ofproto/ofproto-dpif.c

   1 /*
   2  * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18
  19 #include "ofproto/ofproto-dpif.h"
  20 #include "ofproto/ofproto-provider.h"
  21
  22 #include <errno.h>
  23
  24 #include "bfd.h"
  25 #include "bond.h"
  26 #include "bundle.h"
  27 #include "byte-order.h"
  28 #include "connmgr.h"
  29 #include "coverage.h"
  30 #include "cfm.h"
  31 #include "dpif.h"
  32 #include "dynamic-string.h"
  33 #include "fail-open.h"
  34 #include "guarded-list.h"
  35 #include "hmapx.h"
  36 #include "lacp.h"
  37 #include "learn.h"
  38 #include "mac-learning.h"
  39 #include "meta-flow.h"
  40 #include "multipath.h"
  41 #include "netdev-vport.h"
  42 #include "netdev.h"
  43 #include "netlink.h"
  44 #include "nx-match.h"
  45 #include "odp-util.h"
  46 #include "odp-execute.h"
  47 #include "ofp-util.h"
  48 #include "ofpbuf.h"
  49 #include "ofp-actions.h"
  50 #include "ofp-parse.h"
  51 #include "ofp-print.h"
  52 #include "ofproto-dpif-governor.h"
  53 #include "ofproto-dpif-ipfix.h"
  54 #include "ofproto-dpif-mirror.h"
  55 #include "ofproto-dpif-sflow.h"
  56 #include "ofproto-dpif-upcall.h"
  57 #include "ofproto-dpif-xlate.h"
  58 #include "poll-loop.h"
  59 #include "simap.h"
  60 #include "smap.h"
  61 #include "timer.h"
  62 #include "tunnel.h"
  63 #include "unaligned.h"
  64 #include "unixctl.h"
  65 #include "vlan-bitmap.h"
  66 #include "vlog.h"
  67
  68 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  69
  70 COVERAGE_DEFINE(ofproto_dpif_expired);
  71 COVERAGE_DEFINE(facet_changed_rule);
  72 COVERAGE_DEFINE(facet_revalidate);
  73 COVERAGE_DEFINE(facet_unexpected);
  74 COVERAGE_DEFINE(facet_suppress);
  75 COVERAGE_DEFINE(subfacet_install_fail);
  76 COVERAGE_DEFINE(packet_in_overflow);
  77 COVERAGE_DEFINE(flow_mod_overflow);
  78
  79 /* Number of implemented OpenFlow tables. */
  80 enum { N_TABLES = 255 };
  81 enum { TBL_INTERNAL = N_TABLES - 1 };    /* Used for internal hidden rules. */
  82 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
  83
  84 struct flow_miss;
  85 struct facet;
  86
  87 struct rule_dpif {
  88     struct rule up;
  89
  90     /* These statistics:
  91      *
  92      *   - Do include packets and bytes from facets that have been deleted or
  93      *     whose own statistics have been folded into the rule.
  94      *
  95      *   - Do include packets and bytes sent "by hand" that were accounted to
  96      *     the rule without any facet being involved (this is a rare corner
  97      *     case in rule_execute()).
  98      *
  99      *   - Do not include packet or bytes that can be obtained from any facet's
 100      *     packet_count or byte_count member or that can be obtained from the
 101      *     datapath by, e.g., dpif_flow_get() for any subfacet.
 102      */
 103     struct ovs_mutex stats_mutex;
 104     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 105     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 106 };
 107
 108 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
 109 static struct rule_dpif *rule_dpif_cast(const struct rule *);
 110
 111 struct ofbundle {
 112     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
 113     struct ofproto_dpif *ofproto; /* Owning ofproto. */
 114     void *aux;                  /* Key supplied by ofproto's client. */
 115     char *name;                 /* Identifier for log messages. */
 116
 117     /* Configuration. */
 118     struct list ports;          /* Contains "struct ofport"s. */
 119     enum port_vlan_mode vlan_mode; /* VLAN mode */
 120     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
 121     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
 122                                  * NULL if all VLANs are trunked. */
 123     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 124     struct bond *bond;          /* Nonnull iff more than one port. */
 125     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 126
 127     /* Status. */
 128     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 129 };
 130
 131 static void bundle_remove(struct ofport *);
 132 static void bundle_update(struct ofbundle *);
 133 static void bundle_destroy(struct ofbundle *);
 134 static void bundle_del_port(struct ofport_dpif *);
 135 static void bundle_run(struct ofbundle *);
 136 static void bundle_wait(struct ofbundle *);
 137
 138 static void stp_run(struct ofproto_dpif *ofproto);
 139 static void stp_wait(struct ofproto_dpif *ofproto);
 140 static int set_stp_port(struct ofport *,
 141                         const struct ofproto_port_stp_settings *);
 142
 143 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
 144                               enum slow_path_reason,
 145                               uint64_t *stub, size_t stub_size,
 146                               const struct nlattr **actionsp,
 147                               size_t *actions_lenp);
 148
 149 /* A subfacet (see "struct subfacet" below) has three possible installation
 150  * states:
 151  *
 152  *   - SF_NOT_INSTALLED: Not installed in the datapath.  This will only be the
 153  *     case just after the subfacet is created, just before the subfacet is
 154  *     destroyed, or if the datapath returns an error when we try to install a
 155  *     subfacet.
 156  *
 157  *   - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
 158  *
 159  *   - SF_SLOW_PATH: An action that sends every packet for the subfacet through
 160  *     ofproto_dpif is installed in the datapath.
 161  */
 162 enum subfacet_path {
 163     SF_NOT_INSTALLED,           /* No datapath flow for this subfacet. */
 164     SF_FAST_PATH,               /* Full actions are installed. */
 165     SF_SLOW_PATH,               /* Send-to-userspace action is installed. */
 166 };
 167
 168 /* A dpif flow and actions associated with a facet.
 169  *
 170  * See also the large comment on struct facet. */
 171 struct subfacet {
 172     /* Owners. */
 173     struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
 174     struct list list_node;      /* In struct facet's 'facets' list. */
 175     struct facet *facet;        /* Owning facet. */
 176     struct dpif_backer *backer; /* Owning backer. */
 177
 178     enum odp_key_fitness key_fitness;
 179     struct nlattr *key;
 180     int key_len;
 181
 182     long long int used;         /* Time last used; time created if not used. */
 183     long long int created;      /* Time created. */
 184
 185     uint64_t dp_packet_count;   /* Last known packet count in the datapath. */
 186     uint64_t dp_byte_count;     /* Last known byte count in the datapath. */
 187
 188     enum subfacet_path path;    /* Installed in datapath? */
 189 };
 190
 191 #define SUBFACET_DESTROY_MAX_BATCH 50
 192
 193 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *);
 194 static struct subfacet *subfacet_find(struct dpif_backer *,
 195                                       const struct nlattr *key, size_t key_len,
 196                                       uint32_t key_hash);
 197 static void subfacet_destroy(struct subfacet *);
 198 static void subfacet_destroy__(struct subfacet *);
 199 static void subfacet_destroy_batch(struct dpif_backer *,
 200                                    struct subfacet **, int n);
 201 static void subfacet_reset_dp_stats(struct subfacet *,
 202                                     struct dpif_flow_stats *);
 203 static void subfacet_update_stats(struct subfacet *,
 204                                   const struct dpif_flow_stats *);
 205 static int subfacet_install(struct subfacet *,
 206                             const struct ofpbuf *odp_actions,
 207                             struct dpif_flow_stats *);
 208 static void subfacet_uninstall(struct subfacet *);
 209
 210 /* A unique, non-overlapping instantiation of an OpenFlow flow.
 211  *
 212  * A facet associates a "struct flow", which represents the Open vSwitch
 213  * userspace idea of an exact-match flow, with one or more subfacets.
 214  * While the facet is created based on an exact-match flow, it is stored
 215  * within the ofproto based on the wildcards that could be expressed
 216  * based on the flow table and other configuration.  (See the 'wc'
 217  * description in "struct xlate_out" for more details.)
 218  *
 219  * Each subfacet tracks the datapath's idea of the flow equivalent to
 220  * the facet.  When the kernel module (or other dpif implementation) and
 221  * Open vSwitch userspace agree on the definition of a flow key, there
 222  * is exactly one subfacet per facet.  If the dpif implementation
 223  * supports more-specific flow matching than userspace, however, a facet
 224  * can have more than one subfacet.  Examples include the dpif
 225  * implementation not supporting the same wildcards as userspace or some
 226  * distinction in flow that userspace simply doesn't understand.
 227  *
 228  * Flow expiration works in terms of subfacets, so a facet must have at
 229  * least one subfacet or it will never expire, leaking memory. */
 230 struct facet {
 231     /* Owner. */
 232     struct ofproto_dpif *ofproto;
 233
 234     /* Owned data. */
 235     struct list subfacets;
 236     long long int used;         /* Time last used; time created if not used. */
 237
 238     /* Key. */
 239     struct flow flow;           /* Flow of the creating subfacet. */
 240     struct cls_rule cr;         /* In 'ofproto_dpif's facets classifier. */
 241
 242     /* These statistics:
 243      *
 244      *   - Do include packets and bytes sent "by hand", e.g. with
 245      *     dpif_execute().
 246      *
 247      *   - Do include packets and bytes that were obtained from the datapath
 248      *     when a subfacet's statistics were reset (e.g. dpif_flow_put() with
 249      *     DPIF_FP_ZERO_STATS).
 250      *
 251      *   - Do not include packets or bytes that can be obtained from the
 252      *     datapath for any existing subfacet.
 253      */
 254     uint64_t packet_count;       /* Number of packets received. */
 255     uint64_t byte_count;         /* Number of bytes received. */
 256
 257     /* Resubmit statistics. */
 258     uint64_t prev_packet_count;  /* Number of packets from last stats push. */
 259     uint64_t prev_byte_count;    /* Number of bytes from last stats push. */
 260     long long int prev_used;     /* Used time from last stats push. */
 261
 262     /* Accounting. */
 263     uint64_t accounted_bytes;    /* Bytes processed by facet_account(). */
 264     struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
 265     uint8_t tcp_flags;           /* TCP flags seen for this 'rule'. */
 266
 267     struct xlate_out xout;
 268
 269     /* Storage for a single subfacet, to reduce malloc() time and space
 270      * overhead.  (A facet always has at least one subfacet and in the common
 271      * case has exactly one subfacet.  However, 'one_subfacet' may not
 272      * always be valid, since it could have been removed after newer
 273      * subfacets were pushed onto the 'subfacets' list.) */
 274     struct subfacet one_subfacet;
 275
 276     long long int learn_rl;      /* Rate limiter for facet_learn(). */
 277 };
 278
 279 static struct facet *facet_create(const struct flow_miss *);
 280 static void facet_remove(struct facet *);
 281 static void facet_free(struct facet *);
 282
 283 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
 284 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
 285                                         const struct flow *);
 286 static bool facet_revalidate(struct facet *);
 287 static bool facet_check_consistency(struct facet *);
 288
 289 static void facet_flush_stats(struct facet *);
 290
 291 static void facet_reset_counters(struct facet *);
 292 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
 293                             struct dpif_flow_stats *, bool may_learn);
 294 static void facet_push_stats(struct facet *, bool may_learn);
 295 static void facet_learn(struct facet *);
 296 static void facet_account(struct facet *);
 297 static void push_all_stats(void);
 298
 299 static bool facet_is_controller_flow(struct facet *);
 300
 301 struct ofport_dpif {
 302     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 303     struct ofport up;
 304
 305     odp_port_t odp_port;
 306     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 307     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
 308     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 309     struct bfd *bfd;            /* BFD, if any. */
 310     bool may_enable;            /* May be enabled in bonds. */
 311     bool is_tunnel;             /* This port is a tunnel. */
 312     long long int carrier_seq;  /* Carrier status changes. */
 313     struct ofport_dpif *peer;   /* Peer if patch port. */
 314
 315     /* Spanning tree. */
 316     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 317     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 318     long long int stp_state_entered;
 319
 320     /* Queue to DSCP mapping. */
 321     struct ofproto_port_queue *qdscp;
 322     size_t n_qdscp;
 323
 324     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 325      *
 326      * This is deprecated.  It is only for compatibility with broken device
 327      * drivers in old versions of Linux that do not properly support VLANs when
 328      * VLAN devices are not used.  When broken device drivers are no longer in
 329      * widespread use, we will delete these interfaces. */
 330     ofp_port_t realdev_ofp_port;
 331     int vlandev_vid;
 332 };
 333
 334 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 335  *
 336  * This is deprecated.  It is only for compatibility with broken device drivers
 337  * in old versions of Linux that do not properly support VLANs when VLAN
 338  * devices are not used.  When broken device drivers are no longer in
 339  * widespread use, we will delete these interfaces. */
 340 struct vlan_splinter {
 341     struct hmap_node realdev_vid_node;
 342     struct hmap_node vlandev_node;
 343     ofp_port_t realdev_ofp_port;
 344     ofp_port_t vlandev_ofp_port;
 345     int vid;
 346 };
 347
 348 static void vsp_remove(struct ofport_dpif *);
 349 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 350
 351 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 352                                        ofp_port_t);
 353
 354 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 355                                        odp_port_t);
 356
 357 static struct ofport_dpif *
 358 ofport_dpif_cast(const struct ofport *ofport)
 359 {
 360     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 361 }
 362
 363 static void port_run(struct ofport_dpif *);
 364 static void port_run_fast(struct ofport_dpif *);
 365 static void port_wait(struct ofport_dpif *);
 366 static int set_bfd(struct ofport *, const struct smap *);
 367 static int set_cfm(struct ofport *, const struct cfm_settings *);
 368 static void ofport_update_peer(struct ofport_dpif *);
 369 static void run_fast_rl(void);
 370 static int run_fast(struct ofproto *);
 371
 372 struct dpif_completion {
 373     struct list list_node;
 374     struct ofoperation *op;
 375 };
 376
 377 /* Reasons that we might need to revalidate every facet, and corresponding
 378  * coverage counters.
 379  *
 380  * A value of 0 means that there is no need to revalidate.
 381  *
 382  * It would be nice to have some cleaner way to integrate with coverage
 383  * counters, but with only a few reasons I guess this is good enough for
 384  * now. */
 385 enum revalidate_reason {
 386     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 387     REV_STP,                   /* Spanning tree protocol port status change. */
 388     REV_BOND,                  /* Bonding changed. */
 389     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 390     REV_FLOW_TABLE,            /* Flow table changed. */
 391     REV_MAC_LEARNING,          /* Mac learning changed. */
 392     REV_INCONSISTENCY          /* Facet self-check failed. */
 393 };
 394 COVERAGE_DEFINE(rev_reconfigure);
 395 COVERAGE_DEFINE(rev_stp);
 396 COVERAGE_DEFINE(rev_bond);
 397 COVERAGE_DEFINE(rev_port_toggled);
 398 COVERAGE_DEFINE(rev_flow_table);
 399 COVERAGE_DEFINE(rev_mac_learning);
 400 COVERAGE_DEFINE(rev_inconsistency);
 401
 402 struct avg_subfacet_rates {
 403     double add_rate;   /* Moving average of new flows created per minute. */
 404     double del_rate;   /* Moving average of flows deleted per minute. */
 405 };
 406
 407 /* All datapaths of a given type share a single dpif backer instance. */
 408 struct dpif_backer {
 409     char *type;
 410     int refcount;
 411     struct dpif *dpif;
 412     struct udpif *udpif;
 413     struct timer next_expiration;
 414
 415     struct ovs_rwlock odp_to_ofport_lock;
 416     struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
 417
 418     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 419
 420     /* Facet revalidation flags applying to facets which use this backer. */
 421     enum revalidate_reason need_revalidate; /* Revalidate every facet. */
 422
 423     struct hmap drop_keys; /* Set of dropped odp keys. */
 424     bool recv_set_enable; /* Enables or disables receiving packets. */
 425
 426     struct hmap subfacets;
 427     struct governor *governor;
 428
 429     /* Subfacet statistics.
 430      *
 431      * These keep track of the total number of subfacets added and deleted and
 432      * flow life span.  They are useful for computing the flow rates stats
 433      * exposed via "ovs-appctl dpif/show".  The goal is to learn about
 434      * traffic patterns in ways that we can use later to improve Open vSwitch
 435      * performance in new situations.  */
 436     long long int created;           /* Time when it is created. */
 437     unsigned max_n_subfacet;         /* Maximum number of flows */
 438     unsigned avg_n_subfacet;         /* Average number of flows. */
 439     long long int avg_subfacet_life; /* Average life span of subfacets. */
 440
 441     /* The average number of subfacets... */
 442     struct avg_subfacet_rates hourly;   /* ...over the last hour. */
 443     struct avg_subfacet_rates daily;    /* ...over the last day. */
 444     struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
 445     long long int last_minute;          /* Last time 'hourly' was updated. */
 446
 447     /* Number of subfacets added or deleted since 'last_minute'. */
 448     unsigned subfacet_add_count;
 449     unsigned subfacet_del_count;
 450
 451     /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
 452     unsigned long long int total_subfacet_add_count;
 453     unsigned long long int total_subfacet_del_count;
 454
 455     /* Number of upcall handling threads. */
 456     unsigned int n_handler_threads;
 457 };
 458
 459 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 460 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 461
 462 static void drop_key_clear(struct dpif_backer *);
 463 static void update_moving_averages(struct dpif_backer *backer);
 464
 465 struct ofproto_dpif {
 466     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 467     struct ofproto up;
 468     struct dpif_backer *backer;
 469
 470     /* Special OpenFlow rules. */
 471     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 472     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 473     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 474
 475     /* Bridging. */
 476     struct netflow *netflow;
 477     struct dpif_sflow *sflow;
 478     struct dpif_ipfix *ipfix;
 479     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 480     struct mac_learning *ml;
 481     bool has_bonded_bundles;
 482     struct mbridge *mbridge;
 483
 484     /* Facets. */
 485     struct classifier facets;     /* Contains 'struct facet's. */
 486     long long int consistency_rl;
 487
 488     struct netdev_stats stats; /* To account packets generated and consumed in
 489                                 * userspace. */
 490
 491     /* Spanning tree. */
 492     struct stp *stp;
 493     long long int stp_last_tick;
 494
 495     /* VLAN splinters. */
 496     struct ovs_mutex vsp_mutex;
 497     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 498     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 499
 500     /* Ports. */
 501     struct sset ports;             /* Set of standard port names. */
 502     struct sset ghost_ports;       /* Ports with no datapath port. */
 503     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 504     int port_poll_errno;           /* Last errno for port_poll() reply. */
 505
 506     /* Per ofproto's dpif stats. */
 507     uint64_t n_hit;
 508     uint64_t n_missed;
 509
 510     /* Work queues. */
 511     struct guarded_list pins;      /* Contains "struct ofputil_packet_in"s. */
 512 };
 513
 514 /* By default, flows in the datapath are wildcarded (megaflows).  They
 515  * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
 516 static bool enable_megaflows = true;
 517
 518 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 519 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 520
 521 static void ofproto_dpif_unixctl_init(void);
 522
 523 static inline struct ofproto_dpif *
 524 ofproto_dpif_cast(const struct ofproto *ofproto)
 525 {
 526     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 527     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 528 }
 529
 530 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
 531                                         ofp_port_t ofp_port);
 532 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
 533                           const struct ofpbuf *packet, struct ds *);
 534
 535 /* Upcalls. */
 536 static void handle_upcalls(struct dpif_backer *);
 537
 538 /* Flow expiration. */
 539 static int expire(struct dpif_backer *);
 540
 541 /* NetFlow. */
 542 static void send_netflow_active_timeouts(struct ofproto_dpif *);
 543
 544 /* Utilities. */
 545 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
 546
 547 /* Global variables. */
 548 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 549
 550 /* Initial mappings of port to bridge mappings. */
 551 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 552
 553 /* Executes 'fm'.  The caller retains ownership of 'fm' and everything in
 554  * it. */
 555 void
 556 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 557                       struct ofputil_flow_mod *fm)
 558 {
 559     ofproto_flow_mod(&ofproto->up, fm);
 560 }
 561
 562 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 563  * Takes ownership of 'pin' and pin->packet. */
 564 void
 565 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 566                             struct ofputil_packet_in *pin)
 567 {
 568     if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
 569         COVERAGE_INC(packet_in_overflow);
 570         free(CONST_CAST(void *, pin->packet));
 571         free(pin);
 572     }
 573 }
 574 \f
 575 /* Factory functions. */
 576
 577 static void
 578 init(const struct shash *iface_hints)
 579 {
 580     struct shash_node *node;
 581
 582     /* Make a local copy, since we don't own 'iface_hints' elements. */
 583     SHASH_FOR_EACH(node, iface_hints) {
 584         const struct iface_hint *orig_hint = node->data;
 585         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 586
 587         new_hint->br_name = xstrdup(orig_hint->br_name);
 588         new_hint->br_type = xstrdup(orig_hint->br_type);
 589         new_hint->ofp_port = orig_hint->ofp_port;
 590
 591         shash_add(&init_ofp_ports, node->name, new_hint);
 592     }
 593 }
 594
 595 static void
 596 enumerate_types(struct sset *types)
 597 {
 598     dp_enumerate_types(types);
 599 }
 600
 601 static int
 602 enumerate_names(const char *type, struct sset *names)
 603 {
 604     struct ofproto_dpif *ofproto;
 605
 606     sset_clear(names);
 607     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 608         if (strcmp(type, ofproto->up.type)) {
 609             continue;
 610         }
 611         sset_add(names, ofproto->up.name);
 612     }
 613
 614     return 0;
 615 }
 616
 617 static int
 618 del(const char *type, const char *name)
 619 {
 620     struct dpif *dpif;
 621     int error;
 622
 623     error = dpif_open(name, type, &dpif);
 624     if (!error) {
 625         error = dpif_delete(dpif);
 626         dpif_close(dpif);
 627     }
 628     return error;
 629 }
 630 \f
 631 static const char *
 632 port_open_type(const char *datapath_type, const char *port_type)
 633 {
 634     return dpif_port_open_type(datapath_type, port_type);
 635 }
 636
 637 /* Type functions. */
 638
 639 static void process_dpif_port_changes(struct dpif_backer *);
 640 static void process_dpif_all_ports_changed(struct dpif_backer *);
 641 static void process_dpif_port_change(struct dpif_backer *,
 642                                      const char *devname);
 643 static void process_dpif_port_error(struct dpif_backer *, int error);
 644
 645 static struct ofproto_dpif *
 646 lookup_ofproto_dpif_by_port_name(const char *name)
 647 {
 648     struct ofproto_dpif *ofproto;
 649
 650     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 651         if (sset_contains(&ofproto->ports, name)) {
 652             return ofproto;
 653         }
 654     }
 655
 656     return NULL;
 657 }
 658
 659 static int
 660 type_run(const char *type)
 661 {
 662     static long long int push_timer = LLONG_MIN;
 663     struct dpif_backer *backer;
 664
 665     backer = shash_find_data(&all_dpif_backers, type);
 666     if (!backer) {
 667         /* This is not necessarily a problem, since backers are only
 668          * created on demand. */
 669         return 0;
 670     }
 671
 672     dpif_run(backer->dpif);
 673
 674     /* The most natural place to push facet statistics is when they're pulled
 675      * from the datapath.  However, when there are many flows in the datapath,
 676      * this expensive operation can occur so frequently, that it reduces our
 677      * ability to quickly set up flows.  To reduce the cost, we push statistics
 678      * here instead. */
 679     if (time_msec() > push_timer) {
 680         push_timer = time_msec() + 2000;
 681         push_all_stats();
 682     }
 683
 684     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 685      * and the configuration has now changed to "false", enable receiving
 686      * packets from the datapath. */
 687     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 688         int error;
 689
 690         backer->recv_set_enable = true;
 691
 692         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 693         if (error) {
 694             udpif_recv_set(backer->udpif, 0, false);
 695             VLOG_ERR("Failed to enable receiving packets in dpif.");
 696             return error;
 697         }
 698         udpif_recv_set(backer->udpif, n_handler_threads,
 699                        backer->recv_set_enable);
 700         dpif_flow_flush(backer->dpif);
 701         backer->need_revalidate = REV_RECONFIGURE;
 702     }
 703
 704     /* If the n_handler_threads is reconfigured, call udpif_recv_set()
 705      * to reset the handler threads. */
 706     if (backer->n_handler_threads != n_handler_threads) {
 707         udpif_recv_set(backer->udpif, n_handler_threads,
 708                        backer->recv_set_enable);
 709         backer->n_handler_threads = n_handler_threads;
 710     }
 711
 712     if (backer->need_revalidate) {
 713         struct ofproto_dpif *ofproto;
 714         struct simap_node *node;
 715         struct simap tmp_backers;
 716
 717         /* Handle tunnel garbage collection. */
 718         simap_init(&tmp_backers);
 719         simap_swap(&backer->tnl_backers, &tmp_backers);
 720
 721         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 722             struct ofport_dpif *iter;
 723
 724             if (backer != ofproto->backer) {
 725                 continue;
 726             }
 727
 728             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 729                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 730                 const char *dp_port;
 731
 732                 if (!iter->is_tunnel) {
 733                     continue;
 734                 }
 735
 736                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 737                                                      namebuf, sizeof namebuf);
 738                 node = simap_find(&tmp_backers, dp_port);
 739                 if (node) {
 740                     simap_put(&backer->tnl_backers, dp_port, node->data);
 741                     simap_delete(&tmp_backers, node);
 742                     node = simap_find(&backer->tnl_backers, dp_port);
 743                 } else {
 744                     node = simap_find(&backer->tnl_backers, dp_port);
 745                     if (!node) {
 746                         odp_port_t odp_port = ODPP_NONE;
 747
 748                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 749                                            &odp_port)) {
 750                             simap_put(&backer->tnl_backers, dp_port,
 751                                       odp_to_u32(odp_port));
 752                             node = simap_find(&backer->tnl_backers, dp_port);
 753                         }
 754                     }
 755                 }
 756
 757                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 758                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 759                                          iter->odp_port)) {
 760                     backer->need_revalidate = REV_RECONFIGURE;
 761                 }
 762             }
 763         }
 764
 765         SIMAP_FOR_EACH (node, &tmp_backers) {
 766             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 767         }
 768         simap_destroy(&tmp_backers);
 769
 770         switch (backer->need_revalidate) {
 771         case REV_RECONFIGURE:   COVERAGE_INC(rev_reconfigure);   break;
 772         case REV_STP:           COVERAGE_INC(rev_stp);           break;
 773         case REV_BOND:          COVERAGE_INC(rev_bond);          break;
 774         case REV_PORT_TOGGLED:  COVERAGE_INC(rev_port_toggled);  break;
 775         case REV_FLOW_TABLE:    COVERAGE_INC(rev_flow_table);    break;
 776         case REV_MAC_LEARNING:  COVERAGE_INC(rev_mac_learning);  break;
 777         case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
 778         }
 779         backer->need_revalidate = 0;
 780
 781         /* Clear the drop_keys in case we should now be accepting some
 782          * formerly dropped flows. */
 783         drop_key_clear(backer);
 784
 785         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 786             struct facet *facet, *next;
 787             struct ofport_dpif *ofport;
 788             struct cls_cursor cursor;
 789             struct ofbundle *bundle;
 790
 791             if (ofproto->backer != backer) {
 792                 continue;
 793             }
 794
 795             ovs_rwlock_wrlock(&xlate_rwlock);
 796             xlate_ofproto_set(ofproto, ofproto->up.name,
 797                               ofproto->backer->dpif, ofproto->miss_rule,
 798                               ofproto->no_packet_in_rule, ofproto->ml,
 799                               ofproto->stp, ofproto->mbridge,
 800                               ofproto->sflow, ofproto->ipfix,
 801                               ofproto->up.frag_handling,
 802                               ofproto->up.forward_bpdu,
 803                               connmgr_has_in_band(ofproto->up.connmgr),
 804                               ofproto->netflow != NULL);
 805
 806             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 807                 xlate_bundle_set(ofproto, bundle, bundle->name,
 808                                  bundle->vlan_mode, bundle->vlan,
 809                                  bundle->trunks, bundle->use_priority_tags,
 810                                  bundle->bond, bundle->lacp,
 811                                  bundle->floodable);
 812             }
 813
 814             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 815                 int stp_port = ofport->stp_port
 816                     ? stp_port_no(ofport->stp_port)
 817                     : -1;
 818                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 819                                  ofport->up.ofp_port, ofport->odp_port,
 820                                  ofport->up.netdev, ofport->cfm,
 821                                  ofport->bfd, ofport->peer, stp_port,
 822                                  ofport->qdscp, ofport->n_qdscp,
 823                                  ofport->up.pp.config, ofport->is_tunnel,
 824                                  ofport->may_enable);
 825             }
 826             ovs_rwlock_unlock(&xlate_rwlock);
 827
 828             /* Only ofproto-dpif cares about the facet classifier so we just
 829              * lock cls_cursor_init() to appease the thread safety analysis. */
 830             ovs_rwlock_rdlock(&ofproto->facets.rwlock);
 831             cls_cursor_init(&cursor, &ofproto->facets, NULL);
 832             ovs_rwlock_unlock(&ofproto->facets.rwlock);
 833             CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
 834                 facet_revalidate(facet);
 835                 run_fast_rl();
 836             }
 837         }
 838
 839         udpif_revalidate(backer->udpif);
 840     }
 841
 842     if (!backer->recv_set_enable) {
 843         /* Wake up before a max of 1000ms. */
 844         timer_set_duration(&backer->next_expiration, 1000);
 845     } else if (timer_expired(&backer->next_expiration)) {
 846         int delay = expire(backer);
 847         timer_set_duration(&backer->next_expiration, delay);
 848     }
 849
 850     process_dpif_port_changes(backer);
 851
 852     if (backer->governor) {
 853         size_t n_subfacets;
 854
 855         governor_run(backer->governor);
 856
 857         /* If the governor has shrunk to its minimum size and the number of
 858          * subfacets has dwindled, then drop the governor entirely.
