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 "hmapx.h"
  35 #include "lacp.h"
  36 #include "learn.h"
  37 #include "mac-learning.h"
  38 #include "meta-flow.h"
  39 #include "multipath.h"
  40 #include "netdev-vport.h"
  41 #include "netdev.h"
  42 #include "netlink.h"
  43 #include "nx-match.h"
  44 #include "odp-util.h"
  45 #include "odp-execute.h"
  46 #include "ofp-util.h"
  47 #include "ofpbuf.h"
  48 #include "ofp-actions.h"
  49 #include "ofp-parse.h"
  50 #include "ofp-print.h"
  51 #include "ofproto-dpif-governor.h"
  52 #include "ofproto-dpif-ipfix.h"
  53 #include "ofproto-dpif-mirror.h"
  54 #include "ofproto-dpif-sflow.h"
  55 #include "ofproto-dpif-upcall.h"
  56 #include "ofproto-dpif-xlate.h"
  57 #include "poll-loop.h"
  58 #include "simap.h"
  59 #include "smap.h"
  60 #include "timer.h"
  61 #include "tunnel.h"
  62 #include "unaligned.h"
  63 #include "unixctl.h"
  64 #include "vlan-bitmap.h"
  65 #include "vlog.h"
  66
  67 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  68
  69 COVERAGE_DEFINE(ofproto_dpif_expired);
  70 COVERAGE_DEFINE(facet_changed_rule);
  71 COVERAGE_DEFINE(facet_revalidate);
  72 COVERAGE_DEFINE(facet_unexpected);
  73 COVERAGE_DEFINE(facet_suppress);
  74 COVERAGE_DEFINE(subfacet_install_fail);
  75 COVERAGE_DEFINE(packet_in_overflow);
  76 COVERAGE_DEFINE(flow_mod_overflow);
  77
  78 /* Number of implemented OpenFlow tables. */
  79 enum { N_TABLES = 255 };
  80 enum { TBL_INTERNAL = N_TABLES - 1 };    /* Used for internal hidden rules. */
  81 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
  82
  83 struct flow_miss;
  84 struct facet;
  85
  86 struct rule_dpif {
  87     struct rule up;
  88
  89     /* These statistics:
  90      *
  91      *   - Do include packets and bytes from facets that have been deleted or
  92      *     whose own statistics have been folded into the rule.
  93      *
  94      *   - Do include packets and bytes sent "by hand" that were accounted to
  95      *     the rule without any facet being involved (this is a rare corner
  96      *     case in rule_execute()).
  97      *
  98      *   - Do not include packet or bytes that can be obtained from any facet's
  99      *     packet_count or byte_count member or that can be obtained from the
 100      *     datapath by, e.g., dpif_flow_get() for any subfacet.
 101      */
 102     struct ovs_mutex stats_mutex;
 103     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 104     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 105 };
 106
 107 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
 108 static struct rule_dpif *rule_dpif_cast(const struct rule *);
 109
 110 struct ofbundle {
 111     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
 112     struct ofproto_dpif *ofproto; /* Owning ofproto. */
 113     void *aux;                  /* Key supplied by ofproto's client. */
 114     char *name;                 /* Identifier for log messages. */
 115
 116     /* Configuration. */
 117     struct list ports;          /* Contains "struct ofport"s. */
 118     enum port_vlan_mode vlan_mode; /* VLAN mode */
 119     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
 120     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
 121                                  * NULL if all VLANs are trunked. */
 122     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 123     struct bond *bond;          /* Nonnull iff more than one port. */
 124     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 125
 126     /* Status. */
 127     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 128 };
 129
 130 static void bundle_remove(struct ofport *);
 131 static void bundle_update(struct ofbundle *);
 132 static void bundle_destroy(struct ofbundle *);
 133 static void bundle_del_port(struct ofport_dpif *);
 134 static void bundle_run(struct ofbundle *);
 135 static void bundle_wait(struct ofbundle *);
 136
 137 static void stp_run(struct ofproto_dpif *ofproto);
 138 static void stp_wait(struct ofproto_dpif *ofproto);
 139 static int set_stp_port(struct ofport *,
 140                         const struct ofproto_port_stp_settings *);
 141
 142 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
 143                               enum slow_path_reason,
 144                               uint64_t *stub, size_t stub_size,
 145                               const struct nlattr **actionsp,
 146                               size_t *actions_lenp);
 147
 148 /* A subfacet (see "struct subfacet" below) has three possible installation
 149  * states:
 150  *
 151  *   - SF_NOT_INSTALLED: Not installed in the datapath.  This will only be the
 152  *     case just after the subfacet is created, just before the subfacet is
 153  *     destroyed, or if the datapath returns an error when we try to install a
 154  *     subfacet.
 155  *
 156  *   - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
 157  *
 158  *   - SF_SLOW_PATH: An action that sends every packet for the subfacet through
 159  *     ofproto_dpif is installed in the datapath.
 160  */
 161 enum subfacet_path {
 162     SF_NOT_INSTALLED,           /* No datapath flow for this subfacet. */
 163     SF_FAST_PATH,               /* Full actions are installed. */
 164     SF_SLOW_PATH,               /* Send-to-userspace action is installed. */
 165 };
 166
 167 /* A dpif flow and actions associated with a facet.
 168  *
 169  * See also the large comment on struct facet. */
 170 struct subfacet {
 171     /* Owners. */
 172     struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
 173     struct list list_node;      /* In struct facet's 'facets' list. */
 174     struct facet *facet;        /* Owning facet. */
 175     struct dpif_backer *backer; /* Owning backer. */
 176
 177     enum odp_key_fitness key_fitness;
 178     struct nlattr *key;
 179     int key_len;
 180
 181     long long int used;         /* Time last used; time created if not used. */
 182     long long int created;      /* Time created. */
 183
 184     uint64_t dp_packet_count;   /* Last known packet count in the datapath. */
 185     uint64_t dp_byte_count;     /* Last known byte count in the datapath. */
 186
 187     enum subfacet_path path;    /* Installed in datapath? */
 188 };
 189
 190 #define SUBFACET_DESTROY_MAX_BATCH 50
 191
 192 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *);
 193 static struct subfacet *subfacet_find(struct dpif_backer *,
 194                                       const struct nlattr *key, size_t key_len,
 195                                       uint32_t key_hash);
 196 static void subfacet_destroy(struct subfacet *);
 197 static void subfacet_destroy__(struct subfacet *);
 198 static void subfacet_destroy_batch(struct dpif_backer *,
 199                                    struct subfacet **, int n);
 200 static void subfacet_reset_dp_stats(struct subfacet *,
 201                                     struct dpif_flow_stats *);
 202 static void subfacet_update_stats(struct subfacet *,
 203                                   const struct dpif_flow_stats *);
 204 static int subfacet_install(struct subfacet *,
 205                             const struct ofpbuf *odp_actions,
 206                             struct dpif_flow_stats *);
 207 static void subfacet_uninstall(struct subfacet *);
 208
 209 /* A unique, non-overlapping instantiation of an OpenFlow flow.
 210  *
 211  * A facet associates a "struct flow", which represents the Open vSwitch
 212  * userspace idea of an exact-match flow, with one or more subfacets.
 213  * While the facet is created based on an exact-match flow, it is stored
 214  * within the ofproto based on the wildcards that could be expressed
 215  * based on the flow table and other configuration.  (See the 'wc'
 216  * description in "struct xlate_out" for more details.)
 217  *
 218  * Each subfacet tracks the datapath's idea of the flow equivalent to
 219  * the facet.  When the kernel module (or other dpif implementation) and
 220  * Open vSwitch userspace agree on the definition of a flow key, there
 221  * is exactly one subfacet per facet.  If the dpif implementation
 222  * supports more-specific flow matching than userspace, however, a facet
 223  * can have more than one subfacet.  Examples include the dpif
 224  * implementation not supporting the same wildcards as userspace or some
 225  * distinction in flow that userspace simply doesn't understand.
 226  *
 227  * Flow expiration works in terms of subfacets, so a facet must have at
 228  * least one subfacet or it will never expire, leaking memory. */
 229 struct facet {
 230     /* Owner. */
 231     struct ofproto_dpif *ofproto;
 232
 233     /* Owned data. */
 234     struct list subfacets;
 235     long long int used;         /* Time last used; time created if not used. */
 236
 237     /* Key. */
 238     struct flow flow;           /* Flow of the creating subfacet. */
 239     struct cls_rule cr;         /* In 'ofproto_dpif's facets classifier. */
 240
 241     /* These statistics:
 242      *
 243      *   - Do include packets and bytes sent "by hand", e.g. with
 244      *     dpif_execute().
 245      *
 246      *   - Do include packets and bytes that were obtained from the datapath
 247      *     when a subfacet's statistics were reset (e.g. dpif_flow_put() with
 248      *     DPIF_FP_ZERO_STATS).
 249      *
 250      *   - Do not include packets or bytes that can be obtained from the
 251      *     datapath for any existing subfacet.
 252      */
 253     uint64_t packet_count;       /* Number of packets received. */
 254     uint64_t byte_count;         /* Number of bytes received. */
 255
 256     /* Resubmit statistics. */
 257     uint64_t prev_packet_count;  /* Number of packets from last stats push. */
 258     uint64_t prev_byte_count;    /* Number of bytes from last stats push. */
 259     long long int prev_used;     /* Used time from last stats push. */
 260
 261     /* Accounting. */
 262     uint64_t accounted_bytes;    /* Bytes processed by facet_account(). */
 263     struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
 264     uint8_t tcp_flags;           /* TCP flags seen for this 'rule'. */
 265
 266     struct xlate_out xout;
 267
 268     /* Storage for a single subfacet, to reduce malloc() time and space
 269      * overhead.  (A facet always has at least one subfacet and in the common
 270      * case has exactly one subfacet.  However, 'one_subfacet' may not
 271      * always be valid, since it could have been removed after newer
 272      * subfacets were pushed onto the 'subfacets' list.) */
 273     struct subfacet one_subfacet;
 274
 275     long long int learn_rl;      /* Rate limiter for facet_learn(). */
 276 };
 277
 278 static struct facet *facet_create(const struct flow_miss *);
 279 static void facet_remove(struct facet *);
 280 static void facet_free(struct facet *);
 281
 282 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
 283 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
 284                                         const struct flow *);
 285 static bool facet_revalidate(struct facet *);
 286 static bool facet_check_consistency(struct facet *);
 287
 288 static void facet_flush_stats(struct facet *);
 289
 290 static void facet_reset_counters(struct facet *);
 291 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
 292                             struct dpif_flow_stats *, bool may_learn);
 293 static void facet_push_stats(struct facet *, bool may_learn);
 294 static void facet_learn(struct facet *);
 295 static void facet_account(struct facet *);
 296 static void push_all_stats(void);
 297
 298 static bool facet_is_controller_flow(struct facet *);
 299
 300 struct ofport_dpif {
 301     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 302     struct ofport up;
 303
 304     odp_port_t odp_port;
 305     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 306     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
 307     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 308     struct bfd *bfd;            /* BFD, if any. */
 309     bool may_enable;            /* May be enabled in bonds. */
 310     bool is_tunnel;             /* This port is a tunnel. */
 311     long long int carrier_seq;  /* Carrier status changes. */
 312     struct ofport_dpif *peer;   /* Peer if patch port. */
 313
 314     /* Spanning tree. */
 315     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 316     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 317     long long int stp_state_entered;
 318
 319     /* Queue to DSCP mapping. */
 320     struct ofproto_port_queue *qdscp;
 321     size_t n_qdscp;
 322
 323     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 324      *
 325      * This is deprecated.  It is only for compatibility with broken device
 326      * drivers in old versions of Linux that do not properly support VLANs when
 327      * VLAN devices are not used.  When broken device drivers are no longer in
 328      * widespread use, we will delete these interfaces. */
 329     ofp_port_t realdev_ofp_port;
 330     int vlandev_vid;
 331 };
 332
 333 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 334  *
 335  * This is deprecated.  It is only for compatibility with broken device drivers
 336  * in old versions of Linux that do not properly support VLANs when VLAN
 337  * devices are not used.  When broken device drivers are no longer in
 338  * widespread use, we will delete these interfaces. */
 339 struct vlan_splinter {
 340     struct hmap_node realdev_vid_node;
 341     struct hmap_node vlandev_node;
 342     ofp_port_t realdev_ofp_port;
 343     ofp_port_t vlandev_ofp_port;
 344     int vid;
 345 };
 346
 347 static void vsp_remove(struct ofport_dpif *);
 348 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 349
 350 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 351                                        ofp_port_t);
 352
 353 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 354                                        odp_port_t);
 355
 356 static struct ofport_dpif *
 357 ofport_dpif_cast(const struct ofport *ofport)
 358 {
 359     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 360 }
 361
 362 static void port_run(struct ofport_dpif *);
 363 static void port_run_fast(struct ofport_dpif *);
 364 static void port_wait(struct ofport_dpif *);
 365 static int set_bfd(struct ofport *, const struct smap *);
 366 static int set_cfm(struct ofport *, const struct cfm_settings *);
 367 static void ofport_update_peer(struct ofport_dpif *);
 368 static void run_fast_rl(void);
 369 static int run_fast(struct ofproto *);
 370
 371 struct dpif_completion {
 372     struct list list_node;
 373     struct ofoperation *op;
 374 };
 375
 376 /* Reasons that we might need to revalidate every facet, and corresponding
 377  * coverage counters.
 378  *
 379  * A value of 0 means that there is no need to revalidate.
 380  *
 381  * It would be nice to have some cleaner way to integrate with coverage
 382  * counters, but with only a few reasons I guess this is good enough for
 383  * now. */
 384 enum revalidate_reason {
 385     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 386     REV_STP,                   /* Spanning tree protocol port status change. */
 387     REV_BOND,                  /* Bonding changed. */
 388     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 389     REV_FLOW_TABLE,            /* Flow table changed. */
 390     REV_MAC_LEARNING,          /* Mac learning changed. */
 391     REV_INCONSISTENCY          /* Facet self-check failed. */
 392 };
 393 COVERAGE_DEFINE(rev_reconfigure);
 394 COVERAGE_DEFINE(rev_stp);
 395 COVERAGE_DEFINE(rev_bond);
 396 COVERAGE_DEFINE(rev_port_toggled);
 397 COVERAGE_DEFINE(rev_flow_table);
 398 COVERAGE_DEFINE(rev_mac_learning);
 399 COVERAGE_DEFINE(rev_inconsistency);
 400
 401 struct avg_subfacet_rates {
 402     double add_rate;   /* Moving average of new flows created per minute. */
 403     double del_rate;   /* Moving average of flows deleted per minute. */
 404 };
 405
 406 /* All datapaths of a given type share a single dpif backer instance. */
 407 struct dpif_backer {
 408     char *type;
 409     int refcount;
 410     struct dpif *dpif;
 411     struct udpif *udpif;
 412     struct timer next_expiration;
 413
 414     struct ovs_rwlock odp_to_ofport_lock;
 415     struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
 416
 417     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 418
 419     /* Facet revalidation flags applying to facets which use this backer. */
 420     enum revalidate_reason need_revalidate; /* Revalidate every facet. */
 421
 422     struct hmap drop_keys; /* Set of dropped odp keys. */
 423     bool recv_set_enable; /* Enables or disables receiving packets. */
 424
 425     struct hmap subfacets;
 426     struct governor *governor;
 427
 428     /* Subfacet statistics.
 429      *
 430      * These keep track of the total number of subfacets added and deleted and
 431      * flow life span.  They are useful for computing the flow rates stats
 432      * exposed via "ovs-appctl dpif/show".  The goal is to learn about
 433      * traffic patterns in ways that we can use later to improve Open vSwitch
 434      * performance in new situations.  */
 435     long long int created;           /* Time when it is created. */
 436     unsigned max_n_subfacet;         /* Maximum number of flows */
 437     unsigned avg_n_subfacet;         /* Average number of flows. */
 438     long long int avg_subfacet_life; /* Average life span of subfacets. */
 439
 440     /* The average number of subfacets... */
 441     struct avg_subfacet_rates hourly;   /* ...over the last hour. */
 442     struct avg_subfacet_rates daily;    /* ...over the last day. */
 443     struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
 444     long long int last_minute;          /* Last time 'hourly' was updated. */
 445
 446     /* Number of subfacets added or deleted since 'last_minute'. */
 447     unsigned subfacet_add_count;
 448     unsigned subfacet_del_count;
 449
 450     /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
 451     unsigned long long int total_subfacet_add_count;
 452     unsigned long long int total_subfacet_del_count;
 453
 454     /* Number of upcall handling threads. */
 455     unsigned int n_handler_threads;
 456 };
 457
 458 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 459 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 460
 461 static void drop_key_clear(struct dpif_backer *);
 462 static void update_moving_averages(struct dpif_backer *backer);
 463
 464 struct ofproto_dpif {
 465     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 466     struct ofproto up;
 467     struct dpif_backer *backer;
 468
 469     /* Special OpenFlow rules. */
 470     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 471     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 472     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 473
 474     /* Bridging. */
 475     struct netflow *netflow;
 476     struct dpif_sflow *sflow;
 477     struct dpif_ipfix *ipfix;
 478     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 479     struct mac_learning *ml;
 480     bool has_bonded_bundles;
 481     struct mbridge *mbridge;
 482
 483     /* Facets. */
 484     struct classifier facets;     /* Contains 'struct facet's. */
 485     long long int consistency_rl;
 486
 487     /* Support for debugging async flow mods. */
 488     struct list completions;
 489
 490     struct netdev_stats stats; /* To account packets generated and consumed in
 491                                 * userspace. */
 492
 493     /* Spanning tree. */
 494     struct stp *stp;
 495     long long int stp_last_tick;
 496
 497     /* VLAN splinters. */
 498     struct ovs_mutex vsp_mutex;
 499     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 500     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 501
 502     /* Ports. */
 503     struct sset ports;             /* Set of standard port names. */
 504     struct sset ghost_ports;       /* Ports with no datapath port. */
 505     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 506     int port_poll_errno;           /* Last errno for port_poll() reply. */
 507
 508     /* Per ofproto's dpif stats. */
 509     uint64_t n_hit;
 510     uint64_t n_missed;
 511
 512     /* Work queues. */
 513     struct ovs_mutex flow_mod_mutex;
 514     struct list flow_mods OVS_GUARDED;
 515     size_t n_flow_mods OVS_GUARDED;
 516
 517     struct ovs_mutex pin_mutex;
 518     struct list pins OVS_GUARDED;
 519     size_t n_pins OVS_GUARDED;
 520 };
 521
 522 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"?  (Useful only
 523  * for debugging the asynchronous flow_mod implementation.) */
 524 static bool clogged;
 525
 526 /* By default, flows in the datapath are wildcarded (megaflows).  They
 527  * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
 528 static bool enable_megaflows = true;
 529
 530 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 531 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 532
 533 static void ofproto_dpif_unixctl_init(void);
 534
 535 static inline struct ofproto_dpif *
 536 ofproto_dpif_cast(const struct ofproto *ofproto)
 537 {
 538     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 539     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 540 }
 541
 542 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
 543                                         ofp_port_t ofp_port);
 544 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
 545                           const struct ofpbuf *packet, struct ds *);
 546
 547 /* Upcalls. */
 548 static void handle_upcalls(struct dpif_backer *);
 549
 550 /* Flow expiration. */
 551 static int expire(struct dpif_backer *);
 552
 553 /* NetFlow. */
 554 static void send_netflow_active_timeouts(struct ofproto_dpif *);
 555
 556 /* Utilities. */
 557 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
 558
 559 /* Global variables. */
 560 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 561
 562 /* Initial mappings of port to bridge mappings. */
 563 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 564
 565 /* Executes and takes ownership of 'fm'. */
 566 void
 567 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 568                       struct ofputil_flow_mod *fm)
 569 {
 570     ovs_mutex_lock(&ofproto->flow_mod_mutex);
 571     if (ofproto->n_flow_mods > 1024) {
 572         ovs_mutex_unlock(&ofproto->flow_mod_mutex);
 573         COVERAGE_INC(flow_mod_overflow);
 574         free(fm->ofpacts);
 575         free(fm);
 576         return;
 577     }
 578
 579     list_push_back(&ofproto->flow_mods, &fm->list_node);
 580     ofproto->n_flow_mods++;
 581     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
 582 }
 583
 584 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 585  * Takes ownership of 'pin' and pin->packet. */
 586 void
 587 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 588                             struct ofputil_packet_in *pin)
 589 {
 590     ovs_mutex_lock(&ofproto->pin_mutex);
 591     if (ofproto->n_pins > 1024) {
 592         ovs_mutex_unlock(&ofproto->pin_mutex);
 593         COVERAGE_INC(packet_in_overflow);
 594         free(CONST_CAST(void *, pin->packet));
 595         free(pin);
 596         return;
 597     }
 598
 599     list_push_back(&ofproto->pins, &pin->list_node);
 600     ofproto->n_pins++;
 601     ovs_mutex_unlock(&ofproto->pin_mutex);
 602 }
 603 \f
 604 /* Factory functions. */
 605
 606 static void
 607 init(const struct shash *iface_hints)
 608 {
 609     struct shash_node *node;
 610
 611     /* Make a local copy, since we don't own 'iface_hints' elements. */
 612     SHASH_FOR_EACH(node, iface_hints) {
 613         const struct iface_hint *orig_hint = node->data;
 614         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 615
 616         new_hint->br_name = xstrdup(orig_hint->br_name);
 617         new_hint->br_type = xstrdup(orig_hint->br_type);
 618         new_hint->ofp_port = orig_hint->ofp_port;
 619
 620         shash_add(&init_ofp_ports, node->name, new_hint);
 621     }
 622 }
 623
 624 static void
 625 enumerate_types(struct sset *types)
 626 {
 627     dp_enumerate_types(types);
 628 }
 629
 630 static int
 631 enumerate_names(const char *type, struct sset *names)
 632 {
 633     struct ofproto_dpif *ofproto;
 634
 635     sset_clear(names);
 636     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 637         if (strcmp(type, ofproto->up.type)) {
 638             continue;
 639         }
 640         sset_add(names, ofproto->up.name);
 641     }
 642
 643     return 0;
 644 }
 645
 646 static int
 647 del(const char *type, const char *name)
 648 {
 649     struct dpif *dpif;
 650     int error;
 651
 652     error = dpif_open(name, type, &dpif);
 653     if (!error) {
 654         error = dpif_delete(dpif);
 655         dpif_close(dpif);
 656     }
 657     return error;
 658 }
 659 \f
 660 static const char *
 661 port_open_type(const char *datapath_type, const char *port_type)
 662 {
 663     return dpif_port_open_type(datapath_type, port_type);
 664 }
 665
 666 /* Type functions. */
 667
 668 static void process_dpif_port_changes(struct dpif_backer *);
 669 static void process_dpif_all_ports_changed(struct dpif_backer *);
 670 static void process_dpif_port_change(struct dpif_backer *,
 671                                      const char *devname);
 672 static void process_dpif_port_error(struct dpif_backer *, int error);
 673
 674 static struct ofproto_dpif *
 675 lookup_ofproto_dpif_by_port_name(const char *name)
 676 {
 677     struct ofproto_dpif *ofproto;
 678
 679     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 680         if (sset_contains(&ofproto->ports, name)) {
 681             return ofproto;
 682         }
 683     }
 684
 685     return NULL;
 686 }
 687
 688 static int
 689 type_run(const char *type)
 690 {
 691     static long long int push_timer = LLONG_MIN;
 692     struct dpif_backer *backer;
 693
 694     backer = shash_find_data(&all_dpif_backers, type);
 695     if (!backer) {
 696         /* This is not necessarily a problem, since backers are only
 697          * created on demand. */
 698         return 0;
 699     }
 700
 701     dpif_run(backer->dpif);
 702
 703     /* The most natural place to push facet statistics is when they're pulled
 704      * from the datapath.  However, when there are many flows in the datapath,
 705      * this expensive operation can occur so frequently, that it reduces our
 706      * ability to quickly set up flows.  To reduce the cost, we push statistics
 707      * here instead. */
 708     if (time_msec() > push_timer) {
 709         push_timer = time_msec() + 2000;
 710         push_all_stats();
 711     }
 712
 713     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 714      * and the configuration has now changed to "false", enable receiving
 715      * packets from the datapath. */
 716     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 717         int error;
 718
 719         backer->recv_set_enable = true;
 720
 721         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 722         if (error) {
 723             udpif_recv_set(backer->udpif, 0, false);
 724             VLOG_ERR("Failed to enable receiving packets in dpif.");
 725             return error;
 726         }
 727         udpif_recv_set(backer->udpif, n_handler_threads,
 728                        backer->recv_set_enable);
 729         dpif_flow_flush(backer->dpif);
 730         backer->need_revalidate = REV_RECONFIGURE;
 731     }
 732
 733     /* If the n_handler_threads is reconfigured, call udpif_recv_set()
 734      * to reset the handler threads. */
 735     if (backer->n_handler_threads != n_handler_threads) {
 736         udpif_recv_set(backer->udpif, n_handler_threads,
 737                        backer->recv_set_enable);
 738         backer->n_handler_threads = n_handler_threads;
 739     }
 740
 741     if (backer->need_revalidate) {
 742         struct ofproto_dpif *ofproto;
 743         struct simap_node *node;
 744         struct simap tmp_backers;
 745
 746         /* Handle tunnel garbage collection. */
 747         simap_init(&tmp_backers);
 748         simap_swap(&backer->tnl_backers, &tmp_backers);
 749
 750         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 751             struct ofport_dpif *iter;
 752
 753             if (backer != ofproto->backer) {
 754                 continue;
 755             }
 756
 757             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 758                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 759                 const char *dp_port;
 760
 761                 if (!iter->is_tunnel) {
 762                     continue;
 763                 }
 764
 765                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 766                                                      namebuf, sizeof namebuf);
 767                 node = simap_find(&tmp_backers, dp_port);
 768                 if (node) {
 769                     simap_put(&backer->tnl_backers, dp_port, node->data);
 770                     simap_delete(&tmp_backers, node);
 771                     node = simap_find(&backer->tnl_backers, dp_port);
 772                 } else {
 773                     node = simap_find(&backer->tnl_backers, dp_port);
 774                     if (!node) {
 775                         odp_port_t odp_port = ODPP_NONE;
 776
 777                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 778                                            &odp_port)) {
 779                             simap_put(&backer->tnl_backers, dp_port,
 780                                       odp_to_u32(odp_port));
 781                             node = simap_find(&backer->tnl_backers, dp_port);
 782                         }
 783                     }
 784                 }
 785
 786                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 787                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 788                                          iter->odp_port)) {
 789                     backer->need_revalidate = REV_RECONFIGURE;
 790                 }
 791             }
 792         }
 793
 794         SIMAP_FOR_EACH (node, &tmp_backers) {
 795             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 796         }
 797         simap_destroy(&tmp_backers);
 798
 799         switch (backer->need_revalidate) {
 800         case REV_RECONFIGURE:   COVERAGE_INC(rev_reconfigure);   break;
 801         case REV_STP:           COVERAGE_INC(rev_stp);           break;
 802         case REV_BOND:          COVERAGE_INC(rev_bond);          break;
 803         case REV_PORT_TOGGLED:  COVERAGE_INC(rev_port_toggled);  break;
 804         case REV_FLOW_TABLE:    COVERAGE_INC(rev_flow_table);    break;
 805         case REV_MAC_LEARNING:  COVERAGE_INC(rev_mac_learning);  break;
 806         case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
 807         }
 808         backer->need_revalidate = 0;
 809
 810         /* Clear the drop_keys in case we should now be accepting some
 811          * formerly dropped flows. */
 812         drop_key_clear(backer);
 813
 814         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 815             struct facet *facet, *next;
 816             struct ofport_dpif *ofport;
 817             struct cls_cursor cursor;
 818             struct ofbundle *bundle;
 819
 820             if (ofproto->backer != backer) {
 821                 continue;
 822             }
 823
 824             ovs_rwlock_wrlock(&xlate_rwlock);
 825             xlate_ofproto_set(ofproto, ofproto->up.name,
 826                               ofproto->backer->dpif, ofproto->miss_rule,
 827                               ofproto->no_packet_in_rule, ofproto->ml,
 828                               ofproto->stp, ofproto->mbridge,
 829                               ofproto->sflow, ofproto->ipfix,
 830                               ofproto->up.frag_handling,
 831                               ofproto->up.forward_bpdu,
 832                               connmgr_has_in_band(ofproto->up.connmgr),
 833                               ofproto->netflow != NULL);
 834
 835             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 836                 xlate_bundle_set(ofproto, bundle, bundle->name,
 837                                  bundle->vlan_mode, bundle->vlan,
 838                                  bundle->trunks, bundle->use_priority_tags,
 839                                  bundle->bond, bundle->lacp,
 840                                  bundle->floodable);
 841             }
 842
 843             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 844                 int stp_port = ofport->stp_port
 845                     ? stp_port_no(ofport->stp_port)
 846                     : -1;
 847                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 848                                  ofport->up.ofp_port, ofport->odp_port,
 849                                  ofport->up.netdev, ofport->cfm,
 850                                  ofport->bfd, ofport->peer, stp_port,
 851                                  ofport->qdscp, ofport->n_qdscp,
 852                                  ofport->up.pp.config, ofport->is_tunnel,
 853                                  ofport->may_enable);
 854             }
 855             ovs_rwlock_unlock(&xlate_rwlock);
 856
 857             /* Only ofproto-dpif cares about the facet classifier so we just
 858              * lock cls_cursor_init() to appease the thread safety analysis. */
 859             ovs_rwlock_rdlock(&ofproto->facets.rwlock);
 860             cls_cursor_init(&cursor, &ofproto->facets, NULL);
 861             ovs_rwlock_unlock(&ofproto->facets.rwlock);
 862             CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
 863                 facet_revalidate(facet);
 864                 run_fast_rl();
 865             }
 866         }
 867
 868         udpif_revalidate(backer->udpif);
 869     }
 870
 871     if (!backer->recv_set_enable) {
 872         /* Wake up before a max of 1000ms. */
 873         timer_set_duration(&backer->next_expiration, 1000);
 874     } else if (timer_expired(&backer->next_expiration)) {
 875         int delay = expire(backer);
 876         timer_set_duration(&backer->next_expiration, delay);
 877     }
 878
 879     process_dpif_port_changes(backer);
 880
 881     if (backer->governor) {
 882         size_t n_subfacets;
 883
 884         governor_run(backer->governor);
 885
 886         /* If the governor has shrunk to its minimum size and the number of
 887          * subfacets has dwindled, then drop the governor entirely.
