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     udpif_destroy(backer->udpif);
1094
1095     simap_destroy(&backer->tnl_backers);
1096     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1097     hmap_destroy(&backer->odp_to_ofport_map);
1098     node = shash_find(&all_dpif_backers, backer->type);
1099     free(backer->type);
1100     shash_delete(&all_dpif_backers, node);
1101     dpif_close(backer->dpif);
1102
1103     ovs_assert(hmap_is_empty(&backer->subfacets));
1104     hmap_destroy(&backer->subfacets);
1105     governor_destroy(backer->governor);
1106
1107     free(backer);
1108 }
1109
1110 /* Datapath port slated for removal from datapath. */
1111 struct odp_garbage {
1112     struct list list_node;
1113     odp_port_t odp_port;
1114 };
1115
1116 static int
1117 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1118 {
1119     struct dpif_backer *backer;
1120     struct dpif_port_dump port_dump;
1121     struct dpif_port port;
1122     struct shash_node *node;
1123     struct list garbage_list;
1124     struct odp_garbage *garbage, *next;
1125     struct sset names;
1126     char *backer_name;
1127     const char *name;
1128     int error;
1129
1130     backer = shash_find_data(&all_dpif_backers, type);
1131     if (backer) {
1132         backer->refcount++;
1133         *backerp = backer;
1134         return 0;
1135     }
1136
1137     backer_name = xasprintf("ovs-%s", type);
1138
1139     /* Remove any existing datapaths, since we assume we're the only
1140      * userspace controlling the datapath. */
1141     sset_init(&names);
1142     dp_enumerate_names(type, &names);
1143     SSET_FOR_EACH(name, &names) {
1144         struct dpif *old_dpif;
1145
1146         /* Don't remove our backer if it exists. */
1147         if (!strcmp(name, backer_name)) {
1148             continue;
1149         }
1150
1151         if (dpif_open(name, type, &old_dpif)) {
1152             VLOG_WARN("couldn't open old datapath %s to remove it", name);
1153         } else {
1154             dpif_delete(old_dpif);
1155             dpif_close(old_dpif);
1156         }
1157     }
1158     sset_destroy(&names);
1159
1160     backer = xmalloc(sizeof *backer);
1161
1162     error = dpif_create_and_open(backer_name, type, &backer->dpif);
1163     free(backer_name);
1164     if (error) {
1165         VLOG_ERR("failed to open datapath of type %s: %s", type,
1166                  ovs_strerror(error));
1167         free(backer);
1168         return error;
1169     }
1170     backer->udpif = udpif_create(backer, backer->dpif);
1171
1172     backer->type = xstrdup(type);
1173     backer->governor = NULL;
1174     backer->refcount = 1;
1175     hmap_init(&backer->odp_to_ofport_map);
1176     ovs_rwlock_init(&backer->odp_to_ofport_lock);
1177     hmap_init(&backer->drop_keys);
1178     hmap_init(&backer->subfacets);
1179     timer_set_duration(&backer->next_expiration, 1000);
1180     backer->need_revalidate = 0;
1181     simap_init(&backer->tnl_backers);
1182     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1183     *backerp = backer;
1184
1185     if (backer->recv_set_enable) {
1186         dpif_flow_flush(backer->dpif);
1187     }
1188
1189     /* Loop through the ports already on the datapath and remove any
1190      * that we don't need anymore. */
1191     list_init(&garbage_list);
1192     dpif_port_dump_start(&port_dump, backer->dpif);
1193     while (dpif_port_dump_next(&port_dump, &port)) {
1194         node = shash_find(&init_ofp_ports, port.name);
1195         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1196             garbage = xmalloc(sizeof *garbage);
1197             garbage->odp_port = port.port_no;
1198             list_push_front(&garbage_list, &garbage->list_node);
1199         }
1200     }
1201     dpif_port_dump_done(&port_dump);
1202
1203     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1204         dpif_port_del(backer->dpif, garbage->odp_port);
1205         list_remove(&garbage->list_node);
1206         free(garbage);
1207     }
1208
1209     shash_add(&all_dpif_backers, type, backer);
1210
1211     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1212     if (error) {
1213         VLOG_ERR("failed to listen on datapath of type %s: %s",
1214                  type, ovs_strerror(error));
1215         close_dpif_backer(backer);
1216         return error;
1217     }
1218     udpif_recv_set(backer->udpif, n_handler_threads,
1219                    backer->recv_set_enable);
1220     backer->n_handler_threads = n_handler_threads;
1221
1222     backer->max_n_subfacet = 0;
1223     backer->created = time_msec();
1224     backer->last_minute = backer->created;
1225     memset(&backer->hourly, 0, sizeof backer->hourly);
1226     memset(&backer->daily, 0, sizeof backer->daily);
1227     memset(&backer->lifetime, 0, sizeof backer->lifetime);
1228     backer->subfacet_add_count = 0;
1229     backer->subfacet_del_count = 0;
1230     backer->total_subfacet_add_count = 0;
1231     backer->total_subfacet_del_count = 0;
1232     backer->avg_n_subfacet = 0;
1233     backer->avg_subfacet_life = 0;
1234
1235     return error;
1236 }
1237
1238 static int
1239 construct(struct ofproto *ofproto_)
1240 {
1241     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1242     struct shash_node *node, *next;
1243     uint32_t max_ports;
1244     int error;
1245
1246     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1247     if (error) {
1248         return error;
1249     }
1250
1251     max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1252     ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX)));
1253
1254     ofproto->netflow = NULL;
1255     ofproto->sflow = NULL;
1256     ofproto->ipfix = NULL;
1257     ofproto->stp = NULL;
1258     hmap_init(&ofproto->bundles);
1259     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1260     ofproto->mbridge = mbridge_create();
1261     ofproto->has_bonded_bundles = false;
1262     ovs_mutex_init(&ofproto->vsp_mutex);
1263
1264     classifier_init(&ofproto->facets);
1265     ofproto->consistency_rl = LLONG_MIN;
1266
1267     guarded_list_init(&ofproto->pins);
1268
1269     ofproto_dpif_unixctl_init();
1270
1271     hmap_init(&ofproto->vlandev_map);
1272     hmap_init(&ofproto->realdev_vid_map);
1273
1274     sset_init(&ofproto->ports);
1275     sset_init(&ofproto->ghost_ports);
1276     sset_init(&ofproto->port_poll_set);
1277     ofproto->port_poll_errno = 0;
1278
1279     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1280         struct iface_hint *iface_hint = node->data;
1281
1282         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1283             /* Check if the datapath already has this port. */
1284             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1285                 sset_add(&ofproto->ports, node->name);
1286             }
1287
1288             free(iface_hint->br_name);
1289             free(iface_hint->br_type);
1290             free(iface_hint);
1291             shash_delete(&init_ofp_ports, node);
1292         }
1293     }
1294
1295     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1296                 hash_string(ofproto->up.name, 0));
1297     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1298
1299     ofproto_init_tables(ofproto_, N_TABLES);
1300     error = add_internal_flows(ofproto);
1301     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1302
1303     ofproto->n_hit = 0;
1304     ofproto->n_missed = 0;
1305
1306     return error;
1307 }
1308
1309 static int
1310 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1311                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1312 {
1313     struct ofputil_flow_mod fm;
1314     int error;
1315
1316     match_init_catchall(&fm.match);
1317     fm.priority = 0;
1318     match_set_reg(&fm.match, 0, id);
1319     fm.new_cookie = htonll(0);
1320     fm.cookie = htonll(0);
1321     fm.cookie_mask = htonll(0);
1322     fm.modify_cookie = false;
1323     fm.table_id = TBL_INTERNAL;
1324     fm.command = OFPFC_ADD;
1325     fm.idle_timeout = 0;
1326     fm.hard_timeout = 0;
1327     fm.buffer_id = 0;
1328     fm.out_port = 0;
1329     fm.flags = 0;
1330     fm.ofpacts = ofpacts->data;
1331     fm.ofpacts_len = ofpacts->size;
1332
1333     error = ofproto_flow_mod(&ofproto->up, &fm);
1334     if (error) {
1335         VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1336                     id, ofperr_to_string(error));
1337         return error;
1338     }
1339
1340     if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1341                                   rulep)) {
1342         rule_dpif_unref(*rulep);
1343     } else {
1344         NOT_REACHED();
1345     }
1346
1347     return 0;
1348 }
1349
1350 static int
1351 add_internal_flows(struct ofproto_dpif *ofproto)
1352 {
1353     struct ofpact_controller *controller;
1354     uint64_t ofpacts_stub[128 / 8];
1355     struct ofpbuf ofpacts;
1356     int error;
1357     int id;
1358
1359     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1360     id = 1;
1361
1362     controller = ofpact_put_CONTROLLER(&ofpacts);
1363     controller->max_len = UINT16_MAX;
1364     controller->controller_id = 0;
1365     controller->reason = OFPR_NO_MATCH;
1366     ofpact_pad(&ofpacts);
1367
1368     error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1369     if (error) {
1370         return error;
1371     }
1372
1373     ofpbuf_clear(&ofpacts);
1374     error = add_internal_flow(ofproto, id++, &ofpacts,
1375                               &ofproto->no_packet_in_rule);
1376     if (error) {
1377         return error;
1378     }
1379
1380     error = add_internal_flow(ofproto, id++, &ofpacts,
1381                               &ofproto->drop_frags_rule);
1382     return error;
1383 }
1384
1385 static void
1386 destruct(struct ofproto *ofproto_)
1387 {
1388     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1389     struct rule_dpif *rule, *next_rule;
1390     struct ofputil_packet_in *pin, *next_pin;
1391     struct facet *facet, *next_facet;
1392     struct cls_cursor cursor;
1393     struct oftable *table;
1394     struct list pins;
1395
1396     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1397     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1398     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1399     CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1400         facet_remove(facet);
1401     }
1402
1403     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1404     ovs_rwlock_wrlock(&xlate_rwlock);
1405     xlate_remove_ofproto(ofproto);
1406     ovs_rwlock_unlock(&xlate_rwlock);
1407
1408     /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a
1409      * use-after-free error. */
1410     udpif_revalidate(ofproto->backer->udpif);
1411
1412     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1413
1414     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1415         struct cls_cursor cursor;
1416
1417         ovs_rwlock_rdlock(&table->cls.rwlock);
1418         cls_cursor_init(&cursor, &table->cls, NULL);
1419         ovs_rwlock_unlock(&table->cls.rwlock);
1420         CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1421             ofproto_rule_delete(&ofproto->up, &rule->up);
1422         }
1423     }
1424
1425     guarded_list_pop_all(&ofproto->pins, &pins);
1426     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1427         list_remove(&pin->list_node);
1428         free(CONST_CAST(void *, pin->packet));
1429         free(pin);
1430     }
1431     guarded_list_destroy(&ofproto->pins);
1432
1433     mbridge_unref(ofproto->mbridge);
1434
1435     netflow_destroy(ofproto->netflow);
1436     dpif_sflow_unref(ofproto->sflow);
1437     hmap_destroy(&ofproto->bundles);
1438     mac_learning_unref(ofproto->ml);
1439
1440     classifier_destroy(&ofproto->facets);
1441
1442     hmap_destroy(&ofproto->vlandev_map);
1443     hmap_destroy(&ofproto->realdev_vid_map);
1444
1445     sset_destroy(&ofproto->ports);
1446     sset_destroy(&ofproto->ghost_ports);
1447     sset_destroy(&ofproto->port_poll_set);
1448
1449     ovs_mutex_destroy(&ofproto->vsp_mutex);
1450
1451     close_dpif_backer(ofproto->backer);
1452 }
1453
1454 static int
1455 run_fast(struct ofproto *ofproto_)
1456 {
1457     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1458     struct ofputil_packet_in *pin, *next_pin;
1459     struct ofport_dpif *ofport;
1460     struct list pins;
1461
1462     /* Do not perform any periodic activity required by 'ofproto' while
1463      * waiting for flow restore to complete. */
1464     if (ofproto_get_flow_restore_wait()) {
1465         return 0;
1466     }
1467
1468     guarded_list_pop_all(&ofproto->pins, &pins);
1469     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1470         connmgr_send_packet_in(ofproto->up.connmgr, pin);
1471         list_remove(&pin->list_node);
1472         free(CONST_CAST(void *, pin->packet));
1473         free(pin);
1474     }
1475
1476     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1477         port_run_fast(ofport);
1478     }
1479
1480     return 0;
1481 }
1482
1483 static int
1484 run(struct ofproto *ofproto_)
1485 {
1486     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1487     struct ofport_dpif *ofport;
1488     struct ofbundle *bundle;
1489     int error;
1490
1491     if (mbridge_need_revalidate(ofproto->mbridge)) {
1492         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1493         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1494         mac_learning_flush(ofproto->ml);
1495         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1496     }
1497
1498     /* Do not perform any periodic activity below required by 'ofproto' while
1499      * waiting for flow restore to complete. */
1500     if (ofproto_get_flow_restore_wait()) {
1501         return 0;
1502     }
1503
1504     error = run_fast(ofproto_);
1505     if (error) {
1506         return error;
1507     }
1508
1509     if (ofproto->netflow) {
1510         if (netflow_run(ofproto->netflow)) {
1511             send_netflow_active_timeouts(ofproto);
1512         }
1513     }
1514     if (ofproto->sflow) {
1515         dpif_sflow_run(ofproto->sflow);
1516     }
1517     if (ofproto->ipfix) {
1518         dpif_ipfix_run(ofproto->ipfix);
1519     }
1520
1521     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1522         port_run(ofport);
1523     }
1524     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1525         bundle_run(bundle);
1526     }
1527
1528     stp_run(ofproto);
1529     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1530     if (mac_learning_run(ofproto->ml)) {
1531         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1532     }
1533     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1534
1535     /* Check the consistency of a random facet, to aid debugging. */
1536     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1537     if (time_msec() >= ofproto->consistency_rl
1538         && !classifier_is_empty(&ofproto->facets)
1539         && !ofproto->backer->need_revalidate) {
1540         struct cls_table *table;
1541         struct cls_rule *cr;
1542         struct facet *facet;
1543
1544         ofproto->consistency_rl = time_msec() + 250;
1545
1546         table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1547                              struct cls_table, hmap_node);
1548         cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1549                           hmap_node);
1550         facet = CONTAINER_OF(cr, struct facet, cr);
1551
1552         if (!facet_check_consistency(facet)) {
1553             ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1554         }
1555     }
1556     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1557
1558     return 0;
1559 }
1560
1561 static void
1562 wait(struct ofproto *ofproto_)
1563 {
1564     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1565     struct ofport_dpif *ofport;
1566     struct ofbundle *bundle;
1567
1568     if (ofproto_get_flow_restore_wait()) {
1569         return;
1570     }
1571
1572     if (ofproto->sflow) {
1573         dpif_sflow_wait(ofproto->sflow);
1574     }
1575     if (ofproto->ipfix) {
1576         dpif_ipfix_wait(ofproto->ipfix);
1577     }
1578     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1579         port_wait(ofport);
1580     }
1581     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1582         bundle_wait(bundle);
1583     }
1584     if (ofproto->netflow) {
1585         netflow_wait(ofproto->netflow);
1586     }
1587     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1588     mac_learning_wait(ofproto->ml);
1589     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1590     stp_wait(ofproto);
1591     if (ofproto->backer->need_revalidate) {
1592         /* Shouldn't happen, but if it does just go around again. */
1593         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1594         poll_immediate_wake();
1595     }
1596 }
1597
1598 static void
1599 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1600 {
1601     const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1602     struct cls_cursor cursor;
1603     size_t n_subfacets = 0;
1604     struct facet *facet;
1605
1606     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1607     simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1608     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1609
1610     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1611     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1612     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1613         n_subfacets += list_size(&facet->subfacets);
1614     }
1615     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1616     simap_increase(usage, "subfacets", n_subfacets);
1617 }
1618
1619 static void
1620 flush(struct ofproto *ofproto_)
1621 {
1622     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1623     struct subfacet *subfacet, *next_subfacet;
1624     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1625     int n_batch;
1626
1627     n_batch = 0;
1628     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1629                         &ofproto->backer->subfacets) {
1630         if (subfacet->facet->ofproto != ofproto) {
1631             continue;
1632         }
1633
1634         if (subfacet->path != SF_NOT_INSTALLED) {
1635             batch[n_batch++] = subfacet;
1636             if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1637                 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1638                 n_batch = 0;
1639             }
1640         } else {
1641             subfacet_destroy(subfacet);
1642         }
1643     }
1644
1645     if (n_batch > 0) {
1646         subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1647     }
1648 }
1649
1650 static void
1651 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1652              bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1653 {
1654     *arp_match_ip = true;
1655     *actions = (OFPUTIL_A_OUTPUT |
1656                 OFPUTIL_A_SET_VLAN_VID |
1657                 OFPUTIL_A_SET_VLAN_PCP |
1658                 OFPUTIL_A_STRIP_VLAN |
1659                 OFPUTIL_A_SET_DL_SRC |
1660                 OFPUTIL_A_SET_DL_DST |
1661                 OFPUTIL_A_SET_NW_SRC |
1662                 OFPUTIL_A_SET_NW_DST |
1663                 OFPUTIL_A_SET_NW_TOS |
1664                 OFPUTIL_A_SET_TP_SRC |
1665                 OFPUTIL_A_SET_TP_DST |
1666                 OFPUTIL_A_ENQUEUE);
1667 }
1668
1669 static void
1670 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1671 {
1672     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1673     struct dpif_dp_stats s;
1674     uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1675     uint64_t n_lookup;
1676
1677     strcpy(ots->name, "classifier");
1678
1679     dpif_get_dp_stats(ofproto->backer->dpif, &s);
1680     rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1681     rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1682     rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1683
1684     n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1685     ots->lookup_count = htonll(n_lookup);
1686     ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1687 }
1688
1689 static struct ofport *
1690 port_alloc(void)
1691 {
1692     struct ofport_dpif *port = xmalloc(sizeof *port);
1693     return &port->up;
1694 }
1695
1696 static void
1697 port_dealloc(struct ofport *port_)
1698 {
1699     struct ofport_dpif *port = ofport_dpif_cast(port_);
1700     free(port);
1701 }
1702
1703 static int
1704 port_construct(struct ofport *port_)
1705 {
1706     struct ofport_dpif *port = ofport_dpif_cast(port_);
1707     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1708     const struct netdev *netdev = port->up.netdev;
1709     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1710     struct dpif_port dpif_port;
1711     int error;
1712
1713     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1714     port->bundle = NULL;
1715     port->cfm = NULL;
1716     port->bfd = NULL;
1717     port->may_enable = true;
1718     port->stp_port = NULL;
1719     port->stp_state = STP_DISABLED;
1720     port->is_tunnel = false;
1721     port->peer = NULL;
1722     port->qdscp = NULL;
1723     port->n_qdscp = 0;
1724     port->realdev_ofp_port = 0;
1725     port->vlandev_vid = 0;
1726     port->carrier_seq = netdev_get_carrier_resets(netdev);
1727
1728     if (netdev_vport_is_patch(netdev)) {
1729         /* By bailing out here, we don't submit the port to the sFlow module
1730          * to be considered for counter polling export.  This is correct
1731          * because the patch port represents an interface that sFlow considers
1732          * to be "internal" to the switch as a whole, and therefore not an
1733          * candidate for counter polling. */
1734         port->odp_port = ODPP_NONE;
1735         ofport_update_peer(port);
1736         return 0;
1737     }
1738
1739     error = dpif_port_query_by_name(ofproto->backer->dpif,
1740                                     netdev_vport_get_dpif_port(netdev, namebuf,
1741                                                                sizeof namebuf),
1742                                     &dpif_port);
1743     if (error) {
1744         return error;
1745     }
1746
1747     port->odp_port = dpif_port.port_no;
1748
1749     if (netdev_get_tunnel_config(netdev)) {
1750         tnl_port_add(port, port->up.netdev, port->odp_port);
1751         port->is_tunnel = true;
1752     } else {
1753         /* Sanity-check that a mapping doesn't already exist.  This
1754          * shouldn't happen for non-tunnel ports. */
1755         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1756             VLOG_ERR("port %s already has an OpenFlow port number",
1757                      dpif_port.name);
1758             dpif_port_destroy(&dpif_port);
1759             return EBUSY;
1760         }
1761
1762         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1763         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1764                     hash_odp_port(port->odp_port));
1765         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1766     }
1767     dpif_port_destroy(&dpif_port);
1768
1769     if (ofproto->sflow) {
1770         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1771     }
1772
1773     return 0;
1774 }
1775
1776 static void
1777 port_destruct(struct ofport *port_)
1778 {
1779     struct ofport_dpif *port = ofport_dpif_cast(port_);
1780     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1781     const char *devname = netdev_get_name(port->up.netdev);
1782     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1783     const char *dp_port_name;
1784
1785     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1786     ovs_rwlock_wrlock(&xlate_rwlock);
1787     xlate_ofport_remove(port);
1788     ovs_rwlock_unlock(&xlate_rwlock);
1789
