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     odp_port_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_, u16_to_ofp(MIN(odp_to_u32(max_ports),
1261                                                     ofp_to_u16(OFPP_MAX))));
1262
1263     ofproto->netflow = NULL;
1264     ofproto->sflow = NULL;
1265     ofproto->ipfix = NULL;
1266     ofproto->stp = NULL;
1267     hmap_init(&ofproto->bundles);
1268     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1269     ofproto->mbridge = mbridge_create();
1270     ofproto->has_bonded_bundles = false;
1271     ovs_mutex_init(&ofproto->vsp_mutex);
1272
1273     classifier_init(&ofproto->facets);
1274     ofproto->consistency_rl = LLONG_MIN;
1275
1276     list_init(&ofproto->completions);
1277
1278     ovs_mutex_init(&ofproto->flow_mod_mutex);
1279     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1280     list_init(&ofproto->flow_mods);
1281     ofproto->n_flow_mods = 0;
1282     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1283
1284     ovs_mutex_init(&ofproto->pin_mutex);
1285     ovs_mutex_lock(&ofproto->pin_mutex);
1286     list_init(&ofproto->pins);
1287     ofproto->n_pins = 0;
1288     ovs_mutex_unlock(&ofproto->pin_mutex);
1289
1290     ofproto_dpif_unixctl_init();
1291
1292     hmap_init(&ofproto->vlandev_map);
1293     hmap_init(&ofproto->realdev_vid_map);
1294
1295     sset_init(&ofproto->ports);
1296     sset_init(&ofproto->ghost_ports);
1297     sset_init(&ofproto->port_poll_set);
1298     ofproto->port_poll_errno = 0;
1299
1300     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1301         struct iface_hint *iface_hint = node->data;
1302
1303         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1304             /* Check if the datapath already has this port. */
1305             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1306                 sset_add(&ofproto->ports, node->name);
1307             }
1308
1309             free(iface_hint->br_name);
1310             free(iface_hint->br_type);
1311             free(iface_hint);
1312             shash_delete(&init_ofp_ports, node);
1313         }
1314     }
1315
1316     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1317                 hash_string(ofproto->up.name, 0));
1318     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1319
1320     ofproto_init_tables(ofproto_, N_TABLES);
1321     error = add_internal_flows(ofproto);
1322     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1323
1324     ofproto->n_hit = 0;
1325     ofproto->n_missed = 0;
1326
1327     return error;
1328 }
1329
1330 static int
1331 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1332                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1333 {
1334     struct ofputil_flow_mod fm;
1335     int error;
1336
1337     match_init_catchall(&fm.match);
1338     fm.priority = 0;
1339     match_set_reg(&fm.match, 0, id);
1340     fm.new_cookie = htonll(0);
1341     fm.cookie = htonll(0);
1342     fm.cookie_mask = htonll(0);
1343     fm.modify_cookie = false;
1344     fm.table_id = TBL_INTERNAL;
1345     fm.command = OFPFC_ADD;
1346     fm.idle_timeout = 0;
1347     fm.hard_timeout = 0;
1348     fm.buffer_id = 0;
1349     fm.out_port = 0;
1350     fm.flags = 0;
1351     fm.ofpacts = ofpacts->data;
1352     fm.ofpacts_len = ofpacts->size;
1353
1354     error = ofproto_flow_mod(&ofproto->up, &fm);
1355     if (error) {
1356         VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1357                     id, ofperr_to_string(error));
1358         return error;
1359     }
1360
1361     if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1362                                   rulep)) {
1363         ovs_rwlock_unlock(&(*rulep)->up.evict);
1364     } else {
1365         NOT_REACHED();
1366     }
1367
1368     return 0;
1369 }
1370
1371 static int
1372 add_internal_flows(struct ofproto_dpif *ofproto)
1373 {
1374     struct ofpact_controller *controller;
1375     uint64_t ofpacts_stub[128 / 8];
1376     struct ofpbuf ofpacts;
1377     int error;
1378     int id;
1379
1380     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1381     id = 1;
1382
1383     controller = ofpact_put_CONTROLLER(&ofpacts);
1384     controller->max_len = UINT16_MAX;
1385     controller->controller_id = 0;
1386     controller->reason = OFPR_NO_MATCH;
1387     ofpact_pad(&ofpacts);
1388
1389     error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1390     if (error) {
1391         return error;
1392     }
1393
1394     ofpbuf_clear(&ofpacts);
1395     error = add_internal_flow(ofproto, id++, &ofpacts,
1396                               &ofproto->no_packet_in_rule);
1397     if (error) {
1398         return error;
1399     }
1400
1401     error = add_internal_flow(ofproto, id++, &ofpacts,
1402                               &ofproto->drop_frags_rule);
1403     return error;
1404 }
1405
1406 static void
1407 complete_operations(struct ofproto_dpif *ofproto)
1408 {
1409     struct dpif_completion *c, *next;
1410
1411     LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1412         ofoperation_complete(c->op, 0);
1413         list_remove(&c->list_node);
1414         free(c);
1415     }
1416 }
1417
1418 static void
1419 destruct(struct ofproto *ofproto_)
1420 {
1421     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1422     struct rule_dpif *rule, *next_rule;
1423     struct ofputil_packet_in *pin, *next_pin;
1424     struct ofputil_flow_mod *fm, *next_fm;
1425     struct oftable *table;
1426
1427     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1428     ovs_rwlock_wrlock(&xlate_rwlock);
1429     xlate_remove_ofproto(ofproto);
1430     ovs_rwlock_unlock(&xlate_rwlock);
1431
1432     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1433     complete_operations(ofproto);
1434
1435     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1436         struct cls_cursor cursor;
1437
1438         ovs_rwlock_wrlock(&table->cls.rwlock);
1439         cls_cursor_init(&cursor, &table->cls, NULL);
1440         CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1441             ofproto_rule_destroy(&ofproto->up, &table->cls, &rule->up);
1442         }
1443         ovs_rwlock_unlock(&table->cls.rwlock);
1444     }
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     if (ofproto->n_flow_mods) {
1504         flow_mods = ofproto->flow_mods;
1505         list_moved(&flow_mods);
1506         list_init(&ofproto->flow_mods);
1507         ofproto->n_flow_mods = 0;
1508     } else {
1509         list_init(&flow_mods);
1510     }
1511     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1512
1513     LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &flow_mods) {
1514         int error = ofproto_flow_mod(&ofproto->up, fm);
1515         if (error && !VLOG_DROP_WARN(&rl)) {
1516             VLOG_WARN("learning action failed to modify flow table (%s)",
1517                       ofperr_get_name(error));
1518         }
1519
1520         list_remove(&fm->list_node);
1521         free(fm->ofpacts);
1522         free(fm);
1523     }
1524
1525     ovs_mutex_lock(&ofproto->pin_mutex);
1526     if (ofproto->n_pins) {
1527         pins = ofproto->pins;
1528         list_moved(&pins);
1529         list_init(&ofproto->pins);
1530         ofproto->n_pins = 0;
1531     } else {
1532         list_init(&pins);
1533     }
1534     ovs_mutex_unlock(&ofproto->pin_mutex);
1535
1536     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1537         connmgr_send_packet_in(ofproto->up.connmgr, pin);
1538         list_remove(&pin->list_node);
1539         free(CONST_CAST(void *, pin->packet));
1540         free(pin);
1541     }
1542
1543     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1544         port_run_fast(ofport);
1545     }
1546
1547     return 0;
1548 }
1549
1550 static int
1551 run(struct ofproto *ofproto_)
1552 {
1553     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1554     struct ofport_dpif *ofport;
1555     struct ofbundle *bundle;
1556     int error;
1557
1558     if (!clogged) {
1559         complete_operations(ofproto);
1560     }
1561
1562     if (mbridge_need_revalidate(ofproto->mbridge)) {
1563         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1564         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1565         mac_learning_flush(ofproto->ml);
1566         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1567     }
1568
1569     /* Do not perform any periodic activity below required by 'ofproto' while
1570      * waiting for flow restore to complete. */
1571     if (ofproto_get_flow_restore_wait()) {
1572         return 0;
1573     }
1574
1575     error = run_fast(ofproto_);
1576     if (error) {
1577         return error;
1578     }
1579
1580     if (ofproto->netflow) {
1581         if (netflow_run(ofproto->netflow)) {
1582             send_netflow_active_timeouts(ofproto);
1583         }
1584     }
1585     if (ofproto->sflow) {
1586         dpif_sflow_run(ofproto->sflow);
1587     }
1588     if (ofproto->ipfix) {
1589         dpif_ipfix_run(ofproto->ipfix);
1590     }
1591
1592     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1593         port_run(ofport);
1594     }
1595     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1596         bundle_run(bundle);
1597     }
1598
1599     stp_run(ofproto);
1600     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1601     if (mac_learning_run(ofproto->ml)) {
1602         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1603     }
1604     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1605
1606     /* Check the consistency of a random facet, to aid debugging. */
1607     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1608     if (time_msec() >= ofproto->consistency_rl
1609         && !classifier_is_empty(&ofproto->facets)
1610         && !ofproto->backer->need_revalidate) {
1611         struct cls_table *table;
1612         struct cls_rule *cr;
1613         struct facet *facet;
1614
1615         ofproto->consistency_rl = time_msec() + 250;
1616
1617         table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1618                              struct cls_table, hmap_node);
1619         cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1620                           hmap_node);
1621         facet = CONTAINER_OF(cr, struct facet, cr);
1622
1623         if (!facet_check_consistency(facet)) {
1624             ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1625         }
1626     }
1627     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1628
1629     return 0;
1630 }
1631
1632 static void
1633 wait(struct ofproto *ofproto_)
1634 {
1635     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1636     struct ofport_dpif *ofport;
1637     struct ofbundle *bundle;
1638
1639     if (!clogged && !list_is_empty(&ofproto->completions)) {
1640         poll_immediate_wake();
1641     }
1642
1643     if (ofproto_get_flow_restore_wait()) {
1644         return;
1645     }
1646
1647     if (ofproto->sflow) {
1648         dpif_sflow_wait(ofproto->sflow);
1649     }
1650     if (ofproto->ipfix) {
1651         dpif_ipfix_wait(ofproto->ipfix);
1652     }
1653     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1654         port_wait(ofport);
1655     }
1656     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1657         bundle_wait(bundle);
1658     }
1659     if (ofproto->netflow) {
1660         netflow_wait(ofproto->netflow);
1661     }
1662     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1663     mac_learning_wait(ofproto->ml);
1664     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1665     stp_wait(ofproto);
1666     if (ofproto->backer->need_revalidate) {
1667         /* Shouldn't happen, but if it does just go around again. */
1668         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1669         poll_immediate_wake();
1670     }
1671 }
1672
1673 static void
1674 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1675 {
1676     const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1677     struct cls_cursor cursor;
1678     size_t n_subfacets = 0;
1679     struct facet *facet;
1680
1681     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1682     simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1683     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1684
1685     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1686     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1687     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1688         n_subfacets += list_size(&facet->subfacets);
1689     }
1690     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1691     simap_increase(usage, "subfacets", n_subfacets);
1692 }
1693
1694 static void
1695 flush(struct ofproto *ofproto_)
1696 {
1697     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1698     struct subfacet *subfacet, *next_subfacet;
1699     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1700     int n_batch;
1701
1702     n_batch = 0;
1703     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1704                         &ofproto->backer->subfacets) {
1705         if (subfacet->facet->ofproto != ofproto) {
1706             continue;
1707         }
1708
1709         if (subfacet->path != SF_NOT_INSTALLED) {
1710             batch[n_batch++] = subfacet;
1711             if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1712                 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1713                 n_batch = 0;
1714             }
1715         } else {
1716             subfacet_destroy(subfacet);
1717         }
1718     }
1719
1720     if (n_batch > 0) {
1721         subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1722     }
1723 }
1724
1725 static void
1726 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1727              bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1728 {
1729     *arp_match_ip = true;
1730     *actions = (OFPUTIL_A_OUTPUT |
1731                 OFPUTIL_A_SET_VLAN_VID |
1732                 OFPUTIL_A_SET_VLAN_PCP |
1733                 OFPUTIL_A_STRIP_VLAN |
1734                 OFPUTIL_A_SET_DL_SRC |
1735                 OFPUTIL_A_SET_DL_DST |
1736                 OFPUTIL_A_SET_NW_SRC |
1737                 OFPUTIL_A_SET_NW_DST |
1738                 OFPUTIL_A_SET_NW_TOS |
1739                 OFPUTIL_A_SET_TP_SRC |
1740                 OFPUTIL_A_SET_TP_DST |
1741                 OFPUTIL_A_ENQUEUE);
1742 }
1743
1744 static void
1745 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1746 {
1747     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1748     struct dpif_dp_stats s;
1749     uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1750     uint64_t n_lookup;
1751
1752     strcpy(ots->name, "classifier");
1753
1754     dpif_get_dp_stats(ofproto->backer->dpif, &s);
1755     rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1756     rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1757     rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1758
1759     n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1760     ots->lookup_count = htonll(n_lookup);
1761     ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1762 }
1763
1764 static struct ofport *
1765 port_alloc(void)
1766 {
1767     struct ofport_dpif *port = xmalloc(sizeof *port);
1768     return &port->up;
1769 }
1770
1771 static void
1772 port_dealloc(struct ofport *port_)
1773 {
1774     struct ofport_dpif *port = ofport_dpif_cast(port_);
1775     free(port);
1776 }
1777
1778 static int
1779 port_construct(struct ofport *port_)
1780 {
1781     struct ofport_dpif *port = ofport_dpif_cast(port_);
1782     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1783     const struct netdev *netdev = port->up.netdev;
1784     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1785     struct dpif_port dpif_port;
1786     int error;
1787
1788     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1789     port->bundle = NULL;
1790     port->cfm = NULL;
1791     port->bfd = NULL;
1792     port->may_enable = true;
1793     port->stp_port = NULL;
1794     port->stp_state = STP_DISABLED;
1795     port->is_tunnel = false;
1796     port->peer = NULL;
1797     port->qdscp = NULL;
1798     port->n_qdscp = 0;
1799     port->realdev_ofp_port = 0;
1800     port->vlandev_vid = 0;
1801     port->carrier_seq = netdev_get_carrier_resets(netdev);
1802
1803     if (netdev_vport_is_patch(netdev)) {
1804         /* By bailing out here, we don't submit the port to the sFlow module
1805          * to be considered for counter polling export.  This is correct
1806          * because the patch port represents an interface that sFlow considers
1807          * to be "internal" to the switch as a whole, and therefore not an
1808          * candidate for counter polling. */
1809         port->odp_port = ODPP_NONE;
1810         ofport_update_peer(port);
1811         return 0;
1812     }
1813
1814     error = dpif_port_query_by_name(ofproto->backer->dpif,
1815                                     netdev_vport_get_dpif_port(netdev, namebuf,
1816                                                                sizeof namebuf),
1817                                     &dpif_port);
1818     if (error) {
1819         return error;
1820     }
1821
1822     port->odp_port = dpif_port.port_no;
1823
1824     if (netdev_get_tunnel_config(netdev)) {
1825         tnl_port_add(port, port->up.netdev, port->odp_port);
1826         port->is_tunnel = true;
1827     } else {
1828         /* Sanity-check that a mapping doesn't already exist.  This
1829          * shouldn't happen for non-tunnel ports. */
1830         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1831             VLOG_ERR("port %s already has an OpenFlow port number",
1832                      dpif_port.name);
1833             dpif_port_destroy(&dpif_port);
1834             return EBUSY;
1835         }
1836
1837         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1838         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1839                     hash_odp_port(port->odp_port));
1840         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1841     }
1842     dpif_port_destroy(&dpif_port);
1843
1844     if (ofproto->sflow) {
1845         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1846     }
1847
1848     return 0;
1849 }
1850
1851 static void
1852 port_destruct(struct ofport *port_)
1853 {
1854     struct ofport_dpif *port = ofport_dpif_cast(port_);
1855     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1856     const char *devname = netdev_get_name(port->up.netdev);
1857     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1858     const char *dp_port_name;
1859
1860     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1861     ovs_rwlock_wrlock(&xlate_rwlock);
1862     xlate_ofport_remove(port);
1863     ovs_rwlock_unlock(&xlate_rwlock);
1864
1865     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1866                                               sizeof namebuf);
1867     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1868         /* The underlying device is still there, so delete it.  This
1869          * happens when the ofproto is being destroyed, since the caller
1870          * assumes that removal of attached ports will happen as part of
1871          * destruction. */
1872         if (!port->is_tunnel) {
1873             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1874         }
1875     }
1876
1877     if (port->peer) {
1878         port->peer->peer = NULL;
1879         port->peer = NULL;
1880     }
1881
1882     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1883         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1884         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1885         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1886     }
1887
1888     tnl_port_del(port);
1889     sset_find_and_delete(&ofproto->ports, devname);
1890     sset_find_and_delete(&ofproto->ghost_ports, devname);
1891     bundle_remove(port_);
1892     set_cfm(port_, NULL);
1893     set_bfd(port_, NULL);
1894     if (ofproto->sflow) {
1895         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1896     }
1897
1898     free(port->qdscp);
1899 }
1900
1901 static void
1902 port_modified(struct ofport *port_)
1903 {
1904     struct ofport_dpif *port = ofport_dpif_cast(port_);
1905
1906     if (port->bundle && port->bundle->bond) {
1907         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1908     }
1909
1910     if (port->cfm) {
1911         cfm_set_netdev(port->cfm, port->up.netdev);
1912     }
1913
1914     if (port->bfd) {
1915         bfd_set_netdev(port->bfd, port->up.netdev);
1916     }
1917
1918     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1919                                                 port->odp_port)) {
1920         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1921             REV_RECONFIGURE;
1922     }
1923
1924     ofport_update_peer(port);
1925 }
1926
1927 static void
1928 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1929 {
1930     struct ofport_dpif *port = ofport_dpif_cast(port_);
1931     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1932     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1933
1934     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1935                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1936                    OFPUTIL_PC_NO_PACKET_IN)) {
1937         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1938
1939         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1940             bundle_update(port->bundle);
1941         }
1942     }
1943 }
1944
1945 static int
1946 set_sflow(struct ofproto *ofproto_,
1947           const struct ofproto_sflow_options *sflow_options)
1948 {
1949     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1950     struct dpif_sflow *ds = ofproto->sflow;
1951
1952     if (sflow_options) {
1953         if (!ds) {
1954             struct ofport_dpif *ofport;
1955
1956             ds = ofproto->sflow = dpif_sflow_create();
1957             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1958                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1959             }