 888          *
 889          * For hysteresis, the number of subfacets to drop the governor is
 890          * smaller than the number needed to trigger its creation. */
 891         n_subfacets = hmap_count(&backer->subfacets);
 892         if (n_subfacets * 4 < flow_eviction_threshold
 893             && governor_is_idle(backer->governor)) {
 894             governor_destroy(backer->governor);
 895             backer->governor = NULL;
 896         }
 897     }
 898
 899     return 0;
 900 }
 901
 902 /* Check for and handle port changes in 'backer''s dpif. */
 903 static void
 904 process_dpif_port_changes(struct dpif_backer *backer)
 905 {
 906     for (;;) {
 907         char *devname;
 908         int error;
 909
 910         error = dpif_port_poll(backer->dpif, &devname);
 911         switch (error) {
 912         case EAGAIN:
 913             return;
 914
 915         case ENOBUFS:
 916             process_dpif_all_ports_changed(backer);
 917             break;
 918
 919         case 0:
 920             process_dpif_port_change(backer, devname);
 921             free(devname);
 922             break;
 923
 924         default:
 925             process_dpif_port_error(backer, error);
 926             break;
 927         }
 928     }
 929 }
 930
 931 static void
 932 process_dpif_all_ports_changed(struct dpif_backer *backer)
 933 {
 934     struct ofproto_dpif *ofproto;
 935     struct dpif_port dpif_port;
 936     struct dpif_port_dump dump;
 937     struct sset devnames;
 938     const char *devname;
 939
 940     sset_init(&devnames);
 941     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 942         if (ofproto->backer == backer) {
 943             struct ofport *ofport;
 944
 945             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 946                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 947             }
 948         }
 949     }
 950     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 951         sset_add(&devnames, dpif_port.name);
 952     }
 953
 954     SSET_FOR_EACH (devname, &devnames) {
 955         process_dpif_port_change(backer, devname);
 956     }
 957     sset_destroy(&devnames);
 958 }
 959
 960 static void
 961 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 962 {
 963     struct ofproto_dpif *ofproto;
 964     struct dpif_port port;
 965
 966     /* Don't report on the datapath's device. */
 967     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 968         return;
 969     }
 970
 971     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 972                    &all_ofproto_dpifs) {
 973         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 974             return;
 975         }
 976     }
 977
 978     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 979     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 980         /* The port was removed.  If we know the datapath,
 981          * report it through poll_set().  If we don't, it may be
 982          * notifying us of a removal we initiated, so ignore it.
 983          * If there's a pending ENOBUFS, let it stand, since
 984          * everything will be reevaluated. */
 985         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 986             sset_add(&ofproto->port_poll_set, devname);
 987             ofproto->port_poll_errno = 0;
 988         }
 989     } else if (!ofproto) {
 990         /* The port was added, but we don't know with which
 991          * ofproto we should associate it.  Delete it. */
 992         dpif_port_del(backer->dpif, port.port_no);
 993     } else {
 994         struct ofport_dpif *ofport;
 995
 996         ofport = ofport_dpif_cast(shash_find_data(
 997                                       &ofproto->up.port_by_name, devname));
 998         if (ofport
 999             && ofport->odp_port != port.port_no
1000             && !odp_port_to_ofport(backer, port.port_no))
1001         {
1002             /* 'ofport''s datapath port number has changed from
1003              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
1004              * structures to match. */
1005             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
1006             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
1007             ofport->odp_port = port.port_no;
1008             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
1009                         hash_odp_port(port.port_no));
1010             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
1011             backer->need_revalidate = REV_RECONFIGURE;
1012         }
1013     }
1014     dpif_port_destroy(&port);
1015 }
1016
1017 /* Propagate 'error' to all ofprotos based on 'backer'. */
1018 static void
1019 process_dpif_port_error(struct dpif_backer *backer, int error)
1020 {
1021     struct ofproto_dpif *ofproto;
1022
1023     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1024         if (ofproto->backer == backer) {
1025             sset_clear(&ofproto->port_poll_set);
1026             ofproto->port_poll_errno = error;
1027         }
1028     }
1029 }
1030
1031 static int
1032 dpif_backer_run_fast(struct dpif_backer *backer)
1033 {
1034     udpif_run(backer->udpif);
1035     handle_upcalls(backer);
1036
1037     return 0;
1038 }
1039
1040 static int
1041 type_run_fast(const char *type)
1042 {
1043     struct dpif_backer *backer;
1044
1045     backer = shash_find_data(&all_dpif_backers, type);
1046     if (!backer) {
1047         /* This is not necessarily a problem, since backers are only
1048          * created on demand. */
1049         return 0;
1050     }
1051
1052     return dpif_backer_run_fast(backer);
1053 }
1054
1055 static void
1056 run_fast_rl(void)
1057 {
1058     static long long int port_rl = LLONG_MIN;
1059
1060     if (time_msec() >= port_rl) {
1061         struct ofproto_dpif *ofproto;
1062
1063         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1064             run_fast(&ofproto->up);
1065         }
1066         port_rl = time_msec() + 200;
1067     }
1068 }
1069
1070 static void
1071 type_wait(const char *type)
1072 {
1073     struct dpif_backer *backer;
1074
1075     backer = shash_find_data(&all_dpif_backers, type);
1076     if (!backer) {
1077         /* This is not necessarily a problem, since backers are only
1078          * created on demand. */
1079         return;
1080     }
1081
1082     if (backer->governor) {
1083         governor_wait(backer->governor);
1084     }
1085
1086     timer_wait(&backer->next_expiration);
1087     dpif_wait(backer->dpif);
1088     udpif_wait(backer->udpif);
1089 }
1090 \f
1091 /* Basic life-cycle. */
1092
1093 static int add_internal_flows(struct ofproto_dpif *);
1094
1095 static struct ofproto *
1096 alloc(void)
1097 {
1098     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1099     return &ofproto->up;
1100 }
1101
1102 static void
1103 dealloc(struct ofproto *ofproto_)
1104 {
1105     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1106     free(ofproto);
1107 }
1108
1109 static void
1110 close_dpif_backer(struct dpif_backer *backer)
1111 {
1112     struct shash_node *node;
1113
1114     ovs_assert(backer->refcount > 0);
1115
1116     if (--backer->refcount) {
1117         return;
1118     }
1119
1120     drop_key_clear(backer);
1121     hmap_destroy(&backer->drop_keys);
1122
1123     simap_destroy(&backer->tnl_backers);
1124     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1125     hmap_destroy(&backer->odp_to_ofport_map);
1126     node = shash_find(&all_dpif_backers, backer->type);
1127     free(backer->type);
1128     shash_delete(&all_dpif_backers, node);
1129     udpif_destroy(backer->udpif);
1130     dpif_close(backer->dpif);
1131
1132     ovs_assert(hmap_is_empty(&backer->subfacets));
1133     hmap_destroy(&backer->subfacets);
1134     governor_destroy(backer->governor);
1135
1136     free(backer);
1137 }
1138
1139 /* Datapath port slated for removal from datapath. */
1140 struct odp_garbage {
1141     struct list list_node;
1142     odp_port_t odp_port;
1143 };
1144
1145 static int
1146 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1147 {
1148     struct dpif_backer *backer;
1149     struct dpif_port_dump port_dump;
1150     struct dpif_port port;
1151     struct shash_node *node;
1152     struct list garbage_list;
1153     struct odp_garbage *garbage, *next;
1154     struct sset names;
1155     char *backer_name;
1156     const char *name;
1157     int error;
1158
1159     backer = shash_find_data(&all_dpif_backers, type);
1160     if (backer) {
1161         backer->refcount++;
1162         *backerp = backer;
1163         return 0;
1164     }
1165
1166     backer_name = xasprintf("ovs-%s", type);
1167
1168     /* Remove any existing datapaths, since we assume we're the only
1169      * userspace controlling the datapath. */
1170     sset_init(&names);
1171     dp_enumerate_names(type, &names);
1172     SSET_FOR_EACH(name, &names) {
1173         struct dpif *old_dpif;
1174
1175         /* Don't remove our backer if it exists. */
1176         if (!strcmp(name, backer_name)) {
1177             continue;
1178         }
1179
1180         if (dpif_open(name, type, &old_dpif)) {
1181             VLOG_WARN("couldn't open old datapath %s to remove it", name);
1182         } else {
1183             dpif_delete(old_dpif);
1184             dpif_close(old_dpif);
1185         }
1186     }
1187     sset_destroy(&names);
1188
1189     backer = xmalloc(sizeof *backer);
1190
1191     error = dpif_create_and_open(backer_name, type, &backer->dpif);
1192     free(backer_name);
1193     if (error) {
1194         VLOG_ERR("failed to open datapath of type %s: %s", type,
1195                  ovs_strerror(error));
1196         free(backer);
1197         return error;
1198     }
1199     backer->udpif = udpif_create(backer, backer->dpif);
1200
1201     backer->type = xstrdup(type);
1202     backer->governor = NULL;
1203     backer->refcount = 1;
1204     hmap_init(&backer->odp_to_ofport_map);
1205     ovs_rwlock_init(&backer->odp_to_ofport_lock);
1206     hmap_init(&backer->drop_keys);
1207     hmap_init(&backer->subfacets);
1208     timer_set_duration(&backer->next_expiration, 1000);
1209     backer->need_revalidate = 0;
1210     simap_init(&backer->tnl_backers);
1211     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1212     *backerp = backer;
1213
1214     if (backer->recv_set_enable) {
1215         dpif_flow_flush(backer->dpif);
1216     }
1217
1218     /* Loop through the ports already on the datapath and remove any
1219      * that we don't need anymore. */
1220     list_init(&garbage_list);
1221     dpif_port_dump_start(&port_dump, backer->dpif);
1222     while (dpif_port_dump_next(&port_dump, &port)) {
1223         node = shash_find(&init_ofp_ports, port.name);
1224         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1225             garbage = xmalloc(sizeof *garbage);
1226             garbage->odp_port = port.port_no;
1227             list_push_front(&garbage_list, &garbage->list_node);
1228         }
1229     }
1230     dpif_port_dump_done(&port_dump);
1231
1232     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1233         dpif_port_del(backer->dpif, garbage->odp_port);
1234         list_remove(&garbage->list_node);
1235         free(garbage);
1236     }
1237
1238     shash_add(&all_dpif_backers, type, backer);
1239
1240     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1241     if (error) {
1242         VLOG_ERR("failed to listen on datapath of type %s: %s",
1243                  type, ovs_strerror(error));
1244         close_dpif_backer(backer);
1245         return error;
1246     }
1247     udpif_recv_set(backer->udpif, n_handler_threads,
1248                    backer->recv_set_enable);
1249     backer->n_handler_threads = n_handler_threads;
1250
1251     backer->max_n_subfacet = 0;
1252     backer->created = time_msec();
1253     backer->last_minute = backer->created;
1254     memset(&backer->hourly, 0, sizeof backer->hourly);
1255     memset(&backer->daily, 0, sizeof backer->daily);
1256     memset(&backer->lifetime, 0, sizeof backer->lifetime);
1257     backer->subfacet_add_count = 0;
1258     backer->subfacet_del_count = 0;
1259     backer->total_subfacet_add_count = 0;
1260     backer->total_subfacet_del_count = 0;
1261     backer->avg_n_subfacet = 0;
1262     backer->avg_subfacet_life = 0;
1263
1264     return error;
1265 }
1266
1267 static int
1268 construct(struct ofproto *ofproto_)
1269 {
1270     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1271     struct shash_node *node, *next;
1272     uint32_t max_ports;
1273     int error;
1274
1275     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1276     if (error) {
1277         return error;
1278     }
1279
1280     max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1281     ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX)));
1282
1283     ofproto->netflow = NULL;
1284     ofproto->sflow = NULL;
1285     ofproto->ipfix = NULL;
1286     ofproto->stp = NULL;
1287     hmap_init(&ofproto->bundles);
1288     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1289     ofproto->mbridge = mbridge_create();
1290     ofproto->has_bonded_bundles = false;
1291     ovs_mutex_init(&ofproto->vsp_mutex);
1292
1293     classifier_init(&ofproto->facets);
1294     ofproto->consistency_rl = LLONG_MIN;
1295
1296     list_init(&ofproto->completions);
1297
1298     ovs_mutex_init(&ofproto->flow_mod_mutex);
1299     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1300     list_init(&ofproto->flow_mods);
1301     ofproto->n_flow_mods = 0;
1302     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1303
1304     ovs_mutex_init(&ofproto->pin_mutex);
1305     ovs_mutex_lock(&ofproto->pin_mutex);
1306     list_init(&ofproto->pins);
1307     ofproto->n_pins = 0;
1308     ovs_mutex_unlock(&ofproto->pin_mutex);
1309
1310     ofproto_dpif_unixctl_init();
1311
1312     hmap_init(&ofproto->vlandev_map);
1313     hmap_init(&ofproto->realdev_vid_map);
1314
1315     sset_init(&ofproto->ports);
1316     sset_init(&ofproto->ghost_ports);
1317     sset_init(&ofproto->port_poll_set);
1318     ofproto->port_poll_errno = 0;
1319
1320     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1321         struct iface_hint *iface_hint = node->data;
1322
1323         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1324             /* Check if the datapath already has this port. */
1325             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1326                 sset_add(&ofproto->ports, node->name);
1327             }
1328
1329             free(iface_hint->br_name);
1330             free(iface_hint->br_type);
1331             free(iface_hint);
1332             shash_delete(&init_ofp_ports, node);
1333         }
1334     }
1335
1336     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1337                 hash_string(ofproto->up.name, 0));
1338     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1339
1340     ofproto_init_tables(ofproto_, N_TABLES);
1341     error = add_internal_flows(ofproto);
1342     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1343
1344     ofproto->n_hit = 0;
1345     ofproto->n_missed = 0;
1346
1347     return error;
1348 }
1349
1350 static int
1351 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1352                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1353 {
1354     struct ofputil_flow_mod fm;
1355     int error;
1356
1357     match_init_catchall(&fm.match);
1358     fm.priority = 0;
1359     match_set_reg(&fm.match, 0, id);
1360     fm.new_cookie = htonll(0);
1361     fm.cookie = htonll(0);
1362     fm.cookie_mask = htonll(0);
1363     fm.modify_cookie = false;
1364     fm.table_id = TBL_INTERNAL;
1365     fm.command = OFPFC_ADD;
1366     fm.idle_timeout = 0;
1367     fm.hard_timeout = 0;
1368     fm.buffer_id = 0;
1369     fm.out_port = 0;
1370     fm.flags = 0;
1371     fm.ofpacts = ofpacts->data;
1372     fm.ofpacts_len = ofpacts->size;
1373
1374     error = ofproto_flow_mod(&ofproto->up, &fm);
1375     if (error) {
1376         VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1377                     id, ofperr_to_string(error));
1378         return error;
1379     }
1380
1381     if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1382                                   rulep)) {
1383         rule_dpif_release(*rulep);
1384     } else {
1385         NOT_REACHED();
1386     }
1387
1388     return 0;
1389 }
1390
1391 static int
1392 add_internal_flows(struct ofproto_dpif *ofproto)
1393 {
1394     struct ofpact_controller *controller;
1395     uint64_t ofpacts_stub[128 / 8];
1396     struct ofpbuf ofpacts;
1397     int error;
1398     int id;
1399
1400     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1401     id = 1;
1402
1403     controller = ofpact_put_CONTROLLER(&ofpacts);
1404     controller->max_len = UINT16_MAX;
1405     controller->controller_id = 0;
1406     controller->reason = OFPR_NO_MATCH;
1407     ofpact_pad(&ofpacts);
1408
1409     error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1410     if (error) {
1411         return error;
1412     }
1413
1414     ofpbuf_clear(&ofpacts);
1415     error = add_internal_flow(ofproto, id++, &ofpacts,
1416                               &ofproto->no_packet_in_rule);
1417     if (error) {
1418         return error;
1419     }
1420
1421     error = add_internal_flow(ofproto, id++, &ofpacts,
1422                               &ofproto->drop_frags_rule);
1423     return error;
1424 }
1425
1426 static void
1427 complete_operations(struct ofproto_dpif *ofproto)
1428 {
1429     struct dpif_completion *c, *next;
1430
1431     LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1432         ofoperation_complete(c->op, 0);
1433         list_remove(&c->list_node);
1434         free(c);
1435     }
1436 }
1437
1438 static void
1439 destruct(struct ofproto *ofproto_)
1440 {
1441     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1442     struct rule_dpif *rule, *next_rule;
1443     struct ofputil_packet_in *pin, *next_pin;
1444     struct ofputil_flow_mod *fm, *next_fm;
1445     struct facet *facet, *next_facet;
1446     struct cls_cursor cursor;
1447     struct oftable *table;
1448
1449     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1450     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1451     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1452     CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1453         facet_remove(facet);
1454     }
1455
1456     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1457     ovs_rwlock_wrlock(&xlate_rwlock);
1458     xlate_remove_ofproto(ofproto);
1459     ovs_rwlock_unlock(&xlate_rwlock);
1460
1461     flow_miss_batch_ofproto_destroyed(ofproto->backer->udpif, ofproto);
1462
1463     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1464     complete_operations(ofproto);
1465
1466     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1467         struct cls_cursor cursor;
1468
1469         ovs_rwlock_wrlock(&table->cls.rwlock);
1470         cls_cursor_init(&cursor, &table->cls, NULL);
1471         CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1472             ofproto_rule_delete(&ofproto->up, &table->cls, &rule->up);
1473         }
1474         ovs_rwlock_unlock(&table->cls.rwlock);
1475     }
1476     complete_operations(ofproto);
1477
1478     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1479     LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &ofproto->flow_mods) {
1480         list_remove(&fm->list_node);
1481         ofproto->n_flow_mods--;
1482         free(fm->ofpacts);
1483         free(fm);
1484     }
1485     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1486     ovs_mutex_destroy(&ofproto->flow_mod_mutex);
1487
1488     ovs_mutex_lock(&ofproto->pin_mutex);
1489     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &ofproto->pins) {
1490         list_remove(&pin->list_node);
1491         ofproto->n_pins--;
1492         free(CONST_CAST(void *, pin->packet));
1493         free(pin);
1494     }
1495     ovs_mutex_unlock(&ofproto->pin_mutex);
1496     ovs_mutex_destroy(&ofproto->pin_mutex);
1497
1498     mbridge_unref(ofproto->mbridge);
1499
1500     netflow_destroy(ofproto->netflow);
1501     dpif_sflow_unref(ofproto->sflow);
1502     hmap_destroy(&ofproto->bundles);
1503     mac_learning_unref(ofproto->ml);
1504
1505     classifier_destroy(&ofproto->facets);
1506
1507     hmap_destroy(&ofproto->vlandev_map);
1508     hmap_destroy(&ofproto->realdev_vid_map);
1509
1510     sset_destroy(&ofproto->ports);
1511     sset_destroy(&ofproto->ghost_ports);
1512     sset_destroy(&ofproto->port_poll_set);
1513
1514     ovs_mutex_destroy(&ofproto->vsp_mutex);
1515
1516     close_dpif_backer(ofproto->backer);
1517 }
1518
1519 static int
1520 run_fast(struct ofproto *ofproto_)
1521 {
1522     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1523     struct ofputil_packet_in *pin, *next_pin;
1524     struct ofputil_flow_mod *fm, *next_fm;
1525     struct list flow_mods, pins;
1526     struct ofport_dpif *ofport;
1527
1528     /* Do not perform any periodic activity required by 'ofproto' while
1529      * waiting for flow restore to complete. */
1530     if (ofproto_get_flow_restore_wait()) {
1531         return 0;
1532     }
1533
1534     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1535     list_move(&flow_mods, &ofproto->flow_mods);
1536     list_init(&ofproto->flow_mods);
1537     ofproto->n_flow_mods = 0;
1538     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1539
1540     LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &flow_mods) {
1541         int error = ofproto_flow_mod(&ofproto->up, fm);
1542         if (error && !VLOG_DROP_WARN(&rl)) {
1543             VLOG_WARN("learning action failed to modify flow table (%s)",
1544                       ofperr_get_name(error));
1545         }
1546
1547         list_remove(&fm->list_node);
1548         free(fm->ofpacts);
1549         free(fm);
1550     }
1551
1552     ovs_mutex_lock(&ofproto->pin_mutex);
1553     list_move(&pins, &ofproto->pins);
1554     list_init(&ofproto->pins);
1555     ofproto->n_pins = 0;
1556     ovs_mutex_unlock(&ofproto->pin_mutex);
1557
1558     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1559         connmgr_send_packet_in(ofproto->up.connmgr, pin);
1560         list_remove(&pin->list_node);
1561         free(CONST_CAST(void *, pin->packet));
1562         free(pin);
1563     }
1564
1565     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1566         port_run_fast(ofport);
1567     }
1568
1569     return 0;
1570 }
1571
1572 static int
1573 run(struct ofproto *ofproto_)
1574 {
1575     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1576     struct ofport_dpif *ofport;
1577     struct ofbundle *bundle;
1578     int error;
1579
1580     if (!clogged) {
1581         complete_operations(ofproto);
1582     }
1583
1584     if (mbridge_need_revalidate(ofproto->mbridge)) {
1585         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1586         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1587         mac_learning_flush(ofproto->ml);
1588         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1589     }
1590
1591     /* Do not perform any periodic activity below required by 'ofproto' while
1592      * waiting for flow restore to complete. */
1593     if (ofproto_get_flow_restore_wait()) {
1594         return 0;
1595     }
1596
1597     error = run_fast(ofproto_);
1598     if (error) {
1599         return error;
1600     }
1601
1602     if (ofproto->netflow) {
1603         if (netflow_run(ofproto->netflow)) {
1604             send_netflow_active_timeouts(ofproto);
1605         }
1606     }
1607     if (ofproto->sflow) {
1608         dpif_sflow_run(ofproto->sflow);
1609     }
1610     if (ofproto->ipfix) {
1611         dpif_ipfix_run(ofproto->ipfix);
1612     }
1613
1614     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1615         port_run(ofport);
1616     }
1617     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1618         bundle_run(bundle);
1619     }
1620
1621     stp_run(ofproto);
1622     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1623     if (mac_learning_run(ofproto->ml)) {
1624         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1625     }
1626     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1627
1628     /* Check the consistency of a random facet, to aid debugging. */
1629     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1630     if (time_msec() >= ofproto->consistency_rl
1631         && !classifier_is_empty(&ofproto->facets)
1632         && !ofproto->backer->need_revalidate) {
1633         struct cls_table *table;