1790     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1791                                               sizeof namebuf);
1792     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1793         /* The underlying device is still there, so delete it.  This
1794          * happens when the ofproto is being destroyed, since the caller
1795          * assumes that removal of attached ports will happen as part of
1796          * destruction. */
1797         if (!port->is_tunnel) {
1798             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1799         }
1800     }
1801
1802     if (port->peer) {
1803         port->peer->peer = NULL;
1804         port->peer = NULL;
1805     }
1806
1807     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1808         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1809         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1810         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1811     }
1812
1813     tnl_port_del(port);
1814     sset_find_and_delete(&ofproto->ports, devname);
1815     sset_find_and_delete(&ofproto->ghost_ports, devname);
1816     bundle_remove(port_);
1817     set_cfm(port_, NULL);
1818     set_bfd(port_, NULL);
1819     if (ofproto->sflow) {
1820         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1821     }
1822
1823     free(port->qdscp);
1824 }
1825
1826 static void
1827 port_modified(struct ofport *port_)
1828 {
1829     struct ofport_dpif *port = ofport_dpif_cast(port_);
1830
1831     if (port->bundle && port->bundle->bond) {
1832         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1833     }
1834
1835     if (port->cfm) {
1836         cfm_set_netdev(port->cfm, port->up.netdev);
1837     }
1838
1839     if (port->bfd) {
1840         bfd_set_netdev(port->bfd, port->up.netdev);
1841     }
1842
1843     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1844                                                 port->odp_port)) {
1845         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1846             REV_RECONFIGURE;
1847     }
1848
1849     ofport_update_peer(port);
1850 }
1851
1852 static void
1853 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1854 {
1855     struct ofport_dpif *port = ofport_dpif_cast(port_);
1856     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1857     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1858
1859     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1860                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1861                    OFPUTIL_PC_NO_PACKET_IN)) {
1862         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1863
1864         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1865             bundle_update(port->bundle);
1866         }
1867     }
1868 }
1869
1870 static int
1871 set_sflow(struct ofproto *ofproto_,
1872           const struct ofproto_sflow_options *sflow_options)
1873 {
1874     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1875     struct dpif_sflow *ds = ofproto->sflow;
1876
1877     if (sflow_options) {
1878         if (!ds) {
1879             struct ofport_dpif *ofport;
1880
1881             ds = ofproto->sflow = dpif_sflow_create();
1882             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1883                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1884             }
1885             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1886         }
1887         dpif_sflow_set_options(ds, sflow_options);
1888     } else {
1889         if (ds) {
1890             dpif_sflow_unref(ds);
1891             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1892             ofproto->sflow = NULL;
1893         }
1894     }
1895     return 0;
1896 }
1897
1898 static int
1899 set_ipfix(
1900     struct ofproto *ofproto_,
1901     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1902     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1903     size_t n_flow_exporters_options)
1904 {
1905     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1906     struct dpif_ipfix *di = ofproto->ipfix;
1907     bool has_options = bridge_exporter_options || flow_exporters_options;
1908
1909     if (has_options && !di) {
1910         di = ofproto->ipfix = dpif_ipfix_create();
1911     }
1912
1913     if (di) {
1914         /* Call set_options in any case to cleanly flush the flow
1915          * caches in the last exporters that are to be destroyed. */
1916         dpif_ipfix_set_options(
1917             di, bridge_exporter_options, flow_exporters_options,
1918             n_flow_exporters_options);
1919
1920         if (!has_options) {
1921             dpif_ipfix_unref(di);
1922             ofproto->ipfix = NULL;
1923         }
1924     }
1925
1926     return 0;
1927 }
1928
1929 static int
1930 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1931 {
1932     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1933     int error;
1934
1935     if (!s) {
1936         error = 0;
1937     } else {
1938         if (!ofport->cfm) {
1939             struct ofproto_dpif *ofproto;
1940
1941             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1942             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1943             ofport->cfm = cfm_create(ofport->up.netdev);
1944         }
1945
1946         if (cfm_configure(ofport->cfm, s)) {
1947             return 0;
1948         }
1949
1950         error = EINVAL;
1951     }
1952     cfm_unref(ofport->cfm);
1953     ofport->cfm = NULL;
1954     return error;
1955 }
1956
1957 static bool
1958 get_cfm_status(const struct ofport *ofport_,
1959                struct ofproto_cfm_status *status)
1960 {
1961     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1962
1963     if (ofport->cfm) {
1964         status->faults = cfm_get_fault(ofport->cfm);
1965         status->remote_opstate = cfm_get_opup(ofport->cfm);
1966         status->health = cfm_get_health(ofport->cfm);
1967         cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1968         return true;
1969     } else {
1970         return false;
1971     }
1972 }
1973
1974 static int
1975 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1976 {
1977     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1978     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1979     struct bfd *old;
1980
1981     old = ofport->bfd;
1982     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1983                                 cfg, ofport->up.netdev);
1984     if (ofport->bfd != old) {
1985         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1986     }
1987
1988     return 0;
1989 }
1990
1991 static int
1992 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1993 {
1994     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1995
1996     if (ofport->bfd) {
1997         bfd_get_status(ofport->bfd, smap);
1998         return 0;
1999     } else {
2000         return ENOENT;
2001     }
2002 }
2003 \f
2004 /* Spanning Tree. */
2005
2006 static void
2007 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2008 {
2009     struct ofproto_dpif *ofproto = ofproto_;
2010     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2011     struct ofport_dpif *ofport;
2012
2013     ofport = stp_port_get_aux(sp);
2014     if (!ofport) {
2015         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2016                      ofproto->up.name, port_num);
2017     } else {
2018         struct eth_header *eth = pkt->l2;
2019
2020         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2021         if (eth_addr_is_zero(eth->eth_src)) {
2022             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2023                          "with unknown MAC", ofproto->up.name, port_num);
2024         } else {
2025             send_packet(ofport, pkt);
2026         }
2027     }
2028     ofpbuf_delete(pkt);
2029 }
2030
2031 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2032 static int
2033 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2034 {
2035     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2036
2037     /* Only revalidate flows if the configuration changed. */
2038     if (!s != !ofproto->stp) {
2039         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2040     }
2041
2042     if (s) {
2043         if (!ofproto->stp) {
2044             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2045                                       send_bpdu_cb, ofproto);
2046             ofproto->stp_last_tick = time_msec();
2047         }
2048
2049         stp_set_bridge_id(ofproto->stp, s->system_id);
2050         stp_set_bridge_priority(ofproto->stp, s->priority);
2051         stp_set_hello_time(ofproto->stp, s->hello_time);
2052         stp_set_max_age(ofproto->stp, s->max_age);
2053         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2054     }  else {
2055         struct ofport *ofport;
2056
2057         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2058             set_stp_port(ofport, NULL);
2059         }
2060
2061         stp_unref(ofproto->stp);
2062         ofproto->stp = NULL;
2063     }
2064
2065     return 0;
2066 }
2067
2068 static int
2069 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2070 {
2071     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2072
2073     if (ofproto->stp) {
2074         s->enabled = true;
2075         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2076         s->designated_root = stp_get_designated_root(ofproto->stp);
2077         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2078     } else {
2079         s->enabled = false;
2080     }
2081
2082     return 0;
2083 }
2084
2085 static void
2086 update_stp_port_state(struct ofport_dpif *ofport)
2087 {
2088     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2089     enum stp_state state;
2090
2091     /* Figure out new state. */
2092     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2093                              : STP_DISABLED;
2094
2095     /* Update state. */
2096     if (ofport->stp_state != state) {
2097         enum ofputil_port_state of_state;
2098         bool fwd_change;
2099
2100         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2101                     netdev_get_name(ofport->up.netdev),
2102                     stp_state_name(ofport->stp_state),
2103                     stp_state_name(state));
2104         if (stp_learn_in_state(ofport->stp_state)
2105                 != stp_learn_in_state(state)) {
2106             /* xxx Learning action flows should also be flushed. */
2107             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2108             mac_learning_flush(ofproto->ml);
2109             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2110         }
2111         fwd_change = stp_forward_in_state(ofport->stp_state)
2112                         != stp_forward_in_state(state);
2113
2114         ofproto->backer->need_revalidate = REV_STP;
2115         ofport->stp_state = state;
2116         ofport->stp_state_entered = time_msec();
2117
2118         if (fwd_change && ofport->bundle) {
2119             bundle_update(ofport->bundle);
2120         }
2121
2122         /* Update the STP state bits in the OpenFlow port description. */
2123         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2124         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2125                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2126                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2127                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2128                      : 0);
2129         ofproto_port_set_state(&ofport->up, of_state);
2130     }
2131 }
2132
2133 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2134  * caller is responsible for assigning STP port numbers and ensuring
2135  * there are no duplicates. */
2136 static int
2137 set_stp_port(struct ofport *ofport_,
2138              const struct ofproto_port_stp_settings *s)
2139 {
2140     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2141     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2142     struct stp_port *sp = ofport->stp_port;
2143
2144     if (!s || !s->enable) {
2145         if (sp) {
2146             ofport->stp_port = NULL;
2147             stp_port_disable(sp);
2148             update_stp_port_state(ofport);
2149         }
2150         return 0;
2151     } else if (sp && stp_port_no(sp) != s->port_num
2152             && ofport == stp_port_get_aux(sp)) {
2153         /* The port-id changed, so disable the old one if it's not
2154          * already in use by another port. */
2155         stp_port_disable(sp);
2156     }
2157
2158     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2159     stp_port_enable(sp);
2160
2161     stp_port_set_aux(sp, ofport);
2162     stp_port_set_priority(sp, s->priority);
2163     stp_port_set_path_cost(sp, s->path_cost);
2164
2165     update_stp_port_state(ofport);
2166
2167     return 0;
2168 }
2169
2170 static int
2171 get_stp_port_status(struct ofport *ofport_,
2172                     struct ofproto_port_stp_status *s)
2173 {
2174     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2175     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2176     struct stp_port *sp = ofport->stp_port;
2177
2178     if (!ofproto->stp || !sp) {
2179         s->enabled = false;
2180         return 0;
2181     }
2182
2183     s->enabled = true;
2184     s->port_id = stp_port_get_id(sp);
2185     s->state = stp_port_get_state(sp);
2186     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2187     s->role = stp_port_get_role(sp);
2188     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2189
2190     return 0;
2191 }
2192
2193 static void
2194 stp_run(struct ofproto_dpif *ofproto)
2195 {
2196     if (ofproto->stp) {
2197         long long int now = time_msec();
2198         long long int elapsed = now - ofproto->stp_last_tick;
2199         struct stp_port *sp;
2200
2201         if (elapsed > 0) {
2202             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2203             ofproto->stp_last_tick = now;
2204         }
2205         while (stp_get_changed_port(ofproto->stp, &sp)) {
2206             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2207
2208             if (ofport) {
2209                 update_stp_port_state(ofport);
2210             }
2211         }
2212
2213         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2214             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2215             mac_learning_flush(ofproto->ml);
2216             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2217         }
2218     }
2219 }
2220
2221 static void
2222 stp_wait(struct ofproto_dpif *ofproto)
2223 {
2224     if (ofproto->stp) {
2225         poll_timer_wait(1000);
2226     }
2227 }
2228 \f
2229 static int
2230 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2231            size_t n_qdscp)
2232 {
2233     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2234     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2235
2236     if (ofport->n_qdscp != n_qdscp
2237         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2238                               n_qdscp * sizeof *qdscp))) {
2239         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2240         free(ofport->qdscp);
2241         ofport->qdscp = n_qdscp
2242             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2243             : NULL;
2244         ofport->n_qdscp = n_qdscp;
2245     }
2246
2247     return 0;
2248 }
2249 \f
2250 /* Bundles. */
2251
2252 /* Expires all MAC learning entries associated with 'bundle' and forces its
2253  * ofproto to revalidate every flow.
2254  *
2255  * Normally MAC learning entries are removed only from the ofproto associated
2256  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2257  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2258  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2259  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2260  * with the host from which it migrated. */
2261 static void
2262 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2263 {
2264     struct ofproto_dpif *ofproto = bundle->ofproto;
2265     struct mac_learning *ml = ofproto->ml;
2266     struct mac_entry *mac, *next_mac;
2267
2268     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2269     ovs_rwlock_wrlock(&ml->rwlock);
2270     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2271         if (mac->port.p == bundle) {
2272             if (all_ofprotos) {
2273                 struct ofproto_dpif *o;
2274
2275                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2276                     if (o != ofproto) {
2277                         struct mac_entry *e;
2278
2279                         ovs_rwlock_wrlock(&o->ml->rwlock);
2280                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2281                         if (e) {
2282                             mac_learning_expire(o->ml, e);
2283                         }
2284                         ovs_rwlock_unlock(&o->ml->rwlock);
2285                     }
2286                 }
2287             }
2288
2289             mac_learning_expire(ml, mac);
2290         }
2291     }
2292     ovs_rwlock_unlock(&ml->rwlock);
2293 }
2294
2295 static struct ofbundle *
2296 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2297 {
2298     struct ofbundle *bundle;
2299
2300     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2301                              &ofproto->bundles) {
2302         if (bundle->aux == aux) {
2303             return bundle;
2304         }
2305     }
2306     return NULL;
2307 }
2308
2309 static void
2310 bundle_update(struct ofbundle *bundle)
2311 {
2312     struct ofport_dpif *port;
2313
2314     bundle->floodable = true;
2315     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2316         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2317             || !stp_forward_in_state(port->stp_state)) {
2318             bundle->floodable = false;
2319             break;
2320         }
2321     }
2322 }
2323
2324 static void
2325 bundle_del_port(struct ofport_dpif *port)
2326 {
2327     struct ofbundle *bundle = port->bundle;
2328
2329     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2330
2331     list_remove(&port->bundle_node);
2332     port->bundle = NULL;
2333
2334     if (bundle->lacp) {
2335         lacp_slave_unregister(bundle->lacp, port);
2336     }
2337     if (bundle->bond) {
2338         bond_slave_unregister(bundle->bond, port);
2339     }
2340
2341     bundle_update(bundle);
2342 }
2343
2344 static bool
2345 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2346                 struct lacp_slave_settings *lacp)
2347 {
2348     struct ofport_dpif *port;
2349
2350     port = get_ofp_port(bundle->ofproto, ofp_port);
2351     if (!port) {
2352         return false;
2353     }
2354
2355     if (port->bundle != bundle) {
2356         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2357         if (port->bundle) {
2358             bundle_remove(&port->up);
2359         }
2360
2361         port->bundle = bundle;
2362         list_push_back(&bundle->ports, &port->bundle_node);
2363         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2364             || !stp_forward_in_state(port->stp_state)) {
2365             bundle->floodable = false;
2366         }
2367     }
2368     if (lacp) {
2369         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2370         lacp_slave_register(bundle->lacp, port, lacp);
2371     }
2372
2373     return true;
2374 }
2375
2376 static void
2377 bundle_destroy(struct ofbundle *bundle)
2378 {
2379     struct ofproto_dpif *ofproto;
2380     struct ofport_dpif *port, *next_port;
2381
2382     if (!bundle) {
2383         return;
2384     }
2385
2386     ofproto = bundle->ofproto;
2387     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2388
2389     ovs_rwlock_wrlock(&xlate_rwlock);
2390     xlate_bundle_remove(bundle);
2391     ovs_rwlock_unlock(&xlate_rwlock);
2392
2393     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2394         bundle_del_port(port);
2395     }
2396
2397     bundle_flush_macs(bundle, true);
2398     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2399     free(bundle->name);
2400     free(bundle->trunks);
2401     lacp_unref(bundle->lacp);
2402     bond_unref(bundle->bond);
2403     free(bundle);
2404 }
2405
2406 static int
2407 bundle_set(struct ofproto *ofproto_, void *aux,
2408            const struct ofproto_bundle_settings *s)
2409 {
2410     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2411     bool need_flush = false;
2412     struct ofport_dpif *port;
2413     struct ofbundle *bundle;
2414     unsigned long *trunks;
2415     int vlan;
2416     size_t i;
2417     bool ok;
2418
2419     if (!s) {
2420         bundle_destroy(bundle_lookup(ofproto, aux));
2421         return 0;
2422     }
2423
2424     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2425     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2426
2427     bundle = bundle_lookup(ofproto, aux);
2428     if (!bundle) {
2429         bundle = xmalloc(sizeof *bundle);
2430
2431         bundle->ofproto = ofproto;
2432         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2433                     hash_pointer(aux, 0));
2434         bundle->aux = aux;
2435         bundle->name = NULL;
2436
2437         list_init(&bundle->ports);
2438         bundle->vlan_mode = PORT_VLAN_TRUNK;
2439         bundle->vlan = -1;
2440         bundle->trunks = NULL;
2441         bundle->use_priority_tags = s->use_priority_tags;
2442         bundle->lacp = NULL;
2443         bundle->bond = NULL;
2444
2445         bundle->floodable = true;
2446         mbridge_register_bundle(ofproto->mbridge, bundle);
2447     }
2448
2449     if (!bundle->name || strcmp(s->name, bundle->name)) {
2450         free(bundle->name);
2451         bundle->name = xstrdup(s->name);
2452     }
2453
2454     /* LACP. */
2455     if (s->lacp) {
2456         if (!bundle->lacp) {
2457             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2458             bundle->lacp = lacp_create();
2459         }
2460         lacp_configure(bundle->lacp, s->lacp);
2461     } else {
2462         lacp_unref(bundle->lacp);
2463         bundle->lacp = NULL;
2464     }
2465
2466     /* Update set of ports. */
2467     ok = true;
2468     for (i = 0; i < s->n_slaves; i++) {
2469         if (!bundle_add_port(bundle, s->slaves[i],
2470                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2471             ok = false;
2472         }
2473     }
2474     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2475         struct ofport_dpif *next_port;
2476
2477         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2478             for (i = 0; i < s->n_slaves; i++) {
2479                 if (s->slaves[i] == port->up.ofp_port) {
2480                     goto found;
2481                 }
2482             }
2483
2484             bundle_del_port(port);
2485         found: ;
2486         }
2487     }
2488     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2489
2490     if (list_is_empty(&bundle->ports)) {
2491         bundle_destroy(bundle);
2492         return EINVAL;
2493     }
2494
2495     /* Set VLAN tagging mode */
2496     if (s->vlan_mode != bundle->vlan_mode
2497         || s->use_priority_tags != bundle->use_priority_tags) {
2498         bundle->vlan_mode = s->vlan_mode;
2499         bundle->use_priority_tags = s->use_priority_tags;
2500         need_flush = true;
2501     }
2502
2503     /* Set VLAN tag. */
2504     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2505             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2506             : 0);
2507     if (vlan != bundle->vlan) {
2508         bundle->vlan = vlan;
2509         need_flush = true;
2510     }
2511
2512     /* Get trunked VLANs. */
2513     switch (s->vlan_mode) {
2514     case PORT_VLAN_ACCESS:
2515         trunks = NULL;
