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