ofproto/ofproto-dpif.c

   1 /*
   2  * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18
  19 #include "ofproto/ofproto-dpif.h"
  20 #include "ofproto/ofproto-provider.h"
  21
  22 #include <errno.h>
  23
  24 #include "bfd.h"
  25 #include "bond.h"
  26 #include "bundle.h"
  27 #include "byte-order.h"
  28 #include "connmgr.h"
  29 #include "coverage.h"
  30 #include "cfm.h"
  31 #include "dpif.h"
  32 #include "dynamic-string.h"
  33 #include "fail-open.h"
  34 #include "hmapx.h"
  35 #include "lacp.h"
  36 #include "learn.h"
  37 #include "mac-learning.h"
  38 #include "meta-flow.h"
  39 #include "multipath.h"
  40 #include "netdev-vport.h"
  41 #include "netdev.h"
  42 #include "netlink.h"
  43 #include "nx-match.h"
  44 #include "odp-util.h"
  45 #include "odp-execute.h"
  46 #include "ofp-util.h"
  47 #include "ofpbuf.h"
  48 #include "ofp-actions.h"
  49 #include "ofp-parse.h"
  50 #include "ofp-print.h"
  51 #include "ofproto-dpif-governor.h"
  52 #include "ofproto-dpif-ipfix.h"
  53 #include "ofproto-dpif-mirror.h"
  54 #include "ofproto-dpif-sflow.h"
  55 #include "ofproto-dpif-upcall.h"
  56 #include "ofproto-dpif-xlate.h"
  57 #include "poll-loop.h"
  58 #include "simap.h"
  59 #include "smap.h"
  60 #include "timer.h"
  61 #include "tunnel.h"
  62 #include "unaligned.h"
  63 #include "unixctl.h"
  64 #include "vlan-bitmap.h"
  65 #include "vlog.h"
  66
  67 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  68
  69 COVERAGE_DEFINE(ofproto_dpif_expired);
  70 COVERAGE_DEFINE(facet_changed_rule);
  71 COVERAGE_DEFINE(facet_revalidate);
  72 COVERAGE_DEFINE(facet_unexpected);
  73 COVERAGE_DEFINE(facet_suppress);
  74 COVERAGE_DEFINE(subfacet_install_fail);
  75 COVERAGE_DEFINE(packet_in_overflow);
  76 COVERAGE_DEFINE(flow_mod_overflow);
  77
  78 /* Number of implemented OpenFlow tables. */
  79 enum { N_TABLES = 255 };
  80 enum { TBL_INTERNAL = N_TABLES - 1 };    /* Used for internal hidden rules. */
  81 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
  82
  83 struct flow_miss;
  84 struct facet;
  85
  86 struct rule_dpif {
  87     struct rule up;
  88
  89     /* These statistics:
  90      *
  91      *   - Do include packets and bytes from facets that have been deleted or
  92      *     whose own statistics have been folded into the rule.
  93      *
  94      *   - Do include packets and bytes sent "by hand" that were accounted to
  95      *     the rule without any facet being involved (this is a rare corner
  96      *     case in rule_execute()).
  97      *
  98      *   - Do not include packet or bytes that can be obtained from any facet's
  99      *     packet_count or byte_count member or that can be obtained from the
 100      *     datapath by, e.g., dpif_flow_get() for any subfacet.
 101      */
 102     struct ovs_mutex stats_mutex;
 103     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 104     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 105 };
 106
 107 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
 108 static struct rule_dpif *rule_dpif_cast(const struct rule *);
 109
 110 struct ofbundle {
 111     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
 112     struct ofproto_dpif *ofproto; /* Owning ofproto. */
 113     void *aux;                  /* Key supplied by ofproto's client. */
 114     char *name;                 /* Identifier for log messages. */
 115
 116     /* Configuration. */
 117     struct list ports;          /* Contains "struct ofport"s. */
 118     enum port_vlan_mode vlan_mode; /* VLAN mode */
 119     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
 120     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
 121                                  * NULL if all VLANs are trunked. */
 122     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 123     struct bond *bond;          /* Nonnull iff more than one port. */
 124     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 125
 126     /* Status. */
 127     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 128 };
 129
 130 static void bundle_remove(struct ofport *);
 131 static void bundle_update(struct ofbundle *);
 132 static void bundle_destroy(struct ofbundle *);
 133 static void bundle_del_port(struct ofport_dpif *);
 134 static void bundle_run(struct ofbundle *);
 135 static void bundle_wait(struct ofbundle *);
 136
 137 static void stp_run(struct ofproto_dpif *ofproto);
 138 static void stp_wait(struct ofproto_dpif *ofproto);
 139 static int set_stp_port(struct ofport *,
 140                         const struct ofproto_port_stp_settings *);
 141
 142 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
 143                               enum slow_path_reason,
 144                               uint64_t *stub, size_t stub_size,
 145                               const struct nlattr **actionsp,
 146                               size_t *actions_lenp);
 147
 148 /* A subfacet (see "struct subfacet" below) has three possible installation
 149  * states:
 150  *
 151  *   - SF_NOT_INSTALLED: Not installed in the datapath.  This will only be the
 152  *     case just after the subfacet is created, just before the subfacet is
 153  *     destroyed, or if the datapath returns an error when we try to install a
 154  *     subfacet.
 155  *
 156  *   - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
 157  *
 158  *   - SF_SLOW_PATH: An action that sends every packet for the subfacet through
 159  *     ofproto_dpif is installed in the datapath.
 160  */
 161 enum subfacet_path {
 162     SF_NOT_INSTALLED,           /* No datapath flow for this subfacet. */
 163     SF_FAST_PATH,               /* Full actions are installed. */
 164     SF_SLOW_PATH,               /* Send-to-userspace action is installed. */
 165 };
 166
 167 /* A dpif flow and actions associated with a facet.
 168  *
 169  * See also the large comment on struct facet. */
 170 struct subfacet {
 171     /* Owners. */
 172     struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
 173     struct list list_node;      /* In struct facet's 'facets' list. */
 174     struct facet *facet;        /* Owning facet. */
 175     struct dpif_backer *backer; /* Owning backer. */
 176
 177     enum odp_key_fitness key_fitness;
 178     struct nlattr *key;
 179     int key_len;
 180
 181     long long int used;         /* Time last used; time created if not used. */
 182     long long int created;      /* Time created. */
 183
 184     uint64_t dp_packet_count;   /* Last known packet count in the datapath. */
 185     uint64_t dp_byte_count;     /* Last known byte count in the datapath. */
 186
 187     enum subfacet_path path;    /* Installed in datapath? */
 188 };
 189
 190 #define SUBFACET_DESTROY_MAX_BATCH 50
 191
 192 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *);
 193 static struct subfacet *subfacet_find(struct dpif_backer *,
 194                                       const struct nlattr *key, size_t key_len,
 195                                       uint32_t key_hash);
 196 static void subfacet_destroy(struct subfacet *);
 197 static void subfacet_destroy__(struct subfacet *);
 198 static void subfacet_destroy_batch(struct dpif_backer *,
 199                                    struct subfacet **, int n);
 200 static void subfacet_reset_dp_stats(struct subfacet *,
 201                                     struct dpif_flow_stats *);
 202 static void subfacet_update_stats(struct subfacet *,
 203                                   const struct dpif_flow_stats *);
 204 static int subfacet_install(struct subfacet *,
 205                             const struct ofpbuf *odp_actions,
 206                             struct dpif_flow_stats *);
 207 static void subfacet_uninstall(struct subfacet *);
 208
 209 /* A unique, non-overlapping instantiation of an OpenFlow flow.
 210  *
 211  * A facet associates a "struct flow", which represents the Open vSwitch
 212  * userspace idea of an exact-match flow, with one or more subfacets.
 213  * While the facet is created based on an exact-match flow, it is stored
 214  * within the ofproto based on the wildcards that could be expressed
 215  * based on the flow table and other configuration.  (See the 'wc'
 216  * description in "struct xlate_out" for more details.)
 217  *
 218  * Each subfacet tracks the datapath's idea of the flow equivalent to
 219  * the facet.  When the kernel module (or other dpif implementation) and
 220  * Open vSwitch userspace agree on the definition of a flow key, there
 221  * is exactly one subfacet per facet.  If the dpif implementation
 222  * supports more-specific flow matching than userspace, however, a facet
 223  * can have more than one subfacet.  Examples include the dpif
 224  * implementation not supporting the same wildcards as userspace or some
 225  * distinction in flow that userspace simply doesn't understand.
 226  *
 227  * Flow expiration works in terms of subfacets, so a facet must have at
 228  * least one subfacet or it will never expire, leaking memory. */
 229 struct facet {
 230     /* Owner. */
 231     struct ofproto_dpif *ofproto;
 232
 233     /* Owned data. */
 234     struct list subfacets;
 235     long long int used;         /* Time last used; time created if not used. */
 236
 237     /* Key. */
 238     struct flow flow;           /* Flow of the creating subfacet. */
 239     struct cls_rule cr;         /* In 'ofproto_dpif's facets classifier. */
 240
 241     /* These statistics:
 242      *
 243      *   - Do include packets and bytes sent "by hand", e.g. with
 244      *     dpif_execute().
 245      *
 246      *   - Do include packets and bytes that were obtained from the datapath
 247      *     when a subfacet's statistics were reset (e.g. dpif_flow_put() with
 248      *     DPIF_FP_ZERO_STATS).
 249      *
 250      *   - Do not include packets or bytes that can be obtained from the
 251      *     datapath for any existing subfacet.
 252      */
 253     uint64_t packet_count;       /* Number of packets received. */
 254     uint64_t byte_count;         /* Number of bytes received. */
 255
 256     /* Resubmit statistics. */
 257     uint64_t prev_packet_count;  /* Number of packets from last stats push. */
 258     uint64_t prev_byte_count;    /* Number of bytes from last stats push. */
 259     long long int prev_used;     /* Used time from last stats push. */
 260
 261     /* Accounting. */
 262     uint64_t accounted_bytes;    /* Bytes processed by facet_account(). */
 263     struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
 264     uint8_t tcp_flags;           /* TCP flags seen for this 'rule'. */
 265
 266     struct xlate_out xout;
 267
 268     /* Storage for a single subfacet, to reduce malloc() time and space
 269      * overhead.  (A facet always has at least one subfacet and in the common
 270      * case has exactly one subfacet.  However, 'one_subfacet' may not
 271      * always be valid, since it could have been removed after newer
 272      * subfacets were pushed onto the 'subfacets' list.) */
 273     struct subfacet one_subfacet;
 274
 275     long long int learn_rl;      /* Rate limiter for facet_learn(). */
 276 };
 277
 278 static struct facet *facet_create(const struct flow_miss *);
 279 static void facet_remove(struct facet *);
 280 static void facet_free(struct facet *);
 281
 282 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
 283 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
 284                                         const struct flow *);
 285 static bool facet_revalidate(struct facet *);
 286 static bool facet_check_consistency(struct facet *);
 287
 288 static void facet_flush_stats(struct facet *);
 289
 290 static void facet_reset_counters(struct facet *);
 291 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
 292                             struct dpif_flow_stats *, bool may_learn);
 293 static void facet_push_stats(struct facet *, bool may_learn);
 294 static void facet_learn(struct facet *);
 295 static void facet_account(struct facet *);
 296 static void push_all_stats(void);
 297
 298 static bool facet_is_controller_flow(struct facet *);
 299
 300 struct ofport_dpif {
 301     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 302     struct ofport up;
 303
 304     odp_port_t odp_port;
 305     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 306     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
 307     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 308     struct bfd *bfd;            /* BFD, if any. */
 309     bool may_enable;            /* May be enabled in bonds. */
 310     bool is_tunnel;             /* This port is a tunnel. */
 311     long long int carrier_seq;  /* Carrier status changes. */
 312     struct ofport_dpif *peer;   /* Peer if patch port. */
 313
 314     /* Spanning tree. */
 315     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 316     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 317     long long int stp_state_entered;
 318
 319     /* Queue to DSCP mapping. */
 320     struct ofproto_port_queue *qdscp;
 321     size_t n_qdscp;
 322
 323     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 324      *
 325      * This is deprecated.  It is only for compatibility with broken device
 326      * drivers in old versions of Linux that do not properly support VLANs when
 327      * VLAN devices are not used.  When broken device drivers are no longer in
 328      * widespread use, we will delete these interfaces. */
 329     ofp_port_t realdev_ofp_port;
 330     int vlandev_vid;
 331 };
 332
 333 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 334  *
 335  * This is deprecated.  It is only for compatibility with broken device drivers
 336  * in old versions of Linux that do not properly support VLANs when VLAN
 337  * devices are not used.  When broken device drivers are no longer in
 338  * widespread use, we will delete these interfaces. */
 339 struct vlan_splinter {
 340     struct hmap_node realdev_vid_node;
 341     struct hmap_node vlandev_node;
 342     ofp_port_t realdev_ofp_port;
 343     ofp_port_t vlandev_ofp_port;
 344     int vid;
 345 };
 346
 347 static void vsp_remove(struct ofport_dpif *);
 348 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 349
 350 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 351                                        ofp_port_t);
 352
 353 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 354                                        odp_port_t);
 355
 356 static struct ofport_dpif *
 357 ofport_dpif_cast(const struct ofport *ofport)
 358 {
 359     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 360 }
 361
 362 static void port_run(struct ofport_dpif *);
 363 static void port_run_fast(struct ofport_dpif *);
 364 static void port_wait(struct ofport_dpif *);
 365 static int set_bfd(struct ofport *, const struct smap *);
 366 static int set_cfm(struct ofport *, const struct cfm_settings *);
 367 static void ofport_update_peer(struct ofport_dpif *);
 368 static void run_fast_rl(void);
 369 static int run_fast(struct ofproto *);
 370
 371 struct dpif_completion {
 372     struct list list_node;
 373     struct ofoperation *op;
 374 };
 375
 376 /* Reasons that we might need to revalidate every facet, and corresponding
 377  * coverage counters.
 378  *
 379  * A value of 0 means that there is no need to revalidate.
 380  *
 381  * It would be nice to have some cleaner way to integrate with coverage
 382  * counters, but with only a few reasons I guess this is good enough for
 383  * now. */
 384 enum revalidate_reason {
 385     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 386     REV_STP,                   /* Spanning tree protocol port status change. */
 387     REV_BOND,                  /* Bonding changed. */
 388     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 389     REV_FLOW_TABLE,            /* Flow table changed. */
 390     REV_MAC_LEARNING,          /* Mac learning changed. */
 391     REV_INCONSISTENCY          /* Facet self-check failed. */
 392 };
 393 COVERAGE_DEFINE(rev_reconfigure);
 394 COVERAGE_DEFINE(rev_stp);
 395 COVERAGE_DEFINE(rev_bond);
 396 COVERAGE_DEFINE(rev_port_toggled);
 397 COVERAGE_DEFINE(rev_flow_table);
 398 COVERAGE_DEFINE(rev_mac_learning);
 399 COVERAGE_DEFINE(rev_inconsistency);
 400
 401 struct avg_subfacet_rates {
 402     double add_rate;   /* Moving average of new flows created per minute. */
 403     double del_rate;   /* Moving average of flows deleted per minute. */
 404 };
 405
 406 /* All datapaths of a given type share a single dpif backer instance. */
 407 struct dpif_backer {
 408     char *type;
 409     int refcount;
 410     struct dpif *dpif;
 411     struct udpif *udpif;
 412     struct timer next_expiration;
 413
 414     struct ovs_rwlock odp_to_ofport_lock;
 415     struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
 416
 417     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 418
 419     /* Facet revalidation flags applying to facets which use this backer. */
 420     enum revalidate_reason need_revalidate; /* Revalidate every facet. */
 421
 422     struct hmap drop_keys; /* Set of dropped odp keys. */
 423     bool recv_set_enable; /* Enables or disables receiving packets. */
 424
 425     struct hmap subfacets;
 426     struct governor *governor;
 427
 428     /* Subfacet statistics.
 429      *
 430      * These keep track of the total number of subfacets added and deleted and
 431      * flow life span.  They are useful for computing the flow rates stats
 432      * exposed via "ovs-appctl dpif/show".  The goal is to learn about
 433      * traffic patterns in ways that we can use later to improve Open vSwitch
 434      * performance in new situations.  */
 435     long long int created;           /* Time when it is created. */
 436     unsigned max_n_subfacet;         /* Maximum number of flows */
 437     unsigned avg_n_subfacet;         /* Average number of flows. */
 438     long long int avg_subfacet_life; /* Average life span of subfacets. */
 439
 440     /* The average number of subfacets... */
 441     struct avg_subfacet_rates hourly;   /* ...over the last hour. */
 442     struct avg_subfacet_rates daily;    /* ...over the last day. */
 443     struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
 444     long long int last_minute;          /* Last time 'hourly' was updated. */
 445
 446     /* Number of subfacets added or deleted since 'last_minute'. */
 447     unsigned subfacet_add_count;
 448     unsigned subfacet_del_count;
 449
 450     /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
 451     unsigned long long int total_subfacet_add_count;
 452     unsigned long long int total_subfacet_del_count;
 453
 454     /* Number of upcall handling threads. */
 455     unsigned int n_handler_threads;
 456 };
 457
 458 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 459 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 460
 461 static void drop_key_clear(struct dpif_backer *);
 462 static void update_moving_averages(struct dpif_backer *backer);
 463
 464 struct ofproto_dpif {
 465     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 466     struct ofproto up;
 467     struct dpif_backer *backer;
 468
 469     /* Special OpenFlow rules. */
 470     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 471     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 472     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 473
 474     /* Bridging. */
 475     struct netflow *netflow;
 476     struct dpif_sflow *sflow;
 477     struct dpif_ipfix *ipfix;
 478     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 479     struct mac_learning *ml;
 480     bool has_bonded_bundles;
 481     struct mbridge *mbridge;
 482
 483     /* Facets. */
 484     struct classifier facets;     /* Contains 'struct facet's. */
 485     long long int consistency_rl;
 486
 487     struct netdev_stats stats; /* To account packets generated and consumed in
 488                                 * userspace. */
 489
 490     /* Spanning tree. */
 491     struct stp *stp;
 492     long long int stp_last_tick;
 493
 494     /* VLAN splinters. */
 495     struct ovs_mutex vsp_mutex;
 496     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 497     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 498
 499     /* Ports. */
 500     struct sset ports;             /* Set of standard port names. */
 501     struct sset ghost_ports;       /* Ports with no datapath port. */
 502     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 503     int port_poll_errno;           /* Last errno for port_poll() reply. */
 504
 505     /* Per ofproto's dpif stats. */
 506     uint64_t n_hit;
 507     uint64_t n_missed;
 508
 509     /* Work queues. */
 510     struct ovs_mutex flow_mod_mutex;
 511     struct list flow_mods OVS_GUARDED;
 512     size_t n_flow_mods OVS_GUARDED;
 513
 514     struct ovs_mutex pin_mutex;
 515     struct list pins OVS_GUARDED;
 516     size_t n_pins OVS_GUARDED;
 517 };
 518
 519 /* By default, flows in the datapath are wildcarded (megaflows).  They
 520  * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
 521 static bool enable_megaflows = true;
 522
 523 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 524 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 525
 526 static void ofproto_dpif_unixctl_init(void);
 527
 528 static inline struct ofproto_dpif *
 529 ofproto_dpif_cast(const struct ofproto *ofproto)
 530 {
 531     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 532     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 533 }
 534
 535 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
 536                                         ofp_port_t ofp_port);
 537 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
 538                           const struct ofpbuf *packet, struct ds *);
 539
 540 /* Upcalls. */
 541 static void handle_upcalls(struct dpif_backer *);
 542
 543 /* Flow expiration. */
 544 static int expire(struct dpif_backer *);
 545
 546 /* NetFlow. */
 547 static void send_netflow_active_timeouts(struct ofproto_dpif *);
 548
 549 /* Utilities. */
 550 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
 551
 552 /* Global variables. */
 553 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 554
 555 /* Initial mappings of port to bridge mappings. */
 556 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 557
 558 /* Executes and takes ownership of 'fm'. */
 559 void
 560 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 561                       struct ofputil_flow_mod *fm)
 562 {
 563     ovs_mutex_lock(&ofproto->flow_mod_mutex);
 564     if (ofproto->n_flow_mods > 1024) {
 565         ovs_mutex_unlock(&ofproto->flow_mod_mutex);
 566         COVERAGE_INC(flow_mod_overflow);
 567         free(fm->ofpacts);
 568         free(fm);
 569         return;
 570     }
 571
 572     list_push_back(&ofproto->flow_mods, &fm->list_node);
 573     ofproto->n_flow_mods++;
 574     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
 575 }
 576
 577 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 578  * Takes ownership of 'pin' and pin->packet. */
 579 void
 580 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 581                             struct ofputil_packet_in *pin)
 582 {
 583     ovs_mutex_lock(&ofproto->pin_mutex);
 584     if (ofproto->n_pins > 1024) {
 585         ovs_mutex_unlock(&ofproto->pin_mutex);
 586         COVERAGE_INC(packet_in_overflow);
 587         free(CONST_CAST(void *, pin->packet));
 588         free(pin);
 589         return;
 590     }
 591
 592     list_push_back(&ofproto->pins, &pin->list_node);
 593     ofproto->n_pins++;
 594     ovs_mutex_unlock(&ofproto->pin_mutex);
 595 }
 596 \f
 597 /* Factory functions. */
 598
 599 static void
 600 init(const struct shash *iface_hints)
 601 {
 602     struct shash_node *node;
 603
 604     /* Make a local copy, since we don't own 'iface_hints' elements. */
 605     SHASH_FOR_EACH(node, iface_hints) {
 606         const struct iface_hint *orig_hint = node->data;
 607         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 608
 609         new_hint->br_name = xstrdup(orig_hint->br_name);
 610         new_hint->br_type = xstrdup(orig_hint->br_type);
 611         new_hint->ofp_port = orig_hint->ofp_port;
 612
 613         shash_add(&init_ofp_ports, node->name, new_hint);
 614     }
 615 }
 616
 617 static void
 618 enumerate_types(struct sset *types)
 619 {
 620     dp_enumerate_types(types);
 621 }
 622
 623 static int
 624 enumerate_names(const char *type, struct sset *names)
 625 {
 626     struct ofproto_dpif *ofproto;
 627
 628     sset_clear(names);
 629     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 630         if (strcmp(type, ofproto->up.type)) {
 631             continue;
 632         }
 633         sset_add(names, ofproto->up.name);
 634     }
 635
 636     return 0;
 637 }
 638
 639 static int
 640 del(const char *type, const char *name)
 641 {
 642     struct dpif *dpif;
 643     int error;
 644
 645     error = dpif_open(name, type, &dpif);
 646     if (!error) {
 647         error = dpif_delete(dpif);
 648         dpif_close(dpif);
 649     }
 650     return error;
 651 }
 652 \f
 653 static const char *
 654 port_open_type(const char *datapath_type, const char *port_type)
 655 {
 656     return dpif_port_open_type(datapath_type, port_type);
 657 }
 658
 659 /* Type functions. */
 660
 661 static void process_dpif_port_changes(struct dpif_backer *);
 662 static void process_dpif_all_ports_changed(struct dpif_backer *);
 663 static void process_dpif_port_change(struct dpif_backer *,
 664                                      const char *devname);
 665 static void process_dpif_port_error(struct dpif_backer *, int error);
 666
 667 static struct ofproto_dpif *
 668 lookup_ofproto_dpif_by_port_name(const char *name)
 669 {
 670     struct ofproto_dpif *ofproto;
 671
 672     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 673         if (sset_contains(&ofproto->ports, name)) {
 674             return ofproto;
 675         }
 676     }
 677
 678     return NULL;
 679 }
 680
 681 static int
 682 type_run(const char *type)
 683 {
 684     static long long int push_timer = LLONG_MIN;
 685     struct dpif_backer *backer;
 686
 687     backer = shash_find_data(&all_dpif_backers, type);
 688     if (!backer) {
 689         /* This is not necessarily a problem, since backers are only
 690          * created on demand. */
 691         return 0;
 692     }
 693
 694     dpif_run(backer->dpif);
 695
 696     /* The most natural place to push facet statistics is when they're pulled
 697      * from the datapath.  However, when there are many flows in the datapath,
 698      * this expensive operation can occur so frequently, that it reduces our
 699      * ability to quickly set up flows.  To reduce the cost, we push statistics
 700      * here instead. */
 701     if (time_msec() > push_timer) {
 702         push_timer = time_msec() + 2000;
 703         push_all_stats();
 704     }
 705
 706     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 707      * and the configuration has now changed to "false", enable receiving
 708      * packets from the datapath. */
 709     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 710         int error;
 711
 712         backer->recv_set_enable = true;
 713
 714         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 715         if (error) {
 716             udpif_recv_set(backer->udpif, 0, false);
 717             VLOG_ERR("Failed to enable receiving packets in dpif.");
 718             return error;
 719         }
 720         udpif_recv_set(backer->udpif, n_handler_threads,
 721                        backer->recv_set_enable);
 722         dpif_flow_flush(backer->dpif);
 723         backer->need_revalidate = REV_RECONFIGURE;
 724     }
 725
 726     /* If the n_handler_threads is reconfigured, call udpif_recv_set()
 727      * to reset the handler threads. */
 728     if (backer->n_handler_threads != n_handler_threads) {
 729         udpif_recv_set(backer->udpif, n_handler_threads,
 730                        backer->recv_set_enable);
 731         backer->n_handler_threads = n_handler_threads;
 732     }
 733
 734     if (backer->need_revalidate) {
 735         struct ofproto_dpif *ofproto;
 736         struct simap_node *node;
 737         struct simap tmp_backers;
 738
 739         /* Handle tunnel garbage collection. */
 740         simap_init(&tmp_backers);
 741         simap_swap(&backer->tnl_backers, &tmp_backers);
 742
 743         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 744             struct ofport_dpif *iter;
 745
 746             if (backer != ofproto->backer) {
 747                 continue;
 748             }
 749
 750             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 751                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 752                 const char *dp_port;
 753
 754                 if (!iter->is_tunnel) {
 755                     continue;
 756                 }
 757
 758                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 759                                                      namebuf, sizeof namebuf);
 760                 node = simap_find(&tmp_backers, dp_port);
 761                 if (node) {
 762                     simap_put(&backer->tnl_backers, dp_port, node->data);
 763                     simap_delete(&tmp_backers, node);
 764                     node = simap_find(&backer->tnl_backers, dp_port);
 765                 } else {
 766                     node = simap_find(&backer->tnl_backers, dp_port);
 767                     if (!node) {
 768                         odp_port_t odp_port = ODPP_NONE;
 769
 770                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 771                                            &odp_port)) {
 772                             simap_put(&backer->tnl_backers, dp_port,
 773                                       odp_to_u32(odp_port));
 774                             node = simap_find(&backer->tnl_backers, dp_port);
 775                         }
 776                     }
 777                 }
 778
 779                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 780                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 781                                          iter->odp_port)) {
 782                     backer->need_revalidate = REV_RECONFIGURE;
 783                 }
 784             }
 785         }
 786
 787         SIMAP_FOR_EACH (node, &tmp_backers) {
 788             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 789         }
 790         simap_destroy(&tmp_backers);
 791
 792         switch (backer->need_revalidate) {
 793         case REV_RECONFIGURE:   COVERAGE_INC(rev_reconfigure);   break;
 794         case REV_STP:           COVERAGE_INC(rev_stp);           break;
 795         case REV_BOND:          COVERAGE_INC(rev_bond);          break;
 796         case REV_PORT_TOGGLED:  COVERAGE_INC(rev_port_toggled);  break;
 797         case REV_FLOW_TABLE:    COVERAGE_INC(rev_flow_table);    break;
 798         case REV_MAC_LEARNING:  COVERAGE_INC(rev_mac_learning);  break;
 799         case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
 800         }
 801         backer->need_revalidate = 0;
 802
 803         /* Clear the drop_keys in case we should now be accepting some
 804          * formerly dropped flows. */
 805         drop_key_clear(backer);
 806
 807         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 808             struct facet *facet, *next;
 809             struct ofport_dpif *ofport;
 810             struct cls_cursor cursor;
 811             struct ofbundle *bundle;
 812
 813             if (ofproto->backer != backer) {
 814                 continue;
 815             }
 816
 817             ovs_rwlock_wrlock(&xlate_rwlock);
 818             xlate_ofproto_set(ofproto, ofproto->up.name,
 819                               ofproto->backer->dpif, ofproto->miss_rule,
 820                               ofproto->no_packet_in_rule, ofproto->ml,
 821                               ofproto->stp, ofproto->mbridge,
 822                               ofproto->sflow, ofproto->ipfix,
 823                               ofproto->up.frag_handling,
 824                               ofproto->up.forward_bpdu,
 825                               connmgr_has_in_band(ofproto->up.connmgr),
 826                               ofproto->netflow != NULL);
 827
 828             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 829                 xlate_bundle_set(ofproto, bundle, bundle->name,
 830                                  bundle->vlan_mode, bundle->vlan,
 831                                  bundle->trunks, bundle->use_priority_tags,
 832                                  bundle->bond, bundle->lacp,
 833                                  bundle->floodable);
 834             }
 835
 836             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 837                 int stp_port = ofport->stp_port
 838                     ? stp_port_no(ofport->stp_port)
 839                     : -1;
 840                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 841                                  ofport->up.ofp_port, ofport->odp_port,
 842                                  ofport->up.netdev, ofport->cfm,
 843                                  ofport->bfd, ofport->peer, stp_port,
 844                                  ofport->qdscp, ofport->n_qdscp,
 845                                  ofport->up.pp.config, ofport->is_tunnel,
 846                                  ofport->may_enable);
 847             }
 848             ovs_rwlock_unlock(&xlate_rwlock);
 849
 850             /* Only ofproto-dpif cares about the facet classifier so we just
 851              * lock cls_cursor_init() to appease the thread safety analysis. */
 852             ovs_rwlock_rdlock(&ofproto->facets.rwlock);
 853             cls_cursor_init(&cursor, &ofproto->facets, NULL);
 854             ovs_rwlock_unlock(&ofproto->facets.rwlock);
 855             CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
 856                 facet_revalidate(facet);
 857                 run_fast_rl();
 858             }
 859         }
 860
 861         udpif_revalidate(backer->udpif);
 862     }
 863
 864     if (!backer->recv_set_enable) {
 865         /* Wake up before a max of 1000ms. */
 866         timer_set_duration(&backer->next_expiration, 1000);
 867     } else if (timer_expired(&backer->next_expiration)) {
 868         int delay = expire(backer);
 869         timer_set_duration(&backer->next_expiration, delay);
 870     }
 871
 872     process_dpif_port_changes(backer);
 873
 874     if (backer->governor) {
 875         size_t n_subfacets;
 876
 877         governor_run(backer->governor);
 878
 879         /* If the governor has shrunk to its minimum size and the number of
 880          * subfacets has dwindled, then drop the governor entirely.