1960             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1961         }
1962         dpif_sflow_set_options(ds, sflow_options);
1963     } else {
1964         if (ds) {
1965             dpif_sflow_unref(ds);
1966             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1967             ofproto->sflow = NULL;
1968         }
1969     }
1970     return 0;
1971 }
1972
1973 static int
1974 set_ipfix(
1975     struct ofproto *ofproto_,
1976     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1977     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1978     size_t n_flow_exporters_options)
1979 {
1980     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1981     struct dpif_ipfix *di = ofproto->ipfix;
1982     bool has_options = bridge_exporter_options || flow_exporters_options;
1983
1984     if (has_options && !di) {
1985         di = ofproto->ipfix = dpif_ipfix_create();
1986     }
1987
1988     if (di) {
1989         /* Call set_options in any case to cleanly flush the flow
1990          * caches in the last exporters that are to be destroyed. */
1991         dpif_ipfix_set_options(
1992             di, bridge_exporter_options, flow_exporters_options,
1993             n_flow_exporters_options);
1994
1995         if (!has_options) {
1996             dpif_ipfix_unref(di);
1997             ofproto->ipfix = NULL;
1998         }
1999     }
2000
2001     return 0;
2002 }
2003
2004 static int
2005 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
2006 {
2007     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2008     int error;
2009
2010     if (!s) {
2011         error = 0;
2012     } else {
2013         if (!ofport->cfm) {
2014             struct ofproto_dpif *ofproto;
2015
2016             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2017             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2018             ofport->cfm = cfm_create(ofport->up.netdev);
2019         }
2020
2021         if (cfm_configure(ofport->cfm, s)) {
2022             return 0;
2023         }
2024
2025         error = EINVAL;
2026     }
2027     cfm_unref(ofport->cfm);
2028     ofport->cfm = NULL;
2029     return error;
2030 }
2031
2032 static bool
2033 get_cfm_status(const struct ofport *ofport_,
2034                struct ofproto_cfm_status *status)
2035 {
2036     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2037
2038     if (ofport->cfm) {
2039         status->faults = cfm_get_fault(ofport->cfm);
2040         status->remote_opstate = cfm_get_opup(ofport->cfm);
2041         status->health = cfm_get_health(ofport->cfm);
2042         cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2043         return true;
2044     } else {
2045         return false;
2046     }
2047 }
2048
2049 static int
2050 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2051 {
2052     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2053     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2054     struct bfd *old;
2055
2056     old = ofport->bfd;
2057     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2058                                 cfg, ofport->up.netdev);
2059     if (ofport->bfd != old) {
2060         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2061     }
2062
2063     return 0;
2064 }
2065
2066 static int
2067 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2068 {
2069     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2070
2071     if (ofport->bfd) {
2072         bfd_get_status(ofport->bfd, smap);
2073         return 0;
2074     } else {
2075         return ENOENT;
2076     }
2077 }
2078 \f
2079 /* Spanning Tree. */
2080
2081 static void
2082 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2083 {
2084     struct ofproto_dpif *ofproto = ofproto_;
2085     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2086     struct ofport_dpif *ofport;
2087
2088     ofport = stp_port_get_aux(sp);
2089     if (!ofport) {
2090         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2091                      ofproto->up.name, port_num);
2092     } else {
2093         struct eth_header *eth = pkt->l2;
2094
2095         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2096         if (eth_addr_is_zero(eth->eth_src)) {
2097             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2098                          "with unknown MAC", ofproto->up.name, port_num);
2099         } else {
2100             send_packet(ofport, pkt);
2101         }
2102     }
2103     ofpbuf_delete(pkt);
2104 }
2105
2106 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2107 static int
2108 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2109 {
2110     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2111
2112     /* Only revalidate flows if the configuration changed. */
2113     if (!s != !ofproto->stp) {
2114         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2115     }
2116
2117     if (s) {
2118         if (!ofproto->stp) {
2119             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2120                                       send_bpdu_cb, ofproto);
2121             ofproto->stp_last_tick = time_msec();
2122         }
2123
2124         stp_set_bridge_id(ofproto->stp, s->system_id);
2125         stp_set_bridge_priority(ofproto->stp, s->priority);
2126         stp_set_hello_time(ofproto->stp, s->hello_time);
2127         stp_set_max_age(ofproto->stp, s->max_age);
2128         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2129     }  else {
2130         struct ofport *ofport;
2131
2132         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2133             set_stp_port(ofport, NULL);
2134         }
2135
2136         stp_unref(ofproto->stp);
2137         ofproto->stp = NULL;
2138     }
2139
2140     return 0;
2141 }
2142
2143 static int
2144 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2145 {
2146     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2147
2148     if (ofproto->stp) {
2149         s->enabled = true;
2150         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2151         s->designated_root = stp_get_designated_root(ofproto->stp);
2152         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2153     } else {
2154         s->enabled = false;
2155     }
2156
2157     return 0;
2158 }
2159
2160 static void
2161 update_stp_port_state(struct ofport_dpif *ofport)
2162 {
2163     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2164     enum stp_state state;
2165
2166     /* Figure out new state. */
2167     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2168                              : STP_DISABLED;
2169
2170     /* Update state. */
2171     if (ofport->stp_state != state) {
2172         enum ofputil_port_state of_state;
2173         bool fwd_change;
2174
2175         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2176                     netdev_get_name(ofport->up.netdev),
2177                     stp_state_name(ofport->stp_state),
2178                     stp_state_name(state));
2179         if (stp_learn_in_state(ofport->stp_state)
2180                 != stp_learn_in_state(state)) {
2181             /* xxx Learning action flows should also be flushed. */
2182             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2183             mac_learning_flush(ofproto->ml);
2184             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2185         }
2186         fwd_change = stp_forward_in_state(ofport->stp_state)
2187                         != stp_forward_in_state(state);
2188
2189         ofproto->backer->need_revalidate = REV_STP;
2190         ofport->stp_state = state;
2191         ofport->stp_state_entered = time_msec();
2192
2193         if (fwd_change && ofport->bundle) {
2194             bundle_update(ofport->bundle);
2195         }
2196
2197         /* Update the STP state bits in the OpenFlow port description. */
2198         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2199         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2200                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2201                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2202                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2203                      : 0);
2204         ofproto_port_set_state(&ofport->up, of_state);
2205     }
2206 }
2207
2208 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2209  * caller is responsible for assigning STP port numbers and ensuring
2210  * there are no duplicates. */
2211 static int
2212 set_stp_port(struct ofport *ofport_,
2213              const struct ofproto_port_stp_settings *s)
2214 {
2215     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2216     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2217     struct stp_port *sp = ofport->stp_port;
2218
2219     if (!s || !s->enable) {
2220         if (sp) {
2221             ofport->stp_port = NULL;
2222             stp_port_disable(sp);
2223             update_stp_port_state(ofport);
2224         }
2225         return 0;
2226     } else if (sp && stp_port_no(sp) != s->port_num
2227             && ofport == stp_port_get_aux(sp)) {
2228         /* The port-id changed, so disable the old one if it's not
2229          * already in use by another port. */
2230         stp_port_disable(sp);
2231     }
2232
2233     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2234     stp_port_enable(sp);
2235
2236     stp_port_set_aux(sp, ofport);
2237     stp_port_set_priority(sp, s->priority);
2238     stp_port_set_path_cost(sp, s->path_cost);
2239
2240     update_stp_port_state(ofport);
2241
2242     return 0;
2243 }
2244
2245 static int
2246 get_stp_port_status(struct ofport *ofport_,
2247                     struct ofproto_port_stp_status *s)
2248 {
2249     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2250     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2251     struct stp_port *sp = ofport->stp_port;
2252
2253     if (!ofproto->stp || !sp) {
2254         s->enabled = false;
2255         return 0;
2256     }
2257
2258     s->enabled = true;
2259     s->port_id = stp_port_get_id(sp);
2260     s->state = stp_port_get_state(sp);
2261     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2262     s->role = stp_port_get_role(sp);
2263     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2264
2265     return 0;
2266 }
2267
2268 static void
2269 stp_run(struct ofproto_dpif *ofproto)
2270 {
2271     if (ofproto->stp) {
2272         long long int now = time_msec();
2273         long long int elapsed = now - ofproto->stp_last_tick;
2274         struct stp_port *sp;
2275
2276         if (elapsed > 0) {
2277             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2278             ofproto->stp_last_tick = now;
2279         }
2280         while (stp_get_changed_port(ofproto->stp, &sp)) {
2281             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2282
2283             if (ofport) {
2284                 update_stp_port_state(ofport);
2285             }
2286         }
2287
2288         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2289             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2290             mac_learning_flush(ofproto->ml);
2291             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2292         }
2293     }
2294 }
2295
2296 static void
2297 stp_wait(struct ofproto_dpif *ofproto)
2298 {
2299     if (ofproto->stp) {
2300         poll_timer_wait(1000);
2301     }
2302 }
2303 \f
2304 static int
2305 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2306            size_t n_qdscp)
2307 {
2308     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2309     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2310
2311     if (ofport->n_qdscp != n_qdscp
2312         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2313                               n_qdscp * sizeof *qdscp))) {
2314         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2315         free(ofport->qdscp);
2316         ofport->qdscp = n_qdscp
2317             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2318             : NULL;
2319         ofport->n_qdscp = n_qdscp;
2320     }
2321
2322     return 0;
2323 }
2324 \f
2325 /* Bundles. */
2326
2327 /* Expires all MAC learning entries associated with 'bundle' and forces its
2328  * ofproto to revalidate every flow.
2329  *
2330  * Normally MAC learning entries are removed only from the ofproto associated
2331  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2332  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2333  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2334  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2335  * with the host from which it migrated. */
2336 static void
2337 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2338 {
2339     struct ofproto_dpif *ofproto = bundle->ofproto;
2340     struct mac_learning *ml = ofproto->ml;
2341     struct mac_entry *mac, *next_mac;
2342
2343     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2344     ovs_rwlock_wrlock(&ml->rwlock);
2345     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2346         if (mac->port.p == bundle) {
2347             if (all_ofprotos) {
2348                 struct ofproto_dpif *o;
2349
2350                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2351                     if (o != ofproto) {
2352                         struct mac_entry *e;
2353
2354                         ovs_rwlock_wrlock(&o->ml->rwlock);
2355                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2356                         if (e) {
2357                             mac_learning_expire(o->ml, e);
2358                         }
2359                         ovs_rwlock_unlock(&o->ml->rwlock);
2360                     }
2361                 }
2362             }
2363
2364             mac_learning_expire(ml, mac);
2365         }
2366     }
2367     ovs_rwlock_unlock(&ml->rwlock);
2368 }
2369
2370 static struct ofbundle *
2371 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2372 {
2373     struct ofbundle *bundle;
2374
2375     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2376                              &ofproto->bundles) {
2377         if (bundle->aux == aux) {
2378             return bundle;
2379         }
2380     }
2381     return NULL;
2382 }
2383
2384 static void
2385 bundle_update(struct ofbundle *bundle)
2386 {
2387     struct ofport_dpif *port;
2388
2389     bundle->floodable = true;
2390     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2391         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2392             || !stp_forward_in_state(port->stp_state)) {
2393             bundle->floodable = false;
2394             break;
2395         }
2396     }
2397 }
2398
2399 static void
2400 bundle_del_port(struct ofport_dpif *port)
2401 {
2402     struct ofbundle *bundle = port->bundle;
2403
2404     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2405
2406     list_remove(&port->bundle_node);
2407     port->bundle = NULL;
2408
2409     if (bundle->lacp) {
2410         lacp_slave_unregister(bundle->lacp, port);
2411     }
2412     if (bundle->bond) {
2413         bond_slave_unregister(bundle->bond, port);
2414     }
2415
2416     bundle_update(bundle);
2417 }
2418
2419 static bool
2420 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2421                 struct lacp_slave_settings *lacp)
2422 {
2423     struct ofport_dpif *port;
2424
2425     port = get_ofp_port(bundle->ofproto, ofp_port);
2426     if (!port) {
2427         return false;
2428     }
2429
2430     if (port->bundle != bundle) {
2431         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2432         if (port->bundle) {
2433             bundle_remove(&port->up);
2434         }
2435
2436         port->bundle = bundle;
2437         list_push_back(&bundle->ports, &port->bundle_node);
2438         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2439             || !stp_forward_in_state(port->stp_state)) {
2440             bundle->floodable = false;
2441         }
2442     }
2443     if (lacp) {
2444         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2445         lacp_slave_register(bundle->lacp, port, lacp);
2446     }
2447
2448     return true;
2449 }
2450
2451 static void
2452 bundle_destroy(struct ofbundle *bundle)
2453 {
2454     struct ofproto_dpif *ofproto;
2455     struct ofport_dpif *port, *next_port;
2456
2457     if (!bundle) {
2458         return;
2459     }
2460
2461     ofproto = bundle->ofproto;
2462     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2463
2464     ovs_rwlock_wrlock(&xlate_rwlock);
2465     xlate_bundle_remove(bundle);
2466     ovs_rwlock_unlock(&xlate_rwlock);
2467
2468     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2469         bundle_del_port(port);
2470     }
2471
2472     bundle_flush_macs(bundle, true);
2473     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2474     free(bundle->name);
2475     free(bundle->trunks);
2476     lacp_unref(bundle->lacp);
2477     bond_unref(bundle->bond);
2478     free(bundle);
2479 }
2480
2481 static int
2482 bundle_set(struct ofproto *ofproto_, void *aux,
2483            const struct ofproto_bundle_settings *s)
2484 {
2485     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2486     bool need_flush = false;
2487     struct ofport_dpif *port;
2488     struct ofbundle *bundle;
2489     unsigned long *trunks;
2490     int vlan;
2491     size_t i;
2492     bool ok;
2493
2494     if (!s) {
2495         bundle_destroy(bundle_lookup(ofproto, aux));
2496         return 0;
2497     }
2498
2499     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2500     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2501
2502     bundle = bundle_lookup(ofproto, aux);
2503     if (!bundle) {
2504         bundle = xmalloc(sizeof *bundle);
2505
2506         bundle->ofproto = ofproto;
2507         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2508                     hash_pointer(aux, 0));
2509         bundle->aux = aux;
2510         bundle->name = NULL;
2511
2512         list_init(&bundle->ports);
2513         bundle->vlan_mode = PORT_VLAN_TRUNK;
2514         bundle->vlan = -1;
2515         bundle->trunks = NULL;
2516         bundle->use_priority_tags = s->use_priority_tags;
2517         bundle->lacp = NULL;
2518         bundle->bond = NULL;
2519
2520         bundle->floodable = true;
2521         mbridge_register_bundle(ofproto->mbridge, bundle);
2522     }
2523
2524     if (!bundle->name || strcmp(s->name, bundle->name)) {
2525         free(bundle->name);
2526         bundle->name = xstrdup(s->name);
2527     }
2528
2529     /* LACP. */
2530     if (s->lacp) {
2531         if (!bundle->lacp) {
2532             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2533             bundle->lacp = lacp_create();
2534         }
2535         lacp_configure(bundle->lacp, s->lacp);
2536     } else {
2537         lacp_unref(bundle->lacp);
2538         bundle->lacp = NULL;
2539     }
2540
2541     /* Update set of ports. */
2542     ok = true;
2543     for (i = 0; i < s->n_slaves; i++) {
2544         if (!bundle_add_port(bundle, s->slaves[i],
2545                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2546             ok = false;
2547         }
2548     }
2549     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2550         struct ofport_dpif *next_port;
2551
2552         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2553             for (i = 0; i < s->n_slaves; i++) {
2554                 if (s->slaves[i] == port->up.ofp_port) {
2555                     goto found;
2556                 }
2557             }
2558
2559             bundle_del_port(port);
2560         found: ;
2561         }
2562     }
2563     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2564
2565     if (list_is_empty(&bundle->ports)) {
2566         bundle_destroy(bundle);
2567         return EINVAL;
2568     }
2569
2570     /* Set VLAN tagging mode */
2571     if (s->vlan_mode != bundle->vlan_mode
2572         || s->use_priority_tags != bundle->use_priority_tags) {
2573         bundle->vlan_mode = s->vlan_mode;
2574         bundle->use_priority_tags = s->use_priority_tags;
2575         need_flush = true;
2576     }
2577
2578     /* Set VLAN tag. */
2579     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2580             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2581             : 0);
2582     if (vlan != bundle->vlan) {
2583         bundle->vlan = vlan;
2584         need_flush = true;
2585     }
2586
2587     /* Get trunked VLANs. */
2588     switch (s->vlan_mode) {
2589     case PORT_VLAN_ACCESS:
2590         trunks = NULL;
2591         break;
2592
2593     case PORT_VLAN_TRUNK:
2594         trunks = CONST_CAST(unsigned long *, s->trunks);
2595         break;
2596
2597     case PORT_VLAN_NATIVE_UNTAGGED:
2598     case PORT_VLAN_NATIVE_TAGGED:
2599         if (vlan != 0 && (!s->trunks
2600                           || !bitmap_is_set(s->trunks, vlan)
2601                           || bitmap_is_set(s->trunks, 0))) {
2602             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2603             if (s->trunks) {
2604                 trunks = bitmap_clone(s->trunks, 4096);
2605             } else {
2606                 trunks = bitmap_allocate1(4096);
2607             }
2608             bitmap_set1(trunks, vlan);
2609             bitmap_set0(trunks, 0);
2610         } else {
2611             trunks = CONST_CAST(unsigned long *, s->trunks);
2612         }
2613         break;
2614
2615     default:
2616         NOT_REACHED();
2617     }
2618     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2619         free(bundle->trunks);
2620         if (trunks == s->trunks) {
2621             bundle->trunks = vlan_bitmap_clone(trunks);
2622         } else {
2623             bundle->trunks = trunks;
2624             trunks = NULL;
2625         }
2626         need_flush = true;
2627     }
2628     if (trunks != s->trunks) {
2629         free(trunks);
2630     }
2631
2632     /* Bonding. */
2633     if (!list_is_short(&bundle->ports)) {
2634         bundle->ofproto->has_bonded_bundles = true;
2635         if (bundle->bond) {
2636             if (bond_reconfigure(bundle->bond, s->bond)) {
2637                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2638             }
