ofproto/ofproto-dpif.c

   1 /*
   2  * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18
  19 #include "ofproto/ofproto-dpif.h"
  20 #include "ofproto/ofproto-provider.h"
  21
  22 #include <errno.h>
  23
  24 #include "bfd.h"
  25 #include "bond.h"
  26 #include "bundle.h"
  27 #include "byte-order.h"
  28 #include "connmgr.h"
  29 #include "coverage.h"
  30 #include "cfm.h"
  31 #include "dpif.h"
  32 #include "dynamic-string.h"
  33 #include "fail-open.h"
  34 #include "guarded-list.h"
  35 #include "hmapx.h"
  36 #include "lacp.h"
  37 #include "learn.h"
  38 #include "mac-learning.h"
  39 #include "meta-flow.h"
  40 #include "multipath.h"
  41 #include "netdev-vport.h"
  42 #include "netdev.h"
  43 #include "netlink.h"
  44 #include "nx-match.h"
  45 #include "odp-util.h"
  46 #include "odp-execute.h"
  47 #include "ofp-util.h"
  48 #include "ofpbuf.h"
  49 #include "ofp-actions.h"
  50 #include "ofp-parse.h"
  51 #include "ofp-print.h"
  52 #include "ofproto-dpif-governor.h"
  53 #include "ofproto-dpif-ipfix.h"
  54 #include "ofproto-dpif-mirror.h"
  55 #include "ofproto-dpif-monitor.h"
  56 #include "ofproto-dpif-sflow.h"
  57 #include "ofproto-dpif-upcall.h"
  58 #include "ofproto-dpif-xlate.h"
  59 #include "poll-loop.h"
  60 #include "simap.h"
  61 #include "smap.h"
  62 #include "timer.h"
  63 #include "tunnel.h"
  64 #include "unaligned.h"
  65 #include "unixctl.h"
  66 #include "vlan-bitmap.h"
  67 #include "vlog.h"
  68
  69 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  70
  71 COVERAGE_DEFINE(ofproto_dpif_expired);
  72 COVERAGE_DEFINE(facet_revalidate);
  73 COVERAGE_DEFINE(facet_unexpected);
  74 COVERAGE_DEFINE(facet_create);
  75 COVERAGE_DEFINE(facet_remove);
  76 COVERAGE_DEFINE(subfacet_create);
  77 COVERAGE_DEFINE(subfacet_destroy);
  78 COVERAGE_DEFINE(subfacet_install_fail);
  79 COVERAGE_DEFINE(packet_in_overflow);
  80
  81 /* Number of implemented OpenFlow tables. */
  82 enum { N_TABLES = 255 };
  83 enum { TBL_INTERNAL = N_TABLES - 1 };    /* Used for internal hidden rules. */
  84 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
  85
  86 struct flow_miss;
  87 struct facet;
  88
  89 struct rule_dpif {
  90     struct rule up;
  91
  92     /* These statistics:
  93      *
  94      *   - Do include packets and bytes from facets that have been deleted or
  95      *     whose own statistics have been folded into the rule.
  96      *
  97      *   - Do include packets and bytes sent "by hand" that were accounted to
  98      *     the rule without any facet being involved (this is a rare corner
  99      *     case in rule_execute()).
 100      *
 101      *   - Do not include packet or bytes that can be obtained from any facet's
 102      *     packet_count or byte_count member or that can be obtained from the
 103      *     datapath by, e.g., dpif_flow_get() for any subfacet.
 104      */
 105     struct ovs_mutex stats_mutex;
 106     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 107     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 108 };
 109
 110 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
 111 static struct rule_dpif *rule_dpif_cast(const struct rule *);
 112
 113 struct group_dpif {
 114     struct ofgroup up;
 115
 116     /* These statistics:
 117      *
 118      *   - Do include packets and bytes from facets that have been deleted or
 119      *     whose own statistics have been folded into the rule.
 120      *
 121      *   - Do include packets and bytes sent "by hand" that were accounted to
 122      *     the rule without any facet being involved (this is a rare corner
 123      *     case in rule_execute()).
 124      *
 125      *   - Do not include packet or bytes that can be obtained from any facet's
 126      *     packet_count or byte_count member or that can be obtained from the
 127      *     datapath by, e.g., dpif_flow_get() for any subfacet.
 128      */
 129     struct ovs_mutex stats_mutex;
 130     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 131     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 132     struct bucket_counter *bucket_stats OVS_GUARDED;  /* Bucket statistics. */
 133 };
 134
 135 struct ofbundle {
 136     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
 137     struct ofproto_dpif *ofproto; /* Owning ofproto. */
 138     void *aux;                  /* Key supplied by ofproto's client. */
 139     char *name;                 /* Identifier for log messages. */
 140
 141     /* Configuration. */
 142     struct list ports;          /* Contains "struct ofport"s. */
 143     enum port_vlan_mode vlan_mode; /* VLAN mode */
 144     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
 145     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
 146                                  * NULL if all VLANs are trunked. */
 147     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 148     struct bond *bond;          /* Nonnull iff more than one port. */
 149     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 150
 151     /* Status. */
 152     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 153 };
 154
 155 static void bundle_remove(struct ofport *);
 156 static void bundle_update(struct ofbundle *);
 157 static void bundle_destroy(struct ofbundle *);
 158 static void bundle_del_port(struct ofport_dpif *);
 159 static void bundle_run(struct ofbundle *);
 160 static void bundle_wait(struct ofbundle *);
 161
 162 static void stp_run(struct ofproto_dpif *ofproto);
 163 static void stp_wait(struct ofproto_dpif *ofproto);
 164 static int set_stp_port(struct ofport *,
 165                         const struct ofproto_port_stp_settings *);
 166
 167 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
 168                               enum slow_path_reason,
 169                               uint64_t *stub, size_t stub_size,
 170                               const struct nlattr **actionsp,
 171                               size_t *actions_lenp);
 172
 173 /* A subfacet (see "struct subfacet" below) has three possible installation
 174  * states:
 175  *
 176  *   - SF_NOT_INSTALLED: Not installed in the datapath.  This will only be the
 177  *     case just after the subfacet is created, just before the subfacet is
 178  *     destroyed, or if the datapath returns an error when we try to install a
 179  *     subfacet.
 180  *
 181  *   - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
 182  *
 183  *   - SF_SLOW_PATH: An action that sends every packet for the subfacet through
 184  *     ofproto_dpif is installed in the datapath.
 185  */
 186 enum subfacet_path {
 187     SF_NOT_INSTALLED,           /* No datapath flow for this subfacet. */
 188     SF_FAST_PATH,               /* Full actions are installed. */
 189     SF_SLOW_PATH,               /* Send-to-userspace action is installed. */
 190 };
 191
 192 /* A dpif flow and actions associated with a facet.
 193  *
 194  * See also the large comment on struct facet. */
 195 struct subfacet {
 196     /* Owners. */
 197     struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
 198     struct list list_node;      /* In struct facet's 'facets' list. */
 199     struct facet *facet;        /* Owning facet. */
 200     struct dpif_backer *backer; /* Owning backer. */
 201
 202     struct nlattr *key;
 203     int key_len;
 204
 205     long long int used;         /* Time last used; time created if not used. */
 206     long long int created;      /* Time created. */
 207
 208     uint64_t dp_packet_count;   /* Last known packet count in the datapath. */
 209     uint64_t dp_byte_count;     /* Last known byte count in the datapath. */
 210
 211     enum subfacet_path path;    /* Installed in datapath? */
 212 };
 213
 214 #define SUBFACET_DESTROY_MAX_BATCH 50
 215
 216 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *);
 217 static struct subfacet *subfacet_find(struct dpif_backer *,
 218                                       const struct nlattr *key, size_t key_len,
 219                                       uint32_t key_hash);
 220 static void subfacet_destroy(struct subfacet *);
 221 static void subfacet_destroy__(struct subfacet *);
 222 static void subfacet_destroy_batch(struct dpif_backer *,
 223                                    struct subfacet **, int n);
 224 static void subfacet_reset_dp_stats(struct subfacet *,
 225                                     struct dpif_flow_stats *);
 226 static void subfacet_update_stats(struct subfacet *,
 227                                   const struct dpif_flow_stats *);
 228 static int subfacet_install(struct subfacet *,
 229                             const struct ofpbuf *odp_actions,
 230                             struct dpif_flow_stats *);
 231 static void subfacet_uninstall(struct subfacet *);
 232
 233 /* A unique, non-overlapping instantiation of an OpenFlow flow.
 234  *
 235  * A facet associates a "struct flow", which represents the Open vSwitch
 236  * userspace idea of an exact-match flow, with one or more subfacets.
 237  * While the facet is created based on an exact-match flow, it is stored
 238  * within the ofproto based on the wildcards that could be expressed
 239  * based on the flow table and other configuration.  (See the 'wc'
 240  * description in "struct xlate_out" for more details.)
 241  *
 242  * Each subfacet tracks the datapath's idea of the flow equivalent to
 243  * the facet.  When the kernel module (or other dpif implementation) and
 244  * Open vSwitch userspace agree on the definition of a flow key, there
 245  * is exactly one subfacet per facet.  If the dpif implementation
 246  * supports more-specific flow matching than userspace, however, a facet
 247  * can have more than one subfacet.  Examples include the dpif
 248  * implementation not supporting the same wildcards as userspace or some
 249  * distinction in flow that userspace simply doesn't understand.
 250  *
 251  * Flow expiration works in terms of subfacets, so a facet must have at
 252  * least one subfacet or it will never expire, leaking memory. */
 253 struct facet {
 254     /* Owner. */
 255     struct ofproto_dpif *ofproto;
 256
 257     /* Owned data. */
 258     struct list subfacets;
 259     long long int used;         /* Time last used; time created if not used. */
 260
 261     /* Key. */
 262     struct flow flow;           /* Flow of the creating subfacet. */
 263     struct cls_rule cr;         /* In 'ofproto_dpif's facets classifier. */
 264
 265     /* These statistics:
 266      *
 267      *   - Do include packets and bytes sent "by hand", e.g. with
 268      *     dpif_execute().
 269      *
 270      *   - Do include packets and bytes that were obtained from the datapath
 271      *     when a subfacet's statistics were reset (e.g. dpif_flow_put() with
 272      *     DPIF_FP_ZERO_STATS).
 273      *
 274      *   - Do not include packets or bytes that can be obtained from the
 275      *     datapath for any existing subfacet.
 276      */
 277     uint64_t packet_count;       /* Number of packets received. */
 278     uint64_t byte_count;         /* Number of bytes received. */
 279
 280     /* Resubmit statistics. */
 281     uint64_t prev_packet_count;  /* Number of packets from last stats push. */
 282     uint64_t prev_byte_count;    /* Number of bytes from last stats push. */
 283     long long int prev_used;     /* Used time from last stats push. */
 284
 285     /* Accounting. */
 286     uint64_t accounted_bytes;    /* Bytes processed by facet_account(). */
 287     struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
 288     uint16_t tcp_flags;          /* TCP flags seen for this 'rule'. */
 289
 290     struct xlate_out xout;
 291
 292     /* Storage for a single subfacet, to reduce malloc() time and space
 293      * overhead.  (A facet always has at least one subfacet and in the common
 294      * case has exactly one subfacet.  However, 'one_subfacet' may not
 295      * always be valid, since it could have been removed after newer
 296      * subfacets were pushed onto the 'subfacets' list.) */
 297     struct subfacet one_subfacet;
 298
 299     long long int learn_rl;      /* Rate limiter for facet_learn(). */
 300 };
 301
 302 static struct facet *facet_create(const struct flow_miss *);
 303 static void facet_remove(struct facet *);
 304 static void facet_free(struct facet *);
 305
 306 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
 307 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
 308                                         const struct flow *);
 309 static bool facet_revalidate(struct facet *);
 310 static bool facet_check_consistency(struct facet *);
 311
 312 static void facet_flush_stats(struct facet *);
 313
 314 static void facet_reset_counters(struct facet *);
 315 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
 316                             struct dpif_flow_stats *, bool may_learn);
 317 static void facet_push_stats(struct facet *, bool may_learn);
 318 static void facet_learn(struct facet *);
 319 static void facet_account(struct facet *);
 320 static void push_all_stats(void);
 321
 322 static bool facet_is_controller_flow(struct facet *);
 323
 324 struct ofport_dpif {
 325     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 326     struct ofport up;
 327
 328     odp_port_t odp_port;
 329     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 330     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
 331     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 332     struct bfd *bfd;            /* BFD, if any. */
 333     bool may_enable;            /* May be enabled in bonds. */
 334     bool is_tunnel;             /* This port is a tunnel. */
 335     long long int carrier_seq;  /* Carrier status changes. */
 336     struct ofport_dpif *peer;   /* Peer if patch port. */
 337
 338     /* Spanning tree. */
 339     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 340     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 341     long long int stp_state_entered;
 342
 343     /* Queue to DSCP mapping. */
 344     struct ofproto_port_queue *qdscp;
 345     size_t n_qdscp;
 346
 347     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 348      *
 349      * This is deprecated.  It is only for compatibility with broken device
 350      * drivers in old versions of Linux that do not properly support VLANs when
 351      * VLAN devices are not used.  When broken device drivers are no longer in
 352      * widespread use, we will delete these interfaces. */
 353     ofp_port_t realdev_ofp_port;
 354     int vlandev_vid;
 355 };
 356
 357 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 358  *
 359  * This is deprecated.  It is only for compatibility with broken device drivers
 360  * in old versions of Linux that do not properly support VLANs when VLAN
 361  * devices are not used.  When broken device drivers are no longer in
 362  * widespread use, we will delete these interfaces. */
 363 struct vlan_splinter {
 364     struct hmap_node realdev_vid_node;
 365     struct hmap_node vlandev_node;
 366     ofp_port_t realdev_ofp_port;
 367     ofp_port_t vlandev_ofp_port;
 368     int vid;
 369 };
 370
 371 static void vsp_remove(struct ofport_dpif *);
 372 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 373
 374 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 375                                        ofp_port_t);
 376
 377 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 378                                        odp_port_t);
 379
 380 static struct ofport_dpif *
 381 ofport_dpif_cast(const struct ofport *ofport)
 382 {
 383     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 384 }
 385
 386 static void port_run(struct ofport_dpif *);
 387 static int set_bfd(struct ofport *, const struct smap *);
 388 static int set_cfm(struct ofport *, const struct cfm_settings *);
 389 static void ofport_update_peer(struct ofport_dpif *);
 390 static void run_fast_rl(void);
 391 static int run_fast(struct ofproto *);
 392
 393 struct dpif_completion {
 394     struct list list_node;
 395     struct ofoperation *op;
 396 };
 397
 398 /* Reasons that we might need to revalidate every facet, and corresponding
 399  * coverage counters.
 400  *
 401  * A value of 0 means that there is no need to revalidate.
 402  *
 403  * It would be nice to have some cleaner way to integrate with coverage
 404  * counters, but with only a few reasons I guess this is good enough for
 405  * now. */
 406 enum revalidate_reason {
 407     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 408     REV_STP,                   /* Spanning tree protocol port status change. */
 409     REV_BOND,                  /* Bonding changed. */
 410     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 411     REV_FLOW_TABLE,            /* Flow table changed. */
 412     REV_MAC_LEARNING,          /* Mac learning changed. */
 413     REV_INCONSISTENCY          /* Facet self-check failed. */
 414 };
 415 COVERAGE_DEFINE(rev_reconfigure);
 416 COVERAGE_DEFINE(rev_stp);
 417 COVERAGE_DEFINE(rev_bond);
 418 COVERAGE_DEFINE(rev_port_toggled);
 419 COVERAGE_DEFINE(rev_flow_table);
 420 COVERAGE_DEFINE(rev_mac_learning);
 421 COVERAGE_DEFINE(rev_inconsistency);
 422
 423 struct avg_subfacet_rates {
 424     double add_rate;   /* Moving average of new flows created per minute. */
 425     double del_rate;   /* Moving average of flows deleted per minute. */
 426 };
 427
 428 /* All datapaths of a given type share a single dpif backer instance. */
 429 struct dpif_backer {
 430     char *type;
 431     int refcount;
 432     struct dpif *dpif;
 433     struct udpif *udpif;
 434     struct timer next_expiration;
 435
 436     struct ovs_rwlock odp_to_ofport_lock;
 437     struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
 438
 439     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 440
 441     /* Facet revalidation flags applying to facets which use this backer. */
 442     enum revalidate_reason need_revalidate; /* Revalidate every facet. */
 443
 444     struct hmap drop_keys; /* Set of dropped odp keys. */
 445     bool recv_set_enable; /* Enables or disables receiving packets. */
 446
 447     struct hmap subfacets;
 448     struct governor *governor;
 449
 450     /* Subfacet statistics.
 451      *
 452      * These keep track of the total number of subfacets added and deleted and
 453      * flow life span.  They are useful for computing the flow rates stats
 454      * exposed via "ovs-appctl dpif/show".  The goal is to learn about
 455      * traffic patterns in ways that we can use later to improve Open vSwitch
 456      * performance in new situations.  */
 457     long long int created;           /* Time when it is created. */
 458     unsigned max_n_subfacet;         /* Maximum number of flows */
 459     unsigned avg_n_subfacet;         /* Average number of flows. */
 460     long long int avg_subfacet_life; /* Average life span of subfacets. */
 461
 462     /* Number of upcall handling threads. */
 463     unsigned int n_handler_threads;
 464 };
 465
 466 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 467 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 468
 469 static void drop_key_clear(struct dpif_backer *);
 470
 471 struct ofproto_dpif {
 472     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 473     struct ofproto up;
 474     struct dpif_backer *backer;
 475
 476     /* Special OpenFlow rules. */
 477     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 478     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 479     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 480
 481     /* Bridging. */
 482     struct netflow *netflow;
 483     struct dpif_sflow *sflow;
 484     struct dpif_ipfix *ipfix;
 485     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 486     struct mac_learning *ml;
 487     bool has_bonded_bundles;
 488     struct mbridge *mbridge;
 489
 490     /* Facets. */
 491     struct classifier facets;     /* Contains 'struct facet's. */
 492     long long int consistency_rl;
 493
 494     struct ovs_mutex stats_mutex;
 495     struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
 496                                             * consumed in userspace. */
 497
 498     /* Spanning tree. */
 499     struct stp *stp;
 500     long long int stp_last_tick;
 501
 502     /* VLAN splinters. */
 503     struct ovs_mutex vsp_mutex;
 504     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 505     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 506
 507     /* Ports. */
 508     struct sset ports;             /* Set of standard port names. */
 509     struct sset ghost_ports;       /* Ports with no datapath port. */
 510     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 511     int port_poll_errno;           /* Last errno for port_poll() reply. */
 512
 513     /* Per ofproto's dpif stats. */
 514     uint64_t n_hit;
 515     uint64_t n_missed;
 516
 517     /* Work queues. */
 518     struct guarded_list pins;      /* Contains "struct ofputil_packet_in"s. */
 519 };
 520
 521 /* By default, flows in the datapath are wildcarded (megaflows).  They
 522  * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
 523 static bool enable_megaflows = true;
 524
 525 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 526 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 527
 528 static void ofproto_dpif_unixctl_init(void);
 529
 530 static inline struct ofproto_dpif *
 531 ofproto_dpif_cast(const struct ofproto *ofproto)
 532 {
 533     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 534     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 535 }
 536
 537 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
 538                                         ofp_port_t ofp_port);
 539 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
 540                           const struct ofpbuf *packet,
 541                           const struct ofpact[], size_t ofpacts_len,
 542                           struct ds *);
 543
 544 /* Upcalls. */
 545 static void handle_upcalls(struct dpif_backer *);
 546
 547 /* Flow expiration. */
 548 static int expire(struct dpif_backer *);
 549
 550 /* NetFlow. */
 551 static void send_netflow_active_timeouts(struct ofproto_dpif *);
 552
 553 /* Global variables. */
 554 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 555
 556 /* Initial mappings of port to bridge mappings. */
 557 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 558
 559 /* Executes 'fm'.  The caller retains ownership of 'fm' and everything in
 560  * it. */
 561 void
 562 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 563                       struct ofputil_flow_mod *fm)
 564 {
 565     ofproto_flow_mod(&ofproto->up, fm);
 566 }
 567
 568 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 569  * Takes ownership of 'pin' and pin->packet. */
 570 void
 571 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 572                             struct ofproto_packet_in *pin)
 573 {
 574     if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
 575         COVERAGE_INC(packet_in_overflow);
 576         free(CONST_CAST(void *, pin->up.packet));
 577         free(pin);
 578     }
 579 }
 580 \f
 581 /* Factory functions. */
 582
 583 static void
 584 init(const struct shash *iface_hints)
 585 {
 586     struct shash_node *node;
 587
 588     /* Make a local copy, since we don't own 'iface_hints' elements. */
 589     SHASH_FOR_EACH(node, iface_hints) {
 590         const struct iface_hint *orig_hint = node->data;
 591         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 592
 593         new_hint->br_name = xstrdup(orig_hint->br_name);
 594         new_hint->br_type = xstrdup(orig_hint->br_type);
 595         new_hint->ofp_port = orig_hint->ofp_port;
 596
 597         shash_add(&init_ofp_ports, node->name, new_hint);
 598     }
 599 }
 600
 601 static void
 602 enumerate_types(struct sset *types)
 603 {
 604     dp_enumerate_types(types);
 605 }
 606
 607 static int
 608 enumerate_names(const char *type, struct sset *names)
 609 {
 610     struct ofproto_dpif *ofproto;
 611
 612     sset_clear(names);
 613     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 614         if (strcmp(type, ofproto->up.type)) {
 615             continue;
 616         }
 617         sset_add(names, ofproto->up.name);
 618     }
 619
 620     return 0;
 621 }
 622
 623 static int
 624 del(const char *type, const char *name)
 625 {
 626     struct dpif *dpif;
 627     int error;
 628
 629     error = dpif_open(name, type, &dpif);
 630     if (!error) {
 631         error = dpif_delete(dpif);
 632         dpif_close(dpif);
 633     }
 634     return error;
 635 }
 636 \f
 637 static const char *
 638 port_open_type(const char *datapath_type, const char *port_type)
 639 {
 640     return dpif_port_open_type(datapath_type, port_type);
 641 }
 642
 643 /* Type functions. */
 644
 645 static void process_dpif_port_changes(struct dpif_backer *);
 646 static void process_dpif_all_ports_changed(struct dpif_backer *);
 647 static void process_dpif_port_change(struct dpif_backer *,
 648                                      const char *devname);
 649 static void process_dpif_port_error(struct dpif_backer *, int error);
 650
 651 static struct ofproto_dpif *
 652 lookup_ofproto_dpif_by_port_name(const char *name)
 653 {
 654     struct ofproto_dpif *ofproto;
 655
 656     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 657         if (sset_contains(&ofproto->ports, name)) {
 658             return ofproto;
 659         }
 660     }
 661
 662     return NULL;
 663 }
 664
 665 static int
 666 type_run(const char *type)
 667 {
 668     static long long int push_timer = LLONG_MIN;
 669     struct dpif_backer *backer;
 670
 671     backer = shash_find_data(&all_dpif_backers, type);
 672     if (!backer) {
 673         /* This is not necessarily a problem, since backers are only
 674          * created on demand. */
 675         return 0;
 676     }
 677
 678     dpif_run(backer->dpif);
 679
 680     /* The most natural place to push facet statistics is when they're pulled
 681      * from the datapath.  However, when there are many flows in the datapath,
 682      * this expensive operation can occur so frequently, that it reduces our
 683      * ability to quickly set up flows.  To reduce the cost, we push statistics
 684      * here instead. */
 685     if (time_msec() > push_timer) {
 686         push_timer = time_msec() + 2000;
 687         push_all_stats();
 688     }
 689
 690     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 691      * and the configuration has now changed to "false", enable receiving
 692      * packets from the datapath. */
 693     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 694         int error;
 695
 696         backer->recv_set_enable = true;
 697
 698         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 699         if (error) {
 700             udpif_recv_set(backer->udpif, 0, false);
 701             VLOG_ERR("Failed to enable receiving packets in dpif.");
 702             return error;
 703         }
 704         udpif_recv_set(backer->udpif, n_handler_threads,
 705                        backer->recv_set_enable);
 706         dpif_flow_flush(backer->dpif);
 707         backer->need_revalidate = REV_RECONFIGURE;
 708     }
 709
 710     /* If the n_handler_threads is reconfigured, call udpif_recv_set()
 711      * to reset the handler threads. */
 712     if (backer->n_handler_threads != n_handler_threads) {
 713         udpif_recv_set(backer->udpif, n_handler_threads,
 714                        backer->recv_set_enable);
 715         backer->n_handler_threads = n_handler_threads;
 716     }
 717
 718     if (backer->need_revalidate) {
 719         struct ofproto_dpif *ofproto;
 720         struct simap_node *node;
 721         struct simap tmp_backers;
 722
 723         /* Handle tunnel garbage collection. */
 724         simap_init(&tmp_backers);
 725         simap_swap(&backer->tnl_backers, &tmp_backers);
 726
 727         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 728             struct ofport_dpif *iter;
 729
 730             if (backer != ofproto->backer) {
 731                 continue;
 732             }
 733
 734             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 735                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 736                 const char *dp_port;
 737
 738                 if (!iter->is_tunnel) {
 739                     continue;
 740                 }
 741
 742                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 743                                                      namebuf, sizeof namebuf);
 744                 node = simap_find(&tmp_backers, dp_port);
 745                 if (node) {
 746                     simap_put(&backer->tnl_backers, dp_port, node->data);
 747                     simap_delete(&tmp_backers, node);
 748                     node = simap_find(&backer->tnl_backers, dp_port);
 749                 } else {
 750                     node = simap_find(&backer->tnl_backers, dp_port);
 751                     if (!node) {
 752                         odp_port_t odp_port = ODPP_NONE;
 753
 754                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 755                                            &odp_port)) {
 756                             simap_put(&backer->tnl_backers, dp_port,
 757                                       odp_to_u32(odp_port));
 758                             node = simap_find(&backer->tnl_backers, dp_port);
 759                         }
 760                     }
 761                 }
 762
 763                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 764                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 765                                          iter->odp_port)) {
 766                     backer->need_revalidate = REV_RECONFIGURE;
 767                 }
 768             }
 769         }
 770
 771         SIMAP_FOR_EACH (node, &tmp_backers) {
 772             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 773         }
 774         simap_destroy(&tmp_backers);
 775
 776         switch (backer->need_revalidate) {
 777         case REV_RECONFIGURE:   COVERAGE_INC(rev_reconfigure);   break;
 778         case REV_STP:           COVERAGE_INC(rev_stp);           break;
 779         case REV_BOND:          COVERAGE_INC(rev_bond);          break;
 780         case REV_PORT_TOGGLED:  COVERAGE_INC(rev_port_toggled);  break;
 781         case REV_FLOW_TABLE:    COVERAGE_INC(rev_flow_table);    break;
 782         case REV_MAC_LEARNING:  COVERAGE_INC(rev_mac_learning);  break;
 783         case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
 784         }
 785         backer->need_revalidate = 0;
 786
 787         /* Clear the drop_keys in case we should now be accepting some
 788          * formerly dropped flows. */
 789         drop_key_clear(backer);
 790
 791         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 792             struct facet *facet, *next;
 793             struct ofport_dpif *ofport;
 794             struct cls_cursor cursor;
 795             struct ofbundle *bundle;
 796
 797             if (ofproto->backer != backer) {
 798                 continue;
 799             }
 800
 801             ovs_rwlock_wrlock(&xlate_rwlock);
 802             xlate_ofproto_set(ofproto, ofproto->up.name,
 803                               ofproto->backer->dpif, ofproto->miss_rule,
 804                               ofproto->no_packet_in_rule, ofproto->ml,
 805                               ofproto->stp, ofproto->mbridge,
 806                               ofproto->sflow, ofproto->ipfix,
 807                               ofproto->up.frag_handling,
 808                               ofproto->up.forward_bpdu,
 809                               connmgr_has_in_band(ofproto->up.connmgr),
 810                               ofproto->netflow != NULL);
 811
 812             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 813                 xlate_bundle_set(ofproto, bundle, bundle->name,
 814                                  bundle->vlan_mode, bundle->vlan,
 815                                  bundle->trunks, bundle->use_priority_tags,
 816                                  bundle->bond, bundle->lacp,
 817                                  bundle->floodable);
 818             }
 819
 820             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 821                 int stp_port = ofport->stp_port
 822                     ? stp_port_no(ofport->stp_port)
 823                     : -1;
 824                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 825                                  ofport->up.ofp_port, ofport->odp_port,
 826                                  ofport->up.netdev, ofport->cfm,
 827                                  ofport->bfd, ofport->peer, stp_port,
 828                                  ofport->qdscp, ofport->n_qdscp,
 829                                  ofport->up.pp.config, ofport->up.pp.state,
 830                                  ofport->is_tunnel, ofport->may_enable);
 831             }
 832             ovs_rwlock_unlock(&xlate_rwlock);
 833
 834             /* Only ofproto-dpif cares about the facet classifier so we just
 835              * lock cls_cursor_init() to appease the thread safety analysis. */
 836             ovs_rwlock_rdlock(&ofproto->facets.rwlock);
 837             cls_cursor_init(&cursor, &ofproto->facets, NULL);
 838             ovs_rwlock_unlock(&ofproto->facets.rwlock);
 839             CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
 840                 facet_revalidate(facet);
 841                 run_fast_rl();
 842             }
 843         }
 844
 845         udpif_revalidate(backer->udpif);
 846     }
 847
 848     if (!backer->recv_set_enable) {
 849         /* Wake up before a max of 1000ms. */
 850         timer_set_duration(&backer->next_expiration, 1000);
 851     } else if (timer_expired(&backer->next_expiration)) {
 852         int delay = expire(backer);
 853         timer_set_duration(&backer->next_expiration, delay);
 854     }
 855
 856     process_dpif_port_changes(backer);
 857
 858     if (backer->governor) {
 859         size_t n_subfacets;
 860
 861         governor_run(backer->governor);
 862
 863         /* If the governor has shrunk to its minimum size and the number of
 864          * subfacets has dwindled, then drop the governor entirely.
