ofproto/ofproto-dpif.c

   1 /*
   2  * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18
  19 #include "ofproto/ofproto-dpif.h"
  20 #include "ofproto/ofproto-provider.h"
  21
  22 #include <errno.h>
  23
  24 #include "bfd.h"
  25 #include "bond.h"
  26 #include "bundle.h"
  27 #include "byte-order.h"
  28 #include "connmgr.h"
  29 #include "coverage.h"
  30 #include "cfm.h"
  31 #include "dpif.h"
  32 #include "dynamic-string.h"
  33 #include "fail-open.h"
  34 #include "guarded-list.h"
  35 #include "hmapx.h"
  36 #include "lacp.h"
  37 #include "learn.h"
  38 #include "mac-learning.h"
  39 #include "meta-flow.h"
  40 #include "multipath.h"
  41 #include "netdev-vport.h"
  42 #include "netdev.h"
  43 #include "netlink.h"
  44 #include "nx-match.h"
  45 #include "odp-util.h"
  46 #include "odp-execute.h"
  47 #include "ofp-util.h"
  48 #include "ofpbuf.h"
  49 #include "ofp-actions.h"
  50 #include "ofp-parse.h"
  51 #include "ofp-print.h"
  52 #include "ofproto-dpif-governor.h"
  53 #include "ofproto-dpif-ipfix.h"
  54 #include "ofproto-dpif-mirror.h"
  55 #include "ofproto-dpif-sflow.h"
  56 #include "ofproto-dpif-upcall.h"
  57 #include "ofproto-dpif-xlate.h"
  58 #include "poll-loop.h"
  59 #include "simap.h"
  60 #include "smap.h"
  61 #include "timer.h"
  62 #include "tunnel.h"
  63 #include "unaligned.h"
  64 #include "unixctl.h"
  65 #include "vlan-bitmap.h"
  66 #include "vlog.h"
  67
  68 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  69
  70 COVERAGE_DEFINE(ofproto_dpif_expired);
  71 COVERAGE_DEFINE(facet_changed_rule);
  72 COVERAGE_DEFINE(facet_revalidate);
  73 COVERAGE_DEFINE(facet_unexpected);
  74 COVERAGE_DEFINE(facet_suppress);
  75 COVERAGE_DEFINE(subfacet_install_fail);
  76 COVERAGE_DEFINE(packet_in_overflow);
  77 COVERAGE_DEFINE(flow_mod_overflow);
  78
  79 /* Number of implemented OpenFlow tables. */
  80 enum { N_TABLES = 255 };
  81 enum { TBL_INTERNAL = N_TABLES - 1 };    /* Used for internal hidden rules. */
  82 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
  83
  84 struct flow_miss;
  85 struct facet;
  86
  87 struct rule_dpif {
  88     struct rule up;
  89
  90     /* These statistics:
  91      *
  92      *   - Do include packets and bytes from facets that have been deleted or
  93      *     whose own statistics have been folded into the rule.
  94      *
  95      *   - Do include packets and bytes sent "by hand" that were accounted to
  96      *     the rule without any facet being involved (this is a rare corner
  97      *     case in rule_execute()).
  98      *
  99      *   - Do not include packet or bytes that can be obtained from any facet's
 100      *     packet_count or byte_count member or that can be obtained from the
 101      *     datapath by, e.g., dpif_flow_get() for any subfacet.
 102      */
 103     struct ovs_mutex stats_mutex;
 104     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 105     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 106 };
 107
 108 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
 109 static struct rule_dpif *rule_dpif_cast(const struct rule *);
 110
 111 struct ofbundle {
 112     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
 113     struct ofproto_dpif *ofproto; /* Owning ofproto. */
 114     void *aux;                  /* Key supplied by ofproto's client. */
 115     char *name;                 /* Identifier for log messages. */
 116
 117     /* Configuration. */
 118     struct list ports;          /* Contains "struct ofport"s. */
 119     enum port_vlan_mode vlan_mode; /* VLAN mode */
 120     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
 121     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
 122                                  * NULL if all VLANs are trunked. */
 123     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 124     struct bond *bond;          /* Nonnull iff more than one port. */
 125     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 126
 127     /* Status. */
 128     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 129 };
 130
 131 static void bundle_remove(struct ofport *);
 132 static void bundle_update(struct ofbundle *);
 133 static void bundle_destroy(struct ofbundle *);
 134 static void bundle_del_port(struct ofport_dpif *);
 135 static void bundle_run(struct ofbundle *);
 136 static void bundle_wait(struct ofbundle *);
 137
 138 static void stp_run(struct ofproto_dpif *ofproto);
 139 static void stp_wait(struct ofproto_dpif *ofproto);
 140 static int set_stp_port(struct ofport *,
 141                         const struct ofproto_port_stp_settings *);
 142
 143 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
 144                               enum slow_path_reason,
 145                               uint64_t *stub, size_t stub_size,
 146                               const struct nlattr **actionsp,
 147                               size_t *actions_lenp);
 148
 149 /* A subfacet (see "struct subfacet" below) has three possible installation
 150  * states:
 151  *
 152  *   - SF_NOT_INSTALLED: Not installed in the datapath.  This will only be the
 153  *     case just after the subfacet is created, just before the subfacet is
 154  *     destroyed, or if the datapath returns an error when we try to install a
 155  *     subfacet.
 156  *
 157  *   - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
 158  *
 159  *   - SF_SLOW_PATH: An action that sends every packet for the subfacet through
 160  *     ofproto_dpif is installed in the datapath.
 161  */
 162 enum subfacet_path {
 163     SF_NOT_INSTALLED,           /* No datapath flow for this subfacet. */
 164     SF_FAST_PATH,               /* Full actions are installed. */
 165     SF_SLOW_PATH,               /* Send-to-userspace action is installed. */
 166 };
 167
 168 /* A dpif flow and actions associated with a facet.
 169  *
 170  * See also the large comment on struct facet. */
 171 struct subfacet {
 172     /* Owners. */
 173     struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
 174     struct list list_node;      /* In struct facet's 'facets' list. */
 175     struct facet *facet;        /* Owning facet. */
 176     struct dpif_backer *backer; /* Owning backer. */
 177
 178     enum odp_key_fitness key_fitness;
 179     struct nlattr *key;
 180     int key_len;
 181
 182     long long int used;         /* Time last used; time created if not used. */
 183     long long int created;      /* Time created. */
 184
 185     uint64_t dp_packet_count;   /* Last known packet count in the datapath. */
 186     uint64_t dp_byte_count;     /* Last known byte count in the datapath. */
 187
 188     enum subfacet_path path;    /* Installed in datapath? */
 189 };
 190
 191 #define SUBFACET_DESTROY_MAX_BATCH 50
 192
 193 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *);
 194 static struct subfacet *subfacet_find(struct dpif_backer *,
 195                                       const struct nlattr *key, size_t key_len,
 196                                       uint32_t key_hash);
 197 static void subfacet_destroy(struct subfacet *);
 198 static void subfacet_destroy__(struct subfacet *);
 199 static void subfacet_destroy_batch(struct dpif_backer *,
 200                                    struct subfacet **, int n);
 201 static void subfacet_reset_dp_stats(struct subfacet *,
 202                                     struct dpif_flow_stats *);
 203 static void subfacet_update_stats(struct subfacet *,
 204                                   const struct dpif_flow_stats *);
 205 static int subfacet_install(struct subfacet *,
 206                             const struct ofpbuf *odp_actions,
 207                             struct dpif_flow_stats *);
 208 static void subfacet_uninstall(struct subfacet *);
 209
 210 /* A unique, non-overlapping instantiation of an OpenFlow flow.
 211  *
 212  * A facet associates a "struct flow", which represents the Open vSwitch
 213  * userspace idea of an exact-match flow, with one or more subfacets.
 214  * While the facet is created based on an exact-match flow, it is stored
 215  * within the ofproto based on the wildcards that could be expressed
 216  * based on the flow table and other configuration.  (See the 'wc'
 217  * description in "struct xlate_out" for more details.)
 218  *
 219  * Each subfacet tracks the datapath's idea of the flow equivalent to
 220  * the facet.  When the kernel module (or other dpif implementation) and
 221  * Open vSwitch userspace agree on the definition of a flow key, there
 222  * is exactly one subfacet per facet.  If the dpif implementation
 223  * supports more-specific flow matching than userspace, however, a facet
 224  * can have more than one subfacet.  Examples include the dpif
 225  * implementation not supporting the same wildcards as userspace or some
 226  * distinction in flow that userspace simply doesn't understand.
 227  *
 228  * Flow expiration works in terms of subfacets, so a facet must have at
 229  * least one subfacet or it will never expire, leaking memory. */
 230 struct facet {
 231     /* Owner. */
 232     struct ofproto_dpif *ofproto;
 233
 234     /* Owned data. */
 235     struct list subfacets;
 236     long long int used;         /* Time last used; time created if not used. */
 237
 238     /* Key. */
 239     struct flow flow;           /* Flow of the creating subfacet. */
 240     struct cls_rule cr;         /* In 'ofproto_dpif's facets classifier. */
 241
 242     /* These statistics:
 243      *
 244      *   - Do include packets and bytes sent "by hand", e.g. with
 245      *     dpif_execute().
 246      *
 247      *   - Do include packets and bytes that were obtained from the datapath
 248      *     when a subfacet's statistics were reset (e.g. dpif_flow_put() with
 249      *     DPIF_FP_ZERO_STATS).
 250      *
 251      *   - Do not include packets or bytes that can be obtained from the
 252      *     datapath for any existing subfacet.
 253      */
 254     uint64_t packet_count;       /* Number of packets received. */
 255     uint64_t byte_count;         /* Number of bytes received. */
 256
 257     /* Resubmit statistics. */
 258     uint64_t prev_packet_count;  /* Number of packets from last stats push. */
 259     uint64_t prev_byte_count;    /* Number of bytes from last stats push. */
 260     long long int prev_used;     /* Used time from last stats push. */
 261
 262     /* Accounting. */
 263     uint64_t accounted_bytes;    /* Bytes processed by facet_account(). */
 264     struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
 265     uint8_t tcp_flags;           /* TCP flags seen for this 'rule'. */
 266
 267     struct xlate_out xout;
 268
 269     /* Storage for a single subfacet, to reduce malloc() time and space
 270      * overhead.  (A facet always has at least one subfacet and in the common
 271      * case has exactly one subfacet.  However, 'one_subfacet' may not
 272      * always be valid, since it could have been removed after newer
 273      * subfacets were pushed onto the 'subfacets' list.) */
 274     struct subfacet one_subfacet;
 275
 276     long long int learn_rl;      /* Rate limiter for facet_learn(). */
 277 };
 278
 279 static struct facet *facet_create(const struct flow_miss *);
 280 static void facet_remove(struct facet *);
 281 static void facet_free(struct facet *);
 282
 283 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
 284 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
 285                                         const struct flow *);
 286 static bool facet_revalidate(struct facet *);
 287 static bool facet_check_consistency(struct facet *);
 288
 289 static void facet_flush_stats(struct facet *);
 290
 291 static void facet_reset_counters(struct facet *);
 292 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
 293                             struct dpif_flow_stats *, bool may_learn);
 294 static void facet_push_stats(struct facet *, bool may_learn);
 295 static void facet_learn(struct facet *);
 296 static void facet_account(struct facet *);
 297 static void push_all_stats(void);
 298
 299 static bool facet_is_controller_flow(struct facet *);
 300
 301 struct ofport_dpif {
 302     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 303     struct ofport up;
 304
 305     odp_port_t odp_port;
 306     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 307     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
 308     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 309     struct bfd *bfd;            /* BFD, if any. */
 310     bool may_enable;            /* May be enabled in bonds. */
 311     bool is_tunnel;             /* This port is a tunnel. */
 312     long long int carrier_seq;  /* Carrier status changes. */
 313     struct ofport_dpif *peer;   /* Peer if patch port. */
 314
 315     /* Spanning tree. */
 316     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 317     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 318     long long int stp_state_entered;
 319
 320     /* Queue to DSCP mapping. */
 321     struct ofproto_port_queue *qdscp;
 322     size_t n_qdscp;
 323
 324     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 325      *
 326      * This is deprecated.  It is only for compatibility with broken device
 327      * drivers in old versions of Linux that do not properly support VLANs when
 328      * VLAN devices are not used.  When broken device drivers are no longer in
 329      * widespread use, we will delete these interfaces. */
 330     ofp_port_t realdev_ofp_port;
 331     int vlandev_vid;
 332 };
 333
 334 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 335  *
 336  * This is deprecated.  It is only for compatibility with broken device drivers
 337  * in old versions of Linux that do not properly support VLANs when VLAN
 338  * devices are not used.  When broken device drivers are no longer in
 339  * widespread use, we will delete these interfaces. */
 340 struct vlan_splinter {
 341     struct hmap_node realdev_vid_node;
 342     struct hmap_node vlandev_node;
 343     ofp_port_t realdev_ofp_port;
 344     ofp_port_t vlandev_ofp_port;
 345     int vid;
 346 };
 347
 348 static void vsp_remove(struct ofport_dpif *);
 349 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 350
 351 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 352                                        ofp_port_t);
 353
 354 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 355                                        odp_port_t);
 356
 357 static struct ofport_dpif *
 358 ofport_dpif_cast(const struct ofport *ofport)
 359 {
 360     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 361 }
 362
 363 static void port_run(struct ofport_dpif *);
 364 static void port_run_fast(struct ofport_dpif *);
 365 static void port_wait(struct ofport_dpif *);
 366 static int set_bfd(struct ofport *, const struct smap *);
 367 static int set_cfm(struct ofport *, const struct cfm_settings *);
 368 static void ofport_update_peer(struct ofport_dpif *);
 369 static void run_fast_rl(void);
 370 static int run_fast(struct ofproto *);
 371
 372 struct dpif_completion {
 373     struct list list_node;
 374     struct ofoperation *op;
 375 };
 376
 377 /* Reasons that we might need to revalidate every facet, and corresponding
 378  * coverage counters.
 379  *
 380  * A value of 0 means that there is no need to revalidate.
 381  *
 382  * It would be nice to have some cleaner way to integrate with coverage
 383  * counters, but with only a few reasons I guess this is good enough for
 384  * now. */
 385 enum revalidate_reason {
 386     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 387     REV_STP,                   /* Spanning tree protocol port status change. */
 388     REV_BOND,                  /* Bonding changed. */
 389     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 390     REV_FLOW_TABLE,            /* Flow table changed. */
 391     REV_MAC_LEARNING,          /* Mac learning changed. */
 392     REV_INCONSISTENCY          /* Facet self-check failed. */
 393 };
 394 COVERAGE_DEFINE(rev_reconfigure);
 395 COVERAGE_DEFINE(rev_stp);
 396 COVERAGE_DEFINE(rev_bond);
 397 COVERAGE_DEFINE(rev_port_toggled);
 398 COVERAGE_DEFINE(rev_flow_table);
 399 COVERAGE_DEFINE(rev_mac_learning);
 400 COVERAGE_DEFINE(rev_inconsistency);
 401
 402 struct avg_subfacet_rates {
 403     double add_rate;   /* Moving average of new flows created per minute. */
 404     double del_rate;   /* Moving average of flows deleted per minute. */
 405 };
 406
 407 /* All datapaths of a given type share a single dpif backer instance. */
 408 struct dpif_backer {
 409     char *type;
 410     int refcount;
 411     struct dpif *dpif;
 412     struct udpif *udpif;
 413     struct timer next_expiration;
 414
 415     struct ovs_rwlock odp_to_ofport_lock;
 416     struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
 417
 418     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 419
 420     /* Facet revalidation flags applying to facets which use this backer. */
 421     enum revalidate_reason need_revalidate; /* Revalidate every facet. */
 422
 423     struct hmap drop_keys; /* Set of dropped odp keys. */
 424     bool recv_set_enable; /* Enables or disables receiving packets. */
 425
 426     struct hmap subfacets;
 427     struct governor *governor;
 428
 429     /* Subfacet statistics.
 430      *
 431      * These keep track of the total number of subfacets added and deleted and
 432      * flow life span.  They are useful for computing the flow rates stats
 433      * exposed via "ovs-appctl dpif/show".  The goal is to learn about
 434      * traffic patterns in ways that we can use later to improve Open vSwitch
 435      * performance in new situations.  */
 436     long long int created;           /* Time when it is created. */
 437     unsigned max_n_subfacet;         /* Maximum number of flows */
 438     unsigned avg_n_subfacet;         /* Average number of flows. */
 439     long long int avg_subfacet_life; /* Average life span of subfacets. */
 440
 441     /* The average number of subfacets... */
 442     struct avg_subfacet_rates hourly;   /* ...over the last hour. */
 443     struct avg_subfacet_rates daily;    /* ...over the last day. */
 444     struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
 445     long long int last_minute;          /* Last time 'hourly' was updated. */
 446
 447     /* Number of subfacets added or deleted since 'last_minute'. */
 448     unsigned subfacet_add_count;
 449     unsigned subfacet_del_count;
 450
 451     /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
 452     unsigned long long int total_subfacet_add_count;
 453     unsigned long long int total_subfacet_del_count;
 454
 455     /* Number of upcall handling threads. */
 456     unsigned int n_handler_threads;
 457 };
 458
 459 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 460 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 461
 462 static void drop_key_clear(struct dpif_backer *);
 463 static void update_moving_averages(struct dpif_backer *backer);
 464
 465 struct ofproto_dpif {
 466     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 467     struct ofproto up;
 468     struct dpif_backer *backer;
 469
 470     /* Special OpenFlow rules. */
 471     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 472     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 473     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 474
 475     /* Bridging. */
 476     struct netflow *netflow;
 477     struct dpif_sflow *sflow;
 478     struct dpif_ipfix *ipfix;
 479     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 480     struct mac_learning *ml;
 481     bool has_bonded_bundles;
 482     struct mbridge *mbridge;
 483
 484     /* Facets. */
 485     struct classifier facets;     /* Contains 'struct facet's. */
 486     long long int consistency_rl;
 487
 488     struct netdev_stats stats; /* To account packets generated and consumed in
 489                                 * userspace. */
 490
 491     /* Spanning tree. */
 492     struct stp *stp;
 493     long long int stp_last_tick;
 494
 495     /* VLAN splinters. */
 496     struct ovs_mutex vsp_mutex;
 497     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 498     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 499
 500     /* Ports. */
 501     struct sset ports;             /* Set of standard port names. */
 502     struct sset ghost_ports;       /* Ports with no datapath port. */
 503     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 504     int port_poll_errno;           /* Last errno for port_poll() reply. */
 505
 506     /* Per ofproto's dpif stats. */
 507     uint64_t n_hit;
 508     uint64_t n_missed;
 509
 510     /* Work queues. */
 511     struct guarded_list pins;      /* Contains "struct ofputil_packet_in"s. */
 512 };
 513
 514 /* By default, flows in the datapath are wildcarded (megaflows).  They
 515  * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
 516 static bool enable_megaflows = true;
 517
 518 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 519 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 520
 521 static void ofproto_dpif_unixctl_init(void);
 522
 523 static inline struct ofproto_dpif *
 524 ofproto_dpif_cast(const struct ofproto *ofproto)
 525 {
 526     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 527     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 528 }
 529
 530 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
 531                                         ofp_port_t ofp_port);
 532 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
 533                           const struct ofpbuf *packet, struct ds *);
 534
 535 /* Upcalls. */
 536 static void handle_upcalls(struct dpif_backer *);
 537
 538 /* Flow expiration. */
 539 static int expire(struct dpif_backer *);
 540
 541 /* NetFlow. */
 542 static void send_netflow_active_timeouts(struct ofproto_dpif *);
 543
 544 /* Utilities. */
 545 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
 546
 547 /* Global variables. */
 548 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 549
 550 /* Initial mappings of port to bridge mappings. */
 551 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 552
 553 /* Executes 'fm'.  The caller retains ownership of 'fm' and everything in
 554  * it. */
 555 void
 556 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 557                       struct ofputil_flow_mod *fm)
 558 {
 559     ofproto_flow_mod(&ofproto->up, fm);
 560 }
 561
 562 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 563  * Takes ownership of 'pin' and pin->packet. */
 564 void
 565 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 566                             struct ofputil_packet_in *pin)
 567 {
 568     if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
 569         COVERAGE_INC(packet_in_overflow);
 570         free(CONST_CAST(void *, pin->packet));
 571         free(pin);
 572     }
 573 }
 574 \f
 575 /* Factory functions. */
 576
 577 static void
 578 init(const struct shash *iface_hints)
 579 {
 580     struct shash_node *node;
 581
 582     /* Make a local copy, since we don't own 'iface_hints' elements. */
 583     SHASH_FOR_EACH(node, iface_hints) {
 584         const struct iface_hint *orig_hint = node->data;
 585         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 586
 587         new_hint->br_name = xstrdup(orig_hint->br_name);
 588         new_hint->br_type = xstrdup(orig_hint->br_type);
 589         new_hint->ofp_port = orig_hint->ofp_port;
 590
 591         shash_add(&init_ofp_ports, node->name, new_hint);
 592     }
 593 }
 594
 595 static void
 596 enumerate_types(struct sset *types)
 597 {
 598     dp_enumerate_types(types);
 599 }
 600
 601 static int
 602 enumerate_names(const char *type, struct sset *names)
 603 {
 604     struct ofproto_dpif *ofproto;
 605
 606     sset_clear(names);
 607     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 608         if (strcmp(type, ofproto->up.type)) {
 609             continue;
 610         }
 611         sset_add(names, ofproto->up.name);
 612     }
 613
 614     return 0;
 615 }
 616
 617 static int
 618 del(const char *type, const char *name)
 619 {
 620     struct dpif *dpif;
 621     int error;
 622
 623     error = dpif_open(name, type, &dpif);
 624     if (!error) {
 625         error = dpif_delete(dpif);
 626         dpif_close(dpif);
 627     }
 628     return error;
 629 }
 630 \f
 631 static const char *
 632 port_open_type(const char *datapath_type, const char *port_type)
 633 {
 634     return dpif_port_open_type(datapath_type, port_type);
 635 }
 636
 637 /* Type functions. */
 638
 639 static void process_dpif_port_changes(struct dpif_backer *);
 640 static void process_dpif_all_ports_changed(struct dpif_backer *);
 641 static void process_dpif_port_change(struct dpif_backer *,
 642                                      const char *devname);
 643 static void process_dpif_port_error(struct dpif_backer *, int error);
 644
 645 static struct ofproto_dpif *
 646 lookup_ofproto_dpif_by_port_name(const char *name)
 647 {
 648     struct ofproto_dpif *ofproto;
 649
 650     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 651         if (sset_contains(&ofproto->ports, name)) {
 652             return ofproto;
 653         }
 654     }
 655
 656     return NULL;
 657 }
 658
 659 static int
 660 type_run(const char *type)
 661 {
 662     static long long int push_timer = LLONG_MIN;
 663     struct dpif_backer *backer;
 664
 665     backer = shash_find_data(&all_dpif_backers, type);
 666     if (!backer) {
 667         /* This is not necessarily a problem, since backers are only
 668          * created on demand. */
 669         return 0;
 670     }
 671
 672     dpif_run(backer->dpif);
 673
 674     /* The most natural place to push facet statistics is when they're pulled
 675      * from the datapath.  However, when there are many flows in the datapath,
 676      * this expensive operation can occur so frequently, that it reduces our
 677      * ability to quickly set up flows.  To reduce the cost, we push statistics
 678      * here instead. */
 679     if (time_msec() > push_timer) {
 680         push_timer = time_msec() + 2000;
 681         push_all_stats();
 682     }
 683
 684     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 685      * and the configuration has now changed to "false", enable receiving
 686      * packets from the datapath. */
 687     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 688         int error;
 689
 690         backer->recv_set_enable = true;
 691
 692         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 693         if (error) {
 694             udpif_recv_set(backer->udpif, 0, false);
 695             VLOG_ERR("Failed to enable receiving packets in dpif.");
 696             return error;
 697         }
 698         udpif_recv_set(backer->udpif, n_handler_threads,
 699                        backer->recv_set_enable);
 700         dpif_flow_flush(backer->dpif);
 701         backer->need_revalidate = REV_RECONFIGURE;
 702     }
 703
 704     /* If the n_handler_threads is reconfigured, call udpif_recv_set()
 705      * to reset the handler threads. */
 706     if (backer->n_handler_threads != n_handler_threads) {
 707         udpif_recv_set(backer->udpif, n_handler_threads,
 708                        backer->recv_set_enable);
 709         backer->n_handler_threads = n_handler_threads;
 710     }
 711
 712     if (backer->need_revalidate) {
 713         struct ofproto_dpif *ofproto;
 714         struct simap_node *node;
 715         struct simap tmp_backers;
 716
 717         /* Handle tunnel garbage collection. */
 718         simap_init(&tmp_backers);
 719         simap_swap(&backer->tnl_backers, &tmp_backers);
 720
 721         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 722             struct ofport_dpif *iter;
 723
 724             if (backer != ofproto->backer) {
 725                 continue;
 726             }
 727
 728             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 729                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 730                 const char *dp_port;
 731
 732                 if (!iter->is_tunnel) {
 733                     continue;
 734                 }
 735
 736                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 737                                                      namebuf, sizeof namebuf);
 738                 node = simap_find(&tmp_backers, dp_port);
 739                 if (node) {
 740                     simap_put(&backer->tnl_backers, dp_port, node->data);
 741                     simap_delete(&tmp_backers, node);
 742                     node = simap_find(&backer->tnl_backers, dp_port);
 743                 } else {
 744                     node = simap_find(&backer->tnl_backers, dp_port);
 745                     if (!node) {
 746                         odp_port_t odp_port = ODPP_NONE;
 747
 748                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 749                                            &odp_port)) {
 750                             simap_put(&backer->tnl_backers, dp_port,
 751                                       odp_to_u32(odp_port));
 752                             node = simap_find(&backer->tnl_backers, dp_port);
 753                         }
 754                     }
 755                 }
 756
 757                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 758                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 759                                          iter->odp_port)) {
 760                     backer->need_revalidate = REV_RECONFIGURE;
 761                 }
 762             }
 763         }
 764
 765         SIMAP_FOR_EACH (node, &tmp_backers) {
 766             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 767         }
 768         simap_destroy(&tmp_backers);
 769
 770         switch (backer->need_revalidate) {
 771         case REV_RECONFIGURE:   COVERAGE_INC(rev_reconfigure);   break;
 772         case REV_STP:           COVERAGE_INC(rev_stp);           break;
 773         case REV_BOND:          COVERAGE_INC(rev_bond);          break;
 774         case REV_PORT_TOGGLED:  COVERAGE_INC(rev_port_toggled);  break;
 775         case REV_FLOW_TABLE:    COVERAGE_INC(rev_flow_table);    break;
 776         case REV_MAC_LEARNING:  COVERAGE_INC(rev_mac_learning);  break;
 777         case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
 778         }
 779         backer->need_revalidate = 0;
 780
 781         /* Clear the drop_keys in case we should now be accepting some
 782          * formerly dropped flows. */
 783         drop_key_clear(backer);
 784
 785         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 786             struct facet *facet, *next;
 787             struct ofport_dpif *ofport;
 788             struct cls_cursor cursor;
 789             struct ofbundle *bundle;
 790
 791             if (ofproto->backer != backer) {
 792                 continue;
 793             }
 794
 795             ovs_rwlock_wrlock(&xlate_rwlock);
 796             xlate_ofproto_set(ofproto, ofproto->up.name,
 797                               ofproto->backer->dpif, ofproto->miss_rule,
 798                               ofproto->no_packet_in_rule, ofproto->ml,
 799                               ofproto->stp, ofproto->mbridge,
 800                               ofproto->sflow, ofproto->ipfix,
 801                               ofproto->up.frag_handling,
 802                               ofproto->up.forward_bpdu,
 803                               connmgr_has_in_band(ofproto->up.connmgr),
 804                               ofproto->netflow != NULL);
 805
 806             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 807                 xlate_bundle_set(ofproto, bundle, bundle->name,
 808                                  bundle->vlan_mode, bundle->vlan,
 809                                  bundle->trunks, bundle->use_priority_tags,
 810                                  bundle->bond, bundle->lacp,
 811                                  bundle->floodable);
 812             }
 813
 814             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 815                 int stp_port = ofport->stp_port
 816                     ? stp_port_no(ofport->stp_port)
 817                     : -1;
 818                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 819                                  ofport->up.ofp_port, ofport->odp_port,
 820                                  ofport->up.netdev, ofport->cfm,
 821                                  ofport->bfd, ofport->peer, stp_port,
 822                                  ofport->qdscp, ofport->n_qdscp,
 823                                  ofport->up.pp.config, ofport->is_tunnel,
 824                                  ofport->may_enable);
 825             }
 826             ovs_rwlock_unlock(&xlate_rwlock);
 827
 828             /* Only ofproto-dpif cares about the facet classifier so we just
 829              * lock cls_cursor_init() to appease the thread safety analysis. */
 830             ovs_rwlock_rdlock(&ofproto->facets.rwlock);
 831             cls_cursor_init(&cursor, &ofproto->facets, NULL);
 832             ovs_rwlock_unlock(&ofproto->facets.rwlock);
 833             CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
 834                 facet_revalidate(facet);
 835                 run_fast_rl();
 836             }
 837         }
 838
 839         udpif_revalidate(backer->udpif);
 840     }
 841
 842     if (!backer->recv_set_enable) {
 843         /* Wake up before a max of 1000ms. */
 844         timer_set_duration(&backer->next_expiration, 1000);
 845     } else if (timer_expired(&backer->next_expiration)) {
 846         int delay = expire(backer);
 847         timer_set_duration(&backer->next_expiration, delay);
 848     }
 849
 850     process_dpif_port_changes(backer);
 851
 852     if (backer->governor) {
 853         size_t n_subfacets;
 854
 855         governor_run(backer->governor);
 856
 857         /* If the governor has shrunk to its minimum size and the number of
 858          * subfacets has dwindled, then drop the governor entirely.