2516         break;
2517
2518     case PORT_VLAN_TRUNK:
2519         trunks = CONST_CAST(unsigned long *, s->trunks);
2520         break;
2521
2522     case PORT_VLAN_NATIVE_UNTAGGED:
2523     case PORT_VLAN_NATIVE_TAGGED:
2524         if (vlan != 0 && (!s->trunks
2525                           || !bitmap_is_set(s->trunks, vlan)
2526                           || bitmap_is_set(s->trunks, 0))) {
2527             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2528             if (s->trunks) {
2529                 trunks = bitmap_clone(s->trunks, 4096);
2530             } else {
2531                 trunks = bitmap_allocate1(4096);
2532             }
2533             bitmap_set1(trunks, vlan);
2534             bitmap_set0(trunks, 0);
2535         } else {
2536             trunks = CONST_CAST(unsigned long *, s->trunks);
2537         }
2538         break;
2539
2540     default:
2541         NOT_REACHED();
2542     }
2543     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2544         free(bundle->trunks);
2545         if (trunks == s->trunks) {
2546             bundle->trunks = vlan_bitmap_clone(trunks);
2547         } else {
2548             bundle->trunks = trunks;
2549             trunks = NULL;
2550         }
2551         need_flush = true;
2552     }
2553     if (trunks != s->trunks) {
2554         free(trunks);
2555     }
2556
2557     /* Bonding. */
2558     if (!list_is_short(&bundle->ports)) {
2559         bundle->ofproto->has_bonded_bundles = true;
2560         if (bundle->bond) {
2561             if (bond_reconfigure(bundle->bond, s->bond)) {
2562                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2563             }
2564         } else {
2565             bundle->bond = bond_create(s->bond);
2566             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2567         }
2568
2569         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2570             bond_slave_register(bundle->bond, port, port->up.netdev);
2571         }
2572     } else {
2573         bond_unref(bundle->bond);
2574         bundle->bond = NULL;
2575     }
2576
2577     /* If we changed something that would affect MAC learning, un-learn
2578      * everything on this port and force flow revalidation. */
2579     if (need_flush) {
2580         bundle_flush_macs(bundle, false);
2581     }
2582
2583     return 0;
2584 }
2585
2586 static void
2587 bundle_remove(struct ofport *port_)
2588 {
2589     struct ofport_dpif *port = ofport_dpif_cast(port_);
2590     struct ofbundle *bundle = port->bundle;
2591
2592     if (bundle) {
2593         bundle_del_port(port);
2594         if (list_is_empty(&bundle->ports)) {
2595             bundle_destroy(bundle);
2596         } else if (list_is_short(&bundle->ports)) {
2597             bond_unref(bundle->bond);
2598             bundle->bond = NULL;
2599         }
2600     }
2601 }
2602
2603 static void
2604 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2605 {
2606     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2607     struct ofport_dpif *port = port_;
2608     uint8_t ea[ETH_ADDR_LEN];
2609     int error;
2610
2611     error = netdev_get_etheraddr(port->up.netdev, ea);
2612     if (!error) {
2613         struct ofpbuf packet;
2614         void *packet_pdu;
2615
2616         ofpbuf_init(&packet, 0);
2617         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2618                                  pdu_size);
2619         memcpy(packet_pdu, pdu, pdu_size);
2620
2621         send_packet(port, &packet);
2622         ofpbuf_uninit(&packet);
2623     } else {
2624         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2625                     "%s (%s)", port->bundle->name,
2626                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2627     }
2628 }
2629
2630 static void
2631 bundle_send_learning_packets(struct ofbundle *bundle)
2632 {
2633     struct ofproto_dpif *ofproto = bundle->ofproto;
2634     struct ofpbuf *learning_packet;
2635     int error, n_packets, n_errors;
2636     struct mac_entry *e;
2637     struct list packets;
2638
2639     list_init(&packets);
2640     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2641     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2642         if (e->port.p != bundle) {
2643             void *port_void;
2644
2645             learning_packet = bond_compose_learning_packet(bundle->bond,
2646                                                            e->mac, e->vlan,
2647                                                            &port_void);
2648             learning_packet->private_p = port_void;
2649             list_push_back(&packets, &learning_packet->list_node);
2650         }
2651     }
2652     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2653
2654     error = n_packets = n_errors = 0;
2655     LIST_FOR_EACH (learning_packet, list_node, &packets) {
2656         int ret;
2657
2658         ret = send_packet(learning_packet->private_p, learning_packet);
2659         if (ret) {
2660             error = ret;
2661             n_errors++;
2662         }
2663         n_packets++;
2664     }
2665     ofpbuf_list_delete(&packets);
2666
2667     if (n_errors) {
2668         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2669         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2670                      "packets, last error was: %s",
2671                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2672     } else {
2673         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2674                  bundle->name, n_packets);
2675     }
2676 }
2677
2678 static void
2679 bundle_run(struct ofbundle *bundle)
2680 {
2681     if (bundle->lacp) {
2682         lacp_run(bundle->lacp, send_pdu_cb);
2683     }
2684     if (bundle->bond) {
2685         struct ofport_dpif *port;
2686
2687         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2688             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2689         }
2690
2691         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2692             bundle->ofproto->backer->need_revalidate = REV_BOND;
2693         }
2694
2695         if (bond_should_send_learning_packets(bundle->bond)) {
2696             bundle_send_learning_packets(bundle);
2697         }
2698     }
2699 }
2700
2701 static void
2702 bundle_wait(struct ofbundle *bundle)
2703 {
2704     if (bundle->lacp) {
2705         lacp_wait(bundle->lacp);
2706     }
2707     if (bundle->bond) {
2708         bond_wait(bundle->bond);
2709     }
2710 }
2711 \f
2712 /* Mirrors. */
2713
2714 static int
2715 mirror_set__(struct ofproto *ofproto_, void *aux,
2716              const struct ofproto_mirror_settings *s)
2717 {
2718     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2719     struct ofbundle **srcs, **dsts;
2720     int error;
2721     size_t i;
2722
2723     if (!s) {
2724         mirror_destroy(ofproto->mbridge, aux);
2725         return 0;
2726     }
2727
2728     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2729     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2730
2731     for (i = 0; i < s->n_srcs; i++) {
2732         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2733     }
2734
2735     for (i = 0; i < s->n_dsts; i++) {
2736         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2737     }
2738
2739     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2740                        s->n_dsts, s->src_vlans,
2741                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2742     free(srcs);
2743     free(dsts);
2744     return error;
2745 }
2746
2747 static int
2748 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2749                    uint64_t *packets, uint64_t *bytes)
2750 {
2751     push_all_stats();
2752     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2753                             bytes);
2754 }
2755
2756 static int
2757 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2758 {
2759     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2760     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2761     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2762         mac_learning_flush(ofproto->ml);
2763     }
2764     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2765     return 0;
2766 }
2767
2768 static bool
2769 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2770 {
2771     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2772     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2773     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2774 }
2775
2776 static void
2777 forward_bpdu_changed(struct ofproto *ofproto_)
2778 {
2779     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2780     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2781 }
2782
2783 static void
2784 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2785                      size_t max_entries)
2786 {
2787     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2788     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2789     mac_learning_set_idle_time(ofproto->ml, idle_time);
2790     mac_learning_set_max_entries(ofproto->ml, max_entries);
2791     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2792 }
2793 \f
2794 /* Ports. */
2795
2796 static struct ofport_dpif *
2797 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2798 {
2799     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2800     return ofport ? ofport_dpif_cast(ofport) : NULL;
2801 }
2802
2803 static struct ofport_dpif *
2804 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2805 {
2806     struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2807     return port && &ofproto->up == port->up.ofproto ? port : NULL;
2808 }
2809
2810 static void
2811 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2812                             struct ofproto_port *ofproto_port,
2813                             struct dpif_port *dpif_port)
2814 {
2815     ofproto_port->name = dpif_port->name;
2816     ofproto_port->type = dpif_port->type;
2817     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2818 }
2819
2820 static void
2821 ofport_update_peer(struct ofport_dpif *ofport)
2822 {
2823     const struct ofproto_dpif *ofproto;
2824     struct dpif_backer *backer;
2825     char *peer_name;
2826
2827     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2828         return;
2829     }
2830
2831     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2832     backer->need_revalidate = REV_RECONFIGURE;
2833
2834     if (ofport->peer) {
2835         ofport->peer->peer = NULL;
2836         ofport->peer = NULL;
2837     }
2838
2839     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2840     if (!peer_name) {
2841         return;
2842     }
2843
2844     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2845         struct ofport *peer_ofport;
2846         struct ofport_dpif *peer;
2847         char *peer_peer;
2848
2849         if (ofproto->backer != backer) {
2850             continue;
2851         }
2852
2853         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2854         if (!peer_ofport) {
2855             continue;
2856         }
2857
2858         peer = ofport_dpif_cast(peer_ofport);
2859         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2860         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2861                                  peer_peer)) {
2862             ofport->peer = peer;
2863             ofport->peer->peer = ofport;
2864         }
2865         free(peer_peer);
2866
2867         break;
2868     }
2869     free(peer_name);
2870 }
2871
2872 static void
2873 port_run_fast(struct ofport_dpif *ofport)
2874 {
2875     if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2876         struct ofpbuf packet;
2877
2878         ofpbuf_init(&packet, 0);
2879         cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2880         send_packet(ofport, &packet);
2881         ofpbuf_uninit(&packet);
2882     }
2883
2884     if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2885         struct ofpbuf packet;
2886
2887         ofpbuf_init(&packet, 0);
2888         bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2889         send_packet(ofport, &packet);
2890         ofpbuf_uninit(&packet);
2891     }
2892 }
2893
2894 static void
2895 port_run(struct ofport_dpif *ofport)
2896 {
2897     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2898     bool carrier_changed = carrier_seq != ofport->carrier_seq;
2899     bool enable = netdev_get_carrier(ofport->up.netdev);
2900     bool cfm_enable = false;
2901     bool bfd_enable = false;
2902
2903     ofport->carrier_seq = carrier_seq;
2904
2905     port_run_fast(ofport);
2906
2907     if (ofport->cfm) {
2908         int cfm_opup = cfm_get_opup(ofport->cfm);
2909
2910         cfm_run(ofport->cfm);
2911         cfm_enable = !cfm_get_fault(ofport->cfm);
2912
2913         if (cfm_opup >= 0) {
2914             cfm_enable = cfm_enable && cfm_opup;
2915         }
2916     }
2917
2918     if (ofport->bfd) {
2919         bfd_run(ofport->bfd);
2920         bfd_enable = bfd_forwarding(ofport->bfd);
2921     }
2922
2923     if (ofport->bfd || ofport->cfm) {
2924         enable = enable && (cfm_enable || bfd_enable);
2925     }
2926
2927     if (ofport->bundle) {
2928         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2929         if (carrier_changed) {
2930             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2931         }
2932     }
2933
2934     if (ofport->may_enable != enable) {
2935         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2936         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2937     }
2938
2939     ofport->may_enable = enable;
2940 }
2941
2942 static void
2943 port_wait(struct ofport_dpif *ofport)
2944 {
2945     if (ofport->cfm) {
2946         cfm_wait(ofport->cfm);
2947     }
2948
2949     if (ofport->bfd) {
2950         bfd_wait(ofport->bfd);
2951     }
2952 }
2953
2954 static int
2955 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2956                    struct ofproto_port *ofproto_port)
2957 {
2958     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2959     struct dpif_port dpif_port;
2960     int error;
2961
2962     if (sset_contains(&ofproto->ghost_ports, devname)) {
2963         const char *type = netdev_get_type_from_name(devname);
2964
2965         /* We may be called before ofproto->up.port_by_name is populated with
2966          * the appropriate ofport.  For this reason, we must get the name and
2967          * type from the netdev layer directly. */
2968         if (type) {
2969             const struct ofport *ofport;
2970
2971             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2972             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2973             ofproto_port->name = xstrdup(devname);
2974             ofproto_port->type = xstrdup(type);
2975             return 0;
2976         }
2977         return ENODEV;
2978     }
2979
2980     if (!sset_contains(&ofproto->ports, devname)) {
2981         return ENODEV;
2982     }
2983     error = dpif_port_query_by_name(ofproto->backer->dpif,
2984                                     devname, &dpif_port);
2985     if (!error) {
2986         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2987     }
2988     return error;
2989 }
2990
2991 static int
2992 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2993 {
2994     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2995     const char *devname = netdev_get_name(netdev);
2996     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2997     const char *dp_port_name;
2998
2999     if (netdev_vport_is_patch(netdev)) {
3000         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3001         return 0;
3002     }
3003
3004     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3005     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3006         odp_port_t port_no = ODPP_NONE;
3007         int error;
3008
3009         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3010         if (error) {
3011             return error;
3012         }
3013         if (netdev_get_tunnel_config(netdev)) {
3014             simap_put(&ofproto->backer->tnl_backers,
3015                       dp_port_name, odp_to_u32(port_no));
3016         }
3017     }
3018
3019     if (netdev_get_tunnel_config(netdev)) {
3020         sset_add(&ofproto->ghost_ports, devname);
3021     } else {
3022         sset_add(&ofproto->ports, devname);
3023     }
3024     return 0;
3025 }
3026
3027 static int
3028 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3029 {
3030     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3031     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3032     int error = 0;
3033
3034     if (!ofport) {
3035         return 0;
3036     }
3037
3038     sset_find_and_delete(&ofproto->ghost_ports,
3039                          netdev_get_name(ofport->up.netdev));
3040     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3041     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3042         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3043         if (!error) {
3044             /* The caller is going to close ofport->up.netdev.  If this is a
3045              * bonded port, then the bond is using that netdev, so remove it
3046              * from the bond.  The client will need to reconfigure everything
3047              * after deleting ports, so then the slave will get re-added. */
3048             bundle_remove(&ofport->up);
3049         }
3050     }
3051     return error;
3052 }
3053
3054 static int
3055 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3056 {
3057     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3058     int error;
3059
3060     push_all_stats();
3061
3062     error = netdev_get_stats(ofport->up.netdev, stats);
3063
3064     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3065         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3066
3067         /* ofproto->stats.tx_packets represents packets that we created
3068          * internally and sent to some port (e.g. packets sent with
3069          * send_packet()).  Account for them as if they had come from
3070          * OFPP_LOCAL and got forwarded. */
3071
3072         if (stats->rx_packets != UINT64_MAX) {
3073             stats->rx_packets += ofproto->stats.tx_packets;
3074         }
3075
3076         if (stats->rx_bytes != UINT64_MAX) {
3077             stats->rx_bytes += ofproto->stats.tx_bytes;
3078         }
3079
3080         /* ofproto->stats.rx_packets represents packets that were received on
3081          * some port and we processed internally and dropped (e.g. STP).
3082          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3083
3084         if (stats->tx_packets != UINT64_MAX) {
3085             stats->tx_packets += ofproto->stats.rx_packets;
3086         }
3087
3088         if (stats->tx_bytes != UINT64_MAX) {
3089             stats->tx_bytes += ofproto->stats.rx_bytes;
3090         }
3091     }
3092
3093     return error;
3094 }
3095
3096 struct port_dump_state {
3097     uint32_t bucket;
3098     uint32_t offset;
3099     bool ghost;
3100
3101     struct ofproto_port port;
3102     bool has_port;
3103 };
3104
3105 static int
3106 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3107 {
3108     *statep = xzalloc(sizeof(struct port_dump_state));
3109     return 0;
3110 }
3111
3112 static int
3113 port_dump_next(const struct ofproto *ofproto_, void *state_,
3114                struct ofproto_port *port)
3115 {
3116     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3117     struct port_dump_state *state = state_;
3118     const struct sset *sset;
3119     struct sset_node *node;
3120
3121     if (state->has_port) {
3122         ofproto_port_destroy(&state->port);
3123         state->has_port = false;
3124     }
3125     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3126     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3127         int error;
3128
3129         error = port_query_by_name(ofproto_, node->name, &state->port);
3130         if (!error) {
3131             *port = state->port;
3132             state->has_port = true;
3133             return 0;
3134         } else if (error != ENODEV) {
3135             return error;
3136         }
3137     }
3138
3139     if (!state->ghost) {
3140         state->ghost = true;
3141         state->bucket = 0;
3142         state->offset = 0;
3143         return port_dump_next(ofproto_, state_, port);
3144     }
3145
3146     return EOF;
3147 }
3148
3149 static int
3150 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3151 {
3152     struct port_dump_state *state = state_;
3153
3154     if (state->has_port) {
3155         ofproto_port_destroy(&state->port);
3156     }
3157     free(state);
3158     return 0;
3159 }
3160
3161 static int
3162 port_poll(const struct ofproto *ofproto_, char **devnamep)
3163 {
3164     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3165
3166     if (ofproto->port_poll_errno) {
3167         int error = ofproto->port_poll_errno;
3168         ofproto->port_poll_errno = 0;
3169         return error;
3170     }
3171
3172     if (sset_is_empty(&ofproto->port_poll_set)) {
3173         return EAGAIN;
3174     }
3175
3176     *devnamep = sset_pop(&ofproto->port_poll_set);
3177     return 0;
3178 }
3179
3180 static void
3181 port_poll_wait(const struct ofproto *ofproto_)
3182 {
3183     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3184     dpif_port_poll_wait(ofproto->backer->dpif);
3185 }
3186
3187 static int
3188 port_is_lacp_current(const struct ofport *ofport_)
3189 {
3190     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3191     return (ofport->bundle && ofport->bundle->lacp
3192             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3193             : -1);
3194 }
3195 \f
3196 /* Upcall handling. */
3197
3198 struct flow_miss_op {
3199     struct dpif_op dpif_op;
3200
3201     uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3202     struct xlate_out xout;
3203     bool xout_garbage;           /* 'xout' needs to be uninitialized? */
3204
3205     struct ofpbuf mask;          /* Flow mask for "put" ops. */
3206     struct odputil_keybuf maskbuf;
3207
3208     /* If this is a "put" op, then a pointer to the subfacet that should
3209      * be marked as uninstalled if the operation fails. */
3210     struct subfacet *subfacet;
3211 };
3212
3213 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3214  * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3215  * and (usually) installing a datapath flow.  The answer is usually "yes" (a
3216  * return value of true).  However, for short flows the cost of bookkeeping is
3217  * much higher than the benefits, so when the datapath holds a large number of
3218  * flows we impose some heuristics to decide which flows are likely to be worth
3219  * tracking. */
3220 static bool
3221 flow_miss_should_make_facet(struct flow_miss *miss)
3222 {
3223     struct dpif_backer *backer = miss->ofproto->backer;
3224     uint32_t hash;
3225
3226     switch (flow_miss_model) {
3227     case OFPROTO_HANDLE_MISS_AUTO:
3228         break;
3229     case OFPROTO_HANDLE_MISS_WITH_FACETS:
3230         return true;
3231     case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3232         return false;
3233     }
3234
3235     if (!backer->governor) {
3236         size_t n_subfacets;
3237
3238         n_subfacets = hmap_count(&backer->subfacets);
3239         if (n_subfacets * 2 <= flow_eviction_threshold) {
3240             return true;
3241         }
3242
3243         backer->governor = governor_create();
3244     }
3245
3246     hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3247     return governor_should_install_flow(backer->governor, hash,
3248                                         list_size(&miss->packets));
3249 }
3250
3251 /* Handles 'miss', which matches 'facet'.  May add any required datapath
3252  * operations to 'ops', incrementing '*n_ops' for each new op.
