; i++) {
3361         dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3362     }
3363     dpif_operate(backer->dpif, dpif_ops, n_ops);
3364
3365     for (i = 0; i < n_ops; i++) {
3366         if (dpif_ops[i]->error != 0
3367             && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3368             && flow_miss_ops[i].subfacet) {
3369             struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3370
3371             COVERAGE_INC(subfacet_install_fail);
3372
3373             /* Zero-out subfacet counters when installation failed, but
3374              * datapath reported hits.  This should not happen and
3375              * indicates a bug, since if the datapath flow exists, we
3376              * should not be attempting to create a new subfacet.  A
3377              * buggy datapath could trigger this, so just zero out the
3378              * counters and log an error. */
3379             if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3380                 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3381                             "datapath reported hits");
3382                 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3383             }
3384
3385             subfacet->path = SF_NOT_INSTALLED;
3386         }
3387     }
3388 }
3389
3390 static void
3391 handle_upcalls(struct dpif_backer *backer)
3392 {
3393     struct flow_miss_batch *fmb;
3394     int n_processed;
3395
3396     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3397         struct drop_key *drop_key = drop_key_next(backer->udpif);
3398         if (!drop_key) {
3399             break;
3400         }
3401
3402         if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3403             hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3404                         hash_bytes(drop_key->key, drop_key->key_len, 0));
3405             dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3406                           drop_key->key, drop_key->key_len,
3407                           NULL, 0, NULL, 0, NULL);
3408         } else {
3409             drop_key_destroy(drop_key);
3410         }
3411     }
3412
3413     fmb = flow_miss_batch_next(backer->udpif);
3414     if (fmb) {
3415         handle_flow_misses(backer, fmb);
3416         flow_miss_batch_destroy(fmb);
3417     }
3418 }
3419 \f
3420 /* Flow expiration. */
3421
3422 static int subfacet_max_idle(const struct dpif_backer *);
3423 static void update_stats(struct dpif_backer *);
3424 static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex);
3425 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3426
3427 /* This function is called periodically by run().  Its job is to collect
3428  * updates for the flows that have been installed into the datapath, most
3429  * importantly when they last were used, and then use that information to
3430  * expire flows that have not been used recently.
3431  *
3432  * Returns the number of milliseconds after which it should be called again. */
3433 static int
3434 expire(struct dpif_backer *backer)
3435 {
3436     struct ofproto_dpif *ofproto;
3437     size_t n_subfacets;
3438     int max_idle;
3439
3440     /* Periodically clear out the drop keys in an effort to keep them
3441      * relatively few. */
3442     drop_key_clear(backer);
3443
3444     /* Update stats for each flow in the backer. */
3445     update_stats(backer);
3446
3447     n_subfacets = hmap_count(&backer->subfacets);
3448     if (n_subfacets) {
3449         struct subfacet *subfacet;
3450         long long int total, now;
3451
3452         total = 0;
3453         now = time_msec();
3454         HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3455             total += now - subfacet->created;
3456         }
3457         backer->avg_subfacet_life += total / n_subfacets;
3458     }
3459     backer->avg_subfacet_life /= 2;
3460
3461     backer->avg_n_subfacet += n_subfacets;
3462     backer->avg_n_subfacet /= 2;
3463
3464     backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3465
3466     max_idle = subfacet_max_idle(backer);
3467     expire_subfacets(backer, max_idle);
3468
3469     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3470         struct rule *rule, *next_rule;
3471
3472         if (ofproto->backer != backer) {
3473             continue;
3474         }
3475
3476         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3477          * has passed. */
3478         ovs_mutex_lock(&ofproto_mutex);
3479         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3480                             &ofproto->up.expirable) {
3481             rule_expire(rule_dpif_cast(rule));
3482         }
3483         ovs_mutex_unlock(&ofproto_mutex);
3484
3485         /* All outstanding data in existing flows has been accounted, so it's a
3486          * good time to do bond rebalancing. */
3487         if (ofproto->has_bonded_bundles) {
3488             struct ofbundle *bundle;
3489
3490             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3491                 if (bundle->bond) {
3492                     bond_rebalance(bundle->bond);
3493                 }
3494             }
3495         }
3496     }
3497
3498     return MIN(max_idle, 1000);
3499 }
3500
3501 /* Updates flow table statistics given that the datapath just reported 'stats'
3502  * as 'subfacet''s statistics. */
3503 static void
3504 update_subfacet_stats(struct subfacet *subfacet,
3505                       const struct dpif_flow_stats *stats)
3506 {
3507     struct facet *facet = subfacet->facet;
3508     struct dpif_flow_stats diff;
3509
3510     diff.tcp_flags = stats->tcp_flags;
3511     diff.used = stats->used;
3512
3513     if (stats->n_packets >= subfacet->dp_packet_count) {
3514         diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3515     } else {
3516         VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3517         diff.n_packets = 0;
3518     }
3519
3520     if (stats->n_bytes >= subfacet->dp_byte_count) {
3521         diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3522     } else {
3523         VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3524         diff.n_bytes = 0;
3525     }
3526
3527     facet->ofproto->n_hit += diff.n_packets;
3528     subfacet->dp_packet_count = stats->n_packets;
3529     subfacet->dp_byte_count = stats->n_bytes;
3530     subfacet_update_stats(subfacet, &diff);
3531
3532     if (facet->accounted_bytes < facet->byte_count) {
3533         facet_learn(facet);
3534         facet_account(facet);
3535         facet->accounted_bytes = facet->byte_count;
3536     }
3537 }
3538
3539 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3540  * about, or a flow that shouldn't be installed but was anyway.  Delete it. */
3541 static void
3542 delete_unexpected_flow(struct dpif_backer *backer,
3543                        const struct nlattr *key, size_t key_len)
3544 {
3545     if (!VLOG_DROP_WARN(&rl)) {
3546         struct ds s;
3547
3548         ds_init(&s);
3549         odp_flow_key_format(key, key_len, &s);
3550         VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3551         ds_destroy(&s);
3552     }
3553
3554     COVERAGE_INC(facet_unexpected);
3555     dpif_flow_del(backer->dpif, key, key_len, NULL);
3556 }
3557
3558 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3559  *
3560  * This function also pushes statistics updates to rules which each facet
3561  * resubmits into.  Generally these statistics will be accurate.  However, if a
3562  * facet changes the rule it resubmits into at some time in between
3563  * update_stats() runs, it is possible that statistics accrued to the
3564  * old rule will be incorrectly attributed to the new rule.  This could be
3565  * avoided by calling update_stats() whenever rules are created or
3566  * deleted.  However, the performance impact of making so many calls to the
3567  * datapath do not justify the benefit of having perfectly accurate statistics.
3568  *
3569  * In addition, this function maintains per ofproto flow hit counts. The patch
3570  * port is not treated specially. e.g. A packet ingress from br0 patched into
3571  * br1 will increase the hit count of br0 by 1, however, does not affect
3572  * the hit or miss counts of br1.
3573  */
3574 static void
3575 update_stats(struct dpif_backer *backer)
3576 {
3577     const struct dpif_flow_stats *stats;
3578     struct dpif_flow_dump dump;
3579     const struct nlattr *key, *mask;
3580     size_t key_len, mask_len;
3581
3582     dpif_flow_dump_start(&dump, backer->dpif);
3583     while (dpif_flow_dump_next(&dump, &key, &key_len,
3584                                &mask, &mask_len, NULL, NULL, &stats)) {
3585         struct subfacet *subfacet;
3586         uint32_t key_hash;
3587
3588         key_hash = odp_flow_key_hash(key, key_len);
3589         subfacet = subfacet_find(backer, key, key_len, key_hash);
3590         switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3591         case SF_FAST_PATH:
3592             update_subfacet_stats(subfacet, stats);
3593             break;
3594
3595         case SF_SLOW_PATH:
3596             /* Stats are updated per-packet. */
3597             break;
3598
3599         case SF_NOT_INSTALLED:
3600         default:
3601             delete_unexpected_flow(backer, key, key_len);
3602             break;
3603         }
3604         run_fast_rl();
3605     }
3606     dpif_flow_dump_done(&dump);
3607 }
3608
3609 /* Calculates and returns the number of milliseconds of idle time after which
3610  * subfacets should expire from the datapath.  When a subfacet expires, we fold
3611  * its statistics into its facet, and when a facet's last subfacet expires, we
3612  * fold its statistic into its rule. */
3613 static int
3614 subfacet_max_idle(const struct dpif_backer *backer)
3615 {
3616     /*
3617      * Idle time histogram.
3618      *
3619      * Most of the time a switch has a relatively small number of subfacets.
3620      * When this is the case we might as well keep statistics for all of them
3621      * in userspace and to cache them in the kernel datapath for performance as
3622      * well.
3623      *
3624      * As the number of subfacets increases, the memory required to maintain
3625      * statistics about them in userspace and in the kernel becomes
3626      * significant.  However, with a large number of subfacets it is likely
3627      * that only a few of them are "heavy hitters" that consume a large amount
3628      * of bandwidth.  At this point, only heavy hitters are worth caching in
3629      * the kernel and maintaining in userspaces; other subfacets we can
3630      * discard.
3631      *
3632      * The technique used to compute the idle time is to build a histogram with
3633      * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each.  Each subfacet
3634      * that is installed in the kernel gets dropped in the appropriate bucket.
3635      * After the histogram has been built, we compute the cutoff so that only
3636      * the most-recently-used 1% of subfacets (but at least
3637      * flow_eviction_threshold flows) are kept cached.  At least
3638      * the most-recently-used bucket of subfacets is kept, so actually an
3639      * arbitrary number of subfacets can be kept in any given expiration run
3640      * (though the next run will delete most of those unless they receive
3641      * additional data).
3642      *
3643      * This requires a second pass through the subfacets, in addition to the
3644      * pass made by update_stats(), because the former function never looks at
3645      * uninstallable subfacets.