2639         } else {
2640             bundle->bond = bond_create(s->bond);
2641             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2642         }
2643
2644         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2645             bond_slave_register(bundle->bond, port, port->up.netdev);
2646         }
2647     } else {
2648         bond_unref(bundle->bond);
2649         bundle->bond = NULL;
2650     }
2651
2652     /* If we changed something that would affect MAC learning, un-learn
2653      * everything on this port and force flow revalidation. */
2654     if (need_flush) {
2655         bundle_flush_macs(bundle, false);
2656     }
2657
2658     return 0;
2659 }
2660
2661 static void
2662 bundle_remove(struct ofport *port_)
2663 {
2664     struct ofport_dpif *port = ofport_dpif_cast(port_);
2665     struct ofbundle *bundle = port->bundle;
2666
2667     if (bundle) {
2668         bundle_del_port(port);
2669         if (list_is_empty(&bundle->ports)) {
2670             bundle_destroy(bundle);
2671         } else if (list_is_short(&bundle->ports)) {
2672             bond_unref(bundle->bond);
2673             bundle->bond = NULL;
2674         }
2675     }
2676 }
2677
2678 static void
2679 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2680 {
2681     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2682     struct ofport_dpif *port = port_;
2683     uint8_t ea[ETH_ADDR_LEN];
2684     int error;
2685
2686     error = netdev_get_etheraddr(port->up.netdev, ea);
2687     if (!error) {
2688         struct ofpbuf packet;
2689         void *packet_pdu;
2690
2691         ofpbuf_init(&packet, 0);
2692         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2693                                  pdu_size);
2694         memcpy(packet_pdu, pdu, pdu_size);
2695
2696         send_packet(port, &packet);
2697         ofpbuf_uninit(&packet);
2698     } else {
2699         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2700                     "%s (%s)", port->bundle->name,
2701                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2702     }
2703 }
2704
2705 static void
2706 bundle_send_learning_packets(struct ofbundle *bundle)
2707 {
2708     struct ofproto_dpif *ofproto = bundle->ofproto;
2709     int error, n_packets, n_errors;
2710     struct mac_entry *e;
2711
2712     error = n_packets = n_errors = 0;
2713     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2714     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2715         if (e->port.p != bundle) {
2716             struct ofpbuf *learning_packet;
2717             struct ofport_dpif *port;
2718             void *port_void;
2719             int ret;
2720
2721             /* The assignment to "port" is unnecessary but makes "grep"ing for
2722              * struct ofport_dpif more effective. */
2723             learning_packet = bond_compose_learning_packet(bundle->bond,
2724                                                            e->mac, e->vlan,
2725                                                            &port_void);
2726             port = port_void;
2727             ret = send_packet(port, learning_packet);
2728             ofpbuf_delete(learning_packet);
2729             if (ret) {
2730                 error = ret;
2731                 n_errors++;
2732             }
2733             n_packets++;
2734         }
2735     }
2736     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2737
2738     if (n_errors) {
2739         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2740         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2741                      "packets, last error was: %s",
2742                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2743     } else {
2744         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2745                  bundle->name, n_packets);
2746     }
2747 }
2748
2749 static void
2750 bundle_run(struct ofbundle *bundle)
2751 {
2752     if (bundle->lacp) {
2753         lacp_run(bundle->lacp, send_pdu_cb);
2754     }
2755     if (bundle->bond) {
2756         struct ofport_dpif *port;
2757
2758         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2759             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2760         }
2761
2762         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2763             bundle->ofproto->backer->need_revalidate = REV_BOND;
2764         }
2765
2766         if (bond_should_send_learning_packets(bundle->bond)) {
2767             bundle_send_learning_packets(bundle);
2768         }
2769     }
2770 }
2771
2772 static void
2773 bundle_wait(struct ofbundle *bundle)
2774 {
2775     if (bundle->lacp) {
2776         lacp_wait(bundle->lacp);
2777     }
2778     if (bundle->bond) {
2779         bond_wait(bundle->bond);
2780     }
2781 }
2782 \f
2783 /* Mirrors. */
2784
2785 static int
2786 mirror_set__(struct ofproto *ofproto_, void *aux,
2787              const struct ofproto_mirror_settings *s)
2788 {
2789     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2790     struct ofbundle **srcs, **dsts;
2791     int error;
2792     size_t i;
2793
2794     if (!s) {
2795         mirror_destroy(ofproto->mbridge, aux);
2796         return 0;
2797     }
2798
2799     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2800     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2801
2802     for (i = 0; i < s->n_srcs; i++) {
2803         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2804     }
2805
2806     for (i = 0; i < s->n_dsts; i++) {
2807         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2808     }
2809
2810     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2811                        s->n_dsts, s->src_vlans,
2812                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2813     free(srcs);
2814     free(dsts);
2815     return error;
2816 }
2817
2818 static int
2819 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2820                    uint64_t *packets, uint64_t *bytes)
2821 {
2822     push_all_stats();
2823     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2824                             bytes);
2825 }
2826
2827 static int
2828 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2829 {
2830     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2831     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2832     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2833         mac_learning_flush(ofproto->ml);
2834     }
2835     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2836     return 0;
2837 }
2838
2839 static bool
2840 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2841 {
2842     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2843     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2844     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2845 }
2846
2847 static void
2848 forward_bpdu_changed(struct ofproto *ofproto_)
2849 {
2850     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2851     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2852 }
2853
2854 static void
2855 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2856                      size_t max_entries)
2857 {
2858     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2859     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2860     mac_learning_set_idle_time(ofproto->ml, idle_time);
2861     mac_learning_set_max_entries(ofproto->ml, max_entries);
2862     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2863 }
2864 \f
2865 /* Ports. */
2866
2867 static struct ofport_dpif *
2868 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2869 {
2870     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2871     return ofport ? ofport_dpif_cast(ofport) : NULL;
2872 }
2873
2874 static struct ofport_dpif *
2875 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2876 {
2877     struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2878     return port && &ofproto->up == port->up.ofproto ? port : NULL;
2879 }
2880
2881 static void
2882 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2883                             struct ofproto_port *ofproto_port,
2884                             struct dpif_port *dpif_port)
2885 {
2886     ofproto_port->name = dpif_port->name;
2887     ofproto_port->type = dpif_port->type;
2888     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2889 }
2890
2891 static void
2892 ofport_update_peer(struct ofport_dpif *ofport)
2893 {
2894     const struct ofproto_dpif *ofproto;
2895     struct dpif_backer *backer;
2896     char *peer_name;
2897
2898     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2899         return;
2900     }
2901
2902     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2903     backer->need_revalidate = REV_RECONFIGURE;
2904
2905     if (ofport->peer) {
2906         ofport->peer->peer = NULL;
2907         ofport->peer = NULL;
2908     }
2909
2910     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2911     if (!peer_name) {
2912         return;
2913     }
2914
2915     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2916         struct ofport *peer_ofport;
2917         struct ofport_dpif *peer;
2918         char *peer_peer;
2919
2920         if (ofproto->backer != backer) {
2921             continue;
2922         }
2923
2924         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2925         if (!peer_ofport) {
2926             continue;
2927         }
2928
2929         peer = ofport_dpif_cast(peer_ofport);
2930         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2931         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2932                                  peer_peer)) {
2933             ofport->peer = peer;
2934             ofport->peer->peer = ofport;
2935         }
2936         free(peer_peer);
2937
2938         break;
2939     }
2940     free(peer_name);
2941 }
2942
2943 static void
2944 port_run_fast(struct ofport_dpif *ofport)
2945 {
2946     if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2947         struct ofpbuf packet;
2948
2949         ofpbuf_init(&packet, 0);
2950         cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2951         send_packet(ofport, &packet);
2952         ofpbuf_uninit(&packet);
2953     }
2954
2955     if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2956         struct ofpbuf packet;
2957
2958         ofpbuf_init(&packet, 0);
2959         bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2960         send_packet(ofport, &packet);
2961         ofpbuf_uninit(&packet);
2962     }
2963 }
2964
2965 static void
2966 port_run(struct ofport_dpif *ofport)
2967 {
2968     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2969     bool carrier_changed = carrier_seq != ofport->carrier_seq;
2970     bool enable = netdev_get_carrier(ofport->up.netdev);
2971     bool cfm_enable = false;
2972     bool bfd_enable = false;
2973
2974     ofport->carrier_seq = carrier_seq;
2975
2976     port_run_fast(ofport);
2977
2978     if (ofport->cfm) {
2979         int cfm_opup = cfm_get_opup(ofport->cfm);
2980
2981         cfm_run(ofport->cfm);
2982         cfm_enable = !cfm_get_fault(ofport->cfm);
2983
2984         if (cfm_opup >= 0) {
2985             cfm_enable = cfm_enable && cfm_opup;
2986         }
2987     }
2988
2989     if (ofport->bfd) {
2990         bfd_run(ofport->bfd);
2991         bfd_enable = bfd_forwarding(ofport->bfd);
2992     }
2993
2994     if (ofport->bfd || ofport->cfm) {
2995         enable = enable && (cfm_enable || bfd_enable);
2996     }
2997
2998     if (ofport->bundle) {
2999         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3000         if (carrier_changed) {
3001             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3002         }
3003     }
3004
3005     if (ofport->may_enable != enable) {
3006         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3007         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3008     }
3009
3010     ofport->may_enable = enable;
3011 }
3012
3013 static void
3014 port_wait(struct ofport_dpif *ofport)
3015 {
3016     if (ofport->cfm) {
3017         cfm_wait(ofport->cfm);
3018     }
3019
3020     if (ofport->bfd) {
3021         bfd_wait(ofport->bfd);
3022     }
3023 }
3024
3025 static int
3026 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3027                    struct ofproto_port *ofproto_port)
3028 {
3029     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3030     struct dpif_port dpif_port;
3031     int error;
3032
3033     if (sset_contains(&ofproto->ghost_ports, devname)) {
3034         const char *type = netdev_get_type_from_name(devname);
3035
3036         /* We may be called before ofproto->up.port_by_name is populated with
3037          * the appropriate ofport.  For this reason, we must get the name and
3038          * type from the netdev layer directly. */
3039         if (type) {
3040             const struct ofport *ofport;
3041
3042             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3043             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3044             ofproto_port->name = xstrdup(devname);
3045             ofproto_port->type = xstrdup(type);
3046             return 0;
3047         }
3048         return ENODEV;
3049     }
3050
3051     if (!sset_contains(&ofproto->ports, devname)) {
3052         return ENODEV;
3053     }
3054     error = dpif_port_query_by_name(ofproto->backer->dpif,
3055                                     devname, &dpif_port);
3056     if (!error) {
3057         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3058     }
3059     return error;
3060 }
3061
3062 static int
3063 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3064 {
3065     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3066     const char *devname = netdev_get_name(netdev);
3067     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3068     const char *dp_port_name;
3069
3070     if (netdev_vport_is_patch(netdev)) {
3071         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3072         return 0;
3073     }
3074
3075     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3076     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3077         odp_port_t port_no = ODPP_NONE;
3078         int error;
3079
3080         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3081         if (error) {
3082             return error;
3083         }
3084         if (netdev_get_tunnel_config(netdev)) {
3085             simap_put(&ofproto->backer->tnl_backers,
3086                       dp_port_name, odp_to_u32(port_no));
3087         }
3088     }
3089
3090     if (netdev_get_tunnel_config(netdev)) {
3091         sset_add(&ofproto->ghost_ports, devname);
3092     } else {
3093         sset_add(&ofproto->ports, devname);
3094     }
3095     return 0;
3096 }
3097
3098 static int
3099 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3100 {
3101     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3102     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3103     int error = 0;
3104
3105     if (!ofport) {
3106         return 0;
3107     }
3108
3109     sset_find_and_delete(&ofproto->ghost_ports,
3110                          netdev_get_name(ofport->up.netdev));
3111     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3112     if (!ofport->is_tunnel) {
3113         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3114         if (!error) {
3115             /* The caller is going to close ofport->up.netdev.  If this is a
3116              * bonded port, then the bond is using that netdev, so remove it
3117              * from the bond.  The client will need to reconfigure everything
3118              * after deleting ports, so then the slave will get re-added. */
3119             bundle_remove(&ofport->up);
3120         }
3121     }
3122     return error;
3123 }
3124
3125 static int
3126 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3127 {
3128     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3129     int error;
3130
3131     push_all_stats();
3132
3133     error = netdev_get_stats(ofport->up.netdev, stats);
3134
3135     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3136         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3137
3138         /* ofproto->stats.tx_packets represents packets that we created
3139          * internally and sent to some port (e.g. packets sent with
3140          * send_packet()).  Account for them as if they had come from
3141          * OFPP_LOCAL and got forwarded. */
3142
3143         if (stats->rx_packets != UINT64_MAX) {
3144             stats->rx_packets += ofproto->stats.tx_packets;
3145         }
3146
3147         if (stats->rx_bytes != UINT64_MAX) {
3148             stats->rx_bytes += ofproto->stats.tx_bytes;
3149         }
3150
3151         /* ofproto->stats.rx_packets represents packets that were received on
3152          * some port and we processed internally and dropped (e.g. STP).
3153          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3154
3155         if (stats->tx_packets != UINT64_MAX) {
3156             stats->tx_packets += ofproto->stats.rx_packets;
3157         }
3158
3159         if (stats->tx_bytes != UINT64_MAX) {
3160             stats->tx_bytes += ofproto->stats.rx_bytes;
3161         }
3162     }
3163
3164     return error;
3165 }
3166
3167 struct port_dump_state {
3168     uint32_t bucket;
3169     uint32_t offset;
3170     bool ghost;
3171
3172     struct ofproto_port port;
3173     bool has_port;
3174 };
3175
3176 static int
3177 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3178 {
3179     *statep = xzalloc(sizeof(struct port_dump_state));
3180     return 0;
3181 }
3182
3183 static int
3184 port_dump_next(const struct ofproto *ofproto_, void *state_,
3185                struct ofproto_port *port)
3186 {
3187     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3188     struct port_dump_state *state = state_;
3189     const struct sset *sset;
3190     struct sset_node *node;
3191
3192     if (state->has_port) {
3193         ofproto_port_destroy(&state->port);
3194         state->has_port = false;
3195     }
3196     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3197     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3198         int error;
3199
3200         error = port_query_by_name(ofproto_, node->name, &state->port);
3201         if (!error) {
3202             *port = state->port;
3203             state->has_port = true;
3204             return 0;
3205         } else if (error != ENODEV) {
3206             return error;
3207         }
3208     }
3209
3210     if (!state->ghost) {
3211         state->ghost = true;
3212         state->bucket = 0;
3213         state->offset = 0;
3214         return port_dump_next(ofproto_, state_, port);
3215     }
3216
3217     return EOF;
3218 }
3219
3220 static int
3221 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3222 {
3223     struct port_dump_state *state = state_;
3224
3225     if (state->has_port) {
3226         ofproto_port_destroy(&state->port);
3227     }
3228     free(state);
3229     return 0;
3230 }
3231
3232 static int
3233 port_poll(const struct ofproto *ofproto_, char **devnamep)
3234 {
3235     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3236
3237     if (ofproto->port_poll_errno) {
3238         int error = ofproto->port_poll_errno;
3239         ofproto->port_poll_errno = 0;
3240         return error;
3241     }
3242
3243     if (sset_is_empty(&ofproto->port_poll_set)) {
3244         return EAGAIN;
3245     }
3246
3247     *devnamep = sset_pop(&ofproto->port_poll_set);
3248     return 0;
3249 }
3250
3251 static void
3252 port_poll_wait(const struct ofproto *ofproto_)
3253 {
3254     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3255     dpif_port_poll_wait(ofproto->backer->dpif);
3256 }
3257
3258 static int
3259 port_is_lacp_current(const struct ofport *ofport_)
3260 {
3261     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3262     return (ofport->bundle && ofport->bundle->lacp
3263             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3264             : -1);
3265 }
3266 \f
3267 /* Upcall handling. */
3268
3269 struct flow_miss_op {
3270     struct dpif_op dpif_op;
3271
3272     uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3273     struct xlate_out xout;
3274     bool xout_garbage;           /* 'xout' needs to be uninitialized? */
3275
3276     struct ofpbuf mask;          /* Flow mask for "put" ops. */
3277     struct odputil_keybuf maskbuf;
3278
3279     /* If this is a "put" op, then a pointer to the subfacet that should
3280      * be marked as uninstalled if the operation fails. */
3281     struct subfacet *subfacet;
3282 };
3283
3284 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3285  * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3286  * and (usually) installing a datapath flow.  The answer is usually "yes" (a
3287  * return value of true).  However, for short flows the cost of bookkeeping is
3288  * much higher than the benefits, so when the datapath holds a large number of
3289  * flows we impose some heuristics to decide which flows are likely to be worth
3290  * tracking. */
3291 static bool
3292 flow_miss_should_make_facet(struct flow_miss *miss)
3293 {
3294     struct dpif_backer *backer = miss->ofproto->backer;
3295     uint32_t hash;
3296
3297     switch (flow_miss_model) {
3298     case OFPROTO_HANDLE_MISS_AUTO:
3299         break;
3300     case OFPROTO_HANDLE_MISS_WITH_FACETS:
3301         return true;
3302     case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3303         return false;
3304     }
3305
3306     if (!backer->governor) {
3307         size_t n_subfacets;
3308
3309         n_subfacets = hmap_count(&backer->subfacets);
3310         if (n_subfacets * 2 <= flow_eviction_threshold) {
3311             return true;
3312         }
3313
3314         backer->governor = governor_create();
3315     }
3316
3317     hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3318     return governor_should_install_flow(backer->governor, hash,
3319                                         list_size(&miss->packets));
3320 }
3321
3322 /* Handles 'miss', which matches 'facet'.  May add any required datapath
3323  * operations to 'ops', incrementing '*n_ops' for each new op.