3253  *
3254  * All of the packets in 'miss' are considered to have arrived at time
3255  * 'miss->stats.used'.  This is really important only for new facets: if we
3256  * just called time_msec() here, then the new subfacet or its packets could
3257  * look (occasionally) as though it was used some time after the facet was
3258  * used.  That can make a one-packet flow look like it has a nonzero duration,
3259  * which looks odd in e.g. NetFlow statistics. */
3260 static void
3261 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3262                             struct flow_miss_op *ops, size_t *n_ops)
3263 {
3264     enum subfacet_path want_path;
3265     struct subfacet *subfacet;
3266
3267     facet->packet_count += miss->stats.n_packets;
3268     facet->prev_packet_count += miss->stats.n_packets;
3269     facet->byte_count += miss->stats.n_bytes;
3270     facet->prev_byte_count += miss->stats.n_bytes;
3271
3272     want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3273
3274     /* Don't install the flow if it's the result of the "userspace"
3275      * action for an already installed facet.  This can occur when a
3276      * datapath flow with wildcards has a "userspace" action and flows
3277      * sent to userspace result in a different subfacet, which will then
3278      * be rejected as overlapping by the datapath. */
3279     if (miss->upcall_type == DPIF_UC_ACTION
3280         && !list_is_empty(&facet->subfacets)) {
3281         return;
3282     }
3283
3284     subfacet = subfacet_create(facet, miss);
3285     if (subfacet->path != want_path) {
3286         struct flow_miss_op *op = &ops[(*n_ops)++];
3287         struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3288
3289         subfacet->path = want_path;
3290
3291         ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3292         if (enable_megaflows) {
3293             odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3294                                    &miss->flow, UINT32_MAX);
3295         }
3296
3297         op->xout_garbage = false;
3298         op->dpif_op.type = DPIF_OP_FLOW_PUT;
3299         op->subfacet = subfacet;
3300         put->flags = DPIF_FP_CREATE;
3301         put->key = miss->key;
3302         put->key_len = miss->key_len;
3303         put->mask = op->mask.data;
3304         put->mask_len = op->mask.size;
3305
3306         if (want_path == SF_FAST_PATH) {
3307             put->actions = facet->xout.odp_actions.data;
3308             put->actions_len = facet->xout.odp_actions.size;
3309         } else {
3310             compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3311                               op->slow_stub, sizeof op->slow_stub,
3312                               &put->actions, &put->actions_len);
3313         }
3314         put->stats = NULL;
3315     }
3316 }
3317
3318 /* Handles flow miss 'miss'.  May add any required datapath operations
3319  * to 'ops', incrementing '*n_ops' for each new op. */
3320 static void
3321 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3322                  size_t *n_ops)
3323 {
3324     struct facet *facet;
3325
3326     miss->ofproto->n_missed += list_size(&miss->packets);
3327
3328     facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3329     if (!facet) {
3330         /* There does not exist a bijection between 'struct flow' and datapath
3331          * flow keys with fitness ODP_FIT_TO_LITTLE.  This breaks a fundamental
3332          * assumption used throughout the facet and subfacet handling code.
3333          * Since we have to handle these misses in userspace anyway, we simply
3334          * skip facet creation, avoiding the problem altogether. */
3335         if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3336             || !flow_miss_should_make_facet(miss)) {
3337             return;
3338         }
3339
3340         facet = facet_create(miss);
3341     }
3342     handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3343 }
3344
3345 static struct drop_key *
3346 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3347                 size_t key_len)
3348 {
3349     struct drop_key *drop_key;
3350
3351     HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3352                              &backer->drop_keys) {
3353         if (drop_key->key_len == key_len
3354             && !memcmp(drop_key->key, key, key_len)) {
3355             return drop_key;
3356         }
3357     }
3358     return NULL;
3359 }
3360
3361 static void
3362 drop_key_clear(struct dpif_backer *backer)
3363 {
3364     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3365     struct drop_key *drop_key, *next;
3366
3367     HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3368         int error;
3369
3370         error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3371                               NULL);
3372         if (error && !VLOG_DROP_WARN(&rl)) {
3373             struct ds ds = DS_EMPTY_INITIALIZER;
3374             odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3375             VLOG_WARN("Failed to delete drop key (%s) (%s)",
3376                       ovs_strerror(error), ds_cstr(&ds));
3377             ds_destroy(&ds);
3378         }
3379
3380         hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3381         drop_key_destroy(drop_key);
3382     }
3383
3384     udpif_drop_key_clear(backer->udpif);
3385 }
3386
3387 static void
3388 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3389 {
3390     struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3391     struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3392     struct flow_miss *miss;
3393     size_t n_ops, i;
3394
3395     /* Process each element in the to-do list, constructing the set of
3396      * operations to batch. */
3397     n_ops = 0;
3398     HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3399         handle_flow_miss(miss, flow_miss_ops, &n_ops);
3400     }
3401     ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3402
3403     /* Execute batch. */
3404     for (i = 0; i < n_ops; i++) {
3405         dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3406     }
3407     dpif_operate(backer->dpif, dpif_ops, n_ops);
3408
3409     for (i = 0; i < n_ops; i++) {
3410         if (dpif_ops[i]->error != 0
3411             && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3412             && flow_miss_ops[i].subfacet) {
3413             struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3414
3415             COVERAGE_INC(subfacet_install_fail);
3416
3417             /* Zero-out subfacet counters when installation failed, but
3418              * datapath reported hits.  This should not happen and
3419              * indicates a bug, since if the datapath flow exists, we
3420              * should not be attempting to create a new subfacet.  A
3421              * buggy datapath could trigger this, so just zero out the
3422              * counters and log an error. */
3423             if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3424                 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3425                             "datapath reported hits");
3426                 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3427             }
3428
3429             subfacet->path = SF_NOT_INSTALLED;
3430         }
3431     }
3432 }
3433
3434 static void
3435 handle_sflow_upcall(struct dpif_backer *backer,
3436                     const struct dpif_upcall *upcall)
3437 {
3438     struct ofproto_dpif *ofproto;
3439     union user_action_cookie cookie;
3440     struct flow flow;
3441     odp_port_t odp_in_port;
3442
3443     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3444                       &flow, NULL, &ofproto, &odp_in_port)
3445         || !ofproto->sflow) {
3446         return;
3447     }
3448
3449     memset(&cookie, 0, sizeof cookie);
3450     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3451     dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3452                         odp_in_port, &cookie);
3453 }
3454
3455 static void
3456 handle_flow_sample_upcall(struct dpif_backer *backer,
3457                           const struct dpif_upcall *upcall)
3458 {
3459     struct ofproto_dpif *ofproto;
3460     union user_action_cookie cookie;
3461     struct flow flow;
3462
3463     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3464                       &flow, NULL, &ofproto, NULL)
3465         || !ofproto->ipfix) {
3466         return;
3467     }
3468
3469     memset(&cookie, 0, sizeof cookie);
3470     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3471
3472     /* The flow reflects exactly the contents of the packet.  Sample
3473      * the packet using it. */
3474     dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3475                            cookie.flow_sample.collector_set_id,
3476                            cookie.flow_sample.probability,
3477                            cookie.flow_sample.obs_domain_id,
3478                            cookie.flow_sample.obs_point_id);
3479 }
3480
3481 static void
3482 handle_ipfix_upcall(struct dpif_backer *backer,
3483                     const struct dpif_upcall *upcall)
3484 {
3485     struct ofproto_dpif *ofproto;
3486     struct flow flow;
3487
3488     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3489                       &flow, NULL, &ofproto, NULL)
3490         || !ofproto->ipfix) {
3491         return;
3492     }
3493
3494     /* The flow reflects exactly the contents of the packet.  Sample
3495      * the packet using it. */
3496     dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3497 }
3498
3499 static void
3500 handle_upcalls(struct dpif_backer *backer)
3501 {
3502     struct flow_miss_batch *fmb;
3503     int n_processed;
3504
3505     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3506         struct upcall *upcall = upcall_next(backer->udpif);
3507
3508         if (!upcall) {
3509             break;
3510         }
3511
3512         switch (upcall->type) {
3513         case SFLOW_UPCALL:
3514             handle_sflow_upcall(backer, &upcall->dpif_upcall);
3515             break;
3516
3517         case FLOW_SAMPLE_UPCALL:
3518             handle_flow_sample_upcall(backer, &upcall->dpif_upcall);
3519             break;
3520
3521         case IPFIX_UPCALL:
3522             handle_ipfix_upcall(backer, &upcall->dpif_upcall);
3523             break;
3524
3525         case BAD_UPCALL:
3526             break;
3527
3528         case MISS_UPCALL:
3529             NOT_REACHED();
3530         }
3531
3532         upcall_destroy(upcall);
3533     }
3534
3535     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3536         struct drop_key *drop_key = drop_key_next(backer->udpif);
3537         if (!drop_key) {
3538             break;
3539         }
3540
3541         if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3542             hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3543                         hash_bytes(drop_key->key, drop_key->key_len, 0));
3544             dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3545                           drop_key->key, drop_key->key_len,
3546                           NULL, 0, NULL, 0, NULL);
3547         } else {
3548             drop_key_destroy(drop_key);
3549         }
3550     }
3551
3552     fmb = flow_miss_batch_next(backer->udpif);
3553     if (fmb) {
3554         handle_flow_misses(backer, fmb);
3555         flow_miss_batch_destroy(fmb);
3556     }
3557 }
3558 \f
3559 /* Flow expiration. */
3560
3561 static int subfacet_max_idle(const struct dpif_backer *);
3562 static void update_stats(struct dpif_backer *);
3563 static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex);
3564 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3565
3566 /* This function is called periodically by run().  Its job is to collect
3567  * updates for the flows that have been installed into the datapath, most
3568  * importantly when they last were used, and then use that information to
3569  * expire flows that have not been used recently.
3570  *
3571  * Returns the number of milliseconds after which it should be called again. */
3572 static int
3573 expire(struct dpif_backer *backer)
3574 {
3575     struct ofproto_dpif *ofproto;
3576     size_t n_subfacets;
3577     int max_idle;
3578
3579     /* Periodically clear out the drop keys in an effort to keep them
3580      * relatively few. */
3581     drop_key_clear(backer);
3582
3583     /* Update stats for each flow in the backer. */
3584     update_stats(backer);
3585
3586     n_subfacets = hmap_count(&backer->subfacets);
3587     if (n_subfacets) {
3588         struct subfacet *subfacet;
3589         long long int total, now;
3590
3591         total = 0;
3592         now = time_msec();
3593         HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3594             total += now - subfacet->created;
3595         }
3596         backer->avg_subfacet_life += total / n_subfacets;
3597     }
3598     backer->avg_subfacet_life /= 2;
3599
3600     backer->avg_n_subfacet += n_subfacets;
3601     backer->avg_n_subfacet /= 2;
3602
3603     backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3604
3605     max_idle = subfacet_max_idle(backer);
3606     expire_subfacets(backer, max_idle);
3607
3608     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3609         struct rule *rule, *next_rule;
3610
3611         if (ofproto->backer != backer) {
3612             continue;
3613         }
3614
3615         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3616          * has passed. */
3617         ovs_mutex_lock(&ofproto_mutex);
3618         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3619                             &ofproto->up.expirable) {
3620             rule_expire(rule_dpif_cast(rule));
3621         }
3622         ovs_mutex_unlock(&ofproto_mutex);
3623
3624         /* All outstanding data in existing flows has been accounted, so it's a
3625          * good time to do bond rebalancing. */
3626         if (ofproto->has_bonded_bundles) {
3627             struct ofbundle *bundle;
3628
3629             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3630                 if (bundle->bond) {
3631                     bond_rebalance(bundle->bond);
3632                 }
3633             }
3634         }
3635     }
3636
3637     return MIN(max_idle, 1000);
3638 }
3639
3640 /* Updates flow table statistics given that the datapath just reported 'stats'
3641  * as 'subfacet''s statistics. */
3642 static void
3643 update_subfacet_stats(struct subfacet *subfacet,
3644                       const struct dpif_flow_stats *stats)
3645 {
3646     struct facet *facet = subfacet->facet;
3647     struct dpif_flow_stats diff;
3648
3649     diff.tcp_flags = stats->tcp_flags;
3650     diff.used = stats->used;
3651
3652     if (stats->n_packets >= subfacet->dp_packet_count) {
3653         diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3654     } else {
3655         VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3656         diff.n_packets = 0;
3657     }
3658
3659     if (stats->n_bytes >= subfacet->dp_byte_count) {
3660         diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3661     } else {
3662         VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3663         diff.n_bytes = 0;
3664     }
3665
3666     facet->ofproto->n_hit += diff.n_packets;
3667     subfacet->dp_packet_count = stats->n_packets;
3668     subfacet->dp_byte_count = stats->n_bytes;
3669     subfacet_update_stats(subfacet, &diff);
3670
3671     if (facet->accounted_bytes < facet->byte_count) {
3672         facet_learn(facet);
3673         facet_account(facet);
3674         facet->accounted_bytes = facet->byte_count;
3675     }
3676 }
3677
3678 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3679  * about, or a flow that shouldn't be installed but was anyway.  Delete it. */
3680 static void
3681 delete_unexpected_flow(struct dpif_backer *backer,
3682                        const struct nlattr *key, size_t key_len)
3683 {
3684     if (!VLOG_DROP_WARN(&rl)) {
3685         struct ds s;
3686
3687         ds_init(&s);
3688         odp_flow_key_format(key, key_len, &s);
3689         VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3690         ds_destroy(&s);
3691     }
3692
3693     COVERAGE_INC(facet_unexpected);
3694     dpif_flow_del(backer->dpif, key, key_len, NULL);
3695 }
3696
3697 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3698  *
3699  * This function also pushes statistics updates to rules which each facet
3700  * resubmits into.  Generally these statistics will be accurate.  However, if a
3701  * facet changes the rule it resubmits into at some time in between
3702  * update_stats() runs, it is possible that statistics accrued to the
3703  * old rule will be incorrectly attributed to the new rule.  This could be
3704  * avoided by calling update_stats() whenever rules are created or
3705  * deleted.  However, the performance impact of making so many calls to the
3706  * datapath do not justify the benefit of having perfectly accurate statistics.
3707  *
3708  * In addition, this function maintains per ofproto flow hit counts. The patch
3709  * port is not treated specially. e.g. A packet ingress from br0 patched into
3710  * br1 will increase the hit count of br0 by 1, however, does not affect
3711  * the hit or miss counts of br1.
3712  */
3713 static void
3714 update_stats(struct dpif_backer *backer)
3715 {
3716     const struct dpif_flow_stats *stats;
3717     struct dpif_flow_dump dump;
3718     const struct nlattr *key, *mask;
3719     size_t key_len, mask_len;
3720
3721     dpif_flow_dump_start(&dump, backer->dpif);
3722     while (dpif_flow_dump_next(&dump, &key, &key_len,
3723                                &mask, &mask_len, NULL, NULL, &stats)) {
3724         struct subfacet *subfacet;
3725         uint32_t key_hash;
3726
3727         key_hash = odp_flow_key_hash(key, key_len);
3728         subfacet = subfacet_find(backer, key, key_len, key_hash);
3729         switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3730         case SF_FAST_PATH:
3731             update_subfacet_stats(subfacet, stats);
3732             break;
3733
3734         case SF_SLOW_PATH:
3735             /* Stats are updated per-packet. */
3736             break;
3737
3738         case SF_NOT_INSTALLED:
3739         default:
3740             delete_unexpected_flow(backer, key, key_len);
3741             break;
3742         }
3743         run_fast_rl();
3744     }
3745     dpif_flow_dump_done(&dump);
3746
3747     update_moving_averages(backer);
3748 }
3749
3750 /* Calculates and returns the number of milliseconds of idle time after which
3751  * subfacets should expire from the datapath.  When a subfacet expires, we fold
3752  * its statistics into its facet, and when a facet's last subfacet expires, we
3753  * fold its statistic into its rule. */
3754 static int
3755 subfacet_max_idle(const struct dpif_backer *backer)
3756 {
3757     /*
3758      * Idle time histogram.
3759      *
3760      * Most of the time a switch has a relatively small number of subfacets.
3761      * When this is the case we might as well keep statistics for all of them
3762      * in userspace and to cache them in the kernel datapath for performance as
3763      * well.
3764      *
3765      * As the number of subfacets increases, the memory required to maintain
3766      * statistics about them in userspace and in the kernel becomes
3767      * significant.  However, with a large number of subfacets it is likely
3768      * that only a few of them are "heavy hitters" that consume a large amount
3769      * of bandwidth.  At this point, only heavy hitters are worth caching in
3770      * the kernel and maintaining in userspaces; other subfacets we can
3771      * discard.
3772      *
3773      * The technique used to compute the idle time is to build a histogram with
3774      * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each.  Each subfacet
3775      * that is installed in the kernel gets dropped in the appropriate bucket.
3776      * After the histogram has been built, we compute the cutoff so that only
3777      * the most-recently-used 1% of subfacets (but at least
3778      * flow_eviction_threshold flows) are kept cached.  At least
3779      * the most-recently-used bucket of subfacets is kept, so actually an
3780      * arbitrary number of subfacets can be kept in any given expiration run
3781      * (though the next run will delete most of those unless they receive
3782      * additional data).
3783      *
3784      * This requires a second pass through the subfacets, in addition to the
3785      * pass made by update_stats(), because the former function never looks at
3786      * uninstallable subfacets.