1634         struct cls_rule *cr;
1635         struct facet *facet;
1636
1637         ofproto->consistency_rl = time_msec() + 250;
1638
1639         table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1640                              struct cls_table, hmap_node);
1641         cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1642                           hmap_node);
1643         facet = CONTAINER_OF(cr, struct facet, cr);
1644
1645         if (!facet_check_consistency(facet)) {
1646             ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1647         }
1648     }
1649     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1650
1651     return 0;
1652 }
1653
1654 static void
1655 wait(struct ofproto *ofproto_)
1656 {
1657     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1658     struct ofport_dpif *ofport;
1659     struct ofbundle *bundle;
1660
1661     if (!clogged && !list_is_empty(&ofproto->completions)) {
1662         poll_immediate_wake();
1663     }
1664
1665     if (ofproto_get_flow_restore_wait()) {
1666         return;
1667     }
1668
1669     if (ofproto->sflow) {
1670         dpif_sflow_wait(ofproto->sflow);
1671     }
1672     if (ofproto->ipfix) {
1673         dpif_ipfix_wait(ofproto->ipfix);
1674     }
1675     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1676         port_wait(ofport);
1677     }
1678     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1679         bundle_wait(bundle);
1680     }
1681     if (ofproto->netflow) {
1682         netflow_wait(ofproto->netflow);
1683     }
1684     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1685     mac_learning_wait(ofproto->ml);
1686     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1687     stp_wait(ofproto);
1688     if (ofproto->backer->need_revalidate) {
1689         /* Shouldn't happen, but if it does just go around again. */
1690         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1691         poll_immediate_wake();
1692     }
1693 }
1694
1695 static void
1696 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1697 {
1698     const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1699     struct cls_cursor cursor;
1700     size_t n_subfacets = 0;
1701     struct facet *facet;
1702
1703     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1704     simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1705     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1706
1707     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1708     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1709     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1710         n_subfacets += list_size(&facet->subfacets);
1711     }
1712     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1713     simap_increase(usage, "subfacets", n_subfacets);
1714 }
1715
1716 static void
1717 flush(struct ofproto *ofproto_)
1718 {
1719     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1720     struct subfacet *subfacet, *next_subfacet;
1721     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1722     int n_batch;
1723
1724     n_batch = 0;
1725     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1726                         &ofproto->backer->subfacets) {
1727         if (subfacet->facet->ofproto != ofproto) {
1728             continue;
1729         }
1730
1731         if (subfacet->path != SF_NOT_INSTALLED) {
1732             batch[n_batch++] = subfacet;
1733             if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1734                 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1735                 n_batch = 0;
1736             }
1737         } else {
1738             subfacet_destroy(subfacet);
1739         }
1740     }
1741
1742     if (n_batch > 0) {
1743         subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1744     }
1745 }
1746
1747 static void
1748 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1749              bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1750 {
1751     *arp_match_ip = true;
1752     *actions = (OFPUTIL_A_OUTPUT |
1753                 OFPUTIL_A_SET_VLAN_VID |
1754                 OFPUTIL_A_SET_VLAN_PCP |
1755                 OFPUTIL_A_STRIP_VLAN |
1756                 OFPUTIL_A_SET_DL_SRC |
1757                 OFPUTIL_A_SET_DL_DST |
1758                 OFPUTIL_A_SET_NW_SRC |
1759                 OFPUTIL_A_SET_NW_DST |
1760                 OFPUTIL_A_SET_NW_TOS |
1761                 OFPUTIL_A_SET_TP_SRC |
1762                 OFPUTIL_A_SET_TP_DST |
1763                 OFPUTIL_A_ENQUEUE);
1764 }
1765
1766 static void
1767 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1768 {
1769     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1770     struct dpif_dp_stats s;
1771     uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1772     uint64_t n_lookup;
1773
1774     strcpy(ots->name, "classifier");
1775
1776     dpif_get_dp_stats(ofproto->backer->dpif, &s);
1777     rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1778     rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1779     rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1780
1781     n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1782     ots->lookup_count = htonll(n_lookup);
1783     ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1784 }
1785
1786 static struct ofport *
1787 port_alloc(void)
1788 {
1789     struct ofport_dpif *port = xmalloc(sizeof *port);
1790     return &port->up;
1791 }
1792
1793 static void
1794 port_dealloc(struct ofport *port_)
1795 {
1796     struct ofport_dpif *port = ofport_dpif_cast(port_);
1797     free(port);
1798 }
1799
1800 static int
1801 port_construct(struct ofport *port_)
1802 {
1803     struct ofport_dpif *port = ofport_dpif_cast(port_);
1804     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1805     const struct netdev *netdev = port->up.netdev;
1806     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1807     struct dpif_port dpif_port;
1808     int error;
1809
1810     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1811     port->bundle = NULL;
1812     port->cfm = NULL;
1813     port->bfd = NULL;
1814     port->may_enable = true;
1815     port->stp_port = NULL;
1816     port->stp_state = STP_DISABLED;
1817     port->is_tunnel = false;
1818     port->peer = NULL;
1819     port->qdscp = NULL;
1820     port->n_qdscp = 0;
1821     port->realdev_ofp_port = 0;
1822     port->vlandev_vid = 0;
1823     port->carrier_seq = netdev_get_carrier_resets(netdev);
1824
1825     if (netdev_vport_is_patch(netdev)) {
1826         /* By bailing out here, we don't submit the port to the sFlow module
1827          * to be considered for counter polling export.  This is correct
1828          * because the patch port represents an interface that sFlow considers
1829          * to be "internal" to the switch as a whole, and therefore not an
1830          * candidate for counter polling. */
1831         port->odp_port = ODPP_NONE;
1832         ofport_update_peer(port);
1833         return 0;
1834     }
1835
1836     error = dpif_port_query_by_name(ofproto->backer->dpif,
1837                                     netdev_vport_get_dpif_port(netdev, namebuf,
1838                                                                sizeof namebuf),
1839                                     &dpif_port);
1840     if (error) {
1841         return error;
1842     }
1843
1844     port->odp_port = dpif_port.port_no;
1845
1846     if (netdev_get_tunnel_config(netdev)) {
1847         tnl_port_add(port, port->up.netdev, port->odp_port);
1848         port->is_tunnel = true;
1849     } else {
1850         /* Sanity-check that a mapping doesn't already exist.  This
1851          * shouldn't happen for non-tunnel ports. */
1852         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1853             VLOG_ERR("port %s already has an OpenFlow port number",
1854                      dpif_port.name);
1855             dpif_port_destroy(&dpif_port);
1856             return EBUSY;
1857         }
1858
1859         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1860         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1861                     hash_odp_port(port->odp_port));
1862         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1863     }
1864     dpif_port_destroy(&dpif_port);
1865
1866     if (ofproto->sflow) {
1867         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1868     }
1869
1870     return 0;
1871 }
1872
1873 static void
1874 port_destruct(struct ofport *port_)
1875 {
1876     struct ofport_dpif *port = ofport_dpif_cast(port_);
1877     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1878     const char *devname = netdev_get_name(port->up.netdev);
1879     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1880     const char *dp_port_name;
1881
1882     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1883     ovs_rwlock_wrlock(&xlate_rwlock);
1884     xlate_ofport_remove(port);
1885     ovs_rwlock_unlock(&xlate_rwlock);
1886
1887     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1888                                               sizeof namebuf);
1889     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1890         /* The underlying device is still there, so delete it.  This
1891          * happens when the ofproto is being destroyed, since the caller
1892          * assumes that removal of attached ports will happen as part of
1893          * destruction. */
1894         if (!port->is_tunnel) {
1895             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1896         }
1897     }
1898
1899     if (port->peer) {
1900         port->peer->peer = NULL;
1901         port->peer = NULL;
1902     }
1903
1904     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1905         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1906         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1907         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1908     }
1909
1910     tnl_port_del(port);
1911     sset_find_and_delete(&ofproto->ports, devname);
1912     sset_find_and_delete(&ofproto->ghost_ports, devname);
1913     bundle_remove(port_);
1914     set_cfm(port_, NULL);
1915     set_bfd(port_, NULL);
1916     if (ofproto->sflow) {
1917         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1918     }
1919
1920     free(port->qdscp);
1921 }
1922
1923 static void
1924 port_modified(struct ofport *port_)
1925 {
1926     struct ofport_dpif *port = ofport_dpif_cast(port_);
1927
1928     if (port->bundle && port->bundle->bond) {
1929         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1930     }
1931
1932     if (port->cfm) {
1933         cfm_set_netdev(port->cfm, port->up.netdev);
1934     }
1935
1936     if (port->bfd) {
1937         bfd_set_netdev(port->bfd, port->up.netdev);
1938     }
1939
1940     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1941                                                 port->odp_port)) {
1942         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1943             REV_RECONFIGURE;
1944     }
1945
1946     ofport_update_peer(port);
1947 }
1948
1949 static void
1950 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1951 {
1952     struct ofport_dpif *port = ofport_dpif_cast(port_);
1953     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1954     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1955
1956     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1957                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1958                    OFPUTIL_PC_NO_PACKET_IN)) {
1959         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1960
1961         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1962             bundle_update(port->bundle);
1963         }
1964     }
1965 }
1966
1967 static int
1968 set_sflow(struct ofproto *ofproto_,
1969           const struct ofproto_sflow_options *sflow_options)
1970 {
1971     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1972     struct dpif_sflow *ds = ofproto->sflow;
1973
1974     if (sflow_options) {
1975         if (!ds) {
1976             struct ofport_dpif *ofport;
1977
1978             ds = ofproto->sflow = dpif_sflow_create();
1979             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1980                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1981             }
1982             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1983         }
1984         dpif_sflow_set_options(ds, sflow_options);
1985     } else {
1986         if (ds) {
1987             dpif_sflow_unref(ds);
1988             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1989             ofproto->sflow = NULL;
1990         }
1991     }
1992     return 0;
1993 }
1994
1995 static int
1996 set_ipfix(
1997     struct ofproto *ofproto_,
1998     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1999     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
2000     size_t n_flow_exporters_options)
2001 {
2002     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2003     struct dpif_ipfix *di = ofproto->ipfix;
2004     bool has_options = bridge_exporter_options || flow_exporters_options;
2005
2006     if (has_options && !di) {
2007         di = ofproto->ipfix = dpif_ipfix_create();
2008     }
2009
2010     if (di) {
2011         /* Call set_options in any case to cleanly flush the flow
2012          * caches in the last exporters that are to be destroyed. */
2013         dpif_ipfix_set_options(
2014             di, bridge_exporter_options, flow_exporters_options,
2015             n_flow_exporters_options);
2016
2017         if (!has_options) {
2018             dpif_ipfix_unref(di);
2019             ofproto->ipfix = NULL;
2020         }
2021     }
2022
2023     return 0;
2024 }
2025
2026 static int
2027 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
2028 {
2029     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2030     int error;
2031
2032     if (!s) {
2033         error = 0;
2034     } else {
2035         if (!ofport->cfm) {
2036             struct ofproto_dpif *ofproto;
2037
2038             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2039             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2040             ofport->cfm = cfm_create(ofport->up.netdev);
2041         }
2042
2043         if (cfm_configure(ofport->cfm, s)) {
2044             return 0;
2045         }
2046
2047         error = EINVAL;
2048     }
2049     cfm_unref(ofport->cfm);
2050     ofport->cfm = NULL;
2051     return error;
2052 }
2053
2054 static bool
2055 get_cfm_status(const struct ofport *ofport_,
2056                struct ofproto_cfm_status *status)
2057 {
2058     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2059
2060     if (ofport->cfm) {
2061         status->faults = cfm_get_fault(ofport->cfm);
2062         status->remote_opstate = cfm_get_opup(ofport->cfm);
2063         status->health = cfm_get_health(ofport->cfm);
2064         cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2065         return true;
2066     } else {
2067         return false;
2068     }
2069 }
2070
2071 static int
2072 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2073 {
2074     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2075     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2076     struct bfd *old;
2077
2078     old = ofport->bfd;
2079     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2080                                 cfg, ofport->up.netdev);
2081     if (ofport->bfd != old) {
2082         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2083     }
2084
2085     return 0;
2086 }
2087
2088 static int
2089 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2090 {
2091     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2092
2093     if (ofport->bfd) {
2094         bfd_get_status(ofport->bfd, smap);
2095         return 0;
2096     } else {
2097         return ENOENT;
2098     }
2099 }
2100 \f
2101 /* Spanning Tree. */
2102
2103 static void
2104 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2105 {
2106     struct ofproto_dpif *ofproto = ofproto_;
2107     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2108     struct ofport_dpif *ofport;
2109
2110     ofport = stp_port_get_aux(sp);
2111     if (!ofport) {
2112         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2113                      ofproto->up.name, port_num);
2114     } else {
2115         struct eth_header *eth = pkt->l2;
2116
2117         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2118         if (eth_addr_is_zero(eth->eth_src)) {
2119             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2120                          "with unknown MAC", ofproto->up.name, port_num);
2121         } else {
2122             send_packet(ofport, pkt);
2123         }
2124     }
2125     ofpbuf_delete(pkt);
2126 }
2127
2128 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2129 static int
2130 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2131 {
2132     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2133
2134     /* Only revalidate flows if the configuration changed. */
2135     if (!s != !ofproto->stp) {
2136         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2137     }
2138
2139     if (s) {
2140         if (!ofproto->stp) {
2141             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2142                                       send_bpdu_cb, ofproto);
2143             ofproto->stp_last_tick = time_msec();
2144         }
2145
2146         stp_set_bridge_id(ofproto->stp, s->system_id);
2147         stp_set_bridge_priority(ofproto->stp, s->priority);
2148         stp_set_hello_time(ofproto->stp, s->hello_time);
2149         stp_set_max_age(ofproto->stp, s->max_age);
2150         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2151     }  else {
2152         struct ofport *ofport;
2153
2154         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2155             set_stp_port(ofport, NULL);
2156         }
2157
2158         stp_unref(ofproto->stp);
2159         ofproto->stp = NULL;
2160     }
2161
2162     return 0;
2163 }
2164
2165 static int
2166 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2167 {
2168     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2169
2170     if (ofproto->stp) {
2171         s->enabled = true;
2172         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2173         s->designated_root = stp_get_designated_root(ofproto->stp);
2174         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2175     } else {
2176         s->enabled = false;
2177     }
2178
2179     return 0;
2180 }
2181
2182 static void
2183 update_stp_port_state(struct ofport_dpif *ofport)
2184 {
2185     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2186     enum stp_state state;
2187
2188     /* Figure out new state. */
2189     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2190                              : STP_DISABLED;
2191
2192     /* Update state. */
2193     if (ofport->stp_state != state) {
2194         enum ofputil_port_state of_state;
2195         bool fwd_change;
2196
2197         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2198                     netdev_get_name(ofport->up.netdev),
2199                     stp_state_name(ofport->stp_state),
2200                     stp_state_name(state));
2201         if (stp_learn_in_state(ofport->stp_state)
2202                 != stp_learn_in_state(state)) {
2203             /* xxx Learning action flows should also be flushed. */
2204             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2205             mac_learning_flush(ofproto->ml);
2206             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2207         }
2208         fwd_change = stp_forward_in_state(ofport->stp_state)
2209                         != stp_forward_in_state(state);
2210
2211         ofproto->backer->need_revalidate = REV_STP;
2212         ofport->stp_state = state;
2213         ofport->stp_state_entered = time_msec();
2214
2215         if (fwd_change && ofport->bundle) {
2216             bundle_update(ofport->bundle);
2217         }
2218
2219         /* Update the STP state bits in the OpenFlow port description. */
2220         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2221         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2222                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2223                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2224                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2225                      : 0);
2226         ofproto_port_set_state(&ofport->up, of_state);
2227     }
2228 }
2229
2230 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2231  * caller is responsible for assigning STP port numbers and ensuring
2232  * there are no duplicates. */
2233 static int
2234 set_stp_port(struct ofport *ofport_,
2235              const struct ofproto_port_stp_settings *s)
2236 {
2237     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2238     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2239     struct stp_port *sp = ofport->stp_port;
2240
2241     if (!s || !s->enable) {
2242         if (sp) {
2243             ofport->stp_port = NULL;
2244             stp_port_disable(sp);
2245             update_stp_port_state(ofport);
2246         }
2247         return 0;
2248     } else if (sp && stp_port_no(sp) != s->port_num
2249             && ofport == stp_port_get_aux(sp)) {
2250         /* The port-id changed, so disable the old one if it's not
2251          * already in use by another port. */
2252         stp_port_disable(sp);
2253     }
2254
2255     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2256     stp_port_enable(sp);
2257
2258     stp_port_set_aux(sp, ofport);
2259     stp_port_set_priority(sp, s->priority);
2260     stp_port_set_path_cost(sp, s->path_cost);
2261
2262     update_stp_port_state(ofport);
2263
2264     return 0;
2265 }
2266
2267 static int
2268 get_stp_port_status(struct ofport *ofport_,
2269                     struct ofproto_port_stp_status *s)
2270 {
2271     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2272     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2273     struct stp_port *sp = ofport->stp_port;
2274
2275     if (!ofproto->stp || !sp) {
2276         s->enabled = false;
2277         return 0;
2278     }
2279
2280     s->enabled = true;
2281     s->port_id = stp_port_get_id(sp);
2282     s->state = stp_port_get_state(sp);
2283     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2284     s->role = stp_port_get_role(sp);
2285     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2286
2287     return 0;
2288 }
2289
2290 static void
2291 stp_run(struct ofproto_dpif *ofproto)
2292 {
2293     if (ofproto->stp) {
2294         long long int now = time_msec();
2295         long long int elapsed = now - ofproto->stp_last_tick;
2296         struct stp_port *sp;
2297
2298         if (elapsed > 0) {
2299             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2300             ofproto->stp_last_tick = now;
2301         }
2302         while (stp_get_changed_port(ofproto->stp, &sp)) {
2303             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2304
2305             if (ofport) {
2306                 update_stp_port_state(ofport);
2307             }
2308         }
2309
2310         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2311             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2312             mac_learning_flush(ofproto->ml);
2313             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2314         }
2315     }
2316 }
2317
2318 static void
2319 stp_wait(struct ofproto_dpif *ofproto)
2320 {
2321     if (ofproto->stp) {
2322         poll_timer_wait(1000);
2323     }
2324 }
2325 \f
2326 static int
2327 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2328            size_t n_qdscp)
2329 {
2330     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2331     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2332
2333     if (ofport->n_qdscp != n_qdscp
2334         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2335                               n_qdscp * sizeof *qdscp))) {
2336         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2337         free(ofport->qdscp);
2338         ofport->qdscp = n_qdscp
2339             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2340             : NULL;
2341         ofport->n_qdscp = n_qdscp;
2342     }
2343
2344     return 0;
2345 }
2346 \f
2347 /* Bundles. */
2348
2349 /* Expires all MAC learning entries associated with 'bundle' and forces its
2350  * ofproto to revalidate every flow.