 865          *
 866          * For hysteresis, the number of subfacets to drop the governor is
 867          * smaller than the number needed to trigger its creation. */
 868         n_subfacets = hmap_count(&backer->subfacets);
 869         if (n_subfacets * 4 < flow_eviction_threshold
 870             && governor_is_idle(backer->governor)) {
 871             governor_destroy(backer->governor);
 872             backer->governor = NULL;
 873         }
 874     }
 875
 876     return 0;
 877 }
 878
 879 /* Check for and handle port changes in 'backer''s dpif. */
 880 static void
 881 process_dpif_port_changes(struct dpif_backer *backer)
 882 {
 883     for (;;) {
 884         char *devname;
 885         int error;
 886
 887         error = dpif_port_poll(backer->dpif, &devname);
 888         switch (error) {
 889         case EAGAIN:
 890             return;
 891
 892         case ENOBUFS:
 893             process_dpif_all_ports_changed(backer);
 894             break;
 895
 896         case 0:
 897             process_dpif_port_change(backer, devname);
 898             free(devname);
 899             break;
 900
 901         default:
 902             process_dpif_port_error(backer, error);
 903             break;
 904         }
 905     }
 906 }
 907
 908 static void
 909 process_dpif_all_ports_changed(struct dpif_backer *backer)
 910 {
 911     struct ofproto_dpif *ofproto;
 912     struct dpif_port dpif_port;
 913     struct dpif_port_dump dump;
 914     struct sset devnames;
 915     const char *devname;
 916
 917     sset_init(&devnames);
 918     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 919         if (ofproto->backer == backer) {
 920             struct ofport *ofport;
 921
 922             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 923                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 924             }
 925         }
 926     }
 927     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 928         sset_add(&devnames, dpif_port.name);
 929     }
 930
 931     SSET_FOR_EACH (devname, &devnames) {
 932         process_dpif_port_change(backer, devname);
 933     }
 934     sset_destroy(&devnames);
 935 }
 936
 937 static void
 938 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 939 {
 940     struct ofproto_dpif *ofproto;
 941     struct dpif_port port;
 942
 943     /* Don't report on the datapath's device. */
 944     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 945         return;
 946     }
 947
 948     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 949                    &all_ofproto_dpifs) {
 950         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 951             return;
 952         }
 953     }
 954
 955     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 956     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 957         /* The port was removed.  If we know the datapath,
 958          * report it through poll_set().  If we don't, it may be
 959          * notifying us of a removal we initiated, so ignore it.
 960          * If there's a pending ENOBUFS, let it stand, since
 961          * everything will be reevaluated. */
 962         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 963             sset_add(&ofproto->port_poll_set, devname);
 964             ofproto->port_poll_errno = 0;
 965         }
 966     } else if (!ofproto) {
 967         /* The port was added, but we don't know with which
 968          * ofproto we should associate it.  Delete it. */
 969         dpif_port_del(backer->dpif, port.port_no);
 970     } else {
 971         struct ofport_dpif *ofport;
 972
 973         ofport = ofport_dpif_cast(shash_find_data(
 974                                       &ofproto->up.port_by_name, devname));
 975         if (ofport
 976             && ofport->odp_port != port.port_no
 977             && !odp_port_to_ofport(backer, port.port_no))
 978         {
 979             /* 'ofport''s datapath port number has changed from
 980              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 981              * structures to match. */
 982             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 983             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 984             ofport->odp_port = port.port_no;
 985             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 986                         hash_odp_port(port.port_no));
 987             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 988             backer->need_revalidate = REV_RECONFIGURE;
 989         }
 990     }
 991     dpif_port_destroy(&port);
 992 }
 993
 994 /* Propagate 'error' to all ofprotos based on 'backer'. */
 995 static void
 996 process_dpif_port_error(struct dpif_backer *backer, int error)
 997 {
 998     struct ofproto_dpif *ofproto;
 999
1000     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1001         if (ofproto->backer == backer) {
1002             sset_clear(&ofproto->port_poll_set);
1003             ofproto->port_poll_errno = error;
1004         }
1005     }
1006 }
1007
1008 static int
1009 dpif_backer_run_fast(struct dpif_backer *backer)
1010 {
1011     handle_upcalls(backer);
1012
1013     return 0;
1014 }
1015
1016 static int
1017 type_run_fast(const char *type)
1018 {
1019     struct dpif_backer *backer;
1020
1021     backer = shash_find_data(&all_dpif_backers, type);
1022     if (!backer) {
1023         /* This is not necessarily a problem, since backers are only
1024          * created on demand. */
1025         return 0;
1026     }
1027
1028     return dpif_backer_run_fast(backer);
1029 }
1030
1031 static void
1032 run_fast_rl(void)
1033 {
1034     static long long int port_rl = LLONG_MIN;
1035
1036     if (time_msec() >= port_rl) {
1037         struct ofproto_dpif *ofproto;
1038
1039         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1040             run_fast(&ofproto->up);
1041         }
1042         port_rl = time_msec() + 200;
1043     }
1044 }
1045
1046 static void
1047 type_wait(const char *type)
1048 {
1049     struct dpif_backer *backer;
1050
1051     backer = shash_find_data(&all_dpif_backers, type);
1052     if (!backer) {
1053         /* This is not necessarily a problem, since backers are only
1054          * created on demand. */
1055         return;
1056     }
1057
1058     if (backer->governor) {
1059         governor_wait(backer->governor);
1060     }
1061
1062     timer_wait(&backer->next_expiration);
1063     dpif_wait(backer->dpif);
1064     udpif_wait(backer->udpif);
1065 }
1066 \f
1067 /* Basic life-cycle. */
1068
1069 static int add_internal_flows(struct ofproto_dpif *);
1070
1071 static struct ofproto *
1072 alloc(void)
1073 {
1074     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1075     return &ofproto->up;
1076 }
1077
1078 static void
1079 dealloc(struct ofproto *ofproto_)
1080 {
1081     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1082     free(ofproto);
1083 }
1084
1085 static void
1086 close_dpif_backer(struct dpif_backer *backer)
1087 {
1088     ovs_assert(backer->refcount > 0);
1089
1090     if (--backer->refcount) {
1091         return;
1092     }
1093
1094     drop_key_clear(backer);
1095     hmap_destroy(&backer->drop_keys);
1096
1097     udpif_destroy(backer->udpif);
1098
1099     simap_destroy(&backer->tnl_backers);
1100     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1101     hmap_destroy(&backer->odp_to_ofport_map);
1102     shash_find_and_delete(&all_dpif_backers, backer->type);
1103     free(backer->type);
1104     dpif_close(backer->dpif);
1105
1106     ovs_assert(hmap_is_empty(&backer->subfacets));
1107     hmap_destroy(&backer->subfacets);
1108     governor_destroy(backer->governor);
1109
1110     free(backer);
1111 }
1112
1113 /* Datapath port slated for removal from datapath. */
1114 struct odp_garbage {
1115     struct list list_node;
1116     odp_port_t odp_port;
1117 };
1118
1119 static int
1120 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1121 {
1122     struct dpif_backer *backer;
1123     struct dpif_port_dump port_dump;
1124     struct dpif_port port;
1125     struct shash_node *node;
1126     struct list garbage_list;
1127     struct odp_garbage *garbage, *next;
1128     struct sset names;
1129     char *backer_name;
1130     const char *name;
1131     int error;
1132
1133     backer = shash_find_data(&all_dpif_backers, type);
1134     if (backer) {
1135         backer->refcount++;
1136         *backerp = backer;
1137         return 0;
1138     }
1139
1140     backer_name = xasprintf("ovs-%s", type);
1141
1142     /* Remove any existing datapaths, since we assume we're the only
1143      * userspace controlling the datapath. */
1144     sset_init(&names);
1145     dp_enumerate_names(type, &names);
1146     SSET_FOR_EACH(name, &names) {
1147         struct dpif *old_dpif;
1148
1149         /* Don't remove our backer if it exists. */
1150         if (!strcmp(name, backer_name)) {
1151             continue;
1152         }
1153
1154         if (dpif_open(name, type, &old_dpif)) {
1155             VLOG_WARN("couldn't open old datapath %s to remove it", name);
1156         } else {
1157             dpif_delete(old_dpif);
1158             dpif_close(old_dpif);
1159         }
1160     }
1161     sset_destroy(&names);
1162
1163     backer = xmalloc(sizeof *backer);
1164
1165     error = dpif_create_and_open(backer_name, type, &backer->dpif);
1166     free(backer_name);
1167     if (error) {
1168         VLOG_ERR("failed to open datapath of type %s: %s", type,
1169                  ovs_strerror(error));
1170         free(backer);
1171         return error;
1172     }
1173     backer->udpif = udpif_create(backer, backer->dpif);
1174
1175     backer->type = xstrdup(type);
1176     backer->governor = NULL;
1177     backer->refcount = 1;
1178     hmap_init(&backer->odp_to_ofport_map);
1179     ovs_rwlock_init(&backer->odp_to_ofport_lock);
1180     hmap_init(&backer->drop_keys);
1181     hmap_init(&backer->subfacets);
1182     timer_set_duration(&backer->next_expiration, 1000);
1183     backer->need_revalidate = 0;
1184     simap_init(&backer->tnl_backers);
1185     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1186     *backerp = backer;
1187
1188     if (backer->recv_set_enable) {
1189         dpif_flow_flush(backer->dpif);
1190     }
1191
1192     /* Loop through the ports already on the datapath and remove any
1193      * that we don't need anymore. */
1194     list_init(&garbage_list);
1195     dpif_port_dump_start(&port_dump, backer->dpif);
1196     while (dpif_port_dump_next(&port_dump, &port)) {
1197         node = shash_find(&init_ofp_ports, port.name);
1198         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1199             garbage = xmalloc(sizeof *garbage);
1200             garbage->odp_port = port.port_no;
1201             list_push_front(&garbage_list, &garbage->list_node);
1202         }
1203     }
1204     dpif_port_dump_done(&port_dump);
1205
1206     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1207         dpif_port_del(backer->dpif, garbage->odp_port);
1208         list_remove(&garbage->list_node);
1209         free(garbage);
1210     }
1211
1212     shash_add(&all_dpif_backers, type, backer);
1213
1214     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1215     if (error) {
1216         VLOG_ERR("failed to listen on datapath of type %s: %s",
1217                  type, ovs_strerror(error));
1218         close_dpif_backer(backer);
1219         return error;
1220     }
1221     udpif_recv_set(backer->udpif, n_handler_threads,
1222                    backer->recv_set_enable);
1223     backer->n_handler_threads = n_handler_threads;
1224
1225     backer->max_n_subfacet = 0;
1226     backer->created = time_msec();
1227     backer->avg_n_subfacet = 0;
1228     backer->avg_subfacet_life = 0;
1229
1230     return error;
1231 }
1232
1233 static int
1234 construct(struct ofproto *ofproto_)
1235 {
1236     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1237     struct shash_node *node, *next;
1238     int error;
1239
1240     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1241     if (error) {
1242         return error;
1243     }
1244
1245     ofproto->netflow = NULL;
1246     ofproto->sflow = NULL;
1247     ofproto->ipfix = NULL;
1248     ofproto->stp = NULL;
1249     hmap_init(&ofproto->bundles);
1250     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1251     ofproto->mbridge = mbridge_create();
1252     ofproto->has_bonded_bundles = false;
1253     ovs_mutex_init(&ofproto->stats_mutex);
1254     ovs_mutex_init(&ofproto->vsp_mutex);
1255
1256     classifier_init(&ofproto->facets);
1257     ofproto->consistency_rl = LLONG_MIN;
1258
1259     guarded_list_init(&ofproto->pins);
1260
1261     ofproto_dpif_unixctl_init();
1262
1263     hmap_init(&ofproto->vlandev_map);
1264     hmap_init(&ofproto->realdev_vid_map);
1265
1266     sset_init(&ofproto->ports);
1267     sset_init(&ofproto->ghost_ports);
1268     sset_init(&ofproto->port_poll_set);
1269     ofproto->port_poll_errno = 0;
1270
1271     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1272         struct iface_hint *iface_hint = node->data;
1273
1274         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1275             /* Check if the datapath already has this port. */
1276             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1277                 sset_add(&ofproto->ports, node->name);
1278             }
1279
1280             free(iface_hint->br_name);
1281             free(iface_hint->br_type);
1282             free(iface_hint);
1283             shash_delete(&init_ofp_ports, node);
1284         }
1285     }
1286
1287     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1288                 hash_string(ofproto->up.name, 0));
1289     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1290
1291     ofproto_init_tables(ofproto_, N_TABLES);
1292     error = add_internal_flows(ofproto);
1293     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1294
1295     ofproto->n_hit = 0;
1296     ofproto->n_missed = 0;
1297
1298     return error;
1299 }
1300
1301 static int
1302 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1303                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1304 {
1305     struct ofputil_flow_mod fm;
1306     int error;
1307
1308     match_init_catchall(&fm.match);
1309     fm.priority = 0;
1310     match_set_reg(&fm.match, 0, id);
1311     fm.new_cookie = htonll(0);
1312     fm.cookie = htonll(0);
1313     fm.cookie_mask = htonll(0);
1314     fm.modify_cookie = false;
1315     fm.table_id = TBL_INTERNAL;
1316     fm.command = OFPFC_ADD;
1317     fm.idle_timeout = 0;
1318     fm.hard_timeout = 0;
1319     fm.buffer_id = 0;
1320     fm.out_port = 0;
1321     fm.flags = 0;
1322     fm.ofpacts = ofpacts->data;
1323     fm.ofpacts_len = ofpacts->size;
1324
1325     error = ofproto_flow_mod(&ofproto->up, &fm);
1326     if (error) {
1327         VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1328                     id, ofperr_to_string(error));
1329         return error;
1330     }
1331
1332     if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1333                                   rulep)) {
1334         rule_dpif_unref(*rulep);
1335     } else {
1336         NOT_REACHED();
1337     }
1338
1339     return 0;
1340 }
1341
1342 static int
1343 add_internal_flows(struct ofproto_dpif *ofproto)
1344 {
1345     struct ofpact_controller *controller;
1346     uint64_t ofpacts_stub[128 / 8];
1347     struct ofpbuf ofpacts;
1348     int error;
1349     int id;
1350
1351     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1352     id = 1;
1353
1354     controller = ofpact_put_CONTROLLER(&ofpacts);
1355     controller->max_len = UINT16_MAX;
1356     controller->controller_id = 0;
1357     controller->reason = OFPR_NO_MATCH;
1358     ofpact_pad(&ofpacts);
1359
1360     error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1361     if (error) {
1362         return error;
1363     }
1364
1365     ofpbuf_clear(&ofpacts);
1366     error = add_internal_flow(ofproto, id++, &ofpacts,
1367                               &ofproto->no_packet_in_rule);
1368     if (error) {
1369         return error;
1370     }
1371
1372     error = add_internal_flow(ofproto, id++, &ofpacts,
1373                               &ofproto->drop_frags_rule);
1374     return error;
1375 }
1376
1377 static void
1378 destruct(struct ofproto *ofproto_)
1379 {
1380     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1381     struct rule_dpif *rule, *next_rule;
1382     struct ofproto_packet_in *pin, *next_pin;
1383     struct facet *facet, *next_facet;
1384     struct cls_cursor cursor;
1385     struct oftable *table;
1386     struct list pins;
1387
1388     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1389     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1390     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1391     CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1392         facet_remove(facet);
1393     }
1394
1395     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1396     ovs_rwlock_wrlock(&xlate_rwlock);
1397     xlate_remove_ofproto(ofproto);
1398     ovs_rwlock_unlock(&xlate_rwlock);
1399
1400     /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a
1401      * use-after-free error. */
1402     udpif_revalidate(ofproto->backer->udpif);
1403
1404     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1405
1406     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1407         struct cls_cursor cursor;
1408
1409         ovs_rwlock_rdlock(&table->cls.rwlock);
1410         cls_cursor_init(&cursor, &table->cls, NULL);
1411         ovs_rwlock_unlock(&table->cls.rwlock);
1412         CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1413             ofproto_rule_delete(&ofproto->up, &rule->up);
1414         }
1415     }
1416
1417     guarded_list_pop_all(&ofproto->pins, &pins);
1418     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1419         list_remove(&pin->list_node);
1420         free(CONST_CAST(void *, pin->up.packet));
1421         free(pin);
1422     }
1423     guarded_list_destroy(&ofproto->pins);
1424
1425     mbridge_unref(ofproto->mbridge);
1426
1427     netflow_destroy(ofproto->netflow);
1428     dpif_sflow_unref(ofproto->sflow);
1429     hmap_destroy(&ofproto->bundles);
1430     mac_learning_unref(ofproto->ml);
1431
1432     classifier_destroy(&ofproto->facets);
1433
1434     hmap_destroy(&ofproto->vlandev_map);
1435     hmap_destroy(&ofproto->realdev_vid_map);
1436
1437     sset_destroy(&ofproto->ports);
1438     sset_destroy(&ofproto->ghost_ports);
1439     sset_destroy(&ofproto->port_poll_set);
1440
1441     ovs_mutex_destroy(&ofproto->stats_mutex);
1442     ovs_mutex_destroy(&ofproto->vsp_mutex);
1443
1444     close_dpif_backer(ofproto->backer);
1445 }
1446
1447 static int
1448 run_fast(struct ofproto *ofproto_)
1449 {
1450     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1451     struct ofproto_packet_in *pin, *next_pin;
1452     struct list pins;
1453
1454     /* Do not perform any periodic activity required by 'ofproto' while
1455      * waiting for flow restore to complete. */
1456     if (ofproto_get_flow_restore_wait()) {
1457         return 0;
1458     }
1459
1460     guarded_list_pop_all(&ofproto->pins, &pins);
1461     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1462         connmgr_send_packet_in(ofproto->up.connmgr, pin);
1463         list_remove(&pin->list_node);
1464         free(CONST_CAST(void *, pin->up.packet));
1465         free(pin);
1466     }
1467
1468     return 0;
1469 }
1470
1471 static int
1472 run(struct ofproto *ofproto_)
1473 {
1474     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1475     struct ofport_dpif *ofport;
1476     struct ofbundle *bundle;
1477     int error;
1478
1479     if (mbridge_need_revalidate(ofproto->mbridge)) {
1480         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1481         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1482         mac_learning_flush(ofproto->ml);
1483         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1484     }
1485
1486     /* Do not perform any periodic activity below required by 'ofproto' while
1487      * waiting for flow restore to complete. */
1488     if (ofproto_get_flow_restore_wait()) {
1489         return 0;
1490     }
1491
1492     error = run_fast(ofproto_);
1493     if (error) {
1494         return error;
1495     }
1496
1497     if (ofproto->netflow) {
1498         if (netflow_run(ofproto->netflow)) {
1499             send_netflow_active_timeouts(ofproto);
1500         }
1501     }
1502     if (ofproto->sflow) {
1503         dpif_sflow_run(ofproto->sflow);
1504     }
1505     if (ofproto->ipfix) {
1506         dpif_ipfix_run(ofproto->ipfix);
1507     }
1508
1509     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1510         port_run(ofport);
1511     }
1512     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1513         bundle_run(bundle);
1514     }
1515
1516     stp_run(ofproto);
1517     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1518     if (mac_learning_run(ofproto->ml)) {
1519         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1520     }
1521     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1522
1523     /* Check the consistency of a random facet, to aid debugging. */
1524     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1525     if (time_msec() >= ofproto->consistency_rl
1526         && !classifier_is_empty(&ofproto->facets)
1527         && !ofproto->backer->need_revalidate) {
1528         struct cls_subtable *table;
1529         struct cls_rule *cr;
1530         struct facet *facet;
1531
1532         ofproto->consistency_rl = time_msec() + 250;
1533
1534         table = CONTAINER_OF(hmap_random_node(&ofproto->facets.subtables),
1535                              struct cls_subtable, hmap_node);
1536         cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1537                           hmap_node);
1538         facet = CONTAINER_OF(cr, struct facet, cr);
1539
1540         if (!facet_check_consistency(facet)) {
1541             ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1542         }
1543     }
1544     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1545
1546     return 0;
1547 }
1548
1549 static void
1550 wait(struct ofproto *ofproto_)
1551 {
1552     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1553     struct ofbundle *bundle;
1554
1555     if (ofproto_get_flow_restore_wait()) {
1556         return;
1557     }
1558
1559     if (ofproto->sflow) {
1560         dpif_sflow_wait(ofproto->sflow);
1561     }
1562     if (ofproto->ipfix) {
1563         dpif_ipfix_wait(ofproto->ipfix);
1564     }
1565     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1566         bundle_wait(bundle);
1567     }
1568     if (ofproto->netflow) {
1569         netflow_wait(ofproto->netflow);
1570     }
1571     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1572     mac_learning_wait(ofproto->ml);
1573     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1574     stp_wait(ofproto);
1575     if (ofproto->backer->need_revalidate) {
1576         /* Shouldn't happen, but if it does just go around again. */
