ofproto/ofproto-dpif.c

   1 /*
   2  * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18
  19 #include "ofproto/ofproto-dpif.h"
  20 #include "ofproto/ofproto-provider.h"
  21
  22 #include <errno.h>
  23
  24 #include "bfd.h"
  25 #include "bond.h"
  26 #include "bundle.h"
  27 #include "byte-order.h"
  28 #include "connmgr.h"
  29 #include "coverage.h"
  30 #include "cfm.h"
  31 #include "dpif.h"
  32 #include "dynamic-string.h"
  33 #include "fail-open.h"
  34 #include "hmapx.h"
  35 #include "lacp.h"
  36 #include "learn.h"
  37 #include "mac-learning.h"
  38 #include "meta-flow.h"
  39 #include "multipath.h"
  40 #include "netdev-vport.h"
  41 #include "netdev.h"
  42 #include "netlink.h"
  43 #include "nx-match.h"
  44 #include "odp-util.h"
  45 #include "odp-execute.h"
  46 #include "ofp-util.h"
  47 #include "ofpbuf.h"
  48 #include "ofp-actions.h"
  49 #include "ofp-parse.h"
  50 #include "ofp-print.h"
  51 #include "ofproto-dpif-governor.h"
  52 #include "ofproto-dpif-ipfix.h"
  53 #include "ofproto-dpif-mirror.h"
  54 #include "ofproto-dpif-sflow.h"
  55 #include "ofproto-dpif-upcall.h"
  56 #include "ofproto-dpif-xlate.h"
  57 #include "poll-loop.h"
  58 #include "simap.h"
  59 #include "smap.h"
  60 #include "timer.h"
  61 #include "tunnel.h"
  62 #include "unaligned.h"
  63 #include "unixctl.h"
  64 #include "vlan-bitmap.h"
  65 #include "vlog.h"
  66
  67 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  68
  69 COVERAGE_DEFINE(ofproto_dpif_expired);
  70 COVERAGE_DEFINE(facet_changed_rule);
  71 COVERAGE_DEFINE(facet_revalidate);
  72 COVERAGE_DEFINE(facet_unexpected);
  73 COVERAGE_DEFINE(facet_suppress);
  74 COVERAGE_DEFINE(subfacet_install_fail);
  75 COVERAGE_DEFINE(packet_in_overflow);
  76 COVERAGE_DEFINE(flow_mod_overflow);
  77
  78 /* Number of implemented OpenFlow tables. */
  79 enum { N_TABLES = 255 };
  80 enum { TBL_INTERNAL = N_TABLES - 1 };    /* Used for internal hidden rules. */
  81 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
  82
  83 struct flow_miss;
  84 struct facet;
  85
  86 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
  87
  88 struct ofbundle {
  89     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
  90     struct ofproto_dpif *ofproto; /* Owning ofproto. */
  91     void *aux;                  /* Key supplied by ofproto's client. */
  92     char *name;                 /* Identifier for log messages. */
  93
  94     /* Configuration. */
  95     struct list ports;          /* Contains "struct ofport"s. */
  96     enum port_vlan_mode vlan_mode; /* VLAN mode */
  97     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
  98     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
  99                                  * NULL if all VLANs are trunked. */
 100     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 101     struct bond *bond;          /* Nonnull iff more than one port. */
 102     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 103
 104     /* Status. */
 105     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 106 };
 107
 108 static void bundle_remove(struct ofport *);
 109 static void bundle_update(struct ofbundle *);
 110 static void bundle_destroy(struct ofbundle *);
 111 static void bundle_del_port(struct ofport_dpif *);
 112 static void bundle_run(struct ofbundle *);
 113 static void bundle_wait(struct ofbundle *);
 114
 115 static void stp_run(struct ofproto_dpif *ofproto);
 116 static void stp_wait(struct ofproto_dpif *ofproto);
 117 static int set_stp_port(struct ofport *,
 118                         const struct ofproto_port_stp_settings *);
 119
 120 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
 121                               enum slow_path_reason,
 122                               uint64_t *stub, size_t stub_size,
 123                               const struct nlattr **actionsp,
 124                               size_t *actions_lenp);
 125
 126 /* A subfacet (see "struct subfacet" below) has three possible installation
 127  * states:
 128  *
 129  *   - SF_NOT_INSTALLED: Not installed in the datapath.  This will only be the
 130  *     case just after the subfacet is created, just before the subfacet is
 131  *     destroyed, or if the datapath returns an error when we try to install a
 132  *     subfacet.
 133  *
 134  *   - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
 135  *
 136  *   - SF_SLOW_PATH: An action that sends every packet for the subfacet through
 137  *     ofproto_dpif is installed in the datapath.
 138  */
 139 enum subfacet_path {
 140     SF_NOT_INSTALLED,           /* No datapath flow for this subfacet. */
 141     SF_FAST_PATH,               /* Full actions are installed. */
 142     SF_SLOW_PATH,               /* Send-to-userspace action is installed. */
 143 };
 144
 145 /* A dpif flow and actions associated with a facet.
 146  *
 147  * See also the large comment on struct facet. */
 148 struct subfacet {
 149     /* Owners. */
 150     struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
 151     struct list list_node;      /* In struct facet's 'facets' list. */
 152     struct facet *facet;        /* Owning facet. */
 153     struct dpif_backer *backer; /* Owning backer. */
 154
 155     enum odp_key_fitness key_fitness;
 156     struct nlattr *key;
 157     int key_len;
 158
 159     long long int used;         /* Time last used; time created if not used. */
 160     long long int created;      /* Time created. */
 161
 162     uint64_t dp_packet_count;   /* Last known packet count in the datapath. */
 163     uint64_t dp_byte_count;     /* Last known byte count in the datapath. */
 164
 165     enum subfacet_path path;    /* Installed in datapath? */
 166 };
 167
 168 #define SUBFACET_DESTROY_MAX_BATCH 50
 169
 170 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *);
 171 static struct subfacet *subfacet_find(struct dpif_backer *,
 172                                       const struct nlattr *key, size_t key_len,
 173                                       uint32_t key_hash);
 174 static void subfacet_destroy(struct subfacet *);
 175 static void subfacet_destroy__(struct subfacet *);
 176 static void subfacet_destroy_batch(struct dpif_backer *,
 177                                    struct subfacet **, int n);
 178 static void subfacet_reset_dp_stats(struct subfacet *,
 179                                     struct dpif_flow_stats *);
 180 static void subfacet_update_stats(struct subfacet *,
 181                                   const struct dpif_flow_stats *);
 182 static int subfacet_install(struct subfacet *,
 183                             const struct ofpbuf *odp_actions,
 184                             struct dpif_flow_stats *);
 185 static void subfacet_uninstall(struct subfacet *);
 186
 187 /* A unique, non-overlapping instantiation of an OpenFlow flow.
 188  *
 189  * A facet associates a "struct flow", which represents the Open vSwitch
 190  * userspace idea of an exact-match flow, with one or more subfacets.
 191  * While the facet is created based on an exact-match flow, it is stored
 192  * within the ofproto based on the wildcards that could be expressed
 193  * based on the flow table and other configuration.  (See the 'wc'
 194  * description in "struct xlate_out" for more details.)
 195  *
 196  * Each subfacet tracks the datapath's idea of the flow equivalent to
 197  * the facet.  When the kernel module (or other dpif implementation) and
 198  * Open vSwitch userspace agree on the definition of a flow key, there
 199  * is exactly one subfacet per facet.  If the dpif implementation
 200  * supports more-specific flow matching than userspace, however, a facet
 201  * can have more than one subfacet.  Examples include the dpif
 202  * implementation not supporting the same wildcards as userspace or some
 203  * distinction in flow that userspace simply doesn't understand.
