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     /* Owners. */
 231     struct hmap_node hmap_node;  /* In owning ofproto's 'facets' hmap. */
 232     struct ofproto_dpif *ofproto;
 233
 234     /* Owned data. */
 235     struct list subfacets;
 236     long long int used;         /* Time last used; time created if not used. */
 237
 238     /* Key. */
 239     struct flow flow;           /* Flow of the creating subfacet. */
 240     struct cls_rule cr;         /* In 'ofproto_dpif's facets classifier. */
 241
 242     /* These statistics:
 243      *
 244      *   - Do include packets and bytes sent "by hand", e.g. with
 245      *     dpif_execute().
 246      *
 247      *   - Do include packets and bytes that were obtained from the datapath
 248      *     when a subfacet's statistics were reset (e.g. dpif_flow_put() with
 249      *     DPIF_FP_ZERO_STATS).
 250      *
 251      *   - Do not include packets or bytes that can be obtained from the
 252      *     datapath for any existing subfacet.
 253      */
 254     uint64_t packet_count;       /* Number of packets received. */
 255     uint64_t byte_count;         /* Number of bytes received. */
 256
 257     /* Resubmit statistics. */
 258     uint64_t prev_packet_count;  /* Number of packets from last stats push. */
 259     uint64_t prev_byte_count;    /* Number of bytes from last stats push. */
 260     long long int prev_used;     /* Used time from last stats push. */
 261
 262     /* Accounting. */
 263     uint64_t accounted_bytes;    /* Bytes processed by facet_account(). */
 264     struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
 265     uint8_t tcp_flags;           /* TCP flags seen for this 'rule'. */
 266
 267     struct xlate_out xout;
 268
 269     /* Storage for a single subfacet, to reduce malloc() time and space
 270      * overhead.  (A facet always has at least one subfacet and in the common
 271      * case has exactly one subfacet.  However, 'one_subfacet' may not
 272      * always be valid, since it could have been removed after newer
 273      * subfacets were pushed onto the 'subfacets' list.) */
 274     struct subfacet one_subfacet;
 275
 276     long long int learn_rl;      /* Rate limiter for facet_learn(). */
 277 };
 278
 279 static struct facet *facet_create(const struct flow_miss *);
 280 static void facet_remove(struct facet *);
 281 static void facet_free(struct facet *);
 282
 283 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
 284 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
 285                                         const struct flow *);
 286 static bool facet_revalidate(struct facet *);
 287 static bool facet_check_consistency(struct facet *);
 288
 289 static void facet_flush_stats(struct facet *);
 290
 291 static void facet_reset_counters(struct facet *);
 292 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
 293                             struct dpif_flow_stats *, bool may_learn);
 294 static void facet_push_stats(struct facet *, bool may_learn);
 295 static void facet_learn(struct facet *);
 296 static void facet_account(struct facet *);
 297 static void push_all_stats(void);
 298
 299 static bool facet_is_controller_flow(struct facet *);
 300
 301 struct ofport_dpif {
 302     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 303     struct ofport up;
 304
 305     odp_port_t odp_port;
 306     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 307     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
 308     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 309     struct bfd *bfd;            /* BFD, if any. */
 310     bool may_enable;            /* May be enabled in bonds. */
 311     bool is_tunnel;             /* This port is a tunnel. */
 312     long long int carrier_seq;  /* Carrier status changes. */
 313     struct ofport_dpif *peer;   /* Peer if patch port. */
 314
 315     /* Spanning tree. */
 316     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 317     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 318     long long int stp_state_entered;
 319
 320     /* Queue to DSCP mapping. */
 321     struct ofproto_port_queue *qdscp;
 322     size_t n_qdscp;
 323
 324     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 325      *
 326      * This is deprecated.  It is only for compatibility with broken device
 327      * drivers in old versions of Linux that do not properly support VLANs when
 328      * VLAN devices are not used.  When broken device drivers are no longer in
 329      * widespread use, we will delete these interfaces. */
 330     ofp_port_t realdev_ofp_port;
 331     int vlandev_vid;
 332 };
 333
 334 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 335  *
 336  * This is deprecated.  It is only for compatibility with broken device drivers
 337  * in old versions of Linux that do not properly support VLANs when VLAN
 338  * devices are not used.  When broken device drivers are no longer in
 339  * widespread use, we will delete these interfaces. */
 340 struct vlan_splinter {
 341     struct hmap_node realdev_vid_node;
 342     struct hmap_node vlandev_node;
 343     ofp_port_t realdev_ofp_port;
 344     ofp_port_t vlandev_ofp_port;
 345     int vid;
 346 };
 347
 348 static void vsp_remove(struct ofport_dpif *);
 349 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 350
 351 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 352                                        ofp_port_t);
 353
 354 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 355                                        odp_port_t);
 356
 357 static struct ofport_dpif *
 358 ofport_dpif_cast(const struct ofport *ofport)
 359 {
 360     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 361 }
 362
 363 static void port_run(struct ofport_dpif *);
 364 static void port_run_fast(struct ofport_dpif *);
 365 static void port_wait(struct ofport_dpif *);
 366 static int set_bfd(struct ofport *, const struct smap *);
 367 static int set_cfm(struct ofport *, const struct cfm_settings *);
 368 static void ofport_update_peer(struct ofport_dpif *);
 369 static void run_fast_rl(void);
 370 static int run_fast(struct ofproto *);
 371
 372 struct dpif_completion {
 373     struct list list_node;
 374     struct ofoperation *op;
 375 };
 376
 377 /* Reasons that we might need to revalidate every facet, and corresponding
 378  * coverage counters.
 379  *
 380  * A value of 0 means that there is no need to revalidate.
 381  *
 382  * It would be nice to have some cleaner way to integrate with coverage
 383  * counters, but with only a few reasons I guess this is good enough for
 384  * now. */
 385 enum revalidate_reason {
 386     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 387     REV_STP,                   /* Spanning tree protocol port status change. */
 388     REV_BOND,                  /* Bonding changed. */
 389     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 390     REV_FLOW_TABLE,            /* Flow table changed. */
 391     REV_MAC_LEARNING,          /* Mac learning changed. */
 392     REV_INCONSISTENCY          /* Facet self-check failed. */
 393 };
 394 COVERAGE_DEFINE(rev_reconfigure);
 395 COVERAGE_DEFINE(rev_stp);
 396 COVERAGE_DEFINE(rev_bond);
 397 COVERAGE_DEFINE(rev_port_toggled);
 398 COVERAGE_DEFINE(rev_flow_table);
 399 COVERAGE_DEFINE(rev_mac_learning);
 400 COVERAGE_DEFINE(rev_inconsistency);
 401
 402 struct avg_subfacet_rates {
 403     double add_rate;   /* Moving average of new flows created per minute. */
 404     double del_rate;   /* Moving average of flows deleted per minute. */
 405 };
 406
 407 /* All datapaths of a given type share a single dpif backer instance. */
 408 struct dpif_backer {
 409     char *type;
 410     int refcount;
 411     struct dpif *dpif;
 412     struct udpif *udpif;
 413     struct timer next_expiration;
 414
 415     struct ovs_rwlock odp_to_ofport_lock;
 416     struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
 417
 418     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 419
 420     /* Facet revalidation flags applying to facets which use this backer. */
 421     enum revalidate_reason need_revalidate; /* Revalidate every facet. */
 422
 423     struct hmap drop_keys; /* Set of dropped odp keys. */
 424     bool recv_set_enable; /* Enables or disables receiving packets. */
 425
 426     struct hmap subfacets;
 427     struct governor *governor;
 428
 429     /* Subfacet statistics.
 430      *
 431      * These keep track of the total number of subfacets added and deleted and
 432      * flow life span.  They are useful for computing the flow rates stats
 433      * exposed via "ovs-appctl dpif/show".  The goal is to learn about
 434      * traffic patterns in ways that we can use later to improve Open vSwitch
 435      * performance in new situations.  */
 436     long long int created;           /* Time when it is created. */
 437     unsigned max_n_subfacet;         /* Maximum number of flows */
 438     unsigned avg_n_subfacet;         /* Average number of flows. */
 439     long long int avg_subfacet_life; /* Average life span of subfacets. */
 440
 441     /* The average number of subfacets... */
 442     struct avg_subfacet_rates hourly;   /* ...over the last hour. */
 443     struct avg_subfacet_rates daily;    /* ...over the last day. */
 444     struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
 445     long long int last_minute;          /* Last time 'hourly' was updated. */
 446
 447     /* Number of subfacets added or deleted since 'last_minute'. */
 448     unsigned subfacet_add_count;
 449     unsigned subfacet_del_count;
 450
 451     /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
 452     unsigned long long int total_subfacet_add_count;
 453     unsigned long long int total_subfacet_del_count;
 454
 455     /* Number of upcall handling threads. */
 456     unsigned int n_handler_threads;
 457 };
 458
 459 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 460 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 461
 462 static void drop_key_clear(struct dpif_backer *);
 463 static void update_moving_averages(struct dpif_backer *backer);
 464
 465 struct ofproto_dpif {
 466     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 467     struct ofproto up;
 468     struct dpif_backer *backer;
 469
 470     /* Special OpenFlow rules. */
 471     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 472     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 473     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 474
 475     /* Bridging. */
 476     struct netflow *netflow;
 477     struct dpif_sflow *sflow;
 478     struct dpif_ipfix *ipfix;
 479     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 480     struct mac_learning *ml;
 481     bool has_bonded_bundles;
 482     struct mbridge *mbridge;
 483
 484     /* Facets. */
 485     struct classifier facets;     /* Contains 'struct facet's. */
 486     long long int consistency_rl;
 487
 488     /* Support for debugging async flow mods. */
 489     struct list completions;
 490
 491     struct netdev_stats stats; /* To account packets generated and consumed in
 492                                 * userspace. */
 493
 494     /* Spanning tree. */
 495     struct stp *stp;
 496     long long int stp_last_tick;
 497
 498     /* VLAN splinters. */
 499     struct ovs_mutex vsp_mutex;
 500     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 501     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 502
 503     /* Ports. */
 504     struct sset ports;             /* Set of standard port names. */
 505     struct sset ghost_ports;       /* Ports with no datapath port. */
 506     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 507     int port_poll_errno;           /* Last errno for port_poll() reply. */
 508
 509     /* Per ofproto's dpif stats. */
 510     uint64_t n_hit;
 511     uint64_t n_missed;
 512
 513     /* Work queues. */
 514     struct ovs_mutex flow_mod_mutex;
 515     struct list flow_mods OVS_GUARDED;
 516     size_t n_flow_mods OVS_GUARDED;
 517
 518     struct ovs_mutex pin_mutex;
 519     struct list pins OVS_GUARDED;
 520     size_t n_pins OVS_GUARDED;
 521 };
 522
 523 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"?  (Useful only
 524  * for debugging the asynchronous flow_mod implementation.) */
 525 static bool clogged;
 526
 527 /* By default, flows in the datapath are wildcarded (megaflows).  They
 528  * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
 529 static bool enable_megaflows = true;
 530
 531 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 532 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 533
 534 static void ofproto_dpif_unixctl_init(void);
 535
 536 static inline struct ofproto_dpif *
 537 ofproto_dpif_cast(const struct ofproto *ofproto)
 538 {
 539     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 540     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 541 }
 542
 543 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
 544                                         ofp_port_t ofp_port);
 545 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
 546                           const struct ofpbuf *packet, struct ds *);
 547
 548 /* Upcalls. */
 549 static void handle_upcalls(struct dpif_backer *);
 550
 551 /* Flow expiration. */
 552 static int expire(struct dpif_backer *);
 553
 554 /* NetFlow. */
 555 static void send_netflow_active_timeouts(struct ofproto_dpif *);
 556
 557 /* Utilities. */
 558 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
 559
 560 /* Global variables. */
 561 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 562
 563 /* Initial mappings of port to bridge mappings. */
 564 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 565
 566 /* Executes and takes ownership of 'fm'. */
 567 void
 568 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 569                       struct ofputil_flow_mod *fm)
 570 {
 571     ovs_mutex_lock(&ofproto->flow_mod_mutex);
 572     if (ofproto->n_flow_mods > 1024) {
 573         ovs_mutex_unlock(&ofproto->flow_mod_mutex);
 574         COVERAGE_INC(flow_mod_overflow);
 575         free(fm->ofpacts);
 576         free(fm);
 577         return;
 578     }
 579
 580     list_push_back(&ofproto->flow_mods, &fm->list_node);
 581     ofproto->n_flow_mods++;
 582     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
 583 }
 584
 585 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 586  * Takes ownership of 'pin' and pin->packet. */
 587 void
 588 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 589                             struct ofputil_packet_in *pin)
 590 {
 591     ovs_mutex_lock(&ofproto->pin_mutex);
 592     if (ofproto->n_pins > 1024) {
 593         ovs_mutex_unlock(&ofproto->pin_mutex);
 594         COVERAGE_INC(packet_in_overflow);
 595         free(CONST_CAST(void *, pin->packet));
 596         free(pin);
 597         return;
 598     }
 599
 600     list_push_back(&ofproto->pins, &pin->list_node);
 601     ofproto->n_pins++;
 602     ovs_mutex_unlock(&ofproto->pin_mutex);
 603 }
 604 \f
 605 /* Factory functions. */
 606
 607 static void
 608 init(const struct shash *iface_hints)
 609 {
 610     struct shash_node *node;
 611
 612     /* Make a local copy, since we don't own 'iface_hints' elements. */
 613     SHASH_FOR_EACH(node, iface_hints) {
 614         const struct iface_hint *orig_hint = node->data;
 615         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 616
 617         new_hint->br_name = xstrdup(orig_hint->br_name);
 618         new_hint->br_type = xstrdup(orig_hint->br_type);
 619         new_hint->ofp_port = orig_hint->ofp_port;
 620
 621         shash_add(&init_ofp_ports, node->name, new_hint);
 622     }
 623 }
 624
 625 static void
 626 enumerate_types(struct sset *types)
 627 {
 628     dp_enumerate_types(types);
 629 }
 630
 631 static int
 632 enumerate_names(const char *type, struct sset *names)
 633 {
 634     struct ofproto_dpif *ofproto;
 635
 636     sset_clear(names);
 637     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 638         if (strcmp(type, ofproto->up.type)) {
 639             continue;
 640         }
 641         sset_add(names, ofproto->up.name);
 642     }
 643
 644     return 0;
 645 }
 646
 647 static int
 648 del(const char *type, const char *name)
 649 {
 650     struct dpif *dpif;
 651     int error;
 652
 653     error = dpif_open(name, type, &dpif);
 654     if (!error) {
 655         error = dpif_delete(dpif);
 656         dpif_close(dpif);
 657     }
 658     return error;
 659 }
 660 \f
 661 static const char *
 662 port_open_type(const char *datapath_type, const char *port_type)
 663 {
 664     return dpif_port_open_type(datapath_type, port_type);
 665 }
 666
 667 /* Type functions. */
 668
 669 static void process_dpif_port_changes(struct dpif_backer *);
 670 static void process_dpif_all_ports_changed(struct dpif_backer *);
 671 static void process_dpif_port_change(struct dpif_backer *,
 672                                      const char *devname);
 673 static void process_dpif_port_error(struct dpif_backer *, int error);
 674
 675 static struct ofproto_dpif *
 676 lookup_ofproto_dpif_by_port_name(const char *name)
 677 {
 678     struct ofproto_dpif *ofproto;
 679
 680     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 681         if (sset_contains(&ofproto->ports, name)) {
 682             return ofproto;
 683         }
 684     }
 685
 686     return NULL;
 687 }
 688
 689 static int
 690 type_run(const char *type)
 691 {
 692     static long long int push_timer = LLONG_MIN;
 693     struct dpif_backer *backer;
 694
 695     backer = shash_find_data(&all_dpif_backers, type);
 696     if (!backer) {
 697         /* This is not necessarily a problem, since backers are only
 698          * created on demand. */
 699         return 0;
 700     }
 701
 702     dpif_run(backer->dpif);
 703
 704     /* The most natural place to push facet statistics is when they're pulled
 705      * from the datapath.  However, when there are many flows in the datapath,
 706      * this expensive operation can occur so frequently, that it reduces our
 707      * ability to quickly set up flows.  To reduce the cost, we push statistics
 708      * here instead. */
 709     if (time_msec() > push_timer) {
 710         push_timer = time_msec() + 2000;
 711         push_all_stats();
 712     }
 713
 714     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 715      * and the configuration has now changed to "false", enable receiving
 716      * packets from the datapath. */
 717     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 718         int error;
 719
 720         backer->recv_set_enable = true;
 721
 722         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 723         if (error) {
 724             udpif_recv_set(backer->udpif, 0, false);
 725             VLOG_ERR("Failed to enable receiving packets in dpif.");
 726             return error;
 727         }
 728         udpif_recv_set(backer->udpif, n_handler_threads,
 729                        backer->recv_set_enable);
 730         dpif_flow_flush(backer->dpif);
 731         backer->need_revalidate = REV_RECONFIGURE;
 732     }
 733
 734     /* If the n_handler_threads is reconfigured, call udpif_recv_set()
 735      * to reset the handler threads. */
 736     if (backer->n_handler_threads != n_handler_threads) {
 737         udpif_recv_set(backer->udpif, n_handler_threads,
 738                        backer->recv_set_enable);
 739         backer->n_handler_threads = n_handler_threads;
 740     }
 741
 742     if (backer->need_revalidate) {
 743         struct ofproto_dpif *ofproto;
 744         struct simap_node *node;
 745         struct simap tmp_backers;
 746
 747         /* Handle tunnel garbage collection. */
 748         simap_init(&tmp_backers);
 749         simap_swap(&backer->tnl_backers, &tmp_backers);
 750
 751         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 752             struct ofport_dpif *iter;
 753
 754             if (backer != ofproto->backer) {
 755                 continue;
 756             }
 757
 758             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 759                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 760                 const char *dp_port;
 761
 762                 if (!iter->is_tunnel) {
 763                     continue;
 764                 }
 765
 766                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 767                                                      namebuf, sizeof namebuf);
 768                 node = simap_find(&tmp_backers, dp_port);
 769                 if (node) {
 770                     simap_put(&backer->tnl_backers, dp_port, node->data);
 771                     simap_delete(&tmp_backers, node);
 772                     node = simap_find(&backer->tnl_backers, dp_port);
 773                 } else {
 774                     node = simap_find(&backer->tnl_backers, dp_port);
 775                     if (!node) {
 776                         odp_port_t odp_port = ODPP_NONE;
 777
 778                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 779                                            &odp_port)) {
 780                             simap_put(&backer->tnl_backers, dp_port,
 781                                       odp_to_u32(odp_port));
 782                             node = simap_find(&backer->tnl_backers, dp_port);
 783                         }
 784                     }
 785                 }
 786
 787                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 788                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 789                                          iter->odp_port)) {
 790                     backer->need_revalidate = REV_RECONFIGURE;
 791                 }
 792             }
 793         }
 794
 795         SIMAP_FOR_EACH (node, &tmp_backers) {
 796             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 797         }
 798         simap_destroy(&tmp_backers);
 799
 800         switch (backer->need_revalidate) {
 801         case REV_RECONFIGURE:   COVERAGE_INC(rev_reconfigure);   break;
 802         case REV_STP:           COVERAGE_INC(rev_stp);           break;
 803         case REV_BOND:          COVERAGE_INC(rev_bond);          break;
 804         case REV_PORT_TOGGLED:  COVERAGE_INC(rev_port_toggled);  break;
 805         case REV_FLOW_TABLE:    COVERAGE_INC(rev_flow_table);    break;
 806         case REV_MAC_LEARNING:  COVERAGE_INC(rev_mac_learning);  break;
 807         case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
 808         }
 809         backer->need_revalidate = 0;
 810
 811         /* Clear the drop_keys in case we should now be accepting some
 812          * formerly dropped flows. */
 813         drop_key_clear(backer);
 814
 815         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 816             struct facet *facet, *next;
 817             struct ofport_dpif *ofport;
 818             struct cls_cursor cursor;
 819             struct ofbundle *bundle;
 820
 821             if (ofproto->backer != backer) {
 822                 continue;
 823             }
 824
 825             ovs_rwlock_wrlock(&xlate_rwlock);
 826             xlate_ofproto_set(ofproto, ofproto->up.name,
 827                               ofproto->backer->dpif, ofproto->miss_rule,
 828                               ofproto->no_packet_in_rule, ofproto->ml,
 829                               ofproto->stp, ofproto->mbridge,
 830                               ofproto->sflow, ofproto->ipfix,
 831                               ofproto->up.frag_handling,
 832                               ofproto->up.forward_bpdu,
 833                               connmgr_has_in_band(ofproto->up.connmgr),
 834                               ofproto->netflow != NULL);
 835
 836             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 837                 xlate_bundle_set(ofproto, bundle, bundle->name,
 838                                  bundle->vlan_mode, bundle->vlan,
 839                                  bundle->trunks, bundle->use_priority_tags,
 840                                  bundle->bond, bundle->lacp,
 841                                  bundle->floodable);
 842             }
 843
 844             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 845                 int stp_port = ofport->stp_port
 846                     ? stp_port_no(ofport->stp_port)
 847                     : -1;
 848                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 849                                  ofport->up.ofp_port, ofport->odp_port,
 850                                  ofport->up.netdev, ofport->cfm,
 851                                  ofport->bfd, ofport->peer, stp_port,
 852                                  ofport->qdscp, ofport->n_qdscp,
 853                                  ofport->up.pp.config, ofport->is_tunnel,
 854                                  ofport->may_enable);
 855             }
 856             ovs_rwlock_unlock(&xlate_rwlock);
 857
 858             /* Only ofproto-dpif cares about the facet classifier so we just
 859              * lock cls_cursor_init() to appease the thread safety analysis. */
 860             ovs_rwlock_rdlock(&ofproto->facets.rwlock);
 861             cls_cursor_init(&cursor, &ofproto->facets, NULL);
 862             ovs_rwlock_unlock(&ofproto->facets.rwlock);
 863             CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
 864                 facet_revalidate(facet);
 865                 run_fast_rl();
 866             }
 867         }
 868
 869         udpif_revalidate(backer->udpif);
 870     }
 871
 872     if (!backer->recv_set_enable) {
 873         /* Wake up before a max of 1000ms. */
 874         timer_set_duration(&backer->next_expiration, 1000);
 875     } else if (timer_expired(&backer->next_expiration)) {
 876         int delay = expire(backer);
 877         timer_set_duration(&backer->next_expiration, delay);
 878     }
 879
 880     process_dpif_port_changes(backer);
 881
 882     if (backer->governor) {
 883         size_t n_subfacets;
 884
 885         governor_run(backer->governor);
 886
 887         /* If the governor has shrunk to its minimum size and the number of
 888          * subfacets has dwindled, then drop the governor entirely.
