ofproto/ofproto-dpif.c

   1 /*
   2  * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18
  19 #include "ofproto/ofproto-dpif.h"
  20 #include "ofproto/ofproto-provider.h"
  21
  22 #include <errno.h>
  23
  24 #include "bfd.h"
  25 #include "bond.h"
  26 #include "bundle.h"
  27 #include "byte-order.h"
  28 #include "connmgr.h"
  29 #include "coverage.h"
  30 #include "cfm.h"
  31 #include "dpif.h"
  32 #include "dynamic-string.h"
  33 #include "fail-open.h"
  34 #include "hmapx.h"
  35 #include "lacp.h"
  36 #include "learn.h"
  37 #include "mac-learning.h"
  38 #include "meta-flow.h"
  39 #include "multipath.h"
  40 #include "netdev-vport.h"
  41 #include "netdev.h"
  42 #include "netlink.h"
  43 #include "nx-match.h"
  44 #include "odp-util.h"
  45 #include "odp-execute.h"
  46 #include "ofp-util.h"
  47 #include "ofpbuf.h"
  48 #include "ofp-actions.h"
  49 #include "ofp-parse.h"
  50 #include "ofp-print.h"
  51 #include "ofproto-dpif-governor.h"
  52 #include "ofproto-dpif-ipfix.h"
  53 #include "ofproto-dpif-mirror.h"
  54 #include "ofproto-dpif-sflow.h"
  55 #include "ofproto-dpif-upcall.h"
  56 #include "ofproto-dpif-xlate.h"
  57 #include "poll-loop.h"
  58 #include "simap.h"
  59 #include "smap.h"
  60 #include "timer.h"
  61 #include "tunnel.h"
  62 #include "unaligned.h"
  63 #include "unixctl.h"
  64 #include "vlan-bitmap.h"
  65 #include "vlog.h"
  66
  67 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  68
  69 COVERAGE_DEFINE(ofproto_dpif_expired);
  70 COVERAGE_DEFINE(facet_changed_rule);
  71 COVERAGE_DEFINE(facet_revalidate);
  72 COVERAGE_DEFINE(facet_unexpected);
  73 COVERAGE_DEFINE(facet_suppress);
  74 COVERAGE_DEFINE(subfacet_install_fail);
  75 COVERAGE_DEFINE(packet_in_overflow);
  76 COVERAGE_DEFINE(flow_mod_overflow);
  77
  78 /* Number of implemented OpenFlow tables. */
  79 enum { N_TABLES = 255 };
  80 enum { TBL_INTERNAL = N_TABLES - 1 };    /* Used for internal hidden rules. */
  81 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
  82
  83 struct flow_miss;
  84 struct facet;
  85
  86 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
  87
  88 struct ofbundle {
  89     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
  90     struct ofproto_dpif *ofproto; /* Owning ofproto. */
  91     void *aux;                  /* Key supplied by ofproto's client. */
  92     char *name;                 /* Identifier for log messages. */
  93
  94     /* Configuration. */
  95     struct list ports;          /* Contains "struct ofport"s. */
  96     enum port_vlan_mode vlan_mode; /* VLAN mode */
  97     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
  98     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
  99                                  * NULL if all VLANs are trunked. */
 100     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 101     struct bond *bond;          /* Nonnull iff more than one port. */
 102     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 103
 104     /* Status. */
 105     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 106 };
 107
 108 static void bundle_remove(struct ofport *);
 109 static void bundle_update(struct ofbundle *);
 110 static void bundle_destroy(struct ofbundle *);
 111 static void bundle_del_port(struct ofport_dpif *);
 112 static void bundle_run(struct ofbundle *);
 113 static void bundle_wait(struct ofbundle *);
 114
 115 static void stp_run(struct ofproto_dpif *ofproto);
 116 static void stp_wait(struct ofproto_dpif *ofproto);
 117 static int set_stp_port(struct ofport *,
 118                         const struct ofproto_port_stp_settings *);
 119
 120 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
 121                               enum slow_path_reason,
 122                               uint64_t *stub, size_t stub_size,
 123                               const struct nlattr **actionsp,
 124                               size_t *actions_lenp);
 125
 126 /* A subfacet (see "struct subfacet" below) has three possible installation
 127  * states:
 128  *
 129  *   - SF_NOT_INSTALLED: Not installed in the datapath.  This will only be the
 130  *     case just after the subfacet is created, just before the subfacet is
 131  *     destroyed, or if the datapath returns an error when we try to install a
 132  *     subfacet.
 133  *
 134  *   - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
 135  *
 136  *   - SF_SLOW_PATH: An action that sends every packet for the subfacet through
 137  *     ofproto_dpif is installed in the datapath.
 138  */
 139 enum subfacet_path {
 140     SF_NOT_INSTALLED,           /* No datapath flow for this subfacet. */
 141     SF_FAST_PATH,               /* Full actions are installed. */
 142     SF_SLOW_PATH,               /* Send-to-userspace action is installed. */
 143 };
 144
 145 /* A dpif flow and actions associated with a facet.
 146  *
 147  * See also the large comment on struct facet. */
 148 struct subfacet {
 149     /* Owners. */
 150     struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
 151     struct list list_node;      /* In struct facet's 'facets' list. */
 152     struct facet *facet;        /* Owning facet. */
 153     struct dpif_backer *backer; /* Owning backer. */
 154
 155     enum odp_key_fitness key_fitness;
 156     struct nlattr *key;
 157     int key_len;
 158
 159     long long int used;         /* Time last used; time created if not used. */
 160     long long int created;      /* Time created. */
 161
 162     uint64_t dp_packet_count;   /* Last known packet count in the datapath. */
 163     uint64_t dp_byte_count;     /* Last known byte count in the datapath. */
 164
 165     enum subfacet_path path;    /* Installed in datapath? */
 166 };
 167
 168 #define SUBFACET_DESTROY_MAX_BATCH 50
 169
 170 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *);
 171 static struct subfacet *subfacet_find(struct dpif_backer *,
 172                                       const struct nlattr *key, size_t key_len,
 173                                       uint32_t key_hash);
 174 static void subfacet_destroy(struct subfacet *);
 175 static void subfacet_destroy__(struct subfacet *);
 176 static void subfacet_destroy_batch(struct dpif_backer *,
 177                                    struct subfacet **, int n);
 178 static void subfacet_reset_dp_stats(struct subfacet *,
 179                                     struct dpif_flow_stats *);
 180 static void subfacet_update_stats(struct subfacet *,
 181                                   const struct dpif_flow_stats *);
 182 static int subfacet_install(struct subfacet *,
 183                             const struct ofpbuf *odp_actions,
 184                             struct dpif_flow_stats *);
 185 static void subfacet_uninstall(struct subfacet *);
 186
 187 /* A unique, non-overlapping instantiation of an OpenFlow flow.
 188  *
 189  * A facet associates a "struct flow", which represents the Open vSwitch
 190  * userspace idea of an exact-match flow, with one or more subfacets.
 191  * While the facet is created based on an exact-match flow, it is stored
 192  * within the ofproto based on the wildcards that could be expressed
 193  * based on the flow table and other configuration.  (See the 'wc'
 194  * description in "struct xlate_out" for more details.)
 195  *
 196  * Each subfacet tracks the datapath's idea of the flow equivalent to
 197  * the facet.  When the kernel module (or other dpif implementation) and
 198  * Open vSwitch userspace agree on the definition of a flow key, there
 199  * is exactly one subfacet per facet.  If the dpif implementation
 200  * supports more-specific flow matching than userspace, however, a facet
 201  * can have more than one subfacet.  Examples include the dpif
 202  * implementation not supporting the same wildcards as userspace or some
 203  * distinction in flow that userspace simply doesn't understand.
 204  *
 205  * Flow expiration works in terms of subfacets, so a facet must have at
 206  * least one subfacet or it will never expire, leaking memory. */
 207 struct facet {
 208     /* Owners. */
 209     struct hmap_node hmap_node;  /* In owning ofproto's 'facets' hmap. */
 210     struct ofproto_dpif *ofproto;
 211
 212     /* Owned data. */
 213     struct list subfacets;
 214     long long int used;         /* Time last used; time created if not used. */
 215
 216     /* Key. */
 217     struct flow flow;           /* Flow of the creating subfacet. */
 218     struct cls_rule cr;         /* In 'ofproto_dpif's facets classifier. */
 219
 220     /* These statistics:
 221      *
 222      *   - Do include packets and bytes sent "by hand", e.g. with
 223      *     dpif_execute().
 224      *
 225      *   - Do include packets and bytes that were obtained from the datapath
 226      *     when a subfacet's statistics were reset (e.g. dpif_flow_put() with
 227      *     DPIF_FP_ZERO_STATS).
 228      *
 229      *   - Do not include packets or bytes that can be obtained from the
 230      *     datapath for any existing subfacet.
 231      */
 232     uint64_t packet_count;       /* Number of packets received. */
 233     uint64_t byte_count;         /* Number of bytes received. */
 234
 235     /* Resubmit statistics. */
 236     uint64_t prev_packet_count;  /* Number of packets from last stats push. */
 237     uint64_t prev_byte_count;    /* Number of bytes from last stats push. */
 238     long long int prev_used;     /* Used time from last stats push. */
 239
 240     /* Accounting. */
 241     uint64_t accounted_bytes;    /* Bytes processed by facet_account(). */
 242     struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
 243     uint8_t tcp_flags;           /* TCP flags seen for this 'rule'. */
 244
 245     struct xlate_out xout;
 246
 247     /* Storage for a single subfacet, to reduce malloc() time and space
 248      * overhead.  (A facet always has at least one subfacet and in the common
 249      * case has exactly one subfacet.  However, 'one_subfacet' may not
 250      * always be valid, since it could have been removed after newer
 251      * subfacets were pushed onto the 'subfacets' list.) */
 252     struct subfacet one_subfacet;
 253
 254     long long int learn_rl;      /* Rate limiter for facet_learn(). */
 255 };
 256
 257 static struct facet *facet_create(const struct flow_miss *);
 258 static void facet_remove(struct facet *);
 259 static void facet_free(struct facet *);
 260
 261 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
 262 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
 263                                         const struct flow *);
 264 static bool facet_revalidate(struct facet *);
 265 static bool facet_check_consistency(struct facet *);
 266
 267 static void facet_flush_stats(struct facet *);
 268
 269 static void facet_reset_counters(struct facet *);
 270 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
 271                             struct dpif_flow_stats *, bool may_learn);
 272 static void facet_push_stats(struct facet *, bool may_learn);
 273 static void facet_learn(struct facet *);
 274 static void facet_account(struct facet *);
 275 static void push_all_stats(void);
 276
 277 static bool facet_is_controller_flow(struct facet *);
 278
 279 struct ofport_dpif {
 280     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 281     struct ofport up;
 282
 283     odp_port_t odp_port;
 284     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 285     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
 286     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 287     struct bfd *bfd;            /* BFD, if any. */
 288     bool may_enable;            /* May be enabled in bonds. */
 289     bool is_tunnel;             /* This port is a tunnel. */
 290     long long int carrier_seq;  /* Carrier status changes. */
 291     struct ofport_dpif *peer;   /* Peer if patch port. */
 292
 293     /* Spanning tree. */
 294     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 295     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 296     long long int stp_state_entered;
 297
 298     /* Queue to DSCP mapping. */
 299     struct ofproto_port_queue *qdscp;
 300     size_t n_qdscp;
 301
 302     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 303      *
 304      * This is deprecated.  It is only for compatibility with broken device
 305      * drivers in old versions of Linux that do not properly support VLANs when
 306      * VLAN devices are not used.  When broken device drivers are no longer in
 307      * widespread use, we will delete these interfaces. */
 308     ofp_port_t realdev_ofp_port;
 309     int vlandev_vid;
 310 };
 311
 312 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 313  *
 314  * This is deprecated.  It is only for compatibility with broken device drivers
 315  * in old versions of Linux that do not properly support VLANs when VLAN
 316  * devices are not used.  When broken device drivers are no longer in
 317  * widespread use, we will delete these interfaces. */
 318 struct vlan_splinter {
 319     struct hmap_node realdev_vid_node;
 320     struct hmap_node vlandev_node;
 321     ofp_port_t realdev_ofp_port;
 322     ofp_port_t vlandev_ofp_port;
 323     int vid;
 324 };
 325
 326 static void vsp_remove(struct ofport_dpif *);
 327 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 328
 329 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 330                                        ofp_port_t);
 331
 332 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 333                                        odp_port_t);
 334
 335 static struct ofport_dpif *
 336 ofport_dpif_cast(const struct ofport *ofport)
 337 {
 338     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 339 }
 340
 341 static void port_run(struct ofport_dpif *);
 342 static void port_run_fast(struct ofport_dpif *);
 343 static void port_wait(struct ofport_dpif *);
 344 static int set_bfd(struct ofport *, const struct smap *);
 345 static int set_cfm(struct ofport *, const struct cfm_settings *);
 346 static void ofport_update_peer(struct ofport_dpif *);
 347 static void run_fast_rl(void);
 348 static int run_fast(struct ofproto *);
 349
 350 struct dpif_completion {
 351     struct list list_node;
 352     struct ofoperation *op;
 353 };
 354
 355 /* Reasons that we might need to revalidate every facet, and corresponding
 356  * coverage counters.
 357  *
 358  * A value of 0 means that there is no need to revalidate.
 359  *
 360  * It would be nice to have some cleaner way to integrate with coverage
 361  * counters, but with only a few reasons I guess this is good enough for
 362  * now. */
 363 enum revalidate_reason {
 364     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 365     REV_STP,                   /* Spanning tree protocol port status change. */
 366     REV_BOND,                  /* Bonding changed. */
 367     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 368     REV_FLOW_TABLE,            /* Flow table changed. */
 369     REV_MAC_LEARNING,          /* Mac learning changed. */
 370     REV_INCONSISTENCY          /* Facet self-check failed. */
 371 };
 372 COVERAGE_DEFINE(rev_reconfigure);
 373 COVERAGE_DEFINE(rev_stp);
 374 COVERAGE_DEFINE(rev_bond);
 375 COVERAGE_DEFINE(rev_port_toggled);
 376 COVERAGE_DEFINE(rev_flow_table);
 377 COVERAGE_DEFINE(rev_mac_learning);
 378 COVERAGE_DEFINE(rev_inconsistency);
 379
 380 struct avg_subfacet_rates {
 381     double add_rate;   /* Moving average of new flows created per minute. */
 382     double del_rate;   /* Moving average of flows deleted per minute. */
 383 };
 384
 385 /* All datapaths of a given type share a single dpif backer instance. */
 386 struct dpif_backer {
 387     char *type;
 388     int refcount;
 389     struct dpif *dpif;
 390     struct udpif *udpif;
 391     struct timer next_expiration;
 392
 393     struct ovs_rwlock odp_to_ofport_lock;
 394     struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
 395
 396     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 397
 398     /* Facet revalidation flags applying to facets which use this backer. */
 399     enum revalidate_reason need_revalidate; /* Revalidate every facet. */
 400
 401     struct hmap drop_keys; /* Set of dropped odp keys. */
 402     bool recv_set_enable; /* Enables or disables receiving packets. */
 403
 404     struct hmap subfacets;
 405     struct governor *governor;
 406
 407     /* Subfacet statistics.
 408      *
 409      * These keep track of the total number of subfacets added and deleted and
 410      * flow life span.  They are useful for computing the flow rates stats
 411      * exposed via "ovs-appctl dpif/show".  The goal is to learn about
 412      * traffic patterns in ways that we can use later to improve Open vSwitch
 413      * performance in new situations.  */
 414     long long int created;           /* Time when it is created. */
 415     unsigned max_n_subfacet;         /* Maximum number of flows */
 416     unsigned avg_n_subfacet;         /* Average number of flows. */
 417     long long int avg_subfacet_life; /* Average life span of subfacets. */
 418
 419     /* The average number of subfacets... */
 420     struct avg_subfacet_rates hourly;   /* ...over the last hour. */
 421     struct avg_subfacet_rates daily;    /* ...over the last day. */
 422     struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
 423     long long int last_minute;          /* Last time 'hourly' was updated. */
 424
 425     /* Number of subfacets added or deleted since 'last_minute'. */
 426     unsigned subfacet_add_count;
 427     unsigned subfacet_del_count;
 428
 429     /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
 430     unsigned long long int total_subfacet_add_count;
 431     unsigned long long int total_subfacet_del_count;
 432
 433     /* Number of upcall handling threads. */
 434     unsigned int n_handler_threads;
 435 };
 436
 437 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 438 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 439
 440 static void drop_key_clear(struct dpif_backer *);
 441 static void update_moving_averages(struct dpif_backer *backer);
 442
 443 struct ofproto_dpif {
 444     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 445     struct ofproto up;
 446     struct dpif_backer *backer;
 447
 448     /* Special OpenFlow rules. */
 449     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 450     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 451     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 452
 453     /* Bridging. */
 454     struct netflow *netflow;
 455     struct dpif_sflow *sflow;
 456     struct dpif_ipfix *ipfix;
 457     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 458     struct mac_learning *ml;
 459     bool has_bonded_bundles;
 460     struct mbridge *mbridge;
 461
 462     /* Facets. */
 463     struct classifier facets;     /* Contains 'struct facet's. */
 464     long long int consistency_rl;
 465
 466     /* Support for debugging async flow mods. */
 467     struct list completions;
 468
 469     struct netdev_stats stats; /* To account packets generated and consumed in
 470                                 * userspace. */
 471
 472     /* Spanning tree. */
 473     struct stp *stp;
 474     long long int stp_last_tick;
 475
 476     /* VLAN splinters. */
 477     struct ovs_mutex vsp_mutex;
 478     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 479     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 480
 481     /* Ports. */
 482     struct sset ports;             /* Set of standard port names. */
 483     struct sset ghost_ports;       /* Ports with no datapath port. */
 484     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 485     int port_poll_errno;           /* Last errno for port_poll() reply. */
 486
 487     /* Per ofproto's dpif stats. */
 488     uint64_t n_hit;
 489     uint64_t n_missed;
 490
 491     /* Work queues. */
 492     struct ovs_mutex flow_mod_mutex;
 493     struct list flow_mods OVS_GUARDED;
 494     size_t n_flow_mods OVS_GUARDED;
 495
 496     struct ovs_mutex pin_mutex;
 497     struct list pins OVS_GUARDED;
 498     size_t n_pins OVS_GUARDED;
 499 };
 500
 501 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"?  (Useful only
 502  * for debugging the asynchronous flow_mod implementation.) */
 503 static bool clogged;
 504
 505 /* By default, flows in the datapath are wildcarded (megaflows).  They
 506  * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
 507 static bool enable_megaflows = true;
 508
 509 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 510 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 511
 512 static void ofproto_dpif_unixctl_init(void);
 513
 514 static inline struct ofproto_dpif *
 515 ofproto_dpif_cast(const struct ofproto *ofproto)
 516 {
 517     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 518     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 519 }
 520
 521 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
 522                                         ofp_port_t ofp_port);
 523 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
 524                           const struct ofpbuf *packet, struct ds *);
 525
 526 /* Upcalls. */
 527 static void handle_upcalls(struct dpif_backer *);
 528
 529 /* Flow expiration. */
 530 static int expire(struct dpif_backer *);
 531
 532 /* NetFlow. */
 533 static void send_netflow_active_timeouts(struct ofproto_dpif *);
 534
 535 /* Utilities. */
 536 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
 537
 538 /* Global variables. */
 539 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 540
 541 /* Initial mappings of port to bridge mappings. */
 542 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 543
 544 /* Executes and takes ownership of 'fm'. */
 545 void
 546 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 547                       struct ofputil_flow_mod *fm)
 548 {
 549     ovs_mutex_lock(&ofproto->flow_mod_mutex);
 550     if (ofproto->n_flow_mods > 1024) {
 551         ovs_mutex_unlock(&ofproto->flow_mod_mutex);
 552         COVERAGE_INC(flow_mod_overflow);
 553         free(fm->ofpacts);
 554         free(fm);
 555         return;
 556     }
 557
 558     list_push_back(&ofproto->flow_mods, &fm->list_node);
 559     ofproto->n_flow_mods++;
 560     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
 561 }
 562
 563 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 564  * Takes ownership of 'pin' and pin->packet. */
 565 void
 566 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 567                             struct ofputil_packet_in *pin)
 568 {
 569     ovs_mutex_lock(&ofproto->pin_mutex);
 570     if (ofproto->n_pins > 1024) {
 571         ovs_mutex_unlock(&ofproto->pin_mutex);
 572         COVERAGE_INC(packet_in_overflow);
 573         free(CONST_CAST(void *, pin->packet));
 574         free(pin);
 575         return;
 576     }
 577
 578     list_push_back(&ofproto->pins, &pin->list_node);
 579     ofproto->n_pins++;
 580     ovs_mutex_unlock(&ofproto->pin_mutex);
 581 }
 582 \f
 583 /* Factory functions. */
 584
 585 static void
 586 init(const struct shash *iface_hints)
 587 {
 588     struct shash_node *node;
 589
 590     /* Make a local copy, since we don't own 'iface_hints' elements. */
 591     SHASH_FOR_EACH(node, iface_hints) {
 592         const struct iface_hint *orig_hint = node->data;
 593         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 594
 595         new_hint->br_name = xstrdup(orig_hint->br_name);
 596         new_hint->br_type = xstrdup(orig_hint->br_type);
 597         new_hint->ofp_port = orig_hint->ofp_port;
 598
 599         shash_add(&init_ofp_ports, node->name, new_hint);
 600     }
 601 }
 602
 603 static void
 604 enumerate_types(struct sset *types)
 605 {
 606     dp_enumerate_types(types);
 607 }
 608
 609 static int
 610 enumerate_names(const char *type, struct sset *names)
 611 {
 612     struct ofproto_dpif *ofproto;
 613
 614     sset_clear(names);
 615     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 616         if (strcmp(type, ofproto->up.type)) {
 617             continue;
 618         }
 619         sset_add(names, ofproto->up.name);
 620     }
 621
 622     return 0;
 623 }
 624
 625 static int
 626 del(const char *type, const char *name)
 627 {
 628     struct dpif *dpif;
 629     int error;
 630
 631     error = dpif_open(name, type, &dpif);
 632     if (!error) {
 633         error = dpif_delete(dpif);
 634         dpif_close(dpif);
 635     }
 636     return error;
 637 }
 638 \f
 639 static const char *
 640 port_open_type(const char *datapath_type, const char *port_type)
 641 {
 642     return dpif_port_open_type(datapath_type, port_type);
 643 }
 644
 645 /* Type functions. */
 646
 647 static void process_dpif_port_changes(struct dpif_backer *);
 648 static void process_dpif_all_ports_changed(struct dpif_backer *);
 649 static void process_dpif_port_change(struct dpif_backer *,
 650                                      const char *devname);
 651 static void process_dpif_port_error(struct dpif_backer *, int error);
 652
 653 static struct ofproto_dpif *
 654 lookup_ofproto_dpif_by_port_name(const char *name)
 655 {
 656     struct ofproto_dpif *ofproto;
 657
 658     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 659         if (sset_contains(&ofproto->ports, name)) {
 660             return ofproto;
 661         }
 662     }
 663
 664     return NULL;
 665 }
 666
 667 static int
 668 type_run(const char *type)
 669 {
 670     static long long int push_timer = LLONG_MIN;
 671     struct dpif_backer *backer;
 672
 673     backer = shash_find_data(&all_dpif_backers, type);
 674     if (!backer) {
 675         /* This is not necessarily a problem, since backers are only
 676          * created on demand. */
 677         return 0;
 678     }
 679
 680     dpif_run(backer->dpif);
 681
 682     /* The most natural place to push facet statistics is when they're pulled
 683      * from the datapath.  However, when there are many flows in the datapath,
 684      * this expensive operation can occur so frequently, that it reduces our
 685      * ability to quickly set up flows.  To reduce the cost, we push statistics
 686      * here instead. */
 687     if (time_msec() > push_timer) {
 688         push_timer = time_msec() + 2000;
 689         push_all_stats();
 690     }
 691
 692     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 693      * and the configuration has now changed to "false", enable receiving
 694      * packets from the datapath. */
 695     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 696         int error;
 697
 698         backer->recv_set_enable = true;
 699
 700         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 701         if (error) {
 702             udpif_recv_set(backer->udpif, 0, false);
 703             VLOG_ERR("Failed to enable receiving packets in dpif.");
 704             return error;
 705         }
 706         udpif_recv_set(backer->udpif, n_handler_threads,
 707                        backer->recv_set_enable);
 708         dpif_flow_flush(backer->dpif);
 709         backer->need_revalidate = REV_RECONFIGURE;
 710     }
 711
 712     /* If the n_handler_threads is reconfigured, call udpif_recv_set()
 713      * to reset the handler threads. */
 714     if (backer->n_handler_threads != n_handler_threads) {
 715         udpif_recv_set(backer->udpif, n_handler_threads,
 716                        backer->recv_set_enable);
 717         backer->n_handler_threads = n_handler_threads;
 718     }
 719
 720     if (backer->need_revalidate) {
 721         struct ofproto_dpif *ofproto;
 722         struct simap_node *node;
 723         struct simap tmp_backers;
 724
 725         /* Handle tunnel garbage collection. */
 726         simap_init(&tmp_backers);
 727         simap_swap(&backer->tnl_backers, &tmp_backers);
 728
 729         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 730             struct ofport_dpif *iter;
 731
 732             if (backer != ofproto->backer) {
 733                 continue;
 734             }
 735
 736             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 737                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 738                 const char *dp_port;
 739
 740                 if (!iter->is_tunnel) {
 741                     continue;
 742                 }
 743
 744                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 745                                                      namebuf, sizeof namebuf);
 746                 node = simap_find(&tmp_backers, dp_port);
 747                 if (node) {
 748                     simap_put(&backer->tnl_backers, dp_port, node->data);
 749                     simap_delete(&tmp_backers, node);
 750                     node = simap_find(&backer->tnl_backers, dp_port);
 751                 } else {
 752                     node = simap_find(&backer->tnl_backers, dp_port);
 753                     if (!node) {
 754                         odp_port_t odp_port = ODPP_NONE;
 755
 756                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 757                                            &odp_port)) {
 758                             simap_put(&backer->tnl_backers, dp_port,
 759                                       odp_to_u32(odp_port));
 760                             node = simap_find(&backer->tnl_backers, dp_port);
 761                         }
 762                     }
 763                 }
 764
 765                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 766                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 767                                          iter->odp_port)) {
 768                     backer->need_revalidate = REV_RECONFIGURE;
 769                 }
 770             }
 771         }
 772
 773         SIMAP_FOR_EACH (node, &tmp_backers) {
 774             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 775         }
 776         simap_destroy(&tmp_backers);
 777
 778         switch (backer->need_revalidate) {
 779         case REV_RECONFIGURE:   COVERAGE_INC(rev_reconfigure);   break;
 780         case REV_STP:           COVERAGE_INC(rev_stp);           break;
 781         case REV_BOND:          COVERAGE_INC(rev_bond);          break;
 782         case REV_PORT_TOGGLED:  COVERAGE_INC(rev_port_toggled);  break;
 783         case REV_FLOW_TABLE:    COVERAGE_INC(rev_flow_table);    break;
 784         case REV_MAC_LEARNING:  COVERAGE_INC(rev_mac_learning);  break;
 785         case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
 786         }
 787         backer->need_revalidate = 0;
 788
 789         /* Clear the drop_keys in case we should now be accepting some
 790          * formerly dropped flows. */
 791         drop_key_clear(backer);
 792
 793         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 794             struct facet *facet, *next;
 795             struct ofport_dpif *ofport;
 796             struct cls_cursor cursor;
 797             struct ofbundle *bundle;
 798
 799             if (ofproto->backer != backer) {
 800                 continue;
 801             }
 802
 803             ovs_rwlock_wrlock(&xlate_rwlock);
 804             xlate_ofproto_set(ofproto, ofproto->up.name,
 805                               ofproto->backer->dpif, ofproto->miss_rule,
 806                               ofproto->no_packet_in_rule, ofproto->ml,
 807                               ofproto->stp, ofproto->mbridge,
 808                               ofproto->sflow, ofproto->ipfix,
 809                               ofproto->up.frag_handling,
 810                               ofproto->up.forward_bpdu,
 811                               connmgr_has_in_band(ofproto->up.connmgr),
 812                               ofproto->netflow != NULL);
 813
 814             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 815                 xlate_bundle_set(ofproto, bundle, bundle->name,
 816                                  bundle->vlan_mode, bundle->vlan,
 817                                  bundle->trunks, bundle->use_priority_tags,
 818                                  bundle->bond, bundle->lacp,
 819                                  bundle->floodable);
 820             }
 821
 822             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 823                 int stp_port = ofport->stp_port
 824                     ? stp_port_no(ofport->stp_port)
 825                     : -1;
 826                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 827                                  ofport->up.ofp_port, ofport->odp_port,
 828                                  ofport->up.netdev, ofport->cfm,
 829                                  ofport->bfd, ofport->peer, stp_port,
 830                                  ofport->qdscp, ofport->n_qdscp,
 831                                  ofport->up.pp.config, ofport->is_tunnel,
 832                                  ofport->may_enable);
 833             }
 834             ovs_rwlock_unlock(&xlate_rwlock);
 835
 836             /* Only ofproto-dpif cares about the facet classifier so we just
 837              * lock cls_cursor_init() to appease the thread safety analysis. */
 838             ovs_rwlock_rdlock(&ofproto->facets.rwlock);
 839             cls_cursor_init(&cursor, &ofproto->facets, NULL);
 840             ovs_rwlock_unlock(&ofproto->facets.rwlock);
 841             CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
 842                 facet_revalidate(facet);
 843                 run_fast_rl();
 844             }
 845         }
 846
 847         udpif_revalidate(backer->udpif);
 848     }
 849
 850     if (!backer->recv_set_enable) {
 851         /* Wake up before a max of 1000ms. */
 852         timer_set_duration(&backer->next_expiration, 1000);
 853     } else if (timer_expired(&backer->next_expiration)) {
 854         int delay = expire(backer);
 855         timer_set_duration(&backer->next_expiration, delay);
 856     }
 857
 858     process_dpif_port_changes(backer);
 859
 860     if (backer->governor) {
 861         size_t n_subfacets;
 862
 863         governor_run(backer->governor);
 864
 865         /* If the governor has shrunk to its minimum size and the number of
 866          * subfacets has dwindled, then drop the governor entirely.