 881          *
 882          * For hysteresis, the number of subfacets to drop the governor is
 883          * smaller than the number needed to trigger its creation. */
 884         n_subfacets = hmap_count(&backer->subfacets);
 885         if (n_subfacets * 4 < flow_eviction_threshold
 886             && governor_is_idle(backer->governor)) {
 887             governor_destroy(backer->governor);
 888             backer->governor = NULL;
 889         }
 890     }
 891
 892     return 0;
 893 }
 894
 895 /* Check for and handle port changes in 'backer''s dpif. */
 896 static void
 897 process_dpif_port_changes(struct dpif_backer *backer)
 898 {
 899     for (;;) {
 900         char *devname;
 901         int error;
 902
 903         error = dpif_port_poll(backer->dpif, &devname);
 904         switch (error) {
 905         case EAGAIN:
 906             return;
 907
 908         case ENOBUFS:
 909             process_dpif_all_ports_changed(backer);
 910             break;
 911
 912         case 0:
 913             process_dpif_port_change(backer, devname);
 914             free(devname);
 915             break;
 916
 917         default:
 918             process_dpif_port_error(backer, error);
 919             break;
 920         }
 921     }
 922 }
 923
 924 static void
 925 process_dpif_all_ports_changed(struct dpif_backer *backer)
 926 {
 927     struct ofproto_dpif *ofproto;
 928     struct dpif_port dpif_port;
 929     struct dpif_port_dump dump;
 930     struct sset devnames;
 931     const char *devname;
 932
 933     sset_init(&devnames);
 934     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 935         if (ofproto->backer == backer) {
 936             struct ofport *ofport;
 937
 938             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 939                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 940             }
 941         }
 942     }
 943     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 944         sset_add(&devnames, dpif_port.name);
 945     }
 946
 947     SSET_FOR_EACH (devname, &devnames) {
 948         process_dpif_port_change(backer, devname);
 949     }
 950     sset_destroy(&devnames);
 951 }
 952
 953 static void
 954 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 955 {
 956     struct ofproto_dpif *ofproto;
 957     struct dpif_port port;
 958
 959     /* Don't report on the datapath's device. */
 960     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 961         return;
 962     }
 963
 964     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 965                    &all_ofproto_dpifs) {
 966         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 967             return;
 968         }
 969     }
 970
 971     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 972     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 973         /* The port was removed.  If we know the datapath,
 974          * report it through poll_set().  If we don't, it may be
 975          * notifying us of a removal we initiated, so ignore it.
 976          * If there's a pending ENOBUFS, let it stand, since
 977          * everything will be reevaluated. */
 978         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 979             sset_add(&ofproto->port_poll_set, devname);
 980             ofproto->port_poll_errno = 0;
 981         }
 982     } else if (!ofproto) {
 983         /* The port was added, but we don't know with which
 984          * ofproto we should associate it.  Delete it. */
 985         dpif_port_del(backer->dpif, port.port_no);
 986     } else {
 987         struct ofport_dpif *ofport;
 988
 989         ofport = ofport_dpif_cast(shash_find_data(
 990                                       &ofproto->up.port_by_name, devname));
 991         if (ofport
 992             && ofport->odp_port != port.port_no
 993             && !odp_port_to_ofport(backer, port.port_no))
 994         {
 995             /* 'ofport''s datapath port number has changed from
 996              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 997              * structures to match. */
 998             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 999             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
1000             ofport->odp_port = port.port_no;
1001             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
1002                         hash_odp_port(port.port_no));
1003             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
1004             backer->need_revalidate = REV_RECONFIGURE;
1005         }
1006     }
1007     dpif_port_destroy(&port);
1008 }
1009
1010 /* Propagate 'error' to all ofprotos based on 'backer'. */
1011 static void
1012 process_dpif_port_error(struct dpif_backer *backer, int error)
1013 {
1014     struct ofproto_dpif *ofproto;
1015
1016     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1017         if (ofproto->backer == backer) {
1018             sset_clear(&ofproto->port_poll_set);
1019             ofproto->port_poll_errno = error;
1020         }
1021     }
1022 }
1023
1024 static int
1025 dpif_backer_run_fast(struct dpif_backer *backer)
1026 {
1027     udpif_run(backer->udpif);
1028     handle_upcalls(backer);
1029
1030     return 0;
1031 }
1032
1033 static int
1034 type_run_fast(const char *type)
1035 {
1036     struct dpif_backer *backer;
1037
1038     backer = shash_find_data(&all_dpif_backers, type);
1039     if (!backer) {
1040         /* This is not necessarily a problem, since backers are only
1041          * created on demand. */
1042         return 0;
1043     }
1044
1045     return dpif_backer_run_fast(backer);
1046 }
1047
1048 static void
1049 run_fast_rl(void)
1050 {
1051     static long long int port_rl = LLONG_MIN;
1052
1053     if (time_msec() >= port_rl) {
1054         struct ofproto_dpif *ofproto;
1055
1056         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1057             run_fast(&ofproto->up);
1058         }
1059         port_rl = time_msec() + 200;
1060     }
1061 }
1062
1063 static void
1064 type_wait(const char *type)
1065 {
1066     struct dpif_backer *backer;
1067
1068     backer = shash_find_data(&all_dpif_backers, type);
1069     if (!backer) {
1070         /* This is not necessarily a problem, since backers are only
1071          * created on demand. */
1072         return;
1073     }
1074
1075     if (backer->governor) {
1076         governor_wait(backer->governor);
1077     }
1078
1079     timer_wait(&backer->next_expiration);
1080     dpif_wait(backer->dpif);
1081     udpif_wait(backer->udpif);
1082 }
1083 \f
1084 /* Basic life-cycle. */
1085
1086 static int add_internal_flows(struct ofproto_dpif *);
1087
1088 static struct ofproto *
1089 alloc(void)
1090 {
1091     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1092     return &ofproto->up;
1093 }
1094
1095 static void
1096 dealloc(struct ofproto *ofproto_)
1097 {
1098     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1099     free(ofproto);
1100 }
1101
1102 static void
1103 close_dpif_backer(struct dpif_backer *backer)
1104 {
1105     struct shash_node *node;
1106
1107     ovs_assert(backer->refcount > 0);
1108
1109     if (--backer->refcount) {
1110         return;
1111     }
1112
1113     drop_key_clear(backer);
1114     hmap_destroy(&backer->drop_keys);
1115
1116     simap_destroy(&backer->tnl_backers);
1117     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1118     hmap_destroy(&backer->odp_to_ofport_map);
1119     node = shash_find(&all_dpif_backers, backer->type);
1120     free(backer->type);
1121     shash_delete(&all_dpif_backers, node);
1122     udpif_destroy(backer->udpif);
1123     dpif_close(backer->dpif);
1124
1125     ovs_assert(hmap_is_empty(&backer->subfacets));
1126     hmap_destroy(&backer->subfacets);
1127     governor_destroy(backer->governor);
1128
1129     free(backer);
1130 }
1131
1132 /* Datapath port slated for removal from datapath. */
1133 struct odp_garbage {
1134     struct list list_node;
1135     odp_port_t odp_port;
1136 };
1137
1138 static int
1139 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1140 {
1141     struct dpif_backer *backer;
1142     struct dpif_port_dump port_dump;
1143     struct dpif_port port;
1144     struct shash_node *node;
1145     struct list garbage_list;
1146     struct odp_garbage *garbage, *next;
1147     struct sset names;
1148     char *backer_name;
1149     const char *name;
1150     int error;
1151
1152     backer = shash_find_data(&all_dpif_backers, type);
1153     if (backer) {
1154         backer->refcount++;
1155         *backerp = backer;
1156         return 0;
1157     }
1158
1159     backer_name = xasprintf("ovs-%s", type);
1160
1161     /* Remove any existing datapaths, since we assume we're the only
1162      * userspace controlling the datapath. */
1163     sset_init(&names);
1164     dp_enumerate_names(type, &names);
1165     SSET_FOR_EACH(name, &names) {
1166         struct dpif *old_dpif;
1167
1168         /* Don't remove our backer if it exists. */
1169         if (!strcmp(name, backer_name)) {
1170             continue;
1171         }
1172
1173         if (dpif_open(name, type, &old_dpif)) {
1174             VLOG_WARN("couldn't open old datapath %s to remove it", name);
1175         } else {
1176             dpif_delete(old_dpif);
1177             dpif_close(old_dpif);
1178         }
1179     }
1180     sset_destroy(&names);
1181
1182     backer = xmalloc(sizeof *backer);
1183
1184     error = dpif_create_and_open(backer_name, type, &backer->dpif);
1185     free(backer_name);
1186     if (error) {
1187         VLOG_ERR("failed to open datapath of type %s: %s", type,
1188                  ovs_strerror(error));
1189         free(backer);
1190         return error;
1191     }
1192     backer->udpif = udpif_create(backer, backer->dpif);
1193
1194     backer->type = xstrdup(type);
1195     backer->governor = NULL;
1196     backer->refcount = 1;
1197     hmap_init(&backer->odp_to_ofport_map);
1198     ovs_rwlock_init(&backer->odp_to_ofport_lock);
1199     hmap_init(&backer->drop_keys);
1200     hmap_init(&backer->subfacets);
1201     timer_set_duration(&backer->next_expiration, 1000);
1202     backer->need_revalidate = 0;
1203     simap_init(&backer->tnl_backers);
1204     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1205     *backerp = backer;
1206
1207     if (backer->recv_set_enable) {
1208         dpif_flow_flush(backer->dpif);
1209     }
1210
1211     /* Loop through the ports already on the datapath and remove any
1212      * that we don't need anymore. */
1213     list_init(&garbage_list);
1214     dpif_port_dump_start(&port_dump, backer->dpif);
1215     while (dpif_port_dump_next(&port_dump, &port)) {
1216         node = shash_find(&init_ofp_ports, port.name);
1217         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1218             garbage = xmalloc(sizeof *garbage);
1219             garbage->odp_port = port.port_no;
1220             list_push_front(&garbage_list, &garbage->list_node);
1221         }
1222     }
1223     dpif_port_dump_done(&port_dump);
1224
1225     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1226         dpif_port_del(backer->dpif, garbage->odp_port);
1227         list_remove(&garbage->list_node);
1228         free(garbage);
1229     }
1230
1231     shash_add(&all_dpif_backers, type, backer);
1232
1233     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1234     if (error) {
1235         VLOG_ERR("failed to listen on datapath of type %s: %s",
1236                  type, ovs_strerror(error));
1237         close_dpif_backer(backer);
1238         return error;
1239     }
1240     udpif_recv_set(backer->udpif, n_handler_threads,
1241                    backer->recv_set_enable);
1242     backer->n_handler_threads = n_handler_threads;
1243
1244     backer->max_n_subfacet = 0;
1245     backer->created = time_msec();
1246     backer->last_minute = backer->created;
1247     memset(&backer->hourly, 0, sizeof backer->hourly);
1248     memset(&backer->daily, 0, sizeof backer->daily);
1249     memset(&backer->lifetime, 0, sizeof backer->lifetime);
1250     backer->subfacet_add_count = 0;
1251     backer->subfacet_del_count = 0;
1252     backer->total_subfacet_add_count = 0;
1253     backer->total_subfacet_del_count = 0;
1254     backer->avg_n_subfacet = 0;
1255     backer->avg_subfacet_life = 0;
1256
1257     return error;
1258 }
1259
1260 static int
1261 construct(struct ofproto *ofproto_)
1262 {
1263     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1264     struct shash_node *node, *next;
1265     uint32_t max_ports;
1266     int error;
1267
1268     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1269     if (error) {
1270         return error;
1271     }
1272
1273     max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1274     ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX)));
1275
1276     ofproto->netflow = NULL;
1277     ofproto->sflow = NULL;
1278     ofproto->ipfix = NULL;
1279     ofproto->stp = NULL;
1280     hmap_init(&ofproto->bundles);
1281     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1282     ofproto->mbridge = mbridge_create();
1283     ofproto->has_bonded_bundles = false;
1284     ovs_mutex_init(&ofproto->vsp_mutex);
1285
1286     classifier_init(&ofproto->facets);
1287     ofproto->consistency_rl = LLONG_MIN;
1288
1289     ovs_mutex_init(&ofproto->flow_mod_mutex);
1290     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1291     list_init(&ofproto->flow_mods);
1292     ofproto->n_flow_mods = 0;
1293     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1294
1295     ovs_mutex_init(&ofproto->pin_mutex);
1296     ovs_mutex_lock(&ofproto->pin_mutex);
1297     list_init(&ofproto->pins);
1298     ofproto->n_pins = 0;
1299     ovs_mutex_unlock(&ofproto->pin_mutex);
1300
1301     ofproto_dpif_unixctl_init();
1302
1303     hmap_init(&ofproto->vlandev_map);
1304     hmap_init(&ofproto->realdev_vid_map);
1305
1306     sset_init(&ofproto->ports);
1307     sset_init(&ofproto->ghost_ports);
1308     sset_init(&ofproto->port_poll_set);
1309     ofproto->port_poll_errno = 0;
1310
1311     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1312         struct iface_hint *iface_hint = node->data;
1313
1314         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1315             /* Check if the datapath already has this port. */
1316             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1317                 sset_add(&ofproto->ports, node->name);
1318             }
1319
1320             free(iface_hint->br_name);
1321             free(iface_hint->br_type);
1322             free(iface_hint);
1323             shash_delete(&init_ofp_ports, node);
1324         }
1325     }
1326
1327     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1328                 hash_string(ofproto->up.name, 0));
1329     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1330
1331     ofproto_init_tables(ofproto_, N_TABLES);
1332     error = add_internal_flows(ofproto);
1333     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1334
1335     ofproto->n_hit = 0;
1336     ofproto->n_missed = 0;
1337
1338     return error;
1339 }
1340
1341 static int
1342 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1343                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1344 {
1345     struct ofputil_flow_mod fm;
1346     int error;
1347
1348     match_init_catchall(&fm.match);
1349     fm.priority = 0;
1350     match_set_reg(&fm.match, 0, id);
1351     fm.new_cookie = htonll(0);
1352     fm.cookie = htonll(0);
1353     fm.cookie_mask = htonll(0);
1354     fm.modify_cookie = false;
1355     fm.table_id = TBL_INTERNAL;
1356     fm.command = OFPFC_ADD;
1357     fm.idle_timeout = 0;
1358     fm.hard_timeout = 0;
1359     fm.buffer_id = 0;
1360     fm.out_port = 0;
1361     fm.flags = 0;
1362     fm.ofpacts = ofpacts->data;
1363     fm.ofpacts_len = ofpacts->size;
1364
1365     error = ofproto_flow_mod(&ofproto->up, &fm);
1366     if (error) {
1367         VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1368                     id, ofperr_to_string(error));
1369         return error;
1370     }
1371
1372     if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1373                                   rulep)) {
1374         rule_dpif_release(*rulep);
1375     } else {
1376         NOT_REACHED();
1377     }
1378
1379     return 0;
1380 }
1381
1382 static int
1383 add_internal_flows(struct ofproto_dpif *ofproto)
1384 {
1385     struct ofpact_controller *controller;
1386     uint64_t ofpacts_stub[128 / 8];
1387     struct ofpbuf ofpacts;
1388     int error;
1389     int id;
1390
1391     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1392     id = 1;
1393
1394     controller = ofpact_put_CONTROLLER(&ofpacts);
1395     controller->max_len = UINT16_MAX;
1396     controller->controller_id = 0;
1397     controller->reason = OFPR_NO_MATCH;
1398     ofpact_pad(&ofpacts);
1399
1400     error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1401     if (error) {
1402         return error;
1403     }
1404
1405     ofpbuf_clear(&ofpacts);
1406     error = add_internal_flow(ofproto, id++, &ofpacts,
1407                               &ofproto->no_packet_in_rule);
1408     if (error) {
1409         return error;
1410     }
1411
1412     error = add_internal_flow(ofproto, id++, &ofpacts,
1413                               &ofproto->drop_frags_rule);
1414     return error;
1415 }
1416
1417 static void
1418 destruct(struct ofproto *ofproto_)
1419 {
1420     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1421     struct rule_dpif *rule, *next_rule;
1422     struct ofputil_packet_in *pin, *next_pin;
1423     struct ofputil_flow_mod *fm, *next_fm;
1424     struct facet *facet, *next_facet;
1425     struct cls_cursor cursor;
1426     struct oftable *table;
1427
1428     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1429     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1430     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1431     CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1432         facet_remove(facet);
1433     }
1434
1435     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1436     ovs_rwlock_wrlock(&xlate_rwlock);
1437     xlate_remove_ofproto(ofproto);
1438     ovs_rwlock_unlock(&xlate_rwlock);
1439
1440     flow_miss_batch_ofproto_destroyed(ofproto->backer->udpif, ofproto);
1441
1442     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1443
1444     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1445         struct cls_cursor cursor;
1446
1447         ovs_rwlock_wrlock(&table->cls.rwlock);
1448         cls_cursor_init(&cursor, &table->cls, NULL);
1449         CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1450             ofproto_rule_delete(&ofproto->up, &table->cls, &rule->up);
1451         }
1452         ovs_rwlock_unlock(&table->cls.rwlock);
1453     }
1454
1455     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1456     LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &ofproto->flow_mods) {
1457         list_remove(&fm->list_node);
1458         ofproto->n_flow_mods--;
1459         free(fm->ofpacts);
1460         free(fm);
1461     }
1462     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1463     ovs_mutex_destroy(&ofproto->flow_mod_mutex);
1464
1465     ovs_mutex_lock(&ofproto->pin_mutex);
1466     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &ofproto->pins) {
1467         list_remove(&pin->list_node);
1468         ofproto->n_pins--;
1469         free(CONST_CAST(void *, pin->packet));
1470         free(pin);
1471     }
1472     ovs_mutex_unlock(&ofproto->pin_mutex);
1473     ovs_mutex_destroy(&ofproto->pin_mutex);
1474
1475     mbridge_unref(ofproto->mbridge);
1476
1477     netflow_destroy(ofproto->netflow);
1478     dpif_sflow_unref(ofproto->sflow);
1479     hmap_destroy(&ofproto->bundles);
1480     mac_learning_unref(ofproto->ml);
1481
1482     classifier_destroy(&ofproto->facets);
1483
1484     hmap_destroy(&ofproto->vlandev_map);
1485     hmap_destroy(&ofproto->realdev_vid_map);
1486
1487     sset_destroy(&ofproto->ports);
1488     sset_destroy(&ofproto->ghost_ports);
1489     sset_destroy(&ofproto->port_poll_set);
1490
1491     ovs_mutex_destroy(&ofproto->vsp_mutex);
1492
1493     close_dpif_backer(ofproto->backer);
1494 }
1495
1496 static int
1497 run_fast(struct ofproto *ofproto_)
1498 {
1499     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1500     struct ofputil_packet_in *pin, *next_pin;
1501     struct ofputil_flow_mod *fm, *next_fm;
1502     struct list flow_mods, pins;
1503     struct ofport_dpif *ofport;
1504
1505     /* Do not perform any periodic activity required by 'ofproto' while
1506      * waiting for flow restore to complete. */
1507     if (ofproto_get_flow_restore_wait()) {
1508         return 0;
1509     }
1510
1511     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1512     list_move(&flow_mods, &ofproto->flow_mods);
1513     list_init(&ofproto->flow_mods);
1514     ofproto->n_flow_mods = 0;
1515     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1516
1517     LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &flow_mods) {
1518         int error = ofproto_flow_mod(&ofproto->up, fm);
1519         if (error && !VLOG_DROP_WARN(&rl)) {
1520             VLOG_WARN("learning action failed to modify flow table (%s)",
1521                       ofperr_get_name(error));
1522         }
1523
1524         list_remove(&fm->list_node);
1525         free(fm->ofpacts);
1526         free(fm);
1527     }
1528
1529     ovs_mutex_lock(&ofproto->pin_mutex);
1530     list_move(&pins, &ofproto->pins);
1531     list_init(&ofproto->pins);
1532     ofproto->n_pins = 0;
1533     ovs_mutex_unlock(&ofproto->pin_mutex);
1534
1535     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1536         connmgr_send_packet_in(ofproto->up.connmgr, pin);
1537         list_remove(&pin->list_node);
1538         free(CONST_CAST(void *, pin->packet));
1539         free(pin);
1540     }
1541
1542     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1543         port_run_fast(ofport);
1544     }
1545
1546     return 0;
1547 }
1548
1549 static int
1550 run(struct ofproto *ofproto_)
1551 {
1552     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1553     struct ofport_dpif *ofport;
1554     struct ofbundle *bundle;
1555     int error;
1556
1557     if (mbridge_need_revalidate(ofproto->mbridge)) {
1558         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1559         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1560         mac_learning_flush(ofproto->ml);
1561         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1562     }
1563
1564     /* Do not perform any periodic activity below required by 'ofproto' while
1565      * waiting for flow restore to complete. */
1566     if (ofproto_get_flow_restore_wait()) {
1567         return 0;
1568     }
1569
1570     error = run_fast(ofproto_);
1571     if (error) {
1572         return error;
1573     }
1574
1575     if (ofproto->netflow) {
1576         if (netflow_run(ofproto->netflow)) {
1577             send_netflow_active_timeouts(ofproto);
1578         }
1579     }
1580     if (ofproto->sflow) {
1581         dpif_sflow_run(ofproto->sflow);
1582     }
1583     if (ofproto->ipfix) {
1584         dpif_ipfix_run(ofproto->ipfix);
1585     }
1586
1587     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1588         port_run(ofport);
1589     }
1590     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1591         bundle_run(bundle);
1592     }
1593
1594     stp_run(ofproto);
1595     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1596     if (mac_learning_run(ofproto->ml)) {
1597         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1598     }
1599     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1600
1601     /* Check the consistency of a random facet, to aid debugging. */
1602     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1603     if (time_msec() >= ofproto->consistency_rl
1604         && !classifier_is_empty(&ofproto->facets)
1605         && !ofproto->backer->need_revalidate) {
1606         struct cls_table *table;
1607         struct cls_rule *cr;
1608         struct facet *facet;
1609
1610         ofproto->consistency_rl = time_msec() + 250;
1611
1612         table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1613                              struct cls_table, hmap_node);