 859          *
 860          * For hysteresis, the number of subfacets to drop the governor is
 861          * smaller than the number needed to trigger its creation. */
 862         n_subfacets = hmap_count(&backer->subfacets);
 863         if (n_subfacets * 4 < flow_eviction_threshold
 864             && governor_is_idle(backer->governor)) {
 865             governor_destroy(backer->governor);
 866             backer->governor = NULL;
 867         }
 868     }
 869
 870     return 0;
 871 }
 872
 873 /* Check for and handle port changes in 'backer''s dpif. */
 874 static void
 875 process_dpif_port_changes(struct dpif_backer *backer)
 876 {
 877     for (;;) {
 878         char *devname;
 879         int error;
 880
 881         error = dpif_port_poll(backer->dpif, &devname);
 882         switch (error) {
 883         case EAGAIN:
 884             return;
 885
 886         case ENOBUFS:
 887             process_dpif_all_ports_changed(backer);
 888             break;
 889
 890         case 0:
 891             process_dpif_port_change(backer, devname);
 892             free(devname);
 893             break;
 894
 895         default:
 896             process_dpif_port_error(backer, error);
 897             break;
 898         }
 899     }
 900 }
 901
 902 static void
 903 process_dpif_all_ports_changed(struct dpif_backer *backer)
 904 {
 905     struct ofproto_dpif *ofproto;
 906     struct dpif_port dpif_port;
 907     struct dpif_port_dump dump;
 908     struct sset devnames;
 909     const char *devname;
 910
 911     sset_init(&devnames);
 912     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 913         if (ofproto->backer == backer) {
 914             struct ofport *ofport;
 915
 916             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 917                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 918             }
 919         }
 920     }
 921     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 922         sset_add(&devnames, dpif_port.name);
 923     }
 924
 925     SSET_FOR_EACH (devname, &devnames) {
 926         process_dpif_port_change(backer, devname);
 927     }
 928     sset_destroy(&devnames);
 929 }
 930
 931 static void
 932 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 933 {
 934     struct ofproto_dpif *ofproto;
 935     struct dpif_port port;
 936
 937     /* Don't report on the datapath's device. */
 938     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 939         return;
 940     }
 941
 942     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 943                    &all_ofproto_dpifs) {
 944         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 945             return;
 946         }
 947     }
 948
 949     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 950     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 951         /* The port was removed.  If we know the datapath,
 952          * report it through poll_set().  If we don't, it may be
 953          * notifying us of a removal we initiated, so ignore it.
 954          * If there's a pending ENOBUFS, let it stand, since
 955          * everything will be reevaluated. */
 956         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 957             sset_add(&ofproto->port_poll_set, devname);
 958             ofproto->port_poll_errno = 0;
 959         }
 960     } else if (!ofproto) {
 961         /* The port was added, but we don't know with which
 962          * ofproto we should associate it.  Delete it. */
 963         dpif_port_del(backer->dpif, port.port_no);
 964     } else {
 965         struct ofport_dpif *ofport;
 966
 967         ofport = ofport_dpif_cast(shash_find_data(
 968                                       &ofproto->up.port_by_name, devname));
 969         if (ofport
 970             && ofport->odp_port != port.port_no
 971             && !odp_port_to_ofport(backer, port.port_no))
 972         {
 973             /* 'ofport''s datapath port number has changed from
 974              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 975              * structures to match. */
 976             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 977             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 978             ofport->odp_port = port.port_no;
 979             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 980                         hash_odp_port(port.port_no));
 981             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 982             backer->need_revalidate = REV_RECONFIGURE;
 983         }
 984     }
 985     dpif_port_destroy(&port);
 986 }
 987
 988 /* Propagate 'error' to all ofprotos based on 'backer'. */
 989 static void
 990 process_dpif_port_error(struct dpif_backer *backer, int error)
 991 {
 992     struct ofproto_dpif *ofproto;
 993
 994     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 995         if (ofproto->backer == backer) {
 996             sset_clear(&ofproto->port_poll_set);
 997             ofproto->port_poll_errno = error;
 998         }
 999     }
1000 }
1001
1002 static int
1003 dpif_backer_run_fast(struct dpif_backer *backer)
1004 {
1005     handle_upcalls(backer);
1006
1007     return 0;
1008 }
1009
1010 static int
1011 type_run_fast(const char *type)
1012 {
1013     struct dpif_backer *backer;
1014
1015     backer = shash_find_data(&all_dpif_backers, type);
1016     if (!backer) {
1017         /* This is not necessarily a problem, since backers are only
1018          * created on demand. */
1019         return 0;
1020     }
1021
1022     return dpif_backer_run_fast(backer);
1023 }
1024
1025 static void
1026 run_fast_rl(void)
1027 {
1028     static long long int port_rl = LLONG_MIN;
1029
1030     if (time_msec() >= port_rl) {
1031         struct ofproto_dpif *ofproto;
1032
1033         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1034             run_fast(&ofproto->up);
1035         }
1036         port_rl = time_msec() + 200;
1037     }
1038 }
1039
1040 static void
1041 type_wait(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;
1050     }
1051
1052     if (backer->governor) {
1053         governor_wait(backer->governor);
1054     }
1055
1056     timer_wait(&backer->next_expiration);
1057     dpif_wait(backer->dpif);
1058     udpif_wait(backer->udpif);
1059 }
1060 \f
1061 /* Basic life-cycle. */
1062
1063 static int add_internal_flows(struct ofproto_dpif *);
1064
1065 static struct ofproto *
1066 alloc(void)
1067 {
1068     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1069     return &ofproto->up;
1070 }
1071
1072 static void
1073 dealloc(struct ofproto *ofproto_)
1074 {
1075     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1076     free(ofproto);
1077 }
1078
1079 static void
1080 close_dpif_backer(struct dpif_backer *backer)
1081 {
1082     ovs_assert(backer->refcount > 0);
1083
1084     if (--backer->refcount) {
1085         return;
1086     }
1087
1088     drop_key_clear(backer);
1089     hmap_destroy(&backer->drop_keys);
1090
1091     udpif_destroy(backer->udpif);
1092
1093     simap_destroy(&backer->tnl_backers);
1094     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1095     hmap_destroy(&backer->odp_to_ofport_map);
1096     shash_find_and_delete(&all_dpif_backers, backer->type);
1097     free(backer->type);
1098     dpif_close(backer->dpif);
1099
1100     ovs_assert(hmap_is_empty(&backer->subfacets));
1101     hmap_destroy(&backer->subfacets);
1102     governor_destroy(backer->governor);
1103
1104     free(backer);
1105 }
1106
1107 /* Datapath port slated for removal from datapath. */
1108 struct odp_garbage {
1109     struct list list_node;
1110     odp_port_t odp_port;
1111 };
1112
1113 static int
1114 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1115 {
1116     struct dpif_backer *backer;
1117     struct dpif_port_dump port_dump;
1118     struct dpif_port port;
1119     struct shash_node *node;
1120     struct list garbage_list;
1121     struct odp_garbage *garbage, *next;
1122     struct sset names;
1123     char *backer_name;
1124     const char *name;
1125     int error;
1126
1127     backer = shash_find_data(&all_dpif_backers, type);
1128     if (backer) {
1129         backer->refcount++;
1130         *backerp = backer;
1131         return 0;
1132     }
1133
1134     backer_name = xasprintf("ovs-%s", type);
1135
1136     /* Remove any existing datapaths, since we assume we're the only
1137      * userspace controlling the datapath. */
1138     sset_init(&names);
1139     dp_enumerate_names(type, &names);
1140     SSET_FOR_EACH(name, &names) {
1141         struct dpif *old_dpif;
1142
1143         /* Don't remove our backer if it exists. */
1144         if (!strcmp(name, backer_name)) {
1145             continue;
1146         }
1147
1148         if (dpif_open(name, type, &old_dpif)) {
1149             VLOG_WARN("couldn't open old datapath %s to remove it", name);
1150         } else {
1151             dpif_delete(old_dpif);
1152             dpif_close(old_dpif);
1153         }
1154     }
1155     sset_destroy(&names);
1156
1157     backer = xmalloc(sizeof *backer);
1158
1159     error = dpif_create_and_open(backer_name, type, &backer->dpif);
1160     free(backer_name);
1161     if (error) {
1162         VLOG_ERR("failed to open datapath of type %s: %s", type,
1163                  ovs_strerror(error));
1164         free(backer);
1165         return error;
1166     }
1167     backer->udpif = udpif_create(backer, backer->dpif);
1168
1169     backer->type = xstrdup(type);
1170     backer->governor = NULL;
1171     backer->refcount = 1;
1172     hmap_init(&backer->odp_to_ofport_map);
1173     ovs_rwlock_init(&backer->odp_to_ofport_lock);
1174     hmap_init(&backer->drop_keys);
1175     hmap_init(&backer->subfacets);
1176     timer_set_duration(&backer->next_expiration, 1000);
1177     backer->need_revalidate = 0;
1178     simap_init(&backer->tnl_backers);
1179     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1180     *backerp = backer;
1181
1182     if (backer->recv_set_enable) {
1183         dpif_flow_flush(backer->dpif);
1184     }
1185
1186     /* Loop through the ports already on the datapath and remove any
1187      * that we don't need anymore. */
1188     list_init(&garbage_list);
1189     dpif_port_dump_start(&port_dump, backer->dpif);
1190     while (dpif_port_dump_next(&port_dump, &port)) {
1191         node = shash_find(&init_ofp_ports, port.name);
1192         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1193             garbage = xmalloc(sizeof *garbage);
1194             garbage->odp_port = port.port_no;
1195             list_push_front(&garbage_list, &garbage->list_node);
1196         }
1197     }
1198     dpif_port_dump_done(&port_dump);
1199
1200     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1201         dpif_port_del(backer->dpif, garbage->odp_port);
1202         list_remove(&garbage->list_node);
1203         free(garbage);
1204     }
1205
1206     shash_add(&all_dpif_backers, type, backer);
1207
1208     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1209     if (error) {
1210         VLOG_ERR("failed to listen on datapath of type %s: %s",
1211                  type, ovs_strerror(error));
1212         close_dpif_backer(backer);
1213         return error;
1214     }
1215     udpif_recv_set(backer->udpif, n_handler_threads,
1216                    backer->recv_set_enable);
1217     backer->n_handler_threads = n_handler_threads;
1218
1219     backer->max_n_subfacet = 0;
1220     backer->created = time_msec();
1221     backer->last_minute = backer->created;
1222     memset(&backer->hourly, 0, sizeof backer->hourly);
1223     memset(&backer->daily, 0, sizeof backer->daily);
1224     memset(&backer->lifetime, 0, sizeof backer->lifetime);
1225     backer->subfacet_add_count = 0;
1226     backer->subfacet_del_count = 0;
1227     backer->total_subfacet_add_count = 0;
1228     backer->total_subfacet_del_count = 0;
1229     backer->avg_n_subfacet = 0;
1230     backer->avg_subfacet_life = 0;
1231
1232     return error;
1233 }
1234
1235 static int
1236 construct(struct ofproto *ofproto_)
1237 {
1238     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1239     struct shash_node *node, *next;
1240     uint32_t max_ports;
1241     int error;
1242
1243     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1244     if (error) {
1245         return error;
1246     }
1247
1248     max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1249     ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX)));
1250
1251     ofproto->netflow = NULL;
1252     ofproto->sflow = NULL;
1253     ofproto->ipfix = NULL;
1254     ofproto->stp = NULL;
1255     hmap_init(&ofproto->bundles);
1256     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1257     ofproto->mbridge = mbridge_create();
1258     ofproto->has_bonded_bundles = false;
1259     ovs_mutex_init(&ofproto->vsp_mutex);
1260
1261     classifier_init(&ofproto->facets);
1262     ofproto->consistency_rl = LLONG_MIN;
1263
1264     guarded_list_init(&ofproto->pins);
1265
1266     ofproto_dpif_unixctl_init();
1267
1268     hmap_init(&ofproto->vlandev_map);
1269     hmap_init(&ofproto->realdev_vid_map);
1270
1271     sset_init(&ofproto->ports);
1272     sset_init(&ofproto->ghost_ports);
1273     sset_init(&ofproto->port_poll_set);
1274     ofproto->port_poll_errno = 0;
1275
1276     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1277         struct iface_hint *iface_hint = node->data;
1278
1279         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1280             /* Check if the datapath already has this port. */
1281             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1282                 sset_add(&ofproto->ports, node->name);
1283             }
1284
1285             free(iface_hint->br_name);
1286             free(iface_hint->br_type);
1287             free(iface_hint);
1288             shash_delete(&init_ofp_ports, node);
1289         }
1290     }
1291
1292     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1293                 hash_string(ofproto->up.name, 0));
1294     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1295
1296     ofproto_init_tables(ofproto_, N_TABLES);
1297     error = add_internal_flows(ofproto);
1298     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1299
1300     ofproto->n_hit = 0;
1301     ofproto->n_missed = 0;
1302
1303     return error;
1304 }
1305
1306 static int
1307 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1308                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1309 {
1310     struct ofputil_flow_mod fm;
1311     int error;
1312
1313     match_init_catchall(&fm.match);
1314     fm.priority = 0;
1315     match_set_reg(&fm.match, 0, id);
1316     fm.new_cookie = htonll(0);
1317     fm.cookie = htonll(0);
1318     fm.cookie_mask = htonll(0);
1319     fm.modify_cookie = false;
1320     fm.table_id = TBL_INTERNAL;
1321     fm.command = OFPFC_ADD;
1322     fm.idle_timeout = 0;
1323     fm.hard_timeout = 0;
1324     fm.buffer_id = 0;
1325     fm.out_port = 0;
1326     fm.flags = 0;
1327     fm.ofpacts = ofpacts->data;
1328     fm.ofpacts_len = ofpacts->size;
1329
1330     error = ofproto_flow_mod(&ofproto->up, &fm);
1331     if (error) {
1332         VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1333                     id, ofperr_to_string(error));
1334         return error;
1335     }
1336
1337     if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1338                                   rulep)) {
1339         rule_dpif_unref(*rulep);
1340     } else {
1341         NOT_REACHED();
1342     }
1343
1344     return 0;
1345 }
1346
1347 static int
1348 add_internal_flows(struct ofproto_dpif *ofproto)
1349 {
1350     struct ofpact_controller *controller;
1351     uint64_t ofpacts_stub[128 / 8];
1352     struct ofpbuf ofpacts;
1353     int error;
1354     int id;
1355
1356     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1357     id = 1;
1358
1359     controller = ofpact_put_CONTROLLER(&ofpacts);
1360     controller->max_len = UINT16_MAX;
1361     controller->controller_id = 0;
1362     controller->reason = OFPR_NO_MATCH;
1363     ofpact_pad(&ofpacts);
1364
1365     error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1366     if (error) {
1367         return error;
1368     }
1369
1370     ofpbuf_clear(&ofpacts);
1371     error = add_internal_flow(ofproto, id++, &ofpacts,
1372                               &ofproto->no_packet_in_rule);
1373     if (error) {
1374         return error;
1375     }
1376
1377     error = add_internal_flow(ofproto, id++, &ofpacts,
1378                               &ofproto->drop_frags_rule);
1379     return error;
1380 }
1381
1382 static void
1383 destruct(struct ofproto *ofproto_)
1384 {
1385     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1386     struct rule_dpif *rule, *next_rule;
1387     struct ofputil_packet_in *pin, *next_pin;
1388     struct facet *facet, *next_facet;
1389     struct cls_cursor cursor;
1390     struct oftable *table;
1391     struct list pins;
1392
1393     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1394     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1395     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1396     CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1397         facet_remove(facet);
1398     }
1399
1400     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1401     ovs_rwlock_wrlock(&xlate_rwlock);
1402     xlate_remove_ofproto(ofproto);
1403     ovs_rwlock_unlock(&xlate_rwlock);
1404
1405     /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a
1406      * use-after-free error. */
1407     udpif_revalidate(ofproto->backer->udpif);
1408
1409     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1410
1411     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1412         struct cls_cursor cursor;
1413
1414         ovs_rwlock_rdlock(&table->cls.rwlock);
1415         cls_cursor_init(&cursor, &table->cls, NULL);
1416         ovs_rwlock_unlock(&table->cls.rwlock);
1417         CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1418             ofproto_rule_delete(&ofproto->up, &rule->up);
1419         }
1420     }
1421
1422     guarded_list_pop_all(&ofproto->pins, &pins);
1423     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1424         list_remove(&pin->list_node);
1425         free(CONST_CAST(void *, pin->packet));
1426         free(pin);
1427     }
1428     guarded_list_destroy(&ofproto->pins);
1429
1430     mbridge_unref(ofproto->mbridge);
1431
1432     netflow_destroy(ofproto->netflow);
1433     dpif_sflow_unref(ofproto->sflow);
1434     hmap_destroy(&ofproto->bundles);
1435     mac_learning_unref(ofproto->ml);
1436
1437     classifier_destroy(&ofproto->facets);
1438
1439     hmap_destroy(&ofproto->vlandev_map);
1440     hmap_destroy(&ofproto->realdev_vid_map);
1441
1442     sset_destroy(&ofproto->ports);
1443     sset_destroy(&ofproto->ghost_ports);
1444     sset_destroy(&ofproto->port_poll_set);
1445
1446     ovs_mutex_destroy(&ofproto->vsp_mutex);
1447
1448     close_dpif_backer(ofproto->backer);
1449 }
1450
1451 static int
1452 run_fast(struct ofproto *ofproto_)
1453 {
1454     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1455     struct ofputil_packet_in *pin, *next_pin;
1456     struct ofport_dpif *ofport;
1457     struct list pins;
1458
1459     /* Do not perform any periodic activity required by 'ofproto' while
1460      * waiting for flow restore to complete. */
1461     if (ofproto_get_flow_restore_wait()) {
1462         return 0;
1463     }
1464
1465     guarded_list_pop_all(&ofproto->pins, &pins);
1466     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1467         connmgr_send_packet_in(ofproto->up.connmgr, pin);
1468         list_remove(&pin->list_node);
1469         free(CONST_CAST(void *, pin->packet));
1470         free(pin);
1471     }
1472
1473     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1474         port_run_fast(ofport);
1475     }
1476
1477     return 0;
1478 }
1479
1480 static int
1481 run(struct ofproto *ofproto_)
1482 {
1483     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1484     struct ofport_dpif *ofport;
1485     struct ofbundle *bundle;
1486     int error;
1487
1488     if (mbridge_need_revalidate(ofproto->mbridge)) {
1489         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1490         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1491         mac_learning_flush(ofproto->ml);
1492         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1493     }
1494
1495     /* Do not perform any periodic activity below required by 'ofproto' while
1496      * waiting for flow restore to complete. */
1497     if (ofproto_get_flow_restore_wait()) {
1498         return 0;
1499     }
1500
1501     error = run_fast(ofproto_);
1502     if (error) {
1503         return error;
1504     }
1505
1506     if (ofproto->netflow) {
1507         if (netflow_run(ofproto->netflow)) {
1508             send_netflow_active_timeouts(ofproto);
1509         }
1510     }
1511     if (ofproto->sflow) {
1512         dpif_sflow_run(ofproto->sflow);
1513     }
1514     if (ofproto->ipfix) {
1515         dpif_ipfix_run(ofproto->ipfix);
1516     }
1517
1518     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1519         port_run(ofport);
1520     }
1521     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1522         bundle_run(bundle);
1523     }
1524
1525     stp_run(ofproto);
1526     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1527     if (mac_learning_run(ofproto->ml)) {
1528         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1529     }
1530     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1531
1532     /* Check the consistency of a random facet, to aid debugging. */
1533     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1534     if (time_msec() >= ofproto->consistency_rl
1535         && !classifier_is_empty(&ofproto->facets)
1536         && !ofproto->backer->need_revalidate) {
1537         struct cls_table *table;
1538         struct cls_rule *cr;
1539         struct facet *facet;
1540
1541         ofproto->consistency_rl = time_msec() + 250;
1542
1543         table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1544                              struct cls_table, hmap_node);
1545         cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1546                           hmap_node);
1547         facet = CONTAINER_OF(cr, struct facet, cr);
1548
1549         if (!facet_check_consistency(facet)) {
1550             ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1551         }
1552     }
1553     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1554
1555     return 0;
1556 }
1557
1558 static void
1559 wait(struct ofproto *ofproto_)
1560 {
1561     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1562     struct ofport_dpif *ofport;
1563     struct ofbundle *bundle;
1564
1565     if (ofproto_get_flow_restore_wait()) {
1566         return;
1567     }
1568
1569     if (ofproto->sflow) {
1570         dpif_sflow_wait(ofproto->sflow);
1571     }
1572     if (ofproto->ipfix) {
1573         dpif_ipfix_wait(ofproto->ipfix);
1574     }
1575     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1576         port_wait(ofport);
1577     }