2351  *
2352  * Normally MAC learning entries are removed only from the ofproto associated
2353  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2354  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2355  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2356  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2357  * with the host from which it migrated. */
2358 static void
2359 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2360 {
2361     struct ofproto_dpif *ofproto = bundle->ofproto;
2362     struct mac_learning *ml = ofproto->ml;
2363     struct mac_entry *mac, *next_mac;
2364
2365     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2366     ovs_rwlock_wrlock(&ml->rwlock);
2367     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2368         if (mac->port.p == bundle) {
2369             if (all_ofprotos) {
2370                 struct ofproto_dpif *o;
2371
2372                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2373                     if (o != ofproto) {
2374                         struct mac_entry *e;
2375
2376                         ovs_rwlock_wrlock(&o->ml->rwlock);
2377                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2378                         if (e) {
2379                             mac_learning_expire(o->ml, e);
2380                         }
2381                         ovs_rwlock_unlock(&o->ml->rwlock);
2382                     }
2383                 }
2384             }
2385
2386             mac_learning_expire(ml, mac);
2387         }
2388     }
2389     ovs_rwlock_unlock(&ml->rwlock);
2390 }
2391
2392 static struct ofbundle *
2393 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2394 {
2395     struct ofbundle *bundle;
2396
2397     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2398                              &ofproto->bundles) {
2399         if (bundle->aux == aux) {
2400             return bundle;
2401         }
2402     }
2403     return NULL;
2404 }
2405
2406 static void
2407 bundle_update(struct ofbundle *bundle)
2408 {
2409     struct ofport_dpif *port;
2410
2411     bundle->floodable = true;
2412     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2413         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2414             || !stp_forward_in_state(port->stp_state)) {
2415             bundle->floodable = false;
2416             break;
2417         }
2418     }
2419 }
2420
2421 static void
2422 bundle_del_port(struct ofport_dpif *port)
2423 {
2424     struct ofbundle *bundle = port->bundle;
2425
2426     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2427
2428     list_remove(&port->bundle_node);
2429     port->bundle = NULL;
2430
2431     if (bundle->lacp) {
2432         lacp_slave_unregister(bundle->lacp, port);
2433     }
2434     if (bundle->bond) {
2435         bond_slave_unregister(bundle->bond, port);
2436     }
2437
2438     bundle_update(bundle);
2439 }
2440
2441 static bool
2442 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2443                 struct lacp_slave_settings *lacp)
2444 {
2445     struct ofport_dpif *port;
2446
2447     port = get_ofp_port(bundle->ofproto, ofp_port);
2448     if (!port) {
2449         return false;
2450     }
2451
2452     if (port->bundle != bundle) {
2453         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2454         if (port->bundle) {
2455             bundle_remove(&port->up);
2456         }
2457
2458         port->bundle = bundle;
2459         list_push_back(&bundle->ports, &port->bundle_node);
2460         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2461             || !stp_forward_in_state(port->stp_state)) {
2462             bundle->floodable = false;
2463         }
2464     }
2465     if (lacp) {
2466         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2467         lacp_slave_register(bundle->lacp, port, lacp);
2468     }
2469
2470     return true;
2471 }
2472
2473 static void
2474 bundle_destroy(struct ofbundle *bundle)
2475 {
2476     struct ofproto_dpif *ofproto;
2477     struct ofport_dpif *port, *next_port;
2478
2479     if (!bundle) {
2480         return;
2481     }
2482
2483     ofproto = bundle->ofproto;
2484     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2485
2486     ovs_rwlock_wrlock(&xlate_rwlock);
2487     xlate_bundle_remove(bundle);
2488     ovs_rwlock_unlock(&xlate_rwlock);
2489
2490     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2491         bundle_del_port(port);
2492     }
2493
2494     bundle_flush_macs(bundle, true);
2495     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2496     free(bundle->name);
2497     free(bundle->trunks);
2498     lacp_unref(bundle->lacp);
2499     bond_unref(bundle->bond);
2500     free(bundle);
2501 }
2502
2503 static int
2504 bundle_set(struct ofproto *ofproto_, void *aux,
2505            const struct ofproto_bundle_settings *s)
2506 {
2507     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2508     bool need_flush = false;
2509     struct ofport_dpif *port;
2510     struct ofbundle *bundle;
2511     unsigned long *trunks;
2512     int vlan;
2513     size_t i;
2514     bool ok;
2515
2516     if (!s) {
2517         bundle_destroy(bundle_lookup(ofproto, aux));
2518         return 0;
2519     }
2520
2521     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2522     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2523
2524     bundle = bundle_lookup(ofproto, aux);
2525     if (!bundle) {
2526         bundle = xmalloc(sizeof *bundle);
2527
2528         bundle->ofproto = ofproto;
2529         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2530                     hash_pointer(aux, 0));
2531         bundle->aux = aux;
2532         bundle->name = NULL;
2533
2534         list_init(&bundle->ports);
2535         bundle->vlan_mode = PORT_VLAN_TRUNK;
2536         bundle->vlan = -1;
2537         bundle->trunks = NULL;
2538         bundle->use_priority_tags = s->use_priority_tags;
2539         bundle->lacp = NULL;
2540         bundle->bond = NULL;
2541
2542         bundle->floodable = true;
2543         mbridge_register_bundle(ofproto->mbridge, bundle);
2544     }
2545
2546     if (!bundle->name || strcmp(s->name, bundle->name)) {
2547         free(bundle->name);
2548         bundle->name = xstrdup(s->name);
2549     }
2550
2551     /* LACP. */
2552     if (s->lacp) {
2553         if (!bundle->lacp) {
2554             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2555             bundle->lacp = lacp_create();
2556         }
2557         lacp_configure(bundle->lacp, s->lacp);
2558     } else {
2559         lacp_unref(bundle->lacp);
2560         bundle->lacp = NULL;
2561     }
2562
2563     /* Update set of ports. */
2564     ok = true;
2565     for (i = 0; i < s->n_slaves; i++) {
2566         if (!bundle_add_port(bundle, s->slaves[i],
2567                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2568             ok = false;
2569         }
2570     }
2571     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2572         struct ofport_dpif *next_port;
2573
2574         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2575             for (i = 0; i < s->n_slaves; i++) {
2576                 if (s->slaves[i] == port->up.ofp_port) {
2577                     goto found;
2578                 }
2579             }
2580
2581             bundle_del_port(port);
2582         found: ;
2583         }
2584     }
2585     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2586
2587     if (list_is_empty(&bundle->ports)) {
2588         bundle_destroy(bundle);
2589         return EINVAL;
2590     }
2591
2592     /* Set VLAN tagging mode */
2593     if (s->vlan_mode != bundle->vlan_mode
2594         || s->use_priority_tags != bundle->use_priority_tags) {
2595         bundle->vlan_mode = s->vlan_mode;
2596         bundle->use_priority_tags = s->use_priority_tags;
2597         need_flush = true;
2598     }
2599
2600     /* Set VLAN tag. */
2601     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2602             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2603             : 0);
2604     if (vlan != bundle->vlan) {
2605         bundle->vlan = vlan;
2606         need_flush = true;
2607     }
2608
2609     /* Get trunked VLANs. */
2610     switch (s->vlan_mode) {
2611     case PORT_VLAN_ACCESS:
2612         trunks = NULL;
2613         break;
2614
2615     case PORT_VLAN_TRUNK:
2616         trunks = CONST_CAST(unsigned long *, s->trunks);
2617         break;
2618
2619     case PORT_VLAN_NATIVE_UNTAGGED:
2620     case PORT_VLAN_NATIVE_TAGGED:
2621         if (vlan != 0 && (!s->trunks
2622                           || !bitmap_is_set(s->trunks, vlan)
2623                           || bitmap_is_set(s->trunks, 0))) {
2624             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2625             if (s->trunks) {
2626                 trunks = bitmap_clone(s->trunks, 4096);
2627             } else {
2628                 trunks = bitmap_allocate1(4096);
2629             }
2630             bitmap_set1(trunks, vlan);
2631             bitmap_set0(trunks, 0);
2632         } else {
2633             trunks = CONST_CAST(unsigned long *, s->trunks);
2634         }
2635         break;
2636
2637     default:
2638         NOT_REACHED();
2639     }
2640     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2641         free(bundle->trunks);
2642         if (trunks == s->trunks) {
2643             bundle->trunks = vlan_bitmap_clone(trunks);
2644         } else {
2645             bundle->trunks = trunks;
2646             trunks = NULL;
2647         }
2648         need_flush = true;
2649     }
2650     if (trunks != s->trunks) {
2651         free(trunks);
2652     }
2653
2654     /* Bonding. */
2655     if (!list_is_short(&bundle->ports)) {
2656         bundle->ofproto->has_bonded_bundles = true;
2657         if (bundle->bond) {
2658             if (bond_reconfigure(bundle->bond, s->bond)) {
2659                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2660             }
2661         } else {
2662             bundle->bond = bond_create(s->bond);
2663             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2664         }
2665
2666         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2667             bond_slave_register(bundle->bond, port, port->up.netdev);
2668         }
2669     } else {
2670         bond_unref(bundle->bond);
2671         bundle->bond = NULL;
2672     }
2673
2674     /* If we changed something that would affect MAC learning, un-learn
2675      * everything on this port and force flow revalidation. */
2676     if (need_flush) {
2677         bundle_flush_macs(bundle, false);
2678     }
2679
2680     return 0;
2681 }
2682
2683 static void
2684 bundle_remove(struct ofport *port_)
2685 {
2686     struct ofport_dpif *port = ofport_dpif_cast(port_);
2687     struct ofbundle *bundle = port->bundle;
2688
2689     if (bundle) {
2690         bundle_del_port(port);
2691         if (list_is_empty(&bundle->ports)) {
2692             bundle_destroy(bundle);
2693         } else if (list_is_short(&bundle->ports)) {
2694             bond_unref(bundle->bond);
2695             bundle->bond = NULL;
2696         }
2697     }
2698 }
2699
2700 static void
2701 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2702 {
2703     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2704     struct ofport_dpif *port = port_;
2705     uint8_t ea[ETH_ADDR_LEN];
2706     int error;
2707
2708     error = netdev_get_etheraddr(port->up.netdev, ea);
2709     if (!error) {
2710         struct ofpbuf packet;
2711         void *packet_pdu;
2712
2713         ofpbuf_init(&packet, 0);
2714         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2715                                  pdu_size);
2716         memcpy(packet_pdu, pdu, pdu_size);
2717
2718         send_packet(port, &packet);
2719         ofpbuf_uninit(&packet);
2720     } else {
2721         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2722                     "%s (%s)", port->bundle->name,
2723                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2724     }
2725 }
2726
2727 static void
2728 bundle_send_learning_packets(struct ofbundle *bundle)
2729 {
2730     struct ofproto_dpif *ofproto = bundle->ofproto;
2731     int error, n_packets, n_errors;
2732     struct mac_entry *e;
2733
2734     error = n_packets = n_errors = 0;
2735     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2736     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2737         if (e->port.p != bundle) {
2738             struct ofpbuf *learning_packet;
2739             struct ofport_dpif *port;
2740             void *port_void;
2741             int ret;
2742
2743             /* The assignment to "port" is unnecessary but makes "grep"ing for
2744              * struct ofport_dpif more effective. */
2745             learning_packet = bond_compose_learning_packet(bundle->bond,
2746                                                            e->mac, e->vlan,
2747                                                            &port_void);
2748             port = port_void;
2749             ret = send_packet(port, learning_packet);
2750             ofpbuf_delete(learning_packet);
2751             if (ret) {
2752                 error = ret;
2753                 n_errors++;
2754             }
2755             n_packets++;
2756         }
2757     }
2758     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2759
2760     if (n_errors) {
2761         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2762         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2763                      "packets, last error was: %s",
2764                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2765     } else {
2766         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2767                  bundle->name, n_packets);
2768     }
2769 }
2770
2771 static void
2772 bundle_run(struct ofbundle *bundle)
2773 {
2774     if (bundle->lacp) {
2775         lacp_run(bundle->lacp, send_pdu_cb);
2776     }
2777     if (bundle->bond) {
2778         struct ofport_dpif *port;
2779
2780         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2781             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2782         }
2783
2784         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2785             bundle->ofproto->backer->need_revalidate = REV_BOND;
2786         }
2787
2788         if (bond_should_send_learning_packets(bundle->bond)) {
2789             bundle_send_learning_packets(bundle);
2790         }
2791     }
2792 }
2793
2794 static void
2795 bundle_wait(struct ofbundle *bundle)
2796 {
2797     if (bundle->lacp) {
2798         lacp_wait(bundle->lacp);
2799     }
2800     if (bundle->bond) {
2801         bond_wait(bundle->bond);
2802     }
2803 }
2804 \f
2805 /* Mirrors. */
2806
2807 static int
2808 mirror_set__(struct ofproto *ofproto_, void *aux,
2809              const struct ofproto_mirror_settings *s)
2810 {
2811     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2812     struct ofbundle **srcs, **dsts;
2813     int error;
2814     size_t i;
2815
2816     if (!s) {
2817         mirror_destroy(ofproto->mbridge, aux);
2818         return 0;
2819     }
2820
2821     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2822     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2823
2824     for (i = 0; i < s->n_srcs; i++) {
2825         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2826     }
2827
2828     for (i = 0; i < s->n_dsts; i++) {
2829         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2830     }
2831
2832     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2833                        s->n_dsts, s->src_vlans,
2834                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2835     free(srcs);
2836     free(dsts);
2837     return error;
2838 }
2839
2840 static int
2841 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2842                    uint64_t *packets, uint64_t *bytes)
2843 {
2844     push_all_stats();
2845     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2846                             bytes);
2847 }
2848
2849 static int
2850 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2851 {
2852     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2853     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2854     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2855         mac_learning_flush(ofproto->ml);
2856     }
2857     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2858     return 0;
2859 }
2860
2861 static bool
2862 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2863 {
2864     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2865     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2866     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2867 }
2868
2869 static void
2870 forward_bpdu_changed(struct ofproto *ofproto_)
2871 {
2872     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2873     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2874 }
2875
2876 static void
2877 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2878                      size_t max_entries)
2879 {
2880     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2881     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2882     mac_learning_set_idle_time(ofproto->ml, idle_time);
2883     mac_learning_set_max_entries(ofproto->ml, max_entries);
2884     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2885 }
2886 \f
2887 /* Ports. */
2888
2889 static struct ofport_dpif *
2890 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2891 {
2892     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2893     return ofport ? ofport_dpif_cast(ofport) : NULL;
2894 }
2895
2896 static struct ofport_dpif *
2897 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2898 {
2899     struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2900     return port && &ofproto->up == port->up.ofproto ? port : NULL;
2901 }
2902
2903 static void
2904 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2905                             struct ofproto_port *ofproto_port,
2906                             struct dpif_port *dpif_port)
2907 {
2908     ofproto_port->name = dpif_port->name;
2909     ofproto_port->type = dpif_port->type;
2910     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2911 }
2912
2913 static void
2914 ofport_update_peer(struct ofport_dpif *ofport)
2915 {
2916     const struct ofproto_dpif *ofproto;
2917     struct dpif_backer *backer;
2918     char *peer_name;
2919
2920     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2921         return;
2922     }
2923
2924     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2925     backer->need_revalidate = REV_RECONFIGURE;
2926
2927     if (ofport->peer) {
2928         ofport->peer->peer = NULL;
2929         ofport->peer = NULL;
2930     }
2931
2932     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2933     if (!peer_name) {
2934         return;
2935     }
2936
2937     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2938         struct ofport *peer_ofport;
2939         struct ofport_dpif *peer;
2940         char *peer_peer;
2941
2942         if (ofproto->backer != backer) {
2943             continue;
2944         }
2945
2946         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2947         if (!peer_ofport) {
2948             continue;
2949         }
2950
2951         peer = ofport_dpif_cast(peer_ofport);
2952         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2953         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2954                                  peer_peer)) {
2955             ofport->peer = peer;
2956             ofport->peer->peer = ofport;
2957         }
2958         free(peer_peer);
2959
2960         break;
2961     }
2962     free(peer_name);
2963 }
2964
2965 static void
2966 port_run_fast(struct ofport_dpif *ofport)
2967 {
2968     if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2969         struct ofpbuf packet;
2970
2971         ofpbuf_init(&packet, 0);
2972         cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2973         send_packet(ofport, &packet);
2974         ofpbuf_uninit(&packet);
2975     }
2976
2977     if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2978         struct ofpbuf packet;
2979
2980         ofpbuf_init(&packet, 0);
2981         bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2982         send_packet(ofport, &packet);
2983         ofpbuf_uninit(&packet);
2984     }
2985 }
2986
2987 static void
2988 port_run(struct ofport_dpif *ofport)
2989 {
2990     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2991     bool carrier_changed = carrier_seq != ofport->carrier_seq;
2992     bool enable = netdev_get_carrier(ofport->up.netdev);
2993     bool cfm_enable = false;
2994     bool bfd_enable = false;
2995
2996     ofport->carrier_seq = carrier_seq;
2997
2998     port_run_fast(ofport);
2999
3000     if (ofport->cfm) {
3001         int cfm_opup = cfm_get_opup(ofport->cfm);
3002
3003         cfm_run(ofport->cfm);
3004         cfm_enable = !cfm_get_fault(ofport->cfm);
3005
3006         if (cfm_opup >= 0) {
3007             cfm_enable = cfm_enable && cfm_opup;
3008         }
3009     }
3010
3011     if (ofport->bfd) {
3012         bfd_run(ofport->bfd);
3013         bfd_enable = bfd_forwarding(ofport->bfd);
3014     }
3015
3016     if (ofport->bfd || ofport->cfm) {
3017         enable = enable && (cfm_enable || bfd_enable);
3018     }
3019
3020     if (ofport->bundle) {
3021         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3022         if (carrier_changed) {
3023             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3024         }
3025     }
3026
3027     if (ofport->may_enable != enable) {
3028         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3029         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3030     }
3031
3032     ofport->may_enable = enable;
3033 }
3034
3035 static void
3036 port_wait(struct ofport_dpif *ofport)
3037 {
3038     if (ofport->cfm) {
3039         cfm_wait(ofport->cfm);
3040     }
3041
3042     if (ofport->bfd) {
3043         bfd_wait(ofport->bfd);
3044     }
3045 }
3046
3047 static int
3048 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3049                    struct ofproto_port *ofproto_port)
3050 {
3051     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3052     struct dpif_port dpif_port;
3053     int error;
3054
3055     if (sset_contains(&ofproto->ghost_ports, devname)) {
3056         const char *type = netdev_get_type_from_name(devname);
3057
3058         /* We may be called before ofproto->up.port_by_name is populated with
3059          * the appropriate ofport.  For this reason, we must get the name and
3060          * type from the netdev layer directly. */
3061         if (type) {
3062             const struct ofport *ofport;
3063
3064             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3065             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3066             ofproto_port->name = xstrdup(devname);
3067             ofproto_port->type = xstrdup(type);
3068             return 0;
3069         }
3070         return ENODEV;
3071     }
3072
3073     if (!sset_contains(&ofproto->ports, devname)) {
3074         return ENODEV;
3075     }
3076     error = dpif_port_query_by_name(ofproto->backer->dpif,
3077                                     devname, &dpif_port);
3078     if (!error) {
3079         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3080     }
3081     return error;
3082 }
3083
3084 static int
3085 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3086 {
3087     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3088     const char *devname = netdev_get_name(netdev);
3089     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3090     const char *dp_port_name;
3091
3092     if (netdev_vport_is_patch(netdev)) {
3093         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3094         return 0;
3095     }
3096
3097     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3098     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3099         odp_port_t port_no = ODPP_NONE;
3100         int error;
3101
3102         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3103         if (error) {
3104             return error;
3105         }
3106         if (netdev_get_tunnel_config(netdev)) {
3107             simap_put(&ofproto->backer->tnl_backers,
3108                       dp_port_name, odp_to_u32(port_no));
3109         }
3110     }
3111
3112     if (netdev_get_tunnel_config(netdev)) {
3113         sset_add(&ofproto->ghost_ports, devname);
3114     } else {
3115         sset_add(&ofproto->ports, devname);
3116     }
3117     return 0;
3118 }
3119
3120 static int
3121 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3122 {
3123     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3124     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3125     int error = 0;
3126
3127     if (!ofport) {
3128         return 0;
3129     }
3130
3131     sset_find_and_delete(&ofproto->ghost_ports,
3132                          netdev_get_name(ofport->up.netdev));
3133     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3134     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3135         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3136         if (!error) {
3137             /* The caller is going to close ofport->up.netdev.  If this is a
3138              * bonded port, then the bond is using that netdev, so remove it
3139              * from the bond.  The client will need to reconfigure everything
3140              * after deleting ports, so then the slave will get re-added. */
3141             bundle_remove(&ofport->up);
3142         }
3143     }
3144     return error;
3145 }
3146
3147 static int
3148 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3149 {
3150     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3151     int error;
3152
3153     push_all_stats();
3154
3155     error = netdev_get_stats(ofport->up.netdev, stats);
3156
3157     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3158         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3159
3160         /* ofproto->stats.tx_packets represents packets that we created
3161          * internally and sent to some port (e.g. packets sent with
3162          * send_packet()).  Account for them as if they had come from
3163          * OFPP_LOCAL and got forwarded. */
3164
3165         if (stats->rx_packets != UINT64_MAX) {
3166             stats->rx_packets += ofproto->stats.tx_packets;
3167         }
3168
3169         if (stats->rx_bytes != UINT64_MAX) {
3170             stats->rx_bytes += ofproto->stats.tx_bytes;
3171         }
3172
3173         /* ofproto->stats.rx_packets represents packets that were received on
3174          * some port and we processed internally and dropped (e.g. STP).
3175          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3176
3177         if (stats->tx_packets != UINT64_MAX) {
3178             stats->tx_packets += ofproto->stats.rx_packets;
3179         }
3180
3181         if (stats->tx_bytes != UINT64_MAX) {
3182             stats->tx_bytes += ofproto->stats.rx_bytes;
3183         }
3184     }
3185
3186     return error;
3187 }
3188
3189 struct port_dump_state {
3190     uint32_t bucket;
3191     uint32_t offset;
3192     bool ghost;
3193
3194     struct ofproto_port port;
3195     bool has_port;
3196 };
3197
3198 static int
3199 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3200 {
3201     *statep = xzalloc(sizeof(struct port_dump_state));
3202     return 0;
3203 }
3204
3205 static int
3206 port_dump_next(const struct ofproto *ofproto_, void *state_,
3207                struct ofproto_port *port)
3208 {
3209     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3210     struct port_dump_state *state = state_;
3211     const struct sset *sset;
3212     struct sset_node *node;
3213
3214     if (state->has_port) {
3215         ofproto_port_destroy(&state->port);
3216         state->has_port = false;
3217     }
3218     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3219     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3220         int error;
3221
3222         error = port_query_by_name(ofproto_, node->name, &state->port);
3223         if (!error) {
3224             *port = state->port;
3225             state->has_port = true;
3226             return 0;
3227         } else if (error != ENODEV) {
3228             return error;
3229         }
3230     }
3231
3232     if (!state->ghost) {
3233         state->ghost = true;
3234         state->bucket = 0;
3235         state->offset = 0;
3236         return port_dump_next(ofproto_, state_, port);
3237     }
3238
3239     return EOF;
3240 }
3241
3242 static int
3243 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3244 {
3245     struct port_dump_state *state = state_;
3246
3247     if (state->has_port) {
3248         ofproto_port_destroy(&state->port);
3249     }
3250     free(state);
3251     return 0;
3252 }
3253
3254 static int
3255 port_poll(const struct ofproto *ofproto_, char **devnamep)
3256 {
3257     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3258
3259     if (ofproto->port_poll_errno) {
3260         int error = ofproto->port_poll_errno;
3261         ofproto->port_poll_errno = 0;
3262         return error;
3263     }
3264
3265     if (sset_is_empty(&ofproto->port_poll_set)) {
3266         return EAGAIN;
3267     }
3268
3269     *devnamep = sset_pop(&ofproto->port_poll_set);
3270     return 0;
3271 }
3272
3273 static void
3274 port_poll_wait(const struct ofproto *ofproto_)
3275 {
3276     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3277     dpif_port_poll_wait(ofproto->backer->dpif);
3278 }
3279
3280 static int
3281 port_is_lacp_current(const struct ofport *ofport_)
3282 {
3283     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3284     return (ofport->bundle && ofport->bundle->lacp
3285             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3286             : -1);
3287 }
3288 \f
3289 /* Upcall handling. */
3290
3291 struct flow_miss_op {
3292     struct dpif_op dpif_op;
3293
3294     uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3295     struct xlate_out xout;
3296     bool xout_garbage;           /* 'xout' needs to be uninitialized? */
3297
3298     struct ofpbuf mask;          /* Flow mask for "put" ops. */
3299     struct odputil_keybuf maskbuf;
3300
3301     /* If this is a "put" op, then a pointer to the subfacet that should
3302      * be marked as uninstalled if the operation fails. */
3303     struct subfacet *subfacet;
3304 };
3305
3306 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3307  * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3308  * and (usually) installing a datapath flow.  The answer is usually "yes" (a
3309  * return value of true).  However, for short flows the cost of bookkeeping is
3310  * much higher than the benefits, so when the datapath holds a large number of
3311  * flows we impose some heuristics to decide which flows are likely to be worth
3312  * tracking. */
3313 static bool
3314 flow_miss_should_make_facet(struct flow_miss *miss)
3315 {
3316     struct dpif_backer *backer = miss->ofproto->backer;
3317     uint32_t hash;
3318
3319     switch (flow_miss_model) {
3320     case OFPROTO_HANDLE_MISS_AUTO:
3321         break;
3322     case OFPROTO_HANDLE_MISS_WITH_FACETS:
3323         return true;
3324     case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3325         return false;
3326     }
3327
3328     if (!backer->governor) {
3329         size_t n_subfacets;
3330
3331         n_subfacets = hmap_count(&backer->subfacets);
3332         if (n_subfacets * 2 <= flow_eviction_threshold) {
3333             return true;
3334         }
3335
3336         backer->governor = governor_create();
3337     }
3338
3339     hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3340     return governor_should_install_flow(backer->governor, hash,
3341                                         list_size(&miss->packets));
3342 }
3343
3344 /* Handles 'miss', which matches 'facet'.  May add any required datapath
3345  * operations to 'ops', incrementing '*n_ops' for each new op.