 204  *
 205  * Flow expiration works in terms of subfacets, so a facet must have at
 206  * least one subfacet or it will never expire, leaking memory. */
 207 struct facet {
 208     /* Owners. */
 209     struct hmap_node hmap_node;  /* In owning ofproto's 'facets' hmap. */
 210     struct ofproto_dpif *ofproto;
 211
 212     /* Owned data. */
 213     struct list subfacets;
 214     long long int used;         /* Time last used; time created if not used. */
 215
 216     /* Key. */
 217     struct flow flow;           /* Flow of the creating subfacet. */
 218     struct cls_rule cr;         /* In 'ofproto_dpif's facets classifier. */
 219
 220     /* These statistics:
 221      *
 222      *   - Do include packets and bytes sent "by hand", e.g. with
 223      *     dpif_execute().
 224      *
 225      *   - Do include packets and bytes that were obtained from the datapath
 226      *     when a subfacet's statistics were reset (e.g. dpif_flow_put() with
 227      *     DPIF_FP_ZERO_STATS).
 228      *
 229      *   - Do not include packets or bytes that can be obtained from the
 230      *     datapath for any existing subfacet.
 231      */
 232     uint64_t packet_count;       /* Number of packets received. */
 233     uint64_t byte_count;         /* Number of bytes received. */
 234
 235     /* Resubmit statistics. */
 236     uint64_t prev_packet_count;  /* Number of packets from last stats push. */
 237     uint64_t prev_byte_count;    /* Number of bytes from last stats push. */
 238     long long int prev_used;     /* Used time from last stats push. */
 239
 240     /* Accounting. */
 241     uint64_t accounted_bytes;    /* Bytes processed by facet_account(). */
 242     struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
 243     uint8_t tcp_flags;           /* TCP flags seen for this 'rule'. */
 244
 245     struct xlate_out xout;
 246
 247     /* Storage for a single subfacet, to reduce malloc() time and space
 248      * overhead.  (A facet always has at least one subfacet and in the common
 249      * case has exactly one subfacet.  However, 'one_subfacet' may not
 250      * always be valid, since it could have been removed after newer
 251      * subfacets were pushed onto the 'subfacets' list.) */
 252     struct subfacet one_subfacet;
 253
 254     long long int learn_rl;      /* Rate limiter for facet_learn(). */
 255 };
 256
 257 static struct facet *facet_create(const struct flow_miss *);
 258 static void facet_remove(struct facet *);
 259 static void facet_free(struct facet *);
 260
 261 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
 262 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
 263                                         const struct flow *);
 264 static bool facet_revalidate(struct facet *);
 265 static bool facet_check_consistency(struct facet *);
 266
 267 static void facet_flush_stats(struct facet *);
 268
 269 static void facet_reset_counters(struct facet *);
 270 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
 271                             struct dpif_flow_stats *, bool may_learn);
 272 static void facet_push_stats(struct facet *, bool may_learn);
 273 static void facet_learn(struct facet *);
 274 static void facet_account(struct facet *);
 275 static void push_all_stats(void);
 276
 277 static bool facet_is_controller_flow(struct facet *);
 278
 279 struct ofport_dpif {
 280     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 281     struct ofport up;
 282
 283     odp_port_t odp_port;
 284     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 285     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
 286     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 287     struct bfd *bfd;            /* BFD, if any. */
 288     bool may_enable;            /* May be enabled in bonds. */
 289     bool is_tunnel;             /* This port is a tunnel. */
 290     long long int carrier_seq;  /* Carrier status changes. */
 291     struct ofport_dpif *peer;   /* Peer if patch port. */
 292
 293     /* Spanning tree. */
 294     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 295     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 296     long long int stp_state_entered;
 297
 298     /* Queue to DSCP mapping. */
 299     struct ofproto_port_queue *qdscp;
 300     size_t n_qdscp;
 301
 302     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 303      *
 304      * This is deprecated.  It is only for compatibility with broken device
 305      * drivers in old versions of Linux that do not properly support VLANs when
 306      * VLAN devices are not used.  When broken device drivers are no longer in
 307      * widespread use, we will delete these interfaces. */
 308     ofp_port_t realdev_ofp_port;
 309     int vlandev_vid;
 310 };
 311
 312 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 313  *
 314  * This is deprecated.  It is only for compatibility with broken device drivers
 315  * in old versions of Linux that do not properly support VLANs when VLAN
 316  * devices are not used.  When broken device drivers are no longer in
 317  * widespread use, we will delete these interfaces. */
 318 struct vlan_splinter {
 319     struct hmap_node realdev_vid_node;
 320     struct hmap_node vlandev_node;
 321     ofp_port_t realdev_ofp_port;
 322     ofp_port_t vlandev_ofp_port;
 323     int vid;
 324 };
 325
 326 static void vsp_remove(struct ofport_dpif *);
 327 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 328
 329 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 330                                        ofp_port_t);
 331
 332 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 333                                        odp_port_t);
 334
 335 static struct ofport_dpif *
 336 ofport_dpif_cast(const struct ofport *ofport)
 337 {
 338     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 339 }
 340
 341 static void port_run(struct ofport_dpif *);
 342 static void port_run_fast(struct ofport_dpif *);
 343 static void port_wait(struct ofport_dpif *);
 344 static int set_bfd(struct ofport *, const struct smap *);
 345 static int set_cfm(struct ofport *, const struct cfm_settings *);
 346 static void ofport_update_peer(struct ofport_dpif *);
 347 static void run_fast_rl(void);
 348 static int run_fast(struct ofproto *);
 349
 350 struct dpif_completion {
 351     struct list list_node;
 352     struct ofoperation *op;
 353 };
 354
 355 /* Reasons that we might need to revalidate every facet, and corresponding
 356  * coverage counters.
 357  *
 358  * A value of 0 means that there is no need to revalidate.
 359  *
 360  * It would be nice to have some cleaner way to integrate with coverage
 361  * counters, but with only a few reasons I guess this is good enough for
 362  * now. */
 363 enum revalidate_reason {
 364     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 365     REV_STP,                   /* Spanning tree protocol port status change. */
 366     REV_BOND,                  /* Bonding changed. */
 367     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 368     REV_FLOW_TABLE,            /* Flow table changed. */
 369     REV_MAC_LEARNING,          /* Mac learning changed. */
 370     REV_INCONSISTENCY          /* Facet self-check failed. */
 371 };
 372 COVERAGE_DEFINE(rev_reconfigure);
 373 COVERAGE_DEFINE(rev_stp);
 374 COVERAGE_DEFINE(rev_bond);
 375 COVERAGE_DEFINE(rev_port_toggled);
 376 COVERAGE_DEFINE(rev_flow_table);
 377 COVERAGE_DEFINE(rev_mac_learning);
 378 COVERAGE_DEFINE(rev_inconsistency);
 379
 380 struct avg_subfacet_rates {
 381     double add_rate;   /* Moving average of new flows created per minute. */
 382     double del_rate;   /* Moving average of flows deleted per minute. */
 383 };
 384
 385 /* All datapaths of a given type share a single dpif backer instance. */
 386 struct dpif_backer {
 387     char *type;
 388     int refcount;
 389     struct dpif *dpif;
 390     struct udpif *udpif;
 391     struct timer next_expiration;
 392
 393     struct ovs_rwlock odp_to_ofport_lock;
 394     struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
 395
 396     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 397
 398     /* Facet revalidation flags applying to facets which use this backer. */
 399     enum revalidate_reason need_revalidate; /* Revalidate every facet. */
 400
 401     struct hmap drop_keys; /* Set of dropped odp keys. */
 402     bool recv_set_enable; /* Enables or disables receiving packets. */
 403
 404     struct hmap subfacets;
 405     struct governor *governor;
 406
 407     /* Subfacet statistics.