1614         cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1615                           hmap_node);
1616         facet = CONTAINER_OF(cr, struct facet, cr);
1617
1618         if (!facet_check_consistency(facet)) {
1619             ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1620         }
1621     }
1622     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1623
1624     return 0;
1625 }
1626
1627 static void
1628 wait(struct ofproto *ofproto_)
1629 {
1630     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1631     struct ofport_dpif *ofport;
1632     struct ofbundle *bundle;
1633
1634     if (ofproto_get_flow_restore_wait()) {
1635         return;
1636     }
1637
1638     if (ofproto->sflow) {
1639         dpif_sflow_wait(ofproto->sflow);
1640     }
1641     if (ofproto->ipfix) {
1642         dpif_ipfix_wait(ofproto->ipfix);
1643     }
1644     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1645         port_wait(ofport);
1646     }
1647     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1648         bundle_wait(bundle);
1649     }
1650     if (ofproto->netflow) {
1651         netflow_wait(ofproto->netflow);
1652     }
1653     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1654     mac_learning_wait(ofproto->ml);
1655     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1656     stp_wait(ofproto);
1657     if (ofproto->backer->need_revalidate) {
1658         /* Shouldn't happen, but if it does just go around again. */
1659         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1660         poll_immediate_wake();
1661     }
1662 }
1663
1664 static void
1665 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1666 {
1667     const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1668     struct cls_cursor cursor;
1669     size_t n_subfacets = 0;
1670     struct facet *facet;
1671
1672     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1673     simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1674     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1675
1676     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1677     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1678     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1679         n_subfacets += list_size(&facet->subfacets);
1680     }
1681     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1682     simap_increase(usage, "subfacets", n_subfacets);
1683 }
1684
1685 static void
1686 flush(struct ofproto *ofproto_)
1687 {
1688     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1689     struct subfacet *subfacet, *next_subfacet;
1690     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1691     int n_batch;
1692
1693     n_batch = 0;
1694     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1695                         &ofproto->backer->subfacets) {
1696         if (subfacet->facet->ofproto != ofproto) {
1697             continue;
1698         }
1699
1700         if (subfacet->path != SF_NOT_INSTALLED) {
1701             batch[n_batch++] = subfacet;
1702             if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1703                 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1704                 n_batch = 0;
1705             }
1706         } else {
1707             subfacet_destroy(subfacet);
1708         }
1709     }
1710
1711     if (n_batch > 0) {
1712         subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1713     }
1714 }
1715
1716 static void
1717 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1718              bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1719 {
1720     *arp_match_ip = true;
1721     *actions = (OFPUTIL_A_OUTPUT |
1722                 OFPUTIL_A_SET_VLAN_VID |
1723                 OFPUTIL_A_SET_VLAN_PCP |
1724                 OFPUTIL_A_STRIP_VLAN |
1725                 OFPUTIL_A_SET_DL_SRC |
1726                 OFPUTIL_A_SET_DL_DST |
1727                 OFPUTIL_A_SET_NW_SRC |
1728                 OFPUTIL_A_SET_NW_DST |
1729                 OFPUTIL_A_SET_NW_TOS |
1730                 OFPUTIL_A_SET_TP_SRC |
1731                 OFPUTIL_A_SET_TP_DST |
1732                 OFPUTIL_A_ENQUEUE);
1733 }
1734
1735 static void
1736 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1737 {
1738     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1739     struct dpif_dp_stats s;
1740     uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1741     uint64_t n_lookup;
1742
1743     strcpy(ots->name, "classifier");
1744
1745     dpif_get_dp_stats(ofproto->backer->dpif, &s);
1746     rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1747     rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1748     rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1749
1750     n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1751     ots->lookup_count = htonll(n_lookup);
1752     ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1753 }
1754
1755 static struct ofport *
1756 port_alloc(void)
1757 {
1758     struct ofport_dpif *port = xmalloc(sizeof *port);
1759     return &port->up;
1760 }
1761
1762 static void
1763 port_dealloc(struct ofport *port_)
1764 {
1765     struct ofport_dpif *port = ofport_dpif_cast(port_);
1766     free(port);
1767 }
1768
1769 static int
1770 port_construct(struct ofport *port_)
1771 {
1772     struct ofport_dpif *port = ofport_dpif_cast(port_);
1773     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1774     const struct netdev *netdev = port->up.netdev;
1775     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1776     struct dpif_port dpif_port;
1777     int error;
1778
1779     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1780     port->bundle = NULL;
1781     port->cfm = NULL;
1782     port->bfd = NULL;
1783     port->may_enable = true;
1784     port->stp_port = NULL;
1785     port->stp_state = STP_DISABLED;
1786     port->is_tunnel = false;
1787     port->peer = NULL;
1788     port->qdscp = NULL;
1789     port->n_qdscp = 0;
1790     port->realdev_ofp_port = 0;
1791     port->vlandev_vid = 0;
1792     port->carrier_seq = netdev_get_carrier_resets(netdev);
1793
1794     if (netdev_vport_is_patch(netdev)) {
1795         /* By bailing out here, we don't submit the port to the sFlow module
1796          * to be considered for counter polling export.  This is correct
1797          * because the patch port represents an interface that sFlow considers
1798          * to be "internal" to the switch as a whole, and therefore not an
1799          * candidate for counter polling. */
1800         port->odp_port = ODPP_NONE;
1801         ofport_update_peer(port);
1802         return 0;
1803     }
1804
1805     error = dpif_port_query_by_name(ofproto->backer->dpif,
1806                                     netdev_vport_get_dpif_port(netdev, namebuf,
1807                                                                sizeof namebuf),
1808                                     &dpif_port);
1809     if (error) {
1810         return error;
1811     }
1812
1813     port->odp_port = dpif_port.port_no;
1814
1815     if (netdev_get_tunnel_config(netdev)) {
1816         tnl_port_add(port, port->up.netdev, port->odp_port);
1817         port->is_tunnel = true;
1818     } else {
1819         /* Sanity-check that a mapping doesn't already exist.  This
1820          * shouldn't happen for non-tunnel ports. */
1821         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1822             VLOG_ERR("port %s already has an OpenFlow port number",
1823                      dpif_port.name);
1824             dpif_port_destroy(&dpif_port);
1825             return EBUSY;
1826         }
1827
1828         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1829         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1830                     hash_odp_port(port->odp_port));
1831         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1832     }
1833     dpif_port_destroy(&dpif_port);
1834
1835     if (ofproto->sflow) {
1836         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1837     }
1838
1839     return 0;
1840 }
1841
1842 static void
1843 port_destruct(struct ofport *port_)
1844 {
1845     struct ofport_dpif *port = ofport_dpif_cast(port_);
1846     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1847     const char *devname = netdev_get_name(port->up.netdev);
1848     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1849     const char *dp_port_name;
1850
1851     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1852     ovs_rwlock_wrlock(&xlate_rwlock);
1853     xlate_ofport_remove(port);
1854     ovs_rwlock_unlock(&xlate_rwlock);
1855
1856     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1857                                               sizeof namebuf);
1858     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1859         /* The underlying device is still there, so delete it.  This
1860          * happens when the ofproto is being destroyed, since the caller
1861          * assumes that removal of attached ports will happen as part of
1862          * destruction. */
1863         if (!port->is_tunnel) {
1864             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1865         }
1866     }
1867
1868     if (port->peer) {
1869         port->peer->peer = NULL;
1870         port->peer = NULL;
1871     }
1872
1873     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1874         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1875         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1876         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1877     }
1878
1879     tnl_port_del(port);
1880     sset_find_and_delete(&ofproto->ports, devname);
1881     sset_find_and_delete(&ofproto->ghost_ports, devname);
1882     bundle_remove(port_);
1883     set_cfm(port_, NULL);
1884     set_bfd(port_, NULL);
1885     if (ofproto->sflow) {
1886         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1887     }
1888
1889     free(port->qdscp);
1890 }
1891
1892 static void
1893 port_modified(struct ofport *port_)
1894 {
1895     struct ofport_dpif *port = ofport_dpif_cast(port_);
1896
1897     if (port->bundle && port->bundle->bond) {
1898         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1899     }
1900
1901     if (port->cfm) {
1902         cfm_set_netdev(port->cfm, port->up.netdev);
1903     }
1904
1905     if (port->bfd) {
1906         bfd_set_netdev(port->bfd, port->up.netdev);
1907     }
1908
1909     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1910                                                 port->odp_port)) {
1911         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1912             REV_RECONFIGURE;
1913     }
1914
1915     ofport_update_peer(port);
1916 }
1917
1918 static void
1919 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1920 {
1921     struct ofport_dpif *port = ofport_dpif_cast(port_);
1922     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1923     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1924
1925     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1926                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1927                    OFPUTIL_PC_NO_PACKET_IN)) {
1928         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1929
1930         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1931             bundle_update(port->bundle);
1932         }
1933     }
1934 }
1935
1936 static int
1937 set_sflow(struct ofproto *ofproto_,
1938           const struct ofproto_sflow_options *sflow_options)
1939 {
1940     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1941     struct dpif_sflow *ds = ofproto->sflow;
1942
1943     if (sflow_options) {
1944         if (!ds) {
1945             struct ofport_dpif *ofport;
1946
1947             ds = ofproto->sflow = dpif_sflow_create();
1948             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1949                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1950             }
1951             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1952         }
1953         dpif_sflow_set_options(ds, sflow_options);
1954     } else {
1955         if (ds) {
1956             dpif_sflow_unref(ds);
1957             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1958             ofproto->sflow = NULL;
1959         }
1960     }
1961     return 0;
1962 }
1963
1964 static int
1965 set_ipfix(
1966     struct ofproto *ofproto_,
1967     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1968     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1969     size_t n_flow_exporters_options)
1970 {
1971     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1972     struct dpif_ipfix *di = ofproto->ipfix;
1973     bool has_options = bridge_exporter_options || flow_exporters_options;
1974
1975     if (has_options && !di) {
1976         di = ofproto->ipfix = dpif_ipfix_create();
1977     }
1978
1979     if (di) {
1980         /* Call set_options in any case to cleanly flush the flow
1981          * caches in the last exporters that are to be destroyed. */
1982         dpif_ipfix_set_options(
1983             di, bridge_exporter_options, flow_exporters_options,
1984             n_flow_exporters_options);
1985
1986         if (!has_options) {
1987             dpif_ipfix_unref(di);
1988             ofproto->ipfix = NULL;
1989         }
1990     }
1991
1992     return 0;
1993 }
1994
1995 static int
1996 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1997 {
1998     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1999     int error;
2000
2001     if (!s) {
2002         error = 0;
2003     } else {
2004         if (!ofport->cfm) {
2005             struct ofproto_dpif *ofproto;
2006
2007             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2008             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2009             ofport->cfm = cfm_create(ofport->up.netdev);
2010         }
2011
2012         if (cfm_configure(ofport->cfm, s)) {
2013             return 0;
2014         }
2015
2016         error = EINVAL;
2017     }
2018     cfm_unref(ofport->cfm);
2019     ofport->cfm = NULL;
2020     return error;
2021 }
2022
2023 static bool
2024 get_cfm_status(const struct ofport *ofport_,
2025                struct ofproto_cfm_status *status)
2026 {
2027     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2028
2029     if (ofport->cfm) {
2030         status->faults = cfm_get_fault(ofport->cfm);
2031         status->remote_opstate = cfm_get_opup(ofport->cfm);
2032         status->health = cfm_get_health(ofport->cfm);
2033         cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2034         return true;
2035     } else {
2036         return false;
2037     }
2038 }
2039
2040 static int
2041 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2042 {
2043     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2044     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2045     struct bfd *old;
2046
2047     old = ofport->bfd;
2048     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2049                                 cfg, ofport->up.netdev);
2050     if (ofport->bfd != old) {
2051         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2052     }
2053
2054     return 0;
2055 }
2056
2057 static int
2058 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2059 {
2060     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2061
2062     if (ofport->bfd) {
2063         bfd_get_status(ofport->bfd, smap);
2064         return 0;
2065     } else {
2066         return ENOENT;
2067     }
2068 }
2069 \f
2070 /* Spanning Tree. */
2071
2072 static void
2073 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2074 {
2075     struct ofproto_dpif *ofproto = ofproto_;
2076     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2077     struct ofport_dpif *ofport;
2078
2079     ofport = stp_port_get_aux(sp);
2080     if (!ofport) {
2081         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2082                      ofproto->up.name, port_num);
2083     } else {
2084         struct eth_header *eth = pkt->l2;
2085
2086         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2087         if (eth_addr_is_zero(eth->eth_src)) {
2088             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2089                          "with unknown MAC", ofproto->up.name, port_num);
2090         } else {
2091             send_packet(ofport, pkt);
2092         }
2093     }
2094     ofpbuf_delete(pkt);
2095 }
2096
2097 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2098 static int
2099 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2100 {
2101     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2102
2103     /* Only revalidate flows if the configuration changed. */
2104     if (!s != !ofproto->stp) {
2105         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2106     }
2107
2108     if (s) {
2109         if (!ofproto->stp) {
2110             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2111                                       send_bpdu_cb, ofproto);
2112             ofproto->stp_last_tick = time_msec();
2113         }
2114
2115         stp_set_bridge_id(ofproto->stp, s->system_id);
2116         stp_set_bridge_priority(ofproto->stp, s->priority);
2117         stp_set_hello_time(ofproto->stp, s->hello_time);
2118         stp_set_max_age(ofproto->stp, s->max_age);
2119         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2120     }  else {
2121         struct ofport *ofport;
2122
2123         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2124             set_stp_port(ofport, NULL);
2125         }
2126
2127         stp_unref(ofproto->stp);
2128         ofproto->stp = NULL;
2129     }
2130
2131     return 0;
2132 }
2133
2134 static int
2135 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2136 {
2137     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2138
2139     if (ofproto->stp) {
2140         s->enabled = true;
2141         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2142         s->designated_root = stp_get_designated_root(ofproto->stp);
2143         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2144     } else {
2145         s->enabled = false;
2146     }
2147
2148     return 0;
2149 }
2150
2151 static void
2152 update_stp_port_state(struct ofport_dpif *ofport)
2153 {
2154     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2155     enum stp_state state;
2156
2157     /* Figure out new state. */
2158     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2159                              : STP_DISABLED;
2160
2161     /* Update state. */
2162     if (ofport->stp_state != state) {
2163         enum ofputil_port_state of_state;
2164         bool fwd_change;
2165
2166         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2167                     netdev_get_name(ofport->up.netdev),
2168                     stp_state_name(ofport->stp_state),
2169                     stp_state_name(state));
2170         if (stp_learn_in_state(ofport->stp_state)
2171                 != stp_learn_in_state(state)) {
2172             /* xxx Learning action flows should also be flushed. */
2173             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2174             mac_learning_flush(ofproto->ml);
2175             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2176         }
2177         fwd_change = stp_forward_in_state(ofport->stp_state)
2178                         != stp_forward_in_state(state);
2179
2180         ofproto->backer->need_revalidate = REV_STP;
2181         ofport->stp_state = state;
2182         ofport->stp_state_entered = time_msec();
2183
2184         if (fwd_change && ofport->bundle) {
2185             bundle_update(ofport->bundle);
2186         }
2187
2188         /* Update the STP state bits in the OpenFlow port description. */
2189         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2190         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2191                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2192                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2193                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2194                      : 0);
2195         ofproto_port_set_state(&ofport->up, of_state);
2196     }
2197 }
2198
2199 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2200  * caller is responsible for assigning STP port numbers and ensuring
2201  * there are no duplicates. */
2202 static int
2203 set_stp_port(struct ofport *ofport_,
2204              const struct ofproto_port_stp_settings *s)
2205 {
2206     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2207     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2208     struct stp_port *sp = ofport->stp_port;
2209
2210     if (!s || !s->enable) {
2211         if (sp) {
2212             ofport->stp_port = NULL;
2213             stp_port_disable(sp);
2214             update_stp_port_state(ofport);
2215         }
2216         return 0;
2217     } else if (sp && stp_port_no(sp) != s->port_num
2218             && ofport == stp_port_get_aux(sp)) {
2219         /* The port-id changed, so disable the old one if it's not
2220          * already in use by another port. */
2221         stp_port_disable(sp);
2222     }
2223
2224     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2225     stp_port_enable(sp);
2226
2227     stp_port_set_aux(sp, ofport);
2228     stp_port_set_priority(sp, s->priority);
2229     stp_port_set_path_cost(sp, s->path_cost);
2230
2231     update_stp_port_state(ofport);
2232
2233     return 0;
2234 }
2235
2236 static int
2237 get_stp_port_status(struct ofport *ofport_,
2238                     struct ofproto_port_stp_status *s)
2239 {
2240     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2241     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2242     struct stp_port *sp = ofport->stp_port;
2243
2244     if (!ofproto->stp || !sp) {
2245         s->enabled = false;
2246         return 0;
2247     }
2248
2249     s->enabled = true;
2250     s->port_id = stp_port_get_id(sp);
2251     s->state = stp_port_get_state(sp);
2252     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2253     s->role = stp_port_get_role(sp);
2254     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2255
2256     return 0;
2257 }
2258
2259 static void
2260 stp_run(struct ofproto_dpif *ofproto)
2261 {
2262     if (ofproto->stp) {
2263         long long int now = time_msec();
2264         long long int elapsed = now - ofproto->stp_last_tick;
2265         struct stp_port *sp;
2266
2267         if (elapsed > 0) {
2268             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2269             ofproto->stp_last_tick = now;
2270         }
2271         while (stp_get_changed_port(ofproto->stp, &sp)) {
2272             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2273
2274             if (ofport) {
2275                 update_stp_port_state(ofport);
2276             }
2277         }
2278
2279         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2280             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2281             mac_learning_flush(ofproto->ml);
2282             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2283         }
2284     }
2285 }
2286
2287 static void
2288 stp_wait(struct ofproto_dpif *ofproto)
2289 {
2290     if (ofproto->stp) {
2291         poll_timer_wait(1000);
2292     }
2293 }
2294 \f
2295 static int
2296 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2297            size_t n_qdscp)
2298 {
2299     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2300     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2301
2302     if (ofport->n_qdscp != n_qdscp
2303         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2304                               n_qdscp * sizeof *qdscp))) {
2305         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2306         free(ofport->qdscp);
2307         ofport->qdscp = n_qdscp
2308             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2309             : NULL;
2310         ofport->n_qdscp = n_qdscp;
2311     }
2312
2313     return 0;
2314 }
2315 \f
2316 /* Bundles. */
2317
2318 /* Expires all MAC learning entries associated with 'bundle' and forces its
2319  * ofproto to revalidate every flow.