3324  *
3325  * All of the packets in 'miss' are considered to have arrived at time
3326  * 'miss->stats.used'.  This is really important only for new facets: if we
3327  * just called time_msec() here, then the new subfacet or its packets could
3328  * look (occasionally) as though it was used some time after the facet was
3329  * used.  That can make a one-packet flow look like it has a nonzero duration,
3330  * which looks odd in e.g. NetFlow statistics. */
3331 static void
3332 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3333                             struct flow_miss_op *ops, size_t *n_ops)
3334 {
3335     enum subfacet_path want_path;
3336     struct subfacet *subfacet;
3337
3338     facet->packet_count += miss->stats.n_packets;
3339     facet->prev_packet_count += miss->stats.n_packets;
3340     facet->byte_count += miss->stats.n_bytes;
3341     facet->prev_byte_count += miss->stats.n_bytes;
3342
3343     subfacet = subfacet_create(facet, miss);
3344     want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3345
3346     /* Don't install the flow if it's the result of the "userspace"
3347      * action for an already installed facet.  This can occur when a
3348      * datapath flow with wildcards has a "userspace" action and flows
3349      * sent to userspace result in a different subfacet, which will then
3350      * be rejected as overlapping by the datapath. */
3351     if (miss->upcall_type == DPIF_UC_ACTION
3352         && !list_is_empty(&facet->subfacets)) {
3353         return;
3354     }
3355
3356     subfacet = subfacet_create(facet, miss);
3357     if (subfacet->path != want_path) {
3358         struct flow_miss_op *op = &ops[(*n_ops)++];
3359         struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3360
3361         subfacet->path = want_path;
3362
3363         ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3364         if (enable_megaflows) {
3365             odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3366                                    &miss->flow, UINT32_MAX);
3367         }
3368
3369         op->xout_garbage = false;
3370         op->dpif_op.type = DPIF_OP_FLOW_PUT;
3371         op->subfacet = subfacet;
3372         put->flags = DPIF_FP_CREATE;
3373         put->key = miss->key;
3374         put->key_len = miss->key_len;
3375         put->mask = op->mask.data;
3376         put->mask_len = op->mask.size;
3377
3378         if (want_path == SF_FAST_PATH) {
3379             put->actions = facet->xout.odp_actions.data;
3380             put->actions_len = facet->xout.odp_actions.size;
3381         } else {
3382             compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3383                               op->slow_stub, sizeof op->slow_stub,
3384                               &put->actions, &put->actions_len);
3385         }
3386         put->stats = NULL;
3387     }
3388 }
3389
3390 /* Handles flow miss 'miss'.  May add any required datapath operations
3391  * to 'ops', incrementing '*n_ops' for each new op. */
3392 static void
3393 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3394                  size_t *n_ops)
3395 {
3396     struct facet *facet;
3397
3398     miss->ofproto->n_missed += list_size(&miss->packets);
3399
3400     facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3401     if (!facet) {
3402         /* There does not exist a bijection between 'struct flow' and datapath
3403          * flow keys with fitness ODP_FIT_TO_LITTLE.  This breaks a fundamental
3404          * assumption used throughout the facet and subfacet handling code.
3405          * Since we have to handle these misses in userspace anyway, we simply
3406          * skip facet creation, avoiding the problem altogether. */
3407         if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3408             || !flow_miss_should_make_facet(miss)) {
3409             return;
3410         }
3411
3412         facet = facet_create(miss);
3413     }
3414     handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3415 }
3416
3417 static struct drop_key *
3418 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3419                 size_t key_len)
3420 {
3421     struct drop_key *drop_key;
3422
3423     HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3424                              &backer->drop_keys) {
3425         if (drop_key->key_len == key_len
3426             && !memcmp(drop_key->key, key, key_len)) {
3427             return drop_key;
3428         }
3429     }
3430     return NULL;
3431 }
3432
3433 static void
3434 drop_key_clear(struct dpif_backer *backer)
3435 {
3436     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3437     struct drop_key *drop_key, *next;
3438
3439     HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3440         int error;
3441
3442         error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3443                               NULL);
3444         if (error && !VLOG_DROP_WARN(&rl)) {
3445             struct ds ds = DS_EMPTY_INITIALIZER;
3446             odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3447             VLOG_WARN("Failed to delete drop key (%s) (%s)",
3448                       ovs_strerror(error), ds_cstr(&ds));
3449             ds_destroy(&ds);
3450         }
3451
3452         hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3453         drop_key_destroy(drop_key);
3454     }
3455
3456     udpif_drop_key_clear(backer->udpif);
3457 }
3458
3459 static void
3460 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3461 {
3462     struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3463     struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3464     struct flow_miss *miss;
3465     size_t n_ops, i;
3466
3467     /* Process each element in the to-do list, constructing the set of
3468      * operations to batch. */
3469     n_ops = 0;
3470     HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3471         handle_flow_miss(miss, flow_miss_ops, &n_ops);
3472     }
3473     ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3474
3475     /* Execute batch. */
3476     for (i = 0; i < n_ops; i++) {
3477         dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3478     }
3479     dpif_operate(backer->dpif, dpif_ops, n_ops);
3480
3481     for (i = 0; i < n_ops; i++) {
3482         if (dpif_ops[i]->error != 0
3483             && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3484             && flow_miss_ops[i].subfacet) {
3485             struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3486
3487             COVERAGE_INC(subfacet_install_fail);
3488
3489             /* Zero-out subfacet counters when installation failed, but
3490              * datapath reported hits.  This should not happen and
3491              * indicates a bug, since if the datapath flow exists, we
3492              * should not be attempting to create a new subfacet.  A
3493              * buggy datapath could trigger this, so just zero out the
3494              * counters and log an error. */
3495             if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3496                 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3497                             "datapath reported hits");
3498                 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3499             }
3500
3501             subfacet->path = SF_NOT_INSTALLED;
3502         }
3503     }
3504 }
3505
3506 static void
3507 handle_sflow_upcall(struct dpif_backer *backer,
3508                     const struct dpif_upcall *upcall)
3509 {
3510     struct ofproto_dpif *ofproto;
3511     union user_action_cookie cookie;
3512     struct flow flow;
3513     odp_port_t odp_in_port;
3514
3515     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3516                       &flow, NULL, &ofproto, &odp_in_port)
3517         || !ofproto->sflow) {
3518         return;
3519     }
3520
3521     memset(&cookie, 0, sizeof cookie);
3522     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3523     dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3524                         odp_in_port, &cookie);
3525 }
3526
3527 static void
3528 handle_flow_sample_upcall(struct dpif_backer *backer,
3529                           const struct dpif_upcall *upcall)
3530 {
3531     struct ofproto_dpif *ofproto;
3532     union user_action_cookie cookie;
3533     struct flow flow;
3534
3535     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3536                       &flow, NULL, &ofproto, NULL)
3537         || !ofproto->ipfix) {
3538         return;
3539     }
3540
3541     memset(&cookie, 0, sizeof cookie);
3542     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3543
3544     /* The flow reflects exactly the contents of the packet.  Sample
3545      * the packet using it. */
3546     dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3547                            cookie.flow_sample.collector_set_id,
3548                            cookie.flow_sample.probability,
3549                            cookie.flow_sample.obs_domain_id,
3550                            cookie.flow_sample.obs_point_id);
3551 }
3552
3553 static void
3554 handle_ipfix_upcall(struct dpif_backer *backer,
3555                     const struct dpif_upcall *upcall)
3556 {
3557     struct ofproto_dpif *ofproto;
3558     struct flow flow;
3559
3560     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3561                       &flow, NULL, &ofproto, NULL)
3562         || !ofproto->ipfix) {
3563         return;
3564     }
3565
3566     /* The flow reflects exactly the contents of the packet.  Sample
3567      * the packet using it. */
3568     dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3569 }
3570
3571 static void
3572 handle_upcalls(struct dpif_backer *backer)
3573 {
3574     struct flow_miss_batch *fmb;
3575     int n_processed;
3576
3577     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3578         struct upcall *upcall = upcall_next(backer->udpif);
3579
3580         if (!upcall) {
3581             break;
3582         }
3583
3584         switch (upcall->type) {
3585         case SFLOW_UPCALL:
3586             handle_sflow_upcall(backer, &upcall->dpif_upcall);
3587             break;
3588
3589         case FLOW_SAMPLE_UPCALL:
3590             handle_flow_sample_upcall(backer, &upcall->dpif_upcall);
3591             break;
3592
3593         case IPFIX_UPCALL:
3594             handle_ipfix_upcall(backer, &upcall->dpif_upcall);
3595             break;
3596
3597         case BAD_UPCALL:
3598             break;
3599
3600         case MISS_UPCALL:
3601             NOT_REACHED();
3602         }
3603
3604         upcall_destroy(upcall);
3605     }
3606
3607     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3608         struct drop_key *drop_key = drop_key_next(backer->udpif);
3609         if (!drop_key) {
3610             break;
3611         }
3612
3613         if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3614             hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3615                         hash_bytes(drop_key->key, drop_key->key_len, 0));
3616             dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3617                           drop_key->key, drop_key->key_len,
3618                           NULL, 0, NULL, 0, NULL);
3619         } else {
3620             drop_key_destroy(drop_key);
3621         }
3622     }
3623
3624     fmb = flow_miss_batch_next(backer->udpif);
3625     if (fmb) {
3626         handle_flow_misses(backer, fmb);
3627         flow_miss_batch_destroy(fmb);
3628     }
3629 }
3630 \f
3631 /* Flow expiration. */
3632
3633 static int subfacet_max_idle(const struct dpif_backer *);
3634 static void update_stats(struct dpif_backer *);
3635 static void rule_expire(struct rule_dpif *);
3636 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3637
3638 /* This function is called periodically by run().  Its job is to collect
3639  * updates for the flows that have been installed into the datapath, most
3640  * importantly when they last were used, and then use that information to
3641  * expire flows that have not been used recently.
3642  *
3643  * Returns the number of milliseconds after which it should be called again. */
3644 static int
3645 expire(struct dpif_backer *backer)
3646 {
3647     struct ofproto_dpif *ofproto;
3648     size_t n_subfacets;
3649     int max_idle;
3650
3651     /* Periodically clear out the drop keys in an effort to keep them
3652      * relatively few. */
3653     drop_key_clear(backer);
3654
3655     /* Update stats for each flow in the backer. */
3656     update_stats(backer);
3657
3658     n_subfacets = hmap_count(&backer->subfacets);
3659     if (n_subfacets) {
3660         struct subfacet *subfacet;
3661         long long int total, now;
3662
3663         total = 0;
3664         now = time_msec();
3665         HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3666             total += now - subfacet->created;
3667         }
3668         backer->avg_subfacet_life += total / n_subfacets;
3669     }
3670     backer->avg_subfacet_life /= 2;
3671
3672     backer->avg_n_subfacet += n_subfacets;
3673     backer->avg_n_subfacet /= 2;
3674
3675     backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3676
3677     max_idle = subfacet_max_idle(backer);
3678     expire_subfacets(backer, max_idle);
3679
3680     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3681         struct rule *rule, *next_rule;
3682
3683         if (ofproto->backer != backer) {
3684             continue;
3685         }
3686
3687         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3688          * has passed. */
3689         ovs_mutex_lock(&ofproto->up.expirable_mutex);
3690         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3691                             &ofproto->up.expirable) {
3692             rule_expire(rule_dpif_cast(rule));
3693         }
3694         ovs_mutex_unlock(&ofproto->up.expirable_mutex);
3695
3696         /* All outstanding data in existing flows has been accounted, so it's a
3697          * good time to do bond rebalancing. */
3698         if (ofproto->has_bonded_bundles) {
3699             struct ofbundle *bundle;
3700
3701             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3702                 if (bundle->bond) {
3703                     bond_rebalance(bundle->bond);
3704                 }
3705             }
3706         }
3707     }
3708
3709     return MIN(max_idle, 1000);
3710 }
3711
3712 /* Updates flow table statistics given that the datapath just reported 'stats'
3713  * as 'subfacet''s statistics. */
3714 static void
3715 update_subfacet_stats(struct subfacet *subfacet,
3716                       const struct dpif_flow_stats *stats)
3717 {
3718     struct facet *facet = subfacet->facet;
3719     struct dpif_flow_stats diff;
3720
3721     diff.tcp_flags = stats->tcp_flags;
3722     diff.used = stats->used;
3723
3724     if (stats->n_packets >= subfacet->dp_packet_count) {
3725         diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3726     } else {
3727         VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3728         diff.n_packets = 0;
3729     }
3730
3731     if (stats->n_bytes >= subfacet->dp_byte_count) {
3732         diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3733     } else {
3734         VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3735         diff.n_bytes = 0;
3736     }
3737
3738     facet->ofproto->n_hit += diff.n_packets;
3739     subfacet->dp_packet_count = stats->n_packets;
3740     subfacet->dp_byte_count = stats->n_bytes;
3741     subfacet_update_stats(subfacet, &diff);
3742
3743     if (facet->accounted_bytes < facet->byte_count) {
3744         facet_learn(facet);
3745         facet_account(facet);
3746         facet->accounted_bytes = facet->byte_count;
3747     }
3748 }
3749
3750 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3751  * about, or a flow that shouldn't be installed but was anyway.  Delete it. */
3752 static void
3753 delete_unexpected_flow(struct dpif_backer *backer,
3754                        const struct nlattr *key, size_t key_len)
3755 {
3756     if (!VLOG_DROP_WARN(&rl)) {
3757         struct ds s;
3758
3759         ds_init(&s);
3760         odp_flow_key_format(key, key_len, &s);
3761         VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3762         ds_destroy(&s);
3763     }
3764
3765     COVERAGE_INC(facet_unexpected);
3766     dpif_flow_del(backer->dpif, key, key_len, NULL);
3767 }
3768
3769 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3770  *
3771  * This function also pushes statistics updates to rules which each facet
3772  * resubmits into.  Generally these statistics will be accurate.  However, if a
3773  * facet changes the rule it resubmits into at some time in between
3774  * update_stats() runs, it is possible that statistics accrued to the
3775  * old rule will be incorrectly attributed to the new rule.  This could be
3776  * avoided by calling update_stats() whenever rules are created or
3777  * deleted.  However, the performance impact of making so many calls to the
3778  * datapath do not justify the benefit of having perfectly accurate statistics.
3779  *
3780  * In addition, this function maintains per ofproto flow hit counts. The patch
3781  * port is not treated specially. e.g. A packet ingress from br0 patched into
3782  * br1 will increase the hit count of br0 by 1, however, does not affect
3783  * the hit or miss counts of br1.
3784  */
3785 static void
3786 update_stats(struct dpif_backer *backer)
3787 {
3788     const struct dpif_flow_stats *stats;
3789     struct dpif_flow_dump dump;
3790     const struct nlattr *key, *mask;
3791     size_t key_len, mask_len;
3792
3793     dpif_flow_dump_start(&dump, backer->dpif);
3794     while (dpif_flow_dump_next(&dump, &key, &key_len,
3795                                &mask, &mask_len, NULL, NULL, &stats)) {
3796         struct subfacet *subfacet;
3797         uint32_t key_hash;
3798
3799         key_hash = odp_flow_key_hash(key, key_len);
3800         subfacet = subfacet_find(backer, key, key_len, key_hash);
3801         switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3802         case SF_FAST_PATH:
3803             update_subfacet_stats(subfacet, stats);
3804             break;
3805
3806         case SF_SLOW_PATH:
3807             /* Stats are updated per-packet. */
3808             break;
3809
3810         case SF_NOT_INSTALLED:
3811         default:
3812             delete_unexpected_flow(backer, key, key_len);
3813             break;
3814         }
3815         run_fast_rl();
3816     }
3817     dpif_flow_dump_done(&dump);
3818
3819     update_moving_averages(backer);
3820 }
3821
3822 /* Calculates and returns the number of milliseconds of idle time after which
3823  * subfacets should expire from the datapath.  When a subfacet expires, we fold
3824  * its statistics into its facet, and when a facet's last subfacet expires, we
3825  * fold its statistic into its rule. */
3826 static int
3827 subfacet_max_idle(const struct dpif_backer *backer)
3828 {
3829     /*
3830      * Idle time histogram.
3831      *
3832      * Most of the time a switch has a relatively small number of subfacets.
3833      * When this is the case we might as well keep statistics for all of them
3834      * in userspace and to cache them in the kernel datapath for performance as
3835      * well.
3836      *
3837      * As the number of subfacets increases, the memory required to maintain
3838      * statistics about them in userspace and in the kernel becomes
3839      * significant.  However, with a large number of subfacets it is likely
3840      * that only a few of them are "heavy hitters" that consume a large amount
3841      * of bandwidth.  At this point, only heavy hitters are worth caching in
3842      * the kernel and maintaining in userspaces; other subfacets we can
3843      * discard.
3844      *
3845      * The technique used to compute the idle time is to build a histogram with
3846      * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each.  Each subfacet
3847      * that is installed in the kernel gets dropped in the appropriate bucket.
3848      * After the histogram has been built, we compute the cutoff so that only
3849      * the most-recently-used 1% of subfacets (but at least
3850      * flow_eviction_threshold flows) are kept cached.  At least
3851      * the most-recently-used bucket of subfacets is kept, so actually an
3852      * arbitrary number of subfacets can be kept in any given expiration run
3853      * (though the next run will delete most of those unless they receive
3854      * additional data).
3855      *
3856      * This requires a second pass through the subfacets, in addition to the
3857      * pass made by update_stats(), because the former function never looks at
3858      * uninstallable subfacets.
3859      */
3860     enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3861     enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3862     int buckets[N_BUCKETS] = { 0 };
3863     int total, subtotal, bucket;
3864     struct subfacet *subfacet;
3865     long long int now;
3866     int i;
3867
3868     total = hmap_count(&backer->subfacets);
3869     if (total <= flow_eviction_threshold) {
3870         return N_BUCKETS * BUCKET_WIDTH;
3871     }
3872
3873     /* Build histogram. */
3874     now = time_msec();
3875     HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3876         long long int idle = now - subfacet->used;
3877         int bucket = (idle <= 0 ? 0
3878                       : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3879                       : (unsigned int) idle / BUCKET_WIDTH);
3880         buckets[bucket]++;
3881     }
3882
3883     /* Find the first bucket whose flows should be expired. */
3884     subtotal = bucket = 0;
3885     do {
3886         subtotal += buckets[bucket++];
3887     } while (bucket < N_BUCKETS &&
3888              subtotal < MAX(flow_eviction_threshold, total / 100));
3889
3890     if (VLOG_IS_DBG_ENABLED()) {
3891         struct ds s;
3892
3893         ds_init(&s);
3894         ds_put_cstr(&s, "keep");
3895         for (i = 0; i < N_BUCKETS; i++) {
3896             if (i == bucket) {
3897                 ds_put_cstr(&s, ", drop");
3898             }
3899             if (buckets[i]) {
3900                 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3901             }
3902         }
3903         VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3904         ds_destroy(&s);
3905     }
3906
3907     return bucket * BUCKET_WIDTH;
3908 }
3909
3910 static void
3911 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3912 {
3913     /* Cutoff time for most flows. */
3914     long long int normal_cutoff = time_msec() - dp_max_idle;
3915
3916     /* We really want to keep flows for special protocols around, so use a more
3917      * conservative cutoff. */
3918     long long int special_cutoff = time_msec() - 10000;
3919
3920     struct subfacet *subfacet, *next_subfacet;
3921     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3922     int n_batch;
3923
3924     n_batch = 0;
3925     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3926                         &backer->subfacets) {
3927         long long int cutoff;
3928
3929         cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3930                                                 | SLOW_STP)
3931                   ? special_cutoff
3932                   : normal_cutoff);
3933         if (subfacet->used < cutoff) {
3934             if (subfacet->path != SF_NOT_INSTALLED) {
3935                 batch[n_batch++] = subfacet;
3936                 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3937                     subfacet_destroy_batch(backer, batch, n_batch);
3938                     n_batch = 0;
3939                 }
3940             } else {
3941                 subfacet_destroy(subfacet);
3942             }
3943         }
3944     }
3945
3946     if (n_batch > 0) {
3947         subfacet_destroy_batch(backer, batch, n_batch);
3948     }
3949 }
3950
3951 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3952  * then delete it entirely. */
3953 static void
3954 rule_expire(struct rule_dpif *rule)
3955 {
3956     uint16_t idle_timeout, hard_timeout;
3957     long long int now;
3958     uint8_t reason;
3959
3960     if (rule->up.pending) {
3961         /* We'll have to expire it later. */
3962         return;
3963     }
3964
3965     ovs_mutex_lock(&rule->up.timeout_mutex);
3966     hard_timeout = rule->up.hard_timeout;
3967     idle_timeout = rule->up.idle_timeout;
3968     ovs_mutex_unlock(&rule->up.timeout_mutex);
3969
3970     /* Has 'rule' expired? */
3971     now = time_msec();
3972     if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3973         reason = OFPRR_HARD_TIMEOUT;
3974     } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3975         reason = OFPRR_IDLE_TIMEOUT;
3976     } else {
3977         return;
3978     }
3979
3980     if (!ovs_rwlock_trywrlock(&rule->up.evict)) {
3981         COVERAGE_INC(ofproto_dpif_expired);
3982
3983         /* Get rid of the rule. */
3984         ofproto_rule_expire(&rule->up, reason);
3985     }
3986 }
3987 \f
3988 /* Facets. */
3989
3990 /* Creates and returns a new facet based on 'miss'.