 408      *
 409      * These keep track of the total number of subfacets added and deleted and
 410      * flow life span.  They are useful for computing the flow rates stats
 411      * exposed via "ovs-appctl dpif/show".  The goal is to learn about
 412      * traffic patterns in ways that we can use later to improve Open vSwitch
 413      * performance in new situations.  */
 414     long long int created;           /* Time when it is created. */
 415     unsigned max_n_subfacet;         /* Maximum number of flows */
 416     unsigned avg_n_subfacet;         /* Average number of flows. */
 417     long long int avg_subfacet_life; /* Average life span of subfacets. */
 418
 419     /* The average number of subfacets... */
 420     struct avg_subfacet_rates hourly;   /* ...over the last hour. */
 421     struct avg_subfacet_rates daily;    /* ...over the last day. */
 422     struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
 423     long long int last_minute;          /* Last time 'hourly' was updated. */
 424
 425     /* Number of subfacets added or deleted since 'last_minute'. */
 426     unsigned subfacet_add_count;
 427     unsigned subfacet_del_count;
 428
 429     /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
 430     unsigned long long int total_subfacet_add_count;
 431     unsigned long long int total_subfacet_del_count;
 432
 433     /* Number of upcall handling threads. */
 434     unsigned int n_handler_threads;
 435 };
 436
 437 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 438 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 439
 440 static void drop_key_clear(struct dpif_backer *);
 441 static void update_moving_averages(struct dpif_backer *backer);
 442
 443 struct ofproto_dpif {
 444     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 445     struct ofproto up;
 446     struct dpif_backer *backer;
 447
 448     /* Special OpenFlow rules. */
 449     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 450     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 451     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 452
 453     /* Bridging. */
 454     struct netflow *netflow;
 455     struct dpif_sflow *sflow;
 456     struct dpif_ipfix *ipfix;
 457     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 458     struct mac_learning *ml;
 459     bool has_bonded_bundles;
 460     struct mbridge *mbridge;
 461
 462     /* Facets. */
 463     struct classifier facets;     /* Contains 'struct facet's. */
 464     long long int consistency_rl;
 465
 466     /* Support for debugging async flow mods. */
 467     struct list completions;
 468
 469     struct netdev_stats stats; /* To account packets generated and consumed in
 470                                 * userspace. */
 471
 472     /* Spanning tree. */
 473     struct stp *stp;
 474     long long int stp_last_tick;
 475
 476     /* VLAN splinters. */
 477     struct ovs_mutex vsp_mutex;
 478     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 479     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 480
 481     /* Ports. */
 482     struct sset ports;             /* Set of standard port names. */
 483     struct sset ghost_ports;       /* Ports with no datapath port. */
 484     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 485     int port_poll_errno;           /* Last errno for port_poll() reply. */
 486
 487     /* Per ofproto's dpif stats. */
 488     uint64_t n_hit;
 489     uint64_t n_missed;
 490
 491     /* Work queues. */
 492     struct ovs_mutex flow_mod_mutex;
 493     struct list flow_mods OVS_GUARDED;
 494     size_t n_flow_mods OVS_GUARDED;
 495
 496     struct ovs_mutex pin_mutex;
 497     struct list pins OVS_GUARDED;
 498     size_t n_pins OVS_GUARDED;
 499 };
 500
 501 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"?  (Useful only
 502  * for debugging the asynchronous flow_mod implementation.) */
 503 static bool clogged;
 504
 505 /* By default, flows in the datapath are wildcarded (megaflows).  They
 506  * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
 507 static bool enable_megaflows = true;
 508
 509 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 510 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 511
 512 static void ofproto_dpif_unixctl_init(void);
 513
 514 static inline struct ofproto_dpif *
 515 ofproto_dpif_cast(const struct ofproto *ofproto)
 516 {
 517     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 518     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 519 }
 520
 521 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
 522                                         ofp_port_t ofp_port);
 523 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
 524                           const struct ofpbuf *packet, struct ds *);
 525
 526 /* Upcalls. */
 527 static void handle_upcalls(struct dpif_backer *);
 528
 529 /* Flow expiration. */
 530 static int expire(struct dpif_backer *);
 531
 532 /* NetFlow. */
 533 static void send_netflow_active_timeouts(struct ofproto_dpif *);
 534
 535 /* Utilities. */
 536 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
 537
 538 /* Global variables. */
 539 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 540
 541 /* Initial mappings of port to bridge mappings. */
 542 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 543
 544 /* Executes and takes ownership of 'fm'. */
 545 void
 546 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 547                       struct ofputil_flow_mod *fm)
 548 {
 549     ovs_mutex_lock(&ofproto->flow_mod_mutex);
 550     if (ofproto->n_flow_mods > 1024) {
 551         ovs_mutex_unlock(&ofproto->flow_mod_mutex);
 552         COVERAGE_INC(flow_mod_overflow);
 553         free(fm->ofpacts);
 554         free(fm);
 555         return;
 556     }
 557
 558     list_push_back(&ofproto->flow_mods, &fm->list_node);
 559     ofproto->n_flow_mods++;
 560     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
 561 }
 562
 563 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 564  * Takes ownership of 'pin' and pin->packet. */
 565 void
 566 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 567                             struct ofputil_packet_in *pin)
 568 {
 569     ovs_mutex_lock(&ofproto->pin_mutex);
 570     if (ofproto->n_pins > 1024) {
 571         ovs_mutex_unlock(&ofproto->pin_mutex);
 572         COVERAGE_INC(packet_in_overflow);
 573         free(CONST_CAST(void *, pin->packet));
 574         free(pin);
 575         return;
 576     }
 577
 578     list_push_back(&ofproto->pins, &pin->list_node);
 579     ofproto->n_pins++;
 580     ovs_mutex_unlock(&ofproto->pin_mutex);
 581 }
 582 \f
 583 /* Factory functions. */
 584
 585 static void
 586 init(const struct shash *iface_hints)
 587 {
 588     struct shash_node *node;
 589
 590     /* Make a local copy, since we don't own 'iface_hints' elements. */
 591     SHASH_FOR_EACH(node, iface_hints) {
 592         const struct iface_hint *orig_hint = node->data;
 593         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 594
 595         new_hint->br_name = xstrdup(orig_hint->br_name);
 596         new_hint->br_type = xstrdup(orig_hint->br_type);
 597         new_hint->ofp_port = orig_hint->ofp_port;
 598
 599         shash_add(&init_ofp_ports, node->name, new_hint);
 600     }
 601 }
 602
 603 static void
 604 enumerate_types(struct sset *types)
 605 {
 606     dp_enumerate_types(types);
 607 }
 608
 609 static int
 610 enumerate_names(const char *type, struct sset *names)
 611 {
 612     struct ofproto_dpif *ofproto;
 613
 614     sset_clear(names);
 615     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 616         if (strcmp(type, ofproto->up.type)) {
 617             continue;
 618         }
 619         sset_add(names, ofproto->up.name);
 620     }
 621
 622     return 0;
 623 }
 624
 625 static int
 626 del(const char *type, const char *name)
 627 {
 628     struct dpif *dpif;
 629     int error;
 630
 631     error = dpif_open(name, type, &dpif);
 632     if (!error) {
 633         error = dpif_delete(dpif);
 634         dpif_close(dpif);
 635     }
 636     return error;
 637 }
 638 \f
 639 static const char *
 640 port_open_type(const char *datapath_type, const char *port_type)
 641 {
 642     return dpif_port_open_type(datapath_type, port_type);
 643 }
 644
 645 /* Type functions. */
 646
 647 static void process_dpif_port_changes(struct dpif_backer *);
 648 static void process_dpif_all_ports_changed(struct dpif_backer *);
 649 static void process_dpif_port_change(struct dpif_backer *,
 650                                      const char *devname);
 651 static void process_dpif_port_error(struct dpif_backer *, int error);
 652
 653 static struct ofproto_dpif *
 654 lookup_ofproto_dpif_by_port_name(const char *name)
 655 {
 656     struct ofproto_dpif *ofproto;
 657
 658     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 659         if (sset_contains(&ofproto->ports, name)) {