3787      */
3788     enum { BUCKET_WIDTH = 100 };
3789     enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3790     int buckets[N_BUCKETS] = { 0 };
3791     int total, subtotal, bucket;
3792     struct subfacet *subfacet;
3793     long long int now;
3794     int i;
3795
3796     total = hmap_count(&backer->subfacets);
3797     if (total <= flow_eviction_threshold) {
3798         return N_BUCKETS * BUCKET_WIDTH;
3799     }
3800
3801     /* Build histogram. */
3802     now = time_msec();
3803     HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3804         long long int idle = now - subfacet->used;
3805         int bucket = (idle <= 0 ? 0
3806                       : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3807                       : (unsigned int) idle / BUCKET_WIDTH);
3808         buckets[bucket]++;
3809     }
3810
3811     /* Find the first bucket whose flows should be expired. */
3812     subtotal = bucket = 0;
3813     do {
3814         subtotal += buckets[bucket++];
3815     } while (bucket < N_BUCKETS &&
3816              subtotal < MAX(flow_eviction_threshold, total / 100));
3817
3818     if (VLOG_IS_DBG_ENABLED()) {
3819         struct ds s;
3820
3821         ds_init(&s);
3822         ds_put_cstr(&s, "keep");
3823         for (i = 0; i < N_BUCKETS; i++) {
3824             if (i == bucket) {
3825                 ds_put_cstr(&s, ", drop");
3826             }
3827             if (buckets[i]) {
3828                 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3829             }
3830         }
3831         VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3832         ds_destroy(&s);
3833     }
3834
3835     return bucket * BUCKET_WIDTH;
3836 }
3837
3838 static void
3839 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3840 {
3841     /* Cutoff time for most flows. */
3842     long long int normal_cutoff = time_msec() - dp_max_idle;
3843
3844     /* We really want to keep flows for special protocols around, so use a more
3845      * conservative cutoff. */
3846     long long int special_cutoff = time_msec() - 10000;
3847
3848     struct subfacet *subfacet, *next_subfacet;
3849     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3850     int n_batch;
3851
3852     n_batch = 0;
3853     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3854                         &backer->subfacets) {
3855         long long int cutoff;
3856
3857         cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3858                                                 | SLOW_STP)
3859                   ? special_cutoff
3860                   : normal_cutoff);
3861         if (subfacet->used < cutoff) {
3862             if (subfacet->path != SF_NOT_INSTALLED) {
3863                 batch[n_batch++] = subfacet;
3864                 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3865                     subfacet_destroy_batch(backer, batch, n_batch);
3866                     n_batch = 0;
3867                 }
3868             } else {
3869                 subfacet_destroy(subfacet);
3870             }
3871         }
3872     }
3873
3874     if (n_batch > 0) {
3875         subfacet_destroy_batch(backer, batch, n_batch);
3876     }
3877 }
3878
3879 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3880  * then delete it entirely. */
3881 static void
3882 rule_expire(struct rule_dpif *rule)
3883     OVS_REQUIRES(ofproto_mutex)
3884 {
3885     uint16_t idle_timeout, hard_timeout;
3886     long long int now = time_msec();
3887     int reason;
3888
3889     ovs_assert(!rule->up.pending);
3890
3891     /* Has 'rule' expired? */
3892     ovs_mutex_lock(&rule->up.mutex);
3893     hard_timeout = rule->up.hard_timeout;
3894     idle_timeout = rule->up.idle_timeout;
3895     if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3896         reason = OFPRR_HARD_TIMEOUT;
3897     } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3898         reason = OFPRR_IDLE_TIMEOUT;
3899     } else {
3900         reason = -1;
3901     }
3902     ovs_mutex_unlock(&rule->up.mutex);
3903
3904     if (reason >= 0) {
3905         COVERAGE_INC(ofproto_dpif_expired);
3906         ofproto_rule_expire(&rule->up, reason);
3907     }
3908 }
3909 \f
3910 /* Facets. */
3911
3912 /* Creates and returns a new facet based on 'miss'.
3913  *
3914  * The caller must already have determined that no facet with an identical
3915  * 'miss->flow' exists in 'miss->ofproto'.
3916  *
3917  * 'rule' and 'xout' must have been created based on 'miss'.
3918  *
3919  * 'facet'' statistics are initialized based on 'stats'.
3920  *
3921  * The facet will initially have no subfacets.  The caller should create (at
3922  * least) one subfacet with subfacet_create(). */
3923 static struct facet *
3924 facet_create(const struct flow_miss *miss)
3925 {
3926     struct ofproto_dpif *ofproto = miss->ofproto;
3927     struct facet *facet;
3928     struct match match;
3929
3930     facet = xzalloc(sizeof *facet);
3931     facet->ofproto = miss->ofproto;
3932     facet->used = miss->stats.used;
3933     facet->flow = miss->flow;
3934     facet->learn_rl = time_msec() + 500;
3935
3936     list_init(&facet->subfacets);
3937     netflow_flow_init(&facet->nf_flow);
3938     netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3939
3940     xlate_out_copy(&facet->xout, &miss->xout);
3941
3942     match_init(&match, &facet->flow, &facet->xout.wc);
3943     cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3944     ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3945     classifier_insert(&ofproto->facets, &facet->cr);
3946     ovs_rwlock_unlock(&ofproto->facets.rwlock);
3947
3948     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
3949     return facet;
3950 }
3951
3952 static void
3953 facet_free(struct facet *facet)
3954 {
3955     if (facet) {
3956         xlate_out_uninit(&facet->xout);
3957         free(facet);
3958     }
3959 }
3960
3961 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3962  * 'packet', which arrived on 'in_port'. */
3963 static bool
3964 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3965                     const struct nlattr *odp_actions, size_t actions_len,
3966                     struct ofpbuf *packet)
3967 {
3968     struct odputil_keybuf keybuf;
3969     struct ofpbuf key;
3970     int error;
3971
3972     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3973     odp_flow_key_from_flow(&key, flow,
3974                            ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
3975
3976     error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
3977                          odp_actions, actions_len, packet);
3978     return !error;
3979 }
3980
3981 /* Remove 'facet' from its ofproto and free up the associated memory:
3982  *
3983  *   - If 'facet' was installed in the datapath, uninstalls it and updates its
3984  *     rule's statistics, via subfacet_uninstall().
3985  *
3986  *   - Removes 'facet' from its rule and from ofproto->facets.
3987  */
3988 static void
3989 facet_remove(struct facet *facet)
3990 {
3991     struct subfacet *subfacet, *next_subfacet;
3992
3993     ovs_assert(!list_is_empty(&facet->subfacets));
3994
3995     /* First uninstall all of the subfacets to get final statistics. */
3996     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3997         subfacet_uninstall(subfacet);
3998     }
3999
4000     /* Flush the final stats to the rule.
4001      *
4002      * This might require us to have at least one subfacet around so that we
4003      * can use its actions for accounting in facet_account(), which is why we
4004      * have uninstalled but not yet destroyed the subfacets. */
4005     facet_flush_stats(facet);
4006
4007     /* Now we're really all done so destroy everything. */
4008     LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4009                         &facet->subfacets) {
4010         subfacet_destroy__(subfacet);
4011     }
4012     ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4013     classifier_remove(&facet->ofproto->facets, &facet->cr);
4014     ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4015     cls_rule_destroy(&facet->cr);
4016     facet_free(facet);
4017 }
4018
4019 /* Feed information from 'facet' back into the learning table to keep it in
4020  * sync with what is actually flowing through the datapath. */
4021 static void
4022 facet_learn(struct facet *facet)
4023 {
4024     long long int now = time_msec();
4025
4026     if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4027         return;
4028     }
4029
4030     facet->learn_rl = now + 500;
4031
4032     if (!facet->xout.has_learn
4033         && !facet->xout.has_normal
4034         && (!facet->xout.has_fin_timeout
4035             || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4036         return;
4037     }
4038
4039     facet_push_stats(facet, true);
4040 }
4041
4042 static void
4043 facet_account(struct facet *facet)
4044 {
4045     const struct nlattr *a;
4046     unsigned int left;
4047     ovs_be16 vlan_tci;
4048     uint64_t n_bytes;
4049
4050     if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4051         return;
4052     }
4053     n_bytes = facet->byte_count - facet->accounted_bytes;
4054
4055     /* This loop feeds byte counters to bond_account() for rebalancing to use
4056      * as a basis.  We also need to track the actual VLAN on which the packet
4057      * is going to be sent to ensure that it matches the one passed to
4058      * bond_choose_output_slave().  (Otherwise, we will account to the wrong
4059      * hash bucket.)
4060      *
4061      * We use the actions from an arbitrary subfacet because they should all
4062      * be equally valid for our purpose. */
4063     vlan_tci = facet->flow.vlan_tci;
4064     NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4065                              facet->xout.odp_actions.size) {
4066         const struct ovs_action_push_vlan *vlan;
4067         struct ofport_dpif *port;
4068
4069         switch (nl_attr_type(a)) {
4070         case OVS_ACTION_ATTR_OUTPUT:
4071             port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4072             if (port && port->bundle && port->bundle->bond) {
4073                 bond_account(port->bundle->bond, &facet->flow,
4074                              vlan_tci_to_vid(vlan_tci), n_bytes);
4075             }
4076             break;
4077
4078         case OVS_ACTION_ATTR_POP_VLAN:
4079             vlan_tci = htons(0);
4080             break;
4081
4082         case OVS_ACTION_ATTR_PUSH_VLAN:
4083             vlan = nl_attr_get(a);
4084             vlan_tci = vlan->vlan_tci;
4085             break;
4086         }
4087     }
4088 }
4089
4090 /* Returns true if the only action for 'facet' is to send to the controller.
4091  * (We don't report NetFlow expiration messages for such facets because they
4092  * are just part of the control logic for the network, not real traffic). */
4093 static bool
4094 facet_is_controller_flow(struct facet *facet)
4095 {
4096     if (facet) {
4097         struct ofproto_dpif *ofproto = facet->ofproto;
4098         const struct ofpact *ofpacts;
4099         struct rule_actions *actions;
4100         struct rule_dpif *rule;
4101         size_t ofpacts_len;
4102         bool is_controller;
4103
4104         rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4105         actions = rule_dpif_get_actions(rule);
4106         rule_dpif_unref(rule);
4107
4108         ofpacts_len = actions->ofpacts_len;
4109         ofpacts = actions->ofpacts;
4110         is_controller = ofpacts_len > 0
4111             && ofpacts->type == OFPACT_CONTROLLER
4112             && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4113         rule_actions_unref(actions);
4114
4115         return is_controller;
4116     }
4117     return false;
4118 }
4119
4120 /* Folds all of 'facet''s statistics into its rule.  Also updates the
4121  * accounting ofhook and emits a NetFlow expiration if appropriate.  All of
4122  * 'facet''s statistics in the datapath should have been zeroed and folded into
4123  * its packet and byte counts before this function is called. */
4124 static void
4125 facet_flush_stats(struct facet *facet)
4126 {
4127     struct ofproto_dpif *ofproto = facet->ofproto;
4128     struct subfacet *subfacet;
4129
4130     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4131         ovs_assert(!subfacet->dp_byte_count);
4132         ovs_assert(!subfacet->dp_packet_count);
4133     }
4134
4135     facet_push_stats(facet, false);
4136     if (facet->accounted_bytes < facet->byte_count) {
4137         facet_account(facet);
4138         facet->accounted_bytes = facet->byte_count;
4139     }
4140
4141     if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4142         struct ofexpired expired;
4143         expired.flow = facet->flow;
4144         expired.packet_count = facet->packet_count;
4145         expired.byte_count = facet->byte_count;
4146         expired.used = facet->used;
4147         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4148     }
4149
4150     /* Reset counters to prevent double counting if 'facet' ever gets
4151      * reinstalled. */
4152     facet_reset_counters(facet);
4153
4154     netflow_flow_clear(&facet->nf_flow);
4155     facet->tcp_flags = 0;
4156 }
4157
4158 /* Searches 'ofproto''s table of facets for one which would be responsible for
4159  * 'flow'.  Returns it if found, otherwise a null pointer.
4160  *
4161  * The returned facet might need revalidation; use facet_lookup_valid()
4162  * instead if that is important. */
4163 static struct facet *
4164 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4165 {
4166     struct cls_rule *cr;
4167
4168     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4169     cr = classifier_lookup(&ofproto->facets, flow, NULL);
4170     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4171     return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4172 }
4173
4174 /* Searches 'ofproto''s table of facets for one capable that covers
4175  * 'flow'.  Returns it if found, otherwise a null pointer.
4176  *
4177  * The returned facet is guaranteed to be valid. */
4178 static struct facet *
4179 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4180 {
4181     struct facet *facet;
4182
4183     facet = facet_find(ofproto, flow);
4184     if (facet
4185         && ofproto->backer->need_revalidate
4186         && !facet_revalidate(facet)) {
4187         return NULL;
4188     }
4189
4190     return facet;
4191 }
4192
4193 static bool
4194 facet_check_consistency(struct facet *facet)
4195 {
4196     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4197
4198     struct xlate_out xout;
4199     struct xlate_in xin;
4200
4201     struct rule_dpif *rule;
4202     bool ok;
4203
4204     /* Check the datapath actions for consistency. */
4205     rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4206     xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4207     xlate_actions(&xin, &xout);
4208     rule_dpif_unref(rule);
4209
4210     ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4211         && facet->xout.slow == xout.slow;
4212     if (!ok && !VLOG_DROP_WARN(&rl)) {
4213         struct ds s = DS_EMPTY_INITIALIZER;
4214
4215         flow_format(&s, &facet->flow);
4216         ds_put_cstr(&s, ": inconsistency in facet");
4217
4218         if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4219             ds_put_cstr(&s, " (actions were: ");
4220             format_odp_actions(&s, facet->xout.odp_actions.data,
4221                                facet->xout.odp_actions.size);
4222             ds_put_cstr(&s, ") (correct actions: ");
4223             format_odp_actions(&s, xout.odp_actions.data,
4224                                xout.odp_actions.size);
4225             ds_put_char(&s, ')');
4226         }
4227
4228         if (facet->xout.slow != xout.slow) {
4229             ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4230         }
4231
4232         ds_destroy(&s);
4233     }
4234     xlate_out_uninit(&xout);
4235
4236     return ok;
4237 }
4238
4239 /* Re-searches the classifier for 'facet':
4240  *
4241  *   - If the rule found is different from 'facet''s current rule, moves
4242  *     'facet' to the new rule and recompiles its actions.
4243  *
4244  *   - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4245  *     where it is and recompiles its actions anyway.
4246  *
4247  *   - If any of 'facet''s subfacets correspond to a new flow according to
4248  *     xlate_receive(), 'facet' is removed.
4249  *
4250  *   Returns true if 'facet' is still valid.  False if 'facet' was removed. */
4251 static bool
4252 facet_revalidate(struct facet *facet)
4253 {
4254     struct ofproto_dpif *ofproto = facet->ofproto;
4255     struct rule_dpif *new_rule;
4256     struct subfacet *subfacet;
4257     struct flow_wildcards wc;
4258     struct xlate_out xout;
4259     struct xlate_in xin;
4260
4261     COVERAGE_INC(facet_revalidate);
4262
4263     /* Check that child subfacets still correspond to this facet.  Tunnel
4264      * configuration changes could cause a subfacet's OpenFlow in_port to
4265      * change. */
4266     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4267         struct ofproto_dpif *recv_ofproto;
4268         struct flow recv_flow;
4269         int error;
4270
4271         error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4272                               subfacet->key_len, &recv_flow, NULL,
4273                               &recv_ofproto, NULL);
4274         if (error
4275             || recv_ofproto != ofproto
4276             || facet != facet_find(ofproto, &recv_flow)) {
4277             facet_remove(facet);
4278             return false;
4279         }
4280     }
4281
4282     flow_wildcards_init_catchall(&wc);
4283     rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4284
4285     /* Calculate new datapath actions.
4286      *
4287      * We do not modify any 'facet' state yet, because we might need to, e.g.,
4288      * emit a NetFlow expiration and, if so, we need to have the old state
4289      * around to properly compose it. */
4290     xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4291     xlate_actions(&xin, &xout);
4292     flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4293
4294     /* A facet's slow path reason should only change under dramatic
4295      * circumstances.  Rather than try to update everything, it's simpler to
4296      * remove the facet and start over.