 859          *
 860          * For hysteresis, the number of subfacets to drop the governor is
 861          * smaller than the number needed to trigger its creation. */
 862         n_subfacets = hmap_count(&backer->subfacets);
 863         if (n_subfacets * 4 < flow_eviction_threshold
 864             && governor_is_idle(backer->governor)) {
 865             governor_destroy(backer->governor);
 866             backer->governor = NULL;
 867         }
 868     }
 869
 870     return 0;
 871 }
 872
 873 /* Check for and handle port changes in 'backer''s dpif. */
 874 static void
 875 process_dpif_port_changes(struct dpif_backer *backer)
 876 {
 877     for (;;) {
 878         char *devname;
 879         int error;
 880
 881         error = dpif_port_poll(backer->dpif, &devname);
 882         switch (error) {
 883         case EAGAIN:
 884             return;
 885
 886         case ENOBUFS:
 887             process_dpif_all_ports_changed(backer);
 888             break;
 889
 890         case 0:
 891             process_dpif_port_change(backer, devname);
 892             free(devname);
 893             break;
 894
 895         default:
 896             process_dpif_port_error(backer, error);
 897             break;
 898         }
 899     }
 900 }
 901
 902 static void
 903 process_dpif_all_ports_changed(struct dpif_backer *backer)
 904 {
 905     struct ofproto_dpif *ofproto;
 906     struct dpif_port dpif_port;
 907     struct dpif_port_dump dump;
 908     struct sset devnames;
 909     const char *devname;
 910
 911     sset_init(&devnames);
 912     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 913         if (ofproto->backer == backer) {
 914             struct ofport *ofport;
 915
 916             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 917                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 918             }
 919         }
 920     }
 921     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 922         sset_add(&devnames, dpif_port.name);
 923     }
 924
 925     SSET_FOR_EACH (devname, &devnames) {
 926         process_dpif_port_change(backer, devname);
 927     }
 928     sset_destroy(&devnames);
 929 }
 930
 931 static void
 932 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 933 {
 934     struct ofproto_dpif *ofproto;
 935     struct dpif_port port;
 936
 937     /* Don't report on the datapath's device. */
 938     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 939         return;
 940     }
 941
 942     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 943                    &all_ofproto_dpifs) {
 944         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 945             return;
 946         }
 947     }
 948
 949     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 950     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 951         /* The port was removed.  If we know the datapath,
 952          * report it through poll_set().  If we don't, it may be
 953          * notifying us of a removal we initiated, so ignore it.
 954          * If there's a pending ENOBUFS, let it stand, since
 955          * everything will be reevaluated. */
 956         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 957             sset_add(&ofproto->port_poll_set, devname);
 958             ofproto->port_poll_errno = 0;
 959         }
 960     } else if (!ofproto) {
 961         /* The port was added, but we don't know with which
 962          * ofproto we should associate it.  Delete it. */
 963         dpif_port_del(backer->dpif, port.port_no);
 964     } else {
 965         struct ofport_dpif *ofport;
 966
 967         ofport = ofport_dpif_cast(shash_find_data(
 968                                       &ofproto->up.port_by_name, devname));
 969         if (ofport
 970             && ofport->odp_port != port.port_no
 971             && !odp_port_to_ofport(backer, port.port_no))
 972         {
 973             /* 'ofport''s datapath port number has changed from
 974              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 975              * structures to match. */
 976             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 977             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 978             ofport->odp_port = port.port_no;
 979             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 980                         hash_odp_port(port.port_no));
 981             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 982             backer->need_revalidate = REV_RECONFIGURE;
 983         }
 984     }
 985     dpif_port_destroy(&port);
 986 }
 987
 988 /* Propagate 'error' to all ofprotos based on 'backer'. */
 989 static void
 990 process_dpif_port_error(struct dpif_backer *backer, int error)
 991 {
 992     struct ofproto_dpif *ofproto;
 993
 994     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 995         if (ofproto->backer == backer) {
 996             sset_clear(&ofproto->port_poll_set);
 997             ofproto->port_poll_errno = error;
 998         }
 999     }
1000 }
1001
1002 static int
1003 dpif_backer_run_fast(struct dpif_backer *backer)
1004 {
1005     handle_upcalls(backer);
1006
1007     return 0;
1008 }
1009
1010 static int
1011 type_run_fast(const char *type)
1012 {
1013     struct dpif_backer *backer;
1014
1015     backer = shash_find_data(&all_dpif_backers, type);
1016     if (!backer) {
1017         /* This is not necessarily a problem, since backers are only
1018          * created on demand. */
1019         return 0;
1020     }
1021
1022     return dpif_backer_run_fast(backer);
1023 }
1024
1025 static void
1026 run_fast_rl(void)
1027 {
1028     static long long int port_rl = LLONG_MIN;
1029
1030     if (time_msec() >= port_rl) {
1031         struct ofproto_dpif *ofproto;
1032
1033         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1034             run_fast(&ofproto->up);
1035         }
1036         port_rl = time_msec() + 200;
1037     }
1038 }
1039
1040 static void
1041 type_wait(const char *type)
1042 {
1043     struct dpif_backer *backer;
1044
1045     backer = shash_find_data(&all_dpif_backers, type);
1046     if (!backer) {
1047         /* This is not necessarily a problem, since backers are only
1048          * created on demand. */
1049         return;
1050     }
1051
1052     if (backer->governor) {
1053         governor_wait(backer->governor);
1054     }
1055
1056     timer_wait(&backer->next_expiration);
1057     dpif_wait(backer->dpif);
1058     udpif_wait(backer->udpif);
1059 }
1060 \f
1061 /* Basic life-cycle. */
1062
1063 static int add_internal_flows(struct ofproto_dpif *);
1064
1065 static struct ofproto *
1066 alloc(void)
1067 {
1068     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1069     return &ofproto->up;
1070 }
1071
1072 static void
1073 dealloc(struct ofproto *ofproto_)
1074 {
1075     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1076     free(ofproto);
1077 }
1078
1079 static void
1080 close_dpif_backer(struct dpif_backer *backer)
1081 {
1082     struct shash_node *node;
1083
1084     ovs_assert(backer->refcount > 0);
1085
1086     if (--backer->refcount) {
1087         return;
1088     }
1089
1090     drop_key_clear(backer);
1091     hmap_destroy(&backer->drop_keys);
1092
1093     simap_destroy(&backer->tnl_backers);
1094     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1095     hmap_destroy(&backer->odp_to_ofport_map);
1096     node = shash_find(&all_dpif_backers, backer->type);
1097     free(backer->type);
1098     shash_delete(&all_dpif_backers, node);
1099     udpif_destroy(backer->udpif);
1100     dpif_close(backer->dpif);
1101
1102     ovs_assert(hmap_is_empty(&backer->subfacets));
1103     hmap_destroy(&backer->subfacets);
1104     governor_destroy(backer->governor);
1105
1106     free(backer);
1107 }
1108
1109 /* Datapath port slated for removal from datapath. */
1110 struct odp_garbage {
1111     struct list list_node;
1112     odp_port_t odp_port;
1113 };
1114
1115 static int
1116 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1117 {
1118     struct dpif_backer *backer;
1119     struct dpif_port_dump port_dump;
1120     struct dpif_port port;
1121     struct shash_node *node;
1122     struct list garbage_list;
1123     struct odp_garbage *garbage, *next;
1124     struct sset names;
1125     char *backer_name;
1126     const char *name;
1127     int error;
1128
1129     backer = shash_find_data(&all_dpif_backers, type);
1130     if (backer) {
1131         backer->refcount++;
1132         *backerp = backer;
1133         return 0;
1134     }
1135
1136     backer_name = xasprintf("ovs-%s", type);
1137
1138     /* Remove any existing datapaths, since we assume we're the only
1139      * userspace controlling the datapath. */
1140     sset_init(&names);
1141     dp_enumerate_names(type, &names);
1142     SSET_FOR_EACH(name, &names) {
1143         struct dpif *old_dpif;
1144
1145         /* Don't remove our backer if it exists. */
1146         if (!strcmp(name, backer_name)) {
1147             continue;
1148         }
1149
1150         if (dpif_open(name, type, &old_dpif)) {
1151             VLOG_WARN("couldn't open old datapath %s to remove it", name);
1152         } else {
1153             dpif_delete(old_dpif);
1154             dpif_close(old_dpif);
1155         }
1156     }
1157     sset_destroy(&names);
1158
1159     backer = xmalloc(sizeof *backer);
1160
1161     error = dpif_create_and_open(backer_name, type, &backer->dpif);
1162     free(backer_name);
1163     if (error) {
1164         VLOG_ERR("failed to open datapath of type %s: %s", type,
1165                  ovs_strerror(error));
1166         free(backer);
1167         return error;
1168     }
1169     backer->udpif = udpif_create(backer, backer->dpif);
1170
1171     backer->type = xstrdup(type);
1172     backer->governor = NULL;
1173     backer->refcount = 1;
1174     hmap_init(&backer->odp_to_ofport_map);
1175     ovs_rwlock_init(&backer->odp_to_ofport_lock);
1176     hmap_init(&backer->drop_keys);
1177     hmap_init(&backer->subfacets);
1178     timer_set_duration(&backer->next_expiration, 1000);
1179     backer->need_revalidate = 0;
1180     simap_init(&backer->tnl_backers);
1181     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1182     *backerp = backer;
1183
1184     if (backer->recv_set_enable) {
1185         dpif_flow_flush(backer->dpif);
1186     }
1187
1188     /* Loop through the ports already on the datapath and remove any
1189      * that we don't need anymore. */
1190     list_init(&garbage_list);
1191     dpif_port_dump_start(&port_dump, backer->dpif);
1192     while (dpif_port_dump_next(&port_dump, &port)) {
1193         node = shash_find(&init_ofp_ports, port.name);
1194         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1195             garbage = xmalloc(sizeof *garbage);
1196             garbage->odp_port = port.port_no;
1197             list_push_front(&garbage_list, &garbage->list_node);
1198         }
1199     }
1200     dpif_port_dump_done(&port_dump);
1201
1202     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1203         dpif_port_del(backer->dpif, garbage->odp_port);
1204         list_remove(&garbage->list_node);
1205         free(garbage);
1206     }
1207
1208     shash_add(&all_dpif_backers, type, backer);
1209
1210     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1211     if (error) {
1212         VLOG_ERR("failed to listen on datapath of type %s: %s",
1213                  type, ovs_strerror(error));
1214         close_dpif_backer(backer);
1215         return error;
1216     }
1217     udpif_recv_set(backer->udpif, n_handler_threads,
1218                    backer->recv_set_enable);
1219     backer->n_handler_threads = n_handler_threads;
1220
1221     backer->max_n_subfacet = 0;
1222     backer->created = time_msec();
1223     backer->last_minute = backer->created;
1224     memset(&backer->hourly, 0, sizeof backer->hourly);
1225     memset(&backer->daily, 0, sizeof backer->daily);
1226     memset(&backer->lifetime, 0, sizeof backer->lifetime);
1227     backer->subfacet_add_count = 0;
1228     backer->subfacet_del_count = 0;
1229     backer->total_subfacet_add_count = 0;
1230     backer->total_subfacet_del_count = 0;
1231     backer->avg_n_subfacet = 0;
1232     backer->avg_subfacet_life = 0;
1233
1234     return error;
1235 }
1236
1237 static int
1238 construct(struct ofproto *ofproto_)
1239 {
1240     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1241     struct shash_node *node, *next;
1242     uint32_t max_ports;
1243     int error;
1244
1245     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1246     if (error) {
1247         return error;
1248     }
1249
1250     max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1251     ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX)));
1252
1253     ofproto->netflow = NULL;
1254     ofproto->sflow = NULL;
1255     ofproto->ipfix = NULL;
1256     ofproto->stp = NULL;
1257     hmap_init(&ofproto->bundles);
1258     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1259     ofproto->mbridge = mbridge_create();
1260     ofproto->has_bonded_bundles = false;
1261     ovs_mutex_init(&ofproto->vsp_mutex);
1262
1263     classifier_init(&ofproto->facets);
1264     ofproto->consistency_rl = LLONG_MIN;
1265
1266     guarded_list_init(&ofproto->pins);
1267
1268     ofproto_dpif_unixctl_init();
1269
1270     hmap_init(&ofproto->vlandev_map);
1271     hmap_init(&ofproto->realdev_vid_map);
1272
1273     sset_init(&ofproto->ports);
1274     sset_init(&ofproto->ghost_ports);
1275     sset_init(&ofproto->port_poll_set);
1276     ofproto->port_poll_errno = 0;
1277
1278     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1279         struct iface_hint *iface_hint = node->data;
1280
1281         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1282             /* Check if the datapath already has this port. */
1283             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1284                 sset_add(&ofproto->ports, node->name);
1285             }
1286
1287             free(iface_hint->br_name);
1288             free(iface_hint->br_type);
1289             free(iface_hint);
1290             shash_delete(&init_ofp_ports, node);
1291         }
1292     }
1293
1294     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1295                 hash_string(ofproto->up.name, 0));
1296     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1297
1298     ofproto_init_tables(ofproto_, N_TABLES);
1299     error = add_internal_flows(ofproto);
1300     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1301
1302     ofproto->n_hit = 0;
1303     ofproto->n_missed = 0;
1304
1305     return error;
1306 }
1307
1308 static int
1309 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1310                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1311 {
1312     struct ofputil_flow_mod fm;
1313     int error;
1314
1315     match_init_catchall(&fm.match);
1316     fm.priority = 0;
1317     match_set_reg(&fm.match, 0, id);
1318     fm.new_cookie = htonll(0);
1319     fm.cookie = htonll(0);
1320     fm.cookie_mask = htonll(0);
1321     fm.modify_cookie = false;
1322     fm.table_id = TBL_INTERNAL;
1323     fm.command = OFPFC_ADD;
1324     fm.idle_timeout = 0;
1325     fm.hard_timeout = 0;
1326     fm.buffer_id = 0;
1327     fm.out_port = 0;
1328     fm.flags = 0;
1329     fm.ofpacts = ofpacts->data;
1330     fm.ofpacts_len = ofpacts->size;
1331
1332     error = ofproto_flow_mod(&ofproto->up, &fm);
1333     if (error) {
1334         VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1335                     id, ofperr_to_string(error));
1336         return error;
1337     }
1338
1339     if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1340                                   rulep)) {
1341         rule_dpif_unref(*rulep);
1342     } else {
1343         NOT_REACHED();
1344     }
1345
1346     return 0;
1347 }
1348
1349 static int
1350 add_internal_flows(struct ofproto_dpif *ofproto)
1351 {
1352     struct ofpact_controller *controller;
1353     uint64_t ofpacts_stub[128 / 8];
1354     struct ofpbuf ofpacts;
1355     int error;
1356     int id;
1357
1358     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1359     id = 1;
1360
1361     controller = ofpact_put_CONTROLLER(&ofpacts);
1362     controller->max_len = UINT16_MAX;
1363     controller->controller_id = 0;
1364     controller->reason = OFPR_NO_MATCH;
1365     ofpact_pad(&ofpacts);
1366
1367     error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1368     if (error) {
1369         return error;
1370     }
1371
1372     ofpbuf_clear(&ofpacts);
1373     error = add_internal_flow(ofproto, id++, &ofpacts,
1374                               &ofproto->no_packet_in_rule);
1375     if (error) {
1376         return error;
1377     }
1378
1379     error = add_internal_flow(ofproto, id++, &ofpacts,
1380                               &ofproto->drop_frags_rule);
1381     return error;
1382 }
1383
1384 static void
1385 destruct(struct ofproto *ofproto_)
1386 {
1387     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1388     struct rule_dpif *rule, *next_rule;
1389     struct ofputil_packet_in *pin, *next_pin;
1390     struct facet *facet, *next_facet;
1391     struct cls_cursor cursor;
1392     struct oftable *table;
1393     struct list pins;
1394
1395     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1396     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1397     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1398     CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1399         facet_remove(facet);
1400     }
1401
1402     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1403     ovs_rwlock_wrlock(&xlate_rwlock);
1404     xlate_remove_ofproto(ofproto);
1405     ovs_rwlock_unlock(&xlate_rwlock);
1406
1407     /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a
1408      * use-after-free error. */
1409     udpif_revalidate(ofproto->backer->udpif);
1410
1411     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1412
1413     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1414         struct cls_cursor cursor;
1415
1416         ovs_rwlock_rdlock(&table->cls.rwlock);
1417         cls_cursor_init(&cursor, &table->cls, NULL);
1418         ovs_rwlock_unlock(&table->cls.rwlock);
1419         CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1420             ofproto_rule_delete(&ofproto->up, &rule->up);
1421         }
1422     }
1423
1424     guarded_list_pop_all(&ofproto->pins, &pins);
1425     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1426         list_remove(&pin->list_node);
1427         free(CONST_CAST(void *, pin->packet));
1428         free(pin);
1429     }
1430     guarded_list_destroy(&ofproto->pins);
1431
1432     mbridge_unref(ofproto->mbridge);
1433
1434     netflow_destroy(ofproto->netflow);
1435     dpif_sflow_unref(ofproto->sflow);
1436     hmap_destroy(&ofproto->bundles);
1437     mac_learning_unref(ofproto->ml);
1438
1439     classifier_destroy(&ofproto->facets);
1440
1441     hmap_destroy(&ofproto->vlandev_map);
1442     hmap_destroy(&ofproto->realdev_vid_map);
1443
1444     sset_destroy(&ofproto->ports);
1445     sset_destroy(&ofproto->ghost_ports);
1446     sset_destroy(&ofproto->port_poll_set);
1447
1448     ovs_mutex_destroy(&ofproto->vsp_mutex);
1449
1450     close_dpif_backer(ofproto->backer);
1451 }
1452
1453 static int
1454 run_fast(struct ofproto *ofproto_)
1455 {
1456     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1457     struct ofputil_packet_in *pin, *next_pin;
1458     struct ofport_dpif *ofport;
1459     struct list pins;
1460
1461     /* Do not perform any periodic activity required by 'ofproto' while
1462      * waiting for flow restore to complete. */
1463     if (ofproto_get_flow_restore_wait()) {
1464         return 0;
1465     }
1466
1467     guarded_list_pop_all(&ofproto->pins, &pins);
1468     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1469         connmgr_send_packet_in(ofproto->up.connmgr, pin);
1470         list_remove(&pin->list_node);
1471         free(CONST_CAST(void *, pin->packet));
1472         free(pin);
1473     }
1474
1475     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1476         port_run_fast(ofport);
1477     }
1478
1479     return 0;
1480 }
1481
1482 static int
1483 run(struct ofproto *ofproto_)
1484 {
1485     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1486     struct ofport_dpif *ofport;
1487     struct ofbundle *bundle;
1488     int error;
1489
1490     if (mbridge_need_revalidate(ofproto->mbridge)) {
1491         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1492         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1493         mac_learning_flush(ofproto->ml);
1494         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1495     }
1496
1497     /* Do not perform any periodic activity below required by 'ofproto' while
1498      * waiting for flow restore to complete. */
1499     if (ofproto_get_flow_restore_wait()) {
1500         return 0;
1501     }
1502
1503     error = run_fast(ofproto_);
1504     if (error) {
1505         return error;
1506     }
1507
1508     if (ofproto->netflow) {
1509         if (netflow_run(ofproto->netflow)) {
1510             send_netflow_active_timeouts(ofproto);
1511         }
1512     }
1513     if (ofproto->sflow) {
1514         dpif_sflow_run(ofproto->sflow);
1515     }
1516     if (ofproto->ipfix) {
1517         dpif_ipfix_run(ofproto->ipfix);
1518     }
1519
1520     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1521         port_run(ofport);
1522     }
1523     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1524         bundle_run(bundle);
1525     }
1526
1527     stp_run(ofproto);
1528     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1529     if (mac_learning_run(ofproto->ml)) {
1530         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1531     }
1532     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1533
1534     /* Check the consistency of a random facet, to aid debugging. */
1535     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1536     if (time_msec() >= ofproto->consistency_rl
1537         && !classifier_is_empty(&ofproto->facets)
1538         && !ofproto->backer->need_revalidate) {
1539         struct cls_table *table;
1540         struct cls_rule *cr;
1541         struct facet *facet;
1542
1543         ofproto->consistency_rl = time_msec() + 250;
1544
1545         table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1546                              struct cls_table, hmap_node);
1547         cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1548                           hmap_node);
1549         facet = CONTAINER_OF(cr, struct facet, cr);
1550
1551         if (!facet_check_consistency(facet)) {
1552             ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1553         }
1554     }
1555     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1556
1557     return 0;
1558 }
1559
1560 static void
1561 wait(struct ofproto *ofproto_)
1562 {
1563     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1564     struct ofport_dpif *ofport;
1565     struct ofbundle *bundle;
1566
1567     if (ofproto_get_flow_restore_wait()) {
1568         return;
1569     }
1570
1571     if (ofproto->sflow) {
1572         dpif_sflow_wait(ofproto->sflow);
1573     }
1574     if (ofproto->ipfix) {
1575         dpif_ipfix_wait(ofproto->ipfix);