3646      */
3647     enum { BUCKET_WIDTH = 100 };
3648     enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3649     int buckets[N_BUCKETS] = { 0 };
3650     int total, subtotal, bucket;
3651     struct subfacet *subfacet;
3652     long long int now;
3653     int i;
3654
3655     total = hmap_count(&backer->subfacets);
3656     if (total <= flow_eviction_threshold) {
3657         return N_BUCKETS * BUCKET_WIDTH;
3658     }
3659
3660     /* Build histogram. */
3661     now = time_msec();
3662     HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3663         long long int idle = now - subfacet->used;
3664         int bucket = (idle <= 0 ? 0
3665                       : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3666                       : (unsigned int) idle / BUCKET_WIDTH);
3667         buckets[bucket]++;
3668     }
3669
3670     /* Find the first bucket whose flows should be expired. */
3671     subtotal = bucket = 0;
3672     do {
3673         subtotal += buckets[bucket++];
3674     } while (bucket < N_BUCKETS &&
3675              subtotal < MAX(flow_eviction_threshold, total / 100));
3676
3677     if (VLOG_IS_DBG_ENABLED()) {
3678         struct ds s;
3679
3680         ds_init(&s);
3681         ds_put_cstr(&s, "keep");
3682         for (i = 0; i < N_BUCKETS; i++) {
3683             if (i == bucket) {
3684                 ds_put_cstr(&s, ", drop");
3685             }
3686             if (buckets[i]) {
3687                 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3688             }
3689         }
3690         VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3691         ds_destroy(&s);
3692     }
3693
3694     return bucket * BUCKET_WIDTH;
3695 }
3696
3697 static void
3698 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3699 {
3700     /* Cutoff time for most flows. */
3701     long long int normal_cutoff = time_msec() - dp_max_idle;
3702
3703     /* We really want to keep flows for special protocols around, so use a more
3704      * conservative cutoff. */
3705     long long int special_cutoff = time_msec() - 10000;
3706
3707     struct subfacet *subfacet, *next_subfacet;
3708     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3709     int n_batch;
3710
3711     n_batch = 0;
3712     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3713                         &backer->subfacets) {
3714         long long int cutoff;
3715
3716         cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3717                                                 | SLOW_STP)
3718                   ? special_cutoff
3719                   : normal_cutoff);
3720         if (subfacet->used < cutoff) {
3721             if (subfacet->path != SF_NOT_INSTALLED) {
3722                 batch[n_batch++] = subfacet;
3723                 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3724                     subfacet_destroy_batch(backer, batch, n_batch);
3725                     n_batch = 0;
3726                 }
3727             } else {
3728                 subfacet_destroy(subfacet);
3729             }
3730         }
3731     }
3732
3733     if (n_batch > 0) {
3734         subfacet_destroy_batch(backer, batch, n_batch);
3735     }
3736 }
3737
3738 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3739  * then delete it entirely. */
3740 static void
3741 rule_expire(struct rule_dpif *rule)
3742     OVS_REQUIRES(ofproto_mutex)
3743 {
3744     uint16_t idle_timeout, hard_timeout;
3745     long long int now = time_msec();
3746     int reason;
3747
3748     ovs_assert(!rule->up.pending);
3749
3750     /* Has 'rule' expired? */
3751     ovs_mutex_lock(&rule->up.mutex);
3752     hard_timeout = rule->up.hard_timeout;
3753     idle_timeout = rule->up.idle_timeout;
3754     if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3755         reason = OFPRR_HARD_TIMEOUT;
3756     } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3757         reason = OFPRR_IDLE_TIMEOUT;
3758     } else {
3759         reason = -1;
3760     }
3761     ovs_mutex_unlock(&rule->up.mutex);
3762
3763     if (reason >= 0) {
3764         COVERAGE_INC(ofproto_dpif_expired);
3765         ofproto_rule_expire(&rule->up, reason);
3766     }
3767 }
3768 \f
3769 /* Facets. */
3770
3771 /* Creates and returns a new facet based on 'miss'.
3772  *
3773  * The caller must already have determined that no facet with an identical
3774  * 'miss->flow' exists in 'miss->ofproto'.
3775  *
3776  * 'rule' and 'xout' must have been created based on 'miss'.
3777  *
3778  * 'facet'' statistics are initialized based on 'stats'.
3779  *
3780  * The facet will initially have no subfacets.  The caller should create (at
3781  * least) one subfacet with subfacet_create(). */
3782 static struct facet *
3783 facet_create(const struct flow_miss *miss)
3784 {
3785     struct ofproto_dpif *ofproto = miss->ofproto;
3786     struct facet *facet;
3787     struct match match;
3788
3789     COVERAGE_INC(facet_create);
3790     facet = xzalloc(sizeof *facet);
3791     facet->ofproto = miss->ofproto;
3792     facet->used = miss->stats.used;
3793     facet->flow = miss->flow;
3794     facet->learn_rl = time_msec() + 500;
3795
3796     list_init(&facet->subfacets);
3797     netflow_flow_init(&facet->nf_flow);
3798     netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3799
3800     xlate_out_copy(&facet->xout, &miss->xout);
3801
3802     match_init(&match, &facet->flow, &facet->xout.wc);
3803     cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3804     ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3805     classifier_insert(&ofproto->facets, &facet->cr);
3806     ovs_rwlock_unlock(&ofproto->facets.rwlock);
3807
3808     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
3809     return facet;
3810 }
3811
3812 static void
3813 facet_free(struct facet *facet)
3814 {
3815     if (facet) {
3816         xlate_out_uninit(&facet->xout);
3817         free(facet);
3818     }
3819 }
3820
3821 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3822  * 'flow' must reflect the data in 'packet'. */
3823 int
3824 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3825                              const struct flow *flow,
3826                              struct rule_dpif *rule,
3827                              const struct ofpact *ofpacts, size_t ofpacts_len,
3828                              struct ofpbuf *packet)
3829 {
3830     struct odputil_keybuf keybuf;
3831     struct dpif_flow_stats stats;
3832     struct xlate_out xout;
3833     struct xlate_in xin;
3834     ofp_port_t in_port;
3835     struct ofpbuf key;
3836     int error;
3837
3838     ovs_assert((rule != NULL) != (ofpacts != NULL));
3839
3840     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3841     if (rule) {
3842         rule_dpif_credit_stats(rule, &stats);
3843     }
3844
3845     xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
3846     xin.ofpacts = ofpacts;
3847     xin.ofpacts_len = ofpacts_len;
3848     xin.resubmit_stats = &stats;
3849     xlate_actions(&xin, &xout);
3850
3851     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3852     in_port = flow->in_port.ofp_port;
3853     if (in_port == OFPP_NONE) {
3854         in_port = OFPP_LOCAL;
3855     }
3856     odp_flow_key_from_flow(&key, flow, ofp_port_to_odp_port(ofproto, in_port));
3857
3858     error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
3859                          xout.odp_actions.data, xout.odp_actions.size, packet,
3860                          (xout.slow & SLOW_ACTION) != 0);
3861     xlate_out_uninit(&xout);
3862
3863     return error;
3864 }
3865
3866 /* Remove 'facet' from its ofproto and free up the associated memory:
3867  *
3868  *   - If 'facet' was installed in the datapath, uninstalls it and updates its
3869  *     rule's statistics, via subfacet_uninstall().
3870  *
3871  *   - Removes 'facet' from its rule and from ofproto->facets.
3872  */
3873 static void
3874 facet_remove(struct facet *facet)
3875 {
3876     struct subfacet *subfacet, *next_subfacet;
3877
3878     COVERAGE_INC(facet_remove);
3879     ovs_assert(!list_is_empty(&facet->subfacets));
3880
3881     /* First uninstall all of the subfacets to get final statistics. */
3882     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3883         subfacet_uninstall(subfacet);
3884     }
3885
3886     /* Flush the final stats to the rule.
3887      *
3888      * This might require us to have at least one subfacet around so that we
3889      * can use its actions for accounting in facet_account(), which is why we
3890      * have uninstalled but not yet destroyed the subfacets. */
3891     facet_flush_stats(facet);
3892
3893     /* Now we're really all done so destroy everything. */
3894     LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3895                         &facet->subfacets) {
3896         subfacet_destroy__(subfacet);
3897     }
3898     ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
3899     classifier_remove(&facet->ofproto->facets, &facet->cr);
3900     ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
3901     cls_rule_destroy(&facet->cr);
3902     facet_free(facet);
3903 }
3904
3905 /* Feed information from 'facet' back into the learning table to keep it in
3906  * sync with what is actually flowing through the datapath. */
3907 static void
3908 facet_learn(struct facet *facet)
3909 {
3910     long long int now = time_msec();
3911
3912     if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
3913         return;
3914     }
3915
3916     facet->learn_rl = now + 500;
3917
3918     if (!facet->xout.has_learn
3919         && !facet->xout.has_normal
3920         && (!facet->xout.has_fin_timeout
3921             || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3922         return;
3923     }
3924
3925     facet_push_stats(facet, true);
3926 }
3927
3928 static void
3929 facet_account(struct facet *facet)
3930 {
3931     const struct nlattr *a;
3932     unsigned int left;
3933     ovs_be16 vlan_tci;
3934     uint64_t n_bytes;
3935
3936     if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
3937         return;
3938     }
3939     n_bytes = facet->byte_count - facet->accounted_bytes;
3940
3941     /* This loop feeds byte counters to bond_account() for rebalancing to use
3942      * as a basis.  We also need to track the actual VLAN on which the packet
3943      * is going to be sent to ensure that it matches the one passed to
3944      * bond_choose_output_slave().  (Otherwise, we will account to the wrong
3945      * hash bucket.)
3946      *
3947      * We use the actions from an arbitrary subfacet because they should all
3948      * be equally valid for our purpose. */
3949     vlan_tci = facet->flow.vlan_tci;
3950     NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
3951                              facet->xout.odp_actions.size) {
3952         const struct ovs_action_push_vlan *vlan;
3953         struct ofport_dpif *port;
3954
3955         switch (nl_attr_type(a)) {
3956         case OVS_ACTION_ATTR_OUTPUT:
3957             port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
3958             if (port && port->bundle && port->bundle->bond) {
3959                 bond_account(port->bundle->bond, &facet->flow,
3960                              vlan_tci_to_vid(vlan_tci), n_bytes);
3961             }
3962             break;
3963
3964         case OVS_ACTION_ATTR_POP_VLAN:
3965             vlan_tci = htons(0);
3966             break;
3967
3968         case OVS_ACTION_ATTR_PUSH_VLAN:
3969             vlan = nl_attr_get(a);
3970             vlan_tci = vlan->vlan_tci;
3971             break;
3972         }
3973     }
3974 }
3975
3976 /* Returns true if the only action for 'facet' is to send to the controller.
3977  * (We don't report NetFlow expiration messages for such facets because they
3978  * are just part of the control logic for the network, not real traffic). */
3979 static bool
3980 facet_is_controller_flow(struct facet *facet)
3981 {
3982     if (facet) {
3983         struct ofproto_dpif *ofproto = facet->ofproto;
3984         const struct ofpact *ofpacts;
3985         struct rule_actions *actions;
3986         struct rule_dpif *rule;
3987         size_t ofpacts_len;
3988         bool is_controller;
3989
3990         rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
3991         actions = rule_dpif_get_actions(rule);
3992         rule_dpif_unref(rule);
3993
3994         ofpacts_len = actions->ofpacts_len;
3995         ofpacts = actions->ofpacts;
3996         is_controller = ofpacts_len > 0
3997             && ofpacts->type == OFPACT_CONTROLLER
3998             && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
3999         rule_actions_unref(actions);
4000
4001         return is_controller;
4002     }
4003     return false;
4004 }
4005
4006 /* Folds all of 'facet''s statistics into its rule.  Also updates the
4007  * accounting ofhook and emits a NetFlow expiration if appropriate.  All of
4008  * 'facet''s statistics in the datapath should have been zeroed and folded into
4009  * its packet and byte counts before this function is called. */
4010 static void
4011 facet_flush_stats(struct facet *facet)
4012 {
4013     struct ofproto_dpif *ofproto = facet->ofproto;
4014     struct subfacet *subfacet;
4015
4016     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4017         ovs_assert(!subfacet->dp_byte_count);
4018         ovs_assert(!subfacet->dp_packet_count);
4019     }
4020
4021     facet_push_stats(facet, false);
4022     if (facet->accounted_bytes < facet->byte_count) {
4023         facet_account(facet);
4024         facet->accounted_bytes = facet->byte_count;
4025     }
4026
4027     if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4028         struct ofexpired expired;
4029         expired.flow = facet->flow;
4030         expired.packet_count = facet->packet_count;
4031         expired.byte_count = facet->byte_count;
4032         expired.used = facet->used;
4033         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4034     }
4035
4036     /* Reset counters to prevent double counting if 'facet' ever gets
4037      * reinstalled. */
4038     facet_reset_counters(facet);
4039
4040     netflow_flow_clear(&facet->nf_flow);
4041     facet->tcp_flags = 0;
4042 }
4043
4044 /* Searches 'ofproto''s table of facets for one which would be responsible for
4045  * 'flow'.  Returns it if found, otherwise a null pointer.