4297      *
4298      * More importantly, if a facet's wildcards change, it will be relatively
4299      * difficult to figure out if its subfacets still belong to it, and if not
4300      * which facet they may belong to.  Again, to avoid the complexity, we
4301      * simply give up instead. */
4302     if (facet->xout.slow != xout.slow
4303         || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4304         facet_remove(facet);
4305         xlate_out_uninit(&xout);
4306         rule_dpif_unref(new_rule);
4307         return false;
4308     }
4309
4310     if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4311         LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4312             if (subfacet->path == SF_FAST_PATH) {
4313                 struct dpif_flow_stats stats;
4314
4315                 subfacet_install(subfacet, &xout.odp_actions, &stats);
4316                 subfacet_update_stats(subfacet, &stats);
4317             }
4318         }
4319
4320         facet_flush_stats(facet);
4321
4322         ofpbuf_clear(&facet->xout.odp_actions);
4323         ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4324                    xout.odp_actions.size);
4325     }
4326
4327     /* Update 'facet' now that we've taken care of all the old state. */
4328     facet->xout.slow = xout.slow;
4329     facet->xout.has_learn = xout.has_learn;
4330     facet->xout.has_normal = xout.has_normal;
4331     facet->xout.has_fin_timeout = xout.has_fin_timeout;
4332     facet->xout.nf_output_iface = xout.nf_output_iface;
4333     facet->xout.mirrors = xout.mirrors;
4334     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4335
4336     ovs_mutex_lock(&new_rule->up.mutex);
4337     facet->used = MAX(facet->used, new_rule->up.created);
4338     ovs_mutex_unlock(&new_rule->up.mutex);
4339
4340     xlate_out_uninit(&xout);
4341     rule_dpif_unref(new_rule);
4342     return true;
4343 }
4344
4345 static void
4346 facet_reset_counters(struct facet *facet)
4347 {
4348     facet->packet_count = 0;
4349     facet->byte_count = 0;
4350     facet->prev_packet_count = 0;
4351     facet->prev_byte_count = 0;
4352     facet->accounted_bytes = 0;
4353 }
4354
4355 static void
4356 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4357                 struct dpif_flow_stats *stats, bool may_learn)
4358 {
4359     struct ofport_dpif *in_port;
4360     struct rule_dpif *rule;
4361     struct xlate_in xin;
4362
4363     in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4364     if (in_port && in_port->is_tunnel) {
4365         netdev_vport_inc_rx(in_port->up.netdev, stats);
4366     }
4367
4368     rule_dpif_lookup(ofproto, flow, NULL, &rule);
4369     rule_dpif_credit_stats(rule, stats);
4370     xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4371     xin.resubmit_stats = stats;
4372     xin.may_learn = may_learn;
4373     xlate_actions_for_side_effects(&xin);
4374     rule_dpif_unref(rule);
4375 }
4376
4377 static void
4378 facet_push_stats(struct facet *facet, bool may_learn)
4379 {
4380     struct dpif_flow_stats stats;
4381
4382     ovs_assert(facet->packet_count >= facet->prev_packet_count);
4383     ovs_assert(facet->byte_count >= facet->prev_byte_count);
4384     ovs_assert(facet->used >= facet->prev_used);
4385
4386     stats.n_packets = facet->packet_count - facet->prev_packet_count;
4387     stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4388     stats.used = facet->used;
4389     stats.tcp_flags = facet->tcp_flags;
4390
4391     if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4392         facet->prev_packet_count = facet->packet_count;
4393         facet->prev_byte_count = facet->byte_count;
4394         facet->prev_used = facet->used;
4395
4396         netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4397                                  facet->used);
4398         netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4399         mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4400                             stats.n_packets, stats.n_bytes);
4401         flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4402     }
4403 }
4404
4405 static void
4406 push_all_stats__(bool run_fast)
4407 {
4408     static long long int rl = LLONG_MIN;
4409     struct ofproto_dpif *ofproto;
4410
4411     if (time_msec() < rl) {
4412         return;
4413     }
4414
4415     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4416         struct cls_cursor cursor;
4417         struct facet *facet;
4418
4419         ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4420         cls_cursor_init(&cursor, &ofproto->facets, NULL);
4421         CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4422             facet_push_stats(facet, false);
4423             if (run_fast) {
4424                 run_fast_rl();
4425             }
4426         }
4427         ovs_rwlock_unlock(&ofproto->facets.rwlock);
4428     }
4429
4430     rl = time_msec() + 100;
4431 }
4432
4433 static void
4434 push_all_stats(void)
4435 {
4436     push_all_stats__(true);
4437 }
4438
4439 void
4440 rule_dpif_credit_stats(struct rule_dpif *rule,
4441                        const struct dpif_flow_stats *stats)
4442 {
4443     ovs_mutex_lock(&rule->stats_mutex);
4444     rule->packet_count += stats->n_packets;
4445     rule->byte_count += stats->n_bytes;
4446     rule->up.used = MAX(rule->up.used, stats->used);
4447     ovs_mutex_unlock(&rule->stats_mutex);
4448 }
4449
4450 bool
4451 rule_dpif_fail_open(const struct rule_dpif *rule)
4452 {
4453     return rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4454 }
4455
4456 ovs_be64
4457 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4458     OVS_REQUIRES(rule->up.mutex)
4459 {
4460     return rule->up.flow_cookie;
4461 }
4462
4463 void
4464 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4465                      uint16_t hard_timeout)
4466 {
4467     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4468 }
4469
4470 /* Returns 'rule''s actions.  The caller owns a reference on the returned
4471  * actions and must eventually release it (with rule_actions_unref()) to avoid
4472  * a memory leak. */
4473 struct rule_actions *
4474 rule_dpif_get_actions(const struct rule_dpif *rule)
4475 {
4476     return rule_get_actions(&rule->up);
4477 }
4478 \f
4479 /* Subfacets. */
4480
4481 static struct subfacet *
4482 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4483               size_t key_len, uint32_t key_hash)
4484 {
4485     struct subfacet *subfacet;
4486
4487     HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4488                              &backer->subfacets) {
4489         if (subfacet->key_len == key_len
4490             && !memcmp(key, subfacet->key, key_len)) {
4491             return subfacet;
4492         }
4493     }
4494
4495     return NULL;
4496 }
4497
4498 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4499  * 'key_fitness', 'key', and 'key_len' members in 'miss'.  Returns the
4500  * existing subfacet if there is one, otherwise creates and returns a
4501  * new subfacet. */
4502 static struct subfacet *
4503 subfacet_create(struct facet *facet, struct flow_miss *miss)
4504 {
4505     struct dpif_backer *backer = miss->ofproto->backer;
4506     enum odp_key_fitness key_fitness = miss->key_fitness;
4507     const struct nlattr *key = miss->key;
4508     size_t key_len = miss->key_len;
4509     uint32_t key_hash;
4510     struct subfacet *subfacet;
4511
4512     key_hash = odp_flow_key_hash(key, key_len);
4513
4514     if (list_is_empty(&facet->subfacets)) {
4515         subfacet = &facet->one_subfacet;
4516     } else {
4517         subfacet = subfacet_find(backer, key, key_len, key_hash);
4518         if (subfacet) {
4519             if (subfacet->facet == facet) {
4520                 return subfacet;
4521             }
4522
4523             /* This shouldn't happen. */
4524             VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4525             subfacet_destroy(subfacet);
4526         }
4527
4528         subfacet = xmalloc(sizeof *subfacet);
4529     }
4530
4531     hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4532     list_push_back(&facet->subfacets, &subfacet->list_node);
4533     subfacet->facet = facet;
4534     subfacet->key_fitness = key_fitness;
4535     subfacet->key = xmemdup(key, key_len);
4536     subfacet->key_len = key_len;
4537     subfacet->used = miss->stats.used;
4538     subfacet->created = subfacet->used;
4539     subfacet->dp_packet_count = 0;
4540     subfacet->dp_byte_count = 0;
4541     subfacet->path = SF_NOT_INSTALLED;
4542     subfacet->backer = backer;
4543
4544     backer->subfacet_add_count++;
4545     return subfacet;
4546 }
4547
4548 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4549  * its facet within 'ofproto', and frees it. */
4550 static void
4551 subfacet_destroy__(struct subfacet *subfacet)
4552 {
4553     struct facet *facet = subfacet->facet;
4554     struct ofproto_dpif *ofproto = facet->ofproto;
4555
4556     /* Update ofproto stats before uninstall the subfacet. */
4557     ofproto->backer->subfacet_del_count++;
4558
4559     subfacet_uninstall(subfacet);
4560     hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4561     list_remove(&subfacet->list_node);
4562     free(subfacet->key);
4563     if (subfacet != &facet->one_subfacet) {
4564         free(subfacet);
4565     }
4566 }
4567
4568 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4569  * last remaining subfacet in its facet destroys the facet too. */
4570 static void
4571 subfacet_destroy(struct subfacet *subfacet)
4572 {
4573     struct facet *facet = subfacet->facet;
4574
4575     if (list_is_singleton(&facet->subfacets)) {
4576         /* facet_remove() needs at least one subfacet (it will remove it). */
4577         facet_remove(facet);
4578     } else {
4579         subfacet_destroy__(subfacet);
4580     }
4581 }
4582
4583 static void
4584 subfacet_destroy_batch(struct dpif_backer *backer,
4585                        struct subfacet **subfacets, int n)
4586 {
4587     struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4588     struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4589     struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4590     int i;
4591
4592     for (i = 0; i < n; i++) {
4593         ops[i].type = DPIF_OP_FLOW_DEL;
4594         ops[i].u.flow_del.key = subfacets[i]->key;
4595         ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4596         ops[i].u.flow_del.stats = &stats[i];
4597         opsp[i] = &ops[i];
4598     }
4599
4600     dpif_operate(backer->dpif, opsp, n);
4601     for (i = 0; i < n; i++) {
4602         subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4603         subfacets[i]->path = SF_NOT_INSTALLED;
4604         subfacet_destroy(subfacets[i]);
4605         run_fast_rl();
4606     }
4607 }
4608
4609 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4610  * bytes of actions in 'actions'.  If 'stats' is non-null, statistics counters
4611  * in the datapath will be zeroed and 'stats' will be updated with traffic new
4612  * since 'subfacet' was last updated.
4613  *
4614  * Returns 0 if successful, otherwise a positive errno value. */
4615 static int
4616 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4617                  struct dpif_flow_stats *stats)
4618 {
4619     struct facet *facet = subfacet->facet;
4620     enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4621     const struct nlattr *actions = odp_actions->data;
4622     size_t actions_len = odp_actions->size;
4623     struct odputil_keybuf maskbuf;
4624     struct ofpbuf mask;
4625
4626     uint64_t slow_path_stub[128 / 8];
4627     enum dpif_flow_put_flags flags;
4628     int ret;
4629
4630     flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4631                                                : DPIF_FP_MODIFY;
4632     if (stats) {
4633         flags |= DPIF_FP_ZERO_STATS;
4634     }
4635
4636     if (path == SF_SLOW_PATH) {
4637         compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4638                           slow_path_stub, sizeof slow_path_stub,
4639                           &actions, &actions_len);
4640     }
4641
4642     ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4643     if (enable_megaflows) {
4644         odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4645                                &facet->flow, UINT32_MAX);
4646     }
4647
4648     ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4649                         subfacet->key_len,  mask.data, mask.size,
4650                         actions, actions_len, stats);
4651
4652     if (stats) {
4653         subfacet_reset_dp_stats(subfacet, stats);
4654     }
4655
4656     if (ret) {
4657         COVERAGE_INC(subfacet_install_fail);
4658     } else {
4659         subfacet->path = path;
4660     }
4661     return ret;
4662 }
4663
4664 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4665 static void
4666 subfacet_uninstall(struct subfacet *subfacet)
4667 {
4668     if (subfacet->path != SF_NOT_INSTALLED) {
4669         struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4670         struct dpif_flow_stats stats;
4671         int error;
4672
4673         error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4674                               subfacet->key_len, &stats);
4675         subfacet_reset_dp_stats(subfacet, &stats);
4676         if (!error) {
4677             subfacet_update_stats(subfacet, &stats);
4678         }
4679         subfacet->path = SF_NOT_INSTALLED;
4680     } else {
4681         ovs_assert(subfacet->dp_packet_count == 0);
4682         ovs_assert(subfacet->dp_byte_count == 0);
4683     }
4684 }
4685
4686 /* Resets 'subfacet''s datapath statistics counters.  This should be called
4687  * when 'subfacet''s statistics are cleared in the datapath.  If 'stats' is
4688  * non-null, it should contain the statistics returned by dpif when 'subfacet'
4689  * was reset in the datapath.  'stats' will be modified to include only
4690  * statistics new since 'subfacet' was last updated. */
4691 static void
4692 subfacet_reset_dp_stats(struct subfacet *subfacet,
4693                         struct dpif_flow_stats *stats)
4694 {
4695     if (stats
4696         && subfacet->dp_packet_count <= stats->n_packets
4697         && subfacet->dp_byte_count <= stats->n_bytes) {
4698         stats->n_packets -= subfacet->dp_packet_count;
4699         stats->n_bytes -= subfacet->dp_byte_count;
4700     }
4701
4702     subfacet->dp_packet_count = 0;
4703     subfacet->dp_byte_count = 0;
4704 }
4705
4706 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4707  *
4708  * Because of the meaning of a subfacet's counters, it only makes sense to do
4709  * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4710  * represents a packet that was sent by hand or if it represents statistics
4711  * that have been cleared out of the datapath. */
4712 static void
4713 subfacet_update_stats(struct subfacet *subfacet,
4714                       const struct dpif_flow_stats *stats)
4715 {
4716     if (stats->n_packets || stats->used > subfacet->used) {
4717         struct facet *facet = subfacet->facet;
4718
4719         subfacet->used = MAX(subfacet->used, stats->used);
4720         facet->used = MAX(facet->used, stats->used);
4721         facet->packet_count += stats->n_packets;
4722         facet->byte_count += stats->n_bytes;
4723         facet->tcp_flags |= stats->tcp_flags;
4724     }
4725 }
4726 \f
4727 /* Rules. */
4728
4729 /* Lookup 'flow' in 'ofproto''s classifier.  If 'wc' is non-null, sets
4730  * the fields that were relevant as part of the lookup. */
4731 void
4732 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4733                  struct flow_wildcards *wc, struct rule_dpif **rule)
4734 {
4735     struct ofport_dpif *port;
4736
4737     if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4738         return;
4739     }
4740     port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4741     if (!port) {
4742         VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4743                      flow->in_port.ofp_port);
4744     }
4745
4746     choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4747                      ofproto->no_packet_in_rule, rule);
4748 }
4749
4750 bool
4751 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4752                           const struct flow *flow, struct flow_wildcards *wc,
4753                           uint8_t table_id, struct rule_dpif **rule)
4754 {
4755     const struct cls_rule *cls_rule;
4756     struct classifier *cls;
4757     bool frag;
4758
4759     *rule = NULL;
4760     if (table_id >= N_TABLES) {
4761         return false;
4762     }
4763
4764     if (wc) {
4765         memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4766         wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4767     }
4768
4769     cls = &ofproto->up.tables[table_id].cls;
4770     ovs_rwlock_rdlock(&cls->rwlock);
4771     frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4772     if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4773         /* We must pretend that transport ports are unavailable. */
4774         struct flow ofpc_normal_flow = *flow;
4775         ofpc_normal_flow.tp_src = htons(0);
4776         ofpc_normal_flow.tp_dst = htons(0);
4777         cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4778     } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4779         cls_rule = &ofproto->drop_frags_rule->up.cr;
4780         if (wc) {
4781             flow_wildcards_init_exact(wc);
4782         }
4783     } else {
4784         cls_rule = classifier_lookup(cls, flow, wc);
4785     }
4786
4787     *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4788     rule_dpif_ref(*rule);
4789     ovs_rwlock_unlock(&cls->rwlock);
4790
4791     return *rule != NULL;
4792 }
4793
4794 /* Given a port configuration (specified as zero if there's no port), chooses
4795  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4796  * flow table miss. */
4797 void
4798 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4799                  struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4800 {
4801     *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4802     rule_dpif_ref(*rule);
4803 }
4804
4805 void
4806 rule_dpif_ref(struct rule_dpif *rule)
4807 {
4808     if (rule) {
4809         ofproto_rule_ref(&rule->up);
4810     }
4811 }
4812
4813 void
4814 rule_dpif_unref(struct rule_dpif *rule)
4815 {
4816     if (rule) {
4817         ofproto_rule_unref(&rule->up);
4818     }
4819 }
4820
4821 static void
4822 complete_operation(struct rule_dpif *rule)
4823     OVS_REQUIRES(ofproto_mutex)
4824 {
4825     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4826
4827     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4828     ofoperation_complete(rule->up.pending, 0);
4829 }
4830
4831 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4832 {
4833     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4834 }
4835
4836 static struct rule *
4837 rule_alloc(void)
4838 {
4839     struct rule_dpif *rule = xmalloc(sizeof *rule);
4840     return &rule->up;
4841 }
4842
4843 static void
4844 rule_dealloc(struct rule *rule_)
4845 {
4846     struct rule_dpif *rule = rule_dpif_cast(rule_);
4847     free(rule);
4848 }
4849
4850 static enum ofperr
4851 rule_construct(struct rule *rule_)
4852 {
4853     struct rule_dpif *rule = rule_dpif_cast(rule_);
4854     ovs_mutex_init(&rule->stats_mutex);
4855     ovs_mutex_lock(&rule->stats_mutex);
4856     rule->packet_count = 0;
4857     rule->byte_count = 0;
4858     ovs_mutex_unlock(&rule->stats_mutex);
4859     return 0;
4860 }
4861
4862 static void
4863 rule_insert(struct rule *rule_)
4864     OVS_REQUIRES(ofproto_mutex)
4865 {
4866     struct rule_dpif *rule = rule_dpif_cast(rule_);
4867     complete_operation(rule);
4868 }
4869
4870 static void
4871 rule_delete(struct rule *rule_)
4872     OVS_REQUIRES(ofproto_mutex)
4873 {
4874     struct rule_dpif *rule = rule_dpif_cast(rule_);
4875     complete_operation(rule);
4876 }
4877
4878 static void
4879 rule_destruct(struct rule *rule_)
4880 {
4881     struct rule_dpif *rule = rule_dpif_cast(rule_);
4882     ovs_mutex_destroy(&rule->stats_mutex);
4883 }
4884
4885 static void
4886 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4887 {
4888     struct rule_dpif *rule = rule_dpif_cast(rule_);
4889
4890     /* push_all_stats() can handle flow misses which, when using the learn
4891      * action, can cause rules to be added and deleted.  This can corrupt our
4892      * caller's datastructures which assume that rule_get_stats() doesn't have
4893      * an impact on the flow table. To be safe, we disable miss handling. */
4894     push_all_stats__(false);
4895
4896     /* Start from historical data for 'rule' itself that are no longer tracked
4897      * in facets.  This counts, for example, facets that have expired. */
4898     ovs_mutex_lock(&rule->stats_mutex);
4899     *packets = rule->packet_count;
4900     *bytes = rule->byte_count;
4901     ovs_mutex_unlock(&rule->stats_mutex);
4902 }
4903
4904 static void
4905 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4906                   struct ofpbuf *packet)
4907 {
4908     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4909     struct dpif_flow_stats stats;
4910     struct xlate_out xout;
4911     struct xlate_in xin;
4912
4913     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4914     rule_dpif_credit_stats(rule, &stats);
4915
4916     xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4917     xin.resubmit_stats = &stats;
4918     xlate_actions(&xin, &xout);
4919
4920     execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4921                         xout.odp_actions.size, packet);
4922
4923     xlate_out_uninit(&xout);
4924 }
4925
4926 static enum ofperr
4927 rule_execute(struct rule *rule, const struct flow *flow,
4928              struct ofpbuf *packet)
4929 {
4930     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4931     ofpbuf_delete(packet);
4932     return 0;
4933 }
4934
4935 static void
4936 rule_modify_actions(struct rule *rule_, bool reset_counters)
4937     OVS_REQUIRES(ofproto_mutex)
4938 {
4939     struct rule_dpif *rule = rule_dpif_cast(rule_);
4940
4941     if (reset_counters) {
4942         ovs_mutex_lock(&rule->stats_mutex);
4943         rule->packet_count = 0;
4944         rule->byte_count = 0;
4945         ovs_mutex_unlock(&rule->stats_mutex);
4946     }
4947
4948     complete_operation(rule);
4949 }
4950 \f
4951 /* Sends 'packet' out 'ofport'.
4952  * May modify 'packet'.
4953  * Returns 0 if successful, otherwise a positive errno value. */
4954 static int
4955 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4956 {
4957     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4958     uint64_t odp_actions_stub[1024 / 8];
4959     struct ofpbuf key, odp_actions;
4960     struct dpif_flow_stats stats;
4961     struct odputil_keybuf keybuf;
4962     struct ofpact_output output;
4963     struct xlate_out xout;
4964     struct xlate_in xin;
4965     struct flow flow;
4966     union flow_in_port in_port_;
4967     int error;
4968
4969     ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4970     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4971
4972     /* Use OFPP_NONE as the in_port to avoid special packet processing. */
4973     in_port_.ofp_port = OFPP_NONE;
4974     flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
4975     odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
4976                                                              OFPP_LOCAL));
4977     dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
4978
4979     ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
4980     output.port = ofport->up.ofp_port;
4981     output.max_len = 0;
4982
4983     xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
4984     xin.ofpacts_len = sizeof output;
4985     xin.ofpacts = &output.ofpact;
4986     xin.resubmit_stats = &stats;
4987     xlate_actions(&xin, &xout);
4988
4989     error = dpif_execute(ofproto->backer->dpif,
4990                          key.data, key.size,
4991                          xout.odp_actions.data, xout.odp_actions.size,
4992                          packet);
4993     xlate_out_uninit(&xout);
4994
4995     if (error) {
4996         VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
4997                      ofproto->up.name, netdev_get_name(ofport->up.netdev),
4998                      ovs_strerror(error));
4999     }
5000
5001     ofproto->stats.tx_packets++;
5002     ofproto->stats.tx_bytes += packet->size;
5003     return error;
5004 }
5005
5006 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5007  * The action will state 'slow' as the reason that the action is in the slow
5008  * path.  (This is purely informational: it allows a human viewing "ovs-dpctl
5009  * dump-flows" output to see why a flow is in the slow path.)
5010  *
5011  * The 'stub_size' bytes in 'stub' will be used to store the action.
5012  * 'stub_size' must be large enough for the action.