 867          *
 868          * For hysteresis, the number of subfacets to drop the governor is
 869          * smaller than the number needed to trigger its creation. */
 870         n_subfacets = hmap_count(&backer->subfacets);
 871         if (n_subfacets * 4 < flow_eviction_threshold
 872             && governor_is_idle(backer->governor)) {
 873             governor_destroy(backer->governor);
 874             backer->governor = NULL;
 875         }
 876     }
 877
 878     return 0;
 879 }
 880
 881 /* Check for and handle port changes in 'backer''s dpif. */
 882 static void
 883 process_dpif_port_changes(struct dpif_backer *backer)
 884 {
 885     for (;;) {
 886         char *devname;
 887         int error;
 888
 889         error = dpif_port_poll(backer->dpif, &devname);
 890         switch (error) {
 891         case EAGAIN:
 892             return;
 893
 894         case ENOBUFS:
 895             process_dpif_all_ports_changed(backer);
 896             break;
 897
 898         case 0:
 899             process_dpif_port_change(backer, devname);
 900             free(devname);
 901             break;
 902
 903         default:
 904             process_dpif_port_error(backer, error);
 905             break;
 906         }
 907     }
 908 }
 909
 910 static void
 911 process_dpif_all_ports_changed(struct dpif_backer *backer)
 912 {
 913     struct ofproto_dpif *ofproto;
 914     struct dpif_port dpif_port;
 915     struct dpif_port_dump dump;
 916     struct sset devnames;
 917     const char *devname;
 918
 919     sset_init(&devnames);
 920     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 921         if (ofproto->backer == backer) {
 922             struct ofport *ofport;
 923
 924             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 925                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 926             }
 927         }
 928     }
 929     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 930         sset_add(&devnames, dpif_port.name);
 931     }
 932
 933     SSET_FOR_EACH (devname, &devnames) {
 934         process_dpif_port_change(backer, devname);
 935     }
 936     sset_destroy(&devnames);
 937 }
 938
 939 static void
 940 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 941 {
 942     struct ofproto_dpif *ofproto;
 943     struct dpif_port port;
 944
 945     /* Don't report on the datapath's device. */
 946     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 947         return;
 948     }
 949
 950     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 951                    &all_ofproto_dpifs) {
 952         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 953             return;
 954         }
 955     }
 956
 957     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 958     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 959         /* The port was removed.  If we know the datapath,
 960          * report it through poll_set().  If we don't, it may be
 961          * notifying us of a removal we initiated, so ignore it.
 962          * If there's a pending ENOBUFS, let it stand, since
 963          * everything will be reevaluated. */
 964         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 965             sset_add(&ofproto->port_poll_set, devname);
 966             ofproto->port_poll_errno = 0;
 967         }
 968     } else if (!ofproto) {
 969         /* The port was added, but we don't know with which
 970          * ofproto we should associate it.  Delete it. */
 971         dpif_port_del(backer->dpif, port.port_no);
 972     } else {
 973         struct ofport_dpif *ofport;
 974
 975         ofport = ofport_dpif_cast(shash_find_data(
 976                                       &ofproto->up.port_by_name, devname));
 977         if (ofport
 978             && ofport->odp_port != port.port_no
 979             && !odp_port_to_ofport(backer, port.port_no))
 980         {
 981             /* 'ofport''s datapath port number has changed from
 982              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 983              * structures to match. */
 984             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 985             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 986             ofport->odp_port = port.port_no;
 987             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 988                         hash_odp_port(port.port_no));
 989             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 990             backer->need_revalidate = REV_RECONFIGURE;
 991         }
 992     }
 993     dpif_port_destroy(&port);
 994 }
 995
 996 /* Propagate 'error' to all ofprotos based on 'backer'. */
 997 static void
 998 process_dpif_port_error(struct dpif_backer *backer, int error)
 999 {
1000     struct ofproto_dpif *ofproto;
1001
1002     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1003         if (ofproto->backer == backer) {
1004             sset_clear(&ofproto->port_poll_set);
1005             ofproto->port_poll_errno = error;
1006         }
1007     }
1008 }
1009
1010 static int
1011 dpif_backer_run_fast(struct dpif_backer *backer)
1012 {
1013     udpif_run(backer->udpif);
1014     handle_upcalls(backer);
1015
1016     return 0;
1017 }
1018
1019 static int
1020 type_run_fast(const char *type)
1021 {
1022     struct dpif_backer *backer;
1023
1024     backer = shash_find_data(&all_dpif_backers, type);
1025     if (!backer) {
1026         /* This is not necessarily a problem, since backers are only
1027          * created on demand. */
1028         return 0;
1029     }
1030
1031     return dpif_backer_run_fast(backer);
1032 }
1033
1034 static void
1035 run_fast_rl(void)
1036 {
1037     static long long int port_rl = LLONG_MIN;
1038
1039     if (time_msec() >= port_rl) {
1040         struct ofproto_dpif *ofproto;
1041
1042         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1043             run_fast(&ofproto->up);
1044         }
1045         port_rl = time_msec() + 200;
1046     }
1047 }
1048
1049 static void
1050 type_wait(const char *type)
1051 {
1052     struct dpif_backer *backer;
1053
1054     backer = shash_find_data(&all_dpif_backers, type);
1055     if (!backer) {
1056         /* This is not necessarily a problem, since backers are only
1057          * created on demand. */
1058         return;
1059     }
1060
1061     if (backer->governor) {
1062         governor_wait(backer->governor);
1063     }
1064
1065     timer_wait(&backer->next_expiration);
1066     dpif_wait(backer->dpif);
1067     udpif_wait(backer->udpif);
1068 }
1069 \f
1070 /* Basic life-cycle. */
1071
1072 static int add_internal_flows(struct ofproto_dpif *);
1073
1074 static struct ofproto *
1075 alloc(void)
1076 {
1077     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1078     return &ofproto->up;
1079 }
1080
1081 static void
1082 dealloc(struct ofproto *ofproto_)
1083 {
1084     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1085     free(ofproto);
1086 }
1087
1088 static void
1089 close_dpif_backer(struct dpif_backer *backer)
1090 {
1091     struct shash_node *node;
1092
1093     ovs_assert(backer->refcount > 0);
1094
1095     if (--backer->refcount) {
1096         return;
1097     }
1098
1099     drop_key_clear(backer);
1100     hmap_destroy(&backer->drop_keys);
1101
1102     simap_destroy(&backer->tnl_backers);
1103     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1104     hmap_destroy(&backer->odp_to_ofport_map);
1105     node = shash_find(&all_dpif_backers, backer->type);
1106     free(backer->type);
1107     shash_delete(&all_dpif_backers, node);
1108     udpif_destroy(backer->udpif);
1109     dpif_close(backer->dpif);
1110
1111     ovs_assert(hmap_is_empty(&backer->subfacets));
1112     hmap_destroy(&backer->subfacets);
1113     governor_destroy(backer->governor);
1114
1115     free(backer);
1116 }
1117
1118 /* Datapath port slated for removal from datapath. */
1119 struct odp_garbage {
1120     struct list list_node;
1121     odp_port_t odp_port;
1122 };
1123
1124 static int
1125 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1126 {
1127     struct dpif_backer *backer;
1128     struct dpif_port_dump port_dump;
1129     struct dpif_port port;
1130     struct shash_node *node;
1131     struct list garbage_list;
1132     struct odp_garbage *garbage, *next;
1133     struct sset names;
1134     char *backer_name;
1135     const char *name;
1136     int error;
1137
1138     backer = shash_find_data(&all_dpif_backers, type);
1139     if (backer) {
1140         backer->refcount++;
1141         *backerp = backer;
1142         return 0;
1143     }
1144
1145     backer_name = xasprintf("ovs-%s", type);
1146
1147     /* Remove any existing datapaths, since we assume we're the only
1148      * userspace controlling the datapath. */
1149     sset_init(&names);
1150     dp_enumerate_names(type, &names);
1151     SSET_FOR_EACH(name, &names) {
1152         struct dpif *old_dpif;
1153
1154         /* Don't remove our backer if it exists. */
1155         if (!strcmp(name, backer_name)) {
1156             continue;
1157         }
1158
1159         if (dpif_open(name, type, &old_dpif)) {
1160             VLOG_WARN("couldn't open old datapath %s to remove it", name);
1161         } else {
1162             dpif_delete(old_dpif);
1163             dpif_close(old_dpif);
1164         }
1165     }
1166     sset_destroy(&names);
1167
1168     backer = xmalloc(sizeof *backer);
1169
1170     error = dpif_create_and_open(backer_name, type, &backer->dpif);
1171     free(backer_name);
1172     if (error) {
1173         VLOG_ERR("failed to open datapath of type %s: %s", type,
1174                  ovs_strerror(error));
1175         free(backer);
1176         return error;
1177     }
1178     backer->udpif = udpif_create(backer, backer->dpif);
1179
1180     backer->type = xstrdup(type);
1181     backer->governor = NULL;
1182     backer->refcount = 1;
1183     hmap_init(&backer->odp_to_ofport_map);
1184     ovs_rwlock_init(&backer->odp_to_ofport_lock);
1185     hmap_init(&backer->drop_keys);
1186     hmap_init(&backer->subfacets);
1187     timer_set_duration(&backer->next_expiration, 1000);
1188     backer->need_revalidate = 0;
1189     simap_init(&backer->tnl_backers);
1190     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1191     *backerp = backer;
1192
1193     if (backer->recv_set_enable) {
1194         dpif_flow_flush(backer->dpif);
1195     }
1196
1197     /* Loop through the ports already on the datapath and remove any
1198      * that we don't need anymore. */
1199     list_init(&garbage_list);
1200     dpif_port_dump_start(&port_dump, backer->dpif);
1201     while (dpif_port_dump_next(&port_dump, &port)) {
1202         node = shash_find(&init_ofp_ports, port.name);
1203         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1204             garbage = xmalloc(sizeof *garbage);
1205             garbage->odp_port = port.port_no;
1206             list_push_front(&garbage_list, &garbage->list_node);
1207         }
1208     }
1209     dpif_port_dump_done(&port_dump);
1210
1211     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1212         dpif_port_del(backer->dpif, garbage->odp_port);
1213         list_remove(&garbage->list_node);
1214         free(garbage);
1215     }
1216
1217     shash_add(&all_dpif_backers, type, backer);
1218
1219     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1220     if (error) {
1221         VLOG_ERR("failed to listen on datapath of type %s: %s",
1222                  type, ovs_strerror(error));
1223         close_dpif_backer(backer);
1224         return error;
1225     }
1226     udpif_recv_set(backer->udpif, n_handler_threads,
1227                    backer->recv_set_enable);
1228     backer->n_handler_threads = n_handler_threads;
1229
1230     backer->max_n_subfacet = 0;
1231     backer->created = time_msec();
1232     backer->last_minute = backer->created;
1233     memset(&backer->hourly, 0, sizeof backer->hourly);
1234     memset(&backer->daily, 0, sizeof backer->daily);
1235     memset(&backer->lifetime, 0, sizeof backer->lifetime);
1236     backer->subfacet_add_count = 0;
1237     backer->subfacet_del_count = 0;
1238     backer->total_subfacet_add_count = 0;
1239     backer->total_subfacet_del_count = 0;
1240     backer->avg_n_subfacet = 0;
1241     backer->avg_subfacet_life = 0;
1242
1243     return error;
1244 }
1245
1246 static int
1247 construct(struct ofproto *ofproto_)
1248 {
1249     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1250     struct shash_node *node, *next;
1251     uint32_t max_ports;
1252     int error;
1253
1254     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1255     if (error) {
1256         return error;
1257     }
1258
1259     max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1260     ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX)));
1261
1262     ofproto->netflow = NULL;
1263     ofproto->sflow = NULL;
1264     ofproto->ipfix = NULL;
1265     ofproto->stp = NULL;
1266     hmap_init(&ofproto->bundles);
1267     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1268     ofproto->mbridge = mbridge_create();
1269     ofproto->has_bonded_bundles = false;
1270     ovs_mutex_init(&ofproto->vsp_mutex);
1271
1272     classifier_init(&ofproto->facets);
1273     ofproto->consistency_rl = LLONG_MIN;
1274
1275     list_init(&ofproto->completions);
1276
1277     ovs_mutex_init(&ofproto->flow_mod_mutex);
1278     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1279     list_init(&ofproto->flow_mods);
1280     ofproto->n_flow_mods = 0;
1281     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1282
1283     ovs_mutex_init(&ofproto->pin_mutex);
1284     ovs_mutex_lock(&ofproto->pin_mutex);
1285     list_init(&ofproto->pins);
1286     ofproto->n_pins = 0;
1287     ovs_mutex_unlock(&ofproto->pin_mutex);
1288
1289     ofproto_dpif_unixctl_init();
1290
1291     hmap_init(&ofproto->vlandev_map);
1292     hmap_init(&ofproto->realdev_vid_map);
1293
1294     sset_init(&ofproto->ports);
1295     sset_init(&ofproto->ghost_ports);
1296     sset_init(&ofproto->port_poll_set);
1297     ofproto->port_poll_errno = 0;
1298
1299     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1300         struct iface_hint *iface_hint = node->data;
1301
1302         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1303             /* Check if the datapath already has this port. */
1304             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1305                 sset_add(&ofproto->ports, node->name);
1306             }
1307
1308             free(iface_hint->br_name);
1309             free(iface_hint->br_type);
1310             free(iface_hint);
1311             shash_delete(&init_ofp_ports, node);
1312         }
1313     }
1314
1315     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1316                 hash_string(ofproto->up.name, 0));
1317     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1318
1319     ofproto_init_tables(ofproto_, N_TABLES);
1320     error = add_internal_flows(ofproto);
1321     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1322
1323     ofproto->n_hit = 0;
1324     ofproto->n_missed = 0;
1325
1326     return error;
1327 }
1328
1329 static int
1330 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1331                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1332 {
1333     struct ofputil_flow_mod fm;
1334     int error;
1335
1336     match_init_catchall(&fm.match);
1337     fm.priority = 0;
1338     match_set_reg(&fm.match, 0, id);
1339     fm.new_cookie = htonll(0);
1340     fm.cookie = htonll(0);
1341     fm.cookie_mask = htonll(0);
1342     fm.modify_cookie = false;
1343     fm.table_id = TBL_INTERNAL;
1344     fm.command = OFPFC_ADD;
1345     fm.idle_timeout = 0;
1346     fm.hard_timeout = 0;
1347     fm.buffer_id = 0;
1348     fm.out_port = 0;
1349     fm.flags = 0;
1350     fm.ofpacts = ofpacts->data;
1351     fm.ofpacts_len = ofpacts->size;
1352
1353     error = ofproto_flow_mod(&ofproto->up, &fm);
1354     if (error) {
1355         VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1356                     id, ofperr_to_string(error));
1357         return error;
1358     }
1359
1360     if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1361                                   rulep)) {
1362         ovs_rwlock_unlock(&(*rulep)->up.evict);
1363     } else {
1364         NOT_REACHED();
1365     }
1366
1367     return 0;
1368 }
1369
1370 static int
1371 add_internal_flows(struct ofproto_dpif *ofproto)
1372 {
1373     struct ofpact_controller *controller;
1374     uint64_t ofpacts_stub[128 / 8];
1375     struct ofpbuf ofpacts;
1376     int error;
1377     int id;
1378
1379     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1380     id = 1;
1381
1382     controller = ofpact_put_CONTROLLER(&ofpacts);
1383     controller->max_len = UINT16_MAX;
1384     controller->controller_id = 0;
1385     controller->reason = OFPR_NO_MATCH;
1386     ofpact_pad(&ofpacts);
1387
1388     error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1389     if (error) {
1390         return error;
1391     }
1392
1393     ofpbuf_clear(&ofpacts);
1394     error = add_internal_flow(ofproto, id++, &ofpacts,
1395                               &ofproto->no_packet_in_rule);
1396     if (error) {
1397         return error;
1398     }
1399
1400     error = add_internal_flow(ofproto, id++, &ofpacts,
1401                               &ofproto->drop_frags_rule);
1402     return error;
1403 }
1404
1405 static void
1406 complete_operations(struct ofproto_dpif *ofproto)
1407 {
1408     struct dpif_completion *c, *next;
1409
1410     LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1411         ofoperation_complete(c->op, 0);
1412         list_remove(&c->list_node);
1413         free(c);
1414     }
1415 }
1416
1417 static void
1418 destruct(struct ofproto *ofproto_)
1419 {
1420     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1421     struct rule_dpif *rule, *next_rule;
1422     struct ofputil_packet_in *pin, *next_pin;
1423     struct ofputil_flow_mod *fm, *next_fm;
1424     struct facet *facet, *next_facet;
1425     struct cls_cursor cursor;
1426     struct oftable *table;
1427
1428     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1429     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1430     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1431     CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1432         facet_remove(facet);
1433     }
1434
1435     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1436     ovs_rwlock_wrlock(&xlate_rwlock);
1437     xlate_remove_ofproto(ofproto);
1438     ovs_rwlock_unlock(&xlate_rwlock);
1439
1440     flow_miss_batch_ofproto_destroyed(ofproto->backer->udpif, ofproto);
1441
1442     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1443     complete_operations(ofproto);
1444
1445     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1446         struct cls_cursor cursor;
1447
1448         ovs_rwlock_wrlock(&table->cls.rwlock);
1449         cls_cursor_init(&cursor, &table->cls, NULL);
1450         CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1451             ofproto_rule_delete(&ofproto->up, &table->cls, &rule->up);
1452         }
1453         ovs_rwlock_unlock(&table->cls.rwlock);
1454     }
1455     complete_operations(ofproto);
1456
1457     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1458     LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &ofproto->flow_mods) {
1459         list_remove(&fm->list_node);
1460         ofproto->n_flow_mods--;
1461         free(fm->ofpacts);
1462         free(fm);
1463     }
1464     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1465     ovs_mutex_destroy(&ofproto->flow_mod_mutex);
1466
1467     ovs_mutex_lock(&ofproto->pin_mutex);
1468     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &ofproto->pins) {
1469         list_remove(&pin->list_node);
1470         ofproto->n_pins--;
1471         free(CONST_CAST(void *, pin->packet));
1472         free(pin);
1473     }
1474     ovs_mutex_unlock(&ofproto->pin_mutex);
1475     ovs_mutex_destroy(&ofproto->pin_mutex);
1476
1477     mbridge_unref(ofproto->mbridge);
1478
1479     netflow_destroy(ofproto->netflow);
1480     dpif_sflow_unref(ofproto->sflow);
1481     hmap_destroy(&ofproto->bundles);
1482     mac_learning_unref(ofproto->ml);
1483
1484     classifier_destroy(&ofproto->facets);
1485
1486     hmap_destroy(&ofproto->vlandev_map);
1487     hmap_destroy(&ofproto->realdev_vid_map);
1488
1489     sset_destroy(&ofproto->ports);
1490     sset_destroy(&ofproto->ghost_ports);
1491     sset_destroy(&ofproto->port_poll_set);
1492
1493     ovs_mutex_destroy(&ofproto->vsp_mutex);
1494
1495     close_dpif_backer(ofproto->backer);
1496 }
1497
1498 static int
1499 run_fast(struct ofproto *ofproto_)
1500 {
1501     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1502     struct ofputil_packet_in *pin, *next_pin;
1503     struct ofputil_flow_mod *fm, *next_fm;
1504     struct list flow_mods, pins;
1505     struct ofport_dpif *ofport;
1506
1507     /* Do not perform any periodic activity required by 'ofproto' while
1508      * waiting for flow restore to complete. */
1509     if (ofproto_get_flow_restore_wait()) {
1510         return 0;
1511     }
1512
1513     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1514     if (ofproto->n_flow_mods) {
1515         flow_mods = ofproto->flow_mods;
1516         list_moved(&flow_mods);
1517         list_init(&ofproto->flow_mods);
1518         ofproto->n_flow_mods = 0;
1519     } else {
1520         list_init(&flow_mods);
1521     }
1522     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1523
1524     LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &flow_mods) {
1525         int error = ofproto_flow_mod(&ofproto->up, fm);
1526         if (error && !VLOG_DROP_WARN(&rl)) {
1527             VLOG_WARN("learning action failed to modify flow table (%s)",
1528                       ofperr_get_name(error));
1529         }
1530
1531         list_remove(&fm->list_node);
1532         free(fm->ofpacts);
1533         free(fm);
1534     }
1535
1536     ovs_mutex_lock(&ofproto->pin_mutex);
1537     if (ofproto->n_pins) {
1538         pins = ofproto->pins;
1539         list_moved(&pins);
1540         list_init(&ofproto->pins);
1541         ofproto->n_pins = 0;
1542     } else {
1543         list_init(&pins);
1544     }
1545     ovs_mutex_unlock(&ofproto->pin_mutex);
1546
1547     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1548         connmgr_send_packet_in(ofproto->up.connmgr, pin);
1549         list_remove(&pin->list_node);
1550         free(CONST_CAST(void *, pin->packet));
1551         free(pin);
1552     }
1553
1554     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1555         port_run_fast(ofport);
1556     }
1557
1558     return 0;
1559 }
1560
1561 static int
1562 run(struct ofproto *ofproto_)
1563 {
1564     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1565     struct ofport_dpif *ofport;
1566     struct ofbundle *bundle;
1567     int error;
1568
1569     if (!clogged) {
1570         complete_operations(ofproto);
1571     }
1572
1573     if (mbridge_need_revalidate(ofproto->mbridge)) {
1574         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1575         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1576         mac_learning_flush(ofproto->ml);
1577         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1578     }
1579
1580     /* Do not perform any periodic activity below required by 'ofproto' while
1581      * waiting for flow restore to complete. */
1582     if (ofproto_get_flow_restore_wait()) {
1583         return 0;
1584     }
1585
1586     error = run_fast(ofproto_);
1587     if (error) {
1588         return error;
1589     }
1590
1591     if (ofproto->netflow) {
1592         if (netflow_run(ofproto->netflow)) {
1593             send_netflow_active_timeouts(ofproto);
1594         }
1595     }
1596     if (ofproto->sflow) {
1597         dpif_sflow_run(ofproto->sflow);
1598     }
1599     if (ofproto->ipfix) {
1600         dpif_ipfix_run(ofproto->ipfix);
1601     }
1602
1603     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1604         port_run(ofport);
1605     }
1606     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1607         bundle_run(bundle);
1608     }
1609
1610     stp_run(ofproto);
1611     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1612     if (mac_learning_run(ofproto->ml)) {
1613         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1614     }
1615     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1616
1617     /* Check the consistency of a random facet, to aid debugging. */
1618     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1619     if (time_msec() >= ofproto->consistency_rl