 660             return ofproto;
 661         }
 662     }
 663
 664     return NULL;
 665 }
 666
 667 static int
 668 type_run(const char *type)
 669 {
 670     static long long int push_timer = LLONG_MIN;
 671     struct dpif_backer *backer;
 672
 673     backer = shash_find_data(&all_dpif_backers, type);
 674     if (!backer) {
 675         /* This is not necessarily a problem, since backers are only
 676          * created on demand. */
 677         return 0;
 678     }
 679
 680     dpif_run(backer->dpif);
 681
 682     /* The most natural place to push facet statistics is when they're pulled
 683      * from the datapath.  However, when there are many flows in the datapath,
 684      * this expensive operation can occur so frequently, that it reduces our
 685      * ability to quickly set up flows.  To reduce the cost, we push statistics
 686      * here instead. */
 687     if (time_msec() > push_timer) {
 688         push_timer = time_msec() + 2000;
 689         push_all_stats();
 690     }
 691
 692     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 693      * and the configuration has now changed to "false", enable receiving
 694      * packets from the datapath. */
 695     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 696         int error;
 697
 698         backer->recv_set_enable = true;
 699
 700         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 701         if (error) {
 702             udpif_recv_set(backer->udpif, 0, false);
 703             VLOG_ERR("Failed to enable receiving packets in dpif.");
 704             return error;
 705         }
 706         udpif_recv_set(backer->udpif, n_handler_threads,
 707                        backer->recv_set_enable);
 708         dpif_flow_flush(backer->dpif);
 709         backer->need_revalidate = REV_RECONFIGURE;
 710     }
 711
 712     /* If the n_handler_threads is reconfigured, call udpif_recv_set()
 713      * to reset the handler threads. */
 714     if (backer->n_handler_threads != n_handler_threads) {
 715         udpif_recv_set(backer->udpif, n_handler_threads,
 716                        backer->recv_set_enable);
 717         backer->n_handler_threads = n_handler_threads;
 718     }
 719
 720     if (backer->need_revalidate) {
 721         struct ofproto_dpif *ofproto;
 722         struct simap_node *node;
 723         struct simap tmp_backers;
 724
 725         /* Handle tunnel garbage collection. */
 726         simap_init(&tmp_backers);
 727         simap_swap(&backer->tnl_backers, &tmp_backers);
 728
 729         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 730             struct ofport_dpif *iter;
 731
 732             if (backer != ofproto->backer) {
 733                 continue;
 734             }
 735
 736             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 737                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 738                 const char *dp_port;
 739
 740                 if (!iter->is_tunnel) {
 741                     continue;
 742                 }
 743
 744                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 745                                                      namebuf, sizeof namebuf);
 746                 node = simap_find(&tmp_backers, dp_port);
 747                 if (node) {
 748                     simap_put(&backer->tnl_backers, dp_port, node->data);
 749                     simap_delete(&tmp_backers, node);
 750                     node = simap_find(&backer->tnl_backers, dp_port);
 751                 } else {
 752                     node = simap_find(&backer->tnl_backers, dp_port);
 753                     if (!node) {
 754                         odp_port_t odp_port = ODPP_NONE;
 755
 756                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 757                                            &odp_port)) {
 758                             simap_put(&backer->tnl_backers, dp_port,
 759                                       odp_to_u32(odp_port));
 760                             node = simap_find(&backer->tnl_backers, dp_port);
 761                         }
 762                     }
 763                 }
 764
 765                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 766                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 767                                          iter->odp_port)) {
 768                     backer->need_revalidate = REV_RECONFIGURE;
 769                 }
 770             }
 771         }
 772
 773         SIMAP_FOR_EACH (node, &tmp_backers) {
 774             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 775         }
 776         simap_destroy(&tmp_backers);
 777
 778         switch (backer->need_revalidate) {
 779         case REV_RECONFIGURE:   COVERAGE_INC(rev_reconfigure);   break;
 780         case REV_STP:           COVERAGE_INC(rev_stp);           break;
 781         case REV_BOND:          COVERAGE_INC(rev_bond);          break;
 782         case REV_PORT_TOGGLED:  COVERAGE_INC(rev_port_toggled);  break;
 783         case REV_FLOW_TABLE:    COVERAGE_INC(rev_flow_table);    break;
 784         case REV_MAC_LEARNING:  COVERAGE_INC(rev_mac_learning);  break;
 785         case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
 786         }
 787         backer->need_revalidate = 0;
 788
 789         /* Clear the drop_keys in case we should now be accepting some
 790          * formerly dropped flows. */
 791         drop_key_clear(backer);
 792
 793         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 794             struct facet *facet, *next;
 795             struct ofport_dpif *ofport;
 796             struct cls_cursor cursor;
 797             struct ofbundle *bundle;
 798
 799             if (ofproto->backer != backer) {
 800                 continue;
 801             }
 802
 803             ovs_rwlock_wrlock(&xlate_rwlock);
 804             xlate_ofproto_set(ofproto, ofproto->up.name,
 805                               ofproto->backer->dpif, ofproto->miss_rule,
 806                               ofproto->no_packet_in_rule, ofproto->ml,
 807                               ofproto->stp, ofproto->mbridge,
 808                               ofproto->sflow, ofproto->ipfix,
 809                               ofproto->up.frag_handling,
 810                               ofproto->up.forward_bpdu,
 811                               connmgr_has_in_band(ofproto->up.connmgr),
 812                               ofproto->netflow != NULL);
 813
 814             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 815                 xlate_bundle_set(ofproto, bundle, bundle->name,
 816                                  bundle->vlan_mode, bundle->vlan,
 817                                  bundle->trunks, bundle->use_priority_tags,
 818                                  bundle->bond, bundle->lacp,
 819                                  bundle->floodable);
 820             }
 821
 822             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 823                 int stp_port = ofport->stp_port
 824                     ? stp_port_no(ofport->stp_port)
 825                     : 0;
 826                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 827                                  ofport->up.ofp_port, ofport->odp_port,
 828                                  ofport->up.netdev, ofport->cfm,
 829                                  ofport->bfd, ofport->peer, stp_port,
 830                                  ofport->qdscp, ofport->n_qdscp,
 831                                  ofport->up.pp.config, ofport->is_tunnel,
 832                                  ofport->may_enable);
 833             }
 834             ovs_rwlock_unlock(&xlate_rwlock);
 835
 836             /* Only ofproto-dpif cares about the facet classifier so we just
 837              * lock cls_cursor_init() to appease the thread safety analysis. */
 838             ovs_rwlock_rdlock(&ofproto->facets.rwlock);
 839             cls_cursor_init(&cursor, &ofproto->facets, NULL);
 840             ovs_rwlock_unlock(&ofproto->facets.rwlock);
 841             CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
 842                 facet_revalidate(facet);
 843                 run_fast_rl();
 844             }
 845         }
 846
 847         udpif_revalidate(backer->udpif);
 848     }
 849
 850     if (!backer->recv_set_enable) {
 851         /* Wake up before a max of 1000ms. */
 852         timer_set_duration(&backer->next_expiration, 1000);
 853     } else if (timer_expired(&backer->next_expiration)) {
 854         int delay = expire(backer);
 855         timer_set_duration(&backer->next_expiration, delay);
 856     }
 857
 858     process_dpif_port_changes(backer);
 859
 860     if (backer->governor) {
 861         size_t n_subfacets;
 862
 863         governor_run(backer->governor);
 864
 865         /* If the governor has shrunk to its minimum size and the number of
 866          * subfacets has dwindled, then drop the governor entirely.