 889          *
 890          * For hysteresis, the number of subfacets to drop the governor is
 891          * smaller than the number needed to trigger its creation. */
 892         n_subfacets = hmap_count(&backer->subfacets);
 893         if (n_subfacets * 4 < flow_eviction_threshold
 894             && governor_is_idle(backer->governor)) {
 895             governor_destroy(backer->governor);
 896             backer->governor = NULL;
 897         }
 898     }
 899
 900     return 0;
 901 }
 902
 903 /* Check for and handle port changes in 'backer''s dpif. */
 904 static void
 905 process_dpif_port_changes(struct dpif_backer *backer)
 906 {
 907     for (;;) {
 908         char *devname;
 909         int error;
 910
 911         error = dpif_port_poll(backer->dpif, &devname);
 912         switch (error) {
 913         case EAGAIN:
 914             return;
 915
 916         case ENOBUFS:
 917             process_dpif_all_ports_changed(backer);
 918             break;
 919
 920         case 0:
 921             process_dpif_port_change(backer, devname);
 922             free(devname);
 923             break;
 924
 925         default:
 926             process_dpif_port_error(backer, error);
 927             break;
 928         }
 929     }
 930 }
 931
 932 static void
 933 process_dpif_all_ports_changed(struct dpif_backer *backer)
 934 {
 935     struct ofproto_dpif *ofproto;
 936     struct dpif_port dpif_port;
 937     struct dpif_port_dump dump;
 938     struct sset devnames;
 939     const char *devname;
 940
 941     sset_init(&devnames);
 942     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 943         if (ofproto->backer == backer) {
 944             struct ofport *ofport;
 945
 946             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 947                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 948             }
 949         }
 950     }
 951     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 952         sset_add(&devnames, dpif_port.name);
 953     }
 954
 955     SSET_FOR_EACH (devname, &devnames) {
 956         process_dpif_port_change(backer, devname);
 957     }
 958     sset_destroy(&devnames);
 959 }
 960
 961 static void
 962 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 963 {
 964     struct ofproto_dpif *ofproto;
 965     struct dpif_port port;
 966
 967     /* Don't report on the datapath's device. */
 968     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 969         return;
 970     }
 971
 972     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 973                    &all_ofproto_dpifs) {
 974         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 975             return;
 976         }
 977     }
 978
 979     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 980     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 981         /* The port was removed.  If we know the datapath,
 982          * report it through poll_set().  If we don't, it may be
 983          * notifying us of a removal we initiated, so ignore it.
 984          * If there's a pending ENOBUFS, let it stand, since
 985          * everything will be reevaluated. */
 986         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 987             sset_add(&ofproto->port_poll_set, devname);
 988             ofproto->port_poll_errno = 0;
 989         }
 990     } else if (!ofproto) {
 991         /* The port was added, but we don't know with which
 992          * ofproto we should associate it.  Delete it. */
 993         dpif_port_del(backer->dpif, port.port_no);
 994     } else {
 995         struct ofport_dpif *ofport;
 996
 997         ofport = ofport_dpif_cast(shash_find_data(
 998                                       &ofproto->up.port_by_name, devname));
 999         if (ofport
1000             && ofport->odp_port != port.port_no
1001             && !odp_port_to_ofport(backer, port.port_no))
1002         {
1003             /* 'ofport''s datapath port number has changed from
1004              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
1005              * structures to match. */
1006             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
1007             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
1008             ofport->odp_port = port.port_no;
1009             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
1010                         hash_odp_port(port.port_no));
1011             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
1012             backer->need_revalidate = REV_RECONFIGURE;
1013         }
1014     }
1015     dpif_port_destroy(&port);
1016 }
1017
1018 /* Propagate 'error' to all ofprotos based on 'backer'. */
1019 static void
1020 process_dpif_port_error(struct dpif_backer *backer, int error)
1021 {
1022     struct ofproto_dpif *ofproto;
1023
1024     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1025         if (ofproto->backer == backer) {
1026             sset_clear(&ofproto->port_poll_set);
1027             ofproto->port_poll_errno = error;
1028         }
1029     }
1030 }
1031
1032 static int
1033 dpif_backer_run_fast(struct dpif_backer *backer)
1034 {
1035     udpif_run(backer->udpif);
1036     handle_upcalls(backer);
1037
1038     return 0;
1039 }
1040
1041 static int
1042 type_run_fast(const char *type)
1043 {
1044     struct dpif_backer *backer;
1045
1046     backer = shash_find_data(&all_dpif_backers, type);
1047     if (!backer) {
1048         /* This is not necessarily a problem, since backers are only
1049          * created on demand. */
1050         return 0;
1051     }
1052
1053     return dpif_backer_run_fast(backer);
1054 }
1055
1056 static void
1057 run_fast_rl(void)
1058 {
1059     static long long int port_rl = LLONG_MIN;
1060
1061     if (time_msec() >= port_rl) {
1062         struct ofproto_dpif *ofproto;
1063
1064         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1065             run_fast(&ofproto->up);
1066         }
1067         port_rl = time_msec() + 200;
1068     }
1069 }
1070
1071 static void
1072 type_wait(const char *type)
1073 {
1074     struct dpif_backer *backer;
1075
1076     backer = shash_find_data(&all_dpif_backers, type);
1077     if (!backer) {
1078         /* This is not necessarily a problem, since backers are only
1079          * created on demand. */
1080         return;
1081     }
1082
1083     if (backer->governor) {
1084         governor_wait(backer->governor);
1085     }
1086
1087     timer_wait(&backer->next_expiration);
1088     dpif_wait(backer->dpif);
1089     udpif_wait(backer->udpif);
1090 }
1091 \f
1092 /* Basic life-cycle. */
1093
1094 static int add_internal_flows(struct ofproto_dpif *);
1095
1096 static struct ofproto *
1097 alloc(void)
1098 {
1099     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1100     return &ofproto->up;
1101 }
1102
1103 static void
1104 dealloc(struct ofproto *ofproto_)
1105 {
1106     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1107     free(ofproto);
1108 }
1109
1110 static void
1111 close_dpif_backer(struct dpif_backer *backer)
1112 {
1113     struct shash_node *node;
1114
1115     ovs_assert(backer->refcount > 0);
1116
1117     if (--backer->refcount) {
1118         return;
1119     }
1120
1121     drop_key_clear(backer);
1122     hmap_destroy(&backer->drop_keys);
1123
1124     simap_destroy(&backer->tnl_backers);
1125     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1126     hmap_destroy(&backer->odp_to_ofport_map);
1127     node = shash_find(&all_dpif_backers, backer->type);
1128     free(backer->type);
1129     shash_delete(&all_dpif_backers, node);
1130     udpif_destroy(backer->udpif);
1131     dpif_close(backer->dpif);
1132
1133     ovs_assert(hmap_is_empty(&backer->subfacets));
1134     hmap_destroy(&backer->subfacets);
1135     governor_destroy(backer->governor);
1136
1137     free(backer);
1138 }
1139
1140 /* Datapath port slated for removal from datapath. */
1141 struct odp_garbage {
1142     struct list list_node;
1143     odp_port_t odp_port;
1144 };
1145
1146 static int
1147 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1148 {
1149     struct dpif_backer *backer;
1150     struct dpif_port_dump port_dump;
1151     struct dpif_port port;
1152     struct shash_node *node;
1153     struct list garbage_list;
1154     struct odp_garbage *garbage, *next;
1155     struct sset names;
1156     char *backer_name;
1157     const char *name;
1158     int error;
1159
1160     backer = shash_find_data(&all_dpif_backers, type);
1161     if (backer) {
1162         backer->refcount++;
1163         *backerp = backer;
1164         return 0;
1165     }
1166
1167     backer_name = xasprintf("ovs-%s", type);
1168
1169     /* Remove any existing datapaths, since we assume we're the only
1170      * userspace controlling the datapath. */
1171     sset_init(&names);
1172     dp_enumerate_names(type, &names);
1173     SSET_FOR_EACH(name, &names) {
1174         struct dpif *old_dpif;
1175
1176         /* Don't remove our backer if it exists. */
1177         if (!strcmp(name, backer_name)) {
1178             continue;
1179         }
1180
1181         if (dpif_open(name, type, &old_dpif)) {
1182             VLOG_WARN("couldn't open old datapath %s to remove it", name);
1183         } else {
1184             dpif_delete(old_dpif);
1185             dpif_close(old_dpif);
1186         }
1187     }
1188     sset_destroy(&names);
1189
1190     backer = xmalloc(sizeof *backer);
1191
1192     error = dpif_create_and_open(backer_name, type, &backer->dpif);
1193     free(backer_name);
1194     if (error) {
1195         VLOG_ERR("failed to open datapath of type %s: %s", type,
1196                  ovs_strerror(error));
1197         free(backer);
1198         return error;
1199     }
1200     backer->udpif = udpif_create(backer, backer->dpif);
1201
1202     backer->type = xstrdup(type);
1203     backer->governor = NULL;
1204     backer->refcount = 1;
1205     hmap_init(&backer->odp_to_ofport_map);
1206     ovs_rwlock_init(&backer->odp_to_ofport_lock);
1207     hmap_init(&backer->drop_keys);
1208     hmap_init(&backer->subfacets);
1209     timer_set_duration(&backer->next_expiration, 1000);
1210     backer->need_revalidate = 0;
1211     simap_init(&backer->tnl_backers);
1212     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1213     *backerp = backer;
1214
1215     if (backer->recv_set_enable) {
1216         dpif_flow_flush(backer->dpif);
1217     }
1218
1219     /* Loop through the ports already on the datapath and remove any
1220      * that we don't need anymore. */
1221     list_init(&garbage_list);
1222     dpif_port_dump_start(&port_dump, backer->dpif);
1223     while (dpif_port_dump_next(&port_dump, &port)) {
1224         node = shash_find(&init_ofp_ports, port.name);
1225         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1226             garbage = xmalloc(sizeof *garbage);
1227             garbage->odp_port = port.port_no;
1228             list_push_front(&garbage_list, &garbage->list_node);
1229         }
1230     }
1231     dpif_port_dump_done(&port_dump);
1232
1233     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1234         dpif_port_del(backer->dpif, garbage->odp_port);
1235         list_remove(&garbage->list_node);
1236         free(garbage);
1237     }
1238
1239     shash_add(&all_dpif_backers, type, backer);
1240
1241     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1242     if (error) {
1243         VLOG_ERR("failed to listen on datapath of type %s: %s",
1244                  type, ovs_strerror(error));
1245         close_dpif_backer(backer);
1246         return error;
1247     }
1248     udpif_recv_set(backer->udpif, n_handler_threads,
1249                    backer->recv_set_enable);
1250     backer->n_handler_threads = n_handler_threads;
1251
1252     backer->max_n_subfacet = 0;
1253     backer->created = time_msec();
1254     backer->last_minute = backer->created;
1255     memset(&backer->hourly, 0, sizeof backer->hourly);
1256     memset(&backer->daily, 0, sizeof backer->daily);
1257     memset(&backer->lifetime, 0, sizeof backer->lifetime);
1258     backer->subfacet_add_count = 0;
1259     backer->subfacet_del_count = 0;
1260     backer->total_subfacet_add_count = 0;
1261     backer->total_subfacet_del_count = 0;
1262     backer->avg_n_subfacet = 0;
1263     backer->avg_subfacet_life = 0;
1264
1265     return error;
1266 }
1267
1268 static int
1269 construct(struct ofproto *ofproto_)
1270 {
1271     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1272     struct shash_node *node, *next;
1273     uint32_t max_ports;
1274     int error;
1275
1276     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1277     if (error) {
1278         return error;
1279     }
1280
1281     max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1282     ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX)));
1283
1284     ofproto->netflow = NULL;
1285     ofproto->sflow = NULL;
1286     ofproto->ipfix = NULL;
1287     ofproto->stp = NULL;
1288     hmap_init(&ofproto->bundles);
1289     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1290     ofproto->mbridge = mbridge_create();
1291     ofproto->has_bonded_bundles = false;
1292     ovs_mutex_init(&ofproto->vsp_mutex);
1293
1294     classifier_init(&ofproto->facets);
1295     ofproto->consistency_rl = LLONG_MIN;
1296
1297     list_init(&ofproto->completions);
1298
1299     ovs_mutex_init(&ofproto->flow_mod_mutex);
1300     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1301     list_init(&ofproto->flow_mods);
1302     ofproto->n_flow_mods = 0;
1303     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1304
1305     ovs_mutex_init(&ofproto->pin_mutex);
1306     ovs_mutex_lock(&ofproto->pin_mutex);
1307     list_init(&ofproto->pins);
1308     ofproto->n_pins = 0;
1309     ovs_mutex_unlock(&ofproto->pin_mutex);
1310
1311     ofproto_dpif_unixctl_init();
1312
1313     hmap_init(&ofproto->vlandev_map);
1314     hmap_init(&ofproto->realdev_vid_map);
1315
1316     sset_init(&ofproto->ports);
1317     sset_init(&ofproto->ghost_ports);
1318     sset_init(&ofproto->port_poll_set);
1319     ofproto->port_poll_errno = 0;
1320
1321     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1322         struct iface_hint *iface_hint = node->data;
1323
1324         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1325             /* Check if the datapath already has this port. */
1326             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1327                 sset_add(&ofproto->ports, node->name);
1328             }
1329
1330             free(iface_hint->br_name);
1331             free(iface_hint->br_type);
1332             free(iface_hint);
1333             shash_delete(&init_ofp_ports, node);
1334         }
1335     }
1336
1337     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1338                 hash_string(ofproto->up.name, 0));
1339     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1340
1341     ofproto_init_tables(ofproto_, N_TABLES);
1342     error = add_internal_flows(ofproto);
1343     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1344
1345     ofproto->n_hit = 0;
1346     ofproto->n_missed = 0;
1347
1348     return error;
1349 }
1350
1351 static int
1352 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1353                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1354 {
1355     struct ofputil_flow_mod fm;
1356     int error;
1357
1358     match_init_catchall(&fm.match);
1359     fm.priority = 0;
1360     match_set_reg(&fm.match, 0, id);
1361     fm.new_cookie = htonll(0);
1362     fm.cookie = htonll(0);
1363     fm.cookie_mask = htonll(0);
1364     fm.modify_cookie = false;
1365     fm.table_id = TBL_INTERNAL;
1366     fm.command = OFPFC_ADD;
1367     fm.idle_timeout = 0;
1368     fm.hard_timeout = 0;
1369     fm.buffer_id = 0;
1370     fm.out_port = 0;
1371     fm.flags = 0;
1372     fm.ofpacts = ofpacts->data;
1373     fm.ofpacts_len = ofpacts->size;
1374
1375     error = ofproto_flow_mod(&ofproto->up, &fm);
1376     if (error) {
1377         VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1378                     id, ofperr_to_string(error));
1379         return error;
1380     }
1381
1382     if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1383                                   rulep)) {
1384         rule_dpif_release(*rulep);
1385     } else {
1386         NOT_REACHED();
1387     }
1388
1389     return 0;
1390 }
1391
1392 static int
1393 add_internal_flows(struct ofproto_dpif *ofproto)
1394 {
1395     struct ofpact_controller *controller;
1396     uint64_t ofpacts_stub[128 / 8];
1397     struct ofpbuf ofpacts;
1398     int error;
1399     int id;
1400
1401     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1402     id = 1;
1403
1404     controller = ofpact_put_CONTROLLER(&ofpacts);
1405     controller->max_len = UINT16_MAX;
1406     controller->controller_id = 0;
1407     controller->reason = OFPR_NO_MATCH;
1408     ofpact_pad(&ofpacts);
1409
1410     error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1411     if (error) {
1412         return error;
1413     }
1414
1415     ofpbuf_clear(&ofpacts);
1416     error = add_internal_flow(ofproto, id++, &ofpacts,
1417                               &ofproto->no_packet_in_rule);
1418     if (error) {
1419         return error;
1420     }
1421
1422     error = add_internal_flow(ofproto, id++, &ofpacts,
1423                               &ofproto->drop_frags_rule);
1424     return error;
1425 }
1426
1427 static void
1428 complete_operations(struct ofproto_dpif *ofproto)
1429 {
1430     struct dpif_completion *c, *next;
1431
1432     LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1433         ofoperation_complete(c->op, 0);
1434         list_remove(&c->list_node);
1435         free(c);
1436     }
1437 }
1438
1439 static void
1440 destruct(struct ofproto *ofproto_)
1441 {
1442     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1443     struct rule_dpif *rule, *next_rule;
1444     struct ofputil_packet_in *pin, *next_pin;
1445     struct ofputil_flow_mod *fm, *next_fm;
1446     struct facet *facet, *next_facet;
1447     struct cls_cursor cursor;
1448     struct oftable *table;
1449
1450     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1451     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1452     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1453     CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1454         facet_remove(facet);
1455     }
1456
1457     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1458     ovs_rwlock_wrlock(&xlate_rwlock);
1459     xlate_remove_ofproto(ofproto);
1460     ovs_rwlock_unlock(&xlate_rwlock);
1461
1462     flow_miss_batch_ofproto_destroyed(ofproto->backer->udpif, ofproto);
1463
1464     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1465     complete_operations(ofproto);
1466
1467     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1468         struct cls_cursor cursor;
1469
1470         ovs_rwlock_wrlock(&table->cls.rwlock);
1471         cls_cursor_init(&cursor, &table->cls, NULL);
1472         CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1473             ofproto_rule_delete(&ofproto->up, &table->cls, &rule->up);
1474         }
1475         ovs_rwlock_unlock(&table->cls.rwlock);
1476     }
1477     complete_operations(ofproto);
1478
1479     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1480     LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &ofproto->flow_mods) {
1481         list_remove(&fm->list_node);
1482         ofproto->n_flow_mods--;
1483         free(fm->ofpacts);
1484         free(fm);
1485     }
1486     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1487     ovs_mutex_destroy(&ofproto->flow_mod_mutex);
1488
1489     ovs_mutex_lock(&ofproto->pin_mutex);
1490     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &ofproto->pins) {
1491         list_remove(&pin->list_node);
1492         ofproto->n_pins--;
1493         free(CONST_CAST(void *, pin->packet));
1494         free(pin);
1495     }
1496     ovs_mutex_unlock(&ofproto->pin_mutex);
1497     ovs_mutex_destroy(&ofproto->pin_mutex);
1498
1499     mbridge_unref(ofproto->mbridge);
1500
1501     netflow_destroy(ofproto->netflow);
1502     dpif_sflow_unref(ofproto->sflow);
1503     hmap_destroy(&ofproto->bundles);
1504     mac_learning_unref(ofproto->ml);
1505
1506     classifier_destroy(&ofproto->facets);
1507
1508     hmap_destroy(&ofproto->vlandev_map);
1509     hmap_destroy(&ofproto->realdev_vid_map);
1510
1511     sset_destroy(&ofproto->ports);
1512     sset_destroy(&ofproto->ghost_ports);
1513     sset_destroy(&ofproto->port_poll_set);
1514
1515     ovs_mutex_destroy(&ofproto->vsp_mutex);
1516
1517     close_dpif_backer(ofproto->backer);
1518 }
1519
1520 static int
1521 run_fast(struct ofproto *ofproto_)
1522 {
1523     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1524     struct ofputil_packet_in *pin, *next_pin;
1525     struct ofputil_flow_mod *fm, *next_fm;
1526     struct list flow_mods, pins;
1527     struct ofport_dpif *ofport;
1528
1529     /* Do not perform any periodic activity required by 'ofproto' while
1530      * waiting for flow restore to complete. */
1531     if (ofproto_get_flow_restore_wait()) {
1532         return 0;
1533     }
1534
1535     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1536     if (ofproto->n_flow_mods) {
1537         flow_mods = ofproto->flow_mods;
1538         list_moved(&flow_mods);
1539         list_init(&ofproto->flow_mods);
1540         ofproto->n_flow_mods = 0;
1541     } else {
1542         list_init(&flow_mods);
1543     }
1544     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1545
1546     LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &flow_mods) {
1547         int error = ofproto_flow_mod(&ofproto->up, fm);
1548         if (error && !VLOG_DROP_WARN(&rl)) {
1549             VLOG_WARN("learning action failed to modify flow table (%s)",
1550                       ofperr_get_name(error));
1551         }
1552
1553         list_remove(&fm->list_node);
1554         free(fm->ofpacts);
1555         free(fm);
1556     }
1557
1558     ovs_mutex_lock(&ofproto->pin_mutex);
1559     if (ofproto->n_pins) {
1560         pins = ofproto->pins;
1561         list_moved(&pins);
1562         list_init(&ofproto->pins);
1563         ofproto->n_pins = 0;
1564     } else {
1565         list_init(&pins);
1566     }
1567     ovs_mutex_unlock(&ofproto->pin_mutex);
1568
1569     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1570         connmgr_send_packet_in(ofproto->up.connmgr, pin);
1571         list_remove(&pin->list_node);
1572         free(CONST_CAST(void *, pin->packet));
1573         free(pin);
1574     }
1575
1576     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1577         port_run_fast(ofport);
1578     }
1579
1580     return 0;
1581 }
1582
1583 static int
1584 run(struct ofproto *ofproto_)
1585 {
1586     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1587     struct ofport_dpif *ofport;
1588     struct ofbundle *bundle;
1589     int error;
1590
1591     if (!clogged) {
1592         complete_operations(ofproto);
1593     }
1594
1595     if (mbridge_need_revalidate(ofproto->mbridge)) {
1596         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1597         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1598         mac_learning_flush(ofproto->ml);
1599         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1600     }
1601
1602     /* Do not perform any periodic activity below required by 'ofproto' while
1603      * waiting for flow restore to complete. */
1604     if (ofproto_get_flow_restore_wait()) {
1605         return 0;
1606     }
1607
1608     error = run_fast(ofproto_);
1609     if (error) {
1610         return error;
1611     }
1612
1613     if (ofproto->netflow) {
1614         if (netflow_run(ofproto->netflow)) {
1615             send_netflow_active_timeouts(ofproto);
1616         }
1617     }
1618     if (ofproto->sflow) {
1619         dpif_sflow_run(ofproto->sflow);
1620     }
1621     if (ofproto->ipfix) {
1622         dpif_ipfix_run(ofproto->ipfix);
1623     }
1624
1625     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1626         port_run(ofport);
1627     }
1628     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1629         bundle_run(bundle);
1630     }
1631
1632     stp_run(ofproto);
1633     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1634     if (mac_learning_run(ofproto->ml)) {
1635         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1636     }
1637     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1638
1639     /* Check the consistency of a random facet, to aid debugging. */
1640     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1641     if (time_msec() >= ofproto->consistency_rl