2320  *
2321  * Normally MAC learning entries are removed only from the ofproto associated
2322  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2323  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2324  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2325  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2326  * with the host from which it migrated. */
2327 static void
2328 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2329 {
2330     struct ofproto_dpif *ofproto = bundle->ofproto;
2331     struct mac_learning *ml = ofproto->ml;
2332     struct mac_entry *mac, *next_mac;
2333
2334     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2335     ovs_rwlock_wrlock(&ml->rwlock);
2336     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2337         if (mac->port.p == bundle) {
2338             if (all_ofprotos) {
2339                 struct ofproto_dpif *o;
2340
2341                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2342                     if (o != ofproto) {
2343                         struct mac_entry *e;
2344
2345                         ovs_rwlock_wrlock(&o->ml->rwlock);
2346                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2347                         if (e) {
2348                             mac_learning_expire(o->ml, e);
2349                         }
2350                         ovs_rwlock_unlock(&o->ml->rwlock);
2351                     }
2352                 }
2353             }
2354
2355             mac_learning_expire(ml, mac);
2356         }
2357     }
2358     ovs_rwlock_unlock(&ml->rwlock);
2359 }
2360
2361 static struct ofbundle *
2362 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2363 {
2364     struct ofbundle *bundle;
2365
2366     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2367                              &ofproto->bundles) {
2368         if (bundle->aux == aux) {
2369             return bundle;
2370         }
2371     }
2372     return NULL;
2373 }
2374
2375 static void
2376 bundle_update(struct ofbundle *bundle)
2377 {
2378     struct ofport_dpif *port;
2379
2380     bundle->floodable = true;
2381     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2382         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2383             || !stp_forward_in_state(port->stp_state)) {
2384             bundle->floodable = false;
2385             break;
2386         }
2387     }
2388 }
2389
2390 static void
2391 bundle_del_port(struct ofport_dpif *port)
2392 {
2393     struct ofbundle *bundle = port->bundle;
2394
2395     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2396
2397     list_remove(&port->bundle_node);
2398     port->bundle = NULL;
2399
2400     if (bundle->lacp) {
2401         lacp_slave_unregister(bundle->lacp, port);
2402     }
2403     if (bundle->bond) {
2404         bond_slave_unregister(bundle->bond, port);
2405     }
2406
2407     bundle_update(bundle);
2408 }
2409
2410 static bool
2411 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2412                 struct lacp_slave_settings *lacp)
2413 {
2414     struct ofport_dpif *port;
2415
2416     port = get_ofp_port(bundle->ofproto, ofp_port);
2417     if (!port) {
2418         return false;
2419     }
2420
2421     if (port->bundle != bundle) {
2422         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2423         if (port->bundle) {
2424             bundle_remove(&port->up);
2425         }
2426
2427         port->bundle = bundle;
2428         list_push_back(&bundle->ports, &port->bundle_node);
2429         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2430             || !stp_forward_in_state(port->stp_state)) {
2431             bundle->floodable = false;
2432         }
2433     }
2434     if (lacp) {
2435         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2436         lacp_slave_register(bundle->lacp, port, lacp);
2437     }
2438
2439     return true;
2440 }
2441
2442 static void
2443 bundle_destroy(struct ofbundle *bundle)
2444 {
2445     struct ofproto_dpif *ofproto;
2446     struct ofport_dpif *port, *next_port;
2447
2448     if (!bundle) {
2449         return;
2450     }
2451
2452     ofproto = bundle->ofproto;
2453     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2454
2455     ovs_rwlock_wrlock(&xlate_rwlock);
2456     xlate_bundle_remove(bundle);
2457     ovs_rwlock_unlock(&xlate_rwlock);
2458
2459     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2460         bundle_del_port(port);
2461     }
2462
2463     bundle_flush_macs(bundle, true);
2464     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2465     free(bundle->name);
2466     free(bundle->trunks);
2467     lacp_unref(bundle->lacp);
2468     bond_unref(bundle->bond);
2469     free(bundle);
2470 }
2471
2472 static int
2473 bundle_set(struct ofproto *ofproto_, void *aux,
2474            const struct ofproto_bundle_settings *s)
2475 {
2476     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2477     bool need_flush = false;
2478     struct ofport_dpif *port;
2479     struct ofbundle *bundle;
2480     unsigned long *trunks;
2481     int vlan;
2482     size_t i;
2483     bool ok;
2484
2485     if (!s) {
2486         bundle_destroy(bundle_lookup(ofproto, aux));
2487         return 0;
2488     }
2489
2490     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2491     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2492
2493     bundle = bundle_lookup(ofproto, aux);
2494     if (!bundle) {
2495         bundle = xmalloc(sizeof *bundle);
2496
2497         bundle->ofproto = ofproto;
2498         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2499                     hash_pointer(aux, 0));
2500         bundle->aux = aux;
2501         bundle->name = NULL;
2502
2503         list_init(&bundle->ports);
2504         bundle->vlan_mode = PORT_VLAN_TRUNK;
2505         bundle->vlan = -1;
2506         bundle->trunks = NULL;
2507         bundle->use_priority_tags = s->use_priority_tags;
2508         bundle->lacp = NULL;
2509         bundle->bond = NULL;
2510
2511         bundle->floodable = true;
2512         mbridge_register_bundle(ofproto->mbridge, bundle);
2513     }
2514
2515     if (!bundle->name || strcmp(s->name, bundle->name)) {
2516         free(bundle->name);
2517         bundle->name = xstrdup(s->name);
2518     }
2519
2520     /* LACP. */
2521     if (s->lacp) {
2522         if (!bundle->lacp) {
2523             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2524             bundle->lacp = lacp_create();
2525         }
2526         lacp_configure(bundle->lacp, s->lacp);
2527     } else {
2528         lacp_unref(bundle->lacp);
2529         bundle->lacp = NULL;
2530     }
2531
2532     /* Update set of ports. */
2533     ok = true;
2534     for (i = 0; i < s->n_slaves; i++) {
2535         if (!bundle_add_port(bundle, s->slaves[i],
2536                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2537             ok = false;
2538         }
2539     }
2540     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2541         struct ofport_dpif *next_port;
2542
2543         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2544             for (i = 0; i < s->n_slaves; i++) {
2545                 if (s->slaves[i] == port->up.ofp_port) {
2546                     goto found;
2547                 }
2548             }
2549
2550             bundle_del_port(port);
2551         found: ;
2552         }
2553     }
2554     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2555
2556     if (list_is_empty(&bundle->ports)) {
2557         bundle_destroy(bundle);
2558         return EINVAL;
2559     }
2560
2561     /* Set VLAN tagging mode */
2562     if (s->vlan_mode != bundle->vlan_mode
2563         || s->use_priority_tags != bundle->use_priority_tags) {
2564         bundle->vlan_mode = s->vlan_mode;
2565         bundle->use_priority_tags = s->use_priority_tags;
2566         need_flush = true;
2567     }
2568
2569     /* Set VLAN tag. */
2570     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2571             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2572             : 0);
2573     if (vlan != bundle->vlan) {
2574         bundle->vlan = vlan;
2575         need_flush = true;
2576     }
2577
2578     /* Get trunked VLANs. */
2579     switch (s->vlan_mode) {
2580     case PORT_VLAN_ACCESS:
2581         trunks = NULL;
2582         break;
2583
2584     case PORT_VLAN_TRUNK:
2585         trunks = CONST_CAST(unsigned long *, s->trunks);
2586         break;
2587
2588     case PORT_VLAN_NATIVE_UNTAGGED:
2589     case PORT_VLAN_NATIVE_TAGGED:
2590         if (vlan != 0 && (!s->trunks
2591                           || !bitmap_is_set(s->trunks, vlan)
2592                           || bitmap_is_set(s->trunks, 0))) {
2593             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2594             if (s->trunks) {
2595                 trunks = bitmap_clone(s->trunks, 4096);
2596             } else {
2597                 trunks = bitmap_allocate1(4096);
2598             }
2599             bitmap_set1(trunks, vlan);
2600             bitmap_set0(trunks, 0);
2601         } else {
2602             trunks = CONST_CAST(unsigned long *, s->trunks);
2603         }
2604         break;
2605
2606     default:
2607         NOT_REACHED();
2608     }
2609     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2610         free(bundle->trunks);
2611         if (trunks == s->trunks) {
2612             bundle->trunks = vlan_bitmap_clone(trunks);
2613         } else {
2614             bundle->trunks = trunks;
2615             trunks = NULL;
2616         }
2617         need_flush = true;
2618     }
2619     if (trunks != s->trunks) {
2620         free(trunks);
2621     }
2622
2623     /* Bonding. */
2624     if (!list_is_short(&bundle->ports)) {
2625         bundle->ofproto->has_bonded_bundles = true;
2626         if (bundle->bond) {
2627             if (bond_reconfigure(bundle->bond, s->bond)) {
2628                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2629             }
2630         } else {
2631             bundle->bond = bond_create(s->bond);
2632             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2633         }
2634
2635         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2636             bond_slave_register(bundle->bond, port, port->up.netdev);
2637         }
2638     } else {
2639         bond_unref(bundle->bond);
2640         bundle->bond = NULL;
2641     }
2642
2643     /* If we changed something that would affect MAC learning, un-learn
2644      * everything on this port and force flow revalidation. */
2645     if (need_flush) {
2646         bundle_flush_macs(bundle, false);
2647     }
2648
2649     return 0;
2650 }
2651
2652 static void
2653 bundle_remove(struct ofport *port_)
2654 {
2655     struct ofport_dpif *port = ofport_dpif_cast(port_);
2656     struct ofbundle *bundle = port->bundle;
2657
2658     if (bundle) {
2659         bundle_del_port(port);
2660         if (list_is_empty(&bundle->ports)) {
2661             bundle_destroy(bundle);
2662         } else if (list_is_short(&bundle->ports)) {
2663             bond_unref(bundle->bond);
2664             bundle->bond = NULL;
2665         }
2666     }
2667 }
2668
2669 static void
2670 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2671 {
2672     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2673     struct ofport_dpif *port = port_;
2674     uint8_t ea[ETH_ADDR_LEN];
2675     int error;
2676
2677     error = netdev_get_etheraddr(port->up.netdev, ea);
2678     if (!error) {
2679         struct ofpbuf packet;
2680         void *packet_pdu;
2681
2682         ofpbuf_init(&packet, 0);
2683         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2684                                  pdu_size);
2685         memcpy(packet_pdu, pdu, pdu_size);
2686
2687         send_packet(port, &packet);
2688         ofpbuf_uninit(&packet);
2689     } else {
2690         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2691                     "%s (%s)", port->bundle->name,
2692                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2693     }
2694 }
2695
2696 static void
2697 bundle_send_learning_packets(struct ofbundle *bundle)
2698 {
2699     struct ofproto_dpif *ofproto = bundle->ofproto;
2700     struct ofpbuf *learning_packet;
2701     int error, n_packets, n_errors;
2702     struct mac_entry *e;
2703     struct list packets;
2704
2705     list_init(&packets);
2706     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2707     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2708         if (e->port.p != bundle) {
2709             void *port_void;
2710
2711             learning_packet = bond_compose_learning_packet(bundle->bond,
2712                                                            e->mac, e->vlan,
2713                                                            &port_void);
2714             learning_packet->private_p = port_void;
2715             list_push_back(&packets, &learning_packet->list_node);
2716         }
2717     }
2718     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2719
2720     error = n_packets = n_errors = 0;
2721     LIST_FOR_EACH (learning_packet, list_node, &packets) {
2722         int ret;
2723
2724         ret = send_packet(learning_packet->private_p, learning_packet);
2725         if (ret) {
2726             error = ret;
2727             n_errors++;
2728         }
2729         n_packets++;
2730     }
2731     ofpbuf_list_delete(&packets);
2732
2733     if (n_errors) {
2734         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2735         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2736                      "packets, last error was: %s",
2737                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2738     } else {
2739         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2740                  bundle->name, n_packets);
2741     }
2742 }
2743
2744 static void
2745 bundle_run(struct ofbundle *bundle)
2746 {
2747     if (bundle->lacp) {
2748         lacp_run(bundle->lacp, send_pdu_cb);
2749     }
2750     if (bundle->bond) {
2751         struct ofport_dpif *port;
2752
2753         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2754             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2755         }
2756
2757         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2758             bundle->ofproto->backer->need_revalidate = REV_BOND;
2759         }
2760
2761         if (bond_should_send_learning_packets(bundle->bond)) {
2762             bundle_send_learning_packets(bundle);
2763         }
2764     }
2765 }
2766
2767 static void
2768 bundle_wait(struct ofbundle *bundle)
2769 {
2770     if (bundle->lacp) {
2771         lacp_wait(bundle->lacp);
2772     }
2773     if (bundle->bond) {
2774         bond_wait(bundle->bond);
2775     }
2776 }
2777 \f
2778 /* Mirrors. */
2779
2780 static int
2781 mirror_set__(struct ofproto *ofproto_, void *aux,
2782              const struct ofproto_mirror_settings *s)
2783 {
2784     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2785     struct ofbundle **srcs, **dsts;
2786     int error;
2787     size_t i;
2788
2789     if (!s) {
2790         mirror_destroy(ofproto->mbridge, aux);
2791         return 0;
2792     }
2793
2794     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2795     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2796
2797     for (i = 0; i < s->n_srcs; i++) {
2798         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2799     }
2800
2801     for (i = 0; i < s->n_dsts; i++) {
2802         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2803     }
2804
2805     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2806                        s->n_dsts, s->src_vlans,
2807                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2808     free(srcs);
2809     free(dsts);
2810     return error;
2811 }
2812
2813 static int
2814 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2815                    uint64_t *packets, uint64_t *bytes)
2816 {
2817     push_all_stats();
2818     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2819                             bytes);
2820 }
2821
2822 static int
2823 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2824 {
2825     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2826     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2827     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2828         mac_learning_flush(ofproto->ml);
2829     }
2830     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2831     return 0;
2832 }
2833
2834 static bool
2835 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2836 {
2837     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2838     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2839     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2840 }
2841
2842 static void
2843 forward_bpdu_changed(struct ofproto *ofproto_)
2844 {
2845     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2846     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2847 }
2848
2849 static void
2850 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2851                      size_t max_entries)
2852 {
2853     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2854     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2855     mac_learning_set_idle_time(ofproto->ml, idle_time);
2856     mac_learning_set_max_entries(ofproto->ml, max_entries);
2857     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2858 }
2859 \f
2860 /* Ports. */
2861
2862 static struct ofport_dpif *
2863 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2864 {
2865     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2866     return ofport ? ofport_dpif_cast(ofport) : NULL;
2867 }
2868
2869 static struct ofport_dpif *
2870 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2871 {
2872     struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2873     return port && &ofproto->up == port->up.ofproto ? port : NULL;
2874 }
2875
2876 static void
2877 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2878                             struct ofproto_port *ofproto_port,
2879                             struct dpif_port *dpif_port)
2880 {
2881     ofproto_port->name = dpif_port->name;
2882     ofproto_port->type = dpif_port->type;
2883     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2884 }
2885
2886 static void
2887 ofport_update_peer(struct ofport_dpif *ofport)
2888 {
2889     const struct ofproto_dpif *ofproto;
2890     struct dpif_backer *backer;
2891     char *peer_name;
2892
2893     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2894         return;
2895     }
2896
2897     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2898     backer->need_revalidate = REV_RECONFIGURE;
2899
2900     if (ofport->peer) {
2901         ofport->peer->peer = NULL;
2902         ofport->peer = NULL;
2903     }
2904
2905     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2906     if (!peer_name) {
2907         return;
2908     }
2909
2910     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2911         struct ofport *peer_ofport;
2912         struct ofport_dpif *peer;
2913         char *peer_peer;
2914
2915         if (ofproto->backer != backer) {
2916             continue;
2917         }
2918
2919         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2920         if (!peer_ofport) {
2921             continue;
2922         }
2923
2924         peer = ofport_dpif_cast(peer_ofport);
2925         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2926         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2927                                  peer_peer)) {
2928             ofport->peer = peer;
2929             ofport->peer->peer = ofport;
2930         }
2931         free(peer_peer);
2932
2933         break;
2934     }
2935     free(peer_name);
2936 }
2937
2938 static void
2939 port_run_fast(struct ofport_dpif *ofport)
2940 {
2941     if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2942         struct ofpbuf packet;
2943
2944         ofpbuf_init(&packet, 0);
2945         cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2946         send_packet(ofport, &packet);
2947         ofpbuf_uninit(&packet);
2948     }
2949
2950     if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2951         struct ofpbuf packet;
2952
2953         ofpbuf_init(&packet, 0);
2954         bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2955         send_packet(ofport, &packet);
2956         ofpbuf_uninit(&packet);
2957     }
2958 }
2959
2960 static void
2961 port_run(struct ofport_dpif *ofport)
2962 {
2963     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2964     bool carrier_changed = carrier_seq != ofport->carrier_seq;
2965     bool enable = netdev_get_carrier(ofport->up.netdev);
2966     bool cfm_enable = false;
2967     bool bfd_enable = false;
2968
2969     ofport->carrier_seq = carrier_seq;
2970
2971     port_run_fast(ofport);
2972
2973     if (ofport->cfm) {
2974         int cfm_opup = cfm_get_opup(ofport->cfm);
2975
2976         cfm_run(ofport->cfm);
2977         cfm_enable = !cfm_get_fault(ofport->cfm);
2978
2979         if (cfm_opup >= 0) {
2980             cfm_enable = cfm_enable && cfm_opup;
2981         }
2982     }
2983
2984     if (ofport->bfd) {
2985         bfd_run(ofport->bfd);
2986         bfd_enable = bfd_forwarding(ofport->bfd);
2987     }
2988
2989     if (ofport->bfd || ofport->cfm) {
2990         enable = enable && (cfm_enable || bfd_enable);
2991     }
2992
2993     if (ofport->bundle) {
2994         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2995         if (carrier_changed) {
2996             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2997         }
2998     }
2999
3000     if (ofport->may_enable != enable) {
3001         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3002         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3003     }
3004
3005     ofport->may_enable = enable;
3006 }
3007
3008 static void
3009 port_wait(struct ofport_dpif *ofport)
3010 {
3011     if (ofport->cfm) {
3012         cfm_wait(ofport->cfm);
3013     }
3014
3015     if (ofport->bfd) {
3016         bfd_wait(ofport->bfd);
3017     }
3018 }
3019
3020 static int
3021 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3022                    struct ofproto_port *ofproto_port)
3023 {
3024     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3025     struct dpif_port dpif_port;
3026     int error;
3027
3028     if (sset_contains(&ofproto->ghost_ports, devname)) {
3029         const char *type = netdev_get_type_from_name(devname);
3030
3031         /* We may be called before ofproto->up.port_by_name is populated with
3032          * the appropriate ofport.  For this reason, we must get the name and
3033          * type from the netdev layer directly. */
3034         if (type) {
3035             const struct ofport *ofport;
3036
3037             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3038             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3039             ofproto_port->name = xstrdup(devname);
3040             ofproto_port->type = xstrdup(type);
3041             return 0;
3042         }
3043         return ENODEV;
3044     }
3045
3046     if (!sset_contains(&ofproto->ports, devname)) {
3047         return ENODEV;
3048     }
3049     error = dpif_port_query_by_name(ofproto->backer->dpif,
3050                                     devname, &dpif_port);
3051     if (!error) {
3052         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3053     }
3054     return error;
3055 }
3056
3057 static int
3058 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3059 {
3060     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3061     const char *devname = netdev_get_name(netdev);
3062     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3063     const char *dp_port_name;
3064
3065     if (netdev_vport_is_patch(netdev)) {
3066         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3067         return 0;
3068     }
3069
3070     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3071     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3072         odp_port_t port_no = ODPP_NONE;
3073         int error;
3074
3075         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3076         if (error) {
3077             return error;
3078         }
3079         if (netdev_get_tunnel_config(netdev)) {
3080             simap_put(&ofproto->backer->tnl_backers,
3081                       dp_port_name, odp_to_u32(port_no));
3082         }
3083     }
3084
3085     if (netdev_get_tunnel_config(netdev)) {
3086         sset_add(&ofproto->ghost_ports, devname);
3087     } else {
3088         sset_add(&ofproto->ports, devname);
3089     }
3090     return 0;
3091 }
3092
3093 static int
3094 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3095 {
3096     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3097     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3098     int error = 0;
3099
3100     if (!ofport) {
3101         return 0;
3102     }
3103
3104     sset_find_and_delete(&ofproto->ghost_ports,
3105                          netdev_get_name(ofport->up.netdev));
3106     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3107     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3108         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3109         if (!error) {
3110             /* The caller is going to close ofport->up.netdev.  If this is a
3111              * bonded port, then the bond is using that netdev, so remove it
3112              * from the bond.  The client will need to reconfigure everything
3113              * after deleting ports, so then the slave will get re-added. */
3114             bundle_remove(&ofport->up);
3115         }
3116     }
3117     return error;
3118 }
3119
3120 static int
3121 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3122 {
3123     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3124     int error;
3125
3126     push_all_stats();
3127
3128     error = netdev_get_stats(ofport->up.netdev, stats);
3129
3130     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3131         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3132
3133         /* ofproto->stats.tx_packets represents packets that we created
3134          * internally and sent to some port (e.g. packets sent with
3135          * send_packet()).  Account for them as if they had come from
3136          * OFPP_LOCAL and got forwarded. */
3137
3138         if (stats->rx_packets != UINT64_MAX) {
3139             stats->rx_packets += ofproto->stats.tx_packets;
3140         }
3141
3142         if (stats->rx_bytes != UINT64_MAX) {
3143             stats->rx_bytes += ofproto->stats.tx_bytes;
3144         }
3145
3146         /* ofproto->stats.rx_packets represents packets that were received on
3147          * some port and we processed internally and dropped (e.g. STP).
3148          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3149
3150         if (stats->tx_packets != UINT64_MAX) {
3151             stats->tx_packets += ofproto->stats.rx_packets;
3152         }
3153
3154         if (stats->tx_bytes != UINT64_MAX) {
3155             stats->tx_bytes += ofproto->stats.rx_bytes;
3156         }
3157     }
3158
3159     return error;
3160 }
3161
3162 struct port_dump_state {
3163     uint32_t bucket;
3164     uint32_t offset;
3165     bool ghost;
3166
3167     struct ofproto_port port;
3168     bool has_port;
3169 };
3170
3171 static int
3172 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3173 {
3174     *statep = xzalloc(sizeof(struct port_dump_state));
3175     return 0;
3176 }
3177
3178 static int
3179 port_dump_next(const struct ofproto *ofproto_, void *state_,
3180                struct ofproto_port *port)
3181 {
3182     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3183     struct port_dump_state *state = state_;
3184     const struct sset *sset;
3185     struct sset_node *node;
3186
3187     if (state->has_port) {
3188         ofproto_port_destroy(&state->port);
3189         state->has_port = false;
3190     }
3191     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3192     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3193         int error;
3194
3195         error = port_query_by_name(ofproto_, node->name, &state->port);
3196         if (!error) {
3197             *port = state->port;
3198             state->has_port = true;
3199             return 0;
3200         } else if (error != ENODEV) {
3201             return error;
3202         }
3203     }
3204
3205     if (!state->ghost) {
3206         state->ghost = true;
3207         state->bucket = 0;
3208         state->offset = 0;
3209         return port_dump_next(ofproto_, state_, port);
3210     }
3211
3212     return EOF;
3213 }
3214
3215 static int
3216 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3217 {
3218     struct port_dump_state *state = state_;
3219
3220     if (state->has_port) {
3221         ofproto_port_destroy(&state->port);
3222     }
3223     free(state);
3224     return 0;
3225 }
3226
3227 static int
3228 port_poll(const struct ofproto *ofproto_, char **devnamep)
3229 {
3230     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3231
3232     if (ofproto->port_poll_errno) {
3233         int error = ofproto->port_poll_errno;
3234         ofproto->port_poll_errno = 0;
3235         return error;
3236     }
3237
3238     if (sset_is_empty(&ofproto->port_poll_set)) {
3239         return EAGAIN;
3240     }
3241
3242     *devnamep = sset_pop(&ofproto->port_poll_set);
3243     return 0;
3244 }
3245
3246 static void
3247 port_poll_wait(const struct ofproto *ofproto_)
3248 {
3249     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3250     dpif_port_poll_wait(ofproto->backer->dpif);
3251 }
3252
3253 static int
3254 port_is_lacp_current(const struct ofport *ofport_)
3255 {
3256     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3257     return (ofport->bundle && ofport->bundle->lacp
3258             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3259             : -1);
3260 }
3261 \f
3262 /* Upcall handling. */
3263
3264 struct flow_miss_op {
3265     struct dpif_op dpif_op;
3266
3267     uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3268     struct xlate_out xout;
3269     bool xout_garbage;           /* 'xout' needs to be uninitialized? */
3270
3271     struct ofpbuf mask;          /* Flow mask for "put" ops. */
3272     struct odputil_keybuf maskbuf;
3273
3274     /* If this is a "put" op, then a pointer to the subfacet that should
3275      * be marked as uninstalled if the operation fails. */
3276     struct subfacet *subfacet;
3277 };
3278
3279 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3280  * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3281  * and (usually) installing a datapath flow.  The answer is usually "yes" (a
3282  * return value of true).  However, for short flows the cost of bookkeeping is
3283  * much higher than the benefits, so when the datapath holds a large number of
3284  * flows we impose some heuristics to decide which flows are likely to be worth
3285  * tracking. */
3286 static bool
3287 flow_miss_should_make_facet(struct flow_miss *miss)
3288 {
3289     struct dpif_backer *backer = miss->ofproto->backer;
3290     uint32_t hash;
3291
3292     switch (flow_miss_model) {
3293     case OFPROTO_HANDLE_MISS_AUTO:
3294         break;
3295     case OFPROTO_HANDLE_MISS_WITH_FACETS:
3296         return true;
3297     case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3298         return false;
3299     }
3300
3301     if (!backer->governor) {
3302         size_t n_subfacets;
3303
3304         n_subfacets = hmap_count(&backer->subfacets);
3305         if (n_subfacets * 2 <= flow_eviction_threshold) {
3306             return true;
3307         }
3308
3309         backer->governor = governor_create();
3310     }
3311
3312     hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3313     return governor_should_install_flow(backer->governor, hash,
3314                                         list_size(&miss->packets));
3315 }
3316
3317 /* Handles 'miss', which matches 'facet'.  May add any required datapath
3318  * operations to 'ops', incrementing '*n_ops' for each new op.