1577         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1578         poll_immediate_wake();
1579     }
1580 }
1581
1582 static void
1583 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1584 {
1585     const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1586     struct cls_cursor cursor;
1587     size_t n_subfacets = 0;
1588     struct facet *facet;
1589
1590     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1591     simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1592     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1593
1594     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1595     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1596     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1597         n_subfacets += list_size(&facet->subfacets);
1598     }
1599     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1600     simap_increase(usage, "subfacets", n_subfacets);
1601 }
1602
1603 static void
1604 flush(struct ofproto *ofproto_)
1605 {
1606     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1607     struct subfacet *subfacet, *next_subfacet;
1608     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1609     int n_batch;
1610
1611     n_batch = 0;
1612     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1613                         &ofproto->backer->subfacets) {
1614         if (subfacet->facet->ofproto != ofproto) {
1615             continue;
1616         }
1617
1618         if (subfacet->path != SF_NOT_INSTALLED) {
1619             batch[n_batch++] = subfacet;
1620             if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1621                 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1622                 n_batch = 0;
1623             }
1624         } else {
1625             subfacet_destroy(subfacet);
1626         }
1627     }
1628
1629     if (n_batch > 0) {
1630         subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1631     }
1632 }
1633
1634 static void
1635 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1636              bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1637 {
1638     *arp_match_ip = true;
1639     *actions = (OFPUTIL_A_OUTPUT |
1640                 OFPUTIL_A_SET_VLAN_VID |
1641                 OFPUTIL_A_SET_VLAN_PCP |
1642                 OFPUTIL_A_STRIP_VLAN |
1643                 OFPUTIL_A_SET_DL_SRC |
1644                 OFPUTIL_A_SET_DL_DST |
1645                 OFPUTIL_A_SET_NW_SRC |
1646                 OFPUTIL_A_SET_NW_DST |
1647                 OFPUTIL_A_SET_NW_TOS |
1648                 OFPUTIL_A_SET_TP_SRC |
1649                 OFPUTIL_A_SET_TP_DST |
1650                 OFPUTIL_A_ENQUEUE);
1651 }
1652
1653 static void
1654 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1655 {
1656     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1657     struct dpif_dp_stats s;
1658     uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1659     uint64_t n_lookup;
1660
1661     strcpy(ots->name, "classifier");
1662
1663     dpif_get_dp_stats(ofproto->backer->dpif, &s);
1664     rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1665     rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1666     rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1667
1668     n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1669     ots->lookup_count = htonll(n_lookup);
1670     ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1671 }
1672
1673 static struct ofport *
1674 port_alloc(void)
1675 {
1676     struct ofport_dpif *port = xmalloc(sizeof *port);
1677     return &port->up;
1678 }
1679
1680 static void
1681 port_dealloc(struct ofport *port_)
1682 {
1683     struct ofport_dpif *port = ofport_dpif_cast(port_);
1684     free(port);
1685 }
1686
1687 static int
1688 port_construct(struct ofport *port_)
1689 {
1690     struct ofport_dpif *port = ofport_dpif_cast(port_);
1691     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1692     const struct netdev *netdev = port->up.netdev;
1693     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1694     struct dpif_port dpif_port;
1695     int error;
1696
1697     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1698     port->bundle = NULL;
1699     port->cfm = NULL;
1700     port->bfd = NULL;
1701     port->may_enable = true;
1702     port->stp_port = NULL;
1703     port->stp_state = STP_DISABLED;
1704     port->is_tunnel = false;
1705     port->peer = NULL;
1706     port->qdscp = NULL;
1707     port->n_qdscp = 0;
1708     port->realdev_ofp_port = 0;
1709     port->vlandev_vid = 0;
1710     port->carrier_seq = netdev_get_carrier_resets(netdev);
1711
1712     if (netdev_vport_is_patch(netdev)) {
1713         /* By bailing out here, we don't submit the port to the sFlow module
1714          * to be considered for counter polling export.  This is correct
1715          * because the patch port represents an interface that sFlow considers
1716          * to be "internal" to the switch as a whole, and therefore not an
1717          * candidate for counter polling. */
1718         port->odp_port = ODPP_NONE;
1719         ofport_update_peer(port);
1720         return 0;
1721     }
1722
1723     error = dpif_port_query_by_name(ofproto->backer->dpif,
1724                                     netdev_vport_get_dpif_port(netdev, namebuf,
1725                                                                sizeof namebuf),
1726                                     &dpif_port);
1727     if (error) {
1728         return error;
1729     }
1730
1731     port->odp_port = dpif_port.port_no;
1732
1733     if (netdev_get_tunnel_config(netdev)) {
1734         tnl_port_add(port, port->up.netdev, port->odp_port);
1735         port->is_tunnel = true;
1736     } else {
1737         /* Sanity-check that a mapping doesn't already exist.  This
1738          * shouldn't happen for non-tunnel ports. */
1739         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1740             VLOG_ERR("port %s already has an OpenFlow port number",
1741                      dpif_port.name);
1742             dpif_port_destroy(&dpif_port);
1743             return EBUSY;
1744         }
1745
1746         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1747         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1748                     hash_odp_port(port->odp_port));
1749         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1750     }
1751     dpif_port_destroy(&dpif_port);
1752
1753     if (ofproto->sflow) {
1754         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1755     }
1756
1757     return 0;
1758 }
1759
1760 static void
1761 port_destruct(struct ofport *port_)
1762 {
1763     struct ofport_dpif *port = ofport_dpif_cast(port_);
1764     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1765     const char *devname = netdev_get_name(port->up.netdev);
1766     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1767     const char *dp_port_name;
1768
1769     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1770     ovs_rwlock_wrlock(&xlate_rwlock);
1771     xlate_ofport_remove(port);
1772     ovs_rwlock_unlock(&xlate_rwlock);
1773
1774     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1775                                               sizeof namebuf);
1776     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1777         /* The underlying device is still there, so delete it.  This
1778          * happens when the ofproto is being destroyed, since the caller
1779          * assumes that removal of attached ports will happen as part of
1780          * destruction. */
1781         if (!port->is_tunnel) {
1782             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1783         }
1784     }
1785
1786     if (port->peer) {
1787         port->peer->peer = NULL;
1788         port->peer = NULL;
1789     }
1790
1791     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1792         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1793         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1794         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1795     }
1796
1797     tnl_port_del(port);
1798     sset_find_and_delete(&ofproto->ports, devname);
1799     sset_find_and_delete(&ofproto->ghost_ports, devname);
1800     bundle_remove(port_);
1801     set_cfm(port_, NULL);
1802     set_bfd(port_, NULL);
1803     if (ofproto->sflow) {
1804         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1805     }
1806
1807     free(port->qdscp);
1808 }
1809
1810 static void
1811 port_modified(struct ofport *port_)
1812 {
1813     struct ofport_dpif *port = ofport_dpif_cast(port_);
1814
1815     if (port->bundle && port->bundle->bond) {
1816         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1817     }
1818
1819     if (port->cfm) {
1820         cfm_set_netdev(port->cfm, port->up.netdev);
1821     }
1822
1823     if (port->bfd) {
1824         bfd_set_netdev(port->bfd, port->up.netdev);
1825     }
1826
1827     ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1828                                      port->up.pp.hw_addr);
1829
1830     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1831                                                 port->odp_port)) {
1832         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1833             REV_RECONFIGURE;
1834     }
1835
1836     ofport_update_peer(port);
1837 }
1838
1839 static void
1840 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1841 {
1842     struct ofport_dpif *port = ofport_dpif_cast(port_);
1843     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1844     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1845
1846     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1847                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1848                    OFPUTIL_PC_NO_PACKET_IN)) {
1849         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1850
1851         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1852             bundle_update(port->bundle);
1853         }
1854     }
1855 }
1856
1857 static int
1858 set_sflow(struct ofproto *ofproto_,
1859           const struct ofproto_sflow_options *sflow_options)
1860 {
1861     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1862     struct dpif_sflow *ds = ofproto->sflow;
1863
1864     if (sflow_options) {
1865         if (!ds) {
1866             struct ofport_dpif *ofport;
1867
1868             ds = ofproto->sflow = dpif_sflow_create();
1869             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1870                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1871             }
1872             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1873         }
1874         dpif_sflow_set_options(ds, sflow_options);
1875     } else {
1876         if (ds) {
1877             dpif_sflow_unref(ds);
1878             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1879             ofproto->sflow = NULL;
1880         }
1881     }
1882     return 0;
1883 }
1884
1885 static int
1886 set_ipfix(
1887     struct ofproto *ofproto_,
1888     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1889     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1890     size_t n_flow_exporters_options)
1891 {
1892     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1893     struct dpif_ipfix *di = ofproto->ipfix;
1894     bool has_options = bridge_exporter_options || flow_exporters_options;
1895
1896     if (has_options && !di) {
1897         di = ofproto->ipfix = dpif_ipfix_create();
1898     }
1899
1900     if (di) {
1901         /* Call set_options in any case to cleanly flush the flow
1902          * caches in the last exporters that are to be destroyed. */
1903         dpif_ipfix_set_options(
1904             di, bridge_exporter_options, flow_exporters_options,
1905             n_flow_exporters_options);
1906
1907         if (!has_options) {
1908             dpif_ipfix_unref(di);
1909             ofproto->ipfix = NULL;
1910         }
1911     }
1912
1913     return 0;
1914 }
1915
1916 static int
1917 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1918 {
1919     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1920     int error = 0;
1921
1922     if (s) {
1923         if (!ofport->cfm) {
1924             struct ofproto_dpif *ofproto;
1925
1926             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1927             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1928             ofport->cfm = cfm_create(ofport->up.netdev);
1929         }
1930
1931         if (cfm_configure(ofport->cfm, s)) {
1932             error = 0;
1933             goto out;
1934         }
1935
1936         error = EINVAL;
1937     }
1938     cfm_unref(ofport->cfm);
1939     ofport->cfm = NULL;
1940 out:
1941     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1942                                      ofport->up.pp.hw_addr);
1943     return error;
1944 }
1945
1946 static bool
1947 get_cfm_status(const struct ofport *ofport_,
1948                struct ofproto_cfm_status *status)
1949 {
1950     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1951
1952     if (ofport->cfm) {
1953         status->faults = cfm_get_fault(ofport->cfm);
1954         status->flap_count = cfm_get_flap_count(ofport->cfm);
1955         status->remote_opstate = cfm_get_opup(ofport->cfm);
1956         status->health = cfm_get_health(ofport->cfm);
1957         cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1958         return true;
1959     } else {
1960         return false;
1961     }
1962 }
1963
1964 static int
1965 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1966 {
1967     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1968     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1969     struct bfd *old;
1970
1971     old = ofport->bfd;
1972     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1973                                 cfg, ofport->up.netdev);
1974     if (ofport->bfd != old) {
1975         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1976     }
1977     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1978                                      ofport->up.pp.hw_addr);
1979     return 0;
1980 }
1981
1982 static int
1983 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1984 {
1985     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1986
1987     if (ofport->bfd) {
1988         bfd_get_status(ofport->bfd, smap);
1989         return 0;
1990     } else {
1991         return ENOENT;
1992     }
1993 }
1994 \f
1995 /* Spanning Tree. */
1996
1997 static void
1998 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1999 {
2000     struct ofproto_dpif *ofproto = ofproto_;
2001     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2002     struct ofport_dpif *ofport;
2003
2004     ofport = stp_port_get_aux(sp);
2005     if (!ofport) {
2006         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2007                      ofproto->up.name, port_num);
2008     } else {
2009         struct eth_header *eth = pkt->l2;
2010
2011         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2012         if (eth_addr_is_zero(eth->eth_src)) {
2013             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2014                          "with unknown MAC", ofproto->up.name, port_num);
2015         } else {
2016             ofproto_dpif_send_packet(ofport, pkt);
2017         }
2018     }
2019     ofpbuf_delete(pkt);
2020 }
2021
2022 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2023 static int
2024 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2025 {
2026     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2027
2028     /* Only revalidate flows if the configuration changed. */
2029     if (!s != !ofproto->stp) {
2030         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2031     }
2032
2033     if (s) {
2034         if (!ofproto->stp) {
2035             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2036                                       send_bpdu_cb, ofproto);
2037             ofproto->stp_last_tick = time_msec();
2038         }
2039
2040         stp_set_bridge_id(ofproto->stp, s->system_id);
2041         stp_set_bridge_priority(ofproto->stp, s->priority);
2042         stp_set_hello_time(ofproto->stp, s->hello_time);
2043         stp_set_max_age(ofproto->stp, s->max_age);
2044         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2045     }  else {
2046         struct ofport *ofport;
2047
2048         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2049             set_stp_port(ofport, NULL);
2050         }
2051
2052         stp_unref(ofproto->stp);
2053         ofproto->stp = NULL;
2054     }
2055
2056     return 0;
2057 }
2058
2059 static int
2060 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2061 {
2062     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2063
2064     if (ofproto->stp) {
2065         s->enabled = true;
2066         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2067         s->designated_root = stp_get_designated_root(ofproto->stp);
2068         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2069     } else {
2070         s->enabled = false;
2071     }
2072
2073     return 0;
2074 }
2075
2076 static void
2077 update_stp_port_state(struct ofport_dpif *ofport)
2078 {
2079     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2080     enum stp_state state;
2081
2082     /* Figure out new state. */
2083     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2084                              : STP_DISABLED;
2085
2086     /* Update state. */
2087     if (ofport->stp_state != state) {
2088         enum ofputil_port_state of_state;
2089         bool fwd_change;
2090
2091         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2092                     netdev_get_name(ofport->up.netdev),
2093                     stp_state_name(ofport->stp_state),
2094                     stp_state_name(state));
2095         if (stp_learn_in_state(ofport->stp_state)
2096                 != stp_learn_in_state(state)) {
2097             /* xxx Learning action flows should also be flushed. */
2098             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2099             mac_learning_flush(ofproto->ml);
2100             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2101         }
2102         fwd_change = stp_forward_in_state(ofport->stp_state)
2103                         != stp_forward_in_state(state);
2104
2105         ofproto->backer->need_revalidate = REV_STP;
2106         ofport->stp_state = state;
2107         ofport->stp_state_entered = time_msec();
2108
2109         if (fwd_change && ofport->bundle) {
2110             bundle_update(ofport->bundle);
2111         }
2112
2113         /* Update the STP state bits in the OpenFlow port description. */
2114         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2115         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2116                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2117                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2118                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2119                      : 0);
2120         ofproto_port_set_state(&ofport->up, of_state);
2121     }
2122 }
2123
2124 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2125  * caller is responsible for assigning STP port numbers and ensuring
2126  * there are no duplicates. */
2127 static int
2128 set_stp_port(struct ofport *ofport_,
2129              const struct ofproto_port_stp_settings *s)
2130 {
2131     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2132     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2133     struct stp_port *sp = ofport->stp_port;
2134
2135     if (!s || !s->enable) {
2136         if (sp) {
2137             ofport->stp_port = NULL;
2138             stp_port_disable(sp);
2139             update_stp_port_state(ofport);
2140         }
2141         return 0;
2142     } else if (sp && stp_port_no(sp) != s->port_num
2143             && ofport == stp_port_get_aux(sp)) {
2144         /* The port-id changed, so disable the old one if it's not
2145          * already in use by another port. */
2146         stp_port_disable(sp);
2147     }
2148
2149     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2150     stp_port_enable(sp);
2151
2152     stp_port_set_aux(sp, ofport);
2153     stp_port_set_priority(sp, s->priority);
2154     stp_port_set_path_cost(sp, s->path_cost);
2155
2156     update_stp_port_state(ofport);
2157
2158     return 0;
2159 }
2160
2161 static int
2162 get_stp_port_status(struct ofport *ofport_,
2163                     struct ofproto_port_stp_status *s)
2164 {
2165     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2166     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2167     struct stp_port *sp = ofport->stp_port;
2168
2169     if (!ofproto->stp || !sp) {
2170         s->enabled = false;
2171         return 0;
2172     }
2173
2174     s->enabled = true;
2175     s->port_id = stp_port_get_id(sp);
2176     s->state = stp_port_get_state(sp);
2177     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2178     s->role = stp_port_get_role(sp);
2179     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2180
2181     return 0;
2182 }
2183
2184 static void
2185 stp_run(struct ofproto_dpif *ofproto)
2186 {
2187     if (ofproto->stp) {
2188         long long int now = time_msec();
2189         long long int elapsed = now - ofproto->stp_last_tick;
2190         struct stp_port *sp;
2191
2192         if (elapsed > 0) {
2193             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2194             ofproto->stp_last_tick = now;
2195         }
2196         while (stp_get_changed_port(ofproto->stp, &sp)) {
2197             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2198
2199             if (ofport) {
2200                 update_stp_port_state(ofport);
2201             }
2202         }
2203
2204         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2205             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2206             mac_learning_flush(ofproto->ml);
2207             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2208         }
2209     }
2210 }
2211
2212 static void
2213 stp_wait(struct ofproto_dpif *ofproto)
2214 {
2215     if (ofproto->stp) {
2216         poll_timer_wait(1000);
2217     }
2218 }
2219 \f
2220 static int
2221 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2222            size_t n_qdscp)
2223 {
2224     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2225     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2226
2227     if (ofport->n_qdscp != n_qdscp
2228         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2229                               n_qdscp * sizeof *qdscp))) {
2230         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2231         free(ofport->qdscp);
2232         ofport->qdscp = n_qdscp
2233             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2234             : NULL;
2235         ofport->n_qdscp = n_qdscp;
2236     }
2237
2238     return 0;
2239 }
2240 \f
2241 /* Bundles. */
2242
2243 /* Expires all MAC learning entries associated with 'bundle' and forces its
2244  * ofproto to revalidate every flow.