1620         && !classifier_is_empty(&ofproto->facets)
1621         && !ofproto->backer->need_revalidate) {
1622         struct cls_table *table;
1623         struct cls_rule *cr;
1624         struct facet *facet;
1625
1626         ofproto->consistency_rl = time_msec() + 250;
1627
1628         table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1629                              struct cls_table, hmap_node);
1630         cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1631                           hmap_node);
1632         facet = CONTAINER_OF(cr, struct facet, cr);
1633
1634         if (!facet_check_consistency(facet)) {
1635             ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1636         }
1637     }
1638     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1639
1640     return 0;
1641 }
1642
1643 static void
1644 wait(struct ofproto *ofproto_)
1645 {
1646     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1647     struct ofport_dpif *ofport;
1648     struct ofbundle *bundle;
1649
1650     if (!clogged && !list_is_empty(&ofproto->completions)) {
1651         poll_immediate_wake();
1652     }
1653
1654     if (ofproto_get_flow_restore_wait()) {
1655         return;
1656     }
1657
1658     if (ofproto->sflow) {
1659         dpif_sflow_wait(ofproto->sflow);
1660     }
1661     if (ofproto->ipfix) {
1662         dpif_ipfix_wait(ofproto->ipfix);
1663     }
1664     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1665         port_wait(ofport);
1666     }
1667     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1668         bundle_wait(bundle);
1669     }
1670     if (ofproto->netflow) {
1671         netflow_wait(ofproto->netflow);
1672     }
1673     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1674     mac_learning_wait(ofproto->ml);
1675     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1676     stp_wait(ofproto);
1677     if (ofproto->backer->need_revalidate) {
1678         /* Shouldn't happen, but if it does just go around again. */
1679         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1680         poll_immediate_wake();
1681     }
1682 }
1683
1684 static void
1685 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1686 {
1687     const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1688     struct cls_cursor cursor;
1689     size_t n_subfacets = 0;
1690     struct facet *facet;
1691
1692     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1693     simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1694     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1695
1696     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1697     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1698     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1699         n_subfacets += list_size(&facet->subfacets);
1700     }
1701     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1702     simap_increase(usage, "subfacets", n_subfacets);
1703 }
1704
1705 static void
1706 flush(struct ofproto *ofproto_)
1707 {
1708     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1709     struct subfacet *subfacet, *next_subfacet;
1710     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1711     int n_batch;
1712
1713     n_batch = 0;
1714     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1715                         &ofproto->backer->subfacets) {
1716         if (subfacet->facet->ofproto != ofproto) {
1717             continue;
1718         }
1719
1720         if (subfacet->path != SF_NOT_INSTALLED) {
1721             batch[n_batch++] = subfacet;
1722             if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1723                 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1724                 n_batch = 0;
1725             }
1726         } else {
1727             subfacet_destroy(subfacet);
1728         }
1729     }
1730
1731     if (n_batch > 0) {
1732         subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1733     }
1734 }
1735
1736 static void
1737 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1738              bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1739 {
1740     *arp_match_ip = true;
1741     *actions = (OFPUTIL_A_OUTPUT |
1742                 OFPUTIL_A_SET_VLAN_VID |
1743                 OFPUTIL_A_SET_VLAN_PCP |
1744                 OFPUTIL_A_STRIP_VLAN |
1745                 OFPUTIL_A_SET_DL_SRC |
1746                 OFPUTIL_A_SET_DL_DST |
1747                 OFPUTIL_A_SET_NW_SRC |
1748                 OFPUTIL_A_SET_NW_DST |
1749                 OFPUTIL_A_SET_NW_TOS |
1750                 OFPUTIL_A_SET_TP_SRC |
1751                 OFPUTIL_A_SET_TP_DST |
1752                 OFPUTIL_A_ENQUEUE);
1753 }
1754
1755 static void
1756 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1757 {
1758     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1759     struct dpif_dp_stats s;
1760     uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1761     uint64_t n_lookup;
1762
1763     strcpy(ots->name, "classifier");
1764
1765     dpif_get_dp_stats(ofproto->backer->dpif, &s);
1766     rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1767     rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1768     rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1769
1770     n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1771     ots->lookup_count = htonll(n_lookup);
1772     ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1773 }
1774
1775 static struct ofport *
1776 port_alloc(void)
1777 {
1778     struct ofport_dpif *port = xmalloc(sizeof *port);
1779     return &port->up;
1780 }
1781
1782 static void
1783 port_dealloc(struct ofport *port_)
1784 {
1785     struct ofport_dpif *port = ofport_dpif_cast(port_);
1786     free(port);
1787 }
1788
1789 static int
1790 port_construct(struct ofport *port_)
1791 {
1792     struct ofport_dpif *port = ofport_dpif_cast(port_);
1793     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1794     const struct netdev *netdev = port->up.netdev;
1795     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1796     struct dpif_port dpif_port;
1797     int error;
1798
1799     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1800     port->bundle = NULL;
1801     port->cfm = NULL;
1802     port->bfd = NULL;
1803     port->may_enable = true;
1804     port->stp_port = NULL;
1805     port->stp_state = STP_DISABLED;
1806     port->is_tunnel = false;
1807     port->peer = NULL;
1808     port->qdscp = NULL;
1809     port->n_qdscp = 0;
1810     port->realdev_ofp_port = 0;
1811     port->vlandev_vid = 0;
1812     port->carrier_seq = netdev_get_carrier_resets(netdev);
1813
1814     if (netdev_vport_is_patch(netdev)) {
1815         /* By bailing out here, we don't submit the port to the sFlow module
1816          * to be considered for counter polling export.  This is correct
1817          * because the patch port represents an interface that sFlow considers
1818          * to be "internal" to the switch as a whole, and therefore not an
1819          * candidate for counter polling. */
1820         port->odp_port = ODPP_NONE;
1821         ofport_update_peer(port);
1822         return 0;
1823     }
1824
1825     error = dpif_port_query_by_name(ofproto->backer->dpif,
1826                                     netdev_vport_get_dpif_port(netdev, namebuf,
1827                                                                sizeof namebuf),
1828                                     &dpif_port);
1829     if (error) {
1830         return error;
1831     }
1832
1833     port->odp_port = dpif_port.port_no;
1834
1835     if (netdev_get_tunnel_config(netdev)) {
1836         tnl_port_add(port, port->up.netdev, port->odp_port);
1837         port->is_tunnel = true;
1838     } else {
1839         /* Sanity-check that a mapping doesn't already exist.  This
1840          * shouldn't happen for non-tunnel ports. */
1841         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1842             VLOG_ERR("port %s already has an OpenFlow port number",
1843                      dpif_port.name);
1844             dpif_port_destroy(&dpif_port);
1845             return EBUSY;
1846         }
1847
1848         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1849         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1850                     hash_odp_port(port->odp_port));
1851         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1852     }
1853     dpif_port_destroy(&dpif_port);
1854
1855     if (ofproto->sflow) {
1856         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1857     }
1858
1859     return 0;
1860 }
1861
1862 static void
1863 port_destruct(struct ofport *port_)
1864 {
1865     struct ofport_dpif *port = ofport_dpif_cast(port_);
1866     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1867     const char *devname = netdev_get_name(port->up.netdev);
1868     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1869     const char *dp_port_name;
1870
1871     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1872     ovs_rwlock_wrlock(&xlate_rwlock);
1873     xlate_ofport_remove(port);
1874     ovs_rwlock_unlock(&xlate_rwlock);
1875
1876     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1877                                               sizeof namebuf);
1878     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1879         /* The underlying device is still there, so delete it.  This
1880          * happens when the ofproto is being destroyed, since the caller
1881          * assumes that removal of attached ports will happen as part of
1882          * destruction. */
1883         if (!port->is_tunnel) {
1884             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1885         }
1886     }
1887
1888     if (port->peer) {
1889         port->peer->peer = NULL;
1890         port->peer = NULL;
1891     }
1892
1893     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1894         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1895         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1896         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1897     }
1898
1899     tnl_port_del(port);
1900     sset_find_and_delete(&ofproto->ports, devname);
1901     sset_find_and_delete(&ofproto->ghost_ports, devname);
1902     bundle_remove(port_);
1903     set_cfm(port_, NULL);
1904     set_bfd(port_, NULL);
1905     if (ofproto->sflow) {
1906         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1907     }
1908
1909     free(port->qdscp);
1910 }
1911
1912 static void
1913 port_modified(struct ofport *port_)
1914 {
1915     struct ofport_dpif *port = ofport_dpif_cast(port_);
1916
1917     if (port->bundle && port->bundle->bond) {
1918         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1919     }
1920
1921     if (port->cfm) {
1922         cfm_set_netdev(port->cfm, port->up.netdev);
1923     }
1924
1925     if (port->bfd) {
1926         bfd_set_netdev(port->bfd, port->up.netdev);
1927     }
1928
1929     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1930                                                 port->odp_port)) {
1931         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1932             REV_RECONFIGURE;
1933     }
1934
1935     ofport_update_peer(port);
1936 }
1937
1938 static void
1939 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1940 {
1941     struct ofport_dpif *port = ofport_dpif_cast(port_);
1942     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1943     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1944
1945     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1946                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1947                    OFPUTIL_PC_NO_PACKET_IN)) {
1948         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1949
1950         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1951             bundle_update(port->bundle);
1952         }
1953     }
1954 }
1955
1956 static int
1957 set_sflow(struct ofproto *ofproto_,
1958           const struct ofproto_sflow_options *sflow_options)
1959 {
1960     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1961     struct dpif_sflow *ds = ofproto->sflow;
1962
1963     if (sflow_options) {
1964         if (!ds) {
1965             struct ofport_dpif *ofport;
1966
1967             ds = ofproto->sflow = dpif_sflow_create();
1968             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1969                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1970             }
1971             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1972         }
1973         dpif_sflow_set_options(ds, sflow_options);
1974     } else {
1975         if (ds) {
1976             dpif_sflow_unref(ds);
1977             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1978             ofproto->sflow = NULL;
1979         }
1980     }
1981     return 0;
1982 }
1983
1984 static int
1985 set_ipfix(
1986     struct ofproto *ofproto_,
1987     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1988     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1989     size_t n_flow_exporters_options)
1990 {
1991     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1992     struct dpif_ipfix *di = ofproto->ipfix;
1993     bool has_options = bridge_exporter_options || flow_exporters_options;
1994
1995     if (has_options && !di) {
1996         di = ofproto->ipfix = dpif_ipfix_create();
1997     }
1998
1999     if (di) {
2000         /* Call set_options in any case to cleanly flush the flow
2001          * caches in the last exporters that are to be destroyed. */
2002         dpif_ipfix_set_options(
2003             di, bridge_exporter_options, flow_exporters_options,
2004             n_flow_exporters_options);
2005
2006         if (!has_options) {
2007             dpif_ipfix_unref(di);
2008             ofproto->ipfix = NULL;
2009         }
2010     }
2011
2012     return 0;
2013 }
2014
2015 static int
2016 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
2017 {
2018     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2019     int error;
2020
2021     if (!s) {
2022         error = 0;
2023     } else {
2024         if (!ofport->cfm) {
2025             struct ofproto_dpif *ofproto;
2026
2027             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2028             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2029             ofport->cfm = cfm_create(ofport->up.netdev);
2030         }
2031
2032         if (cfm_configure(ofport->cfm, s)) {
2033             return 0;
2034         }
2035
2036         error = EINVAL;
2037     }
2038     cfm_unref(ofport->cfm);
2039     ofport->cfm = NULL;
2040     return error;
2041 }
2042
2043 static bool
2044 get_cfm_status(const struct ofport *ofport_,
2045                struct ofproto_cfm_status *status)
2046 {
2047     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2048
2049     if (ofport->cfm) {
2050         status->faults = cfm_get_fault(ofport->cfm);
2051         status->remote_opstate = cfm_get_opup(ofport->cfm);
2052         status->health = cfm_get_health(ofport->cfm);
2053         cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2054         return true;
2055     } else {
2056         return false;
2057     }
2058 }
2059
2060 static int
2061 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2062 {
2063     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2064     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2065     struct bfd *old;
2066
2067     old = ofport->bfd;
2068     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2069                                 cfg, ofport->up.netdev);
2070     if (ofport->bfd != old) {
2071         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2072     }
2073
2074     return 0;
2075 }
2076
2077 static int
2078 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2079 {
2080     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2081
2082     if (ofport->bfd) {
2083         bfd_get_status(ofport->bfd, smap);
2084         return 0;
2085     } else {
2086         return ENOENT;
2087     }
2088 }
2089 \f
2090 /* Spanning Tree. */
2091
2092 static void
2093 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2094 {
2095     struct ofproto_dpif *ofproto = ofproto_;
2096     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2097     struct ofport_dpif *ofport;
2098
2099     ofport = stp_port_get_aux(sp);
2100     if (!ofport) {
2101         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2102                      ofproto->up.name, port_num);
2103     } else {
2104         struct eth_header *eth = pkt->l2;
2105
2106         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2107         if (eth_addr_is_zero(eth->eth_src)) {
2108             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2109                          "with unknown MAC", ofproto->up.name, port_num);
2110         } else {
2111             send_packet(ofport, pkt);
2112         }
2113     }
2114     ofpbuf_delete(pkt);
2115 }
2116
2117 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2118 static int
2119 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2120 {
2121     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2122
2123     /* Only revalidate flows if the configuration changed. */
2124     if (!s != !ofproto->stp) {
2125         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2126     }
2127
2128     if (s) {
2129         if (!ofproto->stp) {
2130             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2131                                       send_bpdu_cb, ofproto);
2132             ofproto->stp_last_tick = time_msec();
2133         }
2134
2135         stp_set_bridge_id(ofproto->stp, s->system_id);
2136         stp_set_bridge_priority(ofproto->stp, s->priority);
2137         stp_set_hello_time(ofproto->stp, s->hello_time);
2138         stp_set_max_age(ofproto->stp, s->max_age);
2139         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2140     }  else {
2141         struct ofport *ofport;
2142
2143         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2144             set_stp_port(ofport, NULL);
2145         }
2146
2147         stp_unref(ofproto->stp);
2148         ofproto->stp = NULL;
2149     }
2150
2151     return 0;
2152 }
2153
2154 static int
2155 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2156 {
2157     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2158
2159     if (ofproto->stp) {
2160         s->enabled = true;
2161         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2162         s->designated_root = stp_get_designated_root(ofproto->stp);
2163         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2164     } else {
2165         s->enabled = false;
2166     }
2167
2168     return 0;
2169 }
2170
2171 static void
2172 update_stp_port_state(struct ofport_dpif *ofport)
2173 {
2174     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2175     enum stp_state state;
2176
2177     /* Figure out new state. */
2178     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2179                              : STP_DISABLED;
2180
2181     /* Update state. */
2182     if (ofport->stp_state != state) {
2183         enum ofputil_port_state of_state;
2184         bool fwd_change;
2185
2186         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2187                     netdev_get_name(ofport->up.netdev),
2188                     stp_state_name(ofport->stp_state),
2189                     stp_state_name(state));
2190         if (stp_learn_in_state(ofport->stp_state)
2191                 != stp_learn_in_state(state)) {
2192             /* xxx Learning action flows should also be flushed. */
2193             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2194             mac_learning_flush(ofproto->ml);
2195             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2196         }
2197         fwd_change = stp_forward_in_state(ofport->stp_state)
2198                         != stp_forward_in_state(state);
2199
2200         ofproto->backer->need_revalidate = REV_STP;
2201         ofport->stp_state = state;
2202         ofport->stp_state_entered = time_msec();
2203
2204         if (fwd_change && ofport->bundle) {
2205             bundle_update(ofport->bundle);
2206         }
2207
2208         /* Update the STP state bits in the OpenFlow port description. */
2209         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2210         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2211                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2212                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2213                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2214                      : 0);
2215         ofproto_port_set_state(&ofport->up, of_state);
2216     }
2217 }
2218
2219 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2220  * caller is responsible for assigning STP port numbers and ensuring
2221  * there are no duplicates. */
2222 static int
2223 set_stp_port(struct ofport *ofport_,
2224              const struct ofproto_port_stp_settings *s)
2225 {
2226     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2227     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2228     struct stp_port *sp = ofport->stp_port;
2229
2230     if (!s || !s->enable) {
2231         if (sp) {
2232             ofport->stp_port = NULL;
2233             stp_port_disable(sp);
2234             update_stp_port_state(ofport);
2235         }
2236         return 0;
2237     } else if (sp && stp_port_no(sp) != s->port_num
2238             && ofport == stp_port_get_aux(sp)) {
2239         /* The port-id changed, so disable the old one if it's not
2240          * already in use by another port. */
2241         stp_port_disable(sp);
2242     }
2243
2244     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2245     stp_port_enable(sp);
2246
2247     stp_port_set_aux(sp, ofport);
2248     stp_port_set_priority(sp, s->priority);
2249     stp_port_set_path_cost(sp, s->path_cost);
2250
2251     update_stp_port_state(ofport);
2252
2253     return 0;
2254 }
2255
2256 static int
2257 get_stp_port_status(struct ofport *ofport_,
2258                     struct ofproto_port_stp_status *s)
2259 {
2260     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2261     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2262     struct stp_port *sp = ofport->stp_port;
2263
2264     if (!ofproto->stp || !sp) {
2265         s->enabled = false;
2266         return 0;
2267     }
2268
2269     s->enabled = true;
2270     s->port_id = stp_port_get_id(sp);
2271     s->state = stp_port_get_state(sp);
2272     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2273     s->role = stp_port_get_role(sp);
2274     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2275
2276     return 0;
2277 }
2278
2279 static void
2280 stp_run(struct ofproto_dpif *ofproto)
2281 {
2282     if (ofproto->stp) {
2283         long long int now = time_msec();
2284         long long int elapsed = now - ofproto->stp_last_tick;
2285         struct stp_port *sp;
2286
2287         if (elapsed > 0) {
2288             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2289             ofproto->stp_last_tick = now;
2290         }
2291         while (stp_get_changed_port(ofproto->stp, &sp)) {
2292             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2293
2294             if (ofport) {
2295                 update_stp_port_state(ofport);
2296             }
2297         }
2298
2299         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2300             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2301             mac_learning_flush(ofproto->ml);
2302             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2303         }
2304     }
2305 }
2306
2307 static void
2308 stp_wait(struct ofproto_dpif *ofproto)
2309 {
2310     if (ofproto->stp) {
2311         poll_timer_wait(1000);
2312     }
2313 }
2314 \f
2315 static int
2316 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2317            size_t n_qdscp)
2318 {
2319     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2320     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2321
2322     if (ofport->n_qdscp != n_qdscp
2323         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2324                               n_qdscp * sizeof *qdscp))) {
2325         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2326         free(ofport->qdscp);
2327         ofport->qdscp = n_qdscp
2328             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2329             : NULL;
2330         ofport->n_qdscp = n_qdscp;
2331     }
2332
2333     return 0;
2334 }
2335 \f
2336 /* Bundles. */
2337
2338 /* Expires all MAC learning entries associated with 'bundle' and forces its
2339  * ofproto to revalidate every flow.