3319  *
3320  * All of the packets in 'miss' are considered to have arrived at time
3321  * 'miss->stats.used'.  This is really important only for new facets: if we
3322  * just called time_msec() here, then the new subfacet or its packets could
3323  * look (occasionally) as though it was used some time after the facet was
3324  * used.  That can make a one-packet flow look like it has a nonzero duration,
3325  * which looks odd in e.g. NetFlow statistics. */
3326 static void
3327 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3328                             struct flow_miss_op *ops, size_t *n_ops)
3329 {
3330     enum subfacet_path want_path;
3331     struct subfacet *subfacet;
3332
3333     facet->packet_count += miss->stats.n_packets;
3334     facet->prev_packet_count += miss->stats.n_packets;
3335     facet->byte_count += miss->stats.n_bytes;
3336     facet->prev_byte_count += miss->stats.n_bytes;
3337
3338     want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3339
3340     /* Don't install the flow if it's the result of the "userspace"
3341      * action for an already installed facet.  This can occur when a
3342      * datapath flow with wildcards has a "userspace" action and flows
3343      * sent to userspace result in a different subfacet, which will then
3344      * be rejected as overlapping by the datapath. */
3345     if (miss->upcall_type == DPIF_UC_ACTION
3346         && !list_is_empty(&facet->subfacets)) {
3347         return;
3348     }
3349
3350     subfacet = subfacet_create(facet, miss);
3351     if (subfacet->path != want_path) {
3352         struct flow_miss_op *op = &ops[(*n_ops)++];
3353         struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3354
3355         subfacet->path = want_path;
3356
3357         ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3358         if (enable_megaflows) {
3359             odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3360                                    &miss->flow, UINT32_MAX);
3361         }
3362
3363         op->xout_garbage = false;
3364         op->dpif_op.type = DPIF_OP_FLOW_PUT;
3365         op->subfacet = subfacet;
3366         put->flags = DPIF_FP_CREATE;
3367         put->key = miss->key;
3368         put->key_len = miss->key_len;
3369         put->mask = op->mask.data;
3370         put->mask_len = op->mask.size;
3371
3372         if (want_path == SF_FAST_PATH) {
3373             put->actions = facet->xout.odp_actions.data;
3374             put->actions_len = facet->xout.odp_actions.size;
3375         } else {
3376             compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3377                               op->slow_stub, sizeof op->slow_stub,
3378                               &put->actions, &put->actions_len);
3379         }
3380         put->stats = NULL;
3381     }
3382 }
3383
3384 /* Handles flow miss 'miss'.  May add any required datapath operations
3385  * to 'ops', incrementing '*n_ops' for each new op. */
3386 static void
3387 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3388                  size_t *n_ops)
3389 {
3390     struct facet *facet;
3391
3392     miss->ofproto->n_missed += list_size(&miss->packets);
3393
3394     facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3395     if (!facet) {
3396         /* There does not exist a bijection between 'struct flow' and datapath
3397          * flow keys with fitness ODP_FIT_TO_LITTLE.  This breaks a fundamental
3398          * assumption used throughout the facet and subfacet handling code.
3399          * Since we have to handle these misses in userspace anyway, we simply
3400          * skip facet creation, avoiding the problem altogether. */
3401         if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3402             || !flow_miss_should_make_facet(miss)) {
3403             return;
3404         }
3405
3406         facet = facet_create(miss);
3407     }
3408     handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3409 }
3410
3411 static struct drop_key *
3412 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3413                 size_t key_len)
3414 {
3415     struct drop_key *drop_key;
3416
3417     HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3418                              &backer->drop_keys) {
3419         if (drop_key->key_len == key_len
3420             && !memcmp(drop_key->key, key, key_len)) {
3421             return drop_key;
3422         }
3423     }
3424     return NULL;
3425 }
3426
3427 static void
3428 drop_key_clear(struct dpif_backer *backer)
3429 {
3430     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3431     struct drop_key *drop_key, *next;
3432
3433     HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3434         int error;
3435
3436         error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3437                               NULL);
3438         if (error && !VLOG_DROP_WARN(&rl)) {
3439             struct ds ds = DS_EMPTY_INITIALIZER;
3440             odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3441             VLOG_WARN("Failed to delete drop key (%s) (%s)",
3442                       ovs_strerror(error), ds_cstr(&ds));
3443             ds_destroy(&ds);
3444         }
3445
3446         hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3447         drop_key_destroy(drop_key);
3448     }
3449
3450     udpif_drop_key_clear(backer->udpif);
3451 }
3452
3453 static void
3454 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3455 {
3456     struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3457     struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3458     struct flow_miss *miss;
3459     size_t n_ops, i;
3460
3461     /* Process each element in the to-do list, constructing the set of
3462      * operations to batch. */
3463     n_ops = 0;
3464     HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3465         handle_flow_miss(miss, flow_miss_ops, &n_ops);
3466     }
3467     ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3468
3469     /* Execute batch. */
3470     for (i = 0; i < n_ops; i++) {
3471         dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3472     }
3473     dpif_operate(backer->dpif, dpif_ops, n_ops);
3474
3475     for (i = 0; i < n_ops; i++) {
3476         if (dpif_ops[i]->error != 0
3477             && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3478             && flow_miss_ops[i].subfacet) {
3479             struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3480
3481             COVERAGE_INC(subfacet_install_fail);
3482
3483             /* Zero-out subfacet counters when installation failed, but
3484              * datapath reported hits.  This should not happen and
3485              * indicates a bug, since if the datapath flow exists, we
3486              * should not be attempting to create a new subfacet.  A
3487              * buggy datapath could trigger this, so just zero out the
3488              * counters and log an error. */
3489             if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3490                 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3491                             "datapath reported hits");
3492                 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3493             }
3494
3495             subfacet->path = SF_NOT_INSTALLED;
3496         }
3497     }
3498 }
3499
3500 static void
3501 handle_sflow_upcall(struct dpif_backer *backer,
3502                     const struct dpif_upcall *upcall)
3503 {
3504     struct ofproto_dpif *ofproto;
3505     union user_action_cookie cookie;
3506     struct flow flow;
3507     odp_port_t odp_in_port;
3508
3509     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3510                       &flow, NULL, &ofproto, &odp_in_port)
3511         || !ofproto->sflow) {
3512         return;
3513     }
3514
3515     memset(&cookie, 0, sizeof cookie);
3516     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3517     dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3518                         odp_in_port, &cookie);
3519 }
3520
3521 static void
3522 handle_flow_sample_upcall(struct dpif_backer *backer,
3523                           const struct dpif_upcall *upcall)
3524 {
3525     struct ofproto_dpif *ofproto;
3526     union user_action_cookie cookie;
3527     struct flow flow;
3528
3529     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3530                       &flow, NULL, &ofproto, NULL)
3531         || !ofproto->ipfix) {
3532         return;
3533     }
3534
3535     memset(&cookie, 0, sizeof cookie);
3536     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3537
3538     /* The flow reflects exactly the contents of the packet.  Sample
3539      * the packet using it. */
3540     dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3541                            cookie.flow_sample.collector_set_id,
3542                            cookie.flow_sample.probability,
3543                            cookie.flow_sample.obs_domain_id,
3544                            cookie.flow_sample.obs_point_id);
3545 }
3546
3547 static void
3548 handle_ipfix_upcall(struct dpif_backer *backer,
3549                     const struct dpif_upcall *upcall)
3550 {
3551     struct ofproto_dpif *ofproto;
3552     struct flow flow;
3553
3554     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3555                       &flow, NULL, &ofproto, NULL)
3556         || !ofproto->ipfix) {
3557         return;
3558     }
3559
3560     /* The flow reflects exactly the contents of the packet.  Sample
3561      * the packet using it. */
3562     dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3563 }
3564
3565 static void
3566 handle_upcalls(struct dpif_backer *backer)
3567 {
3568     struct flow_miss_batch *fmb;
3569     int n_processed;
3570
3571     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3572         struct upcall *upcall = upcall_next(backer->udpif);
3573
3574         if (!upcall) {
3575             break;
3576         }
3577
3578         switch (upcall->type) {
3579         case SFLOW_UPCALL:
3580             handle_sflow_upcall(backer, &upcall->dpif_upcall);
3581             break;
3582
3583         case FLOW_SAMPLE_UPCALL:
3584             handle_flow_sample_upcall(backer, &upcall->dpif_upcall);
3585             break;
3586
3587         case IPFIX_UPCALL:
3588             handle_ipfix_upcall(backer, &upcall->dpif_upcall);
3589             break;
3590
3591         case BAD_UPCALL:
3592             break;
3593
3594         case MISS_UPCALL:
3595             NOT_REACHED();
3596         }
3597
3598         upcall_destroy(upcall);
3599     }
3600
3601     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3602         struct drop_key *drop_key = drop_key_next(backer->udpif);
3603         if (!drop_key) {
3604             break;
3605         }
3606
3607         if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3608             hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3609                         hash_bytes(drop_key->key, drop_key->key_len, 0));
3610             dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3611                           drop_key->key, drop_key->key_len,
3612                           NULL, 0, NULL, 0, NULL);
3613         } else {
3614             drop_key_destroy(drop_key);
3615         }
3616     }
3617
3618     fmb = flow_miss_batch_next(backer->udpif);
3619     if (fmb) {
3620         handle_flow_misses(backer, fmb);
3621         flow_miss_batch_destroy(fmb);
3622     }
3623 }
3624 \f
3625 /* Flow expiration. */
3626
3627 static int subfacet_max_idle(const struct dpif_backer *);
3628 static void update_stats(struct dpif_backer *);
3629 static void rule_expire(struct rule_dpif *);
3630 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3631
3632 /* This function is called periodically by run().  Its job is to collect
3633  * updates for the flows that have been installed into the datapath, most
3634  * importantly when they last were used, and then use that information to
3635  * expire flows that have not been used recently.
3636  *
3637  * Returns the number of milliseconds after which it should be called again. */
3638 static int
3639 expire(struct dpif_backer *backer)
3640 {
3641     struct ofproto_dpif *ofproto;
3642     size_t n_subfacets;
3643     int max_idle;
3644
3645     /* Periodically clear out the drop keys in an effort to keep them
3646      * relatively few. */
3647     drop_key_clear(backer);
3648
3649     /* Update stats for each flow in the backer. */
3650     update_stats(backer);
3651
3652     n_subfacets = hmap_count(&backer->subfacets);
3653     if (n_subfacets) {
3654         struct subfacet *subfacet;
3655         long long int total, now;
3656
3657         total = 0;
3658         now = time_msec();
3659         HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3660             total += now - subfacet->created;
3661         }
3662         backer->avg_subfacet_life += total / n_subfacets;
3663     }
3664     backer->avg_subfacet_life /= 2;
3665
3666     backer->avg_n_subfacet += n_subfacets;
3667     backer->avg_n_subfacet /= 2;
3668
3669     backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3670
3671     max_idle = subfacet_max_idle(backer);
3672     expire_subfacets(backer, max_idle);
3673
3674     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3675         struct rule *rule, *next_rule;
3676
3677         if (ofproto->backer != backer) {
3678             continue;
3679         }
3680
3681         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3682          * has passed. */
3683         ovs_mutex_lock(&ofproto->up.expirable_mutex);
3684         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3685                             &ofproto->up.expirable) {
3686             rule_expire(rule_dpif_cast(rule));
3687         }
3688         ovs_mutex_unlock(&ofproto->up.expirable_mutex);
3689
3690         /* All outstanding data in existing flows has been accounted, so it's a
3691          * good time to do bond rebalancing. */
3692         if (ofproto->has_bonded_bundles) {
3693             struct ofbundle *bundle;
3694
3695             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3696                 if (bundle->bond) {
3697                     bond_rebalance(bundle->bond);
3698                 }
3699             }
3700         }
3701     }
3702
3703     return MIN(max_idle, 1000);
3704 }
3705
3706 /* Updates flow table statistics given that the datapath just reported 'stats'
3707  * as 'subfacet''s statistics. */
3708 static void
3709 update_subfacet_stats(struct subfacet *subfacet,
3710                       const struct dpif_flow_stats *stats)
3711 {
3712     struct facet *facet = subfacet->facet;
3713     struct dpif_flow_stats diff;
3714
3715     diff.tcp_flags = stats->tcp_flags;
3716     diff.used = stats->used;
3717
3718     if (stats->n_packets >= subfacet->dp_packet_count) {
3719         diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3720     } else {
3721         VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3722         diff.n_packets = 0;
3723     }
3724
3725     if (stats->n_bytes >= subfacet->dp_byte_count) {
3726         diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3727     } else {
3728         VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3729         diff.n_bytes = 0;
3730     }
3731
3732     facet->ofproto->n_hit += diff.n_packets;
3733     subfacet->dp_packet_count = stats->n_packets;
3734     subfacet->dp_byte_count = stats->n_bytes;
3735     subfacet_update_stats(subfacet, &diff);
3736
3737     if (facet->accounted_bytes < facet->byte_count) {
3738         facet_learn(facet);
3739         facet_account(facet);
3740         facet->accounted_bytes = facet->byte_count;
3741     }
3742 }
3743
3744 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3745  * about, or a flow that shouldn't be installed but was anyway.  Delete it. */
3746 static void
3747 delete_unexpected_flow(struct dpif_backer *backer,
3748                        const struct nlattr *key, size_t key_len)
3749 {
3750     if (!VLOG_DROP_WARN(&rl)) {
3751         struct ds s;
3752
3753         ds_init(&s);
3754         odp_flow_key_format(key, key_len, &s);
3755         VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3756         ds_destroy(&s);
3757     }
3758
3759     COVERAGE_INC(facet_unexpected);
3760     dpif_flow_del(backer->dpif, key, key_len, NULL);
3761 }
3762
3763 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3764  *
3765  * This function also pushes statistics updates to rules which each facet
3766  * resubmits into.  Generally these statistics will be accurate.  However, if a
3767  * facet changes the rule it resubmits into at some time in between
3768  * update_stats() runs, it is possible that statistics accrued to the
3769  * old rule will be incorrectly attributed to the new rule.  This could be
3770  * avoided by calling update_stats() whenever rules are created or
3771  * deleted.  However, the performance impact of making so many calls to the
3772  * datapath do not justify the benefit of having perfectly accurate statistics.
3773  *
3774  * In addition, this function maintains per ofproto flow hit counts. The patch
3775  * port is not treated specially. e.g. A packet ingress from br0 patched into
3776  * br1 will increase the hit count of br0 by 1, however, does not affect
3777  * the hit or miss counts of br1.
3778  */
3779 static void
3780 update_stats(struct dpif_backer *backer)
3781 {
3782     const struct dpif_flow_stats *stats;
3783     struct dpif_flow_dump dump;
3784     const struct nlattr *key, *mask;
3785     size_t key_len, mask_len;
3786
3787     dpif_flow_dump_start(&dump, backer->dpif);
3788     while (dpif_flow_dump_next(&dump, &key, &key_len,
3789                                &mask, &mask_len, NULL, NULL, &stats)) {
3790         struct subfacet *subfacet;
3791         uint32_t key_hash;
3792
3793         key_hash = odp_flow_key_hash(key, key_len);
3794         subfacet = subfacet_find(backer, key, key_len, key_hash);
3795         switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3796         case SF_FAST_PATH:
3797             update_subfacet_stats(subfacet, stats);
3798             break;
3799
3800         case SF_SLOW_PATH:
3801             /* Stats are updated per-packet. */
3802             break;
3803
3804         case SF_NOT_INSTALLED:
3805         default:
3806             delete_unexpected_flow(backer, key, key_len);
3807             break;
3808         }
3809         run_fast_rl();
3810     }
3811     dpif_flow_dump_done(&dump);
3812
3813     update_moving_averages(backer);
3814 }
3815
3816 /* Calculates and returns the number of milliseconds of idle time after which
3817  * subfacets should expire from the datapath.  When a subfacet expires, we fold
3818  * its statistics into its facet, and when a facet's last subfacet expires, we
3819  * fold its statistic into its rule. */
3820 static int
3821 subfacet_max_idle(const struct dpif_backer *backer)
3822 {
3823     /*
3824      * Idle time histogram.
3825      *
3826      * Most of the time a switch has a relatively small number of subfacets.
3827      * When this is the case we might as well keep statistics for all of them
3828      * in userspace and to cache them in the kernel datapath for performance as
3829      * well.
3830      *
3831      * As the number of subfacets increases, the memory required to maintain
3832      * statistics about them in userspace and in the kernel becomes
3833      * significant.  However, with a large number of subfacets it is likely
3834      * that only a few of them are "heavy hitters" that consume a large amount
3835      * of bandwidth.  At this point, only heavy hitters are worth caching in
3836      * the kernel and maintaining in userspaces; other subfacets we can
3837      * discard.
3838      *
3839      * The technique used to compute the idle time is to build a histogram with
3840      * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each.  Each subfacet
3841      * that is installed in the kernel gets dropped in the appropriate bucket.
3842      * After the histogram has been built, we compute the cutoff so that only
3843      * the most-recently-used 1% of subfacets (but at least
3844      * flow_eviction_threshold flows) are kept cached.  At least
3845      * the most-recently-used bucket of subfacets is kept, so actually an
3846      * arbitrary number of subfacets can be kept in any given expiration run
3847      * (though the next run will delete most of those unless they receive
3848      * additional data).
3849      *
3850      * This requires a second pass through the subfacets, in addition to the
3851      * pass made by update_stats(), because the former function never looks at
3852      * uninstallable subfacets.
3853      */
3854     enum { BUCKET_WIDTH = 100 };
3855     enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3856     int buckets[N_BUCKETS] = { 0 };
3857     int total, subtotal, bucket;
3858     struct subfacet *subfacet;
3859     long long int now;
3860     int i;
3861
3862     total = hmap_count(&backer->subfacets);
3863     if (total <= flow_eviction_threshold) {
3864         return N_BUCKETS * BUCKET_WIDTH;
3865     }
3866
3867     /* Build histogram. */
3868     now = time_msec();
3869     HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3870         long long int idle = now - subfacet->used;
3871         int bucket = (idle <= 0 ? 0
3872                       : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3873                       : (unsigned int) idle / BUCKET_WIDTH);
3874         buckets[bucket]++;
3875     }
3876
3877     /* Find the first bucket whose flows should be expired. */
3878     subtotal = bucket = 0;
3879     do {
3880         subtotal += buckets[bucket++];
3881     } while (bucket < N_BUCKETS &&
3882              subtotal < MAX(flow_eviction_threshold, total / 100));
3883
3884     if (VLOG_IS_DBG_ENABLED()) {
3885         struct ds s;
3886
3887         ds_init(&s);
3888         ds_put_cstr(&s, "keep");
3889         for (i = 0; i < N_BUCKETS; i++) {
3890             if (i == bucket) {
3891                 ds_put_cstr(&s, ", drop");
3892             }
3893             if (buckets[i]) {
3894                 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3895             }
3896         }
3897         VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3898         ds_destroy(&s);
3899     }
3900
3901     return bucket * BUCKET_WIDTH;
3902 }
3903
3904 static void
3905 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3906 {
3907     /* Cutoff time for most flows. */
3908     long long int normal_cutoff = time_msec() - dp_max_idle;
3909
3910     /* We really want to keep flows for special protocols around, so use a more
3911      * conservative cutoff. */
3912     long long int special_cutoff = time_msec() - 10000;
3913
3914     struct subfacet *subfacet, *next_subfacet;
3915     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3916     int n_batch;
3917
3918     n_batch = 0;
3919     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3920                         &backer->subfacets) {
3921         long long int cutoff;
3922
3923         cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3924                                                 | SLOW_STP)
3925                   ? special_cutoff
3926                   : normal_cutoff);
3927         if (subfacet->used < cutoff) {
3928             if (subfacet->path != SF_NOT_INSTALLED) {
3929                 batch[n_batch++] = subfacet;
3930                 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3931                     subfacet_destroy_batch(backer, batch, n_batch);
3932                     n_batch = 0;
3933                 }
3934             } else {
3935                 subfacet_destroy(subfacet);
3936             }
3937         }
3938     }
3939
3940     if (n_batch > 0) {
3941         subfacet_destroy_batch(backer, batch, n_batch);
3942     }
3943 }
3944
3945 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3946  * then delete it entirely. */
3947 static void
3948 rule_expire(struct rule_dpif *rule)
3949 {
3950     uint16_t idle_timeout, hard_timeout;
3951     long long int now;
3952     uint8_t reason;
3953
3954     ovs_assert(!rule->up.pending);
3955
3956     ovs_mutex_lock(&rule->up.timeout_mutex);
3957     hard_timeout = rule->up.hard_timeout;
3958     idle_timeout = rule->up.idle_timeout;
3959     ovs_mutex_unlock(&rule->up.timeout_mutex);
3960
3961     /* Has 'rule' expired? */
3962     now = time_msec();
3963     if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3964         reason = OFPRR_HARD_TIMEOUT;
3965     } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3966         reason = OFPRR_IDLE_TIMEOUT;
3967     } else {
3968         return;
3969     }
3970
3971     COVERAGE_INC(ofproto_dpif_expired);
3972     ofproto_rule_expire(&rule->up, reason);
3973 }
3974 \f
3975 /* Facets. */
3976
3977 /* Creates and returns a new facet based on 'miss'.
3978  *
3979  * The caller must already have determined that no facet with an identical
3980  * 'miss->flow' exists in 'miss->ofproto'.
3981  *
3982  * 'rule' and 'xout' must have been created based on 'miss'.
3983  *
3984  * 'facet'' statistics are initialized based on 'stats'.
3985  *
3986  * The facet will initially have no subfacets.  The caller should create (at
3987  * least) one subfacet with subfacet_create(). */
3988 static struct facet *
3989 facet_create(const struct flow_miss *miss)
3990 {
3991     struct ofproto_dpif *ofproto = miss->ofproto;
3992     struct facet *facet;
3993     struct match match;
3994
3995     facet = xzalloc(sizeof *facet);
3996     facet->ofproto = miss->ofproto;
3997     facet->used = miss->stats.used;
3998     facet->flow = miss->flow;
3999     facet->learn_rl = time_msec() + 500;
4000
4001     list_init(&facet->subfacets);
4002     netflow_flow_init(&facet->nf_flow);
4003     netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4004
4005     xlate_out_copy(&facet->xout, &miss->xout);
4006
4007     match_init(&match, &facet->flow, &facet->xout.wc);
4008     cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4009     ovs_rwlock_wrlock(&ofproto->facets.rwlock);
4010     classifier_insert(&ofproto->facets, &facet->cr);
4011     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4012
4013     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4014     return facet;
4015 }
4016
4017 static void
4018 facet_free(struct facet *facet)
4019 {
4020     if (facet) {
4021         xlate_out_uninit(&facet->xout);
4022         free(facet);
4023     }
4024 }
4025
4026 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4027  * 'packet', which arrived on 'in_port'. */
4028 static bool
4029 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4030                     const struct nlattr *odp_actions, size_t actions_len,
4031                     struct ofpbuf *packet)
4032 {
4033     struct odputil_keybuf keybuf;
4034     struct ofpbuf key;
4035     int error;
4036
4037     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4038     odp_flow_key_from_flow(&key, flow,
4039                            ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4040
4041     error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4042                          odp_actions, actions_len, packet);
4043     return !error;
4044 }
4045
4046 /* Remove 'facet' from its ofproto and free up the associated memory:
4047  *
4048  *   - If 'facet' was installed in the datapath, uninstalls it and updates its
4049  *     rule's statistics, via subfacet_uninstall().