 867          *
 868          * For hysteresis, the number of subfacets to drop the governor is
 869          * smaller than the number needed to trigger its creation. */
 870         n_subfacets = hmap_count(&backer->subfacets);
 871         if (n_subfacets * 4 < flow_eviction_threshold
 872             && governor_is_idle(backer->governor)) {
 873             governor_destroy(backer->governor);
 874             backer->governor = NULL;
 875         }
 876     }
 877
 878     return 0;
 879 }
 880
 881 /* Check for and handle port changes in 'backer''s dpif. */
 882 static void
 883 process_dpif_port_changes(struct dpif_backer *backer)
 884 {
 885     for (;;) {
 886         char *devname;
 887         int error;
 888
 889         error = dpif_port_poll(backer->dpif, &devname);
 890         switch (error) {
 891         case EAGAIN:
 892             return;
 893
 894         case ENOBUFS:
 895             process_dpif_all_ports_changed(backer);
 896             break;
 897
 898         case 0:
 899             process_dpif_port_change(backer, devname);
 900             free(devname);
 901             break;
 902
 903         default:
 904             process_dpif_port_error(backer, error);
 905             break;
 906         }
 907     }
 908 }
 909
 910 static void
 911 process_dpif_all_ports_changed(struct dpif_backer *backer)
 912 {
 913     struct ofproto_dpif *ofproto;
 914     struct dpif_port dpif_port;
 915     struct dpif_port_dump dump;
 916     struct sset devnames;
 917     const char *devname;
 918
 919     sset_init(&devnames);
 920     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 921         if (ofproto->backer == backer) {
 922             struct ofport *ofport;
 923
 924             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 925                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 926             }
 927         }
 928     }
 929     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 930         sset_add(&devnames, dpif_port.name);
 931     }
 932
 933     SSET_FOR_EACH (devname, &devnames) {
 934         process_dpif_port_change(backer, devname);
 935     }
 936     sset_destroy(&devnames);
 937 }
 938
 939 static void
 940 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 941 {
 942     struct ofproto_dpif *ofproto;
 943     struct dpif_port port;
 944
 945     /* Don't report on the datapath's device. */
 946     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 947         return;
 948     }
 949
 950     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 951                    &all_ofproto_dpifs) {
 952         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 953             return;
 954         }
 955     }
 956
 957     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 958     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 959         /* The port was removed.  If we know the datapath,
 960          * report it through poll_set().  If we don't, it may be
 961          * notifying us of a removal we initiated, so ignore it.
 962          * If there's a pending ENOBUFS, let it stand, since
 963          * everything will be reevaluated. */
 964         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 965             sset_add(&ofproto->port_poll_set, devname);
 966             ofproto->port_poll_errno = 0;
 967         }
 968     } else if (!ofproto) {
 969         /* The port was added, but we don't know with which
 970          * ofproto we should associate it.  Delete it. */
 971         dpif_port_del(backer->dpif, port.port_no);
 972     } else {
 973         struct ofport_dpif *ofport;
 974
 975         ofport = ofport_dpif_cast(shash_find_data(
 976                                       &ofproto->up.port_by_name, devname));
 977         if (ofport
 978             && ofport->odp_port != port.port_no
 979             && !odp_port_to_ofport(backer, port.port_no))
 980         {
 981             /* 'ofport''s datapath port number has changed from
 982              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 983              * structures to match. */
 984             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 985             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 986             ofport->odp_port = port.port_no;
 987             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 988                         hash_odp_port(port.port_no));
 989             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 990             backer->need_revalidate = REV_RECONFIGURE;
 991         }
 992     }
 993     dpif_port_destroy(&port);
 994 }
 995
 996 /* Propagate 'error' to all ofprotos based on 'backer'. */
 997 static void
 998 process_dpif_port_error(struct dpif_backer *backer, int error)
 999 {
1000     struct ofproto_dpif *ofproto;
1001
1002     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1003         if (ofproto->backer == backer) {
1004             sset_clear(&ofproto->port_poll_set);
1005             ofproto->port_poll_errno = error;
1006         }
1007     }
1008 }
1009
1010 static int
1011 dpif_backer_run_fast(struct dpif_backer *backer)
1012 {
1013     udpif_run(backer->udpif);
1014     handle_upcalls(backer);
1015
1016     return 0;
1017 }
1018
1019 static int
1020 type_run_fast(const char *type)
1021 {
1022     struct dpif_backer *backer;
1023
1024     backer = shash_find_data(&all_dpif_backers, type);
1025     if (!backer) {
1026         /* This is not necessarily a problem, since backers are only
1027          * created on demand. */
1028         return 0;
1029     }
1030
1031     return dpif_backer_run_fast(backer);
1032 }
1033
1034 static void
1035 run_fast_rl(void)
1036 {
1037     static long long int port_rl = LLONG_MIN;
1038
1039     if (time_msec() >= port_rl) {
1040         struct ofproto_dpif *ofproto;
1041
1042         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1043             run_fast(&ofproto->up);
1044         }
1045         port_rl = time_msec() + 200;
1046     }
1047 }
1048
1049 static void
1050 type_wait(const char *type)
1051 {
1052     struct dpif_backer *backer;
1053
1054     backer = shash_find_data(&all_dpif_backers, type);
1055     if (!backer) {
1056         /* This is not necessarily a problem, since backers are only
1057          * created on demand. */
1058         return;
1059     }
1060
1061     if (backer->governor) {
1062         governor_wait(backer->governor);
1063     }
1064
1065     timer_wait(&backer->next_expiration);
1066     dpif_wait(backer->dpif);
1067     udpif_wait(backer->udpif);
1068 }
1069 \f
1070 /* Basic life-cycle. */
1071
1072 static int add_internal_flows(struct ofproto_dpif *);
1073
1074 static struct ofproto *
1075 alloc(void)
1076 {
1077     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1078     return &ofproto->up;
1079 }
1080
1081 static void
1082 dealloc(struct ofproto *ofproto_)
1083 {
1084     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1085     free(ofproto);
1086 }
1087
1088 static void
1089 close_dpif_backer(struct dpif_backer *backer)
1090 {
1091     struct shash_node *node;
1092
1093     ovs_assert(backer->refcount > 0);
1094
1095     if (--backer->refcount) {
1096         return;
1097     }
1098
1099     drop_key_clear(backer);
1100     hmap_destroy(&backer->drop_keys);
1101
1102     simap_destroy(&backer->tnl_backers);
1103     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1104     hmap_destroy(&backer->odp_to_ofport_map);
1105     node = shash_find(&all_dpif_backers, backer->type);
1106     free(backer->type);
1107     shash_delete(&all_dpif_backers, node);
1108     udpif_destroy(backer->udpif);
1109     dpif_close(backer->dpif);
1110
1111     ovs_assert(hmap_is_empty(&backer->subfacets));
1112     hmap_destroy(&backer->subfacets);
1113     governor_destroy(backer->governor);
1114
1115     free(backer);
1116 }
1117
1118 /* Datapath port slated for removal from datapath. */
1119 struct odp_garbage {
1120     struct list list_node;
1121     odp_port_t odp_port;
1122 };
1123
1124 static int
1125 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1126 {
1127     struct dpif_backer *backer;
1128     struct dpif_port_dump port_dump;
1129     struct dpif_port port;
1130     struct shash_node *node;
1131     struct list garbage_list;
1132     struct odp_garbage *garbage, *next;
1133     struct sset names;
1134     char *backer_name;
1135     const char *name;
1136     int error;
1137
1138     backer = shash_find_data(&all_dpif_backers, type);
1139     if (backer) {
1140         backer->refcount++;
1141         *backerp = backer;
1142         return 0;
1143     }
1144
1145     backer_name = xasprintf("ovs-%s", type);
1146
1147     /* Remove any existing datapaths, since we assume we're the only
1148      * userspace controlling the datapath. */
1149     sset_init(&names);
1150     dp_enumerate_names(type, &names);
1151     SSET_FOR_EACH(name, &names) {
1152         struct dpif *old_dpif;
1153
1154         /* Don't remove our backer if it exists. */
1155         if (!strcmp(name, backer_name)) {
1156             continue;
1157         }
1158
1159         if (dpif_open(name, type, &old_dpif)) {