4046  *
4047  * The returned facet might need revalidation; use facet_lookup_valid()
4048  * instead if that is important. */
4049 static struct facet *
4050 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4051 {
4052     struct cls_rule *cr;
4053
4054     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4055     cr = classifier_lookup(&ofproto->facets, flow, NULL);
4056     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4057     return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4058 }
4059
4060 /* Searches 'ofproto''s table of facets for one capable that covers
4061  * 'flow'.  Returns it if found, otherwise a null pointer.
4062  *
4063  * The returned facet is guaranteed to be valid. */
4064 static struct facet *
4065 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4066 {
4067     struct facet *facet;
4068
4069     facet = facet_find(ofproto, flow);
4070     if (facet
4071         && ofproto->backer->need_revalidate
4072         && !facet_revalidate(facet)) {
4073         return NULL;
4074     }
4075
4076     return facet;
4077 }
4078
4079 static bool
4080 facet_check_consistency(struct facet *facet)
4081 {
4082     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4083
4084     struct xlate_out xout;
4085     struct xlate_in xin;
4086     bool ok;
4087
4088     /* Check the datapath actions for consistency. */
4089     xlate_in_init(&xin, facet->ofproto, &facet->flow, NULL, 0, NULL);
4090     xlate_actions(&xin, &xout);
4091
4092     ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4093         && facet->xout.slow == xout.slow;
4094     if (!ok && !VLOG_DROP_WARN(&rl)) {
4095         struct ds s = DS_EMPTY_INITIALIZER;
4096
4097         flow_format(&s, &facet->flow);
4098         ds_put_cstr(&s, ": inconsistency in facet");
4099
4100         if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4101             ds_put_cstr(&s, " (actions were: ");
4102             format_odp_actions(&s, facet->xout.odp_actions.data,
4103                                facet->xout.odp_actions.size);
4104             ds_put_cstr(&s, ") (correct actions: ");
4105             format_odp_actions(&s, xout.odp_actions.data,
4106                                xout.odp_actions.size);
4107             ds_put_char(&s, ')');
4108         }
4109
4110         if (facet->xout.slow != xout.slow) {
4111             ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4112         }
4113
4114         ds_destroy(&s);
4115     }
4116     xlate_out_uninit(&xout);
4117
4118     return ok;
4119 }
4120
4121 /* Re-searches the classifier for 'facet':
4122  *
4123  *   - If the rule found is different from 'facet''s current rule, moves
4124  *     'facet' to the new rule and recompiles its actions.
4125  *
4126  *   - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4127  *     where it is and recompiles its actions anyway.
4128  *
4129  *   - If any of 'facet''s subfacets correspond to a new flow according to
4130  *     xlate_receive(), 'facet' is removed.
4131  *
4132  *   Returns true if 'facet' is still valid.  False if 'facet' was removed. */
4133 static bool
4134 facet_revalidate(struct facet *facet)
4135 {
4136     struct ofproto_dpif *ofproto = facet->ofproto;
4137     struct rule_dpif *new_rule;
4138     struct subfacet *subfacet;
4139     struct flow_wildcards wc;
4140     struct xlate_out xout;
4141     struct xlate_in xin;
4142
4143     COVERAGE_INC(facet_revalidate);
4144
4145     /* Check that child subfacets still correspond to this facet.  Tunnel
4146      * configuration changes could cause a subfacet's OpenFlow in_port to
4147      * change. */
4148     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4149         struct ofproto_dpif *recv_ofproto;
4150         struct flow recv_flow;
4151         int error;
4152
4153         error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4154                               subfacet->key_len, &recv_flow, NULL,
4155                               &recv_ofproto, NULL);
4156         if (error
4157             || recv_ofproto != ofproto
4158             || facet != facet_find(ofproto, &recv_flow)) {
4159             facet_remove(facet);
4160             return false;
4161         }
4162     }
4163
4164     flow_wildcards_init_catchall(&wc);
4165     rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4166
4167     /* Calculate new datapath actions.
4168      *
4169      * We do not modify any 'facet' state yet, because we might need to, e.g.,
4170      * emit a NetFlow expiration and, if so, we need to have the old state
4171      * around to properly compose it. */
4172     xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4173     xlate_actions(&xin, &xout);
4174     flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4175
4176     /* A facet's slow path reason should only change under dramatic
4177      * circumstances.  Rather than try to update everything, it's simpler to
4178      * remove the facet and start over.
4179      *
4180      * More importantly, if a facet's wildcards change, it will be relatively
4181      * difficult to figure out if its subfacets still belong to it, and if not
4182      * which facet they may belong to.  Again, to avoid the complexity, we
4183      * simply give up instead. */
4184     if (facet->xout.slow != xout.slow
4185         || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4186         facet_remove(facet);
4187         xlate_out_uninit(&xout);
4188         rule_dpif_unref(new_rule);
4189         return false;
4190     }
4191
4192     if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4193         LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4194             if (subfacet->path == SF_FAST_PATH) {
4195                 struct dpif_flow_stats stats;
4196
4197                 subfacet_install(subfacet, &xout.odp_actions, &stats);
4198                 subfacet_update_stats(subfacet, &stats);
4199             }
4200         }
4201
4202         facet_flush_stats(facet);
4203
4204         ofpbuf_clear(&facet->xout.odp_actions);
4205         ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4206                    xout.odp_actions.size);
4207     }
4208
4209     /* Update 'facet' now that we've taken care of all the old state. */
4210     facet->xout.slow = xout.slow;
4211     facet->xout.has_learn = xout.has_learn;
4212     facet->xout.has_normal = xout.has_normal;
4213     facet->xout.has_fin_timeout = xout.has_fin_timeout;
4214     facet->xout.nf_output_iface = xout.nf_output_iface;
4215     facet->xout.mirrors = xout.mirrors;
4216     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4217
4218     ovs_mutex_lock(&new_rule->up.mutex);
4219     facet->used = MAX(facet->used, new_rule->up.created);
4220     ovs_mutex_unlock(&new_rule->up.mutex);
4221
4222     xlate_out_uninit(&xout);
4223     rule_dpif_unref(new_rule);
4224     return true;
4225 }
4226
4227 static void
4228 facet_reset_counters(struct facet *facet)
4229 {
4230     facet->packet_count = 0;
4231     facet->byte_count = 0;
4232     facet->prev_packet_count = 0;
4233     facet->prev_byte_count = 0;
4234     facet->accounted_bytes = 0;
4235 }
4236
4237 static void
4238 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4239                 struct dpif_flow_stats *stats, bool may_learn)
4240 {
4241     struct ofport_dpif *in_port;
4242     struct xlate_in xin;
4243
4244     in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4245     if (in_port && in_port->is_tunnel) {
4246         netdev_vport_inc_rx(in_port->up.netdev, stats);
4247     }
4248
4249     xlate_in_init(&xin, ofproto, flow, NULL, stats->tcp_flags, NULL);
4250     xin.resubmit_stats = stats;
4251     xin.may_learn = may_learn;
4252     xlate_actions_for_side_effects(&xin);
4253 }
4254
4255 static void
4256 facet_push_stats(struct facet *facet, bool may_learn)
4257 {
4258     struct dpif_flow_stats stats;
4259
4260     ovs_assert(facet->packet_count >= facet->prev_packet_count);
4261     ovs_assert(facet->byte_count >= facet->prev_byte_count);
4262     ovs_assert(facet->used >= facet->prev_used);
4263
4264     stats.n_packets = facet->packet_count - facet->prev_packet_count;
4265     stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4266     stats.used = facet->used;
4267     stats.tcp_flags = facet->tcp_flags;
4268
4269     if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4270         facet->prev_packet_count = facet->packet_count;
4271         facet->prev_byte_count = facet->byte_count;
4272         facet->prev_used = facet->used;
4273
4274         netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4275                                  facet->used);
4276         netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4277         mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4278                             stats.n_packets, stats.n_bytes);
4279         flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4280     }
4281 }
4282
4283 static void
4284 push_all_stats__(bool run_fast)
4285 {
4286     static long long int rl = LLONG_MIN;
4287     struct ofproto_dpif *ofproto;
4288
4289     if (time_msec() < rl) {
4290         return;
4291     }
4292
4293     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4294         struct cls_cursor cursor;
4295         struct facet *facet;
4296
4297         ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4298         cls_cursor_init(&cursor, &ofproto->facets, NULL);
4299         CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4300             facet_push_stats(facet, false);
4301             if (run_fast) {
4302                 run_fast_rl();
4303             }
4304         }
4305         ovs_rwlock_unlock(&ofproto->facets.rwlock);
4306     }
4307
4308     rl = time_msec() + 100;
4309 }
4310
4311 static void
4312 push_all_stats(void)
4313 {
4314     push_all_stats__(true);
4315 }
4316
4317 void
4318 rule_dpif_credit_stats(struct rule_dpif *rule,
4319                        const struct dpif_flow_stats *stats)
4320 {
4321     ovs_mutex_lock(&rule->stats_mutex);
4322     rule->packet_count += stats->n_packets;
4323     rule->byte_count += stats->n_bytes;
4324     rule->up.used = MAX(rule->up.used, stats->used);
4325     ovs_mutex_unlock(&rule->stats_mutex);
4326 }
4327
4328 bool
4329 rule_dpif_is_fail_open(const struct rule_dpif *rule)
4330 {
4331     return is_fail_open_rule(&rule->up);
4332 }
4333
4334 bool
4335 rule_dpif_is_table_miss(const struct rule_dpif *rule)
4336 {
4337     return rule_is_table_miss(&rule->up);
4338 }
4339
4340 ovs_be64
4341 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4342     OVS_REQUIRES(rule->up.mutex)
4343 {
4344     return rule->up.flow_cookie;
4345 }
4346
4347 void
4348 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4349                      uint16_t hard_timeout)
4350 {
4351     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4352 }
4353
4354 /* Returns 'rule''s actions.  The caller owns a reference on the returned
4355  * actions and must eventually release it (with rule_actions_unref()) to avoid
4356  * a memory leak. */
4357 struct rule_actions *
4358 rule_dpif_get_actions(const struct rule_dpif *rule)
4359 {
4360     return rule_get_actions(&rule->up);
4361 }
4362 \f
4363 /* Subfacets. */
4364
4365 static struct subfacet *
4366 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4367               size_t key_len, uint32_t key_hash)
4368 {
4369     struct subfacet *subfacet;
4370
4371     HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4372                              &backer->subfacets) {
4373         if (subfacet->key_len == key_len
4374             && !memcmp(key, subfacet->key, key_len)) {
4375             return subfacet;
4376         }
4377     }
4378
4379     return NULL;
4380 }
4381
4382 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4383  * 'key_fitness', 'key', and 'key_len' members in 'miss'.  Returns the
4384  * existing subfacet if there is one, otherwise creates and returns a
4385  * new subfacet. */
4386 static struct subfacet *
4387 subfacet_create(struct facet *facet, struct flow_miss *miss)
4388 {
4389     struct dpif_backer *backer = miss->ofproto->backer;
4390     const struct nlattr *key = miss->key;
4391     size_t key_len = miss->key_len;
4392     uint32_t key_hash;
4393     struct subfacet *subfacet;
4394
4395     key_hash = odp_flow_key_hash(key, key_len);
4396
4397     if (list_is_empty(&facet->subfacets)) {
4398         subfacet = &facet->one_subfacet;
4399     } else {
4400         subfacet = subfacet_find(backer, key, key_len, key_hash);
4401         if (subfacet) {
4402             if (subfacet->facet == facet) {
4403                 return subfacet;
4404             }
4405
4406             /* This shouldn't happen. */
4407             VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4408             subfacet_destroy(subfacet);
4409         }
4410
4411         subfacet = xmalloc(sizeof *subfacet);
4412     }
4413
4414     COVERAGE_INC(subfacet_create);
4415     hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4416     list_push_back(&facet->subfacets, &subfacet->list_node);
4417     subfacet->facet = facet;
4418     subfacet->key = xmemdup(key, key_len);
4419     subfacet->key_len = key_len;
4420     subfacet->used = miss->stats.used;
4421     subfacet->created = subfacet->used;
4422     subfacet->dp_packet_count = 0;
4423     subfacet->dp_byte_count = 0;
4424     subfacet->path = SF_NOT_INSTALLED;
4425     subfacet->backer = backer;
4426
4427     return subfacet;
4428 }
4429
4430 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4431  * its facet within 'ofproto', and frees it. */
4432 static void
4433 subfacet_destroy__(struct subfacet *subfacet)
4434 {
4435     struct facet *facet = subfacet->facet;
4436
4437     COVERAGE_INC(subfacet_destroy);
4438     subfacet_uninstall(subfacet);
4439     hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4440     list_remove(&subfacet->list_node);
4441     free(subfacet->key);
4442     if (subfacet != &facet->one_subfacet) {
4443         free(subfacet);
4444     }
4445 }
4446
4447 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4448  * last remaining subfacet in its facet destroys the facet too. */
4449 static void
4450 subfacet_destroy(struct subfacet *subfacet)
4451 {
4452     struct facet *facet = subfacet->facet;
4453
4454     if (list_is_singleton(&facet->subfacets)) {
4455         /* facet_remove() needs at least one subfacet (it will remove it). */
4456         facet_remove(facet);
4457     } else {
4458         subfacet_destroy__(subfacet);
4459     }
4460 }
4461
4462 static void
4463 subfacet_destroy_batch(struct dpif_backer *backer,
4464                        struct subfacet **subfacets, int n)
4465 {
4466     struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4467     struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4468     struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4469     int i;
4470
4471     for (i = 0; i < n; i++) {
4472         ops[i].type = DPIF_OP_FLOW_DEL;
4473         ops[i].u.flow_del.key = subfacets[i]->key;
4474         ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4475         ops[i].u.flow_del.stats = &stats[i];
4476         opsp[i] = &ops[i];
4477     }
4478
4479     dpif_operate(backer->dpif, opsp, n);
4480     for (i = 0; i < n; i++) {
4481         subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4482         subfacets[i]->path = SF_NOT_INSTALLED;
4483         subfacet_destroy(subfacets[i]);
4484         run_fast_rl();
4485     }
4486 }
4487
4488 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4489  * bytes of actions in 'actions'.  If 'stats' is non-null, statistics counters
4490  * in the datapath will be zeroed and 'stats' will be updated with traffic new
4491  * since 'subfacet' was last updated.