2340  *
2341  * Normally MAC learning entries are removed only from the ofproto associated
2342  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2343  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2344  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2345  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2346  * with the host from which it migrated. */
2347 static void
2348 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2349 {
2350     struct ofproto_dpif *ofproto = bundle->ofproto;
2351     struct mac_learning *ml = ofproto->ml;
2352     struct mac_entry *mac, *next_mac;
2353
2354     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2355     ovs_rwlock_wrlock(&ml->rwlock);
2356     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2357         if (mac->port.p == bundle) {
2358             if (all_ofprotos) {
2359                 struct ofproto_dpif *o;
2360
2361                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2362                     if (o != ofproto) {
2363                         struct mac_entry *e;
2364
2365                         ovs_rwlock_wrlock(&o->ml->rwlock);
2366                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2367                         if (e) {
2368                             mac_learning_expire(o->ml, e);
2369                         }
2370                         ovs_rwlock_unlock(&o->ml->rwlock);
2371                     }
2372                 }
2373             }
2374
2375             mac_learning_expire(ml, mac);
2376         }
2377     }
2378     ovs_rwlock_unlock(&ml->rwlock);
2379 }
2380
2381 static struct ofbundle *
2382 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2383 {
2384     struct ofbundle *bundle;
2385
2386     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2387                              &ofproto->bundles) {
2388         if (bundle->aux == aux) {
2389             return bundle;
2390         }
2391     }
2392     return NULL;
2393 }
2394
2395 static void
2396 bundle_update(struct ofbundle *bundle)
2397 {
2398     struct ofport_dpif *port;
2399
2400     bundle->floodable = true;
2401     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2402         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2403             || !stp_forward_in_state(port->stp_state)) {
2404             bundle->floodable = false;
2405             break;
2406         }
2407     }
2408 }
2409
2410 static void
2411 bundle_del_port(struct ofport_dpif *port)
2412 {
2413     struct ofbundle *bundle = port->bundle;
2414
2415     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2416
2417     list_remove(&port->bundle_node);
2418     port->bundle = NULL;
2419
2420     if (bundle->lacp) {
2421         lacp_slave_unregister(bundle->lacp, port);
2422     }
2423     if (bundle->bond) {
2424         bond_slave_unregister(bundle->bond, port);
2425     }
2426
2427     bundle_update(bundle);
2428 }
2429
2430 static bool
2431 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2432                 struct lacp_slave_settings *lacp)
2433 {
2434     struct ofport_dpif *port;
2435
2436     port = get_ofp_port(bundle->ofproto, ofp_port);
2437     if (!port) {
2438         return false;
2439     }
2440
2441     if (port->bundle != bundle) {
2442         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2443         if (port->bundle) {
2444             bundle_remove(&port->up);
2445         }
2446
2447         port->bundle = bundle;
2448         list_push_back(&bundle->ports, &port->bundle_node);
2449         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2450             || !stp_forward_in_state(port->stp_state)) {
2451             bundle->floodable = false;
2452         }
2453     }
2454     if (lacp) {
2455         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2456         lacp_slave_register(bundle->lacp, port, lacp);
2457     }
2458
2459     return true;
2460 }
2461
2462 static void
2463 bundle_destroy(struct ofbundle *bundle)
2464 {
2465     struct ofproto_dpif *ofproto;
2466     struct ofport_dpif *port, *next_port;
2467
2468     if (!bundle) {
2469         return;
2470     }
2471
2472     ofproto = bundle->ofproto;
2473     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2474
2475     ovs_rwlock_wrlock(&xlate_rwlock);
2476     xlate_bundle_remove(bundle);
2477     ovs_rwlock_unlock(&xlate_rwlock);
2478
2479     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2480         bundle_del_port(port);
2481     }
2482
2483     bundle_flush_macs(bundle, true);
2484     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2485     free(bundle->name);
2486     free(bundle->trunks);
2487     lacp_unref(bundle->lacp);
2488     bond_unref(bundle->bond);
2489     free(bundle);
2490 }
2491
2492 static int
2493 bundle_set(struct ofproto *ofproto_, void *aux,
2494            const struct ofproto_bundle_settings *s)
2495 {
2496     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2497     bool need_flush = false;
2498     struct ofport_dpif *port;
2499     struct ofbundle *bundle;
2500     unsigned long *trunks;
2501     int vlan;
2502     size_t i;
2503     bool ok;
2504
2505     if (!s) {
2506         bundle_destroy(bundle_lookup(ofproto, aux));
2507         return 0;
2508     }
2509
2510     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2511     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2512
2513     bundle = bundle_lookup(ofproto, aux);
2514     if (!bundle) {
2515         bundle = xmalloc(sizeof *bundle);
2516
2517         bundle->ofproto = ofproto;
2518         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2519                     hash_pointer(aux, 0));
2520         bundle->aux = aux;
2521         bundle->name = NULL;
2522
2523         list_init(&bundle->ports);
2524         bundle->vlan_mode = PORT_VLAN_TRUNK;
2525         bundle->vlan = -1;
2526         bundle->trunks = NULL;
2527         bundle->use_priority_tags = s->use_priority_tags;
2528         bundle->lacp = NULL;
2529         bundle->bond = NULL;
2530
2531         bundle->floodable = true;
2532         mbridge_register_bundle(ofproto->mbridge, bundle);
2533     }
2534
2535     if (!bundle->name || strcmp(s->name, bundle->name)) {
2536         free(bundle->name);
2537         bundle->name = xstrdup(s->name);
2538     }
2539
2540     /* LACP. */
2541     if (s->lacp) {
2542         if (!bundle->lacp) {
2543             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2544             bundle->lacp = lacp_create();
2545         }
2546         lacp_configure(bundle->lacp, s->lacp);
2547     } else {
2548         lacp_unref(bundle->lacp);
2549         bundle->lacp = NULL;
2550     }
2551
2552     /* Update set of ports. */
2553     ok = true;
2554     for (i = 0; i < s->n_slaves; i++) {
2555         if (!bundle_add_port(bundle, s->slaves[i],
2556                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2557             ok = false;
2558         }
2559     }
2560     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2561         struct ofport_dpif *next_port;
2562
2563         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2564             for (i = 0; i < s->n_slaves; i++) {
2565                 if (s->slaves[i] == port->up.ofp_port) {
2566                     goto found;
2567                 }
2568             }
2569
2570             bundle_del_port(port);
2571         found: ;
2572         }
2573     }
2574     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2575
2576     if (list_is_empty(&bundle->ports)) {
2577         bundle_destroy(bundle);
2578         return EINVAL;
2579     }
2580
2581     /* Set VLAN tagging mode */
2582     if (s->vlan_mode != bundle->vlan_mode
2583         || s->use_priority_tags != bundle->use_priority_tags) {
2584         bundle->vlan_mode = s->vlan_mode;
2585         bundle->use_priority_tags = s->use_priority_tags;
2586         need_flush = true;
2587     }
2588
2589     /* Set VLAN tag. */
2590     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2591             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2592             : 0);
2593     if (vlan != bundle->vlan) {
2594         bundle->vlan = vlan;
2595         need_flush = true;
2596     }
2597
2598     /* Get trunked VLANs. */
2599     switch (s->vlan_mode) {
2600     case PORT_VLAN_ACCESS:
2601         trunks = NULL;
2602         break;
2603
2604     case PORT_VLAN_TRUNK:
2605         trunks = CONST_CAST(unsigned long *, s->trunks);
2606         break;
2607
2608     case PORT_VLAN_NATIVE_UNTAGGED:
2609     case PORT_VLAN_NATIVE_TAGGED:
2610         if (vlan != 0 && (!s->trunks
2611                           || !bitmap_is_set(s->trunks, vlan)
2612                           || bitmap_is_set(s->trunks, 0))) {
2613             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2614             if (s->trunks) {
2615                 trunks = bitmap_clone(s->trunks, 4096);
2616             } else {
2617                 trunks = bitmap_allocate1(4096);
2618             }
2619             bitmap_set1(trunks, vlan);
2620             bitmap_set0(trunks, 0);
2621         } else {
2622             trunks = CONST_CAST(unsigned long *, s->trunks);
2623         }
2624         break;
2625
2626     default:
2627         NOT_REACHED();
2628     }
2629     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2630         free(bundle->trunks);
2631         if (trunks == s->trunks) {
2632             bundle->trunks = vlan_bitmap_clone(trunks);
2633         } else {
2634             bundle->trunks = trunks;
2635             trunks = NULL;
2636         }
2637         need_flush = true;
2638     }
2639     if (trunks != s->trunks) {
2640         free(trunks);
2641     }
2642
2643     /* Bonding. */
2644     if (!list_is_short(&bundle->ports)) {
2645         bundle->ofproto->has_bonded_bundles = true;
2646         if (bundle->bond) {
2647             if (bond_reconfigure(bundle->bond, s->bond)) {
2648                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2649             }
2650         } else {
2651             bundle->bond = bond_create(s->bond);
2652             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2653         }
2654
2655         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2656             bond_slave_register(bundle->bond, port, port->up.netdev);
2657         }
2658     } else {
2659         bond_unref(bundle->bond);
2660         bundle->bond = NULL;
2661     }
2662
2663     /* If we changed something that would affect MAC learning, un-learn
2664      * everything on this port and force flow revalidation. */
2665     if (need_flush) {
2666         bundle_flush_macs(bundle, false);
2667     }
2668
2669     return 0;
2670 }
2671
2672 static void
2673 bundle_remove(struct ofport *port_)
2674 {
2675     struct ofport_dpif *port = ofport_dpif_cast(port_);
2676     struct ofbundle *bundle = port->bundle;
2677
2678     if (bundle) {
2679         bundle_del_port(port);
2680         if (list_is_empty(&bundle->ports)) {
2681             bundle_destroy(bundle);
2682         } else if (list_is_short(&bundle->ports)) {
2683             bond_unref(bundle->bond);
2684             bundle->bond = NULL;
2685         }
2686     }
2687 }
2688
2689 static void
2690 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2691 {
2692     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2693     struct ofport_dpif *port = port_;
2694     uint8_t ea[ETH_ADDR_LEN];
2695     int error;
2696
2697     error = netdev_get_etheraddr(port->up.netdev, ea);
2698     if (!error) {
2699         struct ofpbuf packet;
2700         void *packet_pdu;
2701
2702         ofpbuf_init(&packet, 0);
2703         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2704                                  pdu_size);
2705         memcpy(packet_pdu, pdu, pdu_size);
2706
2707         send_packet(port, &packet);
2708         ofpbuf_uninit(&packet);
2709     } else {
2710         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2711                     "%s (%s)", port->bundle->name,
2712                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2713     }
2714 }
2715
2716 static void
2717 bundle_send_learning_packets(struct ofbundle *bundle)
2718 {
2719     struct ofproto_dpif *ofproto = bundle->ofproto;
2720     int error, n_packets, n_errors;
2721     struct mac_entry *e;
2722
2723     error = n_packets = n_errors = 0;
2724     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2725     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2726         if (e->port.p != bundle) {
2727             struct ofpbuf *learning_packet;
2728             struct ofport_dpif *port;
2729             void *port_void;
2730             int ret;
2731
2732             /* The assignment to "port" is unnecessary but makes "grep"ing for
2733              * struct ofport_dpif more effective. */
2734             learning_packet = bond_compose_learning_packet(bundle->bond,
2735                                                            e->mac, e->vlan,
2736                                                            &port_void);
2737             port = port_void;
2738             ret = send_packet(port, learning_packet);
2739             ofpbuf_delete(learning_packet);
2740             if (ret) {
2741                 error = ret;
2742                 n_errors++;
2743             }
2744             n_packets++;
2745         }
2746     }
2747     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2748
2749     if (n_errors) {
2750         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2751         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2752                      "packets, last error was: %s",
2753                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2754     } else {
2755         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2756                  bundle->name, n_packets);
2757     }
2758 }
2759
2760 static void
2761 bundle_run(struct ofbundle *bundle)
2762 {
2763     if (bundle->lacp) {
2764         lacp_run(bundle->lacp, send_pdu_cb);
2765     }
2766     if (bundle->bond) {
2767         struct ofport_dpif *port;
2768
2769         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2770             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2771         }
2772
2773         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2774             bundle->ofproto->backer->need_revalidate = REV_BOND;
2775         }
2776
2777         if (bond_should_send_learning_packets(bundle->bond)) {
2778             bundle_send_learning_packets(bundle);
2779         }
2780     }
2781 }
2782
2783 static void
2784 bundle_wait(struct ofbundle *bundle)
2785 {
2786     if (bundle->lacp) {
2787         lacp_wait(bundle->lacp);
2788     }
2789     if (bundle->bond) {
2790         bond_wait(bundle->bond);
2791     }
2792 }
2793 \f
2794 /* Mirrors. */
2795
2796 static int
2797 mirror_set__(struct ofproto *ofproto_, void *aux,
2798              const struct ofproto_mirror_settings *s)
2799 {
2800     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2801     struct ofbundle **srcs, **dsts;
2802     int error;
2803     size_t i;
2804
2805     if (!s) {
2806         mirror_destroy(ofproto->mbridge, aux);
2807         return 0;
2808     }
2809
2810     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2811     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2812
2813     for (i = 0; i < s->n_srcs; i++) {
2814         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2815     }
2816
2817     for (i = 0; i < s->n_dsts; i++) {
2818         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2819     }
2820
2821     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2822                        s->n_dsts, s->src_vlans,
2823                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2824     free(srcs);
2825     free(dsts);
2826     return error;
2827 }
2828
2829 static int
2830 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2831                    uint64_t *packets, uint64_t *bytes)
2832 {
2833     push_all_stats();
2834     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2835                             bytes);
2836 }
2837
2838 static int
2839 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2840 {
2841     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2842     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2843     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2844         mac_learning_flush(ofproto->ml);
2845     }
2846     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2847     return 0;
2848 }
2849
2850 static bool
2851 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2852 {
2853     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2854     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2855     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2856 }
2857
2858 static void
2859 forward_bpdu_changed(struct ofproto *ofproto_)
2860 {
2861     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2862     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2863 }
2864
2865 static void
2866 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2867                      size_t max_entries)
2868 {
2869     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2870     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2871     mac_learning_set_idle_time(ofproto->ml, idle_time);
2872     mac_learning_set_max_entries(ofproto->ml, max_entries);
2873     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2874 }
2875 \f
2876 /* Ports. */
2877
2878 static struct ofport_dpif *
2879 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2880 {
2881     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2882     return ofport ? ofport_dpif_cast(ofport) : NULL;
2883 }
2884
2885 static struct ofport_dpif *
2886 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2887 {
2888     struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2889     return port && &ofproto->up == port->up.ofproto ? port : NULL;
2890 }
2891
2892 static void
2893 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2894                             struct ofproto_port *ofproto_port,
2895                             struct dpif_port *dpif_port)
2896 {
2897     ofproto_port->name = dpif_port->name;
2898     ofproto_port->type = dpif_port->type;
2899     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2900 }
2901
2902 static void
2903 ofport_update_peer(struct ofport_dpif *ofport)
2904 {
2905     const struct ofproto_dpif *ofproto;
2906     struct dpif_backer *backer;
2907     char *peer_name;
2908
2909     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2910         return;
2911     }
2912
2913     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2914     backer->need_revalidate = REV_RECONFIGURE;
2915
2916     if (ofport->peer) {
2917         ofport->peer->peer = NULL;
2918         ofport->peer = NULL;
2919     }
2920
2921     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2922     if (!peer_name) {
2923         return;
2924     }
2925
2926     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2927         struct ofport *peer_ofport;
2928         struct ofport_dpif *peer;
2929         char *peer_peer;
2930
2931         if (ofproto->backer != backer) {
2932             continue;
2933         }
2934
2935         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2936         if (!peer_ofport) {
2937             continue;
2938         }
2939
2940         peer = ofport_dpif_cast(peer_ofport);
2941         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2942         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2943                                  peer_peer)) {
2944             ofport->peer = peer;
2945             ofport->peer->peer = ofport;
2946         }
2947         free(peer_peer);
2948
2949         break;
2950     }
2951     free(peer_name);
2952 }
2953
2954 static void
2955 port_run_fast(struct ofport_dpif *ofport)
2956 {
2957     if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2958         struct ofpbuf packet;
2959
2960         ofpbuf_init(&packet, 0);
2961         cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2962         send_packet(ofport, &packet);
2963         ofpbuf_uninit(&packet);
2964     }
2965
2966     if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2967         struct ofpbuf packet;
2968
2969         ofpbuf_init(&packet, 0);
2970         bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2971         send_packet(ofport, &packet);
2972         ofpbuf_uninit(&packet);
2973     }
2974 }
2975
2976 static void
2977 port_run(struct ofport_dpif *ofport)
2978 {
2979     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2980     bool carrier_changed = carrier_seq != ofport->carrier_seq;
2981     bool enable = netdev_get_carrier(ofport->up.netdev);
2982     bool cfm_enable = false;
2983     bool bfd_enable = false;
2984
2985     ofport->carrier_seq = carrier_seq;
2986
2987     port_run_fast(ofport);
2988
2989     if (ofport->cfm) {
2990         int cfm_opup = cfm_get_opup(ofport->cfm);
2991
2992         cfm_run(ofport->cfm);
2993         cfm_enable = !cfm_get_fault(ofport->cfm);
2994
2995         if (cfm_opup >= 0) {
2996             cfm_enable = cfm_enable && cfm_opup;
2997         }
2998     }
2999
3000     if (ofport->bfd) {
3001         bfd_run(ofport->bfd);
3002         bfd_enable = bfd_forwarding(ofport->bfd);
3003     }
3004
3005     if (ofport->bfd || ofport->cfm) {
3006         enable = enable && (cfm_enable || bfd_enable);
3007     }
3008
3009     if (ofport->bundle) {
3010         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3011         if (carrier_changed) {
3012             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3013         }
3014     }
3015
3016     if (ofport->may_enable != enable) {
3017         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3018         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3019     }
3020
3021     ofport->may_enable = enable;
3022 }
3023
3024 static void
3025 port_wait(struct ofport_dpif *ofport)
3026 {
3027     if (ofport->cfm) {
3028         cfm_wait(ofport->cfm);
3029     }
3030
3031     if (ofport->bfd) {
3032         bfd_wait(ofport->bfd);
3033     }
3034 }
3035
3036 static int
3037 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3038                    struct ofproto_port *ofproto_port)
3039 {
3040     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3041     struct dpif_port dpif_port;
3042     int error;
3043
3044     if (sset_contains(&ofproto->ghost_ports, devname)) {
3045         const char *type = netdev_get_type_from_name(devname);
3046
3047         /* We may be called before ofproto->up.port_by_name is populated with
3048          * the appropriate ofport.  For this reason, we must get the name and
3049          * type from the netdev layer directly. */
3050         if (type) {
3051             const struct ofport *ofport;
3052
3053             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3054             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3055             ofproto_port->name = xstrdup(devname);
3056             ofproto_port->type = xstrdup(type);
3057             return 0;
3058         }
3059         return ENODEV;
3060     }
3061
3062     if (!sset_contains(&ofproto->ports, devname)) {
3063         return ENODEV;
3064     }
3065     error = dpif_port_query_by_name(ofproto->backer->dpif,
3066                                     devname, &dpif_port);
3067     if (!error) {
3068         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3069     }
3070     return error;
3071 }
3072
3073 static int
3074 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3075 {
3076     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3077     const char *devname = netdev_get_name(netdev);
3078     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3079     const char *dp_port_name;
3080
3081     if (netdev_vport_is_patch(netdev)) {
3082         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3083         return 0;
3084     }
3085
3086     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3087     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3088         odp_port_t port_no = ODPP_NONE;
3089         int error;
3090
3091         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3092         if (error) {
3093             return error;
3094         }
3095         if (netdev_get_tunnel_config(netdev)) {
3096             simap_put(&ofproto->backer->tnl_backers,
3097                       dp_port_name, odp_to_u32(port_no));
3098         }
3099     }
3100
3101     if (netdev_get_tunnel_config(netdev)) {
3102         sset_add(&ofproto->ghost_ports, devname);
3103     } else {
3104         sset_add(&ofproto->ports, devname);
3105     }
3106     return 0;
3107 }
3108
3109 static int
3110 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3111 {
3112     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3113     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3114     int error = 0;
3115
3116     if (!ofport) {
3117         return 0;
3118     }
3119
3120     sset_find_and_delete(&ofproto->ghost_ports,
3121                          netdev_get_name(ofport->up.netdev));
3122     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3123     if (!ofport->is_tunnel) {
3124         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3125         if (!error) {
3126             /* The caller is going to close ofport->up.netdev.  If this is a
3127              * bonded port, then the bond is using that netdev, so remove it
3128              * from the bond.  The client will need to reconfigure everything
3129              * after deleting ports, so then the slave will get re-added. */
3130             bundle_remove(&ofport->up);
3131         }
3132     }
3133     return error;
3134 }
3135
3136 static int
3137 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3138 {
3139     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3140     int error;
3141
3142     push_all_stats();
3143
3144     error = netdev_get_stats(ofport->up.netdev, stats);
3145
3146     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3147         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3148
3149         /* ofproto->stats.tx_packets represents packets that we created
3150          * internally and sent to some port (e.g. packets sent with
3151          * send_packet()).  Account for them as if they had come from
3152          * OFPP_LOCAL and got forwarded. */
3153
3154         if (stats->rx_packets != UINT64_MAX) {
3155             stats->rx_packets += ofproto->stats.tx_packets;
3156         }
3157
3158         if (stats->rx_bytes != UINT64_MAX) {
3159             stats->rx_bytes += ofproto->stats.tx_bytes;
3160         }
3161
3162         /* ofproto->stats.rx_packets represents packets that were received on
3163          * some port and we processed internally and dropped (e.g. STP).
3164          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3165
3166         if (stats->tx_packets != UINT64_MAX) {
3167             stats->tx_packets += ofproto->stats.rx_packets;
3168         }
3169
3170         if (stats->tx_bytes != UINT64_MAX) {
3171             stats->tx_bytes += ofproto->stats.rx_bytes;
3172         }
3173     }
3174
3175     return error;
3176 }
3177
3178 struct port_dump_state {
3179     uint32_t bucket;
3180     uint32_t offset;
3181     bool ghost;
3182
3183     struct ofproto_port port;
3184     bool has_port;
3185 };
3186
3187 static int
3188 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3189 {
3190     *statep = xzalloc(sizeof(struct port_dump_state));
3191     return 0;
3192 }
3193
3194 static int
3195 port_dump_next(const struct ofproto *ofproto_, void *state_,
3196                struct ofproto_port *port)
3197 {
3198     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3199     struct port_dump_state *state = state_;
3200     const struct sset *sset;
3201     struct sset_node *node;
3202
3203     if (state->has_port) {
3204         ofproto_port_destroy(&state->port);
3205         state->has_port = false;
3206     }
3207     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3208     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3209         int error;
3210
3211         error = port_query_by_name(ofproto_, node->name, &state->port);
3212         if (!error) {
3213             *port = state->port;
3214             state->has_port = true;
3215             return 0;
3216         } else if (error != ENODEV) {
3217             return error;
3218         }
3219     }
3220
3221     if (!state->ghost) {
3222         state->ghost = true;
3223         state->bucket = 0;
3224         state->offset = 0;
3225         return port_dump_next(ofproto_, state_, port);
3226     }
3227
3228     return EOF;
3229 }
3230
3231 static int
3232 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3233 {
3234     struct port_dump_state *state = state_;
3235
3236     if (state->has_port) {
3237         ofproto_port_destroy(&state->port);
3238     }
3239     free(state);
3240     return 0;
3241 }
3242
3243 static int
3244 port_poll(const struct ofproto *ofproto_, char **devnamep)
3245 {
3246     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3247
3248     if (ofproto->port_poll_errno) {
3249         int error = ofproto->port_poll_errno;
3250         ofproto->port_poll_errno = 0;
3251         return error;
3252     }
3253
3254     if (sset_is_empty(&ofproto->port_poll_set)) {
3255         return EAGAIN;
3256     }
3257
3258     *devnamep = sset_pop(&ofproto->port_poll_set);
3259     return 0;
3260 }
3261
3262 static void
3263 port_poll_wait(const struct ofproto *ofproto_)
3264 {
3265     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3266     dpif_port_poll_wait(ofproto->backer->dpif);
3267 }
3268
3269 static int
3270 port_is_lacp_current(const struct ofport *ofport_)
3271 {
3272     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3273     return (ofport->bundle && ofport->bundle->lacp
3274             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3275             : -1);
3276 }
3277 \f
3278 /* Upcall handling. */
3279
3280 struct flow_miss_op {
3281     struct dpif_op dpif_op;
3282
3283     uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3284     struct xlate_out xout;
3285     bool xout_garbage;           /* 'xout' needs to be uninitialized? */
3286
3287     struct ofpbuf mask;          /* Flow mask for "put" ops. */
3288     struct odputil_keybuf maskbuf;
3289
3290     /* If this is a "put" op, then a pointer to the subfacet that should
3291      * be marked as uninstalled if the operation fails. */
3292     struct subfacet *subfacet;
3293 };
3294
3295 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3296  * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3297  * and (usually) installing a datapath flow.  The answer is usually "yes" (a
3298  * return value of true).  However, for short flows the cost of bookkeeping is
3299  * much higher than the benefits, so when the datapath holds a large number of
3300  * flows we impose some heuristics to decide which flows are likely to be worth
3301  * tracking. */
3302 static bool
3303 flow_miss_should_make_facet(struct flow_miss *miss)
3304 {
3305     struct dpif_backer *backer = miss->ofproto->backer;
3306     uint32_t hash;
3307
3308     switch (flow_miss_model) {
3309     case OFPROTO_HANDLE_MISS_AUTO:
3310         break;
3311     case OFPROTO_HANDLE_MISS_WITH_FACETS:
3312         return true;
3313     case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3314         return false;
3315     }
3316
3317     if (!backer->governor) {
3318         size_t n_subfacets;
3319
3320         n_subfacets = hmap_count(&backer->subfacets);
3321         if (n_subfacets * 2 <= flow_eviction_threshold) {
3322             return true;
3323         }
3324
3325         backer->governor = governor_create();
3326     }
3327
3328     hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3329     return governor_should_install_flow(backer->governor, hash,
3330                                         list_size(&miss->packets));
3331 }
3332
3333 /* Handles 'miss', which matches 'facet'.  May add any required datapath
3334  * operations to 'ops', incrementing '*n_ops' for each new op.