1578     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1579         bundle_wait(bundle);
1580     }
1581     if (ofproto->netflow) {
1582         netflow_wait(ofproto->netflow);
1583     }
1584     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1585     mac_learning_wait(ofproto->ml);
1586     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1587     stp_wait(ofproto);
1588     if (ofproto->backer->need_revalidate) {
1589         /* Shouldn't happen, but if it does just go around again. */
1590         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1591         poll_immediate_wake();
1592     }
1593 }
1594
1595 static void
1596 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1597 {
1598     const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1599     struct cls_cursor cursor;
1600     size_t n_subfacets = 0;
1601     struct facet *facet;
1602
1603     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1604     simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1605     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1606
1607     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1608     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1609     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1610         n_subfacets += list_size(&facet->subfacets);
1611     }
1612     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1613     simap_increase(usage, "subfacets", n_subfacets);
1614 }
1615
1616 static void
1617 flush(struct ofproto *ofproto_)
1618 {
1619     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1620     struct subfacet *subfacet, *next_subfacet;
1621     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1622     int n_batch;
1623
1624     n_batch = 0;
1625     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1626                         &ofproto->backer->subfacets) {
1627         if (subfacet->facet->ofproto != ofproto) {
1628             continue;
1629         }
1630
1631         if (subfacet->path != SF_NOT_INSTALLED) {
1632             batch[n_batch++] = subfacet;
1633             if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1634                 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1635                 n_batch = 0;
1636             }
1637         } else {
1638             subfacet_destroy(subfacet);
1639         }
1640     }
1641
1642     if (n_batch > 0) {
1643         subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1644     }
1645 }
1646
1647 static void
1648 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1649              bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1650 {
1651     *arp_match_ip = true;
1652     *actions = (OFPUTIL_A_OUTPUT |
1653                 OFPUTIL_A_SET_VLAN_VID |
1654                 OFPUTIL_A_SET_VLAN_PCP |
1655                 OFPUTIL_A_STRIP_VLAN |
1656                 OFPUTIL_A_SET_DL_SRC |
1657                 OFPUTIL_A_SET_DL_DST |
1658                 OFPUTIL_A_SET_NW_SRC |
1659                 OFPUTIL_A_SET_NW_DST |
1660                 OFPUTIL_A_SET_NW_TOS |
1661                 OFPUTIL_A_SET_TP_SRC |
1662                 OFPUTIL_A_SET_TP_DST |
1663                 OFPUTIL_A_ENQUEUE);
1664 }
1665
1666 static void
1667 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1668 {
1669     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1670     struct dpif_dp_stats s;
1671     uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1672     uint64_t n_lookup;
1673
1674     strcpy(ots->name, "classifier");
1675
1676     dpif_get_dp_stats(ofproto->backer->dpif, &s);
1677     rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1678     rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1679     rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1680
1681     n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1682     ots->lookup_count = htonll(n_lookup);
1683     ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1684 }
1685
1686 static struct ofport *
1687 port_alloc(void)
1688 {
1689     struct ofport_dpif *port = xmalloc(sizeof *port);
1690     return &port->up;
1691 }
1692
1693 static void
1694 port_dealloc(struct ofport *port_)
1695 {
1696     struct ofport_dpif *port = ofport_dpif_cast(port_);
1697     free(port);
1698 }
1699
1700 static int
1701 port_construct(struct ofport *port_)
1702 {
1703     struct ofport_dpif *port = ofport_dpif_cast(port_);
1704     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1705     const struct netdev *netdev = port->up.netdev;
1706     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1707     struct dpif_port dpif_port;
1708     int error;
1709
1710     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1711     port->bundle = NULL;
1712     port->cfm = NULL;
1713     port->bfd = NULL;
1714     port->may_enable = true;
1715     port->stp_port = NULL;
1716     port->stp_state = STP_DISABLED;
1717     port->is_tunnel = false;
1718     port->peer = NULL;
1719     port->qdscp = NULL;
1720     port->n_qdscp = 0;
1721     port->realdev_ofp_port = 0;
1722     port->vlandev_vid = 0;
1723     port->carrier_seq = netdev_get_carrier_resets(netdev);
1724
1725     if (netdev_vport_is_patch(netdev)) {
1726         /* By bailing out here, we don't submit the port to the sFlow module
1727          * to be considered for counter polling export.  This is correct
1728          * because the patch port represents an interface that sFlow considers
1729          * to be "internal" to the switch as a whole, and therefore not an
1730          * candidate for counter polling. */
1731         port->odp_port = ODPP_NONE;
1732         ofport_update_peer(port);
1733         return 0;
1734     }
1735
1736     error = dpif_port_query_by_name(ofproto->backer->dpif,
1737                                     netdev_vport_get_dpif_port(netdev, namebuf,
1738                                                                sizeof namebuf),
1739                                     &dpif_port);
1740     if (error) {
1741         return error;
1742     }
1743
1744     port->odp_port = dpif_port.port_no;
1745
1746     if (netdev_get_tunnel_config(netdev)) {
1747         tnl_port_add(port, port->up.netdev, port->odp_port);
1748         port->is_tunnel = true;
1749     } else {
1750         /* Sanity-check that a mapping doesn't already exist.  This
1751          * shouldn't happen for non-tunnel ports. */
1752         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1753             VLOG_ERR("port %s already has an OpenFlow port number",
1754                      dpif_port.name);
1755             dpif_port_destroy(&dpif_port);
1756             return EBUSY;
1757         }
1758
1759         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1760         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1761                     hash_odp_port(port->odp_port));
1762         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1763     }
1764     dpif_port_destroy(&dpif_port);
1765
1766     if (ofproto->sflow) {
1767         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1768     }
1769
1770     return 0;
1771 }
1772
1773 static void
1774 port_destruct(struct ofport *port_)
1775 {
1776     struct ofport_dpif *port = ofport_dpif_cast(port_);
1777     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1778     const char *devname = netdev_get_name(port->up.netdev);
1779     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1780     const char *dp_port_name;
1781
1782     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1783     ovs_rwlock_wrlock(&xlate_rwlock);
1784     xlate_ofport_remove(port);
1785     ovs_rwlock_unlock(&xlate_rwlock);
1786
1787     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1788                                               sizeof namebuf);
1789     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1790         /* The underlying device is still there, so delete it.  This
1791          * happens when the ofproto is being destroyed, since the caller
1792          * assumes that removal of attached ports will happen as part of
1793          * destruction. */
1794         if (!port->is_tunnel) {
1795             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1796         }
1797     }
1798
1799     if (port->peer) {
1800         port->peer->peer = NULL;
1801         port->peer = NULL;
1802     }
1803
1804     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1805         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1806         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1807         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1808     }
1809
1810     tnl_port_del(port);
1811     sset_find_and_delete(&ofproto->ports, devname);
1812     sset_find_and_delete(&ofproto->ghost_ports, devname);
1813     bundle_remove(port_);
1814     set_cfm(port_, NULL);
1815     set_bfd(port_, NULL);
1816     if (ofproto->sflow) {
1817         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1818     }
1819
1820     free(port->qdscp);
1821 }
1822
1823 static void
1824 port_modified(struct ofport *port_)
1825 {
1826     struct ofport_dpif *port = ofport_dpif_cast(port_);
1827
1828     if (port->bundle && port->bundle->bond) {
1829         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1830     }
1831
1832     if (port->cfm) {
1833         cfm_set_netdev(port->cfm, port->up.netdev);
1834     }
1835
1836     if (port->bfd) {
1837         bfd_set_netdev(port->bfd, port->up.netdev);
1838     }
1839
1840     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1841                                                 port->odp_port)) {
1842         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1843             REV_RECONFIGURE;
1844     }
1845
1846     ofport_update_peer(port);
1847 }
1848
1849 static void
1850 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1851 {
1852     struct ofport_dpif *port = ofport_dpif_cast(port_);
1853     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1854     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1855
1856     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1857                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1858                    OFPUTIL_PC_NO_PACKET_IN)) {
1859         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1860
1861         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1862             bundle_update(port->bundle);
1863         }
1864     }
1865 }
1866
1867 static int
1868 set_sflow(struct ofproto *ofproto_,
1869           const struct ofproto_sflow_options *sflow_options)
1870 {
1871     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1872     struct dpif_sflow *ds = ofproto->sflow;
1873
1874     if (sflow_options) {
1875         if (!ds) {
1876             struct ofport_dpif *ofport;
1877
1878             ds = ofproto->sflow = dpif_sflow_create();
1879             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1880                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1881             }
1882             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1883         }
1884         dpif_sflow_set_options(ds, sflow_options);
1885     } else {
1886         if (ds) {
1887             dpif_sflow_unref(ds);
1888             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1889             ofproto->sflow = NULL;
1890         }
1891     }
1892     return 0;
1893 }
1894
1895 static int
1896 set_ipfix(
1897     struct ofproto *ofproto_,
1898     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1899     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1900     size_t n_flow_exporters_options)
1901 {
1902     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1903     struct dpif_ipfix *di = ofproto->ipfix;
1904     bool has_options = bridge_exporter_options || flow_exporters_options;
1905
1906     if (has_options && !di) {
1907         di = ofproto->ipfix = dpif_ipfix_create();
1908     }
1909
1910     if (di) {
1911         /* Call set_options in any case to cleanly flush the flow
1912          * caches in the last exporters that are to be destroyed. */
1913         dpif_ipfix_set_options(
1914             di, bridge_exporter_options, flow_exporters_options,
1915             n_flow_exporters_options);
1916
1917         if (!has_options) {
1918             dpif_ipfix_unref(di);
1919             ofproto->ipfix = NULL;
1920         }
1921     }
1922
1923     return 0;
1924 }
1925
1926 static int
1927 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1928 {
1929     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1930     int error;
1931
1932     if (!s) {
1933         error = 0;
1934     } else {
1935         if (!ofport->cfm) {
1936             struct ofproto_dpif *ofproto;
1937
1938             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1939             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1940             ofport->cfm = cfm_create(ofport->up.netdev);
1941         }
1942
1943         if (cfm_configure(ofport->cfm, s)) {
1944             return 0;
1945         }
1946
1947         error = EINVAL;
1948     }
1949     cfm_unref(ofport->cfm);
1950     ofport->cfm = NULL;
1951     return error;
1952 }
1953
1954 static bool
1955 get_cfm_status(const struct ofport *ofport_,
1956                struct ofproto_cfm_status *status)
1957 {
1958     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1959
1960     if (ofport->cfm) {
1961         status->faults = cfm_get_fault(ofport->cfm);
1962         status->remote_opstate = cfm_get_opup(ofport->cfm);
1963         status->health = cfm_get_health(ofport->cfm);
1964         cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1965         return true;
1966     } else {
1967         return false;
1968     }
1969 }
1970
1971 static int
1972 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1973 {
1974     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1975     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1976     struct bfd *old;
1977
1978     old = ofport->bfd;
1979     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1980                                 cfg, ofport->up.netdev);
1981     if (ofport->bfd != old) {
1982         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1983     }
1984
1985     return 0;
1986 }
1987
1988 static int
1989 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1990 {
1991     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1992
1993     if (ofport->bfd) {
1994         bfd_get_status(ofport->bfd, smap);
1995         return 0;
1996     } else {
1997         return ENOENT;
1998     }
1999 }
2000 \f
2001 /* Spanning Tree. */
2002
2003 static void
2004 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2005 {
2006     struct ofproto_dpif *ofproto = ofproto_;
2007     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2008     struct ofport_dpif *ofport;
2009
2010     ofport = stp_port_get_aux(sp);
2011     if (!ofport) {
2012         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2013                      ofproto->up.name, port_num);
2014     } else {
2015         struct eth_header *eth = pkt->l2;
2016
2017         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2018         if (eth_addr_is_zero(eth->eth_src)) {
2019             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2020                          "with unknown MAC", ofproto->up.name, port_num);
2021         } else {
2022             send_packet(ofport, pkt);
2023         }
2024     }
2025     ofpbuf_delete(pkt);
2026 }
2027
2028 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2029 static int
2030 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2031 {
2032     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2033
2034     /* Only revalidate flows if the configuration changed. */
2035     if (!s != !ofproto->stp) {
2036         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2037     }
2038
2039     if (s) {
2040         if (!ofproto->stp) {
2041             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2042                                       send_bpdu_cb, ofproto);
2043             ofproto->stp_last_tick = time_msec();
2044         }
2045
2046         stp_set_bridge_id(ofproto->stp, s->system_id);
2047         stp_set_bridge_priority(ofproto->stp, s->priority);
2048         stp_set_hello_time(ofproto->stp, s->hello_time);
2049         stp_set_max_age(ofproto->stp, s->max_age);
2050         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2051     }  else {
2052         struct ofport *ofport;
2053
2054         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2055             set_stp_port(ofport, NULL);
2056         }
2057
2058         stp_unref(ofproto->stp);
2059         ofproto->stp = NULL;
2060     }
2061
2062     return 0;
2063 }
2064
2065 static int
2066 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2067 {
2068     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2069
2070     if (ofproto->stp) {
2071         s->enabled = true;
2072         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2073         s->designated_root = stp_get_designated_root(ofproto->stp);
2074         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2075     } else {
2076         s->enabled = false;
2077     }
2078
2079     return 0;
2080 }
2081
2082 static void
2083 update_stp_port_state(struct ofport_dpif *ofport)
2084 {
2085     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2086     enum stp_state state;
2087
2088     /* Figure out new state. */
2089     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2090                              : STP_DISABLED;
2091
2092     /* Update state. */
2093     if (ofport->stp_state != state) {
2094         enum ofputil_port_state of_state;
2095         bool fwd_change;
2096
2097         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2098                     netdev_get_name(ofport->up.netdev),
2099                     stp_state_name(ofport->stp_state),
2100                     stp_state_name(state));
2101         if (stp_learn_in_state(ofport->stp_state)
2102                 != stp_learn_in_state(state)) {
2103             /* xxx Learning action flows should also be flushed. */
2104             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2105             mac_learning_flush(ofproto->ml);
2106             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2107         }
2108         fwd_change = stp_forward_in_state(ofport->stp_state)
2109                         != stp_forward_in_state(state);
2110
2111         ofproto->backer->need_revalidate = REV_STP;
2112         ofport->stp_state = state;
2113         ofport->stp_state_entered = time_msec();
2114
2115         if (fwd_change && ofport->bundle) {
2116             bundle_update(ofport->bundle);
2117         }
2118
2119         /* Update the STP state bits in the OpenFlow port description. */
2120         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2121         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2122                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2123                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2124                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2125                      : 0);
2126         ofproto_port_set_state(&ofport->up, of_state);
2127     }
2128 }
2129
2130 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2131  * caller is responsible for assigning STP port numbers and ensuring
2132  * there are no duplicates. */
2133 static int
2134 set_stp_port(struct ofport *ofport_,
2135              const struct ofproto_port_stp_settings *s)
2136 {
2137     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2138     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2139     struct stp_port *sp = ofport->stp_port;
2140
2141     if (!s || !s->enable) {
2142         if (sp) {
2143             ofport->stp_port = NULL;
2144             stp_port_disable(sp);
2145             update_stp_port_state(ofport);
2146         }
2147         return 0;
2148     } else if (sp && stp_port_no(sp) != s->port_num
2149             && ofport == stp_port_get_aux(sp)) {
2150         /* The port-id changed, so disable the old one if it's not
2151          * already in use by another port. */
2152         stp_port_disable(sp);
2153     }
2154
2155     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2156     stp_port_enable(sp);
2157
2158     stp_port_set_aux(sp, ofport);
2159     stp_port_set_priority(sp, s->priority);
2160     stp_port_set_path_cost(sp, s->path_cost);
2161
2162     update_stp_port_state(ofport);
2163
2164     return 0;
2165 }
2166
2167 static int
2168 get_stp_port_status(struct ofport *ofport_,
2169                     struct ofproto_port_stp_status *s)
2170 {
2171     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2172     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2173     struct stp_port *sp = ofport->stp_port;
2174
2175     if (!ofproto->stp || !sp) {
2176         s->enabled = false;
2177         return 0;
2178     }
2179
2180     s->enabled = true;
2181     s->port_id = stp_port_get_id(sp);
2182     s->state = stp_port_get_state(sp);
2183     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2184     s->role = stp_port_get_role(sp);
2185     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2186
2187     return 0;
2188 }
2189
2190 static void
2191 stp_run(struct ofproto_dpif *ofproto)
2192 {
2193     if (ofproto->stp) {
2194         long long int now = time_msec();
2195         long long int elapsed = now - ofproto->stp_last_tick;
2196         struct stp_port *sp;
2197
2198         if (elapsed > 0) {
2199             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2200             ofproto->stp_last_tick = now;
2201         }
2202         while (stp_get_changed_port(ofproto->stp, &sp)) {
2203             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2204
2205             if (ofport) {
2206                 update_stp_port_state(ofport);
2207             }
2208         }
2209
2210         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2211             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2212             mac_learning_flush(ofproto->ml);
2213             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2214         }
2215     }
2216 }
2217
2218 static void
2219 stp_wait(struct ofproto_dpif *ofproto)
2220 {
2221     if (ofproto->stp) {
2222         poll_timer_wait(1000);
2223     }
2224 }
2225 \f
2226 static int
2227 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2228            size_t n_qdscp)
2229 {
2230     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2231     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2232
2233     if (ofport->n_qdscp != n_qdscp
2234         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2235                               n_qdscp * sizeof *qdscp))) {
2236         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2237         free(ofport->qdscp);
2238         ofport->qdscp = n_qdscp
2239             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2240             : NULL;
2241         ofport->n_qdscp = n_qdscp;
2242     }
2243
2244     return 0;
2245 }
2246 \f
2247 /* Bundles. */
2248
2249 /* Expires all MAC learning entries associated with 'bundle' and forces its
2250  * ofproto to revalidate every flow.