4492  *
4493  * Returns 0 if successful, otherwise a positive errno value. */
4494 static int
4495 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4496                  struct dpif_flow_stats *stats)
4497 {
4498     struct facet *facet = subfacet->facet;
4499     enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4500     const struct nlattr *actions = odp_actions->data;
4501     size_t actions_len = odp_actions->size;
4502     struct odputil_keybuf maskbuf;
4503     struct ofpbuf mask;
4504
4505     uint64_t slow_path_stub[128 / 8];
4506     enum dpif_flow_put_flags flags;
4507     int ret;
4508
4509     flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4510                                                : DPIF_FP_MODIFY;
4511     if (stats) {
4512         flags |= DPIF_FP_ZERO_STATS;
4513     }
4514
4515     if (path == SF_SLOW_PATH) {
4516         compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4517                           slow_path_stub, sizeof slow_path_stub,
4518                           &actions, &actions_len);
4519     }
4520
4521     ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4522     if (enable_megaflows) {
4523         odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4524                                &facet->flow, UINT32_MAX);
4525     }
4526
4527     ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4528                         subfacet->key_len,  mask.data, mask.size,
4529                         actions, actions_len, stats);
4530
4531     if (stats) {
4532         subfacet_reset_dp_stats(subfacet, stats);
4533     }
4534
4535     if (ret) {
4536         COVERAGE_INC(subfacet_install_fail);
4537     } else {
4538         subfacet->path = path;
4539     }
4540     return ret;
4541 }
4542
4543 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4544 static void
4545 subfacet_uninstall(struct subfacet *subfacet)
4546 {
4547     if (subfacet->path != SF_NOT_INSTALLED) {
4548         struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4549         struct dpif_flow_stats stats;
4550         int error;
4551
4552         error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4553                               subfacet->key_len, &stats);
4554         subfacet_reset_dp_stats(subfacet, &stats);
4555         if (!error) {
4556             subfacet_update_stats(subfacet, &stats);
4557         }
4558         subfacet->path = SF_NOT_INSTALLED;
4559     } else {
4560         ovs_assert(subfacet->dp_packet_count == 0);
4561         ovs_assert(subfacet->dp_byte_count == 0);
4562     }
4563 }
4564
4565 /* Resets 'subfacet''s datapath statistics counters.  This should be called
4566  * when 'subfacet''s statistics are cleared in the datapath.  If 'stats' is
4567  * non-null, it should contain the statistics returned by dpif when 'subfacet'
4568  * was reset in the datapath.  'stats' will be modified to include only
4569  * statistics new since 'subfacet' was last updated. */
4570 static void
4571 subfacet_reset_dp_stats(struct subfacet *subfacet,
4572                         struct dpif_flow_stats *stats)
4573 {
4574     if (stats
4575         && subfacet->dp_packet_count <= stats->n_packets
4576         && subfacet->dp_byte_count <= stats->n_bytes) {
4577         stats->n_packets -= subfacet->dp_packet_count;
4578         stats->n_bytes -= subfacet->dp_byte_count;
4579     }
4580
4581     subfacet->dp_packet_count = 0;
4582     subfacet->dp_byte_count = 0;
4583 }
4584
4585 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4586  *
4587  * Because of the meaning of a subfacet's counters, it only makes sense to do
4588  * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4589  * represents a packet that was sent by hand or if it represents statistics
4590  * that have been cleared out of the datapath. */
4591 static void
4592 subfacet_update_stats(struct subfacet *subfacet,
4593                       const struct dpif_flow_stats *stats)
4594 {
4595     if (stats->n_packets || stats->used > subfacet->used) {
4596         struct facet *facet = subfacet->facet;
4597
4598         subfacet->used = MAX(subfacet->used, stats->used);
4599         facet->used = MAX(facet->used, stats->used);
4600         facet->packet_count += stats->n_packets;
4601         facet->byte_count += stats->n_bytes;
4602         facet->tcp_flags |= stats->tcp_flags;
4603     }
4604 }
4605 \f
4606 /* Rules. */
4607
4608 /* Lookup 'flow' in 'ofproto''s classifier.  If 'wc' is non-null, sets
4609  * the fields that were relevant as part of the lookup. */
4610 void
4611 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4612                  struct flow_wildcards *wc, struct rule_dpif **rule)
4613 {
4614     struct ofport_dpif *port;
4615
4616     if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4617         return;
4618     }
4619     port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4620     if (!port) {
4621         VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4622                      flow->in_port.ofp_port);
4623     }
4624
4625     choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4626                      ofproto->no_packet_in_rule, rule);
4627 }
4628
4629 bool
4630 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4631                           const struct flow *flow, struct flow_wildcards *wc,
4632                           uint8_t table_id, struct rule_dpif **rule)
4633 {
4634     const struct cls_rule *cls_rule;
4635     struct classifier *cls;
4636     bool frag;
4637
4638     *rule = NULL;
4639     if (table_id >= N_TABLES) {
4640         return false;
4641     }
4642
4643     if (wc) {
4644         memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4645         wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4646     }
4647
4648     cls = &ofproto->up.tables[table_id].cls;
4649     ovs_rwlock_rdlock(&cls->rwlock);
4650     frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4651     if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4652         /* We must pretend that transport ports are unavailable. */
4653         struct flow ofpc_normal_flow = *flow;
4654         ofpc_normal_flow.tp_src = htons(0);
4655         ofpc_normal_flow.tp_dst = htons(0);
4656         cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4657     } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4658         cls_rule = &ofproto->drop_frags_rule->up.cr;
4659         /* Frag mask in wc already set above. */
4660     } else {
4661         cls_rule = classifier_lookup(cls, flow, wc);
4662     }
4663
4664     *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4665     rule_dpif_ref(*rule);
4666     ovs_rwlock_unlock(&cls->rwlock);
4667
4668     return *rule != NULL;
4669 }
4670
4671 /* Given a port configuration (specified as zero if there's no port), chooses
4672  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4673  * flow table miss. */
4674 void
4675 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4676                  struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4677 {
4678     *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4679     rule_dpif_ref(*rule);
4680 }
4681
4682 void
4683 rule_dpif_ref(struct rule_dpif *rule)
4684 {
4685     if (rule) {
4686         ofproto_rule_ref(&rule->up);
4687     }
4688 }
4689
4690 void
4691 rule_dpif_unref(struct rule_dpif *rule)
4692 {
4693     if (rule) {
4694         ofproto_rule_unref(&rule->up);
4695     }
4696 }
4697
4698 static void
4699 complete_operation(struct rule_dpif *rule)
4700     OVS_REQUIRES(ofproto_mutex)
4701 {
4702     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4703
4704     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4705     ofoperation_complete(rule->up.pending, 0);
4706 }
4707
4708 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4709 {
4710     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4711 }
4712
4713 static struct rule *
4714 rule_alloc(void)
4715 {
4716     struct rule_dpif *rule = xmalloc(sizeof *rule);
4717     return &rule->up;
4718 }
4719
4720 static void
4721 rule_dealloc(struct rule *rule_)
4722 {
4723     struct rule_dpif *rule = rule_dpif_cast(rule_);
4724     free(rule);
4725 }
4726
4727 static enum ofperr
4728 rule_construct(struct rule *rule_)
4729 {
4730     struct rule_dpif *rule = rule_dpif_cast(rule_);
4731     ovs_mutex_init(&rule->stats_mutex);
4732     ovs_mutex_lock(&rule->stats_mutex);
4733     rule->packet_count = 0;
4734     rule->byte_count = 0;
4735     ovs_mutex_unlock(&rule->stats_mutex);
4736     return 0;
4737 }
4738
4739 static void
4740 rule_insert(struct rule *rule_)
4741     OVS_REQUIRES(ofproto_mutex)
4742 {
4743     struct rule_dpif *rule = rule_dpif_cast(rule_);
4744     complete_operation(rule);
4745 }
4746
4747 static void
4748 rule_delete(struct rule *rule_)
4749     OVS_REQUIRES(ofproto_mutex)
4750 {
4751     struct rule_dpif *rule = rule_dpif_cast(rule_);
4752     complete_operation(rule);
4753 }
4754
4755 static void
4756 rule_destruct(struct rule *rule_)
4757 {
4758     struct rule_dpif *rule = rule_dpif_cast(rule_);
4759     ovs_mutex_destroy(&rule->stats_mutex);
4760 }
4761
4762 static void
4763 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4764 {
4765     struct rule_dpif *rule = rule_dpif_cast(rule_);
4766
4767     /* push_all_stats() can handle flow misses which, when using the learn
4768      * action, can cause rules to be added and deleted.  This can corrupt our
4769      * caller's datastructures which assume that rule_get_stats() doesn't have
4770      * an impact on the flow table. To be safe, we disable miss handling. */
4771     push_all_stats__(false);
4772
4773     /* Start from historical data for 'rule' itself that are no longer tracked
4774      * in facets.  This counts, for example, facets that have expired. */
4775     ovs_mutex_lock(&rule->stats_mutex);
4776     *packets = rule->packet_count;
4777     *bytes = rule->byte_count;
4778     ovs_mutex_unlock(&rule->stats_mutex);
4779 }
4780
4781 static void
4782 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4783                   struct ofpbuf *packet)
4784 {
4785     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4786
4787     ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
4788 }
4789
4790 static enum ofperr
4791 rule_execute(struct rule *rule, const struct flow *flow,
4792              struct ofpbuf *packet)
4793 {
4794     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4795     ofpbuf_delete(packet);
4796     return 0;
4797 }
4798
4799 static void
4800 rule_modify_actions(struct rule *rule_, bool reset_counters)
4801     OVS_REQUIRES(ofproto_mutex)
4802 {
4803     struct rule_dpif *rule = rule_dpif_cast(rule_);
4804
4805     if (reset_counters) {
4806         ovs_mutex_lock(&rule->stats_mutex);
4807         rule->packet_count = 0;
4808         rule->byte_count = 0;
4809         ovs_mutex_unlock(&rule->stats_mutex);
4810     }
4811
4812     complete_operation(rule);
4813 }
4814
4815 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4816 {
