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