3991  *
3992  * The caller must already have determined that no facet with an identical
3993  * 'miss->flow' exists in 'miss->ofproto'.
3994  *
3995  * 'rule' and 'xout' must have been created based on 'miss'.
3996  *
3997  * 'facet'' statistics are initialized based on 'stats'.
3998  *
3999  * The facet will initially have no subfacets.  The caller should create (at
4000  * least) one subfacet with subfacet_create(). */
4001 static struct facet *
4002 facet_create(const struct flow_miss *miss)
4003 {
4004     struct ofproto_dpif *ofproto = miss->ofproto;
4005     struct facet *facet;
4006     struct match match;
4007
4008     facet = xzalloc(sizeof *facet);
4009     facet->ofproto = miss->ofproto;
4010     facet->used = miss->stats.used;
4011     facet->flow = miss->flow;
4012     facet->learn_rl = time_msec() + 500;
4013
4014     list_init(&facet->subfacets);
4015     netflow_flow_init(&facet->nf_flow);
4016     netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4017
4018     xlate_out_copy(&facet->xout, &miss->xout);
4019
4020     match_init(&match, &facet->flow, &facet->xout.wc);
4021     cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4022     ovs_rwlock_wrlock(&ofproto->facets.rwlock);
4023     classifier_insert(&ofproto->facets, &facet->cr);
4024     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4025
4026     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4027     return facet;
4028 }
4029
4030 static void
4031 facet_free(struct facet *facet)
4032 {
4033     if (facet) {
4034         xlate_out_uninit(&facet->xout);
4035         free(facet);
4036     }
4037 }
4038
4039 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4040  * 'packet', which arrived on 'in_port'. */
4041 static bool
4042 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4043                     const struct nlattr *odp_actions, size_t actions_len,
4044                     struct ofpbuf *packet)
4045 {
4046     struct odputil_keybuf keybuf;
4047     struct ofpbuf key;
4048     int error;
4049
4050     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4051     odp_flow_key_from_flow(&key, flow,
4052                            ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4053
4054     error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4055                          odp_actions, actions_len, packet);
4056     return !error;
4057 }
4058
4059 /* Remove 'facet' from its ofproto and free up the associated memory:
4060  *
4061  *   - If 'facet' was installed in the datapath, uninstalls it and updates its
4062  *     rule's statistics, via subfacet_uninstall().
4063  *
4064  *   - Removes 'facet' from its rule and from ofproto->facets.
4065  */
4066 static void
4067 facet_remove(struct facet *facet)
4068 {
4069     struct subfacet *subfacet, *next_subfacet;
4070
4071     ovs_assert(!list_is_empty(&facet->subfacets));
4072
4073     /* First uninstall all of the subfacets to get final statistics. */
4074     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4075         subfacet_uninstall(subfacet);
4076     }
4077
4078     /* Flush the final stats to the rule.
4079      *
4080      * This might require us to have at least one subfacet around so that we
4081      * can use its actions for accounting in facet_account(), which is why we
4082      * have uninstalled but not yet destroyed the subfacets. */
4083     facet_flush_stats(facet);
4084
4085     /* Now we're really all done so destroy everything. */
4086     LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4087                         &facet->subfacets) {
4088         subfacet_destroy__(subfacet);
4089     }
4090     ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4091     classifier_remove(&facet->ofproto->facets, &facet->cr);
4092     ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4093     cls_rule_destroy(&facet->cr);
4094     facet_free(facet);
4095 }
4096
4097 /* Feed information from 'facet' back into the learning table to keep it in
4098  * sync with what is actually flowing through the datapath. */
4099 static void
4100 facet_learn(struct facet *facet)
4101 {
4102     long long int now = time_msec();
4103
4104     if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4105         return;
4106     }
4107
4108     facet->learn_rl = now + 500;
4109
4110     if (!facet->xout.has_learn
4111         && !facet->xout.has_normal
4112         && (!facet->xout.has_fin_timeout
4113             || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4114         return;
4115     }
4116
4117     facet_push_stats(facet, true);
4118 }
4119
4120 static void
4121 facet_account(struct facet *facet)
4122 {
4123     const struct nlattr *a;
4124     unsigned int left;
4125     ovs_be16 vlan_tci;
4126     uint64_t n_bytes;
4127
4128     if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4129         return;
4130     }
4131     n_bytes = facet->byte_count - facet->accounted_bytes;
4132
4133     /* This loop feeds byte counters to bond_account() for rebalancing to use
4134      * as a basis.  We also need to track the actual VLAN on which the packet
4135      * is going to be sent to ensure that it matches the one passed to
4136      * bond_choose_output_slave().  (Otherwise, we will account to the wrong
4137      * hash bucket.)
4138      *
4139      * We use the actions from an arbitrary subfacet because they should all
4140      * be equally valid for our purpose. */
4141     vlan_tci = facet->flow.vlan_tci;
4142     NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4143                              facet->xout.odp_actions.size) {
4144         const struct ovs_action_push_vlan *vlan;
4145         struct ofport_dpif *port;
4146
4147         switch (nl_attr_type(a)) {
4148         case OVS_ACTION_ATTR_OUTPUT:
4149             port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4150             if (port && port->bundle && port->bundle->bond) {
4151                 bond_account(port->bundle->bond, &facet->flow,
4152                              vlan_tci_to_vid(vlan_tci), n_bytes);
4153             }
4154             break;
4155
4156         case OVS_ACTION_ATTR_POP_VLAN:
4157             vlan_tci = htons(0);
4158             break;
4159
4160         case OVS_ACTION_ATTR_PUSH_VLAN:
4161             vlan = nl_attr_get(a);
4162             vlan_tci = vlan->vlan_tci;
4163             break;
4164         }
4165     }
4166 }
4167
4168 /* Returns true if the only action for 'facet' is to send to the controller.
4169  * (We don't report NetFlow expiration messages for such facets because they
4170  * are just part of the control logic for the network, not real traffic). */
4171 static bool
4172 facet_is_controller_flow(struct facet *facet)
4173 {
4174     if (facet) {
4175         struct ofproto_dpif *ofproto = facet->ofproto;
4176         const struct ofpact *ofpacts;
4177         struct rule_dpif *rule;
4178         size_t ofpacts_len;
4179         bool is_controller;
4180
4181         rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4182         ofpacts_len = rule->up.ofpacts_len;
4183         ofpacts = rule->up.ofpacts;
4184         is_controller = ofpacts_len > 0
4185             && ofpacts->type == OFPACT_CONTROLLER
4186             && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4187         rule_release(rule);
4188         return is_controller;
4189     }
4190     return false;
4191 }
4192
4193 /* Folds all of 'facet''s statistics into its rule.  Also updates the
4194  * accounting ofhook and emits a NetFlow expiration if appropriate.  All of
4195  * 'facet''s statistics in the datapath should have been zeroed and folded into
4196  * its packet and byte counts before this function is called. */
4197 static void
4198 facet_flush_stats(struct facet *facet)
4199 {
4200     struct ofproto_dpif *ofproto = facet->ofproto;
4201     struct subfacet *subfacet;
4202
4203     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4204         ovs_assert(!subfacet->dp_byte_count);
4205         ovs_assert(!subfacet->dp_packet_count);
4206     }
4207
4208     facet_push_stats(facet, false);
4209     if (facet->accounted_bytes < facet->byte_count) {
4210         facet_account(facet);
4211         facet->accounted_bytes = facet->byte_count;
4212     }
4213
4214     if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4215         struct ofexpired expired;
4216         expired.flow = facet->flow;
4217         expired.packet_count = facet->packet_count;
4218         expired.byte_count = facet->byte_count;
4219         expired.used = facet->used;
4220         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4221     }
4222
4223     /* Reset counters to prevent double counting if 'facet' ever gets
4224      * reinstalled. */
4225     facet_reset_counters(facet);
4226
4227     netflow_flow_clear(&facet->nf_flow);
4228     facet->tcp_flags = 0;
4229 }
4230
4231 /* Searches 'ofproto''s table of facets for one which would be responsible for
4232  * 'flow'.  Returns it if found, otherwise a null pointer.
4233  *
4234  * The returned facet might need revalidation; use facet_lookup_valid()
4235  * instead if that is important. */
4236 static struct facet *
4237 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4238 {
4239     struct cls_rule *cr;
4240
4241     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4242     cr = classifier_lookup(&ofproto->facets, flow, NULL);
4243     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4244     return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4245 }
4246
4247 /* Searches 'ofproto''s table of facets for one capable that covers
4248  * 'flow'.  Returns it if found, otherwise a null pointer.
4249  *
4250  * The returned facet is guaranteed to be valid. */
4251 static struct facet *
4252 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4253 {
4254     struct facet *facet;
4255
4256     facet = facet_find(ofproto, flow);
4257     if (facet
4258         && ofproto->backer->need_revalidate
4259         && !facet_revalidate(facet)) {
4260         return NULL;
4261     }
4262
4263     return facet;
4264 }
4265
4266 static bool
4267 facet_check_consistency(struct facet *facet)
4268 {
4269     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4270
4271     struct xlate_out xout;
4272     struct xlate_in xin;
4273
4274     struct rule_dpif *rule;
4275     bool ok;
4276
4277     /* Check the datapath actions for consistency. */
4278     rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4279     xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4280     xlate_actions(&xin, &xout);
4281     rule_release(rule);
4282
4283     ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4284         && facet->xout.slow == xout.slow;
4285     if (!ok && !VLOG_DROP_WARN(&rl)) {
4286         struct ds s = DS_EMPTY_INITIALIZER;
4287
4288         flow_format(&s, &facet->flow);
4289         ds_put_cstr(&s, ": inconsistency in facet");
4290
4291         if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4292             ds_put_cstr(&s, " (actions were: ");
4293             format_odp_actions(&s, facet->xout.odp_actions.data,
4294                                facet->xout.odp_actions.size);
4295             ds_put_cstr(&s, ") (correct actions: ");
4296             format_odp_actions(&s, xout.odp_actions.data,
4297                                xout.odp_actions.size);
4298             ds_put_char(&s, ')');
4299         }
4300
4301         if (facet->xout.slow != xout.slow) {
4302             ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4303         }
4304
4305         ds_destroy(&s);
4306     }
4307     xlate_out_uninit(&xout);
4308
4309     return ok;
4310 }
4311
4312 /* Re-searches the classifier for 'facet':
4313  *
4314  *   - If the rule found is different from 'facet''s current rule, moves
4315  *     'facet' to the new rule and recompiles its actions.
4316  *
4317  *   - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4318  *     where it is and recompiles its actions anyway.
4319  *
4320  *   - If any of 'facet''s subfacets correspond to a new flow according to
4321  *     xlate_receive(), 'facet' is removed.
4322  *
4323  *   Returns true if 'facet' is still valid.  False if 'facet' was removed. */
4324 static bool
4325 facet_revalidate(struct facet *facet)
4326 {
4327     struct ofproto_dpif *ofproto = facet->ofproto;
4328     struct rule_dpif *new_rule;
4329     struct subfacet *subfacet;
4330     struct flow_wildcards wc;
4331     struct xlate_out xout;
4332     struct xlate_in xin;
4333
4334     COVERAGE_INC(facet_revalidate);
4335
4336     /* Check that child subfacets still correspond to this facet.  Tunnel
4337      * configuration changes could cause a subfacet's OpenFlow in_port to
4338      * change. */
4339     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4340         struct ofproto_dpif *recv_ofproto;
4341         struct flow recv_flow;
4342         int error;
4343
4344         error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4345                               subfacet->key_len, &recv_flow, NULL,
4346                               &recv_ofproto, NULL);
4347         if (error
4348             || recv_ofproto != ofproto
4349             || facet != facet_find(ofproto, &recv_flow)) {
4350             facet_remove(facet);
4351             return false;
4352         }
4353     }
4354
4355     flow_wildcards_init_catchall(&wc);
4356     rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4357
4358     /* Calculate new datapath actions.
4359      *
4360      * We do not modify any 'facet' state yet, because we might need to, e.g.,
4361      * emit a NetFlow expiration and, if so, we need to have the old state
4362      * around to properly compose it. */
4363     xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4364     xlate_actions(&xin, &xout);
4365     flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4366
4367     /* A facet's slow path reason should only change under dramatic
4368      * circumstances.  Rather than try to update everything, it's simpler to
4369      * remove the facet and start over.
4370      *
4371      * More importantly, if a facet's wildcards change, it will be relatively
4372      * difficult to figure out if its subfacets still belong to it, and if not
4373      * which facet they may belong to.  Again, to avoid the complexity, we
4374      * simply give up instead. */
4375     if (facet->xout.slow != xout.slow
4376         || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4377         facet_remove(facet);
4378         xlate_out_uninit(&xout);
4379         rule_release(new_rule);
4380         return false;
4381     }
4382
4383     if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4384         LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4385             if (subfacet->path == SF_FAST_PATH) {
4386                 struct dpif_flow_stats stats;
4387
4388                 subfacet_install(subfacet, &xout.odp_actions, &stats);
4389                 subfacet_update_stats(subfacet, &stats);
4390             }
4391         }
4392
4393         facet_flush_stats(facet);
4394
4395         ofpbuf_clear(&facet->xout.odp_actions);
4396         ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4397                    xout.odp_actions.size);
4398     }
4399
4400     /* Update 'facet' now that we've taken care of all the old state. */
4401     facet->xout.slow = xout.slow;
4402     facet->xout.has_learn = xout.has_learn;
4403     facet->xout.has_normal = xout.has_normal;
4404     facet->xout.has_fin_timeout = xout.has_fin_timeout;
4405     facet->xout.nf_output_iface = xout.nf_output_iface;
4406     facet->xout.mirrors = xout.mirrors;
4407     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4408     facet->used = MAX(facet->used, new_rule->up.created);
4409
4410     xlate_out_uninit(&xout);
4411     rule_release(new_rule);
4412     return true;
4413 }
4414
4415 static void
4416 facet_reset_counters(struct facet *facet)
4417 {
4418     facet->packet_count = 0;
4419     facet->byte_count = 0;
4420     facet->prev_packet_count = 0;
4421     facet->prev_byte_count = 0;
4422     facet->accounted_bytes = 0;
4423 }
4424
4425 static void
4426 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4427                 struct dpif_flow_stats *stats, bool may_learn)
4428 {
4429     struct ofport_dpif *in_port;
4430     struct rule_dpif *rule;
4431     struct xlate_in xin;
4432
4433     in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4434     if (in_port && in_port->is_tunnel) {
4435         netdev_vport_inc_rx(in_port->up.netdev, stats);
4436     }
4437
4438     rule_dpif_lookup(ofproto, flow, NULL, &rule);
4439     rule_credit_stats(rule, stats);
4440     xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4441     xin.resubmit_stats = stats;
4442     xin.may_learn = may_learn;
4443     xlate_actions_for_side_effects(&xin);
4444     rule_release(rule);
4445 }
4446
4447 static void
4448 facet_push_stats(struct facet *facet, bool may_learn)
4449 {
4450     struct dpif_flow_stats stats;
4451
4452     ovs_assert(facet->packet_count >= facet->prev_packet_count);
4453     ovs_assert(facet->byte_count >= facet->prev_byte_count);
4454     ovs_assert(facet->used >= facet->prev_used);
4455
4456     stats.n_packets = facet->packet_count - facet->prev_packet_count;
4457     stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4458     stats.used = facet->used;
4459     stats.tcp_flags = facet->tcp_flags;
4460
4461     if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4462         facet->prev_packet_count = facet->packet_count;
4463         facet->prev_byte_count = facet->byte_count;
4464         facet->prev_used = facet->used;
4465
4466         netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4467                                  facet->used);
4468         netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4469         mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4470                             stats.n_packets, stats.n_bytes);
4471         flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4472     }
4473 }
4474
4475 static void
4476 push_all_stats__(bool run_fast)
4477 {
4478     static long long int rl = LLONG_MIN;
4479     struct ofproto_dpif *ofproto;
4480
4481     if (time_msec() < rl) {
4482         return;
4483     }
4484
4485     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4486         struct cls_cursor cursor;
4487         struct facet *facet;
4488
4489         ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4490         cls_cursor_init(&cursor, &ofproto->facets, NULL);
4491         CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4492             facet_push_stats(facet, false);
4493             if (run_fast) {
4494                 run_fast_rl();
4495             }
4496         }
4497         ovs_rwlock_unlock(&ofproto->facets.rwlock);
4498     }
4499
4500     rl = time_msec() + 100;
4501 }
4502
4503 static void
4504 push_all_stats(void)
4505 {
4506     push_all_stats__(true);
4507 }
4508
4509 void
4510 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4511 {
4512     ovs_mutex_lock(&rule->stats_mutex);
4513     rule->packet_count += stats->n_packets;
4514     rule->byte_count += stats->n_bytes;
4515     ofproto_rule_update_used(&rule->up, stats->used);
4516     ovs_mutex_unlock(&rule->stats_mutex);
4517 }
4518 \f
4519 /* Subfacets. */
4520
4521 static struct subfacet *
4522 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4523               size_t key_len, uint32_t key_hash)
4524 {
4525     struct subfacet *subfacet;
4526
4527     HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4528                              &backer->subfacets) {
4529         if (subfacet->key_len == key_len
4530             && !memcmp(key, subfacet->key, key_len)) {
4531             return subfacet;
4532         }
4533     }
4534
4535     return NULL;
4536 }
4537
4538 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4539  * 'key_fitness', 'key', and 'key_len' members in 'miss'.  Returns the
4540  * existing subfacet if there is one, otherwise creates and returns a
4541  * new subfacet. */
4542 static struct subfacet *
4543 subfacet_create(struct facet *facet, struct flow_miss *miss)
4544 {
4545     struct dpif_backer *backer = miss->ofproto->backer;
4546     enum odp_key_fitness key_fitness = miss->key_fitness;
4547     const struct nlattr *key = miss->key;
4548     size_t key_len = miss->key_len;
4549     uint32_t key_hash;
4550     struct subfacet *subfacet;
4551
4552     key_hash = odp_flow_key_hash(key, key_len);
4553
4554     if (list_is_empty(&facet->subfacets)) {
4555         subfacet = &facet->one_subfacet;
4556     } else {
4557         subfacet = subfacet_find(backer, key, key_len, key_hash);
4558         if (subfacet) {
4559             if (subfacet->facet == facet) {
4560                 return subfacet;
4561             }
4562
4563             /* This shouldn't happen. */
4564             VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4565             subfacet_destroy(subfacet);
4566         }
4567
4568         subfacet = xmalloc(sizeof *subfacet);
4569     }
4570
4571     hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4572     list_push_back(&facet->subfacets, &subfacet->list_node);
4573     subfacet->facet = facet;
4574     subfacet->key_fitness = key_fitness;
4575     subfacet->key = xmemdup(key, key_len);
4576     subfacet->key_len = key_len;
4577     subfacet->used = miss->stats.used;
4578     subfacet->created = subfacet->used;
4579     subfacet->dp_packet_count = 0;
4580     subfacet->dp_byte_count = 0;
4581     subfacet->path = SF_NOT_INSTALLED;
4582     subfacet->backer = backer;
4583
4584     backer->subfacet_add_count++;
4585     return subfacet;
4586 }
4587
4588 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4589  * its facet within 'ofproto', and frees it. */
4590 static void
4591 subfacet_destroy__(struct subfacet *subfacet)
4592 {
4593     struct facet *facet = subfacet->facet;
4594     struct ofproto_dpif *ofproto = facet->ofproto;
4595
4596     /* Update ofproto stats before uninstall the subfacet. */
4597     ofproto->backer->subfacet_del_count++;
4598
4599     subfacet_uninstall(subfacet);
4600     hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4601     list_remove(&subfacet->list_node);
4602     free(subfacet->key);
4603     if (subfacet != &facet->one_subfacet) {
4604         free(subfacet);
4605     }
4606 }
4607
4608 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4609  * last remaining subfacet in its facet destroys the facet too. */
4610 static void
4611 subfacet_destroy(struct subfacet *subfacet)
4612 {
4613     struct facet *facet = subfacet->facet;
4614
4615     if (list_is_singleton(&facet->subfacets)) {
4616         /* facet_remove() needs at least one subfacet (it will remove it). */
4617         facet_remove(facet);
4618     } else {
4619         subfacet_destroy__(subfacet);
4620     }
4621 }
4622
4623 static void
4624 subfacet_destroy_batch(struct dpif_backer *backer,
4625                        struct subfacet **subfacets, int n)
4626 {
4627     struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4628     struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4629     struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4630     int i;
4631
4632     for (i = 0; i < n; i++) {
4633         ops[i].type = DPIF_OP_FLOW_DEL;
4634         ops[i].u.flow_del.key = subfacets[i]->key;
4635         ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4636         ops[i].u.flow_del.stats = &stats[i];
4637         opsp[i] = &ops[i];
4638     }
4639
4640     dpif_operate(backer->dpif, opsp, n);
4641     for (i = 0; i < n; i++) {
4642         subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4643         subfacets[i]->path = SF_NOT_INSTALLED;
4644         subfacet_destroy(subfacets[i]);
4645         run_fast_rl();
4646     }
4647 }
4648
4649 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4650  * bytes of actions in 'actions'.  If 'stats' is non-null, statistics counters
4651  * in the datapath will be zeroed and 'stats' will be updated with traffic new
4652  * since 'subfacet' was last updated.