1576     }
1577     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1578         port_wait(ofport);
1579     }
1580     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1581         bundle_wait(bundle);
1582     }
1583     if (ofproto->netflow) {
1584         netflow_wait(ofproto->netflow);
1585     }
1586     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1587     mac_learning_wait(ofproto->ml);
1588     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1589     stp_wait(ofproto);
1590     if (ofproto->backer->need_revalidate) {
1591         /* Shouldn't happen, but if it does just go around again. */
1592         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1593         poll_immediate_wake();
1594     }
1595 }
1596
1597 static void
1598 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1599 {
1600     const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1601     struct cls_cursor cursor;
1602     size_t n_subfacets = 0;
1603     struct facet *facet;
1604
1605     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1606     simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1607     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1608
1609     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1610     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1611     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1612         n_subfacets += list_size(&facet->subfacets);
1613     }
1614     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1615     simap_increase(usage, "subfacets", n_subfacets);
1616 }
1617
1618 static void
1619 flush(struct ofproto *ofproto_)
1620 {
1621     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1622     struct subfacet *subfacet, *next_subfacet;
1623     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1624     int n_batch;
1625
1626     n_batch = 0;
1627     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1628                         &ofproto->backer->subfacets) {
1629         if (subfacet->facet->ofproto != ofproto) {
1630             continue;
1631         }
1632
1633         if (subfacet->path != SF_NOT_INSTALLED) {
1634             batch[n_batch++] = subfacet;
1635             if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1636                 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1637                 n_batch = 0;
1638             }
1639         } else {
1640             subfacet_destroy(subfacet);
1641         }
1642     }
1643
1644     if (n_batch > 0) {
1645         subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1646     }
1647 }
1648
1649 static void
1650 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1651              bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1652 {
1653     *arp_match_ip = true;
1654     *actions = (OFPUTIL_A_OUTPUT |
1655                 OFPUTIL_A_SET_VLAN_VID |
1656                 OFPUTIL_A_SET_VLAN_PCP |
1657                 OFPUTIL_A_STRIP_VLAN |
1658                 OFPUTIL_A_SET_DL_SRC |
1659                 OFPUTIL_A_SET_DL_DST |
1660                 OFPUTIL_A_SET_NW_SRC |
1661                 OFPUTIL_A_SET_NW_DST |
1662                 OFPUTIL_A_SET_NW_TOS |
1663                 OFPUTIL_A_SET_TP_SRC |
1664                 OFPUTIL_A_SET_TP_DST |
1665                 OFPUTIL_A_ENQUEUE);
1666 }
1667
1668 static void
1669 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1670 {
1671     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1672     struct dpif_dp_stats s;
1673     uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1674     uint64_t n_lookup;
1675
1676     strcpy(ots->name, "classifier");
1677
1678     dpif_get_dp_stats(ofproto->backer->dpif, &s);
1679     rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1680     rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1681     rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1682
1683     n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1684     ots->lookup_count = htonll(n_lookup);
1685     ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1686 }
1687
1688 static struct ofport *
1689 port_alloc(void)
1690 {
1691     struct ofport_dpif *port = xmalloc(sizeof *port);
1692     return &port->up;
1693 }
1694
1695 static void
1696 port_dealloc(struct ofport *port_)
1697 {
1698     struct ofport_dpif *port = ofport_dpif_cast(port_);
1699     free(port);
1700 }
1701
1702 static int
1703 port_construct(struct ofport *port_)
1704 {
1705     struct ofport_dpif *port = ofport_dpif_cast(port_);
1706     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1707     const struct netdev *netdev = port->up.netdev;
1708     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1709     struct dpif_port dpif_port;
1710     int error;
1711
1712     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1713     port->bundle = NULL;
1714     port->cfm = NULL;
1715     port->bfd = NULL;
1716     port->may_enable = true;
1717     port->stp_port = NULL;
1718     port->stp_state = STP_DISABLED;
1719     port->is_tunnel = false;
1720     port->peer = NULL;
1721     port->qdscp = NULL;
1722     port->n_qdscp = 0;
1723     port->realdev_ofp_port = 0;
1724     port->vlandev_vid = 0;
1725     port->carrier_seq = netdev_get_carrier_resets(netdev);
1726
1727     if (netdev_vport_is_patch(netdev)) {
1728         /* By bailing out here, we don't submit the port to the sFlow module
1729          * to be considered for counter polling export.  This is correct
1730          * because the patch port represents an interface that sFlow considers
1731          * to be "internal" to the switch as a whole, and therefore not an
1732          * candidate for counter polling. */
1733         port->odp_port = ODPP_NONE;
1734         ofport_update_peer(port);
1735         return 0;
1736     }
1737
1738     error = dpif_port_query_by_name(ofproto->backer->dpif,
1739                                     netdev_vport_get_dpif_port(netdev, namebuf,
1740                                                                sizeof namebuf),
1741                                     &dpif_port);
1742     if (error) {
1743         return error;
1744     }
1745
1746     port->odp_port = dpif_port.port_no;
1747
1748     if (netdev_get_tunnel_config(netdev)) {
1749         tnl_port_add(port, port->up.netdev, port->odp_port);
1750         port->is_tunnel = true;
1751     } else {
1752         /* Sanity-check that a mapping doesn't already exist.  This
1753          * shouldn't happen for non-tunnel ports. */
1754         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1755             VLOG_ERR("port %s already has an OpenFlow port number",
1756                      dpif_port.name);
1757             dpif_port_destroy(&dpif_port);
1758             return EBUSY;
1759         }
1760
1761         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1762         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1763                     hash_odp_port(port->odp_port));
1764         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1765     }
1766     dpif_port_destroy(&dpif_port);
1767
1768     if (ofproto->sflow) {
1769         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1770     }
1771
1772     return 0;
1773 }
1774
1775 static void
1776 port_destruct(struct ofport *port_)
1777 {
1778     struct ofport_dpif *port = ofport_dpif_cast(port_);
1779     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1780     const char *devname = netdev_get_name(port->up.netdev);
1781     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1782     const char *dp_port_name;
1783
1784     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1785     ovs_rwlock_wrlock(&xlate_rwlock);
1786     xlate_ofport_remove(port);
1787     ovs_rwlock_unlock(&xlate_rwlock);
1788
1789     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1790                                               sizeof namebuf);
1791     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1792         /* The underlying device is still there, so delete it.  This
1793          * happens when the ofproto is being destroyed, since the caller
1794          * assumes that removal of attached ports will happen as part of
1795          * destruction. */
1796         if (!port->is_tunnel) {
1797             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1798         }
1799     }
1800
1801     if (port->peer) {
1802         port->peer->peer = NULL;
1803         port->peer = NULL;
1804     }
1805
1806     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1807         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1808         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1809         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1810     }
1811
1812     tnl_port_del(port);
1813     sset_find_and_delete(&ofproto->ports, devname);
1814     sset_find_and_delete(&ofproto->ghost_ports, devname);
1815     bundle_remove(port_);
1816     set_cfm(port_, NULL);
1817     set_bfd(port_, NULL);
1818     if (ofproto->sflow) {
1819         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1820     }
1821
1822     free(port->qdscp);
1823 }
1824
1825 static void
1826 port_modified(struct ofport *port_)
1827 {
1828     struct ofport_dpif *port = ofport_dpif_cast(port_);
1829
1830     if (port->bundle && port->bundle->bond) {
1831         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1832     }
1833
1834     if (port->cfm) {
1835         cfm_set_netdev(port->cfm, port->up.netdev);
1836     }
1837
1838     if (port->bfd) {
1839         bfd_set_netdev(port->bfd, port->up.netdev);
1840     }
1841
1842     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1843                                                 port->odp_port)) {
1844         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1845             REV_RECONFIGURE;
1846     }
1847
1848     ofport_update_peer(port);
1849 }
1850
1851 static void
1852 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1853 {
1854     struct ofport_dpif *port = ofport_dpif_cast(port_);
1855     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1856     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1857
1858     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1859                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1860                    OFPUTIL_PC_NO_PACKET_IN)) {
1861         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1862
1863         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1864             bundle_update(port->bundle);
1865         }
1866     }
1867 }
1868
1869 static int
1870 set_sflow(struct ofproto *ofproto_,
1871           const struct ofproto_sflow_options *sflow_options)
1872 {
1873     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1874     struct dpif_sflow *ds = ofproto->sflow;
1875
1876     if (sflow_options) {
1877         if (!ds) {
1878             struct ofport_dpif *ofport;
1879
1880             ds = ofproto->sflow = dpif_sflow_create();
1881             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1882                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1883             }
1884             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1885         }
1886         dpif_sflow_set_options(ds, sflow_options);
1887     } else {
1888         if (ds) {
1889             dpif_sflow_unref(ds);
1890             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1891             ofproto->sflow = NULL;
1892         }
1893     }
1894     return 0;
1895 }
1896
1897 static int
1898 set_ipfix(
1899     struct ofproto *ofproto_,
1900     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1901     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1902     size_t n_flow_exporters_options)
1903 {
1904     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1905     struct dpif_ipfix *di = ofproto->ipfix;
1906     bool has_options = bridge_exporter_options || flow_exporters_options;
1907
1908     if (has_options && !di) {
1909         di = ofproto->ipfix = dpif_ipfix_create();
1910     }
1911
1912     if (di) {
1913         /* Call set_options in any case to cleanly flush the flow
1914          * caches in the last exporters that are to be destroyed. */
1915         dpif_ipfix_set_options(
1916             di, bridge_exporter_options, flow_exporters_options,
1917             n_flow_exporters_options);
1918
1919         if (!has_options) {
1920             dpif_ipfix_unref(di);
1921             ofproto->ipfix = NULL;
1922         }
1923     }
1924
1925     return 0;
1926 }
1927
1928 static int
1929 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1930 {
1931     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1932     int error;
1933
1934     if (!s) {
1935         error = 0;
1936     } else {
1937         if (!ofport->cfm) {
1938             struct ofproto_dpif *ofproto;
1939
1940             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1941             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1942             ofport->cfm = cfm_create(ofport->up.netdev);
1943         }
1944
1945         if (cfm_configure(ofport->cfm, s)) {
1946             return 0;
1947         }
1948
1949         error = EINVAL;
1950     }
1951     cfm_unref(ofport->cfm);
1952     ofport->cfm = NULL;
1953     return error;
1954 }
1955
1956 static bool
1957 get_cfm_status(const struct ofport *ofport_,
1958                struct ofproto_cfm_status *status)
1959 {
1960     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1961
1962     if (ofport->cfm) {
1963         status->faults = cfm_get_fault(ofport->cfm);
1964         status->remote_opstate = cfm_get_opup(ofport->cfm);
1965         status->health = cfm_get_health(ofport->cfm);
1966         cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1967         return true;
1968     } else {
1969         return false;
1970     }
1971 }
1972
1973 static int
1974 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1975 {
1976     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1977     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1978     struct bfd *old;
1979
1980     old = ofport->bfd;
1981     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1982                                 cfg, ofport->up.netdev);
1983     if (ofport->bfd != old) {
1984         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1985     }
1986
1987     return 0;
1988 }
1989
1990 static int
1991 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1992 {
1993     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1994
1995     if (ofport->bfd) {
1996         bfd_get_status(ofport->bfd, smap);
1997         return 0;
1998     } else {
1999         return ENOENT;
2000     }
2001 }
2002 \f
2003 /* Spanning Tree. */
2004
2005 static void
2006 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2007 {
2008     struct ofproto_dpif *ofproto = ofproto_;
2009     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2010     struct ofport_dpif *ofport;
2011
2012     ofport = stp_port_get_aux(sp);
2013     if (!ofport) {
2014         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2015                      ofproto->up.name, port_num);
2016     } else {
2017         struct eth_header *eth = pkt->l2;
2018
2019         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2020         if (eth_addr_is_zero(eth->eth_src)) {
2021             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2022                          "with unknown MAC", ofproto->up.name, port_num);
2023         } else {
2024             send_packet(ofport, pkt);
2025         }
2026     }
2027     ofpbuf_delete(pkt);
2028 }
2029
2030 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2031 static int
2032 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2033 {
2034     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2035
2036     /* Only revalidate flows if the configuration changed. */
2037     if (!s != !ofproto->stp) {
2038         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2039     }
2040
2041     if (s) {
2042         if (!ofproto->stp) {
2043             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2044                                       send_bpdu_cb, ofproto);
2045             ofproto->stp_last_tick = time_msec();
2046         }
2047
2048         stp_set_bridge_id(ofproto->stp, s->system_id);
2049         stp_set_bridge_priority(ofproto->stp, s->priority);
2050         stp_set_hello_time(ofproto->stp, s->hello_time);
2051         stp_set_max_age(ofproto->stp, s->max_age);
2052         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2053     }  else {
2054         struct ofport *ofport;
2055
2056         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2057             set_stp_port(ofport, NULL);
2058         }
2059
2060         stp_unref(ofproto->stp);
2061         ofproto->stp = NULL;
2062     }
2063
2064     return 0;
2065 }
2066
2067 static int
2068 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2069 {
2070     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2071
2072     if (ofproto->stp) {
2073         s->enabled = true;
2074         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2075         s->designated_root = stp_get_designated_root(ofproto->stp);
2076         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2077     } else {
2078         s->enabled = false;
2079     }
2080
2081     return 0;
2082 }
2083
2084 static void
2085 update_stp_port_state(struct ofport_dpif *ofport)
2086 {
2087     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2088     enum stp_state state;
2089
2090     /* Figure out new state. */
2091     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2092                              : STP_DISABLED;
2093
2094     /* Update state. */
2095     if (ofport->stp_state != state) {
2096         enum ofputil_port_state of_state;
2097         bool fwd_change;
2098
2099         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2100                     netdev_get_name(ofport->up.netdev),
2101                     stp_state_name(ofport->stp_state),
2102                     stp_state_name(state));
2103         if (stp_learn_in_state(ofport->stp_state)
2104                 != stp_learn_in_state(state)) {
2105             /* xxx Learning action flows should also be flushed. */
2106             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2107             mac_learning_flush(ofproto->ml);
2108             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2109         }
2110         fwd_change = stp_forward_in_state(ofport->stp_state)
2111                         != stp_forward_in_state(state);
2112
2113         ofproto->backer->need_revalidate = REV_STP;
2114         ofport->stp_state = state;
2115         ofport->stp_state_entered = time_msec();
2116
2117         if (fwd_change && ofport->bundle) {
2118             bundle_update(ofport->bundle);
2119         }
2120
2121         /* Update the STP state bits in the OpenFlow port description. */
2122         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2123         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2124                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2125                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2126                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2127                      : 0);
2128         ofproto_port_set_state(&ofport->up, of_state);
2129     }
2130 }
2131
2132 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2133  * caller is responsible for assigning STP port numbers and ensuring
2134  * there are no duplicates. */
2135 static int
2136 set_stp_port(struct ofport *ofport_,
2137              const struct ofproto_port_stp_settings *s)
2138 {
2139     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2140     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2141     struct stp_port *sp = ofport->stp_port;
2142
2143     if (!s || !s->enable) {
2144         if (sp) {
2145             ofport->stp_port = NULL;
2146             stp_port_disable(sp);
2147             update_stp_port_state(ofport);
2148         }
2149         return 0;
2150     } else if (sp && stp_port_no(sp) != s->port_num
2151             && ofport == stp_port_get_aux(sp)) {
2152         /* The port-id changed, so disable the old one if it's not
2153          * already in use by another port. */
2154         stp_port_disable(sp);
2155     }
2156
2157     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2158     stp_port_enable(sp);
2159
2160     stp_port_set_aux(sp, ofport);
2161     stp_port_set_priority(sp, s->priority);
2162     stp_port_set_path_cost(sp, s->path_cost);
2163
2164     update_stp_port_state(ofport);
2165
2166     return 0;
2167 }
2168
2169 static int
2170 get_stp_port_status(struct ofport *ofport_,
2171                     struct ofproto_port_stp_status *s)
2172 {
2173     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2174     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2175     struct stp_port *sp = ofport->stp_port;
2176
2177     if (!ofproto->stp || !sp) {
2178         s->enabled = false;
2179         return 0;
2180     }
2181
2182     s->enabled = true;
2183     s->port_id = stp_port_get_id(sp);
2184     s->state = stp_port_get_state(sp);
2185     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2186     s->role = stp_port_get_role(sp);
2187     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2188
2189     return 0;
2190 }
2191
2192 static void
2193 stp_run(struct ofproto_dpif *ofproto)
2194 {
2195     if (ofproto->stp) {
2196         long long int now = time_msec();
2197         long long int elapsed = now - ofproto->stp_last_tick;
2198         struct stp_port *sp;
2199
2200         if (elapsed > 0) {
2201             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2202             ofproto->stp_last_tick = now;
2203         }
2204         while (stp_get_changed_port(ofproto->stp, &sp)) {
2205             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2206
2207             if (ofport) {
2208                 update_stp_port_state(ofport);
2209             }
2210         }
2211
2212         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2213             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2214             mac_learning_flush(ofproto->ml);
2215             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2216         }
2217     }
2218 }
2219
2220 static void
2221 stp_wait(struct ofproto_dpif *ofproto)
2222 {
2223     if (ofproto->stp) {
2224         poll_timer_wait(1000);
2225     }
2226 }
2227 \f
2228 static int
2229 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2230            size_t n_qdscp)
2231 {
2232     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2233     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2234
2235     if (ofport->n_qdscp != n_qdscp
2236         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2237                               n_qdscp * sizeof *qdscp))) {
2238         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2239         free(ofport->qdscp);
2240         ofport->qdscp = n_qdscp
2241             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2242             : NULL;
2243         ofport->n_qdscp = n_qdscp;
2244     }
2245
2246     return 0;
2247 }
2248 \f
2249 /* Bundles. */
2250
2251 /* Expires all MAC learning entries associated with 'bundle' and forces its
2252  * ofproto to revalidate every flow.