5013  *
5014  * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5015  * respectively. */
5016 static void
5017 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5018                   enum slow_path_reason slow,
5019                   uint64_t *stub, size_t stub_size,
5020                   const struct nlattr **actionsp, size_t *actions_lenp)
5021 {
5022     union user_action_cookie cookie;
5023     struct ofpbuf buf;
5024
5025     cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5026     cookie.slow_path.unused = 0;
5027     cookie.slow_path.reason = slow;
5028
5029     ofpbuf_use_stack(&buf, stub, stub_size);
5030     if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5031         uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5032                                          ODPP_NONE);
5033         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5034     } else {
5035         odp_port_t odp_port;
5036         uint32_t pid;
5037
5038         odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5039         pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5040         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5041     }
5042     *actionsp = buf.data;
5043     *actions_lenp = buf.size;
5044 }
5045 \f
5046 static bool
5047 set_frag_handling(struct ofproto *ofproto_,
5048                   enum ofp_config_flags frag_handling)
5049 {
5050     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5051     if (frag_handling != OFPC_FRAG_REASM) {
5052         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5053         return true;
5054     } else {
5055         return false;
5056     }
5057 }
5058
5059 static enum ofperr
5060 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5061            const struct flow *flow,
5062            const struct ofpact *ofpacts, size_t ofpacts_len)
5063 {
5064     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5065     struct odputil_keybuf keybuf;
5066     struct dpif_flow_stats stats;
5067     struct xlate_out xout;
5068     struct xlate_in xin;
5069     struct ofpbuf key;
5070
5071
5072     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5073     odp_flow_key_from_flow(&key, flow,
5074                            ofp_port_to_odp_port(ofproto,
5075                                       flow->in_port.ofp_port));
5076
5077     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5078
5079     xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5080     xin.resubmit_stats = &stats;
5081     xin.ofpacts_len = ofpacts_len;
5082     xin.ofpacts = ofpacts;
5083
5084     xlate_actions(&xin, &xout);
5085     dpif_execute(ofproto->backer->dpif, key.data, key.size,
5086                  xout.odp_actions.data, xout.odp_actions.size, packet);
5087     xlate_out_uninit(&xout);
5088
5089     return 0;
5090 }
5091 \f
5092 /* NetFlow. */
5093
5094 static int
5095 set_netflow(struct ofproto *ofproto_,
5096             const struct netflow_options *netflow_options)
5097 {
5098     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5099
5100     if (netflow_options) {
5101         if (!ofproto->netflow) {
5102             ofproto->netflow = netflow_create();
5103             ofproto->backer->need_revalidate = REV_RECONFIGURE;
5104         }
5105         return netflow_set_options(ofproto->netflow, netflow_options);
5106     } else if (ofproto->netflow) {
5107         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5108         netflow_destroy(ofproto->netflow);
5109         ofproto->netflow = NULL;
5110     }
5111
5112     return 0;
5113 }
5114
5115 static void
5116 get_netflow_ids(const struct ofproto *ofproto_,
5117                 uint8_t *engine_type, uint8_t *engine_id)
5118 {
5119     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5120
5121     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5122 }
5123
5124 static void
5125 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5126 {
5127     if (!facet_is_controller_flow(facet) &&
5128         netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5129         struct subfacet *subfacet;
5130         struct ofexpired expired;
5131
5132         LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5133             if (subfacet->path == SF_FAST_PATH) {
5134                 struct dpif_flow_stats stats;
5135
5136                 subfacet_install(subfacet, &facet->xout.odp_actions,
5137                                  &stats);
5138                 subfacet_update_stats(subfacet, &stats);
5139             }
5140         }
5141
5142         expired.flow = facet->flow;
5143         expired.packet_count = facet->packet_count;
5144         expired.byte_count = facet->byte_count;
5145         expired.used = facet->used;
5146         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5147     }
5148 }
5149
5150 static void
5151 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5152 {
5153     struct cls_cursor cursor;
5154     struct facet *facet;
5155
5156     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5157     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5158     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5159         send_active_timeout(ofproto, facet);
5160     }
5161     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5162 }
5163 \f
5164 static struct ofproto_dpif *
5165 ofproto_dpif_lookup(const char *name)
5166 {
5167     struct ofproto_dpif *ofproto;
5168
5169     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5170                              hash_string(name, 0), &all_ofproto_dpifs) {
5171         if (!strcmp(ofproto->up.name, name)) {
5172             return ofproto;
5173         }
5174     }
5175     return NULL;
5176 }
5177
5178 static void
5179 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5180                           const char *argv[], void *aux OVS_UNUSED)
5181 {
5182     struct ofproto_dpif *ofproto;
5183
5184     if (argc > 1) {
5185         ofproto = ofproto_dpif_lookup(argv[1]);
5186         if (!ofproto) {
5187             unixctl_command_reply_error(conn, "no such bridge");
5188             return;
5189         }
5190         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5191         mac_learning_flush(ofproto->ml);
5192         ovs_rwlock_unlock(&ofproto->ml->rwlock);
5193     } else {
5194         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5195             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5196             mac_learning_flush(ofproto->ml);
5197             ovs_rwlock_unlock(&ofproto->ml->rwlock);
5198         }
5199     }
5200
5201     unixctl_command_reply(conn, "table successfully flushed");
5202 }
5203
5204 static struct ofport_dpif *
5205 ofbundle_get_a_port(const struct ofbundle *bundle)
5206 {
5207     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5208                         bundle_node);
5209 }
5210
5211 static void
5212 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5213                          const char *argv[], void *aux OVS_UNUSED)
5214 {
5215     struct ds ds = DS_EMPTY_INITIALIZER;
5216     const struct ofproto_dpif *ofproto;
5217     const struct mac_entry *e;
5218
5219     ofproto = ofproto_dpif_lookup(argv[1]);
5220     if (!ofproto) {
5221         unixctl_command_reply_error(conn, "no such bridge");
5222         return;
5223     }
5224
5225     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
5226     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5227     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5228         struct ofbundle *bundle = e->port.p;
5229         char name[OFP_MAX_PORT_NAME_LEN];
5230
5231         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5232                                name, sizeof name);
5233         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
5234                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
5235                       mac_entry_age(ofproto->ml, e));
5236     }
5237     ovs_rwlock_unlock(&ofproto->ml->rwlock);
5238     unixctl_command_reply(conn, ds_cstr(&ds));
5239     ds_destroy(&ds);
5240 }
5241
5242 struct trace_ctx {
5243     struct xlate_out xout;
5244     struct xlate_in xin;
5245     struct flow flow;
5246     struct ds *result;
5247 };
5248
5249 static void
5250 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5251 {
5252     struct rule_actions *actions;
5253     ovs_be64 cookie;
5254
5255     ds_put_char_multiple(result, '\t', level);
5256     if (!rule) {
5257         ds_put_cstr(result, "No match\n");
5258         return;
5259     }
5260
5261     ovs_mutex_lock(&rule->up.mutex);
5262     cookie = rule->up.flow_cookie;
5263     ovs_mutex_unlock(&rule->up.mutex);
5264
5265     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5266                   rule ? rule->up.table_id : 0, ntohll(cookie));
5267     cls_rule_format(&rule->up.cr, result);
5268     ds_put_char(result, '\n');
5269
5270     actions = rule_dpif_get_actions(rule);
5271
5272     ds_put_char_multiple(result, '\t', level);
5273     ds_put_cstr(result, "OpenFlow ");
5274     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
5275     ds_put_char(result, '\n');
5276
5277     rule_actions_unref(actions);
5278 }
5279
5280 static void
5281 trace_format_flow(struct ds *result, int level, const char *title,
5282                   struct trace_ctx *trace)
5283 {
5284     ds_put_char_multiple(result, '\t', level);
5285     ds_put_format(result, "%s: ", title);
5286     if (flow_equal(&trace->xin.flow, &trace->flow)) {
5287         ds_put_cstr(result, "unchanged");
5288     } else {
5289         flow_format(result, &trace->xin.flow);
5290         trace->flow = trace->xin.flow;
5291     }
5292     ds_put_char(result, '\n');
5293 }
5294
5295 static void
5296 trace_format_regs(struct ds *result, int level, const char *title,
5297                   struct trace_ctx *trace)
5298 {
5299     size_t i;
5300
5301     ds_put_char_multiple(result, '\t', level);
5302     ds_put_format(result, "%s:", title);
5303     for (i = 0; i < FLOW_N_REGS; i++) {
5304         ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5305     }
5306     ds_put_char(result, '\n');
5307 }
5308
5309 static void
5310 trace_format_odp(struct ds *result, int level, const char *title,
5311                  struct trace_ctx *trace)
5312 {
5313     struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5314
5315     ds_put_char_multiple(result, '\t', level);
5316     ds_put_format(result, "%s: ", title);
5317     format_odp_actions(result, odp_actions->data, odp_actions->size);
5318     ds_put_char(result, '\n');
5319 }
5320
5321 static void
5322 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5323 {
5324     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5325     struct ds *result = trace->result;
5326
5327     ds_put_char(result, '\n');
5328     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5329     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5330     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
5331     trace_format_rule(result, recurse + 1, rule);
5332 }
5333
5334 static void
5335 trace_report(struct xlate_in *xin, const char *s, int recurse)
5336 {
5337     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5338     struct ds *result = trace->result;
5339
5340     ds_put_char_multiple(result, '\t', recurse);
5341     ds_put_cstr(result, s);
5342     ds_put_char(result, '\n');
5343 }
5344
5345 static void
5346 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5347                       void *aux OVS_UNUSED)
5348 {
5349     const struct dpif_backer *backer;
5350     struct ofproto_dpif *ofproto;
5351     struct ofpbuf odp_key, odp_mask;
5352     struct ofpbuf *packet;
5353     struct ds result;
5354     struct flow flow;
5355     char *s;
5356
5357     packet = NULL;
5358     backer = NULL;
5359     ds_init(&result);
5360     ofpbuf_init(&odp_key, 0);
5361     ofpbuf_init(&odp_mask, 0);
5362
5363     /* Handle "-generate" or a hex string as the last argument. */
5364     if (!strcmp(argv[argc - 1], "-generate")) {
5365         packet = ofpbuf_new(0);
5366         argc--;
5367     } else {
5368         const char *error = eth_from_hex(argv[argc - 1], &packet);
5369         if (!error) {
5370             argc--;
5371         } else if (argc == 4) {
5372             /* The 3-argument form must end in "-generate' or a hex string. */
5373             unixctl_command_reply_error(conn, error);
5374             goto exit;
5375         }
5376     }
5377
5378     /* Parse the flow and determine whether a datapath or
5379      * bridge is specified. If function odp_flow_key_from_string()
5380      * returns 0, the flow is a odp_flow. If function
5381      * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5382     if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5383         /* If the odp_flow is the second argument,
5384          * the datapath name is the first argument. */
5385         if (argc == 3) {
5386             const char *dp_type;
5387             if (!strncmp(argv[1], "ovs-", 4)) {
5388                 dp_type = argv[1] + 4;
5389             } else {
5390                 dp_type = argv[1];
5391             }
5392             backer = shash_find_data(&all_dpif_backers, dp_type);
5393             if (!backer) {
5394                 unixctl_command_reply_error(conn, "Cannot find datapath "
5395                                "of this name");
5396                 goto exit;
5397             }
5398         } else {
5399             /* No datapath name specified, so there should be only one
5400              * datapath. */
5401             struct shash_node *node;
5402             if (shash_count(&all_dpif_backers) != 1) {
5403                 unixctl_command_reply_error(conn, "Must specify datapath "
5404                          "name, there is more than one type of datapath");
5405                 goto exit;
5406             }
5407             node = shash_first(&all_dpif_backers);
5408             backer = node->data;
5409         }
5410
5411         if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5412                           NULL, &ofproto, NULL)) {
5413             unixctl_command_reply_error(conn, "Invalid datapath flow");
5414             goto exit;
5415         }
5416         ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5417     } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5418         if (argc != 3) {
5419             unixctl_command_reply_error(conn, "Must specify bridge name");
5420             goto exit;
5421         }
5422
5423         ofproto = ofproto_dpif_lookup(argv[1]);
5424         if (!ofproto) {
5425             unixctl_command_reply_error(conn, "Unknown bridge name");
5426             goto exit;
5427         }
5428     } else {
5429         unixctl_command_reply_error(conn, "Bad flow syntax");
5430         goto exit;
5431     }
5432
5433     /* Generate a packet, if requested. */
5434     if (packet) {
5435         if (!packet->size) {
5436             flow_compose(packet, &flow);
5437         } else {
5438             union flow_in_port in_port_;
5439
5440             in_port_ = flow.in_port;
5441             ds_put_cstr(&result, "Packet: ");
5442             s = ofp_packet_to_string(packet->data, packet->size);
5443             ds_put_cstr(&result, s);
5444             free(s);
5445
5446             /* Use the metadata from the flow and the packet argument
5447              * to reconstruct the flow. */
5448             flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5449                          &in_port_, &flow);
5450         }
5451     }
5452
5453     ofproto_trace(ofproto, &flow, packet, &result);
5454     unixctl_command_reply(conn, ds_cstr(&result));
5455
5456 exit:
5457     ds_destroy(&result);
5458     ofpbuf_delete(packet);
5459     ofpbuf_uninit(&odp_key);
5460     ofpbuf_uninit(&odp_mask);
5461 }
5462
5463 static void
5464 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5465               const struct ofpbuf *packet, struct ds *ds)
5466 {
5467     struct rule_dpif *rule;
5468     struct flow_wildcards wc;
5469
5470     ds_put_cstr(ds, "Flow: ");
5471     flow_format(ds, flow);
5472     ds_put_char(ds, '\n');
5473
5474     flow_wildcards_init_catchall(&wc);
5475     rule_dpif_lookup(ofproto, flow, &wc, &rule);
5476
5477     trace_format_rule(ds, 0, rule);
5478     if (rule == ofproto->miss_rule) {
5479         ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5480     } else if (rule == ofproto->no_packet_in_rule) {
5481         ds_put_cstr(ds, "\nNo match, packets dropped because "
5482                     "OFPPC_NO_PACKET_IN is set on in_port.\n");
5483     } else if (rule == ofproto->drop_frags_rule) {
5484         ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5485                     "and the fragment handling mode is \"drop\".\n");
5486     }
5487
5488     if (rule) {
5489         uint64_t odp_actions_stub[1024 / 8];
5490         struct ofpbuf odp_actions;
5491         struct trace_ctx trace;
5492         struct match match;
5493         uint8_t tcp_flags;
5494
5495         tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5496         trace.result = ds;
5497         trace.flow = *flow;
5498         ofpbuf_use_stub(&odp_actions,
5499                         odp_actions_stub, sizeof odp_actions_stub);
5500         xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5501         trace.xin.resubmit_hook = trace_resubmit;
5502         trace.xin.report_hook = trace_report;
5503
5504         xlate_actions(&trace.xin, &trace.xout);
5505         flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5506
5507         ds_put_char(ds, '\n');
5508         trace_format_flow(ds, 0, "Final flow", &trace);
5509
5510         match_init(&match, flow, &trace.xout.wc);
5511         ds_put_cstr(ds, "Relevant fields: ");
5512         match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5513         ds_put_char(ds, '\n');
5514
5515         ds_put_cstr(ds, "Datapath actions: ");
5516         format_odp_actions(ds, trace.xout.odp_actions.data,
5517                            trace.xout.odp_actions.size);
5518
5519         if (trace.xout.slow) {
5520             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5521                         "slow path because it:");
5522             switch (trace.xout.slow) {
5523             case SLOW_CFM:
5524                 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5525                 break;
5526             case SLOW_LACP:
5527                 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5528                 break;
5529             case SLOW_STP:
5530                 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5531                 break;
5532             case SLOW_BFD:
5533                 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5534                 break;
5535             case SLOW_CONTROLLER:
5536                 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5537                             "to the OpenFlow controller.");
5538                 break;
5539             case __SLOW_MAX:
5540                 NOT_REACHED();
5541             }
5542         }
5543
5544         xlate_out_uninit(&trace.xout);
5545     }
5546
5547     rule_dpif_unref(rule);
5548 }
5549
5550 /* Runs a self-check of flow translations in 'ofproto'.  Appends a message to
5551  * 'reply' describing the results. */
5552 static void
5553 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5554 {
5555     struct cls_cursor cursor;
5556     struct facet *facet;
5557     int errors;
5558
5559     errors = 0;
5560     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5561     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5562     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5563         if (!facet_check_consistency(facet)) {
5564             errors++;
5565         }
5566     }
5567     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5568     if (errors) {
5569         ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5570     }
5571
5572     if (errors) {
5573         ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5574                       ofproto->up.name, errors);
5575     } else {
5576         ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5577     }
5578 }
5579
5580 static void
5581 ofproto_dpif_self_check(struct unixctl_conn *conn,
5582                         int argc, const char *argv[], void *aux OVS_UNUSED)
5583 {
5584     struct ds reply = DS_EMPTY_INITIALIZER;
5585     struct ofproto_dpif *ofproto;
5586
5587     if (argc > 1) {
5588         ofproto = ofproto_dpif_lookup(argv[1]);
5589         if (!ofproto) {
5590             unixctl_command_reply_error(conn, "Unknown ofproto (use "
5591                                         "ofproto/list for help)");
5592             return;
5593         }
5594         ofproto_dpif_self_check__(ofproto, &reply);
5595     } else {
5596         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5597             ofproto_dpif_self_check__(ofproto, &reply);
5598         }
5599     }
5600
5601     unixctl_command_reply(conn, ds_cstr(&reply));
5602     ds_destroy(&reply);
5603 }
5604
5605 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
5606  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
5607  * to destroy 'ofproto_shash' and free the returned value. */
5608 static const struct shash_node **
5609 get_ofprotos(struct shash *ofproto_shash)
5610 {
5611     const struct ofproto_dpif *ofproto;
5612
5613     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5614         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5615         shash_add_nocopy(ofproto_shash, name, ofproto);
5616     }
5617
5618     return shash_sort(ofproto_shash);
5619 }
5620
5621 static void
5622 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5623                               const char *argv[] OVS_UNUSED,
5624                               void *aux OVS_UNUSED)
5625 {
5626     struct ds ds = DS_EMPTY_INITIALIZER;
5627     struct shash ofproto_shash;
5628     const struct shash_node **sorted_ofprotos;
5629     int i;
5630
5631     shash_init(&ofproto_shash);
5632     sorted_ofprotos = get_ofprotos(&ofproto_shash);
5633     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5634         const struct shash_node *node = sorted_ofprotos[i];
5635         ds_put_format(&ds, "%s\n", node->name);
5636     }
5637
5638     shash_destroy(&ofproto_shash);
5639     free(sorted_ofprotos);
5640
5641     unixctl_command_reply(conn, ds_cstr(&ds));
5642     ds_destroy(&ds);
5643 }
5644
5645 static void
5646 show_dp_rates(struct ds *ds, const char *heading,
5647               const struct avg_subfacet_rates *rates)
5648 {
5649     ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5650                   heading, rates->add_rate, rates->del_rate);
5651 }
5652
5653 static void
5654 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5655 {
5656     const struct shash_node **ofprotos;
5657     struct ofproto_dpif *ofproto;
5658     struct shash ofproto_shash;
5659     uint64_t n_hit, n_missed;
5660     long long int minutes;
5661     size_t i;
5662
5663     n_hit = n_missed = 0;
5664     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5665         if (ofproto->backer == backer) {
5666             n_missed += ofproto->n_missed;
5667             n_hit += ofproto->n_hit;
5668         }
5669     }
5670
5671     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5672                   dpif_name(backer->dpif), n_hit, n_missed);
5673     ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5674                   " life span: %lldms\n", hmap_count(&backer->subfacets),
5675                   backer->avg_n_subfacet, backer->max_n_subfacet,
5676                   backer->avg_subfacet_life);
5677
5678     minutes = (time_msec() - backer->created) / (1000 * 60);
5679     if (minutes >= 60) {
5680         show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5681     }
5682     if (minutes >= 60 * 24) {
5683         show_dp_rates(ds, "\tdaily avg:",  &backer->daily);
5684     }
5685     show_dp_rates(ds, "\toverall avg:",  &backer->lifetime);
5686
5687     shash_init(&ofproto_shash);
5688     ofprotos = get_ofprotos(&ofproto_shash);
5689     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5690         struct ofproto_dpif *ofproto = ofprotos[i]->data;
5691         const struct shash_node **ports;
5692         size_t j;
5693
5694         if (ofproto->backer != backer) {
5695             continue;
5696         }
5697
5698         ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5699                       ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5700
5701         ports = shash_sort(&ofproto->up.port_by_name);
5702         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5703             const struct shash_node *node = ports[j];
5704             struct ofport *ofport = node->data;
5705             struct smap config;
5706             odp_port_t odp_port;
5707
5708             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5709                           ofport->ofp_port);
5710
5711             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5712             if (odp_port != ODPP_NONE) {
5713                 ds_put_format(ds, "%"PRIu32":", odp_port);
5714             } else {
5715                 ds_put_cstr(ds, "none:");
5716             }
5717
5718             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5719
5720             smap_init(&config);
5721             if (!netdev_get_config(ofport->netdev, &config)) {
5722                 const struct smap_node **nodes;
5723                 size_t i;
5724
5725                 nodes = smap_sort(&config);
5726                 for (i = 0; i < smap_count(&config); i++) {
5727                     const struct smap_node *node = nodes[i];
5728                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5729                                   node->key, node->value);
5730                 }
5731                 free(nodes);
5732             }
5733             smap_destroy(&config);
5734
5735             ds_put_char(ds, ')');
5736             ds_put_char(ds, '\n');
5737         }
5738         free(ports);
5739     }
5740     shash_destroy(&ofproto_shash);
5741     free(ofprotos);
5742 }
5743
5744 static void
5745 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5746                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5747 {
5748     struct ds ds = DS_EMPTY_INITIALIZER;
5749     const struct shash_node **backers;
5750     int i;
5751
5752     backers = shash_sort(&all_dpif_backers);
5753     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5754         dpif_show_backer(backers[i]->data, &ds);
5755     }
5756     free(backers);
5757
5758     unixctl_command_reply(conn, ds_cstr(&ds));
5759     ds_destroy(&ds);
5760 }
5761
5762 /* Dump the megaflow (facet) cache.  This is useful to check the
5763  * correctness of flow wildcarding, since the same mechanism is used for
5764  * both xlate caching and kernel wildcarding.