4653  *
4654  * Returns 0 if successful, otherwise a positive errno value. */
4655 static int
4656 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4657                  struct dpif_flow_stats *stats)
4658 {
4659     struct facet *facet = subfacet->facet;
4660     enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4661     const struct nlattr *actions = odp_actions->data;
4662     size_t actions_len = odp_actions->size;
4663     struct odputil_keybuf maskbuf;
4664     struct ofpbuf mask;
4665
4666     uint64_t slow_path_stub[128 / 8];
4667     enum dpif_flow_put_flags flags;
4668     int ret;
4669
4670     flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4671                                                : DPIF_FP_MODIFY;
4672     if (stats) {
4673         flags |= DPIF_FP_ZERO_STATS;
4674     }
4675
4676     if (path == SF_SLOW_PATH) {
4677         compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4678                           slow_path_stub, sizeof slow_path_stub,
4679                           &actions, &actions_len);
4680     }
4681
4682     ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4683     if (enable_megaflows) {
4684         odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4685                                &facet->flow, UINT32_MAX);
4686     }
4687
4688     ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4689                         subfacet->key_len,  mask.data, mask.size,
4690                         actions, actions_len, stats);
4691
4692     if (stats) {
4693         subfacet_reset_dp_stats(subfacet, stats);
4694     }
4695
4696     if (ret) {
4697         COVERAGE_INC(subfacet_install_fail);
4698     } else {
4699         subfacet->path = path;
4700     }
4701     return ret;
4702 }
4703
4704 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4705 static void
4706 subfacet_uninstall(struct subfacet *subfacet)
4707 {
4708     if (subfacet->path != SF_NOT_INSTALLED) {
4709         struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4710         struct dpif_flow_stats stats;
4711         int error;
4712
4713         error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4714                               subfacet->key_len, &stats);
4715         subfacet_reset_dp_stats(subfacet, &stats);
4716         if (!error) {
4717             subfacet_update_stats(subfacet, &stats);
4718         }
4719         subfacet->path = SF_NOT_INSTALLED;
4720     } else {
4721         ovs_assert(subfacet->dp_packet_count == 0);
4722         ovs_assert(subfacet->dp_byte_count == 0);
4723     }
4724 }
4725
4726 /* Resets 'subfacet''s datapath statistics counters.  This should be called
4727  * when 'subfacet''s statistics are cleared in the datapath.  If 'stats' is
4728  * non-null, it should contain the statistics returned by dpif when 'subfacet'
4729  * was reset in the datapath.  'stats' will be modified to include only
4730  * statistics new since 'subfacet' was last updated. */
4731 static void
4732 subfacet_reset_dp_stats(struct subfacet *subfacet,
4733                         struct dpif_flow_stats *stats)
4734 {
4735     if (stats
4736         && subfacet->dp_packet_count <= stats->n_packets
4737         && subfacet->dp_byte_count <= stats->n_bytes) {
4738         stats->n_packets -= subfacet->dp_packet_count;
4739         stats->n_bytes -= subfacet->dp_byte_count;
4740     }
4741
4742     subfacet->dp_packet_count = 0;
4743     subfacet->dp_byte_count = 0;
4744 }
4745
4746 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4747  *
4748  * Because of the meaning of a subfacet's counters, it only makes sense to do
4749  * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4750  * represents a packet that was sent by hand or if it represents statistics
4751  * that have been cleared out of the datapath. */
4752 static void
4753 subfacet_update_stats(struct subfacet *subfacet,
4754                       const struct dpif_flow_stats *stats)
4755 {
4756     if (stats->n_packets || stats->used > subfacet->used) {
4757         struct facet *facet = subfacet->facet;
4758
4759         subfacet->used = MAX(subfacet->used, stats->used);
4760         facet->used = MAX(facet->used, stats->used);
4761         facet->packet_count += stats->n_packets;
4762         facet->byte_count += stats->n_bytes;
4763         facet->tcp_flags |= stats->tcp_flags;
4764     }
4765 }
4766 \f
4767 /* Rules. */
4768
4769 /* Lookup 'flow' in 'ofproto''s classifier.  If 'wc' is non-null, sets
4770  * the fields that were relevant as part of the lookup. */
4771 void
4772 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4773                  struct flow_wildcards *wc, struct rule_dpif **rule)
4774 {
4775     struct ofport_dpif *port;
4776
4777     if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4778         return;
4779     }
4780     port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4781     if (!port) {
4782         VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4783                      flow->in_port.ofp_port);
4784     }
4785
4786     *rule = choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4787                              ofproto->no_packet_in_rule);
4788     ovs_rwlock_rdlock(&(*rule)->up.evict);
4789 }
4790
4791 bool
4792 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4793                           const struct flow *flow, struct flow_wildcards *wc,
4794                           uint8_t table_id, struct rule_dpif **rule)
4795     OVS_ACQ_RDLOCK((*rule)->up.evict)
4796 {
4797     struct cls_rule *cls_rule;
4798     struct classifier *cls;
4799     bool frag;
4800
4801     *rule = NULL;
4802     if (table_id >= N_TABLES) {
4803         return false;
4804     }
4805
4806     if (wc) {
4807         memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4808         wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4809     }
4810
4811     cls = &ofproto->up.tables[table_id].cls;
4812     ovs_rwlock_rdlock(&cls->rwlock);
4813     frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4814     if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4815         /* We must pretend that transport ports are unavailable. */
4816         struct flow ofpc_normal_flow = *flow;
4817         ofpc_normal_flow.tp_src = htons(0);
4818         ofpc_normal_flow.tp_dst = htons(0);
4819         cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4820     } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4821         cls_rule = &ofproto->drop_frags_rule->up.cr;
4822         if (wc) {
4823             flow_wildcards_init_exact(wc);
4824         }
4825     } else {
4826         cls_rule = classifier_lookup(cls, flow, wc);
4827     }
4828
4829     *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4830     if (*rule && ovs_rwlock_tryrdlock(&(*rule)->up.evict)) {
4831         /* The rule is in the process of being removed.  Best we can do is
4832          * pretend it isn't there. */
4833         *rule = NULL;
4834     }
4835     ovs_rwlock_unlock(&cls->rwlock);
4836
4837     return *rule != NULL;
4838 }
4839
4840 /* Given a port configuration (specified as zero if there's no port), chooses
4841  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4842  * flow table miss. */
4843 struct rule_dpif *
4844 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4845                  struct rule_dpif *no_packet_in_rule)
4846 {
4847     return config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4848 }
4849
4850 void
4851 rule_release(struct rule_dpif *rule)
4852 {
4853     if (rule) {
4854         ovs_rwlock_unlock(&rule->up.evict);
4855     }
4856 }
4857
4858 static void
4859 complete_operation(struct rule_dpif *rule)
4860 {
4861     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4862
4863     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4864     if (clogged) {
4865         struct dpif_completion *c = xmalloc(sizeof *c);
4866         c->op = rule->up.pending;
4867         list_push_back(&ofproto->completions, &c->list_node);
4868     } else {
4869         ofoperation_complete(rule->up.pending, 0);
4870     }
4871 }
4872
4873 static struct rule *
4874 rule_alloc(void)
4875 {
4876     struct rule_dpif *rule = xmalloc(sizeof *rule);
4877     return &rule->up;
4878 }
4879
4880 static void
4881 rule_dealloc(struct rule *rule_)
4882 {
4883     struct rule_dpif *rule = rule_dpif_cast(rule_);
4884     free(rule);
4885 }
4886
4887 static enum ofperr
4888 rule_construct(struct rule *rule_)
4889 {
4890     struct rule_dpif *rule = rule_dpif_cast(rule_);
4891     ovs_mutex_init(&rule->stats_mutex);
4892     ovs_mutex_lock(&rule->stats_mutex);
4893     rule->packet_count = 0;
4894     rule->byte_count = 0;
4895     ovs_mutex_unlock(&rule->stats_mutex);
4896     complete_operation(rule);
4897     return 0;
4898 }
4899
4900 static void
4901 rule_destruct(struct rule *rule_)
4902 {
4903     struct rule_dpif *rule = rule_dpif_cast(rule_);
4904     complete_operation(rule);
4905     ovs_mutex_destroy(&rule->stats_mutex);
4906 }
4907
4908 static void
4909 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4910 {
4911     struct rule_dpif *rule = rule_dpif_cast(rule_);
4912
4913     /* push_all_stats() can handle flow misses which, when using the learn
4914      * action, can cause rules to be added and deleted.  This can corrupt our
4915      * caller's datastructures which assume that rule_get_stats() doesn't have
4916      * an impact on the flow table. To be safe, we disable miss handling. */
4917     push_all_stats__(false);
4918
4919     /* Start from historical data for 'rule' itself that are no longer tracked
4920      * in facets.  This counts, for example, facets that have expired. */
4921     ovs_mutex_lock(&rule->stats_mutex);
4922     *packets = rule->packet_count;
4923     *bytes = rule->byte_count;
4924     ovs_mutex_unlock(&rule->stats_mutex);
4925 }
4926
4927 static void
4928 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4929                   struct ofpbuf *packet)
4930 {
4931     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4932     struct dpif_flow_stats stats;
4933     struct xlate_out xout;
4934     struct xlate_in xin;
4935
4936     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4937     rule_credit_stats(rule, &stats);
4938
4939     xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4940     xin.resubmit_stats = &stats;
4941     xlate_actions(&xin, &xout);
4942
4943     execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4944                         xout.odp_actions.size, packet);
4945
4946     xlate_out_uninit(&xout);
4947 }
4948
4949 static enum ofperr
4950 rule_execute(struct rule *rule, const struct flow *flow,
4951              struct ofpbuf *packet)
4952 {
4953     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4954     ofpbuf_delete(packet);
4955     return 0;
4956 }
4957
4958 static void
4959 rule_modify_actions(struct rule *rule_)
4960 {
4961     struct rule_dpif *rule = rule_dpif_cast(rule_);
4962
4963     complete_operation(rule);
4964 }
4965 \f
4966 /* Sends 'packet' out 'ofport'.
4967  * May modify 'packet'.
4968  * Returns 0 if successful, otherwise a positive errno value. */
4969 static int
4970 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4971 {
4972     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4973     uint64_t odp_actions_stub[1024 / 8];
4974     struct ofpbuf key, odp_actions;
4975     struct dpif_flow_stats stats;
4976     struct odputil_keybuf keybuf;
4977     struct ofpact_output output;
4978     struct xlate_out xout;
4979     struct xlate_in xin;
4980     struct flow flow;
4981     union flow_in_port in_port_;
4982     int error;
4983
4984     ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4985     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4986
4987     /* Use OFPP_NONE as the in_port to avoid special packet processing. */
4988     in_port_.ofp_port = OFPP_NONE;
4989     flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
4990     odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
4991                                                              OFPP_LOCAL));
4992     dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
4993
4994     ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
4995     output.port = ofport->up.ofp_port;
4996     output.max_len = 0;
4997
4998     xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
4999     xin.ofpacts_len = sizeof output;
5000     xin.ofpacts = &output.ofpact;
5001     xin.resubmit_stats = &stats;
5002     xlate_actions(&xin, &xout);
5003
5004     error = dpif_execute(ofproto->backer->dpif,
5005                          key.data, key.size,
5006                          xout.odp_actions.data, xout.odp_actions.size,
5007                          packet);
5008     xlate_out_uninit(&xout);
5009
5010     if (error) {
5011         VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5012                      ofproto->up.name, netdev_get_name(ofport->up.netdev),
5013                      ovs_strerror(error));
5014     }
5015
5016     ofproto->stats.tx_packets++;
5017     ofproto->stats.tx_bytes += packet->size;
5018     return error;
5019 }
5020
5021 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5022  * The action will state 'slow' as the reason that the action is in the slow
5023  * path.  (This is purely informational: it allows a human viewing "ovs-dpctl
5024  * dump-flows" output to see why a flow is in the slow path.)
5025  *
5026  * The 'stub_size' bytes in 'stub' will be used to store the action.
5027  * 'stub_size' must be large enough for the action.