2245  *
2246  * Normally MAC learning entries are removed only from the ofproto associated
2247  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2248  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2249  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2250  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2251  * with the host from which it migrated. */
2252 static void
2253 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2254 {
2255     struct ofproto_dpif *ofproto = bundle->ofproto;
2256     struct mac_learning *ml = ofproto->ml;
2257     struct mac_entry *mac, *next_mac;
2258
2259     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2260     ovs_rwlock_wrlock(&ml->rwlock);
2261     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2262         if (mac->port.p == bundle) {
2263             if (all_ofprotos) {
2264                 struct ofproto_dpif *o;
2265
2266                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2267                     if (o != ofproto) {
2268                         struct mac_entry *e;
2269
2270                         ovs_rwlock_wrlock(&o->ml->rwlock);
2271                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2272                         if (e) {
2273                             mac_learning_expire(o->ml, e);
2274                         }
2275                         ovs_rwlock_unlock(&o->ml->rwlock);
2276                     }
2277                 }
2278             }
2279
2280             mac_learning_expire(ml, mac);
2281         }
2282     }
2283     ovs_rwlock_unlock(&ml->rwlock);
2284 }
2285
2286 static struct ofbundle *
2287 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2288 {
2289     struct ofbundle *bundle;
2290
2291     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2292                              &ofproto->bundles) {
2293         if (bundle->aux == aux) {
2294             return bundle;
2295         }
2296     }
2297     return NULL;
2298 }
2299
2300 static void
2301 bundle_update(struct ofbundle *bundle)
2302 {
2303     struct ofport_dpif *port;
2304
2305     bundle->floodable = true;
2306     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2307         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2308             || !stp_forward_in_state(port->stp_state)) {
2309             bundle->floodable = false;
2310             break;
2311         }
2312     }
2313 }
2314
2315 static void
2316 bundle_del_port(struct ofport_dpif *port)
2317 {
2318     struct ofbundle *bundle = port->bundle;
2319
2320     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2321
2322     list_remove(&port->bundle_node);
2323     port->bundle = NULL;
2324
2325     if (bundle->lacp) {
2326         lacp_slave_unregister(bundle->lacp, port);
2327     }
2328     if (bundle->bond) {
2329         bond_slave_unregister(bundle->bond, port);
2330     }
2331
2332     bundle_update(bundle);
2333 }
2334
2335 static bool
2336 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2337                 struct lacp_slave_settings *lacp)
2338 {
2339     struct ofport_dpif *port;
2340
2341     port = get_ofp_port(bundle->ofproto, ofp_port);
2342     if (!port) {
2343         return false;
2344     }
2345
2346     if (port->bundle != bundle) {
2347         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2348         if (port->bundle) {
2349             bundle_remove(&port->up);
2350         }
2351
2352         port->bundle = bundle;
2353         list_push_back(&bundle->ports, &port->bundle_node);
2354         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2355             || !stp_forward_in_state(port->stp_state)) {
2356             bundle->floodable = false;
2357         }
2358     }
2359     if (lacp) {
2360         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2361         lacp_slave_register(bundle->lacp, port, lacp);
2362     }
2363
2364     return true;
2365 }
2366
2367 static void
2368 bundle_destroy(struct ofbundle *bundle)
2369 {
2370     struct ofproto_dpif *ofproto;
2371     struct ofport_dpif *port, *next_port;
2372
2373     if (!bundle) {
2374         return;
2375     }
2376
2377     ofproto = bundle->ofproto;
2378     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2379
2380     ovs_rwlock_wrlock(&xlate_rwlock);
2381     xlate_bundle_remove(bundle);
2382     ovs_rwlock_unlock(&xlate_rwlock);
2383
2384     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2385         bundle_del_port(port);
2386     }
2387
2388     bundle_flush_macs(bundle, true);
2389     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2390     free(bundle->name);
2391     free(bundle->trunks);
2392     lacp_unref(bundle->lacp);
2393     bond_unref(bundle->bond);
2394     free(bundle);
2395 }
2396
2397 static int
2398 bundle_set(struct ofproto *ofproto_, void *aux,
2399            const struct ofproto_bundle_settings *s)
2400 {
2401     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2402     bool need_flush = false;
2403     struct ofport_dpif *port;
2404     struct ofbundle *bundle;
2405     unsigned long *trunks;
2406     int vlan;
2407     size_t i;
2408     bool ok;
2409
2410     if (!s) {
2411         bundle_destroy(bundle_lookup(ofproto, aux));
2412         return 0;
2413     }
2414
2415     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2416     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2417
2418     bundle = bundle_lookup(ofproto, aux);
2419     if (!bundle) {
2420         bundle = xmalloc(sizeof *bundle);
2421
2422         bundle->ofproto = ofproto;
2423         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2424                     hash_pointer(aux, 0));
2425         bundle->aux = aux;
2426         bundle->name = NULL;
2427
2428         list_init(&bundle->ports);
2429         bundle->vlan_mode = PORT_VLAN_TRUNK;
2430         bundle->vlan = -1;
2431         bundle->trunks = NULL;
2432         bundle->use_priority_tags = s->use_priority_tags;
2433         bundle->lacp = NULL;
2434         bundle->bond = NULL;
2435
2436         bundle->floodable = true;
2437         mbridge_register_bundle(ofproto->mbridge, bundle);
2438     }
2439
2440     if (!bundle->name || strcmp(s->name, bundle->name)) {
2441         free(bundle->name);
2442         bundle->name = xstrdup(s->name);
2443     }
2444
2445     /* LACP. */
2446     if (s->lacp) {
2447         if (!bundle->lacp) {
2448             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2449             bundle->lacp = lacp_create();
2450         }
2451         lacp_configure(bundle->lacp, s->lacp);
2452     } else {
2453         lacp_unref(bundle->lacp);
2454         bundle->lacp = NULL;
2455     }
2456
2457     /* Update set of ports. */
2458     ok = true;
2459     for (i = 0; i < s->n_slaves; i++) {
2460         if (!bundle_add_port(bundle, s->slaves[i],
2461                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2462             ok = false;
2463         }
2464     }
2465     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2466         struct ofport_dpif *next_port;
2467
2468         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2469             for (i = 0; i < s->n_slaves; i++) {
2470                 if (s->slaves[i] == port->up.ofp_port) {
2471                     goto found;
2472                 }
2473             }
2474
2475             bundle_del_port(port);
2476         found: ;
2477         }
2478     }
2479     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2480
2481     if (list_is_empty(&bundle->ports)) {
2482         bundle_destroy(bundle);
2483         return EINVAL;
2484     }
2485
2486     /* Set VLAN tagging mode */
2487     if (s->vlan_mode != bundle->vlan_mode
2488         || s->use_priority_tags != bundle->use_priority_tags) {
2489         bundle->vlan_mode = s->vlan_mode;
2490         bundle->use_priority_tags = s->use_priority_tags;
2491         need_flush = true;
2492     }
2493
2494     /* Set VLAN tag. */
2495     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2496             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2497             : 0);
2498     if (vlan != bundle->vlan) {
2499         bundle->vlan = vlan;
2500         need_flush = true;
2501     }
2502
2503     /* Get trunked VLANs. */
2504     switch (s->vlan_mode) {
2505     case PORT_VLAN_ACCESS:
2506         trunks = NULL;
2507         break;
2508
2509     case PORT_VLAN_TRUNK:
2510         trunks = CONST_CAST(unsigned long *, s->trunks);
2511         break;
2512
2513     case PORT_VLAN_NATIVE_UNTAGGED:
2514     case PORT_VLAN_NATIVE_TAGGED:
2515         if (vlan != 0 && (!s->trunks
2516                           || !bitmap_is_set(s->trunks, vlan)
2517                           || bitmap_is_set(s->trunks, 0))) {
2518             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2519             if (s->trunks) {
2520                 trunks = bitmap_clone(s->trunks, 4096);
2521             } else {
2522                 trunks = bitmap_allocate1(4096);
2523             }
2524             bitmap_set1(trunks, vlan);
2525             bitmap_set0(trunks, 0);
2526         } else {
2527             trunks = CONST_CAST(unsigned long *, s->trunks);
2528         }
2529         break;
2530
2531     default:
2532         NOT_REACHED();
2533     }
2534     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2535         free(bundle->trunks);
2536         if (trunks == s->trunks) {
2537             bundle->trunks = vlan_bitmap_clone(trunks);
2538         } else {
2539             bundle->trunks = trunks;
2540             trunks = NULL;
2541         }
2542         need_flush = true;
2543     }
2544     if (trunks != s->trunks) {
2545         free(trunks);
2546     }
2547
2548     /* Bonding. */
2549     if (!list_is_short(&bundle->ports)) {
2550         bundle->ofproto->has_bonded_bundles = true;
2551         if (bundle->bond) {
2552             if (bond_reconfigure(bundle->bond, s->bond)) {
2553                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2554             }
2555         } else {
2556             bundle->bond = bond_create(s->bond);
2557             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2558         }
2559
2560         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2561             bond_slave_register(bundle->bond, port, port->up.netdev);
2562         }
2563     } else {
2564         bond_unref(bundle->bond);
2565         bundle->bond = NULL;
2566     }
2567
2568     /* If we changed something that would affect MAC learning, un-learn
2569      * everything on this port and force flow revalidation. */
2570     if (need_flush) {
2571         bundle_flush_macs(bundle, false);
2572     }
2573
2574     return 0;
2575 }
2576
2577 static void
2578 bundle_remove(struct ofport *port_)
2579 {
2580     struct ofport_dpif *port = ofport_dpif_cast(port_);
2581     struct ofbundle *bundle = port->bundle;
2582
2583     if (bundle) {
2584         bundle_del_port(port);
2585         if (list_is_empty(&bundle->ports)) {
2586             bundle_destroy(bundle);
2587         } else if (list_is_short(&bundle->ports)) {
2588             bond_unref(bundle->bond);
2589             bundle->bond = NULL;
2590         }
2591     }
2592 }
2593
2594 static void
2595 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2596 {
2597     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2598     struct ofport_dpif *port = port_;
2599     uint8_t ea[ETH_ADDR_LEN];
2600     int error;
2601
2602     error = netdev_get_etheraddr(port->up.netdev, ea);
2603     if (!error) {
2604         struct ofpbuf packet;
2605         void *packet_pdu;
2606
2607         ofpbuf_init(&packet, 0);
2608         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2609                                  pdu_size);
2610         memcpy(packet_pdu, pdu, pdu_size);
2611
2612         ofproto_dpif_send_packet(port, &packet);
2613         ofpbuf_uninit(&packet);
2614     } else {
2615         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2616                     "%s (%s)", port->bundle->name,
2617                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2618     }
2619 }
2620
2621 static void
2622 bundle_send_learning_packets(struct ofbundle *bundle)
2623 {
2624     struct ofproto_dpif *ofproto = bundle->ofproto;
2625     struct ofpbuf *learning_packet;
2626     int error, n_packets, n_errors;
2627     struct mac_entry *e;
2628     struct list packets;
2629
2630     list_init(&packets);
2631     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2632     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2633         if (e->port.p != bundle) {
2634             void *port_void;
2635
2636             learning_packet = bond_compose_learning_packet(bundle->bond,
2637                                                            e->mac, e->vlan,
2638                                                            &port_void);
2639             learning_packet->private_p = port_void;
2640             list_push_back(&packets, &learning_packet->list_node);
2641         }
2642     }
2643     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2644
2645     error = n_packets = n_errors = 0;
2646     LIST_FOR_EACH (learning_packet, list_node, &packets) {
2647         int ret;
2648
2649         ret = ofproto_dpif_send_packet(learning_packet->private_p, learning_packet);
2650         if (ret) {
2651             error = ret;
2652             n_errors++;
2653         }
2654         n_packets++;
2655     }
2656     ofpbuf_list_delete(&packets);
2657
2658     if (n_errors) {
2659         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2660         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2661                      "packets, last error was: %s",
2662                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2663     } else {
2664         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2665                  bundle->name, n_packets);
2666     }
2667 }
2668
2669 static void
2670 bundle_run(struct ofbundle *bundle)
2671 {
2672     if (bundle->lacp) {
2673         lacp_run(bundle->lacp, send_pdu_cb);
2674     }
2675     if (bundle->bond) {
2676         struct ofport_dpif *port;
2677
2678         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2679             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2680         }
2681
2682         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2683             bundle->ofproto->backer->need_revalidate = REV_BOND;
2684         }
2685
2686         if (bond_should_send_learning_packets(bundle->bond)) {
2687             bundle_send_learning_packets(bundle);
2688         }
2689     }
2690 }
2691
2692 static void
2693 bundle_wait(struct ofbundle *bundle)
2694 {
2695     if (bundle->lacp) {
2696         lacp_wait(bundle->lacp);
2697     }
2698     if (bundle->bond) {
2699         bond_wait(bundle->bond);
2700     }
2701 }
2702 \f
2703 /* Mirrors. */
2704
2705 static int
2706 mirror_set__(struct ofproto *ofproto_, void *aux,
2707              const struct ofproto_mirror_settings *s)
2708 {
2709     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2710     struct ofbundle **srcs, **dsts;
2711     int error;
2712     size_t i;
2713
2714     if (!s) {
2715         mirror_destroy(ofproto->mbridge, aux);
2716         return 0;
2717     }
2718
2719     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2720     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2721
2722     for (i = 0; i < s->n_srcs; i++) {
2723         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2724     }
2725
2726     for (i = 0; i < s->n_dsts; i++) {
2727         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2728     }
2729
2730     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2731                        s->n_dsts, s->src_vlans,
2732                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2733     free(srcs);
2734     free(dsts);
2735     return error;
2736 }
2737
2738 static int
2739 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2740                    uint64_t *packets, uint64_t *bytes)
2741 {
2742     push_all_stats();
2743     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2744                             bytes);
2745 }
2746
2747 static int
2748 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2749 {
2750     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2751     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2752     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2753         mac_learning_flush(ofproto->ml);
2754     }
2755     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2756     return 0;
2757 }
2758
2759 static bool
2760 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2761 {
2762     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2763     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2764     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2765 }
2766
2767 static void
2768 forward_bpdu_changed(struct ofproto *ofproto_)
2769 {
2770     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2771     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2772 }
2773
2774 static void
2775 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2776                      size_t max_entries)
2777 {
2778     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2779     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2780     mac_learning_set_idle_time(ofproto->ml, idle_time);
2781     mac_learning_set_max_entries(ofproto->ml, max_entries);
2782     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2783 }
2784 \f
2785 /* Ports. */
2786
2787 static struct ofport_dpif *
2788 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2789 {
2790     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2791     return ofport ? ofport_dpif_cast(ofport) : NULL;
2792 }
2793
2794 static struct ofport_dpif *
2795 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2796 {
2797     struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2798     return port && &ofproto->up == port->up.ofproto ? port : NULL;
2799 }
2800
2801 static void
2802 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2803                             struct ofproto_port *ofproto_port,
2804                             struct dpif_port *dpif_port)
2805 {
2806     ofproto_port->name = dpif_port->name;
2807     ofproto_port->type = dpif_port->type;
2808     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2809 }
2810
2811 static void
2812 ofport_update_peer(struct ofport_dpif *ofport)
2813 {
2814     const struct ofproto_dpif *ofproto;
2815     struct dpif_backer *backer;
2816     char *peer_name;
2817
2818     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2819         return;
2820     }
2821
2822     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2823     backer->need_revalidate = REV_RECONFIGURE;
2824
2825     if (ofport->peer) {
2826         ofport->peer->peer = NULL;
2827         ofport->peer = NULL;
2828     }
2829
2830     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2831     if (!peer_name) {
2832         return;
2833     }
2834
2835     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2836         struct ofport *peer_ofport;
2837         struct ofport_dpif *peer;
2838         char *peer_peer;
2839
2840         if (ofproto->backer != backer) {
2841             continue;
2842         }
2843
2844         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2845         if (!peer_ofport) {
2846             continue;
2847         }
2848
2849         peer = ofport_dpif_cast(peer_ofport);
2850         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2851         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2852                                  peer_peer)) {
2853             ofport->peer = peer;
2854             ofport->peer->peer = ofport;
2855         }
2856         free(peer_peer);
2857
2858         break;
2859     }
2860     free(peer_name);
2861 }
2862
2863 static void
2864 port_run(struct ofport_dpif *ofport)
2865 {
2866     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2867     bool carrier_changed = carrier_seq != ofport->carrier_seq;
2868     bool enable = netdev_get_carrier(ofport->up.netdev);
2869     bool cfm_enable = false;
2870     bool bfd_enable = false;
2871
2872     ofport->carrier_seq = carrier_seq;
2873
2874     if (ofport->cfm) {
2875         int cfm_opup = cfm_get_opup(ofport->cfm);
2876
2877         cfm_enable = !cfm_get_fault(ofport->cfm);
2878
2879         if (cfm_opup >= 0) {
2880             cfm_enable = cfm_enable && cfm_opup;
2881         }
2882     }
2883
2884     if (ofport->bfd) {
2885         bfd_enable = bfd_forwarding(ofport->bfd);
2886     }
2887
2888     if (ofport->bfd || ofport->cfm) {
2889         enable = enable && (cfm_enable || bfd_enable);
2890     }
2891
2892     if (ofport->bundle) {
2893         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2894         if (carrier_changed) {
2895             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2896         }
2897     }
2898
2899     if (ofport->may_enable != enable) {
2900         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2901         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2902     }
2903
2904     ofport->may_enable = enable;
2905 }
2906
2907 static int
2908 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2909                    struct ofproto_port *ofproto_port)
2910 {
2911     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2912     struct dpif_port dpif_port;
2913     int error;
2914
2915     if (sset_contains(&ofproto->ghost_ports, devname)) {
2916         const char *type = netdev_get_type_from_name(devname);
2917
2918         /* We may be called before ofproto->up.port_by_name is populated with
2919          * the appropriate ofport.  For this reason, we must get the name and
2920          * type from the netdev layer directly. */
2921         if (type) {
2922             const struct ofport *ofport;
2923
2924             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2925             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2926             ofproto_port->name = xstrdup(devname);
2927             ofproto_port->type = xstrdup(type);
2928             return 0;
2929         }
2930         return ENODEV;
2931     }
2932
2933     if (!sset_contains(&ofproto->ports, devname)) {
2934         return ENODEV;
2935     }
2936     error = dpif_port_query_by_name(ofproto->backer->dpif,
2937                                     devname, &dpif_port);
2938     if (!error) {
2939         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2940     }
2941     return error;
2942 }
2943
2944 static int
2945 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2946 {
2947     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2948     const char *devname = netdev_get_name(netdev);
2949     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2950     const char *dp_port_name;
2951
2952     if (netdev_vport_is_patch(netdev)) {
2953         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2954         return 0;
2955     }
2956
2957     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2958     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2959         odp_port_t port_no = ODPP_NONE;
2960         int error;
2961
2962         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2963         if (error) {
2964             return error;
2965         }
2966         if (netdev_get_tunnel_config(netdev)) {
2967             simap_put(&ofproto->backer->tnl_backers,
2968                       dp_port_name, odp_to_u32(port_no));
2969         }
2970     }
2971
2972     if (netdev_get_tunnel_config(netdev)) {
2973         sset_add(&ofproto->ghost_ports, devname);
2974     } else {
2975         sset_add(&ofproto->ports, devname);
2976     }
2977     return 0;
2978 }
2979
2980 static int
2981 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
2982 {
2983     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2984     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
2985     int error = 0;
2986
2987     if (!ofport) {
2988         return 0;
2989     }
2990
2991     sset_find_and_delete(&ofproto->ghost_ports,
2992                          netdev_get_name(ofport->up.netdev));
2993     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2994     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
2995         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
2996         if (!error) {
2997             /* The caller is going to close ofport->up.netdev.  If this is a
2998              * bonded port, then the bond is using that netdev, so remove it
2999              * from the bond.  The client will need to reconfigure everything
3000              * after deleting ports, so then the slave will get re-added. */
3001             bundle_remove(&ofport->up);
3002         }
3003     }
3004     return error;
3005 }
3006
3007 static int
3008 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3009 {
3010     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3011     int error;
3012
3013     push_all_stats();
3014
3015     error = netdev_get_stats(ofport->up.netdev, stats);
3016
3017     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3018         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3019
3020         ovs_mutex_lock(&ofproto->stats_mutex);
3021         /* ofproto->stats.tx_packets represents packets that we created
3022          * internally and sent to some port (e.g. packets sent with
3023          * ofproto_dpif_send_packet()).  Account for them as if they had
3024          * come from OFPP_LOCAL and got forwarded. */
3025
3026         if (stats->rx_packets != UINT64_MAX) {
3027             stats->rx_packets += ofproto->stats.tx_packets;
3028         }
3029
3030         if (stats->rx_bytes != UINT64_MAX) {
3031             stats->rx_bytes += ofproto->stats.tx_bytes;
3032         }
3033
3034         /* ofproto->stats.rx_packets represents packets that were received on
3035          * some port and we processed internally and dropped (e.g. STP).
3036          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3037
3038         if (stats->tx_packets != UINT64_MAX) {
3039             stats->tx_packets += ofproto->stats.rx_packets;
3040         }
3041
3042         if (stats->tx_bytes != UINT64_MAX) {
3043             stats->tx_bytes += ofproto->stats.rx_bytes;
3044         }
3045         ovs_mutex_unlock(&ofproto->stats_mutex);
3046     }
3047
3048     return error;
3049 }
3050
3051 struct port_dump_state {
3052     uint32_t bucket;
3053     uint32_t offset;
3054     bool ghost;
3055
3056     struct ofproto_port port;
3057     bool has_port;
3058 };
3059
3060 static int
3061 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3062 {
3063     *statep = xzalloc(sizeof(struct port_dump_state));
3064     return 0;
3065 }
3066
3067 static int
3068 port_dump_next(const struct ofproto *ofproto_, void *state_,
3069                struct ofproto_port *port)
3070 {
3071     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3072     struct port_dump_state *state = state_;
3073     const struct sset *sset;
3074     struct sset_node *node;
3075
3076     if (state->has_port) {
3077         ofproto_port_destroy(&state->port);
3078         state->has_port = false;
3079     }
3080     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3081     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3082         int error;
3083
3084         error = port_query_by_name(ofproto_, node->name, &state->port);
3085         if (!error) {
3086             *port = state->port;
3087             state->has_port = true;
3088             return 0;
3089         } else if (error != ENODEV) {
3090             return error;
3091         }
3092     }
3093
3094     if (!state->ghost) {
3095         state->ghost = true;
3096         state->bucket = 0;
3097         state->offset = 0;
3098         return port_dump_next(ofproto_, state_, port);
3099     }
3100
3101     return EOF;
3102 }
3103
3104 static int
3105 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3106 {
3107     struct port_dump_state *state = state_;
3108
3109     if (state->has_port) {
3110         ofproto_port_destroy(&state->port);
3111     }
3112     free(state);
3113     return 0;
3114 }
3115
3116 static int
3117 port_poll(const struct ofproto *ofproto_, char **devnamep)
3118 {
3119     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3120
3121     if (ofproto->port_poll_errno) {
3122         int error = ofproto->port_poll_errno;
3123         ofproto->port_poll_errno = 0;
3124         return error;
3125     }
3126
3127     if (sset_is_empty(&ofproto->port_poll_set)) {
3128         return EAGAIN;
3129     }
3130
3131     *devnamep = sset_pop(&ofproto->port_poll_set);
3132     return 0;
3133 }
3134
3135 static void
3136 port_poll_wait(const struct ofproto *ofproto_)
3137 {
3138     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3139     dpif_port_poll_wait(ofproto->backer->dpif);
3140 }
3141
3142 static int
3143 port_is_lacp_current(const struct ofport *ofport_)
3144 {
3145     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3146     return (ofport->bundle && ofport->bundle->lacp
3147             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3148             : -1);
3149 }
3150 \f
3151 /* Upcall handling. */
3152
3153 struct flow_miss_op {
3154     struct dpif_op dpif_op;
3155
3156     uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3157     struct xlate_out xout;
3158     bool xout_garbage;           /* 'xout' needs to be uninitialized? */
3159
3160     struct ofpbuf mask;          /* Flow mask for "put" ops. */
3161     struct odputil_keybuf maskbuf;
3162
3163     /* If this is a "put" op, then a pointer to the subfacet that should
3164      * be marked as uninstalled if the operation fails. */
3165     struct subfacet *subfacet;
3166 };
3167
3168 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3169  * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3170  * and (usually) installing a datapath flow.  The answer is usually "yes" (a
3171  * return value of true).  However, for short flows the cost of bookkeeping is
3172  * much higher than the benefits, so when the datapath holds a large number of
3173  * flows we impose some heuristics to decide which flows are likely to be worth
3174  * tracking. */
3175 static bool
3176 flow_miss_should_make_facet(struct flow_miss *miss)
3177 {
3178     struct dpif_backer *backer = miss->ofproto->backer;
3179     uint32_t hash;
3180
3181     switch (flow_miss_model) {
3182     case OFPROTO_HANDLE_MISS_AUTO:
3183         break;
3184     case OFPROTO_HANDLE_MISS_WITH_FACETS:
3185         return true;
3186     case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3187         return false;
3188     }
3189
3190     if (!backer->governor) {
3191         size_t n_subfacets;
3192
3193         n_subfacets = hmap_count(&backer->subfacets);
3194         if (n_subfacets * 2 <= flow_eviction_threshold) {
3195             return true;
3196         }
3197
3198         backer->governor = governor_create();
3199     }
3200
3201     hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3202     return governor_should_install_flow(backer->governor, hash,
3203                                         miss->stats.n_packets);
3204 }
3205
3206 /* Handles 'miss', which matches 'facet'.  May add any required datapath
3207  * operations to 'ops', incrementing '*n_ops' for each new op.
3208  *
3209  * All of the packets in 'miss' are considered to have arrived at time
3210  * 'miss->stats.used'.  This is really important only for new facets: if we
3211  * just called time_msec() here, then the new subfacet or its packets could
3212  * look (occasionally) as though it was used some time after the facet was
3213  * used.  That can make a one-packet flow look like it has a nonzero duration,
3214  * which looks odd in e.g. NetFlow statistics. */
3215 static void
3216 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3217                             struct flow_miss_op *ops, size_t *n_ops)
3218 {
3219     enum subfacet_path want_path;
3220     struct subfacet *subfacet;
3221
3222     facet->packet_count += miss->stats.n_packets;
3223     facet->prev_packet_count += miss->stats.n_packets;
3224     facet->byte_count += miss->stats.n_bytes;
3225     facet->prev_byte_count += miss->stats.n_bytes;
3226
3227     want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3228
3229     /* Don't install the flow if it's the result of the "userspace"
3230      * action for an already installed facet.  This can occur when a
3231      * datapath flow with wildcards has a "userspace" action and flows
3232      * sent to userspace result in a different subfacet, which will then
3233      * be rejected as overlapping by the datapath. */
3234     if (miss->upcall_type == DPIF_UC_ACTION
3235         && !list_is_empty(&facet->subfacets)) {
3236         return;
3237     }
3238
3239     subfacet = subfacet_create(facet, miss);
3240     if (subfacet->path != want_path) {
3241         struct flow_miss_op *op = &ops[(*n_ops)++];
3242         struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3243
3244         subfacet->path = want_path;
3245
3246         ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3247         if (enable_megaflows) {
3248             odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3249                                    &miss->flow, UINT32_MAX);
3250         }
3251
3252         op->xout_garbage = false;
3253         op->dpif_op.type = DPIF_OP_FLOW_PUT;
3254         op->subfacet = subfacet;
3255         put->flags = DPIF_FP_CREATE;
3256         put->key = miss->key;
3257         put->key_len = miss->key_len;
3258         put->mask = op->mask.data;
3259         put->mask_len = op->mask.size;
3260
3261         if (want_path == SF_FAST_PATH) {
3262             put->actions = facet->xout.odp_actions.data;
3263             put->actions_len = facet->xout.odp_actions.size;
3264         } else {
3265             compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3266                               op->slow_stub, sizeof op->slow_stub,
3267                               &put->actions, &put->actions_len);
3268         }
3269         put->stats = NULL;
3270     }
3271 }
3272
3273 /* Handles flow miss 'miss'.  May add any required datapath operations
3274  * to 'ops', incrementing '*n_ops' for each new op. */
3275 static void
3276 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3277                  size_t *n_ops)
3278 {
3279     struct facet *facet;
3280
3281     miss->ofproto->n_missed += miss->stats.n_packets;
3282
3283     facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3284     if (!facet) {
3285         /* There does not exist a bijection between 'struct flow' and datapath
3286          * flow keys with fitness ODP_FIT_TO_LITTLE.  This breaks a fundamental
3287          * assumption used throughout the facet and subfacet handling code.