5765  *
5766  * It's important to note that in the output the flow description uses
5767  * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5768  *
5769  * This command is only needed for advanced debugging, so it's not
5770  * documented in the man page. */
5771 static void
5772 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5773                                     int argc OVS_UNUSED, const char *argv[],
5774                                     void *aux OVS_UNUSED)
5775 {
5776     struct ds ds = DS_EMPTY_INITIALIZER;
5777     const struct ofproto_dpif *ofproto;
5778     long long int now = time_msec();
5779     struct cls_cursor cursor;
5780     struct facet *facet;
5781
5782     ofproto = ofproto_dpif_lookup(argv[1]);
5783     if (!ofproto) {
5784         unixctl_command_reply_error(conn, "no such bridge");
5785         return;
5786     }
5787
5788     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5789     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5790     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5791         cls_rule_format(&facet->cr, &ds);
5792         ds_put_cstr(&ds, ", ");
5793         ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5794         ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5795         ds_put_cstr(&ds, "Datapath actions: ");
5796         if (facet->xout.slow) {
5797             uint64_t slow_path_stub[128 / 8];
5798             const struct nlattr *actions;
5799             size_t actions_len;
5800
5801             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5802                               slow_path_stub, sizeof slow_path_stub,
5803                               &actions, &actions_len);
5804             format_odp_actions(&ds, actions, actions_len);
5805         } else {
5806             format_odp_actions(&ds, facet->xout.odp_actions.data,
5807                                facet->xout.odp_actions.size);
5808         }
5809         ds_put_cstr(&ds, "\n");
5810     }
5811     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5812
5813     ds_chomp(&ds, '\n');
5814     unixctl_command_reply(conn, ds_cstr(&ds));
5815     ds_destroy(&ds);
5816 }
5817
5818 /* Disable using the megaflows.
5819  *
5820  * This command is only needed for advanced debugging, so it's not
5821  * documented in the man page. */
5822 static void
5823 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5824                                        int argc OVS_UNUSED,
5825                                        const char *argv[] OVS_UNUSED,
5826                                        void *aux OVS_UNUSED)
5827 {
5828     struct ofproto_dpif *ofproto;
5829
5830     enable_megaflows = false;
5831
5832     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5833         flush(&ofproto->up);
5834     }
5835
5836     unixctl_command_reply(conn, "megaflows disabled");
5837 }
5838
5839 /* Re-enable using megaflows.
5840  *
5841  * This command is only needed for advanced debugging, so it's not
5842  * documented in the man page. */
5843 static void
5844 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5845                                       int argc OVS_UNUSED,
5846                                       const char *argv[] OVS_UNUSED,
5847                                       void *aux OVS_UNUSED)
5848 {
5849     struct ofproto_dpif *ofproto;
5850
5851     enable_megaflows = true;
5852
5853     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5854         flush(&ofproto->up);
5855     }
5856
5857     unixctl_command_reply(conn, "megaflows enabled");
5858 }
5859
5860 static void
5861 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5862                                 int argc OVS_UNUSED, const char *argv[],
5863                                 void *aux OVS_UNUSED)
5864 {
5865     struct ds ds = DS_EMPTY_INITIALIZER;
5866     const struct ofproto_dpif *ofproto;
5867     struct subfacet *subfacet;
5868
5869     ofproto = ofproto_dpif_lookup(argv[1]);
5870     if (!ofproto) {
5871         unixctl_command_reply_error(conn, "no such bridge");
5872         return;
5873     }
5874
5875     update_stats(ofproto->backer);
5876
5877     HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5878         struct facet *facet = subfacet->facet;
5879         struct odputil_keybuf maskbuf;
5880         struct ofpbuf mask;
5881
5882         if (facet->ofproto != ofproto) {
5883             continue;
5884         }
5885
5886         ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5887         if (enable_megaflows) {
5888             odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5889                                    &facet->flow, UINT32_MAX);
5890         }
5891
5892         odp_flow_format(subfacet->key, subfacet->key_len,
5893                         mask.data, mask.size, &ds, false);
5894
5895         ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5896                       subfacet->dp_packet_count, subfacet->dp_byte_count);
5897         if (subfacet->used) {
5898             ds_put_format(&ds, "%.3fs",
5899                           (time_msec() - subfacet->used) / 1000.0);
5900         } else {
5901             ds_put_format(&ds, "never");
5902         }
5903         if (subfacet->facet->tcp_flags) {
5904             ds_put_cstr(&ds, ", flags:");
5905             packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5906         }
5907
5908         ds_put_cstr(&ds, ", actions:");
5909         if (facet->xout.slow) {
5910             uint64_t slow_path_stub[128 / 8];
5911             const struct nlattr *actions;
5912             size_t actions_len;
5913
5914             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5915                               slow_path_stub, sizeof slow_path_stub,
5916                               &actions, &actions_len);
5917             format_odp_actions(&ds, actions, actions_len);
5918         } else {
5919             format_odp_actions(&ds, facet->xout.odp_actions.data,
5920                                facet->xout.odp_actions.size);
5921         }
5922         ds_put_char(&ds, '\n');
5923     }
5924
5925     unixctl_command_reply(conn, ds_cstr(&ds));
5926     ds_destroy(&ds);
5927 }
5928
5929 static void
5930 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5931                                int argc OVS_UNUSED, const char *argv[],
5932                                void *aux OVS_UNUSED)
5933 {
5934     struct ds ds = DS_EMPTY_INITIALIZER;
5935     struct ofproto_dpif *ofproto;
5936
5937     ofproto = ofproto_dpif_lookup(argv[1]);
5938     if (!ofproto) {
5939         unixctl_command_reply_error(conn, "no such bridge");
5940         return;
5941     }
5942
5943     flush(&ofproto->up);
5944
5945     unixctl_command_reply(conn, ds_cstr(&ds));
5946     ds_destroy(&ds);
5947 }
5948
5949 static void
5950 ofproto_dpif_unixctl_init(void)
5951 {
5952     static bool registered;
5953     if (registered) {
5954         return;
5955     }
5956     registered = true;
5957
5958     unixctl_command_register(
5959         "ofproto/trace",
5960         "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
5961         1, 3, ofproto_unixctl_trace, NULL);
5962     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5963                              ofproto_unixctl_fdb_flush, NULL);
5964     unixctl_command_register("fdb/show", "bridge", 1, 1,
5965                              ofproto_unixctl_fdb_show, NULL);
5966     unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
5967                              ofproto_dpif_self_check, NULL);
5968     unixctl_command_register("dpif/dump-dps", "", 0, 0,
5969                              ofproto_unixctl_dpif_dump_dps, NULL);
5970     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5971                              NULL);
5972     unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
5973                              ofproto_unixctl_dpif_dump_flows, NULL);
5974     unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
5975                              ofproto_unixctl_dpif_del_flows, NULL);
5976     unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
5977                              ofproto_unixctl_dpif_dump_megaflows, NULL);
5978     unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
5979                              ofproto_unixctl_dpif_disable_megaflows, NULL);
5980     unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
5981                              ofproto_unixctl_dpif_enable_megaflows, NULL);
5982 }
5983 \f
5984 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5985  *
5986  * This is deprecated.  It is only for compatibility with broken device drivers
5987  * in old versions of Linux that do not properly support VLANs when VLAN
5988  * devices are not used.  When broken device drivers are no longer in
5989  * widespread use, we will delete these interfaces. */
5990
5991 static int
5992 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5993 {
5994     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5995     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5996
5997     if (realdev_ofp_port == ofport->realdev_ofp_port
5998         && vid == ofport->vlandev_vid) {
5999         return 0;
6000     }
6001
6002     ofproto->backer->need_revalidate = REV_RECONFIGURE;
6003
6004     if (ofport->realdev_ofp_port) {
6005         vsp_remove(ofport);
6006     }
6007     if (realdev_ofp_port && ofport->bundle) {
6008         /* vlandevs are enslaved to their realdevs, so they are not allowed to
6009          * themselves be part of a bundle. */
6010         bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6011     }
6012
6013     ofport->realdev_ofp_port = realdev_ofp_port;
6014     ofport->vlandev_vid = vid;
6015
6016     if (realdev_ofp_port) {
6017         vsp_add(ofport, realdev_ofp_port, vid);
6018     }
6019
6020     return 0;
6021 }
6022
6023 static uint32_t
6024 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6025 {
6026     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6027 }
6028
6029 bool
6030 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6031     OVS_EXCLUDED(ofproto->vsp_mutex)
6032 {
6033     bool ret;
6034
6035     ovs_mutex_lock(&ofproto->vsp_mutex);
6036     ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6037     ovs_mutex_unlock(&ofproto->vsp_mutex);
6038     return ret;
6039 }
6040
6041 static ofp_port_t
6042 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6043                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6044     OVS_REQUIRES(ofproto->vsp_mutex)
6045 {
6046     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6047         int vid = vlan_tci_to_vid(vlan_tci);
6048         const struct vlan_splinter *vsp;
6049
6050         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6051                                  hash_realdev_vid(realdev_ofp_port, vid),
6052                                  &ofproto->realdev_vid_map) {
6053             if (vsp->realdev_ofp_port == realdev_ofp_port
6054                 && vsp->vid == vid) {
6055                 return vsp->vlandev_ofp_port;
6056             }
6057         }
6058     }
6059     return realdev_ofp_port;
6060 }
6061
6062 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6063  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6064  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
6065  * 'vlan_tci' 9, it would return the port number of eth0.9.
6066  *
6067  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6068  * function just returns its 'realdev_ofp_port' argument. */
6069 ofp_port_t
6070 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6071                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6072     OVS_EXCLUDED(ofproto->vsp_mutex)
6073 {
6074     ofp_port_t ret;
6075
6076     ovs_mutex_lock(&ofproto->vsp_mutex);
6077     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6078     ovs_mutex_unlock(&ofproto->vsp_mutex);
6079     return ret;
6080 }
6081
6082 static struct vlan_splinter *
6083 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6084 {
6085     struct vlan_splinter *vsp;
6086
6087     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6088                              hash_ofp_port(vlandev_ofp_port),
6089                              &ofproto->vlandev_map) {
6090         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6091             return vsp;
6092         }
6093     }
6094
6095     return NULL;
6096 }
6097
6098 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6099  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6100  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
6101  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6102  * eth0 and store 9 in '*vid'.
6103  *
6104  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6105  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
6106  * always does.*/
6107 static ofp_port_t
6108 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6109                        ofp_port_t vlandev_ofp_port, int *vid)
6110     OVS_REQUIRES(ofproto->vsp_mutex)
6111 {
6112     if (!hmap_is_empty(&ofproto->vlandev_map)) {
6113         const struct vlan_splinter *vsp;
6114
6115         vsp = vlandev_find(ofproto, vlandev_ofp_port);
6116         if (vsp) {
6117             if (vid) {
6118                 *vid = vsp->vid;
6119             }
6120             return vsp->realdev_ofp_port;
6121         }
6122     }
6123     return 0;
6124 }
6125
6126 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6127  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
6128  * 'flow->in_port' to the "real" device backing the VLAN device, sets
6129  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Otherwise (which is
6130  * always the case unless VLAN splinters are enabled), returns false without
6131  * making any changes. */
6132 bool
6133 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6134     OVS_EXCLUDED(ofproto->vsp_mutex)
6135 {
6136     ofp_port_t realdev;
6137     int vid;
6138
6139     ovs_mutex_lock(&ofproto->vsp_mutex);
6140     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6141     ovs_mutex_unlock(&ofproto->vsp_mutex);
6142     if (!realdev) {
6143         return false;
6144     }
6145
6146     /* Cause the flow to be processed as if it came in on the real device with
6147      * the VLAN device's VLAN ID. */
6148     flow->in_port.ofp_port = realdev;
6149     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6150     return true;
6151 }
6152
6153 static void
6154 vsp_remove(struct ofport_dpif *port)
6155 {
6156     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6157     struct vlan_splinter *vsp;
6158
6159     ovs_mutex_lock(&ofproto->vsp_mutex);
6160     vsp = vlandev_find(ofproto, port->up.ofp_port);
6161     if (vsp) {
6162         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6163         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6164         free(vsp);
6165
6166         port->realdev_ofp_port = 0;
6167     } else {
6168         VLOG_ERR("missing vlan device record");
6169     }
6170     ovs_mutex_unlock(&ofproto->vsp_mutex);
6171 }
6172
6173 static void
6174 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6175 {
6176     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6177
6178     ovs_mutex_lock(&ofproto->vsp_mutex);
6179     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6180         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6181             == realdev_ofp_port)) {
6182         struct vlan_splinter *vsp;
6183
6184         vsp = xmalloc(sizeof *vsp);
6185         vsp->realdev_ofp_port = realdev_ofp_port;
6186         vsp->vlandev_ofp_port = port->up.ofp_port;
6187         vsp->vid = vid;
6188
6189         port->realdev_ofp_port = realdev_ofp_port;
6190
6191         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6192                     hash_ofp_port(port->up.ofp_port));
6193         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6194                     hash_realdev_vid(realdev_ofp_port, vid));
6195     } else {
6196         VLOG_ERR("duplicate vlan device record");
6197     }
6198     ovs_mutex_unlock(&ofproto->vsp_mutex);
6199 }
6200
6201 static odp_port_t
6202 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6203 {
6204     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6205     return ofport ? ofport->odp_port : ODPP_NONE;
6206 }
6207
6208 struct ofport_dpif *
6209 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6210 {
6211     struct ofport_dpif *port;
6212
6213     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6214     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6215                              &backer->odp_to_ofport_map) {
6216         if (port->odp_port == odp_port) {
6217             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6218             return port;
6219         }
6220     }
6221
6222     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6223     return NULL;
6224 }
6225
6226 static ofp_port_t
6227 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6228 {
6229     struct ofport_dpif *port;
6230
6231     port = odp_port_to_ofport(ofproto->backer, odp_port);
6232     if (port && &ofproto->up == port->up.ofproto) {
6233         return port->up.ofp_port;
6234     } else {
6235         return OFPP_NONE;
6236     }
6237 }
6238
6239 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6240  * most heavily weighted element.  'base' designates the rate of decay: after
6241  * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6242  * (about .37). */
6243 static void
6244 exp_mavg(double *avg, int base, double new)
6245 {
6246     *avg = (*avg * (base - 1) + new) / base;
6247 }
6248
6249 static void
6250 update_moving_averages(struct dpif_backer *backer)
6251 {
6252     const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6253     long long int minutes = (time_msec() - backer->created) / min_ms;
6254
6255     if (minutes > 0) {
6256         backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6257             / minutes;
6258         backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6259             / minutes;
6260     } else {
6261         backer->lifetime.add_rate = 0.0;
6262         backer->lifetime.del_rate = 0.0;
6263     }
6264
6265     /* Update hourly averages on the minute boundaries. */
6266     if (time_msec() - backer->last_minute >= min_ms) {
6267         exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6268         exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6269
6270         /* Update daily averages on the hour boundaries. */
6271         if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6272             exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6273             exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6274         }
6275
6276         backer->total_subfacet_add_count += backer->subfacet_add_count;
6277         backer->total_subfacet_del_count += backer->subfacet_del_count;
6278         backer->subfacet_add_count = 0;
6279         backer->subfacet_del_count = 0;
6280         backer->last_minute += min_ms;
6281     }
6282 }
6283
6284 const struct ofproto_class ofproto_dpif_class = {
6285     init,
6286     enumerate_types,
6287     enumerate_names,
6288     del,
6289     port_open_type,
6290     type_run,
6291     type_run_fast,
6292     type_wait,
6293     alloc,
6294     construct,
6295     destruct,
6296     dealloc,
6297     run,
6298     run_fast,
6299     wait,
6300     get_memory_usage,
6301     flush,
6302     get_features,
6303     get_tables,
6304     port_alloc,
6305     port_construct,
6306     port_destruct,
6307     port_dealloc,
6308     port_modified,
6309     port_reconfigured,
6310     port_query_by_name,
6311     port_add,
6312     port_del,
6313     port_get_stats,
6314     port_dump_start,
6315     port_dump_next,
6316     port_dump_done,
6317     port_poll,
6318     port_poll_wait,
6319     port_is_lacp_current,
6320     NULL,                       /* rule_choose_table */
6321     rule_alloc,
6322     rule_construct,
6323     rule_insert,
6324     rule_delete,
6325     rule_destruct,
6326     rule_dealloc,
6327     rule_get_stats,
6328     rule_execute,
6329     rule_modify_actions,
6330     set_frag_handling,
6331     packet_out,
6332     set_netflow,
6333     get_netflow_ids,
6334     set_sflow,
6335     set_ipfix,
6336     set_cfm,
6337     get_cfm_status,
6338     set_bfd,
6339     get_bfd_status,
6340     set_stp,
6341     get_stp_status,
6342     set_stp_port,
6343     get_stp_port_status,
6344     set_queues,
6345     bundle_set,
6346     bundle_remove,
6347     mirror_set__,
6348     mirror_get_stats__,
6349     set_flood_vlans,
6350     is_mirror_output_bundle,
6351     forward_bpdu_changed,
6352     set_mac_table_config,
6353     set_realdev,
6354     NULL,                       /* meter_get_features */
6355     NULL,                       /* meter_set */
6356     NULL,                       /* meter_get */
6357     NULL,                       /* meter_del */
6358     NULL,                       /* group_alloc */
6359     NULL,                       /* group_construct */
6360     NULL,                       /* group_destruct */
6361     NULL,                       /* group_dealloc */
6362     NULL,                       /* group_modify */
6363     NULL,                       /* group_get_stats */
6364 };