3346  *
3347  * All of the packets in 'miss' are considered to have arrived at time
3348  * 'miss->stats.used'.  This is really important only for new facets: if we
3349  * just called time_msec() here, then the new subfacet or its packets could
3350  * look (occasionally) as though it was used some time after the facet was
3351  * used.  That can make a one-packet flow look like it has a nonzero duration,
3352  * which looks odd in e.g. NetFlow statistics. */
3353 static void
3354 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3355                             struct flow_miss_op *ops, size_t *n_ops)
3356 {
3357     enum subfacet_path want_path;
3358     struct subfacet *subfacet;
3359
3360     facet->packet_count += miss->stats.n_packets;
3361     facet->prev_packet_count += miss->stats.n_packets;
3362     facet->byte_count += miss->stats.n_bytes;
3363     facet->prev_byte_count += miss->stats.n_bytes;
3364
3365     want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3366
3367     /* Don't install the flow if it's the result of the "userspace"
3368      * action for an already installed facet.  This can occur when a
3369      * datapath flow with wildcards has a "userspace" action and flows
3370      * sent to userspace result in a different subfacet, which will then
3371      * be rejected as overlapping by the datapath. */
3372     if (miss->upcall_type == DPIF_UC_ACTION
3373         && !list_is_empty(&facet->subfacets)) {
3374         return;
3375     }
3376
3377     subfacet = subfacet_create(facet, miss);
3378     if (subfacet->path != want_path) {
3379         struct flow_miss_op *op = &ops[(*n_ops)++];
3380         struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3381
3382         subfacet->path = want_path;
3383
3384         ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3385         if (enable_megaflows) {
3386             odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3387                                    &miss->flow, UINT32_MAX);
3388         }
3389
3390         op->xout_garbage = false;
3391         op->dpif_op.type = DPIF_OP_FLOW_PUT;
3392         op->subfacet = subfacet;
3393         put->flags = DPIF_FP_CREATE;
3394         put->key = miss->key;
3395         put->key_len = miss->key_len;
3396         put->mask = op->mask.data;
3397         put->mask_len = op->mask.size;
3398
3399         if (want_path == SF_FAST_PATH) {
3400             put->actions = facet->xout.odp_actions.data;
3401             put->actions_len = facet->xout.odp_actions.size;
3402         } else {
3403             compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3404                               op->slow_stub, sizeof op->slow_stub,
3405                               &put->actions, &put->actions_len);
3406         }
3407         put->stats = NULL;
3408     }
3409 }
3410
3411 /* Handles flow miss 'miss'.  May add any required datapath operations
3412  * to 'ops', incrementing '*n_ops' for each new op. */
3413 static void
3414 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3415                  size_t *n_ops)
3416 {
3417     struct facet *facet;
3418
3419     miss->ofproto->n_missed += list_size(&miss->packets);
3420
3421     facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3422     if (!facet) {
3423         /* There does not exist a bijection between 'struct flow' and datapath
3424          * flow keys with fitness ODP_FIT_TO_LITTLE.  This breaks a fundamental
3425          * assumption used throughout the facet and subfacet handling code.
3426          * Since we have to handle these misses in userspace anyway, we simply
3427          * skip facet creation, avoiding the problem altogether. */
3428         if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3429             || !flow_miss_should_make_facet(miss)) {
3430             return;
3431         }
3432
3433         facet = facet_create(miss);
3434     }
3435     handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3436 }
3437
3438 static struct drop_key *
3439 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3440                 size_t key_len)
3441 {
3442     struct drop_key *drop_key;
3443
3444     HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3445                              &backer->drop_keys) {
3446         if (drop_key->key_len == key_len
3447             && !memcmp(drop_key->key, key, key_len)) {
3448             return drop_key;
3449         }
3450     }
3451     return NULL;
3452 }
3453
3454 static void
3455 drop_key_clear(struct dpif_backer *backer)
3456 {
3457     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3458     struct drop_key *drop_key, *next;
3459
3460     HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3461         int error;
3462
3463         error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3464                               NULL);
3465         if (error && !VLOG_DROP_WARN(&rl)) {
3466             struct ds ds = DS_EMPTY_INITIALIZER;
3467             odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3468             VLOG_WARN("Failed to delete drop key (%s) (%s)",
3469                       ovs_strerror(error), ds_cstr(&ds));
3470             ds_destroy(&ds);
3471         }
3472
3473         hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3474         drop_key_destroy(drop_key);
3475     }
3476
3477     udpif_drop_key_clear(backer->udpif);
3478 }
3479
3480 static void
3481 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3482 {
3483     struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3484     struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3485     struct flow_miss *miss;
3486     size_t n_ops, i;
3487
3488     /* Process each element in the to-do list, constructing the set of
3489      * operations to batch. */
3490     n_ops = 0;
3491     HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3492         handle_flow_miss(miss, flow_miss_ops, &n_ops);
3493     }
3494     ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3495
3496     /* Execute batch. */
3497     for (i = 0; i < n_ops; i++) {
3498         dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3499     }
3500     dpif_operate(backer->dpif, dpif_ops, n_ops);
3501
3502     for (i = 0; i < n_ops; i++) {
3503         if (dpif_ops[i]->error != 0
3504             && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3505             && flow_miss_ops[i].subfacet) {
3506             struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3507
3508             COVERAGE_INC(subfacet_install_fail);
3509
3510             /* Zero-out subfacet counters when installation failed, but
3511              * datapath reported hits.  This should not happen and
3512              * indicates a bug, since if the datapath flow exists, we
3513              * should not be attempting to create a new subfacet.  A
3514              * buggy datapath could trigger this, so just zero out the
3515              * counters and log an error. */
3516             if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3517                 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3518                             "datapath reported hits");
3519                 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3520             }
3521
3522             subfacet->path = SF_NOT_INSTALLED;
3523         }
3524     }
3525 }
3526
3527 static void
3528 handle_sflow_upcall(struct dpif_backer *backer,
3529                     const struct dpif_upcall *upcall)
3530 {
3531     struct ofproto_dpif *ofproto;
3532     union user_action_cookie cookie;
3533     struct flow flow;
3534     odp_port_t odp_in_port;
3535
3536     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3537                       &flow, NULL, &ofproto, &odp_in_port)
3538         || !ofproto->sflow) {
3539         return;
3540     }
3541
3542     memset(&cookie, 0, sizeof cookie);
3543     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3544     dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3545                         odp_in_port, &cookie);
3546 }
3547
3548 static void
3549 handle_flow_sample_upcall(struct dpif_backer *backer,
3550                           const struct dpif_upcall *upcall)
3551 {
3552     struct ofproto_dpif *ofproto;
3553     union user_action_cookie cookie;
3554     struct flow flow;
3555
3556     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3557                       &flow, NULL, &ofproto, NULL)
3558         || !ofproto->ipfix) {
3559         return;
3560     }
3561
3562     memset(&cookie, 0, sizeof cookie);
3563     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3564
3565     /* The flow reflects exactly the contents of the packet.  Sample
3566      * the packet using it. */
3567     dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3568                            cookie.flow_sample.collector_set_id,
3569                            cookie.flow_sample.probability,
3570                            cookie.flow_sample.obs_domain_id,
3571                            cookie.flow_sample.obs_point_id);
3572 }
3573
3574 static void
3575 handle_ipfix_upcall(struct dpif_backer *backer,
3576                     const struct dpif_upcall *upcall)
3577 {
3578     struct ofproto_dpif *ofproto;
3579     struct flow flow;
3580
3581     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3582                       &flow, NULL, &ofproto, NULL)
3583         || !ofproto->ipfix) {
3584         return;
3585     }
3586
3587     /* The flow reflects exactly the contents of the packet.  Sample
3588      * the packet using it. */
3589     dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3590 }
3591
3592 static void
3593 handle_upcalls(struct dpif_backer *backer)
3594 {
3595     struct flow_miss_batch *fmb;
3596     int n_processed;
3597
3598     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3599         struct upcall *upcall = upcall_next(backer->udpif);
3600
3601         if (!upcall) {
3602             break;
3603         }
3604
3605         switch (upcall->type) {
3606         case SFLOW_UPCALL:
3607             handle_sflow_upcall(backer, &upcall->dpif_upcall);
3608             break;
3609
3610         case FLOW_SAMPLE_UPCALL:
3611             handle_flow_sample_upcall(backer, &upcall->dpif_upcall);
3612             break;
3613
3614         case IPFIX_UPCALL:
3615             handle_ipfix_upcall(backer, &upcall->dpif_upcall);
3616             break;
3617
3618         case BAD_UPCALL:
3619             break;
3620
3621         case MISS_UPCALL:
3622             NOT_REACHED();
3623         }
3624
3625         upcall_destroy(upcall);
3626     }
3627
3628     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3629         struct drop_key *drop_key = drop_key_next(backer->udpif);
3630         if (!drop_key) {
3631             break;
3632         }
3633
3634         if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3635             hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3636                         hash_bytes(drop_key->key, drop_key->key_len, 0));
3637             dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3638                           drop_key->key, drop_key->key_len,
3639                           NULL, 0, NULL, 0, NULL);
3640         } else {
3641             drop_key_destroy(drop_key);
3642         }
3643     }
3644
3645     fmb = flow_miss_batch_next(backer->udpif);
3646     if (fmb) {
3647         handle_flow_misses(backer, fmb);
3648         flow_miss_batch_destroy(fmb);
3649     }
3650 }
3651 \f
3652 /* Flow expiration. */
3653
3654 static int subfacet_max_idle(const struct dpif_backer *);
3655 static void update_stats(struct dpif_backer *);
3656 static void rule_expire(struct rule_dpif *);
3657 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3658
3659 /* This function is called periodically by run().  Its job is to collect
3660  * updates for the flows that have been installed into the datapath, most
3661  * importantly when they last were used, and then use that information to
3662  * expire flows that have not been used recently.
3663  *
3664  * Returns the number of milliseconds after which it should be called again. */
3665 static int
3666 expire(struct dpif_backer *backer)
3667 {
3668     struct ofproto_dpif *ofproto;
3669     size_t n_subfacets;
3670     int max_idle;
3671
3672     /* Periodically clear out the drop keys in an effort to keep them
3673      * relatively few. */
3674     drop_key_clear(backer);
3675
3676     /* Update stats for each flow in the backer. */
3677     update_stats(backer);
3678
3679     n_subfacets = hmap_count(&backer->subfacets);
3680     if (n_subfacets) {
3681         struct subfacet *subfacet;
3682         long long int total, now;
3683
3684         total = 0;
3685         now = time_msec();
3686         HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3687             total += now - subfacet->created;
3688         }
3689         backer->avg_subfacet_life += total / n_subfacets;
3690     }
3691     backer->avg_subfacet_life /= 2;
3692
3693     backer->avg_n_subfacet += n_subfacets;
3694     backer->avg_n_subfacet /= 2;
3695
3696     backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3697
3698     max_idle = subfacet_max_idle(backer);
3699     expire_subfacets(backer, max_idle);
3700
3701     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3702         struct rule *rule, *next_rule;
3703
3704         if (ofproto->backer != backer) {
3705             continue;
3706         }
3707
3708         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3709          * has passed. */
3710         ovs_mutex_lock(&ofproto->up.expirable_mutex);
3711         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3712                             &ofproto->up.expirable) {
3713             rule_expire(rule_dpif_cast(rule));
3714         }
3715         ovs_mutex_unlock(&ofproto->up.expirable_mutex);
3716
3717         /* All outstanding data in existing flows has been accounted, so it's a
3718          * good time to do bond rebalancing. */
3719         if (ofproto->has_bonded_bundles) {
3720             struct ofbundle *bundle;
3721
3722             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3723                 if (bundle->bond) {
3724                     bond_rebalance(bundle->bond);
3725                 }
3726             }
3727         }
3728     }
3729
3730     return MIN(max_idle, 1000);
3731 }
3732
3733 /* Updates flow table statistics given that the datapath just reported 'stats'
3734  * as 'subfacet''s statistics. */
3735 static void
3736 update_subfacet_stats(struct subfacet *subfacet,
3737                       const struct dpif_flow_stats *stats)
3738 {
3739     struct facet *facet = subfacet->facet;
3740     struct dpif_flow_stats diff;
3741
3742     diff.tcp_flags = stats->tcp_flags;
3743     diff.used = stats->used;
3744
3745     if (stats->n_packets >= subfacet->dp_packet_count) {
3746         diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3747     } else {
3748         VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3749         diff.n_packets = 0;
3750     }
3751
3752     if (stats->n_bytes >= subfacet->dp_byte_count) {
3753         diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3754     } else {
3755         VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3756         diff.n_bytes = 0;
3757     }
3758
3759     facet->ofproto->n_hit += diff.n_packets;
3760     subfacet->dp_packet_count = stats->n_packets;
3761     subfacet->dp_byte_count = stats->n_bytes;
3762     subfacet_update_stats(subfacet, &diff);
3763
3764     if (facet->accounted_bytes < facet->byte_count) {
3765         facet_learn(facet);
3766         facet_account(facet);
3767         facet->accounted_bytes = facet->byte_count;
3768     }
3769 }
3770
3771 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3772  * about, or a flow that shouldn't be installed but was anyway.  Delete it. */
3773 static void
3774 delete_unexpected_flow(struct dpif_backer *backer,
3775                        const struct nlattr *key, size_t key_len)
3776 {
3777     if (!VLOG_DROP_WARN(&rl)) {
3778         struct ds s;
3779
3780         ds_init(&s);
3781         odp_flow_key_format(key, key_len, &s);
3782         VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3783         ds_destroy(&s);
3784     }
3785
3786     COVERAGE_INC(facet_unexpected);
3787     dpif_flow_del(backer->dpif, key, key_len, NULL);
3788 }
3789
3790 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3791  *
3792  * This function also pushes statistics updates to rules which each facet
3793  * resubmits into.  Generally these statistics will be accurate.  However, if a
3794  * facet changes the rule it resubmits into at some time in between
3795  * update_stats() runs, it is possible that statistics accrued to the
3796  * old rule will be incorrectly attributed to the new rule.  This could be
3797  * avoided by calling update_stats() whenever rules are created or
3798  * deleted.  However, the performance impact of making so many calls to the
3799  * datapath do not justify the benefit of having perfectly accurate statistics.
3800  *
3801  * In addition, this function maintains per ofproto flow hit counts. The patch
3802  * port is not treated specially. e.g. A packet ingress from br0 patched into
3803  * br1 will increase the hit count of br0 by 1, however, does not affect
3804  * the hit or miss counts of br1.
3805  */
3806 static void
3807 update_stats(struct dpif_backer *backer)
3808 {
3809     const struct dpif_flow_stats *stats;
3810     struct dpif_flow_dump dump;
3811     const struct nlattr *key, *mask;
3812     size_t key_len, mask_len;
3813
3814     dpif_flow_dump_start(&dump, backer->dpif);
3815     while (dpif_flow_dump_next(&dump, &key, &key_len,
3816                                &mask, &mask_len, NULL, NULL, &stats)) {
3817         struct subfacet *subfacet;
3818         uint32_t key_hash;
3819
3820         key_hash = odp_flow_key_hash(key, key_len);
3821         subfacet = subfacet_find(backer, key, key_len, key_hash);
3822         switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3823         case SF_FAST_PATH:
3824             update_subfacet_stats(subfacet, stats);
3825             break;
3826
3827         case SF_SLOW_PATH:
3828             /* Stats are updated per-packet. */
3829             break;
3830
3831         case SF_NOT_INSTALLED:
3832         default:
3833             delete_unexpected_flow(backer, key, key_len);
3834             break;
3835         }
3836         run_fast_rl();
3837     }
3838     dpif_flow_dump_done(&dump);
3839
3840     update_moving_averages(backer);
3841 }
3842
3843 /* Calculates and returns the number of milliseconds of idle time after which
3844  * subfacets should expire from the datapath.  When a subfacet expires, we fold
3845  * its statistics into its facet, and when a facet's last subfacet expires, we
3846  * fold its statistic into its rule. */
3847 static int
3848 subfacet_max_idle(const struct dpif_backer *backer)
3849 {
3850     /*
3851      * Idle time histogram.
3852      *
3853      * Most of the time a switch has a relatively small number of subfacets.
3854      * When this is the case we might as well keep statistics for all of them
3855      * in userspace and to cache them in the kernel datapath for performance as
3856      * well.
3857      *
3858      * As the number of subfacets increases, the memory required to maintain
3859      * statistics about them in userspace and in the kernel becomes
3860      * significant.  However, with a large number of subfacets it is likely
3861      * that only a few of them are "heavy hitters" that consume a large amount
3862      * of bandwidth.  At this point, only heavy hitters are worth caching in
3863      * the kernel and maintaining in userspaces; other subfacets we can
3864      * discard.
3865      *
3866      * The technique used to compute the idle time is to build a histogram with
3867      * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each.  Each subfacet
3868      * that is installed in the kernel gets dropped in the appropriate bucket.
3869      * After the histogram has been built, we compute the cutoff so that only
3870      * the most-recently-used 1% of subfacets (but at least
3871      * flow_eviction_threshold flows) are kept cached.  At least
3872      * the most-recently-used bucket of subfacets is kept, so actually an
3873      * arbitrary number of subfacets can be kept in any given expiration run
3874      * (though the next run will delete most of those unless they receive
3875      * additional data).
3876      *
3877      * This requires a second pass through the subfacets, in addition to the
3878      * pass made by update_stats(), because the former function never looks at
3879      * uninstallable subfacets.
3880      */
3881     enum { BUCKET_WIDTH = 100 };
3882     enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3883     int buckets[N_BUCKETS] = { 0 };
3884     int total, subtotal, bucket;
3885     struct subfacet *subfacet;
3886     long long int now;
3887     int i;
3888
3889     total = hmap_count(&backer->subfacets);
3890     if (total <= flow_eviction_threshold) {
3891         return N_BUCKETS * BUCKET_WIDTH;
3892     }
3893
3894     /* Build histogram. */
3895     now = time_msec();
3896     HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3897         long long int idle = now - subfacet->used;
3898         int bucket = (idle <= 0 ? 0
3899                       : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3900                       : (unsigned int) idle / BUCKET_WIDTH);
3901         buckets[bucket]++;
3902     }
3903
3904     /* Find the first bucket whose flows should be expired. */
3905     subtotal = bucket = 0;
3906     do {
3907         subtotal += buckets[bucket++];
3908     } while (bucket < N_BUCKETS &&
3909              subtotal < MAX(flow_eviction_threshold, total / 100));
3910
3911     if (VLOG_IS_DBG_ENABLED()) {
3912         struct ds s;
3913
3914         ds_init(&s);
3915         ds_put_cstr(&s, "keep");
3916         for (i = 0; i < N_BUCKETS; i++) {
3917             if (i == bucket) {
3918                 ds_put_cstr(&s, ", drop");
3919             }
3920             if (buckets[i]) {
3921                 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3922             }
3923         }
3924         VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3925         ds_destroy(&s);
3926     }
3927
3928     return bucket * BUCKET_WIDTH;
3929 }
3930
3931 static void
3932 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3933 {
3934     /* Cutoff time for most flows. */
3935     long long int normal_cutoff = time_msec() - dp_max_idle;
3936
3937     /* We really want to keep flows for special protocols around, so use a more
3938      * conservative cutoff. */
3939     long long int special_cutoff = time_msec() - 10000;
3940
3941     struct subfacet *subfacet, *next_subfacet;
3942     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3943     int n_batch;
3944
3945     n_batch = 0;
3946     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3947                         &backer->subfacets) {
3948         long long int cutoff;
3949
3950         cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3951                                                 | SLOW_STP)
3952                   ? special_cutoff
3953                   : normal_cutoff);
3954         if (subfacet->used < cutoff) {
3955             if (subfacet->path != SF_NOT_INSTALLED) {
3956                 batch[n_batch++] = subfacet;
3957                 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3958                     subfacet_destroy_batch(backer, batch, n_batch);
3959                     n_batch = 0;
3960                 }
3961             } else {
3962                 subfacet_destroy(subfacet);
3963             }
3964         }
3965     }
3966
3967     if (n_batch > 0) {
3968         subfacet_destroy_batch(backer, batch, n_batch);
3969     }
3970 }
3971
3972 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3973  * then delete it entirely. */
3974 static void
3975 rule_expire(struct rule_dpif *rule)
3976 {
3977     uint16_t idle_timeout, hard_timeout;
3978     long long int now;
3979     uint8_t reason;
3980
3981     if (rule->up.pending) {
3982         /* We'll have to expire it later. */
3983         return;
3984     }
3985
3986     ovs_mutex_lock(&rule->up.timeout_mutex);
3987     hard_timeout = rule->up.hard_timeout;
3988     idle_timeout = rule->up.idle_timeout;
3989     ovs_mutex_unlock(&rule->up.timeout_mutex);
3990
3991     /* Has 'rule' expired? */
3992     now = time_msec();
3993     if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3994         reason = OFPRR_HARD_TIMEOUT;
3995     } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3996         reason = OFPRR_IDLE_TIMEOUT;
3997     } else {
3998         return;
3999     }
4000
4001     COVERAGE_INC(ofproto_dpif_expired);
4002     ofproto_rule_expire(&rule->up, reason);
4003 }
4004 \f
4005 /* Facets. */
4006
4007 /* Creates and returns a new facet based on 'miss'.
4008  *
4009  * The caller must already have determined that no facet with an identical
4010  * 'miss->flow' exists in 'miss->ofproto'.
4011  *
4012  * 'rule' and 'xout' must have been created based on 'miss'.
4013  *
4014  * 'facet'' statistics are initialized based on 'stats'.
4015  *
4016  * The facet will initially have no subfacets.  The caller should create (at
4017  * least) one subfacet with subfacet_create(). */
4018 static struct facet *
4019 facet_create(const struct flow_miss *miss)
4020 {
4021     struct ofproto_dpif *ofproto = miss->ofproto;
4022     struct facet *facet;
4023     struct match match;
4024
4025     facet = xzalloc(sizeof *facet);
4026     facet->ofproto = miss->ofproto;
4027     facet->used = miss->stats.used;
4028     facet->flow = miss->flow;
4029     facet->learn_rl = time_msec() + 500;
4030
4031     list_init(&facet->subfacets);
4032     netflow_flow_init(&facet->nf_flow);
4033     netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4034
4035     xlate_out_copy(&facet->xout, &miss->xout);
4036
4037     match_init(&match, &facet->flow, &facet->xout.wc);
4038     cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4039     ovs_rwlock_wrlock(&ofproto->facets.rwlock);
4040     classifier_insert(&ofproto->facets, &facet->cr);
4041     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4042
4043     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4044     return facet;
4045 }
4046
4047 static void
4048 facet_free(struct facet *facet)
4049 {
4050     if (facet) {
4051         xlate_out_uninit(&facet->xout);
4052         free(facet);
4053     }
4054 }
4055
4056 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4057  * 'packet', which arrived on 'in_port'. */
4058 static bool
4059 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4060                     const struct nlattr *odp_actions, size_t actions_len,
4061                     struct ofpbuf *packet)
4062 {
4063     struct odputil_keybuf keybuf;
4064     struct ofpbuf key;
4065     int error;
4066
4067     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4068     odp_flow_key_from_flow(&key, flow,
4069                            ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4070
4071     error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4072                          odp_actions, actions_len, packet);
4073     return !error;
4074 }
4075
4076 /* Remove 'facet' from its ofproto and free up the associated memory:
4077  *
4078  *   - If 'facet' was installed in the datapath, uninstalls it and updates its
4079  *     rule's statistics, via subfacet_uninstall().