2251  *
2252  * Normally MAC learning entries are removed only from the ofproto associated
2253  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2254  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2255  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2256  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2257  * with the host from which it migrated. */
2258 static void
2259 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2260 {
2261     struct ofproto_dpif *ofproto = bundle->ofproto;
2262     struct mac_learning *ml = ofproto->ml;
2263     struct mac_entry *mac, *next_mac;
2264
2265     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2266     ovs_rwlock_wrlock(&ml->rwlock);
2267     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2268         if (mac->port.p == bundle) {
2269             if (all_ofprotos) {
2270                 struct ofproto_dpif *o;
2271
2272                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2273                     if (o != ofproto) {
2274                         struct mac_entry *e;
2275
2276                         ovs_rwlock_wrlock(&o->ml->rwlock);
2277                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2278                         if (e) {
2279                             mac_learning_expire(o->ml, e);
2280                         }
2281                         ovs_rwlock_unlock(&o->ml->rwlock);
2282                     }
2283                 }
2284             }
2285
2286             mac_learning_expire(ml, mac);
2287         }
2288     }
2289     ovs_rwlock_unlock(&ml->rwlock);
2290 }
2291
2292 static struct ofbundle *
2293 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2294 {
2295     struct ofbundle *bundle;
2296
2297     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2298                              &ofproto->bundles) {
2299         if (bundle->aux == aux) {
2300             return bundle;
2301         }
2302     }
2303     return NULL;
2304 }
2305
2306 static void
2307 bundle_update(struct ofbundle *bundle)
2308 {
2309     struct ofport_dpif *port;
2310
2311     bundle->floodable = true;
2312     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2313         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2314             || !stp_forward_in_state(port->stp_state)) {
2315             bundle->floodable = false;
2316             break;
2317         }
2318     }
2319 }
2320
2321 static void
2322 bundle_del_port(struct ofport_dpif *port)
2323 {
2324     struct ofbundle *bundle = port->bundle;
2325
2326     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2327
2328     list_remove(&port->bundle_node);
2329     port->bundle = NULL;
2330
2331     if (bundle->lacp) {
2332         lacp_slave_unregister(bundle->lacp, port);
2333     }
2334     if (bundle->bond) {
2335         bond_slave_unregister(bundle->bond, port);
2336     }
2337
2338     bundle_update(bundle);
2339 }
2340
2341 static bool
2342 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2343                 struct lacp_slave_settings *lacp)
2344 {
2345     struct ofport_dpif *port;
2346
2347     port = get_ofp_port(bundle->ofproto, ofp_port);
2348     if (!port) {
2349         return false;
2350     }
2351
2352     if (port->bundle != bundle) {
2353         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2354         if (port->bundle) {
2355             bundle_remove(&port->up);
2356         }
2357
2358         port->bundle = bundle;
2359         list_push_back(&bundle->ports, &port->bundle_node);
2360         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2361             || !stp_forward_in_state(port->stp_state)) {
2362             bundle->floodable = false;
2363         }
2364     }
2365     if (lacp) {
2366         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2367         lacp_slave_register(bundle->lacp, port, lacp);
2368     }
2369
2370     return true;
2371 }
2372
2373 static void
2374 bundle_destroy(struct ofbundle *bundle)
2375 {
2376     struct ofproto_dpif *ofproto;
2377     struct ofport_dpif *port, *next_port;
2378
2379     if (!bundle) {
2380         return;
2381     }
2382
2383     ofproto = bundle->ofproto;
2384     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2385
2386     ovs_rwlock_wrlock(&xlate_rwlock);
2387     xlate_bundle_remove(bundle);
2388     ovs_rwlock_unlock(&xlate_rwlock);
2389
2390     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2391         bundle_del_port(port);
2392     }
2393
2394     bundle_flush_macs(bundle, true);
2395     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2396     free(bundle->name);
2397     free(bundle->trunks);
2398     lacp_unref(bundle->lacp);
2399     bond_unref(bundle->bond);
2400     free(bundle);
2401 }
2402
2403 static int
2404 bundle_set(struct ofproto *ofproto_, void *aux,
2405            const struct ofproto_bundle_settings *s)
2406 {
2407     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2408     bool need_flush = false;
2409     struct ofport_dpif *port;
2410     struct ofbundle *bundle;
2411     unsigned long *trunks;
2412     int vlan;
2413     size_t i;
2414     bool ok;
2415
2416     if (!s) {
2417         bundle_destroy(bundle_lookup(ofproto, aux));
2418         return 0;
2419     }
2420
2421     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2422     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2423
2424     bundle = bundle_lookup(ofproto, aux);
2425     if (!bundle) {
2426         bundle = xmalloc(sizeof *bundle);
2427
2428         bundle->ofproto = ofproto;
2429         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2430                     hash_pointer(aux, 0));
2431         bundle->aux = aux;
2432         bundle->name = NULL;
2433
2434         list_init(&bundle->ports);
2435         bundle->vlan_mode = PORT_VLAN_TRUNK;
2436         bundle->vlan = -1;
2437         bundle->trunks = NULL;
2438         bundle->use_priority_tags = s->use_priority_tags;
2439         bundle->lacp = NULL;
2440         bundle->bond = NULL;
2441
2442         bundle->floodable = true;
2443         mbridge_register_bundle(ofproto->mbridge, bundle);
2444     }
2445
2446     if (!bundle->name || strcmp(s->name, bundle->name)) {
2447         free(bundle->name);
2448         bundle->name = xstrdup(s->name);
2449     }
2450
2451     /* LACP. */
2452     if (s->lacp) {
2453         if (!bundle->lacp) {
2454             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2455             bundle->lacp = lacp_create();
2456         }
2457         lacp_configure(bundle->lacp, s->lacp);
2458     } else {
2459         lacp_unref(bundle->lacp);
2460         bundle->lacp = NULL;
2461     }
2462
2463     /* Update set of ports. */
2464     ok = true;
2465     for (i = 0; i < s->n_slaves; i++) {
2466         if (!bundle_add_port(bundle, s->slaves[i],
2467                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2468             ok = false;
2469         }
2470     }
2471     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2472         struct ofport_dpif *next_port;
2473
2474         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2475             for (i = 0; i < s->n_slaves; i++) {
2476                 if (s->slaves[i] == port->up.ofp_port) {
2477                     goto found;
2478                 }
2479             }
2480
2481             bundle_del_port(port);
2482         found: ;
2483         }
2484     }
2485     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2486
2487     if (list_is_empty(&bundle->ports)) {
2488         bundle_destroy(bundle);
2489         return EINVAL;
2490     }
2491
2492     /* Set VLAN tagging mode */
2493     if (s->vlan_mode != bundle->vlan_mode
2494         || s->use_priority_tags != bundle->use_priority_tags) {
2495         bundle->vlan_mode = s->vlan_mode;
2496         bundle->use_priority_tags = s->use_priority_tags;
2497         need_flush = true;
2498     }
2499
2500     /* Set VLAN tag. */
2501     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2502             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2503             : 0);
2504     if (vlan != bundle->vlan) {
2505         bundle->vlan = vlan;
2506         need_flush = true;
2507     }
2508
2509     /* Get trunked VLANs. */
2510     switch (s->vlan_mode) {
2511     case PORT_VLAN_ACCESS:
2512         trunks = NULL;
2513         break;
2514
2515     case PORT_VLAN_TRUNK:
2516         trunks = CONST_CAST(unsigned long *, s->trunks);
2517         break;
2518
2519     case PORT_VLAN_NATIVE_UNTAGGED:
2520     case PORT_VLAN_NATIVE_TAGGED:
2521         if (vlan != 0 && (!s->trunks
2522                           || !bitmap_is_set(s->trunks, vlan)
2523                           || bitmap_is_set(s->trunks, 0))) {
2524             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2525             if (s->trunks) {
2526                 trunks = bitmap_clone(s->trunks, 4096);
2527             } else {
2528                 trunks = bitmap_allocate1(4096);
2529             }
2530             bitmap_set1(trunks, vlan);
2531             bitmap_set0(trunks, 0);
2532         } else {
2533             trunks = CONST_CAST(unsigned long *, s->trunks);
2534         }
2535         break;
2536
2537     default:
2538         NOT_REACHED();
2539     }
2540     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2541         free(bundle->trunks);
2542         if (trunks == s->trunks) {
2543             bundle->trunks = vlan_bitmap_clone(trunks);
2544         } else {
2545             bundle->trunks = trunks;
2546             trunks = NULL;
2547         }
2548         need_flush = true;
2549     }
2550     if (trunks != s->trunks) {
2551         free(trunks);
2552     }
2553
2554     /* Bonding. */
2555     if (!list_is_short(&bundle->ports)) {
2556         bundle->ofproto->has_bonded_bundles = true;
2557         if (bundle->bond) {
2558             if (bond_reconfigure(bundle->bond, s->bond)) {
2559                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2560             }
2561         } else {
2562             bundle->bond = bond_create(s->bond);
2563             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2564         }
2565
2566         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2567             bond_slave_register(bundle->bond, port, port->up.netdev);
2568         }
2569     } else {
2570         bond_unref(bundle->bond);
2571         bundle->bond = NULL;
2572     }
2573
2574     /* If we changed something that would affect MAC learning, un-learn
2575      * everything on this port and force flow revalidation. */
2576     if (need_flush) {
2577         bundle_flush_macs(bundle, false);
2578     }
2579
2580     return 0;
2581 }
2582
2583 static void
2584 bundle_remove(struct ofport *port_)
2585 {
2586     struct ofport_dpif *port = ofport_dpif_cast(port_);
2587     struct ofbundle *bundle = port->bundle;
2588
2589     if (bundle) {
2590         bundle_del_port(port);
2591         if (list_is_empty(&bundle->ports)) {
2592             bundle_destroy(bundle);
2593         } else if (list_is_short(&bundle->ports)) {
2594             bond_unref(bundle->bond);
2595             bundle->bond = NULL;
2596         }
2597     }
2598 }
2599
2600 static void
2601 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2602 {
2603     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2604     struct ofport_dpif *port = port_;
2605     uint8_t ea[ETH_ADDR_LEN];
2606     int error;
2607
2608     error = netdev_get_etheraddr(port->up.netdev, ea);
2609     if (!error) {
2610         struct ofpbuf packet;
2611         void *packet_pdu;
2612
2613         ofpbuf_init(&packet, 0);
2614         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2615                                  pdu_size);
2616         memcpy(packet_pdu, pdu, pdu_size);
2617
2618         send_packet(port, &packet);
2619         ofpbuf_uninit(&packet);
2620     } else {
2621         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2622                     "%s (%s)", port->bundle->name,
2623                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2624     }
2625 }
2626
2627 static void
2628 bundle_send_learning_packets(struct ofbundle *bundle)
2629 {
2630     struct ofproto_dpif *ofproto = bundle->ofproto;
2631     struct ofpbuf *learning_packet;
2632     int error, n_packets, n_errors;
2633     struct mac_entry *e;
2634     struct list packets;
2635
2636     list_init(&packets);
2637     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2638     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2639         if (e->port.p != bundle) {
2640             void *port_void;
2641
2642             learning_packet = bond_compose_learning_packet(bundle->bond,
2643                                                            e->mac, e->vlan,
2644                                                            &port_void);
2645             learning_packet->private_p = port_void;
2646             list_push_back(&packets, &learning_packet->list_node);
2647         }
2648     }
2649     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2650
2651     error = n_packets = n_errors = 0;
2652     LIST_FOR_EACH (learning_packet, list_node, &packets) {
2653         int ret;
2654
2655         ret = send_packet(learning_packet->private_p, learning_packet);
2656         if (ret) {
2657             error = ret;
2658             n_errors++;
2659         }
2660         n_packets++;
2661     }
2662     ofpbuf_list_delete(&packets);
2663
2664     if (n_errors) {
2665         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2666         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2667                      "packets, last error was: %s",
2668                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2669     } else {
2670         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2671                  bundle->name, n_packets);
2672     }
2673 }
2674
2675 static void
2676 bundle_run(struct ofbundle *bundle)
2677 {
2678     if (bundle->lacp) {
2679         lacp_run(bundle->lacp, send_pdu_cb);
2680     }
2681     if (bundle->bond) {
2682         struct ofport_dpif *port;
2683
2684         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2685             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2686         }
2687
2688         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2689             bundle->ofproto->backer->need_revalidate = REV_BOND;
2690         }
2691
2692         if (bond_should_send_learning_packets(bundle->bond)) {
2693             bundle_send_learning_packets(bundle);
2694         }
2695     }
2696 }
2697
2698 static void
2699 bundle_wait(struct ofbundle *bundle)
2700 {
2701     if (bundle->lacp) {
2702         lacp_wait(bundle->lacp);
2703     }
2704     if (bundle->bond) {
2705         bond_wait(bundle->bond);
2706     }
2707 }
2708 \f
2709 /* Mirrors. */
2710
2711 static int
2712 mirror_set__(struct ofproto *ofproto_, void *aux,
2713              const struct ofproto_mirror_settings *s)
2714 {
2715     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2716     struct ofbundle **srcs, **dsts;
2717     int error;
2718     size_t i;
2719
2720     if (!s) {
2721         mirror_destroy(ofproto->mbridge, aux);
2722         return 0;
2723     }
2724
2725     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2726     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2727
2728     for (i = 0; i < s->n_srcs; i++) {
2729         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2730     }
2731
2732     for (i = 0; i < s->n_dsts; i++) {
2733         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2734     }
2735
2736     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2737                        s->n_dsts, s->src_vlans,
2738                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2739     free(srcs);
2740     free(dsts);
2741     return error;
2742 }
2743
2744 static int
2745 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2746                    uint64_t *packets, uint64_t *bytes)
2747 {
2748     push_all_stats();
2749     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2750                             bytes);
2751 }
2752
2753 static int
2754 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2755 {
2756     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2757     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2758     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2759         mac_learning_flush(ofproto->ml);
2760     }
2761     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2762     return 0;
2763 }
2764
2765 static bool
2766 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2767 {
2768     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2769     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2770     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2771 }
2772
2773 static void
2774 forward_bpdu_changed(struct ofproto *ofproto_)
2775 {
2776     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2777     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2778 }
2779
2780 static void
2781 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2782                      size_t max_entries)
2783 {
2784     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2785     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2786     mac_learning_set_idle_time(ofproto->ml, idle_time);
2787     mac_learning_set_max_entries(ofproto->ml, max_entries);
2788     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2789 }
2790 \f
2791 /* Ports. */
2792
2793 static struct ofport_dpif *
2794 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2795 {
2796     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2797     return ofport ? ofport_dpif_cast(ofport) : NULL;
2798 }
2799
2800 static struct ofport_dpif *
2801 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2802 {
2803     struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2804     return port && &ofproto->up == port->up.ofproto ? port : NULL;
2805 }
2806
2807 static void
2808 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2809                             struct ofproto_port *ofproto_port,
2810                             struct dpif_port *dpif_port)
2811 {
2812     ofproto_port->name = dpif_port->name;
2813     ofproto_port->type = dpif_port->type;
2814     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2815 }
2816
2817 static void
2818 ofport_update_peer(struct ofport_dpif *ofport)
2819 {
2820     const struct ofproto_dpif *ofproto;
2821     struct dpif_backer *backer;
2822     char *peer_name;
2823
2824     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2825         return;
2826     }
2827
2828     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2829     backer->need_revalidate = REV_RECONFIGURE;
2830
2831     if (ofport->peer) {
2832         ofport->peer->peer = NULL;
2833         ofport->peer = NULL;
2834     }
2835
2836     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2837     if (!peer_name) {
2838         return;
2839     }
2840
2841     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2842         struct ofport *peer_ofport;
2843         struct ofport_dpif *peer;
2844         char *peer_peer;
2845
2846         if (ofproto->backer != backer) {
2847             continue;
2848         }
2849
2850         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2851         if (!peer_ofport) {
2852             continue;
2853         }
2854
2855         peer = ofport_dpif_cast(peer_ofport);
2856         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2857         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2858                                  peer_peer)) {
2859             ofport->peer = peer;
2860             ofport->peer->peer = ofport;
2861         }
2862         free(peer_peer);
2863
2864         break;
2865     }
2866     free(peer_name);
2867 }
2868
2869 static void
2870 port_run_fast(struct ofport_dpif *ofport)
2871 {
2872     if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2873         struct ofpbuf packet;
2874
2875         ofpbuf_init(&packet, 0);
2876         cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2877         send_packet(ofport, &packet);
2878         ofpbuf_uninit(&packet);
2879     }
2880
2881     if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2882         struct ofpbuf packet;
2883
2884         ofpbuf_init(&packet, 0);
2885         bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2886         send_packet(ofport, &packet);
2887         ofpbuf_uninit(&packet);
2888     }
2889 }
2890
2891 static void
2892 port_run(struct ofport_dpif *ofport)
2893 {
2894     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2895     bool carrier_changed = carrier_seq != ofport->carrier_seq;
2896     bool enable = netdev_get_carrier(ofport->up.netdev);
2897     bool cfm_enable = false;
2898     bool bfd_enable = false;
2899
2900     ofport->carrier_seq = carrier_seq;
2901
2902     port_run_fast(ofport);
2903
2904     if (ofport->cfm) {
2905         int cfm_opup = cfm_get_opup(ofport->cfm);
2906
2907         cfm_run(ofport->cfm);
2908         cfm_enable = !cfm_get_fault(ofport->cfm);
2909
2910         if (cfm_opup >= 0) {
2911             cfm_enable = cfm_enable && cfm_opup;
2912         }
2913     }
2914
2915     if (ofport->bfd) {
2916         bfd_run(ofport->bfd);
2917         bfd_enable = bfd_forwarding(ofport->bfd);
2918     }
2919
2920     if (ofport->bfd || ofport->cfm) {
2921         enable = enable && (cfm_enable || bfd_enable);
2922     }
2923
2924     if (ofport->bundle) {
2925         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2926         if (carrier_changed) {
2927             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2928         }
2929     }
2930
2931     if (ofport->may_enable != enable) {
2932         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2933         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2934     }
2935
2936     ofport->may_enable = enable;
2937 }
2938
2939 static void
2940 port_wait(struct ofport_dpif *ofport)
2941 {
2942     if (ofport->cfm) {
2943         cfm_wait(ofport->cfm);
2944     }
2945
2946     if (ofport->bfd) {
2947         bfd_wait(ofport->bfd);
2948     }
2949 }
2950
2951 static int
2952 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2953                    struct ofproto_port *ofproto_port)
2954 {
2955     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2956     struct dpif_port dpif_port;
2957     int error;
2958
2959     if (sset_contains(&ofproto->ghost_ports, devname)) {
2960         const char *type = netdev_get_type_from_name(devname);
2961
2962         /* We may be called before ofproto->up.port_by_name is populated with
2963          * the appropriate ofport.  For this reason, we must get the name and
2964          * type from the netdev layer directly. */
2965         if (type) {
2966             const struct ofport *ofport;
2967
2968             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2969             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2970             ofproto_port->name = xstrdup(devname);
2971             ofproto_port->type = xstrdup(type);
2972             return 0;
2973         }
2974         return ENODEV;
2975     }
2976
2977     if (!sset_contains(&ofproto->ports, devname)) {
2978         return ENODEV;
2979     }
2980     error = dpif_port_query_by_name(ofproto->backer->dpif,
2981                                     devname, &dpif_port);
2982     if (!error) {
2983         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2984     }
2985     return error;
2986 }
2987
2988 static int
2989 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2990 {
2991     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2992     const char *devname = netdev_get_name(netdev);
2993     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2994     const char *dp_port_name;
2995
2996     if (netdev_vport_is_patch(netdev)) {
2997         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2998         return 0;
2999     }
3000
3001     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3002     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3003         odp_port_t port_no = ODPP_NONE;
3004         int error;
3005
3006         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3007         if (error) {
3008             return error;
3009         }
3010         if (netdev_get_tunnel_config(netdev)) {
3011             simap_put(&ofproto->backer->tnl_backers,
3012                       dp_port_name, odp_to_u32(port_no));
3013         }
3014     }
3015
3016     if (netdev_get_tunnel_config(netdev)) {
3017         sset_add(&ofproto->ghost_ports, devname);
3018     } else {
3019         sset_add(&ofproto->ports, devname);
3020     }
3021     return 0;
3022 }
3023
3024 static int
3025 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3026 {
3027     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3028     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3029     int error = 0;
3030
3031     if (!ofport) {
3032         return 0;
3033     }
3034
3035     sset_find_and_delete(&ofproto->ghost_ports,
3036                          netdev_get_name(ofport->up.netdev));
3037     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3038     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3039         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3040         if (!error) {
3041             /* The caller is going to close ofport->up.netdev.  If this is a
3042              * bonded port, then the bond is using that netdev, so remove it
3043              * from the bond.  The client will need to reconfigure everything
3044              * after deleting ports, so then the slave will get re-added. */
3045             bundle_remove(&ofport->up);
3046         }
3047     }
3048     return error;
3049 }
3050
3051 static int
3052 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3053 {
3054     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3055     int error;
3056
3057     push_all_stats();
3058
3059     error = netdev_get_stats(ofport->up.netdev, stats);
3060
3061     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3062         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3063
3064         /* ofproto->stats.tx_packets represents packets that we created
3065          * internally and sent to some port (e.g. packets sent with
3066          * send_packet()).  Account for them as if they had come from
3067          * OFPP_LOCAL and got forwarded. */
3068
3069         if (stats->rx_packets != UINT64_MAX) {
3070             stats->rx_packets += ofproto->stats.tx_packets;
3071         }
3072
3073         if (stats->rx_bytes != UINT64_MAX) {
3074             stats->rx_bytes += ofproto->stats.tx_bytes;
3075         }
3076
3077         /* ofproto->stats.rx_packets represents packets that were received on
3078          * some port and we processed internally and dropped (e.g. STP).
3079          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3080
3081         if (stats->tx_packets != UINT64_MAX) {
3082             stats->tx_packets += ofproto->stats.rx_packets;
3083         }
3084
3085         if (stats->tx_bytes != UINT64_MAX) {
3086             stats->tx_bytes += ofproto->stats.rx_bytes;
3087         }
3088     }
3089
3090     return error;
3091 }
3092
3093 struct port_dump_state {
3094     uint32_t bucket;
3095     uint32_t offset;
3096     bool ghost;
3097
3098     struct ofproto_port port;
3099     bool has_port;
3100 };
3101
3102 static int
3103 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3104 {
3105     *statep = xzalloc(sizeof(struct port_dump_state));
3106     return 0;
3107 }
3108
3109 static int
3110 port_dump_next(const struct ofproto *ofproto_, void *state_,
3111                struct ofproto_port *port)
3112 {
3113     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3114     struct port_dump_state *state = state_;
3115     const struct sset *sset;
3116     struct sset_node *node;
3117
3118     if (state->has_port) {
3119         ofproto_port_destroy(&state->port);
3120         state->has_port = false;
3121     }
3122     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3123     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3124         int error;
3125
3126         error = port_query_by_name(ofproto_, node->name, &state->port);
3127         if (!error) {
3128             *port = state->port;
3129             state->has_port = true;
3130             return 0;
3131         } else if (error != ENODEV) {
3132             return error;
3133         }
3134     }
3135
3136     if (!state->ghost) {
3137         state->ghost = true;
3138         state->bucket = 0;
3139         state->offset = 0;
3140         return port_dump_next(ofproto_, state_, port);
3141     }
3142
3143     return EOF;
3144 }
3145
3146 static int
3147 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3148 {
3149     struct port_dump_state *state = state_;
3150
3151     if (state->has_port) {
3152         ofproto_port_destroy(&state->port);
3153     }
3154     free(state);
3155     return 0;
3156 }
3157
3158 static int
3159 port_poll(const struct ofproto *ofproto_, char **devnamep)
3160 {
3161     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3162
3163     if (ofproto->port_poll_errno) {
3164         int error = ofproto->port_poll_errno;
3165         ofproto->port_poll_errno = 0;
3166         return error;
3167     }
3168
3169     if (sset_is_empty(&ofproto->port_poll_set)) {
3170         return EAGAIN;
3171     }
3172
3173     *devnamep = sset_pop(&ofproto->port_poll_set);
3174     return 0;
3175 }
3176
3177 static void
3178 port_poll_wait(const struct ofproto *ofproto_)
3179 {
3180     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3181     dpif_port_poll_wait(ofproto->backer->dpif);
3182 }
3183
3184 static int
3185 port_is_lacp_current(const struct ofport *ofport_)
3186 {
3187     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3188     return (ofport->bundle && ofport->bundle->lacp
3189             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3190             : -1);
3191 }
3192 \f
3193 /* Upcall handling. */
3194
3195 struct flow_miss_op {
3196     struct dpif_op dpif_op;
3197
3198     uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3199     struct xlate_out xout;
3200     bool xout_garbage;           /* 'xout' needs to be uninitialized? */
3201
3202     struct ofpbuf mask;          /* Flow mask for "put" ops. */
3203     struct odputil_keybuf maskbuf;
3204
3205     /* If this is a "put" op, then a pointer to the subfacet that should
3206      * be marked as uninstalled if the operation fails. */
3207     struct subfacet *subfacet;
3208 };
3209
3210 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3211  * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3212  * and (usually) installing a datapath flow.  The answer is usually "yes" (a
3213  * return value of true).  However, for short flows the cost of bookkeeping is
3214  * much higher than the benefits, so when the datapath holds a large number of
3215  * flows we impose some heuristics to decide which flows are likely to be worth
3216  * tracking. */
3217 static bool
3218 flow_miss_should_make_facet(struct flow_miss *miss)
3219 {
3220     struct dpif_backer *backer = miss->ofproto->backer;
3221     uint32_t hash;
3222
3223     switch (flow_miss_model) {
3224     case OFPROTO_HANDLE_MISS_AUTO:
3225         break;
3226     case OFPROTO_HANDLE_MISS_WITH_FACETS:
3227         return true;
3228     case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3229         return false;
3230     }
3231
3232     if (!backer->governor) {
3233         size_t n_subfacets;
3234
3235         n_subfacets = hmap_count(&backer->subfacets);
3236         if (n_subfacets * 2 <= flow_eviction_threshold) {
3237             return true;
3238         }
3239
3240         backer->governor = governor_create();
3241     }
3242
3243     hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3244     return governor_should_install_flow(backer->governor, hash,
3245                                         list_size(&miss->upcalls));
3246 }
3247
3248 /* Handles 'miss', which matches 'facet'.  May add any required datapath
3249  * operations to 'ops', incrementing '*n_ops' for each new op.