1160             VLOG_WARN("couldn't open old datapath %s to remove it", name);
1161         } else {
1162             dpif_delete(old_dpif);
1163             dpif_close(old_dpif);
1164         }
1165     }
1166     sset_destroy(&names);
1167
1168     backer = xmalloc(sizeof *backer);
1169
1170     error = dpif_create_and_open(backer_name, type, &backer->dpif);
1171     free(backer_name);
1172     if (error) {
1173         VLOG_ERR("failed to open datapath of type %s: %s", type,
1174                  ovs_strerror(error));
1175         free(backer);
1176         return error;
1177     }
1178     backer->udpif = udpif_create(backer, backer->dpif);
1179
1180     backer->type = xstrdup(type);
1181     backer->governor = NULL;
1182     backer->refcount = 1;
1183     hmap_init(&backer->odp_to_ofport_map);
1184     ovs_rwlock_init(&backer->odp_to_ofport_lock);
1185     hmap_init(&backer->drop_keys);
1186     hmap_init(&backer->subfacets);
1187     timer_set_duration(&backer->next_expiration, 1000);
1188     backer->need_revalidate = 0;
1189     simap_init(&backer->tnl_backers);
1190     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1191     *backerp = backer;
1192
1193     if (backer->recv_set_enable) {
1194         dpif_flow_flush(backer->dpif);
1195     }
1196
1197     /* Loop through the ports already on the datapath and remove any
1198      * that we don't need anymore. */
1199     list_init(&garbage_list);
1200     dpif_port_dump_start(&port_dump, backer->dpif);
1201     while (dpif_port_dump_next(&port_dump, &port)) {
1202         node = shash_find(&init_ofp_ports, port.name);
1203         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1204             garbage = xmalloc(sizeof *garbage);
1205             garbage->odp_port = port.port_no;
1206             list_push_front(&garbage_list, &garbage->list_node);
1207         }
1208     }
1209     dpif_port_dump_done(&port_dump);
1210
1211     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1212         dpif_port_del(backer->dpif, garbage->odp_port);
1213         list_remove(&garbage->list_node);
1214         free(garbage);
1215     }
1216
1217     shash_add(&all_dpif_backers, type, backer);
1218
1219     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1220     if (error) {
1221         VLOG_ERR("failed to listen on datapath of type %s: %s",
1222                  type, ovs_strerror(error));
1223         close_dpif_backer(backer);
1224         return error;
1225     }
1226     udpif_recv_set(backer->udpif, n_handler_threads,
1227                    backer->recv_set_enable);
1228     backer->n_handler_threads = n_handler_threads;
1229
1230     backer->max_n_subfacet = 0;
1231     backer->created = time_msec();
1232     backer->last_minute = backer->created;
1233     memset(&backer->hourly, 0, sizeof backer->hourly);
1234     memset(&backer->daily, 0, sizeof backer->daily);
1235     memset(&backer->lifetime, 0, sizeof backer->lifetime);
1236     backer->subfacet_add_count = 0;
1237     backer->subfacet_del_count = 0;
1238     backer->total_subfacet_add_count = 0;
1239     backer->total_subfacet_del_count = 0;
1240     backer->avg_n_subfacet = 0;
1241     backer->avg_subfacet_life = 0;
1242
1243     return error;
1244 }
1245
1246 static int
1247 construct(struct ofproto *ofproto_)
1248 {
1249     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1250     struct shash_node *node, *next;
1251     uint32_t max_ports;
1252     int error;
1253
1254     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1255     if (error) {
1256         return error;
1257     }
1258
1259     max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1260     ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX)));
1261
1262     ofproto->netflow = NULL;
1263     ofproto->sflow = NULL;
1264     ofproto->ipfix = NULL;
1265     ofproto->stp = NULL;
1266     hmap_init(&ofproto->bundles);
1267     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1268     ofproto->mbridge = mbridge_create();
1269     ofproto->has_bonded_bundles = false;
1270     ovs_mutex_init(&ofproto->vsp_mutex);
1271
1272     classifier_init(&ofproto->facets);
1273     ofproto->consistency_rl = LLONG_MIN;
1274
1275     list_init(&ofproto->completions);
1276
1277     ovs_mutex_init(&ofproto->flow_mod_mutex);
1278     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1279     list_init(&ofproto->flow_mods);
1280     ofproto->n_flow_mods = 0;
1281     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1282
1283     ovs_mutex_init(&ofproto->pin_mutex);
1284     ovs_mutex_lock(&ofproto->pin_mutex);
1285     list_init(&ofproto->pins);
1286     ofproto->n_pins = 0;
1287     ovs_mutex_unlock(&ofproto->pin_mutex);
1288
1289     ofproto_dpif_unixctl_init();
1290
1291     hmap_init(&ofproto->vlandev_map);
1292     hmap_init(&ofproto->realdev_vid_map);
1293
1294     sset_init(&ofproto->ports);
1295     sset_init(&ofproto->ghost_ports);
1296     sset_init(&ofproto->port_poll_set);
1297     ofproto->port_poll_errno = 0;
1298
1299     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1300         struct iface_hint *iface_hint = node->data;
1301
1302         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1303             /* Check if the datapath already has this port. */
1304             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1305                 sset_add(&ofproto->ports, node->name);
1306             }
1307
1308             free(iface_hint->br_name);
1309             free(iface_hint->br_type);
1310             free(iface_hint);
1311             shash_delete(&init_ofp_ports, node);
1312         }
1313     }
1314
1315     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1316                 hash_string(ofproto->up.name, 0));
1317     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1318
1319     ofproto_init_tables(ofproto_, N_TABLES);
1320     error = add_internal_flows(ofproto);
1321     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1322
1323     ofproto->n_hit = 0;
1324     ofproto->n_missed = 0;
1325
1326     return error;
1327 }
1328
1329 static int
1330 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1331                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1332 {
1333     struct ofputil_flow_mod fm;
1334     int error;
1335
1336     match_init_catchall(&fm.match);
1337     fm.priority = 0;
1338     match_set_reg(&fm.match, 0, id);
1339     fm.new_cookie = htonll(0);
1340     fm.cookie = htonll(0);
1341     fm.cookie_mask = htonll(0);
1342     fm.modify_cookie = false;
1343     fm.table_id = TBL_INTERNAL;
1344     fm.command = OFPFC_ADD;
1345     fm.idle_timeout = 0;
1346     fm.hard_timeout = 0;
1347     fm.buffer_id = 0;
1348     fm.out_port = 0;
1349     fm.flags = 0;
1350     fm.ofpacts = ofpacts->data;
1351     fm.ofpacts_len = ofpacts->size;
1352
1353     error = ofproto_flow_mod(&ofproto->up, &fm);
1354     if (error) {
1355         VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1356                     id, ofperr_to_string(error));
1357         return error;
1358     }
1359
1360     if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1361                                   rulep)) {
1362         ovs_rwlock_unlock(&(*rulep)->up.evict);
1363     } else {
1364         NOT_REACHED();
1365     }
1366
1367     return 0;
1368 }
1369
1370 static int
1371 add_internal_flows(struct ofproto_dpif *ofproto)
1372 {
1373     struct ofpact_controller *controller;
1374     uint64_t ofpacts_stub[128 / 8];
1375     struct ofpbuf ofpacts;
1376     int error;
1377     int id;
1378
1379     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1380     id = 1;
1381
1382     controller = ofpact_put_CONTROLLER(&ofpacts);
1383     controller->max_len = UINT16_MAX;
1384     controller->controller_id = 0;
1385     controller->reason = OFPR_NO_MATCH;
1386     ofpact_pad(&ofpacts);
1387
1388     error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1389     if (error) {
1390         return error;
1391     }
1392
1393     ofpbuf_clear(&ofpacts);
1394     error = add_internal_flow(ofproto, id++, &ofpacts,
1395                               &ofproto->no_packet_in_rule);
1396     if (error) {
1397         return error;
1398     }
1399
1400     error = add_internal_flow(ofproto, id++, &ofpacts,
1401                               &ofproto->drop_frags_rule);
1402     return error;
1403 }
1404
1405 static void
1406 complete_operations(struct ofproto_dpif *ofproto)
1407 {
1408     struct dpif_completion *c, *next;
1409
1410     LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1411         ofoperation_complete(c->op, 0);
1412         list_remove(&c->list_node);
1413         free(c);
1414     }
1415 }
1416
1417 static void
1418 destruct(struct ofproto *ofproto_)
1419 {
1420     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1421     struct rule_dpif *rule, *next_rule;
1422     struct ofputil_packet_in *pin, *next_pin;
1423     struct ofputil_flow_mod *fm, *next_fm;
1424     struct facet *facet, *next_facet;
1425     struct cls_cursor cursor;
1426     struct oftable *table;
1427