4080  *
4081  *   - Removes 'facet' from its rule and from ofproto->facets.
4082  */
4083 static void
4084 facet_remove(struct facet *facet)
4085 {
4086     struct subfacet *subfacet, *next_subfacet;
4087
4088     ovs_assert(!list_is_empty(&facet->subfacets));
4089
4090     /* First uninstall all of the subfacets to get final statistics. */
4091     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4092         subfacet_uninstall(subfacet);
4093     }
4094
4095     /* Flush the final stats to the rule.
4096      *
4097      * This might require us to have at least one subfacet around so that we
4098      * can use its actions for accounting in facet_account(), which is why we
4099      * have uninstalled but not yet destroyed the subfacets. */
4100     facet_flush_stats(facet);
4101
4102     /* Now we're really all done so destroy everything. */
4103     LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4104                         &facet->subfacets) {
4105         subfacet_destroy__(subfacet);
4106     }
4107     ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4108     classifier_remove(&facet->ofproto->facets, &facet->cr);
4109     ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4110     cls_rule_destroy(&facet->cr);
4111     facet_free(facet);
4112 }
4113
4114 /* Feed information from 'facet' back into the learning table to keep it in
4115  * sync with what is actually flowing through the datapath. */
4116 static void
4117 facet_learn(struct facet *facet)
4118 {
4119     long long int now = time_msec();
4120
4121     if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4122         return;
4123     }
4124
4125     facet->learn_rl = now + 500;
4126
4127     if (!facet->xout.has_learn
4128         && !facet->xout.has_normal
4129         && (!facet->xout.has_fin_timeout
4130             || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4131         return;
4132     }
4133
4134     facet_push_stats(facet, true);
4135 }
4136
4137 static void
4138 facet_account(struct facet *facet)
4139 {
4140     const struct nlattr *a;
4141     unsigned int left;
4142     ovs_be16 vlan_tci;
4143     uint64_t n_bytes;
4144
4145     if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4146         return;
4147     }
4148     n_bytes = facet->byte_count - facet->accounted_bytes;
4149
4150     /* This loop feeds byte counters to bond_account() for rebalancing to use
4151      * as a basis.  We also need to track the actual VLAN on which the packet
4152      * is going to be sent to ensure that it matches the one passed to
4153      * bond_choose_output_slave().  (Otherwise, we will account to the wrong
4154      * hash bucket.)
4155      *
4156      * We use the actions from an arbitrary subfacet because they should all
4157      * be equally valid for our purpose. */
4158     vlan_tci = facet->flow.vlan_tci;
4159     NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4160                              facet->xout.odp_actions.size) {
4161         const struct ovs_action_push_vlan *vlan;
4162         struct ofport_dpif *port;
4163
4164         switch (nl_attr_type(a)) {
4165         case OVS_ACTION_ATTR_OUTPUT:
4166             port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4167             if (port && port->bundle && port->bundle->bond) {
4168                 bond_account(port->bundle->bond, &facet->flow,
4169                              vlan_tci_to_vid(vlan_tci), n_bytes);
4170             }
4171             break;
4172
4173         case OVS_ACTION_ATTR_POP_VLAN:
4174             vlan_tci = htons(0);
4175             break;
4176
4177         case OVS_ACTION_ATTR_PUSH_VLAN:
4178             vlan = nl_attr_get(a);
4179             vlan_tci = vlan->vlan_tci;
4180             break;
4181         }
4182     }
4183 }
4184
4185 /* Returns true if the only action for 'facet' is to send to the controller.
4186  * (We don't report NetFlow expiration messages for such facets because they
4187  * are just part of the control logic for the network, not real traffic). */
4188 static bool
4189 facet_is_controller_flow(struct facet *facet)
4190 {
4191     if (facet) {
4192         struct ofproto_dpif *ofproto = facet->ofproto;
4193         const struct ofpact *ofpacts;
4194         struct rule_dpif *rule;
4195         size_t ofpacts_len;
4196         bool is_controller;
4197
4198         rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4199         ofpacts_len = rule->up.ofpacts_len;
4200         ofpacts = rule->up.ofpacts;
4201         is_controller = ofpacts_len > 0
4202             && ofpacts->type == OFPACT_CONTROLLER
4203             && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4204         rule_dpif_release(rule);
4205         return is_controller;
4206     }
4207     return false;
4208 }
4209
4210 /* Folds all of 'facet''s statistics into its rule.  Also updates the
4211  * accounting ofhook and emits a NetFlow expiration if appropriate.  All of
4212  * 'facet''s statistics in the datapath should have been zeroed and folded into
4213  * its packet and byte counts before this function is called. */
4214 static void
4215 facet_flush_stats(struct facet *facet)
4216 {
4217     struct ofproto_dpif *ofproto = facet->ofproto;
4218     struct subfacet *subfacet;
4219
4220     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4221         ovs_assert(!subfacet->dp_byte_count);
4222         ovs_assert(!subfacet->dp_packet_count);
4223     }
4224
4225     facet_push_stats(facet, false);
4226     if (facet->accounted_bytes < facet->byte_count) {
4227         facet_account(facet);
4228         facet->accounted_bytes = facet->byte_count;
4229     }
4230
4231     if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4232         struct ofexpired expired;
4233         expired.flow = facet->flow;
4234         expired.packet_count = facet->packet_count;
4235         expired.byte_count = facet->byte_count;
4236         expired.used = facet->used;
4237         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4238     }
4239
4240     /* Reset counters to prevent double counting if 'facet' ever gets
4241      * reinstalled. */
4242     facet_reset_counters(facet);
4243
4244     netflow_flow_clear(&facet->nf_flow);
4245     facet->tcp_flags = 0;
4246 }
4247
4248 /* Searches 'ofproto''s table of facets for one which would be responsible for
4249  * 'flow'.  Returns it if found, otherwise a null pointer.
4250  *
4251  * The returned facet might need revalidation; use facet_lookup_valid()
4252  * instead if that is important. */
4253 static struct facet *
4254 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4255 {
4256     struct cls_rule *cr;
4257
4258     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4259     cr = classifier_lookup(&ofproto->facets, flow, NULL);
4260     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4261     return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4262 }
4263
4264 /* Searches 'ofproto''s table of facets for one capable that covers
4265  * 'flow'.  Returns it if found, otherwise a null pointer.
4266  *
4267  * The returned facet is guaranteed to be valid. */
4268 static struct facet *
4269 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4270 {
4271     struct facet *facet;
4272
4273     facet = facet_find(ofproto, flow);
4274     if (facet
4275         && ofproto->backer->need_revalidate
4276         && !facet_revalidate(facet)) {
4277         return NULL;
4278     }
4279
4280     return facet;
4281 }
4282
4283 static bool
4284 facet_check_consistency(struct facet *facet)
4285 {
4286     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4287
4288     struct xlate_out xout;
4289     struct xlate_in xin;
4290
4291     struct rule_dpif *rule;
4292     bool ok;
4293
4294     /* Check the datapath actions for consistency. */
4295     rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4296     xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4297     xlate_actions(&xin, &xout);
4298     rule_dpif_release(rule);
4299
4300     ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4301         && facet->xout.slow == xout.slow;
4302     if (!ok && !VLOG_DROP_WARN(&rl)) {
4303         struct ds s = DS_EMPTY_INITIALIZER;
4304
4305         flow_format(&s, &facet->flow);
4306         ds_put_cstr(&s, ": inconsistency in facet");
4307
4308         if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4309             ds_put_cstr(&s, " (actions were: ");
4310             format_odp_actions(&s, facet->xout.odp_actions.data,
4311                                facet->xout.odp_actions.size);
4312             ds_put_cstr(&s, ") (correct actions: ");
4313             format_odp_actions(&s, xout.odp_actions.data,
4314                                xout.odp_actions.size);
4315             ds_put_char(&s, ')');
4316         }
4317
4318         if (facet->xout.slow != xout.slow) {
4319             ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4320         }
4321
4322         ds_destroy(&s);
4323     }
4324     xlate_out_uninit(&xout);
4325
4326     return ok;
4327 }
4328
4329 /* Re-searches the classifier for 'facet':
4330  *
4331  *   - If the rule found is different from 'facet''s current rule, moves
4332  *     'facet' to the new rule and recompiles its actions.
4333  *
4334  *   - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4335  *     where it is and recompiles its actions anyway.
4336  *
4337  *   - If any of 'facet''s subfacets correspond to a new flow according to
4338  *     xlate_receive(), 'facet' is removed.
4339  *
4340  *   Returns true if 'facet' is still valid.  False if 'facet' was removed. */
4341 static bool
4342 facet_revalidate(struct facet *facet)
4343 {
4344     struct ofproto_dpif *ofproto = facet->ofproto;
4345     struct rule_dpif *new_rule;
4346     struct subfacet *subfacet;
4347     struct flow_wildcards wc;
4348     struct xlate_out xout;
4349     struct xlate_in xin;
4350
4351     COVERAGE_INC(facet_revalidate);
4352
4353     /* Check that child subfacets still correspond to this facet.  Tunnel
4354      * configuration changes could cause a subfacet's OpenFlow in_port to
4355      * change. */
4356     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4357         struct ofproto_dpif *recv_ofproto;
4358         struct flow recv_flow;
4359         int error;
4360
4361         error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4362                               subfacet->key_len, &recv_flow, NULL,
4363                               &recv_ofproto, NULL);
4364         if (error
4365             || recv_ofproto != ofproto
4366             || facet != facet_find(ofproto, &recv_flow)) {
4367             facet_remove(facet);
4368             return false;
4369         }
4370     }
4371
4372     flow_wildcards_init_catchall(&wc);
4373     rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4374
4375     /* Calculate new datapath actions.
4376      *
4377      * We do not modify any 'facet' state yet, because we might need to, e.g.,
4378      * emit a NetFlow expiration and, if so, we need to have the old state
4379      * around to properly compose it. */
4380     xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4381     xlate_actions(&xin, &xout);
4382     flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4383
4384     /* A facet's slow path reason should only change under dramatic
4385      * circumstances.  Rather than try to update everything, it's simpler to
4386      * remove the facet and start over.
4387      *
4388      * More importantly, if a facet's wildcards change, it will be relatively
4389      * difficult to figure out if its subfacets still belong to it, and if not
4390      * which facet they may belong to.  Again, to avoid the complexity, we
4391      * simply give up instead. */
4392     if (facet->xout.slow != xout.slow
4393         || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4394         facet_remove(facet);
4395         xlate_out_uninit(&xout);
4396         rule_dpif_release(new_rule);
4397         return false;
4398     }
4399
4400     if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4401         LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4402             if (subfacet->path == SF_FAST_PATH) {
4403                 struct dpif_flow_stats stats;
4404
4405                 subfacet_install(subfacet, &xout.odp_actions, &stats);
4406                 subfacet_update_stats(subfacet, &stats);
4407             }
4408         }
4409
4410         facet_flush_stats(facet);
4411
4412         ofpbuf_clear(&facet->xout.odp_actions);
4413         ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4414                    xout.odp_actions.size);
4415     }
4416
4417     /* Update 'facet' now that we've taken care of all the old state. */
4418     facet->xout.slow = xout.slow;
4419     facet->xout.has_learn = xout.has_learn;
4420     facet->xout.has_normal = xout.has_normal;
4421     facet->xout.has_fin_timeout = xout.has_fin_timeout;
4422     facet->xout.nf_output_iface = xout.nf_output_iface;
4423     facet->xout.mirrors = xout.mirrors;
4424     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4425     facet->used = MAX(facet->used, new_rule->up.created);
4426
4427     xlate_out_uninit(&xout);
4428     rule_dpif_release(new_rule);
4429     return true;
4430 }
4431
4432 static void
4433 facet_reset_counters(struct facet *facet)
4434 {
4435     facet->packet_count = 0;
4436     facet->byte_count = 0;
4437     facet->prev_packet_count = 0;
4438     facet->prev_byte_count = 0;
4439     facet->accounted_bytes = 0;
4440 }
4441
4442 static void
4443 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4444                 struct dpif_flow_stats *stats, bool may_learn)
4445 {
4446     struct ofport_dpif *in_port;
4447     struct rule_dpif *rule;
4448     struct xlate_in xin;
4449
4450     in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4451     if (in_port && in_port->is_tunnel) {
4452         netdev_vport_inc_rx(in_port->up.netdev, stats);
4453     }
4454
4455     rule_dpif_lookup(ofproto, flow, NULL, &rule);
4456     rule_dpif_credit_stats(rule, stats);
4457     xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4458     xin.resubmit_stats = stats;
4459     xin.may_learn = may_learn;
4460     xlate_actions_for_side_effects(&xin);
4461     rule_dpif_release(rule);
4462 }
4463
4464 static void
4465 facet_push_stats(struct facet *facet, bool may_learn)
4466 {
4467     struct dpif_flow_stats stats;
4468
4469     ovs_assert(facet->packet_count >= facet->prev_packet_count);
4470     ovs_assert(facet->byte_count >= facet->prev_byte_count);
4471     ovs_assert(facet->used >= facet->prev_used);
4472
4473     stats.n_packets = facet->packet_count - facet->prev_packet_count;
4474     stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4475     stats.used = facet->used;
4476     stats.tcp_flags = facet->tcp_flags;
4477
4478     if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4479         facet->prev_packet_count = facet->packet_count;
4480         facet->prev_byte_count = facet->byte_count;
4481         facet->prev_used = facet->used;
4482
4483         netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4484                                  facet->used);
4485         netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4486         mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4487                             stats.n_packets, stats.n_bytes);
4488         flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4489     }
4490 }
4491
4492 static void
4493 push_all_stats__(bool run_fast)
4494 {
4495     static long long int rl = LLONG_MIN;
4496     struct ofproto_dpif *ofproto;
4497
4498     if (time_msec() < rl) {
4499         return;
4500     }
4501
4502     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4503         struct cls_cursor cursor;
4504         struct facet *facet;
4505
4506         ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4507         cls_cursor_init(&cursor, &ofproto->facets, NULL);
4508         CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4509             facet_push_stats(facet, false);
4510             if (run_fast) {
4511                 run_fast_rl();
4512             }
4513         }
4514         ovs_rwlock_unlock(&ofproto->facets.rwlock);
4515     }
4516
4517     rl = time_msec() + 100;
4518 }
4519
4520 static void
4521 push_all_stats(void)
4522 {
4523     push_all_stats__(true);
4524 }
4525
4526 void
4527 rule_dpif_credit_stats(struct rule_dpif *rule,
4528                        const struct dpif_flow_stats *stats)
4529 {
4530     ovs_mutex_lock(&rule->stats_mutex);
4531     rule->packet_count += stats->n_packets;
4532     rule->byte_count += stats->n_bytes;
4533     rule->up.used = MAX(rule->up.used, stats->used);
4534     ovs_mutex_unlock(&rule->stats_mutex);
4535 }
4536
4537 bool
4538 rule_dpif_fail_open(const struct rule_dpif *rule)
4539 {
4540     return rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4541 }
4542
4543 ovs_be64
4544 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4545 {
4546     return rule->up.flow_cookie;
4547 }
4548
4549 void
4550 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4551                      uint16_t hard_timeout)
4552 {
4553     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4554 }
4555
4556 void
4557 rule_dpif_get_actions(const struct rule_dpif *rule,
4558                       const struct ofpact **ofpacts, size_t *ofpacts_len)
4559 {
4560     *ofpacts = rule->up.ofpacts;
4561     *ofpacts_len = rule->up.ofpacts_len;
4562 }
4563 \f
4564 /* Subfacets. */
4565
4566 static struct subfacet *
4567 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4568               size_t key_len, uint32_t key_hash)
4569 {
4570     struct subfacet *subfacet;
4571
4572     HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4573                              &backer->subfacets) {
4574         if (subfacet->key_len == key_len
4575             && !memcmp(key, subfacet->key, key_len)) {
4576             return subfacet;
4577         }
4578     }
4579
4580     return NULL;
4581 }
4582
4583 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4584  * 'key_fitness', 'key', and 'key_len' members in 'miss'.  Returns the
4585  * existing subfacet if there is one, otherwise creates and returns a
4586  * new subfacet. */
4587 static struct subfacet *
4588 subfacet_create(struct facet *facet, struct flow_miss *miss)
4589 {
4590     struct dpif_backer *backer = miss->ofproto->backer;
4591     enum odp_key_fitness key_fitness = miss->key_fitness;
4592     const struct nlattr *key = miss->key;
4593     size_t key_len = miss->key_len;
4594     uint32_t key_hash;
4595     struct subfacet *subfacet;
4596
4597     key_hash = odp_flow_key_hash(key, key_len);
4598
4599     if (list_is_empty(&facet->subfacets)) {
4600         subfacet = &facet->one_subfacet;
4601     } else {
4602         subfacet = subfacet_find(backer, key, key_len, key_hash);
4603         if (subfacet) {
4604             if (subfacet->facet == facet) {
4605                 return subfacet;
4606             }
4607
4608             /* This shouldn't happen. */
4609             VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4610             subfacet_destroy(subfacet);
4611         }
4612
4613         subfacet = xmalloc(sizeof *subfacet);
4614     }
4615
4616     hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4617     list_push_back(&facet->subfacets, &subfacet->list_node);
4618     subfacet->facet = facet;
4619     subfacet->key_fitness = key_fitness;
4620     subfacet->key = xmemdup(key, key_len);
4621     subfacet->key_len = key_len;
4622     subfacet->used = miss->stats.used;
4623     subfacet->created = subfacet->used;
4624     subfacet->dp_packet_count = 0;
4625     subfacet->dp_byte_count = 0;
4626     subfacet->path = SF_NOT_INSTALLED;
4627     subfacet->backer = backer;
4628
4629     backer->subfacet_add_count++;
4630     return subfacet;
4631 }
4632
4633 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4634  * its facet within 'ofproto', and frees it. */
4635 static void
4636 subfacet_destroy__(struct subfacet *subfacet)
4637 {
4638     struct facet *facet = subfacet->facet;
4639     struct ofproto_dpif *ofproto = facet->ofproto;
4640
4641     /* Update ofproto stats before uninstall the subfacet. */
4642     ofproto->backer->subfacet_del_count++;
4643
4644     subfacet_uninstall(subfacet);
4645     hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4646     list_remove(&subfacet->list_node);
4647     free(subfacet->key);
4648     if (subfacet != &facet->one_subfacet) {
4649         free(subfacet);
4650     }
4651 }
4652
4653 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4654  * last remaining subfacet in its facet destroys the facet too. */
4655 static void
4656 subfacet_destroy(struct subfacet *subfacet)
4657 {
4658     struct facet *facet = subfacet->facet;
4659
4660     if (list_is_singleton(&facet->subfacets)) {
4661         /* facet_remove() needs at least one subfacet (it will remove it). */
4662         facet_remove(facet);
4663     } else {
4664         subfacet_destroy__(subfacet);
4665     }
4666 }
4667
4668 static void
4669 subfacet_destroy_batch(struct dpif_backer *backer,
4670                        struct subfacet **subfacets, int n)
4671 {
4672     struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4673     struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4674     struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4675     int i;
4676
4677     for (i = 0; i < n; i++) {
4678         ops[i].type = DPIF_OP_FLOW_DEL;
4679         ops[i].u.flow_del.key = subfacets[i]->key;
4680         ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4681         ops[i].u.flow_del.stats = &stats[i];
4682         opsp[i] = &ops[i];
4683     }
4684
4685     dpif_operate(backer->dpif, opsp, n);
4686     for (i = 0; i < n; i++) {
4687         subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4688         subfacets[i]->path = SF_NOT_INSTALLED;
4689         subfacet_destroy(subfacets[i]);
4690         run_fast_rl();
4691     }
4692 }
4693
4694 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4695  * bytes of actions in 'actions'.  If 'stats' is non-null, statistics counters
4696  * in the datapath will be zeroed and 'stats' will be updated with traffic new
4697  * since 'subfacet' was last updated.