4050  *
4051  *   - Removes 'facet' from its rule and from ofproto->facets.
4052  */
4053 static void
4054 facet_remove(struct facet *facet)
4055 {
4056     struct subfacet *subfacet, *next_subfacet;
4057
4058     ovs_assert(!list_is_empty(&facet->subfacets));
4059
4060     /* First uninstall all of the subfacets to get final statistics. */
4061     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4062         subfacet_uninstall(subfacet);
4063     }
4064
4065     /* Flush the final stats to the rule.
4066      *
4067      * This might require us to have at least one subfacet around so that we
4068      * can use its actions for accounting in facet_account(), which is why we
4069      * have uninstalled but not yet destroyed the subfacets. */
4070     facet_flush_stats(facet);
4071
4072     /* Now we're really all done so destroy everything. */
4073     LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4074                         &facet->subfacets) {
4075         subfacet_destroy__(subfacet);
4076     }
4077     ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4078     classifier_remove(&facet->ofproto->facets, &facet->cr);
4079     ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4080     cls_rule_destroy(&facet->cr);
4081     facet_free(facet);
4082 }
4083
4084 /* Feed information from 'facet' back into the learning table to keep it in
4085  * sync with what is actually flowing through the datapath. */
4086 static void
4087 facet_learn(struct facet *facet)
4088 {
4089     long long int now = time_msec();
4090
4091     if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4092         return;
4093     }
4094
4095     facet->learn_rl = now + 500;
4096
4097     if (!facet->xout.has_learn
4098         && !facet->xout.has_normal
4099         && (!facet->xout.has_fin_timeout
4100             || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4101         return;
4102     }
4103
4104     facet_push_stats(facet, true);
4105 }
4106
4107 static void
4108 facet_account(struct facet *facet)
4109 {
4110     const struct nlattr *a;
4111     unsigned int left;
4112     ovs_be16 vlan_tci;
4113     uint64_t n_bytes;
4114
4115     if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4116         return;
4117     }
4118     n_bytes = facet->byte_count - facet->accounted_bytes;
4119
4120     /* This loop feeds byte counters to bond_account() for rebalancing to use
4121      * as a basis.  We also need to track the actual VLAN on which the packet
4122      * is going to be sent to ensure that it matches the one passed to
4123      * bond_choose_output_slave().  (Otherwise, we will account to the wrong
4124      * hash bucket.)
4125      *
4126      * We use the actions from an arbitrary subfacet because they should all
4127      * be equally valid for our purpose. */
4128     vlan_tci = facet->flow.vlan_tci;
4129     NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4130                              facet->xout.odp_actions.size) {
4131         const struct ovs_action_push_vlan *vlan;
4132         struct ofport_dpif *port;
4133
4134         switch (nl_attr_type(a)) {
4135         case OVS_ACTION_ATTR_OUTPUT:
4136             port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4137             if (port && port->bundle && port->bundle->bond) {
4138                 bond_account(port->bundle->bond, &facet->flow,
4139                              vlan_tci_to_vid(vlan_tci), n_bytes);
4140             }
4141             break;
4142
4143         case OVS_ACTION_ATTR_POP_VLAN:
4144             vlan_tci = htons(0);
4145             break;
4146
4147         case OVS_ACTION_ATTR_PUSH_VLAN:
4148             vlan = nl_attr_get(a);
4149             vlan_tci = vlan->vlan_tci;
4150             break;
4151         }
4152     }
4153 }
4154
4155 /* Returns true if the only action for 'facet' is to send to the controller.
4156  * (We don't report NetFlow expiration messages for such facets because they
4157  * are just part of the control logic for the network, not real traffic). */
4158 static bool
4159 facet_is_controller_flow(struct facet *facet)
4160 {
4161     if (facet) {
4162         struct ofproto_dpif *ofproto = facet->ofproto;
4163         const struct ofpact *ofpacts;
4164         struct rule_dpif *rule;
4165         size_t ofpacts_len;
4166         bool is_controller;
4167
4168         rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4169         ofpacts_len = rule->up.ofpacts_len;
4170         ofpacts = rule->up.ofpacts;
4171         is_controller = ofpacts_len > 0
4172             && ofpacts->type == OFPACT_CONTROLLER
4173             && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4174         rule_dpif_release(rule);
4175         return is_controller;
4176     }
4177     return false;
4178 }
4179
4180 /* Folds all of 'facet''s statistics into its rule.  Also updates the
4181  * accounting ofhook and emits a NetFlow expiration if appropriate.  All of
4182  * 'facet''s statistics in the datapath should have been zeroed and folded into
4183  * its packet and byte counts before this function is called. */
4184 static void
4185 facet_flush_stats(struct facet *facet)
4186 {
4187     struct ofproto_dpif *ofproto = facet->ofproto;
4188     struct subfacet *subfacet;
4189
4190     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4191         ovs_assert(!subfacet->dp_byte_count);
4192         ovs_assert(!subfacet->dp_packet_count);
4193     }
4194
4195     facet_push_stats(facet, false);
4196     if (facet->accounted_bytes < facet->byte_count) {
4197         facet_account(facet);
4198         facet->accounted_bytes = facet->byte_count;
4199     }
4200
4201     if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4202         struct ofexpired expired;
4203         expired.flow = facet->flow;
4204         expired.packet_count = facet->packet_count;
4205         expired.byte_count = facet->byte_count;
4206         expired.used = facet->used;
4207         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4208     }
4209
4210     /* Reset counters to prevent double counting if 'facet' ever gets
4211      * reinstalled. */
4212     facet_reset_counters(facet);
4213
4214     netflow_flow_clear(&facet->nf_flow);
4215     facet->tcp_flags = 0;
4216 }
4217
4218 /* Searches 'ofproto''s table of facets for one which would be responsible for
4219  * 'flow'.  Returns it if found, otherwise a null pointer.
4220  *
4221  * The returned facet might need revalidation; use facet_lookup_valid()
4222  * instead if that is important. */
4223 static struct facet *
4224 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4225 {
4226     struct cls_rule *cr;
4227
4228     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4229     cr = classifier_lookup(&ofproto->facets, flow, NULL);
4230     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4231     return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4232 }
4233
4234 /* Searches 'ofproto''s table of facets for one capable that covers
4235  * 'flow'.  Returns it if found, otherwise a null pointer.
4236  *
4237  * The returned facet is guaranteed to be valid. */
4238 static struct facet *
4239 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4240 {
4241     struct facet *facet;
4242
4243     facet = facet_find(ofproto, flow);
4244     if (facet
4245         && ofproto->backer->need_revalidate
4246         && !facet_revalidate(facet)) {
4247         return NULL;
4248     }
4249
4250     return facet;
4251 }
4252
4253 static bool
4254 facet_check_consistency(struct facet *facet)
4255 {
4256     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4257
4258     struct xlate_out xout;
4259     struct xlate_in xin;
4260
4261     struct rule_dpif *rule;
4262     bool ok;
4263
4264     /* Check the datapath actions for consistency. */
4265     rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4266     xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4267     xlate_actions(&xin, &xout);
4268     rule_dpif_release(rule);
4269
4270     ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4271         && facet->xout.slow == xout.slow;
4272     if (!ok && !VLOG_DROP_WARN(&rl)) {
4273         struct ds s = DS_EMPTY_INITIALIZER;
4274
4275         flow_format(&s, &facet->flow);
4276         ds_put_cstr(&s, ": inconsistency in facet");
4277
4278         if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4279             ds_put_cstr(&s, " (actions were: ");
4280             format_odp_actions(&s, facet->xout.odp_actions.data,
4281                                facet->xout.odp_actions.size);
4282             ds_put_cstr(&s, ") (correct actions: ");
4283             format_odp_actions(&s, xout.odp_actions.data,
4284                                xout.odp_actions.size);
4285             ds_put_char(&s, ')');
4286         }
4287
4288         if (facet->xout.slow != xout.slow) {
4289             ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4290         }
4291
4292         ds_destroy(&s);
4293     }
4294     xlate_out_uninit(&xout);
4295
4296     return ok;
4297 }
4298
4299 /* Re-searches the classifier for 'facet':
4300  *
4301  *   - If the rule found is different from 'facet''s current rule, moves
4302  *     'facet' to the new rule and recompiles its actions.
4303  *
4304  *   - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4305  *     where it is and recompiles its actions anyway.
4306  *
4307  *   - If any of 'facet''s subfacets correspond to a new flow according to
4308  *     xlate_receive(), 'facet' is removed.
4309  *
4310  *   Returns true if 'facet' is still valid.  False if 'facet' was removed. */
4311 static bool
4312 facet_revalidate(struct facet *facet)
4313 {
4314     struct ofproto_dpif *ofproto = facet->ofproto;
4315     struct rule_dpif *new_rule;
4316     struct subfacet *subfacet;
4317     struct flow_wildcards wc;
4318     struct xlate_out xout;
4319     struct xlate_in xin;
4320
4321     COVERAGE_INC(facet_revalidate);
4322
4323     /* Check that child subfacets still correspond to this facet.  Tunnel
4324      * configuration changes could cause a subfacet's OpenFlow in_port to
4325      * change. */
4326     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4327         struct ofproto_dpif *recv_ofproto;
4328         struct flow recv_flow;
4329         int error;
4330
4331         error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4332                               subfacet->key_len, &recv_flow, NULL,
4333                               &recv_ofproto, NULL);
4334         if (error
4335             || recv_ofproto != ofproto
4336             || facet != facet_find(ofproto, &recv_flow)) {
4337             facet_remove(facet);
4338             return false;
4339         }
4340     }
4341
4342     flow_wildcards_init_catchall(&wc);
4343     rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4344
4345     /* Calculate new datapath actions.
4346      *
4347      * We do not modify any 'facet' state yet, because we might need to, e.g.,
4348      * emit a NetFlow expiration and, if so, we need to have the old state
4349      * around to properly compose it. */
4350     xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4351     xlate_actions(&xin, &xout);
4352     flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4353
4354     /* A facet's slow path reason should only change under dramatic
4355      * circumstances.  Rather than try to update everything, it's simpler to
4356      * remove the facet and start over.
4357      *
4358      * More importantly, if a facet's wildcards change, it will be relatively
4359      * difficult to figure out if its subfacets still belong to it, and if not
4360      * which facet they may belong to.  Again, to avoid the complexity, we
4361      * simply give up instead. */
4362     if (facet->xout.slow != xout.slow
4363         || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4364         facet_remove(facet);
4365         xlate_out_uninit(&xout);
4366         rule_dpif_release(new_rule);
4367         return false;
4368     }
4369
4370     if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4371         LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4372             if (subfacet->path == SF_FAST_PATH) {
4373                 struct dpif_flow_stats stats;
4374
4375                 subfacet_install(subfacet, &xout.odp_actions, &stats);
4376                 subfacet_update_stats(subfacet, &stats);
4377             }
4378         }
4379
4380         facet_flush_stats(facet);
4381
4382         ofpbuf_clear(&facet->xout.odp_actions);
4383         ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4384                    xout.odp_actions.size);
4385     }
4386
4387     /* Update 'facet' now that we've taken care of all the old state. */
4388     facet->xout.slow = xout.slow;
4389     facet->xout.has_learn = xout.has_learn;
4390     facet->xout.has_normal = xout.has_normal;
4391     facet->xout.has_fin_timeout = xout.has_fin_timeout;
4392     facet->xout.nf_output_iface = xout.nf_output_iface;
4393     facet->xout.mirrors = xout.mirrors;
4394     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4395     facet->used = MAX(facet->used, new_rule->up.created);
4396
4397     xlate_out_uninit(&xout);
4398     rule_dpif_release(new_rule);
4399     return true;
4400 }
4401
4402 static void
4403 facet_reset_counters(struct facet *facet)
4404 {
4405     facet->packet_count = 0;
4406     facet->byte_count = 0;
4407     facet->prev_packet_count = 0;
4408     facet->prev_byte_count = 0;
4409     facet->accounted_bytes = 0;
4410 }
4411
4412 static void
4413 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4414                 struct dpif_flow_stats *stats, bool may_learn)
4415 {
4416     struct ofport_dpif *in_port;
4417     struct rule_dpif *rule;
4418     struct xlate_in xin;
4419
4420     in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4421     if (in_port && in_port->is_tunnel) {
4422         netdev_vport_inc_rx(in_port->up.netdev, stats);
4423     }
4424
4425     rule_dpif_lookup(ofproto, flow, NULL, &rule);
4426     rule_dpif_credit_stats(rule, stats);
4427     xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4428     xin.resubmit_stats = stats;
4429     xin.may_learn = may_learn;
4430     xlate_actions_for_side_effects(&xin);
4431     rule_dpif_release(rule);
4432 }
4433
4434 static void
4435 facet_push_stats(struct facet *facet, bool may_learn)
4436 {
4437     struct dpif_flow_stats stats;
4438
4439     ovs_assert(facet->packet_count >= facet->prev_packet_count);
4440     ovs_assert(facet->byte_count >= facet->prev_byte_count);
4441     ovs_assert(facet->used >= facet->prev_used);
4442
4443     stats.n_packets = facet->packet_count - facet->prev_packet_count;
4444     stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4445     stats.used = facet->used;
4446     stats.tcp_flags = facet->tcp_flags;
4447
4448     if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4449         facet->prev_packet_count = facet->packet_count;
4450         facet->prev_byte_count = facet->byte_count;
4451         facet->prev_used = facet->used;
4452
4453         netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4454                                  facet->used);
4455         netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4456         mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4457                             stats.n_packets, stats.n_bytes);
4458         flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4459     }
4460 }
4461
4462 static void
4463 push_all_stats__(bool run_fast)
4464 {
4465     static long long int rl = LLONG_MIN;
4466     struct ofproto_dpif *ofproto;
4467
4468     if (time_msec() < rl) {
4469         return;
4470     }
4471
4472     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4473         struct cls_cursor cursor;
4474         struct facet *facet;
4475
4476         ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4477         cls_cursor_init(&cursor, &ofproto->facets, NULL);
4478         CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4479             facet_push_stats(facet, false);
4480             if (run_fast) {
4481                 run_fast_rl();
4482             }
4483         }
4484         ovs_rwlock_unlock(&ofproto->facets.rwlock);
4485     }
4486
4487     rl = time_msec() + 100;
4488 }
4489
4490 static void
4491 push_all_stats(void)
4492 {
4493     push_all_stats__(true);
4494 }
4495
4496 void
4497 rule_dpif_credit_stats(struct rule_dpif *rule,
4498                        const struct dpif_flow_stats *stats)
4499 {
4500     ovs_mutex_lock(&rule->stats_mutex);
4501     rule->packet_count += stats->n_packets;
4502     rule->byte_count += stats->n_bytes;
4503     rule->up.used = MAX(rule->up.used, stats->used);
4504     ovs_mutex_unlock(&rule->stats_mutex);
4505 }
4506
4507 bool
4508 rule_dpif_fail_open(const struct rule_dpif *rule)
4509 {
4510     return rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4511 }
4512
4513 ovs_be64
4514 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4515 {
4516     return rule->up.flow_cookie;
4517 }
4518
4519 void
4520 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4521                      uint16_t hard_timeout)
4522 {
4523     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4524 }
4525
4526 void
4527 rule_dpif_get_actions(const struct rule_dpif *rule,
4528                       const struct ofpact **ofpacts, size_t *ofpacts_len)
4529 {
4530     *ofpacts = rule->up.ofpacts;
4531     *ofpacts_len = rule->up.ofpacts_len;
4532 }
4533 \f
4534 /* Subfacets. */
4535
4536 static struct subfacet *
4537 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4538               size_t key_len, uint32_t key_hash)
4539 {
4540     struct subfacet *subfacet;
4541
4542     HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4543                              &backer->subfacets) {
4544         if (subfacet->key_len == key_len
4545             && !memcmp(key, subfacet->key, key_len)) {
4546             return subfacet;
4547         }
4548     }
4549
4550     return NULL;
4551 }
4552
4553 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4554  * 'key_fitness', 'key', and 'key_len' members in 'miss'.  Returns the
4555  * existing subfacet if there is one, otherwise creates and returns a
4556  * new subfacet. */
4557 static struct subfacet *
4558 subfacet_create(struct facet *facet, struct flow_miss *miss)
4559 {
4560     struct dpif_backer *backer = miss->ofproto->backer;
4561     enum odp_key_fitness key_fitness = miss->key_fitness;
4562     const struct nlattr *key = miss->key;
4563     size_t key_len = miss->key_len;
4564     uint32_t key_hash;
4565     struct subfacet *subfacet;
4566
4567     key_hash = odp_flow_key_hash(key, key_len);
4568
4569     if (list_is_empty(&facet->subfacets)) {
4570         subfacet = &facet->one_subfacet;
4571     } else {
4572         subfacet = subfacet_find(backer, key, key_len, key_hash);
4573         if (subfacet) {
4574             if (subfacet->facet == facet) {
4575                 return subfacet;
4576             }
4577
4578             /* This shouldn't happen. */
4579             VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4580             subfacet_destroy(subfacet);
4581         }
4582
4583         subfacet = xmalloc(sizeof *subfacet);
4584     }
4585
4586     hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4587     list_push_back(&facet->subfacets, &subfacet->list_node);
4588     subfacet->facet = facet;
4589     subfacet->key_fitness = key_fitness;
4590     subfacet->key = xmemdup(key, key_len);
4591     subfacet->key_len = key_len;
4592     subfacet->used = miss->stats.used;
4593     subfacet->created = subfacet->used;
4594     subfacet->dp_packet_count = 0;
4595     subfacet->dp_byte_count = 0;
4596     subfacet->path = SF_NOT_INSTALLED;
4597     subfacet->backer = backer;
4598
4599     backer->subfacet_add_count++;
4600     return subfacet;
4601 }
4602
4603 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4604  * its facet within 'ofproto', and frees it. */
4605 static void
4606 subfacet_destroy__(struct subfacet *subfacet)
4607 {
4608     struct facet *facet = subfacet->facet;
4609     struct ofproto_dpif *ofproto = facet->ofproto;
4610
4611     /* Update ofproto stats before uninstall the subfacet. */
4612     ofproto->backer->subfacet_del_count++;
4613
4614     subfacet_uninstall(subfacet);
4615     hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4616     list_remove(&subfacet->list_node);
4617     free(subfacet->key);
4618     if (subfacet != &facet->one_subfacet) {
4619         free(subfacet);
4620     }
4621 }
4622
4623 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4624  * last remaining subfacet in its facet destroys the facet too. */
4625 static void
4626 subfacet_destroy(struct subfacet *subfacet)
4627 {
4628     struct facet *facet = subfacet->facet;
4629
4630     if (list_is_singleton(&facet->subfacets)) {
4631         /* facet_remove() needs at least one subfacet (it will remove it). */
4632         facet_remove(facet);
4633     } else {
4634         subfacet_destroy__(subfacet);
4635     }
4636 }
4637
4638 static void
4639 subfacet_destroy_batch(struct dpif_backer *backer,
4640                        struct subfacet **subfacets, int n)
4641 {
4642     struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4643     struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4644     struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4645     int i;
4646
4647     for (i = 0; i < n; i++) {
4648         ops[i].type = DPIF_OP_FLOW_DEL;
4649         ops[i].u.flow_del.key = subfacets[i]->key;
4650         ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4651         ops[i].u.flow_del.stats = &stats[i];
4652         opsp[i] = &ops[i];
4653     }
4654
4655     dpif_operate(backer->dpif, opsp, n);
4656     for (i = 0; i < n; i++) {
4657         subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4658         subfacets[i]->path = SF_NOT_INSTALLED;
4659         subfacet_destroy(subfacets[i]);
4660         run_fast_rl();
4661     }
4662 }
4663
4664 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4665  * bytes of actions in 'actions'.  If 'stats' is non-null, statistics counters
4666  * in the datapath will be zeroed and 'stats' will be updated with traffic new
4667  * since 'subfacet' was last updated.