4817     return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4818 }
4819
4820 static struct ofgroup *
4821 group_alloc(void)
4822 {
4823     struct group_dpif *group = xzalloc(sizeof *group);
4824     return &group->up;
4825 }
4826
4827 static void
4828 group_dealloc(struct ofgroup *group_)
4829 {
4830     struct group_dpif *group = group_dpif_cast(group_);
4831     free(group);
4832 }
4833
4834 static void
4835 group_construct_stats(struct group_dpif *group)
4836     OVS_REQUIRES(group->stats_mutex)
4837 {
4838     group->packet_count = 0;
4839     group->byte_count = 0;
4840     if (!group->bucket_stats) {
4841         group->bucket_stats = xcalloc(group->up.n_buckets,
4842                                       sizeof *group->bucket_stats);
4843     } else {
4844         memset(group->bucket_stats, 0, group->up.n_buckets *
4845                sizeof *group->bucket_stats);
4846     }
4847 }
4848
4849 static enum ofperr
4850 group_construct(struct ofgroup *group_)
4851 {
4852     struct group_dpif *group = group_dpif_cast(group_);
4853     ovs_mutex_init(&group->stats_mutex);
4854     ovs_mutex_lock(&group->stats_mutex);
4855     group_construct_stats(group);
4856     ovs_mutex_unlock(&group->stats_mutex);
4857     return 0;
4858 }
4859
4860 static void
4861 group_destruct__(struct group_dpif *group)
4862     OVS_REQUIRES(group->stats_mutex)
4863 {
4864     free(group->bucket_stats);
4865     group->bucket_stats = NULL;
4866 }
4867
4868 static void
4869 group_destruct(struct ofgroup *group_)
4870 {
4871     struct group_dpif *group = group_dpif_cast(group_);
4872     ovs_mutex_lock(&group->stats_mutex);
4873     group_destruct__(group);
4874     ovs_mutex_unlock(&group->stats_mutex);
4875     ovs_mutex_destroy(&group->stats_mutex);
4876 }
4877
4878 static enum ofperr
4879 group_modify(struct ofgroup *group_, struct ofgroup *victim_)
4880 {
4881     struct group_dpif *group = group_dpif_cast(group_);
4882     struct group_dpif *victim = group_dpif_cast(victim_);
4883
4884     ovs_mutex_lock(&group->stats_mutex);
4885     if (victim->up.n_buckets < group->up.n_buckets) {
4886         group_destruct__(group);
4887     }
4888     group_construct_stats(group);
4889     ovs_mutex_unlock(&group->stats_mutex);
4890
4891     return 0;
4892 }
4893
4894 static enum ofperr
4895 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4896 {
4897     struct group_dpif *group = group_dpif_cast(group_);
4898
4899     /* Start from historical data for 'group' itself that are no longer tracked
4900      * in facets.  This counts, for example, facets that have expired. */
4901     ovs_mutex_lock(&group->stats_mutex);
4902     ogs->packet_count = group->packet_count;
4903     ogs->byte_count = group->byte_count;
4904     memcpy(ogs->bucket_stats, group->bucket_stats,
4905            group->up.n_buckets * sizeof *group->bucket_stats);
4906     ovs_mutex_unlock(&group->stats_mutex);
4907
4908     return 0;
4909 }
4910
4911 bool
4912 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4913                   struct group_dpif **group)
4914     OVS_TRY_RDLOCK(true, (*group)->up.rwlock)
4915 {
4916     struct ofgroup *ofgroup;
4917     bool found;
4918
4919     *group = NULL;
4920     found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4921     *group = found ?  group_dpif_cast(ofgroup) : NULL;
4922
4923     return found;
4924 }
4925
4926 void
4927 group_dpif_release(struct group_dpif *group)
4928     OVS_RELEASES(group->up.rwlock)
4929 {
4930     ofproto_group_release(&group->up);
4931 }
4932
4933 void
4934 group_dpif_get_buckets(const struct group_dpif *group,
4935                        const struct list **buckets)
4936 {
4937     *buckets = &group->up.buckets;
4938 }
4939
4940 enum ofp11_group_type
4941 group_dpif_get_type(const struct group_dpif *group)
4942 {
4943     return group->up.type;
4944 }
4945 \f
4946 /* Sends 'packet' out 'ofport'.
4947  * May modify 'packet'.
4948  * Returns 0 if successful, otherwise a positive errno value. */
4949 int
4950 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4951 {
4952     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4953     int error;
4954
4955     error = xlate_send_packet(ofport, packet);
4956
4957     ovs_mutex_lock(&ofproto->stats_mutex);
4958     ofproto->stats.tx_packets++;
4959     ofproto->stats.tx_bytes += packet->size;
4960     ovs_mutex_unlock(&ofproto->stats_mutex);
4961     return error;
4962 }
4963
4964 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4965  * The action will state 'slow' as the reason that the action is in the slow
4966  * path.  (This is purely informational: it allows a human viewing "ovs-dpctl
4967  * dump-flows" output to see why a flow is in the slow path.)
4968  *
4969  * The 'stub_size' bytes in 'stub' will be used to store the action.
4970  * 'stub_size' must be large enough for the action.
4971  *
4972  * The action and its size will be stored in '*actionsp' and '*actions_lenp',
4973  * respectively. */
4974 static void
4975 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
4976                   enum slow_path_reason slow,
4977                   uint64_t *stub, size_t stub_size,
4978                   const struct nlattr **actionsp, size_t *actions_lenp)
4979 {
4980     union user_action_cookie cookie;
4981     struct ofpbuf buf;
4982
4983     cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
4984     cookie.slow_path.unused = 0;
4985     cookie.slow_path.reason = slow;
4986
4987     ofpbuf_use_stack(&buf, stub, stub_size);
4988     if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
4989         uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
4990                                          ODPP_NONE);
4991         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
4992     } else {
4993         odp_port_t odp_port;
4994         uint32_t pid;
4995
4996         odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
4997         pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
4998         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
4999     }
5000     *actionsp = buf.data;
5001     *actions_lenp = buf.size;
5002 }
5003 \f
5004 static bool
5005 set_frag_handling(struct ofproto *ofproto_,
5006                   enum ofp_config_flags frag_handling)
5007 {
5008     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5009     if (frag_handling != OFPC_FRAG_REASM) {
5010         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5011         return true;
5012     } else {
5013         return false;
5014     }
5015 }
5016
5017 static enum ofperr
5018 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5019            const struct flow *flow,
5020            const struct ofpact *ofpacts, size_t ofpacts_len)
5021 {
5022     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5023
5024     ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
5025                                  ofpacts_len, packet);
5026     return 0;
5027 }
5028 \f
5029 /* NetFlow. */
5030
5031 static int
5032 set_netflow(struct ofproto *ofproto_,
5033             const struct netflow_options *netflow_options)
5034 {
5035     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5036
5037     if (netflow_options) {
5038         if (!ofproto->netflow) {
5039             ofproto->netflow = netflow_create();
5040             ofproto->backer->need_revalidate = REV_RECONFIGURE;
5041         }
5042         return netflow_set_options(ofproto->netflow, netflow_options);
5043     } else if (ofproto->netflow) {
5044         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5045         netflow_destroy(ofproto->netflow);
5046         ofproto->netflow = NULL;
5047     }
5048
5049     return 0;
5050 }
5051
5052 static void
5053 get_netflow_ids(const struct ofproto *ofproto_,
5054                 uint8_t *engine_type, uint8_t *engine_id)
5055 {
5056     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5057
5058     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5059 }
5060
5061 static void
5062 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5063 {
5064     if (!facet_is_controller_flow(facet) &&
5065         netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5066         struct subfacet *subfacet;
5067         struct ofexpired expired;
5068
5069         LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5070             if (subfacet->path == SF_FAST_PATH) {
5071                 struct dpif_flow_stats stats;
5072
5073                 subfacet_install(subfacet, &facet->xout.odp_actions,
5074                                  &stats);
5075                 subfacet_update_stats(subfacet, &stats);
5076             }
5077         }
5078
5079         expired.flow = facet->flow;
5080         expired.packet_count = facet->packet_count;
5081         expired.byte_count = facet->byte_count;
5082         expired.used = facet->used;
5083         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5084     }
5085 }
5086
5087 static void
5088 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5089 {
5090     struct cls_cursor cursor;
5091     struct facet *facet;
5092
5093     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5094     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5095     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5096         send_active_timeout(ofproto, facet);
5097     }
5098     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5099 }
5100 \f
5101 static struct ofproto_dpif *
5102 ofproto_dpif_lookup(const char *name)
5103 {
5104     struct ofproto_dpif *ofproto;
5105
5106     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5107                              hash_string(name, 0), &all_ofproto_dpifs) {
5108         if (!strcmp(ofproto->up.name, name)) {
5109             return ofproto;
5110         }
5111     }
5112     return NULL;
5113 }
5114
5115 static void
5116 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5117                           const char *argv[], void *aux OVS_UNUSED)
5118 {
5119     struct ofproto_dpif *ofproto;
5120
5121     if (argc > 1) {
5122         ofproto = ofproto_dpif_lookup(argv[1]);
5123         if (!ofproto) {
5124             unixctl_command_reply_error(conn, "no such bridge");
5125             return;
5126         }
5127         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5128         mac_learning_flush(ofproto->ml);
5129         ovs_rwlock_unlock(&ofproto->ml->rwlock);
5130     } else {
5131         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5132             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5133             mac_learning_flush(ofproto->ml);
5134             ovs_rwlock_unlock(&ofproto->ml->rwlock);
5135         }
5136     }
5137
5138     unixctl_command_reply(conn, "table successfully flushed");
5139 }
5140
5141 static struct ofport_dpif *
5142 ofbundle_get_a_port(const struct ofbundle *bundle)
5143 {
5144     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5145                         bundle_node);
5146 }
5147
5148 static void
5149 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5150                          const char *argv[], void *aux OVS_UNUSED)