3335  *
3336  * All of the packets in 'miss' are considered to have arrived at time
3337  * 'miss->stats.used'.  This is really important only for new facets: if we
3338  * just called time_msec() here, then the new subfacet or its packets could
3339  * look (occasionally) as though it was used some time after the facet was
3340  * used.  That can make a one-packet flow look like it has a nonzero duration,
3341  * which looks odd in e.g. NetFlow statistics. */
3342 static void
3343 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3344                             struct flow_miss_op *ops, size_t *n_ops)
3345 {
3346     enum subfacet_path want_path;
3347     struct subfacet *subfacet;
3348
3349     facet->packet_count += miss->stats.n_packets;
3350     facet->prev_packet_count += miss->stats.n_packets;
3351     facet->byte_count += miss->stats.n_bytes;
3352     facet->prev_byte_count += miss->stats.n_bytes;
3353
3354     want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3355
3356     /* Don't install the flow if it's the result of the "userspace"
3357      * action for an already installed facet.  This can occur when a
3358      * datapath flow with wildcards has a "userspace" action and flows
3359      * sent to userspace result in a different subfacet, which will then
3360      * be rejected as overlapping by the datapath. */
3361     if (miss->upcall_type == DPIF_UC_ACTION
3362         && !list_is_empty(&facet->subfacets)) {
3363         return;
3364     }
3365
3366     subfacet = subfacet_create(facet, miss);
3367     if (subfacet->path != want_path) {
3368         struct flow_miss_op *op = &ops[(*n_ops)++];
3369         struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3370
3371         subfacet->path = want_path;
3372
3373         ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3374         if (enable_megaflows) {
3375             odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3376                                    &miss->flow, UINT32_MAX);
3377         }
3378
3379         op->xout_garbage = false;
3380         op->dpif_op.type = DPIF_OP_FLOW_PUT;
3381         op->subfacet = subfacet;
3382         put->flags = DPIF_FP_CREATE;
3383         put->key = miss->key;
3384         put->key_len = miss->key_len;
3385         put->mask = op->mask.data;
3386         put->mask_len = op->mask.size;
3387
3388         if (want_path == SF_FAST_PATH) {
3389             put->actions = facet->xout.odp_actions.data;
3390             put->actions_len = facet->xout.odp_actions.size;
3391         } else {
3392             compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3393                               op->slow_stub, sizeof op->slow_stub,
3394                               &put->actions, &put->actions_len);
3395         }
3396         put->stats = NULL;
3397     }
3398 }
3399
3400 /* Handles flow miss 'miss'.  May add any required datapath operations
3401  * to 'ops', incrementing '*n_ops' for each new op. */
3402 static void
3403 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3404                  size_t *n_ops)
3405 {
3406     struct facet *facet;
3407
3408     miss->ofproto->n_missed += list_size(&miss->packets);
3409
3410     facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3411     if (!facet) {
3412         /* There does not exist a bijection between 'struct flow' and datapath
3413          * flow keys with fitness ODP_FIT_TO_LITTLE.  This breaks a fundamental
3414          * assumption used throughout the facet and subfacet handling code.
3415          * Since we have to handle these misses in userspace anyway, we simply
3416          * skip facet creation, avoiding the problem altogether. */
3417         if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3418             || !flow_miss_should_make_facet(miss)) {
3419             return;
3420         }
3421
3422         facet = facet_create(miss);
3423     }
3424     handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3425 }
3426
3427 static struct drop_key *
3428 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3429                 size_t key_len)
3430 {
3431     struct drop_key *drop_key;
3432
3433     HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3434                              &backer->drop_keys) {
3435         if (drop_key->key_len == key_len
3436             && !memcmp(drop_key->key, key, key_len)) {
3437             return drop_key;
3438         }
3439     }
3440     return NULL;
3441 }
3442
3443 static void
3444 drop_key_clear(struct dpif_backer *backer)
3445 {
3446     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3447     struct drop_key *drop_key, *next;
3448
3449     HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3450         int error;
3451
3452         error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3453                               NULL);
3454         if (error && !VLOG_DROP_WARN(&rl)) {
3455             struct ds ds = DS_EMPTY_INITIALIZER;
3456             odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3457             VLOG_WARN("Failed to delete drop key (%s) (%s)",
3458                       ovs_strerror(error), ds_cstr(&ds));
3459             ds_destroy(&ds);
3460         }
3461
3462         hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3463         drop_key_destroy(drop_key);
3464     }
3465
3466     udpif_drop_key_clear(backer->udpif);
3467 }
3468
3469 static void
3470 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3471 {
3472     struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3473     struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3474     struct flow_miss *miss;
3475     size_t n_ops, i;
3476
3477     /* Process each element in the to-do list, constructing the set of
3478      * operations to batch. */
3479     n_ops = 0;
3480     HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3481         handle_flow_miss(miss, flow_miss_ops, &n_ops);
3482     }
3483     ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3484
3485     /* Execute batch. */
3486     for (i = 0; i < n_ops; i++) {
3487         dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3488     }
3489     dpif_operate(backer->dpif, dpif_ops, n_ops);
3490
3491     for (i = 0; i < n_ops; i++) {
3492         if (dpif_ops[i]->error != 0
3493             && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3494             && flow_miss_ops[i].subfacet) {
3495             struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3496
3497             COVERAGE_INC(subfacet_install_fail);
3498
3499             /* Zero-out subfacet counters when installation failed, but
3500              * datapath reported hits.  This should not happen and
3501              * indicates a bug, since if the datapath flow exists, we
3502              * should not be attempting to create a new subfacet.  A
3503              * buggy datapath could trigger this, so just zero out the
3504              * counters and log an error. */
3505             if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3506                 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3507                             "datapath reported hits");
3508                 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3509             }
3510
3511             subfacet->path = SF_NOT_INSTALLED;
3512         }
3513     }
3514 }
3515
3516 static void
3517 handle_sflow_upcall(struct dpif_backer *backer,
3518                     const struct dpif_upcall *upcall)
3519 {
3520     struct ofproto_dpif *ofproto;
3521     union user_action_cookie cookie;
3522     struct flow flow;
3523     odp_port_t odp_in_port;
3524
3525     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3526                       &flow, NULL, &ofproto, &odp_in_port)
3527         || !ofproto->sflow) {
3528         return;
3529     }
3530
3531     memset(&cookie, 0, sizeof cookie);
3532     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3533     dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3534                         odp_in_port, &cookie);
3535 }
3536
3537 static void
3538 handle_flow_sample_upcall(struct dpif_backer *backer,
3539                           const struct dpif_upcall *upcall)
3540 {
3541     struct ofproto_dpif *ofproto;
3542     union user_action_cookie cookie;
3543     struct flow flow;
3544
3545     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3546                       &flow, NULL, &ofproto, NULL)
3547         || !ofproto->ipfix) {
3548         return;
3549     }
3550
3551     memset(&cookie, 0, sizeof cookie);
3552     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3553
3554     /* The flow reflects exactly the contents of the packet.  Sample
3555      * the packet using it. */
3556     dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3557                            cookie.flow_sample.collector_set_id,
3558                            cookie.flow_sample.probability,
3559                            cookie.flow_sample.obs_domain_id,
3560                            cookie.flow_sample.obs_point_id);
3561 }
3562
3563 static void
3564 handle_ipfix_upcall(struct dpif_backer *backer,
3565                     const struct dpif_upcall *upcall)
3566 {
3567     struct ofproto_dpif *ofproto;
3568     struct flow flow;
3569
3570     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3571                       &flow, NULL, &ofproto, NULL)
3572         || !ofproto->ipfix) {
3573         return;
3574     }
3575
3576     /* The flow reflects exactly the contents of the packet.  Sample
3577      * the packet using it. */
3578     dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3579 }
3580
3581 static void
3582 handle_upcalls(struct dpif_backer *backer)
3583 {
3584     struct flow_miss_batch *fmb;
3585     int n_processed;
3586
3587     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3588         struct upcall *upcall = upcall_next(backer->udpif);
3589
3590         if (!upcall) {
3591             break;
3592         }
3593
3594         switch (upcall->type) {
3595         case SFLOW_UPCALL:
3596             handle_sflow_upcall(backer, &upcall->dpif_upcall);
3597             break;
3598
3599         case FLOW_SAMPLE_UPCALL:
3600             handle_flow_sample_upcall(backer, &upcall->dpif_upcall);
3601             break;
3602
3603         case IPFIX_UPCALL:
3604             handle_ipfix_upcall(backer, &upcall->dpif_upcall);
3605             break;
3606
3607         case BAD_UPCALL:
3608             break;
3609
3610         case MISS_UPCALL:
3611             NOT_REACHED();
3612         }
3613
3614         upcall_destroy(upcall);
3615     }
3616
3617     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3618         struct drop_key *drop_key = drop_key_next(backer->udpif);
3619         if (!drop_key) {
3620             break;
3621         }
3622
3623         if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3624             hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3625                         hash_bytes(drop_key->key, drop_key->key_len, 0));
3626             dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3627                           drop_key->key, drop_key->key_len,
3628                           NULL, 0, NULL, 0, NULL);
3629         } else {
3630             drop_key_destroy(drop_key);
3631         }
3632     }
3633
3634     fmb = flow_miss_batch_next(backer->udpif);
3635     if (fmb) {
3636         handle_flow_misses(backer, fmb);
3637         flow_miss_batch_destroy(fmb);
3638     }
3639 }
3640 \f
3641 /* Flow expiration. */
3642
3643 static int subfacet_max_idle(const struct dpif_backer *);
3644 static void update_stats(struct dpif_backer *);
3645 static void rule_expire(struct rule_dpif *);
3646 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3647
3648 /* This function is called periodically by run().  Its job is to collect
3649  * updates for the flows that have been installed into the datapath, most
3650  * importantly when they last were used, and then use that information to
3651  * expire flows that have not been used recently.
3652  *
3653  * Returns the number of milliseconds after which it should be called again. */
3654 static int
3655 expire(struct dpif_backer *backer)
3656 {
3657     struct ofproto_dpif *ofproto;
3658     size_t n_subfacets;
3659     int max_idle;
3660
3661     /* Periodically clear out the drop keys in an effort to keep them
3662      * relatively few. */
3663     drop_key_clear(backer);
3664
3665     /* Update stats for each flow in the backer. */
3666     update_stats(backer);
3667
3668     n_subfacets = hmap_count(&backer->subfacets);
3669     if (n_subfacets) {
3670         struct subfacet *subfacet;
3671         long long int total, now;
3672
3673         total = 0;
3674         now = time_msec();
3675         HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3676             total += now - subfacet->created;
3677         }
3678         backer->avg_subfacet_life += total / n_subfacets;
3679     }
3680     backer->avg_subfacet_life /= 2;
3681
3682     backer->avg_n_subfacet += n_subfacets;
3683     backer->avg_n_subfacet /= 2;
3684
3685     backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3686
3687     max_idle = subfacet_max_idle(backer);
3688     expire_subfacets(backer, max_idle);
3689
3690     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3691         struct rule *rule, *next_rule;
3692
3693         if (ofproto->backer != backer) {
3694             continue;
3695         }
3696
3697         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3698          * has passed. */
3699         ovs_mutex_lock(&ofproto->up.expirable_mutex);
3700         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3701                             &ofproto->up.expirable) {
3702             rule_expire(rule_dpif_cast(rule));
3703         }
3704         ovs_mutex_unlock(&ofproto->up.expirable_mutex);
3705
3706         /* All outstanding data in existing flows has been accounted, so it's a
3707          * good time to do bond rebalancing. */
3708         if (ofproto->has_bonded_bundles) {
3709             struct ofbundle *bundle;
3710
3711             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3712                 if (bundle->bond) {
3713                     bond_rebalance(bundle->bond);
3714                 }
3715             }
3716         }
3717     }
3718
3719     return MIN(max_idle, 1000);
3720 }
3721
3722 /* Updates flow table statistics given that the datapath just reported 'stats'
3723  * as 'subfacet''s statistics. */
3724 static void
3725 update_subfacet_stats(struct subfacet *subfacet,
3726                       const struct dpif_flow_stats *stats)
3727 {
3728     struct facet *facet = subfacet->facet;
3729     struct dpif_flow_stats diff;
3730
3731     diff.tcp_flags = stats->tcp_flags;
3732     diff.used = stats->used;
3733
3734     if (stats->n_packets >= subfacet->dp_packet_count) {
3735         diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3736     } else {
3737         VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3738         diff.n_packets = 0;
3739     }
3740
3741     if (stats->n_bytes >= subfacet->dp_byte_count) {
3742         diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3743     } else {
3744         VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3745         diff.n_bytes = 0;
3746     }
3747
3748     facet->ofproto->n_hit += diff.n_packets;
3749     subfacet->dp_packet_count = stats->n_packets;
3750     subfacet->dp_byte_count = stats->n_bytes;
3751     subfacet_update_stats(subfacet, &diff);
3752
3753     if (facet->accounted_bytes < facet->byte_count) {
3754         facet_learn(facet);
3755         facet_account(facet);
3756         facet->accounted_bytes = facet->byte_count;
3757     }
3758 }
3759
3760 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3761  * about, or a flow that shouldn't be installed but was anyway.  Delete it. */
3762 static void
3763 delete_unexpected_flow(struct dpif_backer *backer,
3764                        const struct nlattr *key, size_t key_len)
3765 {
3766     if (!VLOG_DROP_WARN(&rl)) {
3767         struct ds s;
3768
3769         ds_init(&s);
3770         odp_flow_key_format(key, key_len, &s);
3771         VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3772         ds_destroy(&s);
3773     }
3774
3775     COVERAGE_INC(facet_unexpected);
3776     dpif_flow_del(backer->dpif, key, key_len, NULL);
3777 }
3778
3779 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3780  *
3781  * This function also pushes statistics updates to rules which each facet
3782  * resubmits into.  Generally these statistics will be accurate.  However, if a
3783  * facet changes the rule it resubmits into at some time in between
3784  * update_stats() runs, it is possible that statistics accrued to the
3785  * old rule will be incorrectly attributed to the new rule.  This could be
3786  * avoided by calling update_stats() whenever rules are created or
3787  * deleted.  However, the performance impact of making so many calls to the
3788  * datapath do not justify the benefit of having perfectly accurate statistics.
3789  *
3790  * In addition, this function maintains per ofproto flow hit counts. The patch
3791  * port is not treated specially. e.g. A packet ingress from br0 patched into
3792  * br1 will increase the hit count of br0 by 1, however, does not affect
3793  * the hit or miss counts of br1.
3794  */
3795 static void
3796 update_stats(struct dpif_backer *backer)
3797 {
3798     const struct dpif_flow_stats *stats;
3799     struct dpif_flow_dump dump;
3800     const struct nlattr *key, *mask;
3801     size_t key_len, mask_len;
3802
3803     dpif_flow_dump_start(&dump, backer->dpif);
3804     while (dpif_flow_dump_next(&dump, &key, &key_len,
3805                                &mask, &mask_len, NULL, NULL, &stats)) {
3806         struct subfacet *subfacet;
3807         uint32_t key_hash;
3808
3809         key_hash = odp_flow_key_hash(key, key_len);
3810         subfacet = subfacet_find(backer, key, key_len, key_hash);
3811         switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3812         case SF_FAST_PATH:
3813             update_subfacet_stats(subfacet, stats);
3814             break;
3815
3816         case SF_SLOW_PATH:
3817             /* Stats are updated per-packet. */
3818             break;
3819
3820         case SF_NOT_INSTALLED:
3821         default:
3822             delete_unexpected_flow(backer, key, key_len);
3823             break;
3824         }
3825         run_fast_rl();
3826     }
3827     dpif_flow_dump_done(&dump);
3828
3829     update_moving_averages(backer);
3830 }
3831
3832 /* Calculates and returns the number of milliseconds of idle time after which
3833  * subfacets should expire from the datapath.  When a subfacet expires, we fold
3834  * its statistics into its facet, and when a facet's last subfacet expires, we
3835  * fold its statistic into its rule. */
3836 static int
3837 subfacet_max_idle(const struct dpif_backer *backer)
3838 {
3839     /*
3840      * Idle time histogram.
3841      *
3842      * Most of the time a switch has a relatively small number of subfacets.
3843      * When this is the case we might as well keep statistics for all of them
3844      * in userspace and to cache them in the kernel datapath for performance as
3845      * well.
3846      *
3847      * As the number of subfacets increases, the memory required to maintain
3848      * statistics about them in userspace and in the kernel becomes
3849      * significant.  However, with a large number of subfacets it is likely
3850      * that only a few of them are "heavy hitters" that consume a large amount
3851      * of bandwidth.  At this point, only heavy hitters are worth caching in
3852      * the kernel and maintaining in userspaces; other subfacets we can
3853      * discard.
3854      *
3855      * The technique used to compute the idle time is to build a histogram with
3856      * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each.  Each subfacet
3857      * that is installed in the kernel gets dropped in the appropriate bucket.
3858      * After the histogram has been built, we compute the cutoff so that only
3859      * the most-recently-used 1% of subfacets (but at least
3860      * flow_eviction_threshold flows) are kept cached.  At least
3861      * the most-recently-used bucket of subfacets is kept, so actually an
3862      * arbitrary number of subfacets can be kept in any given expiration run
3863      * (though the next run will delete most of those unless they receive
3864      * additional data).
3865      *
3866      * This requires a second pass through the subfacets, in addition to the
3867      * pass made by update_stats(), because the former function never looks at
3868      * uninstallable subfacets.
3869      */
3870     enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3871     enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3872     int buckets[N_BUCKETS] = { 0 };
3873     int total, subtotal, bucket;
3874     struct subfacet *subfacet;
3875     long long int now;
3876     int i;
3877
3878     total = hmap_count(&backer->subfacets);
3879     if (total <= flow_eviction_threshold) {
3880         return N_BUCKETS * BUCKET_WIDTH;
3881     }
3882
3883     /* Build histogram. */
3884     now = time_msec();
3885     HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3886         long long int idle = now - subfacet->used;
3887         int bucket = (idle <= 0 ? 0
3888                       : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3889                       : (unsigned int) idle / BUCKET_WIDTH);
3890         buckets[bucket]++;
3891     }
3892
3893     /* Find the first bucket whose flows should be expired. */
3894     subtotal = bucket = 0;
3895     do {
3896         subtotal += buckets[bucket++];
3897     } while (bucket < N_BUCKETS &&
3898              subtotal < MAX(flow_eviction_threshold, total / 100));
3899
3900     if (VLOG_IS_DBG_ENABLED()) {
3901         struct ds s;
3902
3903         ds_init(&s);
3904         ds_put_cstr(&s, "keep");
3905         for (i = 0; i < N_BUCKETS; i++) {
3906             if (i == bucket) {
3907                 ds_put_cstr(&s, ", drop");
3908             }
3909             if (buckets[i]) {
3910                 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3911             }
3912         }
3913         VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3914         ds_destroy(&s);
3915     }
3916
3917     return bucket * BUCKET_WIDTH;
3918 }
3919
3920 static void
3921 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3922 {
3923     /* Cutoff time for most flows. */
3924     long long int normal_cutoff = time_msec() - dp_max_idle;
3925
3926     /* We really want to keep flows for special protocols around, so use a more
3927      * conservative cutoff. */
3928     long long int special_cutoff = time_msec() - 10000;
3929
3930     struct subfacet *subfacet, *next_subfacet;
3931     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3932     int n_batch;
3933
3934     n_batch = 0;
3935     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3936                         &backer->subfacets) {
3937         long long int cutoff;
3938
3939         cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3940                                                 | SLOW_STP)
3941                   ? special_cutoff
3942                   : normal_cutoff);
3943         if (subfacet->used < cutoff) {
3944             if (subfacet->path != SF_NOT_INSTALLED) {
3945                 batch[n_batch++] = subfacet;
3946                 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3947                     subfacet_destroy_batch(backer, batch, n_batch);
3948                     n_batch = 0;
3949                 }
3950             } else {
3951                 subfacet_destroy(subfacet);
3952             }
3953         }
3954     }
3955
3956     if (n_batch > 0) {
3957         subfacet_destroy_batch(backer, batch, n_batch);
3958     }
3959 }
3960
3961 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3962  * then delete it entirely. */
3963 static void
3964 rule_expire(struct rule_dpif *rule)
3965 {
3966     uint16_t idle_timeout, hard_timeout;
3967     long long int now;
3968     uint8_t reason;
3969
3970     if (rule->up.pending) {
3971         /* We'll have to expire it later. */
3972         return;
3973     }
3974
3975     ovs_mutex_lock(&rule->up.timeout_mutex);
3976     hard_timeout = rule->up.hard_timeout;
3977     idle_timeout = rule->up.idle_timeout;
3978     ovs_mutex_unlock(&rule->up.timeout_mutex);
3979
3980     /* Has 'rule' expired? */
3981     now = time_msec();
3982     if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3983         reason = OFPRR_HARD_TIMEOUT;
3984     } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3985         reason = OFPRR_IDLE_TIMEOUT;
3986     } else {
3987         return;
3988     }
3989
3990     COVERAGE_INC(ofproto_dpif_expired);
3991     ofproto_rule_expire(&rule->up, reason);
3992 }
3993 \f
3994 /* Facets. */
3995
3996 /* Creates and returns a new facet based on 'miss'.
3997  *
3998  * The caller must already have determined that no facet with an identical
3999  * 'miss->flow' exists in 'miss->ofproto'.
4000  *
4001  * 'rule' and 'xout' must have been created based on 'miss'.
4002  *
4003  * 'facet'' statistics are initialized based on 'stats'.