4698  *
4699  * Returns 0 if successful, otherwise a positive errno value. */
4700 static int
4701 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4702                  struct dpif_flow_stats *stats)
4703 {
4704     struct facet *facet = subfacet->facet;
4705     enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4706     const struct nlattr *actions = odp_actions->data;
4707     size_t actions_len = odp_actions->size;
4708     struct odputil_keybuf maskbuf;
4709     struct ofpbuf mask;
4710
4711     uint64_t slow_path_stub[128 / 8];
4712     enum dpif_flow_put_flags flags;
4713     int ret;
4714
4715     flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4716                                                : DPIF_FP_MODIFY;
4717     if (stats) {
4718         flags |= DPIF_FP_ZERO_STATS;
4719     }
4720
4721     if (path == SF_SLOW_PATH) {
4722         compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4723                           slow_path_stub, sizeof slow_path_stub,
4724                           &actions, &actions_len);
4725     }
4726
4727     ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4728     if (enable_megaflows) {
4729         odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4730                                &facet->flow, UINT32_MAX);
4731     }
4732
4733     ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4734                         subfacet->key_len,  mask.data, mask.size,
4735                         actions, actions_len, stats);
4736
4737     if (stats) {
4738         subfacet_reset_dp_stats(subfacet, stats);
4739     }
4740
4741     if (ret) {
4742         COVERAGE_INC(subfacet_install_fail);
4743     } else {
4744         subfacet->path = path;
4745     }
4746     return ret;
4747 }
4748
4749 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4750 static void
4751 subfacet_uninstall(struct subfacet *subfacet)
4752 {
4753     if (subfacet->path != SF_NOT_INSTALLED) {
4754         struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4755         struct dpif_flow_stats stats;
4756         int error;
4757
4758         error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4759                               subfacet->key_len, &stats);
4760         subfacet_reset_dp_stats(subfacet, &stats);
4761         if (!error) {
4762             subfacet_update_stats(subfacet, &stats);
4763         }
4764         subfacet->path = SF_NOT_INSTALLED;
4765     } else {
4766         ovs_assert(subfacet->dp_packet_count == 0);
4767         ovs_assert(subfacet->dp_byte_count == 0);
4768     }
4769 }
4770
4771 /* Resets 'subfacet''s datapath statistics counters.  This should be called
4772  * when 'subfacet''s statistics are cleared in the datapath.  If 'stats' is
4773  * non-null, it should contain the statistics returned by dpif when 'subfacet'
4774  * was reset in the datapath.  'stats' will be modified to include only
4775  * statistics new since 'subfacet' was last updated. */
4776 static void
4777 subfacet_reset_dp_stats(struct subfacet *subfacet,
4778                         struct dpif_flow_stats *stats)
4779 {
4780     if (stats
4781         && subfacet->dp_packet_count <= stats->n_packets
4782         && subfacet->dp_byte_count <= stats->n_bytes) {
4783         stats->n_packets -= subfacet->dp_packet_count;
4784         stats->n_bytes -= subfacet->dp_byte_count;
4785     }
4786
4787     subfacet->dp_packet_count = 0;
4788     subfacet->dp_byte_count = 0;
4789 }
4790
4791 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4792  *
4793  * Because of the meaning of a subfacet's counters, it only makes sense to do
4794  * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4795  * represents a packet that was sent by hand or if it represents statistics
4796  * that have been cleared out of the datapath. */
4797 static void
4798 subfacet_update_stats(struct subfacet *subfacet,
4799                       const struct dpif_flow_stats *stats)
4800 {
4801     if (stats->n_packets || stats->used > subfacet->used) {
4802         struct facet *facet = subfacet->facet;
4803
4804         subfacet->used = MAX(subfacet->used, stats->used);
4805         facet->used = MAX(facet->used, stats->used);
4806         facet->packet_count += stats->n_packets;
4807         facet->byte_count += stats->n_bytes;
4808         facet->tcp_flags |= stats->tcp_flags;
4809     }
4810 }
4811 \f
4812 /* Rules. */
4813
4814 /* Lookup 'flow' in 'ofproto''s classifier.  If 'wc' is non-null, sets
4815  * the fields that were relevant as part of the lookup. */
4816 void
4817 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4818                  struct flow_wildcards *wc, struct rule_dpif **rule)
4819 {
4820     struct ofport_dpif *port;
4821
4822     if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4823         return;
4824     }
4825     port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4826     if (!port) {
4827         VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4828                      flow->in_port.ofp_port);
4829     }
4830
4831     choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4832                      ofproto->no_packet_in_rule, rule);
4833 }
4834
4835 bool
4836 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4837                           const struct flow *flow, struct flow_wildcards *wc,
4838                           uint8_t table_id, struct rule_dpif **rule)
4839     OVS_TRY_RDLOCK(true, (*rule)->up.rwlock)
4840 {
4841     struct cls_rule *cls_rule;
4842     struct classifier *cls;
4843     bool frag;
4844
4845     *rule = NULL;
4846     if (table_id >= N_TABLES) {
4847         return false;
4848     }
4849
4850     if (wc) {
4851         memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4852         wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4853     }
4854
4855     cls = &ofproto->up.tables[table_id].cls;
4856     ovs_rwlock_rdlock(&cls->rwlock);
4857     frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4858     if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4859         /* We must pretend that transport ports are unavailable. */
4860         struct flow ofpc_normal_flow = *flow;
4861         ofpc_normal_flow.tp_src = htons(0);
4862         ofpc_normal_flow.tp_dst = htons(0);
4863         cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4864     } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4865         cls_rule = &ofproto->drop_frags_rule->up.cr;
4866         if (wc) {
4867             flow_wildcards_init_exact(wc);
4868         }
4869     } else {
4870         cls_rule = classifier_lookup(cls, flow, wc);
4871     }
4872
4873     *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4874     if (*rule && ovs_rwlock_tryrdlock(&(*rule)->up.rwlock)) {
4875         /* The rule is in the process of being removed.  Best we can do is
4876          * pretend it isn't there. */
4877         *rule = NULL;
4878     }
4879     ovs_rwlock_unlock(&cls->rwlock);
4880
4881     return *rule != NULL;
4882 }
4883
4884 /* Given a port configuration (specified as zero if there's no port), chooses
4885  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4886  * flow table miss. */
4887 void
4888 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4889                  struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4890     OVS_NO_THREAD_SAFETY_ANALYSIS
4891 {
4892     *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4893     ovs_rwlock_rdlock(&(*rule)->up.rwlock);
4894 }
4895
4896 void
4897 rule_dpif_release(struct rule_dpif *rule)
4898     OVS_NO_THREAD_SAFETY_ANALYSIS
4899 {
4900     if (rule) {
4901         ovs_rwlock_unlock(&rule->up.rwlock);
4902     }
4903 }
4904
4905 static void
4906 complete_operation(struct rule_dpif *rule)
4907 {
4908     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4909
4910     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4911     if (clogged) {
4912         struct dpif_completion *c = xmalloc(sizeof *c);
4913         c->op = rule->up.pending;
4914         list_push_back(&ofproto->completions, &c->list_node);
4915     } else {
4916         ofoperation_complete(rule->up.pending, 0);
4917     }
4918 }
4919
4920 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4921 {
4922     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4923 }
4924
4925 static struct rule *
4926 rule_alloc(void)
4927 {
4928     struct rule_dpif *rule = xmalloc(sizeof *rule);
4929     return &rule->up;
4930 }
4931
4932 static void
4933 rule_dealloc(struct rule *rule_)
4934 {
4935     struct rule_dpif *rule = rule_dpif_cast(rule_);
4936     free(rule);
4937 }
4938
4939 static enum ofperr
4940 rule_construct(struct rule *rule_)
4941 {
4942     struct rule_dpif *rule = rule_dpif_cast(rule_);
4943     ovs_mutex_init(&rule->stats_mutex);
4944     ovs_mutex_lock(&rule->stats_mutex);
4945     rule->packet_count = 0;
4946     rule->byte_count = 0;
4947     ovs_mutex_unlock(&rule->stats_mutex);
4948     return 0;
4949 }
4950
4951 static void
4952 rule_insert(struct rule *rule_)
4953 {
4954     struct rule_dpif *rule = rule_dpif_cast(rule_);
4955     complete_operation(rule);
4956 }
4957
4958 static void
4959 rule_delete(struct rule *rule_)
4960 {
4961     struct rule_dpif *rule = rule_dpif_cast(rule_);
4962     complete_operation(rule);
4963 }
4964
4965 static void
4966 rule_destruct(struct rule *rule_)
4967 {
4968     struct rule_dpif *rule = rule_dpif_cast(rule_);
4969     ovs_mutex_destroy(&rule->stats_mutex);
4970 }
4971
4972 static void
4973 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4974 {
4975     struct rule_dpif *rule = rule_dpif_cast(rule_);
4976
4977     /* push_all_stats() can handle flow misses which, when using the learn
4978      * action, can cause rules to be added and deleted.  This can corrupt our
4979      * caller's datastructures which assume that rule_get_stats() doesn't have
4980      * an impact on the flow table. To be safe, we disable miss handling. */
4981     push_all_stats__(false);
4982
4983     /* Start from historical data for 'rule' itself that are no longer tracked
4984      * in facets.  This counts, for example, facets that have expired. */
4985     ovs_mutex_lock(&rule->stats_mutex);
4986     *packets = rule->packet_count;
4987     *bytes = rule->byte_count;
4988     ovs_mutex_unlock(&rule->stats_mutex);
4989 }
4990
4991 static void
4992 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4993                   struct ofpbuf *packet)
4994 {
4995     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4996     struct dpif_flow_stats stats;
4997     struct xlate_out xout;
4998     struct xlate_in xin;
4999
5000     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5001     rule_dpif_credit_stats(rule, &stats);
5002
5003     xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5004     xin.resubmit_stats = &stats;
5005     xlate_actions(&xin, &xout);
5006
5007     execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5008                         xout.odp_actions.size, packet);
5009
5010     xlate_out_uninit(&xout);
5011 }
5012
5013 static enum ofperr
5014 rule_execute(struct rule *rule, const struct flow *flow,
5015              struct ofpbuf *packet)
5016 {
5017     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5018     ofpbuf_delete(packet);
5019     return 0;
5020 }
5021
5022 static void
5023 rule_modify_actions(struct rule *rule_, bool reset_counters)
5024 {
5025     struct rule_dpif *rule = rule_dpif_cast(rule_);
5026
5027     if (reset_counters) {
5028         ovs_mutex_lock(&rule->stats_mutex);
5029         rule->packet_count = 0;
5030         rule->byte_count = 0;
5031         ovs_mutex_unlock(&rule->stats_mutex);
5032     }
5033
5034     complete_operation(rule);
5035 }
5036 \f
5037 /* Sends 'packet' out 'ofport'.
5038  * May modify 'packet'.
5039  * Returns 0 if successful, otherwise a positive errno value. */
5040 static int
5041 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5042 {
5043     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5044     uint64_t odp_actions_stub[1024 / 8];
5045     struct ofpbuf key, odp_actions;
5046     struct dpif_flow_stats stats;
5047     struct odputil_keybuf keybuf;
5048     struct ofpact_output output;
5049     struct xlate_out xout;
5050     struct xlate_in xin;
5051     struct flow flow;
5052     union flow_in_port in_port_;
5053     int error;
5054
5055     ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5056     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5057
5058     /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5059     in_port_.ofp_port = OFPP_NONE;
5060     flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5061     odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5062                                                              OFPP_LOCAL));
5063     dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5064
5065     ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5066     output.port = ofport->up.ofp_port;
5067     output.max_len = 0;
5068
5069     xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5070     xin.ofpacts_len = sizeof output;
5071     xin.ofpacts = &output.ofpact;
5072     xin.resubmit_stats = &stats;
5073     xlate_actions(&xin, &xout);
5074
5075     error = dpif_execute(ofproto->backer->dpif,
5076                          key.data, key.size,
5077                          xout.odp_actions.data, xout.odp_actions.size,
5078                          packet);
5079     xlate_out_uninit(&xout);
5080
5081     if (error) {
5082         VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5083                      ofproto->up.name, netdev_get_name(ofport->up.netdev),
5084                      ovs_strerror(error));
5085     }
5086
5087     ofproto->stats.tx_packets++;
5088     ofproto->stats.tx_bytes += packet->size;
5089     return error;
5090 }
5091
5092 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5093  * The action will state 'slow' as the reason that the action is in the slow
5094  * path.  (This is purely informational: it allows a human viewing "ovs-dpctl
5095  * dump-flows" output to see why a flow is in the slow path.)
5096  *
5097  * The 'stub_size' bytes in 'stub' will be used to store the action.
5098  * 'stub_size' must be large enough for the action.
5099  *
5100  * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5101  * respectively. */
5102 static void
5103 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5104                   enum slow_path_reason slow,
5105                   uint64_t *stub, size_t stub_size,
5106                   const struct nlattr **actionsp, size_t *actions_lenp)
5107 {
5108     union user_action_cookie cookie;
5109     struct ofpbuf buf;
5110
5111     cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5112     cookie.slow_path.unused = 0;
5113     cookie.slow_path.reason = slow;
5114
5115     ofpbuf_use_stack(&buf, stub, stub_size);
5116     if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5117         uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5118                                          ODPP_NONE);
5119         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5120     } else {
5121         odp_port_t odp_port;
5122         uint32_t pid;
5123
5124         odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5125         pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5126         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5127     }
5128     *actionsp = buf.data;
5129     *actions_lenp = buf.size;
5130 }
5131 \f
5132 static bool
5133 set_frag_handling(struct ofproto *ofproto_,
5134                   enum ofp_config_flags frag_handling)
5135 {
5136     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5137     if (frag_handling != OFPC_FRAG_REASM) {
5138         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5139         return true;
5140     } else {
5141         return false;
5142     }
5143 }
5144
5145 static enum ofperr
5146 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5147            const struct flow *flow,
5148            const struct ofpact *ofpacts, size_t ofpacts_len)
5149 {
5150     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5151     struct odputil_keybuf keybuf;
5152     struct dpif_flow_stats stats;
5153     struct xlate_out xout;
5154     struct xlate_in xin;
5155     struct ofpbuf key;
5156
5157
5158     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5159     odp_flow_key_from_flow(&key, flow,
5160                            ofp_port_to_odp_port(ofproto,
5161                                       flow->in_port.ofp_port));
5162
5163     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5164
5165     xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5166     xin.resubmit_stats = &stats;
5167     xin.ofpacts_len = ofpacts_len;
5168     xin.ofpacts = ofpacts;
5169
5170     xlate_actions(&xin, &xout);
5171     dpif_execute(ofproto->backer->dpif, key.data, key.size,
5172                  xout.odp_actions.data, xout.odp_actions.size, packet);
5173     xlate_out_uninit(&xout);
5174
5175     return 0;
5176 }
5177 \f
5178 /* NetFlow. */
5179
5180 static int
5181 set_netflow(struct ofproto *ofproto_,
5182             const struct netflow_options *netflow_options)
5183 {
5184     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5185
5186     if (netflow_options) {
5187         if (!ofproto->netflow) {
5188             ofproto->netflow = netflow_create();
5189             ofproto->backer->need_revalidate = REV_RECONFIGURE;
5190         }
5191         return netflow_set_options(ofproto->netflow, netflow_options);
5192     } else if (ofproto->netflow) {
5193         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5194         netflow_destroy(ofproto->netflow);
5195         ofproto->netflow = NULL;
5196     }
5197
5198     return 0;
5199 }
5200
5201 static void
5202 get_netflow_ids(const struct ofproto *ofproto_,
5203                 uint8_t *engine_type, uint8_t *engine_id)
5204 {
5205     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5206
5207     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5208 }
5209
5210 static void
5211 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5212 {
5213     if (!facet_is_controller_flow(facet) &&
5214         netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5215         struct subfacet *subfacet;
5216         struct ofexpired expired;
5217
5218         LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5219             if (subfacet->path == SF_FAST_PATH) {
5220                 struct dpif_flow_stats stats;
5221
5222                 subfacet_install(subfacet, &facet->xout.odp_actions,
5223                                  &stats);
5224                 subfacet_update_stats(subfacet, &stats);
5225             }
5226         }
5227
5228         expired.flow = facet->flow;
5229         expired.packet_count = facet->packet_count;
5230         expired.byte_count = facet->byte_count;
5231         expired.used = facet->used;
5232         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5233     }
5234 }
5235
5236 static void
5237 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5238 {
5239     struct cls_cursor cursor;
5240     struct facet *facet;
5241
5242     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5243     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5244     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5245         send_active_timeout(ofproto, facet);
5246     }
5247     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5248 }
5249 \f
5250 static struct ofproto_dpif *
5251 ofproto_dpif_lookup(const char *name)
5252 {
5253     struct ofproto_dpif *ofproto;
5254
5255     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5256                              hash_string(name, 0), &all_ofproto_dpifs) {
5257         if (!strcmp(ofproto->up.name, name)) {
5258             return ofproto;
5259         }
5260     }
5261     return NULL;
5262 }
5263
5264 static void
5265 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5266                           const char *argv[], void *aux OVS_UNUSED)
5267 {
5268     struct ofproto_dpif *ofproto;
5269
5270     if (argc > 1) {
5271         ofproto = ofproto_dpif_lookup(argv[1]);
5272         if (!ofproto) {
5273             unixctl_command_reply_error(conn, "no such bridge");
5274             return;
5275         }
5276         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5277         mac_learning_flush(ofproto->ml);
5278         ovs_rwlock_unlock(&ofproto->ml->rwlock);
5279     } else {
5280         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5281             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5282             mac_learning_flush(ofproto->ml);
5283             ovs_rwlock_unlock(&ofproto->ml->rwlock);
5284         }
5285     }
5286
5287     unixctl_command_reply(conn, "table successfully flushed");
5288 }
5289
5290 static struct ofport_dpif *
5291 ofbundle_get_a_port(const struct ofbundle *bundle)
5292 {
5293     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5294                         bundle_node);
5295 }
5296
5297 static void
5298 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5299                          const char *argv[], void *aux OVS_UNUSED)
5300 {
5301     struct ds ds = DS_EMPTY_INITIALIZER;
5302     const struct ofproto_dpif *ofproto;
5303     const struct mac_entry *e;
5304
5305     ofproto = ofproto_dpif_lookup(argv[1]);
5306     if (!ofproto) {
5307         unixctl_command_reply_error(conn, "no such bridge");
5308         return;
5309     }
5310
5311     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
5312     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5313     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5314         struct ofbundle *bundle = e->port.p;
5315         char name[OFP_MAX_PORT_NAME_LEN];
5316
5317         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5318                                name, sizeof name);
5319         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
5320                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
5321                       mac_entry_age(ofproto->ml, e));
5322     }
5323     ovs_rwlock_unlock(&ofproto->ml->rwlock);
5324     unixctl_command_reply(conn, ds_cstr(&ds));
5325     ds_destroy(&ds);
5326 }
5327
5328 struct trace_ctx {
5329     struct xlate_out xout;
5330     struct xlate_in xin;
5331     struct flow flow;
5332     struct ds *result;
5333 };
5334
5335 static void
5336 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5337 {
5338     ds_put_char_multiple(result, '\t', level);
5339     if (!rule) {
5340         ds_put_cstr(result, "No match\n");
5341         return;
5342     }
5343
5344     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5345                   rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5346     cls_rule_format(&rule->up.cr, result);
5347     ds_put_char(result, '\n');
5348
5349     ds_put_char_multiple(result, '\t', level);
5350     ds_put_cstr(result, "OpenFlow ");
5351     ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5352     ds_put_char(result, '\n');
5353 }
5354
5355 static void
5356 trace_format_flow(struct ds *result, int level, const char *title,
5357                   struct trace_ctx *trace)
5358 {
5359     ds_put_char_multiple(result, '\t', level);
5360     ds_put_format(result, "%s: ", title);
5361     if (flow_equal(&trace->xin.flow, &trace->flow)) {
5362         ds_put_cstr(result, "unchanged");
5363     } else {
5364         flow_format(result, &trace->xin.flow);
5365         trace->flow = trace->xin.flow;
5366     }
5367     ds_put_char(result, '\n');
5368 }
5369
5370 static void
5371 trace_format_regs(struct ds *result, int level, const char *title,
5372                   struct trace_ctx *trace)
5373 {
5374     size_t i;
5375
5376     ds_put_char_multiple(result, '\t', level);
5377     ds_put_format(result, "%s:", title);
5378     for (i = 0; i < FLOW_N_REGS; i++) {
5379         ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5380     }
5381     ds_put_char(result, '\n');
5382 }
5383
5384 static void
5385 trace_format_odp(struct ds *result, int level, const char *title,
5386                  struct trace_ctx *trace)
5387 {
5388     struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5389
5390     ds_put_char_multiple(result, '\t', level);
5391     ds_put_format(result, "%s: ", title);
5392     format_odp_actions(result, odp_actions->data, odp_actions->size);
5393     ds_put_char(result, '\n');
5394 }
5395
5396 static void
5397 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5398 {
5399     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5400     struct ds *result = trace->result;
5401
5402     ds_put_char(result, '\n');
5403     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5404     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5405     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
5406     trace_format_rule(result, recurse + 1, rule);
5407 }
5408
5409 static void
5410 trace_report(struct xlate_in *xin, const char *s, int recurse)
5411 {
5412     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5413     struct ds *result = trace->result;
5414
5415     ds_put_char_multiple(result, '\t', recurse);
5416     ds_put_cstr(result, s);
5417     ds_put_char(result, '\n');
5418 }
5419
5420 static void
5421 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5422                       void *aux OVS_UNUSED)
5423 {
5424     const struct dpif_backer *backer;
5425     struct ofproto_dpif *ofproto;
5426     struct ofpbuf odp_key, odp_mask;
5427     struct ofpbuf *packet;
5428     struct ds result;
5429     struct flow flow;
5430     char *s;
5431
5432     packet = NULL;
5433     backer = NULL;
5434     ds_init(&result);
5435     ofpbuf_init(&odp_key, 0);
5436     ofpbuf_init(&odp_mask, 0);
5437
5438     /* Handle "-generate" or a hex string as the last argument. */
5439     if (!strcmp(argv[argc - 1], "-generate")) {
5440         packet = ofpbuf_new(0);
5441         argc--;
5442     } else {
5443         const char *error = eth_from_hex(argv[argc - 1], &packet);
5444         if (!error) {
5445             argc--;
5446         } else if (argc == 4) {
5447             /* The 3-argument form must end in "-generate' or a hex string. */
5448             unixctl_command_reply_error(conn, error);
5449             goto exit;
5450         }
5451     }
5452
5453     /* Parse the flow and determine whether a datapath or
5454      * bridge is specified. If function odp_flow_key_from_string()
5455      * returns 0, the flow is a odp_flow. If function
5456      * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5457     if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5458         /* If the odp_flow is the second argument,
5459          * the datapath name is the first argument. */
5460         if (argc == 3) {
5461             const char *dp_type;
5462             if (!strncmp(argv[1], "ovs-", 4)) {
5463                 dp_type = argv[1] + 4;
5464             } else {
5465                 dp_type = argv[1];
5466             }
5467             backer = shash_find_data(&all_dpif_backers, dp_type);
5468             if (!backer) {
5469                 unixctl_command_reply_error(conn, "Cannot find datapath "
5470                                "of this name");
5471                 goto exit;
5472             }
5473         } else {
5474             /* No datapath name specified, so there should be only one
5475              * datapath. */
5476             struct shash_node *node;
5477             if (shash_count(&all_dpif_backers) != 1) {
5478                 unixctl_command_reply_error(conn, "Must specify datapath "
5479                          "name, there is more than one type of datapath");
5480                 goto exit;
5481             }
5482             node = shash_first(&all_dpif_backers);
5483             backer = node->data;
5484         }
5485
5486         if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5487                           NULL, &ofproto, NULL)) {
5488             unixctl_command_reply_error(conn, "Invalid datapath flow");
5489             goto exit;
5490         }
5491         ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5492     } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5493         if (argc != 3) {
5494             unixctl_command_reply_error(conn, "Must specify bridge name");
5495             goto exit;
5496         }
5497
5498         ofproto = ofproto_dpif_lookup(argv[1]);
5499         if (!ofproto) {
5500             unixctl_command_reply_error(conn, "Unknown bridge name");
5501             goto exit;
5502         }
5503     } else {
5504         unixctl_command_reply_error(conn, "Bad flow syntax");
5505         goto exit;
5506     }
5507
5508     /* Generate a packet, if requested. */
5509     if (packet) {
5510         if (!packet->size) {
5511             flow_compose(packet, &flow);
5512         } else {
5513             union flow_in_port in_port_;
5514
5515             in_port_ = flow.in_port;
5516             ds_put_cstr(&result, "Packet: ");
5517             s = ofp_packet_to_string(packet->data, packet->size);
5518             ds_put_cstr(&result, s);
5519             free(s);
5520
5521             /* Use the metadata from the flow and the packet argument
5522              * to reconstruct the flow. */
5523             flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5524                          &in_port_, &flow);
5525         }
5526     }
5527
5528     ofproto_trace(ofproto, &flow, packet, &result);
5529     unixctl_command_reply(conn, ds_cstr(&result));
5530
5531 exit:
5532     ds_destroy(&result);
5533     ofpbuf_delete(packet);
5534     ofpbuf_uninit(&odp_key);
5535     ofpbuf_uninit(&odp_mask);
5536 }
5537
5538 static void
5539 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5540               const struct ofpbuf *packet, struct ds *ds)
5541 {
5542     struct rule_dpif *rule;
5543     struct flow_wildcards wc;
5544
5545     ds_put_cstr(ds, "Flow: ");
5546     flow_format(ds, flow);
5547     ds_put_char(ds, '\n');
5548
5549     flow_wildcards_init_catchall(&wc);
5550     rule_dpif_lookup(ofproto, flow, &wc, &rule);
5551
5552     trace_format_rule(ds, 0, rule);
5553     if (rule == ofproto->miss_rule) {
5554         ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5555     } else if (rule == ofproto->no_packet_in_rule) {
5556         ds_put_cstr(ds, "\nNo match, packets dropped because "
5557                     "OFPPC_NO_PACKET_IN is set on in_port.\n");
5558     } else if (rule == ofproto->drop_frags_rule) {
5559         ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5560                     "and the fragment handling mode is \"drop\".\n");
5561     }
5562
5563     if (rule) {
5564         uint64_t odp_actions_stub[1024 / 8];
5565         struct ofpbuf odp_actions;
5566         struct trace_ctx trace;
5567         struct match match;
5568         uint8_t tcp_flags;
5569
5570         tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5571         trace.result = ds;
5572         trace.flow = *flow;
5573         ofpbuf_use_stub(&odp_actions,
5574                         odp_actions_stub, sizeof odp_actions_stub);
5575         xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5576         trace.xin.resubmit_hook = trace_resubmit;
5577         trace.xin.report_hook = trace_report;
5578
5579         xlate_actions(&trace.xin, &trace.xout);
5580         flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5581
5582         ds_put_char(ds, '\n');
5583         trace_format_flow(ds, 0, "Final flow", &trace);
5584
5585         match_init(&match, flow, &trace.xout.wc);
5586         ds_put_cstr(ds, "Relevant fields: ");
5587         match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5588         ds_put_char(ds, '\n');
5589
5590         ds_put_cstr(ds, "Datapath actions: ");
5591         format_odp_actions(ds, trace.xout.odp_actions.data,
5592                            trace.xout.odp_actions.size);
5593
5594         if (trace.xout.slow) {
5595             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5596                         "slow path because it:");
5597             switch (trace.xout.slow) {
5598             case SLOW_CFM:
5599                 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5600                 break;
5601             case SLOW_LACP:
5602                 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5603                 break;
5604             case SLOW_STP:
5605                 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5606                 break;
5607             case SLOW_BFD:
5608                 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5609                 break;
5610             case SLOW_CONTROLLER:
5611                 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5612                             "to the OpenFlow controller.");
5613                 break;
5614             case __SLOW_MAX:
5615                 NOT_REACHED();
5616             }
5617         }
5618
5619         xlate_out_uninit(&trace.xout);
5620     }
5621
5622     rule_dpif_release(rule);
5623 }
5624
5625 static void
5626 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5627                   const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5628 {
5629     clogged = true;
5630     unixctl_command_reply(conn, NULL);
5631 }
5632
5633 static void
5634 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5635                     const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5636 {
5637     clogged = false;
5638     unixctl_command_reply(conn, NULL);
5639 }
5640
5641 /* Runs a self-check of flow translations in 'ofproto'.  Appends a message to
5642  * 'reply' describing the results. */
5643 static void
5644 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5645 {
5646     struct cls_cursor cursor;
5647     struct facet *facet;
5648     int errors;
5649
5650     errors = 0;
5651     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5652     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5653     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5654         if (!facet_check_consistency(facet)) {
5655             errors++;
5656         }
5657     }
5658     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5659     if (errors) {
5660         ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5661     }
5662
5663     if (errors) {
5664         ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5665                       ofproto->up.name, errors);
5666     } else {
5667         ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5668     }
5669 }
5670
5671 static void
5672 ofproto_dpif_self_check(struct unixctl_conn *conn,
5673                         int argc, const char *argv[], void *aux OVS_UNUSED)
5674 {
5675     struct ds reply = DS_EMPTY_INITIALIZER;
5676     struct ofproto_dpif *ofproto;
5677
5678     if (argc > 1) {
5679         ofproto = ofproto_dpif_lookup(argv[1]);
5680         if (!ofproto) {
5681             unixctl_command_reply_error(conn, "Unknown ofproto (use "
5682                                         "ofproto/list for help)");
5683             return;
5684         }
5685         ofproto_dpif_self_check__(ofproto, &reply);
5686     } else {
5687         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5688             ofproto_dpif_self_check__(ofproto, &reply);
5689         }
5690     }
5691
5692     unixctl_command_reply(conn, ds_cstr(&reply));
5693     ds_destroy(&reply);
5694 }
5695
5696 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
5697  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
5698  * to destroy 'ofproto_shash' and free the returned value. */
5699 static const struct shash_node **
5700 get_ofprotos(struct shash *ofproto_shash)
5701 {
5702     const struct ofproto_dpif *ofproto;
5703
5704     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5705         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5706         shash_add_nocopy(ofproto_shash, name, ofproto);
5707     }
5708
5709     return shash_sort(ofproto_shash);
5710 }
5711
5712 static void
5713 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5714                               const char *argv[] OVS_UNUSED,
5715                               void *aux OVS_UNUSED)
5716 {
5717     struct ds ds = DS_EMPTY_INITIALIZER;
5718     struct shash ofproto_shash;
5719     const struct shash_node **sorted_ofprotos;
5720     int i;
5721
5722     shash_init(&ofproto_shash);
5723     sorted_ofprotos = get_ofprotos(&ofproto_shash);
5724     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5725         const struct shash_node *node = sorted_ofprotos[i];
5726         ds_put_format(&ds, "%s\n", node->name);
5727     }
5728
5729     shash_destroy(&ofproto_shash);
5730     free(sorted_ofprotos);
5731
5732     unixctl_command_reply(conn, ds_cstr(&ds));
5733     ds_destroy(&ds);
5734 }
5735
5736 static void
5737 show_dp_rates(struct ds *ds, const char *heading,
5738               const struct avg_subfacet_rates *rates)
5739 {
5740     ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5741                   heading, rates->add_rate, rates->del_rate);
5742 }
5743
5744 static void
5745 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5746 {
5747     const struct shash_node **ofprotos;
5748     struct ofproto_dpif *ofproto;
5749     struct shash ofproto_shash;
5750     uint64_t n_hit, n_missed;
5751     long long int minutes;
5752     size_t i;
5753
5754     n_hit = n_missed = 0;
5755     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5756         if (ofproto->backer == backer) {
5757             n_missed += ofproto->n_missed;
5758             n_hit += ofproto->n_hit;
5759         }
5760     }
5761
5762     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5763                   dpif_name(backer->dpif), n_hit, n_missed);
5764     ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5765                   " life span: %lldms\n", hmap_count(&backer->subfacets),
5766                   backer->avg_n_subfacet, backer->max_n_subfacet,
5767                   backer->avg_subfacet_life);
5768
5769     minutes = (time_msec() - backer->created) / (1000 * 60);
5770     if (minutes >= 60) {
5771         show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5772     }
5773     if (minutes >= 60 * 24) {
5774         show_dp_rates(ds, "\tdaily avg:",  &backer->daily);
5775     }
5776     show_dp_rates(ds, "\toverall avg:",  &backer->lifetime);
5777
5778     shash_init(&ofproto_shash);
5779     ofprotos = get_ofprotos(&ofproto_shash);
5780     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5781         struct ofproto_dpif *ofproto = ofprotos[i]->data;
5782         const struct shash_node **ports;
5783         size_t j;
5784
5785         if (ofproto->backer != backer) {
5786             continue;
5787         }
5788
5789         ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5790                       ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5791
5792         ports = shash_sort(&ofproto->up.port_by_name);
5793         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5794             const struct shash_node *node = ports[j];
5795             struct ofport *ofport = node->data;
5796             struct smap config;
5797             odp_port_t odp_port;
5798
5799             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5800                           ofport->ofp_port);
5801
5802             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5803             if (odp_port != ODPP_NONE) {
5804                 ds_put_format(ds, "%"PRIu32":", odp_port);
5805             } else {
5806                 ds_put_cstr(ds, "none:");
5807             }
5808
5809             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5810
5811             smap_init(&config);
5812             if (!netdev_get_config(ofport->netdev, &config)) {
5813                 const struct smap_node **nodes;
5814                 size_t i;
5815
5816                 nodes = smap_sort(&config);
5817                 for (i = 0; i < smap_count(&config); i++) {
5818                     const struct smap_node *node = nodes[i];
5819                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5820                                   node->key, node->value);
5821                 }
5822                 free(nodes);
5823             }
5824             smap_destroy(&config);
5825
5826             ds_put_char(ds, ')');
5827             ds_put_char(ds, '\n');
5828         }
5829         free(ports);
5830     }
5831     shash_destroy(&ofproto_shash);
5832     free(ofprotos);
5833 }
5834
5835 static void
5836 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5837                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5838 {
5839     struct ds ds = DS_EMPTY_INITIALIZER;
5840     const struct shash_node **backers;
5841     int i;
5842
5843     backers = shash_sort(&all_dpif_backers);
5844     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5845         dpif_show_backer(backers[i]->data, &ds);
5846     }
5847     free(backers);
5848
5849     unixctl_command_reply(conn, ds_cstr(&ds));
5850     ds_destroy(&ds);
5851 }
5852
5853 /* Dump the megaflow (facet) cache.  This is useful to check the
5854  * correctness of flow wildcarding, since the same mechanism is used for
5855  * both xlate caching and kernel wildcarding.