2253  *
2254  * Normally MAC learning entries are removed only from the ofproto associated
2255  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2256  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2257  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2258  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2259  * with the host from which it migrated. */
2260 static void
2261 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2262 {
2263     struct ofproto_dpif *ofproto = bundle->ofproto;
2264     struct mac_learning *ml = ofproto->ml;
2265     struct mac_entry *mac, *next_mac;
2266
2267     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2268     ovs_rwlock_wrlock(&ml->rwlock);
2269     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2270         if (mac->port.p == bundle) {
2271             if (all_ofprotos) {
2272                 struct ofproto_dpif *o;
2273
2274                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2275                     if (o != ofproto) {
2276                         struct mac_entry *e;
2277
2278                         ovs_rwlock_wrlock(&o->ml->rwlock);
2279                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2280                         if (e) {
2281                             mac_learning_expire(o->ml, e);
2282                         }
2283                         ovs_rwlock_unlock(&o->ml->rwlock);
2284                     }
2285                 }
2286             }
2287
2288             mac_learning_expire(ml, mac);
2289         }
2290     }
2291     ovs_rwlock_unlock(&ml->rwlock);
2292 }
2293
2294 static struct ofbundle *
2295 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2296 {
2297     struct ofbundle *bundle;
2298
2299     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2300                              &ofproto->bundles) {
2301         if (bundle->aux == aux) {
2302             return bundle;
2303         }
2304     }
2305     return NULL;
2306 }
2307
2308 static void
2309 bundle_update(struct ofbundle *bundle)
2310 {
2311     struct ofport_dpif *port;
2312
2313     bundle->floodable = true;
2314     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2315         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2316             || !stp_forward_in_state(port->stp_state)) {
2317             bundle->floodable = false;
2318             break;
2319         }
2320     }
2321 }
2322
2323 static void
2324 bundle_del_port(struct ofport_dpif *port)
2325 {
2326     struct ofbundle *bundle = port->bundle;
2327
2328     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2329
2330     list_remove(&port->bundle_node);
2331     port->bundle = NULL;
2332
2333     if (bundle->lacp) {
2334         lacp_slave_unregister(bundle->lacp, port);
2335     }
2336     if (bundle->bond) {
2337         bond_slave_unregister(bundle->bond, port);
2338     }
2339
2340     bundle_update(bundle);
2341 }
2342
2343 static bool
2344 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2345                 struct lacp_slave_settings *lacp)
2346 {
2347     struct ofport_dpif *port;
2348
2349     port = get_ofp_port(bundle->ofproto, ofp_port);
2350     if (!port) {
2351         return false;
2352     }
2353
2354     if (port->bundle != bundle) {
2355         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2356         if (port->bundle) {
2357             bundle_remove(&port->up);
2358         }
2359
2360         port->bundle = bundle;
2361         list_push_back(&bundle->ports, &port->bundle_node);
2362         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2363             || !stp_forward_in_state(port->stp_state)) {
2364             bundle->floodable = false;
2365         }
2366     }
2367     if (lacp) {
2368         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2369         lacp_slave_register(bundle->lacp, port, lacp);
2370     }
2371
2372     return true;
2373 }
2374
2375 static void
2376 bundle_destroy(struct ofbundle *bundle)
2377 {
2378     struct ofproto_dpif *ofproto;
2379     struct ofport_dpif *port, *next_port;
2380
2381     if (!bundle) {
2382         return;
2383     }
2384
2385     ofproto = bundle->ofproto;
2386     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2387
2388     ovs_rwlock_wrlock(&xlate_rwlock);
2389     xlate_bundle_remove(bundle);
2390     ovs_rwlock_unlock(&xlate_rwlock);
2391
2392     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2393         bundle_del_port(port);
2394     }
2395
2396     bundle_flush_macs(bundle, true);
2397     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2398     free(bundle->name);
2399     free(bundle->trunks);
2400     lacp_unref(bundle->lacp);
2401     bond_unref(bundle->bond);
2402     free(bundle);
2403 }
2404
2405 static int
2406 bundle_set(struct ofproto *ofproto_, void *aux,
2407            const struct ofproto_bundle_settings *s)
2408 {
2409     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2410     bool need_flush = false;
2411     struct ofport_dpif *port;
2412     struct ofbundle *bundle;
2413     unsigned long *trunks;
2414     int vlan;
2415     size_t i;
2416     bool ok;
2417
2418     if (!s) {
2419         bundle_destroy(bundle_lookup(ofproto, aux));
2420         return 0;
2421     }
2422
2423     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2424     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2425
2426     bundle = bundle_lookup(ofproto, aux);
2427     if (!bundle) {
2428         bundle = xmalloc(sizeof *bundle);
2429
2430         bundle->ofproto = ofproto;
2431         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2432                     hash_pointer(aux, 0));
2433         bundle->aux = aux;
2434         bundle->name = NULL;
2435
2436         list_init(&bundle->ports);
2437         bundle->vlan_mode = PORT_VLAN_TRUNK;
2438         bundle->vlan = -1;
2439         bundle->trunks = NULL;
2440         bundle->use_priority_tags = s->use_priority_tags;
2441         bundle->lacp = NULL;
2442         bundle->bond = NULL;
2443
2444         bundle->floodable = true;
2445         mbridge_register_bundle(ofproto->mbridge, bundle);
2446     }
2447
2448     if (!bundle->name || strcmp(s->name, bundle->name)) {
2449         free(bundle->name);
2450         bundle->name = xstrdup(s->name);
2451     }
2452
2453     /* LACP. */
2454     if (s->lacp) {
2455         if (!bundle->lacp) {
2456             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2457             bundle->lacp = lacp_create();
2458         }
2459         lacp_configure(bundle->lacp, s->lacp);
2460     } else {
2461         lacp_unref(bundle->lacp);
2462         bundle->lacp = NULL;
2463     }
2464
2465     /* Update set of ports. */
2466     ok = true;
2467     for (i = 0; i < s->n_slaves; i++) {
2468         if (!bundle_add_port(bundle, s->slaves[i],
2469                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2470             ok = false;
2471         }
2472     }
2473     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2474         struct ofport_dpif *next_port;
2475
2476         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2477             for (i = 0; i < s->n_slaves; i++) {
2478                 if (s->slaves[i] == port->up.ofp_port) {
2479                     goto found;
2480                 }
2481             }
2482
2483             bundle_del_port(port);
2484         found: ;
2485         }
2486     }
2487     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2488
2489     if (list_is_empty(&bundle->ports)) {
2490         bundle_destroy(bundle);
2491         return EINVAL;
2492     }
2493
2494     /* Set VLAN tagging mode */
2495     if (s->vlan_mode != bundle->vlan_mode
2496         || s->use_priority_tags != bundle->use_priority_tags) {
2497         bundle->vlan_mode = s->vlan_mode;
2498         bundle->use_priority_tags = s->use_priority_tags;
2499         need_flush = true;
2500     }
2501
2502     /* Set VLAN tag. */
2503     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2504             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2505             : 0);
2506     if (vlan != bundle->vlan) {
2507         bundle->vlan = vlan;
2508         need_flush = true;
2509     }
2510
2511     /* Get trunked VLANs. */
2512     switch (s->vlan_mode) {
2513     case PORT_VLAN_ACCESS:
2514         trunks = NULL;
2515         break;
2516
2517     case PORT_VLAN_TRUNK:
2518         trunks = CONST_CAST(unsigned long *, s->trunks);
2519         break;
2520
2521     case PORT_VLAN_NATIVE_UNTAGGED:
2522     case PORT_VLAN_NATIVE_TAGGED:
2523         if (vlan != 0 && (!s->trunks
2524                           || !bitmap_is_set(s->trunks, vlan)
2525                           || bitmap_is_set(s->trunks, 0))) {
2526             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2527             if (s->trunks) {
2528                 trunks = bitmap_clone(s->trunks, 4096);
2529             } else {
2530                 trunks = bitmap_allocate1(4096);
2531             }
2532             bitmap_set1(trunks, vlan);
2533             bitmap_set0(trunks, 0);
2534         } else {
2535             trunks = CONST_CAST(unsigned long *, s->trunks);
2536         }
2537         break;
2538
2539     default:
2540         NOT_REACHED();
2541     }
2542     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2543         free(bundle->trunks);
2544         if (trunks == s->trunks) {
2545             bundle->trunks = vlan_bitmap_clone(trunks);
2546         } else {
2547             bundle->trunks = trunks;
2548             trunks = NULL;
2549         }
2550         need_flush = true;
2551     }
2552     if (trunks != s->trunks) {
2553         free(trunks);
2554     }
2555
2556     /* Bonding. */
2557     if (!list_is_short(&bundle->ports)) {
2558         bundle->ofproto->has_bonded_bundles = true;
2559         if (bundle->bond) {
2560             if (bond_reconfigure(bundle->bond, s->bond)) {
2561                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2562             }
2563         } else {
2564             bundle->bond = bond_create(s->bond);
2565             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2566         }
2567
2568         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2569             bond_slave_register(bundle->bond, port, port->up.netdev);
2570         }
2571     } else {
2572         bond_unref(bundle->bond);
2573         bundle->bond = NULL;
2574     }
2575
2576     /* If we changed something that would affect MAC learning, un-learn
2577      * everything on this port and force flow revalidation. */
2578     if (need_flush) {
2579         bundle_flush_macs(bundle, false);
2580     }
2581
2582     return 0;
2583 }
2584
2585 static void
2586 bundle_remove(struct ofport *port_)
2587 {
2588     struct ofport_dpif *port = ofport_dpif_cast(port_);
2589     struct ofbundle *bundle = port->bundle;
2590
2591     if (bundle) {
2592         bundle_del_port(port);
2593         if (list_is_empty(&bundle->ports)) {
2594             bundle_destroy(bundle);
2595         } else if (list_is_short(&bundle->ports)) {
2596             bond_unref(bundle->bond);
2597             bundle->bond = NULL;
2598         }
2599     }
2600 }
2601
2602 static void
2603 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2604 {
2605     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2606     struct ofport_dpif *port = port_;
2607     uint8_t ea[ETH_ADDR_LEN];
2608     int error;
2609
2610     error = netdev_get_etheraddr(port->up.netdev, ea);
2611     if (!error) {
2612         struct ofpbuf packet;
2613         void *packet_pdu;
2614
2615         ofpbuf_init(&packet, 0);
2616         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2617                                  pdu_size);
2618         memcpy(packet_pdu, pdu, pdu_size);
2619
2620         send_packet(port, &packet);
2621         ofpbuf_uninit(&packet);
2622     } else {
2623         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2624                     "%s (%s)", port->bundle->name,
2625                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2626     }
2627 }
2628
2629 static void
2630 bundle_send_learning_packets(struct ofbundle *bundle)
2631 {
2632     struct ofproto_dpif *ofproto = bundle->ofproto;
2633     struct ofpbuf *learning_packet;
2634     int error, n_packets, n_errors;
2635     struct mac_entry *e;
2636     struct list packets;
2637
2638     list_init(&packets);
2639     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2640     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2641         if (e->port.p != bundle) {
2642             void *port_void;
2643
2644             learning_packet = bond_compose_learning_packet(bundle->bond,
2645                                                            e->mac, e->vlan,
2646                                                            &port_void);
2647             learning_packet->private_p = port_void;
2648             list_push_back(&packets, &learning_packet->list_node);
2649         }
2650     }
2651     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2652
2653     error = n_packets = n_errors = 0;
2654     LIST_FOR_EACH (learning_packet, list_node, &packets) {
2655         int ret;
2656
2657         ret = send_packet(learning_packet->private_p, learning_packet);
2658         if (ret) {
2659             error = ret;
2660             n_errors++;
2661         }
2662         n_packets++;
2663     }
2664     ofpbuf_list_delete(&packets);
2665
2666     if (n_errors) {
2667         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2668         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2669                      "packets, last error was: %s",
2670                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2671     } else {
2672         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2673                  bundle->name, n_packets);
2674     }
2675 }
2676
2677 static void
2678 bundle_run(struct ofbundle *bundle)
2679 {
2680     if (bundle->lacp) {
2681         lacp_run(bundle->lacp, send_pdu_cb);
2682     }
2683     if (bundle->bond) {
2684         struct ofport_dpif *port;
2685
2686         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2687             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2688         }
2689
2690         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2691             bundle->ofproto->backer->need_revalidate = REV_BOND;
2692         }
2693
2694         if (bond_should_send_learning_packets(bundle->bond)) {
2695             bundle_send_learning_packets(bundle);
2696         }
2697     }
2698 }
2699
2700 static void
2701 bundle_wait(struct ofbundle *bundle)
2702 {
2703     if (bundle->lacp) {
2704         lacp_wait(bundle->lacp);
2705     }
2706     if (bundle->bond) {
2707         bond_wait(bundle->bond);
2708     }
2709 }
2710 \f
2711 /* Mirrors. */
2712
2713 static int
2714 mirror_set__(struct ofproto *ofproto_, void *aux,
2715              const struct ofproto_mirror_settings *s)
2716 {
2717     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2718     struct ofbundle **srcs, **dsts;
2719     int error;
2720     size_t i;
2721
2722     if (!s) {
2723         mirror_destroy(ofproto->mbridge, aux);
2724         return 0;
2725     }
2726
2727     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2728     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2729
2730     for (i = 0; i < s->n_srcs; i++) {
2731         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2732     }
2733
2734     for (i = 0; i < s->n_dsts; i++) {
2735         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2736     }
2737
2738     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2739                        s->n_dsts, s->src_vlans,
2740                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2741     free(srcs);
2742     free(dsts);
2743     return error;
2744 }
2745
2746 static int
2747 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2748                    uint64_t *packets, uint64_t *bytes)
2749 {
2750     push_all_stats();
2751     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2752                             bytes);
2753 }
2754
2755 static int
2756 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2757 {
2758     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2759     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2760     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2761         mac_learning_flush(ofproto->ml);
2762     }
2763     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2764     return 0;
2765 }
2766
2767 static bool
2768 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2769 {
2770     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2771     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2772     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2773 }
2774
2775 static void
2776 forward_bpdu_changed(struct ofproto *ofproto_)
2777 {
2778     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2779     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2780 }
2781
2782 static void
2783 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2784                      size_t max_entries)
2785 {
2786     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2787     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2788     mac_learning_set_idle_time(ofproto->ml, idle_time);
2789     mac_learning_set_max_entries(ofproto->ml, max_entries);
2790     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2791 }
2792 \f
2793 /* Ports. */
2794
2795 static struct ofport_dpif *
2796 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2797 {
2798     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2799     return ofport ? ofport_dpif_cast(ofport) : NULL;
2800 }
2801
2802 static struct ofport_dpif *
2803 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2804 {
2805     struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2806     return port && &ofproto->up == port->up.ofproto ? port : NULL;
2807 }
2808
2809 static void
2810 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2811                             struct ofproto_port *ofproto_port,
2812                             struct dpif_port *dpif_port)
2813 {
2814     ofproto_port->name = dpif_port->name;
2815     ofproto_port->type = dpif_port->type;
2816     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2817 }
2818
2819 static void
2820 ofport_update_peer(struct ofport_dpif *ofport)
2821 {
2822     const struct ofproto_dpif *ofproto;
2823     struct dpif_backer *backer;
2824     char *peer_name;
2825
2826     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2827         return;
2828     }
2829
2830     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2831     backer->need_revalidate = REV_RECONFIGURE;
2832
2833     if (ofport->peer) {
2834         ofport->peer->peer = NULL;
2835         ofport->peer = NULL;
2836     }
2837
2838     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2839     if (!peer_name) {
2840         return;
2841     }
2842
2843     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2844         struct ofport *peer_ofport;
2845         struct ofport_dpif *peer;
2846         char *peer_peer;
2847
2848         if (ofproto->backer != backer) {
2849             continue;
2850         }
2851
2852         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2853         if (!peer_ofport) {
2854             continue;
2855         }
2856
2857         peer = ofport_dpif_cast(peer_ofport);
2858         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2859         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2860                                  peer_peer)) {
2861             ofport->peer = peer;
2862             ofport->peer->peer = ofport;
2863         }
2864         free(peer_peer);
2865
2866         break;
2867     }
2868     free(peer_name);
2869 }
2870
2871 static void
2872 port_run_fast(struct ofport_dpif *ofport)
2873 {
2874     if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2875         struct ofpbuf packet;
2876
2877         ofpbuf_init(&packet, 0);
2878         cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2879         send_packet(ofport, &packet);
2880         ofpbuf_uninit(&packet);
2881     }
2882
2883     if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2884         struct ofpbuf packet;
2885
2886         ofpbuf_init(&packet, 0);
2887         bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2888         send_packet(ofport, &packet);
2889         ofpbuf_uninit(&packet);
2890     }
2891 }
2892
2893 static void
2894 port_run(struct ofport_dpif *ofport)
2895 {
2896     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2897     bool carrier_changed = carrier_seq != ofport->carrier_seq;
2898     bool enable = netdev_get_carrier(ofport->up.netdev);
2899     bool cfm_enable = false;
2900     bool bfd_enable = false;
2901
2902     ofport->carrier_seq = carrier_seq;
2903
2904     port_run_fast(ofport);
2905
2906     if (ofport->cfm) {
2907         int cfm_opup = cfm_get_opup(ofport->cfm);
2908
2909         cfm_run(ofport->cfm);
2910         cfm_enable = !cfm_get_fault(ofport->cfm);
2911
2912         if (cfm_opup >= 0) {
2913             cfm_enable = cfm_enable && cfm_opup;
2914         }
2915     }
2916
2917     if (ofport->bfd) {
2918         bfd_run(ofport->bfd);
2919         bfd_enable = bfd_forwarding(ofport->bfd);
2920     }
2921
2922     if (ofport->bfd || ofport->cfm) {
2923         enable = enable && (cfm_enable || bfd_enable);
2924     }
2925
2926     if (ofport->bundle) {
2927         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2928         if (carrier_changed) {
2929             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2930         }
2931     }
2932
2933     if (ofport->may_enable != enable) {
2934         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2935         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2936     }
2937
2938     ofport->may_enable = enable;
2939 }
2940
2941 static void
2942 port_wait(struct ofport_dpif *ofport)
2943 {
2944     if (ofport->cfm) {
2945         cfm_wait(ofport->cfm);
2946     }
2947
2948     if (ofport->bfd) {
2949         bfd_wait(ofport->bfd);
2950     }
2951 }
2952
2953 static int
2954 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2955                    struct ofproto_port *ofproto_port)
2956 {
2957     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2958     struct dpif_port dpif_port;
2959     int error;
2960
2961     if (sset_contains(&ofproto->ghost_ports, devname)) {
2962         const char *type = netdev_get_type_from_name(devname);
2963
2964         /* We may be called before ofproto->up.port_by_name is populated with
2965          * the appropriate ofport.  For this reason, we must get the name and
2966          * type from the netdev layer directly. */
2967         if (type) {
2968             const struct ofport *ofport;
2969
2970             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2971             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2972             ofproto_port->name = xstrdup(devname);
2973             ofproto_port->type = xstrdup(type);
2974             return 0;
2975         }
2976         return ENODEV;
2977     }
2978
2979     if (!sset_contains(&ofproto->ports, devname)) {
2980         return ENODEV;
2981     }
2982     error = dpif_port_query_by_name(ofproto->backer->dpif,
2983                                     devname, &dpif_port);
2984     if (!error) {
2985         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2986     }
2987     return error;
2988 }
2989
2990 static int
2991 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2992 {
2993     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2994     const char *devname = netdev_get_name(netdev);
2995     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2996     const char *dp_port_name;
2997
2998     if (netdev_vport_is_patch(netdev)) {
2999         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3000         return 0;
3001     }
3002
3003     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3004     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3005         odp_port_t port_no = ODPP_NONE;
3006         int error;
3007
3008         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3009         if (error) {
3010             return error;
3011         }
3012         if (netdev_get_tunnel_config(netdev)) {
3013             simap_put(&ofproto->backer->tnl_backers,
3014                       dp_port_name, odp_to_u32(port_no));
3015         }
3016     }
3017
3018     if (netdev_get_tunnel_config(netdev)) {
3019         sset_add(&ofproto->ghost_ports, devname);
3020     } else {
3021         sset_add(&ofproto->ports, devname);
3022     }
3023     return 0;
3024 }
3025
3026 static int
3027 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3028 {
3029     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3030     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3031     int error = 0;
3032
3033     if (!ofport) {
3034         return 0;
3035     }
3036
3037     sset_find_and_delete(&ofproto->ghost_ports,
3038                          netdev_get_name(ofport->up.netdev));
3039     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3040     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3041         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3042         if (!error) {
3043             /* The caller is going to close ofport->up.netdev.  If this is a
3044              * bonded port, then the bond is using that netdev, so remove it
3045              * from the bond.  The client will need to reconfigure everything
3046              * after deleting ports, so then the slave will get re-added. */
3047             bundle_remove(&ofport->up);
3048         }
3049     }
3050     return error;
3051 }
3052
3053 static int
3054 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3055 {
3056     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3057     int error;
3058
3059     push_all_stats();
3060
3061     error = netdev_get_stats(ofport->up.netdev, stats);
3062
3063     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3064         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3065
3066         /* ofproto->stats.tx_packets represents packets that we created
3067          * internally and sent to some port (e.g. packets sent with
3068          * send_packet()).  Account for them as if they had come from
3069          * OFPP_LOCAL and got forwarded. */
3070
3071         if (stats->rx_packets != UINT64_MAX) {
3072             stats->rx_packets += ofproto->stats.tx_packets;
3073         }
3074
3075         if (stats->rx_bytes != UINT64_MAX) {
3076             stats->rx_bytes += ofproto->stats.tx_bytes;
3077         }
3078
3079         /* ofproto->stats.rx_packets represents packets that were received on
3080          * some port and we processed internally and dropped (e.g. STP).
3081          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3082
3083         if (stats->tx_packets != UINT64_MAX) {
3084             stats->tx_packets += ofproto->stats.rx_packets;
3085         }
3086
3087         if (stats->tx_bytes != UINT64_MAX) {
3088             stats->tx_bytes += ofproto->stats.rx_bytes;
3089         }
3090     }
3091
3092     return error;
3093 }
3094
3095 struct port_dump_state {
3096     uint32_t bucket;
3097     uint32_t offset;
3098     bool ghost;
3099
3100     struct ofproto_port port;
3101     bool has_port;
3102 };
3103
3104 static int
3105 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3106 {
3107     *statep = xzalloc(sizeof(struct port_dump_state));
3108     return 0;
3109 }
3110
3111 static int
3112 port_dump_next(const struct ofproto *ofproto_, void *state_,
3113                struct ofproto_port *port)
3114 {
3115     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3116     struct port_dump_state *state = state_;
3117     const struct sset *sset;
3118     struct sset_node *node;
3119
3120     if (state->has_port) {
3121         ofproto_port_destroy(&state->port);
3122         state->has_port = false;
3123     }
3124     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3125     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3126         int error;
3127
3128         error = port_query_by_name(ofproto_, node->name, &state->port);
3129         if (!error) {
3130             *port = state->port;
3131             state->has_port = true;
3132             return 0;
3133         } else if (error != ENODEV) {
3134             return error;
3135         }
3136     }
3137
3138     if (!state->ghost) {
3139         state->ghost = true;
3140         state->bucket = 0;
3141         state->offset = 0;
3142         return port_dump_next(ofproto_, state_, port);
3143     }
3144
3145     return EOF;
3146 }
3147
3148 static int
3149 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3150 {
3151     struct port_dump_state *state = state_;
3152
3153     if (state->has_port) {
3154         ofproto_port_destroy(&state->port);
3155     }
3156     free(state);
3157     return 0;
3158 }
3159
3160 static int
3161 port_poll(const struct ofproto *ofproto_, char **devnamep)
3162 {
3163     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3164
3165     if (ofproto->port_poll_errno) {
3166         int error = ofproto->port_poll_errno;
3167         ofproto->port_poll_errno = 0;
3168         return error;
3169     }
3170
3171     if (sset_is_empty(&ofproto->port_poll_set)) {
3172         return EAGAIN;
3173     }
3174
3175     *devnamep = sset_pop(&ofproto->port_poll_set);
3176     return 0;
3177 }
3178
3179 static void
3180 port_poll_wait(const struct ofproto *ofproto_)
3181 {
3182     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3183     dpif_port_poll_wait(ofproto->backer->dpif);
3184 }
3185
3186 static int
3187 port_is_lacp_current(const struct ofport *ofport_)
3188 {
3189     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3190     return (ofport->bundle && ofport->bundle->lacp
3191             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3192             : -1);
3193 }
3194 \f
3195 /* Upcall handling. */
3196
3197 struct flow_miss_op {
3198     struct dpif_op dpif_op;
3199
3200     uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3201     struct xlate_out xout;
3202     bool xout_garbage;           /* 'xout' needs to be uninitialized? */
3203
3204     struct ofpbuf mask;          /* Flow mask for "put" ops. */
3205     struct odputil_keybuf maskbuf;
3206
3207     /* If this is a "put" op, then a pointer to the subfacet that should
3208      * be marked as uninstalled if the operation fails. */
3209     struct subfacet *subfacet;
3210 };
3211
3212 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3213  * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3214  * and (usually) installing a datapath flow.  The answer is usually "yes" (a
3215  * return value of true).  However, for short flows the cost of bookkeeping is
3216  * much higher than the benefits, so when the datapath holds a large number of
3217  * flows we impose some heuristics to decide which flows are likely to be worth
3218  * tracking. */
3219 static bool
3220 flow_miss_should_make_facet(struct flow_miss *miss)
3221 {
3222     struct dpif_backer *backer = miss->ofproto->backer;
3223     uint32_t hash;
3224
3225     switch (flow_miss_model) {
3226     case OFPROTO_HANDLE_MISS_AUTO:
3227         break;
3228     case OFPROTO_HANDLE_MISS_WITH_FACETS:
3229         return true;
3230     case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3231         return false;
3232     }
3233
3234     if (!backer->governor) {
3235         size_t n_subfacets;
3236
3237         n_subfacets = hmap_count(&backer->subfacets);
3238         if (n_subfacets * 2 <= flow_eviction_threshold) {
3239             return true;
3240         }
3241
3242         backer->governor = governor_create();
3243     }
3244
3245     hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3246     return governor_should_install_flow(backer->governor, hash,
3247                                         list_size(&miss->packets));
3248 }
3249
3250 /* Handles 'miss', which matches 'facet'.  May add any required datapath
3251  * operations to 'ops', incrementing '*n_ops' for each new op.