1642         && !classifier_is_empty(&ofproto->facets)
1643         && !ofproto->backer->need_revalidate) {
1644         struct cls_table *table;
1645         struct cls_rule *cr;
1646         struct facet *facet;
1647
1648         ofproto->consistency_rl = time_msec() + 250;
1649
1650         table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1651                              struct cls_table, hmap_node);
1652         cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1653                           hmap_node);
1654         facet = CONTAINER_OF(cr, struct facet, cr);
1655
1656         if (!facet_check_consistency(facet)) {
1657             ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1658         }
1659     }
1660     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1661
1662     return 0;
1663 }
1664
1665 static void
1666 wait(struct ofproto *ofproto_)
1667 {
1668     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1669     struct ofport_dpif *ofport;
1670     struct ofbundle *bundle;
1671
1672     if (!clogged && !list_is_empty(&ofproto->completions)) {
1673         poll_immediate_wake();
1674     }
1675
1676     if (ofproto_get_flow_restore_wait()) {
1677         return;
1678     }
1679
1680     if (ofproto->sflow) {
1681         dpif_sflow_wait(ofproto->sflow);
1682     }
1683     if (ofproto->ipfix) {
1684         dpif_ipfix_wait(ofproto->ipfix);
1685     }
1686     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1687         port_wait(ofport);
1688     }
1689     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1690         bundle_wait(bundle);
1691     }
1692     if (ofproto->netflow) {
1693         netflow_wait(ofproto->netflow);
1694     }
1695     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1696     mac_learning_wait(ofproto->ml);
1697     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1698     stp_wait(ofproto);
1699     if (ofproto->backer->need_revalidate) {
1700         /* Shouldn't happen, but if it does just go around again. */
1701         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1702         poll_immediate_wake();
1703     }
1704 }
1705
1706 static void
1707 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1708 {
1709     const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1710     struct cls_cursor cursor;
1711     size_t n_subfacets = 0;
1712     struct facet *facet;
1713
1714     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1715     simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1716     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1717
1718     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1719     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1720     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1721         n_subfacets += list_size(&facet->subfacets);
1722     }
1723     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1724     simap_increase(usage, "subfacets", n_subfacets);
1725 }
1726
1727 static void
1728 flush(struct ofproto *ofproto_)
1729 {
1730     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1731     struct subfacet *subfacet, *next_subfacet;
1732     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1733     int n_batch;
1734
1735     n_batch = 0;
1736     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1737                         &ofproto->backer->subfacets) {
1738         if (subfacet->facet->ofproto != ofproto) {
1739             continue;
1740         }
1741
1742         if (subfacet->path != SF_NOT_INSTALLED) {
1743             batch[n_batch++] = subfacet;
1744             if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1745                 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1746                 n_batch = 0;
1747             }
1748         } else {
1749             subfacet_destroy(subfacet);
1750         }
1751     }
1752
1753     if (n_batch > 0) {
1754         subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1755     }
1756 }
1757
1758 static void
1759 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1760              bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1761 {
1762     *arp_match_ip = true;
1763     *actions = (OFPUTIL_A_OUTPUT |
1764                 OFPUTIL_A_SET_VLAN_VID |
1765                 OFPUTIL_A_SET_VLAN_PCP |
1766                 OFPUTIL_A_STRIP_VLAN |
1767                 OFPUTIL_A_SET_DL_SRC |
1768                 OFPUTIL_A_SET_DL_DST |
1769                 OFPUTIL_A_SET_NW_SRC |
1770                 OFPUTIL_A_SET_NW_DST |
1771                 OFPUTIL_A_SET_NW_TOS |
1772                 OFPUTIL_A_SET_TP_SRC |
1773                 OFPUTIL_A_SET_TP_DST |
1774                 OFPUTIL_A_ENQUEUE);
1775 }
1776
1777 static void
1778 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1779 {
1780     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1781     struct dpif_dp_stats s;
1782     uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1783     uint64_t n_lookup;
1784
1785     strcpy(ots->name, "classifier");
1786
1787     dpif_get_dp_stats(ofproto->backer->dpif, &s);
1788     rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1789     rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1790     rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1791
1792     n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1793     ots->lookup_count = htonll(n_lookup);
1794     ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1795 }
1796
1797 static struct ofport *
1798 port_alloc(void)
1799 {
1800     struct ofport_dpif *port = xmalloc(sizeof *port);
1801     return &port->up;
1802 }
1803
1804 static void
1805 port_dealloc(struct ofport *port_)
1806 {
1807     struct ofport_dpif *port = ofport_dpif_cast(port_);
1808     free(port);
1809 }
1810
1811 static int
1812 port_construct(struct ofport *port_)
1813 {
1814     struct ofport_dpif *port = ofport_dpif_cast(port_);
1815     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1816     const struct netdev *netdev = port->up.netdev;
1817     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1818     struct dpif_port dpif_port;
1819     int error;
1820
1821     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1822     port->bundle = NULL;
1823     port->cfm = NULL;
1824     port->bfd = NULL;
1825     port->may_enable = true;
1826     port->stp_port = NULL;
1827     port->stp_state = STP_DISABLED;
1828     port->is_tunnel = false;
1829     port->peer = NULL;
1830     port->qdscp = NULL;
1831     port->n_qdscp = 0;
1832     port->realdev_ofp_port = 0;
1833     port->vlandev_vid = 0;
1834     port->carrier_seq = netdev_get_carrier_resets(netdev);
1835
1836     if (netdev_vport_is_patch(netdev)) {
1837         /* By bailing out here, we don't submit the port to the sFlow module
1838          * to be considered for counter polling export.  This is correct
1839          * because the patch port represents an interface that sFlow considers
1840          * to be "internal" to the switch as a whole, and therefore not an
1841          * candidate for counter polling. */
1842         port->odp_port = ODPP_NONE;
1843         ofport_update_peer(port);
1844         return 0;
1845     }
1846
1847     error = dpif_port_query_by_name(ofproto->backer->dpif,
1848                                     netdev_vport_get_dpif_port(netdev, namebuf,
1849                                                                sizeof namebuf),
1850                                     &dpif_port);
1851     if (error) {
1852         return error;
1853     }
1854
1855     port->odp_port = dpif_port.port_no;
1856
1857     if (netdev_get_tunnel_config(netdev)) {
1858         tnl_port_add(port, port->up.netdev, port->odp_port);
1859         port->is_tunnel = true;
1860     } else {
1861         /* Sanity-check that a mapping doesn't already exist.  This
1862          * shouldn't happen for non-tunnel ports. */
1863         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1864             VLOG_ERR("port %s already has an OpenFlow port number",
1865                      dpif_port.name);
1866             dpif_port_destroy(&dpif_port);
1867             return EBUSY;
1868         }
1869
1870         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1871         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1872                     hash_odp_port(port->odp_port));
1873         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1874     }
1875     dpif_port_destroy(&dpif_port);
1876
1877     if (ofproto->sflow) {
1878         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1879     }
1880
1881     return 0;
1882 }
1883
1884 static void
1885 port_destruct(struct ofport *port_)
1886 {
1887     struct ofport_dpif *port = ofport_dpif_cast(port_);
1888     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1889     const char *devname = netdev_get_name(port->up.netdev);
1890     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1891     const char *dp_port_name;
1892
1893     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1894     ovs_rwlock_wrlock(&xlate_rwlock);
1895     xlate_ofport_remove(port);
1896     ovs_rwlock_unlock(&xlate_rwlock);
1897
1898     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1899                                               sizeof namebuf);
1900     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1901         /* The underlying device is still there, so delete it.  This
1902          * happens when the ofproto is being destroyed, since the caller
1903          * assumes that removal of attached ports will happen as part of
1904          * destruction. */
1905         if (!port->is_tunnel) {
1906             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1907         }
1908     }
1909
1910     if (port->peer) {
1911         port->peer->peer = NULL;
1912         port->peer = NULL;
1913     }
1914
1915     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1916         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1917         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1918         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1919     }
1920
1921     tnl_port_del(port);
1922     sset_find_and_delete(&ofproto->ports, devname);
1923     sset_find_and_delete(&ofproto->ghost_ports, devname);
1924     bundle_remove(port_);
1925     set_cfm(port_, NULL);
1926     set_bfd(port_, NULL);
1927     if (ofproto->sflow) {
1928         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1929     }
1930
1931     free(port->qdscp);
1932 }
1933
1934 static void
1935 port_modified(struct ofport *port_)
1936 {
1937     struct ofport_dpif *port = ofport_dpif_cast(port_);
1938
1939     if (port->bundle && port->bundle->bond) {
1940         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1941     }
1942
1943     if (port->cfm) {
1944         cfm_set_netdev(port->cfm, port->up.netdev);
1945     }
1946
1947     if (port->bfd) {
1948         bfd_set_netdev(port->bfd, port->up.netdev);
1949     }
1950
1951     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1952                                                 port->odp_port)) {
1953         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1954             REV_RECONFIGURE;
1955     }
1956
1957     ofport_update_peer(port);
1958 }
1959
1960 static void
1961 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1962 {
1963     struct ofport_dpif *port = ofport_dpif_cast(port_);
1964     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1965     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1966
1967     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1968                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1969                    OFPUTIL_PC_NO_PACKET_IN)) {
1970         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1971
1972         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1973             bundle_update(port->bundle);
1974         }
1975     }
1976 }
1977
1978 static int
1979 set_sflow(struct ofproto *ofproto_,
1980           const struct ofproto_sflow_options *sflow_options)
1981 {
1982     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1983     struct dpif_sflow *ds = ofproto->sflow;
1984
1985     if (sflow_options) {
1986         if (!ds) {
1987             struct ofport_dpif *ofport;
1988
1989             ds = ofproto->sflow = dpif_sflow_create();
1990             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1991                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1992             }
1993             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1994         }
1995         dpif_sflow_set_options(ds, sflow_options);
1996     } else {
1997         if (ds) {
1998             dpif_sflow_unref(ds);
1999             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2000             ofproto->sflow = NULL;
2001         }
2002     }
2003     return 0;
2004 }
2005
2006 static int
2007 set_ipfix(
2008     struct ofproto *ofproto_,
2009     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
2010     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
2011     size_t n_flow_exporters_options)
2012 {
2013     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2014     struct dpif_ipfix *di = ofproto->ipfix;
2015     bool has_options = bridge_exporter_options || flow_exporters_options;
2016
2017     if (has_options && !di) {
2018         di = ofproto->ipfix = dpif_ipfix_create();
2019     }
2020
2021     if (di) {
2022         /* Call set_options in any case to cleanly flush the flow
2023          * caches in the last exporters that are to be destroyed. */
2024         dpif_ipfix_set_options(
2025             di, bridge_exporter_options, flow_exporters_options,
2026             n_flow_exporters_options);
2027
2028         if (!has_options) {
2029             dpif_ipfix_unref(di);
2030             ofproto->ipfix = NULL;
2031         }
2032     }
2033
2034     return 0;
2035 }
2036
2037 static int
2038 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
2039 {
2040     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2041     int error;
2042
2043     if (!s) {
2044         error = 0;
2045     } else {
2046         if (!ofport->cfm) {
2047             struct ofproto_dpif *ofproto;
2048
2049             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2050             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2051             ofport->cfm = cfm_create(ofport->up.netdev);
2052         }
2053
2054         if (cfm_configure(ofport->cfm, s)) {
2055             return 0;
2056         }
2057
2058         error = EINVAL;
2059     }
2060     cfm_unref(ofport->cfm);
2061     ofport->cfm = NULL;
2062     return error;
2063 }
2064
2065 static bool
2066 get_cfm_status(const struct ofport *ofport_,
2067                struct ofproto_cfm_status *status)
2068 {
2069     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2070
2071     if (ofport->cfm) {
2072         status->faults = cfm_get_fault(ofport->cfm);
2073         status->remote_opstate = cfm_get_opup(ofport->cfm);
2074         status->health = cfm_get_health(ofport->cfm);
2075         cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2076         return true;
2077     } else {
2078         return false;
2079     }
2080 }
2081
2082 static int
2083 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2084 {
2085     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2086     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2087     struct bfd *old;
2088
2089     old = ofport->bfd;
2090     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2091                                 cfg, ofport->up.netdev);
2092     if (ofport->bfd != old) {
2093         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2094     }
2095
2096     return 0;
2097 }
2098
2099 static int
2100 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2101 {
2102     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2103
2104     if (ofport->bfd) {
2105         bfd_get_status(ofport->bfd, smap);
2106         return 0;
2107     } else {
2108         return ENOENT;
2109     }
2110 }
2111 \f
2112 /* Spanning Tree. */
2113
2114 static void
2115 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2116 {
2117     struct ofproto_dpif *ofproto = ofproto_;
2118     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2119     struct ofport_dpif *ofport;
2120
2121     ofport = stp_port_get_aux(sp);
2122     if (!ofport) {
2123         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2124                      ofproto->up.name, port_num);
2125     } else {
2126         struct eth_header *eth = pkt->l2;
2127
2128         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2129         if (eth_addr_is_zero(eth->eth_src)) {
2130             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2131                          "with unknown MAC", ofproto->up.name, port_num);
2132         } else {
2133             send_packet(ofport, pkt);
2134         }
2135     }
2136     ofpbuf_delete(pkt);
2137 }
2138
2139 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2140 static int
2141 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2142 {
2143     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2144
2145     /* Only revalidate flows if the configuration changed. */
2146     if (!s != !ofproto->stp) {
2147         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2148     }
2149
2150     if (s) {
2151         if (!ofproto->stp) {
2152             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2153                                       send_bpdu_cb, ofproto);
2154             ofproto->stp_last_tick = time_msec();
2155         }
2156
2157         stp_set_bridge_id(ofproto->stp, s->system_id);
2158         stp_set_bridge_priority(ofproto->stp, s->priority);
2159         stp_set_hello_time(ofproto->stp, s->hello_time);
2160         stp_set_max_age(ofproto->stp, s->max_age);
2161         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2162     }  else {
2163         struct ofport *ofport;
2164
2165         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2166             set_stp_port(ofport, NULL);
2167         }
2168
2169         stp_unref(ofproto->stp);
2170         ofproto->stp = NULL;
2171     }
2172
2173     return 0;
2174 }
2175
2176 static int
2177 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2178 {
2179     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2180
2181     if (ofproto->stp) {
2182         s->enabled = true;
2183         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2184         s->designated_root = stp_get_designated_root(ofproto->stp);
2185         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2186     } else {
2187         s->enabled = false;
2188     }
2189
2190     return 0;
2191 }
2192
2193 static void
2194 update_stp_port_state(struct ofport_dpif *ofport)
2195 {
2196     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2197     enum stp_state state;
2198
2199     /* Figure out new state. */
2200     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2201                              : STP_DISABLED;
2202
2203     /* Update state. */
2204     if (ofport->stp_state != state) {
2205         enum ofputil_port_state of_state;
2206         bool fwd_change;
2207
2208         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2209                     netdev_get_name(ofport->up.netdev),
2210                     stp_state_name(ofport->stp_state),
2211                     stp_state_name(state));
2212         if (stp_learn_in_state(ofport->stp_state)
2213                 != stp_learn_in_state(state)) {
2214             /* xxx Learning action flows should also be flushed. */
2215             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2216             mac_learning_flush(ofproto->ml);
2217             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2218         }
2219         fwd_change = stp_forward_in_state(ofport->stp_state)
2220                         != stp_forward_in_state(state);
2221
2222         ofproto->backer->need_revalidate = REV_STP;
2223         ofport->stp_state = state;
2224         ofport->stp_state_entered = time_msec();
2225
2226         if (fwd_change && ofport->bundle) {
2227             bundle_update(ofport->bundle);
2228         }
2229
2230         /* Update the STP state bits in the OpenFlow port description. */
2231         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2232         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2233                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2234                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2235                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2236                      : 0);
2237         ofproto_port_set_state(&ofport->up, of_state);
2238     }
2239 }
2240
2241 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2242  * caller is responsible for assigning STP port numbers and ensuring
2243  * there are no duplicates. */
2244 static int
2245 set_stp_port(struct ofport *ofport_,
2246              const struct ofproto_port_stp_settings *s)
2247 {
2248     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2249     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2250     struct stp_port *sp = ofport->stp_port;
2251
2252     if (!s || !s->enable) {
2253         if (sp) {
2254             ofport->stp_port = NULL;
2255             stp_port_disable(sp);
2256             update_stp_port_state(ofport);
2257         }
2258         return 0;
2259     } else if (sp && stp_port_no(sp) != s->port_num
2260             && ofport == stp_port_get_aux(sp)) {
2261         /* The port-id changed, so disable the old one if it's not
2262          * already in use by another port. */
2263         stp_port_disable(sp);
2264     }
2265
2266     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2267     stp_port_enable(sp);
2268
2269     stp_port_set_aux(sp, ofport);
2270     stp_port_set_priority(sp, s->priority);
2271     stp_port_set_path_cost(sp, s->path_cost);
2272
2273     update_stp_port_state(ofport);
2274
2275     return 0;
2276 }
2277
2278 static int
2279 get_stp_port_status(struct ofport *ofport_,
2280                     struct ofproto_port_stp_status *s)
2281 {
2282     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2283     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2284     struct stp_port *sp = ofport->stp_port;
2285
2286     if (!ofproto->stp || !sp) {
2287         s->enabled = false;
2288         return 0;
2289     }
2290
2291     s->enabled = true;
2292     s->port_id = stp_port_get_id(sp);
2293     s->state = stp_port_get_state(sp);
2294     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2295     s->role = stp_port_get_role(sp);
2296     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2297
2298     return 0;
2299 }
2300
2301 static void
2302 stp_run(struct ofproto_dpif *ofproto)
2303 {
2304     if (ofproto->stp) {
2305         long long int now = time_msec();
2306         long long int elapsed = now - ofproto->stp_last_tick;
2307         struct stp_port *sp;
2308
2309         if (elapsed > 0) {
2310             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2311             ofproto->stp_last_tick = now;
2312         }
2313         while (stp_get_changed_port(ofproto->stp, &sp)) {
2314             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2315
2316             if (ofport) {
2317                 update_stp_port_state(ofport);
2318             }
2319         }
2320
2321         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2322             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2323             mac_learning_flush(ofproto->ml);
2324             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2325         }
2326     }
2327 }
2328
2329 static void
2330 stp_wait(struct ofproto_dpif *ofproto)
2331 {
2332     if (ofproto->stp) {
2333         poll_timer_wait(1000);
2334     }
2335 }
2336 \f
2337 static int
2338 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2339            size_t n_qdscp)
2340 {
2341     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2342     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2343
2344     if (ofport->n_qdscp != n_qdscp
2345         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2346                               n_qdscp * sizeof *qdscp))) {
2347         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2348         free(ofport->qdscp);
2349         ofport->qdscp = n_qdscp
2350             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2351             : NULL;
2352         ofport->n_qdscp = n_qdscp;
2353     }
2354
2355     return 0;
2356 }
2357 \f
2358 /* Bundles. */
2359
2360 /* Expires all MAC learning entries associated with 'bundle' and forces its
2361  * ofproto to revalidate every flow.