3250  *
3251  * All of the packets in 'miss' are considered to have arrived at time
3252  * 'miss->stats.used'.  This is really important only for new facets: if we
3253  * just called time_msec() here, then the new subfacet or its packets could
3254  * look (occasionally) as though it was used some time after the facet was
3255  * used.  That can make a one-packet flow look like it has a nonzero duration,
3256  * which looks odd in e.g. NetFlow statistics. */
3257 static void
3258 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3259                             struct flow_miss_op *ops, size_t *n_ops)
3260 {
3261     enum subfacet_path want_path;
3262     struct subfacet *subfacet;
3263
3264     facet->packet_count += miss->stats.n_packets;
3265     facet->prev_packet_count += miss->stats.n_packets;
3266     facet->byte_count += miss->stats.n_bytes;
3267     facet->prev_byte_count += miss->stats.n_bytes;
3268
3269     want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3270
3271     /* Don't install the flow if it's the result of the "userspace"
3272      * action for an already installed facet.  This can occur when a
3273      * datapath flow with wildcards has a "userspace" action and flows
3274      * sent to userspace result in a different subfacet, which will then
3275      * be rejected as overlapping by the datapath. */
3276     if (miss->upcall_type == DPIF_UC_ACTION
3277         && !list_is_empty(&facet->subfacets)) {
3278         return;
3279     }
3280
3281     subfacet = subfacet_create(facet, miss);
3282     if (subfacet->path != want_path) {
3283         struct flow_miss_op *op = &ops[(*n_ops)++];
3284         struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3285
3286         subfacet->path = want_path;
3287
3288         ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3289         if (enable_megaflows) {
3290             odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3291                                    &miss->flow, UINT32_MAX);
3292         }
3293
3294         op->xout_garbage = false;
3295         op->dpif_op.type = DPIF_OP_FLOW_PUT;
3296         op->subfacet = subfacet;
3297         put->flags = DPIF_FP_CREATE;
3298         put->key = miss->key;
3299         put->key_len = miss->key_len;
3300         put->mask = op->mask.data;
3301         put->mask_len = op->mask.size;
3302
3303         if (want_path == SF_FAST_PATH) {
3304             put->actions = facet->xout.odp_actions.data;
3305             put->actions_len = facet->xout.odp_actions.size;
3306         } else {
3307             compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3308                               op->slow_stub, sizeof op->slow_stub,
3309                               &put->actions, &put->actions_len);
3310         }
3311         put->stats = NULL;
3312     }
3313 }
3314
3315 /* Handles flow miss 'miss'.  May add any required datapath operations
3316  * to 'ops', incrementing '*n_ops' for each new op. */
3317 static void
3318 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3319                  size_t *n_ops)
3320 {
3321     struct facet *facet;
3322
3323     miss->ofproto->n_missed += list_size(&miss->upcalls);
3324
3325     facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3326     if (!facet) {
3327         /* There does not exist a bijection between 'struct flow' and datapath
3328          * flow keys with fitness ODP_FIT_TO_LITTLE.  This breaks a fundamental
3329          * assumption used throughout the facet and subfacet handling code.
3330          * Since we have to handle these misses in userspace anyway, we simply
3331          * skip facet creation, avoiding the problem altogether. */
3332         if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3333             || !flow_miss_should_make_facet(miss)) {
3334             return;
3335         }
3336
3337         facet = facet_create(miss);
3338     }
3339     handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3340 }
3341
3342 static struct drop_key *
3343 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3344                 size_t key_len)
3345 {
3346     struct drop_key *drop_key;
3347
3348     HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3349                              &backer->drop_keys) {
3350         if (drop_key->key_len == key_len
3351             && !memcmp(drop_key->key, key, key_len)) {
3352             return drop_key;
3353         }
3354     }
3355     return NULL;
3356 }
3357
3358 static void
3359 drop_key_clear(struct dpif_backer *backer)
3360 {
3361     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3362     struct drop_key *drop_key, *next;
3363
3364     HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3365         int error;
3366
3367         error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3368                               NULL);
3369         if (error && !VLOG_DROP_WARN(&rl)) {
3370             struct ds ds = DS_EMPTY_INITIALIZER;
3371             odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3372             VLOG_WARN("Failed to delete drop key (%s) (%s)",
3373                       ovs_strerror(error), ds_cstr(&ds));
3374             ds_destroy(&ds);
3375         }
3376
3377         hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3378         drop_key_destroy(drop_key);
3379     }
3380
3381     udpif_drop_key_clear(backer->udpif);
3382 }
3383
3384 static void
3385 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3386 {
3387     struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3388     struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3389     struct flow_miss *miss;
3390     size_t n_ops, i;
3391
3392     /* Process each element in the to-do list, constructing the set of
3393      * operations to batch. */
3394     n_ops = 0;
3395     HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3396         handle_flow_miss(miss, flow_miss_ops, &n_ops);
3397     }
3398     ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3399
3400     /* Execute batch. */
3401     for (i = 0; i < n_ops; i++) {
3402         dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3403     }
3404     dpif_operate(backer->dpif, dpif_ops, n_ops);
3405
3406     for (i = 0; i < n_ops; i++) {
3407         if (dpif_ops[i]->error != 0
3408             && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3409             && flow_miss_ops[i].subfacet) {
3410             struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3411
3412             COVERAGE_INC(subfacet_install_fail);
3413
3414             /* Zero-out subfacet counters when installation failed, but
3415              * datapath reported hits.  This should not happen and
3416              * indicates a bug, since if the datapath flow exists, we
3417              * should not be attempting to create a new subfacet.  A
3418              * buggy datapath could trigger this, so just zero out the
3419              * counters and log an error. */
3420             if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3421                 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3422                             "datapath reported hits");
3423                 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3424             }
3425
3426             subfacet->path = SF_NOT_INSTALLED;
3427         }
3428     }
3429 }
3430
3431 static void
3432 handle_sflow_upcall(struct dpif_backer *backer,
3433                     const struct dpif_upcall *upcall)
3434 {
3435     struct ofproto_dpif *ofproto;
3436     union user_action_cookie cookie;
3437     struct flow flow;
3438     odp_port_t odp_in_port;
3439
3440     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3441                       &flow, NULL, &ofproto, &odp_in_port)
3442         || !ofproto->sflow) {
3443         return;
3444     }
3445
3446     memset(&cookie, 0, sizeof cookie);
3447     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3448     dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3449                         odp_in_port, &cookie);
3450 }
3451
3452 static void
3453 handle_flow_sample_upcall(struct dpif_backer *backer,
3454                           const struct dpif_upcall *upcall)
3455 {
3456     struct ofproto_dpif *ofproto;
3457     union user_action_cookie cookie;
3458     struct flow flow;
3459
3460     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3461                       &flow, NULL, &ofproto, NULL)
3462         || !ofproto->ipfix) {
3463         return;
3464     }
3465
3466     memset(&cookie, 0, sizeof cookie);
3467     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3468
3469     /* The flow reflects exactly the contents of the packet.  Sample
3470      * the packet using it. */
3471     dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3472                            cookie.flow_sample.collector_set_id,
3473                            cookie.flow_sample.probability,
3474                            cookie.flow_sample.obs_domain_id,
3475                            cookie.flow_sample.obs_point_id);
3476 }
3477
3478 static void
3479 handle_ipfix_upcall(struct dpif_backer *backer,
3480                     const struct dpif_upcall *upcall)
3481 {
3482     struct ofproto_dpif *ofproto;
3483     struct flow flow;
3484
3485     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3486                       &flow, NULL, &ofproto, NULL)
3487         || !ofproto->ipfix) {
3488         return;
3489     }
3490
3491     /* The flow reflects exactly the contents of the packet.  Sample
3492      * the packet using it. */
3493     dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3494 }
3495
3496 static void
3497 handle_upcalls(struct dpif_backer *backer)
3498 {
3499     struct flow_miss_batch *fmb;
3500     int n_processed;
3501
3502     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3503         struct upcall *upcall = upcall_next(backer->udpif);
3504
3505         if (!upcall) {
3506             break;
3507         }
3508
3509         switch (upcall->type) {
3510         case SFLOW_UPCALL:
3511             handle_sflow_upcall(backer, &upcall->dpif_upcall);
3512             break;
3513
3514         case FLOW_SAMPLE_UPCALL:
3515             handle_flow_sample_upcall(backer, &upcall->dpif_upcall);
3516             break;
3517
3518         case IPFIX_UPCALL:
3519             handle_ipfix_upcall(backer, &upcall->dpif_upcall);
3520             break;
3521
3522         case BAD_UPCALL:
3523             break;
3524
3525         case MISS_UPCALL:
3526             NOT_REACHED();
3527         }
3528
3529         upcall_destroy(upcall);
3530     }
3531
3532     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3533         struct drop_key *drop_key = drop_key_next(backer->udpif);
3534         if (!drop_key) {
3535             break;
3536         }
3537
3538         if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3539             hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3540                         hash_bytes(drop_key->key, drop_key->key_len, 0));
3541             dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3542                           drop_key->key, drop_key->key_len,
3543                           NULL, 0, NULL, 0, NULL);
3544         } else {
3545             drop_key_destroy(drop_key);
3546         }
3547     }
3548
3549     fmb = flow_miss_batch_next(backer->udpif);
3550     if (fmb) {
3551         handle_flow_misses(backer, fmb);
3552         flow_miss_batch_destroy(fmb);
3553     }
3554 }
3555 \f
3556 /* Flow expiration. */
3557
3558 static int subfacet_max_idle(const struct dpif_backer *);
3559 static void update_stats(struct dpif_backer *);
3560 static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex);
3561 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3562
3563 /* This function is called periodically by run().  Its job is to collect
3564  * updates for the flows that have been installed into the datapath, most
3565  * importantly when they last were used, and then use that information to
3566  * expire flows that have not been used recently.
3567  *
3568  * Returns the number of milliseconds after which it should be called again. */
3569 static int
3570 expire(struct dpif_backer *backer)
3571 {
3572     struct ofproto_dpif *ofproto;
3573     size_t n_subfacets;
3574     int max_idle;
3575
3576     /* Periodically clear out the drop keys in an effort to keep them
3577      * relatively few. */
3578     drop_key_clear(backer);
3579
3580     /* Update stats for each flow in the backer. */
3581     update_stats(backer);
3582
3583     n_subfacets = hmap_count(&backer->subfacets);
3584     if (n_subfacets) {
3585         struct subfacet *subfacet;
3586         long long int total, now;
3587
3588         total = 0;
3589         now = time_msec();
3590         HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3591             total += now - subfacet->created;
3592         }
3593         backer->avg_subfacet_life += total / n_subfacets;
3594     }
3595     backer->avg_subfacet_life /= 2;
3596
3597     backer->avg_n_subfacet += n_subfacets;
3598     backer->avg_n_subfacet /= 2;
3599
3600     backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3601
3602     max_idle = subfacet_max_idle(backer);
3603     expire_subfacets(backer, max_idle);
3604
3605     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3606         struct rule *rule, *next_rule;
3607
3608         if (ofproto->backer != backer) {
3609             continue;
3610         }
3611
3612         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3613          * has passed. */
3614         ovs_mutex_lock(&ofproto_mutex);
3615         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3616                             &ofproto->up.expirable) {
3617             rule_expire(rule_dpif_cast(rule));
3618         }
3619         ovs_mutex_unlock(&ofproto_mutex);
3620
3621         /* All outstanding data in existing flows has been accounted, so it's a
3622          * good time to do bond rebalancing. */
3623         if (ofproto->has_bonded_bundles) {
3624             struct ofbundle *bundle;
3625
3626             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3627                 if (bundle->bond) {
3628                     bond_rebalance(bundle->bond);
3629                 }
3630             }
3631         }
3632     }
3633
3634     return MIN(max_idle, 1000);
3635 }
3636
3637 /* Updates flow table statistics given that the datapath just reported 'stats'
3638  * as 'subfacet''s statistics. */
3639 static void
3640 update_subfacet_stats(struct subfacet *subfacet,
3641                       const struct dpif_flow_stats *stats)
3642 {
3643     struct facet *facet = subfacet->facet;
3644     struct dpif_flow_stats diff;
3645
3646     diff.tcp_flags = stats->tcp_flags;
3647     diff.used = stats->used;
3648
3649     if (stats->n_packets >= subfacet->dp_packet_count) {
3650         diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3651     } else {
3652         VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3653         diff.n_packets = 0;
3654     }
3655
3656     if (stats->n_bytes >= subfacet->dp_byte_count) {
3657         diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3658     } else {
3659         VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3660         diff.n_bytes = 0;
3661     }
3662
3663     facet->ofproto->n_hit += diff.n_packets;
3664     subfacet->dp_packet_count = stats->n_packets;
3665     subfacet->dp_byte_count = stats->n_bytes;
3666     subfacet_update_stats(subfacet, &diff);
3667
3668     if (facet->accounted_bytes < facet->byte_count) {
3669         facet_learn(facet);
3670         facet_account(facet);
3671         facet->accounted_bytes = facet->byte_count;
3672     }
3673 }
3674
3675 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3676  * about, or a flow that shouldn't be installed but was anyway.  Delete it. */
3677 static void
3678 delete_unexpected_flow(struct dpif_backer *backer,
3679                        const struct nlattr *key, size_t key_len)
3680 {
3681     if (!VLOG_DROP_WARN(&rl)) {
3682         struct ds s;
3683
3684         ds_init(&s);
3685         odp_flow_key_format(key, key_len, &s);
3686         VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3687         ds_destroy(&s);
3688     }
3689
3690     COVERAGE_INC(facet_unexpected);
3691     dpif_flow_del(backer->dpif, key, key_len, NULL);
3692 }
3693
3694 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3695  *
3696  * This function also pushes statistics updates to rules which each facet
3697  * resubmits into.  Generally these statistics will be accurate.  However, if a
3698  * facet changes the rule it resubmits into at some time in between
3699  * update_stats() runs, it is possible that statistics accrued to the
3700  * old rule will be incorrectly attributed to the new rule.  This could be
3701  * avoided by calling update_stats() whenever rules are created or
3702  * deleted.  However, the performance impact of making so many calls to the
3703  * datapath do not justify the benefit of having perfectly accurate statistics.
3704  *
3705  * In addition, this function maintains per ofproto flow hit counts. The patch
3706  * port is not treated specially. e.g. A packet ingress from br0 patched into
3707  * br1 will increase the hit count of br0 by 1, however, does not affect
3708  * the hit or miss counts of br1.
3709  */
3710 static void
3711 update_stats(struct dpif_backer *backer)
3712 {
3713     const struct dpif_flow_stats *stats;
3714     struct dpif_flow_dump dump;
3715     const struct nlattr *key, *mask;
3716     size_t key_len, mask_len;
3717
3718     dpif_flow_dump_start(&dump, backer->dpif);
3719     while (dpif_flow_dump_next(&dump, &key, &key_len,
3720                                &mask, &mask_len, NULL, NULL, &stats)) {
3721         struct subfacet *subfacet;
3722         uint32_t key_hash;
3723
3724         key_hash = odp_flow_key_hash(key, key_len);
3725         subfacet = subfacet_find(backer, key, key_len, key_hash);
3726         switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3727         case SF_FAST_PATH:
3728             update_subfacet_stats(subfacet, stats);
3729             break;
3730
3731         case SF_SLOW_PATH:
3732             /* Stats are updated per-packet. */
3733             break;
3734
3735         case SF_NOT_INSTALLED:
3736         default:
3737             delete_unexpected_flow(backer, key, key_len);
3738             break;
3739         }
3740         run_fast_rl();
3741     }
3742     dpif_flow_dump_done(&dump);
3743
3744     update_moving_averages(backer);
3745 }
3746
3747 /* Calculates and returns the number of milliseconds of idle time after which
3748  * subfacets should expire from the datapath.  When a subfacet expires, we fold
3749  * its statistics into its facet, and when a facet's last subfacet expires, we
3750  * fold its statistic into its rule. */
3751 static int
3752 subfacet_max_idle(const struct dpif_backer *backer)
3753 {
3754     /*
3755      * Idle time histogram.
3756      *
3757      * Most of the time a switch has a relatively small number of subfacets.
3758      * When this is the case we might as well keep statistics for all of them
3759      * in userspace and to cache them in the kernel datapath for performance as
3760      * well.
3761      *
3762      * As the number of subfacets increases, the memory required to maintain
3763      * statistics about them in userspace and in the kernel becomes
3764      * significant.  However, with a large number of subfacets it is likely
3765      * that only a few of them are "heavy hitters" that consume a large amount
3766      * of bandwidth.  At this point, only heavy hitters are worth caching in
3767      * the kernel and maintaining in userspaces; other subfacets we can
3768      * discard.
3769      *
3770      * The technique used to compute the idle time is to build a histogram with
3771      * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each.  Each subfacet
3772      * that is installed in the kernel gets dropped in the appropriate bucket.
3773      * After the histogram has been built, we compute the cutoff so that only
3774      * the most-recently-used 1% of subfacets (but at least
3775      * flow_eviction_threshold flows) are kept cached.  At least
3776      * the most-recently-used bucket of subfacets is kept, so actually an
3777      * arbitrary number of subfacets can be kept in any given expiration run
3778      * (though the next run will delete most of those unless they receive
3779      * additional data).
3780      *
3781      * This requires a second pass through the subfacets, in addition to the
3782      * pass made by update_stats(), because the former function never looks at
3783      * uninstallable subfacets.
3784      */
3785     enum { BUCKET_WIDTH = 100 };
3786     enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3787     int buckets[N_BUCKETS] = { 0 };
3788     int total, subtotal, bucket;
3789     struct subfacet *subfacet;
3790     long long int now;
3791     int i;
3792
3793     total = hmap_count(&backer->subfacets);
3794     if (total <= flow_eviction_threshold) {
3795         return N_BUCKETS * BUCKET_WIDTH;
3796     }
3797
3798     /* Build histogram. */
3799     now = time_msec();
3800     HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3801         long long int idle = now - subfacet->used;
3802         int bucket = (idle <= 0 ? 0
3803                       : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3804                       : (unsigned int) idle / BUCKET_WIDTH);
3805         buckets[bucket]++;
3806     }
3807
3808     /* Find the first bucket whose flows should be expired. */
3809     subtotal = bucket = 0;
3810     do {
3811         subtotal += buckets[bucket++];
3812     } while (bucket < N_BUCKETS &&
3813              subtotal < MAX(flow_eviction_threshold, total / 100));
3814
3815     if (VLOG_IS_DBG_ENABLED()) {
3816         struct ds s;
3817
3818         ds_init(&s);
3819         ds_put_cstr(&s, "keep");
3820         for (i = 0; i < N_BUCKETS; i++) {
3821             if (i == bucket) {
3822                 ds_put_cstr(&s, ", drop");
3823             }
3824             if (buckets[i]) {
3825                 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3826             }
3827         }
3828         VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3829         ds_destroy(&s);
3830     }
3831
3832     return bucket * BUCKET_WIDTH;
3833 }
3834
3835 static void
3836 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3837 {
3838     /* Cutoff time for most flows. */
3839     long long int normal_cutoff = time_msec() - dp_max_idle;
3840
3841     /* We really want to keep flows for special protocols around, so use a more
3842      * conservative cutoff. */
3843     long long int special_cutoff = time_msec() - 10000;
3844
3845     struct subfacet *subfacet, *next_subfacet;
3846     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3847     int n_batch;
3848
3849     n_batch = 0;
3850     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3851                         &backer->subfacets) {
3852         long long int cutoff;
3853
3854         cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3855                                                 | SLOW_STP)
3856                   ? special_cutoff
3857                   : normal_cutoff);
3858         if (subfacet->used < cutoff) {
3859             if (subfacet->path != SF_NOT_INSTALLED) {
3860                 batch[n_batch++] = subfacet;
3861                 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3862                     subfacet_destroy_batch(backer, batch, n_batch);
3863                     n_batch = 0;
3864                 }
3865             } else {
3866                 subfacet_destroy(subfacet);
3867             }
3868         }
3869     }
3870
3871     if (n_batch > 0) {
3872         subfacet_destroy_batch(backer, batch, n_batch);
3873     }
3874 }
3875
3876 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3877  * then delete it entirely. */
3878 static void
3879 rule_expire(struct rule_dpif *rule)
3880     OVS_REQUIRES(ofproto_mutex)
3881 {
3882     uint16_t idle_timeout, hard_timeout;
3883     long long int now = time_msec();
3884     int reason;
3885
3886     ovs_assert(!rule->up.pending);
3887
3888     /* Has 'rule' expired? */
3889     ovs_mutex_lock(&rule->up.mutex);
3890     hard_timeout = rule->up.hard_timeout;
3891     idle_timeout = rule->up.idle_timeout;
3892     if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3893         reason = OFPRR_HARD_TIMEOUT;
3894     } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3895         reason = OFPRR_IDLE_TIMEOUT;
3896     } else {
3897         reason = -1;
3898     }
3899     ovs_mutex_unlock(&rule->up.mutex);
3900
3901     if (reason >= 0) {
3902         COVERAGE_INC(ofproto_dpif_expired);
3903         ofproto_rule_expire(&rule->up, reason);
3904     }
3905 }
3906 \f
3907 /* Facets. */
3908
3909 /* Creates and returns a new facet based on 'miss'.
3910  *
3911  * The caller must already have determined that no facet with an identical
3912  * 'miss->flow' exists in 'miss->ofproto'.
3913  *
3914  * 'rule' and 'xout' must have been created based on 'miss'.
3915  *
3916  * 'facet'' statistics are initialized based on 'stats'.
3917  *
3918  * The facet will initially have no subfacets.  The caller should create (at
3919  * least) one subfacet with subfacet_create(). */
3920 static struct facet *
3921 facet_create(const struct flow_miss *miss)
3922 {
3923     struct ofproto_dpif *ofproto = miss->ofproto;
3924     struct facet *facet;
3925     struct match match;
3926
3927     facet = xzalloc(sizeof *facet);
3928     facet->ofproto = miss->ofproto;
3929     facet->used = miss->stats.used;
3930     facet->flow = miss->flow;
3931     facet->learn_rl = time_msec() + 500;
3932
3933     list_init(&facet->subfacets);
3934     netflow_flow_init(&facet->nf_flow);
3935     netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3936
3937     xlate_out_copy(&facet->xout, &miss->xout);
3938
3939     match_init(&match, &facet->flow, &facet->xout.wc);
3940     cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3941     ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3942     classifier_insert(&ofproto->facets, &facet->cr);
3943     ovs_rwlock_unlock(&ofproto->facets.rwlock);
3944
3945     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
3946     return facet;
3947 }
3948
3949 static void
3950 facet_free(struct facet *facet)
3951 {
3952     if (facet) {
3953         xlate_out_uninit(&facet->xout);
3954         free(facet);
3955     }
3956 }
3957
3958 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3959  * 'packet', which arrived on 'in_port'. */
3960 static bool
3961 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3962                     const struct nlattr *odp_actions, size_t actions_len,
3963                     struct ofpbuf *packet)
3964 {
3965     struct odputil_keybuf keybuf;
3966     struct ofpbuf key;
3967     int error;
3968
3969     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3970     odp_flow_key_from_flow(&key, flow,
3971                            ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
3972
3973     error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
3974                          odp_actions, actions_len, packet);
3975     return !error;
3976 }
3977
3978 /* Remove 'facet' from its ofproto and free up the associated memory:
3979  *
3980  *   - If 'facet' was installed in the datapath, uninstalls it and updates its
3981  *     rule's statistics, via subfacet_uninstall().
3982  *
3983  *   - Removes 'facet' from its rule and from ofproto->facets.
3984  */
3985 static void
3986 facet_remove(struct facet *facet)
3987 {
3988     struct subfacet *subfacet, *next_subfacet;
3989
3990     ovs_assert(!list_is_empty(&facet->subfacets));
3991
3992     /* First uninstall all of the subfacets to get final statistics. */
3993     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3994         subfacet_uninstall(subfacet);
3995     }
3996
3997     /* Flush the final stats to the rule.