4004  *
4005  * The facet will initially have no subfacets.  The caller should create (at
4006  * least) one subfacet with subfacet_create(). */
4007 static struct facet *
4008 facet_create(const struct flow_miss *miss)
4009 {
4010     struct ofproto_dpif *ofproto = miss->ofproto;
4011     struct facet *facet;
4012     struct match match;
4013
4014     facet = xzalloc(sizeof *facet);
4015     facet->ofproto = miss->ofproto;
4016     facet->used = miss->stats.used;
4017     facet->flow = miss->flow;
4018     facet->learn_rl = time_msec() + 500;
4019
4020     list_init(&facet->subfacets);
4021     netflow_flow_init(&facet->nf_flow);
4022     netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4023
4024     xlate_out_copy(&facet->xout, &miss->xout);
4025
4026     match_init(&match, &facet->flow, &facet->xout.wc);
4027     cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4028     ovs_rwlock_wrlock(&ofproto->facets.rwlock);
4029     classifier_insert(&ofproto->facets, &facet->cr);
4030     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4031
4032     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4033     return facet;
4034 }
4035
4036 static void
4037 facet_free(struct facet *facet)
4038 {
4039     if (facet) {
4040         xlate_out_uninit(&facet->xout);
4041         free(facet);
4042     }
4043 }
4044
4045 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4046  * 'packet', which arrived on 'in_port'. */
4047 static bool
4048 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4049                     const struct nlattr *odp_actions, size_t actions_len,
4050                     struct ofpbuf *packet)
4051 {
4052     struct odputil_keybuf keybuf;
4053     struct ofpbuf key;
4054     int error;
4055
4056     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4057     odp_flow_key_from_flow(&key, flow,
4058                            ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4059
4060     error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4061                          odp_actions, actions_len, packet);
4062     return !error;
4063 }
4064
4065 /* Remove 'facet' from its ofproto and free up the associated memory:
4066  *
4067  *   - If 'facet' was installed in the datapath, uninstalls it and updates its
4068  *     rule's statistics, via subfacet_uninstall().
4069  *
4070  *   - Removes 'facet' from its rule and from ofproto->facets.
4071  */
4072 static void
4073 facet_remove(struct facet *facet)
4074 {
4075     struct subfacet *subfacet, *next_subfacet;
4076
4077     ovs_assert(!list_is_empty(&facet->subfacets));
4078
4079     /* First uninstall all of the subfacets to get final statistics. */
4080     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4081         subfacet_uninstall(subfacet);
4082     }
4083
4084     /* Flush the final stats to the rule.
4085      *
4086      * This might require us to have at least one subfacet around so that we
4087      * can use its actions for accounting in facet_account(), which is why we
4088      * have uninstalled but not yet destroyed the subfacets. */
4089     facet_flush_stats(facet);
4090
4091     /* Now we're really all done so destroy everything. */
4092     LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4093                         &facet->subfacets) {
4094         subfacet_destroy__(subfacet);
4095     }
4096     ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4097     classifier_remove(&facet->ofproto->facets, &facet->cr);
4098     ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4099     cls_rule_destroy(&facet->cr);
4100     facet_free(facet);
4101 }
4102
4103 /* Feed information from 'facet' back into the learning table to keep it in
4104  * sync with what is actually flowing through the datapath. */
4105 static void
4106 facet_learn(struct facet *facet)
4107 {
4108     long long int now = time_msec();
4109
4110     if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4111         return;
4112     }
4113
4114     facet->learn_rl = now + 500;
4115
4116     if (!facet->xout.has_learn
4117         && !facet->xout.has_normal
4118         && (!facet->xout.has_fin_timeout
4119             || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4120         return;
4121     }
4122
4123     facet_push_stats(facet, true);
4124 }
4125
4126 static void
4127 facet_account(struct facet *facet)
4128 {
4129     const struct nlattr *a;
4130     unsigned int left;
4131     ovs_be16 vlan_tci;
4132     uint64_t n_bytes;
4133
4134     if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4135         return;
4136     }
4137     n_bytes = facet->byte_count - facet->accounted_bytes;
4138
4139     /* This loop feeds byte counters to bond_account() for rebalancing to use
4140      * as a basis.  We also need to track the actual VLAN on which the packet
4141      * is going to be sent to ensure that it matches the one passed to
4142      * bond_choose_output_slave().  (Otherwise, we will account to the wrong
4143      * hash bucket.)
4144      *
4145      * We use the actions from an arbitrary subfacet because they should all
4146      * be equally valid for our purpose. */
4147     vlan_tci = facet->flow.vlan_tci;
4148     NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4149                              facet->xout.odp_actions.size) {
4150         const struct ovs_action_push_vlan *vlan;
4151         struct ofport_dpif *port;
4152
4153         switch (nl_attr_type(a)) {
4154         case OVS_ACTION_ATTR_OUTPUT:
4155             port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4156             if (port && port->bundle && port->bundle->bond) {
4157                 bond_account(port->bundle->bond, &facet->flow,
4158                              vlan_tci_to_vid(vlan_tci), n_bytes);
4159             }
4160             break;
4161
4162         case OVS_ACTION_ATTR_POP_VLAN:
4163             vlan_tci = htons(0);
4164             break;
4165
4166         case OVS_ACTION_ATTR_PUSH_VLAN:
4167             vlan = nl_attr_get(a);
4168             vlan_tci = vlan->vlan_tci;
4169             break;
4170         }
4171     }
4172 }
4173
4174 /* Returns true if the only action for 'facet' is to send to the controller.
4175  * (We don't report NetFlow expiration messages for such facets because they
4176  * are just part of the control logic for the network, not real traffic). */
4177 static bool
4178 facet_is_controller_flow(struct facet *facet)
4179 {
4180     if (facet) {
4181         struct ofproto_dpif *ofproto = facet->ofproto;
4182         const struct ofpact *ofpacts;
4183         struct rule_dpif *rule;
4184         size_t ofpacts_len;
4185         bool is_controller;
4186
4187         rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4188         ofpacts_len = rule->up.ofpacts_len;
4189         ofpacts = rule->up.ofpacts;
4190         is_controller = ofpacts_len > 0
4191             && ofpacts->type == OFPACT_CONTROLLER
4192             && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4193         rule_release(rule);
4194         return is_controller;
4195     }
4196     return false;
4197 }
4198
4199 /* Folds all of 'facet''s statistics into its rule.  Also updates the
4200  * accounting ofhook and emits a NetFlow expiration if appropriate.  All of
4201  * 'facet''s statistics in the datapath should have been zeroed and folded into
4202  * its packet and byte counts before this function is called. */
4203 static void
4204 facet_flush_stats(struct facet *facet)
4205 {
4206     struct ofproto_dpif *ofproto = facet->ofproto;
4207     struct subfacet *subfacet;
4208
4209     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4210         ovs_assert(!subfacet->dp_byte_count);
4211         ovs_assert(!subfacet->dp_packet_count);
4212     }
4213
4214     facet_push_stats(facet, false);
4215     if (facet->accounted_bytes < facet->byte_count) {
4216         facet_account(facet);
4217         facet->accounted_bytes = facet->byte_count;
4218     }
4219
4220     if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4221         struct ofexpired expired;
4222         expired.flow = facet->flow;
4223         expired.packet_count = facet->packet_count;
4224         expired.byte_count = facet->byte_count;
4225         expired.used = facet->used;
4226         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4227     }
4228
4229     /* Reset counters to prevent double counting if 'facet' ever gets
4230      * reinstalled. */
4231     facet_reset_counters(facet);
4232
4233     netflow_flow_clear(&facet->nf_flow);
4234     facet->tcp_flags = 0;
4235 }
4236
4237 /* Searches 'ofproto''s table of facets for one which would be responsible for
4238  * 'flow'.  Returns it if found, otherwise a null pointer.
4239  *
4240  * The returned facet might need revalidation; use facet_lookup_valid()
4241  * instead if that is important. */
4242 static struct facet *
4243 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4244 {
4245     struct cls_rule *cr;
4246
4247     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4248     cr = classifier_lookup(&ofproto->facets, flow, NULL);
4249     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4250     return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4251 }
4252
4253 /* Searches 'ofproto''s table of facets for one capable that covers
4254  * 'flow'.  Returns it if found, otherwise a null pointer.
4255  *
4256  * The returned facet is guaranteed to be valid. */
4257 static struct facet *
4258 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4259 {
4260     struct facet *facet;
4261
4262     facet = facet_find(ofproto, flow);
4263     if (facet
4264         && ofproto->backer->need_revalidate
4265         && !facet_revalidate(facet)) {
4266         return NULL;
4267     }
4268
4269     return facet;
4270 }
4271
4272 static bool
4273 facet_check_consistency(struct facet *facet)
4274 {
4275     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4276
4277     struct xlate_out xout;
4278     struct xlate_in xin;
4279
4280     struct rule_dpif *rule;
4281     bool ok;
4282
4283     /* Check the datapath actions for consistency. */
4284     rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4285     xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4286     xlate_actions(&xin, &xout);
4287     rule_release(rule);
4288
4289     ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4290         && facet->xout.slow == xout.slow;
4291     if (!ok && !VLOG_DROP_WARN(&rl)) {
4292         struct ds s = DS_EMPTY_INITIALIZER;
4293
4294         flow_format(&s, &facet->flow);
4295         ds_put_cstr(&s, ": inconsistency in facet");
4296
4297         if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4298             ds_put_cstr(&s, " (actions were: ");
4299             format_odp_actions(&s, facet->xout.odp_actions.data,
4300                                facet->xout.odp_actions.size);
4301             ds_put_cstr(&s, ") (correct actions: ");
4302             format_odp_actions(&s, xout.odp_actions.data,
4303                                xout.odp_actions.size);
4304             ds_put_char(&s, ')');
4305         }
4306
4307         if (facet->xout.slow != xout.slow) {
4308             ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4309         }
4310
4311         ds_destroy(&s);
4312     }
4313     xlate_out_uninit(&xout);
4314
4315     return ok;
4316 }
4317
4318 /* Re-searches the classifier for 'facet':
4319  *
4320  *   - If the rule found is different from 'facet''s current rule, moves
4321  *     'facet' to the new rule and recompiles its actions.
4322  *
4323  *   - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4324  *     where it is and recompiles its actions anyway.
4325  *
4326  *   - If any of 'facet''s subfacets correspond to a new flow according to
4327  *     xlate_receive(), 'facet' is removed.
4328  *
4329  *   Returns true if 'facet' is still valid.  False if 'facet' was removed. */
4330 static bool
4331 facet_revalidate(struct facet *facet)
4332 {
4333     struct ofproto_dpif *ofproto = facet->ofproto;
4334     struct rule_dpif *new_rule;
4335     struct subfacet *subfacet;
4336     struct flow_wildcards wc;
4337     struct xlate_out xout;
4338     struct xlate_in xin;
4339
4340     COVERAGE_INC(facet_revalidate);
4341
4342     /* Check that child subfacets still correspond to this facet.  Tunnel
4343      * configuration changes could cause a subfacet's OpenFlow in_port to
4344      * change. */
4345     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4346         struct ofproto_dpif *recv_ofproto;
4347         struct flow recv_flow;
4348         int error;
4349
4350         error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4351                               subfacet->key_len, &recv_flow, NULL,
4352                               &recv_ofproto, NULL);
4353         if (error
4354             || recv_ofproto != ofproto
4355             || facet != facet_find(ofproto, &recv_flow)) {
4356             facet_remove(facet);
4357             return false;
4358         }
4359     }
4360
4361     flow_wildcards_init_catchall(&wc);
4362     rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4363
4364     /* Calculate new datapath actions.
4365      *
4366      * We do not modify any 'facet' state yet, because we might need to, e.g.,
4367      * emit a NetFlow expiration and, if so, we need to have the old state
4368      * around to properly compose it. */
4369     xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4370     xlate_actions(&xin, &xout);
4371     flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4372
4373     /* A facet's slow path reason should only change under dramatic
4374      * circumstances.  Rather than try to update everything, it's simpler to
4375      * remove the facet and start over.
4376      *
4377      * More importantly, if a facet's wildcards change, it will be relatively
4378      * difficult to figure out if its subfacets still belong to it, and if not
4379      * which facet they may belong to.  Again, to avoid the complexity, we
4380      * simply give up instead. */
4381     if (facet->xout.slow != xout.slow
4382         || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4383         facet_remove(facet);
4384         xlate_out_uninit(&xout);
4385         rule_release(new_rule);
4386         return false;
4387     }
4388
4389     if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4390         LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4391             if (subfacet->path == SF_FAST_PATH) {
4392                 struct dpif_flow_stats stats;
4393
4394                 subfacet_install(subfacet, &xout.odp_actions, &stats);
4395                 subfacet_update_stats(subfacet, &stats);
4396             }
4397         }
4398
4399         facet_flush_stats(facet);
4400
4401         ofpbuf_clear(&facet->xout.odp_actions);
4402         ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4403                    xout.odp_actions.size);
4404     }
4405
4406     /* Update 'facet' now that we've taken care of all the old state. */
4407     facet->xout.slow = xout.slow;
4408     facet->xout.has_learn = xout.has_learn;
4409     facet->xout.has_normal = xout.has_normal;
4410     facet->xout.has_fin_timeout = xout.has_fin_timeout;
4411     facet->xout.nf_output_iface = xout.nf_output_iface;
4412     facet->xout.mirrors = xout.mirrors;
4413     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4414     facet->used = MAX(facet->used, new_rule->up.created);
4415
4416     xlate_out_uninit(&xout);
4417     rule_release(new_rule);
4418     return true;
4419 }
4420
4421 static void
4422 facet_reset_counters(struct facet *facet)
4423 {
4424     facet->packet_count = 0;
4425     facet->byte_count = 0;
4426     facet->prev_packet_count = 0;
4427     facet->prev_byte_count = 0;
4428     facet->accounted_bytes = 0;
4429 }
4430
4431 static void
4432 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4433                 struct dpif_flow_stats *stats, bool may_learn)
4434 {
4435     struct ofport_dpif *in_port;
4436     struct rule_dpif *rule;
4437     struct xlate_in xin;
4438
4439     in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4440     if (in_port && in_port->is_tunnel) {
4441         netdev_vport_inc_rx(in_port->up.netdev, stats);
4442     }
4443
4444     rule_dpif_lookup(ofproto, flow, NULL, &rule);
4445     rule_credit_stats(rule, stats);
4446     xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4447     xin.resubmit_stats = stats;
4448     xin.may_learn = may_learn;
4449     xlate_actions_for_side_effects(&xin);
4450     rule_release(rule);
4451 }
4452
4453 static void
4454 facet_push_stats(struct facet *facet, bool may_learn)
4455 {
4456     struct dpif_flow_stats stats;
4457
4458     ovs_assert(facet->packet_count >= facet->prev_packet_count);
4459     ovs_assert(facet->byte_count >= facet->prev_byte_count);
4460     ovs_assert(facet->used >= facet->prev_used);
4461
4462     stats.n_packets = facet->packet_count - facet->prev_packet_count;
4463     stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4464     stats.used = facet->used;
4465     stats.tcp_flags = facet->tcp_flags;
4466
4467     if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4468         facet->prev_packet_count = facet->packet_count;
4469         facet->prev_byte_count = facet->byte_count;
4470         facet->prev_used = facet->used;
4471
4472         netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4473                                  facet->used);
4474         netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4475         mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4476                             stats.n_packets, stats.n_bytes);
4477         flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4478     }
4479 }
4480
4481 static void
4482 push_all_stats__(bool run_fast)
4483 {
4484     static long long int rl = LLONG_MIN;
4485     struct ofproto_dpif *ofproto;
4486
4487     if (time_msec() < rl) {
4488         return;
4489     }
4490
4491     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4492         struct cls_cursor cursor;
4493         struct facet *facet;
4494
4495         ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4496         cls_cursor_init(&cursor, &ofproto->facets, NULL);
4497         CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4498             facet_push_stats(facet, false);
4499             if (run_fast) {
4500                 run_fast_rl();
4501             }
4502         }
4503         ovs_rwlock_unlock(&ofproto->facets.rwlock);
4504     }
4505
4506     rl = time_msec() + 100;
4507 }
4508
4509 static void
4510 push_all_stats(void)
4511 {
4512     push_all_stats__(true);
4513 }
4514
4515 void
4516 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4517 {
4518     ovs_mutex_lock(&rule->stats_mutex);
4519     rule->packet_count += stats->n_packets;
4520     rule->byte_count += stats->n_bytes;
4521     ofproto_rule_update_used(&rule->up, stats->used);
4522     ovs_mutex_unlock(&rule->stats_mutex);
4523 }
4524 \f
4525 /* Subfacets. */
4526
4527 static struct subfacet *
4528 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4529               size_t key_len, uint32_t key_hash)
4530 {
4531     struct subfacet *subfacet;
4532
4533     HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4534                              &backer->subfacets) {
4535         if (subfacet->key_len == key_len
4536             && !memcmp(key, subfacet->key, key_len)) {
4537             return subfacet;
4538         }
4539     }
4540
4541     return NULL;
4542 }
4543
4544 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4545  * 'key_fitness', 'key', and 'key_len' members in 'miss'.  Returns the
4546  * existing subfacet if there is one, otherwise creates and returns a
4547  * new subfacet. */
4548 static struct subfacet *
4549 subfacet_create(struct facet *facet, struct flow_miss *miss)
4550 {
4551     struct dpif_backer *backer = miss->ofproto->backer;
4552     enum odp_key_fitness key_fitness = miss->key_fitness;
4553     const struct nlattr *key = miss->key;
4554     size_t key_len = miss->key_len;
4555     uint32_t key_hash;
4556     struct subfacet *subfacet;
4557
4558     key_hash = odp_flow_key_hash(key, key_len);
4559
4560     if (list_is_empty(&facet->subfacets)) {
4561         subfacet = &facet->one_subfacet;
4562     } else {
4563         subfacet = subfacet_find(backer, key, key_len, key_hash);
4564         if (subfacet) {
4565             if (subfacet->facet == facet) {
4566                 return subfacet;
4567             }
4568
4569             /* This shouldn't happen. */
4570             VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4571             subfacet_destroy(subfacet);
4572         }
4573
4574         subfacet = xmalloc(sizeof *subfacet);
4575     }
4576
4577     hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4578     list_push_back(&facet->subfacets, &subfacet->list_node);
4579     subfacet->facet = facet;
4580     subfacet->key_fitness = key_fitness;
4581     subfacet->key = xmemdup(key, key_len);
4582     subfacet->key_len = key_len;
4583     subfacet->used = miss->stats.used;
4584     subfacet->created = subfacet->used;
4585     subfacet->dp_packet_count = 0;
4586     subfacet->dp_byte_count = 0;
4587     subfacet->path = SF_NOT_INSTALLED;
4588     subfacet->backer = backer;
4589
4590     backer->subfacet_add_count++;
4591     return subfacet;
4592 }
4593
4594 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4595  * its facet within 'ofproto', and frees it. */
4596 static void
4597 subfacet_destroy__(struct subfacet *subfacet)
4598 {
4599     struct facet *facet = subfacet->facet;
4600     struct ofproto_dpif *ofproto = facet->ofproto;
4601
4602     /* Update ofproto stats before uninstall the subfacet. */
4603     ofproto->backer->subfacet_del_count++;
4604
4605     subfacet_uninstall(subfacet);
4606     hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4607     list_remove(&subfacet->list_node);
4608     free(subfacet->key);
4609     if (subfacet != &facet->one_subfacet) {
4610         free(subfacet);
4611     }
4612 }
4613
4614 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4615  * last remaining subfacet in its facet destroys the facet too. */
4616 static void
4617 subfacet_destroy(struct subfacet *subfacet)
4618 {
4619     struct facet *facet = subfacet->facet;
4620
4621     if (list_is_singleton(&facet->subfacets)) {
4622         /* facet_remove() needs at least one subfacet (it will remove it). */
4623         facet_remove(facet);
4624     } else {
4625         subfacet_destroy__(subfacet);
4626     }
4627 }
4628
4629 static void
4630 subfacet_destroy_batch(struct dpif_backer *backer,
4631                        struct subfacet **subfacets, int n)
4632 {
4633     struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4634     struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4635     struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4636     int i;
4637
4638     for (i = 0; i < n; i++) {
4639         ops[i].type = DPIF_OP_FLOW_DEL;
4640         ops[i].u.flow_del.key = subfacets[i]->key;
4641         ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4642         ops[i].u.flow_del.stats = &stats[i];
4643         opsp[i] = &ops[i];
4644     }
4645
4646     dpif_operate(backer->dpif, opsp, n);
4647     for (i = 0; i < n; i++) {
4648         subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4649         subfacets[i]->path = SF_NOT_INSTALLED;
4650         subfacet_destroy(subfacets[i]);
4651         run_fast_rl();
4652     }
4653 }
4654
4655 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4656  * bytes of actions in 'actions'.  If 'stats' is non-null, statistics counters
4657  * in the datapath will be zeroed and 'stats' will be updated with traffic new
4658  * since 'subfacet' was last updated.