2362  *
2363  * Normally MAC learning entries are removed only from the ofproto associated
2364  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2365  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2366  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2367  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2368  * with the host from which it migrated. */
2369 static void
2370 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2371 {
2372     struct ofproto_dpif *ofproto = bundle->ofproto;
2373     struct mac_learning *ml = ofproto->ml;
2374     struct mac_entry *mac, *next_mac;
2375
2376     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2377     ovs_rwlock_wrlock(&ml->rwlock);
2378     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2379         if (mac->port.p == bundle) {
2380             if (all_ofprotos) {
2381                 struct ofproto_dpif *o;
2382
2383                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2384                     if (o != ofproto) {
2385                         struct mac_entry *e;
2386
2387                         ovs_rwlock_wrlock(&o->ml->rwlock);
2388                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2389                         if (e) {
2390                             mac_learning_expire(o->ml, e);
2391                         }
2392                         ovs_rwlock_unlock(&o->ml->rwlock);
2393                     }
2394                 }
2395             }
2396
2397             mac_learning_expire(ml, mac);
2398         }
2399     }
2400     ovs_rwlock_unlock(&ml->rwlock);
2401 }
2402
2403 static struct ofbundle *
2404 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2405 {
2406     struct ofbundle *bundle;
2407
2408     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2409                              &ofproto->bundles) {
2410         if (bundle->aux == aux) {
2411             return bundle;
2412         }
2413     }
2414     return NULL;
2415 }
2416
2417 static void
2418 bundle_update(struct ofbundle *bundle)
2419 {
2420     struct ofport_dpif *port;
2421
2422     bundle->floodable = true;
2423     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2424         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2425             || !stp_forward_in_state(port->stp_state)) {
2426             bundle->floodable = false;
2427             break;
2428         }
2429     }
2430 }
2431
2432 static void
2433 bundle_del_port(struct ofport_dpif *port)
2434 {
2435     struct ofbundle *bundle = port->bundle;
2436
2437     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2438
2439     list_remove(&port->bundle_node);
2440     port->bundle = NULL;
2441
2442     if (bundle->lacp) {
2443         lacp_slave_unregister(bundle->lacp, port);
2444     }
2445     if (bundle->bond) {
2446         bond_slave_unregister(bundle->bond, port);
2447     }
2448
2449     bundle_update(bundle);
2450 }
2451
2452 static bool
2453 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2454                 struct lacp_slave_settings *lacp)
2455 {
2456     struct ofport_dpif *port;
2457
2458     port = get_ofp_port(bundle->ofproto, ofp_port);
2459     if (!port) {
2460         return false;
2461     }
2462
2463     if (port->bundle != bundle) {
2464         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2465         if (port->bundle) {
2466             bundle_remove(&port->up);
2467         }
2468
2469         port->bundle = bundle;
2470         list_push_back(&bundle->ports, &port->bundle_node);
2471         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2472             || !stp_forward_in_state(port->stp_state)) {
2473             bundle->floodable = false;
2474         }
2475     }
2476     if (lacp) {
2477         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2478         lacp_slave_register(bundle->lacp, port, lacp);
2479     }
2480
2481     return true;
2482 }
2483
2484 static void
2485 bundle_destroy(struct ofbundle *bundle)
2486 {
2487     struct ofproto_dpif *ofproto;
2488     struct ofport_dpif *port, *next_port;
2489
2490     if (!bundle) {
2491         return;
2492     }
2493
2494     ofproto = bundle->ofproto;
2495     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2496
2497     ovs_rwlock_wrlock(&xlate_rwlock);
2498     xlate_bundle_remove(bundle);
2499     ovs_rwlock_unlock(&xlate_rwlock);
2500
2501     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2502         bundle_del_port(port);
2503     }
2504
2505     bundle_flush_macs(bundle, true);
2506     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2507     free(bundle->name);
2508     free(bundle->trunks);
2509     lacp_unref(bundle->lacp);
2510     bond_unref(bundle->bond);
2511     free(bundle);
2512 }
2513
2514 static int
2515 bundle_set(struct ofproto *ofproto_, void *aux,
2516            const struct ofproto_bundle_settings *s)
2517 {
2518     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2519     bool need_flush = false;
2520     struct ofport_dpif *port;
2521     struct ofbundle *bundle;
2522     unsigned long *trunks;
2523     int vlan;
2524     size_t i;
2525     bool ok;
2526
2527     if (!s) {
2528         bundle_destroy(bundle_lookup(ofproto, aux));
2529         return 0;
2530     }
2531
2532     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2533     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2534
2535     bundle = bundle_lookup(ofproto, aux);
2536     if (!bundle) {
2537         bundle = xmalloc(sizeof *bundle);
2538
2539         bundle->ofproto = ofproto;
2540         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2541                     hash_pointer(aux, 0));
2542         bundle->aux = aux;
2543         bundle->name = NULL;
2544
2545         list_init(&bundle->ports);
2546         bundle->vlan_mode = PORT_VLAN_TRUNK;
2547         bundle->vlan = -1;
2548         bundle->trunks = NULL;
2549         bundle->use_priority_tags = s->use_priority_tags;
2550         bundle->lacp = NULL;
2551         bundle->bond = NULL;
2552
2553         bundle->floodable = true;
2554         mbridge_register_bundle(ofproto->mbridge, bundle);
2555     }
2556
2557     if (!bundle->name || strcmp(s->name, bundle->name)) {
2558         free(bundle->name);
2559         bundle->name = xstrdup(s->name);
2560     }
2561
2562     /* LACP. */
2563     if (s->lacp) {
2564         if (!bundle->lacp) {
2565             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2566             bundle->lacp = lacp_create();
2567         }
2568         lacp_configure(bundle->lacp, s->lacp);
2569     } else {
2570         lacp_unref(bundle->lacp);
2571         bundle->lacp = NULL;
2572     }
2573
2574     /* Update set of ports. */
2575     ok = true;
2576     for (i = 0; i < s->n_slaves; i++) {
2577         if (!bundle_add_port(bundle, s->slaves[i],
2578                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2579             ok = false;
2580         }
2581     }
2582     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2583         struct ofport_dpif *next_port;
2584
2585         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2586             for (i = 0; i < s->n_slaves; i++) {
2587                 if (s->slaves[i] == port->up.ofp_port) {
2588                     goto found;
2589                 }
2590             }
2591
2592             bundle_del_port(port);
2593         found: ;
2594         }
2595     }
2596     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2597
2598     if (list_is_empty(&bundle->ports)) {
2599         bundle_destroy(bundle);
2600         return EINVAL;
2601     }
2602
2603     /* Set VLAN tagging mode */
2604     if (s->vlan_mode != bundle->vlan_mode
2605         || s->use_priority_tags != bundle->use_priority_tags) {
2606         bundle->vlan_mode = s->vlan_mode;
2607         bundle->use_priority_tags = s->use_priority_tags;
2608         need_flush = true;
2609     }
2610
2611     /* Set VLAN tag. */
2612     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2613             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2614             : 0);
2615     if (vlan != bundle->vlan) {
2616         bundle->vlan = vlan;
2617         need_flush = true;
2618     }
2619
2620     /* Get trunked VLANs. */
2621     switch (s->vlan_mode) {
2622     case PORT_VLAN_ACCESS:
2623         trunks = NULL;
2624         break;
2625
2626     case PORT_VLAN_TRUNK:
2627         trunks = CONST_CAST(unsigned long *, s->trunks);
2628         break;
2629
2630     case PORT_VLAN_NATIVE_UNTAGGED:
2631     case PORT_VLAN_NATIVE_TAGGED:
2632         if (vlan != 0 && (!s->trunks
2633                           || !bitmap_is_set(s->trunks, vlan)
2634                           || bitmap_is_set(s->trunks, 0))) {
2635             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2636             if (s->trunks) {
2637                 trunks = bitmap_clone(s->trunks, 4096);
2638             } else {
2639                 trunks = bitmap_allocate1(4096);
2640             }
2641             bitmap_set1(trunks, vlan);
2642             bitmap_set0(trunks, 0);
2643         } else {
2644             trunks = CONST_CAST(unsigned long *, s->trunks);
2645         }
2646         break;
2647
2648     default:
2649         NOT_REACHED();
2650     }
2651     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2652         free(bundle->trunks);
2653         if (trunks == s->trunks) {
2654             bundle->trunks = vlan_bitmap_clone(trunks);
2655         } else {
2656             bundle->trunks = trunks;
2657             trunks = NULL;
2658         }
2659         need_flush = true;
2660     }
2661     if (trunks != s->trunks) {
2662         free(trunks);
2663     }
2664
2665     /* Bonding. */
2666     if (!list_is_short(&bundle->ports)) {
2667         bundle->ofproto->has_bonded_bundles = true;
2668         if (bundle->bond) {
2669             if (bond_reconfigure(bundle->bond, s->bond)) {
2670                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2671             }
2672         } else {
2673             bundle->bond = bond_create(s->bond);
2674             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2675         }
2676
2677         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2678             bond_slave_register(bundle->bond, port, port->up.netdev);
2679         }
2680     } else {
2681         bond_unref(bundle->bond);
2682         bundle->bond = NULL;
2683     }
2684
2685     /* If we changed something that would affect MAC learning, un-learn
2686      * everything on this port and force flow revalidation. */
2687     if (need_flush) {
2688         bundle_flush_macs(bundle, false);
2689     }
2690
2691     return 0;
2692 }
2693
2694 static void
2695 bundle_remove(struct ofport *port_)
2696 {
2697     struct ofport_dpif *port = ofport_dpif_cast(port_);
2698     struct ofbundle *bundle = port->bundle;
2699
2700     if (bundle) {
2701         bundle_del_port(port);
2702         if (list_is_empty(&bundle->ports)) {
2703             bundle_destroy(bundle);
2704         } else if (list_is_short(&bundle->ports)) {
2705             bond_unref(bundle->bond);
2706             bundle->bond = NULL;
2707         }
2708     }
2709 }
2710
2711 static void
2712 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2713 {
2714     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2715     struct ofport_dpif *port = port_;
2716     uint8_t ea[ETH_ADDR_LEN];
2717     int error;
2718
2719     error = netdev_get_etheraddr(port->up.netdev, ea);
2720     if (!error) {
2721         struct ofpbuf packet;
2722         void *packet_pdu;
2723
2724         ofpbuf_init(&packet, 0);
2725         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2726                                  pdu_size);
2727         memcpy(packet_pdu, pdu, pdu_size);
2728
2729         send_packet(port, &packet);
2730         ofpbuf_uninit(&packet);
2731     } else {
2732         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2733                     "%s (%s)", port->bundle->name,
2734                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2735     }
2736 }
2737
2738 static void
2739 bundle_send_learning_packets(struct ofbundle *bundle)
2740 {
2741     struct ofproto_dpif *ofproto = bundle->ofproto;
2742     int error, n_packets, n_errors;
2743     struct mac_entry *e;
2744
2745     error = n_packets = n_errors = 0;
2746     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2747     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2748         if (e->port.p != bundle) {
2749             struct ofpbuf *learning_packet;
2750             struct ofport_dpif *port;
2751             void *port_void;
2752             int ret;
2753
2754             /* The assignment to "port" is unnecessary but makes "grep"ing for
2755              * struct ofport_dpif more effective. */
2756             learning_packet = bond_compose_learning_packet(bundle->bond,
2757                                                            e->mac, e->vlan,
2758                                                            &port_void);
2759             port = port_void;
2760             ret = send_packet(port, learning_packet);
2761             ofpbuf_delete(learning_packet);
2762             if (ret) {
2763                 error = ret;
2764                 n_errors++;
2765             }
2766             n_packets++;
2767         }
2768     }
2769     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2770
2771     if (n_errors) {
2772         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2773         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2774                      "packets, last error was: %s",
2775                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2776     } else {
2777         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2778                  bundle->name, n_packets);
2779     }
2780 }
2781
2782 static void
2783 bundle_run(struct ofbundle *bundle)
2784 {
2785     if (bundle->lacp) {
2786         lacp_run(bundle->lacp, send_pdu_cb);
2787     }
2788     if (bundle->bond) {
2789         struct ofport_dpif *port;
2790
2791         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2792             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2793         }
2794
2795         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2796             bundle->ofproto->backer->need_revalidate = REV_BOND;
2797         }
2798
2799         if (bond_should_send_learning_packets(bundle->bond)) {
2800             bundle_send_learning_packets(bundle);
2801         }
2802     }
2803 }
2804
2805 static void
2806 bundle_wait(struct ofbundle *bundle)
2807 {
2808     if (bundle->lacp) {
2809         lacp_wait(bundle->lacp);
2810     }
2811     if (bundle->bond) {
2812         bond_wait(bundle->bond);
2813     }
2814 }
2815 \f
2816 /* Mirrors. */
2817
2818 static int
2819 mirror_set__(struct ofproto *ofproto_, void *aux,
2820              const struct ofproto_mirror_settings *s)
2821 {
2822     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2823     struct ofbundle **srcs, **dsts;
2824     int error;
2825     size_t i;
2826
2827     if (!s) {
2828         mirror_destroy(ofproto->mbridge, aux);
2829         return 0;
2830     }
2831
2832     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2833     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2834
2835     for (i = 0; i < s->n_srcs; i++) {
2836         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2837     }
2838
2839     for (i = 0; i < s->n_dsts; i++) {
2840         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2841     }
2842
2843     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2844                        s->n_dsts, s->src_vlans,
2845                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2846     free(srcs);
2847     free(dsts);
2848     return error;
2849 }
2850
2851 static int
2852 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2853                    uint64_t *packets, uint64_t *bytes)
2854 {
2855     push_all_stats();
2856     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2857                             bytes);
2858 }
2859
2860 static int
2861 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2862 {
2863     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2864     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2865     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2866         mac_learning_flush(ofproto->ml);
2867     }
2868     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2869     return 0;
2870 }
2871
2872 static bool
2873 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2874 {
2875     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2876     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2877     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2878 }
2879
2880 static void
2881 forward_bpdu_changed(struct ofproto *ofproto_)
2882 {
2883     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2884     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2885 }
2886
2887 static void
2888 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2889                      size_t max_entries)
2890 {
2891     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2892     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2893     mac_learning_set_idle_time(ofproto->ml, idle_time);
2894     mac_learning_set_max_entries(ofproto->ml, max_entries);
2895     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2896 }
2897 \f
2898 /* Ports. */
2899
2900 static struct ofport_dpif *
2901 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2902 {
2903     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2904     return ofport ? ofport_dpif_cast(ofport) : NULL;
2905 }
2906
2907 static struct ofport_dpif *
2908 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2909 {
2910     struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2911     return port && &ofproto->up == port->up.ofproto ? port : NULL;
2912 }
2913
2914 static void
2915 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2916                             struct ofproto_port *ofproto_port,
2917                             struct dpif_port *dpif_port)
2918 {
2919     ofproto_port->name = dpif_port->name;
2920     ofproto_port->type = dpif_port->type;
2921     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2922 }
2923
2924 static void
2925 ofport_update_peer(struct ofport_dpif *ofport)
2926 {
2927     const struct ofproto_dpif *ofproto;
2928     struct dpif_backer *backer;
2929     char *peer_name;
2930
2931     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2932         return;
2933     }
2934
2935     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2936     backer->need_revalidate = REV_RECONFIGURE;
2937
2938     if (ofport->peer) {
2939         ofport->peer->peer = NULL;
2940         ofport->peer = NULL;
2941     }
2942
2943     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2944     if (!peer_name) {
2945         return;
2946     }
2947
2948     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2949         struct ofport *peer_ofport;
2950         struct ofport_dpif *peer;
2951         char *peer_peer;
2952
2953         if (ofproto->backer != backer) {
2954             continue;
2955         }
2956
2957         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2958         if (!peer_ofport) {
2959             continue;
2960         }
2961
2962         peer = ofport_dpif_cast(peer_ofport);
2963         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2964         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2965                                  peer_peer)) {
2966             ofport->peer = peer;
2967             ofport->peer->peer = ofport;
2968         }
2969         free(peer_peer);
2970
2971         break;
2972     }
2973     free(peer_name);
2974 }
2975
2976 static void
2977 port_run_fast(struct ofport_dpif *ofport)
2978 {
2979     if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2980         struct ofpbuf packet;
2981
2982         ofpbuf_init(&packet, 0);
2983         cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2984         send_packet(ofport, &packet);
2985         ofpbuf_uninit(&packet);
2986     }
2987
2988     if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2989         struct ofpbuf packet;
2990
2991         ofpbuf_init(&packet, 0);
2992         bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2993         send_packet(ofport, &packet);
2994         ofpbuf_uninit(&packet);
2995     }
2996 }
2997
2998 static void
2999 port_run(struct ofport_dpif *ofport)
3000 {
3001     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3002     bool carrier_changed = carrier_seq != ofport->carrier_seq;
3003     bool enable = netdev_get_carrier(ofport->up.netdev);
3004     bool cfm_enable = false;
3005     bool bfd_enable = false;
3006
3007     ofport->carrier_seq = carrier_seq;
3008
3009     port_run_fast(ofport);
3010
3011     if (ofport->cfm) {
3012         int cfm_opup = cfm_get_opup(ofport->cfm);
3013
3014         cfm_run(ofport->cfm);
3015         cfm_enable = !cfm_get_fault(ofport->cfm);
3016
3017         if (cfm_opup >= 0) {
3018             cfm_enable = cfm_enable && cfm_opup;
3019         }
3020     }
3021
3022     if (ofport->bfd) {
3023         bfd_run(ofport->bfd);
3024         bfd_enable = bfd_forwarding(ofport->bfd);
3025     }
3026
3027     if (ofport->bfd || ofport->cfm) {
3028         enable = enable && (cfm_enable || bfd_enable);
3029     }
3030
3031     if (ofport->bundle) {
3032         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3033         if (carrier_changed) {
3034             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3035         }
3036     }
3037
3038     if (ofport->may_enable != enable) {
3039         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3040         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3041     }
3042
3043     ofport->may_enable = enable;
3044 }
3045
3046 static void
3047 port_wait(struct ofport_dpif *ofport)
3048 {
3049     if (ofport->cfm) {
3050         cfm_wait(ofport->cfm);
3051     }
3052
3053     if (ofport->bfd) {
3054         bfd_wait(ofport->bfd);
3055     }
3056 }
3057
3058 static int
3059 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3060                    struct ofproto_port *ofproto_port)
3061 {
3062     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3063     struct dpif_port dpif_port;
3064     int error;
3065
3066     if (sset_contains(&ofproto->ghost_ports, devname)) {
3067         const char *type = netdev_get_type_from_name(devname);
3068
3069         /* We may be called before ofproto->up.port_by_name is populated with
3070          * the appropriate ofport.  For this reason, we must get the name and
3071          * type from the netdev layer directly. */
3072         if (type) {
3073             const struct ofport *ofport;
3074
3075             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3076             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3077             ofproto_port->name = xstrdup(devname);
3078             ofproto_port->type = xstrdup(type);
3079             return 0;
3080         }
3081         return ENODEV;
3082     }
3083
3084     if (!sset_contains(&ofproto->ports, devname)) {
3085         return ENODEV;
3086     }
3087     error = dpif_port_query_by_name(ofproto->backer->dpif,
3088                                     devname, &dpif_port);
3089     if (!error) {
3090         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3091     }
3092     return error;
3093 }
3094
3095 static int
3096 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3097 {
3098     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3099     const char *devname = netdev_get_name(netdev);
3100     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3101     const char *dp_port_name;
3102
3103     if (netdev_vport_is_patch(netdev)) {
3104         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3105         return 0;
3106     }
3107
3108     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3109     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3110         odp_port_t port_no = ODPP_NONE;
3111         int error;
3112
3113         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3114         if (error) {
3115             return error;
3116         }
3117         if (netdev_get_tunnel_config(netdev)) {
3118             simap_put(&ofproto->backer->tnl_backers,
3119                       dp_port_name, odp_to_u32(port_no));
3120         }
3121     }
3122
3123     if (netdev_get_tunnel_config(netdev)) {
3124         sset_add(&ofproto->ghost_ports, devname);
3125     } else {
3126         sset_add(&ofproto->ports, devname);
3127     }
3128     return 0;
3129 }
3130
3131 static int
3132 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3133 {
3134     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3135     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3136     int error = 0;
3137
3138     if (!ofport) {
3139         return 0;
3140     }
3141
3142     sset_find_and_delete(&ofproto->ghost_ports,
3143                          netdev_get_name(ofport->up.netdev));
3144     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3145     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3146         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3147         if (!error) {
3148             /* The caller is going to close ofport->up.netdev.  If this is a
3149              * bonded port, then the bond is using that netdev, so remove it
3150              * from the bond.  The client will need to reconfigure everything
3151              * after deleting ports, so then the slave will get re-added. */
3152             bundle_remove(&ofport->up);
3153         }
3154     }
3155     return error;
3156 }
3157
3158 static int
3159 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3160 {
3161     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3162     int error;
3163
3164     push_all_stats();
3165
3166     error = netdev_get_stats(ofport->up.netdev, stats);
3167
3168     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3169         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3170
3171         /* ofproto->stats.tx_packets represents packets that we created
3172          * internally and sent to some port (e.g. packets sent with
3173          * send_packet()).  Account for them as if they had come from
3174          * OFPP_LOCAL and got forwarded. */
3175
3176         if (stats->rx_packets != UINT64_MAX) {
3177             stats->rx_packets += ofproto->stats.tx_packets;
3178         }
3179
3180         if (stats->rx_bytes != UINT64_MAX) {
3181             stats->rx_bytes += ofproto->stats.tx_bytes;
3182         }
3183
3184         /* ofproto->stats.rx_packets represents packets that were received on
3185          * some port and we processed internally and dropped (e.g. STP).
3186          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3187
3188         if (stats->tx_packets != UINT64_MAX) {
3189             stats->tx_packets += ofproto->stats.rx_packets;
3190         }
3191
3192         if (stats->tx_bytes != UINT64_MAX) {
3193             stats->tx_bytes += ofproto->stats.rx_bytes;
3194         }
3195     }
3196
3197     return error;
3198 }
3199
3200 struct port_dump_state {
3201     uint32_t bucket;
3202     uint32_t offset;
3203     bool ghost;
3204
3205     struct ofproto_port port;
3206     bool has_port;
3207 };
3208
3209 static int
3210 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3211 {
3212     *statep = xzalloc(sizeof(struct port_dump_state));
3213     return 0;
3214 }
3215
3216 static int
3217 port_dump_next(const struct ofproto *ofproto_, void *state_,
3218                struct ofproto_port *port)
3219 {
3220     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3221     struct port_dump_state *state = state_;
3222     const struct sset *sset;
3223     struct sset_node *node;
3224
3225     if (state->has_port) {
3226         ofproto_port_destroy(&state->port);
3227         state->has_port = false;
3228     }
3229     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3230     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3231         int error;
3232
3233         error = port_query_by_name(ofproto_, node->name, &state->port);
3234         if (!error) {
3235             *port = state->port;
3236             state->has_port = true;
3237             return 0;
3238         } else if (error != ENODEV) {
3239             return error;
3240         }
3241     }
3242
3243     if (!state->ghost) {
3244         state->ghost = true;
3245         state->bucket = 0;
3246         state->offset = 0;
3247         return port_dump_next(ofproto_, state_, port);
3248     }
3249
3250     return EOF;
3251 }
3252
3253 static int
3254 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3255 {
3256     struct port_dump_state *state = state_;
3257
3258     if (state->has_port) {
3259         ofproto_port_destroy(&state->port);
3260     }
3261     free(state);
3262     return 0;
3263 }
3264
3265 static int
3266 port_poll(const struct ofproto *ofproto_, char **devnamep)
3267 {
3268     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3269
3270     if (ofproto->port_poll_errno) {
3271         int error = ofproto->port_poll_errno;
3272         ofproto->port_poll_errno = 0;
3273         return error;
3274     }
3275
3276     if (sset_is_empty(&ofproto->port_poll_set)) {
3277         return EAGAIN;
3278     }
3279
3280     *devnamep = sset_pop(&ofproto->port_poll_set);
3281     return 0;
3282 }
3283
3284 static void
3285 port_poll_wait(const struct ofproto *ofproto_)
3286 {
3287     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3288     dpif_port_poll_wait(ofproto->backer->dpif);
3289 }
3290
3291 static int
3292 port_is_lacp_current(const struct ofport *ofport_)
3293 {
3294     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3295     return (ofport->bundle && ofport->bundle->lacp
3296             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3297             : -1);
3298 }
3299 \f
3300 /* Upcall handling. */
3301
3302 struct flow_miss_op {
3303     struct dpif_op dpif_op;
3304
3305     uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3306     struct xlate_out xout;
3307     bool xout_garbage;           /* 'xout' needs to be uninitialized? */
3308
3309     struct ofpbuf mask;          /* Flow mask for "put" ops. */
3310     struct odputil_keybuf maskbuf;
3311
3312     /* If this is a "put" op, then a pointer to the subfacet that should
3313      * be marked as uninstalled if the operation fails. */
3314     struct subfacet *subfacet;
3315 };
3316
3317 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3318  * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3319  * and (usually) installing a datapath flow.  The answer is usually "yes" (a
3320  * return value of true).  However, for short flows the cost of bookkeeping is
3321  * much higher than the benefits, so when the datapath holds a large number of
3322  * flows we impose some heuristics to decide which flows are likely to be worth
3323  * tracking. */
3324 static bool
3325 flow_miss_should_make_facet(struct flow_miss *miss)
3326 {
3327     struct dpif_backer *backer = miss->ofproto->backer;
3328     uint32_t hash;
3329
3330     switch (flow_miss_model) {
3331     case OFPROTO_HANDLE_MISS_AUTO:
3332         break;
3333     case OFPROTO_HANDLE_MISS_WITH_FACETS:
3334         return true;
3335     case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3336         return false;
3337     }
3338
3339     if (!backer->governor) {
3340         size_t n_subfacets;
3341
3342         n_subfacets = hmap_count(&backer->subfacets);
3343         if (n_subfacets * 2 <= flow_eviction_threshold) {
3344             return true;
3345         }
3346
3347         backer->governor = governor_create();
3348     }
3349
3350     hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3351     return governor_should_install_flow(backer->governor, hash,
3352                                         list_size(&miss->packets));
3353 }
3354
3355 /* Handles 'miss', which matches 'facet'.  May add any required datapath
3356  * operations to 'ops', incrementing '*n_ops' for each new op.
3357  *
3358  * All of the packets in 'miss' are considered to have arrived at time
3359  * 'miss->stats.used'.  This is really important only for new facets: if we
3360  * just called time_msec() here, then the new subfacet or its packets could
3361  * look (occasionally) as though it was used some time after the facet was
3362  * used.  That can make a one-packet flow look like it has a nonzero duration,
3363  * which looks odd in e.g. NetFlow statistics. */
3364 static void
3365 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3366                             struct flow_miss_op *ops, size_t *n_ops)
3367 {
3368     enum subfacet_path want_path;
3369     struct subfacet *subfacet;
3370
3371     facet->packet_count += miss->stats.n_packets;
3372     facet->prev_packet_count += miss->stats.n_packets;
3373     facet->byte_count += miss->stats.n_bytes;
3374     facet->prev_byte_count += miss->stats.n_bytes;
3375
3376     want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3377
3378     /* Don't install the flow if it's the result of the "userspace"
3379      * action for an already installed facet.  This can occur when a
3380      * datapath flow with wildcards has a "userspace" action and flows
3381      * sent to userspace result in a different subfacet, which will then
3382      * be rejected as overlapping by the datapath. */
3383     if (miss->upcall_type == DPIF_UC_ACTION
3384         && !list_is_empty(&facet->subfacets)) {
3385         return;
3386     }
3387
3388     subfacet = subfacet_create(facet, miss);
3389     if (subfacet->path != want_path) {
3390         struct flow_miss_op *op = &ops[(*n_ops)++];
3391         struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3392
3393         subfacet->path = want_path;
3394
3395         ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3396         if (enable_megaflows) {
3397             odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3398                                    &miss->flow, UINT32_MAX);
3399         }
3400
3401         op->xout_garbage = false;
3402         op->dpif_op.type = DPIF_OP_FLOW_PUT;
3403         op->subfacet = subfacet;
3404         put->flags = DPIF_FP_CREATE;
3405         put->key = miss->key;
3406         put->key_len = miss->key_len;
3407         put->mask = op->mask.data;
3408         put->mask_len = op->mask.size;
3409
3410         if (want_path == SF_FAST_PATH) {
3411             put->actions = facet->xout.odp_actions.data;
3412             put->actions_len = facet->xout.odp_actions.size;
3413         } else {
3414             compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3415                               op->slow_stub, sizeof op->slow_stub,
3416                               &put->actions, &put->actions_len);
3417         }
3418         put->stats = NULL;
3419     }
3420 }
3421
3422 /* Handles flow miss 'miss'.  May add any required datapath operations
3423  * to 'ops', incrementing '*n_ops' for each new op. */
3424 static void
3425 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3426                  size_t *n_ops)
3427 {
3428     struct facet *facet;
3429
3430     miss->ofproto->n_missed += list_size(&miss->packets);
3431
3432     facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3433     if (!facet) {
3434         /* There does not exist a bijection between 'struct flow' and datapath
3435          * flow keys with fitness ODP_FIT_TO_LITTLE.  This breaks a fundamental
3436          * assumption used throughout the facet and subfacet handling code.