3288          * Since we have to handle these misses in userspace anyway, we simply
3289          * skip facet creation, avoiding the problem altogether. */
3290         if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3291             || !flow_miss_should_make_facet(miss)) {
3292             return;
3293         }
3294
3295         facet = facet_create(miss);
3296     }
3297     handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3298 }
3299
3300 static struct drop_key *
3301 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3302                 size_t key_len)
3303 {
3304     struct drop_key *drop_key;
3305
3306     HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3307                              &backer->drop_keys) {
3308         if (drop_key->key_len == key_len
3309             && !memcmp(drop_key->key, key, key_len)) {
3310             return drop_key;
3311         }
3312     }
3313     return NULL;
3314 }
3315
3316 static void
3317 drop_key_clear(struct dpif_backer *backer)
3318 {
3319     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3320     struct drop_key *drop_key, *next;
3321
3322     HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3323         int error;
3324
3325         error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3326                               NULL);
3327         if (error && !VLOG_DROP_WARN(&rl)) {
3328             struct ds ds = DS_EMPTY_INITIALIZER;
3329             odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3330             VLOG_WARN("Failed to delete drop key (%s) (%s)",
3331                       ovs_strerror(error), ds_cstr(&ds));
3332             ds_destroy(&ds);
3333         }
3334
3335         hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3336         drop_key_destroy(drop_key);
3337     }
3338
3339     udpif_drop_key_clear(backer->udpif);
3340 }
3341
3342 static void
3343 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3344 {
3345     struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3346     struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3347     struct flow_miss *miss;
3348     size_t n_ops, i;
3349
3350     /* Process each element in the to-do list, constructing the set of
3351      * operations to batch. */
3352     n_ops = 0;
3353     HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3354         handle_flow_miss(miss, flow_miss_ops, &n_ops);
3355     }
3356     ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3357
3358     /* Execute batch. */
3359     for (i = 0; i < n_ops; i++) {
3360         dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3361     }
3362     dpif_operate(backer->dpif, dpif_ops, n_ops);
3363
3364     for (i = 0; i < n_ops; i++) {
3365         if (dpif_ops[i]->error != 0
3366             && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3367             && flow_miss_ops[i].subfacet) {
3368             struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3369
3370             COVERAGE_INC(subfacet_install_fail);
3371
3372             /* Zero-out subfacet counters when installation failed, but
3373              * datapath reported hits.  This should not happen and
3374              * indicates a bug, since if the datapath flow exists, we
3375              * should not be attempting to create a new subfacet.  A
3376              * buggy datapath could trigger this, so just zero out the
3377              * counters and log an error. */
3378             if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3379                 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3380                             "datapath reported hits");
3381                 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3382             }
3383
3384             subfacet->path = SF_NOT_INSTALLED;
3385         }
3386     }
3387 }
3388
3389 static void
3390 handle_upcalls(struct dpif_backer *backer)
3391 {
3392     struct flow_miss_batch *fmb;
3393     int n_processed;
3394
3395     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3396         struct drop_key *drop_key = drop_key_next(backer->udpif);
3397         if (!drop_key) {
3398             break;
3399         }
3400
3401         if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3402             hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3403                         hash_bytes(drop_key->key, drop_key->key_len, 0));
3404             dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3405                           drop_key->key, drop_key->key_len,
3406                           NULL, 0, NULL, 0, NULL);
3407         } else {
3408             drop_key_destroy(drop_key);
3409         }
3410     }
3411
3412     fmb = flow_miss_batch_next(backer->udpif);
3413     if (fmb) {
3414         handle_flow_misses(backer, fmb);
3415         flow_miss_batch_destroy(fmb);
3416     }
3417 }
3418 \f
3419 /* Flow expiration. */
3420
3421 static int subfacet_max_idle(const struct dpif_backer *);
3422 static void update_stats(struct dpif_backer *);
3423 static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex);
3424 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3425
3426 /* This function is called periodically by run().  Its job is to collect
3427  * updates for the flows that have been installed into the datapath, most
3428  * importantly when they last were used, and then use that information to
3429  * expire flows that have not been used recently.
3430  *
3431  * Returns the number of milliseconds after which it should be called again. */
3432 static int
3433 expire(struct dpif_backer *backer)
3434 {
3435     struct ofproto_dpif *ofproto;
3436     size_t n_subfacets;
3437     int max_idle;
3438
3439     /* Periodically clear out the drop keys in an effort to keep them
3440      * relatively few. */
3441     drop_key_clear(backer);
3442
3443     /* Update stats for each flow in the backer. */
3444     update_stats(backer);
3445
3446     n_subfacets = hmap_count(&backer->subfacets);
3447     if (n_subfacets) {
3448         struct subfacet *subfacet;
3449         long long int total, now;
3450
3451         total = 0;
3452         now = time_msec();
3453         HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3454             total += now - subfacet->created;
3455         }
3456         backer->avg_subfacet_life += total / n_subfacets;
3457     }
3458     backer->avg_subfacet_life /= 2;
3459
3460     backer->avg_n_subfacet += n_subfacets;
3461     backer->avg_n_subfacet /= 2;
3462
3463     backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3464
3465     max_idle = subfacet_max_idle(backer);
3466     expire_subfacets(backer, max_idle);
3467
3468     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3469         struct rule *rule, *next_rule;
3470
3471         if (ofproto->backer != backer) {
3472             continue;
3473         }
3474
3475         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3476          * has passed. */
3477         ovs_mutex_lock(&ofproto_mutex);
3478         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3479                             &ofproto->up.expirable) {
3480             rule_expire(rule_dpif_cast(rule));
3481         }
3482         ovs_mutex_unlock(&ofproto_mutex);
3483
3484         /* All outstanding data in existing flows has been accounted, so it's a
3485          * good time to do bond rebalancing. */
3486         if (ofproto->has_bonded_bundles) {
3487             struct ofbundle *bundle;
3488
3489             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3490                 if (bundle->bond) {
3491                     bond_rebalance(bundle->bond);
3492                 }
3493             }
3494         }
3495     }
3496
3497     return MIN(max_idle, 1000);
3498 }
3499
3500 /* Updates flow table statistics given that the datapath just reported 'stats'
3501  * as 'subfacet''s statistics. */
3502 static void
3503 update_subfacet_stats(struct subfacet *subfacet,
3504                       const struct dpif_flow_stats *stats)
3505 {
3506     struct facet *facet = subfacet->facet;
3507     struct dpif_flow_stats diff;
3508
3509     diff.tcp_flags = stats->tcp_flags;
3510     diff.used = stats->used;
3511
3512     if (stats->n_packets >= subfacet->dp_packet_count) {
3513         diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3514     } else {
3515         VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3516         diff.n_packets = 0;
3517     }
3518
3519     if (stats->n_bytes >= subfacet->dp_byte_count) {
3520         diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3521     } else {
3522         VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3523         diff.n_bytes = 0;
3524     }
3525
3526     facet->ofproto->n_hit += diff.n_packets;
3527     subfacet->dp_packet_count = stats->n_packets;
3528     subfacet->dp_byte_count = stats->n_bytes;
3529     subfacet_update_stats(subfacet, &diff);
3530
3531     if (facet->accounted_bytes < facet->byte_count) {
3532         facet_learn(facet);
3533         facet_account(facet);
3534         facet->accounted_bytes = facet->byte_count;
3535     }
3536 }
3537
3538 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3539  * about, or a flow that shouldn't be installed but was anyway.  Delete it. */
3540 static void
3541 delete_unexpected_flow(struct dpif_backer *backer,
3542                        const struct nlattr *key, size_t key_len)
3543 {
3544     if (!VLOG_DROP_WARN(&rl)) {
3545         struct ds s;
3546
3547         ds_init(&s);
3548         odp_flow_key_format(key, key_len, &s);
3549         VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3550         ds_destroy(&s);
3551     }
3552
3553     COVERAGE_INC(facet_unexpected);
3554     dpif_flow_del(backer->dpif, key, key_len, NULL);
3555 }
3556
3557 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3558  *
3559  * This function also pushes statistics updates to rules which each facet
3560  * resubmits into.  Generally these statistics will be accurate.  However, if a
3561  * facet changes the rule it resubmits into at some time in between
3562  * update_stats() runs, it is possible that statistics accrued to the
3563  * old rule will be incorrectly attributed to the new rule.  This could be
3564  * avoided by calling update_stats() whenever rules are created or
3565  * deleted.  However, the performance impact of making so many calls to the
3566  * datapath do not justify the benefit of having perfectly accurate statistics.
3567  *
3568  * In addition, this function maintains per ofproto flow hit counts. The patch
3569  * port is not treated specially. e.g. A packet ingress from br0 patched into
3570  * br1 will increase the hit count of br0 by 1, however, does not affect
3571  * the hit or miss counts of br1.
3572  */
3573 static void
3574 update_stats(struct dpif_backer *backer)
3575 {
3576     const struct dpif_flow_stats *stats;
3577     struct dpif_flow_dump dump;
3578     const struct nlattr *key, *mask;
3579     size_t key_len, mask_len;
3580
3581     dpif_flow_dump_start(&dump, backer->dpif);
3582     while (dpif_flow_dump_next(&dump, &key, &key_len,
3583                                &mask, &mask_len, NULL, NULL, &stats)) {
3584         struct subfacet *subfacet;
3585         uint32_t key_hash;
3586
3587         key_hash = odp_flow_key_hash(key, key_len);
3588         subfacet = subfacet_find(backer, key, key_len, key_hash);
3589         switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3590         case SF_FAST_PATH:
3591             update_subfacet_stats(subfacet, stats);
3592             break;
3593
3594         case SF_SLOW_PATH:
3595             /* Stats are updated per-packet. */
3596             break;
3597
3598         case SF_NOT_INSTALLED:
3599         default:
3600             delete_unexpected_flow(backer, key, key_len);
3601             break;
3602         }
3603         run_fast_rl();
3604     }
3605     dpif_flow_dump_done(&dump);
3606 }
3607
3608 /* Calculates and returns the number of milliseconds of idle time after which
3609  * subfacets should expire from the datapath.  When a subfacet expires, we fold
3610  * its statistics into its facet, and when a facet's last subfacet expires, we
3611  * fold its statistic into its rule. */
3612 static int
3613 subfacet_max_idle(const struct dpif_backer *backer)
3614 {
3615     /*
3616      * Idle time histogram.
3617      *
3618      * Most of the time a switch has a relatively small number of subfacets.
3619      * When this is the case we might as well keep statistics for all of them
3620      * in userspace and to cache them in the kernel datapath for performance as
3621      * well.
3622      *
3623      * As the number of subfacets increases, the memory required to maintain
3624      * statistics about them in userspace and in the kernel becomes
3625      * significant.  However, with a large number of subfacets it is likely
3626      * that only a few of them are "heavy hitters" that consume a large amount
3627      * of bandwidth.  At this point, only heavy hitters are worth caching in
3628      * the kernel and maintaining in userspaces; other subfacets we can
3629      * discard.
3630      *
3631      * The technique used to compute the idle time is to build a histogram with
3632      * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each.  Each subfacet
3633      * that is installed in the kernel gets dropped in the appropriate bucket.
3634      * After the histogram has been built, we compute the cutoff so that only
3635      * the most-recently-used 1% of subfacets (but at least
3636      * flow_eviction_threshold flows) are kept cached.  At least
3637      * the most-recently-used bucket of subfacets is kept, so actually an
3638      * arbitrary number of subfacets can be kept in any given expiration run
3639      * (though the next run will delete most of those unless they receive
3640      * additional data).
3641      *
3642      * This requires a second pass through the subfacets, in addition to the
3643      * pass made by update_stats(), because the former function never looks at
3644      * uninstallable subfacets.
3645      */
3646     enum { BUCKET_WIDTH = 100 };
3647     enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3648     int buckets[N_BUCKETS] = { 0 };
3649     int total, subtotal, bucket;
3650     struct subfacet *subfacet;
3651     long long int now;
3652     int i;
3653
3654     total = hmap_count(&backer->subfacets);
3655     if (total <= flow_eviction_threshold) {
3656         return N_BUCKETS * BUCKET_WIDTH;
3657     }
3658
3659     /* Build histogram. */
3660     now = time_msec();
3661     HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3662         long long int idle = now - subfacet->used;
3663         int bucket = (idle <= 0 ? 0
3664                       : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3665                       : (unsigned int) idle / BUCKET_WIDTH);
3666         buckets[bucket]++;
3667     }
3668
3669     /* Find the first bucket whose flows should be expired. */
3670     subtotal = bucket = 0;
3671     do {
3672         subtotal += buckets[bucket++];
3673     } while (bucket < N_BUCKETS &&
3674              subtotal < MAX(flow_eviction_threshold, total / 100));
3675
3676     if (VLOG_IS_DBG_ENABLED()) {
3677         struct ds s;
3678
3679         ds_init(&s);
3680         ds_put_cstr(&s, "keep");
3681         for (i = 0; i < N_BUCKETS; i++) {
3682             if (i == bucket) {
3683                 ds_put_cstr(&s, ", drop");
3684             }
3685             if (buckets[i]) {
3686                 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3687             }
3688         }
3689         VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3690         ds_destroy(&s);
3691     }
3692
3693     return bucket * BUCKET_WIDTH;
3694 }
3695
3696 static void
3697 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3698 {
3699     /* Cutoff time for most flows. */
3700     long long int normal_cutoff = time_msec() - dp_max_idle;
3701
3702     /* We really want to keep flows for special protocols around, so use a more
3703      * conservative cutoff. */
3704     long long int special_cutoff = time_msec() - 10000;
3705
3706     struct subfacet *subfacet, *next_subfacet;
3707     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3708     int n_batch;
3709
3710     n_batch = 0;
3711     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3712                         &backer->subfacets) {
3713         long long int cutoff;
3714
3715         cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3716                                                 | SLOW_STP)
3717                   ? special_cutoff
3718                   : normal_cutoff);
3719         if (subfacet->used < cutoff) {
3720             if (subfacet->path != SF_NOT_INSTALLED) {
3721                 batch[n_batch++] = subfacet;
3722                 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3723                     subfacet_destroy_batch(backer, batch, n_batch);
3724                     n_batch = 0;
3725                 }
3726             } else {
3727                 subfacet_destroy(subfacet);
3728             }
3729         }
3730     }
3731
3732     if (n_batch > 0) {
3733         subfacet_destroy_batch(backer, batch, n_batch);
3734     }
3735 }
3736
3737 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3738  * then delete it entirely. */
3739 static void
3740 rule_expire(struct rule_dpif *rule)
3741     OVS_REQUIRES(ofproto_mutex)
3742 {
3743     uint16_t idle_timeout, hard_timeout;
3744     long long int now = time_msec();
3745     int reason;
3746
3747     ovs_assert(!rule->up.pending);
3748
3749     /* Has 'rule' expired? */
3750     ovs_mutex_lock(&rule->up.mutex);
3751     hard_timeout = rule->up.hard_timeout;
3752     idle_timeout = rule->up.idle_timeout;
3753     if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3754         reason = OFPRR_HARD_TIMEOUT;
3755     } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3756         reason = OFPRR_IDLE_TIMEOUT;
3757     } else {
3758         reason = -1;
3759     }
3760     ovs_mutex_unlock(&rule->up.mutex);
3761
3762     if (reason >= 0) {
3763         COVERAGE_INC(ofproto_dpif_expired);
3764         ofproto_rule_expire(&rule->up, reason);
3765     }
3766 }
3767 \f
3768 /* Facets. */
3769
3770 /* Creates and returns a new facet based on 'miss'.
3771  *
3772  * The caller must already have determined that no facet with an identical
3773  * 'miss->flow' exists in 'miss->ofproto'.
3774  *
3775  * 'rule' and 'xout' must have been created based on 'miss'.
3776  *
3777  * 'facet'' statistics are initialized based on 'stats'.
3778  *
3779  * The facet will initially have no subfacets.  The caller should create (at
3780  * least) one subfacet with subfacet_create(). */
3781 static struct facet *
3782 facet_create(const struct flow_miss *miss)
3783 {
3784     struct ofproto_dpif *ofproto = miss->ofproto;
3785     struct facet *facet;
3786     struct match match;
3787
3788     COVERAGE_INC(facet_create);
3789     facet = xzalloc(sizeof *facet);
3790     facet->ofproto = miss->ofproto;
3791     facet->used = miss->stats.used;
3792     facet->flow = miss->flow;
3793     facet->learn_rl = time_msec() + 500;
3794
3795     list_init(&facet->subfacets);
3796     netflow_flow_init(&facet->nf_flow);
3797     netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3798
3799     xlate_out_copy(&facet->xout, &miss->xout);
3800
3801     match_init(&match, &facet->flow, &facet->xout.wc);
3802     cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3803     ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3804     classifier_insert(&ofproto->facets, &facet->cr);
3805     ovs_rwlock_unlock(&ofproto->facets.rwlock);
3806
3807     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
3808     return facet;
3809 }
3810
3811 static void
3812 facet_free(struct facet *facet)
3813 {
3814     if (facet) {
3815         xlate_out_uninit(&facet->xout);
3816         free(facet);
3817     }
3818 }
3819
3820 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3821  * 'flow' must reflect the data in 'packet'. */
3822 int
3823 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3824                              const struct flow *flow,
3825                              struct rule_dpif *rule,
3826                              const struct ofpact *ofpacts, size_t ofpacts_len,
3827                              struct ofpbuf *packet)
3828 {
3829     struct odputil_keybuf keybuf;
3830     struct dpif_flow_stats stats;
3831     struct xlate_out xout;
3832     struct xlate_in xin;
3833     ofp_port_t in_port;
3834     struct ofpbuf key;
3835     int error;
3836
3837     ovs_assert((rule != NULL) != (ofpacts != NULL));
3838
3839     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3840     if (rule) {
3841         rule_dpif_credit_stats(rule, &stats);
3842     }
3843
3844     xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
3845     xin.ofpacts = ofpacts;
3846     xin.ofpacts_len = ofpacts_len;
3847     xin.resubmit_stats = &stats;
3848     xlate_actions(&xin, &xout);
3849
3850     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3851     in_port = flow->in_port.ofp_port;
3852     if (in_port == OFPP_NONE) {
3853         in_port = OFPP_LOCAL;
3854     }
3855     odp_flow_key_from_flow(&key, flow, ofp_port_to_odp_port(ofproto, in_port));
3856
3857     error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
3858                          xout.odp_actions.data, xout.odp_actions.size, packet,
3859                          (xout.slow & SLOW_ACTION) != 0);
3860     xlate_out_uninit(&xout);
3861
3862     return error;
3863 }
3864
3865 /* Remove 'facet' from its ofproto and free up the associated memory:
3866  *
3867  *   - If 'facet' was installed in the datapath, uninstalls it and updates its
3868  *     rule's statistics, via subfacet_uninstall().
3869  *
3870  *   - Removes 'facet' from its rule and from ofproto->facets.
3871  */
3872 static void
3873 facet_remove(struct facet *facet)
3874 {
3875     struct subfacet *subfacet, *next_subfacet;
3876
3877     COVERAGE_INC(facet_remove);
3878     ovs_assert(!list_is_empty(&facet->subfacets));
3879
3880     /* First uninstall all of the subfacets to get final statistics. */
3881     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3882         subfacet_uninstall(subfacet);
3883     }
3884
3885     /* Flush the final stats to the rule.
3886      *
3887      * This might require us to have at least one subfacet around so that we
3888      * can use its actions for accounting in facet_account(), which is why we
3889      * have uninstalled but not yet destroyed the subfacets. */
3890     facet_flush_stats(facet);
3891
3892     /* Now we're really all done so destroy everything. */
3893     LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3894                         &facet->subfacets) {
3895         subfacet_destroy__(subfacet);
3896     }
3897     ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
3898     classifier_remove(&facet->ofproto->facets, &facet->cr);
3899     ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
3900     cls_rule_destroy(&facet->cr);
3901     facet_free(facet);
3902 }
3903
3904 /* Feed information from 'facet' back into the learning table to keep it in
3905  * sync with what is actually flowing through the datapath. */
3906 static void
3907 facet_learn(struct facet *facet)
3908 {
3909     long long int now = time_msec();
3910
3911     if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
3912         return;
3913     }
3914
3915     facet->learn_rl = now + 500;
3916
3917     if (!facet->xout.has_learn
3918         && !facet->xout.has_normal
3919         && (!facet->xout.has_fin_timeout
3920             || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3921         return;
3922     }
3923
3924     facet_push_stats(facet, true);
3925 }
3926
3927 static void
3928 facet_account(struct facet *facet)
3929 {
3930     const struct nlattr *a;
3931     unsigned int left;
3932     ovs_be16 vlan_tci;
3933     uint64_t n_bytes;
3934
3935     if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
3936         return;
3937     }
3938     n_bytes = facet->byte_count - facet->accounted_bytes;
3939
3940     /* This loop feeds byte counters to bond_account() for rebalancing to use
3941      * as a basis.  We also need to track the actual VLAN on which the packet
3942      * is going to be sent to ensure that it matches the one passed to
3943      * bond_choose_output_slave().  (Otherwise, we will account to the wrong
3944      * hash bucket.)
3945      *
3946      * We use the actions from an arbitrary subfacet because they should all
3947      * be equally valid for our purpose. */
3948     vlan_tci = facet->flow.vlan_tci;
3949     NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
3950                              facet->xout.odp_actions.size) {
3951         const struct ovs_action_push_vlan *vlan;
3952         struct ofport_dpif *port;
3953
3954         switch (nl_attr_type(a)) {
3955         case OVS_ACTION_ATTR_OUTPUT:
3956             port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
3957             if (port && port->bundle && port->bundle->bond) {
3958                 bond_account(port->bundle->bond, &facet->flow,
3959                              vlan_tci_to_vid(vlan_tci), n_bytes);
3960             }
3961             break;
3962
3963         case OVS_ACTION_ATTR_POP_VLAN:
3964             vlan_tci = htons(0);
3965             break;
3966
3967         case OVS_ACTION_ATTR_PUSH_VLAN:
3968             vlan = nl_attr_get(a);
3969             vlan_tci = vlan->vlan_tci;
3970             break;
3971         }
3972     }
3973 }
3974
3975 /* Returns true if the only action for 'facet' is to send to the controller.