4659  *
4660  * Returns 0 if successful, otherwise a positive errno value. */
4661 static int
4662 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4663                  struct dpif_flow_stats *stats)
4664 {
4665     struct facet *facet = subfacet->facet;
4666     enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4667     const struct nlattr *actions = odp_actions->data;
4668     size_t actions_len = odp_actions->size;
4669     struct odputil_keybuf maskbuf;
4670     struct ofpbuf mask;
4671
4672     uint64_t slow_path_stub[128 / 8];
4673     enum dpif_flow_put_flags flags;
4674     int ret;
4675
4676     flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4677                                                : DPIF_FP_MODIFY;
4678     if (stats) {
4679         flags |= DPIF_FP_ZERO_STATS;
4680     }
4681
4682     if (path == SF_SLOW_PATH) {
4683         compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4684                           slow_path_stub, sizeof slow_path_stub,
4685                           &actions, &actions_len);
4686     }
4687
4688     ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4689     if (enable_megaflows) {
4690         odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4691                                &facet->flow, UINT32_MAX);
4692     }
4693
4694     ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4695                         subfacet->key_len,  mask.data, mask.size,
4696                         actions, actions_len, stats);
4697
4698     if (stats) {
4699         subfacet_reset_dp_stats(subfacet, stats);
4700     }
4701
4702     if (ret) {
4703         COVERAGE_INC(subfacet_install_fail);
4704     } else {
4705         subfacet->path = path;
4706     }
4707     return ret;
4708 }
4709
4710 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4711 static void
4712 subfacet_uninstall(struct subfacet *subfacet)
4713 {
4714     if (subfacet->path != SF_NOT_INSTALLED) {
4715         struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4716         struct dpif_flow_stats stats;
4717         int error;
4718
4719         error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4720                               subfacet->key_len, &stats);
4721         subfacet_reset_dp_stats(subfacet, &stats);
4722         if (!error) {
4723             subfacet_update_stats(subfacet, &stats);
4724         }
4725         subfacet->path = SF_NOT_INSTALLED;
4726     } else {
4727         ovs_assert(subfacet->dp_packet_count == 0);
4728         ovs_assert(subfacet->dp_byte_count == 0);
4729     }
4730 }
4731
4732 /* Resets 'subfacet''s datapath statistics counters.  This should be called
4733  * when 'subfacet''s statistics are cleared in the datapath.  If 'stats' is
4734  * non-null, it should contain the statistics returned by dpif when 'subfacet'
4735  * was reset in the datapath.  'stats' will be modified to include only
4736  * statistics new since 'subfacet' was last updated. */
4737 static void
4738 subfacet_reset_dp_stats(struct subfacet *subfacet,
4739                         struct dpif_flow_stats *stats)
4740 {
4741     if (stats
4742         && subfacet->dp_packet_count <= stats->n_packets
4743         && subfacet->dp_byte_count <= stats->n_bytes) {
4744         stats->n_packets -= subfacet->dp_packet_count;
4745         stats->n_bytes -= subfacet->dp_byte_count;
4746     }
4747
4748     subfacet->dp_packet_count = 0;
4749     subfacet->dp_byte_count = 0;
4750 }
4751
4752 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4753  *
4754  * Because of the meaning of a subfacet's counters, it only makes sense to do
4755  * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4756  * represents a packet that was sent by hand or if it represents statistics
4757  * that have been cleared out of the datapath. */
4758 static void
4759 subfacet_update_stats(struct subfacet *subfacet,
4760                       const struct dpif_flow_stats *stats)
4761 {
4762     if (stats->n_packets || stats->used > subfacet->used) {
4763         struct facet *facet = subfacet->facet;
4764
4765         subfacet->used = MAX(subfacet->used, stats->used);
4766         facet->used = MAX(facet->used, stats->used);
4767         facet->packet_count += stats->n_packets;
4768         facet->byte_count += stats->n_bytes;
4769         facet->tcp_flags |= stats->tcp_flags;
4770     }
4771 }
4772 \f
4773 /* Rules. */
4774
4775 /* Lookup 'flow' in 'ofproto''s classifier.  If 'wc' is non-null, sets
4776  * the fields that were relevant as part of the lookup. */
4777 void
4778 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4779                  struct flow_wildcards *wc, struct rule_dpif **rule)
4780 {
4781     struct ofport_dpif *port;
4782
4783     if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4784         return;
4785     }
4786     port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4787     if (!port) {
4788         VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4789                      flow->in_port.ofp_port);
4790     }
4791
4792     *rule = choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4793                              ofproto->no_packet_in_rule);
4794     ovs_rwlock_rdlock(&(*rule)->up.evict);
4795 }
4796
4797 bool
4798 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4799                           const struct flow *flow, struct flow_wildcards *wc,
4800                           uint8_t table_id, struct rule_dpif **rule)
4801     OVS_TRY_RDLOCK(true, (*rule)->up.evict)
4802 {
4803     struct cls_rule *cls_rule;
4804     struct classifier *cls;
4805     bool frag;
4806
4807     *rule = NULL;
4808     if (table_id >= N_TABLES) {
4809         return false;
4810     }
4811
4812     if (wc) {
4813         memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4814         wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4815     }
4816
4817     cls = &ofproto->up.tables[table_id].cls;
4818     ovs_rwlock_rdlock(&cls->rwlock);
4819     frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4820     if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4821         /* We must pretend that transport ports are unavailable. */
4822         struct flow ofpc_normal_flow = *flow;
4823         ofpc_normal_flow.tp_src = htons(0);
4824         ofpc_normal_flow.tp_dst = htons(0);
4825         cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4826     } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4827         cls_rule = &ofproto->drop_frags_rule->up.cr;
4828         if (wc) {
4829             flow_wildcards_init_exact(wc);
4830         }
4831     } else {
4832         cls_rule = classifier_lookup(cls, flow, wc);
4833     }
4834
4835     *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4836     if (*rule && ovs_rwlock_tryrdlock(&(*rule)->up.evict)) {
4837         /* The rule is in the process of being removed.  Best we can do is
4838          * pretend it isn't there. */
4839         *rule = NULL;
4840     }
4841     ovs_rwlock_unlock(&cls->rwlock);
4842
4843     return *rule != NULL;
4844 }
4845
4846 /* Given a port configuration (specified as zero if there's no port), chooses
4847  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4848  * flow table miss. */
4849 struct rule_dpif *
4850 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4851                  struct rule_dpif *no_packet_in_rule)
4852 {
4853     return config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4854 }
4855
4856 void
4857 rule_release(struct rule_dpif *rule)
4858     OVS_NO_THREAD_SAFETY_ANALYSIS
4859 {
4860     if (rule) {
4861         ovs_rwlock_unlock(&rule->up.evict);
4862     }
4863 }
4864
4865 static void
4866 complete_operation(struct rule_dpif *rule)
4867 {
4868     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4869
4870     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4871     if (clogged) {
4872         struct dpif_completion *c = xmalloc(sizeof *c);
4873         c->op = rule->up.pending;
4874         list_push_back(&ofproto->completions, &c->list_node);
4875     } else {
4876         ofoperation_complete(rule->up.pending, 0);
4877     }
4878 }
4879
4880 static struct rule *
4881 rule_alloc(void)
4882 {
4883     struct rule_dpif *rule = xmalloc(sizeof *rule);
4884     return &rule->up;
4885 }
4886
4887 static void
4888 rule_dealloc(struct rule *rule_)
4889 {
4890     struct rule_dpif *rule = rule_dpif_cast(rule_);
4891     free(rule);
4892 }
4893
4894 static enum ofperr
4895 rule_construct(struct rule *rule_)
4896 {
4897     struct rule_dpif *rule = rule_dpif_cast(rule_);
4898     ovs_mutex_init(&rule->stats_mutex);
4899     ovs_mutex_lock(&rule->stats_mutex);
4900     rule->packet_count = 0;
4901     rule->byte_count = 0;
4902     ovs_mutex_unlock(&rule->stats_mutex);
4903     return 0;
4904 }
4905
4906 static void
4907 rule_insert(struct rule *rule_)
4908 {
4909     struct rule_dpif *rule = rule_dpif_cast(rule_);
4910     complete_operation(rule);
4911 }
4912
4913 static void
4914 rule_delete(struct rule *rule_)
4915 {
4916     struct rule_dpif *rule = rule_dpif_cast(rule_);
4917     complete_operation(rule);
4918 }
4919
4920 static void
4921 rule_destruct(struct rule *rule_)
4922 {
4923     struct rule_dpif *rule = rule_dpif_cast(rule_);
4924     ovs_mutex_destroy(&rule->stats_mutex);
4925 }
4926
4927 static void
4928 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4929 {
4930     struct rule_dpif *rule = rule_dpif_cast(rule_);
4931
4932     /* push_all_stats() can handle flow misses which, when using the learn
4933      * action, can cause rules to be added and deleted.  This can corrupt our
4934      * caller's datastructures which assume that rule_get_stats() doesn't have
4935      * an impact on the flow table. To be safe, we disable miss handling. */
4936     push_all_stats__(false);
4937
4938     /* Start from historical data for 'rule' itself that are no longer tracked
4939      * in facets.  This counts, for example, facets that have expired. */
4940     ovs_mutex_lock(&rule->stats_mutex);
4941     *packets = rule->packet_count;
4942     *bytes = rule->byte_count;
4943     ovs_mutex_unlock(&rule->stats_mutex);
4944 }
4945
4946 static void
4947 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4948                   struct ofpbuf *packet)
4949 {
4950     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4951     struct dpif_flow_stats stats;
4952     struct xlate_out xout;
4953     struct xlate_in xin;
4954
4955     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4956     rule_credit_stats(rule, &stats);
4957
4958     xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4959     xin.resubmit_stats = &stats;
4960     xlate_actions(&xin, &xout);
4961
4962     execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4963                         xout.odp_actions.size, packet);
4964
4965     xlate_out_uninit(&xout);
4966 }
4967
4968 static enum ofperr
4969 rule_execute(struct rule *rule, const struct flow *flow,
4970              struct ofpbuf *packet)
4971 {
4972     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4973     ofpbuf_delete(packet);
4974     return 0;
4975 }
4976
4977 static void
4978 rule_modify_actions(struct rule *rule_, bool reset_counters)
4979 {
4980     struct rule_dpif *rule = rule_dpif_cast(rule_);
4981
4982     if (reset_counters) {
4983         ovs_mutex_lock(&rule->stats_mutex);
4984         rule->packet_count = 0;
4985         rule->byte_count = 0;
4986         ovs_mutex_unlock(&rule->stats_mutex);
4987     }
4988
4989     complete_operation(rule);
4990 }
4991 \f
4992 /* Sends 'packet' out 'ofport'.
4993  * May modify 'packet'.
4994  * Returns 0 if successful, otherwise a positive errno value. */
4995 static int
4996 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4997 {
4998     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4999     uint64_t odp_actions_stub[1024 / 8];
5000     struct ofpbuf key, odp_actions;
5001     struct dpif_flow_stats stats;
5002     struct odputil_keybuf keybuf;
5003     struct ofpact_output output;
5004     struct xlate_out xout;
5005     struct xlate_in xin;
5006     struct flow flow;
5007     union flow_in_port in_port_;
5008     int error;
5009
5010     ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5011     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5012
5013     /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5014     in_port_.ofp_port = OFPP_NONE;
5015     flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5016     odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5017                                                              OFPP_LOCAL));
5018     dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5019
5020     ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5021     output.port = ofport->up.ofp_port;
5022     output.max_len = 0;
5023
5024     xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5025     xin.ofpacts_len = sizeof output;
5026     xin.ofpacts = &output.ofpact;
5027     xin.resubmit_stats = &stats;
5028     xlate_actions(&xin, &xout);
5029
5030     error = dpif_execute(ofproto->backer->dpif,
5031                          key.data, key.size,
5032                          xout.odp_actions.data, xout.odp_actions.size,
5033                          packet);
5034     xlate_out_uninit(&xout);
5035
5036     if (error) {
5037         VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5038                      ofproto->up.name, netdev_get_name(ofport->up.netdev),
5039                      ovs_strerror(error));
5040     }
5041
5042     ofproto->stats.tx_packets++;
5043     ofproto->stats.tx_bytes += packet->size;
5044     return error;
5045 }
5046
5047 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5048  * The action will state 'slow' as the reason that the action is in the slow
5049  * path.  (This is purely informational: it allows a human viewing "ovs-dpctl
5050  * dump-flows" output to see why a flow is in the slow path.)
5051  *
5052  * The 'stub_size' bytes in 'stub' will be used to store the action.
5053  * 'stub_size' must be large enough for the action.
5054  *
5055  * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5056  * respectively. */
5057 static void
5058 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5059                   enum slow_path_reason slow,
5060                   uint64_t *stub, size_t stub_size,
5061                   const struct nlattr **actionsp, size_t *actions_lenp)
5062 {
5063     union user_action_cookie cookie;
5064     struct ofpbuf buf;
5065
5066     cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5067     cookie.slow_path.unused = 0;
5068     cookie.slow_path.reason = slow;
5069
5070     ofpbuf_use_stack(&buf, stub, stub_size);
5071     if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5072         uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5073                                          ODPP_NONE);
5074         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5075     } else {
5076         odp_port_t odp_port;
5077         uint32_t pid;
5078
5079         odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5080         pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5081         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5082     }
5083     *actionsp = buf.data;
5084     *actions_lenp = buf.size;
5085 }
5086 \f
5087 static bool
5088 set_frag_handling(struct ofproto *ofproto_,
5089                   enum ofp_config_flags frag_handling)
5090 {
5091     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5092     if (frag_handling != OFPC_FRAG_REASM) {
5093         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5094         return true;
5095     } else {
5096         return false;
5097     }
5098 }
5099
5100 static enum ofperr
5101 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5102            const struct flow *flow,
5103            const struct ofpact *ofpacts, size_t ofpacts_len)
5104 {
5105     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5106     struct odputil_keybuf keybuf;
5107     struct dpif_flow_stats stats;
5108     struct xlate_out xout;
5109     struct xlate_in xin;
5110     struct ofpbuf key;
5111
5112
5113     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5114     odp_flow_key_from_flow(&key, flow,
5115                            ofp_port_to_odp_port(ofproto,
5116                                       flow->in_port.ofp_port));
5117
5118     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5119
5120     xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5121     xin.resubmit_stats = &stats;
5122     xin.ofpacts_len = ofpacts_len;
5123     xin.ofpacts = ofpacts;
5124
5125     xlate_actions(&xin, &xout);
5126     dpif_execute(ofproto->backer->dpif, key.data, key.size,
5127                  xout.odp_actions.data, xout.odp_actions.size, packet);
5128     xlate_out_uninit(&xout);
5129
5130     return 0;
5131 }
5132 \f
5133 /* NetFlow. */
5134
5135 static int
5136 set_netflow(struct ofproto *ofproto_,
5137             const struct netflow_options *netflow_options)
5138 {
5139     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5140
5141     if (netflow_options) {
5142         if (!ofproto->netflow) {
5143             ofproto->netflow = netflow_create();
5144             ofproto->backer->need_revalidate = REV_RECONFIGURE;
5145         }
5146         return netflow_set_options(ofproto->netflow, netflow_options);
5147     } else if (ofproto->netflow) {
5148         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5149         netflow_destroy(ofproto->netflow);
5150         ofproto->netflow = NULL;
5151     }
5152
5153     return 0;
5154 }
5155
5156 static void
5157 get_netflow_ids(const struct ofproto *ofproto_,
5158                 uint8_t *engine_type, uint8_t *engine_id)
5159 {
5160     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5161
5162     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5163 }
5164
5165 static void
5166 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5167 {
5168     if (!facet_is_controller_flow(facet) &&
5169         netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5170         struct subfacet *subfacet;
5171         struct ofexpired expired;
5172
5173         LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5174             if (subfacet->path == SF_FAST_PATH) {
5175                 struct dpif_flow_stats stats;
5176
5177                 subfacet_install(subfacet, &facet->xout.odp_actions,
5178                                  &stats);
5179                 subfacet_update_stats(subfacet, &stats);
5180             }
5181         }
5182
5183         expired.flow = facet->flow;
5184         expired.packet_count = facet->packet_count;
5185         expired.byte_count = facet->byte_count;
5186         expired.used = facet->used;
5187         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5188     }
5189 }
5190
5191 static void
5192 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5193 {
5194     struct cls_cursor cursor;
5195     struct facet *facet;
5196
5197     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5198     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5199     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5200         send_active_timeout(ofproto, facet);
5201     }
5202     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5203 }
5204 \f
5205 static struct ofproto_dpif *
5206 ofproto_dpif_lookup(const char *name)
5207 {
5208     struct ofproto_dpif *ofproto;
5209
5210     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5211                              hash_string(name, 0), &all_ofproto_dpifs) {
5212         if (!strcmp(ofproto->up.name, name)) {
5213             return ofproto;
5214         }
5215     }
5216     return NULL;
5217 }
5218
5219 static void
5220 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5221                           const char *argv[], void *aux OVS_UNUSED)
5222 {
5223     struct ofproto_dpif *ofproto;
5224
5225     if (argc > 1) {
5226         ofproto = ofproto_dpif_lookup(argv[1]);
5227         if (!ofproto) {
5228             unixctl_command_reply_error(conn, "no such bridge");
5229             return;
5230         }
5231         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5232         mac_learning_flush(ofproto->ml);
5233         ovs_rwlock_unlock(&ofproto->ml->rwlock);
5234     } else {
5235         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5236             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5237             mac_learning_flush(ofproto->ml);
5238             ovs_rwlock_unlock(&ofproto->ml->rwlock);
5239         }
5240     }
5241
5242     unixctl_command_reply(conn, "table successfully flushed");
5243 }
5244
5245 static struct ofport_dpif *
5246 ofbundle_get_a_port(const struct ofbundle *bundle)
5247 {
5248     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5249                         bundle_node);
5250 }
5251
5252 static void
5253 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5254                          const char *argv[], void *aux OVS_UNUSED)
5255 {
5256     struct ds ds = DS_EMPTY_INITIALIZER;
5257     const struct ofproto_dpif *ofproto;
5258     const struct mac_entry *e;
5259
5260     ofproto = ofproto_dpif_lookup(argv[1]);
5261     if (!ofproto) {
5262         unixctl_command_reply_error(conn, "no such bridge");
5263         return;
5264     }
5265
5266     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
5267     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5268     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5269         struct ofbundle *bundle = e->port.p;
5270         char name[OFP_MAX_PORT_NAME_LEN];
5271
5272         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5273                                name, sizeof name);
5274         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
5275                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
5276                       mac_entry_age(ofproto->ml, e));
5277     }
5278     ovs_rwlock_unlock(&ofproto->ml->rwlock);
5279     unixctl_command_reply(conn, ds_cstr(&ds));
5280     ds_destroy(&ds);
5281 }
5282
5283 struct trace_ctx {
5284     struct xlate_out xout;
5285     struct xlate_in xin;
5286     struct flow flow;
5287     struct ds *result;
5288 };
5289
5290 static void
5291 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5292 {
5293     ds_put_char_multiple(result, '\t', level);
5294     if (!rule) {
5295         ds_put_cstr(result, "No match\n");
5296         return;
5297     }
5298
5299     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5300                   rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5301     cls_rule_format(&rule->up.cr, result);
5302     ds_put_char(result, '\n');
5303
5304     ds_put_char_multiple(result, '\t', level);
5305     ds_put_cstr(result, "OpenFlow ");
5306     ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5307     ds_put_char(result, '\n');
5308 }
5309
5310 static void
5311 trace_format_flow(struct ds *result, int level, const char *title,
5312                   struct trace_ctx *trace)
5313 {
5314     ds_put_char_multiple(result, '\t', level);
5315     ds_put_format(result, "%s: ", title);
5316     if (flow_equal(&trace->xin.flow, &trace->flow)) {
5317         ds_put_cstr(result, "unchanged");
5318     } else {
5319         flow_format(result, &trace->xin.flow);
5320         trace->flow = trace->xin.flow;
5321     }
5322     ds_put_char(result, '\n');
5323 }
5324
5325 static void
5326 trace_format_regs(struct ds *result, int level, const char *title,
5327                   struct trace_ctx *trace)
5328 {
5329     size_t i;
5330
5331     ds_put_char_multiple(result, '\t', level);
5332     ds_put_format(result, "%s:", title);
5333     for (i = 0; i < FLOW_N_REGS; i++) {
5334         ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5335     }
5336     ds_put_char(result, '\n');
5337 }
5338
5339 static void
5340 trace_format_odp(struct ds *result, int level, const char *title,
5341                  struct trace_ctx *trace)
5342 {
5343     struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5344
5345     ds_put_char_multiple(result, '\t', level);
5346     ds_put_format(result, "%s: ", title);
5347     format_odp_actions(result, odp_actions->data, odp_actions->size);
5348     ds_put_char(result, '\n');
5349 }
5350
5351 static void
5352 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5353 {
5354     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5355     struct ds *result = trace->result;
5356
5357     ds_put_char(result, '\n');
5358     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5359     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5360     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
5361     trace_format_rule(result, recurse + 1, rule);
5362 }
5363
5364 static void
5365 trace_report(struct xlate_in *xin, const char *s, int recurse)
5366 {
5367     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5368     struct ds *result = trace->result;
5369
5370     ds_put_char_multiple(result, '\t', recurse);
5371     ds_put_cstr(result, s);
5372     ds_put_char(result, '\n');
5373 }
5374
5375 static void
5376 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5377                       void *aux OVS_UNUSED)
5378 {
5379     const struct dpif_backer *backer;
5380     struct ofproto_dpif *ofproto;
5381     struct ofpbuf odp_key, odp_mask;
5382     struct ofpbuf *packet;
5383     struct ds result;
5384     struct flow flow;
5385     char *s;
5386
5387     packet = NULL;
5388     backer = NULL;
5389     ds_init(&result);
5390     ofpbuf_init(&odp_key, 0);
5391     ofpbuf_init(&odp_mask, 0);
5392
5393     /* Handle "-generate" or a hex string as the last argument. */
5394     if (!strcmp(argv[argc - 1], "-generate")) {
5395         packet = ofpbuf_new(0);
5396         argc--;
5397     } else {
5398         const char *error = eth_from_hex(argv[argc - 1], &packet);
5399         if (!error) {
5400             argc--;
5401         } else if (argc == 4) {
5402             /* The 3-argument form must end in "-generate' or a hex string. */
5403             unixctl_command_reply_error(conn, error);
5404             goto exit;
5405         }
5406     }
5407
5408     /* Parse the flow and determine whether a datapath or
5409      * bridge is specified. If function odp_flow_key_from_string()
5410      * returns 0, the flow is a odp_flow. If function
5411      * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5412     if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5413         /* If the odp_flow is the second argument,
5414          * the datapath name is the first argument. */
5415         if (argc == 3) {
5416             const char *dp_type;
5417             if (!strncmp(argv[1], "ovs-", 4)) {
5418                 dp_type = argv[1] + 4;
5419             } else {
5420                 dp_type = argv[1];
5421             }
5422             backer = shash_find_data(&all_dpif_backers, dp_type);
5423             if (!backer) {
5424                 unixctl_command_reply_error(conn, "Cannot find datapath "
5425                                "of this name");
5426                 goto exit;
5427             }
5428         } else {
5429             /* No datapath name specified, so there should be only one
5430              * datapath. */
5431             struct shash_node *node;
5432             if (shash_count(&all_dpif_backers) != 1) {
5433                 unixctl_command_reply_error(conn, "Must specify datapath "
5434                          "name, there is more than one type of datapath");
5435                 goto exit;
5436             }
5437             node = shash_first(&all_dpif_backers);
5438             backer = node->data;
5439         }
5440
5441         if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5442                           NULL, &ofproto, NULL)) {
5443             unixctl_command_reply_error(conn, "Invalid datapath flow");
5444             goto exit;
5445         }
5446         ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5447     } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5448         if (argc != 3) {
5449             unixctl_command_reply_error(conn, "Must specify bridge name");
5450             goto exit;
5451         }
5452
5453         ofproto = ofproto_dpif_lookup(argv[1]);
5454         if (!ofproto) {
5455             unixctl_command_reply_error(conn, "Unknown bridge name");
5456             goto exit;
5457         }
5458     } else {
5459         unixctl_command_reply_error(conn, "Bad flow syntax");
5460         goto exit;
5461     }
5462
5463     /* Generate a packet, if requested. */
5464     if (packet) {
5465         if (!packet->size) {
5466             flow_compose(packet, &flow);
5467         } else {
5468             union flow_in_port in_port_;
5469
5470             in_port_ = flow.in_port;
5471             ds_put_cstr(&result, "Packet: ");
5472             s = ofp_packet_to_string(packet->data, packet->size);
5473             ds_put_cstr(&result, s);
5474             free(s);
5475
5476             /* Use the metadata from the flow and the packet argument
5477              * to reconstruct the flow. */
5478             flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5479                          &in_port_, &flow);
5480         }
5481     }
5482
5483     ofproto_trace(ofproto, &flow, packet, &result);
5484     unixctl_command_reply(conn, ds_cstr(&result));
5485
5486 exit:
5487     ds_destroy(&result);
5488     ofpbuf_delete(packet);
5489     ofpbuf_uninit(&odp_key);
5490     ofpbuf_uninit(&odp_mask);
5491 }
5492
5493 static void
5494 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5495               const struct ofpbuf *packet, struct ds *ds)
5496 {
5497     struct rule_dpif *rule;
5498     struct flow_wildcards wc;
5499
5500     ds_put_cstr(ds, "Flow: ");
5501     flow_format(ds, flow);
5502     ds_put_char(ds, '\n');
5503
5504     flow_wildcards_init_catchall(&wc);
5505     rule_dpif_lookup(ofproto, flow, &wc, &rule);
5506
5507     trace_format_rule(ds, 0, rule);
5508     if (rule == ofproto->miss_rule) {
5509         ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5510     } else if (rule == ofproto->no_packet_in_rule) {
5511         ds_put_cstr(ds, "\nNo match, packets dropped because "
5512                     "OFPPC_NO_PACKET_IN is set on in_port.\n");
5513     } else if (rule == ofproto->drop_frags_rule) {
5514         ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5515                     "and the fragment handling mode is \"drop\".\n");
5516     }
5517
5518     if (rule) {
5519         uint64_t odp_actions_stub[1024 / 8];
5520         struct ofpbuf odp_actions;
5521         struct trace_ctx trace;
5522         struct match match;
5523         uint8_t tcp_flags;
5524
5525         tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5526         trace.result = ds;
5527         trace.flow = *flow;
5528         ofpbuf_use_stub(&odp_actions,
5529                         odp_actions_stub, sizeof odp_actions_stub);
5530         xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5531         trace.xin.resubmit_hook = trace_resubmit;
5532         trace.xin.report_hook = trace_report;
5533
5534         xlate_actions(&trace.xin, &trace.xout);
5535         flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5536
5537         ds_put_char(ds, '\n');
5538         trace_format_flow(ds, 0, "Final flow", &trace);
5539
5540         match_init(&match, flow, &trace.xout.wc);
5541         ds_put_cstr(ds, "Relevant fields: ");
5542         match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5543         ds_put_char(ds, '\n');
5544
5545         ds_put_cstr(ds, "Datapath actions: ");
5546         format_odp_actions(ds, trace.xout.odp_actions.data,
5547                            trace.xout.odp_actions.size);
5548
5549         if (trace.xout.slow) {
5550             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5551                         "slow path because it:");
5552             switch (trace.xout.slow) {
5553             case SLOW_CFM:
5554                 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5555                 break;
5556             case SLOW_LACP:
5557                 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5558                 break;
5559             case SLOW_STP:
5560                 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5561                 break;
5562             case SLOW_BFD:
5563                 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5564                 break;
5565             case SLOW_CONTROLLER:
5566                 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5567                             "to the OpenFlow controller.");
5568                 break;
5569             case __SLOW_MAX:
5570                 NOT_REACHED();
5571             }
5572         }
5573
5574         xlate_out_uninit(&trace.xout);
5575     }
5576
5577     rule_release(rule);
5578 }
5579
5580 static void
5581 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5582                   const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5583 {
5584     clogged = true;
5585     unixctl_command_reply(conn, NULL);
5586 }
5587
5588 static void
5589 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5590                     const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5591 {
5592     clogged = false;
5593     unixctl_command_reply(conn, NULL);
5594 }
5595
5596 /* Runs a self-check of flow translations in 'ofproto'.  Appends a message to
5597  * 'reply' describing the results. */
5598 static void
5599 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5600 {
5601     struct cls_cursor cursor;
5602     struct facet *facet;
5603     int errors;
5604
5605     errors = 0;
5606     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5607     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5608     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5609         if (!facet_check_consistency(facet)) {
5610             errors++;
5611         }
5612     }
5613     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5614     if (errors) {
5615         ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5616     }
5617
5618     if (errors) {
5619         ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5620                       ofproto->up.name, errors);
5621     } else {
5622         ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5623     }
5624 }
5625
5626 static void
5627 ofproto_dpif_self_check(struct unixctl_conn *conn,
5628                         int argc, const char *argv[], void *aux OVS_UNUSED)
5629 {
5630     struct ds reply = DS_EMPTY_INITIALIZER;
5631     struct ofproto_dpif *ofproto;
5632
5633     if (argc > 1) {
5634         ofproto = ofproto_dpif_lookup(argv[1]);
5635         if (!ofproto) {
5636             unixctl_command_reply_error(conn, "Unknown ofproto (use "
5637                                         "ofproto/list for help)");
5638             return;
5639         }
5640         ofproto_dpif_self_check__(ofproto, &reply);
5641     } else {
5642         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5643             ofproto_dpif_self_check__(ofproto, &reply);
5644         }
5645     }
5646
5647     unixctl_command_reply(conn, ds_cstr(&reply));
5648     ds_destroy(&reply);
5649 }
5650
5651 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
5652  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
5653  * to destroy 'ofproto_shash' and free the returned value. */
5654 static const struct shash_node **
5655 get_ofprotos(struct shash *ofproto_shash)
5656 {
5657     const struct ofproto_dpif *ofproto;
5658
5659     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5660         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5661         shash_add_nocopy(ofproto_shash, name, ofproto);
5662     }
5663
5664     return shash_sort(ofproto_shash);
5665 }
5666
5667 static void
5668 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5669                               const char *argv[] OVS_UNUSED,
5670                               void *aux OVS_UNUSED)
5671 {
5672     struct ds ds = DS_EMPTY_INITIALIZER;
5673     struct shash ofproto_shash;
5674     const struct shash_node **sorted_ofprotos;
5675     int i;
5676
5677     shash_init(&ofproto_shash);
5678     sorted_ofprotos = get_ofprotos(&ofproto_shash);
5679     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5680         const struct shash_node *node = sorted_ofprotos[i];
5681         ds_put_format(&ds, "%s\n", node->name);
5682     }
5683
5684     shash_destroy(&ofproto_shash);
5685     free(sorted_ofprotos);
5686
5687     unixctl_command_reply(conn, ds_cstr(&ds));
5688     ds_destroy(&ds);
5689 }
5690
5691 static void
5692 show_dp_rates(struct ds *ds, const char *heading,
5693               const struct avg_subfacet_rates *rates)
5694 {
5695     ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5696                   heading, rates->add_rate, rates->del_rate);
5697 }
5698
5699 static void
5700 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5701 {
5702     const struct shash_node **ofprotos;
5703     struct ofproto_dpif *ofproto;
5704     struct shash ofproto_shash;
5705     uint64_t n_hit, n_missed;
5706     long long int minutes;
5707     size_t i;
5708
5709     n_hit = n_missed = 0;
5710     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5711         if (ofproto->backer == backer) {
5712             n_missed += ofproto->n_missed;
5713             n_hit += ofproto->n_hit;
5714         }
5715     }
5716
5717     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5718                   dpif_name(backer->dpif), n_hit, n_missed);
5719     ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5720                   " life span: %lldms\n", hmap_count(&backer->subfacets),
5721                   backer->avg_n_subfacet, backer->max_n_subfacet,
5722                   backer->avg_subfacet_life);
5723
5724     minutes = (time_msec() - backer->created) / (1000 * 60);
5725     if (minutes >= 60) {
5726         show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5727     }
5728     if (minutes >= 60 * 24) {
5729         show_dp_rates(ds, "\tdaily avg:",  &backer->daily);
5730     }
5731     show_dp_rates(ds, "\toverall avg:",  &backer->lifetime);
5732
5733     shash_init(&ofproto_shash);
5734     ofprotos = get_ofprotos(&ofproto_shash);
5735     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5736         struct ofproto_dpif *ofproto = ofprotos[i]->data;
5737         const struct shash_node **ports;
5738         size_t j;
5739
5740         if (ofproto->backer != backer) {
5741             continue;
5742         }
5743
5744         ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5745                       ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5746
5747         ports = shash_sort(&ofproto->up.port_by_name);
5748         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5749             const struct shash_node *node = ports[j];
5750             struct ofport *ofport = node->data;
5751             struct smap config;
5752             odp_port_t odp_port;
5753
5754             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5755                           ofport->ofp_port);
5756
5757             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5758             if (odp_port != ODPP_NONE) {
5759                 ds_put_format(ds, "%"PRIu32":", odp_port);
5760             } else {
5761                 ds_put_cstr(ds, "none:");
5762             }
5763
5764             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5765
5766             smap_init(&config);
5767             if (!netdev_get_config(ofport->netdev, &config)) {
5768                 const struct smap_node **nodes;
5769                 size_t i;
5770
5771                 nodes = smap_sort(&config);
5772                 for (i = 0; i < smap_count(&config); i++) {
5773                     const struct smap_node *node = nodes[i];
5774                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5775                                   node->key, node->value);
5776                 }
5777                 free(nodes);
5778             }
5779             smap_destroy(&config);
5780
5781             ds_put_char(ds, ')');
5782             ds_put_char(ds, '\n');
5783         }
5784         free(ports);
5785     }
5786     shash_destroy(&ofproto_shash);
5787     free(ofprotos);
5788 }
5789
5790 static void
5791 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5792                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5793 {
5794     struct ds ds = DS_EMPTY_INITIALIZER;
5795     const struct shash_node **backers;
5796     int i;
5797
5798     backers = shash_sort(&all_dpif_backers);
5799     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5800         dpif_show_backer(backers[i]->data, &ds);
5801     }
5802     free(backers);
5803
5804     unixctl_command_reply(conn, ds_cstr(&ds));
5805     ds_destroy(&ds);
5806 }
5807
5808 /* Dump the megaflow (facet) cache.  This is useful to check the
5809  * correctness of flow wildcarding, since the same mechanism is used for
5810  * both xlate caching and kernel wildcarding.