3998      *
3999      * This might require us to have at least one subfacet around so that we
4000      * can use its actions for accounting in facet_account(), which is why we
4001      * have uninstalled but not yet destroyed the subfacets. */
4002     facet_flush_stats(facet);
4003
4004     /* Now we're really all done so destroy everything. */
4005     LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4006                         &facet->subfacets) {
4007         subfacet_destroy__(subfacet);
4008     }
4009     ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4010     classifier_remove(&facet->ofproto->facets, &facet->cr);
4011     ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4012     cls_rule_destroy(&facet->cr);
4013     facet_free(facet);
4014 }
4015
4016 /* Feed information from 'facet' back into the learning table to keep it in
4017  * sync with what is actually flowing through the datapath. */
4018 static void
4019 facet_learn(struct facet *facet)
4020 {
4021     long long int now = time_msec();
4022
4023     if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4024         return;
4025     }
4026
4027     facet->learn_rl = now + 500;
4028
4029     if (!facet->xout.has_learn
4030         && !facet->xout.has_normal
4031         && (!facet->xout.has_fin_timeout
4032             || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4033         return;
4034     }
4035
4036     facet_push_stats(facet, true);
4037 }
4038
4039 static void
4040 facet_account(struct facet *facet)
4041 {
4042     const struct nlattr *a;
4043     unsigned int left;
4044     ovs_be16 vlan_tci;
4045     uint64_t n_bytes;
4046
4047     if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4048         return;
4049     }
4050     n_bytes = facet->byte_count - facet->accounted_bytes;
4051
4052     /* This loop feeds byte counters to bond_account() for rebalancing to use
4053      * as a basis.  We also need to track the actual VLAN on which the packet
4054      * is going to be sent to ensure that it matches the one passed to
4055      * bond_choose_output_slave().  (Otherwise, we will account to the wrong
4056      * hash bucket.)
4057      *
4058      * We use the actions from an arbitrary subfacet because they should all
4059      * be equally valid for our purpose. */
4060     vlan_tci = facet->flow.vlan_tci;
4061     NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4062                              facet->xout.odp_actions.size) {
4063         const struct ovs_action_push_vlan *vlan;
4064         struct ofport_dpif *port;
4065
4066         switch (nl_attr_type(a)) {
4067         case OVS_ACTION_ATTR_OUTPUT:
4068             port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4069             if (port && port->bundle && port->bundle->bond) {
4070                 bond_account(port->bundle->bond, &facet->flow,
4071                              vlan_tci_to_vid(vlan_tci), n_bytes);
4072             }
4073             break;
4074
4075         case OVS_ACTION_ATTR_POP_VLAN:
4076             vlan_tci = htons(0);
4077             break;
4078
4079         case OVS_ACTION_ATTR_PUSH_VLAN:
4080             vlan = nl_attr_get(a);
4081             vlan_tci = vlan->vlan_tci;
4082             break;
4083         }
4084     }
4085 }
4086
4087 /* Returns true if the only action for 'facet' is to send to the controller.
4088  * (We don't report NetFlow expiration messages for such facets because they
4089  * are just part of the control logic for the network, not real traffic). */
4090 static bool
4091 facet_is_controller_flow(struct facet *facet)
4092 {
4093     if (facet) {
4094         struct ofproto_dpif *ofproto = facet->ofproto;
4095         const struct ofpact *ofpacts;
4096         struct rule_actions *actions;
4097         struct rule_dpif *rule;
4098         size_t ofpacts_len;
4099         bool is_controller;
4100
4101         rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4102         actions = rule_dpif_get_actions(rule);
4103         rule_dpif_unref(rule);
4104
4105         ofpacts_len = actions->ofpacts_len;
4106         ofpacts = actions->ofpacts;
4107         is_controller = ofpacts_len > 0
4108             && ofpacts->type == OFPACT_CONTROLLER
4109             && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4110         rule_actions_unref(actions);
4111
4112         return is_controller;
4113     }
4114     return false;
4115 }
4116
4117 /* Folds all of 'facet''s statistics into its rule.  Also updates the
4118  * accounting ofhook and emits a NetFlow expiration if appropriate.  All of
4119  * 'facet''s statistics in the datapath should have been zeroed and folded into
4120  * its packet and byte counts before this function is called. */
4121 static void
4122 facet_flush_stats(struct facet *facet)
4123 {
4124     struct ofproto_dpif *ofproto = facet->ofproto;
4125     struct subfacet *subfacet;
4126
4127     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4128         ovs_assert(!subfacet->dp_byte_count);
4129         ovs_assert(!subfacet->dp_packet_count);
4130     }
4131
4132     facet_push_stats(facet, false);
4133     if (facet->accounted_bytes < facet->byte_count) {
4134         facet_account(facet);
4135         facet->accounted_bytes = facet->byte_count;
4136     }
4137
4138     if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4139         struct ofexpired expired;
4140         expired.flow = facet->flow;
4141         expired.packet_count = facet->packet_count;
4142         expired.byte_count = facet->byte_count;
4143         expired.used = facet->used;
4144         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4145     }
4146
4147     /* Reset counters to prevent double counting if 'facet' ever gets
4148      * reinstalled. */
4149     facet_reset_counters(facet);
4150
4151     netflow_flow_clear(&facet->nf_flow);
4152     facet->tcp_flags = 0;
4153 }
4154
4155 /* Searches 'ofproto''s table of facets for one which would be responsible for
4156  * 'flow'.  Returns it if found, otherwise a null pointer.
4157  *
4158  * The returned facet might need revalidation; use facet_lookup_valid()
4159  * instead if that is important. */
4160 static struct facet *
4161 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4162 {
4163     struct cls_rule *cr;
4164
4165     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4166     cr = classifier_lookup(&ofproto->facets, flow, NULL);
4167     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4168     return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4169 }
4170
4171 /* Searches 'ofproto''s table of facets for one capable that covers
4172  * 'flow'.  Returns it if found, otherwise a null pointer.
4173  *
4174  * The returned facet is guaranteed to be valid. */
4175 static struct facet *
4176 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4177 {
4178     struct facet *facet;
4179
4180     facet = facet_find(ofproto, flow);
4181     if (facet
4182         && ofproto->backer->need_revalidate
4183         && !facet_revalidate(facet)) {
4184         return NULL;
4185     }
4186
4187     return facet;
4188 }
4189
4190 static bool
4191 facet_check_consistency(struct facet *facet)
4192 {
4193     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4194
4195     struct xlate_out xout;
4196     struct xlate_in xin;
4197
4198     struct rule_dpif *rule;
4199     bool ok;
4200
4201     /* Check the datapath actions for consistency. */
4202     rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4203     xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4204     xlate_actions(&xin, &xout);
4205     rule_dpif_unref(rule);
4206
4207     ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4208         && facet->xout.slow == xout.slow;
4209     if (!ok && !VLOG_DROP_WARN(&rl)) {
4210         struct ds s = DS_EMPTY_INITIALIZER;
4211
4212         flow_format(&s, &facet->flow);
4213         ds_put_cstr(&s, ": inconsistency in facet");
4214
4215         if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4216             ds_put_cstr(&s, " (actions were: ");
4217             format_odp_actions(&s, facet->xout.odp_actions.data,
4218                                facet->xout.odp_actions.size);
4219             ds_put_cstr(&s, ") (correct actions: ");
4220             format_odp_actions(&s, xout.odp_actions.data,
4221                                xout.odp_actions.size);
4222             ds_put_char(&s, ')');
4223         }
4224
4225         if (facet->xout.slow != xout.slow) {
4226             ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4227         }
4228
4229         ds_destroy(&s);
4230     }
4231     xlate_out_uninit(&xout);
4232
4233     return ok;
4234 }
4235
4236 /* Re-searches the classifier for 'facet':
4237  *
4238  *   - If the rule found is different from 'facet''s current rule, moves
4239  *     'facet' to the new rule and recompiles its actions.
4240  *
4241  *   - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4242  *     where it is and recompiles its actions anyway.
4243  *
4244  *   - If any of 'facet''s subfacets correspond to a new flow according to
4245  *     xlate_receive(), 'facet' is removed.
4246  *
4247  *   Returns true if 'facet' is still valid.  False if 'facet' was removed. */
4248 static bool
4249 facet_revalidate(struct facet *facet)
4250 {
4251     struct ofproto_dpif *ofproto = facet->ofproto;
4252     struct rule_dpif *new_rule;
4253     struct subfacet *subfacet;
4254     struct flow_wildcards wc;
4255     struct xlate_out xout;
4256     struct xlate_in xin;
4257
4258     COVERAGE_INC(facet_revalidate);
4259
4260     /* Check that child subfacets still correspond to this facet.  Tunnel
4261      * configuration changes could cause a subfacet's OpenFlow in_port to
4262      * change. */
4263     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4264         struct ofproto_dpif *recv_ofproto;
4265         struct flow recv_flow;
4266         int error;
4267
4268         error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4269                               subfacet->key_len, &recv_flow, NULL,
4270                               &recv_ofproto, NULL);
4271         if (error
4272             || recv_ofproto != ofproto
4273             || facet != facet_find(ofproto, &recv_flow)) {
4274             facet_remove(facet);
4275             return false;
4276         }
4277     }
4278
4279     flow_wildcards_init_catchall(&wc);
4280     rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4281
4282     /* Calculate new datapath actions.
4283      *
4284      * We do not modify any 'facet' state yet, because we might need to, e.g.,
4285      * emit a NetFlow expiration and, if so, we need to have the old state
4286      * around to properly compose it. */
4287     xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4288     xlate_actions(&xin, &xout);
4289     flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4290
4291     /* A facet's slow path reason should only change under dramatic
4292      * circumstances.  Rather than try to update everything, it's simpler to
4293      * remove the facet and start over.
4294      *
4295      * More importantly, if a facet's wildcards change, it will be relatively
4296      * difficult to figure out if its subfacets still belong to it, and if not
4297      * which facet they may belong to.  Again, to avoid the complexity, we
4298      * simply give up instead. */
4299     if (facet->xout.slow != xout.slow
4300         || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4301         facet_remove(facet);
4302         xlate_out_uninit(&xout);
4303         rule_dpif_unref(new_rule);
4304         return false;
4305     }
4306
4307     if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4308         LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4309             if (subfacet->path == SF_FAST_PATH) {
4310                 struct dpif_flow_stats stats;
4311
4312                 subfacet_install(subfacet, &xout.odp_actions, &stats);
4313                 subfacet_update_stats(subfacet, &stats);
4314             }
4315         }
4316
4317         facet_flush_stats(facet);
4318
4319         ofpbuf_clear(&facet->xout.odp_actions);
4320         ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4321                    xout.odp_actions.size);
4322     }
4323
4324     /* Update 'facet' now that we've taken care of all the old state. */
4325     facet->xout.slow = xout.slow;
4326     facet->xout.has_learn = xout.has_learn;
4327     facet->xout.has_normal = xout.has_normal;
4328     facet->xout.has_fin_timeout = xout.has_fin_timeout;
4329     facet->xout.nf_output_iface = xout.nf_output_iface;
4330     facet->xout.mirrors = xout.mirrors;
4331     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4332
4333     ovs_mutex_lock(&new_rule->up.mutex);
4334     facet->used = MAX(facet->used, new_rule->up.created);
4335     ovs_mutex_unlock(&new_rule->up.mutex);
4336
4337     xlate_out_uninit(&xout);
4338     rule_dpif_unref(new_rule);
4339     return true;
4340 }
4341
4342 static void
4343 facet_reset_counters(struct facet *facet)
4344 {
4345     facet->packet_count = 0;
4346     facet->byte_count = 0;
4347     facet->prev_packet_count = 0;
4348     facet->prev_byte_count = 0;
4349     facet->accounted_bytes = 0;
4350 }
4351
4352 static void
4353 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4354                 struct dpif_flow_stats *stats, bool may_learn)
4355 {
4356     struct ofport_dpif *in_port;
4357     struct rule_dpif *rule;
4358     struct xlate_in xin;
4359
4360     in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4361     if (in_port && in_port->is_tunnel) {
4362         netdev_vport_inc_rx(in_port->up.netdev, stats);
4363     }
4364
4365     rule_dpif_lookup(ofproto, flow, NULL, &rule);
4366     rule_dpif_credit_stats(rule, stats);
4367     xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4368     xin.resubmit_stats = stats;
4369     xin.may_learn = may_learn;
4370     xlate_actions_for_side_effects(&xin);
4371     rule_dpif_unref(rule);
4372 }
4373
4374 static void
4375 facet_push_stats(struct facet *facet, bool may_learn)
4376 {
4377     struct dpif_flow_stats stats;
4378
4379     ovs_assert(facet->packet_count >= facet->prev_packet_count);
4380     ovs_assert(facet->byte_count >= facet->prev_byte_count);
4381     ovs_assert(facet->used >= facet->prev_used);
4382
4383     stats.n_packets = facet->packet_count - facet->prev_packet_count;
4384     stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4385     stats.used = facet->used;
4386     stats.tcp_flags = facet->tcp_flags;
4387
4388     if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4389         facet->prev_packet_count = facet->packet_count;
4390         facet->prev_byte_count = facet->byte_count;
4391         facet->prev_used = facet->used;
4392
4393         netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4394                                  facet->used);
4395         netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4396         mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4397                             stats.n_packets, stats.n_bytes);
4398         flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4399     }
4400 }
4401
4402 static void
4403 push_all_stats__(bool run_fast)
4404 {
4405     static long long int rl = LLONG_MIN;
4406     struct ofproto_dpif *ofproto;
4407
4408     if (time_msec() < rl) {
4409         return;
4410     }
4411
4412     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4413         struct cls_cursor cursor;
4414         struct facet *facet;
4415
4416         ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4417         cls_cursor_init(&cursor, &ofproto->facets, NULL);
4418         CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4419             facet_push_stats(facet, false);
4420             if (run_fast) {
4421                 run_fast_rl();
4422             }
4423         }
4424         ovs_rwlock_unlock(&ofproto->facets.rwlock);
4425     }
4426
4427     rl = time_msec() + 100;
4428 }
4429
4430 static void
4431 push_all_stats(void)
4432 {
4433     push_all_stats__(true);
4434 }
4435
4436 void
4437 rule_dpif_credit_stats(struct rule_dpif *rule,
4438                        const struct dpif_flow_stats *stats)
4439 {
4440     ovs_mutex_lock(&rule->stats_mutex);
4441     rule->packet_count += stats->n_packets;
4442     rule->byte_count += stats->n_bytes;
4443     rule->up.used = MAX(rule->up.used, stats->used);
4444     ovs_mutex_unlock(&rule->stats_mutex);
4445 }
4446
4447 bool
4448 rule_dpif_fail_open(const struct rule_dpif *rule)
4449 {
4450     return rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4451 }
4452
4453 ovs_be64
4454 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4455     OVS_REQUIRES(rule->up.mutex)
4456 {
4457     return rule->up.flow_cookie;
4458 }
4459
4460 void
4461 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4462                      uint16_t hard_timeout)
4463 {
4464     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4465 }
4466
4467 /* Returns 'rule''s actions.  The caller owns a reference on the returned
4468  * actions and must eventually release it (with rule_actions_unref()) to avoid
4469  * a memory leak. */
4470 struct rule_actions *
4471 rule_dpif_get_actions(const struct rule_dpif *rule)
4472 {
4473     return rule_get_actions(&rule->up);
4474 }
4475 \f
4476 /* Subfacets. */
4477
4478 static struct subfacet *
4479 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4480               size_t key_len, uint32_t key_hash)
4481 {
4482     struct subfacet *subfacet;
4483
4484     HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4485                              &backer->subfacets) {
4486         if (subfacet->key_len == key_len
4487             && !memcmp(key, subfacet->key, key_len)) {
4488             return subfacet;
4489         }
4490     }
4491
4492     return NULL;
4493 }
4494
4495 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4496  * 'key_fitness', 'key', and 'key_len' members in 'miss'.  Returns the
4497  * existing subfacet if there is one, otherwise creates and returns a
4498  * new subfacet. */
4499 static struct subfacet *
4500 subfacet_create(struct facet *facet, struct flow_miss *miss)
4501 {
4502     struct dpif_backer *backer = miss->ofproto->backer;
4503     enum odp_key_fitness key_fitness = miss->key_fitness;
4504     const struct nlattr *key = miss->key;
4505     size_t key_len = miss->key_len;
4506     uint32_t key_hash;
4507     struct subfacet *subfacet;
4508
4509     key_hash = odp_flow_key_hash(key, key_len);
4510
4511     if (list_is_empty(&facet->subfacets)) {
4512         subfacet = &facet->one_subfacet;
4513     } else {
4514         subfacet = subfacet_find(backer, key, key_len, key_hash);
4515         if (subfacet) {
4516             if (subfacet->facet == facet) {
4517                 return subfacet;
4518             }
4519
4520             /* This shouldn't happen. */
4521             VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4522             subfacet_destroy(subfacet);
4523         }
4524
4525         subfacet = xmalloc(sizeof *subfacet);
4526     }
4527
4528     hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4529     list_push_back(&facet->subfacets, &subfacet->list_node);
4530     subfacet->facet = facet;
4531     subfacet->key_fitness = key_fitness;
4532     subfacet->key = xmemdup(key, key_len);
4533     subfacet->key_len = key_len;
4534     subfacet->used = miss->stats.used;
4535     subfacet->created = subfacet->used;
4536     subfacet->dp_packet_count = 0;
4537     subfacet->dp_byte_count = 0;
4538     subfacet->path = SF_NOT_INSTALLED;
4539     subfacet->backer = backer;
4540
4541     backer->subfacet_add_count++;
4542     return subfacet;
4543 }
4544
4545 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4546  * its facet within 'ofproto', and frees it. */
4547 static void
4548 subfacet_destroy__(struct subfacet *subfacet)
4549 {
4550     struct facet *facet = subfacet->facet;
4551     struct ofproto_dpif *ofproto = facet->ofproto;
4552
4553     /* Update ofproto stats before uninstall the subfacet. */
4554     ofproto->backer->subfacet_del_count++;
4555
4556     subfacet_uninstall(subfacet);
4557     hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4558     list_remove(&subfacet->list_node);
4559     free(subfacet->key);
4560     if (subfacet != &facet->one_subfacet) {
4561         free(subfacet);
4562     }
4563 }
4564
4565 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4566  * last remaining subfacet in its facet destroys the facet too. */
4567 static void
4568 subfacet_destroy(struct subfacet *subfacet)
4569 {
4570     struct facet *facet = subfacet->facet;
4571
4572     if (list_is_singleton(&facet->subfacets)) {
4573         /* facet_remove() needs at least one subfacet (it will remove it). */
4574         facet_remove(facet);
4575     } else {
4576         subfacet_destroy__(subfacet);
4577     }
4578 }
4579
4580 static void
4581 subfacet_destroy_batch(struct dpif_backer *backer,
4582                        struct subfacet **subfacets, int n)
4583 {
4584     struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4585     struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4586     struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4587     int i;
4588
4589     for (i = 0; i < n; i++) {
4590         ops[i].type = DPIF_OP_FLOW_DEL;
4591         ops[i].u.flow_del.key = subfacets[i]->key;
4592         ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4593         ops[i].u.flow_del.stats = &stats[i];
4594         opsp[i] = &ops[i];
4595     }
4596
4597     dpif_operate(backer->dpif, opsp, n);
4598     for (i = 0; i < n; i++) {
4599         subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4600         subfacets[i]->path = SF_NOT_INSTALLED;
4601         subfacet_destroy(subfacets[i]);
4602         run_fast_rl();
4603     }
4604 }
4605
4606 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4607  * bytes of actions in 'actions'.  If 'stats' is non-null, statistics counters
4608  * in the datapath will be zeroed and 'stats' will be updated with traffic new
4609  * since 'subfacet' was last updated.