3437          * Since we have to handle these misses in userspace anyway, we simply
3438          * skip facet creation, avoiding the problem altogether. */
3439         if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3440             || !flow_miss_should_make_facet(miss)) {
3441             return;
3442         }
3443
3444         facet = facet_create(miss);
3445     }
3446     handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3447 }
3448
3449 static struct drop_key *
3450 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3451                 size_t key_len)
3452 {
3453     struct drop_key *drop_key;
3454
3455     HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3456                              &backer->drop_keys) {
3457         if (drop_key->key_len == key_len
3458             && !memcmp(drop_key->key, key, key_len)) {
3459             return drop_key;
3460         }
3461     }
3462     return NULL;
3463 }
3464
3465 static void
3466 drop_key_clear(struct dpif_backer *backer)
3467 {
3468     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3469     struct drop_key *drop_key, *next;
3470
3471     HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3472         int error;
3473
3474         error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3475                               NULL);
3476         if (error && !VLOG_DROP_WARN(&rl)) {
3477             struct ds ds = DS_EMPTY_INITIALIZER;
3478             odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3479             VLOG_WARN("Failed to delete drop key (%s) (%s)",
3480                       ovs_strerror(error), ds_cstr(&ds));
3481             ds_destroy(&ds);
3482         }
3483
3484         hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3485         drop_key_destroy(drop_key);
3486     }
3487
3488     udpif_drop_key_clear(backer->udpif);
3489 }
3490
3491 static void
3492 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3493 {
3494     struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3495     struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3496     struct flow_miss *miss;
3497     size_t n_ops, i;
3498
3499     /* Process each element in the to-do list, constructing the set of
3500      * operations to batch. */
3501     n_ops = 0;
3502     HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3503         handle_flow_miss(miss, flow_miss_ops, &n_ops);
3504     }
3505     ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3506
3507     /* Execute batch. */
3508     for (i = 0; i < n_ops; i++) {
3509         dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3510     }
3511     dpif_operate(backer->dpif, dpif_ops, n_ops);
3512
3513     for (i = 0; i < n_ops; i++) {
3514         if (dpif_ops[i]->error != 0
3515             && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3516             && flow_miss_ops[i].subfacet) {
3517             struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3518
3519             COVERAGE_INC(subfacet_install_fail);
3520
3521             /* Zero-out subfacet counters when installation failed, but
3522              * datapath reported hits.  This should not happen and
3523              * indicates a bug, since if the datapath flow exists, we
3524              * should not be attempting to create a new subfacet.  A
3525              * buggy datapath could trigger this, so just zero out the
3526              * counters and log an error. */
3527             if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3528                 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3529                             "datapath reported hits");
3530                 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3531             }
3532
3533             subfacet->path = SF_NOT_INSTALLED;
3534         }
3535     }
3536 }
3537
3538 static void
3539 handle_sflow_upcall(struct dpif_backer *backer,
3540                     const struct dpif_upcall *upcall)
3541 {
3542     struct ofproto_dpif *ofproto;
3543     union user_action_cookie cookie;
3544     struct flow flow;
3545     odp_port_t odp_in_port;
3546
3547     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3548                       &flow, NULL, &ofproto, &odp_in_port)
3549         || !ofproto->sflow) {
3550         return;
3551     }
3552
3553     memset(&cookie, 0, sizeof cookie);
3554     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3555     dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3556                         odp_in_port, &cookie);
3557 }
3558
3559 static void
3560 handle_flow_sample_upcall(struct dpif_backer *backer,
3561                           const struct dpif_upcall *upcall)
3562 {
3563     struct ofproto_dpif *ofproto;
3564     union user_action_cookie cookie;
3565     struct flow flow;
3566
3567     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3568                       &flow, NULL, &ofproto, NULL)
3569         || !ofproto->ipfix) {
3570         return;
3571     }
3572
3573     memset(&cookie, 0, sizeof cookie);
3574     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3575
3576     /* The flow reflects exactly the contents of the packet.  Sample
3577      * the packet using it. */
3578     dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3579                            cookie.flow_sample.collector_set_id,
3580                            cookie.flow_sample.probability,
3581                            cookie.flow_sample.obs_domain_id,
3582                            cookie.flow_sample.obs_point_id);
3583 }
3584
3585 static void
3586 handle_ipfix_upcall(struct dpif_backer *backer,
3587                     const struct dpif_upcall *upcall)
3588 {
3589     struct ofproto_dpif *ofproto;
3590     struct flow flow;
3591
3592     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3593                       &flow, NULL, &ofproto, NULL)
3594         || !ofproto->ipfix) {
3595         return;
3596     }
3597
3598     /* The flow reflects exactly the contents of the packet.  Sample
3599      * the packet using it. */
3600     dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3601 }
3602
3603 static void
3604 handle_upcalls(struct dpif_backer *backer)
3605 {
3606     struct flow_miss_batch *fmb;
3607     int n_processed;
3608
3609     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3610         struct upcall *upcall = upcall_next(backer->udpif);
3611
3612         if (!upcall) {
3613             break;
3614         }
3615
3616         switch (upcall->type) {
3617         case SFLOW_UPCALL:
3618             handle_sflow_upcall(backer, &upcall->dpif_upcall);
3619             break;
3620
3621         case FLOW_SAMPLE_UPCALL:
3622             handle_flow_sample_upcall(backer, &upcall->dpif_upcall);
3623             break;
3624
3625         case IPFIX_UPCALL:
3626             handle_ipfix_upcall(backer, &upcall->dpif_upcall);
3627             break;
3628
3629         case BAD_UPCALL:
3630             break;
3631
3632         case MISS_UPCALL:
3633             NOT_REACHED();
3634         }
3635
3636         upcall_destroy(upcall);
3637     }
3638
3639     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3640         struct drop_key *drop_key = drop_key_next(backer->udpif);
3641         if (!drop_key) {
3642             break;
3643         }
3644
3645         if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3646             hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3647                         hash_bytes(drop_key->key, drop_key->key_len, 0));
3648             dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3649                           drop_key->key, drop_key->key_len,
3650                           NULL, 0, NULL, 0, NULL);
3651         } else {
3652             drop_key_destroy(drop_key);
3653         }
3654     }
3655
3656     fmb = flow_miss_batch_next(backer->udpif);
3657     if (fmb) {
3658         handle_flow_misses(backer, fmb);
3659         flow_miss_batch_destroy(fmb);
3660     }
3661 }
3662 \f
3663 /* Flow expiration. */
3664
3665 static int subfacet_max_idle(const struct dpif_backer *);
3666 static void update_stats(struct dpif_backer *);
3667 static void rule_expire(struct rule_dpif *);
3668 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3669
3670 /* This function is called periodically by run().  Its job is to collect
3671  * updates for the flows that have been installed into the datapath, most
3672  * importantly when they last were used, and then use that information to
3673  * expire flows that have not been used recently.
3674  *
3675  * Returns the number of milliseconds after which it should be called again. */
3676 static int
3677 expire(struct dpif_backer *backer)
3678 {
3679     struct ofproto_dpif *ofproto;
3680     size_t n_subfacets;
3681     int max_idle;
3682
3683     /* Periodically clear out the drop keys in an effort to keep them
3684      * relatively few. */
3685     drop_key_clear(backer);
3686
3687     /* Update stats for each flow in the backer. */
3688     update_stats(backer);
3689
3690     n_subfacets = hmap_count(&backer->subfacets);
3691     if (n_subfacets) {
3692         struct subfacet *subfacet;
3693         long long int total, now;
3694
3695         total = 0;
3696         now = time_msec();
3697         HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3698             total += now - subfacet->created;
3699         }
3700         backer->avg_subfacet_life += total / n_subfacets;
3701     }
3702     backer->avg_subfacet_life /= 2;
3703
3704     backer->avg_n_subfacet += n_subfacets;
3705     backer->avg_n_subfacet /= 2;
3706
3707     backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3708
3709     max_idle = subfacet_max_idle(backer);
3710     expire_subfacets(backer, max_idle);
3711
3712     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3713         struct rule *rule, *next_rule;
3714
3715         if (ofproto->backer != backer) {
3716             continue;
3717         }
3718
3719         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3720          * has passed. */
3721         ovs_mutex_lock(&ofproto->up.expirable_mutex);
3722         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3723                             &ofproto->up.expirable) {
3724             rule_expire(rule_dpif_cast(rule));
3725         }
3726         ovs_mutex_unlock(&ofproto->up.expirable_mutex);
3727
3728         /* All outstanding data in existing flows has been accounted, so it's a
3729          * good time to do bond rebalancing. */
3730         if (ofproto->has_bonded_bundles) {
3731             struct ofbundle *bundle;
3732
3733             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3734                 if (bundle->bond) {
3735                     bond_rebalance(bundle->bond);
3736                 }
3737             }
3738         }
3739     }
3740
3741     return MIN(max_idle, 1000);
3742 }
3743
3744 /* Updates flow table statistics given that the datapath just reported 'stats'
3745  * as 'subfacet''s statistics. */
3746 static void
3747 update_subfacet_stats(struct subfacet *subfacet,
3748                       const struct dpif_flow_stats *stats)
3749 {
3750     struct facet *facet = subfacet->facet;
3751     struct dpif_flow_stats diff;
3752
3753     diff.tcp_flags = stats->tcp_flags;
3754     diff.used = stats->used;
3755
3756     if (stats->n_packets >= subfacet->dp_packet_count) {
3757         diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3758     } else {
3759         VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3760         diff.n_packets = 0;
3761     }
3762
3763     if (stats->n_bytes >= subfacet->dp_byte_count) {
3764         diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3765     } else {
3766         VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3767         diff.n_bytes = 0;
3768     }
3769
3770     facet->ofproto->n_hit += diff.n_packets;
3771     subfacet->dp_packet_count = stats->n_packets;
3772     subfacet->dp_byte_count = stats->n_bytes;
3773     subfacet_update_stats(subfacet, &diff);
3774
3775     if (facet->accounted_bytes < facet->byte_count) {
3776         facet_learn(facet);
3777         facet_account(facet);
3778         facet->accounted_bytes = facet->byte_count;
3779     }
3780 }
3781
3782 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3783  * about, or a flow that shouldn't be installed but was anyway.  Delete it. */
3784 static void
3785 delete_unexpected_flow(struct dpif_backer *backer,
3786                        const struct nlattr *key, size_t key_len)
3787 {
3788     if (!VLOG_DROP_WARN(&rl)) {
3789         struct ds s;
3790
3791         ds_init(&s);
3792         odp_flow_key_format(key, key_len, &s);
3793         VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3794         ds_destroy(&s);
3795     }
3796
3797     COVERAGE_INC(facet_unexpected);
3798     dpif_flow_del(backer->dpif, key, key_len, NULL);
3799 }
3800
3801 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3802  *
3803  * This function also pushes statistics updates to rules which each facet
3804  * resubmits into.  Generally these statistics will be accurate.  However, if a
3805  * facet changes the rule it resubmits into at some time in between
3806  * update_stats() runs, it is possible that statistics accrued to the
3807  * old rule will be incorrectly attributed to the new rule.  This could be
3808  * avoided by calling update_stats() whenever rules are created or
3809  * deleted.  However, the performance impact of making so many calls to the
3810  * datapath do not justify the benefit of having perfectly accurate statistics.
3811  *
3812  * In addition, this function maintains per ofproto flow hit counts. The patch
3813  * port is not treated specially. e.g. A packet ingress from br0 patched into
3814  * br1 will increase the hit count of br0 by 1, however, does not affect
3815  * the hit or miss counts of br1.
3816  */
3817 static void
3818 update_stats(struct dpif_backer *backer)
3819 {
3820     const struct dpif_flow_stats *stats;
3821     struct dpif_flow_dump dump;
3822     const struct nlattr *key, *mask;
3823     size_t key_len, mask_len;
3824
3825     dpif_flow_dump_start(&dump, backer->dpif);
3826     while (dpif_flow_dump_next(&dump, &key, &key_len,
3827                                &mask, &mask_len, NULL, NULL, &stats)) {
3828         struct subfacet *subfacet;
3829         uint32_t key_hash;
3830
3831         key_hash = odp_flow_key_hash(key, key_len);
3832         subfacet = subfacet_find(backer, key, key_len, key_hash);
3833         switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3834         case SF_FAST_PATH:
3835             update_subfacet_stats(subfacet, stats);
3836             break;
3837
3838         case SF_SLOW_PATH:
3839             /* Stats are updated per-packet. */
3840             break;
3841
3842         case SF_NOT_INSTALLED:
3843         default:
3844             delete_unexpected_flow(backer, key, key_len);
3845             break;
3846         }
3847         run_fast_rl();
3848     }
3849     dpif_flow_dump_done(&dump);
3850
3851     update_moving_averages(backer);
3852 }
3853
3854 /* Calculates and returns the number of milliseconds of idle time after which
3855  * subfacets should expire from the datapath.  When a subfacet expires, we fold
3856  * its statistics into its facet, and when a facet's last subfacet expires, we
3857  * fold its statistic into its rule. */
3858 static int
3859 subfacet_max_idle(const struct dpif_backer *backer)
3860 {
3861     /*
3862      * Idle time histogram.
3863      *
3864      * Most of the time a switch has a relatively small number of subfacets.
3865      * When this is the case we might as well keep statistics for all of them
3866      * in userspace and to cache them in the kernel datapath for performance as
3867      * well.
3868      *
3869      * As the number of subfacets increases, the memory required to maintain
3870      * statistics about them in userspace and in the kernel becomes
3871      * significant.  However, with a large number of subfacets it is likely
3872      * that only a few of them are "heavy hitters" that consume a large amount
3873      * of bandwidth.  At this point, only heavy hitters are worth caching in
3874      * the kernel and maintaining in userspaces; other subfacets we can
3875      * discard.
3876      *
3877      * The technique used to compute the idle time is to build a histogram with
3878      * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each.  Each subfacet
3879      * that is installed in the kernel gets dropped in the appropriate bucket.
3880      * After the histogram has been built, we compute the cutoff so that only
3881      * the most-recently-used 1% of subfacets (but at least
3882      * flow_eviction_threshold flows) are kept cached.  At least
3883      * the most-recently-used bucket of subfacets is kept, so actually an
3884      * arbitrary number of subfacets can be kept in any given expiration run
3885      * (though the next run will delete most of those unless they receive
3886      * additional data).
3887      *
3888      * This requires a second pass through the subfacets, in addition to the
3889      * pass made by update_stats(), because the former function never looks at
3890      * uninstallable subfacets.
3891      */
3892     enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3893     enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3894     int buckets[N_BUCKETS] = { 0 };
3895     int total, subtotal, bucket;
3896     struct subfacet *subfacet;
3897     long long int now;
3898     int i;
3899
3900     total = hmap_count(&backer->subfacets);
3901     if (total <= flow_eviction_threshold) {
3902         return N_BUCKETS * BUCKET_WIDTH;
3903     }
3904
3905     /* Build histogram. */
3906     now = time_msec();
3907     HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3908         long long int idle = now - subfacet->used;
3909         int bucket = (idle <= 0 ? 0
3910                       : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3911                       : (unsigned int) idle / BUCKET_WIDTH);
3912         buckets[bucket]++;
3913     }
3914
3915     /* Find the first bucket whose flows should be expired. */
3916     subtotal = bucket = 0;
3917     do {
3918         subtotal += buckets[bucket++];
3919     } while (bucket < N_BUCKETS &&
3920              subtotal < MAX(flow_eviction_threshold, total / 100));
3921
3922     if (VLOG_IS_DBG_ENABLED()) {
3923         struct ds s;
3924
3925         ds_init(&s);
3926         ds_put_cstr(&s, "keep");
3927         for (i = 0; i < N_BUCKETS; i++) {
3928             if (i == bucket) {
3929                 ds_put_cstr(&s, ", drop");
3930             }
3931             if (buckets[i]) {
3932                 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3933             }
3934         }
3935         VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3936         ds_destroy(&s);
3937     }
3938
3939     return bucket * BUCKET_WIDTH;
3940 }
3941
3942 static void
3943 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3944 {
3945     /* Cutoff time for most flows. */
3946     long long int normal_cutoff = time_msec() - dp_max_idle;
3947
3948     /* We really want to keep flows for special protocols around, so use a more
3949      * conservative cutoff. */
3950     long long int special_cutoff = time_msec() - 10000;
3951
3952     struct subfacet *subfacet, *next_subfacet;
3953     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3954     int n_batch;
3955
3956     n_batch = 0;
3957     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3958                         &backer->subfacets) {
3959         long long int cutoff;
3960
3961         cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3962                                                 | SLOW_STP)
3963                   ? special_cutoff
3964                   : normal_cutoff);
3965         if (subfacet->used < cutoff) {
3966             if (subfacet->path != SF_NOT_INSTALLED) {
3967                 batch[n_batch++] = subfacet;
3968                 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3969                     subfacet_destroy_batch(backer, batch, n_batch);
3970                     n_batch = 0;
3971                 }
3972             } else {
3973                 subfacet_destroy(subfacet);
3974             }
3975         }
3976     }
3977
3978     if (n_batch > 0) {
3979         subfacet_destroy_batch(backer, batch, n_batch);
3980     }
3981 }
3982
3983 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3984  * then delete it entirely. */
3985 static void
3986 rule_expire(struct rule_dpif *rule)
3987 {
3988     uint16_t idle_timeout, hard_timeout;
3989     long long int now;
3990     uint8_t reason;
3991
3992     if (rule->up.pending) {
3993         /* We'll have to expire it later. */
3994         return;
3995     }
3996
3997     ovs_mutex_lock(&rule->up.timeout_mutex);
3998     hard_timeout = rule->up.hard_timeout;
3999     idle_timeout = rule->up.idle_timeout;
4000     ovs_mutex_unlock(&rule->up.timeout_mutex);
4001
4002     /* Has 'rule' expired? */
4003     now = time_msec();
4004     if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
4005         reason = OFPRR_HARD_TIMEOUT;
4006     } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
4007         reason = OFPRR_IDLE_TIMEOUT;
4008     } else {
4009         return;
4010     }
4011
4012     COVERAGE_INC(ofproto_dpif_expired);
4013     ofproto_rule_expire(&rule->up, reason);
4014 }
4015 \f
4016 /* Facets. */
4017
4018 /* Creates and returns a new facet based on 'miss'.
4019  *
4020  * The caller must already have determined that no facet with an identical
4021  * 'miss->flow' exists in 'miss->ofproto'.
4022  *
4023  * 'rule' and 'xout' must have been created based on 'miss'.
4024  *
4025  * 'facet'' statistics are initialized based on 'stats'.
4026  *
4027  * The facet will initially have no subfacets.  The caller should create (at
4028  * least) one subfacet with subfacet_create(). */
4029 static struct facet *
4030 facet_create(const struct flow_miss *miss)
4031 {
4032     struct ofproto_dpif *ofproto = miss->ofproto;
4033     struct facet *facet;
4034     struct match match;
4035
4036     facet = xzalloc(sizeof *facet);
4037     facet->ofproto = miss->ofproto;
4038     facet->used = miss->stats.used;
4039     facet->flow = miss->flow;
4040     facet->learn_rl = time_msec() + 500;
4041
4042     list_init(&facet->subfacets);
4043     netflow_flow_init(&facet->nf_flow);
4044     netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4045
4046     xlate_out_copy(&facet->xout, &miss->xout);
4047
4048     match_init(&match, &facet->flow, &facet->xout.wc);
4049     cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4050     ovs_rwlock_wrlock(&ofproto->facets.rwlock);
4051     classifier_insert(&ofproto->facets, &facet->cr);
4052     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4053
4054     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4055     return facet;
4056 }
4057
4058 static void
4059 facet_free(struct facet *facet)
4060 {
4061     if (facet) {
4062         xlate_out_uninit(&facet->xout);
4063         free(facet);
4064     }
4065 }
4066
4067 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4068  * 'packet', which arrived on 'in_port'. */
4069 static bool
4070 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4071                     const struct nlattr *odp_actions, size_t actions_len,
4072                     struct ofpbuf *packet)
4073 {
4074     struct odputil_keybuf keybuf;
4075     struct ofpbuf key;
4076     int error;
4077
4078     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4079     odp_flow_key_from_flow(&key, flow,
4080                            ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4081
4082     error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4083                          odp_actions, actions_len, packet);
4084     return !error;
4085 }
4086
4087 /* Remove 'facet' from its ofproto and free up the associated memory:
4088  *
4089  *   - If 'facet' was installed in the datapath, uninstalls it and updates its
4090  *     rule's statistics, via subfacet_uninstall().
4091  *
4092  *   - Removes 'facet' from its rule and from ofproto->facets.
4093  */
4094 static void
4095 facet_remove(struct facet *facet)
4096 {
4097     struct subfacet *subfacet, *next_subfacet;
4098
4099     ovs_assert(!list_is_empty(&facet->subfacets));
4100
4101     /* First uninstall all of the subfacets to get final statistics. */
4102     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4103         subfacet_uninstall(subfacet);
4104     }
4105
4106     /* Flush the final stats to the rule.
4107      *
4108      * This might require us to have at least one subfacet around so that we
4109      * can use its actions for accounting in facet_account(), which is why we
4110      * have uninstalled but not yet destroyed the subfacets. */
4111     facet_flush_stats(facet);
4112
4113     /* Now we're really all done so destroy everything. */
4114     LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4115                         &facet->subfacets) {
4116         subfacet_destroy__(subfacet);
4117     }
4118     ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4119     classifier_remove(&facet->ofproto->facets, &facet->cr);
4120     ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4121     cls_rule_destroy(&facet->cr);
4122     facet_free(facet);
4123 }
4124
4125 /* Feed information from 'facet' back into the learning table to keep it in
4126  * sync with what is actually flowing through the datapath. */
4127 static void
4128 facet_learn(struct facet *facet)
4129 {
4130     long long int now = time_msec();
4131
4132     if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4133         return;
4134     }
4135
4136     facet->learn_rl = now + 500;
4137
4138     if (!facet->xout.has_learn
4139         && !facet->xout.has_normal
4140         && (!facet->xout.has_fin_timeout
4141             || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4142         return;
4143     }
4144
4145     facet_push_stats(facet, true);
4146 }
4147
4148 static void
4149 facet_account(struct facet *facet)
4150 {
4151     const struct nlattr *a;
4152     unsigned int left;
4153     ovs_be16 vlan_tci;
4154     uint64_t n_bytes;
4155
4156     if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4157         return;
4158     }
4159     n_bytes = facet->byte_count - facet->accounted_bytes;
4160
4161     /* This loop feeds byte counters to bond_account() for rebalancing to use
4162      * as a basis.  We also need to track the actual VLAN on which the packet
4163      * is going to be sent to ensure that it matches the one passed to
4164      * bond_choose_output_slave().  (Otherwise, we will account to the wrong
4165      * hash bucket.)