3252  *
3253  * All of the packets in 'miss' are considered to have arrived at time
3254  * 'miss->stats.used'.  This is really important only for new facets: if we
3255  * just called time_msec() here, then the new subfacet or its packets could
3256  * look (occasionally) as though it was used some time after the facet was
3257  * used.  That can make a one-packet flow look like it has a nonzero duration,
3258  * which looks odd in e.g. NetFlow statistics. */
3259 static void
3260 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3261                             struct flow_miss_op *ops, size_t *n_ops)
3262 {
3263     enum subfacet_path want_path;
3264     struct subfacet *subfacet;
3265
3266     facet->packet_count += miss->stats.n_packets;
3267     facet->prev_packet_count += miss->stats.n_packets;
3268     facet->byte_count += miss->stats.n_bytes;
3269     facet->prev_byte_count += miss->stats.n_bytes;
3270
3271     want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3272
3273     /* Don't install the flow if it's the result of the "userspace"
3274      * action for an already installed facet.  This can occur when a
3275      * datapath flow with wildcards has a "userspace" action and flows
3276      * sent to userspace result in a different subfacet, which will then
3277      * be rejected as overlapping by the datapath. */
3278     if (miss->upcall_type == DPIF_UC_ACTION
3279         && !list_is_empty(&facet->subfacets)) {
3280         return;
3281     }
3282
3283     subfacet = subfacet_create(facet, miss);
3284     if (subfacet->path != want_path) {
3285         struct flow_miss_op *op = &ops[(*n_ops)++];
3286         struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3287
3288         subfacet->path = want_path;
3289
3290         ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3291         if (enable_megaflows) {
3292             odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3293                                    &miss->flow, UINT32_MAX);
3294         }
3295
3296         op->xout_garbage = false;
3297         op->dpif_op.type = DPIF_OP_FLOW_PUT;
3298         op->subfacet = subfacet;
3299         put->flags = DPIF_FP_CREATE;
3300         put->key = miss->key;
3301         put->key_len = miss->key_len;
3302         put->mask = op->mask.data;
3303         put->mask_len = op->mask.size;
3304
3305         if (want_path == SF_FAST_PATH) {
3306             put->actions = facet->xout.odp_actions.data;
3307             put->actions_len = facet->xout.odp_actions.size;
3308         } else {
3309             compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3310                               op->slow_stub, sizeof op->slow_stub,
3311                               &put->actions, &put->actions_len);
3312         }
3313         put->stats = NULL;
3314     }
3315 }
3316
3317 /* Handles flow miss 'miss'.  May add any required datapath operations
3318  * to 'ops', incrementing '*n_ops' for each new op. */
3319 static void
3320 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3321                  size_t *n_ops)
3322 {
3323     struct facet *facet;
3324
3325     miss->ofproto->n_missed += list_size(&miss->packets);
3326
3327     facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3328     if (!facet) {
3329         /* There does not exist a bijection between 'struct flow' and datapath
3330          * flow keys with fitness ODP_FIT_TO_LITTLE.  This breaks a fundamental
3331          * assumption used throughout the facet and subfacet handling code.
3332          * Since we have to handle these misses in userspace anyway, we simply
3333          * skip facet creation, avoiding the problem altogether. */
3334         if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3335             || !flow_miss_should_make_facet(miss)) {
3336             return;
3337         }
3338
3339         facet = facet_create(miss);
3340     }
3341     handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3342 }
3343
3344 static struct drop_key *
3345 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3346                 size_t key_len)
3347 {
3348     struct drop_key *drop_key;
3349
3350     HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3351                              &backer->drop_keys) {
3352         if (drop_key->key_len == key_len
3353             && !memcmp(drop_key->key, key, key_len)) {
3354             return drop_key;
3355         }
3356     }
3357     return NULL;
3358 }
3359
3360 static void
3361 drop_key_clear(struct dpif_backer *backer)
3362 {
3363     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3364     struct drop_key *drop_key, *next;
3365
3366     HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3367         int error;
3368
3369         error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3370                               NULL);
3371         if (error && !VLOG_DROP_WARN(&rl)) {
3372             struct ds ds = DS_EMPTY_INITIALIZER;
3373             odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3374             VLOG_WARN("Failed to delete drop key (%s) (%s)",
3375                       ovs_strerror(error), ds_cstr(&ds));
3376             ds_destroy(&ds);
3377         }
3378
3379         hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3380         drop_key_destroy(drop_key);
3381     }
3382
3383     udpif_drop_key_clear(backer->udpif);
3384 }
3385
3386 static void
3387 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3388 {
3389     struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3390     struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3391     struct flow_miss *miss;
3392     size_t n_ops, i;
3393
3394     /* Process each element in the to-do list, constructing the set of
3395      * operations to batch. */
3396     n_ops = 0;
3397     HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3398         handle_flow_miss(miss, flow_miss_ops, &n_ops);
3399     }
3400     ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3401
3402     /* Execute batch. */
3403     for (i = 0; i < n_ops; i++) {
3404         dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3405     }
3406     dpif_operate(backer->dpif, dpif_ops, n_ops);
3407
3408     for (i = 0; i < n_ops; i++) {
3409         if (dpif_ops[i]->error != 0
3410             && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3411             && flow_miss_ops[i].subfacet) {
3412             struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3413
3414             COVERAGE_INC(subfacet_install_fail);
3415
3416             /* Zero-out subfacet counters when installation failed, but
3417              * datapath reported hits.  This should not happen and
3418              * indicates a bug, since if the datapath flow exists, we
3419              * should not be attempting to create a new subfacet.  A
3420              * buggy datapath could trigger this, so just zero out the
3421              * counters and log an error. */
3422             if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3423                 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3424                             "datapath reported hits");
3425                 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3426             }
3427
3428             subfacet->path = SF_NOT_INSTALLED;
3429         }
3430     }
3431 }
3432
3433 static void
3434 handle_sflow_upcall(struct dpif_backer *backer,
3435                     const struct dpif_upcall *upcall)
3436 {
3437     struct ofproto_dpif *ofproto;
3438     union user_action_cookie cookie;
3439     struct flow flow;
3440     odp_port_t odp_in_port;
3441
3442     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3443                       &flow, NULL, &ofproto, &odp_in_port)
3444         || !ofproto->sflow) {
3445         return;
3446     }
3447
3448     memset(&cookie, 0, sizeof cookie);
3449     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3450     dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3451                         odp_in_port, &cookie);
3452 }
3453
3454 static void
3455 handle_flow_sample_upcall(struct dpif_backer *backer,
3456                           const struct dpif_upcall *upcall)
3457 {
3458     struct ofproto_dpif *ofproto;
3459     union user_action_cookie cookie;
3460     struct flow flow;
3461
3462     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3463                       &flow, NULL, &ofproto, NULL)
3464         || !ofproto->ipfix) {
3465         return;
3466     }
3467
3468     memset(&cookie, 0, sizeof cookie);
3469     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3470
3471     /* The flow reflects exactly the contents of the packet.  Sample
3472      * the packet using it. */
3473     dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3474                            cookie.flow_sample.collector_set_id,
3475                            cookie.flow_sample.probability,
3476                            cookie.flow_sample.obs_domain_id,
3477                            cookie.flow_sample.obs_point_id);
3478 }
3479
3480 static void
3481 handle_ipfix_upcall(struct dpif_backer *backer,
3482                     const struct dpif_upcall *upcall)
3483 {
3484     struct ofproto_dpif *ofproto;
3485     struct flow flow;
3486
3487     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3488                       &flow, NULL, &ofproto, NULL)
3489         || !ofproto->ipfix) {
3490         return;
3491     }
3492
3493     /* The flow reflects exactly the contents of the packet.  Sample
3494      * the packet using it. */
3495     dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3496 }
3497
3498 static void
3499 handle_upcalls(struct dpif_backer *backer)
3500 {
3501     struct flow_miss_batch *fmb;
3502     int n_processed;
3503
3504     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3505         struct upcall *upcall = upcall_next(backer->udpif);
3506
3507         if (!upcall) {
3508             break;
3509         }
3510
3511         switch (upcall->type) {
3512         case SFLOW_UPCALL:
3513             handle_sflow_upcall(backer, &upcall->dpif_upcall);
3514             break;
3515
3516         case FLOW_SAMPLE_UPCALL:
3517             handle_flow_sample_upcall(backer, &upcall->dpif_upcall);
3518             break;
3519
3520         case IPFIX_UPCALL:
3521             handle_ipfix_upcall(backer, &upcall->dpif_upcall);
3522             break;
3523
3524         case BAD_UPCALL:
3525             break;
3526
3527         case MISS_UPCALL:
3528             NOT_REACHED();
3529         }
3530
3531         upcall_destroy(upcall);
3532     }
3533
3534     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3535         struct drop_key *drop_key = drop_key_next(backer->udpif);
3536         if (!drop_key) {
3537             break;
3538         }
3539
3540         if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3541             hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3542                         hash_bytes(drop_key->key, drop_key->key_len, 0));
3543             dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3544                           drop_key->key, drop_key->key_len,
3545                           NULL, 0, NULL, 0, NULL);
3546         } else {
3547             drop_key_destroy(drop_key);
3548         }
3549     }
3550
3551     fmb = flow_miss_batch_next(backer->udpif);
3552     if (fmb) {
3553         handle_flow_misses(backer, fmb);
3554         flow_miss_batch_destroy(fmb);
3555     }
3556 }
3557 \f
3558 /* Flow expiration. */
3559
3560 static int subfacet_max_idle(const struct dpif_backer *);
3561 static void update_stats(struct dpif_backer *);
3562 static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex);
3563 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3564
3565 /* This function is called periodically by run().  Its job is to collect
3566  * updates for the flows that have been installed into the datapath, most
3567  * importantly when they last were used, and then use that information to
3568  * expire flows that have not been used recently.
3569  *
3570  * Returns the number of milliseconds after which it should be called again. */
3571 static int
3572 expire(struct dpif_backer *backer)
3573 {
3574     struct ofproto_dpif *ofproto;
3575     size_t n_subfacets;
3576     int max_idle;
3577
3578     /* Periodically clear out the drop keys in an effort to keep them
3579      * relatively few. */
3580     drop_key_clear(backer);
3581
3582     /* Update stats for each flow in the backer. */
3583     update_stats(backer);
3584
3585     n_subfacets = hmap_count(&backer->subfacets);
3586     if (n_subfacets) {
3587         struct subfacet *subfacet;
3588         long long int total, now;
3589
3590         total = 0;
3591         now = time_msec();
3592         HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3593             total += now - subfacet->created;
3594         }
3595         backer->avg_subfacet_life += total / n_subfacets;
3596     }
3597     backer->avg_subfacet_life /= 2;
3598
3599     backer->avg_n_subfacet += n_subfacets;
3600     backer->avg_n_subfacet /= 2;
3601
3602     backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3603
3604     max_idle = subfacet_max_idle(backer);
3605     expire_subfacets(backer, max_idle);
3606
3607     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3608         struct rule *rule, *next_rule;
3609
3610         if (ofproto->backer != backer) {
3611             continue;
3612         }
3613
3614         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3615          * has passed. */
3616         ovs_mutex_lock(&ofproto_mutex);
3617         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3618                             &ofproto->up.expirable) {
3619             rule_expire(rule_dpif_cast(rule));
3620         }
3621         ovs_mutex_unlock(&ofproto_mutex);
3622
3623         /* All outstanding data in existing flows has been accounted, so it's a
3624          * good time to do bond rebalancing. */
3625         if (ofproto->has_bonded_bundles) {
3626             struct ofbundle *bundle;
3627
3628             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3629                 if (bundle->bond) {
3630                     bond_rebalance(bundle->bond);
3631                 }
3632             }
3633         }
3634     }
3635
3636     return MIN(max_idle, 1000);
3637 }
3638
3639 /* Updates flow table statistics given that the datapath just reported 'stats'
3640  * as 'subfacet''s statistics. */
3641 static void
3642 update_subfacet_stats(struct subfacet *subfacet,
3643                       const struct dpif_flow_stats *stats)
3644 {
3645     struct facet *facet = subfacet->facet;
3646     struct dpif_flow_stats diff;
3647
3648     diff.tcp_flags = stats->tcp_flags;
3649     diff.used = stats->used;
3650
3651     if (stats->n_packets >= subfacet->dp_packet_count) {
3652         diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3653     } else {
3654         VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3655         diff.n_packets = 0;
3656     }
3657
3658     if (stats->n_bytes >= subfacet->dp_byte_count) {
3659         diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3660     } else {
3661         VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3662         diff.n_bytes = 0;
3663     }
3664
3665     facet->ofproto->n_hit += diff.n_packets;
3666     subfacet->dp_packet_count = stats->n_packets;
3667     subfacet->dp_byte_count = stats->n_bytes;
3668     subfacet_update_stats(subfacet, &diff);
3669
3670     if (facet->accounted_bytes < facet->byte_count) {
3671         facet_learn(facet);
3672         facet_account(facet);
3673         facet->accounted_bytes = facet->byte_count;
3674     }
3675 }
3676
3677 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3678  * about, or a flow that shouldn't be installed but was anyway.  Delete it. */
3679 static void
3680 delete_unexpected_flow(struct dpif_backer *backer,
3681                        const struct nlattr *key, size_t key_len)
3682 {
3683     if (!VLOG_DROP_WARN(&rl)) {
3684         struct ds s;
3685
3686         ds_init(&s);
3687         odp_flow_key_format(key, key_len, &s);
3688         VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3689         ds_destroy(&s);
3690     }
3691
3692     COVERAGE_INC(facet_unexpected);
3693     dpif_flow_del(backer->dpif, key, key_len, NULL);
3694 }
3695
3696 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3697  *
3698  * This function also pushes statistics updates to rules which each facet
3699  * resubmits into.  Generally these statistics will be accurate.  However, if a
3700  * facet changes the rule it resubmits into at some time in between
3701  * update_stats() runs, it is possible that statistics accrued to the
3702  * old rule will be incorrectly attributed to the new rule.  This could be
3703  * avoided by calling update_stats() whenever rules are created or
3704  * deleted.  However, the performance impact of making so many calls to the
3705  * datapath do not justify the benefit of having perfectly accurate statistics.
3706  *
3707  * In addition, this function maintains per ofproto flow hit counts. The patch
3708  * port is not treated specially. e.g. A packet ingress from br0 patched into
3709  * br1 will increase the hit count of br0 by 1, however, does not affect
3710  * the hit or miss counts of br1.
3711  */
3712 static void
3713 update_stats(struct dpif_backer *backer)
3714 {
3715     const struct dpif_flow_stats *stats;
3716     struct dpif_flow_dump dump;
3717     const struct nlattr *key, *mask;
3718     size_t key_len, mask_len;
3719
3720     dpif_flow_dump_start(&dump, backer->dpif);
3721     while (dpif_flow_dump_next(&dump, &key, &key_len,
3722                                &mask, &mask_len, NULL, NULL, &stats)) {
3723         struct subfacet *subfacet;
3724         uint32_t key_hash;
3725
3726         key_hash = odp_flow_key_hash(key, key_len);
3727         subfacet = subfacet_find(backer, key, key_len, key_hash);
3728         switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3729         case SF_FAST_PATH:
3730             update_subfacet_stats(subfacet, stats);
3731             break;
3732
3733         case SF_SLOW_PATH:
3734             /* Stats are updated per-packet. */
3735             break;
3736
3737         case SF_NOT_INSTALLED:
3738         default:
3739             delete_unexpected_flow(backer, key, key_len);
3740             break;
3741         }
3742         run_fast_rl();
3743     }
3744     dpif_flow_dump_done(&dump);
3745
3746     update_moving_averages(backer);
3747 }
3748
3749 /* Calculates and returns the number of milliseconds of idle time after which
3750  * subfacets should expire from the datapath.  When a subfacet expires, we fold
3751  * its statistics into its facet, and when a facet's last subfacet expires, we
3752  * fold its statistic into its rule. */
3753 static int
3754 subfacet_max_idle(const struct dpif_backer *backer)
3755 {
3756     /*
3757      * Idle time histogram.
3758      *
3759      * Most of the time a switch has a relatively small number of subfacets.
3760      * When this is the case we might as well keep statistics for all of them
3761      * in userspace and to cache them in the kernel datapath for performance as
3762      * well.
3763      *
3764      * As the number of subfacets increases, the memory required to maintain
3765      * statistics about them in userspace and in the kernel becomes
3766      * significant.  However, with a large number of subfacets it is likely
3767      * that only a few of them are "heavy hitters" that consume a large amount
3768      * of bandwidth.  At this point, only heavy hitters are worth caching in
3769      * the kernel and maintaining in userspaces; other subfacets we can
3770      * discard.
3771      *
3772      * The technique used to compute the idle time is to build a histogram with
3773      * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each.  Each subfacet
3774      * that is installed in the kernel gets dropped in the appropriate bucket.
3775      * After the histogram has been built, we compute the cutoff so that only
3776      * the most-recently-used 1% of subfacets (but at least
3777      * flow_eviction_threshold flows) are kept cached.  At least
3778      * the most-recently-used bucket of subfacets is kept, so actually an
3779      * arbitrary number of subfacets can be kept in any given expiration run
3780      * (though the next run will delete most of those unless they receive
3781      * additional data).
3782      *
3783      * This requires a second pass through the subfacets, in addition to the
3784      * pass made by update_stats(), because the former function never looks at
3785      * uninstallable subfacets.
3786      */
3787     enum { BUCKET_WIDTH = 100 };
3788     enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3789     int buckets[N_BUCKETS] = { 0 };
3790     int total, subtotal, bucket;
3791     struct subfacet *subfacet;
3792     long long int now;
3793     int i;
3794
3795     total = hmap_count(&backer->subfacets);
3796     if (total <= flow_eviction_threshold) {
3797         return N_BUCKETS * BUCKET_WIDTH;
3798     }
3799
3800     /* Build histogram. */
3801     now = time_msec();
3802     HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3803         long long int idle = now - subfacet->used;
3804         int bucket = (idle <= 0 ? 0
3805                       : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3806                       : (unsigned int) idle / BUCKET_WIDTH);
3807         buckets[bucket]++;
3808     }
3809
3810     /* Find the first bucket whose flows should be expired. */
3811     subtotal = bucket = 0;
3812     do {
3813         subtotal += buckets[bucket++];
3814     } while (bucket < N_BUCKETS &&
3815              subtotal < MAX(flow_eviction_threshold, total / 100));
3816
3817     if (VLOG_IS_DBG_ENABLED()) {
3818         struct ds s;
3819
3820         ds_init(&s);
3821         ds_put_cstr(&s, "keep");
3822         for (i = 0; i < N_BUCKETS; i++) {
3823             if (i == bucket) {
3824                 ds_put_cstr(&s, ", drop");
3825             }
3826             if (buckets[i]) {
3827                 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3828             }
3829         }
3830         VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3831         ds_destroy(&s);
3832     }
3833
3834     return bucket * BUCKET_WIDTH;
3835 }
3836
3837 static void
3838 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3839 {
3840     /* Cutoff time for most flows. */
3841     long long int normal_cutoff = time_msec() - dp_max_idle;
3842
3843     /* We really want to keep flows for special protocols around, so use a more
3844      * conservative cutoff. */
3845     long long int special_cutoff = time_msec() - 10000;
3846
3847     struct subfacet *subfacet, *next_subfacet;
3848     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3849     int n_batch;
3850
3851     n_batch = 0;
3852     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3853                         &backer->subfacets) {
3854         long long int cutoff;
3855
3856         cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3857                                                 | SLOW_STP)
3858                   ? special_cutoff
3859                   : normal_cutoff);
3860         if (subfacet->used < cutoff) {
3861             if (subfacet->path != SF_NOT_INSTALLED) {
3862                 batch[n_batch++] = subfacet;
3863                 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3864                     subfacet_destroy_batch(backer, batch, n_batch);
3865                     n_batch = 0;
3866                 }
3867             } else {
3868                 subfacet_destroy(subfacet);
3869             }
3870         }
3871     }
3872
3873     if (n_batch > 0) {
3874         subfacet_destroy_batch(backer, batch, n_batch);
3875     }
3876 }
3877
3878 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3879  * then delete it entirely. */
3880 static void
3881 rule_expire(struct rule_dpif *rule)
3882     OVS_REQUIRES(ofproto_mutex)
3883 {
3884     uint16_t idle_timeout, hard_timeout;
3885     long long int now = time_msec();
3886     int reason;
3887
3888     ovs_assert(!rule->up.pending);
3889
3890     /* Has 'rule' expired? */
3891     ovs_mutex_lock(&rule->up.mutex);
3892     hard_timeout = rule->up.hard_timeout;
3893     idle_timeout = rule->up.idle_timeout;
3894     if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3895         reason = OFPRR_HARD_TIMEOUT;
3896     } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3897         reason = OFPRR_IDLE_TIMEOUT;
3898     } else {
3899         reason = -1;
3900     }
3901     ovs_mutex_unlock(&rule->up.mutex);
3902
3903     if (reason >= 0) {
3904         COVERAGE_INC(ofproto_dpif_expired);
3905         ofproto_rule_expire(&rule->up, reason);
3906     }
3907 }
3908 \f
3909 /* Facets. */
3910
3911 /* Creates and returns a new facet based on 'miss'.
3912  *
3913  * The caller must already have determined that no facet with an identical
3914  * 'miss->flow' exists in 'miss->ofproto'.
3915  *
3916  * 'rule' and 'xout' must have been created based on 'miss'.
3917  *
3918  * 'facet'' statistics are initialized based on 'stats'.
3919  *
3920  * The facet will initially have no subfacets.  The caller should create (at
3921  * least) one subfacet with subfacet_create(). */
3922 static struct facet *
3923 facet_create(const struct flow_miss *miss)
3924 {
3925     struct ofproto_dpif *ofproto = miss->ofproto;
3926     struct facet *facet;
3927     struct match match;
3928
3929     facet = xzalloc(sizeof *facet);
3930     facet->ofproto = miss->ofproto;
3931     facet->used = miss->stats.used;
3932     facet->flow = miss->flow;
3933     facet->learn_rl = time_msec() + 500;
3934
3935     list_init(&facet->subfacets);
3936     netflow_flow_init(&facet->nf_flow);
3937     netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3938
3939     xlate_out_copy(&facet->xout, &miss->xout);
3940
3941     match_init(&match, &facet->flow, &facet->xout.wc);
3942     cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3943     ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3944     classifier_insert(&ofproto->facets, &facet->cr);
3945     ovs_rwlock_unlock(&ofproto->facets.rwlock);
3946
3947     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
3948     return facet;
3949 }
3950
3951 static void
3952 facet_free(struct facet *facet)
3953 {
3954     if (facet) {
3955         xlate_out_uninit(&facet->xout);
3956         free(facet);
3957     }
3958 }
3959
3960 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3961  * 'packet', which arrived on 'in_port'. */
3962 static bool
3963 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3964                     const struct nlattr *odp_actions, size_t actions_len,
3965                     struct ofpbuf *packet)
3966 {
3967     struct odputil_keybuf keybuf;
3968     struct ofpbuf key;
3969     int error;
3970
3971     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3972     odp_flow_key_from_flow(&key, flow,
3973                            ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
3974
3975     error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
3976                          odp_actions, actions_len, packet);
3977     return !error;
3978 }
3979
3980 /* Remove 'facet' from its ofproto and free up the associated memory:
3981  *
3982  *   - If 'facet' was installed in the datapath, uninstalls it and updates its
3983  *     rule's statistics, via subfacet_uninstall().
3984  *
3985  *   - Removes 'facet' from its rule and from ofproto->facets.
3986  */
3987 static void
3988 facet_remove(struct facet *facet)
3989 {
3990     struct subfacet *subfacet, *next_subfacet;
3991
3992     ovs_assert(!list_is_empty(&facet->subfacets));
3993
3994     /* First uninstall all of the subfacets to get final statistics. */
3995     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3996         subfacet_uninstall(subfacet);
3997     }
3998
3999     /* Flush the final stats to the rule.