4668  *
4669  * Returns 0 if successful, otherwise a positive errno value. */
4670 static int
4671 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4672                  struct dpif_flow_stats *stats)
4673 {
4674     struct facet *facet = subfacet->facet;
4675     enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4676     const struct nlattr *actions = odp_actions->data;
4677     size_t actions_len = odp_actions->size;
4678     struct odputil_keybuf maskbuf;
4679     struct ofpbuf mask;
4680
4681     uint64_t slow_path_stub[128 / 8];
4682     enum dpif_flow_put_flags flags;
4683     int ret;
4684
4685     flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4686                                                : DPIF_FP_MODIFY;
4687     if (stats) {
4688         flags |= DPIF_FP_ZERO_STATS;
4689     }
4690
4691     if (path == SF_SLOW_PATH) {
4692         compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4693                           slow_path_stub, sizeof slow_path_stub,
4694                           &actions, &actions_len);
4695     }
4696
4697     ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4698     if (enable_megaflows) {
4699         odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4700                                &facet->flow, UINT32_MAX);
4701     }
4702
4703     ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4704                         subfacet->key_len,  mask.data, mask.size,
4705                         actions, actions_len, stats);
4706
4707     if (stats) {
4708         subfacet_reset_dp_stats(subfacet, stats);
4709     }
4710
4711     if (ret) {
4712         COVERAGE_INC(subfacet_install_fail);
4713     } else {
4714         subfacet->path = path;
4715     }
4716     return ret;
4717 }
4718
4719 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4720 static void
4721 subfacet_uninstall(struct subfacet *subfacet)
4722 {
4723     if (subfacet->path != SF_NOT_INSTALLED) {
4724         struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4725         struct dpif_flow_stats stats;
4726         int error;
4727
4728         error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4729                               subfacet->key_len, &stats);
4730         subfacet_reset_dp_stats(subfacet, &stats);
4731         if (!error) {
4732             subfacet_update_stats(subfacet, &stats);
4733         }
4734         subfacet->path = SF_NOT_INSTALLED;
4735     } else {
4736         ovs_assert(subfacet->dp_packet_count == 0);
4737         ovs_assert(subfacet->dp_byte_count == 0);
4738     }
4739 }
4740
4741 /* Resets 'subfacet''s datapath statistics counters.  This should be called
4742  * when 'subfacet''s statistics are cleared in the datapath.  If 'stats' is
4743  * non-null, it should contain the statistics returned by dpif when 'subfacet'
4744  * was reset in the datapath.  'stats' will be modified to include only
4745  * statistics new since 'subfacet' was last updated. */
4746 static void
4747 subfacet_reset_dp_stats(struct subfacet *subfacet,
4748                         struct dpif_flow_stats *stats)
4749 {
4750     if (stats
4751         && subfacet->dp_packet_count <= stats->n_packets
4752         && subfacet->dp_byte_count <= stats->n_bytes) {
4753         stats->n_packets -= subfacet->dp_packet_count;
4754         stats->n_bytes -= subfacet->dp_byte_count;
4755     }
4756
4757     subfacet->dp_packet_count = 0;
4758     subfacet->dp_byte_count = 0;
4759 }
4760
4761 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4762  *
4763  * Because of the meaning of a subfacet's counters, it only makes sense to do
4764  * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4765  * represents a packet that was sent by hand or if it represents statistics
4766  * that have been cleared out of the datapath. */
4767 static void
4768 subfacet_update_stats(struct subfacet *subfacet,
4769                       const struct dpif_flow_stats *stats)
4770 {
4771     if (stats->n_packets || stats->used > subfacet->used) {
4772         struct facet *facet = subfacet->facet;
4773
4774         subfacet->used = MAX(subfacet->used, stats->used);
4775         facet->used = MAX(facet->used, stats->used);
4776         facet->packet_count += stats->n_packets;
4777         facet->byte_count += stats->n_bytes;
4778         facet->tcp_flags |= stats->tcp_flags;
4779     }
4780 }
4781 \f
4782 /* Rules. */
4783
4784 /* Lookup 'flow' in 'ofproto''s classifier.  If 'wc' is non-null, sets
4785  * the fields that were relevant as part of the lookup. */
4786 void
4787 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4788                  struct flow_wildcards *wc, struct rule_dpif **rule)
4789 {
4790     struct ofport_dpif *port;
4791
4792     if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4793         return;
4794     }
4795     port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4796     if (!port) {
4797         VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4798                      flow->in_port.ofp_port);
4799     }
4800
4801     choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4802                      ofproto->no_packet_in_rule, rule);
4803 }
4804
4805 bool
4806 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4807                           const struct flow *flow, struct flow_wildcards *wc,
4808                           uint8_t table_id, struct rule_dpif **rule)
4809     OVS_TRY_RDLOCK(true, (*rule)->up.rwlock)
4810 {
4811     struct cls_rule *cls_rule;
4812     struct classifier *cls;
4813     bool frag;
4814
4815     *rule = NULL;
4816     if (table_id >= N_TABLES) {
4817         return false;
4818     }
4819
4820     if (wc) {
4821         memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4822         wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4823     }
4824
4825     cls = &ofproto->up.tables[table_id].cls;
4826     ovs_rwlock_rdlock(&cls->rwlock);
4827     frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4828     if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4829         /* We must pretend that transport ports are unavailable. */
4830         struct flow ofpc_normal_flow = *flow;
4831         ofpc_normal_flow.tp_src = htons(0);
4832         ofpc_normal_flow.tp_dst = htons(0);
4833         cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4834     } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4835         cls_rule = &ofproto->drop_frags_rule->up.cr;
4836         if (wc) {
4837             flow_wildcards_init_exact(wc);
4838         }
4839     } else {
4840         cls_rule = classifier_lookup(cls, flow, wc);
4841     }
4842
4843     *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4844     if (*rule && ovs_rwlock_tryrdlock(&(*rule)->up.rwlock)) {
4845         /* The rule is in the process of being removed.  Best we can do is
4846          * pretend it isn't there. */
4847         *rule = NULL;
4848     }
4849     ovs_rwlock_unlock(&cls->rwlock);
4850
4851     return *rule != NULL;
4852 }
4853
4854 /* Given a port configuration (specified as zero if there's no port), chooses
4855  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4856  * flow table miss. */
4857 void
4858 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4859                  struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4860     OVS_NO_THREAD_SAFETY_ANALYSIS
4861 {
4862     *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4863     ovs_rwlock_rdlock(&(*rule)->up.rwlock);
4864 }
4865
4866 void
4867 rule_dpif_release(struct rule_dpif *rule)
4868     OVS_NO_THREAD_SAFETY_ANALYSIS
4869 {
4870     if (rule) {
4871         ovs_rwlock_unlock(&rule->up.rwlock);
4872     }
4873 }
4874
4875 static void
4876 complete_operation(struct rule_dpif *rule)
4877 {
4878     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4879
4880     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4881     ofoperation_complete(rule->up.pending, 0);
4882 }
4883
4884 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4885 {
4886     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4887 }
4888
4889 static struct rule *
4890 rule_alloc(void)
4891 {
4892     struct rule_dpif *rule = xmalloc(sizeof *rule);
4893     return &rule->up;
4894 }
4895
4896 static void
4897 rule_dealloc(struct rule *rule_)
4898 {
4899     struct rule_dpif *rule = rule_dpif_cast(rule_);
4900     free(rule);
4901 }
4902
4903 static enum ofperr
4904 rule_construct(struct rule *rule_)
4905 {
4906     struct rule_dpif *rule = rule_dpif_cast(rule_);
4907     ovs_mutex_init(&rule->stats_mutex);
4908     ovs_mutex_lock(&rule->stats_mutex);
4909     rule->packet_count = 0;
4910     rule->byte_count = 0;
4911     ovs_mutex_unlock(&rule->stats_mutex);
4912     return 0;
4913 }
4914
4915 static void
4916 rule_insert(struct rule *rule_)
4917 {
4918     struct rule_dpif *rule = rule_dpif_cast(rule_);
4919     complete_operation(rule);
4920 }
4921
4922 static void
4923 rule_delete(struct rule *rule_)
4924 {
4925     struct rule_dpif *rule = rule_dpif_cast(rule_);
4926     complete_operation(rule);
4927 }
4928
4929 static void
4930 rule_destruct(struct rule *rule_)
4931 {
4932     struct rule_dpif *rule = rule_dpif_cast(rule_);
4933     ovs_mutex_destroy(&rule->stats_mutex);
4934 }
4935
4936 static void
4937 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4938 {
4939     struct rule_dpif *rule = rule_dpif_cast(rule_);
4940
4941     /* push_all_stats() can handle flow misses which, when using the learn
4942      * action, can cause rules to be added and deleted.  This can corrupt our
4943      * caller's datastructures which assume that rule_get_stats() doesn't have
4944      * an impact on the flow table. To be safe, we disable miss handling. */
4945     push_all_stats__(false);
4946
4947     /* Start from historical data for 'rule' itself that are no longer tracked
4948      * in facets.  This counts, for example, facets that have expired. */
4949     ovs_mutex_lock(&rule->stats_mutex);
4950     *packets = rule->packet_count;
4951     *bytes = rule->byte_count;
4952     ovs_mutex_unlock(&rule->stats_mutex);
4953 }
4954
4955 static void
4956 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4957                   struct ofpbuf *packet)
4958 {
4959     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4960     struct dpif_flow_stats stats;
4961     struct xlate_out xout;
4962     struct xlate_in xin;
4963
4964     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4965     rule_dpif_credit_stats(rule, &stats);
4966
4967     xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4968     xin.resubmit_stats = &stats;
4969     xlate_actions(&xin, &xout);
4970
4971     execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4972                         xout.odp_actions.size, packet);
4973
4974     xlate_out_uninit(&xout);
4975 }
4976
4977 static enum ofperr
4978 rule_execute(struct rule *rule, const struct flow *flow,
4979              struct ofpbuf *packet)
4980 {
4981     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4982     ofpbuf_delete(packet);
4983     return 0;
4984 }
4985
4986 static void
4987 rule_modify_actions(struct rule *rule_, bool reset_counters)
4988 {
4989     struct rule_dpif *rule = rule_dpif_cast(rule_);
4990
4991     if (reset_counters) {
4992         ovs_mutex_lock(&rule->stats_mutex);
4993         rule->packet_count = 0;
4994         rule->byte_count = 0;
4995         ovs_mutex_unlock(&rule->stats_mutex);
4996     }
4997
4998     complete_operation(rule);
4999 }
5000 \f
5001 /* Sends 'packet' out 'ofport'.
5002  * May modify 'packet'.
5003  * Returns 0 if successful, otherwise a positive errno value. */
5004 static int
5005 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5006 {
5007     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5008     uint64_t odp_actions_stub[1024 / 8];
5009     struct ofpbuf key, odp_actions;
5010     struct dpif_flow_stats stats;
5011     struct odputil_keybuf keybuf;
5012     struct ofpact_output output;
5013     struct xlate_out xout;
5014     struct xlate_in xin;
5015     struct flow flow;
5016     union flow_in_port in_port_;
5017     int error;
5018
5019     ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5020     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5021
5022     /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5023     in_port_.ofp_port = OFPP_NONE;
5024     flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5025     odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5026                                                              OFPP_LOCAL));
5027     dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5028
5029     ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5030     output.port = ofport->up.ofp_port;
5031     output.max_len = 0;
5032
5033     xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5034     xin.ofpacts_len = sizeof output;
5035     xin.ofpacts = &output.ofpact;
5036     xin.resubmit_stats = &stats;
5037     xlate_actions(&xin, &xout);
5038
5039     error = dpif_execute(ofproto->backer->dpif,
5040                          key.data, key.size,
5041                          xout.odp_actions.data, xout.odp_actions.size,
5042                          packet);
5043     xlate_out_uninit(&xout);
5044
5045     if (error) {
5046         VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5047                      ofproto->up.name, netdev_get_name(ofport->up.netdev),
5048                      ovs_strerror(error));
5049     }
5050
5051     ofproto->stats.tx_packets++;
5052     ofproto->stats.tx_bytes += packet->size;
5053     return error;
5054 }
5055
5056 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5057  * The action will state 'slow' as the reason that the action is in the slow
5058  * path.  (This is purely informational: it allows a human viewing "ovs-dpctl
5059  * dump-flows" output to see why a flow is in the slow path.)
5060  *
5061  * The 'stub_size' bytes in 'stub' will be used to store the action.
5062  * 'stub_size' must be large enough for the action.
5063  *
5064  * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5065  * respectively. */
5066 static void
5067 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5068                   enum slow_path_reason slow,
5069                   uint64_t *stub, size_t stub_size,
5070                   const struct nlattr **actionsp, size_t *actions_lenp)
5071 {
5072     union user_action_cookie cookie;
5073     struct ofpbuf buf;
5074
5075     cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5076     cookie.slow_path.unused = 0;
5077     cookie.slow_path.reason = slow;
5078
5079     ofpbuf_use_stack(&buf, stub, stub_size);
5080     if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5081         uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5082                                          ODPP_NONE);
5083         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5084     } else {
5085         odp_port_t odp_port;
5086         uint32_t pid;
5087
5088         odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5089         pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5090         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5091     }
5092     *actionsp = buf.data;
5093     *actions_lenp = buf.size;
5094 }
5095 \f
5096 static bool
5097 set_frag_handling(struct ofproto *ofproto_,
5098                   enum ofp_config_flags frag_handling)
5099 {
5100     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5101     if (frag_handling != OFPC_FRAG_REASM) {
5102         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5103         return true;
5104     } else {
5105         return false;
5106     }
5107 }
5108
5109 static enum ofperr
5110 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5111            const struct flow *flow,
5112            const struct ofpact *ofpacts, size_t ofpacts_len)
5113 {
5114     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5115     struct odputil_keybuf keybuf;
5116     struct dpif_flow_stats stats;
5117     struct xlate_out xout;
5118     struct xlate_in xin;
5119     struct ofpbuf key;
5120
5121
5122     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5123     odp_flow_key_from_flow(&key, flow,
5124                            ofp_port_to_odp_port(ofproto,
5125                                       flow->in_port.ofp_port));
5126
5127     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5128
5129     xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5130     xin.resubmit_stats = &stats;
5131     xin.ofpacts_len = ofpacts_len;
5132     xin.ofpacts = ofpacts;
5133
5134     xlate_actions(&xin, &xout);
5135     dpif_execute(ofproto->backer->dpif, key.data, key.size,
5136                  xout.odp_actions.data, xout.odp_actions.size, packet);
5137     xlate_out_uninit(&xout);
5138
5139     return 0;
5140 }
5141 \f
5142 /* NetFlow. */
5143
5144 static int
5145 set_netflow(struct ofproto *ofproto_,
5146             const struct netflow_options *netflow_options)
5147 {
5148     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5149
5150     if (netflow_options) {
5151         if (!ofproto->netflow) {
5152             ofproto->netflow = netflow_create();
5153             ofproto->backer->need_revalidate = REV_RECONFIGURE;
5154         }
5155         return netflow_set_options(ofproto->netflow, netflow_options);
5156     } else if (ofproto->netflow) {
5157         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5158         netflow_destroy(ofproto->netflow);
5159         ofproto->netflow = NULL;
5160     }
5161
5162     return 0;
5163 }
5164
5165 static void
5166 get_netflow_ids(const struct ofproto *ofproto_,
5167                 uint8_t *engine_type, uint8_t *engine_id)
5168 {
5169     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5170
5171     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5172 }
5173
5174 static void
5175 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5176 {
5177     if (!facet_is_controller_flow(facet) &&
5178         netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5179         struct subfacet *subfacet;
5180         struct ofexpired expired;
5181
5182         LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5183             if (subfacet->path == SF_FAST_PATH) {
5184                 struct dpif_flow_stats stats;
5185
5186                 subfacet_install(subfacet, &facet->xout.odp_actions,
5187                                  &stats);
5188                 subfacet_update_stats(subfacet, &stats);
5189             }
5190         }
5191
5192         expired.flow = facet->flow;
5193         expired.packet_count = facet->packet_count;
5194         expired.byte_count = facet->byte_count;
5195         expired.used = facet->used;
5196         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5197     }
5198 }
5199
5200 static void
5201 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5202 {
5203     struct cls_cursor cursor;
5204     struct facet *facet;
5205
5206     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5207     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5208     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5209         send_active_timeout(ofproto, facet);
5210     }
5211     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5212 }
5213 \f
5214 static struct ofproto_dpif *
5215 ofproto_dpif_lookup(const char *name)
5216 {
5217     struct ofproto_dpif *ofproto;
5218
5219     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5220                              hash_string(name, 0), &all_ofproto_dpifs) {
5221         if (!strcmp(ofproto->up.name, name)) {
5222             return ofproto;
5223         }
5224     }
5225     return NULL;
5226 }
5227
5228 static void
5229 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5230                           const char *argv[], void *aux OVS_UNUSED)
5231 {
5232     struct ofproto_dpif *ofproto;
5233
5234     if (argc > 1) {
5235         ofproto = ofproto_dpif_lookup(argv[1]);
5236         if (!ofproto) {
5237             unixctl_command_reply_error(conn, "no such bridge");
5238             return;
5239         }
5240         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5241         mac_learning_flush(ofproto->ml);
5242         ovs_rwlock_unlock(&ofproto->ml->rwlock);
5243     } else {
5244         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5245             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5246             mac_learning_flush(ofproto->ml);
5247             ovs_rwlock_unlock(&ofproto->ml->rwlock);
5248         }
5249     }
5250
5251     unixctl_command_reply(conn, "table successfully flushed");
5252 }
5253
5254 static struct ofport_dpif *
5255 ofbundle_get_a_port(const struct ofbundle *bundle)
5256 {
5257     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5258                         bundle_node);
5259 }
5260
5261 static void
5262 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5263                          const char *argv[], void *aux OVS_UNUSED)
5264 {
5265     struct ds ds = DS_EMPTY_INITIALIZER;
5266     const struct ofproto_dpif *ofproto;
5267     const struct mac_entry *e;
5268
5269     ofproto = ofproto_dpif_lookup(argv[1]);
5270     if (!ofproto) {
5271         unixctl_command_reply_error(conn, "no such bridge");
5272         return;
5273     }
5274
5275     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
5276     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5277     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5278         struct ofbundle *bundle = e->port.p;
5279         char name[OFP_MAX_PORT_NAME_LEN];
5280
5281         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5282                                name, sizeof name);
5283         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
5284                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
5285                       mac_entry_age(ofproto->ml, e));
5286     }
5287     ovs_rwlock_unlock(&ofproto->ml->rwlock);
5288     unixctl_command_reply(conn, ds_cstr(&ds));
5289     ds_destroy(&ds);
5290 }
5291
5292 struct trace_ctx {
5293     struct xlate_out xout;
5294     struct xlate_in xin;
5295     struct flow flow;
5296     struct ds *result;
5297 };
5298
5299 static void
5300 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5301 {
5302     ds_put_char_multiple(result, '\t', level);
5303     if (!rule) {
5304         ds_put_cstr(result, "No match\n");
5305         return;
5306     }
5307
5308     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5309                   rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5310     cls_rule_format(&rule->up.cr, result);
5311     ds_put_char(result, '\n');
5312
5313     ds_put_char_multiple(result, '\t', level);
5314     ds_put_cstr(result, "OpenFlow ");
5315     ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5316     ds_put_char(result, '\n');
5317 }
5318
5319 static void
5320 trace_format_flow(struct ds *result, int level, const char *title,
5321                   struct trace_ctx *trace)
5322 {
5323     ds_put_char_multiple(result, '\t', level);
5324     ds_put_format(result, "%s: ", title);
5325     if (flow_equal(&trace->xin.flow, &trace->flow)) {
5326         ds_put_cstr(result, "unchanged");
5327     } else {
5328         flow_format(result, &trace->xin.flow);
5329         trace->flow = trace->xin.flow;
5330     }
5331     ds_put_char(result, '\n');
5332 }
5333
5334 static void
5335 trace_format_regs(struct ds *result, int level, const char *title,
5336                   struct trace_ctx *trace)
5337 {
5338     size_t i;
5339
5340     ds_put_char_multiple(result, '\t', level);
5341     ds_put_format(result, "%s:", title);
5342     for (i = 0; i < FLOW_N_REGS; i++) {
5343         ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5344     }
5345     ds_put_char(result, '\n');
5346 }
5347
5348 static void
5349 trace_format_odp(struct ds *result, int level, const char *title,
5350                  struct trace_ctx *trace)
5351 {
5352     struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5353
5354     ds_put_char_multiple(result, '\t', level);
5355     ds_put_format(result, "%s: ", title);
5356     format_odp_actions(result, odp_actions->data, odp_actions->size);
5357     ds_put_char(result, '\n');
5358 }
5359
5360 static void
5361 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5362 {
5363     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5364     struct ds *result = trace->result;
5365
5366     ds_put_char(result, '\n');
5367     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5368     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5369     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
5370     trace_format_rule(result, recurse + 1, rule);
5371 }
5372
5373 static void
5374 trace_report(struct xlate_in *xin, const char *s, int recurse)
5375 {
5376     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5377     struct ds *result = trace->result;
5378
5379     ds_put_char_multiple(result, '\t', recurse);
5380     ds_put_cstr(result, s);
5381     ds_put_char(result, '\n');
5382 }
5383
5384 static void
5385 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5386                       void *aux OVS_UNUSED)
5387 {
5388     const struct dpif_backer *backer;
5389     struct ofproto_dpif *ofproto;
5390     struct ofpbuf odp_key, odp_mask;
5391     struct ofpbuf *packet;
5392     struct ds result;
5393     struct flow flow;
5394     char *s;
5395
5396     packet = NULL;
5397     backer = NULL;
5398     ds_init(&result);
5399     ofpbuf_init(&odp_key, 0);
5400     ofpbuf_init(&odp_mask, 0);
5401
5402     /* Handle "-generate" or a hex string as the last argument. */
5403     if (!strcmp(argv[argc - 1], "-generate")) {
5404         packet = ofpbuf_new(0);
5405         argc--;
5406     } else {
5407         const char *error = eth_from_hex(argv[argc - 1], &packet);
5408         if (!error) {
5409             argc--;
5410         } else if (argc == 4) {
5411             /* The 3-argument form must end in "-generate' or a hex string. */
5412             unixctl_command_reply_error(conn, error);
5413             goto exit;
5414         }
5415     }
5416
5417     /* Parse the flow and determine whether a datapath or
5418      * bridge is specified. If function odp_flow_key_from_string()
5419      * returns 0, the flow is a odp_flow. If function
5420      * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5421     if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5422         /* If the odp_flow is the second argument,
5423          * the datapath name is the first argument. */
5424         if (argc == 3) {
5425             const char *dp_type;
5426             if (!strncmp(argv[1], "ovs-", 4)) {
5427                 dp_type = argv[1] + 4;
5428             } else {
5429                 dp_type = argv[1];
5430             }
5431             backer = shash_find_data(&all_dpif_backers, dp_type);
5432             if (!backer) {
5433                 unixctl_command_reply_error(conn, "Cannot find datapath "
5434                                "of this name");
5435                 goto exit;
5436             }
5437         } else {
5438             /* No datapath name specified, so there should be only one
5439              * datapath. */
5440             struct shash_node *node;
5441             if (shash_count(&all_dpif_backers) != 1) {
5442                 unixctl_command_reply_error(conn, "Must specify datapath "
5443                          "name, there is more than one type of datapath");
5444                 goto exit;
5445             }
5446             node = shash_first(&all_dpif_backers);
5447             backer = node->data;
5448         }
5449
5450         if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5451                           NULL, &ofproto, NULL)) {
5452             unixctl_command_reply_error(conn, "Invalid datapath flow");
5453             goto exit;
5454         }
5455         ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5456     } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5457         if (argc != 3) {
5458             unixctl_command_reply_error(conn, "Must specify bridge name");
5459             goto exit;
5460         }
5461
5462         ofproto = ofproto_dpif_lookup(argv[1]);
5463         if (!ofproto) {
5464             unixctl_command_reply_error(conn, "Unknown bridge name");
5465             goto exit;
5466         }
5467     } else {
5468         unixctl_command_reply_error(conn, "Bad flow syntax");
5469         goto exit;
5470     }
5471
5472     /* Generate a packet, if requested. */
5473     if (packet) {
5474         if (!packet->size) {
5475             flow_compose(packet, &flow);
5476         } else {
5477             union flow_in_port in_port_;
5478
5479             in_port_ = flow.in_port;
5480             ds_put_cstr(&result, "Packet: ");
5481             s = ofp_packet_to_string(packet->data, packet->size);
5482             ds_put_cstr(&result, s);
5483             free(s);
5484
5485             /* Use the metadata from the flow and the packet argument
5486              * to reconstruct the flow. */
5487             flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5488                          &in_port_, &flow);
5489         }
5490     }
5491
5492     ofproto_trace(ofproto, &flow, packet, &result);
5493     unixctl_command_reply(conn, ds_cstr(&result));
5494
5495 exit:
5496     ds_destroy(&result);
5497     ofpbuf_delete(packet);
5498     ofpbuf_uninit(&odp_key);
5499     ofpbuf_uninit(&odp_mask);
5500 }
5501
5502 static void
5503 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5504               const struct ofpbuf *packet, struct ds *ds)
5505 {
5506     struct rule_dpif *rule;
5507     struct flow_wildcards wc;
5508
5509     ds_put_cstr(ds, "Flow: ");
5510     flow_format(ds, flow);
5511     ds_put_char(ds, '\n');
5512
5513     flow_wildcards_init_catchall(&wc);
5514     rule_dpif_lookup(ofproto, flow, &wc, &rule);
5515
5516     trace_format_rule(ds, 0, rule);
5517     if (rule == ofproto->miss_rule) {
5518         ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5519     } else if (rule == ofproto->no_packet_in_rule) {
5520         ds_put_cstr(ds, "\nNo match, packets dropped because "
5521                     "OFPPC_NO_PACKET_IN is set on in_port.\n");
5522     } else if (rule == ofproto->drop_frags_rule) {
5523         ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5524                     "and the fragment handling mode is \"drop\".\n");
5525     }
5526
5527     if (rule) {
5528         uint64_t odp_actions_stub[1024 / 8];
5529         struct ofpbuf odp_actions;
5530         struct trace_ctx trace;
5531         struct match match;
5532         uint8_t tcp_flags;
5533
5534         tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5535         trace.result = ds;
5536         trace.flow = *flow;
5537         ofpbuf_use_stub(&odp_actions,
5538                         odp_actions_stub, sizeof odp_actions_stub);
5539         xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5540         trace.xin.resubmit_hook = trace_resubmit;
5541         trace.xin.report_hook = trace_report;
5542
5543         xlate_actions(&trace.xin, &trace.xout);
5544         flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5545
5546         ds_put_char(ds, '\n');
5547         trace_format_flow(ds, 0, "Final flow", &trace);
5548
5549         match_init(&match, flow, &trace.xout.wc);
5550         ds_put_cstr(ds, "Relevant fields: ");
5551         match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5552         ds_put_char(ds, '\n');
5553
5554         ds_put_cstr(ds, "Datapath actions: ");
5555         format_odp_actions(ds, trace.xout.odp_actions.data,
5556                            trace.xout.odp_actions.size);
5557
5558         if (trace.xout.slow) {
5559             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5560                         "slow path because it:");
5561             switch (trace.xout.slow) {
5562             case SLOW_CFM:
5563                 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5564                 break;
5565             case SLOW_LACP:
5566                 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5567                 break;
5568             case SLOW_STP:
5569                 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5570                 break;
5571             case SLOW_BFD:
5572                 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5573                 break;
5574             case SLOW_CONTROLLER:
5575                 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5576                             "to the OpenFlow controller.");
5577                 break;
5578             case __SLOW_MAX:
5579                 NOT_REACHED();
5580             }
5581         }
5582
5583         xlate_out_uninit(&trace.xout);
5584     }
5585
5586     rule_dpif_release(rule);
5587 }
5588
5589 /* Runs a self-check of flow translations in 'ofproto'.  Appends a message to
5590  * 'reply' describing the results. */
5591 static void
5592 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5593 {
5594     struct cls_cursor cursor;
5595     struct facet *facet;
5596     int errors;
5597
5598     errors = 0;
5599     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5600     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5601     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5602         if (!facet_check_consistency(facet)) {
5603             errors++;
5604         }
5605     }
5606     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5607     if (errors) {
5608         ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5609     }
5610
5611     if (errors) {
5612         ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5613                       ofproto->up.name, errors);
5614     } else {
5615         ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5616     }
5617 }
5618
5619 static void
5620 ofproto_dpif_self_check(struct unixctl_conn *conn,
5621                         int argc, const char *argv[], void *aux OVS_UNUSED)
5622 {
5623     struct ds reply = DS_EMPTY_INITIALIZER;
5624     struct ofproto_dpif *ofproto;
5625
5626     if (argc > 1) {
5627         ofproto = ofproto_dpif_lookup(argv[1]);
5628         if (!ofproto) {
5629             unixctl_command_reply_error(conn, "Unknown ofproto (use "
5630                                         "ofproto/list for help)");
5631             return;
5632         }
5633         ofproto_dpif_self_check__(ofproto, &reply);
5634     } else {
5635         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5636             ofproto_dpif_self_check__(ofproto, &reply);
5637         }
5638     }
5639
5640     unixctl_command_reply(conn, ds_cstr(&reply));
5641     ds_destroy(&reply);
5642 }
5643
5644 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
5645  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
5646  * to destroy 'ofproto_shash' and free the returned value. */
5647 static const struct shash_node **
5648 get_ofprotos(struct shash *ofproto_shash)
5649 {
5650     const struct ofproto_dpif *ofproto;
5651
5652     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5653         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5654         shash_add_nocopy(ofproto_shash, name, ofproto);
5655     }
5656
5657     return shash_sort(ofproto_shash);
5658 }
5659
5660 static void
5661 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5662                               const char *argv[] OVS_UNUSED,
5663                               void *aux OVS_UNUSED)
5664 {
5665     struct ds ds = DS_EMPTY_INITIALIZER;
5666     struct shash ofproto_shash;
5667     const struct shash_node **sorted_ofprotos;
5668     int i;
5669
5670     shash_init(&ofproto_shash);
5671     sorted_ofprotos = get_ofprotos(&ofproto_shash);
5672     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5673         const struct shash_node *node = sorted_ofprotos[i];
5674         ds_put_format(&ds, "%s\n", node->name);
5675     }
5676
5677     shash_destroy(&ofproto_shash);
5678     free(sorted_ofprotos);
5679
5680     unixctl_command_reply(conn, ds_cstr(&ds));
5681     ds_destroy(&ds);
5682 }
5683
5684 static void
5685 show_dp_rates(struct ds *ds, const char *heading,
5686               const struct avg_subfacet_rates *rates)
5687 {
5688     ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5689                   heading, rates->add_rate, rates->del_rate);
5690 }
5691
5692 static void
5693 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5694 {
5695     const struct shash_node **ofprotos;
5696     struct ofproto_dpif *ofproto;
5697     struct shash ofproto_shash;
5698     uint64_t n_hit, n_missed;
5699     long long int minutes;
5700     size_t i;
5701
5702     n_hit = n_missed = 0;
5703     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5704         if (ofproto->backer == backer) {
5705             n_missed += ofproto->n_missed;
5706             n_hit += ofproto->n_hit;
5707         }
5708     }
5709
5710     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5711                   dpif_name(backer->dpif), n_hit, n_missed);
5712     ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5713                   " life span: %lldms\n", hmap_count(&backer->subfacets),
5714                   backer->avg_n_subfacet, backer->max_n_subfacet,
5715                   backer->avg_subfacet_life);
5716
5717     minutes = (time_msec() - backer->created) / (1000 * 60);
5718     if (minutes >= 60) {
5719         show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5720     }
5721     if (minutes >= 60 * 24) {
5722         show_dp_rates(ds, "\tdaily avg:",  &backer->daily);
5723     }
5724     show_dp_rates(ds, "\toverall avg:",  &backer->lifetime);
5725
5726     shash_init(&ofproto_shash);
5727     ofprotos = get_ofprotos(&ofproto_shash);
5728     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5729         struct ofproto_dpif *ofproto = ofprotos[i]->data;
5730         const struct shash_node **ports;
5731         size_t j;
5732
5733         if (ofproto->backer != backer) {
5734             continue;
5735         }
5736
5737         ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5738                       ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5739
5740         ports = shash_sort(&ofproto->up.port_by_name);
5741         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5742             const struct shash_node *node = ports[j];
5743             struct ofport *ofport = node->data;
5744             struct smap config;
5745             odp_port_t odp_port;
5746
5747             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5748                           ofport->ofp_port);
5749
5750             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5751             if (odp_port != ODPP_NONE) {
5752                 ds_put_format(ds, "%"PRIu32":", odp_port);
5753             } else {
5754                 ds_put_cstr(ds, "none:");
5755             }
5756
5757             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5758
5759             smap_init(&config);
5760             if (!netdev_get_config(ofport->netdev, &config)) {
5761                 const struct smap_node **nodes;
5762                 size_t i;
5763
5764                 nodes = smap_sort(&config);
5765                 for (i = 0; i < smap_count(&config); i++) {
5766                     const struct smap_node *node = nodes[i];
5767                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5768                                   node->key, node->value);
5769                 }
5770                 free(nodes);
5771             }
5772             smap_destroy(&config);
5773
5774             ds_put_char(ds, ')');
5775             ds_put_char(ds, '\n');
5776         }
5777         free(ports);
5778     }
5779     shash_destroy(&ofproto_shash);
5780     free(ofprotos);
5781 }
5782
5783 static void
5784 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5785                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5786 {
5787     struct ds ds = DS_EMPTY_INITIALIZER;
5788     const struct shash_node **backers;
5789     int i;
5790
5791     backers = shash_sort(&all_dpif_backers);
5792     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5793         dpif_show_backer(backers[i]->data, &ds);
5794     }
5795     free(backers);
5796
5797     unixctl_command_reply(conn, ds_cstr(&ds));
5798     ds_destroy(&ds);
5799 }
5800
5801 /* Dump the megaflow (facet) cache.  This is useful to check the
5802  * correctness of flow wildcarding, since the same mechanism is used for
5803  * both xlate caching and kernel wildcarding.