3976  * (We don't report NetFlow expiration messages for such facets because they
3977  * are just part of the control logic for the network, not real traffic). */
3978 static bool
3979 facet_is_controller_flow(struct facet *facet)
3980 {
3981     if (facet) {
3982         struct ofproto_dpif *ofproto = facet->ofproto;
3983         const struct ofpact *ofpacts;
3984         struct rule_actions *actions;
3985         struct rule_dpif *rule;
3986         size_t ofpacts_len;
3987         bool is_controller;
3988
3989         rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
3990         actions = rule_dpif_get_actions(rule);
3991         rule_dpif_unref(rule);
3992
3993         ofpacts_len = actions->ofpacts_len;
3994         ofpacts = actions->ofpacts;
3995         is_controller = ofpacts_len > 0
3996             && ofpacts->type == OFPACT_CONTROLLER
3997             && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
3998         rule_actions_unref(actions);
3999
4000         return is_controller;
4001     }
4002     return false;
4003 }
4004
4005 /* Folds all of 'facet''s statistics into its rule.  Also updates the
4006  * accounting ofhook and emits a NetFlow expiration if appropriate.  All of
4007  * 'facet''s statistics in the datapath should have been zeroed and folded into
4008  * its packet and byte counts before this function is called. */
4009 static void
4010 facet_flush_stats(struct facet *facet)
4011 {
4012     struct ofproto_dpif *ofproto = facet->ofproto;
4013     struct subfacet *subfacet;
4014
4015     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4016         ovs_assert(!subfacet->dp_byte_count);
4017         ovs_assert(!subfacet->dp_packet_count);
4018     }
4019
4020     facet_push_stats(facet, false);
4021     if (facet->accounted_bytes < facet->byte_count) {
4022         facet_account(facet);
4023         facet->accounted_bytes = facet->byte_count;
4024     }
4025
4026     if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4027         struct ofexpired expired;
4028         expired.flow = facet->flow;
4029         expired.packet_count = facet->packet_count;
4030         expired.byte_count = facet->byte_count;
4031         expired.used = facet->used;
4032         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4033     }
4034
4035     /* Reset counters to prevent double counting if 'facet' ever gets
4036      * reinstalled. */
4037     facet_reset_counters(facet);
4038
4039     netflow_flow_clear(&facet->nf_flow);
4040     facet->tcp_flags = 0;
4041 }
4042
4043 /* Searches 'ofproto''s table of facets for one which would be responsible for
4044  * 'flow'.  Returns it if found, otherwise a null pointer.
4045  *
4046  * The returned facet might need revalidation; use facet_lookup_valid()
4047  * instead if that is important. */
4048 static struct facet *
4049 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4050 {
4051     struct cls_rule *cr;
4052
4053     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4054     cr = classifier_lookup(&ofproto->facets, flow, NULL);
4055     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4056     return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4057 }
4058
4059 /* Searches 'ofproto''s table of facets for one capable that covers
4060  * 'flow'.  Returns it if found, otherwise a null pointer.
4061  *
4062  * The returned facet is guaranteed to be valid. */
4063 static struct facet *
4064 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4065 {
4066     struct facet *facet;
4067
4068     facet = facet_find(ofproto, flow);
4069     if (facet
4070         && ofproto->backer->need_revalidate
4071         && !facet_revalidate(facet)) {
4072         return NULL;
4073     }
4074
4075     return facet;
4076 }
4077
4078 static bool
4079 facet_check_consistency(struct facet *facet)
4080 {
4081     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4082
4083     struct xlate_out xout;
4084     struct xlate_in xin;
4085     bool ok;
4086
4087     /* Check the datapath actions for consistency. */
4088     xlate_in_init(&xin, facet->ofproto, &facet->flow, NULL, 0, NULL);
4089     xlate_actions(&xin, &xout);
4090
4091     ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4092         && facet->xout.slow == xout.slow;
4093     if (!ok && !VLOG_DROP_WARN(&rl)) {
4094         struct ds s = DS_EMPTY_INITIALIZER;
4095
4096         flow_format(&s, &facet->flow);
4097         ds_put_cstr(&s, ": inconsistency in facet");
4098
4099         if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4100             ds_put_cstr(&s, " (actions were: ");
4101             format_odp_actions(&s, facet->xout.odp_actions.data,
4102                                facet->xout.odp_actions.size);
4103             ds_put_cstr(&s, ") (correct actions: ");
4104             format_odp_actions(&s, xout.odp_actions.data,
4105                                xout.odp_actions.size);
4106             ds_put_char(&s, ')');
4107         }
4108
4109         if (facet->xout.slow != xout.slow) {
4110             ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4111         }
4112
4113         ds_destroy(&s);
4114     }
4115     xlate_out_uninit(&xout);
4116
4117     return ok;
4118 }
4119
4120 /* Re-searches the classifier for 'facet':
4121  *
4122  *   - If the rule found is different from 'facet''s current rule, moves
4123  *     'facet' to the new rule and recompiles its actions.
4124  *
4125  *   - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4126  *     where it is and recompiles its actions anyway.
4127  *
4128  *   - If any of 'facet''s subfacets correspond to a new flow according to
4129  *     xlate_receive(), 'facet' is removed.
4130  *
4131  *   Returns true if 'facet' is still valid.  False if 'facet' was removed. */
4132 static bool
4133 facet_revalidate(struct facet *facet)
4134 {
4135     struct ofproto_dpif *ofproto = facet->ofproto;
4136     struct rule_dpif *new_rule;
4137     struct subfacet *subfacet;
4138     struct flow_wildcards wc;
4139     struct xlate_out xout;
4140     struct xlate_in xin;
4141
4142     COVERAGE_INC(facet_revalidate);
4143
4144     /* Check that child subfacets still correspond to this facet.  Tunnel
4145      * configuration changes could cause a subfacet's OpenFlow in_port to
4146      * change. */
4147     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4148         struct ofproto_dpif *recv_ofproto;
4149         struct flow recv_flow;
4150         int error;
4151
4152         error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4153                               subfacet->key_len, &recv_flow, NULL,
4154                               &recv_ofproto, NULL);
4155         if (error
4156             || recv_ofproto != ofproto
4157             || facet != facet_find(ofproto, &recv_flow)) {
4158             facet_remove(facet);
4159             return false;
4160         }
4161     }
4162
4163     flow_wildcards_init_catchall(&wc);
4164     rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4165
4166     /* Calculate new datapath actions.
4167      *
4168      * We do not modify any 'facet' state yet, because we might need to, e.g.,
4169      * emit a NetFlow expiration and, if so, we need to have the old state
4170      * around to properly compose it. */
4171     xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4172     xlate_actions(&xin, &xout);
4173     flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4174
4175     /* A facet's slow path reason should only change under dramatic
4176      * circumstances.  Rather than try to update everything, it's simpler to
4177      * remove the facet and start over.
4178      *
4179      * More importantly, if a facet's wildcards change, it will be relatively
4180      * difficult to figure out if its subfacets still belong to it, and if not
4181      * which facet they may belong to.  Again, to avoid the complexity, we
4182      * simply give up instead. */
4183     if (facet->xout.slow != xout.slow
4184         || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4185         facet_remove(facet);
4186         xlate_out_uninit(&xout);
4187         rule_dpif_unref(new_rule);
4188         return false;
4189     }
4190
4191     if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4192         LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4193             if (subfacet->path == SF_FAST_PATH) {
4194                 struct dpif_flow_stats stats;
4195
4196                 subfacet_install(subfacet, &xout.odp_actions, &stats);
4197                 subfacet_update_stats(subfacet, &stats);
4198             }
4199         }
4200
4201         facet_flush_stats(facet);
4202
4203         ofpbuf_clear(&facet->xout.odp_actions);
4204         ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4205                    xout.odp_actions.size);
4206     }
4207
4208     /* Update 'facet' now that we've taken care of all the old state. */
4209     facet->xout.slow = xout.slow;
4210     facet->xout.has_learn = xout.has_learn;
4211     facet->xout.has_normal = xout.has_normal;
4212     facet->xout.has_fin_timeout = xout.has_fin_timeout;
4213     facet->xout.nf_output_iface = xout.nf_output_iface;
4214     facet->xout.mirrors = xout.mirrors;
4215     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4216
4217     ovs_mutex_lock(&new_rule->up.mutex);
4218     facet->used = MAX(facet->used, new_rule->up.created);
4219     ovs_mutex_unlock(&new_rule->up.mutex);
4220
4221     xlate_out_uninit(&xout);
4222     rule_dpif_unref(new_rule);
4223     return true;
4224 }
4225
4226 static void
4227 facet_reset_counters(struct facet *facet)
4228 {
4229     facet->packet_count = 0;
4230     facet->byte_count = 0;
4231     facet->prev_packet_count = 0;
4232     facet->prev_byte_count = 0;
4233     facet->accounted_bytes = 0;
4234 }
4235
4236 static void
4237 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4238                 struct dpif_flow_stats *stats, bool may_learn)
4239 {
4240     struct ofport_dpif *in_port;
4241     struct xlate_in xin;
4242
4243     in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4244     if (in_port && in_port->is_tunnel) {
4245         netdev_vport_inc_rx(in_port->up.netdev, stats);
4246     }
4247
4248     xlate_in_init(&xin, ofproto, flow, NULL, stats->tcp_flags, NULL);
4249     xin.resubmit_stats = stats;
4250     xin.may_learn = may_learn;
4251     xlate_actions_for_side_effects(&xin);
4252 }
4253
4254 static void
4255 facet_push_stats(struct facet *facet, bool may_learn)
4256 {
4257     struct dpif_flow_stats stats;
4258
4259     ovs_assert(facet->packet_count >= facet->prev_packet_count);
4260     ovs_assert(facet->byte_count >= facet->prev_byte_count);
4261     ovs_assert(facet->used >= facet->prev_used);
4262
4263     stats.n_packets = facet->packet_count - facet->prev_packet_count;
4264     stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4265     stats.used = facet->used;
4266     stats.tcp_flags = facet->tcp_flags;
4267
4268     if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4269         facet->prev_packet_count = facet->packet_count;
4270         facet->prev_byte_count = facet->byte_count;
4271         facet->prev_used = facet->used;
4272
4273         netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4274                                  facet->used);
4275         netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4276         mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4277                             stats.n_packets, stats.n_bytes);
4278         flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4279     }
4280 }
4281
4282 static void
4283 push_all_stats__(bool run_fast)
4284 {
4285     static long long int rl = LLONG_MIN;
4286     struct ofproto_dpif *ofproto;
4287
4288     if (time_msec() < rl) {
4289         return;
4290     }
4291
4292     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4293         struct cls_cursor cursor;
4294         struct facet *facet;
4295
4296         ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4297         cls_cursor_init(&cursor, &ofproto->facets, NULL);
4298         CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4299             facet_push_stats(facet, false);
4300             if (run_fast) {
4301                 run_fast_rl();
4302             }
4303         }
4304         ovs_rwlock_unlock(&ofproto->facets.rwlock);
4305     }
4306
4307     rl = time_msec() + 100;
4308 }
4309
4310 static void
4311 push_all_stats(void)
4312 {
4313     push_all_stats__(true);
4314 }
4315
4316 void
4317 rule_dpif_credit_stats(struct rule_dpif *rule,
4318                        const struct dpif_flow_stats *stats)
4319 {
4320     ovs_mutex_lock(&rule->stats_mutex);
4321     rule->packet_count += stats->n_packets;
4322     rule->byte_count += stats->n_bytes;
4323     rule->up.used = MAX(rule->up.used, stats->used);
4324     ovs_mutex_unlock(&rule->stats_mutex);
4325 }
4326
4327 bool
4328 rule_dpif_is_fail_open(const struct rule_dpif *rule)
4329 {
4330     return is_fail_open_rule(&rule->up);
4331 }
4332
4333 bool
4334 rule_dpif_is_table_miss(const struct rule_dpif *rule)
4335 {
4336     return rule_is_table_miss(&rule->up);
4337 }
4338
4339 ovs_be64
4340 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4341     OVS_REQUIRES(rule->up.mutex)
4342 {
4343     return rule->up.flow_cookie;
4344 }
4345
4346 void
4347 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4348                      uint16_t hard_timeout)
4349 {
4350     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4351 }
4352
4353 /* Returns 'rule''s actions.  The caller owns a reference on the returned
4354  * actions and must eventually release it (with rule_actions_unref()) to avoid
4355  * a memory leak. */
4356 struct rule_actions *
4357 rule_dpif_get_actions(const struct rule_dpif *rule)
4358 {
4359     return rule_get_actions(&rule->up);
4360 }
4361 \f
4362 /* Subfacets. */
4363
4364 static struct subfacet *
4365 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4366               size_t key_len, uint32_t key_hash)
4367 {
4368     struct subfacet *subfacet;
4369
4370     HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4371                              &backer->subfacets) {
4372         if (subfacet->key_len == key_len
4373             && !memcmp(key, subfacet->key, key_len)) {
4374             return subfacet;
4375         }
4376     }
4377
4378     return NULL;
4379 }
4380
4381 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4382  * 'key_fitness', 'key', and 'key_len' members in 'miss'.  Returns the
4383  * existing subfacet if there is one, otherwise creates and returns a
4384  * new subfacet. */
4385 static struct subfacet *
4386 subfacet_create(struct facet *facet, struct flow_miss *miss)
4387 {
4388     struct dpif_backer *backer = miss->ofproto->backer;
4389     const struct nlattr *key = miss->key;
4390     size_t key_len = miss->key_len;
4391     uint32_t key_hash;
4392     struct subfacet *subfacet;
4393
4394     key_hash = odp_flow_key_hash(key, key_len);
4395
4396     if (list_is_empty(&facet->subfacets)) {
4397         subfacet = &facet->one_subfacet;
4398     } else {
4399         subfacet = subfacet_find(backer, key, key_len, key_hash);
4400         if (subfacet) {
4401             if (subfacet->facet == facet) {
4402                 return subfacet;
4403             }
4404
4405             /* This shouldn't happen. */
4406             VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4407             subfacet_destroy(subfacet);
4408         }
4409
4410         subfacet = xmalloc(sizeof *subfacet);
4411     }
4412
4413     COVERAGE_INC(subfacet_create);
4414     hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4415     list_push_back(&facet->subfacets, &subfacet->list_node);
4416     subfacet->facet = facet;
4417     subfacet->key = xmemdup(key, key_len);
4418     subfacet->key_len = key_len;
4419     subfacet->used = miss->stats.used;
4420     subfacet->created = subfacet->used;
4421     subfacet->dp_packet_count = 0;
4422     subfacet->dp_byte_count = 0;
4423     subfacet->path = SF_NOT_INSTALLED;
4424     subfacet->backer = backer;
4425
4426     return subfacet;
4427 }
4428
4429 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4430  * its facet within 'ofproto', and frees it. */
4431 static void
4432 subfacet_destroy__(struct subfacet *subfacet)
4433 {
4434     struct facet *facet = subfacet->facet;
4435
4436     COVERAGE_INC(subfacet_destroy);
4437     subfacet_uninstall(subfacet);
4438     hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4439     list_remove(&subfacet->list_node);
4440     free(subfacet->key);
4441     if (subfacet != &facet->one_subfacet) {
4442         free(subfacet);
4443     }
4444 }
4445
4446 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4447  * last remaining subfacet in its facet destroys the facet too. */
4448 static void
4449 subfacet_destroy(struct subfacet *subfacet)
4450 {
4451     struct facet *facet = subfacet->facet;
4452
4453     if (list_is_singleton(&facet->subfacets)) {
4454         /* facet_remove() needs at least one subfacet (it will remove it). */
4455         facet_remove(facet);
4456     } else {
4457         subfacet_destroy__(subfacet);
4458     }
4459 }
4460
4461 static void
4462 subfacet_destroy_batch(struct dpif_backer *backer,
4463                        struct subfacet **subfacets, int n)
4464 {
4465     struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4466     struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4467     struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4468     int i;
4469
4470     for (i = 0; i < n; i++) {
4471         ops[i].type = DPIF_OP_FLOW_DEL;
4472         ops[i].u.flow_del.key = subfacets[i]->key;
4473         ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4474         ops[i].u.flow_del.stats = &stats[i];
4475         opsp[i] = &ops[i];
4476     }
4477
4478     dpif_operate(backer->dpif, opsp, n);
4479     for (i = 0; i < n; i++) {
4480         subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4481         subfacets[i]->path = SF_NOT_INSTALLED;
4482         subfacet_destroy(subfacets[i]);
4483         run_fast_rl();
4484     }
4485 }
4486
4487 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4488  * bytes of actions in 'actions'.  If 'stats' is non-null, statistics counters
4489  * in the datapath will be zeroed and 'stats' will be updated with traffic new
4490  * since 'subfacet' was last updated.
4491  *
4492  * Returns 0 if successful, otherwise a positive errno value. */
4493 static int
4494 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4495                  struct dpif_flow_stats *stats)
4496 {
4497     struct facet *facet = subfacet->facet;
4498     enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4499     const struct nlattr *actions = odp_actions->data;
4500     size_t actions_len = odp_actions->size;
4501     struct odputil_keybuf maskbuf;
4502     struct ofpbuf mask;
4503
4504     uint64_t slow_path_stub[128 / 8];
4505     enum dpif_flow_put_flags flags;
4506     int ret;
4507
4508     flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4509                                                : DPIF_FP_MODIFY;
4510     if (stats) {
4511         flags |= DPIF_FP_ZERO_STATS;
4512     }
4513
4514     if (path == SF_SLOW_PATH) {
4515         compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4516                           slow_path_stub, sizeof slow_path_stub,
4517                           &actions, &actions_len);
4518     }
4519
4520     ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4521     if (enable_megaflows) {
4522         odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4523                                &facet->flow, UINT32_MAX);
4524     }
4525
4526     ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4527                         subfacet->key_len,  mask.data, mask.size,
4528                         actions, actions_len, stats);
4529
4530     if (stats) {
4531         subfacet_reset_dp_stats(subfacet, stats);
4532     }
4533
4534     if (ret) {
4535         COVERAGE_INC(subfacet_install_fail);
4536     } else {
4537         subfacet->path = path;
4538     }
4539     return ret;
4540 }
4541
4542 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4543 static void
4544 subfacet_uninstall(struct subfacet *subfacet)
4545 {
4546     if (subfacet->path != SF_NOT_INSTALLED) {
4547         struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4548         struct dpif_flow_stats stats;
4549         int error;
4550
4551         error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4552                               subfacet->key_len, &stats);
4553         subfacet_reset_dp_stats(subfacet, &stats);
4554         if (!error) {
4555             subfacet_update_stats(subfacet, &stats);
4556         }
4557         subfacet->path = SF_NOT_INSTALLED;
4558     } else {
4559         ovs_assert(subfacet->dp_packet_count == 0);
4560         ovs_assert(subfacet->dp_byte_count == 0);
4561     }
4562 }
4563
4564 /* Resets 'subfacet''s datapath statistics counters.  This should be called
4565  * when 'subfacet''s statistics are cleared in the datapath.  If 'stats' is
4566  * non-null, it should contain the statistics returned by dpif when 'subfacet'
4567  * was reset in the datapath.  'stats' will be modified to include only
4568  * statistics new since 'subfacet' was last updated. */
4569 static void
4570 subfacet_reset_dp_stats(struct subfacet *subfacet,
4571                         struct dpif_flow_stats *stats)
4572 {
4573     if (stats
4574         && subfacet->dp_packet_count <= stats->n_packets
4575         && subfacet->dp_byte_count <= stats->n_bytes) {
4576         stats->n_packets -= subfacet->dp_packet_count;
4577         stats->n_bytes -= subfacet->dp_byte_count;
4578     }
4579
4580     subfacet->dp_packet_count = 0;
4581     subfacet->dp_byte_count = 0;
4582 }
4583
4584 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4585  *
4586  * Because of the meaning of a subfacet's counters, it only makes sense to do
4587  * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4588  * represents a packet that was sent by hand or if it represents statistics
4589  * that have been cleared out of the datapath. */
4590 static void
4591 subfacet_update_stats(struct subfacet *subfacet,
4592                       const struct dpif_flow_stats *stats)
4593 {
4594     if (stats->n_packets || stats->used > subfacet->used) {
4595         struct facet *facet = subfacet->facet;
4596
4597         subfacet->used = MAX(subfacet->used, stats->used);
4598         facet->used = MAX(facet->used, stats->used);
4599         facet->packet_count += stats->n_packets;
4600         facet->byte_count += stats->n_bytes;
4601         facet->tcp_flags |= stats->tcp_flags;
4602     }
4603 }
4604 \f
4605 /* Rules. */
4606
4607 /* Lookup 'flow' in 'ofproto''s classifier.  If 'wc' is non-null, sets
4608  * the fields that were relevant as part of the lookup. */
4609 void
4610 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4611                  struct flow_wildcards *wc, struct rule_dpif **rule)
4612 {
4613     struct ofport_dpif *port;
4614
4615     if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4616         return;
4617     }
4618     port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4619     if (!port) {
4620         VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4621                      flow->in_port.ofp_port);
4622     }
4623
4624     choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4625                      ofproto->no_packet_in_rule, rule);
4626 }
4627
4628 bool
4629 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4630                           const struct flow *flow, struct flow_wildcards *wc,
4631                           uint8_t table_id, struct rule_dpif **rule)
4632 {
4633     const struct cls_rule *cls_rule;
4634     struct classifier *cls;
4635     bool frag;
4636
4637     *rule = NULL;
4638     if (table_id >= N_TABLES) {
4639         return false;
4640     }
4641
4642     if (wc) {
4643         memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4644         wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4645     }
4646
4647     cls = &ofproto->up.tables[table_id].cls;
4648     ovs_rwlock_rdlock(&cls->rwlock);
4649     frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4650     if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4651         /* We must pretend that transport ports are unavailable. */
4652         struct flow ofpc_normal_flow = *flow;
4653         ofpc_normal_flow.tp_src = htons(0);
4654         ofpc_normal_flow.tp_dst = htons(0);
4655         cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4656     } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4657         cls_rule = &ofproto->drop_frags_rule->up.cr;
4658         /* Frag mask in wc already set above. */
4659     } else {
4660         cls_rule = classifier_lookup(cls, flow, wc);
4661     }
4662
4663     *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4664     rule_dpif_ref(*rule);
4665     ovs_rwlock_unlock(&cls->rwlock);
4666
4667     return *rule != NULL;
4668 }
4669
4670 /* Given a port configuration (specified as zero if there's no port), chooses
4671  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4672  * flow table miss. */
4673 void
4674 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4675                  struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4676 {
4677     *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4678     rule_dpif_ref(*rule);
4679 }
4680
4681 void
4682 rule_dpif_ref(struct rule_dpif *rule)
4683 {
4684     if (rule) {
4685         ofproto_rule_ref(&rule->up);
4686     }
4687 }
4688
4689 void
4690 rule_dpif_unref(struct rule_dpif *rule)
4691 {
4692     if (rule) {
4693         ofproto_rule_unref(&rule->up);
4694     }
4695 }
4696
4697 static void
4698 complete_operation(struct rule_dpif *rule)
4699     OVS_REQUIRES(ofproto_mutex)
4700 {
4701     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4702
4703     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4704     ofoperation_complete(rule->up.pending, 0);
4705 }
4706
4707 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4708 {
4709     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4710 }
4711
4712 static struct rule *
4713 rule_alloc(void)
4714 {
4715     struct rule_dpif *rule = xmalloc(sizeof *rule);
4716     return &rule->up;
4717 }
4718
4719 static void
4720 rule_dealloc(struct rule *rule_)
4721 {
4722     struct rule_dpif *rule = rule_dpif_cast(rule_);
4723     free(rule);
4724 }
4725
4726 static enum ofperr
4727 rule_construct(struct rule *rule_)
4728 {
4729     struct rule_dpif *rule = rule_dpif_cast(rule_);
4730     ovs_mutex_init(&rule->stats_mutex);
4731     ovs_mutex_lock(&rule->stats_mutex);
4732     rule->packet_count = 0;
4733     rule->byte_count = 0;
4734     ovs_mutex_unlock(&rule->stats_mutex);
4735     return 0;
4736 }
4737
4738 static void
4739 rule_insert(struct rule *rule_)
4740     OVS_REQUIRES(ofproto_mutex)
4741 {
4742     struct rule_dpif *rule = rule_dpif_cast(rule_);
4743     complete_operation(rule);
4744 }
4745
4746 static void
4747 rule_delete(struct rule *rule_)
4748     OVS_REQUIRES(ofproto_mutex)
4749 {
4750     struct rule_dpif *rule = rule_dpif_cast(rule_);
4751     complete_operation(rule);
4752 }
4753
4754 static void
4755 rule_destruct(struct rule *rule_)
4756 {
4757     struct rule_dpif *rule = rule_dpif_cast(rule_);
4758     ovs_mutex_destroy(&rule->stats_mutex);
4759 }
4760
4761 static void
4762 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4763 {
4764     struct rule_dpif *rule = rule_dpif_cast(rule_);
4765
4766     /* push_all_stats() can handle flow misses which, when using the learn
4767      * action, can cause rules to be added and deleted.  This can corrupt our
4768      * caller's datastructures which assume that rule_get_stats() doesn't have
4769      * an impact on the flow table. To be safe, we disable miss handling. */
4770     push_all_stats__(false);
4771
4772     /* Start from historical data for 'rule' itself that are no longer tracked
4773      * in facets.  This counts, for example, facets that have expired. */
4774     ovs_mutex_lock(&rule->stats_mutex);
4775     *packets = rule->packet_count;
4776     *bytes = rule->byte_count;
4777     ovs_mutex_unlock(&rule->stats_mutex);
4778 }
4779
4780 static void
4781 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4782                   struct ofpbuf *packet)
4783 {
4784     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4785
4786     ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
4787 }
4788
4789 static enum ofperr
4790 rule_execute(struct rule *rule, const struct flow *flow,
4791              struct ofpbuf *packet)
4792 {
4793     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4794     ofpbuf_delete(packet);
4795     return 0;
4796 }
4797
4798 static void
4799 rule_modify_actions(struct rule *rule_, bool reset_counters)
4800     OVS_REQUIRES(ofproto_mutex)
4801 {
4802     struct rule_dpif *rule = rule_dpif_cast(rule_);
4803
4804     if (reset_counters) {
4805         ovs_mutex_lock(&rule->stats_mutex);
4806         rule->packet_count = 0;
4807         rule->byte_count = 0;
4808         ovs_mutex_unlock(&rule->stats_mutex);
4809     }
4810
4811     complete_operation(rule);
4812 }
4813
4814 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4815 {
4816     return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4817 }
4818
4819 static struct ofgroup *
4820 group_alloc(void)
4821 {
4822     struct group_dpif *group = xzalloc(sizeof *group);
4823     return &group->up;
4824 }
4825
4826 static void
4827 group_dealloc(struct ofgroup *group_)
4828 {
4829     struct group_dpif *group = group_dpif_cast(group_);
4830     free(group);
4831 }
4832
4833 static void
4834 group_construct_stats(struct group_dpif *group)
4835     OVS_REQUIRES(group->stats_mutex)
4836 {
4837     group->packet_count = 0;
4838     group->byte_count = 0;
4839     if (!group->bucket_stats) {
4840         group->bucket_stats = xcalloc(group->up.n_buckets,
4841                                       sizeof *group->bucket_stats);
4842     } else {
4843         memset(group->bucket_stats, 0, group->up.n_buckets *
4844                sizeof *group->bucket_stats);
4845     }
4846 }
4847
4848 static enum ofperr
4849 group_construct(struct ofgroup *group_)
4850 {
4851     struct group_dpif *group = group_dpif_cast(group_);
4852     ovs_mutex_init(&group->stats_mutex);
4853     ovs_mutex_lock(&group->stats_mutex);
4854     group_construct_stats(group);
4855     ovs_mutex_unlock(&group->stats_mutex);
4856     return 0;
4857 }
4858
4859 static void
4860 group_destruct__(struct group_dpif *group)
4861     OVS_REQUIRES(group->stats_mutex)
4862 {
4863     free(group->bucket_stats);
4864     group->bucket_stats = NULL;
4865 }
4866
4867 static void
4868 group_destruct(struct ofgroup *group_)
4869 {
4870     struct group_dpif *group = group_dpif_cast(group_);
4871     ovs_mutex_lock(&group->stats_mutex);
4872     group_destruct__(group);
4873     ovs_mutex_unlock(&group->stats_mutex);
4874     ovs_mutex_destroy(&group->stats_mutex);
4875 }
4876
4877 static enum ofperr
4878 group_modify(struct ofgroup *group_, struct ofgroup *victim_)
4879 {
4880     struct group_dpif *group = group_dpif_cast(group_);
4881     struct group_dpif *victim = group_dpif_cast(victim_);
4882
4883     ovs_mutex_lock(&group->stats_mutex);
4884     if (victim->up.n_buckets < group->up.n_buckets) {
4885         group_destruct__(group);
4886     }
4887     group_construct_stats(group);
4888     ovs_mutex_unlock(&group->stats_mutex);
4889
4890     return 0;
4891 }
4892
4893 static enum ofperr
4894 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4895 {
4896     struct group_dpif *group = group_dpif_cast(group_);
4897
4898     /* Start from historical data for 'group' itself that are no longer tracked
4899      * in facets.  This counts, for example, facets that have expired. */
4900     ovs_mutex_lock(&group->stats_mutex);
4901     ogs->packet_count = group->packet_count;
4902     ogs->byte_count = group->byte_count;
4903     memcpy(ogs->bucket_stats, group->bucket_stats,
4904            group->up.n_buckets * sizeof *group->bucket_stats);
4905     ovs_mutex_unlock(&group->stats_mutex);
4906
4907     return 0;
4908 }
4909
4910 bool
4911 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4912                   struct group_dpif **group)
4913     OVS_TRY_RDLOCK(true, (*group)->up.rwlock)
4914 {
4915     struct ofgroup *ofgroup;
4916     bool found;
4917
4918     *group = NULL;
4919     found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4920     *group = found ?  group_dpif_cast(ofgroup) : NULL;
4921
4922     return found;
4923 }
4924
4925 void
4926 group_dpif_release(struct group_dpif *group)
4927     OVS_RELEASES(group->up.rwlock)
4928 {
4929     ofproto_group_release(&group->up);
4930 }
4931
4932 void
4933 group_dpif_get_buckets(const struct group_dpif *group,
4934                        const struct list **buckets)
4935 {
4936     *buckets = &group->up.buckets;
4937 }
4938
4939 enum ofp11_group_type
4940 group_dpif_get_type(const struct group_dpif *group)
4941 {
4942     return group->up.type;
4943 }
4944 \f
4945 /* Sends 'packet' out 'ofport'.