4000      *
4001      * This might require us to have at least one subfacet around so that we
4002      * can use its actions for accounting in facet_account(), which is why we
4003      * have uninstalled but not yet destroyed the subfacets. */
4004     facet_flush_stats(facet);
4005
4006     /* Now we're really all done so destroy everything. */
4007     LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4008                         &facet->subfacets) {
4009         subfacet_destroy__(subfacet);
4010     }
4011     ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4012     classifier_remove(&facet->ofproto->facets, &facet->cr);
4013     ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4014     cls_rule_destroy(&facet->cr);
4015     facet_free(facet);
4016 }
4017
4018 /* Feed information from 'facet' back into the learning table to keep it in
4019  * sync with what is actually flowing through the datapath. */
4020 static void
4021 facet_learn(struct facet *facet)
4022 {
4023     long long int now = time_msec();
4024
4025     if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4026         return;
4027     }
4028
4029     facet->learn_rl = now + 500;
4030
4031     if (!facet->xout.has_learn
4032         && !facet->xout.has_normal
4033         && (!facet->xout.has_fin_timeout
4034             || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4035         return;
4036     }
4037
4038     facet_push_stats(facet, true);
4039 }
4040
4041 static void
4042 facet_account(struct facet *facet)
4043 {
4044     const struct nlattr *a;
4045     unsigned int left;
4046     ovs_be16 vlan_tci;
4047     uint64_t n_bytes;
4048
4049     if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4050         return;
4051     }
4052     n_bytes = facet->byte_count - facet->accounted_bytes;
4053
4054     /* This loop feeds byte counters to bond_account() for rebalancing to use
4055      * as a basis.  We also need to track the actual VLAN on which the packet
4056      * is going to be sent to ensure that it matches the one passed to
4057      * bond_choose_output_slave().  (Otherwise, we will account to the wrong
4058      * hash bucket.)
4059      *
4060      * We use the actions from an arbitrary subfacet because they should all
4061      * be equally valid for our purpose. */
4062     vlan_tci = facet->flow.vlan_tci;
4063     NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4064                              facet->xout.odp_actions.size) {
4065         const struct ovs_action_push_vlan *vlan;
4066         struct ofport_dpif *port;
4067
4068         switch (nl_attr_type(a)) {
4069         case OVS_ACTION_ATTR_OUTPUT:
4070             port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4071             if (port && port->bundle && port->bundle->bond) {
4072                 bond_account(port->bundle->bond, &facet->flow,
4073                              vlan_tci_to_vid(vlan_tci), n_bytes);
4074             }
4075             break;
4076
4077         case OVS_ACTION_ATTR_POP_VLAN:
4078             vlan_tci = htons(0);
4079             break;
4080
4081         case OVS_ACTION_ATTR_PUSH_VLAN:
4082             vlan = nl_attr_get(a);
4083             vlan_tci = vlan->vlan_tci;
4084             break;
4085         }
4086     }
4087 }
4088
4089 /* Returns true if the only action for 'facet' is to send to the controller.
4090  * (We don't report NetFlow expiration messages for such facets because they
4091  * are just part of the control logic for the network, not real traffic). */
4092 static bool
4093 facet_is_controller_flow(struct facet *facet)
4094 {
4095     if (facet) {
4096         struct ofproto_dpif *ofproto = facet->ofproto;
4097         const struct ofpact *ofpacts;
4098         struct rule_actions *actions;
4099         struct rule_dpif *rule;
4100         size_t ofpacts_len;
4101         bool is_controller;
4102
4103         rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4104         actions = rule_dpif_get_actions(rule);
4105         rule_dpif_unref(rule);
4106
4107         ofpacts_len = actions->ofpacts_len;
4108         ofpacts = actions->ofpacts;
4109         is_controller = ofpacts_len > 0
4110             && ofpacts->type == OFPACT_CONTROLLER
4111             && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4112         rule_actions_unref(actions);
4113
4114         return is_controller;
4115     }
4116     return false;
4117 }
4118
4119 /* Folds all of 'facet''s statistics into its rule.  Also updates the
4120  * accounting ofhook and emits a NetFlow expiration if appropriate.  All of
4121  * 'facet''s statistics in the datapath should have been zeroed and folded into
4122  * its packet and byte counts before this function is called. */
4123 static void
4124 facet_flush_stats(struct facet *facet)
4125 {
4126     struct ofproto_dpif *ofproto = facet->ofproto;
4127     struct subfacet *subfacet;
4128
4129     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4130         ovs_assert(!subfacet->dp_byte_count);
4131         ovs_assert(!subfacet->dp_packet_count);
4132     }
4133
4134     facet_push_stats(facet, false);
4135     if (facet->accounted_bytes < facet->byte_count) {
4136         facet_account(facet);
4137         facet->accounted_bytes = facet->byte_count;
4138     }
4139
4140     if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4141         struct ofexpired expired;
4142         expired.flow = facet->flow;
4143         expired.packet_count = facet->packet_count;
4144         expired.byte_count = facet->byte_count;
4145         expired.used = facet->used;
4146         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4147     }
4148
4149     /* Reset counters to prevent double counting if 'facet' ever gets
4150      * reinstalled. */
4151     facet_reset_counters(facet);
4152
4153     netflow_flow_clear(&facet->nf_flow);
4154     facet->tcp_flags = 0;
4155 }
4156
4157 /* Searches 'ofproto''s table of facets for one which would be responsible for
4158  * 'flow'.  Returns it if found, otherwise a null pointer.
4159  *
4160  * The returned facet might need revalidation; use facet_lookup_valid()
4161  * instead if that is important. */
4162 static struct facet *
4163 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4164 {
4165     struct cls_rule *cr;
4166
4167     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4168     cr = classifier_lookup(&ofproto->facets, flow, NULL);
4169     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4170     return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4171 }
4172
4173 /* Searches 'ofproto''s table of facets for one capable that covers
4174  * 'flow'.  Returns it if found, otherwise a null pointer.
4175  *
4176  * The returned facet is guaranteed to be valid. */
4177 static struct facet *
4178 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4179 {
4180     struct facet *facet;
4181
4182     facet = facet_find(ofproto, flow);
4183     if (facet
4184         && ofproto->backer->need_revalidate
4185         && !facet_revalidate(facet)) {
4186         return NULL;
4187     }
4188
4189     return facet;
4190 }
4191
4192 static bool
4193 facet_check_consistency(struct facet *facet)
4194 {
4195     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4196
4197     struct xlate_out xout;
4198     struct xlate_in xin;
4199
4200     struct rule_dpif *rule;
4201     bool ok;
4202
4203     /* Check the datapath actions for consistency. */
4204     rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4205     xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4206     xlate_actions(&xin, &xout);
4207     rule_dpif_unref(rule);
4208
4209     ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4210         && facet->xout.slow == xout.slow;
4211     if (!ok && !VLOG_DROP_WARN(&rl)) {
4212         struct ds s = DS_EMPTY_INITIALIZER;
4213
4214         flow_format(&s, &facet->flow);
4215         ds_put_cstr(&s, ": inconsistency in facet");
4216
4217         if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4218             ds_put_cstr(&s, " (actions were: ");
4219             format_odp_actions(&s, facet->xout.odp_actions.data,
4220                                facet->xout.odp_actions.size);
4221             ds_put_cstr(&s, ") (correct actions: ");
4222             format_odp_actions(&s, xout.odp_actions.data,
4223                                xout.odp_actions.size);
4224             ds_put_char(&s, ')');
4225         }
4226
4227         if (facet->xout.slow != xout.slow) {
4228             ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4229         }
4230
4231         ds_destroy(&s);
4232     }
4233     xlate_out_uninit(&xout);
4234
4235     return ok;
4236 }
4237
4238 /* Re-searches the classifier for 'facet':
4239  *
4240  *   - If the rule found is different from 'facet''s current rule, moves
4241  *     'facet' to the new rule and recompiles its actions.
4242  *
4243  *   - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4244  *     where it is and recompiles its actions anyway.
4245  *
4246  *   - If any of 'facet''s subfacets correspond to a new flow according to
4247  *     xlate_receive(), 'facet' is removed.
4248  *
4249  *   Returns true if 'facet' is still valid.  False if 'facet' was removed. */
4250 static bool
4251 facet_revalidate(struct facet *facet)
4252 {
4253     struct ofproto_dpif *ofproto = facet->ofproto;
4254     struct rule_dpif *new_rule;
4255     struct subfacet *subfacet;
4256     struct flow_wildcards wc;
4257     struct xlate_out xout;
4258     struct xlate_in xin;
4259
4260     COVERAGE_INC(facet_revalidate);
4261
4262     /* Check that child subfacets still correspond to this facet.  Tunnel
4263      * configuration changes could cause a subfacet's OpenFlow in_port to
4264      * change. */
4265     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4266         struct ofproto_dpif *recv_ofproto;
4267         struct flow recv_flow;
4268         int error;
4269
4270         error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4271                               subfacet->key_len, &recv_flow, NULL,
4272                               &recv_ofproto, NULL);
4273         if (error
4274             || recv_ofproto != ofproto
4275             || facet != facet_find(ofproto, &recv_flow)) {
4276             facet_remove(facet);
4277             return false;
4278         }
4279     }
4280
4281     flow_wildcards_init_catchall(&wc);
4282     rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4283
4284     /* Calculate new datapath actions.
4285      *
4286      * We do not modify any 'facet' state yet, because we might need to, e.g.,
4287      * emit a NetFlow expiration and, if so, we need to have the old state
4288      * around to properly compose it. */
4289     xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4290     xlate_actions(&xin, &xout);
4291     flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4292
4293     /* A facet's slow path reason should only change under dramatic
4294      * circumstances.  Rather than try to update everything, it's simpler to
4295      * remove the facet and start over.
4296      *
4297      * More importantly, if a facet's wildcards change, it will be relatively
4298      * difficult to figure out if its subfacets still belong to it, and if not
4299      * which facet they may belong to.  Again, to avoid the complexity, we
4300      * simply give up instead. */
4301     if (facet->xout.slow != xout.slow
4302         || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4303         facet_remove(facet);
4304         xlate_out_uninit(&xout);
4305         rule_dpif_unref(new_rule);
4306         return false;
4307     }
4308
4309     if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4310         LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4311             if (subfacet->path == SF_FAST_PATH) {
4312                 struct dpif_flow_stats stats;
4313
4314                 subfacet_install(subfacet, &xout.odp_actions, &stats);
4315                 subfacet_update_stats(subfacet, &stats);
4316             }
4317         }
4318
4319         facet_flush_stats(facet);
4320
4321         ofpbuf_clear(&facet->xout.odp_actions);
4322         ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4323                    xout.odp_actions.size);
4324     }
4325
4326     /* Update 'facet' now that we've taken care of all the old state. */
4327     facet->xout.slow = xout.slow;
4328     facet->xout.has_learn = xout.has_learn;
4329     facet->xout.has_normal = xout.has_normal;
4330     facet->xout.has_fin_timeout = xout.has_fin_timeout;
4331     facet->xout.nf_output_iface = xout.nf_output_iface;
4332     facet->xout.mirrors = xout.mirrors;
4333     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4334
4335     ovs_mutex_lock(&new_rule->up.mutex);
4336     facet->used = MAX(facet->used, new_rule->up.created);
4337     ovs_mutex_unlock(&new_rule->up.mutex);
4338
4339     xlate_out_uninit(&xout);
4340     rule_dpif_unref(new_rule);
4341     return true;
4342 }
4343
4344 static void
4345 facet_reset_counters(struct facet *facet)
4346 {
4347     facet->packet_count = 0;
4348     facet->byte_count = 0;
4349     facet->prev_packet_count = 0;
4350     facet->prev_byte_count = 0;
4351     facet->accounted_bytes = 0;
4352 }
4353
4354 static void
4355 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4356                 struct dpif_flow_stats *stats, bool may_learn)
4357 {
4358     struct ofport_dpif *in_port;
4359     struct rule_dpif *rule;
4360     struct xlate_in xin;
4361
4362     in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4363     if (in_port && in_port->is_tunnel) {
4364         netdev_vport_inc_rx(in_port->up.netdev, stats);
4365     }
4366
4367     rule_dpif_lookup(ofproto, flow, NULL, &rule);
4368     rule_dpif_credit_stats(rule, stats);
4369     xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4370     xin.resubmit_stats = stats;
4371     xin.may_learn = may_learn;
4372     xlate_actions_for_side_effects(&xin);
4373     rule_dpif_unref(rule);
4374 }
4375
4376 static void
4377 facet_push_stats(struct facet *facet, bool may_learn)
4378 {
4379     struct dpif_flow_stats stats;
4380
4381     ovs_assert(facet->packet_count >= facet->prev_packet_count);
4382     ovs_assert(facet->byte_count >= facet->prev_byte_count);
4383     ovs_assert(facet->used >= facet->prev_used);
4384
4385     stats.n_packets = facet->packet_count - facet->prev_packet_count;
4386     stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4387     stats.used = facet->used;
4388     stats.tcp_flags = facet->tcp_flags;
4389
4390     if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4391         facet->prev_packet_count = facet->packet_count;
4392         facet->prev_byte_count = facet->byte_count;
4393         facet->prev_used = facet->used;
4394
4395         netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4396                                  facet->used);
4397         netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4398         mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4399                             stats.n_packets, stats.n_bytes);
4400         flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4401     }
4402 }
4403
4404 static void
4405 push_all_stats__(bool run_fast)
4406 {
4407     static long long int rl = LLONG_MIN;
4408     struct ofproto_dpif *ofproto;
4409
4410     if (time_msec() < rl) {
4411         return;
4412     }
4413
4414     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4415         struct cls_cursor cursor;
4416         struct facet *facet;
4417
4418         ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4419         cls_cursor_init(&cursor, &ofproto->facets, NULL);
4420         CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4421             facet_push_stats(facet, false);
4422             if (run_fast) {
4423                 run_fast_rl();
4424             }
4425         }
4426         ovs_rwlock_unlock(&ofproto->facets.rwlock);
4427     }
4428
4429     rl = time_msec() + 100;
4430 }
4431
4432 static void
4433 push_all_stats(void)
4434 {
4435     push_all_stats__(true);
4436 }
4437
4438 void
4439 rule_dpif_credit_stats(struct rule_dpif *rule,
4440                        const struct dpif_flow_stats *stats)
4441 {
4442     ovs_mutex_lock(&rule->stats_mutex);
4443     rule->packet_count += stats->n_packets;
4444     rule->byte_count += stats->n_bytes;
4445     rule->up.used = MAX(rule->up.used, stats->used);
4446     ovs_mutex_unlock(&rule->stats_mutex);
4447 }
4448
4449 bool
4450 rule_dpif_fail_open(const struct rule_dpif *rule)
4451 {
4452     return rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4453 }
4454
4455 ovs_be64
4456 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4457     OVS_REQUIRES(rule->up.mutex)
4458 {
4459     return rule->up.flow_cookie;
4460 }
4461
4462 void
4463 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4464                      uint16_t hard_timeout)
4465 {
4466     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4467 }
4468
4469 /* Returns 'rule''s actions.  The caller owns a reference on the returned
4470  * actions and must eventually release it (with rule_actions_unref()) to avoid
4471  * a memory leak. */
4472 struct rule_actions *
4473 rule_dpif_get_actions(const struct rule_dpif *rule)
4474 {
4475     return rule_get_actions(&rule->up);
4476 }
4477 \f
4478 /* Subfacets. */
4479
4480 static struct subfacet *
4481 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4482               size_t key_len, uint32_t key_hash)
4483 {
4484     struct subfacet *subfacet;
4485
4486     HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4487                              &backer->subfacets) {
4488         if (subfacet->key_len == key_len
4489             && !memcmp(key, subfacet->key, key_len)) {
4490             return subfacet;
4491         }
4492     }
4493
4494     return NULL;
4495 }
4496
4497 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4498  * 'key_fitness', 'key', and 'key_len' members in 'miss'.  Returns the
4499  * existing subfacet if there is one, otherwise creates and returns a
4500  * new subfacet. */
4501 static struct subfacet *
4502 subfacet_create(struct facet *facet, struct flow_miss *miss)
4503 {
4504     struct dpif_backer *backer = miss->ofproto->backer;
4505     enum odp_key_fitness key_fitness = miss->key_fitness;
4506     const struct nlattr *key = miss->key;
4507     size_t key_len = miss->key_len;
4508     uint32_t key_hash;
4509     struct subfacet *subfacet;
4510
4511     key_hash = odp_flow_key_hash(key, key_len);
4512
4513     if (list_is_empty(&facet->subfacets)) {
4514         subfacet = &facet->one_subfacet;
4515     } else {
4516         subfacet = subfacet_find(backer, key, key_len, key_hash);
4517         if (subfacet) {
4518             if (subfacet->facet == facet) {
4519                 return subfacet;
4520             }
4521
4522             /* This shouldn't happen. */
4523             VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4524             subfacet_destroy(subfacet);
4525         }
4526
4527         subfacet = xmalloc(sizeof *subfacet);
4528     }
4529
4530     hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4531     list_push_back(&facet->subfacets, &subfacet->list_node);
4532     subfacet->facet = facet;
4533     subfacet->key_fitness = key_fitness;
4534     subfacet->key = xmemdup(key, key_len);
4535     subfacet->key_len = key_len;
4536     subfacet->used = miss->stats.used;
4537     subfacet->created = subfacet->used;
4538     subfacet->dp_packet_count = 0;
4539     subfacet->dp_byte_count = 0;
4540     subfacet->path = SF_NOT_INSTALLED;
4541     subfacet->backer = backer;
4542
4543     backer->subfacet_add_count++;
4544     return subfacet;
4545 }
4546
4547 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4548  * its facet within 'ofproto', and frees it. */
4549 static void
4550 subfacet_destroy__(struct subfacet *subfacet)
4551 {
4552     struct facet *facet = subfacet->facet;
4553     struct ofproto_dpif *ofproto = facet->ofproto;
4554
4555     /* Update ofproto stats before uninstall the subfacet. */
4556     ofproto->backer->subfacet_del_count++;
4557
4558     subfacet_uninstall(subfacet);
4559     hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4560     list_remove(&subfacet->list_node);
4561     free(subfacet->key);
4562     if (subfacet != &facet->one_subfacet) {
4563         free(subfacet);
4564     }
4565 }
4566
4567 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4568  * last remaining subfacet in its facet destroys the facet too. */
4569 static void
4570 subfacet_destroy(struct subfacet *subfacet)
4571 {
4572     struct facet *facet = subfacet->facet;
4573
4574     if (list_is_singleton(&facet->subfacets)) {
4575         /* facet_remove() needs at least one subfacet (it will remove it). */
4576         facet_remove(facet);
4577     } else {
4578         subfacet_destroy__(subfacet);
4579     }
4580 }
4581
4582 static void
4583 subfacet_destroy_batch(struct dpif_backer *backer,
4584                        struct subfacet **subfacets, int n)
4585 {
4586     struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4587     struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4588     struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4589     int i;
4590
4591     for (i = 0; i < n; i++) {
4592         ops[i].type = DPIF_OP_FLOW_DEL;
4593         ops[i].u.flow_del.key = subfacets[i]->key;
4594         ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4595         ops[i].u.flow_del.stats = &stats[i];
4596         opsp[i] = &ops[i];
4597     }
4598
4599     dpif_operate(backer->dpif, opsp, n);
4600     for (i = 0; i < n; i++) {
4601         subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4602         subfacets[i]->path = SF_NOT_INSTALLED;
4603         subfacet_destroy(subfacets[i]);
4604         run_fast_rl();
4605     }
4606 }
4607
4608 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4609  * bytes of actions in 'actions'.  If 'stats' is non-null, statistics counters
4610  * in the datapath will be zeroed and 'stats' will be updated with traffic new
4611  * since 'subfacet' was last updated.