5804  *
5805  * It's important to note that in the output the flow description uses
5806  * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5807  *
5808  * This command is only needed for advanced debugging, so it's not
5809  * documented in the man page. */
5810 static void
5811 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5812                                     int argc OVS_UNUSED, const char *argv[],
5813                                     void *aux OVS_UNUSED)
5814 {
5815     struct ds ds = DS_EMPTY_INITIALIZER;
5816     const struct ofproto_dpif *ofproto;
5817     long long int now = time_msec();
5818     struct cls_cursor cursor;
5819     struct facet *facet;
5820
5821     ofproto = ofproto_dpif_lookup(argv[1]);
5822     if (!ofproto) {
5823         unixctl_command_reply_error(conn, "no such bridge");
5824         return;
5825     }
5826
5827     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5828     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5829     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5830         cls_rule_format(&facet->cr, &ds);
5831         ds_put_cstr(&ds, ", ");
5832         ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5833         ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5834         ds_put_cstr(&ds, "Datapath actions: ");
5835         if (facet->xout.slow) {
5836             uint64_t slow_path_stub[128 / 8];
5837             const struct nlattr *actions;
5838             size_t actions_len;
5839
5840             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5841                               slow_path_stub, sizeof slow_path_stub,
5842                               &actions, &actions_len);
5843             format_odp_actions(&ds, actions, actions_len);
5844         } else {
5845             format_odp_actions(&ds, facet->xout.odp_actions.data,
5846                                facet->xout.odp_actions.size);
5847         }
5848         ds_put_cstr(&ds, "\n");
5849     }
5850     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5851
5852     ds_chomp(&ds, '\n');
5853     unixctl_command_reply(conn, ds_cstr(&ds));
5854     ds_destroy(&ds);
5855 }
5856
5857 /* Disable using the megaflows.
5858  *
5859  * This command is only needed for advanced debugging, so it's not
5860  * documented in the man page. */
5861 static void
5862 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5863                                        int argc OVS_UNUSED,
5864                                        const char *argv[] OVS_UNUSED,
5865                                        void *aux OVS_UNUSED)
5866 {
5867     struct ofproto_dpif *ofproto;
5868
5869     enable_megaflows = false;
5870
5871     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5872         flush(&ofproto->up);
5873     }
5874
5875     unixctl_command_reply(conn, "megaflows disabled");
5876 }
5877
5878 /* Re-enable using megaflows.
5879  *
5880  * This command is only needed for advanced debugging, so it's not
5881  * documented in the man page. */
5882 static void
5883 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5884                                       int argc OVS_UNUSED,
5885                                       const char *argv[] OVS_UNUSED,
5886                                       void *aux OVS_UNUSED)
5887 {
5888     struct ofproto_dpif *ofproto;
5889
5890     enable_megaflows = true;
5891
5892     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5893         flush(&ofproto->up);
5894     }
5895
5896     unixctl_command_reply(conn, "megaflows enabled");
5897 }
5898
5899 static void
5900 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5901                                 int argc OVS_UNUSED, const char *argv[],
5902                                 void *aux OVS_UNUSED)
5903 {
5904     struct ds ds = DS_EMPTY_INITIALIZER;
5905     const struct ofproto_dpif *ofproto;
5906     struct subfacet *subfacet;
5907
5908     ofproto = ofproto_dpif_lookup(argv[1]);
5909     if (!ofproto) {
5910         unixctl_command_reply_error(conn, "no such bridge");
5911         return;
5912     }
5913
5914     update_stats(ofproto->backer);
5915
5916     HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5917         struct facet *facet = subfacet->facet;
5918         struct odputil_keybuf maskbuf;
5919         struct ofpbuf mask;
5920
5921         if (facet->ofproto != ofproto) {
5922             continue;
5923         }
5924
5925         ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5926         if (enable_megaflows) {
5927             odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5928                                    &facet->flow, UINT32_MAX);
5929         }
5930
5931         odp_flow_format(subfacet->key, subfacet->key_len,
5932                         mask.data, mask.size, &ds, false);
5933
5934         ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5935                       subfacet->dp_packet_count, subfacet->dp_byte_count);
5936         if (subfacet->used) {
5937             ds_put_format(&ds, "%.3fs",
5938                           (time_msec() - subfacet->used) / 1000.0);
5939         } else {
5940             ds_put_format(&ds, "never");
5941         }
5942         if (subfacet->facet->tcp_flags) {
5943             ds_put_cstr(&ds, ", flags:");
5944             packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5945         }
5946
5947         ds_put_cstr(&ds, ", actions:");
5948         if (facet->xout.slow) {
5949             uint64_t slow_path_stub[128 / 8];
5950             const struct nlattr *actions;
5951             size_t actions_len;
5952
5953             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5954                               slow_path_stub, sizeof slow_path_stub,
5955                               &actions, &actions_len);
5956             format_odp_actions(&ds, actions, actions_len);
5957         } else {
5958             format_odp_actions(&ds, facet->xout.odp_actions.data,
5959                                facet->xout.odp_actions.size);
5960         }
5961         ds_put_char(&ds, '\n');
5962     }
5963
5964     unixctl_command_reply(conn, ds_cstr(&ds));
5965     ds_destroy(&ds);
5966 }
5967
5968 static void
5969 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5970                                int argc OVS_UNUSED, const char *argv[],
5971                                void *aux OVS_UNUSED)
5972 {
5973     struct ds ds = DS_EMPTY_INITIALIZER;
5974     struct ofproto_dpif *ofproto;
5975
5976     ofproto = ofproto_dpif_lookup(argv[1]);
5977     if (!ofproto) {
5978         unixctl_command_reply_error(conn, "no such bridge");
5979         return;
5980     }
5981
5982     flush(&ofproto->up);
5983
5984     unixctl_command_reply(conn, ds_cstr(&ds));
5985     ds_destroy(&ds);
5986 }
5987
5988 static void
5989 ofproto_dpif_unixctl_init(void)
5990 {
5991     static bool registered;
5992     if (registered) {
5993         return;
5994     }
5995     registered = true;
5996
5997     unixctl_command_register(
5998         "ofproto/trace",
5999         "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6000         1, 3, ofproto_unixctl_trace, NULL);
6001     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6002                              ofproto_unixctl_fdb_flush, NULL);
6003     unixctl_command_register("fdb/show", "bridge", 1, 1,
6004                              ofproto_unixctl_fdb_show, NULL);
6005     unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6006                              ofproto_dpif_self_check, NULL);
6007     unixctl_command_register("dpif/dump-dps", "", 0, 0,
6008                              ofproto_unixctl_dpif_dump_dps, NULL);
6009     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6010                              NULL);
6011     unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6012                              ofproto_unixctl_dpif_dump_flows, NULL);
6013     unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6014                              ofproto_unixctl_dpif_del_flows, NULL);
6015     unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6016                              ofproto_unixctl_dpif_dump_megaflows, NULL);
6017     unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6018                              ofproto_unixctl_dpif_disable_megaflows, NULL);
6019     unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6020                              ofproto_unixctl_dpif_enable_megaflows, NULL);
6021 }
6022 \f
6023 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6024  *
6025  * This is deprecated.  It is only for compatibility with broken device drivers
6026  * in old versions of Linux that do not properly support VLANs when VLAN
6027  * devices are not used.  When broken device drivers are no longer in
6028  * widespread use, we will delete these interfaces. */
6029
6030 static int
6031 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6032 {
6033     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6034     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6035
6036     if (realdev_ofp_port == ofport->realdev_ofp_port
6037         && vid == ofport->vlandev_vid) {
6038         return 0;
6039     }
6040
6041     ofproto->backer->need_revalidate = REV_RECONFIGURE;
6042
6043     if (ofport->realdev_ofp_port) {
6044         vsp_remove(ofport);
6045     }
6046     if (realdev_ofp_port && ofport->bundle) {
6047         /* vlandevs are enslaved to their realdevs, so they are not allowed to
6048          * themselves be part of a bundle. */
6049         bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6050     }
6051
6052     ofport->realdev_ofp_port = realdev_ofp_port;
6053     ofport->vlandev_vid = vid;
6054
6055     if (realdev_ofp_port) {
6056         vsp_add(ofport, realdev_ofp_port, vid);
6057     }
6058
6059     return 0;
6060 }
6061
6062 static uint32_t
6063 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6064 {
6065     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6066 }
6067
6068 bool
6069 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6070     OVS_EXCLUDED(ofproto->vsp_mutex)
6071 {
6072     bool ret;
6073
6074     ovs_mutex_lock(&ofproto->vsp_mutex);
6075     ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6076     ovs_mutex_unlock(&ofproto->vsp_mutex);
6077     return ret;
6078 }
6079
6080 static ofp_port_t
6081 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6082                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6083     OVS_REQUIRES(ofproto->vsp_mutex)
6084 {
6085     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6086         int vid = vlan_tci_to_vid(vlan_tci);
6087         const struct vlan_splinter *vsp;
6088
6089         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6090                                  hash_realdev_vid(realdev_ofp_port, vid),
6091                                  &ofproto->realdev_vid_map) {
6092             if (vsp->realdev_ofp_port == realdev_ofp_port
6093                 && vsp->vid == vid) {
6094                 return vsp->vlandev_ofp_port;
6095             }
6096         }
6097     }
6098     return realdev_ofp_port;
6099 }
6100
6101 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6102  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6103  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
6104  * 'vlan_tci' 9, it would return the port number of eth0.9.
6105  *
6106  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6107  * function just returns its 'realdev_ofp_port' argument. */
6108 ofp_port_t
6109 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6110                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6111     OVS_EXCLUDED(ofproto->vsp_mutex)
6112 {
6113     ofp_port_t ret;
6114
6115     ovs_mutex_lock(&ofproto->vsp_mutex);
6116     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6117     ovs_mutex_unlock(&ofproto->vsp_mutex);
6118     return ret;
6119 }
6120
6121 static struct vlan_splinter *
6122 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6123 {
6124     struct vlan_splinter *vsp;
6125
6126     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6127                              hash_ofp_port(vlandev_ofp_port),
6128                              &ofproto->vlandev_map) {
6129         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6130             return vsp;
6131         }
6132     }
6133
6134     return NULL;
6135 }
6136
6137 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6138  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6139  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
6140  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6141  * eth0 and store 9 in '*vid'.
6142  *
6143  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6144  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
6145  * always does.*/
6146 static ofp_port_t
6147 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6148                        ofp_port_t vlandev_ofp_port, int *vid)
6149     OVS_REQUIRES(ofproto->vsp_mutex)
6150 {
6151     if (!hmap_is_empty(&ofproto->vlandev_map)) {
6152         const struct vlan_splinter *vsp;
6153
6154         vsp = vlandev_find(ofproto, vlandev_ofp_port);
6155         if (vsp) {
6156             if (vid) {
6157                 *vid = vsp->vid;
6158             }
6159             return vsp->realdev_ofp_port;
6160         }
6161     }
6162     return 0;
6163 }
6164
6165 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6166  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
6167  * 'flow->in_port' to the "real" device backing the VLAN device, sets
6168  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Otherwise (which is
6169  * always the case unless VLAN splinters are enabled), returns false without
6170  * making any changes. */
6171 bool
6172 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6173     OVS_EXCLUDED(ofproto->vsp_mutex)
6174 {
6175     ofp_port_t realdev;
6176     int vid;
6177
6178     ovs_mutex_lock(&ofproto->vsp_mutex);
6179     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6180     ovs_mutex_unlock(&ofproto->vsp_mutex);
6181     if (!realdev) {
6182         return false;
6183     }
6184
6185     /* Cause the flow to be processed as if it came in on the real device with
6186      * the VLAN device's VLAN ID. */
6187     flow->in_port.ofp_port = realdev;
6188     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6189     return true;
6190 }
6191
6192 static void
6193 vsp_remove(struct ofport_dpif *port)
6194 {
6195     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6196     struct vlan_splinter *vsp;
6197
6198     ovs_mutex_lock(&ofproto->vsp_mutex);
6199     vsp = vlandev_find(ofproto, port->up.ofp_port);
6200     if (vsp) {
6201         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6202         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6203         free(vsp);
6204
6205         port->realdev_ofp_port = 0;
6206     } else {
6207         VLOG_ERR("missing vlan device record");
6208     }
6209     ovs_mutex_unlock(&ofproto->vsp_mutex);
6210 }
6211
6212 static void
6213 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6214 {
6215     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6216
6217     ovs_mutex_lock(&ofproto->vsp_mutex);
6218     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6219         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6220             == realdev_ofp_port)) {
6221         struct vlan_splinter *vsp;
6222
6223         vsp = xmalloc(sizeof *vsp);
6224         vsp->realdev_ofp_port = realdev_ofp_port;
6225         vsp->vlandev_ofp_port = port->up.ofp_port;
6226         vsp->vid = vid;
6227
6228         port->realdev_ofp_port = realdev_ofp_port;
6229
6230         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6231                     hash_ofp_port(port->up.ofp_port));
6232         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6233                     hash_realdev_vid(realdev_ofp_port, vid));
6234     } else {
6235         VLOG_ERR("duplicate vlan device record");
6236     }
6237     ovs_mutex_unlock(&ofproto->vsp_mutex);
6238 }
6239
6240 static odp_port_t
6241 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6242 {
6243     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6244     return ofport ? ofport->odp_port : ODPP_NONE;
6245 }
6246
6247 struct ofport_dpif *
6248 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6249 {
6250     struct ofport_dpif *port;
6251
6252     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6253     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6254                              &backer->odp_to_ofport_map) {
6255         if (port->odp_port == odp_port) {
6256             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6257             return port;
6258         }
6259     }
6260
6261     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6262     return NULL;
6263 }
6264
6265 static ofp_port_t
6266 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6267 {
6268     struct ofport_dpif *port;
6269
6270     port = odp_port_to_ofport(ofproto->backer, odp_port);
6271     if (port && &ofproto->up == port->up.ofproto) {
6272         return port->up.ofp_port;
6273     } else {
6274         return OFPP_NONE;
6275     }
6276 }
6277
6278 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6279  * most heavily weighted element.  'base' designates the rate of decay: after
6280  * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6281  * (about .37). */
6282 static void
6283 exp_mavg(double *avg, int base, double new)
6284 {
6285     *avg = (*avg * (base - 1) + new) / base;
6286 }
6287
6288 static void
6289 update_moving_averages(struct dpif_backer *backer)
6290 {
6291     const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6292     long long int minutes = (time_msec() - backer->created) / min_ms;
6293
6294     if (minutes > 0) {
6295         backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6296             / minutes;
6297         backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6298             / minutes;
6299     } else {
6300         backer->lifetime.add_rate = 0.0;
6301         backer->lifetime.del_rate = 0.0;
6302     }
6303
6304     /* Update hourly averages on the minute boundaries. */
6305     if (time_msec() - backer->last_minute >= min_ms) {
6306         exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6307         exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6308
6309         /* Update daily averages on the hour boundaries. */
6310         if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6311             exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6312             exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6313         }
6314
6315         backer->total_subfacet_add_count += backer->subfacet_add_count;
6316         backer->total_subfacet_del_count += backer->subfacet_del_count;
6317         backer->subfacet_add_count = 0;
6318         backer->subfacet_del_count = 0;
6319         backer->last_minute += min_ms;
6320     }
6321 }
6322
6323 const struct ofproto_class ofproto_dpif_class = {
6324     init,
6325     enumerate_types,
6326     enumerate_names,
6327     del,
6328     port_open_type,
6329     type_run,
6330     type_run_fast,
6331     type_wait,
6332     alloc,
6333     construct,
6334     destruct,
6335     dealloc,
6336     run,
6337     run_fast,
6338     wait,
6339     get_memory_usage,
6340     flush,
6341     get_features,
6342     get_tables,
6343     port_alloc,
6344     port_construct,
6345     port_destruct,
6346     port_dealloc,
6347     port_modified,
6348     port_reconfigured,
6349     port_query_by_name,
6350     port_add,
6351     port_del,
6352     port_get_stats,
6353     port_dump_start,
6354     port_dump_next,
6355     port_dump_done,
6356     port_poll,
6357     port_poll_wait,
6358     port_is_lacp_current,
6359     NULL,                       /* rule_choose_table */
6360     rule_alloc,
6361     rule_construct,
6362     rule_insert,
6363     rule_delete,
6364     rule_destruct,
6365     rule_dealloc,
6366     rule_get_stats,
6367     rule_execute,
6368     rule_modify_actions,
6369     set_frag_handling,
6370     packet_out,
6371     set_netflow,
6372     get_netflow_ids,
6373     set_sflow,
6374     set_ipfix,
6375     set_cfm,
6376     get_cfm_status,
6377     set_bfd,
6378     get_bfd_status,
6379     set_stp,
6380     get_stp_status,
6381     set_stp_port,
6382     get_stp_port_status,
6383     set_queues,
6384     bundle_set,
6385     bundle_remove,
6386     mirror_set__,
6387     mirror_get_stats__,
6388     set_flood_vlans,
6389     is_mirror_output_bundle,
6390     forward_bpdu_changed,
6391     set_mac_table_config,
6392     set_realdev,
6393     NULL,                       /* meter_get_features */
6394     NULL,                       /* meter_set */
6395     NULL,                       /* meter_get */
6396     NULL,                       /* meter_del */
6397     NULL,                       /* group_alloc */
6398     NULL,                       /* group_construct */
6399     NULL,                       /* group_destruct */
6400     NULL,                       /* group_dealloc */
6401     NULL,                       /* group_modify */
6402     NULL,                       /* group_get_stats */
6403 };