5856  *
5857  * It's important to note that in the output the flow description uses
5858  * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5859  *
5860  * This command is only needed for advanced debugging, so it's not
5861  * documented in the man page. */
5862 static void
5863 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5864                                     int argc OVS_UNUSED, const char *argv[],
5865                                     void *aux OVS_UNUSED)
5866 {
5867     struct ds ds = DS_EMPTY_INITIALIZER;
5868     const struct ofproto_dpif *ofproto;
5869     long long int now = time_msec();
5870     struct cls_cursor cursor;
5871     struct facet *facet;
5872
5873     ofproto = ofproto_dpif_lookup(argv[1]);
5874     if (!ofproto) {
5875         unixctl_command_reply_error(conn, "no such bridge");
5876         return;
5877     }
5878
5879     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5880     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5881     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5882         cls_rule_format(&facet->cr, &ds);
5883         ds_put_cstr(&ds, ", ");
5884         ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5885         ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5886         ds_put_cstr(&ds, "Datapath actions: ");
5887         if (facet->xout.slow) {
5888             uint64_t slow_path_stub[128 / 8];
5889             const struct nlattr *actions;
5890             size_t actions_len;
5891
5892             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5893                               slow_path_stub, sizeof slow_path_stub,
5894                               &actions, &actions_len);
5895             format_odp_actions(&ds, actions, actions_len);
5896         } else {
5897             format_odp_actions(&ds, facet->xout.odp_actions.data,
5898                                facet->xout.odp_actions.size);
5899         }
5900         ds_put_cstr(&ds, "\n");
5901     }
5902     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5903
5904     ds_chomp(&ds, '\n');
5905     unixctl_command_reply(conn, ds_cstr(&ds));
5906     ds_destroy(&ds);
5907 }
5908
5909 /* Disable using the megaflows.
5910  *
5911  * This command is only needed for advanced debugging, so it's not
5912  * documented in the man page. */
5913 static void
5914 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5915                                        int argc OVS_UNUSED,
5916                                        const char *argv[] OVS_UNUSED,
5917                                        void *aux OVS_UNUSED)
5918 {
5919     struct ofproto_dpif *ofproto;
5920
5921     enable_megaflows = false;
5922
5923     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5924         flush(&ofproto->up);
5925     }
5926
5927     unixctl_command_reply(conn, "megaflows disabled");
5928 }
5929
5930 /* Re-enable using megaflows.
5931  *
5932  * This command is only needed for advanced debugging, so it's not
5933  * documented in the man page. */
5934 static void
5935 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5936                                       int argc OVS_UNUSED,
5937                                       const char *argv[] OVS_UNUSED,
5938                                       void *aux OVS_UNUSED)
5939 {
5940     struct ofproto_dpif *ofproto;
5941
5942     enable_megaflows = true;
5943
5944     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5945         flush(&ofproto->up);
5946     }
5947
5948     unixctl_command_reply(conn, "megaflows enabled");
5949 }
5950
5951 static void
5952 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5953                                 int argc OVS_UNUSED, const char *argv[],
5954                                 void *aux OVS_UNUSED)
5955 {
5956     struct ds ds = DS_EMPTY_INITIALIZER;
5957     const struct ofproto_dpif *ofproto;
5958     struct subfacet *subfacet;
5959
5960     ofproto = ofproto_dpif_lookup(argv[1]);
5961     if (!ofproto) {
5962         unixctl_command_reply_error(conn, "no such bridge");
5963         return;
5964     }
5965
5966     update_stats(ofproto->backer);
5967
5968     HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5969         struct facet *facet = subfacet->facet;
5970         struct odputil_keybuf maskbuf;
5971         struct ofpbuf mask;
5972
5973         if (facet->ofproto != ofproto) {
5974             continue;
5975         }
5976
5977         ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5978         if (enable_megaflows) {
5979             odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5980                                    &facet->flow, UINT32_MAX);
5981         }
5982
5983         odp_flow_format(subfacet->key, subfacet->key_len,
5984                         mask.data, mask.size, &ds, false);
5985
5986         ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5987                       subfacet->dp_packet_count, subfacet->dp_byte_count);
5988         if (subfacet->used) {
5989             ds_put_format(&ds, "%.3fs",
5990                           (time_msec() - subfacet->used) / 1000.0);
5991         } else {
5992             ds_put_format(&ds, "never");
5993         }
5994         if (subfacet->facet->tcp_flags) {
5995             ds_put_cstr(&ds, ", flags:");
5996             packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5997         }
5998
5999         ds_put_cstr(&ds, ", actions:");
6000         if (facet->xout.slow) {
6001             uint64_t slow_path_stub[128 / 8];
6002             const struct nlattr *actions;
6003             size_t actions_len;
6004
6005             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6006                               slow_path_stub, sizeof slow_path_stub,
6007                               &actions, &actions_len);
6008             format_odp_actions(&ds, actions, actions_len);
6009         } else {
6010             format_odp_actions(&ds, facet->xout.odp_actions.data,
6011                                facet->xout.odp_actions.size);
6012         }
6013         ds_put_char(&ds, '\n');
6014     }
6015
6016     unixctl_command_reply(conn, ds_cstr(&ds));
6017     ds_destroy(&ds);
6018 }
6019
6020 static void
6021 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6022                                int argc OVS_UNUSED, const char *argv[],
6023                                void *aux OVS_UNUSED)
6024 {
6025     struct ds ds = DS_EMPTY_INITIALIZER;
6026     struct ofproto_dpif *ofproto;
6027
6028     ofproto = ofproto_dpif_lookup(argv[1]);
6029     if (!ofproto) {
6030         unixctl_command_reply_error(conn, "no such bridge");
6031         return;
6032     }
6033
6034     flush(&ofproto->up);
6035
6036     unixctl_command_reply(conn, ds_cstr(&ds));
6037     ds_destroy(&ds);
6038 }
6039
6040 static void
6041 ofproto_dpif_unixctl_init(void)
6042 {
6043     static bool registered;
6044     if (registered) {
6045         return;
6046     }
6047     registered = true;
6048
6049     unixctl_command_register(
6050         "ofproto/trace",
6051         "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6052         1, 3, ofproto_unixctl_trace, NULL);
6053     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6054                              ofproto_unixctl_fdb_flush, NULL);
6055     unixctl_command_register("fdb/show", "bridge", 1, 1,
6056                              ofproto_unixctl_fdb_show, NULL);
6057     unixctl_command_register("ofproto/clog", "", 0, 0,
6058                              ofproto_dpif_clog, NULL);
6059     unixctl_command_register("ofproto/unclog", "", 0, 0,
6060                              ofproto_dpif_unclog, NULL);
6061     unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6062                              ofproto_dpif_self_check, NULL);
6063     unixctl_command_register("dpif/dump-dps", "", 0, 0,
6064                              ofproto_unixctl_dpif_dump_dps, NULL);
6065     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6066                              NULL);
6067     unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6068                              ofproto_unixctl_dpif_dump_flows, NULL);
6069     unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6070                              ofproto_unixctl_dpif_del_flows, NULL);
6071     unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6072                              ofproto_unixctl_dpif_dump_megaflows, NULL);
6073     unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6074                              ofproto_unixctl_dpif_disable_megaflows, NULL);
6075     unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6076                              ofproto_unixctl_dpif_enable_megaflows, NULL);
6077 }
6078 \f
6079 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6080  *
6081  * This is deprecated.  It is only for compatibility with broken device drivers
6082  * in old versions of Linux that do not properly support VLANs when VLAN
6083  * devices are not used.  When broken device drivers are no longer in
6084  * widespread use, we will delete these interfaces. */
6085
6086 static int
6087 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6088 {
6089     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6090     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6091
6092     if (realdev_ofp_port == ofport->realdev_ofp_port
6093         && vid == ofport->vlandev_vid) {
6094         return 0;
6095     }
6096
6097     ofproto->backer->need_revalidate = REV_RECONFIGURE;
6098
6099     if (ofport->realdev_ofp_port) {
6100         vsp_remove(ofport);
6101     }
6102     if (realdev_ofp_port && ofport->bundle) {
6103         /* vlandevs are enslaved to their realdevs, so they are not allowed to
6104          * themselves be part of a bundle. */
6105         bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6106     }
6107
6108     ofport->realdev_ofp_port = realdev_ofp_port;
6109     ofport->vlandev_vid = vid;
6110
6111     if (realdev_ofp_port) {
6112         vsp_add(ofport, realdev_ofp_port, vid);
6113     }
6114
6115     return 0;
6116 }
6117
6118 static uint32_t
6119 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6120 {
6121     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6122 }
6123
6124 bool
6125 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6126     OVS_EXCLUDED(ofproto->vsp_mutex)
6127 {
6128     bool ret;
6129
6130     ovs_mutex_lock(&ofproto->vsp_mutex);
6131     ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6132     ovs_mutex_unlock(&ofproto->vsp_mutex);
6133     return ret;
6134 }
6135
6136 static ofp_port_t
6137 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6138                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6139     OVS_REQUIRES(ofproto->vsp_mutex)
6140 {
6141     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6142         int vid = vlan_tci_to_vid(vlan_tci);
6143         const struct vlan_splinter *vsp;
6144
6145         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6146                                  hash_realdev_vid(realdev_ofp_port, vid),
6147                                  &ofproto->realdev_vid_map) {
6148             if (vsp->realdev_ofp_port == realdev_ofp_port
6149                 && vsp->vid == vid) {
6150                 return vsp->vlandev_ofp_port;
6151             }
6152         }
6153     }
6154     return realdev_ofp_port;
6155 }
6156
6157 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6158  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6159  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
6160  * 'vlan_tci' 9, it would return the port number of eth0.9.
6161  *
6162  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6163  * function just returns its 'realdev_ofp_port' argument. */
6164 ofp_port_t
6165 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6166                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6167     OVS_EXCLUDED(ofproto->vsp_mutex)
6168 {
6169     ofp_port_t ret;
6170
6171     ovs_mutex_lock(&ofproto->vsp_mutex);
6172     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6173     ovs_mutex_unlock(&ofproto->vsp_mutex);
6174     return ret;
6175 }
6176
6177 static struct vlan_splinter *
6178 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6179 {
6180     struct vlan_splinter *vsp;
6181
6182     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6183                              hash_ofp_port(vlandev_ofp_port),
6184                              &ofproto->vlandev_map) {
6185         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6186             return vsp;
6187         }
6188     }
6189
6190     return NULL;
6191 }
6192
6193 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6194  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6195  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
6196  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6197  * eth0 and store 9 in '*vid'.
6198  *
6199  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6200  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
6201  * always does.*/
6202 static ofp_port_t
6203 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6204                        ofp_port_t vlandev_ofp_port, int *vid)
6205     OVS_REQUIRES(ofproto->vsp_mutex)
6206 {
6207     if (!hmap_is_empty(&ofproto->vlandev_map)) {
6208         const struct vlan_splinter *vsp;
6209
6210         vsp = vlandev_find(ofproto, vlandev_ofp_port);
6211         if (vsp) {
6212             if (vid) {
6213                 *vid = vsp->vid;
6214             }
6215             return vsp->realdev_ofp_port;
6216         }
6217     }
6218     return 0;
6219 }
6220
6221 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6222  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
6223  * 'flow->in_port' to the "real" device backing the VLAN device, sets
6224  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Otherwise (which is
6225  * always the case unless VLAN splinters are enabled), returns false without
6226  * making any changes. */
6227 bool
6228 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6229     OVS_EXCLUDED(ofproto->vsp_mutex)
6230 {
6231     ofp_port_t realdev;
6232     int vid;
6233
6234     ovs_mutex_lock(&ofproto->vsp_mutex);
6235     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6236     ovs_mutex_unlock(&ofproto->vsp_mutex);
6237     if (!realdev) {
6238         return false;
6239     }
6240
6241     /* Cause the flow to be processed as if it came in on the real device with
6242      * the VLAN device's VLAN ID. */
6243     flow->in_port.ofp_port = realdev;
6244     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6245     return true;
6246 }
6247
6248 static void
6249 vsp_remove(struct ofport_dpif *port)
6250 {
6251     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6252     struct vlan_splinter *vsp;
6253
6254     ovs_mutex_lock(&ofproto->vsp_mutex);
6255     vsp = vlandev_find(ofproto, port->up.ofp_port);
6256     if (vsp) {
6257         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6258         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6259         free(vsp);
6260
6261         port->realdev_ofp_port = 0;
6262     } else {
6263         VLOG_ERR("missing vlan device record");
6264     }
6265     ovs_mutex_unlock(&ofproto->vsp_mutex);
6266 }
6267
6268 static void
6269 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6270 {
6271     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6272
6273     ovs_mutex_lock(&ofproto->vsp_mutex);
6274     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6275         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6276             == realdev_ofp_port)) {
6277         struct vlan_splinter *vsp;
6278
6279         vsp = xmalloc(sizeof *vsp);
6280         vsp->realdev_ofp_port = realdev_ofp_port;
6281         vsp->vlandev_ofp_port = port->up.ofp_port;
6282         vsp->vid = vid;
6283
6284         port->realdev_ofp_port = realdev_ofp_port;
6285
6286         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6287                     hash_ofp_port(port->up.ofp_port));
6288         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6289                     hash_realdev_vid(realdev_ofp_port, vid));
6290     } else {
6291         VLOG_ERR("duplicate vlan device record");
6292     }
6293     ovs_mutex_unlock(&ofproto->vsp_mutex);
6294 }
6295
6296 static odp_port_t
6297 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6298 {
6299     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6300     return ofport ? ofport->odp_port : ODPP_NONE;
6301 }
6302
6303 struct ofport_dpif *
6304 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6305 {
6306     struct ofport_dpif *port;
6307
6308     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6309     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6310                              &backer->odp_to_ofport_map) {
6311         if (port->odp_port == odp_port) {
6312             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6313             return port;
6314         }
6315     }
6316
6317     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6318     return NULL;
6319 }
6320
6321 static ofp_port_t
6322 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6323 {
6324     struct ofport_dpif *port;
6325
6326     port = odp_port_to_ofport(ofproto->backer, odp_port);
6327     if (port && &ofproto->up == port->up.ofproto) {
6328         return port->up.ofp_port;
6329     } else {
6330         return OFPP_NONE;
6331     }
6332 }
6333
6334 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6335  * most heavily weighted element.  'base' designates the rate of decay: after
6336  * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6337  * (about .37). */
6338 static void
6339 exp_mavg(double *avg, int base, double new)
6340 {
6341     *avg = (*avg * (base - 1) + new) / base;
6342 }
6343
6344 static void
6345 update_moving_averages(struct dpif_backer *backer)
6346 {
6347     const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6348     long long int minutes = (time_msec() - backer->created) / min_ms;
6349
6350     if (minutes > 0) {
6351         backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6352             / minutes;
6353         backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6354             / minutes;
6355     } else {
6356         backer->lifetime.add_rate = 0.0;
6357         backer->lifetime.del_rate = 0.0;
6358     }
6359
6360     /* Update hourly averages on the minute boundaries. */
6361     if (time_msec() - backer->last_minute >= min_ms) {
6362         exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6363         exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6364
6365         /* Update daily averages on the hour boundaries. */
6366         if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6367             exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6368             exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6369         }
6370
6371         backer->total_subfacet_add_count += backer->subfacet_add_count;
6372         backer->total_subfacet_del_count += backer->subfacet_del_count;
6373         backer->subfacet_add_count = 0;
6374         backer->subfacet_del_count = 0;
6375         backer->last_minute += min_ms;
6376     }
6377 }
6378
6379 const struct ofproto_class ofproto_dpif_class = {
6380     init,
6381     enumerate_types,
6382     enumerate_names,
6383     del,
6384     port_open_type,
6385     type_run,
6386     type_run_fast,
6387     type_wait,
6388     alloc,
6389     construct,
6390     destruct,
6391     dealloc,
6392     run,
6393     run_fast,
6394     wait,
6395     get_memory_usage,
6396     flush,
6397     get_features,
6398     get_tables,
6399     port_alloc,
6400     port_construct,
6401     port_destruct,
6402     port_dealloc,
6403     port_modified,
6404     port_reconfigured,
6405     port_query_by_name,
6406     port_add,
6407     port_del,
6408     port_get_stats,
6409     port_dump_start,
6410     port_dump_next,
6411     port_dump_done,
6412     port_poll,
6413     port_poll_wait,
6414     port_is_lacp_current,
6415     NULL,                       /* rule_choose_table */
6416     rule_alloc,
6417     rule_construct,
6418     rule_insert,
6419     rule_delete,
6420     rule_destruct,
6421     rule_dealloc,
6422     rule_get_stats,
6423     rule_execute,
6424     rule_modify_actions,
6425     set_frag_handling,
6426     packet_out,
6427     set_netflow,
6428     get_netflow_ids,
6429     set_sflow,
6430     set_ipfix,
6431     set_cfm,
6432     get_cfm_status,
6433     set_bfd,
6434     get_bfd_status,
6435     set_stp,
6436     get_stp_status,
6437     set_stp_port,
6438     get_stp_port_status,
6439     set_queues,
6440     bundle_set,
6441     bundle_remove,
6442     mirror_set__,
6443     mirror_get_stats__,
6444     set_flood_vlans,
6445     is_mirror_output_bundle,
6446     forward_bpdu_changed,
6447     set_mac_table_config,
6448     set_realdev,
6449     NULL,                       /* meter_get_features */
6450     NULL,                       /* meter_set */
6451     NULL,                       /* meter_get */
6452     NULL,                       /* meter_del */
6453     NULL,                       /* group_alloc */
6454     NULL,                       /* group_construct */
6455     NULL,                       /* group_destruct */
6456     NULL,                       /* group_dealloc */
6457     NULL,                       /* group_modify */
6458     NULL,                       /* group_get_stats */
6459 };