4946  * May modify 'packet'.
4947  * Returns 0 if successful, otherwise a positive errno value. */
4948 int
4949 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4950 {
4951     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4952     int error;
4953
4954     error = xlate_send_packet(ofport, packet);
4955
4956     ovs_mutex_lock(&ofproto->stats_mutex);
4957     ofproto->stats.tx_packets++;
4958     ofproto->stats.tx_bytes += packet->size;
4959     ovs_mutex_unlock(&ofproto->stats_mutex);
4960     return error;
4961 }
4962
4963 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4964  * The action will state 'slow' as the reason that the action is in the slow
4965  * path.  (This is purely informational: it allows a human viewing "ovs-dpctl
4966  * dump-flows" output to see why a flow is in the slow path.)
4967  *
4968  * The 'stub_size' bytes in 'stub' will be used to store the action.
4969  * 'stub_size' must be large enough for the action.
4970  *
4971  * The action and its size will be stored in '*actionsp' and '*actions_lenp',
4972  * respectively. */
4973 static void
4974 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
4975                   enum slow_path_reason slow,
4976                   uint64_t *stub, size_t stub_size,
4977                   const struct nlattr **actionsp, size_t *actions_lenp)
4978 {
4979     union user_action_cookie cookie;
4980     struct ofpbuf buf;
4981
4982     cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
4983     cookie.slow_path.unused = 0;
4984     cookie.slow_path.reason = slow;
4985
4986     ofpbuf_use_stack(&buf, stub, stub_size);
4987     if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
4988         uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
4989                                          ODPP_NONE);
4990         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
4991     } else {
4992         odp_port_t odp_port;
4993         uint32_t pid;
4994
4995         odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
4996         pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
4997         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
4998     }
4999     *actionsp = buf.data;
5000     *actions_lenp = buf.size;
5001 }
5002 \f
5003 static bool
5004 set_frag_handling(struct ofproto *ofproto_,
5005                   enum ofp_config_flags frag_handling)
5006 {
5007     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5008     if (frag_handling != OFPC_FRAG_REASM) {
5009         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5010         return true;
5011     } else {
5012         return false;
5013     }
5014 }
5015
5016 static enum ofperr
5017 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5018            const struct flow *flow,
5019            const struct ofpact *ofpacts, size_t ofpacts_len)
5020 {
5021     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5022
5023     ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
5024                                  ofpacts_len, packet);
5025     return 0;
5026 }
5027 \f
5028 /* NetFlow. */
5029
5030 static int
5031 set_netflow(struct ofproto *ofproto_,
5032             const struct netflow_options *netflow_options)
5033 {
5034     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5035
5036     if (netflow_options) {
5037         if (!ofproto->netflow) {
5038             ofproto->netflow = netflow_create();
5039             ofproto->backer->need_revalidate = REV_RECONFIGURE;
5040         }
5041         return netflow_set_options(ofproto->netflow, netflow_options);
5042     } else if (ofproto->netflow) {
5043         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5044         netflow_destroy(ofproto->netflow);
5045         ofproto->netflow = NULL;
5046     }
5047
5048     return 0;
5049 }
5050
5051 static void
5052 get_netflow_ids(const struct ofproto *ofproto_,
5053                 uint8_t *engine_type, uint8_t *engine_id)
5054 {
5055     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5056
5057     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5058 }
5059
5060 static void
5061 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5062 {
5063     if (!facet_is_controller_flow(facet) &&
5064         netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5065         struct subfacet *subfacet;
5066         struct ofexpired expired;
5067
5068         LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5069             if (subfacet->path == SF_FAST_PATH) {
5070                 struct dpif_flow_stats stats;
5071
5072                 subfacet_install(subfacet, &facet->xout.odp_actions,
5073                                  &stats);
5074                 subfacet_update_stats(subfacet, &stats);
5075             }
5076         }
5077
5078         expired.flow = facet->flow;
5079         expired.packet_count = facet->packet_count;
5080         expired.byte_count = facet->byte_count;
5081         expired.used = facet->used;
5082         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5083     }
5084 }
5085
5086 static void
5087 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5088 {
5089     struct cls_cursor cursor;
5090     struct facet *facet;
5091
5092     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5093     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5094     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5095         send_active_timeout(ofproto, facet);
5096     }
5097     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5098 }
5099 \f
5100 static struct ofproto_dpif *
5101 ofproto_dpif_lookup(const char *name)
5102 {
5103     struct ofproto_dpif *ofproto;
5104
5105     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5106                              hash_string(name, 0), &all_ofproto_dpifs) {
5107         if (!strcmp(ofproto->up.name, name)) {
5108             return ofproto;
5109         }
5110     }
5111     return NULL;
5112 }
5113
5114 static void
5115 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5116                           const char *argv[], void *aux OVS_UNUSED)
5117 {
5118     struct ofproto_dpif *ofproto;
5119
5120     if (argc > 1) {
5121         ofproto = ofproto_dpif_lookup(argv[1]);
5122         if (!ofproto) {
5123             unixctl_command_reply_error(conn, "no such bridge");
5124             return;
5125         }
5126         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5127         mac_learning_flush(ofproto->ml);
5128         ovs_rwlock_unlock(&ofproto->ml->rwlock);
5129     } else {
5130         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5131             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5132             mac_learning_flush(ofproto->ml);
5133             ovs_rwlock_unlock(&ofproto->ml->rwlock);
5134         }
5135     }
5136
5137     unixctl_command_reply(conn, "table successfully flushed");
5138 }
5139
5140 static struct ofport_dpif *
5141 ofbundle_get_a_port(const struct ofbundle *bundle)
5142 {
5143     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5144                         bundle_node);
5145 }
5146
5147 static void
5148 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5149                          const char *argv[], void *aux OVS_UNUSED)
5150 {
5151     struct ds ds = DS_EMPTY_INITIALIZER;
5152     const struct ofproto_dpif *ofproto;
5153     const struct mac_entry *e;
5154
5155     ofproto = ofproto_dpif_lookup(argv[1]);
5156     if (!ofproto) {
5157         unixctl_command_reply_error(conn, "no such bridge");
5158         return;
5159     }
5160
5161     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
5162     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5163     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5164         struct ofbundle *bundle = e->port.p;
5165         char name[OFP_MAX_PORT_NAME_LEN];
5166
5167         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5168                                name, sizeof name);
5169         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
5170                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
5171                       mac_entry_age(ofproto->ml, e));
5172     }
5173     ovs_rwlock_unlock(&ofproto->ml->rwlock);
5174     unixctl_command_reply(conn, ds_cstr(&ds));
5175     ds_destroy(&ds);
5176 }
5177
5178 struct trace_ctx {
5179     struct xlate_out xout;
5180     struct xlate_in xin;
5181     struct flow flow;
5182     struct ds *result;
5183 };
5184
5185 static void
5186 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5187 {
5188     struct rule_actions *actions;
5189     ovs_be64 cookie;
5190
5191     ds_put_char_multiple(result, '\t', level);
5192     if (!rule) {
5193         ds_put_cstr(result, "No match\n");
5194         return;
5195     }
5196
5197     ovs_mutex_lock(&rule->up.mutex);
5198     cookie = rule->up.flow_cookie;
5199     ovs_mutex_unlock(&rule->up.mutex);
5200
5201     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5202                   rule ? rule->up.table_id : 0, ntohll(cookie));
5203     cls_rule_format(&rule->up.cr, result);
5204     ds_put_char(result, '\n');
5205
5206     actions = rule_dpif_get_actions(rule);
5207
5208     ds_put_char_multiple(result, '\t', level);
5209     ds_put_cstr(result, "OpenFlow actions=");
5210     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
5211     ds_put_char(result, '\n');
5212
5213     rule_actions_unref(actions);
5214 }
5215
5216 static void
5217 trace_format_flow(struct ds *result, int level, const char *title,
5218                   struct trace_ctx *trace)
5219 {
5220     ds_put_char_multiple(result, '\t', level);
5221     ds_put_format(result, "%s: ", title);
5222     if (flow_equal(&trace->xin.flow, &trace->flow)) {
5223         ds_put_cstr(result, "unchanged");
5224     } else {
5225         flow_format(result, &trace->xin.flow);
5226         trace->flow = trace->xin.flow;
5227     }
5228     ds_put_char(result, '\n');
5229 }
5230
5231 static void
5232 trace_format_regs(struct ds *result, int level, const char *title,
5233                   struct trace_ctx *trace)
5234 {
5235     size_t i;
5236
5237     ds_put_char_multiple(result, '\t', level);
5238     ds_put_format(result, "%s:", title);
5239     for (i = 0; i < FLOW_N_REGS; i++) {
5240         ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5241     }
5242     ds_put_char(result, '\n');
5243 }
5244
5245 static void
5246 trace_format_odp(struct ds *result, int level, const char *title,
5247                  struct trace_ctx *trace)
5248 {
5249     struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5250
5251     ds_put_char_multiple(result, '\t', level);
5252     ds_put_format(result, "%s: ", title);
5253     format_odp_actions(result, odp_actions->data, odp_actions->size);
5254     ds_put_char(result, '\n');
5255 }
5256
5257 static void
5258 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5259 {
5260     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5261     struct ds *result = trace->result;
5262
5263     ds_put_char(result, '\n');
5264     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5265     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5266     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
5267     trace_format_rule(result, recurse + 1, rule);
5268 }
5269
5270 static void
5271 trace_report(struct xlate_in *xin, const char *s, int recurse)
5272 {
5273     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5274     struct ds *result = trace->result;
5275
5276     ds_put_char_multiple(result, '\t', recurse);
5277     ds_put_cstr(result, s);
5278     ds_put_char(result, '\n');
5279 }
5280
5281 /* Parses the 'argc' elements of 'argv', ignoring argv[0].  The following
5282  * forms are supported:
5283  *
5284  *     - [dpname] odp_flow [-generate | packet]
5285  *     - bridge br_flow [-generate | packet]
5286  *
5287  * On success, initializes '*ofprotop' and 'flow' and returns NULL.  On failure
5288  * returns a nonnull error message. */
5289 static const char *
5290 parse_flow_and_packet(int argc, const char *argv[],
5291                       struct ofproto_dpif **ofprotop, struct flow *flow,
5292                       struct ofpbuf **packetp)
5293 {
5294     const struct dpif_backer *backer = NULL;
5295     const char *error = NULL;
5296     struct simap port_names = SIMAP_INITIALIZER(&port_names);
5297     struct ofpbuf *packet;
5298     struct ofpbuf odp_key;
5299     struct ofpbuf odp_mask;
5300
5301     ofpbuf_init(&odp_key, 0);
5302     ofpbuf_init(&odp_mask, 0);
5303
5304     /* Handle "-generate" or a hex string as the last argument. */
5305     if (!strcmp(argv[argc - 1], "-generate")) {
5306         packet = ofpbuf_new(0);
5307         argc--;
5308     } else {
5309         error = eth_from_hex(argv[argc - 1], &packet);
5310         if (!error) {
5311             argc--;
5312         } else if (argc == 4) {
5313             /* The 3-argument form must end in "-generate' or a hex string. */
5314             goto exit;
5315         }
5316     }
5317
5318     /* odp_flow can have its in_port specified as a name instead of port no.
5319      * We do not yet know whether a given flow is a odp_flow or a br_flow.
5320      * But, to know whether a flow is odp_flow through odp_flow_from_string(),
5321      * we need to create a simap of name to port no. */
5322     if (argc == 3) {
5323         const char *dp_type;
5324         if (!strncmp(argv[1], "ovs-", 4)) {
5325             dp_type = argv[1] + 4;
5326         } else {
5327             dp_type = argv[1];
5328         }
5329         backer = shash_find_data(&all_dpif_backers, dp_type);
5330     } else if (argc == 2) {
5331         struct shash_node *node;
5332         if (shash_count(&all_dpif_backers) == 1) {
5333             node = shash_first(&all_dpif_backers);
5334             backer = node->data;
5335         }
5336     } else {
5337         error = "Syntax error";
5338         goto exit;
5339     }
5340     if (backer && backer->dpif) {
5341         struct dpif_port dpif_port;
5342         struct dpif_port_dump port_dump;
5343         DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
5344             simap_put(&port_names, dpif_port.name,
5345                       odp_to_u32(dpif_port.port_no));
5346         }
5347     }
5348
5349     /* Parse the flow and determine whether a datapath or
5350      * bridge is specified. If function odp_flow_key_from_string()
5351      * returns 0, the flow is a odp_flow. If function
5352      * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5353     if (!odp_flow_from_string(argv[argc - 1], &port_names,
5354                               &odp_key, &odp_mask)) {
5355         if (!backer) {
5356             error = "Cannot find the datapath";
5357             goto exit;
5358         }
5359
5360         if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, flow,
5361                           NULL, ofprotop, NULL)) {
5362             error = "Invalid datapath flow";
5363             goto exit;
5364         }
5365     } else if (!parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL)) {
5366         if (argc != 3) {
5367             error = "Must specify bridge name";
5368             goto exit;
5369         }
5370
5371         *ofprotop = ofproto_dpif_lookup(argv[1]);
5372         if (!*ofprotop) {
5373             error = "Unknown bridge name";
5374             goto exit;
5375         }
5376     } else {
5377         error = "Bad flow syntax";
5378         goto exit;
5379     }
5380
5381     /* Generate a packet, if requested. */
5382     if (packet) {
5383         if (!packet->size) {
5384             flow_compose(packet, flow);
5385         } else {
5386             union flow_in_port in_port = flow->in_port;
5387
5388             /* Use the metadata from the flow and the packet argument
5389              * to reconstruct the flow. */
5390             flow_extract(packet, flow->skb_priority, flow->pkt_mark, NULL,
5391                          &in_port, flow);
5392         }
5393     }
5394
5395     error = NULL;
5396
5397 exit:
5398     if (error) {
5399         ofpbuf_delete(packet);
5400         packet = NULL;
5401     }
5402     *packetp = packet;
5403     ofpbuf_uninit(&odp_key);
5404     ofpbuf_uninit(&odp_mask);
5405     simap_destroy(&port_names);
5406     return error;
5407 }
5408
5409 static void
5410 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5411                       void *aux OVS_UNUSED)
5412 {
5413     struct ofproto_dpif *ofproto;
5414     struct ofpbuf *packet;
5415     const char *error;
5416     struct flow flow;
5417
5418     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
5419     if (!error) {
5420         struct ds result;
5421
5422         ds_init(&result);
5423         ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
5424         unixctl_command_reply(conn, ds_cstr(&result));
5425         ds_destroy(&result);
5426         ofpbuf_delete(packet);
5427     } else {
5428         unixctl_command_reply_error(conn, error);
5429     }
5430 }
5431
5432 static void
5433 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
5434                               const char *argv[], void *aux OVS_UNUSED)
5435 {
5436     enum ofputil_protocol usable_protocols;
5437     struct ofproto_dpif *ofproto;
5438     bool enforce_consistency;
5439     struct ofpbuf ofpacts;
5440     struct ofpbuf *packet;
5441     struct ds result;
5442     struct flow flow;
5443     uint16_t in_port;
5444
5445     /* Three kinds of error return values! */
5446     enum ofperr retval;
5447     const char *error;
5448     char *rw_error;
5449
5450     packet = NULL;
5451     ds_init(&result);
5452     ofpbuf_init(&ofpacts, 0);
5453
5454     /* Parse actions. */
5455     rw_error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols);
5456     if (rw_error) {
5457         unixctl_command_reply_error(conn, rw_error);
5458         free(rw_error);
5459         goto exit;
5460     }
5461
5462     /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
5463      * consistency between the flow and the actions, so enforce these by
5464      * default if the actions can only work in OF1.1 or later. */
5465     enforce_consistency = !(usable_protocols & OFPUTIL_P_OF10_ANY);
5466     if (!strcmp(argv[1], "-consistent")) {
5467         enforce_consistency = true;
5468         argv++;
5469         argc--;
5470     }
5471
5472     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
5473     if (error) {
5474         unixctl_command_reply_error(conn, error);
5475         goto exit;
5476     }
5477
5478     /* Do the same checks as handle_packet_out() in ofproto.c.
5479      *
5480      * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
5481      * strictly correct because these actions aren't in any table, but it's OK
5482      * because it 'table_id' is used only to check goto_table instructions, but
5483      * packet-outs take a list of actions and therefore it can't include
5484      * instructions.
5485      *
5486      * We skip the "meter" check here because meter is an instruction, not an
5487      * action, and thus cannot appear in ofpacts. */
5488     in_port = ofp_to_u16(flow.in_port.ofp_port);
5489     if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
5490         unixctl_command_reply_error(conn, "invalid in_port");
5491         goto exit;
5492     }
5493     retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow,
5494                            enforce_consistency,
5495                            u16_to_ofp(ofproto->up.max_ports), 0, 0);
5496     if (retval) {
5497         ds_clear(&result);
5498         ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
5499         unixctl_command_reply_error(conn, ds_cstr(&result));
5500         goto exit;
5501     }
5502
5503     ofproto_trace(ofproto, &flow, packet, ofpacts.data, ofpacts.size, &result);
5504     unixctl_command_reply(conn, ds_cstr(&result));
5505
5506 exit:
5507     ds_destroy(&result);
5508     ofpbuf_delete(packet);
5509     ofpbuf_uninit(&ofpacts);
5510 }
5511
5512 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
5513  * description of the results to 'ds'.
5514  *
5515  * The trace follows a packet with the specified 'flow' through the flow
5516  * table.  'packet' may be nonnull to trace an actual packet, with consequent
5517  * side effects (if it is nonnull then its flow must be 'flow').