4612  *
4613  * Returns 0 if successful, otherwise a positive errno value. */
4614 static int
4615 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4616                  struct dpif_flow_stats *stats)
4617 {
4618     struct facet *facet = subfacet->facet;
4619     enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4620     const struct nlattr *actions = odp_actions->data;
4621     size_t actions_len = odp_actions->size;
4622     struct odputil_keybuf maskbuf;
4623     struct ofpbuf mask;
4624
4625     uint64_t slow_path_stub[128 / 8];
4626     enum dpif_flow_put_flags flags;
4627     int ret;
4628
4629     flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4630                                                : DPIF_FP_MODIFY;
4631     if (stats) {
4632         flags |= DPIF_FP_ZERO_STATS;
4633     }
4634
4635     if (path == SF_SLOW_PATH) {
4636         compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4637                           slow_path_stub, sizeof slow_path_stub,
4638                           &actions, &actions_len);
4639     }
4640
4641     ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4642     if (enable_megaflows) {
4643         odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4644                                &facet->flow, UINT32_MAX);
4645     }
4646
4647     ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4648                         subfacet->key_len,  mask.data, mask.size,
4649                         actions, actions_len, stats);
4650
4651     if (stats) {
4652         subfacet_reset_dp_stats(subfacet, stats);
4653     }
4654
4655     if (ret) {
4656         COVERAGE_INC(subfacet_install_fail);
4657     } else {
4658         subfacet->path = path;
4659     }
4660     return ret;
4661 }
4662
4663 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4664 static void
4665 subfacet_uninstall(struct subfacet *subfacet)
4666 {
4667     if (subfacet->path != SF_NOT_INSTALLED) {
4668         struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4669         struct dpif_flow_stats stats;
4670         int error;
4671
4672         error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4673                               subfacet->key_len, &stats);
4674         subfacet_reset_dp_stats(subfacet, &stats);
4675         if (!error) {
4676             subfacet_update_stats(subfacet, &stats);
4677         }
4678         subfacet->path = SF_NOT_INSTALLED;
4679     } else {
4680         ovs_assert(subfacet->dp_packet_count == 0);
4681         ovs_assert(subfacet->dp_byte_count == 0);
4682     }
4683 }
4684
4685 /* Resets 'subfacet''s datapath statistics counters.  This should be called
4686  * when 'subfacet''s statistics are cleared in the datapath.  If 'stats' is
4687  * non-null, it should contain the statistics returned by dpif when 'subfacet'
4688  * was reset in the datapath.  'stats' will be modified to include only
4689  * statistics new since 'subfacet' was last updated. */
4690 static void
4691 subfacet_reset_dp_stats(struct subfacet *subfacet,
4692                         struct dpif_flow_stats *stats)
4693 {
4694     if (stats
4695         && subfacet->dp_packet_count <= stats->n_packets
4696         && subfacet->dp_byte_count <= stats->n_bytes) {
4697         stats->n_packets -= subfacet->dp_packet_count;
4698         stats->n_bytes -= subfacet->dp_byte_count;
4699     }
4700
4701     subfacet->dp_packet_count = 0;
4702     subfacet->dp_byte_count = 0;
4703 }
4704
4705 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4706  *
4707  * Because of the meaning of a subfacet's counters, it only makes sense to do
4708  * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4709  * represents a packet that was sent by hand or if it represents statistics
4710  * that have been cleared out of the datapath. */
4711 static void
4712 subfacet_update_stats(struct subfacet *subfacet,
4713                       const struct dpif_flow_stats *stats)
4714 {
4715     if (stats->n_packets || stats->used > subfacet->used) {
4716         struct facet *facet = subfacet->facet;
4717
4718         subfacet->used = MAX(subfacet->used, stats->used);
4719         facet->used = MAX(facet->used, stats->used);
4720         facet->packet_count += stats->n_packets;
4721         facet->byte_count += stats->n_bytes;
4722         facet->tcp_flags |= stats->tcp_flags;
4723     }
4724 }
4725 \f
4726 /* Rules. */
4727
4728 /* Lookup 'flow' in 'ofproto''s classifier.  If 'wc' is non-null, sets
4729  * the fields that were relevant as part of the lookup. */
4730 void
4731 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4732                  struct flow_wildcards *wc, struct rule_dpif **rule)
4733 {
4734     struct ofport_dpif *port;
4735
4736     if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4737         return;
4738     }
4739     port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4740     if (!port) {
4741         VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4742                      flow->in_port.ofp_port);
4743     }
4744
4745     choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4746                      ofproto->no_packet_in_rule, rule);
4747 }
4748
4749 bool
4750 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4751                           const struct flow *flow, struct flow_wildcards *wc,
4752                           uint8_t table_id, struct rule_dpif **rule)
4753 {
4754     const struct cls_rule *cls_rule;
4755     struct classifier *cls;
4756     bool frag;
4757
4758     *rule = NULL;
4759     if (table_id >= N_TABLES) {
4760         return false;
4761     }
4762
4763     if (wc) {
4764         memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4765         wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4766     }
4767
4768     cls = &ofproto->up.tables[table_id].cls;
4769     ovs_rwlock_rdlock(&cls->rwlock);
4770     frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4771     if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4772         /* We must pretend that transport ports are unavailable. */
4773         struct flow ofpc_normal_flow = *flow;
4774         ofpc_normal_flow.tp_src = htons(0);
4775         ofpc_normal_flow.tp_dst = htons(0);
4776         cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4777     } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4778         cls_rule = &ofproto->drop_frags_rule->up.cr;
4779         if (wc) {
4780             flow_wildcards_init_exact(wc);
4781         }
4782     } else {
4783         cls_rule = classifier_lookup(cls, flow, wc);
4784     }
4785
4786     *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4787     rule_dpif_ref(*rule);
4788     ovs_rwlock_unlock(&cls->rwlock);
4789
4790     return *rule != NULL;
4791 }
4792
4793 /* Given a port configuration (specified as zero if there's no port), chooses
4794  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4795  * flow table miss. */
4796 void
4797 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4798                  struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4799 {
4800     *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4801     rule_dpif_ref(*rule);
4802 }
4803
4804 void
4805 rule_dpif_ref(struct rule_dpif *rule)
4806 {
4807     if (rule) {
4808         ofproto_rule_ref(&rule->up);
4809     }
4810 }
4811
4812 void
4813 rule_dpif_unref(struct rule_dpif *rule)
4814 {
4815     if (rule) {
4816         ofproto_rule_unref(&rule->up);
4817     }
4818 }
4819
4820 static void
4821 complete_operation(struct rule_dpif *rule)
4822     OVS_REQUIRES(ofproto_mutex)
4823 {
4824     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4825
4826     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4827     ofoperation_complete(rule->up.pending, 0);
4828 }
4829
4830 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4831 {
4832     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4833 }
4834
4835 static struct rule *
4836 rule_alloc(void)
4837 {
4838     struct rule_dpif *rule = xmalloc(sizeof *rule);
4839     return &rule->up;
4840 }
4841
4842 static void
4843 rule_dealloc(struct rule *rule_)
4844 {
4845     struct rule_dpif *rule = rule_dpif_cast(rule_);
4846     free(rule);
4847 }
4848
4849 static enum ofperr
4850 rule_construct(struct rule *rule_)
4851 {
4852     struct rule_dpif *rule = rule_dpif_cast(rule_);
4853     ovs_mutex_init(&rule->stats_mutex);
4854     ovs_mutex_lock(&rule->stats_mutex);
4855     rule->packet_count = 0;
4856     rule->byte_count = 0;
4857     ovs_mutex_unlock(&rule->stats_mutex);
4858     return 0;
4859 }
4860
4861 static void
4862 rule_insert(struct rule *rule_)
4863     OVS_REQUIRES(ofproto_mutex)
4864 {
4865     struct rule_dpif *rule = rule_dpif_cast(rule_);
4866     complete_operation(rule);
4867 }
4868
4869 static void
4870 rule_delete(struct rule *rule_)
4871     OVS_REQUIRES(ofproto_mutex)
4872 {
4873     struct rule_dpif *rule = rule_dpif_cast(rule_);
4874     complete_operation(rule);
4875 }
4876
4877 static void
4878 rule_destruct(struct rule *rule_)
4879 {
4880     struct rule_dpif *rule = rule_dpif_cast(rule_);
4881     ovs_mutex_destroy(&rule->stats_mutex);
4882 }
4883
4884 static void
4885 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4886 {
4887     struct rule_dpif *rule = rule_dpif_cast(rule_);
4888
4889     /* push_all_stats() can handle flow misses which, when using the learn
4890      * action, can cause rules to be added and deleted.  This can corrupt our
4891      * caller's datastructures which assume that rule_get_stats() doesn't have
4892      * an impact on the flow table. To be safe, we disable miss handling. */
4893     push_all_stats__(false);
4894
4895     /* Start from historical data for 'rule' itself that are no longer tracked
4896      * in facets.  This counts, for example, facets that have expired. */
4897     ovs_mutex_lock(&rule->stats_mutex);
4898     *packets = rule->packet_count;
4899     *bytes = rule->byte_count;
4900     ovs_mutex_unlock(&rule->stats_mutex);
4901 }
4902
4903 static void
4904 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4905                   struct ofpbuf *packet)
4906 {
4907     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4908     struct dpif_flow_stats stats;
4909     struct xlate_out xout;
4910     struct xlate_in xin;
4911
4912     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4913     rule_dpif_credit_stats(rule, &stats);
4914
4915     xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4916     xin.resubmit_stats = &stats;
4917     xlate_actions(&xin, &xout);
4918
4919     execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4920                         xout.odp_actions.size, packet);
4921
4922     xlate_out_uninit(&xout);
4923 }
4924
4925 static enum ofperr
4926 rule_execute(struct rule *rule, const struct flow *flow,
4927              struct ofpbuf *packet)
4928 {
4929     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4930     ofpbuf_delete(packet);
4931     return 0;
4932 }
4933
4934 static void
4935 rule_modify_actions(struct rule *rule_, bool reset_counters)
4936     OVS_REQUIRES(ofproto_mutex)
4937 {
4938     struct rule_dpif *rule = rule_dpif_cast(rule_);
4939
4940     if (reset_counters) {
4941         ovs_mutex_lock(&rule->stats_mutex);
4942         rule->packet_count = 0;
4943         rule->byte_count = 0;
4944         ovs_mutex_unlock(&rule->stats_mutex);
4945     }
4946
4947     complete_operation(rule);
4948 }
4949 \f
4950 /* Sends 'packet' out 'ofport'.
4951  * May modify 'packet'.
4952  * Returns 0 if successful, otherwise a positive errno value. */
4953 static int
4954 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4955 {
4956     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4957     uint64_t odp_actions_stub[1024 / 8];
4958     struct ofpbuf key, odp_actions;
4959     struct dpif_flow_stats stats;
4960     struct odputil_keybuf keybuf;
4961     struct ofpact_output output;
4962     struct xlate_out xout;
4963     struct xlate_in xin;
4964     struct flow flow;
4965     union flow_in_port in_port_;
4966     int error;
4967
4968     ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4969     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4970
4971     /* Use OFPP_NONE as the in_port to avoid special packet processing. */
4972     in_port_.ofp_port = OFPP_NONE;
4973     flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
4974     odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
4975                                                              OFPP_LOCAL));
4976     dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
4977
4978     ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
4979     output.port = ofport->up.ofp_port;
4980     output.max_len = 0;
4981
4982     xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
4983     xin.ofpacts_len = sizeof output;
4984     xin.ofpacts = &output.ofpact;
4985     xin.resubmit_stats = &stats;
4986     xlate_actions(&xin, &xout);
4987
4988     error = dpif_execute(ofproto->backer->dpif,
4989                          key.data, key.size,
4990                          xout.odp_actions.data, xout.odp_actions.size,
4991                          packet);
4992     xlate_out_uninit(&xout);
4993
4994     if (error) {
4995         VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
4996                      ofproto->up.name, netdev_get_name(ofport->up.netdev),
4997                      ovs_strerror(error));
4998     }
4999
5000     ofproto->stats.tx_packets++;
5001     ofproto->stats.tx_bytes += packet->size;
5002     return error;
5003 }
5004
5005 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5006  * The action will state 'slow' as the reason that the action is in the slow
5007  * path.  (This is purely informational: it allows a human viewing "ovs-dpctl
5008  * dump-flows" output to see why a flow is in the slow path.)
5009  *
5010  * The 'stub_size' bytes in 'stub' will be used to store the action.
5011  * 'stub_size' must be large enough for the action.
5012  *
5013  * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5014  * respectively. */
5015 static void
5016 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5017                   enum slow_path_reason slow,
5018                   uint64_t *stub, size_t stub_size,
5019                   const struct nlattr **actionsp, size_t *actions_lenp)
5020 {
5021     union user_action_cookie cookie;
5022     struct ofpbuf buf;
5023
5024     cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5025     cookie.slow_path.unused = 0;
5026     cookie.slow_path.reason = slow;
5027
5028     ofpbuf_use_stack(&buf, stub, stub_size);
5029     if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5030         uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5031                                          ODPP_NONE);
5032         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5033     } else {
5034         odp_port_t odp_port;
5035         uint32_t pid;
5036
5037         odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5038         pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5039         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5040     }
5041     *actionsp = buf.data;
5042     *actions_lenp = buf.size;
5043 }
5044 \f
5045 static bool
5046 set_frag_handling(struct ofproto *ofproto_,
5047                   enum ofp_config_flags frag_handling)
5048 {
5049     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5050     if (frag_handling != OFPC_FRAG_REASM) {
5051         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5052         return true;
5053     } else {
5054         return false;
5055     }
5056 }
5057
5058 static enum ofperr
5059 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5060            const struct flow *flow,
5061            const struct ofpact *ofpacts, size_t ofpacts_len)
5062 {
5063     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5064     struct odputil_keybuf keybuf;
5065     struct dpif_flow_stats stats;
5066     struct xlate_out xout;
5067     struct xlate_in xin;
5068     struct ofpbuf key;
5069
5070
5071     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5072     odp_flow_key_from_flow(&key, flow,
5073                            ofp_port_to_odp_port(ofproto,
5074                                       flow->in_port.ofp_port));
5075
5076     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5077
5078     xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5079     xin.resubmit_stats = &stats;
5080     xin.ofpacts_len = ofpacts_len;
5081     xin.ofpacts = ofpacts;
5082
5083     xlate_actions(&xin, &xout);
5084     dpif_execute(ofproto->backer->dpif, key.data, key.size,
5085                  xout.odp_actions.data, xout.odp_actions.size, packet);
5086     xlate_out_uninit(&xout);
5087
5088     return 0;
5089 }
5090 \f
5091 /* NetFlow. */
5092
5093 static int
5094 set_netflow(struct ofproto *ofproto_,
5095             const struct netflow_options *netflow_options)
5096 {
5097     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5098
5099     if (netflow_options) {
5100         if (!ofproto->netflow) {
5101             ofproto->netflow = netflow_create();
5102             ofproto->backer->need_revalidate = REV_RECONFIGURE;
5103         }
5104         return netflow_set_options(ofproto->netflow, netflow_options);
5105     } else if (ofproto->netflow) {
5106         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5107         netflow_destroy(ofproto->netflow);
5108         ofproto->netflow = NULL;
5109     }
5110
5111     return 0;
5112 }
5113
5114 static void
5115 get_netflow_ids(const struct ofproto *ofproto_,
5116                 uint8_t *engine_type, uint8_t *engine_id)
5117 {
5118     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5119
5120     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5121 }
5122
5123 static void
5124 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5125 {
5126     if (!facet_is_controller_flow(facet) &&
5127         netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5128         struct subfacet *subfacet;
5129         struct ofexpired expired;
5130
5131         LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5132             if (subfacet->path == SF_FAST_PATH) {
5133                 struct dpif_flow_stats stats;
5134
5135                 subfacet_install(subfacet, &facet->xout.odp_actions,
5136                                  &stats);
5137                 subfacet_update_stats(subfacet, &stats);
5138             }
5139         }
5140
5141         expired.flow = facet->flow;
5142         expired.packet_count = facet->packet_count;
5143         expired.byte_count = facet->byte_count;
5144         expired.used = facet->used;
5145         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5146     }
5147 }
5148
5149 static void
5150 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5151 {
5152     struct cls_cursor cursor;
5153     struct facet *facet;
5154
5155     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5156     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5157     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5158         send_active_timeout(ofproto, facet);
5159     }
5160     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5161 }
5162 \f
5163 static struct ofproto_dpif *
5164 ofproto_dpif_lookup(const char *name)
5165 {
5166     struct ofproto_dpif *ofproto;
5167
5168     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5169                              hash_string(name, 0), &all_ofproto_dpifs) {
5170         if (!strcmp(ofproto->up.name, name)) {
5171             return ofproto;
5172         }
5173     }
5174     return NULL;
5175 }
5176
5177 static void
5178 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5179                           const char *argv[], void *aux OVS_UNUSED)
5180 {
5181     struct ofproto_dpif *ofproto;
5182
5183     if (argc > 1) {
5184         ofproto = ofproto_dpif_lookup(argv[1]);
5185         if (!ofproto) {
5186             unixctl_command_reply_error(conn, "no such bridge");
5187             return;
5188         }
5189         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5190         mac_learning_flush(ofproto->ml);
5191         ovs_rwlock_unlock(&ofproto->ml->rwlock);
5192     } else {
5193         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5194             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5195             mac_learning_flush(ofproto->ml);
5196             ovs_rwlock_unlock(&ofproto->ml->rwlock);
5197         }
5198     }
5199
5200     unixctl_command_reply(conn, "table successfully flushed");
5201 }
5202
5203 static struct ofport_dpif *
5204 ofbundle_get_a_port(const struct ofbundle *bundle)
5205 {
5206     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5207                         bundle_node);
5208 }
5209
5210 static void
5211 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5212                          const char *argv[], void *aux OVS_UNUSED)
5213 {
5214     struct ds ds = DS_EMPTY_INITIALIZER;
5215     const struct ofproto_dpif *ofproto;
5216     const struct mac_entry *e;
5217
5218     ofproto = ofproto_dpif_lookup(argv[1]);
5219     if (!ofproto) {
5220         unixctl_command_reply_error(conn, "no such bridge");
5221         return;
5222     }
5223
5224     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
5225     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5226     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5227         struct ofbundle *bundle = e->port.p;
5228         char name[OFP_MAX_PORT_NAME_LEN];
5229
5230         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5231                                name, sizeof name);
5232         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
5233                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
5234                       mac_entry_age(ofproto->ml, e));
5235     }
5236     ovs_rwlock_unlock(&ofproto->ml->rwlock);
5237     unixctl_command_reply(conn, ds_cstr(&ds));
5238     ds_destroy(&ds);
5239 }
5240
5241 struct trace_ctx {
5242     struct xlate_out xout;
5243     struct xlate_in xin;
5244     struct flow flow;
5245     struct ds *result;
5246 };
5247
5248 static void
5249 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5250 {
5251     struct rule_actions *actions;
5252     ovs_be64 cookie;
5253
5254     ds_put_char_multiple(result, '\t', level);
5255     if (!rule) {
5256         ds_put_cstr(result, "No match\n");
5257         return;
5258     }
5259
5260     ovs_mutex_lock(&rule->up.mutex);
5261     cookie = rule->up.flow_cookie;
5262     ovs_mutex_unlock(&rule->up.mutex);
5263
5264     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5265                   rule ? rule->up.table_id : 0, ntohll(cookie));
5266     cls_rule_format(&rule->up.cr, result);
5267     ds_put_char(result, '\n');
5268
5269     actions = rule_dpif_get_actions(rule);
5270
5271     ds_put_char_multiple(result, '\t', level);
5272     ds_put_cstr(result, "OpenFlow ");
5273     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
5274     ds_put_char(result, '\n');
5275
5276     rule_actions_unref(actions);
5277 }
5278
5279 static void
5280 trace_format_flow(struct ds *result, int level, const char *title,
5281                   struct trace_ctx *trace)
5282 {
5283     ds_put_char_multiple(result, '\t', level);
5284     ds_put_format(result, "%s: ", title);
5285     if (flow_equal(&trace->xin.flow, &trace->flow)) {
5286         ds_put_cstr(result, "unchanged");
5287     } else {
5288         flow_format(result, &trace->xin.flow);
5289         trace->flow = trace->xin.flow;
5290     }
5291     ds_put_char(result, '\n');
5292 }
5293
5294 static void
5295 trace_format_regs(struct ds *result, int level, const char *title,
5296                   struct trace_ctx *trace)
5297 {
5298     size_t i;
5299
5300     ds_put_char_multiple(result, '\t', level);
5301     ds_put_format(result, "%s:", title);
5302     for (i = 0; i < FLOW_N_REGS; i++) {
5303         ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5304     }
5305     ds_put_char(result, '\n');
5306 }
5307
5308 static void
5309 trace_format_odp(struct ds *result, int level, const char *title,
5310                  struct trace_ctx *trace)
5311 {
5312     struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5313
5314     ds_put_char_multiple(result, '\t', level);
5315     ds_put_format(result, "%s: ", title);
5316     format_odp_actions(result, odp_actions->data, odp_actions->size);
5317     ds_put_char(result, '\n');
5318 }
5319
5320 static void
5321 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5322 {
5323     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5324     struct ds *result = trace->result;
5325
5326     ds_put_char(result, '\n');
5327     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5328     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5329     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
5330     trace_format_rule(result, recurse + 1, rule);
5331 }
5332
5333 static void
5334 trace_report(struct xlate_in *xin, const char *s, int recurse)
5335 {
5336     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5337     struct ds *result = trace->result;
5338
5339     ds_put_char_multiple(result, '\t', recurse);
5340     ds_put_cstr(result, s);
5341     ds_put_char(result, '\n');
5342 }
5343
5344 static void
5345 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5346                       void *aux OVS_UNUSED)
5347 {
5348     const struct dpif_backer *backer;
5349     struct ofproto_dpif *ofproto;
5350     struct ofpbuf odp_key, odp_mask;
5351     struct ofpbuf *packet;
5352     struct ds result;
5353     struct flow flow;
5354     char *s;
5355
5356     packet = NULL;
5357     backer = NULL;
5358     ds_init(&result);
5359     ofpbuf_init(&odp_key, 0);
5360     ofpbuf_init(&odp_mask, 0);
5361
5362     /* Handle "-generate" or a hex string as the last argument. */
5363     if (!strcmp(argv[argc - 1], "-generate")) {
5364         packet = ofpbuf_new(0);
5365         argc--;
5366     } else {
5367         const char *error = eth_from_hex(argv[argc - 1], &packet);
5368         if (!error) {
5369             argc--;
5370         } else if (argc == 4) {
5371             /* The 3-argument form must end in "-generate' or a hex string. */
5372             unixctl_command_reply_error(conn, error);
5373             goto exit;
5374         }
5375     }
5376
5377     /* Parse the flow and determine whether a datapath or
5378      * bridge is specified. If function odp_flow_key_from_string()
5379      * returns 0, the flow is a odp_flow. If function
5380      * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5381     if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5382         /* If the odp_flow is the second argument,
5383          * the datapath name is the first argument. */
5384         if (argc == 3) {
5385             const char *dp_type;
5386             if (!strncmp(argv[1], "ovs-", 4)) {
5387                 dp_type = argv[1] + 4;
5388             } else {
5389                 dp_type = argv[1];
5390             }
5391             backer = shash_find_data(&all_dpif_backers, dp_type);
5392             if (!backer) {
5393                 unixctl_command_reply_error(conn, "Cannot find datapath "
5394                                "of this name");
5395                 goto exit;
5396             }
5397         } else {
5398             /* No datapath name specified, so there should be only one
5399              * datapath. */
5400             struct shash_node *node;
5401             if (shash_count(&all_dpif_backers) != 1) {
5402                 unixctl_command_reply_error(conn, "Must specify datapath "
5403                          "name, there is more than one type of datapath");
5404                 goto exit;
5405             }
5406             node = shash_first(&all_dpif_backers);
5407             backer = node->data;
5408         }
5409
5410         if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5411                           NULL, &ofproto, NULL)) {
5412             unixctl_command_reply_error(conn, "Invalid datapath flow");
5413             goto exit;
5414         }
5415         ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5416     } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5417         if (argc != 3) {
5418             unixctl_command_reply_error(conn, "Must specify bridge name");
5419             goto exit;
5420         }
5421
5422         ofproto = ofproto_dpif_lookup(argv[1]);
5423         if (!ofproto) {
5424             unixctl_command_reply_error(conn, "Unknown bridge name");
5425             goto exit;
5426         }
5427     } else {
5428         unixctl_command_reply_error(conn, "Bad flow syntax");
5429         goto exit;
5430     }
5431
5432     /* Generate a packet, if requested. */
5433     if (packet) {
5434         if (!packet->size) {
5435             flow_compose(packet, &flow);
5436         } else {
5437             union flow_in_port in_port_;
5438
5439             in_port_ = flow.in_port;
5440             ds_put_cstr(&result, "Packet: ");
5441             s = ofp_packet_to_string(packet->data, packet->size);
5442             ds_put_cstr(&result, s);
5443             free(s);
5444
5445             /* Use the metadata from the flow and the packet argument
5446              * to reconstruct the flow. */
5447             flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5448                          &in_port_, &flow);
5449         }
5450     }
5451
5452     ofproto_trace(ofproto, &flow, packet, &result);
5453     unixctl_command_reply(conn, ds_cstr(&result));
5454
5455 exit:
5456     ds_destroy(&result);
5457     ofpbuf_delete(packet);
5458     ofpbuf_uninit(&odp_key);
5459     ofpbuf_uninit(&odp_mask);
5460 }
5461
5462 static void
5463 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5464               const struct ofpbuf *packet, struct ds *ds)
5465 {
5466     struct rule_dpif *rule;
5467     struct flow_wildcards wc;
5468
5469     ds_put_cstr(ds, "Flow: ");
5470     flow_format(ds, flow);
5471     ds_put_char(ds, '\n');
5472
5473     flow_wildcards_init_catchall(&wc);
5474     rule_dpif_lookup(ofproto, flow, &wc, &rule);
5475
5476     trace_format_rule(ds, 0, rule);
5477     if (rule == ofproto->miss_rule) {
5478         ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5479     } else if (rule == ofproto->no_packet_in_rule) {
5480         ds_put_cstr(ds, "\nNo match, packets dropped because "
5481                     "OFPPC_NO_PACKET_IN is set on in_port.\n");
5482     } else if (rule == ofproto->drop_frags_rule) {
5483         ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5484                     "and the fragment handling mode is \"drop\".\n");
5485     }
5486
5487     if (rule) {
5488         uint64_t odp_actions_stub[1024 / 8];
5489         struct ofpbuf odp_actions;
5490         struct trace_ctx trace;
5491         struct match match;
5492         uint8_t tcp_flags;
5493
5494         tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5495         trace.result = ds;
5496         trace.flow = *flow;
5497         ofpbuf_use_stub(&odp_actions,
5498                         odp_actions_stub, sizeof odp_actions_stub);
5499         xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5500         trace.xin.resubmit_hook = trace_resubmit;
5501         trace.xin.report_hook = trace_report;
5502
5503         xlate_actions(&trace.xin, &trace.xout);
5504         flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5505
5506         ds_put_char(ds, '\n');
5507         trace_format_flow(ds, 0, "Final flow", &trace);
5508
5509         match_init(&match, flow, &trace.xout.wc);
5510         ds_put_cstr(ds, "Relevant fields: ");
5511         match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5512         ds_put_char(ds, '\n');
5513
5514         ds_put_cstr(ds, "Datapath actions: ");
5515         format_odp_actions(ds, trace.xout.odp_actions.data,
5516                            trace.xout.odp_actions.size);
5517
5518         if (trace.xout.slow) {
5519             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5520                         "slow path because it:");
5521             switch (trace.xout.slow) {
5522             case SLOW_CFM:
5523                 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5524                 break;
5525             case SLOW_LACP:
5526                 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5527                 break;
5528             case SLOW_STP:
5529                 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5530                 break;
5531             case SLOW_BFD:
5532                 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5533                 break;
5534             case SLOW_CONTROLLER:
5535                 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5536                             "to the OpenFlow controller.");
5537                 break;
5538             case __SLOW_MAX:
5539                 NOT_REACHED();
5540             }
5541         }
5542
5543         xlate_out_uninit(&trace.xout);
5544     }
5545
5546     rule_dpif_unref(rule);
5547 }
5548
5549 /* Runs a self-check of flow translations in 'ofproto'.  Appends a message to
5550  * 'reply' describing the results. */
5551 static void
5552 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5553 {
5554     struct cls_cursor cursor;
5555     struct facet *facet;
5556     int errors;
5557
5558     errors = 0;
5559     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5560     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5561     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5562         if (!facet_check_consistency(facet)) {
5563             errors++;
5564         }
5565     }
5566     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5567     if (errors) {
5568         ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5569     }
5570
5571     if (errors) {
5572         ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5573                       ofproto->up.name, errors);
5574     } else {
5575         ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5576     }
5577 }
5578
5579 static void
5580 ofproto_dpif_self_check(struct unixctl_conn *conn,
5581                         int argc, const char *argv[], void *aux OVS_UNUSED)
5582 {
5583     struct ds reply = DS_EMPTY_INITIALIZER;
5584     struct ofproto_dpif *ofproto;
5585
5586     if (argc > 1) {
5587         ofproto = ofproto_dpif_lookup(argv[1]);
5588         if (!ofproto) {
5589             unixctl_command_reply_error(conn, "Unknown ofproto (use "
5590                                         "ofproto/list for help)");
5591             return;
5592         }
5593         ofproto_dpif_self_check__(ofproto, &reply);
5594     } else {
5595         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5596             ofproto_dpif_self_check__(ofproto, &reply);
5597         }
5598     }
5599
5600     unixctl_command_reply(conn, ds_cstr(&reply));
5601     ds_destroy(&reply);
5602 }
5603
5604 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
5605  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
5606  * to destroy 'ofproto_shash' and free the returned value. */
5607 static const struct shash_node **
5608 get_ofprotos(struct shash *ofproto_shash)
5609 {
5610     const struct ofproto_dpif *ofproto;
5611
5612     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5613         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5614         shash_add_nocopy(ofproto_shash, name, ofproto);
5615     }
5616
5617     return shash_sort(ofproto_shash);
5618 }
5619
5620 static void
5621 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5622                               const char *argv[] OVS_UNUSED,
5623                               void *aux OVS_UNUSED)
5624 {
5625     struct ds ds = DS_EMPTY_INITIALIZER;
5626     struct shash ofproto_shash;
5627     const struct shash_node **sorted_ofprotos;
5628     int i;
5629
5630     shash_init(&ofproto_shash);
5631     sorted_ofprotos = get_ofprotos(&ofproto_shash);
5632     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5633         const struct shash_node *node = sorted_ofprotos[i];
5634         ds_put_format(&ds, "%s\n", node->name);
5635     }
5636
5637     shash_destroy(&ofproto_shash);
5638     free(sorted_ofprotos);
5639
5640     unixctl_command_reply(conn, ds_cstr(&ds));
5641     ds_destroy(&ds);
5642 }
5643
5644 static void
5645 show_dp_rates(struct ds *ds, const char *heading,
5646               const struct avg_subfacet_rates *rates)
5647 {
5648     ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5649                   heading, rates->add_rate, rates->del_rate);
5650 }
5651
5652 static void
5653 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5654 {
5655     const struct shash_node **ofprotos;
5656     struct ofproto_dpif *ofproto;
5657     struct shash ofproto_shash;
5658     uint64_t n_hit, n_missed;
5659     long long int minutes;
5660     size_t i;
5661
5662     n_hit = n_missed = 0;
5663     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5664         if (ofproto->backer == backer) {
5665             n_missed += ofproto->n_missed;
5666             n_hit += ofproto->n_hit;
5667         }
5668     }
5669
5670     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5671                   dpif_name(backer->dpif), n_hit, n_missed);
5672     ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5673                   " life span: %lldms\n", hmap_count(&backer->subfacets),
5674                   backer->avg_n_subfacet, backer->max_n_subfacet,
5675                   backer->avg_subfacet_life);
5676
5677     minutes = (time_msec() - backer->created) / (1000 * 60);
5678     if (minutes >= 60) {
5679         show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5680     }
5681     if (minutes >= 60 * 24) {
5682         show_dp_rates(ds, "\tdaily avg:",  &backer->daily);
5683     }
5684     show_dp_rates(ds, "\toverall avg:",  &backer->lifetime);
5685
5686     shash_init(&ofproto_shash);
5687     ofprotos = get_ofprotos(&ofproto_shash);
5688     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5689         struct ofproto_dpif *ofproto = ofprotos[i]->data;
5690         const struct shash_node **ports;
5691         size_t j;
5692
5693         if (ofproto->backer != backer) {
5694             continue;
5695         }
5696
5697         ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5698                       ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5699
5700         ports = shash_sort(&ofproto->up.port_by_name);
5701         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5702             const struct shash_node *node = ports[j];
5703             struct ofport *ofport = node->data;
5704             struct smap config;
5705             odp_port_t odp_port;
5706
5707             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5708                           ofport->ofp_port);
5709
5710             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5711             if (odp_port != ODPP_NONE) {
5712                 ds_put_format(ds, "%"PRIu32":", odp_port);
5713             } else {
5714                 ds_put_cstr(ds, "none:");
5715             }
5716
5717             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5718
5719             smap_init(&config);
5720             if (!netdev_get_config(ofport->netdev, &config)) {
5721                 const struct smap_node **nodes;
5722                 size_t i;
5723
5724                 nodes = smap_sort(&config);
5725                 for (i = 0; i < smap_count(&config); i++) {
5726                     const struct smap_node *node = nodes[i];
5727                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5728                                   node->key, node->value);
5729                 }
5730                 free(nodes);
5731             }
5732             smap_destroy(&config);
5733
5734             ds_put_char(ds, ')');
5735             ds_put_char(ds, '\n');
5736         }
5737         free(ports);
5738     }
5739     shash_destroy(&ofproto_shash);
5740     free(ofprotos);
5741 }
5742
5743 static void
5744 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5745                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5746 {
5747     struct ds ds = DS_EMPTY_INITIALIZER;
5748     const struct shash_node **backers;
5749     int i;
5750
5751     backers = shash_sort(&all_dpif_backers);
5752     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5753         dpif_show_backer(backers[i]->data, &ds);
5754     }
5755     free(backers);
5756
5757     unixctl_command_reply(conn, ds_cstr(&ds));
5758     ds_destroy(&ds);
5759 }
5760
5761 /* Dump the megaflow (facet) cache.  This is useful to check the
5762  * correctness of flow wildcarding, since the same mechanism is used for
5763  * both xlate caching and kernel wildcarding.