4166      *
4167      * We use the actions from an arbitrary subfacet because they should all
4168      * be equally valid for our purpose. */
4169     vlan_tci = facet->flow.vlan_tci;
4170     NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4171                              facet->xout.odp_actions.size) {
4172         const struct ovs_action_push_vlan *vlan;
4173         struct ofport_dpif *port;
4174
4175         switch (nl_attr_type(a)) {
4176         case OVS_ACTION_ATTR_OUTPUT:
4177             port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4178             if (port && port->bundle && port->bundle->bond) {
4179                 bond_account(port->bundle->bond, &facet->flow,
4180                              vlan_tci_to_vid(vlan_tci), n_bytes);
4181             }
4182             break;
4183
4184         case OVS_ACTION_ATTR_POP_VLAN:
4185             vlan_tci = htons(0);
4186             break;
4187
4188         case OVS_ACTION_ATTR_PUSH_VLAN:
4189             vlan = nl_attr_get(a);
4190             vlan_tci = vlan->vlan_tci;
4191             break;
4192         }
4193     }
4194 }
4195
4196 /* Returns true if the only action for 'facet' is to send to the controller.
4197  * (We don't report NetFlow expiration messages for such facets because they
4198  * are just part of the control logic for the network, not real traffic). */
4199 static bool
4200 facet_is_controller_flow(struct facet *facet)
4201 {
4202     if (facet) {
4203         struct ofproto_dpif *ofproto = facet->ofproto;
4204         const struct ofpact *ofpacts;
4205         struct rule_dpif *rule;
4206         size_t ofpacts_len;
4207         bool is_controller;
4208
4209         rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4210         ofpacts_len = rule->up.ofpacts_len;
4211         ofpacts = rule->up.ofpacts;
4212         is_controller = ofpacts_len > 0
4213             && ofpacts->type == OFPACT_CONTROLLER
4214             && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4215         rule_dpif_release(rule);
4216         return is_controller;
4217     }
4218     return false;
4219 }
4220
4221 /* Folds all of 'facet''s statistics into its rule.  Also updates the
4222  * accounting ofhook and emits a NetFlow expiration if appropriate.  All of
4223  * 'facet''s statistics in the datapath should have been zeroed and folded into
4224  * its packet and byte counts before this function is called. */
4225 static void
4226 facet_flush_stats(struct facet *facet)
4227 {
4228     struct ofproto_dpif *ofproto = facet->ofproto;
4229     struct subfacet *subfacet;
4230
4231     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4232         ovs_assert(!subfacet->dp_byte_count);
4233         ovs_assert(!subfacet->dp_packet_count);
4234     }
4235
4236     facet_push_stats(facet, false);
4237     if (facet->accounted_bytes < facet->byte_count) {
4238         facet_account(facet);
4239         facet->accounted_bytes = facet->byte_count;
4240     }
4241
4242     if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4243         struct ofexpired expired;
4244         expired.flow = facet->flow;
4245         expired.packet_count = facet->packet_count;
4246         expired.byte_count = facet->byte_count;
4247         expired.used = facet->used;
4248         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4249     }
4250
4251     /* Reset counters to prevent double counting if 'facet' ever gets
4252      * reinstalled. */
4253     facet_reset_counters(facet);
4254
4255     netflow_flow_clear(&facet->nf_flow);
4256     facet->tcp_flags = 0;
4257 }
4258
4259 /* Searches 'ofproto''s table of facets for one which would be responsible for
4260  * 'flow'.  Returns it if found, otherwise a null pointer.
4261  *
4262  * The returned facet might need revalidation; use facet_lookup_valid()
4263  * instead if that is important. */
4264 static struct facet *
4265 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4266 {
4267     struct cls_rule *cr;
4268
4269     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4270     cr = classifier_lookup(&ofproto->facets, flow, NULL);
4271     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4272     return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4273 }
4274
4275 /* Searches 'ofproto''s table of facets for one capable that covers
4276  * 'flow'.  Returns it if found, otherwise a null pointer.
4277  *
4278  * The returned facet is guaranteed to be valid. */
4279 static struct facet *
4280 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4281 {
4282     struct facet *facet;
4283
4284     facet = facet_find(ofproto, flow);
4285     if (facet
4286         && ofproto->backer->need_revalidate
4287         && !facet_revalidate(facet)) {
4288         return NULL;
4289     }
4290
4291     return facet;
4292 }
4293
4294 static bool
4295 facet_check_consistency(struct facet *facet)
4296 {
4297     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4298
4299     struct xlate_out xout;
4300     struct xlate_in xin;
4301
4302     struct rule_dpif *rule;
4303     bool ok;
4304
4305     /* Check the datapath actions for consistency. */
4306     rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4307     xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4308     xlate_actions(&xin, &xout);
4309     rule_dpif_release(rule);
4310
4311     ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4312         && facet->xout.slow == xout.slow;
4313     if (!ok && !VLOG_DROP_WARN(&rl)) {
4314         struct ds s = DS_EMPTY_INITIALIZER;
4315
4316         flow_format(&s, &facet->flow);
4317         ds_put_cstr(&s, ": inconsistency in facet");
4318
4319         if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4320             ds_put_cstr(&s, " (actions were: ");
4321             format_odp_actions(&s, facet->xout.odp_actions.data,
4322                                facet->xout.odp_actions.size);
4323             ds_put_cstr(&s, ") (correct actions: ");
4324             format_odp_actions(&s, xout.odp_actions.data,
4325                                xout.odp_actions.size);
4326             ds_put_char(&s, ')');
4327         }
4328
4329         if (facet->xout.slow != xout.slow) {
4330             ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4331         }
4332
4333         ds_destroy(&s);
4334     }
4335     xlate_out_uninit(&xout);
4336
4337     return ok;
4338 }
4339
4340 /* Re-searches the classifier for 'facet':
4341  *
4342  *   - If the rule found is different from 'facet''s current rule, moves
4343  *     'facet' to the new rule and recompiles its actions.
4344  *
4345  *   - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4346  *     where it is and recompiles its actions anyway.
4347  *
4348  *   - If any of 'facet''s subfacets correspond to a new flow according to
4349  *     xlate_receive(), 'facet' is removed.
4350  *
4351  *   Returns true if 'facet' is still valid.  False if 'facet' was removed. */
4352 static bool
4353 facet_revalidate(struct facet *facet)
4354 {
4355     struct ofproto_dpif *ofproto = facet->ofproto;
4356     struct rule_dpif *new_rule;
4357     struct subfacet *subfacet;
4358     struct flow_wildcards wc;
4359     struct xlate_out xout;
4360     struct xlate_in xin;
4361
4362     COVERAGE_INC(facet_revalidate);
4363
4364     /* Check that child subfacets still correspond to this facet.  Tunnel
4365      * configuration changes could cause a subfacet's OpenFlow in_port to
4366      * change. */
4367     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4368         struct ofproto_dpif *recv_ofproto;
4369         struct flow recv_flow;
4370         int error;
4371
4372         error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4373                               subfacet->key_len, &recv_flow, NULL,
4374                               &recv_ofproto, NULL);
4375         if (error
4376             || recv_ofproto != ofproto
4377             || facet != facet_find(ofproto, &recv_flow)) {
4378             facet_remove(facet);
4379             return false;
4380         }
4381     }
4382
4383     flow_wildcards_init_catchall(&wc);
4384     rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4385
4386     /* Calculate new datapath actions.
4387      *
4388      * We do not modify any 'facet' state yet, because we might need to, e.g.,
4389      * emit a NetFlow expiration and, if so, we need to have the old state
4390      * around to properly compose it. */
4391     xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4392     xlate_actions(&xin, &xout);
4393     flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4394
4395     /* A facet's slow path reason should only change under dramatic
4396      * circumstances.  Rather than try to update everything, it's simpler to
4397      * remove the facet and start over.
4398      *
4399      * More importantly, if a facet's wildcards change, it will be relatively
4400      * difficult to figure out if its subfacets still belong to it, and if not
4401      * which facet they may belong to.  Again, to avoid the complexity, we
4402      * simply give up instead. */
4403     if (facet->xout.slow != xout.slow
4404         || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4405         facet_remove(facet);
4406         xlate_out_uninit(&xout);
4407         rule_dpif_release(new_rule);
4408         return false;
4409     }
4410
4411     if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4412         LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4413             if (subfacet->path == SF_FAST_PATH) {
4414                 struct dpif_flow_stats stats;
4415
4416                 subfacet_install(subfacet, &xout.odp_actions, &stats);
4417                 subfacet_update_stats(subfacet, &stats);
4418             }
4419         }
4420
4421         facet_flush_stats(facet);
4422
4423         ofpbuf_clear(&facet->xout.odp_actions);
4424         ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4425                    xout.odp_actions.size);
4426     }
4427
4428     /* Update 'facet' now that we've taken care of all the old state. */
4429     facet->xout.slow = xout.slow;
4430     facet->xout.has_learn = xout.has_learn;
4431     facet->xout.has_normal = xout.has_normal;
4432     facet->xout.has_fin_timeout = xout.has_fin_timeout;
4433     facet->xout.nf_output_iface = xout.nf_output_iface;
4434     facet->xout.mirrors = xout.mirrors;
4435     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4436     facet->used = MAX(facet->used, new_rule->up.created);
4437
4438     xlate_out_uninit(&xout);
4439     rule_dpif_release(new_rule);
4440     return true;
4441 }
4442
4443 static void
4444 facet_reset_counters(struct facet *facet)
4445 {
4446     facet->packet_count = 0;
4447     facet->byte_count = 0;
4448     facet->prev_packet_count = 0;
4449     facet->prev_byte_count = 0;
4450     facet->accounted_bytes = 0;
4451 }
4452
4453 static void
4454 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4455                 struct dpif_flow_stats *stats, bool may_learn)
4456 {
4457     struct ofport_dpif *in_port;
4458     struct rule_dpif *rule;
4459     struct xlate_in xin;
4460
4461     in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4462     if (in_port && in_port->is_tunnel) {
4463         netdev_vport_inc_rx(in_port->up.netdev, stats);
4464     }
4465
4466     rule_dpif_lookup(ofproto, flow, NULL, &rule);
4467     rule_dpif_credit_stats(rule, stats);
4468     xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4469     xin.resubmit_stats = stats;
4470     xin.may_learn = may_learn;
4471     xlate_actions_for_side_effects(&xin);
4472     rule_dpif_release(rule);
4473 }
4474
4475 static void
4476 facet_push_stats(struct facet *facet, bool may_learn)
4477 {
4478     struct dpif_flow_stats stats;
4479
4480     ovs_assert(facet->packet_count >= facet->prev_packet_count);
4481     ovs_assert(facet->byte_count >= facet->prev_byte_count);
4482     ovs_assert(facet->used >= facet->prev_used);
4483
4484     stats.n_packets = facet->packet_count - facet->prev_packet_count;
4485     stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4486     stats.used = facet->used;
4487     stats.tcp_flags = facet->tcp_flags;
4488
4489     if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4490         facet->prev_packet_count = facet->packet_count;
4491         facet->prev_byte_count = facet->byte_count;
4492         facet->prev_used = facet->used;
4493
4494         netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4495                                  facet->used);
4496         netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4497         mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4498                             stats.n_packets, stats.n_bytes);
4499         flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4500     }
4501 }
4502
4503 static void
4504 push_all_stats__(bool run_fast)
4505 {
4506     static long long int rl = LLONG_MIN;
4507     struct ofproto_dpif *ofproto;
4508
4509     if (time_msec() < rl) {
4510         return;
4511     }
4512
4513     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4514         struct cls_cursor cursor;
4515         struct facet *facet;
4516
4517         ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4518         cls_cursor_init(&cursor, &ofproto->facets, NULL);
4519         CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4520             facet_push_stats(facet, false);
4521             if (run_fast) {
4522                 run_fast_rl();
4523             }
4524         }
4525         ovs_rwlock_unlock(&ofproto->facets.rwlock);
4526     }
4527
4528     rl = time_msec() + 100;
4529 }
4530
4531 static void
4532 push_all_stats(void)
4533 {
4534     push_all_stats__(true);
4535 }
4536
4537 void
4538 rule_dpif_credit_stats(struct rule_dpif *rule,
4539                        const struct dpif_flow_stats *stats)
4540 {
4541     ovs_mutex_lock(&rule->stats_mutex);
4542     rule->packet_count += stats->n_packets;
4543     rule->byte_count += stats->n_bytes;
4544     rule->up.used = MAX(rule->up.used, stats->used);
4545     ovs_mutex_unlock(&rule->stats_mutex);
4546 }
4547
4548 bool
4549 rule_dpif_fail_open(const struct rule_dpif *rule)
4550 {
4551     return rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4552 }
4553
4554 ovs_be64
4555 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4556 {
4557     return rule->up.flow_cookie;
4558 }
4559
4560 void
4561 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4562                      uint16_t hard_timeout)
4563 {
4564     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4565 }
4566
4567 void
4568 rule_dpif_get_actions(const struct rule_dpif *rule,
4569                       const struct ofpact **ofpacts, size_t *ofpacts_len)
4570 {
4571     *ofpacts = rule->up.ofpacts;
4572     *ofpacts_len = rule->up.ofpacts_len;
4573 }
4574 \f
4575 /* Subfacets. */
4576
4577 static struct subfacet *
4578 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4579               size_t key_len, uint32_t key_hash)
4580 {
4581     struct subfacet *subfacet;
4582
4583     HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4584                              &backer->subfacets) {
4585         if (subfacet->key_len == key_len
4586             && !memcmp(key, subfacet->key, key_len)) {
4587             return subfacet;
4588         }
4589     }
4590
4591     return NULL;
4592 }
4593
4594 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4595  * 'key_fitness', 'key', and 'key_len' members in 'miss'.  Returns the
4596  * existing subfacet if there is one, otherwise creates and returns a
4597  * new subfacet. */
4598 static struct subfacet *
4599 subfacet_create(struct facet *facet, struct flow_miss *miss)
4600 {
4601     struct dpif_backer *backer = miss->ofproto->backer;
4602     enum odp_key_fitness key_fitness = miss->key_fitness;
4603     const struct nlattr *key = miss->key;
4604     size_t key_len = miss->key_len;
4605     uint32_t key_hash;
4606     struct subfacet *subfacet;
4607
4608     key_hash = odp_flow_key_hash(key, key_len);
4609
4610     if (list_is_empty(&facet->subfacets)) {
4611         subfacet = &facet->one_subfacet;
4612     } else {
4613         subfacet = subfacet_find(backer, key, key_len, key_hash);
4614         if (subfacet) {
4615             if (subfacet->facet == facet) {
4616                 return subfacet;
4617             }
4618
4619             /* This shouldn't happen. */
4620             VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4621             subfacet_destroy(subfacet);
4622         }
4623
4624         subfacet = xmalloc(sizeof *subfacet);
4625     }
4626
4627     hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4628     list_push_back(&facet->subfacets, &subfacet->list_node);
4629     subfacet->facet = facet;
4630     subfacet->key_fitness = key_fitness;
4631     subfacet->key = xmemdup(key, key_len);
4632     subfacet->key_len = key_len;
4633     subfacet->used = miss->stats.used;
4634     subfacet->created = subfacet->used;
4635     subfacet->dp_packet_count = 0;
4636     subfacet->dp_byte_count = 0;
4637     subfacet->path = SF_NOT_INSTALLED;
4638     subfacet->backer = backer;
4639
4640     backer->subfacet_add_count++;
4641     return subfacet;
4642 }
4643
4644 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4645  * its facet within 'ofproto', and frees it. */
4646 static void
4647 subfacet_destroy__(struct subfacet *subfacet)
4648 {
4649     struct facet *facet = subfacet->facet;
4650     struct ofproto_dpif *ofproto = facet->ofproto;
4651
4652     /* Update ofproto stats before uninstall the subfacet. */
4653     ofproto->backer->subfacet_del_count++;
4654
4655     subfacet_uninstall(subfacet);
4656     hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4657     list_remove(&subfacet->list_node);
4658     free(subfacet->key);
4659     if (subfacet != &facet->one_subfacet) {
4660         free(subfacet);
4661     }
4662 }
4663
4664 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4665  * last remaining subfacet in its facet destroys the facet too. */
4666 static void
4667 subfacet_destroy(struct subfacet *subfacet)
4668 {
4669     struct facet *facet = subfacet->facet;
4670
4671     if (list_is_singleton(&facet->subfacets)) {
4672         /* facet_remove() needs at least one subfacet (it will remove it). */
4673         facet_remove(facet);
4674     } else {
4675         subfacet_destroy__(subfacet);
4676     }
4677 }
4678
4679 static void
4680 subfacet_destroy_batch(struct dpif_backer *backer,
4681                        struct subfacet **subfacets, int n)
4682 {
4683     struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4684     struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4685     struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4686     int i;
4687
4688     for (i = 0; i < n; i++) {
4689         ops[i].type = DPIF_OP_FLOW_DEL;
4690         ops[i].u.flow_del.key = subfacets[i]->key;
4691         ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4692         ops[i].u.flow_del.stats = &stats[i];
4693         opsp[i] = &ops[i];
4694     }
4695
4696     dpif_operate(backer->dpif, opsp, n);
4697     for (i = 0; i < n; i++) {
4698         subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4699         subfacets[i]->path = SF_NOT_INSTALLED;
4700         subfacet_destroy(subfacets[i]);
4701         run_fast_rl();
4702     }
4703 }
4704
4705 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4706  * bytes of actions in 'actions'.  If 'stats' is non-null, statistics counters
4707  * in the datapath will be zeroed and 'stats' will be updated with traffic new
4708  * since 'subfacet' was last updated.
4709  *
4710  * Returns 0 if successful, otherwise a positive errno value. */
4711 static int
4712 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4713                  struct dpif_flow_stats *stats)
4714 {
4715     struct facet *facet = subfacet->facet;
4716     enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4717     const struct nlattr *actions = odp_actions->data;
4718     size_t actions_len = odp_actions->size;
4719     struct odputil_keybuf maskbuf;
4720     struct ofpbuf mask;
4721
4722     uint64_t slow_path_stub[128 / 8];
4723     enum dpif_flow_put_flags flags;
4724     int ret;
4725
4726     flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4727                                                : DPIF_FP_MODIFY;
4728     if (stats) {
4729         flags |= DPIF_FP_ZERO_STATS;
4730     }
4731
4732     if (path == SF_SLOW_PATH) {
4733         compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4734                           slow_path_stub, sizeof slow_path_stub,
4735                           &actions, &actions_len);
4736     }
4737
4738     ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4739     if (enable_megaflows) {
4740         odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4741                                &facet->flow, UINT32_MAX);
4742     }
4743
4744     ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4745                         subfacet->key_len,  mask.data, mask.size,
4746                         actions, actions_len, stats);
4747
4748     if (stats) {
4749         subfacet_reset_dp_stats(subfacet, stats);
4750     }
4751
4752     if (ret) {
4753         COVERAGE_INC(subfacet_install_fail);
4754     } else {
4755         subfacet->path = path;
4756     }
4757     return ret;
4758 }
4759
4760 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4761 static void
4762 subfacet_uninstall(struct subfacet *subfacet)
4763 {
4764     if (subfacet->path != SF_NOT_INSTALLED) {
4765         struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4766         struct dpif_flow_stats stats;
4767         int error;
4768
4769         error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4770                               subfacet->key_len, &stats);
4771         subfacet_reset_dp_stats(subfacet, &stats);
4772         if (!error) {
4773             subfacet_update_stats(subfacet, &stats);
4774         }
4775         subfacet->path = SF_NOT_INSTALLED;
4776     } else {
4777         ovs_assert(subfacet->dp_packet_count == 0);
4778         ovs_assert(subfacet->dp_byte_count == 0);
4779     }
4780 }
4781
4782 /* Resets 'subfacet''s datapath statistics counters.  This should be called
4783  * when 'subfacet''s statistics are cleared in the datapath.  If 'stats' is
4784  * non-null, it should contain the statistics returned by dpif when 'subfacet'
4785  * was reset in the datapath.  'stats' will be modified to include only
4786  * statistics new since 'subfacet' was last updated. */
4787 static void
4788 subfacet_reset_dp_stats(struct subfacet *subfacet,
4789                         struct dpif_flow_stats *stats)
4790 {
4791     if (stats
4792         && subfacet->dp_packet_count <= stats->n_packets
4793         && subfacet->dp_byte_count <= stats->n_bytes) {
4794         stats->n_packets -= subfacet->dp_packet_count;
4795         stats->n_bytes -= subfacet->dp_byte_count;
4796     }
4797
4798     subfacet->dp_packet_count = 0;
4799     subfacet->dp_byte_count = 0;
4800 }
4801
4802 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4803  *
4804  * Because of the meaning of a subfacet's counters, it only makes sense to do
4805  * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4806  * represents a packet that was sent by hand or if it represents statistics
4807  * that have been cleared out of the datapath. */
4808 static void
4809 subfacet_update_stats(struct subfacet *subfacet,
4810                       const struct dpif_flow_stats *stats)
4811 {
4812     if (stats->n_packets || stats->used > subfacet->used) {
4813         struct facet *facet = subfacet->facet;
4814
4815         subfacet->used = MAX(subfacet->used, stats->used);
4816         facet->used = MAX(facet->used, stats->used);
4817         facet->packet_count += stats->n_packets;
4818         facet->byte_count += stats->n_bytes;
4819         facet->tcp_flags |= stats->tcp_flags;
4820     }
4821 }
4822 \f
4823 /* Rules. */
4824
4825 /* Lookup 'flow' in 'ofproto''s classifier.  If 'wc' is non-null, sets
4826  * the fields that were relevant as part of the lookup. */
4827 void
4828 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4829                  struct flow_wildcards *wc, struct rule_dpif **rule)
4830 {
4831     struct ofport_dpif *port;
4832
4833     if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4834         return;
4835     }
4836     port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4837     if (!port) {
4838         VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4839                      flow->in_port.ofp_port);
4840     }
4841
4842     choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4843                      ofproto->no_packet_in_rule, rule);
4844 }
4845
4846 bool
4847 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4848                           const struct flow *flow, struct flow_wildcards *wc,
4849                           uint8_t table_id, struct rule_dpif **rule)
4850     OVS_TRY_RDLOCK(true, (*rule)->up.evict)
4851 {
4852     struct cls_rule *cls_rule;
4853     struct classifier *cls;
4854     bool frag;
4855
4856     *rule = NULL;
4857     if (table_id >= N_TABLES) {
4858         return false;
4859     }
4860
4861     if (wc) {
4862         memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4863         wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4864     }
4865
4866     cls = &ofproto->up.tables[table_id].cls;
4867     ovs_rwlock_rdlock(&cls->rwlock);
4868     frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4869     if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4870         /* We must pretend that transport ports are unavailable. */
4871         struct flow ofpc_normal_flow = *flow;
4872         ofpc_normal_flow.tp_src = htons(0);
4873         ofpc_normal_flow.tp_dst = htons(0);
4874         cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4875     } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4876         cls_rule = &ofproto->drop_frags_rule->up.cr;
4877         if (wc) {
4878             flow_wildcards_init_exact(wc);
4879         }
4880     } else {
4881         cls_rule = classifier_lookup(cls, flow, wc);
4882     }
4883
4884     *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4885     if (*rule && ovs_rwlock_tryrdlock(&(*rule)->up.evict)) {
4886         /* The rule is in the process of being removed.  Best we can do is
4887          * pretend it isn't there. */
4888         *rule = NULL;
4889     }
4890     ovs_rwlock_unlock(&cls->rwlock);
4891
4892     return *rule != NULL;
4893 }
4894
4895 /* Given a port configuration (specified as zero if there's no port), chooses
4896  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4897  * flow table miss. */
4898 void
4899 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4900                  struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4901     OVS_NO_THREAD_SAFETY_ANALYSIS
4902 {
4903     *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4904     ovs_rwlock_rdlock(&(*rule)->up.evict);
4905 }
4906
4907 void
4908 rule_dpif_release(struct rule_dpif *rule)
4909     OVS_NO_THREAD_SAFETY_ANALYSIS
4910 {
4911     if (rule) {
4912         ovs_rwlock_unlock(&rule->up.evict);
4913     }
4914 }
4915
4916 static void
4917 complete_operation(struct rule_dpif *rule)
4918 {
4919     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4920
4921     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4922     if (clogged) {
4923         struct dpif_completion *c = xmalloc(sizeof *c);
4924         c->op = rule->up.pending;
4925         list_push_back(&ofproto->completions, &c->list_node);
4926     } else {
4927         ofoperation_complete(rule->up.pending, 0);
4928     }
4929 }
4930
4931 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4932 {
4933     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4934 }
4935
4936 static struct rule *
4937 rule_alloc(void)
4938 {
4939     struct rule_dpif *rule = xmalloc(sizeof *rule);
4940     return &rule->up;
4941 }
4942
4943 static void
4944 rule_dealloc(struct rule *rule_)
4945 {
4946     struct rule_dpif *rule = rule_dpif_cast(rule_);
4947     free(rule);
4948 }
4949
4950 static enum ofperr
4951 rule_construct(struct rule *rule_)
4952 {
4953     struct rule_dpif *rule = rule_dpif_cast(rule_);
4954     ovs_mutex_init(&rule->stats_mutex);
4955     ovs_mutex_lock(&rule->stats_mutex);
4956     rule->packet_count = 0;
4957     rule->byte_count = 0;
4958     ovs_mutex_unlock(&rule->stats_mutex);
4959     return 0;
4960 }
4961
4962 static void
4963 rule_insert(struct rule *rule_)
4964 {
4965     struct rule_dpif *rule = rule_dpif_cast(rule_);
4966     complete_operation(rule);
4967 }
4968
4969 static void
4970 rule_delete(struct rule *rule_)
4971 {
4972     struct rule_dpif *rule = rule_dpif_cast(rule_);
4973     complete_operation(rule);
4974 }
4975
4976 static void
4977 rule_destruct(struct rule *rule_)
4978 {
4979     struct rule_dpif *rule = rule_dpif_cast(rule_);
4980     ovs_mutex_destroy(&rule->stats_mutex);
4981 }
4982
4983 static void
4984 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4985 {
4986     struct rule_dpif *rule = rule_dpif_cast(rule_);
4987
4988     /* push_all_stats() can handle flow misses which, when using the learn
4989      * action, can cause rules to be added and deleted.  This can corrupt our
4990      * caller's datastructures which assume that rule_get_stats() doesn't have
4991      * an impact on the flow table. To be safe, we disable miss handling. */
4992     push_all_stats__(false);
4993
4994     /* Start from historical data for 'rule' itself that are no longer tracked
4995      * in facets.  This counts, for example, facets that have expired. */
4996     ovs_mutex_lock(&rule->stats_mutex);
4997     *packets = rule->packet_count;
4998     *bytes = rule->byte_count;
4999     ovs_mutex_unlock(&rule->stats_mutex);
5000 }
5001
5002 static void
5003 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5004                   struct ofpbuf *packet)
5005 {
5006     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5007     struct dpif_flow_stats stats;
5008     struct xlate_out xout;
5009     struct xlate_in xin;
5010
5011     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5012     rule_dpif_credit_stats(rule, &stats);
5013
5014     xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
5015     xin.resubmit_stats = &stats;
5016     xlate_actions(&xin, &xout);
5017
5018     execute_odp_actions(ofproto, flow, xout.odp_actions.data,
5019                         xout.odp_actions.size, packet);
5020
5021     xlate_out_uninit(&xout);
5022 }
5023
5024 static enum ofperr
5025 rule_execute(struct rule *rule, const struct flow *flow,
5026              struct ofpbuf *packet)
5027 {
5028     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5029     ofpbuf_delete(packet);
5030     return 0;
5031 }
5032
5033 static void
5034 rule_modify_actions(struct rule *rule_, bool reset_counters)
5035 {
5036     struct rule_dpif *rule = rule_dpif_cast(rule_);
5037
5038     if (reset_counters) {
5039         ovs_mutex_lock(&rule->stats_mutex);
5040         rule->packet_count = 0;
5041         rule->byte_count = 0;
5042         ovs_mutex_unlock(&rule->stats_mutex);
5043     }
5044
5045     complete_operation(rule);
5046 }
5047 \f
5048 /* Sends 'packet' out 'ofport'.