5151 {
5152     struct ds ds = DS_EMPTY_INITIALIZER;
5153     const struct ofproto_dpif *ofproto;
5154     const struct mac_entry *e;
5155
5156     ofproto = ofproto_dpif_lookup(argv[1]);
5157     if (!ofproto) {
5158         unixctl_command_reply_error(conn, "no such bridge");
5159         return;
5160     }
5161
5162     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
5163     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5164     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5165         struct ofbundle *bundle = e->port.p;
5166         char name[OFP_MAX_PORT_NAME_LEN];
5167
5168         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5169                                name, sizeof name);
5170         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
5171                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
5172                       mac_entry_age(ofproto->ml, e));
5173     }
5174     ovs_rwlock_unlock(&ofproto->ml->rwlock);
5175     unixctl_command_reply(conn, ds_cstr(&ds));
5176     ds_destroy(&ds);
5177 }
5178
5179 struct trace_ctx {
5180     struct xlate_out xout;
5181     struct xlate_in xin;
5182     struct flow flow;
5183     struct ds *result;
5184 };
5185
5186 static void
5187 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5188 {
5189     struct rule_actions *actions;
5190     ovs_be64 cookie;
5191
5192     ds_put_char_multiple(result, '\t', level);
5193     if (!rule) {
5194         ds_put_cstr(result, "No match\n");
5195         return;
5196     }
5197
5198     ovs_mutex_lock(&rule->up.mutex);
5199     cookie = rule->up.flow_cookie;
5200     ovs_mutex_unlock(&rule->up.mutex);
5201
5202     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5203                   rule ? rule->up.table_id : 0, ntohll(cookie));
5204     cls_rule_format(&rule->up.cr, result);
5205     ds_put_char(result, '\n');
5206
5207     actions = rule_dpif_get_actions(rule);
5208
5209     ds_put_char_multiple(result, '\t', level);
5210     ds_put_cstr(result, "OpenFlow actions=");
5211     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
5212     ds_put_char(result, '\n');
5213
5214     rule_actions_unref(actions);
5215 }
5216
5217 static void
5218 trace_format_flow(struct ds *result, int level, const char *title,
5219                   struct trace_ctx *trace)
5220 {
5221     ds_put_char_multiple(result, '\t', level);
5222     ds_put_format(result, "%s: ", title);
5223     if (flow_equal(&trace->xin.flow, &trace->flow)) {
5224         ds_put_cstr(result, "unchanged");
5225     } else {
5226         flow_format(result, &trace->xin.flow);
5227         trace->flow = trace->xin.flow;
5228     }
5229     ds_put_char(result, '\n');
5230 }
5231
5232 static void
5233 trace_format_regs(struct ds *result, int level, const char *title,
5234                   struct trace_ctx *trace)
5235 {
5236     size_t i;
5237
5238     ds_put_char_multiple(result, '\t', level);
5239     ds_put_format(result, "%s:", title);
5240     for (i = 0; i < FLOW_N_REGS; i++) {
5241         ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5242     }
5243     ds_put_char(result, '\n');
5244 }
5245
5246 static void
5247 trace_format_odp(struct ds *result, int level, const char *title,
5248                  struct trace_ctx *trace)
5249 {
5250     struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5251
5252     ds_put_char_multiple(result, '\t', level);
5253     ds_put_format(result, "%s: ", title);
5254     format_odp_actions(result, odp_actions->data, odp_actions->size);
5255     ds_put_char(result, '\n');
5256 }
5257
5258 static void
5259 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5260 {
5261     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5262     struct ds *result = trace->result;
5263
5264     ds_put_char(result, '\n');
5265     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5266     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5267     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
5268     trace_format_rule(result, recurse + 1, rule);
5269 }
5270
5271 static void
5272 trace_report(struct xlate_in *xin, const char *s, int recurse)
5273 {
5274     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5275     struct ds *result = trace->result;
5276
5277     ds_put_char_multiple(result, '\t', recurse);
5278     ds_put_cstr(result, s);
5279     ds_put_char(result, '\n');
5280 }
5281
5282 static void
5283 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5284                       void *aux OVS_UNUSED)
5285 {
5286     const struct dpif_backer *backer = NULL;
5287     struct ofproto_dpif *ofproto;
5288     struct ofpbuf odp_key, odp_mask;
5289     struct ofpbuf *packet;
5290     struct ds result;
5291     struct flow flow;
5292     struct simap port_names;
5293     char *s;
5294
5295     packet = NULL;
5296     backer = NULL;
5297     ds_init(&result);
5298     ofpbuf_init(&odp_key, 0);
5299     ofpbuf_init(&odp_mask, 0);
5300     simap_init(&port_names);
5301
5302     /* Handle "-generate" or a hex string as the last argument. */
5303     if (!strcmp(argv[argc - 1], "-generate")) {
5304         packet = ofpbuf_new(0);
5305         argc--;
5306     } else {
5307         const char *error = eth_from_hex(argv[argc - 1], &packet);
5308         if (!error) {
5309             argc--;
5310         } else if (argc == 4) {
5311             /* The 3-argument form must end in "-generate' or a hex string. */
5312             unixctl_command_reply_error(conn, error);
5313             goto exit;
5314         }
5315     }
5316
5317     /* odp_flow can have its in_port specified as a name instead of port no.
5318      * We do not yet know whether a given flow is a odp_flow or a br_flow.
5319      * But, to know whether a flow is odp_flow through odp_flow_from_string(),
5320      * we need to create a simap of name to port no. */
5321     if (argc == 3) {
5322         const char *dp_type;
5323         if (!strncmp(argv[1], "ovs-", 4)) {
5324             dp_type = argv[1] + 4;
5325         } else {
5326             dp_type = argv[1];
5327         }
5328         backer = shash_find_data(&all_dpif_backers, dp_type);
5329     } else {
5330         struct shash_node *node;
5331         if (shash_count(&all_dpif_backers) == 1) {
5332             node = shash_first(&all_dpif_backers);
5333             backer = node->data;
5334         }
5335     }
5336     if (backer && backer->dpif) {
5337         struct dpif_port dpif_port;
5338         struct dpif_port_dump port_dump;
5339         DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
5340             simap_put(&port_names, dpif_port.name,
5341                       odp_to_u32(dpif_port.port_no));
5342         }
5343     }
5344
5345     /* Parse the flow and determine whether a datapath or
5346      * bridge is specified. If function odp_flow_key_from_string()
5347      * returns 0, the flow is a odp_flow. If function
5348      * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5349     if (!odp_flow_from_string(argv[argc - 1], &port_names, &odp_key, &odp_mask)) {
5350         if (!backer) {
5351             unixctl_command_reply_error(conn, "Cannot find the datapath");
5352             goto exit;
5353         }
5354
5355         if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5356                           NULL, &ofproto, NULL)) {
5357             unixctl_command_reply_error(conn, "Invalid datapath flow");
5358             goto exit;
5359         }
5360         ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5361     } else if (!parse_ofp_exact_flow(&flow, NULL, argv[argc - 1], NULL)) {
5362         if (argc != 3) {
5363             unixctl_command_reply_error(conn, "Must specify bridge name");
5364             goto exit;
5365         }
5366
5367         ofproto = ofproto_dpif_lookup(argv[1]);
5368         if (!ofproto) {
5369             unixctl_command_reply_error(conn, "Unknown bridge name");
5370             goto exit;
5371         }
5372     } else {
5373         unixctl_command_reply_error(conn, "Bad flow syntax");
5374         goto exit;
5375     }
5376
5377     /* Generate a packet, if requested. */
5378     if (packet) {
5379         if (!packet->size) {
5380             flow_compose(packet, &flow);
5381         } else {
5382             union flow_in_port in_port_;
5383
5384             in_port_ = flow.in_port;
5385             ds_put_cstr(&result, "Packet: ");
5386             s = ofp_packet_to_string(packet->data, packet->size);
5387             ds_put_cstr(&result, s);
5388             free(s);
5389
5390             /* Use the metadata from the flow and the packet argument
5391              * to reconstruct the flow. */
5392             flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5393                          &in_port_, &flow);
5394         }
5395     }
5396
5397     ofproto_trace(ofproto, &flow, packet, &result);
5398     unixctl_command_reply(conn, ds_cstr(&result));
5399
5400 exit:
5401     ds_destroy(&result);
5402     ofpbuf_delete(packet);
5403     ofpbuf_uninit(&odp_key);
5404     ofpbuf_uninit(&odp_mask);
5405     simap_destroy(&port_names);
5406 }
5407
5408 static void
5409 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5410               const struct ofpbuf *packet, struct ds *ds)
5411 {
5412     struct rule_dpif *rule;
5413     struct flow_wildcards wc;
5414
5415     ds_put_cstr(ds, "Flow: ");
5416     flow_format(ds, flow);
5417     ds_put_char(ds, '\n');
5418
5419     flow_wildcards_init_catchall(&wc);
5420     rule_dpif_lookup(ofproto, flow, &wc, &rule);
5421
5422     trace_format_rule(ds, 0, rule);
5423     if (rule == ofproto->miss_rule) {
5424         ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5425     } else if (rule == ofproto->no_packet_in_rule) {
5426         ds_put_cstr(ds, "\nNo match, packets dropped because "
5427                     "OFPPC_NO_PACKET_IN is set on in_port.\n");
5428     } else if (rule == ofproto->drop_frags_rule) {
5429         ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5430                     "and the fragment handling mode is \"drop\".\n");
5431     }
5432
5433     if (rule) {
5434         uint64_t odp_actions_stub[1024 / 8];
5435         struct ofpbuf odp_actions;
5436         struct trace_ctx trace;
5437         struct match match;
5438         uint16_t tcp_flags;
5439
5440         tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5441         trace.result = ds;
5442         trace.flow = *flow;
5443         ofpbuf_use_stub(&odp_actions,
5444                         odp_actions_stub, sizeof odp_actions_stub);
5445         xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5446         trace.xin.resubmit_hook = trace_resubmit;
5447         trace.xin.report_hook = trace_report;
5448
5449         xlate_actions(&trace.xin, &trace.xout);
5450         flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5451
5452         ds_put_char(ds, '\n');
5453         trace_format_flow(ds, 0, "Final flow", &trace);
5454
5455         match_init(&match, flow, &trace.xout.wc);