5518  *
5519  * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
5520  * trace, otherwise the actions are determined by a flow table lookup. */
5521 static void
5522 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5523               const struct ofpbuf *packet,
5524               const struct ofpact ofpacts[], size_t ofpacts_len,
5525               struct ds *ds)
5526 {
5527     struct rule_dpif *rule;
5528     struct flow_wildcards wc;
5529
5530     ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
5531     ds_put_cstr(ds, "Flow: ");
5532     flow_format(ds, flow);
5533     ds_put_char(ds, '\n');
5534
5535     flow_wildcards_init_catchall(&wc);
5536     if (ofpacts) {
5537         rule = NULL;
5538     } else {
5539         rule_dpif_lookup(ofproto, flow, &wc, &rule);
5540
5541         trace_format_rule(ds, 0, rule);
5542         if (rule == ofproto->miss_rule) {
5543             ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5544         } else if (rule == ofproto->no_packet_in_rule) {
5545             ds_put_cstr(ds, "\nNo match, packets dropped because "
5546                         "OFPPC_NO_PACKET_IN is set on in_port.\n");
5547         } else if (rule == ofproto->drop_frags_rule) {
5548             ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5549                         "and the fragment handling mode is \"drop\".\n");
5550         }
5551     }
5552
5553     if (rule || ofpacts) {
5554         uint64_t odp_actions_stub[1024 / 8];
5555         struct ofpbuf odp_actions;
5556         struct trace_ctx trace;
5557         struct match match;
5558         uint16_t tcp_flags;
5559
5560         tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5561         trace.result = ds;
5562         trace.flow = *flow;
5563         ofpbuf_use_stub(&odp_actions,
5564                         odp_actions_stub, sizeof odp_actions_stub);
5565         xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5566         if (ofpacts) {
5567             trace.xin.ofpacts = ofpacts;
5568             trace.xin.ofpacts_len = ofpacts_len;
5569         }
5570         trace.xin.resubmit_hook = trace_resubmit;
5571         trace.xin.report_hook = trace_report;
5572
5573         xlate_actions(&trace.xin, &trace.xout);
5574         flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5575
5576         ds_put_char(ds, '\n');
5577         trace_format_flow(ds, 0, "Final flow", &trace);
5578
5579         match_init(&match, flow, &trace.xout.wc);
5580         ds_put_cstr(ds, "Relevant fields: ");
5581         match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5582         ds_put_char(ds, '\n');
5583
5584         ds_put_cstr(ds, "Datapath actions: ");
5585         format_odp_actions(ds, trace.xout.odp_actions.data,
5586                            trace.xout.odp_actions.size);
5587
5588         if (trace.xout.slow) {
5589             enum slow_path_reason slow;
5590
5591             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5592                         "slow path because it:");
5593
5594             slow = trace.xout.slow;
5595             while (slow) {
5596                 enum slow_path_reason bit = rightmost_1bit(slow);
5597
5598                 ds_put_format(ds, "\n\t- %s.",
5599                               slow_path_reason_to_explanation(bit));
5600
5601                 slow &= ~bit;
5602             }
5603         }
5604
5605         xlate_out_uninit(&trace.xout);
5606     }
5607
5608     rule_dpif_unref(rule);
5609 }
5610
5611 /* Runs a self-check of flow translations in 'ofproto'.  Appends a message to
5612  * 'reply' describing the results. */
5613 static void
5614 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5615 {
5616     struct cls_cursor cursor;
5617     struct facet *facet;
5618     int errors;
5619
5620     errors = 0;
5621     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5622     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5623     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5624         if (!facet_check_consistency(facet)) {
5625             errors++;
5626         }
5627     }
5628     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5629     if (errors) {
5630         ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5631     }
5632
5633     if (errors) {
5634         ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5635                       ofproto->up.name, errors);
5636     } else {
5637         ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5638     }
5639 }
5640
5641 static void
5642 ofproto_dpif_self_check(struct unixctl_conn *conn,
5643                         int argc, const char *argv[], void *aux OVS_UNUSED)
5644 {
5645     struct ds reply = DS_EMPTY_INITIALIZER;
5646     struct ofproto_dpif *ofproto;
5647
5648     if (argc > 1) {
5649         ofproto = ofproto_dpif_lookup(argv[1]);
5650         if (!ofproto) {
5651             unixctl_command_reply_error(conn, "Unknown ofproto (use "
5652                                         "ofproto/list for help)");
5653             return;
5654         }
5655         ofproto_dpif_self_check__(ofproto, &reply);
5656     } else {
5657         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5658             ofproto_dpif_self_check__(ofproto, &reply);
5659         }
5660     }
5661
5662     unixctl_command_reply(conn, ds_cstr(&reply));
5663     ds_destroy(&reply);
5664 }
5665
5666 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
5667  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
5668  * to destroy 'ofproto_shash' and free the returned value. */
5669 static const struct shash_node **
5670 get_ofprotos(struct shash *ofproto_shash)
5671 {
5672     const struct ofproto_dpif *ofproto;
5673
5674     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5675         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5676         shash_add_nocopy(ofproto_shash, name, ofproto);
5677     }
5678
5679     return shash_sort(ofproto_shash);
5680 }
5681
5682 static void
5683 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5684                               const char *argv[] OVS_UNUSED,
5685                               void *aux OVS_UNUSED)
5686 {
5687     struct ds ds = DS_EMPTY_INITIALIZER;
5688     struct shash ofproto_shash;
5689     const struct shash_node **sorted_ofprotos;
5690     int i;
5691
5692     shash_init(&ofproto_shash);
5693     sorted_ofprotos = get_ofprotos(&ofproto_shash);
5694     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5695         const struct shash_node *node = sorted_ofprotos[i];
5696         ds_put_format(&ds, "%s\n", node->name);
5697     }
5698
5699     shash_destroy(&ofproto_shash);
5700     free(sorted_ofprotos);
5701
5702     unixctl_command_reply(conn, ds_cstr(&ds));
5703     ds_destroy(&ds);
5704 }
5705
5706 static void
5707 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5708 {
5709     const struct shash_node **ofprotos;
5710     struct ofproto_dpif *ofproto;
5711     struct shash ofproto_shash;
5712     uint64_t n_hit, n_missed;
5713     size_t i;
5714
5715     n_hit = n_missed = 0;
5716     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5717         if (ofproto->backer == backer) {
5718             n_missed += ofproto->n_missed;
5719             n_hit += ofproto->n_hit;
5720         }
5721     }
5722
5723     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5724                   dpif_name(backer->dpif), n_hit, n_missed);
5725     ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5726                   " life span: %lldms\n", hmap_count(&backer->subfacets),
5727                   backer->avg_n_subfacet, backer->max_n_subfacet,
5728                   backer->avg_subfacet_life);
5729
5730     shash_init(&ofproto_shash);
5731     ofprotos = get_ofprotos(&ofproto_shash);
5732     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5733         struct ofproto_dpif *ofproto = ofprotos[i]->data;
5734         const struct shash_node **ports;
5735         size_t j;
5736
5737         if (ofproto->backer != backer) {
5738             continue;
5739         }
5740
5741         ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5742                       ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5743
5744         ports = shash_sort(&ofproto->up.port_by_name);
5745         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5746             const struct shash_node *node = ports[j];
5747             struct ofport *ofport = node->data;
5748             struct smap config;
5749             odp_port_t odp_port;
5750
5751             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5752                           ofport->ofp_port);
5753
5754             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5755             if (odp_port != ODPP_NONE) {
5756                 ds_put_format(ds, "%"PRIu32":", odp_port);
5757             } else {
5758                 ds_put_cstr(ds, "none:");
5759             }
5760
5761             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5762
5763             smap_init(&config);
5764             if (!netdev_get_config(ofport->netdev, &config)) {
5765                 const struct smap_node **nodes;
5766                 size_t i;
5767
5768                 nodes = smap_sort(&config);
5769                 for (i = 0; i < smap_count(&config); i++) {
5770                     const struct smap_node *node = nodes[i];
5771                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5772                                   node->key, node->value);
5773                 }
5774                 free(nodes);
5775             }
5776             smap_destroy(&config);
5777
5778             ds_put_char(ds, ')');
5779             ds_put_char(ds, '\n');
5780         }
5781         free(ports);
5782     }
5783     shash_destroy(&ofproto_shash);
5784     free(ofprotos);
5785 }
5786
5787 static void
5788 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5789                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5790 {
5791     struct ds ds = DS_EMPTY_INITIALIZER;
5792     const struct shash_node **backers;
5793     int i;
5794
5795     backers = shash_sort(&all_dpif_backers);
5796     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5797         dpif_show_backer(backers[i]->data, &ds);
5798     }
5799     free(backers);
5800
5801     unixctl_command_reply(conn, ds_cstr(&ds));
5802     ds_destroy(&ds);
5803 }
5804
5805 /* Dump the megaflow (facet) cache.  This is useful to check the
5806  * correctness of flow wildcarding, since the same mechanism is used for
5807  * both xlate caching and kernel wildcarding.
5808  *
5809  * It's important to note that in the output the flow description uses
5810  * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5811  *
5812  * This command is only needed for advanced debugging, so it's not
5813  * documented in the man page. */
5814 static void
5815 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5816                                     int argc OVS_UNUSED, const char *argv[],
5817                                     void *aux OVS_UNUSED)
5818 {
5819     struct ds ds = DS_EMPTY_INITIALIZER;
5820     const struct ofproto_dpif *ofproto;
5821     long long int now = time_msec();
5822     struct cls_cursor cursor;
5823     struct facet *facet;
5824
5825     ofproto = ofproto_dpif_lookup(argv[1]);
5826     if (!ofproto) {
5827         unixctl_command_reply_error(conn, "no such bridge");
5828         return;
5829     }
5830
5831     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5832     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5833     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5834         cls_rule_format(&facet->cr, &ds);
5835         ds_put_cstr(&ds, ", ");
5836         ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5837         ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5838         ds_put_cstr(&ds, "Datapath actions: ");
5839         if (facet->xout.slow) {
5840             uint64_t slow_path_stub[128 / 8];
5841             const struct nlattr *actions;
5842             size_t actions_len;
5843
5844             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5845                               slow_path_stub, sizeof slow_path_stub,
5846                               &actions, &actions_len);
5847             format_odp_actions(&ds, actions, actions_len);
5848         } else {
5849             format_odp_actions(&ds, facet->xout.odp_actions.data,
5850                                facet->xout.odp_actions.size);
5851         }
5852         ds_put_cstr(&ds, "\n");
5853     }
5854     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5855
5856     ds_chomp(&ds, '\n');
5857     unixctl_command_reply(conn, ds_cstr(&ds));
5858     ds_destroy(&ds);
5859 }
5860
5861 /* Disable using the megaflows.
5862  *
5863  * This command is only needed for advanced debugging, so it's not
5864  * documented in the man page. */
5865 static void
5866 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5867                                        int argc OVS_UNUSED,
5868                                        const char *argv[] OVS_UNUSED,
5869                                        void *aux OVS_UNUSED)
5870 {
5871     struct ofproto_dpif *ofproto;
5872
5873     enable_megaflows = false;
5874
5875     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5876         flush(&ofproto->up);
5877     }
5878
5879     unixctl_command_reply(conn, "megaflows disabled");
5880 }
5881
5882 /* Re-enable using megaflows.
5883  *
5884  * This command is only needed for advanced debugging, so it's not
5885  * documented in the man page. */
5886 static void
5887 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5888                                       int argc OVS_UNUSED,
5889                                       const char *argv[] OVS_UNUSED,
5890                                       void *aux OVS_UNUSED)
5891 {
5892     struct ofproto_dpif *ofproto;
5893
5894     enable_megaflows = true;
5895
5896     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5897         flush(&ofproto->up);
5898     }
5899
5900     unixctl_command_reply(conn, "megaflows enabled");
5901 }
5902
5903 static void
5904 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5905                                 int argc OVS_UNUSED, const char *argv[],
5906                                 void *aux OVS_UNUSED)
5907 {
5908     struct ds ds = DS_EMPTY_INITIALIZER;
5909     const struct ofproto_dpif *ofproto;
5910     struct subfacet *subfacet;
5911
5912     ofproto = ofproto_dpif_lookup(argv[1]);
5913     if (!ofproto) {
5914         unixctl_command_reply_error(conn, "no such bridge");
5915         return;
5916     }
5917
5918     update_stats(ofproto->backer);
5919
5920     HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5921         struct facet *facet = subfacet->facet;
5922         struct odputil_keybuf maskbuf;
5923         struct ofpbuf mask;
5924
5925         if (facet->ofproto != ofproto) {
5926             continue;
5927         }
5928
5929         ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5930         if (enable_megaflows) {
5931             odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5932                                    &facet->flow, UINT32_MAX);
5933         }
5934
5935         odp_flow_format(subfacet->key, subfacet->key_len,
5936                         mask.data, mask.size, NULL, &ds, false);
5937
5938         ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5939                       subfacet->dp_packet_count, subfacet->dp_byte_count);
5940         if (subfacet->used) {
5941             ds_put_format(&ds, "%.3fs",
5942                           (time_msec() - subfacet->used) / 1000.0);
5943         } else {
5944             ds_put_format(&ds, "never");
5945         }
5946         if (subfacet->facet->tcp_flags) {
5947             ds_put_cstr(&ds, ", flags:");
5948             packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5949         }
5950
5951         ds_put_cstr(&ds, ", actions:");
5952         if (facet->xout.slow) {
5953             uint64_t slow_path_stub[128 / 8];
5954             const struct nlattr *actions;
5955             size_t actions_len;
5956
5957             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5958                               slow_path_stub, sizeof slow_path_stub,
5959                               &actions, &actions_len);
5960             format_odp_actions(&ds, actions, actions_len);
5961         } else {
5962             format_odp_actions(&ds, facet->xout.odp_actions.data,
5963                                facet->xout.odp_actions.size);
5964         }
5965         ds_put_char(&ds, '\n');
5966     }
5967
5968     unixctl_command_reply(conn, ds_cstr(&ds));
5969     ds_destroy(&ds);
5970 }
5971
5972 static void
5973 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5974                                int argc OVS_UNUSED, const char *argv[],
5975                                void *aux OVS_UNUSED)
5976 {
5977     struct ds ds = DS_EMPTY_INITIALIZER;
5978     struct ofproto_dpif *ofproto;
5979
5980     ofproto = ofproto_dpif_lookup(argv[1]);
5981     if (!ofproto) {
5982         unixctl_command_reply_error(conn, "no such bridge");
5983         return;
5984     }
5985
5986     flush(&ofproto->up);
5987
5988     unixctl_command_reply(conn, ds_cstr(&ds));
5989     ds_destroy(&ds);
5990 }
5991
5992 static void
5993 ofproto_dpif_unixctl_init(void)
5994 {
5995     static bool registered;
5996     if (registered) {
5997         return;
5998     }
5999     registered = true;
6000
6001     unixctl_command_register(
6002         "ofproto/trace",
6003         "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
6004         1, 3, ofproto_unixctl_trace, NULL);
6005     unixctl_command_register(
6006         "ofproto/trace-packet-out",
6007         "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
6008         2, 6, ofproto_unixctl_trace_actions, NULL);
6009     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6010                              ofproto_unixctl_fdb_flush, NULL);
6011     unixctl_command_register("fdb/show", "bridge", 1, 1,
6012                              ofproto_unixctl_fdb_show, NULL);
6013     unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6014                              ofproto_dpif_self_check, NULL);
6015     unixctl_command_register("dpif/dump-dps", "", 0, 0,
6016                              ofproto_unixctl_dpif_dump_dps, NULL);
6017     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6018                              NULL);
6019     unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6020                              ofproto_unixctl_dpif_dump_flows, NULL);
6021     unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6022                              ofproto_unixctl_dpif_del_flows, NULL);
6023     unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6024                              ofproto_unixctl_dpif_dump_megaflows, NULL);
6025     unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6026                              ofproto_unixctl_dpif_disable_megaflows, NULL);
6027     unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6028                              ofproto_unixctl_dpif_enable_megaflows, NULL);
6029 }
6030 \f
6031 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6032  *
6033  * This is deprecated.  It is only for compatibility with broken device drivers
6034  * in old versions of Linux that do not properly support VLANs when VLAN
6035  * devices are not used.  When broken device drivers are no longer in
6036  * widespread use, we will delete these interfaces. */
6037
6038 static int
6039 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6040 {
6041     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6042     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6043
6044     if (realdev_ofp_port == ofport->realdev_ofp_port
6045         && vid == ofport->vlandev_vid) {
6046         return 0;
6047     }
6048
6049     ofproto->backer->need_revalidate = REV_RECONFIGURE;
6050
6051     if (ofport->realdev_ofp_port) {
6052         vsp_remove(ofport);
6053     }
6054     if (realdev_ofp_port && ofport->bundle) {
6055         /* vlandevs are enslaved to their realdevs, so they are not allowed to
6056          * themselves be part of a bundle. */
6057         bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6058     }
6059
6060     ofport->realdev_ofp_port = realdev_ofp_port;
6061     ofport->vlandev_vid = vid;
6062
6063     if (realdev_ofp_port) {
6064         vsp_add(ofport, realdev_ofp_port, vid);
6065     }
6066
6067     return 0;
6068 }
6069
6070 static uint32_t
6071 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6072 {
6073     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6074 }
6075
6076 bool
6077 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6078     OVS_EXCLUDED(ofproto->vsp_mutex)
6079 {
6080     bool ret;
6081
6082     ovs_mutex_lock(&ofproto->vsp_mutex);
6083     ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6084     ovs_mutex_unlock(&ofproto->vsp_mutex);
6085     return ret;
6086 }
6087
6088 static ofp_port_t
6089 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6090                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6091     OVS_REQUIRES(ofproto->vsp_mutex)
6092 {
6093     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6094         int vid = vlan_tci_to_vid(vlan_tci);
6095         const struct vlan_splinter *vsp;
6096
6097         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6098                                  hash_realdev_vid(realdev_ofp_port, vid),
6099                                  &ofproto->realdev_vid_map) {
6100             if (vsp->realdev_ofp_port == realdev_ofp_port
6101                 && vsp->vid == vid) {
6102                 return vsp->vlandev_ofp_port;
6103             }
6104         }
6105     }
6106     return realdev_ofp_port;
6107 }
6108
6109 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6110  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6111  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
6112  * 'vlan_tci' 9, it would return the port number of eth0.9.
6113  *
6114  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6115  * function just returns its 'realdev_ofp_port' argument. */
6116 ofp_port_t
6117 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6118                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6119     OVS_EXCLUDED(ofproto->vsp_mutex)
6120 {
6121     ofp_port_t ret;
6122
6123     ovs_mutex_lock(&ofproto->vsp_mutex);
6124     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6125     ovs_mutex_unlock(&ofproto->vsp_mutex);
6126     return ret;
6127 }
6128
6129 static struct vlan_splinter *
6130 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6131 {
6132     struct vlan_splinter *vsp;
6133
6134     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6135                              hash_ofp_port(vlandev_ofp_port),
6136                              &ofproto->vlandev_map) {
6137         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6138             return vsp;
6139         }
6140     }
6141
6142     return NULL;
6143 }
6144
6145 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6146  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6147  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
6148  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6149  * eth0 and store 9 in '*vid'.
6150  *
6151  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6152  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
6153  * always does.*/
6154 static ofp_port_t
6155 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6156                        ofp_port_t vlandev_ofp_port, int *vid)
6157     OVS_REQUIRES(ofproto->vsp_mutex)
6158 {
6159     if (!hmap_is_empty(&ofproto->vlandev_map)) {
6160         const struct vlan_splinter *vsp;
6161
6162         vsp = vlandev_find(ofproto, vlandev_ofp_port);
6163         if (vsp) {
6164             if (vid) {
6165                 *vid = vsp->vid;
6166             }
6167             return vsp->realdev_ofp_port;
6168         }
6169     }
6170     return 0;
6171 }
6172
6173 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6174  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
6175  * 'flow->in_port' to the "real" device backing the VLAN device, sets
6176  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Otherwise (which is
6177  * always the case unless VLAN splinters are enabled), returns false without
6178  * making any changes. */
6179 bool
6180 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6181     OVS_EXCLUDED(ofproto->vsp_mutex)
6182 {
6183     ofp_port_t realdev;
6184     int vid;
6185
6186     ovs_mutex_lock(&ofproto->vsp_mutex);
6187     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6188     ovs_mutex_unlock(&ofproto->vsp_mutex);
6189     if (!realdev) {
6190         return false;
6191     }
6192
6193     /* Cause the flow to be processed as if it came in on the real device with
6194      * the VLAN device's VLAN ID. */
6195     flow->in_port.ofp_port = realdev;
6196     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6197     return true;
6198 }
6199
6200 static void
6201 vsp_remove(struct ofport_dpif *port)
6202 {
6203     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6204     struct vlan_splinter *vsp;
6205
6206     ovs_mutex_lock(&ofproto->vsp_mutex);
6207     vsp = vlandev_find(ofproto, port->up.ofp_port);
6208     if (vsp) {
6209         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6210         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6211         free(vsp);
6212
6213         port->realdev_ofp_port = 0;
6214     } else {
6215         VLOG_ERR("missing vlan device record");
6216     }
6217     ovs_mutex_unlock(&ofproto->vsp_mutex);
6218 }
6219
6220 static void
6221 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6222 {
6223     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6224
6225     ovs_mutex_lock(&ofproto->vsp_mutex);
6226     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6227         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6228             == realdev_ofp_port)) {
6229         struct vlan_splinter *vsp;
6230
6231         vsp = xmalloc(sizeof *vsp);
6232         vsp->realdev_ofp_port = realdev_ofp_port;
6233         vsp->vlandev_ofp_port = port->up.ofp_port;
6234         vsp->vid = vid;
6235
6236         port->realdev_ofp_port = realdev_ofp_port;
6237
6238         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6239                     hash_ofp_port(port->up.ofp_port));
6240         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6241                     hash_realdev_vid(realdev_ofp_port, vid));
6242     } else {
6243         VLOG_ERR("duplicate vlan device record");
6244     }
6245     ovs_mutex_unlock(&ofproto->vsp_mutex);
6246 }
6247
6248 static odp_port_t
6249 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6250 {
6251     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6252     return ofport ? ofport->odp_port : ODPP_NONE;
6253 }
6254
6255 struct ofport_dpif *
6256 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6257 {
6258     struct ofport_dpif *port;
6259
6260     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6261     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6262                              &backer->odp_to_ofport_map) {
6263         if (port->odp_port == odp_port) {
6264             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6265             return port;
6266         }
6267     }
6268
6269     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6270     return NULL;
6271 }
6272
6273 static ofp_port_t
6274 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6275 {
6276     struct ofport_dpif *port;
6277
6278     port = odp_port_to_ofport(ofproto->backer, odp_port);
6279     if (port && &ofproto->up == port->up.ofproto) {
6280         return port->up.ofp_port;
6281     } else {
6282         return OFPP_NONE;
6283     }
6284 }
6285
6286 const struct ofproto_class ofproto_dpif_class = {
6287     init,
6288     enumerate_types,
6289     enumerate_names,
6290     del,
6291     port_open_type,
6292     type_run,
6293     type_run_fast,
6294     type_wait,
6295     alloc,
6296     construct,
6297     destruct,
6298     dealloc,
6299     run,
6300     run_fast,
6301     wait,
6302     get_memory_usage,
6303     flush,
6304     get_features,
6305     get_tables,
6306     port_alloc,
6307     port_construct,
6308     port_destruct,
6309     port_dealloc,
6310     port_modified,
6311     port_reconfigured,
6312     port_query_by_name,
6313     port_add,
6314     port_del,
6315     port_get_stats,
6316     port_dump_start,
6317     port_dump_next,
6318     port_dump_done,
6319     port_poll,
6320     port_poll_wait,
6321     port_is_lacp_current,
6322     NULL,                       /* rule_choose_table */
6323     rule_alloc,
6324     rule_construct,
6325     rule_insert,
6326     rule_delete,
6327     rule_destruct,
6328     rule_dealloc,
6329     rule_get_stats,
6330     rule_execute,
6331     rule_modify_actions,
6332     set_frag_handling,
6333     packet_out,
6334     set_netflow,
6335     get_netflow_ids,
6336     set_sflow,
6337     set_ipfix,
6338     set_cfm,
6339     get_cfm_status,
6340     set_bfd,
6341     get_bfd_status,
6342     set_stp,
6343     get_stp_status,
6344     set_stp_port,
6345     get_stp_port_status,
6346     set_queues,
6347     bundle_set,
6348     bundle_remove,
6349     mirror_set__,
6350     mirror_get_stats__,
6351     set_flood_vlans,
6352     is_mirror_output_bundle,
6353     forward_bpdu_changed,
6354     set_mac_table_config,
6355     set_realdev,
6356     NULL,                       /* meter_get_features */
6357     NULL,                       /* meter_set */
6358     NULL,                       /* meter_get */
6359     NULL,                       /* meter_del */
6360     group_alloc,                /* group_alloc */
6361     group_construct,            /* group_construct */
6362     group_destruct,             /* group_destruct */
6363     group_dealloc,              /* group_dealloc */
6364     group_modify,               /* group_modify */
6365     group_get_stats,            /* group_get_stats */
6366 };