4610  *
4611  * Returns 0 if successful, otherwise a positive errno value. */
4612 static int
4613 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4614                  struct dpif_flow_stats *stats)
4615 {
4616     struct facet *facet = subfacet->facet;
4617     enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4618     const struct nlattr *actions = odp_actions->data;
4619     size_t actions_len = odp_actions->size;
4620     struct odputil_keybuf maskbuf;
4621     struct ofpbuf mask;
4622
4623     uint64_t slow_path_stub[128 / 8];
4624     enum dpif_flow_put_flags flags;
4625     int ret;
4626
4627     flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4628                                                : DPIF_FP_MODIFY;
4629     if (stats) {
4630         flags |= DPIF_FP_ZERO_STATS;
4631     }
4632
4633     if (path == SF_SLOW_PATH) {
4634         compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4635                           slow_path_stub, sizeof slow_path_stub,
4636                           &actions, &actions_len);
4637     }
4638
4639     ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4640     if (enable_megaflows) {
4641         odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4642                                &facet->flow, UINT32_MAX);
4643     }
4644
4645     ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4646                         subfacet->key_len,  mask.data, mask.size,
4647                         actions, actions_len, stats);
4648
4649     if (stats) {
4650         subfacet_reset_dp_stats(subfacet, stats);
4651     }
4652
4653     if (ret) {
4654         COVERAGE_INC(subfacet_install_fail);
4655     } else {
4656         subfacet->path = path;
4657     }
4658     return ret;
4659 }
4660
4661 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4662 static void
4663 subfacet_uninstall(struct subfacet *subfacet)
4664 {
4665     if (subfacet->path != SF_NOT_INSTALLED) {
4666         struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4667         struct dpif_flow_stats stats;
4668         int error;
4669
4670         error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4671                               subfacet->key_len, &stats);
4672         subfacet_reset_dp_stats(subfacet, &stats);
4673         if (!error) {
4674             subfacet_update_stats(subfacet, &stats);
4675         }
4676         subfacet->path = SF_NOT_INSTALLED;
4677     } else {
4678         ovs_assert(subfacet->dp_packet_count == 0);
4679         ovs_assert(subfacet->dp_byte_count == 0);
4680     }
4681 }
4682
4683 /* Resets 'subfacet''s datapath statistics counters.  This should be called
4684  * when 'subfacet''s statistics are cleared in the datapath.  If 'stats' is
4685  * non-null, it should contain the statistics returned by dpif when 'subfacet'
4686  * was reset in the datapath.  'stats' will be modified to include only
4687  * statistics new since 'subfacet' was last updated. */
4688 static void
4689 subfacet_reset_dp_stats(struct subfacet *subfacet,
4690                         struct dpif_flow_stats *stats)
4691 {
4692     if (stats
4693         && subfacet->dp_packet_count <= stats->n_packets
4694         && subfacet->dp_byte_count <= stats->n_bytes) {
4695         stats->n_packets -= subfacet->dp_packet_count;
4696         stats->n_bytes -= subfacet->dp_byte_count;
4697     }
4698
4699     subfacet->dp_packet_count = 0;
4700     subfacet->dp_byte_count = 0;
4701 }
4702
4703 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4704  *
4705  * Because of the meaning of a subfacet's counters, it only makes sense to do
4706  * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4707  * represents a packet that was sent by hand or if it represents statistics
4708  * that have been cleared out of the datapath. */
4709 static void
4710 subfacet_update_stats(struct subfacet *subfacet,
4711                       const struct dpif_flow_stats *stats)
4712 {
4713     if (stats->n_packets || stats->used > subfacet->used) {
4714         struct facet *facet = subfacet->facet;
4715
4716         subfacet->used = MAX(subfacet->used, stats->used);
4717         facet->used = MAX(facet->used, stats->used);
4718         facet->packet_count += stats->n_packets;
4719         facet->byte_count += stats->n_bytes;
4720         facet->tcp_flags |= stats->tcp_flags;
4721     }
4722 }
4723 \f
4724 /* Rules. */
4725
4726 /* Lookup 'flow' in 'ofproto''s classifier.  If 'wc' is non-null, sets
4727  * the fields that were relevant as part of the lookup. */
4728 void
4729 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4730                  struct flow_wildcards *wc, struct rule_dpif **rule)
4731 {
4732     struct ofport_dpif *port;
4733
4734     if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4735         return;
4736     }
4737     port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4738     if (!port) {
4739         VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4740                      flow->in_port.ofp_port);
4741     }
4742
4743     choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4744                      ofproto->no_packet_in_rule, rule);
4745 }
4746
4747 bool
4748 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4749                           const struct flow *flow, struct flow_wildcards *wc,
4750                           uint8_t table_id, struct rule_dpif **rule)
4751 {
4752     const struct cls_rule *cls_rule;
4753     struct classifier *cls;
4754     bool frag;
4755
4756     *rule = NULL;
4757     if (table_id >= N_TABLES) {
4758         return false;
4759     }
4760
4761     if (wc) {
4762         memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4763         wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4764     }
4765
4766     cls = &ofproto->up.tables[table_id].cls;
4767     ovs_rwlock_rdlock(&cls->rwlock);
4768     frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4769     if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4770         /* We must pretend that transport ports are unavailable. */
4771         struct flow ofpc_normal_flow = *flow;
4772         ofpc_normal_flow.tp_src = htons(0);
4773         ofpc_normal_flow.tp_dst = htons(0);
4774         cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4775     } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4776         cls_rule = &ofproto->drop_frags_rule->up.cr;
4777         if (wc) {
4778             flow_wildcards_init_exact(wc);
4779         }
4780     } else {
4781         cls_rule = classifier_lookup(cls, flow, wc);
4782     }
4783
4784     *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4785     rule_dpif_ref(*rule);
4786     ovs_rwlock_unlock(&cls->rwlock);
4787
4788     return *rule != NULL;
4789 }
4790
4791 /* Given a port configuration (specified as zero if there's no port), chooses
4792  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4793  * flow table miss. */
4794 void
4795 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4796                  struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4797 {
4798     *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4799     rule_dpif_ref(*rule);
4800 }
4801
4802 void
4803 rule_dpif_ref(struct rule_dpif *rule)
4804 {
4805     if (rule) {
4806         ofproto_rule_ref(&rule->up);
4807     }
4808 }
4809
4810 void
4811 rule_dpif_unref(struct rule_dpif *rule)
4812 {
4813     if (rule) {
4814         ofproto_rule_unref(&rule->up);
4815     }
4816 }
4817
4818 static void
4819 complete_operation(struct rule_dpif *rule)
4820     OVS_REQUIRES(ofproto_mutex)
4821 {
4822     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4823
4824     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4825     ofoperation_complete(rule->up.pending, 0);
4826 }
4827
4828 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4829 {
4830     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4831 }
4832
4833 static struct rule *
4834 rule_alloc(void)
4835 {
4836     struct rule_dpif *rule = xmalloc(sizeof *rule);
4837     return &rule->up;
4838 }
4839
4840 static void
4841 rule_dealloc(struct rule *rule_)
4842 {
4843     struct rule_dpif *rule = rule_dpif_cast(rule_);
4844     free(rule);
4845 }
4846
4847 static enum ofperr
4848 rule_construct(struct rule *rule_)
4849 {
4850     struct rule_dpif *rule = rule_dpif_cast(rule_);
4851     ovs_mutex_init(&rule->stats_mutex);
4852     ovs_mutex_lock(&rule->stats_mutex);
4853     rule->packet_count = 0;
4854     rule->byte_count = 0;
4855     ovs_mutex_unlock(&rule->stats_mutex);
4856     return 0;
4857 }
4858
4859 static void
4860 rule_insert(struct rule *rule_)
4861     OVS_REQUIRES(ofproto_mutex)
4862 {
4863     struct rule_dpif *rule = rule_dpif_cast(rule_);
4864     complete_operation(rule);
4865 }
4866
4867 static void
4868 rule_delete(struct rule *rule_)
4869     OVS_REQUIRES(ofproto_mutex)
4870 {
4871     struct rule_dpif *rule = rule_dpif_cast(rule_);
4872     complete_operation(rule);
4873 }
4874
4875 static void
4876 rule_destruct(struct rule *rule_)
4877 {
4878     struct rule_dpif *rule = rule_dpif_cast(rule_);
4879     ovs_mutex_destroy(&rule->stats_mutex);
4880 }
4881
4882 static void
4883 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4884 {
4885     struct rule_dpif *rule = rule_dpif_cast(rule_);
4886
4887     /* push_all_stats() can handle flow misses which, when using the learn
4888      * action, can cause rules to be added and deleted.  This can corrupt our
4889      * caller's datastructures which assume that rule_get_stats() doesn't have
4890      * an impact on the flow table. To be safe, we disable miss handling. */
4891     push_all_stats__(false);
4892
4893     /* Start from historical data for 'rule' itself that are no longer tracked
4894      * in facets.  This counts, for example, facets that have expired. */
4895     ovs_mutex_lock(&rule->stats_mutex);
4896     *packets = rule->packet_count;
4897     *bytes = rule->byte_count;
4898     ovs_mutex_unlock(&rule->stats_mutex);
4899 }
4900
4901 static void
4902 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4903                   struct ofpbuf *packet)
4904 {
4905     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4906     struct dpif_flow_stats stats;
4907     struct xlate_out xout;
4908     struct xlate_in xin;
4909
4910     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4911     rule_dpif_credit_stats(rule, &stats);
4912
4913     xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4914     xin.resubmit_stats = &stats;
4915     xlate_actions(&xin, &xout);
4916
4917     execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4918                         xout.odp_actions.size, packet);
4919
4920     xlate_out_uninit(&xout);
4921 }
4922
4923 static enum ofperr
4924 rule_execute(struct rule *rule, const struct flow *flow,
4925              struct ofpbuf *packet)
4926 {
4927     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4928     ofpbuf_delete(packet);
4929     return 0;
4930 }
4931
4932 static void
4933 rule_modify_actions(struct rule *rule_, bool reset_counters)
4934     OVS_REQUIRES(ofproto_mutex)
4935 {
4936     struct rule_dpif *rule = rule_dpif_cast(rule_);
4937
4938     if (reset_counters) {
4939         ovs_mutex_lock(&rule->stats_mutex);
4940         rule->packet_count = 0;
4941         rule->byte_count = 0;
4942         ovs_mutex_unlock(&rule->stats_mutex);
4943     }
4944
4945     complete_operation(rule);
4946 }
4947 \f
4948 /* Sends 'packet' out 'ofport'.
4949  * May modify 'packet'.
4950  * Returns 0 if successful, otherwise a positive errno value. */
4951 static int
4952 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4953 {
4954     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4955     uint64_t odp_actions_stub[1024 / 8];
4956     struct ofpbuf key, odp_actions;
4957     struct dpif_flow_stats stats;
4958     struct odputil_keybuf keybuf;
4959     struct ofpact_output output;
4960     struct xlate_out xout;
4961     struct xlate_in xin;
4962     struct flow flow;
4963     union flow_in_port in_port_;
4964     int error;
4965
4966     ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4967     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4968
4969     /* Use OFPP_NONE as the in_port to avoid special packet processing. */
4970     in_port_.ofp_port = OFPP_NONE;
4971     flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
4972     odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
4973                                                              OFPP_LOCAL));
4974     dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
4975
4976     ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
4977     output.port = ofport->up.ofp_port;
4978     output.max_len = 0;
4979
4980     xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
4981     xin.ofpacts_len = sizeof output;
4982     xin.ofpacts = &output.ofpact;
4983     xin.resubmit_stats = &stats;
4984     xlate_actions(&xin, &xout);
4985
4986     error = dpif_execute(ofproto->backer->dpif,
4987                          key.data, key.size,
4988                          xout.odp_actions.data, xout.odp_actions.size,
4989                          packet);
4990     xlate_out_uninit(&xout);
4991
4992     if (error) {
4993         VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
4994                      ofproto->up.name, netdev_get_name(ofport->up.netdev),
4995                      ovs_strerror(error));
4996     }
4997
4998     ofproto->stats.tx_packets++;
4999     ofproto->stats.tx_bytes += packet->size;
5000     return error;
5001 }
5002
5003 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5004  * The action will state 'slow' as the reason that the action is in the slow
5005  * path.  (This is purely informational: it allows a human viewing "ovs-dpctl
5006  * dump-flows" output to see why a flow is in the slow path.)
5007  *
5008  * The 'stub_size' bytes in 'stub' will be used to store the action.
5009  * 'stub_size' must be large enough for the action.
5010  *
5011  * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5012  * respectively. */
5013 static void
5014 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5015                   enum slow_path_reason slow,
5016                   uint64_t *stub, size_t stub_size,
5017                   const struct nlattr **actionsp, size_t *actions_lenp)
5018 {
5019     union user_action_cookie cookie;
5020     struct ofpbuf buf;
5021
5022     cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5023     cookie.slow_path.unused = 0;
5024     cookie.slow_path.reason = slow;
5025
5026     ofpbuf_use_stack(&buf, stub, stub_size);
5027     if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5028         uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5029                                          ODPP_NONE);
5030         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5031     } else {
5032         odp_port_t odp_port;
5033         uint32_t pid;
5034
5035         odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5036         pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5037         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5038     }
5039     *actionsp = buf.data;
5040     *actions_lenp = buf.size;
5041 }
5042 \f
5043 static bool
5044 set_frag_handling(struct ofproto *ofproto_,
5045                   enum ofp_config_flags frag_handling)
5046 {
5047     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5048     if (frag_handling != OFPC_FRAG_REASM) {
5049         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5050         return true;
5051     } else {
5052         return false;
5053     }
5054 }
5055
5056 static enum ofperr
5057 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5058            const struct flow *flow,
5059            const struct ofpact *ofpacts, size_t ofpacts_len)
5060 {
5061     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5062     struct odputil_keybuf keybuf;
5063     struct dpif_flow_stats stats;
5064     struct xlate_out xout;
5065     struct xlate_in xin;
5066     struct ofpbuf key;
5067
5068
5069     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5070     odp_flow_key_from_flow(&key, flow,
5071                            ofp_port_to_odp_port(ofproto,
5072                                       flow->in_port.ofp_port));
5073
5074     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5075
5076     xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5077     xin.resubmit_stats = &stats;
5078     xin.ofpacts_len = ofpacts_len;
5079     xin.ofpacts = ofpacts;
5080
5081     xlate_actions(&xin, &xout);
5082     dpif_execute(ofproto->backer->dpif, key.data, key.size,
5083                  xout.odp_actions.data, xout.odp_actions.size, packet);
5084     xlate_out_uninit(&xout);
5085
5086     return 0;
5087 }
5088 \f
5089 /* NetFlow. */
5090
5091 static int
5092 set_netflow(struct ofproto *ofproto_,
5093             const struct netflow_options *netflow_options)
5094 {
5095     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5096
5097     if (netflow_options) {
5098         if (!ofproto->netflow) {
5099             ofproto->netflow = netflow_create();
5100             ofproto->backer->need_revalidate = REV_RECONFIGURE;
5101         }
5102         return netflow_set_options(ofproto->netflow, netflow_options);
5103     } else if (ofproto->netflow) {
5104         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5105         netflow_destroy(ofproto->netflow);
5106         ofproto->netflow = NULL;
5107     }
5108
5109     return 0;
5110 }
5111
5112 static void
5113 get_netflow_ids(const struct ofproto *ofproto_,
5114                 uint8_t *engine_type, uint8_t *engine_id)
5115 {
5116     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5117
5118     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5119 }
5120
5121 static void
5122 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5123 {
5124     if (!facet_is_controller_flow(facet) &&
5125         netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5126         struct subfacet *subfacet;
5127         struct ofexpired expired;
5128
5129         LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5130             if (subfacet->path == SF_FAST_PATH) {
5131                 struct dpif_flow_stats stats;
5132
5133                 subfacet_install(subfacet, &facet->xout.odp_actions,
5134                                  &stats);
5135                 subfacet_update_stats(subfacet, &stats);
5136             }
5137         }
5138
5139         expired.flow = facet->flow;
5140         expired.packet_count = facet->packet_count;
5141         expired.byte_count = facet->byte_count;
5142         expired.used = facet->used;
5143         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5144     }
5145 }
5146
5147 static void
5148 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5149 {
5150     struct cls_cursor cursor;
5151     struct facet *facet;
5152
5153     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5154     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5155     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5156         send_active_timeout(ofproto, facet);
5157     }
5158     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5159 }
5160 \f
5161 static struct ofproto_dpif *
5162 ofproto_dpif_lookup(const char *name)
5163 {
5164     struct ofproto_dpif *ofproto;
5165
5166     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5167                              hash_string(name, 0), &all_ofproto_dpifs) {
5168         if (!strcmp(ofproto->up.name, name)) {
5169             return ofproto;
5170         }
5171     }
5172     return NULL;
5173 }
5174
5175 static void
5176 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5177                           const char *argv[], void *aux OVS_UNUSED)
5178 {
5179     struct ofproto_dpif *ofproto;
5180
5181     if (argc > 1) {
5182         ofproto = ofproto_dpif_lookup(argv[1]);
5183         if (!ofproto) {
5184             unixctl_command_reply_error(conn, "no such bridge");
5185             return;
5186         }
5187         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5188         mac_learning_flush(ofproto->ml);
5189         ovs_rwlock_unlock(&ofproto->ml->rwlock);
5190     } else {
5191         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5192             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5193             mac_learning_flush(ofproto->ml);
5194             ovs_rwlock_unlock(&ofproto->ml->rwlock);
5195         }
5196     }
5197
5198     unixctl_command_reply(conn, "table successfully flushed");
5199 }
5200
5201 static struct ofport_dpif *
5202 ofbundle_get_a_port(const struct ofbundle *bundle)
5203 {
5204     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5205                         bundle_node);
5206 }
5207
5208 static void
5209 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5210                          const char *argv[], void *aux OVS_UNUSED)
5211 {
5212     struct ds ds = DS_EMPTY_INITIALIZER;
5213     const struct ofproto_dpif *ofproto;
5214     const struct mac_entry *e;
5215
5216     ofproto = ofproto_dpif_lookup(argv[1]);
5217     if (!ofproto) {
5218         unixctl_command_reply_error(conn, "no such bridge");
5219         return;
5220     }
5221
5222     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
5223     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5224     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5225         struct ofbundle *bundle = e->port.p;
5226         char name[OFP_MAX_PORT_NAME_LEN];
5227
5228         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5229                                name, sizeof name);
5230         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
5231                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
5232                       mac_entry_age(ofproto->ml, e));
5233     }
5234     ovs_rwlock_unlock(&ofproto->ml->rwlock);
5235     unixctl_command_reply(conn, ds_cstr(&ds));
5236     ds_destroy(&ds);
5237 }
5238
5239 struct trace_ctx {
5240     struct xlate_out xout;
5241     struct xlate_in xin;
5242     struct flow flow;
5243     struct ds *result;
5244 };
5245
5246 static void
5247 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5248 {
5249     struct rule_actions *actions;
5250     ovs_be64 cookie;
5251
5252     ds_put_char_multiple(result, '\t', level);
5253     if (!rule) {
5254         ds_put_cstr(result, "No match\n");
5255         return;
5256     }
5257
5258     ovs_mutex_lock(&rule->up.mutex);
5259     cookie = rule->up.flow_cookie;
5260     ovs_mutex_unlock(&rule->up.mutex);
5261
5262     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5263                   rule ? rule->up.table_id : 0, ntohll(cookie));
5264     cls_rule_format(&rule->up.cr, result);
5265     ds_put_char(result, '\n');
5266
5267     actions = rule_dpif_get_actions(rule);
5268
5269     ds_put_char_multiple(result, '\t', level);
5270     ds_put_cstr(result, "OpenFlow ");
5271     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
5272     ds_put_char(result, '\n');
5273
5274     rule_actions_unref(actions);
5275 }
5276
5277 static void
5278 trace_format_flow(struct ds *result, int level, const char *title,
5279                   struct trace_ctx *trace)
5280 {
5281     ds_put_char_multiple(result, '\t', level);
5282     ds_put_format(result, "%s: ", title);
5283     if (flow_equal(&trace->xin.flow, &trace->flow)) {
5284         ds_put_cstr(result, "unchanged");
5285     } else {
5286         flow_format(result, &trace->xin.flow);
5287         trace->flow = trace->xin.flow;
5288     }
5289     ds_put_char(result, '\n');
5290 }
5291
5292 static void
5293 trace_format_regs(struct ds *result, int level, const char *title,
5294                   struct trace_ctx *trace)
5295 {
5296     size_t i;
5297
5298     ds_put_char_multiple(result, '\t', level);
5299     ds_put_format(result, "%s:", title);
5300     for (i = 0; i < FLOW_N_REGS; i++) {
5301         ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5302     }
5303     ds_put_char(result, '\n');
5304 }
5305
5306 static void
5307 trace_format_odp(struct ds *result, int level, const char *title,
5308                  struct trace_ctx *trace)
5309 {
5310     struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5311
5312     ds_put_char_multiple(result, '\t', level);
5313     ds_put_format(result, "%s: ", title);
5314     format_odp_actions(result, odp_actions->data, odp_actions->size);
5315     ds_put_char(result, '\n');
5316 }
5317
5318 static void
5319 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5320 {
5321     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5322     struct ds *result = trace->result;
5323
5324     ds_put_char(result, '\n');
5325     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5326     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5327     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
5328     trace_format_rule(result, recurse + 1, rule);
5329 }
5330
5331 static void
5332 trace_report(struct xlate_in *xin, const char *s, int recurse)
5333 {
5334     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5335     struct ds *result = trace->result;
5336
5337     ds_put_char_multiple(result, '\t', recurse);
5338     ds_put_cstr(result, s);
5339     ds_put_char(result, '\n');
5340 }
5341
5342 static void
5343 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5344                       void *aux OVS_UNUSED)
5345 {
5346     const struct dpif_backer *backer;
5347     struct ofproto_dpif *ofproto;
5348     struct ofpbuf odp_key, odp_mask;
5349     struct ofpbuf *packet;
5350     struct ds result;
5351     struct flow flow;
5352     char *s;
5353
5354     packet = NULL;
5355     backer = NULL;
5356     ds_init(&result);
5357     ofpbuf_init(&odp_key, 0);
5358     ofpbuf_init(&odp_mask, 0);
5359
5360     /* Handle "-generate" or a hex string as the last argument. */
5361     if (!strcmp(argv[argc - 1], "-generate")) {
5362         packet = ofpbuf_new(0);
5363         argc--;
5364     } else {
5365         const char *error = eth_from_hex(argv[argc - 1], &packet);
5366         if (!error) {
5367             argc--;
5368         } else if (argc == 4) {
5369             /* The 3-argument form must end in "-generate' or a hex string. */
5370             unixctl_command_reply_error(conn, error);
5371             goto exit;
5372         }
5373     }
5374
5375     /* Parse the flow and determine whether a datapath or
5376      * bridge is specified. If function odp_flow_key_from_string()
5377      * returns 0, the flow is a odp_flow. If function
5378      * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5379     if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5380         /* If the odp_flow is the second argument,
5381          * the datapath name is the first argument. */
5382         if (argc == 3) {
5383             const char *dp_type;
5384             if (!strncmp(argv[1], "ovs-", 4)) {
5385                 dp_type = argv[1] + 4;
5386             } else {
5387                 dp_type = argv[1];
5388             }
5389             backer = shash_find_data(&all_dpif_backers, dp_type);
5390             if (!backer) {
5391                 unixctl_command_reply_error(conn, "Cannot find datapath "
5392                                "of this name");
5393                 goto exit;
5394             }
5395         } else {
5396             /* No datapath name specified, so there should be only one
5397              * datapath. */
5398             struct shash_node *node;
5399             if (shash_count(&all_dpif_backers) != 1) {
5400                 unixctl_command_reply_error(conn, "Must specify datapath "
5401                          "name, there is more than one type of datapath");
5402                 goto exit;
5403             }
5404             node = shash_first(&all_dpif_backers);
5405             backer = node->data;
5406         }
5407
5408         if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5409                           NULL, &ofproto, NULL)) {
5410             unixctl_command_reply_error(conn, "Invalid datapath flow");
5411             goto exit;
5412         }
5413         ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5414     } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5415         if (argc != 3) {
5416             unixctl_command_reply_error(conn, "Must specify bridge name");
5417             goto exit;
5418         }
5419
5420         ofproto = ofproto_dpif_lookup(argv[1]);
5421         if (!ofproto) {
5422             unixctl_command_reply_error(conn, "Unknown bridge name");
5423             goto exit;
5424         }
5425     } else {
5426         unixctl_command_reply_error(conn, "Bad flow syntax");
5427         goto exit;
5428     }
5429
5430     /* Generate a packet, if requested. */
5431     if (packet) {
5432         if (!packet->size) {
5433             flow_compose(packet, &flow);
5434         } else {
5435             union flow_in_port in_port_;
5436
5437             in_port_ = flow.in_port;
5438             ds_put_cstr(&result, "Packet: ");
5439             s = ofp_packet_to_string(packet->data, packet->size);
5440             ds_put_cstr(&result, s);
5441             free(s);
5442
5443             /* Use the metadata from the flow and the packet argument
5444              * to reconstruct the flow. */
5445             flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5446                          &in_port_, &flow);
5447         }
5448     }
5449
5450     ofproto_trace(ofproto, &flow, packet, &result);
5451     unixctl_command_reply(conn, ds_cstr(&result));
5452
5453 exit:
5454     ds_destroy(&result);
5455     ofpbuf_delete(packet);
5456     ofpbuf_uninit(&odp_key);
5457     ofpbuf_uninit(&odp_mask);
5458 }
5459
5460 static void
5461 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5462               const struct ofpbuf *packet, struct ds *ds)
5463 {
5464     struct rule_dpif *rule;
5465     struct flow_wildcards wc;
5466
5467     ds_put_cstr(ds, "Flow: ");
5468     flow_format(ds, flow);
5469     ds_put_char(ds, '\n');
5470
5471     flow_wildcards_init_catchall(&wc);
5472     rule_dpif_lookup(ofproto, flow, &wc, &rule);
5473
5474     trace_format_rule(ds, 0, rule);
5475     if (rule == ofproto->miss_rule) {
5476         ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5477     } else if (rule == ofproto->no_packet_in_rule) {
5478         ds_put_cstr(ds, "\nNo match, packets dropped because "
5479                     "OFPPC_NO_PACKET_IN is set on in_port.\n");
5480     } else if (rule == ofproto->drop_frags_rule) {
5481         ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5482                     "and the fragment handling mode is \"drop\".\n");
5483     }
5484
5485     if (rule) {
5486         uint64_t odp_actions_stub[1024 / 8];
5487         struct ofpbuf odp_actions;
5488         struct trace_ctx trace;
5489         struct match match;
5490         uint8_t tcp_flags;
5491
5492         tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5493         trace.result = ds;
5494         trace.flow = *flow;
5495         ofpbuf_use_stub(&odp_actions,
5496                         odp_actions_stub, sizeof odp_actions_stub);
5497         xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5498         trace.xin.resubmit_hook = trace_resubmit;
5499         trace.xin.report_hook = trace_report;
5500
5501         xlate_actions(&trace.xin, &trace.xout);
5502         flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5503
5504         ds_put_char(ds, '\n');
5505         trace_format_flow(ds, 0, "Final flow", &trace);
5506
5507         match_init(&match, flow, &trace.xout.wc);
5508         ds_put_cstr(ds, "Relevant fields: ");
5509         match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5510         ds_put_char(ds, '\n');
5511
5512         ds_put_cstr(ds, "Datapath actions: ");
5513         format_odp_actions(ds, trace.xout.odp_actions.data,
5514                            trace.xout.odp_actions.size);
5515
5516         if (trace.xout.slow) {
5517             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5518                         "slow path because it:");
5519             switch (trace.xout.slow) {
5520             case SLOW_CFM:
5521                 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5522                 break;
5523             case SLOW_LACP:
5524                 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5525                 break;
5526             case SLOW_STP:
5527                 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5528                 break;
5529             case SLOW_BFD:
5530                 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5531                 break;
5532             case SLOW_CONTROLLER:
5533                 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5534                             "to the OpenFlow controller.");
5535                 break;
5536             case __SLOW_MAX:
5537                 NOT_REACHED();
5538             }
5539         }
5540
5541         xlate_out_uninit(&trace.xout);
5542     }
5543
5544     rule_dpif_unref(rule);
5545 }
5546
5547 /* Runs a self-check of flow translations in 'ofproto'.  Appends a message to
5548  * 'reply' describing the results. */
5549 static void
5550 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5551 {
5552     struct cls_cursor cursor;
5553     struct facet *facet;
5554     int errors;
5555
5556     errors = 0;
5557     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5558     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5559     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5560         if (!facet_check_consistency(facet)) {
5561             errors++;
5562         }
5563     }
5564     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5565     if (errors) {
5566         ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5567     }
5568
5569     if (errors) {
5570         ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5571                       ofproto->up.name, errors);
5572     } else {
5573         ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5574     }
5575 }
5576
5577 static void
5578 ofproto_dpif_self_check(struct unixctl_conn *conn,
5579                         int argc, const char *argv[], void *aux OVS_UNUSED)
5580 {
5581     struct ds reply = DS_EMPTY_INITIALIZER;
5582     struct ofproto_dpif *ofproto;
5583
5584     if (argc > 1) {
5585         ofproto = ofproto_dpif_lookup(argv[1]);
5586         if (!ofproto) {
5587             unixctl_command_reply_error(conn, "Unknown ofproto (use "
5588                                         "ofproto/list for help)");
5589             return;
5590         }
5591         ofproto_dpif_self_check__(ofproto, &reply);
5592     } else {
5593         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5594             ofproto_dpif_self_check__(ofproto, &reply);
5595         }
5596     }
5597
5598     unixctl_command_reply(conn, ds_cstr(&reply));
5599     ds_destroy(&reply);
5600 }
5601
5602 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
5603  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
5604  * to destroy 'ofproto_shash' and free the returned value. */
5605 static const struct shash_node **
5606 get_ofprotos(struct shash *ofproto_shash)
5607 {
5608     const struct ofproto_dpif *ofproto;
5609
5610     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5611         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5612         shash_add_nocopy(ofproto_shash, name, ofproto);
5613     }
5614
5615     return shash_sort(ofproto_shash);
5616 }
5617
5618 static void
5619 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5620                               const char *argv[] OVS_UNUSED,
5621                               void *aux OVS_UNUSED)
5622 {
5623     struct ds ds = DS_EMPTY_INITIALIZER;
5624     struct shash ofproto_shash;
5625     const struct shash_node **sorted_ofprotos;
5626     int i;
5627
5628     shash_init(&ofproto_shash);
5629     sorted_ofprotos = get_ofprotos(&ofproto_shash);
5630     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5631         const struct shash_node *node = sorted_ofprotos[i];
5632         ds_put_format(&ds, "%s\n", node->name);
5633     }
5634
5635     shash_destroy(&ofproto_shash);
5636     free(sorted_ofprotos);
5637
5638     unixctl_command_reply(conn, ds_cstr(&ds));
5639     ds_destroy(&ds);
5640 }
5641
5642 static void
5643 show_dp_rates(struct ds *ds, const char *heading,
5644               const struct avg_subfacet_rates *rates)
5645 {
5646     ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5647                   heading, rates->add_rate, rates->del_rate);
5648 }
5649
5650 static void
5651 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5652 {
5653     const struct shash_node **ofprotos;
5654     struct ofproto_dpif *ofproto;
5655     struct shash ofproto_shash;
5656     uint64_t n_hit, n_missed;
5657     long long int minutes;
5658     size_t i;
5659
5660     n_hit = n_missed = 0;
5661     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5662         if (ofproto->backer == backer) {
5663             n_missed += ofproto->n_missed;
5664             n_hit += ofproto->n_hit;
5665         }
5666     }
5667
5668     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5669                   dpif_name(backer->dpif), n_hit, n_missed);
5670     ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5671                   " life span: %lldms\n", hmap_count(&backer->subfacets),
5672                   backer->avg_n_subfacet, backer->max_n_subfacet,
5673                   backer->avg_subfacet_life);
5674
5675     minutes = (time_msec() - backer->created) / (1000 * 60);
5676     if (minutes >= 60) {
5677         show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5678     }
5679     if (minutes >= 60 * 24) {
5680         show_dp_rates(ds, "\tdaily avg:",  &backer->daily);
5681     }
5682     show_dp_rates(ds, "\toverall avg:",  &backer->lifetime);
5683
5684     shash_init(&ofproto_shash);
5685     ofprotos = get_ofprotos(&ofproto_shash);
5686     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5687         struct ofproto_dpif *ofproto = ofprotos[i]->data;
5688         const struct shash_node **ports;
5689         size_t j;
5690
5691         if (ofproto->backer != backer) {
5692             continue;
5693         }
5694
5695         ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5696                       ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5697
5698         ports = shash_sort(&ofproto->up.port_by_name);
5699         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5700             const struct shash_node *node = ports[j];
5701             struct ofport *ofport = node->data;
5702             struct smap config;
5703             odp_port_t odp_port;
5704
5705             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5706                           ofport->ofp_port);
5707
5708             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5709             if (odp_port != ODPP_NONE) {
5710                 ds_put_format(ds, "%"PRIu32":", odp_port);
5711             } else {
5712                 ds_put_cstr(ds, "none:");
5713             }
5714
5715             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5716
5717             smap_init(&config);
5718             if (!netdev_get_config(ofport->netdev, &config)) {
5719                 const struct smap_node **nodes;
5720                 size_t i;
5721
5722                 nodes = smap_sort(&config);
5723                 for (i = 0; i < smap_count(&config); i++) {
5724                     const struct smap_node *node = nodes[i];
5725                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5726                                   node->key, node->value);
5727                 }
5728                 free(nodes);
5729             }
5730             smap_destroy(&config);
5731
5732             ds_put_char(ds, ')');
5733             ds_put_char(ds, '\n');
5734         }
5735         free(ports);
5736     }
5737     shash_destroy(&ofproto_shash);
5738     free(ofprotos);
5739 }
5740
5741 static void
5742 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5743                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5744 {
5745     struct ds ds = DS_EMPTY_INITIALIZER;
5746     const struct shash_node **backers;
5747     int i;
5748
5749     backers = shash_sort(&all_dpif_backers);
5750     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5751         dpif_show_backer(backers[i]->data, &ds);
5752     }
5753     free(backers);
5754
5755     unixctl_command_reply(conn, ds_cstr(&ds));
5756     ds_destroy(&ds);
5757 }
5758
5759 /* Dump the megaflow (facet) cache.  This is useful to check the
5760  * correctness of flow wildcarding, since the same mechanism is used for
5761  * both xlate caching and kernel wildcarding.