5456         ds_put_cstr(ds, "Relevant fields: ");
5457         match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5458         ds_put_char(ds, '\n');
5459
5460         ds_put_cstr(ds, "Datapath actions: ");
5461         format_odp_actions(ds, trace.xout.odp_actions.data,
5462                            trace.xout.odp_actions.size);
5463
5464         if (trace.xout.slow) {
5465             enum slow_path_reason slow;
5466
5467             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5468                         "slow path because it:");
5469
5470             slow = trace.xout.slow;
5471             while (slow) {
5472                 enum slow_path_reason bit = rightmost_1bit(slow);
5473
5474                 ds_put_format(ds, "\n\t- %s.",
5475                               slow_path_reason_to_explanation(bit));
5476
5477                 slow &= ~bit;
5478             }
5479         }
5480
5481         xlate_out_uninit(&trace.xout);
5482     }
5483
5484     rule_dpif_unref(rule);
5485 }
5486
5487 /* Runs a self-check of flow translations in 'ofproto'.  Appends a message to
5488  * 'reply' describing the results. */
5489 static void
5490 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5491 {
5492     struct cls_cursor cursor;
5493     struct facet *facet;
5494     int errors;
5495
5496     errors = 0;
5497     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5498     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5499     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5500         if (!facet_check_consistency(facet)) {
5501             errors++;
5502         }
5503     }
5504     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5505     if (errors) {
5506         ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5507     }
5508
5509     if (errors) {
5510         ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5511                       ofproto->up.name, errors);
5512     } else {
5513         ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5514     }
5515 }
5516
5517 static void
5518 ofproto_dpif_self_check(struct unixctl_conn *conn,
5519                         int argc, const char *argv[], void *aux OVS_UNUSED)
5520 {
5521     struct ds reply = DS_EMPTY_INITIALIZER;
5522     struct ofproto_dpif *ofproto;
5523
5524     if (argc > 1) {
5525         ofproto = ofproto_dpif_lookup(argv[1]);
5526         if (!ofproto) {
5527             unixctl_command_reply_error(conn, "Unknown ofproto (use "
5528                                         "ofproto/list for help)");
5529             return;
5530         }
5531         ofproto_dpif_self_check__(ofproto, &reply);
5532     } else {
5533         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5534             ofproto_dpif_self_check__(ofproto, &reply);
5535         }
5536     }
5537
5538     unixctl_command_reply(conn, ds_cstr(&reply));
5539     ds_destroy(&reply);
5540 }
5541
5542 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
5543  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
5544  * to destroy 'ofproto_shash' and free the returned value. */
5545 static const struct shash_node **
5546 get_ofprotos(struct shash *ofproto_shash)
5547 {
5548     const struct ofproto_dpif *ofproto;
5549
5550     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5551         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5552         shash_add_nocopy(ofproto_shash, name, ofproto);
5553     }
5554
5555     return shash_sort(ofproto_shash);
5556 }
5557
5558 static void
5559 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5560                               const char *argv[] OVS_UNUSED,
5561                               void *aux OVS_UNUSED)
5562 {
5563     struct ds ds = DS_EMPTY_INITIALIZER;
5564     struct shash ofproto_shash;
5565     const struct shash_node **sorted_ofprotos;
5566     int i;
5567
5568     shash_init(&ofproto_shash);
5569     sorted_ofprotos = get_ofprotos(&ofproto_shash);
5570     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5571         const struct shash_node *node = sorted_ofprotos[i];
5572         ds_put_format(&ds, "%s\n", node->name);
5573     }
5574
5575     shash_destroy(&ofproto_shash);
5576     free(sorted_ofprotos);
5577
5578     unixctl_command_reply(conn, ds_cstr(&ds));
5579     ds_destroy(&ds);
5580 }
5581
5582 static void
5583 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5584 {
5585     const struct shash_node **ofprotos;
5586     struct ofproto_dpif *ofproto;
5587     struct shash ofproto_shash;
5588     uint64_t n_hit, n_missed;
5589     size_t i;
5590
5591     n_hit = n_missed = 0;
5592     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5593         if (ofproto->backer == backer) {
5594             n_missed += ofproto->n_missed;
5595             n_hit += ofproto->n_hit;
5596         }
5597     }
5598
5599     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5600                   dpif_name(backer->dpif), n_hit, n_missed);
5601     ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5602                   " life span: %lldms\n", hmap_count(&backer->subfacets),
5603                   backer->avg_n_subfacet, backer->max_n_subfacet,
5604                   backer->avg_subfacet_life);
5605
5606     shash_init(&ofproto_shash);
5607     ofprotos = get_ofprotos(&ofproto_shash);
5608     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5609         struct ofproto_dpif *ofproto = ofprotos[i]->data;
5610         const struct shash_node **ports;
5611         size_t j;
5612
5613         if (ofproto->backer != backer) {
5614             continue;
5615         }
5616
5617         ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5618                       ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5619
5620         ports = shash_sort(&ofproto->up.port_by_name);
5621         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5622             const struct shash_node *node = ports[j];
5623             struct ofport *ofport = node->data;
5624             struct smap config;
5625             odp_port_t odp_port;
5626
5627             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5628                           ofport->ofp_port);
5629
5630             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5631             if (odp_port != ODPP_NONE) {
5632                 ds_put_format(ds, "%"PRIu32":", odp_port);
5633             } else {
5634                 ds_put_cstr(ds, "none:");
5635             }
5636
5637             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5638
5639             smap_init(&config);
5640             if (!netdev_get_config(ofport->netdev, &config)) {
5641                 const struct smap_node **nodes;
5642                 size_t i;
5643
5644                 nodes = smap_sort(&config);
5645                 for (i = 0; i < smap_count(&config); i++) {
5646                     const struct smap_node *node = nodes[i];
5647                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5648                                   node->key, node->value);
5649                 }
5650                 free(nodes);
5651             }
5652             smap_destroy(&config);
5653
5654             ds_put_char(ds, ')');
5655             ds_put_char(ds, '\n');
5656         }
5657         free(ports);
5658     }
5659     shash_destroy(&ofproto_shash);
5660     free(ofprotos);
5661 }
5662
5663 static void
5664 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5665                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5666 {
5667     struct ds ds = DS_EMPTY_INITIALIZER;
5668     const struct shash_node **backers;
5669     int i;
5670
5671     backers = shash_sort(&all_dpif_backers);
5672     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5673         dpif_show_backer(backers[i]->data, &ds);
5674     }
5675     free(backers);
5676
5677     unixctl_command_reply(conn, ds_cstr(&ds));
5678     ds_destroy(&ds);
5679 }
5680
5681 /* Dump the megaflow (facet) cache.  This is useful to check the
5682  * correctness of flow wildcarding, since the same mechanism is used for
5683  * both xlate caching and kernel wildcarding.
5684  *
5685  * It's important to note that in the output the flow description uses
5686  * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5687  *
5688  * This command is only needed for advanced debugging, so it's not
5689  * documented in the man page. */
5690 static void
5691 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5692                                     int argc OVS_UNUSED, const char *argv[],
5693                                     void *aux OVS_UNUSED)
5694 {
5695     struct ds ds = DS_EMPTY_INITIALIZER;
5696     const struct ofproto_dpif *ofproto;
5697     long long int now = time_msec();
5698     struct cls_cursor cursor;
5699     struct facet *facet;
5700
5701     ofproto = ofproto_dpif_lookup(argv[1]);
5702     if (!ofproto) {
5703         unixctl_command_reply_error(conn, "no such bridge");
5704         return;
5705     }
5706
5707     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5708     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5709     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5710         cls_rule_format(&facet->cr, &ds);
5711         ds_put_cstr(&ds, ", ");
5712         ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5713         ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5714         ds_put_cstr(&ds, "Datapath actions: ");
5715         if (facet->xout.slow) {
5716             uint64_t slow_path_stub[128 / 8];
5717             const struct nlattr *actions;
5718             size_t actions_len;
5719
5720             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5721                               slow_path_stub, sizeof slow_path_stub,
5722                               &actions, &actions_len);
5723             format_odp_actions(&ds, actions, actions_len);
5724         } else {
5725             format_odp_actions(&ds, facet->xout.odp_actions.data,
5726                                facet->xout.odp_actions.size);
5727         }
5728         ds_put_cstr(&ds, "\n");
5729     }
5730     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5731
5732     ds_chomp(&ds, '\n');
5733     unixctl_command_reply(conn, ds_cstr(&ds));
5734     ds_destroy(&ds);
5735 }
5736
5737 /* Disable using the megaflows.
5738  *
5739  * This command is only needed for advanced debugging, so it's not
5740  * documented in the man page. */
5741 static void
5742 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5743                                        int argc OVS_UNUSED,
5744                                        const char *argv[] OVS_UNUSED,
5745                                        void *aux OVS_UNUSED)
5746 {
5747     struct ofproto_dpif *ofproto;
5748
5749     enable_megaflows = false;
5750
5751     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5752         flush(&ofproto->up);
5753     }
5754
5755     unixctl_command_reply(conn, "megaflows disabled");
5756 }
5757
5758 /* Re-enable using megaflows.