1428     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1429     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1430     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1431     CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1432         facet_remove(facet);
1433     }
1434
1435     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1436     ovs_rwlock_wrlock(&xlate_rwlock);
1437     xlate_remove_ofproto(ofproto);
1438     ovs_rwlock_unlock(&xlate_rwlock);
1439
1440     flow_miss_batch_ofproto_destroyed(ofproto->backer->udpif, ofproto);
1441
1442     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1443     complete_operations(ofproto);
1444
1445     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1446         struct cls_cursor cursor;
1447
1448         ovs_rwlock_wrlock(&table->cls.rwlock);
1449         cls_cursor_init(&cursor, &table->cls, NULL);
1450         CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1451             ofproto_rule_delete(&ofproto->up, &table->cls, &rule->up);
1452         }
1453         ovs_rwlock_unlock(&table->cls.rwlock);
1454     }
1455     complete_operations(ofproto);
1456
1457     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1458     LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &ofproto->flow_mods) {
1459         list_remove(&fm->list_node);
1460         ofproto->n_flow_mods--;
1461         free(fm->ofpacts);
1462         free(fm);
1463     }
1464     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1465     ovs_mutex_destroy(&ofproto->flow_mod_mutex);
1466
1467     ovs_mutex_lock(&ofproto->pin_mutex);
1468     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &ofproto->pins) {
1469         list_remove(&pin->list_node);
1470         ofproto->n_pins--;
1471         free(CONST_CAST(void *, pin->packet));
1472         free(pin);
1473     }
1474     ovs_mutex_unlock(&ofproto->pin_mutex);
1475     ovs_mutex_destroy(&ofproto->pin_mutex);
1476
1477     mbridge_unref(ofproto->mbridge);
1478
1479     netflow_destroy(ofproto->netflow);
1480     dpif_sflow_unref(ofproto->sflow);
1481     hmap_destroy(&ofproto->bundles);
1482     mac_learning_unref(ofproto->ml);
1483
1484     classifier_destroy(&ofproto->facets);
1485
1486     hmap_destroy(&ofproto->vlandev_map);
1487     hmap_destroy(&ofproto->realdev_vid_map);
1488
1489     sset_destroy(&ofproto->ports);
1490     sset_destroy(&ofproto->ghost_ports);
1491     sset_destroy(&ofproto->port_poll_set);
1492
1493     ovs_mutex_destroy(&ofproto->vsp_mutex);
1494
1495     close_dpif_backer(ofproto->backer);
1496 }
1497
1498 static int
1499 run_fast(struct ofproto *ofproto_)
1500 {
1501     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1502     struct ofputil_packet_in *pin, *next_pin;
1503     struct ofputil_flow_mod *fm, *next_fm;
1504     struct list flow_mods, pins;
1505     struct ofport_dpif *ofport;
1506
1507     /* Do not perform any periodic activity required by 'ofproto' while
1508      * waiting for flow restore to complete. */
1509     if (ofproto_get_flow_restore_wait()) {
1510         return 0;
1511     }
1512
1513     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1514     list_move(&flow_mods, &ofproto->flow_mods);
1515     list_init(&ofproto->flow_mods);
1516     ofproto->n_flow_mods = 0;
1517     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1518
1519     LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &flow_mods) {
1520         int error = ofproto_flow_mod(&ofproto->up, fm);
1521         if (error && !VLOG_DROP_WARN(&rl)) {
1522             VLOG_WARN("learning action failed to modify flow table (%s)",
1523                       ofperr_get_name(error));
1524         }
1525
1526         list_remove(&fm->list_node);
1527         free(fm->ofpacts);
1528         free(fm);
1529     }
1530
1531     ovs_mutex_lock(&ofproto->pin_mutex);
1532     list_move(&pins, &ofproto->pins);
1533     list_init(&ofproto->pins);
1534     ofproto->n_pins = 0;
1535     ovs_mutex_unlock(&ofproto->pin_mutex);
1536
1537     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1538         connmgr_send_packet_in(ofproto->up.connmgr, pin);
1539         list_remove(&pin->list_node);
1540         free(CONST_CAST(void *, pin->packet));
1541         free(pin);
1542     }
1543
1544     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1545         port_run_fast(ofport);
1546     }
1547
1548     return 0;
1549 }
1550
1551 static int
1552 run(struct ofproto *ofproto_)
1553 {
1554     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1555     struct ofport_dpif *ofport;
1556     struct ofbundle *bundle;
1557     int error;
1558
1559     if (!clogged) {
1560         complete_operations(ofproto);
1561     }
1562
1563     if (mbridge_need_revalidate(ofproto->mbridge)) {
1564         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1565         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1566         mac_learning_flush(ofproto->ml);
1567         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1568     }
1569
1570     /* Do not perform any periodic activity below required by 'ofproto' while
1571      * waiting for flow restore to complete. */
1572     if (ofproto_get_flow_restore_wait()) {
1573         return 0;
1574     }
1575
1576     error = run_fast(ofproto_);
1577     if (error) {
1578         return error;
1579     }
1580
1581     if (ofproto->netflow) {
1582         if (netflow_run(ofproto->netflow)) {
1583             send_netflow_active_timeouts(ofproto);
1584         }
1585     }
1586     if (ofproto->sflow) {
1587         dpif_sflow_run(ofproto->sflow);
1588     }
1589     if (ofproto->ipfix) {
1590         dpif_ipfix_run(ofproto->ipfix);
1591     }
1592
1593     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1594         port_run(ofport);
1595     }
1596     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1597         bundle_run(bundle);
1598     }
1599
1600     stp_run(ofproto);
1601     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1602     if (mac_learning_run(ofproto->ml)) {
1603         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1604     }
1605     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1606
1607     /* Check the consistency of a random facet, to aid debugging. */
1608     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1609     if (time_msec() >= ofproto->consistency_rl
1610         && !classifier_is_empty(&ofproto->facets)
1611         && !ofproto->backer->need_revalidate) {
1612         struct cls_table *table;
1613         struct cls_rule *cr;
1614         struct facet *facet;
1615
1616         ofproto->consistency_rl = time_msec() + 250;
1617
1618         table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1619                              struct cls_table, hmap_node);
1620         cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1621                           hmap_node);
1622         facet = CONTAINER_OF(cr, struct facet, cr);
1623
1624         if (!facet_check_consistency(facet)) {
1625             ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1626         }
1627     }
1628     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1629
1630     return 0;
1631 }
1632
1633 static void
1634 wait(struct ofproto *ofproto_)
1635 {
1636     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1637     struct ofport_dpif *ofport;
1638     struct ofbundle *bundle;
1639
1640     if (!clogged && !list_is_empty(&ofproto->completions)) {
1641         poll_immediate_wake();
1642     }
1643
1644     if (ofproto_get_flow_restore_wait()) {
1645         return;
1646     }
1647
1648     if (ofproto->sflow) {
1649         dpif_sflow_wait(ofproto->sflow);
1650     }
1651     if (ofproto->ipfix) {
1652         dpif_ipfix_wait(ofproto->ipfix);
1653     }
1654     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1655         port_wait(ofport);
1656     }
1657     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1658         bundle_wait(bundle);
1659     }
1660     if (ofproto->netflow) {
1661         netflow_wait(ofproto->netflow);
1662     }
1663     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1664     mac_learning_wait(ofproto->ml);
1665     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1666     stp_wait(ofproto);
1667     if (ofproto->backer->need_revalidate) {
1668         /* Shouldn't happen, but if it does just go around again. */
1669         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1670         poll_immediate_wake();
1671     }
1672 }
1673
1674 static void
1675 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1676 {
1677     const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1678     struct cls_cursor cursor;
1679     size_t n_subfacets = 0;
1680     struct facet *facet;
1681
1682     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1683     simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1684     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1685
1686     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1687     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1688     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1689         n_subfacets += list_size(&facet->subfacets);
1690     }