5028  *
5029  * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5030  * respectively. */
5031 static void
5032 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5033                   enum slow_path_reason slow,
5034                   uint64_t *stub, size_t stub_size,
5035                   const struct nlattr **actionsp, size_t *actions_lenp)
5036 {
5037     union user_action_cookie cookie;
5038     struct ofpbuf buf;
5039
5040     cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5041     cookie.slow_path.unused = 0;
5042     cookie.slow_path.reason = slow;
5043
5044     ofpbuf_use_stack(&buf, stub, stub_size);
5045     if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5046         uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5047                                          ODPP_NONE);
5048         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5049     } else {
5050         odp_port_t odp_port;
5051         uint32_t pid;
5052
5053         odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5054         pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5055         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5056     }
5057     *actionsp = buf.data;
5058     *actions_lenp = buf.size;
5059 }
5060 \f
5061 static bool
5062 set_frag_handling(struct ofproto *ofproto_,
5063                   enum ofp_config_flags frag_handling)
5064 {
5065     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5066     if (frag_handling != OFPC_FRAG_REASM) {
5067         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5068         return true;
5069     } else {
5070         return false;
5071     }
5072 }
5073
5074 static enum ofperr
5075 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5076            const struct flow *flow,
5077            const struct ofpact *ofpacts, size_t ofpacts_len)
5078 {
5079     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5080     struct odputil_keybuf keybuf;
5081     struct dpif_flow_stats stats;
5082     struct xlate_out xout;
5083     struct xlate_in xin;
5084     struct ofpbuf key;
5085
5086
5087     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5088     odp_flow_key_from_flow(&key, flow,
5089                            ofp_port_to_odp_port(ofproto,
5090                                       flow->in_port.ofp_port));
5091
5092     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5093
5094     xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5095     xin.resubmit_stats = &stats;
5096     xin.ofpacts_len = ofpacts_len;
5097     xin.ofpacts = ofpacts;
5098
5099     xlate_actions(&xin, &xout);
5100     dpif_execute(ofproto->backer->dpif, key.data, key.size,
5101                  xout.odp_actions.data, xout.odp_actions.size, packet);
5102     xlate_out_uninit(&xout);
5103
5104     return 0;
5105 }
5106 \f
5107 /* NetFlow. */
5108
5109 static int
5110 set_netflow(struct ofproto *ofproto_,
5111             const struct netflow_options *netflow_options)
5112 {
5113     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5114
5115     if (netflow_options) {
5116         if (!ofproto->netflow) {
5117             ofproto->netflow = netflow_create();
5118             ofproto->backer->need_revalidate = REV_RECONFIGURE;
5119         }
5120         return netflow_set_options(ofproto->netflow, netflow_options);
5121     } else if (ofproto->netflow) {
5122         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5123         netflow_destroy(ofproto->netflow);
5124         ofproto->netflow = NULL;
5125     }
5126
5127     return 0;
5128 }
5129
5130 static void
5131 get_netflow_ids(const struct ofproto *ofproto_,
5132                 uint8_t *engine_type, uint8_t *engine_id)
5133 {
5134     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5135
5136     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5137 }
5138
5139 static void
5140 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5141 {
5142     if (!facet_is_controller_flow(facet) &&
5143         netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5144         struct subfacet *subfacet;
5145         struct ofexpired expired;
5146
5147         LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5148             if (subfacet->path == SF_FAST_PATH) {
5149                 struct dpif_flow_stats stats;
5150
5151                 subfacet_install(subfacet, &facet->xout.odp_actions,
5152                                  &stats);
5153                 subfacet_update_stats(subfacet, &stats);
5154             }
5155         }
5156
5157         expired.flow = facet->flow;
5158         expired.packet_count = facet->packet_count;
5159         expired.byte_count = facet->byte_count;
5160         expired.used = facet->used;
5161         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5162     }
5163 }
5164
5165 static void
5166 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5167 {
5168     struct cls_cursor cursor;
5169     struct facet *facet;
5170
5171     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5172     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5173     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5174         send_active_timeout(ofproto, facet);
5175     }
5176     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5177 }
5178 \f
5179 static struct ofproto_dpif *
5180 ofproto_dpif_lookup(const char *name)
5181 {
5182     struct ofproto_dpif *ofproto;
5183
5184     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5185                              hash_string(name, 0), &all_ofproto_dpifs) {
5186         if (!strcmp(ofproto->up.name, name)) {
5187             return ofproto;
5188         }
5189     }
5190     return NULL;
5191 }
5192
5193 static void
5194 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5195                           const char *argv[], void *aux OVS_UNUSED)
5196 {
5197     struct ofproto_dpif *ofproto;
5198
5199     if (argc > 1) {
5200         ofproto = ofproto_dpif_lookup(argv[1]);
5201         if (!ofproto) {
5202             unixctl_command_reply_error(conn, "no such bridge");
5203             return;
5204         }
5205         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5206         mac_learning_flush(ofproto->ml);
5207         ovs_rwlock_unlock(&ofproto->ml->rwlock);
5208     } else {
5209         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5210             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5211             mac_learning_flush(ofproto->ml);
5212             ovs_rwlock_unlock(&ofproto->ml->rwlock);
5213         }
5214     }
5215
5216     unixctl_command_reply(conn, "table successfully flushed");
5217 }
5218
5219 static struct ofport_dpif *
5220 ofbundle_get_a_port(const struct ofbundle *bundle)
5221 {
5222     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5223                         bundle_node);
5224 }
5225
5226 static void
5227 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5228                          const char *argv[], void *aux OVS_UNUSED)
5229 {
5230     struct ds ds = DS_EMPTY_INITIALIZER;
5231     const struct ofproto_dpif *ofproto;
5232     const struct mac_entry *e;
5233
5234     ofproto = ofproto_dpif_lookup(argv[1]);
5235     if (!ofproto) {
5236         unixctl_command_reply_error(conn, "no such bridge");
5237         return;
5238     }
5239
5240     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
5241     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5242     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5243         struct ofbundle *bundle = e->port.p;
5244         char name[OFP_MAX_PORT_NAME_LEN];
5245
5246         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5247                                name, sizeof name);
5248         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
5249                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
5250                       mac_entry_age(ofproto->ml, e));
5251     }
5252     ovs_rwlock_unlock(&ofproto->ml->rwlock);
5253     unixctl_command_reply(conn, ds_cstr(&ds));
5254     ds_destroy(&ds);
5255 }
5256
5257 struct trace_ctx {
5258     struct xlate_out xout;
5259     struct xlate_in xin;
5260     struct flow flow;
5261     struct ds *result;
5262 };
5263
5264 static void
5265 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5266 {
5267     ds_put_char_multiple(result, '\t', level);
5268     if (!rule) {
5269         ds_put_cstr(result, "No match\n");
5270         return;
5271     }
5272
5273     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5274                   rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5275     cls_rule_format(&rule->up.cr, result);
5276     ds_put_char(result, '\n');
5277
5278     ds_put_char_multiple(result, '\t', level);
5279     ds_put_cstr(result, "OpenFlow ");
5280     ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5281     ds_put_char(result, '\n');
5282 }
5283
5284 static void
5285 trace_format_flow(struct ds *result, int level, const char *title,
5286                   struct trace_ctx *trace)
5287 {
5288     ds_put_char_multiple(result, '\t', level);
5289     ds_put_format(result, "%s: ", title);
5290     if (flow_equal(&trace->xin.flow, &trace->flow)) {
5291         ds_put_cstr(result, "unchanged");
5292     } else {
5293         flow_format(result, &trace->xin.flow);
5294         trace->flow = trace->xin.flow;
5295     }
5296     ds_put_char(result, '\n');
5297 }
5298
5299 static void
5300 trace_format_regs(struct ds *result, int level, const char *title,
5301                   struct trace_ctx *trace)
5302 {
5303     size_t i;
5304
5305     ds_put_char_multiple(result, '\t', level);
5306     ds_put_format(result, "%s:", title);
5307     for (i = 0; i < FLOW_N_REGS; i++) {
5308         ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5309     }
5310     ds_put_char(result, '\n');
5311 }
5312
5313 static void
5314 trace_format_odp(struct ds *result, int level, const char *title,
5315                  struct trace_ctx *trace)
5316 {
5317     struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5318
5319     ds_put_char_multiple(result, '\t', level);
5320     ds_put_format(result, "%s: ", title);
5321     format_odp_actions(result, odp_actions->data, odp_actions->size);
5322     ds_put_char(result, '\n');
5323 }
5324
5325 static void
5326 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5327 {
5328     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5329     struct ds *result = trace->result;
5330
5331     ds_put_char(result, '\n');
5332     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5333     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5334     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
5335     trace_format_rule(result, recurse + 1, rule);
5336 }
5337
5338 static void
5339 trace_report(struct xlate_in *xin, const char *s, int recurse)
5340 {
5341     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5342     struct ds *result = trace->result;
5343
5344     ds_put_char_multiple(result, '\t', recurse);
5345     ds_put_cstr(result, s);
5346     ds_put_char(result, '\n');
5347 }
5348
5349 static void
5350 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5351                       void *aux OVS_UNUSED)
5352 {
5353     const struct dpif_backer *backer;
5354     struct ofproto_dpif *ofproto;
5355     struct ofpbuf odp_key, odp_mask;
5356     struct ofpbuf *packet;
5357     struct ds result;
5358     struct flow flow;
5359     char *s;
5360
5361     packet = NULL;
5362     backer = NULL;
5363     ds_init(&result);
5364     ofpbuf_init(&odp_key, 0);
5365     ofpbuf_init(&odp_mask, 0);
5366
5367     /* Handle "-generate" or a hex string as the last argument. */
5368     if (!strcmp(argv[argc - 1], "-generate")) {
5369         packet = ofpbuf_new(0);
5370         argc--;
5371     } else {
5372         const char *error = eth_from_hex(argv[argc - 1], &packet);
5373         if (!error) {
5374             argc--;
5375         } else if (argc == 4) {
5376             /* The 3-argument form must end in "-generate' or a hex string. */
5377             unixctl_command_reply_error(conn, error);
5378             goto exit;
5379         }
5380     }
5381
5382     /* Parse the flow and determine whether a datapath or
5383      * bridge is specified. If function odp_flow_key_from_string()
5384      * returns 0, the flow is a odp_flow. If function
5385      * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5386     if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5387         /* If the odp_flow is the second argument,
5388          * the datapath name is the first argument. */
5389         if (argc == 3) {
5390             const char *dp_type;
5391             if (!strncmp(argv[1], "ovs-", 4)) {
5392                 dp_type = argv[1] + 4;
5393             } else {
5394                 dp_type = argv[1];
5395             }
5396             backer = shash_find_data(&all_dpif_backers, dp_type);
5397             if (!backer) {
5398                 unixctl_command_reply_error(conn, "Cannot find datapath "
5399                                "of this name");
5400                 goto exit;
5401             }
5402         } else {
5403             /* No datapath name specified, so there should be only one
5404              * datapath. */
5405             struct shash_node *node;
5406             if (shash_count(&all_dpif_backers) != 1) {
5407                 unixctl_command_reply_error(conn, "Must specify datapath "
5408                          "name, there is more than one type of datapath");
5409                 goto exit;
5410             }
5411             node = shash_first(&all_dpif_backers);
5412             backer = node->data;
5413         }
5414
5415         if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5416                           NULL, &ofproto, NULL)) {
5417             unixctl_command_reply_error(conn, "Invalid datapath flow");
5418             goto exit;
5419         }
5420         ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5421     } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5422         if (argc != 3) {
5423             unixctl_command_reply_error(conn, "Must specify bridge name");
5424             goto exit;
5425         }
5426
5427         ofproto = ofproto_dpif_lookup(argv[1]);
5428         if (!ofproto) {
5429             unixctl_command_reply_error(conn, "Unknown bridge name");
5430             goto exit;
5431         }
5432     } else {
5433         unixctl_command_reply_error(conn, "Bad flow syntax");
5434         goto exit;
5435     }
5436
5437     /* Generate a packet, if requested. */
5438     if (packet) {
5439         if (!packet->size) {
5440             flow_compose(packet, &flow);
5441         } else {
5442             union flow_in_port in_port_;
5443
5444             in_port_ = flow.in_port;
5445             ds_put_cstr(&result, "Packet: ");
5446             s = ofp_packet_to_string(packet->data, packet->size);
5447             ds_put_cstr(&result, s);
5448             free(s);
5449
5450             /* Use the metadata from the flow and the packet argument
5451              * to reconstruct the flow. */
5452             flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5453                          &in_port_, &flow);
5454         }
5455     }
5456
5457     ofproto_trace(ofproto, &flow, packet, &result);
5458     unixctl_command_reply(conn, ds_cstr(&result));
5459
5460 exit:
5461     ds_destroy(&result);
5462     ofpbuf_delete(packet);
5463     ofpbuf_uninit(&odp_key);
5464     ofpbuf_uninit(&odp_mask);
5465 }
5466
5467 static void
5468 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5469               const struct ofpbuf *packet, struct ds *ds)
5470 {
5471     struct rule_dpif *rule;
5472     struct flow_wildcards wc;
5473
5474     ds_put_cstr(ds, "Flow: ");
5475     flow_format(ds, flow);
5476     ds_put_char(ds, '\n');
5477
5478     flow_wildcards_init_catchall(&wc);
5479     rule_dpif_lookup(ofproto, flow, &wc, &rule);
5480
5481     trace_format_rule(ds, 0, rule);
5482     if (rule == ofproto->miss_rule) {
5483         ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5484     } else if (rule == ofproto->no_packet_in_rule) {
5485         ds_put_cstr(ds, "\nNo match, packets dropped because "
5486                     "OFPPC_NO_PACKET_IN is set on in_port.\n");
5487     } else if (rule == ofproto->drop_frags_rule) {
5488         ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5489                     "and the fragment handling mode is \"drop\".\n");
5490     }
5491
5492     if (rule) {
5493         uint64_t odp_actions_stub[1024 / 8];
5494         struct ofpbuf odp_actions;
5495         struct trace_ctx trace;
5496         struct match match;
5497         uint8_t tcp_flags;
5498
5499         tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5500         trace.result = ds;
5501         trace.flow = *flow;
5502         ofpbuf_use_stub(&odp_actions,
5503                         odp_actions_stub, sizeof odp_actions_stub);
5504         xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5505         trace.xin.resubmit_hook = trace_resubmit;
5506         trace.xin.report_hook = trace_report;
5507
5508         xlate_actions(&trace.xin, &trace.xout);
5509         flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5510
5511         ds_put_char(ds, '\n');
5512         trace_format_flow(ds, 0, "Final flow", &trace);
5513
5514         match_init(&match, flow, &trace.xout.wc);
5515         ds_put_cstr(ds, "Relevant fields: ");
5516         match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5517         ds_put_char(ds, '\n');
5518
5519         ds_put_cstr(ds, "Datapath actions: ");
5520         format_odp_actions(ds, trace.xout.odp_actions.data,
5521                            trace.xout.odp_actions.size);
5522
5523         if (trace.xout.slow) {
5524             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5525                         "slow path because it:");
5526             switch (trace.xout.slow) {
5527             case SLOW_CFM:
5528                 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5529                 break;
5530             case SLOW_LACP:
5531                 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5532                 break;
5533             case SLOW_STP:
5534                 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5535                 break;
5536             case SLOW_BFD:
5537                 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5538                 break;
5539             case SLOW_CONTROLLER:
5540                 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5541                             "to the OpenFlow controller.");
5542                 break;
5543             case __SLOW_MAX:
5544                 NOT_REACHED();
5545             }
5546         }
5547
5548         xlate_out_uninit(&trace.xout);
5549     }
5550
5551     rule_release(rule);
5552 }
5553
5554 static void
5555 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5556                   const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5557 {
5558     clogged = true;
5559     unixctl_command_reply(conn, NULL);
5560 }
5561
5562 static void
5563 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5564                     const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5565 {
5566     clogged = false;
5567     unixctl_command_reply(conn, NULL);
5568 }
5569
5570 /* Runs a self-check of flow translations in 'ofproto'.  Appends a message to
5571  * 'reply' describing the results. */
5572 static void
5573 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5574 {
5575     struct cls_cursor cursor;
5576     struct facet *facet;
5577     int errors;
5578
5579     errors = 0;
5580     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5581     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5582     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5583         if (!facet_check_consistency(facet)) {
5584             errors++;
5585         }
5586     }
5587     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5588     if (errors) {
5589         ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5590     }
5591
5592     if (errors) {
5593         ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5594                       ofproto->up.name, errors);
5595     } else {
5596         ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5597     }
5598 }
5599
5600 static void
5601 ofproto_dpif_self_check(struct unixctl_conn *conn,
5602                         int argc, const char *argv[], void *aux OVS_UNUSED)
5603 {
5604     struct ds reply = DS_EMPTY_INITIALIZER;
5605     struct ofproto_dpif *ofproto;
5606
5607     if (argc > 1) {
5608         ofproto = ofproto_dpif_lookup(argv[1]);
5609         if (!ofproto) {
5610             unixctl_command_reply_error(conn, "Unknown ofproto (use "
5611                                         "ofproto/list for help)");
5612             return;
5613         }
5614         ofproto_dpif_self_check__(ofproto, &reply);
5615     } else {
5616         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5617             ofproto_dpif_self_check__(ofproto, &reply);
5618         }
5619     }
5620
5621     unixctl_command_reply(conn, ds_cstr(&reply));
5622     ds_destroy(&reply);
5623 }
5624
5625 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
5626  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
5627  * to destroy 'ofproto_shash' and free the returned value. */
5628 static const struct shash_node **
5629 get_ofprotos(struct shash *ofproto_shash)
5630 {
5631     const struct ofproto_dpif *ofproto;
5632
5633     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5634         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5635         shash_add_nocopy(ofproto_shash, name, ofproto);
5636     }
5637
5638     return shash_sort(ofproto_shash);
5639 }
5640
5641 static void
5642 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5643                               const char *argv[] OVS_UNUSED,
5644                               void *aux OVS_UNUSED)
5645 {
5646     struct ds ds = DS_EMPTY_INITIALIZER;
5647     struct shash ofproto_shash;
5648     const struct shash_node **sorted_ofprotos;
5649     int i;
5650
5651     shash_init(&ofproto_shash);
5652     sorted_ofprotos = get_ofprotos(&ofproto_shash);
5653     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5654         const struct shash_node *node = sorted_ofprotos[i];
5655         ds_put_format(&ds, "%s\n", node->name);
5656     }
5657
5658     shash_destroy(&ofproto_shash);
5659     free(sorted_ofprotos);
5660
5661     unixctl_command_reply(conn, ds_cstr(&ds));
5662     ds_destroy(&ds);
5663 }
5664
5665 static void
5666 show_dp_rates(struct ds *ds, const char *heading,
5667               const struct avg_subfacet_rates *rates)
5668 {
5669     ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5670                   heading, rates->add_rate, rates->del_rate);
5671 }
5672
5673 static void
5674 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5675 {
5676     const struct shash_node **ofprotos;
5677     struct ofproto_dpif *ofproto;
5678     struct shash ofproto_shash;
5679     uint64_t n_hit, n_missed;
5680     long long int minutes;
5681     size_t i;
5682
5683     n_hit = n_missed = 0;
5684     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5685         if (ofproto->backer == backer) {
5686             n_missed += ofproto->n_missed;
5687             n_hit += ofproto->n_hit;
5688         }
5689     }
5690
5691     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5692                   dpif_name(backer->dpif), n_hit, n_missed);
5693     ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5694                   " life span: %lldms\n", hmap_count(&backer->subfacets),
5695                   backer->avg_n_subfacet, backer->max_n_subfacet,
5696                   backer->avg_subfacet_life);
5697
5698     minutes = (time_msec() - backer->created) / (1000 * 60);
5699     if (minutes >= 60) {
5700         show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5701     }
5702     if (minutes >= 60 * 24) {
5703         show_dp_rates(ds, "\tdaily avg:",  &backer->daily);
5704     }
5705     show_dp_rates(ds, "\toverall avg:",  &backer->lifetime);
5706
5707     shash_init(&ofproto_shash);
5708     ofprotos = get_ofprotos(&ofproto_shash);
5709     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5710         struct ofproto_dpif *ofproto = ofprotos[i]->data;
5711         const struct shash_node **ports;
5712         size_t j;
5713
5714         if (ofproto->backer != backer) {
5715             continue;
5716         }
5717
5718         ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5719                       ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5720
5721         ports = shash_sort(&ofproto->up.port_by_name);
5722         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5723             const struct shash_node *node = ports[j];
5724             struct ofport *ofport = node->data;
5725             struct smap config;
5726             odp_port_t odp_port;
5727
5728             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5729                           ofport->ofp_port);
5730
5731             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5732             if (odp_port != ODPP_NONE) {
5733                 ds_put_format(ds, "%"PRIu32":", odp_port);
5734             } else {
5735                 ds_put_cstr(ds, "none:");
5736             }
5737
5738             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5739
5740             smap_init(&config);
5741             if (!netdev_get_config(ofport->netdev, &config)) {
5742                 const struct smap_node **nodes;
5743                 size_t i;
5744
5745                 nodes = smap_sort(&config);
5746                 for (i = 0; i < smap_count(&config); i++) {
5747                     const struct smap_node *node = nodes[i];
5748                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5749                                   node->key, node->value);
5750                 }
5751                 free(nodes);
5752             }
5753             smap_destroy(&config);
5754
5755             ds_put_char(ds, ')');
5756             ds_put_char(ds, '\n');
5757         }
5758         free(ports);
5759     }
5760     shash_destroy(&ofproto_shash);
5761     free(ofprotos);
5762 }
5763
5764 static void
5765 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5766                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5767 {
5768     struct ds ds = DS_EMPTY_INITIALIZER;
5769     const struct shash_node **backers;
5770     int i;
5771
5772     backers = shash_sort(&all_dpif_backers);
5773     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5774         dpif_show_backer(backers[i]->data, &ds);
5775     }
5776     free(backers);
5777
5778     unixctl_command_reply(conn, ds_cstr(&ds));
5779     ds_destroy(&ds);
5780 }
5781
5782 /* Dump the megaflow (facet) cache.  This is useful to check the
5783  * correctness of flow wildcarding, since the same mechanism is used for
5784  * both xlate caching and kernel wildcarding.