5049  * May modify 'packet'.
5050  * Returns 0 if successful, otherwise a positive errno value. */
5051 static int
5052 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5053 {
5054     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5055     uint64_t odp_actions_stub[1024 / 8];
5056     struct ofpbuf key, odp_actions;
5057     struct dpif_flow_stats stats;
5058     struct odputil_keybuf keybuf;
5059     struct ofpact_output output;
5060     struct xlate_out xout;
5061     struct xlate_in xin;
5062     struct flow flow;
5063     union flow_in_port in_port_;
5064     int error;
5065
5066     ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5067     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5068
5069     /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5070     in_port_.ofp_port = OFPP_NONE;
5071     flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5072     odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5073                                                              OFPP_LOCAL));
5074     dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5075
5076     ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5077     output.port = ofport->up.ofp_port;
5078     output.max_len = 0;
5079
5080     xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5081     xin.ofpacts_len = sizeof output;
5082     xin.ofpacts = &output.ofpact;
5083     xin.resubmit_stats = &stats;
5084     xlate_actions(&xin, &xout);
5085
5086     error = dpif_execute(ofproto->backer->dpif,
5087                          key.data, key.size,
5088                          xout.odp_actions.data, xout.odp_actions.size,
5089                          packet);
5090     xlate_out_uninit(&xout);
5091
5092     if (error) {
5093         VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5094                      ofproto->up.name, netdev_get_name(ofport->up.netdev),
5095                      ovs_strerror(error));
5096     }
5097
5098     ofproto->stats.tx_packets++;
5099     ofproto->stats.tx_bytes += packet->size;
5100     return error;
5101 }
5102
5103 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5104  * The action will state 'slow' as the reason that the action is in the slow
5105  * path.  (This is purely informational: it allows a human viewing "ovs-dpctl
5106  * dump-flows" output to see why a flow is in the slow path.)
5107  *
5108  * The 'stub_size' bytes in 'stub' will be used to store the action.
5109  * 'stub_size' must be large enough for the action.
5110  *
5111  * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5112  * respectively. */
5113 static void
5114 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5115                   enum slow_path_reason slow,
5116                   uint64_t *stub, size_t stub_size,
5117                   const struct nlattr **actionsp, size_t *actions_lenp)
5118 {
5119     union user_action_cookie cookie;
5120     struct ofpbuf buf;
5121
5122     cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5123     cookie.slow_path.unused = 0;
5124     cookie.slow_path.reason = slow;
5125
5126     ofpbuf_use_stack(&buf, stub, stub_size);
5127     if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5128         uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5129                                          ODPP_NONE);
5130         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5131     } else {
5132         odp_port_t odp_port;
5133         uint32_t pid;
5134
5135         odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5136         pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5137         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5138     }
5139     *actionsp = buf.data;
5140     *actions_lenp = buf.size;
5141 }
5142 \f
5143 static bool
5144 set_frag_handling(struct ofproto *ofproto_,
5145                   enum ofp_config_flags frag_handling)
5146 {
5147     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5148     if (frag_handling != OFPC_FRAG_REASM) {
5149         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5150         return true;
5151     } else {
5152         return false;
5153     }
5154 }
5155
5156 static enum ofperr
5157 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5158            const struct flow *flow,
5159            const struct ofpact *ofpacts, size_t ofpacts_len)
5160 {
5161     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5162     struct odputil_keybuf keybuf;
5163     struct dpif_flow_stats stats;
5164     struct xlate_out xout;
5165     struct xlate_in xin;
5166     struct ofpbuf key;
5167
5168
5169     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5170     odp_flow_key_from_flow(&key, flow,
5171                            ofp_port_to_odp_port(ofproto,
5172                                       flow->in_port.ofp_port));
5173
5174     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5175
5176     xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5177     xin.resubmit_stats = &stats;
5178     xin.ofpacts_len = ofpacts_len;
5179     xin.ofpacts = ofpacts;
5180
5181     xlate_actions(&xin, &xout);
5182     dpif_execute(ofproto->backer->dpif, key.data, key.size,
5183                  xout.odp_actions.data, xout.odp_actions.size, packet);
5184     xlate_out_uninit(&xout);
5185
5186     return 0;
5187 }
5188 \f
5189 /* NetFlow. */
5190
5191 static int
5192 set_netflow(struct ofproto *ofproto_,
5193             const struct netflow_options *netflow_options)
5194 {
5195     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5196
5197     if (netflow_options) {
5198         if (!ofproto->netflow) {
5199             ofproto->netflow = netflow_create();
5200             ofproto->backer->need_revalidate = REV_RECONFIGURE;
5201         }
5202         return netflow_set_options(ofproto->netflow, netflow_options);
5203     } else if (ofproto->netflow) {
5204         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5205         netflow_destroy(ofproto->netflow);
5206         ofproto->netflow = NULL;
5207     }
5208
5209     return 0;
5210 }
5211
5212 static void
5213 get_netflow_ids(const struct ofproto *ofproto_,
5214                 uint8_t *engine_type, uint8_t *engine_id)
5215 {
5216     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5217
5218     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5219 }
5220
5221 static void
5222 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5223 {
5224     if (!facet_is_controller_flow(facet) &&
5225         netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5226         struct subfacet *subfacet;
5227         struct ofexpired expired;
5228
5229         LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5230             if (subfacet->path == SF_FAST_PATH) {
5231                 struct dpif_flow_stats stats;
5232
5233                 subfacet_install(subfacet, &facet->xout.odp_actions,
5234                                  &stats);
5235                 subfacet_update_stats(subfacet, &stats);
5236             }
5237         }
5238
5239         expired.flow = facet->flow;
5240         expired.packet_count = facet->packet_count;
5241         expired.byte_count = facet->byte_count;
5242         expired.used = facet->used;
5243         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5244     }
5245 }
5246
5247 static void
5248 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5249 {
5250     struct cls_cursor cursor;
5251     struct facet *facet;
5252
5253     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5254     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5255     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5256         send_active_timeout(ofproto, facet);
5257     }
5258     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5259 }
5260 \f
5261 static struct ofproto_dpif *
5262 ofproto_dpif_lookup(const char *name)
5263 {
5264     struct ofproto_dpif *ofproto;
5265
5266     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5267                              hash_string(name, 0), &all_ofproto_dpifs) {
5268         if (!strcmp(ofproto->up.name, name)) {
5269             return ofproto;
5270         }
5271     }
5272     return NULL;
5273 }
5274
5275 static void
5276 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5277                           const char *argv[], void *aux OVS_UNUSED)
5278 {
5279     struct ofproto_dpif *ofproto;
5280
5281     if (argc > 1) {
5282         ofproto = ofproto_dpif_lookup(argv[1]);
5283         if (!ofproto) {
5284             unixctl_command_reply_error(conn, "no such bridge");
5285             return;
5286         }
5287         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5288         mac_learning_flush(ofproto->ml);
5289         ovs_rwlock_unlock(&ofproto->ml->rwlock);
5290     } else {
5291         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5292             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5293             mac_learning_flush(ofproto->ml);
5294             ovs_rwlock_unlock(&ofproto->ml->rwlock);
5295         }
5296     }
5297
5298     unixctl_command_reply(conn, "table successfully flushed");
5299 }
5300
5301 static struct ofport_dpif *
5302 ofbundle_get_a_port(const struct ofbundle *bundle)
5303 {
5304     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5305                         bundle_node);
5306 }
5307
5308 static void
5309 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5310                          const char *argv[], void *aux OVS_UNUSED)
5311 {
5312     struct ds ds = DS_EMPTY_INITIALIZER;
5313     const struct ofproto_dpif *ofproto;
5314     const struct mac_entry *e;
5315
5316     ofproto = ofproto_dpif_lookup(argv[1]);
5317     if (!ofproto) {
5318         unixctl_command_reply_error(conn, "no such bridge");
5319         return;
5320     }
5321
5322     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
5323     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5324     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5325         struct ofbundle *bundle = e->port.p;
5326         char name[OFP_MAX_PORT_NAME_LEN];
5327
5328         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5329                                name, sizeof name);
5330         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
5331                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
5332                       mac_entry_age(ofproto->ml, e));
5333     }
5334     ovs_rwlock_unlock(&ofproto->ml->rwlock);
5335     unixctl_command_reply(conn, ds_cstr(&ds));
5336     ds_destroy(&ds);
5337 }
5338
5339 struct trace_ctx {
5340     struct xlate_out xout;
5341     struct xlate_in xin;
5342     struct flow flow;
5343     struct ds *result;
5344 };
5345
5346 static void
5347 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5348 {
5349     ds_put_char_multiple(result, '\t', level);
5350     if (!rule) {
5351         ds_put_cstr(result, "No match\n");
5352         return;
5353     }
5354
5355     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5356                   rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5357     cls_rule_format(&rule->up.cr, result);
5358     ds_put_char(result, '\n');
5359
5360     ds_put_char_multiple(result, '\t', level);
5361     ds_put_cstr(result, "OpenFlow ");
5362     ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5363     ds_put_char(result, '\n');
5364 }
5365
5366 static void
5367 trace_format_flow(struct ds *result, int level, const char *title,
5368                   struct trace_ctx *trace)
5369 {
5370     ds_put_char_multiple(result, '\t', level);
5371     ds_put_format(result, "%s: ", title);
5372     if (flow_equal(&trace->xin.flow, &trace->flow)) {
5373         ds_put_cstr(result, "unchanged");
5374     } else {
5375         flow_format(result, &trace->xin.flow);
5376         trace->flow = trace->xin.flow;
5377     }
5378     ds_put_char(result, '\n');
5379 }
5380
5381 static void
5382 trace_format_regs(struct ds *result, int level, const char *title,
5383                   struct trace_ctx *trace)
5384 {
5385     size_t i;
5386
5387     ds_put_char_multiple(result, '\t', level);
5388     ds_put_format(result, "%s:", title);
5389     for (i = 0; i < FLOW_N_REGS; i++) {
5390         ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5391     }
5392     ds_put_char(result, '\n');
5393 }
5394
5395 static void
5396 trace_format_odp(struct ds *result, int level, const char *title,
5397                  struct trace_ctx *trace)
5398 {
5399     struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5400
5401     ds_put_char_multiple(result, '\t', level);
5402     ds_put_format(result, "%s: ", title);
5403     format_odp_actions(result, odp_actions->data, odp_actions->size);
5404     ds_put_char(result, '\n');
5405 }
5406
5407 static void
5408 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5409 {
5410     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5411     struct ds *result = trace->result;
5412
5413     ds_put_char(result, '\n');
5414     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5415     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5416     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
5417     trace_format_rule(result, recurse + 1, rule);
5418 }
5419
5420 static void
5421 trace_report(struct xlate_in *xin, const char *s, int recurse)
5422 {
5423     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5424     struct ds *result = trace->result;
5425
5426     ds_put_char_multiple(result, '\t', recurse);
5427     ds_put_cstr(result, s);
5428     ds_put_char(result, '\n');
5429 }
5430
5431 static void
5432 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5433                       void *aux OVS_UNUSED)
5434 {
5435     const struct dpif_backer *backer;
5436     struct ofproto_dpif *ofproto;
5437     struct ofpbuf odp_key, odp_mask;
5438     struct ofpbuf *packet;
5439     struct ds result;
5440     struct flow flow;
5441     char *s;
5442
5443     packet = NULL;
5444     backer = NULL;
5445     ds_init(&result);
5446     ofpbuf_init(&odp_key, 0);
5447     ofpbuf_init(&odp_mask, 0);
5448
5449     /* Handle "-generate" or a hex string as the last argument. */
5450     if (!strcmp(argv[argc - 1], "-generate")) {
5451         packet = ofpbuf_new(0);
5452         argc--;
5453     } else {
5454         const char *error = eth_from_hex(argv[argc - 1], &packet);
5455         if (!error) {
5456             argc--;
5457         } else if (argc == 4) {
5458             /* The 3-argument form must end in "-generate' or a hex string. */
5459             unixctl_command_reply_error(conn, error);
5460             goto exit;
5461         }
5462     }
5463
5464     /* Parse the flow and determine whether a datapath or
5465      * bridge is specified. If function odp_flow_key_from_string()
5466      * returns 0, the flow is a odp_flow. If function
5467      * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5468     if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5469         /* If the odp_flow is the second argument,
5470          * the datapath name is the first argument. */
5471         if (argc == 3) {
5472             const char *dp_type;
5473             if (!strncmp(argv[1], "ovs-", 4)) {
5474                 dp_type = argv[1] + 4;
5475             } else {
5476                 dp_type = argv[1];
5477             }
5478             backer = shash_find_data(&all_dpif_backers, dp_type);
5479             if (!backer) {
5480                 unixctl_command_reply_error(conn, "Cannot find datapath "
5481                                "of this name");
5482                 goto exit;
5483             }
5484         } else {
5485             /* No datapath name specified, so there should be only one
5486              * datapath. */
5487             struct shash_node *node;
5488             if (shash_count(&all_dpif_backers) != 1) {
5489                 unixctl_command_reply_error(conn, "Must specify datapath "
5490                          "name, there is more than one type of datapath");
5491                 goto exit;
5492             }
5493             node = shash_first(&all_dpif_backers);
5494             backer = node->data;
5495         }
5496
5497         if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5498                           NULL, &ofproto, NULL)) {
5499             unixctl_command_reply_error(conn, "Invalid datapath flow");
5500             goto exit;
5501         }
5502         ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5503     } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5504         if (argc != 3) {
5505             unixctl_command_reply_error(conn, "Must specify bridge name");
5506             goto exit;
5507         }
5508
5509         ofproto = ofproto_dpif_lookup(argv[1]);
5510         if (!ofproto) {
5511             unixctl_command_reply_error(conn, "Unknown bridge name");
5512             goto exit;
5513         }
5514     } else {
5515         unixctl_command_reply_error(conn, "Bad flow syntax");
5516         goto exit;
5517     }
5518
5519     /* Generate a packet, if requested. */
5520     if (packet) {
5521         if (!packet->size) {
5522             flow_compose(packet, &flow);
5523         } else {
5524             union flow_in_port in_port_;
5525
5526             in_port_ = flow.in_port;
5527             ds_put_cstr(&result, "Packet: ");
5528             s = ofp_packet_to_string(packet->data, packet->size);
5529             ds_put_cstr(&result, s);
5530             free(s);
5531
5532             /* Use the metadata from the flow and the packet argument
5533              * to reconstruct the flow. */
5534             flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5535                          &in_port_, &flow);
5536         }
5537     }
5538
5539     ofproto_trace(ofproto, &flow, packet, &result);
5540     unixctl_command_reply(conn, ds_cstr(&result));
5541
5542 exit:
5543     ds_destroy(&result);
5544     ofpbuf_delete(packet);
5545     ofpbuf_uninit(&odp_key);
5546     ofpbuf_uninit(&odp_mask);
5547 }
5548
5549 static void
5550 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5551               const struct ofpbuf *packet, struct ds *ds)
5552 {
5553     struct rule_dpif *rule;
5554     struct flow_wildcards wc;
5555
5556     ds_put_cstr(ds, "Flow: ");
5557     flow_format(ds, flow);
5558     ds_put_char(ds, '\n');
5559
5560     flow_wildcards_init_catchall(&wc);
5561     rule_dpif_lookup(ofproto, flow, &wc, &rule);
5562
5563     trace_format_rule(ds, 0, rule);
5564     if (rule == ofproto->miss_rule) {
5565         ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5566     } else if (rule == ofproto->no_packet_in_rule) {
5567         ds_put_cstr(ds, "\nNo match, packets dropped because "
5568                     "OFPPC_NO_PACKET_IN is set on in_port.\n");
5569     } else if (rule == ofproto->drop_frags_rule) {
5570         ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5571                     "and the fragment handling mode is \"drop\".\n");
5572     }
5573
5574     if (rule) {
5575         uint64_t odp_actions_stub[1024 / 8];
5576         struct ofpbuf odp_actions;
5577         struct trace_ctx trace;
5578         struct match match;
5579         uint8_t tcp_flags;
5580
5581         tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5582         trace.result = ds;
5583         trace.flow = *flow;
5584         ofpbuf_use_stub(&odp_actions,
5585                         odp_actions_stub, sizeof odp_actions_stub);
5586         xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5587         trace.xin.resubmit_hook = trace_resubmit;
5588         trace.xin.report_hook = trace_report;
5589
5590         xlate_actions(&trace.xin, &trace.xout);
5591         flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5592
5593         ds_put_char(ds, '\n');
5594         trace_format_flow(ds, 0, "Final flow", &trace);
5595
5596         match_init(&match, flow, &trace.xout.wc);
5597         ds_put_cstr(ds, "Relevant fields: ");
5598         match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5599         ds_put_char(ds, '\n');
5600
5601         ds_put_cstr(ds, "Datapath actions: ");
5602         format_odp_actions(ds, trace.xout.odp_actions.data,
5603                            trace.xout.odp_actions.size);
5604
5605         if (trace.xout.slow) {
5606             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5607                         "slow path because it:");
5608             switch (trace.xout.slow) {
5609             case SLOW_CFM:
5610                 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5611                 break;
5612             case SLOW_LACP:
5613                 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5614                 break;
5615             case SLOW_STP:
5616                 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5617                 break;
5618             case SLOW_BFD:
5619                 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5620                 break;
5621             case SLOW_CONTROLLER:
5622                 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5623                             "to the OpenFlow controller.");
5624                 break;
5625             case __SLOW_MAX:
5626                 NOT_REACHED();
5627             }
5628         }
5629
5630         xlate_out_uninit(&trace.xout);
5631     }
5632
5633     rule_dpif_release(rule);
5634 }
5635
5636 static void
5637 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5638                   const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5639 {
5640     clogged = true;
5641     unixctl_command_reply(conn, NULL);
5642 }
5643
5644 static void
5645 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5646                     const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5647 {
5648     clogged = false;
5649     unixctl_command_reply(conn, NULL);
5650 }
5651
5652 /* Runs a self-check of flow translations in 'ofproto'.  Appends a message to
5653  * 'reply' describing the results. */
5654 static void
5655 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5656 {
5657     struct cls_cursor cursor;
5658     struct facet *facet;
5659     int errors;
5660
5661     errors = 0;
5662     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5663     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5664     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5665         if (!facet_check_consistency(facet)) {
5666             errors++;
5667         }
5668     }
5669     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5670     if (errors) {
5671         ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5672     }
5673
5674     if (errors) {
5675         ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5676                       ofproto->up.name, errors);
5677     } else {
5678         ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5679     }
5680 }
5681
5682 static void
5683 ofproto_dpif_self_check(struct unixctl_conn *conn,
5684                         int argc, const char *argv[], void *aux OVS_UNUSED)
5685 {
5686     struct ds reply = DS_EMPTY_INITIALIZER;
5687     struct ofproto_dpif *ofproto;
5688
5689     if (argc > 1) {
5690         ofproto = ofproto_dpif_lookup(argv[1]);
5691         if (!ofproto) {
5692             unixctl_command_reply_error(conn, "Unknown ofproto (use "
5693                                         "ofproto/list for help)");
5694             return;
5695         }
5696         ofproto_dpif_self_check__(ofproto, &reply);
5697     } else {
5698         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5699             ofproto_dpif_self_check__(ofproto, &reply);
5700         }
5701     }
5702
5703     unixctl_command_reply(conn, ds_cstr(&reply));
5704     ds_destroy(&reply);
5705 }
5706
5707 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
5708  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
5709  * to destroy 'ofproto_shash' and free the returned value. */
5710 static const struct shash_node **
5711 get_ofprotos(struct shash *ofproto_shash)
5712 {
5713     const struct ofproto_dpif *ofproto;
5714
5715     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5716         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5717         shash_add_nocopy(ofproto_shash, name, ofproto);
5718     }
5719
5720     return shash_sort(ofproto_shash);
5721 }
5722
5723 static void
5724 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5725                               const char *argv[] OVS_UNUSED,
5726                               void *aux OVS_UNUSED)
5727 {
5728     struct ds ds = DS_EMPTY_INITIALIZER;
5729     struct shash ofproto_shash;
5730     const struct shash_node **sorted_ofprotos;
5731     int i;
5732
5733     shash_init(&ofproto_shash);
5734     sorted_ofprotos = get_ofprotos(&ofproto_shash);
5735     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5736         const struct shash_node *node = sorted_ofprotos[i];
5737         ds_put_format(&ds, "%s\n", node->name);
5738     }
5739
5740     shash_destroy(&ofproto_shash);
5741     free(sorted_ofprotos);
5742
5743     unixctl_command_reply(conn, ds_cstr(&ds));
5744     ds_destroy(&ds);
5745 }
5746
5747 static void
5748 show_dp_rates(struct ds *ds, const char *heading,
5749               const struct avg_subfacet_rates *rates)
5750 {
5751     ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5752                   heading, rates->add_rate, rates->del_rate);
5753 }
5754
5755 static void
5756 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5757 {
5758     const struct shash_node **ofprotos;
5759     struct ofproto_dpif *ofproto;
5760     struct shash ofproto_shash;
5761     uint64_t n_hit, n_missed;
5762     long long int minutes;
5763     size_t i;
5764
5765     n_hit = n_missed = 0;
5766     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5767         if (ofproto->backer == backer) {
5768             n_missed += ofproto->n_missed;
5769             n_hit += ofproto->n_hit;
5770         }
5771     }
5772
5773     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5774                   dpif_name(backer->dpif), n_hit, n_missed);
5775     ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5776                   " life span: %lldms\n", hmap_count(&backer->subfacets),
5777                   backer->avg_n_subfacet, backer->max_n_subfacet,
5778                   backer->avg_subfacet_life);
5779
5780     minutes = (time_msec() - backer->created) / (1000 * 60);
5781     if (minutes >= 60) {
5782         show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5783     }
5784     if (minutes >= 60 * 24) {
5785         show_dp_rates(ds, "\tdaily avg:",  &backer->daily);
5786     }
5787     show_dp_rates(ds, "\toverall avg:",  &backer->lifetime);
5788
5789     shash_init(&ofproto_shash);
5790     ofprotos = get_ofprotos(&ofproto_shash);
5791     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5792         struct ofproto_dpif *ofproto = ofprotos[i]->data;
5793         const struct shash_node **ports;
5794         size_t j;
5795
5796         if (ofproto->backer != backer) {
5797             continue;
5798         }
5799
5800         ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5801                       ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5802
5803         ports = shash_sort(&ofproto->up.port_by_name);
5804         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5805             const struct shash_node *node = ports[j];
5806             struct ofport *ofport = node->data;
5807             struct smap config;
5808             odp_port_t odp_port;
5809
5810             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5811                           ofport->ofp_port);
5812
5813             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5814             if (odp_port != ODPP_NONE) {
5815                 ds_put_format(ds, "%"PRIu32":", odp_port);
5816             } else {
5817                 ds_put_cstr(ds, "none:");
5818             }
5819
5820             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5821
5822             smap_init(&config);
5823             if (!netdev_get_config(ofport->netdev, &config)) {
5824                 const struct smap_node **nodes;
5825                 size_t i;
5826
5827                 nodes = smap_sort(&config);
5828                 for (i = 0; i < smap_count(&config); i++) {
5829                     const struct smap_node *node = nodes[i];
5830                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5831                                   node->key, node->value);
5832                 }
5833                 free(nodes);
5834             }
5835             smap_destroy(&config);
5836
5837             ds_put_char(ds, ')');
5838             ds_put_char(ds, '\n');
5839         }
5840         free(ports);
5841     }
5842     shash_destroy(&ofproto_shash);
5843     free(ofprotos);
5844 }
5845
5846 static void
5847 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5848                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5849 {
5850     struct ds ds = DS_EMPTY_INITIALIZER;
5851     const struct shash_node **backers;
5852     int i;
5853
5854     backers = shash_sort(&all_dpif_backers);
5855     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5856         dpif_show_backer(backers[i]->data, &ds);
5857     }
5858     free(backers);
5859
5860     unixctl_command_reply(conn, ds_cstr(&ds));
5861     ds_destroy(&ds);
5862 }
5863
5864 /* Dump the megaflow (facet) cache.  This is useful to check the
5865  * correctness of flow wildcarding, since the same mechanism is used for
5866  * both xlate caching and kernel wildcarding.