5762  *
5763  * It's important to note that in the output the flow description uses
5764  * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5765  *
5766  * This command is only needed for advanced debugging, so it's not
5767  * documented in the man page. */
5768 static void
5769 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5770                                     int argc OVS_UNUSED, const char *argv[],
5771                                     void *aux OVS_UNUSED)
5772 {
5773     struct ds ds = DS_EMPTY_INITIALIZER;
5774     const struct ofproto_dpif *ofproto;
5775     long long int now = time_msec();
5776     struct cls_cursor cursor;
5777     struct facet *facet;
5778
5779     ofproto = ofproto_dpif_lookup(argv[1]);
5780     if (!ofproto) {
5781         unixctl_command_reply_error(conn, "no such bridge");
5782         return;
5783     }
5784
5785     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5786     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5787     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5788         cls_rule_format(&facet->cr, &ds);
5789         ds_put_cstr(&ds, ", ");
5790         ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5791         ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5792         ds_put_cstr(&ds, "Datapath actions: ");
5793         if (facet->xout.slow) {
5794             uint64_t slow_path_stub[128 / 8];
5795             const struct nlattr *actions;
5796             size_t actions_len;
5797
5798             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5799                               slow_path_stub, sizeof slow_path_stub,
5800                               &actions, &actions_len);
5801             format_odp_actions(&ds, actions, actions_len);
5802         } else {
5803             format_odp_actions(&ds, facet->xout.odp_actions.data,
5804                                facet->xout.odp_actions.size);
5805         }
5806         ds_put_cstr(&ds, "\n");
5807     }
5808     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5809
5810     ds_chomp(&ds, '\n');
5811     unixctl_command_reply(conn, ds_cstr(&ds));
5812     ds_destroy(&ds);
5813 }
5814
5815 /* Disable using the megaflows.
5816  *
5817  * This command is only needed for advanced debugging, so it's not
5818  * documented in the man page. */
5819 static void
5820 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5821                                        int argc OVS_UNUSED,
5822                                        const char *argv[] OVS_UNUSED,
5823                                        void *aux OVS_UNUSED)
5824 {
5825     struct ofproto_dpif *ofproto;
5826
5827     enable_megaflows = false;
5828
5829     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5830         flush(&ofproto->up);
5831     }
5832
5833     unixctl_command_reply(conn, "megaflows disabled");
5834 }
5835
5836 /* Re-enable using megaflows.
5837  *
5838  * This command is only needed for advanced debugging, so it's not
5839  * documented in the man page. */
5840 static void
5841 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5842                                       int argc OVS_UNUSED,
5843                                       const char *argv[] OVS_UNUSED,
5844                                       void *aux OVS_UNUSED)
5845 {
5846     struct ofproto_dpif *ofproto;
5847
5848     enable_megaflows = true;
5849
5850     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5851         flush(&ofproto->up);
5852     }
5853
5854     unixctl_command_reply(conn, "megaflows enabled");
5855 }
5856
5857 static void
5858 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5859                                 int argc OVS_UNUSED, const char *argv[],
5860                                 void *aux OVS_UNUSED)
5861 {
5862     struct ds ds = DS_EMPTY_INITIALIZER;
5863     const struct ofproto_dpif *ofproto;
5864     struct subfacet *subfacet;
5865
5866     ofproto = ofproto_dpif_lookup(argv[1]);
5867     if (!ofproto) {
5868         unixctl_command_reply_error(conn, "no such bridge");
5869         return;
5870     }
5871
5872     update_stats(ofproto->backer);
5873
5874     HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5875         struct facet *facet = subfacet->facet;
5876         struct odputil_keybuf maskbuf;
5877         struct ofpbuf mask;
5878
5879         if (facet->ofproto != ofproto) {
5880             continue;
5881         }
5882
5883         ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5884         if (enable_megaflows) {
5885             odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5886                                    &facet->flow, UINT32_MAX);
5887         }
5888
5889         odp_flow_format(subfacet->key, subfacet->key_len,
5890                         mask.data, mask.size, &ds, false);
5891
5892         ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5893                       subfacet->dp_packet_count, subfacet->dp_byte_count);
5894         if (subfacet->used) {
5895             ds_put_format(&ds, "%.3fs",
5896                           (time_msec() - subfacet->used) / 1000.0);
5897         } else {
5898             ds_put_format(&ds, "never");
5899         }
5900         if (subfacet->facet->tcp_flags) {
5901             ds_put_cstr(&ds, ", flags:");
5902             packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5903         }
5904
5905         ds_put_cstr(&ds, ", actions:");
5906         if (facet->xout.slow) {
5907             uint64_t slow_path_stub[128 / 8];
5908             const struct nlattr *actions;
5909             size_t actions_len;
5910
5911             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5912                               slow_path_stub, sizeof slow_path_stub,
5913                               &actions, &actions_len);
5914             format_odp_actions(&ds, actions, actions_len);
5915         } else {
5916             format_odp_actions(&ds, facet->xout.odp_actions.data,
5917                                facet->xout.odp_actions.size);
5918         }
5919         ds_put_char(&ds, '\n');
5920     }
5921
5922     unixctl_command_reply(conn, ds_cstr(&ds));
5923     ds_destroy(&ds);
5924 }
5925
5926 static void
5927 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5928                                int argc OVS_UNUSED, const char *argv[],
5929                                void *aux OVS_UNUSED)
5930 {
5931     struct ds ds = DS_EMPTY_INITIALIZER;
5932     struct ofproto_dpif *ofproto;
5933
5934     ofproto = ofproto_dpif_lookup(argv[1]);
5935     if (!ofproto) {
5936         unixctl_command_reply_error(conn, "no such bridge");
5937         return;
5938     }
5939
5940     flush(&ofproto->up);
5941
5942     unixctl_command_reply(conn, ds_cstr(&ds));
5943     ds_destroy(&ds);
5944 }
5945
5946 static void
5947 ofproto_dpif_unixctl_init(void)
5948 {
5949     static bool registered;
5950     if (registered) {
5951         return;
5952     }
5953     registered = true;
5954
5955     unixctl_command_register(
5956         "ofproto/trace",
5957         "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
5958         1, 3, ofproto_unixctl_trace, NULL);
5959     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5960                              ofproto_unixctl_fdb_flush, NULL);
5961     unixctl_command_register("fdb/show", "bridge", 1, 1,
5962                              ofproto_unixctl_fdb_show, NULL);
5963     unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
5964                              ofproto_dpif_self_check, NULL);
5965     unixctl_command_register("dpif/dump-dps", "", 0, 0,
5966                              ofproto_unixctl_dpif_dump_dps, NULL);
5967     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5968                              NULL);
5969     unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
5970                              ofproto_unixctl_dpif_dump_flows, NULL);
5971     unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
5972                              ofproto_unixctl_dpif_del_flows, NULL);
5973     unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
5974                              ofproto_unixctl_dpif_dump_megaflows, NULL);
5975     unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
5976                              ofproto_unixctl_dpif_disable_megaflows, NULL);
5977     unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
5978                              ofproto_unixctl_dpif_enable_megaflows, NULL);
5979 }
5980 \f
5981 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5982  *
5983  * This is deprecated.  It is only for compatibility with broken device drivers
5984  * in old versions of Linux that do not properly support VLANs when VLAN
5985  * devices are not used.  When broken device drivers are no longer in
5986  * widespread use, we will delete these interfaces. */
5987
5988 static int
5989 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5990 {
5991     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5992     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5993
5994     if (realdev_ofp_port == ofport->realdev_ofp_port
5995         && vid == ofport->vlandev_vid) {
5996         return 0;
5997     }
5998
5999     ofproto->backer->need_revalidate = REV_RECONFIGURE;
6000
6001     if (ofport->realdev_ofp_port) {
6002         vsp_remove(ofport);
6003     }
6004     if (realdev_ofp_port && ofport->bundle) {
6005         /* vlandevs are enslaved to their realdevs, so they are not allowed to
6006          * themselves be part of a bundle. */
6007         bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6008     }
6009
6010     ofport->realdev_ofp_port = realdev_ofp_port;
6011     ofport->vlandev_vid = vid;
6012
6013     if (realdev_ofp_port) {
6014         vsp_add(ofport, realdev_ofp_port, vid);
6015     }
6016
6017     return 0;
6018 }
6019
6020 static uint32_t
6021 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6022 {
6023     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6024 }
6025
6026 bool
6027 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6028     OVS_EXCLUDED(ofproto->vsp_mutex)
6029 {
6030     bool ret;
6031
6032     ovs_mutex_lock(&ofproto->vsp_mutex);
6033     ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6034     ovs_mutex_unlock(&ofproto->vsp_mutex);
6035     return ret;
6036 }
6037
6038 static ofp_port_t
6039 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6040                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6041     OVS_REQUIRES(ofproto->vsp_mutex)
6042 {
6043     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6044         int vid = vlan_tci_to_vid(vlan_tci);
6045         const struct vlan_splinter *vsp;
6046
6047         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6048                                  hash_realdev_vid(realdev_ofp_port, vid),
6049                                  &ofproto->realdev_vid_map) {
6050             if (vsp->realdev_ofp_port == realdev_ofp_port
6051                 && vsp->vid == vid) {
6052                 return vsp->vlandev_ofp_port;
6053             }
6054         }
6055     }
6056     return realdev_ofp_port;
6057 }
6058
6059 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6060  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6061  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
6062  * 'vlan_tci' 9, it would return the port number of eth0.9.
6063  *
6064  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6065  * function just returns its 'realdev_ofp_port' argument. */
6066 ofp_port_t
6067 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6068                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6069     OVS_EXCLUDED(ofproto->vsp_mutex)
6070 {
6071     ofp_port_t ret;
6072
6073     ovs_mutex_lock(&ofproto->vsp_mutex);
6074     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6075     ovs_mutex_unlock(&ofproto->vsp_mutex);
6076     return ret;
6077 }
6078
6079 static struct vlan_splinter *
6080 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6081 {
6082     struct vlan_splinter *vsp;
6083
6084     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6085                              hash_ofp_port(vlandev_ofp_port),
6086                              &ofproto->vlandev_map) {
6087         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6088             return vsp;
6089         }
6090     }
6091
6092     return NULL;
6093 }
6094
6095 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6096  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6097  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
6098  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6099  * eth0 and store 9 in '*vid'.
6100  *
6101  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6102  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
6103  * always does.*/
6104 static ofp_port_t
6105 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6106                        ofp_port_t vlandev_ofp_port, int *vid)
6107     OVS_REQUIRES(ofproto->vsp_mutex)
6108 {
6109     if (!hmap_is_empty(&ofproto->vlandev_map)) {
6110         const struct vlan_splinter *vsp;
6111
6112         vsp = vlandev_find(ofproto, vlandev_ofp_port);
6113         if (vsp) {
6114             if (vid) {
6115                 *vid = vsp->vid;
6116             }
6117             return vsp->realdev_ofp_port;
6118         }
6119     }
6120     return 0;
6121 }
6122
6123 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6124  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
6125  * 'flow->in_port' to the "real" device backing the VLAN device, sets
6126  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Otherwise (which is
6127  * always the case unless VLAN splinters are enabled), returns false without
6128  * making any changes. */
6129 bool
6130 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6131     OVS_EXCLUDED(ofproto->vsp_mutex)
6132 {
6133     ofp_port_t realdev;
6134     int vid;
6135
6136     ovs_mutex_lock(&ofproto->vsp_mutex);
6137     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6138     ovs_mutex_unlock(&ofproto->vsp_mutex);
6139     if (!realdev) {
6140         return false;
6141     }
6142
6143     /* Cause the flow to be processed as if it came in on the real device with
6144      * the VLAN device's VLAN ID. */
6145     flow->in_port.ofp_port = realdev;
6146     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6147     return true;
6148 }
6149
6150 static void
6151 vsp_remove(struct ofport_dpif *port)
6152 {
6153     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6154     struct vlan_splinter *vsp;
6155
6156     ovs_mutex_lock(&ofproto->vsp_mutex);
6157     vsp = vlandev_find(ofproto, port->up.ofp_port);
6158     if (vsp) {
6159         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6160         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6161         free(vsp);
6162
6163         port->realdev_ofp_port = 0;
6164     } else {
6165         VLOG_ERR("missing vlan device record");
6166     }
6167     ovs_mutex_unlock(&ofproto->vsp_mutex);
6168 }
6169
6170 static void
6171 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6172 {
6173     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6174
6175     ovs_mutex_lock(&ofproto->vsp_mutex);
6176     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6177         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6178             == realdev_ofp_port)) {
6179         struct vlan_splinter *vsp;
6180
6181         vsp = xmalloc(sizeof *vsp);
6182         vsp->realdev_ofp_port = realdev_ofp_port;
6183         vsp->vlandev_ofp_port = port->up.ofp_port;
6184         vsp->vid = vid;
6185
6186         port->realdev_ofp_port = realdev_ofp_port;
6187
6188         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6189                     hash_ofp_port(port->up.ofp_port));
6190         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6191                     hash_realdev_vid(realdev_ofp_port, vid));
6192     } else {
6193         VLOG_ERR("duplicate vlan device record");
6194     }
6195     ovs_mutex_unlock(&ofproto->vsp_mutex);
6196 }
6197
6198 static odp_port_t
6199 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6200 {
6201     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6202     return ofport ? ofport->odp_port : ODPP_NONE;
6203 }
6204
6205 struct ofport_dpif *
6206 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6207 {
6208     struct ofport_dpif *port;
6209
6210     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6211     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6212                              &backer->odp_to_ofport_map) {
6213         if (port->odp_port == odp_port) {
6214             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6215             return port;
6216         }
6217     }
6218
6219     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6220     return NULL;
6221 }
6222
6223 static ofp_port_t
6224 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6225 {
6226     struct ofport_dpif *port;
6227
6228     port = odp_port_to_ofport(ofproto->backer, odp_port);
6229     if (port && &ofproto->up == port->up.ofproto) {
6230         return port->up.ofp_port;
6231     } else {
6232         return OFPP_NONE;
6233     }
6234 }
6235
6236 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6237  * most heavily weighted element.  'base' designates the rate of decay: after
6238  * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6239  * (about .37). */
6240 static void
6241 exp_mavg(double *avg, int base, double new)
6242 {
6243     *avg = (*avg * (base - 1) + new) / base;
6244 }
6245
6246 static void
6247 update_moving_averages(struct dpif_backer *backer)
6248 {
6249     const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6250     long long int minutes = (time_msec() - backer->created) / min_ms;
6251
6252     if (minutes > 0) {
6253         backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6254             / minutes;
6255         backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6256             / minutes;
6257     } else {
6258         backer->lifetime.add_rate = 0.0;
6259         backer->lifetime.del_rate = 0.0;
6260     }
6261
6262     /* Update hourly averages on the minute boundaries. */
6263     if (time_msec() - backer->last_minute >= min_ms) {
6264         exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6265         exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6266
6267         /* Update daily averages on the hour boundaries. */
6268         if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6269             exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6270             exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6271         }
6272
6273         backer->total_subfacet_add_count += backer->subfacet_add_count;
6274         backer->total_subfacet_del_count += backer->subfacet_del_count;
6275         backer->subfacet_add_count = 0;
6276         backer->subfacet_del_count = 0;
6277         backer->last_minute += min_ms;
6278     }
6279 }
6280
6281 const struct ofproto_class ofproto_dpif_class = {
6282     init,
6283     enumerate_types,
6284     enumerate_names,
6285     del,
6286     port_open_type,
6287     type_run,
6288     type_run_fast,
6289     type_wait,
6290     alloc,
6291     construct,
6292     destruct,
6293     dealloc,
6294     run,
6295     run_fast,
6296     wait,
6297     get_memory_usage,
6298     flush,
6299     get_features,
6300     get_tables,
6301     port_alloc,
6302     port_construct,
6303     port_destruct,
6304     port_dealloc,
6305     port_modified,
6306     port_reconfigured,
6307     port_query_by_name,
6308     port_add,
6309     port_del,
6310     port_get_stats,
6311     port_dump_start,
6312     port_dump_next,
6313     port_dump_done,
6314     port_poll,
6315     port_poll_wait,
6316     port_is_lacp_current,
6317     NULL,                       /* rule_choose_table */
6318     rule_alloc,
6319     rule_construct,
6320     rule_insert,
6321     rule_delete,
6322     rule_destruct,
6323     rule_dealloc,
6324     rule_get_stats,
6325     rule_execute,
6326     rule_modify_actions,
6327     set_frag_handling,
6328     packet_out,
6329     set_netflow,
6330     get_netflow_ids,
6331     set_sflow,
6332     set_ipfix,
6333     set_cfm,
6334     get_cfm_status,
6335     set_bfd,
6336     get_bfd_status,
6337     set_stp,
6338     get_stp_status,
6339     set_stp_port,
6340     get_stp_port_status,
6341     set_queues,
6342     bundle_set,
6343     bundle_remove,
6344     mirror_set__,
6345     mirror_get_stats__,
6346     set_flood_vlans,
6347     is_mirror_output_bundle,
6348     forward_bpdu_changed,
6349     set_mac_table_config,
6350     set_realdev,
6351     NULL,                       /* meter_get_features */
6352     NULL,                       /* meter_set */
6353     NULL,                       /* meter_get */
6354     NULL,                       /* meter_del */
6355     NULL,                       /* group_alloc */
6356     NULL,                       /* group_construct */
6357     NULL,                       /* group_destruct */
6358     NULL,                       /* group_dealloc */
6359     NULL,                       /* group_modify */
6360     NULL,                       /* group_get_stats */
6361 };