5785  *
5786  * It's important to note that in the output the flow description uses
5787  * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5788  *
5789  * This command is only needed for advanced debugging, so it's not
5790  * documented in the man page. */
5791 static void
5792 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5793                                     int argc OVS_UNUSED, const char *argv[],
5794                                     void *aux OVS_UNUSED)
5795 {
5796     struct ds ds = DS_EMPTY_INITIALIZER;
5797     const struct ofproto_dpif *ofproto;
5798     long long int now = time_msec();
5799     struct cls_cursor cursor;
5800     struct facet *facet;
5801
5802     ofproto = ofproto_dpif_lookup(argv[1]);
5803     if (!ofproto) {
5804         unixctl_command_reply_error(conn, "no such bridge");
5805         return;
5806     }
5807
5808     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5809     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5810     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5811         cls_rule_format(&facet->cr, &ds);
5812         ds_put_cstr(&ds, ", ");
5813         ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5814         ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5815         ds_put_cstr(&ds, "Datapath actions: ");
5816         if (facet->xout.slow) {
5817             uint64_t slow_path_stub[128 / 8];
5818             const struct nlattr *actions;
5819             size_t actions_len;
5820
5821             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5822                               slow_path_stub, sizeof slow_path_stub,
5823                               &actions, &actions_len);
5824             format_odp_actions(&ds, actions, actions_len);
5825         } else {
5826             format_odp_actions(&ds, facet->xout.odp_actions.data,
5827                                facet->xout.odp_actions.size);
5828         }
5829         ds_put_cstr(&ds, "\n");
5830     }
5831     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5832
5833     ds_chomp(&ds, '\n');
5834     unixctl_command_reply(conn, ds_cstr(&ds));
5835     ds_destroy(&ds);
5836 }
5837
5838 /* Disable using the megaflows.
5839  *
5840  * This command is only needed for advanced debugging, so it's not
5841  * documented in the man page. */
5842 static void
5843 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5844                                        int argc OVS_UNUSED,
5845                                        const char *argv[] OVS_UNUSED,
5846                                        void *aux OVS_UNUSED)
5847 {
5848     struct ofproto_dpif *ofproto;
5849
5850     enable_megaflows = false;
5851
5852     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5853         flush(&ofproto->up);
5854     }
5855
5856     unixctl_command_reply(conn, "megaflows disabled");
5857 }
5858
5859 /* Re-enable using megaflows.
5860  *
5861  * This command is only needed for advanced debugging, so it's not
5862  * documented in the man page. */
5863 static void
5864 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5865                                       int argc OVS_UNUSED,
5866                                       const char *argv[] OVS_UNUSED,
5867                                       void *aux OVS_UNUSED)
5868 {
5869     struct ofproto_dpif *ofproto;
5870
5871     enable_megaflows = true;
5872
5873     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5874         flush(&ofproto->up);
5875     }
5876
5877     unixctl_command_reply(conn, "megaflows enabled");
5878 }
5879
5880 static void
5881 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5882                                 int argc OVS_UNUSED, const char *argv[],
5883                                 void *aux OVS_UNUSED)
5884 {
5885     struct ds ds = DS_EMPTY_INITIALIZER;
5886     const struct ofproto_dpif *ofproto;
5887     struct subfacet *subfacet;
5888
5889     ofproto = ofproto_dpif_lookup(argv[1]);
5890     if (!ofproto) {
5891         unixctl_command_reply_error(conn, "no such bridge");
5892         return;
5893     }
5894
5895     update_stats(ofproto->backer);
5896
5897     HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5898         struct facet *facet = subfacet->facet;
5899         struct odputil_keybuf maskbuf;
5900         struct ofpbuf mask;
5901
5902         if (facet->ofproto != ofproto) {
5903             continue;
5904         }
5905
5906         ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5907         if (enable_megaflows) {
5908             odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5909                                    &facet->flow, UINT32_MAX);
5910         }
5911
5912         odp_flow_format(subfacet->key, subfacet->key_len,
5913                         mask.data, mask.size, &ds, false);
5914
5915         ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5916                       subfacet->dp_packet_count, subfacet->dp_byte_count);
5917         if (subfacet->used) {
5918             ds_put_format(&ds, "%.3fs",
5919                           (time_msec() - subfacet->used) / 1000.0);
5920         } else {
5921             ds_put_format(&ds, "never");
5922         }
5923         if (subfacet->facet->tcp_flags) {
5924             ds_put_cstr(&ds, ", flags:");
5925             packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5926         }
5927
5928         ds_put_cstr(&ds, ", actions:");
5929         if (facet->xout.slow) {
5930             uint64_t slow_path_stub[128 / 8];
5931             const struct nlattr *actions;
5932             size_t actions_len;
5933
5934             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5935                               slow_path_stub, sizeof slow_path_stub,
5936                               &actions, &actions_len);
5937             format_odp_actions(&ds, actions, actions_len);
5938         } else {
5939             format_odp_actions(&ds, facet->xout.odp_actions.data,
5940                                facet->xout.odp_actions.size);
5941         }
5942         ds_put_char(&ds, '\n');
5943     }
5944
5945     unixctl_command_reply(conn, ds_cstr(&ds));
5946     ds_destroy(&ds);
5947 }
5948
5949 static void
5950 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5951                                int argc OVS_UNUSED, const char *argv[],
5952                                void *aux OVS_UNUSED)
5953 {
5954     struct ds ds = DS_EMPTY_INITIALIZER;
5955     struct ofproto_dpif *ofproto;
5956
5957     ofproto = ofproto_dpif_lookup(argv[1]);
5958     if (!ofproto) {
5959         unixctl_command_reply_error(conn, "no such bridge");
5960         return;
5961     }
5962
5963     flush(&ofproto->up);
5964
5965     unixctl_command_reply(conn, ds_cstr(&ds));
5966     ds_destroy(&ds);
5967 }
5968
5969 static void
5970 ofproto_dpif_unixctl_init(void)
5971 {
5972     static bool registered;
5973     if (registered) {
5974         return;
5975     }
5976     registered = true;
5977
5978     unixctl_command_register(
5979         "ofproto/trace",
5980         "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
5981         1, 3, ofproto_unixctl_trace, NULL);
5982     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5983                              ofproto_unixctl_fdb_flush, NULL);
5984     unixctl_command_register("fdb/show", "bridge", 1, 1,
5985                              ofproto_unixctl_fdb_show, NULL);
5986     unixctl_command_register("ofproto/clog", "", 0, 0,
5987                              ofproto_dpif_clog, NULL);
5988     unixctl_command_register("ofproto/unclog", "", 0, 0,
5989                              ofproto_dpif_unclog, NULL);
5990     unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
5991                              ofproto_dpif_self_check, NULL);
5992     unixctl_command_register("dpif/dump-dps", "", 0, 0,
5993                              ofproto_unixctl_dpif_dump_dps, NULL);
5994     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5995                              NULL);
5996     unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
5997                              ofproto_unixctl_dpif_dump_flows, NULL);
5998     unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
5999                              ofproto_unixctl_dpif_del_flows, NULL);
6000     unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6001                              ofproto_unixctl_dpif_dump_megaflows, NULL);
6002     unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6003                              ofproto_unixctl_dpif_disable_megaflows, NULL);
6004     unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6005                              ofproto_unixctl_dpif_enable_megaflows, NULL);
6006 }
6007 \f
6008 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6009  *
6010  * This is deprecated.  It is only for compatibility with broken device drivers
6011  * in old versions of Linux that do not properly support VLANs when VLAN
6012  * devices are not used.  When broken device drivers are no longer in
6013  * widespread use, we will delete these interfaces. */
6014
6015 static int
6016 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6017 {
6018     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6019     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6020
6021     if (realdev_ofp_port == ofport->realdev_ofp_port
6022         && vid == ofport->vlandev_vid) {
6023         return 0;
6024     }
6025
6026     ofproto->backer->need_revalidate = REV_RECONFIGURE;
6027
6028     if (ofport->realdev_ofp_port) {
6029         vsp_remove(ofport);
6030     }
6031     if (realdev_ofp_port && ofport->bundle) {
6032         /* vlandevs are enslaved to their realdevs, so they are not allowed to
6033          * themselves be part of a bundle. */
6034         bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6035     }
6036
6037     ofport->realdev_ofp_port = realdev_ofp_port;
6038     ofport->vlandev_vid = vid;
6039
6040     if (realdev_ofp_port) {
6041         vsp_add(ofport, realdev_ofp_port, vid);
6042     }
6043
6044     return 0;
6045 }
6046
6047 static uint32_t
6048 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6049 {
6050     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6051 }
6052
6053 bool
6054 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6055     OVS_EXCLUDED(ofproto->vsp_mutex)
6056 {
6057     bool ret;
6058
6059     ovs_mutex_lock(&ofproto->vsp_mutex);
6060     ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6061     ovs_mutex_unlock(&ofproto->vsp_mutex);
6062     return ret;
6063 }
6064
6065 static ofp_port_t
6066 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6067                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6068     OVS_REQUIRES(ofproto->vsp_mutex)
6069 {
6070     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6071         int vid = vlan_tci_to_vid(vlan_tci);
6072         const struct vlan_splinter *vsp;
6073
6074         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6075                                  hash_realdev_vid(realdev_ofp_port, vid),
6076                                  &ofproto->realdev_vid_map) {
6077             if (vsp->realdev_ofp_port == realdev_ofp_port
6078                 && vsp->vid == vid) {
6079                 return vsp->vlandev_ofp_port;
6080             }
6081         }
6082     }
6083     return realdev_ofp_port;
6084 }
6085
6086 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6087  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6088  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
6089  * 'vlan_tci' 9, it would return the port number of eth0.9.
6090  *
6091  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6092  * function just returns its 'realdev_ofp_port' argument. */
6093 ofp_port_t
6094 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6095                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6096     OVS_EXCLUDED(ofproto->vsp_mutex)
6097 {
6098     ofp_port_t ret;
6099
6100     ovs_mutex_lock(&ofproto->vsp_mutex);
6101     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6102     ovs_mutex_unlock(&ofproto->vsp_mutex);
6103     return ret;
6104 }
6105
6106 static struct vlan_splinter *
6107 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6108 {
6109     struct vlan_splinter *vsp;
6110
6111     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6112                              hash_ofp_port(vlandev_ofp_port),
6113                              &ofproto->vlandev_map) {
6114         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6115             return vsp;
6116         }
6117     }
6118
6119     return NULL;
6120 }
6121
6122 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6123  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6124  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
6125  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6126  * eth0 and store 9 in '*vid'.
6127  *
6128  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6129  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
6130  * always does.*/
6131 static ofp_port_t
6132 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6133                        ofp_port_t vlandev_ofp_port, int *vid)
6134     OVS_REQUIRES(ofproto->vsp_mutex)
6135 {
6136     if (!hmap_is_empty(&ofproto->vlandev_map)) {
6137         const struct vlan_splinter *vsp;
6138
6139         vsp = vlandev_find(ofproto, vlandev_ofp_port);
6140         if (vsp) {
6141             if (vid) {
6142                 *vid = vsp->vid;
6143             }
6144             return vsp->realdev_ofp_port;
6145         }
6146     }
6147     return 0;
6148 }
6149
6150 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6151  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
6152  * 'flow->in_port' to the "real" device backing the VLAN device, sets
6153  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Otherwise (which is
6154  * always the case unless VLAN splinters are enabled), returns false without
6155  * making any changes. */
6156 bool
6157 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6158     OVS_EXCLUDED(ofproto->vsp_mutex)
6159 {
6160     ofp_port_t realdev;
6161     int vid;
6162
6163     ovs_mutex_lock(&ofproto->vsp_mutex);
6164     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6165     ovs_mutex_unlock(&ofproto->vsp_mutex);
6166     if (!realdev) {
6167         return false;
6168     }
6169
6170     /* Cause the flow to be processed as if it came in on the real device with
6171      * the VLAN device's VLAN ID. */
6172     flow->in_port.ofp_port = realdev;
6173     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6174     return true;
6175 }
6176
6177 static void
6178 vsp_remove(struct ofport_dpif *port)
6179 {
6180     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6181     struct vlan_splinter *vsp;
6182
6183     ovs_mutex_lock(&ofproto->vsp_mutex);
6184     vsp = vlandev_find(ofproto, port->up.ofp_port);
6185     if (vsp) {
6186         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6187         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6188         free(vsp);
6189
6190         port->realdev_ofp_port = 0;
6191     } else {
6192         VLOG_ERR("missing vlan device record");
6193     }
6194     ovs_mutex_unlock(&ofproto->vsp_mutex);
6195 }
6196
6197 static void
6198 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6199 {
6200     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6201
6202     ovs_mutex_lock(&ofproto->vsp_mutex);
6203     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6204         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6205             == realdev_ofp_port)) {
6206         struct vlan_splinter *vsp;
6207
6208         vsp = xmalloc(sizeof *vsp);
6209         vsp->realdev_ofp_port = realdev_ofp_port;
6210         vsp->vlandev_ofp_port = port->up.ofp_port;
6211         vsp->vid = vid;
6212
6213         port->realdev_ofp_port = realdev_ofp_port;
6214
6215         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6216                     hash_ofp_port(port->up.ofp_port));
6217         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6218                     hash_realdev_vid(realdev_ofp_port, vid));
6219     } else {
6220         VLOG_ERR("duplicate vlan device record");
6221     }
6222     ovs_mutex_unlock(&ofproto->vsp_mutex);
6223 }
6224
6225 static odp_port_t
6226 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6227 {
6228     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6229     return ofport ? ofport->odp_port : ODPP_NONE;
6230 }
6231
6232 struct ofport_dpif *
6233 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6234 {
6235     struct ofport_dpif *port;
6236
6237     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6238     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6239                              &backer->odp_to_ofport_map) {
6240         if (port->odp_port == odp_port) {
6241             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6242             return port;
6243         }
6244     }
6245
6246     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6247     return NULL;
6248 }
6249
6250 static ofp_port_t
6251 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6252 {
6253     struct ofport_dpif *port;
6254
6255     port = odp_port_to_ofport(ofproto->backer, odp_port);
6256     if (port && &ofproto->up == port->up.ofproto) {
6257         return port->up.ofp_port;
6258     } else {
6259         return OFPP_NONE;
6260     }
6261 }
6262
6263 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6264  * most heavily weighted element.  'base' designates the rate of decay: after
6265  * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6266  * (about .37). */
6267 static void
6268 exp_mavg(double *avg, int base, double new)
6269 {
6270     *avg = (*avg * (base - 1) + new) / base;
6271 }
6272
6273 static void
6274 update_moving_averages(struct dpif_backer *backer)
6275 {
6276     const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6277     long long int minutes = (time_msec() - backer->created) / min_ms;
6278
6279     if (minutes > 0) {
6280         backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6281             / minutes;
6282         backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6283             / minutes;
6284     } else {
6285         backer->lifetime.add_rate = 0.0;
6286         backer->lifetime.del_rate = 0.0;
6287     }
6288
6289     /* Update hourly averages on the minute boundaries. */
6290     if (time_msec() - backer->last_minute >= min_ms) {
6291         exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6292         exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6293
6294         /* Update daily averages on the hour boundaries. */
6295         if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6296             exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6297             exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6298         }
6299
6300         backer->total_subfacet_add_count += backer->subfacet_add_count;
6301         backer->total_subfacet_del_count += backer->subfacet_del_count;
6302         backer->subfacet_add_count = 0;
6303         backer->subfacet_del_count = 0;
6304         backer->last_minute += min_ms;
6305     }
6306 }
6307
6308 const struct ofproto_class ofproto_dpif_class = {
6309     init,
6310     enumerate_types,
6311     enumerate_names,
6312     del,
6313     port_open_type,
6314     type_run,
6315     type_run_fast,
6316     type_wait,
6317     alloc,
6318     construct,
6319     destruct,
6320     dealloc,
6321     run,
6322     run_fast,
6323     wait,
6324     get_memory_usage,
6325     flush,
6326     get_features,
6327     get_tables,
6328     port_alloc,
6329     port_construct,
6330     port_destruct,
6331     port_dealloc,
6332     port_modified,
6333     port_reconfigured,
6334     port_query_by_name,
6335     port_add,
6336     port_del,
6337     port_get_stats,
6338     port_dump_start,
6339     port_dump_next,
6340     port_dump_done,
6341     port_poll,
6342     port_poll_wait,
6343     port_is_lacp_current,
6344     NULL,                       /* rule_choose_table */
6345     rule_alloc,
6346     rule_construct,
6347     rule_destruct,
6348     rule_dealloc,
6349     rule_get_stats,
6350     rule_execute,
6351     rule_modify_actions,
6352     set_frag_handling,
6353     packet_out,
6354     set_netflow,
6355     get_netflow_ids,
6356     set_sflow,
6357     set_ipfix,
6358     set_cfm,
6359     get_cfm_status,
6360     set_bfd,
6361     get_bfd_status,
6362     set_stp,
6363     get_stp_status,
6364     set_stp_port,
6365     get_stp_port_status,
6366     set_queues,
6367     bundle_set,
6368     bundle_remove,
6369     mirror_set__,
6370     mirror_get_stats__,
6371     set_flood_vlans,
6372     is_mirror_output_bundle,
6373     forward_bpdu_changed,
6374     set_mac_table_config,
6375     set_realdev,
6376     NULL,                       /* meter_get_features */
6377     NULL,                       /* meter_set */
6378     NULL,                       /* meter_get */
6379     NULL,                       /* meter_del */
6380 };