5764  *
5765  * It's important to note that in the output the flow description uses
5766  * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5767  *
5768  * This command is only needed for advanced debugging, so it's not
5769  * documented in the man page. */
5770 static void
5771 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5772                                     int argc OVS_UNUSED, const char *argv[],
5773                                     void *aux OVS_UNUSED)
5774 {
5775     struct ds ds = DS_EMPTY_INITIALIZER;
5776     const struct ofproto_dpif *ofproto;
5777     long long int now = time_msec();
5778     struct cls_cursor cursor;
5779     struct facet *facet;
5780
5781     ofproto = ofproto_dpif_lookup(argv[1]);
5782     if (!ofproto) {
5783         unixctl_command_reply_error(conn, "no such bridge");
5784         return;
5785     }
5786
5787     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5788     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5789     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5790         cls_rule_format(&facet->cr, &ds);
5791         ds_put_cstr(&ds, ", ");
5792         ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5793         ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5794         ds_put_cstr(&ds, "Datapath actions: ");
5795         if (facet->xout.slow) {
5796             uint64_t slow_path_stub[128 / 8];
5797             const struct nlattr *actions;
5798             size_t actions_len;
5799
5800             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5801                               slow_path_stub, sizeof slow_path_stub,
5802                               &actions, &actions_len);
5803             format_odp_actions(&ds, actions, actions_len);
5804         } else {
5805             format_odp_actions(&ds, facet->xout.odp_actions.data,
5806                                facet->xout.odp_actions.size);
5807         }
5808         ds_put_cstr(&ds, "\n");
5809     }
5810     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5811
5812     ds_chomp(&ds, '\n');
5813     unixctl_command_reply(conn, ds_cstr(&ds));
5814     ds_destroy(&ds);
5815 }
5816
5817 /* Disable using the megaflows.
5818  *
5819  * This command is only needed for advanced debugging, so it's not
5820  * documented in the man page. */
5821 static void
5822 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5823                                        int argc OVS_UNUSED,
5824                                        const char *argv[] OVS_UNUSED,
5825                                        void *aux OVS_UNUSED)
5826 {
5827     struct ofproto_dpif *ofproto;
5828
5829     enable_megaflows = false;
5830
5831     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5832         flush(&ofproto->up);
5833     }
5834
5835     unixctl_command_reply(conn, "megaflows disabled");
5836 }
5837
5838 /* Re-enable using megaflows.
5839  *
5840  * This command is only needed for advanced debugging, so it's not
5841  * documented in the man page. */
5842 static void
5843 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5844                                       int argc OVS_UNUSED,
5845                                       const char *argv[] OVS_UNUSED,
5846                                       void *aux OVS_UNUSED)
5847 {
5848     struct ofproto_dpif *ofproto;
5849
5850     enable_megaflows = true;
5851
5852     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5853         flush(&ofproto->up);
5854     }
5855
5856     unixctl_command_reply(conn, "megaflows enabled");
5857 }
5858
5859 static void
5860 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5861                                 int argc OVS_UNUSED, const char *argv[],
5862                                 void *aux OVS_UNUSED)
5863 {
5864     struct ds ds = DS_EMPTY_INITIALIZER;
5865     const struct ofproto_dpif *ofproto;
5866     struct subfacet *subfacet;
5867
5868     ofproto = ofproto_dpif_lookup(argv[1]);
5869     if (!ofproto) {
5870         unixctl_command_reply_error(conn, "no such bridge");
5871         return;
5872     }
5873
5874     update_stats(ofproto->backer);
5875
5876     HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5877         struct facet *facet = subfacet->facet;
5878         struct odputil_keybuf maskbuf;
5879         struct ofpbuf mask;
5880
5881         if (facet->ofproto != ofproto) {
5882             continue;
5883         }
5884
5885         ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5886         if (enable_megaflows) {
5887             odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5888                                    &facet->flow, UINT32_MAX);
5889         }
5890
5891         odp_flow_format(subfacet->key, subfacet->key_len,
5892                         mask.data, mask.size, &ds, false);
5893
5894         ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5895                       subfacet->dp_packet_count, subfacet->dp_byte_count);
5896         if (subfacet->used) {
5897             ds_put_format(&ds, "%.3fs",
5898                           (time_msec() - subfacet->used) / 1000.0);
5899         } else {
5900             ds_put_format(&ds, "never");
5901         }
5902         if (subfacet->facet->tcp_flags) {
5903             ds_put_cstr(&ds, ", flags:");
5904             packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5905         }
5906
5907         ds_put_cstr(&ds, ", actions:");
5908         if (facet->xout.slow) {
5909             uint64_t slow_path_stub[128 / 8];
5910             const struct nlattr *actions;
5911             size_t actions_len;
5912
5913             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5914                               slow_path_stub, sizeof slow_path_stub,
5915                               &actions, &actions_len);
5916             format_odp_actions(&ds, actions, actions_len);
5917         } else {
5918             format_odp_actions(&ds, facet->xout.odp_actions.data,
5919                                facet->xout.odp_actions.size);
5920         }
5921         ds_put_char(&ds, '\n');
5922     }
5923
5924     unixctl_command_reply(conn, ds_cstr(&ds));
5925     ds_destroy(&ds);
5926 }
5927
5928 static void
5929 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5930                                int argc OVS_UNUSED, const char *argv[],
5931                                void *aux OVS_UNUSED)
5932 {
5933     struct ds ds = DS_EMPTY_INITIALIZER;
5934     struct ofproto_dpif *ofproto;
5935
5936     ofproto = ofproto_dpif_lookup(argv[1]);
5937     if (!ofproto) {
5938         unixctl_command_reply_error(conn, "no such bridge");
5939         return;
5940     }
5941
5942     flush(&ofproto->up);
5943
5944     unixctl_command_reply(conn, ds_cstr(&ds));
5945     ds_destroy(&ds);
5946 }
5947
5948 static void
5949 ofproto_dpif_unixctl_init(void)
5950 {
5951     static bool registered;
5952     if (registered) {
5953         return;
5954     }
5955     registered = true;
5956
5957     unixctl_command_register(
5958         "ofproto/trace",
5959         "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
5960         1, 3, ofproto_unixctl_trace, NULL);
5961     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5962                              ofproto_unixctl_fdb_flush, NULL);
5963     unixctl_command_register("fdb/show", "bridge", 1, 1,
5964                              ofproto_unixctl_fdb_show, NULL);
5965     unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
5966                              ofproto_dpif_self_check, NULL);
5967     unixctl_command_register("dpif/dump-dps", "", 0, 0,
5968                              ofproto_unixctl_dpif_dump_dps, NULL);
5969     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5970                              NULL);
5971     unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
5972                              ofproto_unixctl_dpif_dump_flows, NULL);
5973     unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
5974                              ofproto_unixctl_dpif_del_flows, NULL);
5975     unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
5976                              ofproto_unixctl_dpif_dump_megaflows, NULL);
5977     unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
5978                              ofproto_unixctl_dpif_disable_megaflows, NULL);
5979     unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
5980                              ofproto_unixctl_dpif_enable_megaflows, NULL);
5981 }
5982 \f
5983 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5984  *
5985  * This is deprecated.  It is only for compatibility with broken device drivers
5986  * in old versions of Linux that do not properly support VLANs when VLAN
5987  * devices are not used.  When broken device drivers are no longer in
5988  * widespread use, we will delete these interfaces. */
5989
5990 static int
5991 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5992 {
5993     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5994     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5995
5996     if (realdev_ofp_port == ofport->realdev_ofp_port
5997         && vid == ofport->vlandev_vid) {
5998         return 0;
5999     }
6000
6001     ofproto->backer->need_revalidate = REV_RECONFIGURE;
6002
6003     if (ofport->realdev_ofp_port) {
6004         vsp_remove(ofport);
6005     }
6006     if (realdev_ofp_port && ofport->bundle) {
6007         /* vlandevs are enslaved to their realdevs, so they are not allowed to
6008          * themselves be part of a bundle. */
6009         bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6010     }
6011
6012     ofport->realdev_ofp_port = realdev_ofp_port;
6013     ofport->vlandev_vid = vid;
6014
6015     if (realdev_ofp_port) {
6016         vsp_add(ofport, realdev_ofp_port, vid);
6017     }
6018
6019     return 0;
6020 }
6021
6022 static uint32_t
6023 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6024 {
6025     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6026 }
6027
6028 bool
6029 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6030     OVS_EXCLUDED(ofproto->vsp_mutex)
6031 {
6032     bool ret;
6033
6034     ovs_mutex_lock(&ofproto->vsp_mutex);
6035     ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6036     ovs_mutex_unlock(&ofproto->vsp_mutex);
6037     return ret;
6038 }
6039
6040 static ofp_port_t
6041 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6042                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6043     OVS_REQUIRES(ofproto->vsp_mutex)
6044 {
6045     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6046         int vid = vlan_tci_to_vid(vlan_tci);
6047         const struct vlan_splinter *vsp;
6048
6049         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6050                                  hash_realdev_vid(realdev_ofp_port, vid),
6051                                  &ofproto->realdev_vid_map) {
6052             if (vsp->realdev_ofp_port == realdev_ofp_port
6053                 && vsp->vid == vid) {
6054                 return vsp->vlandev_ofp_port;
6055             }
6056         }
6057     }
6058     return realdev_ofp_port;
6059 }
6060
6061 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6062  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6063  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
6064  * 'vlan_tci' 9, it would return the port number of eth0.9.
6065  *
6066  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6067  * function just returns its 'realdev_ofp_port' argument. */
6068 ofp_port_t
6069 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6070                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6071     OVS_EXCLUDED(ofproto->vsp_mutex)
6072 {
6073     ofp_port_t ret;
6074
6075     ovs_mutex_lock(&ofproto->vsp_mutex);
6076     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6077     ovs_mutex_unlock(&ofproto->vsp_mutex);
6078     return ret;
6079 }
6080
6081 static struct vlan_splinter *
6082 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6083 {
6084     struct vlan_splinter *vsp;
6085
6086     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6087                              hash_ofp_port(vlandev_ofp_port),
6088                              &ofproto->vlandev_map) {
6089         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6090             return vsp;
6091         }
6092     }
6093
6094     return NULL;
6095 }
6096
6097 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6098  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6099  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
6100  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6101  * eth0 and store 9 in '*vid'.
6102  *
6103  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6104  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
6105  * always does.*/
6106 static ofp_port_t
6107 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6108                        ofp_port_t vlandev_ofp_port, int *vid)
6109     OVS_REQUIRES(ofproto->vsp_mutex)
6110 {
6111     if (!hmap_is_empty(&ofproto->vlandev_map)) {
6112         const struct vlan_splinter *vsp;
6113
6114         vsp = vlandev_find(ofproto, vlandev_ofp_port);
6115         if (vsp) {
6116             if (vid) {
6117                 *vid = vsp->vid;
6118             }
6119             return vsp->realdev_ofp_port;
6120         }
6121     }
6122     return 0;
6123 }
6124
6125 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6126  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
6127  * 'flow->in_port' to the "real" device backing the VLAN device, sets
6128  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Otherwise (which is
6129  * always the case unless VLAN splinters are enabled), returns false without
6130  * making any changes. */
6131 bool
6132 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6133     OVS_EXCLUDED(ofproto->vsp_mutex)
6134 {
6135     ofp_port_t realdev;
6136     int vid;
6137
6138     ovs_mutex_lock(&ofproto->vsp_mutex);
6139     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6140     ovs_mutex_unlock(&ofproto->vsp_mutex);
6141     if (!realdev) {
6142         return false;
6143     }
6144
6145     /* Cause the flow to be processed as if it came in on the real device with
6146      * the VLAN device's VLAN ID. */
6147     flow->in_port.ofp_port = realdev;
6148     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6149     return true;
6150 }
6151
6152 static void
6153 vsp_remove(struct ofport_dpif *port)
6154 {
6155     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6156     struct vlan_splinter *vsp;
6157
6158     ovs_mutex_lock(&ofproto->vsp_mutex);
6159     vsp = vlandev_find(ofproto, port->up.ofp_port);
6160     if (vsp) {
6161         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6162         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6163         free(vsp);
6164
6165         port->realdev_ofp_port = 0;
6166     } else {
6167         VLOG_ERR("missing vlan device record");
6168     }
6169     ovs_mutex_unlock(&ofproto->vsp_mutex);
6170 }
6171
6172 static void
6173 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6174 {
6175     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6176
6177     ovs_mutex_lock(&ofproto->vsp_mutex);
6178     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6179         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6180             == realdev_ofp_port)) {
6181         struct vlan_splinter *vsp;
6182
6183         vsp = xmalloc(sizeof *vsp);
6184         vsp->realdev_ofp_port = realdev_ofp_port;
6185         vsp->vlandev_ofp_port = port->up.ofp_port;
6186         vsp->vid = vid;
6187
6188         port->realdev_ofp_port = realdev_ofp_port;
6189
6190         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6191                     hash_ofp_port(port->up.ofp_port));
6192         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6193                     hash_realdev_vid(realdev_ofp_port, vid));
6194     } else {
6195         VLOG_ERR("duplicate vlan device record");
6196     }
6197     ovs_mutex_unlock(&ofproto->vsp_mutex);
6198 }
6199
6200 static odp_port_t
6201 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6202 {
6203     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6204     return ofport ? ofport->odp_port : ODPP_NONE;
6205 }
6206
6207 struct ofport_dpif *
6208 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6209 {
6210     struct ofport_dpif *port;
6211
6212     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6213     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6214                              &backer->odp_to_ofport_map) {
6215         if (port->odp_port == odp_port) {
6216             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6217             return port;
6218         }
6219     }
6220
6221     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6222     return NULL;
6223 }
6224
6225 static ofp_port_t
6226 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6227 {
6228     struct ofport_dpif *port;
6229
6230     port = odp_port_to_ofport(ofproto->backer, odp_port);
6231     if (port && &ofproto->up == port->up.ofproto) {
6232         return port->up.ofp_port;
6233     } else {
6234         return OFPP_NONE;
6235     }
6236 }
6237
6238 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6239  * most heavily weighted element.  'base' designates the rate of decay: after
6240  * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6241  * (about .37). */
6242 static void
6243 exp_mavg(double *avg, int base, double new)
6244 {
6245     *avg = (*avg * (base - 1) + new) / base;
6246 }
6247
6248 static void
6249 update_moving_averages(struct dpif_backer *backer)
6250 {
6251     const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6252     long long int minutes = (time_msec() - backer->created) / min_ms;
6253
6254     if (minutes > 0) {
6255         backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6256             / minutes;
6257         backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6258             / minutes;
6259     } else {
6260         backer->lifetime.add_rate = 0.0;
6261         backer->lifetime.del_rate = 0.0;
6262     }
6263
6264     /* Update hourly averages on the minute boundaries. */
6265     if (time_msec() - backer->last_minute >= min_ms) {
6266         exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6267         exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6268
6269         /* Update daily averages on the hour boundaries. */
6270         if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6271             exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6272             exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6273         }
6274
6275         backer->total_subfacet_add_count += backer->subfacet_add_count;
6276         backer->total_subfacet_del_count += backer->subfacet_del_count;
6277         backer->subfacet_add_count = 0;
6278         backer->subfacet_del_count = 0;
6279         backer->last_minute += min_ms;
6280     }
6281 }
6282
6283 const struct ofproto_class ofproto_dpif_class = {
6284     init,
6285     enumerate_types,
6286     enumerate_names,
6287     del,
6288     port_open_type,
6289     type_run,
6290     type_run_fast,
6291     type_wait,
6292     alloc,
6293     construct,
6294     destruct,
6295     dealloc,
6296     run,
6297     run_fast,
6298     wait,
6299     get_memory_usage,
6300     flush,
6301     get_features,
6302     get_tables,
6303     port_alloc,
6304     port_construct,
6305     port_destruct,
6306     port_dealloc,
6307     port_modified,
6308     port_reconfigured,
6309     port_query_by_name,
6310     port_add,
6311     port_del,
6312     port_get_stats,
6313     port_dump_start,
6314     port_dump_next,
6315     port_dump_done,
6316     port_poll,
6317     port_poll_wait,
6318     port_is_lacp_current,
6319     NULL,                       /* rule_choose_table */
6320     rule_alloc,
6321     rule_construct,
6322     rule_insert,
6323     rule_delete,
6324     rule_destruct,
6325     rule_dealloc,
6326     rule_get_stats,
6327     rule_execute,
6328     rule_modify_actions,
6329     set_frag_handling,
6330     packet_out,
6331     set_netflow,
6332     get_netflow_ids,
6333     set_sflow,
6334     set_ipfix,
6335     set_cfm,
6336     get_cfm_status,
6337     set_bfd,
6338     get_bfd_status,
6339     set_stp,
6340     get_stp_status,
6341     set_stp_port,
6342     get_stp_port_status,
6343     set_queues,
6344     bundle_set,
6345     bundle_remove,
6346     mirror_set__,
6347     mirror_get_stats__,
6348     set_flood_vlans,
6349     is_mirror_output_bundle,
6350     forward_bpdu_changed,
6351     set_mac_table_config,
6352     set_realdev,
6353     NULL,                       /* meter_get_features */
6354     NULL,                       /* meter_set */
6355     NULL,                       /* meter_get */
6356     NULL,                       /* meter_del */
6357     NULL,                       /* group_alloc */
6358     NULL,                       /* group_construct */
6359     NULL,                       /* group_destruct */
6360     NULL,                       /* group_dealloc */
6361     NULL,                       /* group_modify */
6362     NULL,                       /* group_get_stats */
6363 };