5759  *
5760  * This command is only needed for advanced debugging, so it's not
5761  * documented in the man page. */
5762 static void
5763 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5764                                       int argc OVS_UNUSED,
5765                                       const char *argv[] OVS_UNUSED,
5766                                       void *aux OVS_UNUSED)
5767 {
5768     struct ofproto_dpif *ofproto;
5769
5770     enable_megaflows = true;
5771
5772     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5773         flush(&ofproto->up);
5774     }
5775
5776     unixctl_command_reply(conn, "megaflows enabled");
5777 }
5778
5779 static void
5780 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5781                                 int argc OVS_UNUSED, const char *argv[],
5782                                 void *aux OVS_UNUSED)
5783 {
5784     struct ds ds = DS_EMPTY_INITIALIZER;
5785     const struct ofproto_dpif *ofproto;
5786     struct subfacet *subfacet;
5787
5788     ofproto = ofproto_dpif_lookup(argv[1]);
5789     if (!ofproto) {
5790         unixctl_command_reply_error(conn, "no such bridge");
5791         return;
5792     }
5793
5794     update_stats(ofproto->backer);
5795
5796     HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5797         struct facet *facet = subfacet->facet;
5798         struct odputil_keybuf maskbuf;
5799         struct ofpbuf mask;
5800
5801         if (facet->ofproto != ofproto) {
5802             continue;
5803         }
5804
5805         ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5806         if (enable_megaflows) {
5807             odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5808                                    &facet->flow, UINT32_MAX);
5809         }
5810
5811         odp_flow_format(subfacet->key, subfacet->key_len,
5812                         mask.data, mask.size, NULL, &ds, false);
5813
5814         ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5815                       subfacet->dp_packet_count, subfacet->dp_byte_count);
5816         if (subfacet->used) {
5817             ds_put_format(&ds, "%.3fs",
5818                           (time_msec() - subfacet->used) / 1000.0);
5819         } else {
5820             ds_put_format(&ds, "never");
5821         }
5822         if (subfacet->facet->tcp_flags) {
5823             ds_put_cstr(&ds, ", flags:");
5824             packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5825         }
5826
5827         ds_put_cstr(&ds, ", actions:");
5828         if (facet->xout.slow) {
5829             uint64_t slow_path_stub[128 / 8];
5830             const struct nlattr *actions;
5831             size_t actions_len;
5832
5833             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5834                               slow_path_stub, sizeof slow_path_stub,
5835                               &actions, &actions_len);
5836             format_odp_actions(&ds, actions, actions_len);
5837         } else {
5838             format_odp_actions(&ds, facet->xout.odp_actions.data,
5839                                facet->xout.odp_actions.size);
5840         }
5841         ds_put_char(&ds, '\n');
5842     }
5843
5844     unixctl_command_reply(conn, ds_cstr(&ds));
5845     ds_destroy(&ds);
5846 }
5847
5848 static void
5849 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5850                                int argc OVS_UNUSED, const char *argv[],
5851                                void *aux OVS_UNUSED)
5852 {
5853     struct ds ds = DS_EMPTY_INITIALIZER;
5854     struct ofproto_dpif *ofproto;
5855
5856     ofproto = ofproto_dpif_lookup(argv[1]);
5857     if (!ofproto) {
5858         unixctl_command_reply_error(conn, "no such bridge");
5859         return;
5860     }
5861
5862     flush(&ofproto->up);
5863
5864     unixctl_command_reply(conn, ds_cstr(&ds));
5865     ds_destroy(&ds);
5866 }
5867
5868 static void
5869 ofproto_dpif_unixctl_init(void)
5870 {
5871     static bool registered;
5872     if (registered) {
5873         return;
5874     }
5875     registered = true;
5876
5877     unixctl_command_register(
5878         "ofproto/trace",
5879         "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
5880         1, 3, ofproto_unixctl_trace, NULL);
5881     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5882                              ofproto_unixctl_fdb_flush, NULL);
5883     unixctl_command_register("fdb/show", "bridge", 1, 1,
5884                              ofproto_unixctl_fdb_show, NULL);
5885     unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
5886                              ofproto_dpif_self_check, NULL);
5887     unixctl_command_register("dpif/dump-dps", "", 0, 0,
5888                              ofproto_unixctl_dpif_dump_dps, NULL);
5889     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5890                              NULL);
5891     unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
5892                              ofproto_unixctl_dpif_dump_flows, NULL);
5893     unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
5894                              ofproto_unixctl_dpif_del_flows, NULL);
5895     unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
5896                              ofproto_unixctl_dpif_dump_megaflows, NULL);
5897     unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
5898                              ofproto_unixctl_dpif_disable_megaflows, NULL);
5899     unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
5900                              ofproto_unixctl_dpif_enable_megaflows, NULL);
5901 }
5902 \f
5903 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5904  *
5905  * This is deprecated.  It is only for compatibility with broken device drivers
5906  * in old versions of Linux that do not properly support VLANs when VLAN
5907  * devices are not used.  When broken device drivers are no longer in
5908  * widespread use, we will delete these interfaces. */
5909
5910 static int
5911 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5912 {
5913     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5914     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5915
5916     if (realdev_ofp_port == ofport->realdev_ofp_port
5917         && vid == ofport->vlandev_vid) {
5918         return 0;
5919     }
5920
5921     ofproto->backer->need_revalidate = REV_RECONFIGURE;
5922
5923     if (ofport->realdev_ofp_port) {
5924         vsp_remove(ofport);
5925     }
5926     if (realdev_ofp_port && ofport->bundle) {
5927         /* vlandevs are enslaved to their realdevs, so they are not allowed to
5928          * themselves be part of a bundle. */
5929         bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5930     }
5931
5932     ofport->realdev_ofp_port = realdev_ofp_port;
5933     ofport->vlandev_vid = vid;
5934
5935     if (realdev_ofp_port) {
5936         vsp_add(ofport, realdev_ofp_port, vid);
5937     }
5938
5939     return 0;
5940 }
5941
5942 static uint32_t
5943 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
5944 {
5945     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
5946 }
5947
5948 bool
5949 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
5950     OVS_EXCLUDED(ofproto->vsp_mutex)
5951 {
5952     bool ret;
5953
5954     ovs_mutex_lock(&ofproto->vsp_mutex);
5955     ret = !hmap_is_empty(&ofproto->realdev_vid_map);
5956     ovs_mutex_unlock(&ofproto->vsp_mutex);
5957     return ret;
5958 }
5959
5960 static ofp_port_t
5961 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
5962                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5963     OVS_REQUIRES(ofproto->vsp_mutex)
5964 {
5965     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5966         int vid = vlan_tci_to_vid(vlan_tci);
5967         const struct vlan_splinter *vsp;
5968
5969         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5970                                  hash_realdev_vid(realdev_ofp_port, vid),
5971                                  &ofproto->realdev_vid_map) {
5972             if (vsp->realdev_ofp_port == realdev_ofp_port
5973                 && vsp->vid == vid) {
5974                 return vsp->vlandev_ofp_port;
5975             }
5976         }
5977     }
5978     return realdev_ofp_port;
5979 }
5980
5981 /* Returns the OFP port number of the Linux VLAN device that corresponds to
5982  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
5983  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
5984  * 'vlan_tci' 9, it would return the port number of eth0.9.
5985  *
5986  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
5987  * function just returns its 'realdev_ofp_port' argument. */
5988 ofp_port_t
5989 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5990                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5991     OVS_EXCLUDED(ofproto->vsp_mutex)
5992 {
5993     ofp_port_t ret;
5994
5995     ovs_mutex_lock(&ofproto->vsp_mutex);
5996     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
5997     ovs_mutex_unlock(&ofproto->vsp_mutex);
5998     return ret;
5999 }
6000
6001 static struct vlan_splinter *
6002 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6003 {
6004     struct vlan_splinter *vsp;
6005
6006     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6007                              hash_ofp_port(vlandev_ofp_port),
6008                              &ofproto->vlandev_map) {
6009         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6010             return vsp;
6011         }
6012     }
6013
6014     return NULL;
6015 }
6016
6017 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6018  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6019  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
6020  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6021  * eth0 and store 9 in '*vid'.
6022  *
6023  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6024  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
6025  * always does.*/
6026 static ofp_port_t
6027 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6028                        ofp_port_t vlandev_ofp_port, int *vid)
6029     OVS_REQUIRES(ofproto->vsp_mutex)
6030 {
6031     if (!hmap_is_empty(&ofproto->vlandev_map)) {
6032         const struct vlan_splinter *vsp;
6033
6034         vsp = vlandev_find(ofproto, vlandev_ofp_port);
6035         if (vsp) {
6036             if (vid) {
6037                 *vid = vsp->vid;
6038             }
6039             return vsp->realdev_ofp_port;
6040         }
6041     }
6042     return 0;
6043 }
6044
6045 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6046  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
6047  * 'flow->in_port' to the "real" device backing the VLAN device, sets
6048  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Otherwise (which is
6049  * always the case unless VLAN splinters are enabled), returns false without
6050  * making any changes. */
6051 bool
6052 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6053     OVS_EXCLUDED(ofproto->vsp_mutex)
6054 {
6055     ofp_port_t realdev;
6056     int vid;
6057
6058     ovs_mutex_lock(&ofproto->vsp_mutex);
6059     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6060     ovs_mutex_unlock(&ofproto->vsp_mutex);
6061     if (!realdev) {
6062         return false;
6063     }
6064
6065     /* Cause the flow to be processed as if it came in on the real device with
6066      * the VLAN device's VLAN ID. */
6067     flow->in_port.ofp_port = realdev;
6068     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6069     return true;
6070 }
6071
6072 static void
6073 vsp_remove(struct ofport_dpif *port)
6074 {
6075     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6076     struct vlan_splinter *vsp;
6077
6078     ovs_mutex_lock(&ofproto->vsp_mutex);
6079     vsp = vlandev_find(ofproto, port->up.ofp_port);
6080     if (vsp) {
6081         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6082         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6083         free(vsp);
6084
6085         port->realdev_ofp_port = 0;
6086     } else {
6087         VLOG_ERR("missing vlan device record");
6088     }
6089     ovs_mutex_unlock(&ofproto->vsp_mutex);
6090 }
6091
6092 static void
6093 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6094 {
6095     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6096
6097     ovs_mutex_lock(&ofproto->vsp_mutex);
6098     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6099         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6100             == realdev_ofp_port)) {
6101         struct vlan_splinter *vsp;
6102
6103         vsp = xmalloc(sizeof *vsp);
6104         vsp->realdev_ofp_port = realdev_ofp_port;
6105         vsp->vlandev_ofp_port = port->up.ofp_port;
6106         vsp->vid = vid;
6107
6108         port->realdev_ofp_port = realdev_ofp_port;
6109
6110         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6111                     hash_ofp_port(port->up.ofp_port));
6112         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6113                     hash_realdev_vid(realdev_ofp_port, vid));
6114     } else {
6115         VLOG_ERR("duplicate vlan device record");
6116     }
6117     ovs_mutex_unlock(&ofproto->vsp_mutex);
6118 }
6119
6120 static odp_port_t
6121 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6122 {
6123     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6124     return ofport ? ofport->odp_port : ODPP_NONE;
6125 }
6126
6127 struct ofport_dpif *
6128 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6129 {
6130     struct ofport_dpif *port;
6131
6132     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6133     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6134                              &backer->odp_to_ofport_map) {
6135         if (port->odp_port == odp_port) {
6136             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6137             return port;
6138         }
6139     }
6140
6141     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6142     return NULL;
6143 }
6144
6145 static ofp_port_t
6146 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6147 {
6148     struct ofport_dpif *port;
6149
6150     port = odp_port_to_ofport(ofproto->backer, odp_port);
6151     if (port && &ofproto->up == port->up.ofproto) {
6152         return port->up.ofp_port;
6153     } else {
6154         return OFPP_NONE;
6155     }
6156 }
6157
6158 const struct ofproto_class ofproto_dpif_class = {
6159     init,
6160     enumerate_types,
6161     enumerate_names,
6162     del,
6163     port_open_type,
6164     type_run,
6165     type_run_fast,
6166     type_wait,
6167     alloc,
6168     construct,
6169     destruct,
6170     dealloc,
6171     run,
6172     run_fast,
6173     wait,
6174     get_memory_usage,
6175     flush,
6176     get_features,
6177     get_tables,
6178     port_alloc,
6179     port_construct,
6180     port_destruct,
6181     port_dealloc,
6182     port_modified,
6183     port_reconfigured,
6184     port_query_by_name,
6185     port_add,
6186     port_del,
6187     port_get_stats,
6188     port_dump_start,
6189     port_dump_next,
6190     port_dump_done,
6191     port_poll,
6192     port_poll_wait,
6193     port_is_lacp_current,
6194     NULL,                       /* rule_choose_table */
6195     rule_alloc,
6196     rule_construct,
6197     rule_insert,
6198     rule_delete,
6199     rule_destruct,
6200     rule_dealloc,
6201     rule_get_stats,
6202     rule_execute,
6203     rule_modify_actions,
6204     set_frag_handling,
6205     packet_out,
6206     set_netflow,
6207     get_netflow_ids,
6208     set_sflow,
6209     set_ipfix,
6210     set_cfm,
6211     get_cfm_status,
6212     set_bfd,
6213     get_bfd_status,
6214     set_stp,
6215     get_stp_status,
6216     set_stp_port,
6217     get_stp_port_status,
6218     set_queues,
6219     bundle_set,
6220     bundle_remove,
6221     mirror_set__,
6222     mirror_get_stats__,
6223     set_flood_vlans,
6224     is_mirror_output_bundle,
6225     forward_bpdu_changed,
6226     set_mac_table_config,
6227     set_realdev,
6228     NULL,                       /* meter_get_features */
6229     NULL,                       /* meter_set */
6230     NULL,                       /* meter_get */
6231     NULL,                       /* meter_del */
6232     group_alloc,                /* group_alloc */
6233     group_construct,            /* group_construct */
6234     group_destruct,             /* group_destruct */
6235     group_dealloc,              /* group_dealloc */
6236     group_modify,               /* group_modify */
6237     group_get_stats,            /* group_get_stats */
6238 };