5867  *
5868  * It's important to note that in the output the flow description uses
5869  * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5870  *
5871  * This command is only needed for advanced debugging, so it's not
5872  * documented in the man page. */
5873 static void
5874 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5875                                     int argc OVS_UNUSED, const char *argv[],
5876                                     void *aux OVS_UNUSED)
5877 {
5878     struct ds ds = DS_EMPTY_INITIALIZER;
5879     const struct ofproto_dpif *ofproto;
5880     long long int now = time_msec();
5881     struct cls_cursor cursor;
5882     struct facet *facet;
5883
5884     ofproto = ofproto_dpif_lookup(argv[1]);
5885     if (!ofproto) {
5886         unixctl_command_reply_error(conn, "no such bridge");
5887         return;
5888     }
5889
5890     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5891     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5892     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5893         cls_rule_format(&facet->cr, &ds);
5894         ds_put_cstr(&ds, ", ");
5895         ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5896         ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5897         ds_put_cstr(&ds, "Datapath actions: ");
5898         if (facet->xout.slow) {
5899             uint64_t slow_path_stub[128 / 8];
5900             const struct nlattr *actions;
5901             size_t actions_len;
5902
5903             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5904                               slow_path_stub, sizeof slow_path_stub,
5905                               &actions, &actions_len);
5906             format_odp_actions(&ds, actions, actions_len);
5907         } else {
5908             format_odp_actions(&ds, facet->xout.odp_actions.data,
5909                                facet->xout.odp_actions.size);
5910         }
5911         ds_put_cstr(&ds, "\n");
5912     }
5913     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5914
5915     ds_chomp(&ds, '\n');
5916     unixctl_command_reply(conn, ds_cstr(&ds));
5917     ds_destroy(&ds);
5918 }
5919
5920 /* Disable using the megaflows.
5921  *
5922  * This command is only needed for advanced debugging, so it's not
5923  * documented in the man page. */
5924 static void
5925 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5926                                        int argc OVS_UNUSED,
5927                                        const char *argv[] OVS_UNUSED,
5928                                        void *aux OVS_UNUSED)
5929 {
5930     struct ofproto_dpif *ofproto;
5931
5932     enable_megaflows = false;
5933
5934     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5935         flush(&ofproto->up);
5936     }
5937
5938     unixctl_command_reply(conn, "megaflows disabled");
5939 }
5940
5941 /* Re-enable using megaflows.
5942  *
5943  * This command is only needed for advanced debugging, so it's not
5944  * documented in the man page. */
5945 static void
5946 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5947                                       int argc OVS_UNUSED,
5948                                       const char *argv[] OVS_UNUSED,
5949                                       void *aux OVS_UNUSED)
5950 {
5951     struct ofproto_dpif *ofproto;
5952
5953     enable_megaflows = true;
5954
5955     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5956         flush(&ofproto->up);
5957     }
5958
5959     unixctl_command_reply(conn, "megaflows enabled");
5960 }
5961
5962 static void
5963 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5964                                 int argc OVS_UNUSED, const char *argv[],
5965                                 void *aux OVS_UNUSED)
5966 {
5967     struct ds ds = DS_EMPTY_INITIALIZER;
5968     const struct ofproto_dpif *ofproto;
5969     struct subfacet *subfacet;
5970
5971     ofproto = ofproto_dpif_lookup(argv[1]);
5972     if (!ofproto) {
5973         unixctl_command_reply_error(conn, "no such bridge");
5974         return;
5975     }
5976
5977     update_stats(ofproto->backer);
5978
5979     HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5980         struct facet *facet = subfacet->facet;
5981         struct odputil_keybuf maskbuf;
5982         struct ofpbuf mask;
5983
5984         if (facet->ofproto != ofproto) {
5985             continue;
5986         }
5987
5988         ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5989         if (enable_megaflows) {
5990             odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5991                                    &facet->flow, UINT32_MAX);
5992         }
5993
5994         odp_flow_format(subfacet->key, subfacet->key_len,
5995                         mask.data, mask.size, &ds, false);
5996
5997         ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5998                       subfacet->dp_packet_count, subfacet->dp_byte_count);
5999         if (subfacet->used) {
6000             ds_put_format(&ds, "%.3fs",
6001                           (time_msec() - subfacet->used) / 1000.0);
6002         } else {
6003             ds_put_format(&ds, "never");
6004         }
6005         if (subfacet->facet->tcp_flags) {
6006             ds_put_cstr(&ds, ", flags:");
6007             packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6008         }
6009
6010         ds_put_cstr(&ds, ", actions:");
6011         if (facet->xout.slow) {
6012             uint64_t slow_path_stub[128 / 8];
6013             const struct nlattr *actions;
6014             size_t actions_len;
6015
6016             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6017                               slow_path_stub, sizeof slow_path_stub,
6018                               &actions, &actions_len);
6019             format_odp_actions(&ds, actions, actions_len);
6020         } else {
6021             format_odp_actions(&ds, facet->xout.odp_actions.data,
6022                                facet->xout.odp_actions.size);
6023         }
6024         ds_put_char(&ds, '\n');
6025     }
6026
6027     unixctl_command_reply(conn, ds_cstr(&ds));
6028     ds_destroy(&ds);
6029 }
6030
6031 static void
6032 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6033                                int argc OVS_UNUSED, const char *argv[],
6034                                void *aux OVS_UNUSED)
6035 {
6036     struct ds ds = DS_EMPTY_INITIALIZER;
6037     struct ofproto_dpif *ofproto;
6038
6039     ofproto = ofproto_dpif_lookup(argv[1]);
6040     if (!ofproto) {
6041         unixctl_command_reply_error(conn, "no such bridge");
6042         return;
6043     }
6044
6045     flush(&ofproto->up);
6046
6047     unixctl_command_reply(conn, ds_cstr(&ds));
6048     ds_destroy(&ds);
6049 }
6050
6051 static void
6052 ofproto_dpif_unixctl_init(void)
6053 {
6054     static bool registered;
6055     if (registered) {
6056         return;
6057     }
6058     registered = true;
6059
6060     unixctl_command_register(
6061         "ofproto/trace",
6062         "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
6063         1, 3, ofproto_unixctl_trace, NULL);
6064     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6065                              ofproto_unixctl_fdb_flush, NULL);
6066     unixctl_command_register("fdb/show", "bridge", 1, 1,
6067                              ofproto_unixctl_fdb_show, NULL);
6068     unixctl_command_register("ofproto/clog", "", 0, 0,
6069                              ofproto_dpif_clog, NULL);
6070     unixctl_command_register("ofproto/unclog", "", 0, 0,
6071                              ofproto_dpif_unclog, NULL);
6072     unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6073                              ofproto_dpif_self_check, NULL);
6074     unixctl_command_register("dpif/dump-dps", "", 0, 0,
6075                              ofproto_unixctl_dpif_dump_dps, NULL);
6076     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6077                              NULL);
6078     unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6079                              ofproto_unixctl_dpif_dump_flows, NULL);
6080     unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6081                              ofproto_unixctl_dpif_del_flows, NULL);
6082     unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6083                              ofproto_unixctl_dpif_dump_megaflows, NULL);
6084     unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6085                              ofproto_unixctl_dpif_disable_megaflows, NULL);
6086     unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6087                              ofproto_unixctl_dpif_enable_megaflows, NULL);
6088 }
6089 \f
6090 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6091  *
6092  * This is deprecated.  It is only for compatibility with broken device drivers
6093  * in old versions of Linux that do not properly support VLANs when VLAN
6094  * devices are not used.  When broken device drivers are no longer in
6095  * widespread use, we will delete these interfaces. */
6096
6097 static int
6098 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6099 {
6100     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6101     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6102
6103     if (realdev_ofp_port == ofport->realdev_ofp_port
6104         && vid == ofport->vlandev_vid) {
6105         return 0;
6106     }
6107
6108     ofproto->backer->need_revalidate = REV_RECONFIGURE;
6109
6110     if (ofport->realdev_ofp_port) {
6111         vsp_remove(ofport);
6112     }
6113     if (realdev_ofp_port && ofport->bundle) {
6114         /* vlandevs are enslaved to their realdevs, so they are not allowed to
6115          * themselves be part of a bundle. */
6116         bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6117     }
6118
6119     ofport->realdev_ofp_port = realdev_ofp_port;
6120     ofport->vlandev_vid = vid;
6121
6122     if (realdev_ofp_port) {
6123         vsp_add(ofport, realdev_ofp_port, vid);
6124     }
6125
6126     return 0;
6127 }
6128
6129 static uint32_t
6130 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6131 {
6132     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6133 }
6134
6135 bool
6136 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6137     OVS_EXCLUDED(ofproto->vsp_mutex)
6138 {
6139     bool ret;
6140
6141     ovs_mutex_lock(&ofproto->vsp_mutex);
6142     ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6143     ovs_mutex_unlock(&ofproto->vsp_mutex);
6144     return ret;
6145 }
6146
6147 static ofp_port_t
6148 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6149                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6150     OVS_REQUIRES(ofproto->vsp_mutex)
6151 {
6152     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6153         int vid = vlan_tci_to_vid(vlan_tci);
6154         const struct vlan_splinter *vsp;
6155
6156         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6157                                  hash_realdev_vid(realdev_ofp_port, vid),
6158                                  &ofproto->realdev_vid_map) {
6159             if (vsp->realdev_ofp_port == realdev_ofp_port
6160                 && vsp->vid == vid) {
6161                 return vsp->vlandev_ofp_port;
6162             }
6163         }
6164     }
6165     return realdev_ofp_port;
6166 }
6167
6168 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6169  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6170  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
6171  * 'vlan_tci' 9, it would return the port number of eth0.9.
6172  *
6173  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6174  * function just returns its 'realdev_ofp_port' argument. */
6175 ofp_port_t
6176 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6177                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6178     OVS_EXCLUDED(ofproto->vsp_mutex)
6179 {
6180     ofp_port_t ret;
6181
6182     ovs_mutex_lock(&ofproto->vsp_mutex);
6183     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6184     ovs_mutex_unlock(&ofproto->vsp_mutex);
6185     return ret;
6186 }
6187
6188 static struct vlan_splinter *
6189 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6190 {
6191     struct vlan_splinter *vsp;
6192
6193     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6194                              hash_ofp_port(vlandev_ofp_port),
6195                              &ofproto->vlandev_map) {
6196         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6197             return vsp;
6198         }
6199     }
6200
6201     return NULL;
6202 }
6203
6204 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6205  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6206  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
6207  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6208  * eth0 and store 9 in '*vid'.
6209  *
6210  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6211  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
6212  * always does.*/
6213 static ofp_port_t
6214 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6215                        ofp_port_t vlandev_ofp_port, int *vid)
6216     OVS_REQUIRES(ofproto->vsp_mutex)
6217 {
6218     if (!hmap_is_empty(&ofproto->vlandev_map)) {
6219         const struct vlan_splinter *vsp;
6220
6221         vsp = vlandev_find(ofproto, vlandev_ofp_port);
6222         if (vsp) {
6223             if (vid) {
6224                 *vid = vsp->vid;
6225             }
6226             return vsp->realdev_ofp_port;
6227         }
6228     }
6229     return 0;
6230 }
6231
6232 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6233  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
6234  * 'flow->in_port' to the "real" device backing the VLAN device, sets
6235  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Otherwise (which is
6236  * always the case unless VLAN splinters are enabled), returns false without
6237  * making any changes. */
6238 bool
6239 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6240     OVS_EXCLUDED(ofproto->vsp_mutex)
6241 {
6242     ofp_port_t realdev;
6243     int vid;
6244
6245     ovs_mutex_lock(&ofproto->vsp_mutex);
6246     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6247     ovs_mutex_unlock(&ofproto->vsp_mutex);
6248     if (!realdev) {
6249         return false;
6250     }
6251
6252     /* Cause the flow to be processed as if it came in on the real device with
6253      * the VLAN device's VLAN ID. */
6254     flow->in_port.ofp_port = realdev;
6255     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6256     return true;
6257 }
6258
6259 static void
6260 vsp_remove(struct ofport_dpif *port)
6261 {
6262     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6263     struct vlan_splinter *vsp;
6264
6265     ovs_mutex_lock(&ofproto->vsp_mutex);
6266     vsp = vlandev_find(ofproto, port->up.ofp_port);
6267     if (vsp) {
6268         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6269         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6270         free(vsp);
6271
6272         port->realdev_ofp_port = 0;
6273     } else {
6274         VLOG_ERR("missing vlan device record");
6275     }
6276     ovs_mutex_unlock(&ofproto->vsp_mutex);
6277 }
6278
6279 static void
6280 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6281 {
6282     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6283
6284     ovs_mutex_lock(&ofproto->vsp_mutex);
6285     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6286         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6287             == realdev_ofp_port)) {
6288         struct vlan_splinter *vsp;
6289
6290         vsp = xmalloc(sizeof *vsp);
6291         vsp->realdev_ofp_port = realdev_ofp_port;
6292         vsp->vlandev_ofp_port = port->up.ofp_port;
6293         vsp->vid = vid;
6294
6295         port->realdev_ofp_port = realdev_ofp_port;
6296
6297         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6298                     hash_ofp_port(port->up.ofp_port));
6299         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6300                     hash_realdev_vid(realdev_ofp_port, vid));
6301     } else {
6302         VLOG_ERR("duplicate vlan device record");
6303     }
6304     ovs_mutex_unlock(&ofproto->vsp_mutex);
6305 }
6306
6307 static odp_port_t
6308 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6309 {
6310     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6311     return ofport ? ofport->odp_port : ODPP_NONE;
6312 }
6313
6314 struct ofport_dpif *
6315 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6316 {
6317     struct ofport_dpif *port;
6318
6319     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6320     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6321                              &backer->odp_to_ofport_map) {
6322         if (port->odp_port == odp_port) {
6323             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6324             return port;
6325         }
6326     }
6327
6328     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6329     return NULL;
6330 }
6331
6332 static ofp_port_t
6333 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6334 {
6335     struct ofport_dpif *port;
6336
6337     port = odp_port_to_ofport(ofproto->backer, odp_port);
6338     if (port && &ofproto->up == port->up.ofproto) {
6339         return port->up.ofp_port;
6340     } else {
6341         return OFPP_NONE;
6342     }
6343 }
6344
6345 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6346  * most heavily weighted element.  'base' designates the rate of decay: after
6347  * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6348  * (about .37). */
6349 static void
6350 exp_mavg(double *avg, int base, double new)
6351 {
6352     *avg = (*avg * (base - 1) + new) / base;
6353 }
6354
6355 static void
6356 update_moving_averages(struct dpif_backer *backer)
6357 {
6358     const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6359     long long int minutes = (time_msec() - backer->created) / min_ms;
6360
6361     if (minutes > 0) {
6362         backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6363             / minutes;
6364         backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6365             / minutes;
6366     } else {
6367         backer->lifetime.add_rate = 0.0;
6368         backer->lifetime.del_rate = 0.0;
6369     }
6370
6371     /* Update hourly averages on the minute boundaries. */
6372     if (time_msec() - backer->last_minute >= min_ms) {
6373         exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6374         exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6375
6376         /* Update daily averages on the hour boundaries. */
6377         if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6378             exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6379             exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6380         }
6381
6382         backer->total_subfacet_add_count += backer->subfacet_add_count;
6383         backer->total_subfacet_del_count += backer->subfacet_del_count;
6384         backer->subfacet_add_count = 0;
6385         backer->subfacet_del_count = 0;
6386         backer->last_minute += min_ms;
6387     }
6388 }
6389
6390 const struct ofproto_class ofproto_dpif_class = {
6391     init,
6392     enumerate_types,
6393     enumerate_names,
6394     del,
6395     port_open_type,
6396     type_run,
6397     type_run_fast,
6398     type_wait,
6399     alloc,
6400     construct,
6401     destruct,
6402     dealloc,
6403     run,
6404     run_fast,
6405     wait,
6406     get_memory_usage,
6407     flush,
6408     get_features,
6409     get_tables,
6410     port_alloc,
6411     port_construct,
6412     port_destruct,
6413     port_dealloc,
6414     port_modified,
6415     port_reconfigured,
6416     port_query_by_name,
6417     port_add,
6418     port_del,
6419     port_get_stats,
6420     port_dump_start,
6421     port_dump_next,
6422     port_dump_done,
6423     port_poll,
6424     port_poll_wait,
6425     port_is_lacp_current,
6426     NULL,                       /* rule_choose_table */
6427     rule_alloc,
6428     rule_construct,
6429     rule_insert,
6430     rule_delete,
6431     rule_destruct,
6432     rule_dealloc,
6433     rule_get_stats,
6434     rule_execute,
6435     rule_modify_actions,
6436     set_frag_handling,
6437     packet_out,
6438     set_netflow,
6439     get_netflow_ids,
6440     set_sflow,
6441     set_ipfix,
6442     set_cfm,
6443     get_cfm_status,
6444     set_bfd,
6445     get_bfd_status,
6446     set_stp,
6447     get_stp_status,
6448     set_stp_port,
6449     get_stp_port_status,
6450     set_queues,
6451     bundle_set,
6452     bundle_remove,
6453     mirror_set__,
6454     mirror_get_stats__,
6455     set_flood_vlans,
6456     is_mirror_output_bundle,
6457     forward_bpdu_changed,
6458     set_mac_table_config,
6459     set_realdev,
6460     NULL,                       /* meter_get_features */
6461     NULL,                       /* meter_set */
6462     NULL,                       /* meter_get */
6463     NULL,                       /* meter_del */
6464 };