5811  *
5812  * It's important to note that in the output the flow description uses
5813  * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5814  *
5815  * This command is only needed for advanced debugging, so it's not
5816  * documented in the man page. */
5817 static void
5818 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5819                                     int argc OVS_UNUSED, const char *argv[],
5820                                     void *aux OVS_UNUSED)
5821 {
5822     struct ds ds = DS_EMPTY_INITIALIZER;
5823     const struct ofproto_dpif *ofproto;
5824     long long int now = time_msec();
5825     struct cls_cursor cursor;
5826     struct facet *facet;
5827
5828     ofproto = ofproto_dpif_lookup(argv[1]);
5829     if (!ofproto) {
5830         unixctl_command_reply_error(conn, "no such bridge");
5831         return;
5832     }
5833
5834     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5835     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5836     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5837         cls_rule_format(&facet->cr, &ds);
5838         ds_put_cstr(&ds, ", ");
5839         ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5840         ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5841         ds_put_cstr(&ds, "Datapath actions: ");
5842         if (facet->xout.slow) {
5843             uint64_t slow_path_stub[128 / 8];
5844             const struct nlattr *actions;
5845             size_t actions_len;
5846
5847             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5848                               slow_path_stub, sizeof slow_path_stub,
5849                               &actions, &actions_len);
5850             format_odp_actions(&ds, actions, actions_len);
5851         } else {
5852             format_odp_actions(&ds, facet->xout.odp_actions.data,
5853                                facet->xout.odp_actions.size);
5854         }
5855         ds_put_cstr(&ds, "\n");
5856     }
5857     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5858
5859     ds_chomp(&ds, '\n');
5860     unixctl_command_reply(conn, ds_cstr(&ds));
5861     ds_destroy(&ds);
5862 }
5863
5864 /* Disable using the megaflows.
5865  *
5866  * This command is only needed for advanced debugging, so it's not
5867  * documented in the man page. */
5868 static void
5869 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5870                                        int argc OVS_UNUSED,
5871                                        const char *argv[] OVS_UNUSED,
5872                                        void *aux OVS_UNUSED)
5873 {
5874     struct ofproto_dpif *ofproto;
5875
5876     enable_megaflows = false;
5877
5878     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5879         flush(&ofproto->up);
5880     }
5881
5882     unixctl_command_reply(conn, "megaflows disabled");
5883 }
5884
5885 /* Re-enable using megaflows.
5886  *
5887  * This command is only needed for advanced debugging, so it's not
5888  * documented in the man page. */
5889 static void
5890 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5891                                       int argc OVS_UNUSED,
5892                                       const char *argv[] OVS_UNUSED,
5893                                       void *aux OVS_UNUSED)
5894 {
5895     struct ofproto_dpif *ofproto;
5896
5897     enable_megaflows = true;
5898
5899     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5900         flush(&ofproto->up);
5901     }
5902
5903     unixctl_command_reply(conn, "megaflows enabled");
5904 }
5905
5906 static void
5907 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5908                                 int argc OVS_UNUSED, const char *argv[],
5909                                 void *aux OVS_UNUSED)
5910 {
5911     struct ds ds = DS_EMPTY_INITIALIZER;
5912     const struct ofproto_dpif *ofproto;
5913     struct subfacet *subfacet;
5914
5915     ofproto = ofproto_dpif_lookup(argv[1]);
5916     if (!ofproto) {
5917         unixctl_command_reply_error(conn, "no such bridge");
5918         return;
5919     }
5920
5921     update_stats(ofproto->backer);
5922
5923     HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5924         struct facet *facet = subfacet->facet;
5925         struct odputil_keybuf maskbuf;
5926         struct ofpbuf mask;
5927
5928         if (facet->ofproto != ofproto) {
5929             continue;
5930         }
5931
5932         ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5933         if (enable_megaflows) {
5934             odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5935                                    &facet->flow, UINT32_MAX);
5936         }
5937
5938         odp_flow_format(subfacet->key, subfacet->key_len,
5939                         mask.data, mask.size, &ds, false);
5940
5941         ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5942                       subfacet->dp_packet_count, subfacet->dp_byte_count);
5943         if (subfacet->used) {
5944             ds_put_format(&ds, "%.3fs",
5945                           (time_msec() - subfacet->used) / 1000.0);
5946         } else {
5947             ds_put_format(&ds, "never");
5948         }
5949         if (subfacet->facet->tcp_flags) {
5950             ds_put_cstr(&ds, ", flags:");
5951             packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5952         }
5953
5954         ds_put_cstr(&ds, ", actions:");
5955         if (facet->xout.slow) {
5956             uint64_t slow_path_stub[128 / 8];
5957             const struct nlattr *actions;
5958             size_t actions_len;
5959
5960             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5961                               slow_path_stub, sizeof slow_path_stub,
5962                               &actions, &actions_len);
5963             format_odp_actions(&ds, actions, actions_len);
5964         } else {
5965             format_odp_actions(&ds, facet->xout.odp_actions.data,
5966                                facet->xout.odp_actions.size);
5967         }
5968         ds_put_char(&ds, '\n');
5969     }
5970
5971     unixctl_command_reply(conn, ds_cstr(&ds));
5972     ds_destroy(&ds);
5973 }
5974
5975 static void
5976 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5977                                int argc OVS_UNUSED, const char *argv[],
5978                                void *aux OVS_UNUSED)
5979 {
5980     struct ds ds = DS_EMPTY_INITIALIZER;
5981     struct ofproto_dpif *ofproto;
5982
5983     ofproto = ofproto_dpif_lookup(argv[1]);
5984     if (!ofproto) {
5985         unixctl_command_reply_error(conn, "no such bridge");
5986         return;
5987     }
5988
5989     flush(&ofproto->up);
5990
5991     unixctl_command_reply(conn, ds_cstr(&ds));
5992     ds_destroy(&ds);
5993 }
5994
5995 static void
5996 ofproto_dpif_unixctl_init(void)
5997 {
5998     static bool registered;
5999     if (registered) {
6000         return;
6001     }
6002     registered = true;
6003
6004     unixctl_command_register(
6005         "ofproto/trace",
6006         "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6007         1, 3, ofproto_unixctl_trace, NULL);
6008     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6009                              ofproto_unixctl_fdb_flush, NULL);
6010     unixctl_command_register("fdb/show", "bridge", 1, 1,
6011                              ofproto_unixctl_fdb_show, NULL);
6012     unixctl_command_register("ofproto/clog", "", 0, 0,
6013                              ofproto_dpif_clog, NULL);
6014     unixctl_command_register("ofproto/unclog", "", 0, 0,
6015                              ofproto_dpif_unclog, NULL);
6016     unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6017                              ofproto_dpif_self_check, NULL);
6018     unixctl_command_register("dpif/dump-dps", "", 0, 0,
6019                              ofproto_unixctl_dpif_dump_dps, NULL);
6020     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6021                              NULL);
6022     unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6023                              ofproto_unixctl_dpif_dump_flows, NULL);
6024     unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6025                              ofproto_unixctl_dpif_del_flows, NULL);
6026     unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6027                              ofproto_unixctl_dpif_dump_megaflows, NULL);
6028     unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6029                              ofproto_unixctl_dpif_disable_megaflows, NULL);
6030     unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6031                              ofproto_unixctl_dpif_enable_megaflows, NULL);
6032 }
6033 \f
6034 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6035  *
6036  * This is deprecated.  It is only for compatibility with broken device drivers
6037  * in old versions of Linux that do not properly support VLANs when VLAN
6038  * devices are not used.  When broken device drivers are no longer in
6039  * widespread use, we will delete these interfaces. */
6040
6041 static int
6042 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6043 {
6044     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6045     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6046
6047     if (realdev_ofp_port == ofport->realdev_ofp_port
6048         && vid == ofport->vlandev_vid) {
6049         return 0;
6050     }
6051
6052     ofproto->backer->need_revalidate = REV_RECONFIGURE;
6053
6054     if (ofport->realdev_ofp_port) {
6055         vsp_remove(ofport);
6056     }
6057     if (realdev_ofp_port && ofport->bundle) {
6058         /* vlandevs are enslaved to their realdevs, so they are not allowed to
6059          * themselves be part of a bundle. */
6060         bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6061     }
6062
6063     ofport->realdev_ofp_port = realdev_ofp_port;
6064     ofport->vlandev_vid = vid;
6065
6066     if (realdev_ofp_port) {
6067         vsp_add(ofport, realdev_ofp_port, vid);
6068     }
6069
6070     return 0;
6071 }
6072
6073 static uint32_t
6074 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6075 {
6076     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6077 }
6078
6079 bool
6080 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6081     OVS_EXCLUDED(ofproto->vsp_mutex)
6082 {
6083     bool ret;
6084
6085     ovs_mutex_lock(&ofproto->vsp_mutex);
6086     ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6087     ovs_mutex_unlock(&ofproto->vsp_mutex);
6088     return ret;
6089 }
6090
6091 static ofp_port_t
6092 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6093                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6094     OVS_REQUIRES(ofproto->vsp_mutex)
6095 {
6096     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6097         int vid = vlan_tci_to_vid(vlan_tci);
6098         const struct vlan_splinter *vsp;
6099
6100         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6101                                  hash_realdev_vid(realdev_ofp_port, vid),
6102                                  &ofproto->realdev_vid_map) {
6103             if (vsp->realdev_ofp_port == realdev_ofp_port
6104                 && vsp->vid == vid) {
6105                 return vsp->vlandev_ofp_port;
6106             }
6107         }
6108     }
6109     return realdev_ofp_port;
6110 }
6111
6112 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6113  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6114  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
6115  * 'vlan_tci' 9, it would return the port number of eth0.9.
6116  *
6117  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6118  * function just returns its 'realdev_ofp_port' argument. */
6119 ofp_port_t
6120 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6121                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6122     OVS_EXCLUDED(ofproto->vsp_mutex)
6123 {
6124     ofp_port_t ret;
6125
6126     ovs_mutex_lock(&ofproto->vsp_mutex);
6127     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6128     ovs_mutex_unlock(&ofproto->vsp_mutex);
6129     return ret;
6130 }
6131
6132 static struct vlan_splinter *
6133 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6134 {
6135     struct vlan_splinter *vsp;
6136
6137     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6138                              hash_ofp_port(vlandev_ofp_port),
6139                              &ofproto->vlandev_map) {
6140         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6141             return vsp;
6142         }
6143     }
6144
6145     return NULL;
6146 }
6147
6148 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6149  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6150  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
6151  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6152  * eth0 and store 9 in '*vid'.
6153  *
6154  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6155  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
6156  * always does.*/
6157 static ofp_port_t
6158 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6159                        ofp_port_t vlandev_ofp_port, int *vid)
6160     OVS_REQUIRES(ofproto->vsp_mutex)
6161 {
6162     if (!hmap_is_empty(&ofproto->vlandev_map)) {
6163         const struct vlan_splinter *vsp;
6164
6165         vsp = vlandev_find(ofproto, vlandev_ofp_port);
6166         if (vsp) {
6167             if (vid) {
6168                 *vid = vsp->vid;
6169             }
6170             return vsp->realdev_ofp_port;
6171         }
6172     }
6173     return 0;
6174 }
6175
6176 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6177  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
6178  * 'flow->in_port' to the "real" device backing the VLAN device, sets
6179  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Otherwise (which is
6180  * always the case unless VLAN splinters are enabled), returns false without
6181  * making any changes. */
6182 bool
6183 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6184     OVS_EXCLUDED(ofproto->vsp_mutex)
6185 {
6186     ofp_port_t realdev;
6187     int vid;
6188
6189     ovs_mutex_lock(&ofproto->vsp_mutex);
6190     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6191     ovs_mutex_unlock(&ofproto->vsp_mutex);
6192     if (!realdev) {
6193         return false;
6194     }
6195
6196     /* Cause the flow to be processed as if it came in on the real device with
6197      * the VLAN device's VLAN ID. */
6198     flow->in_port.ofp_port = realdev;
6199     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6200     return true;
6201 }
6202
6203 static void
6204 vsp_remove(struct ofport_dpif *port)
6205 {
6206     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6207     struct vlan_splinter *vsp;
6208
6209     ovs_mutex_lock(&ofproto->vsp_mutex);
6210     vsp = vlandev_find(ofproto, port->up.ofp_port);
6211     if (vsp) {
6212         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6213         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6214         free(vsp);
6215
6216         port->realdev_ofp_port = 0;
6217     } else {
6218         VLOG_ERR("missing vlan device record");
6219     }
6220     ovs_mutex_unlock(&ofproto->vsp_mutex);
6221 }
6222
6223 static void
6224 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6225 {
6226     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6227
6228     ovs_mutex_lock(&ofproto->vsp_mutex);
6229     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6230         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6231             == realdev_ofp_port)) {
6232         struct vlan_splinter *vsp;
6233
6234         vsp = xmalloc(sizeof *vsp);
6235         vsp->realdev_ofp_port = realdev_ofp_port;
6236         vsp->vlandev_ofp_port = port->up.ofp_port;
6237         vsp->vid = vid;
6238
6239         port->realdev_ofp_port = realdev_ofp_port;
6240
6241         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6242                     hash_ofp_port(port->up.ofp_port));
6243         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6244                     hash_realdev_vid(realdev_ofp_port, vid));
6245     } else {
6246         VLOG_ERR("duplicate vlan device record");
6247     }
6248     ovs_mutex_unlock(&ofproto->vsp_mutex);
6249 }
6250
6251 static odp_port_t
6252 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6253 {
6254     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6255     return ofport ? ofport->odp_port : ODPP_NONE;
6256 }
6257
6258 struct ofport_dpif *
6259 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6260 {
6261     struct ofport_dpif *port;
6262
6263     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6264     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6265                              &backer->odp_to_ofport_map) {
6266         if (port->odp_port == odp_port) {
6267             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6268             return port;
6269         }
6270     }
6271
6272     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6273     return NULL;
6274 }
6275
6276 static ofp_port_t
6277 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6278 {
6279     struct ofport_dpif *port;
6280
6281     port = odp_port_to_ofport(ofproto->backer, odp_port);
6282     if (port && &ofproto->up == port->up.ofproto) {
6283         return port->up.ofp_port;
6284     } else {
6285         return OFPP_NONE;
6286     }
6287 }
6288
6289 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6290  * most heavily weighted element.  'base' designates the rate of decay: after
6291  * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6292  * (about .37). */
6293 static void
6294 exp_mavg(double *avg, int base, double new)
6295 {
6296     *avg = (*avg * (base - 1) + new) / base;
6297 }
6298
6299 static void
6300 update_moving_averages(struct dpif_backer *backer)
6301 {
6302     const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6303     long long int minutes = (time_msec() - backer->created) / min_ms;
6304
6305     if (minutes > 0) {
6306         backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6307             / minutes;
6308         backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6309             / minutes;
6310     } else {
6311         backer->lifetime.add_rate = 0.0;
6312         backer->lifetime.del_rate = 0.0;
6313     }
6314
6315     /* Update hourly averages on the minute boundaries. */
6316     if (time_msec() - backer->last_minute >= min_ms) {
6317         exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6318         exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6319
6320         /* Update daily averages on the hour boundaries. */
6321         if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6322             exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6323             exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6324         }
6325
6326         backer->total_subfacet_add_count += backer->subfacet_add_count;
6327         backer->total_subfacet_del_count += backer->subfacet_del_count;
6328         backer->subfacet_add_count = 0;
6329         backer->subfacet_del_count = 0;
6330         backer->last_minute += min_ms;
6331     }
6332 }
6333
6334 const struct ofproto_class ofproto_dpif_class = {
6335     init,
6336     enumerate_types,
6337     enumerate_names,
6338     del,
6339     port_open_type,
6340     type_run,
6341     type_run_fast,
6342     type_wait,
6343     alloc,
6344     construct,
6345     destruct,
6346     dealloc,
6347     run,
6348     run_fast,
6349     wait,
6350     get_memory_usage,
6351     flush,
6352     get_features,
6353     get_tables,
6354     port_alloc,
6355     port_construct,
6356     port_destruct,
6357     port_dealloc,
6358     port_modified,
6359     port_reconfigured,
6360     port_query_by_name,
6361     port_add,
6362     port_del,
6363     port_get_stats,
6364     port_dump_start,
6365     port_dump_next,
6366     port_dump_done,
6367     port_poll,
6368     port_poll_wait,
6369     port_is_lacp_current,
6370     NULL,                       /* rule_choose_table */
6371     rule_alloc,
6372     rule_construct,
6373     rule_insert,
6374     rule_delete,
6375     rule_destruct,
6376     rule_dealloc,
6377     rule_get_stats,
6378     rule_execute,
6379     rule_modify_actions,
6380     set_frag_handling,
6381     packet_out,
6382     set_netflow,
6383     get_netflow_ids,
6384     set_sflow,
6385     set_ipfix,
6386     set_cfm,
6387     get_cfm_status,
6388     set_bfd,
6389     get_bfd_status,
6390     set_stp,
6391     get_stp_status,
6392     set_stp_port,
6393     get_stp_port_status,
6394     set_queues,
6395     bundle_set,
6396     bundle_remove,
6397     mirror_set__,
6398     mirror_get_stats__,
6399     set_flood_vlans,
6400     is_mirror_output_bundle,
6401     forward_bpdu_changed,
6402     set_mac_table_config,
6403     set_realdev,
6404     NULL,                       /* meter_get_features */
6405     NULL,                       /* meter_set */
6406     NULL,                       /* meter_get */
6407     NULL,                       /* meter_del */
6408 };