1691     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1692     simap_increase(usage, "subfacets", n_subfacets);
1693 }
1694
1695 static void
1696 flush(struct ofproto *ofproto_)
1697 {
1698     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1699     struct subfacet *subfacet, *next_subfacet;
1700     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1701     int n_batch;
1702
1703     n_batch = 0;
1704     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1705                         &ofproto->backer->subfacets) {
1706         if (subfacet->facet->ofproto != ofproto) {
1707             continue;
1708         }
1709
1710         if (subfacet->path != SF_NOT_INSTALLED) {
1711             batch[n_batch++] = subfacet;
1712             if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1713                 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1714                 n_batch = 0;
1715             }
1716         } else {
1717             subfacet_destroy(subfacet);
1718         }
1719     }
1720
1721     if (n_batch > 0) {
1722         subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1723     }
1724 }
1725
1726 static void
1727 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1728              bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1729 {
1730     *arp_match_ip = true;
1731     *actions = (OFPUTIL_A_OUTPUT |
1732                 OFPUTIL_A_SET_VLAN_VID |
1733                 OFPUTIL_A_SET_VLAN_PCP |
1734                 OFPUTIL_A_STRIP_VLAN |
1735                 OFPUTIL_A_SET_DL_SRC |
1736                 OFPUTIL_A_SET_DL_DST |
1737                 OFPUTIL_A_SET_NW_SRC |
1738                 OFPUTIL_A_SET_NW_DST |
1739                 OFPUTIL_A_SET_NW_TOS |
1740                 OFPUTIL_A_SET_TP_SRC |
1741                 OFPUTIL_A_SET_TP_DST |
1742                 OFPUTIL_A_ENQUEUE);
1743 }
1744
1745 static void
1746 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1747 {
1748     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1749     struct dpif_dp_stats s;
1750     uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1751     uint64_t n_lookup;
1752
1753     strcpy(ots->name, "classifier");
1754
1755     dpif_get_dp_stats(ofproto->backer->dpif, &s);
1756     rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1757     rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1758     rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1759
1760     n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1761     ots->lookup_count = htonll(n_lookup);
1762     ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1763 }
1764
1765 static struct ofport *
1766 port_alloc(void)
1767 {
1768     struct ofport_dpif *port = xmalloc(sizeof *port);
1769     return &port->up;
1770 }
1771
1772 static void
1773 port_dealloc(struct ofport *port_)
1774 {
1775     struct ofport_dpif *port = ofport_dpif_cast(port_);
1776     free(port);
1777 }
1778
1779 static int
1780 port_construct(struct ofport *port_)
1781 {
1782     struct ofport_dpif *port = ofport_dpif_cast(port_);
1783     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1784     const struct netdev *netdev = port->up.netdev;
1785     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1786     struct dpif_port dpif_port;
1787     int error;
1788
1789     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1790     port->bundle = NULL;
1791     port->cfm = NULL;
1792     port->bfd = NULL;
1793     port->may_enable = true;
1794     port->stp_port = NULL;
1795     port->stp_state = STP_DISABLED;
1796     port->is_tunnel = false;
1797     port->peer = NULL;
1798     port->qdscp = NULL;
1799     port->n_qdscp = 0;
1800     port->realdev_ofp_port = 0;
1801     port->vlandev_vid = 0;
1802     port->carrier_seq = netdev_get_carrier_resets(netdev);
1803
1804     if (netdev_vport_is_patch(netdev)) {
1805         /* By bailing out here, we don't submit the port to the sFlow module
1806          * to be considered for counter polling export.  This is correct
1807          * because the patch port represents an interface that sFlow considers
1808          * to be "internal" to the switch as a whole, and therefore not an
1809          * candidate for counter polling. */
1810         port->odp_port = ODPP_NONE;
1811         ofport_update_peer(port);
1812         return 0;
1813     }
1814
1815     error = dpif_port_query_by_name(ofproto->backer->dpif,
1816                                     netdev_vport_get_dpif_port(netdev, namebuf,
1817                                                                sizeof namebuf),
1818                                     &dpif_port);
1819     if (error) {
1820         return error;
1821     }
1822
1823     port->odp_port = dpif_port.port_no;
1824
1825     if (netdev_get_tunnel_config(netdev)) {
1826         tnl_port_add(port, port->up.netdev, port->odp_port);
1827         port->is_tunnel = true;
1828     } else {
1829         /* Sanity-check that a mapping doesn't already exist.  This
1830          * shouldn't happen for non-tunnel ports. */
1831         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1832             VLOG_ERR("port %s already has an OpenFlow port number",
1833                      dpif_port.name);
1834             dpif_port_destroy(&dpif_port);
1835             return EBUSY;
1836         }
1837
1838         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1839         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1840                     hash_odp_port(port->odp_port));
1841         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1842     }
1843     dpif_port_destroy(&dpif_port);
1844
1845     if (ofproto->sflow) {
1846         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1847     }
1848
1849     return 0;
1850 }
1851
1852 static void
1853 port_destruct(struct ofport *port_)
1854 {
1855     struct ofport_dpif *port = ofport_dpif_cast(port_);
1856     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1857     const char *devname = netdev_get_name(port->up.netdev);
1858     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1859     const char *dp_port_name;
1860
1861     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1862     ovs_rwlock_wrlock(&xlate_rwlock);
1863     xlate_ofport_remove(port);
1864     ovs_rwlock_unlock(&xlate_rwlock);
1865
1866     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1867                                               sizeof namebuf);
1868     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1869         /* The underlying device is still there, so delete it.  This
1870          * happens when the ofproto is being destroyed, since the caller
1871          * assumes that removal of attached ports will happen as part of
1872          * destruction. */
1873         if (!port->is_tunnel) {
1874             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1875         }
1876     }
1877
1878     if (port->peer) {
1879         port->peer->peer = NULL;
1880         port->peer = NULL;
1881     }
1882
1883     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1884         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1885         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1886         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1887     }
1888
1889     tnl_port_del(port);
1890     sset_find_and_delete(&ofproto->ports, devname);
1891     sset_find_and_delete(&ofproto->ghost_ports, devname);
1892     bundle_remove(port_);
1893     set_cfm(port_, NULL);
1894     set_bfd(port_, NULL);
1895     if (ofproto->sflow) {
1896         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1897     }
1898
1899     free(port->qdscp);
1900 }
1901
1902 static void
1903 port_modified(struct ofport *port_)
1904 {
1905     struct ofport_dpif *port = ofport_dpif_cast(port_);
1906
1907     if (port->bundle && port->bundle->bond) {
1908         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1909     }
1910
1911     if (port->cfm) {
1912         cfm_set_netdev(port->cfm, port->up.netdev);
1913     }
1914
1915     if (port->bfd) {
1916         bfd_set_netdev(port->bfd, port->up.netdev);
1917     }
1918
1919     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1920                                                 port->odp_port)) {
1921         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1922             REV_RECONFIGURE;
1923     }
1924
1925     ofport_update_peer(port);
1926 }
1927
1928 static void
1929 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1930 {
1931     struct ofport_dpif *port = ofport_dpif_cast(port_);
1932     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1933     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1934
1935     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1936                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1937                    OFPUTIL_PC_NO_PACKET_IN)) {
1938         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1939
1940         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1941             bundle_update(port->bundle);
1942         }
1943     }
1944 }
1945
1946 static int
1947 set_sflow(struct ofproto *ofproto_,
1948           const struct ofproto_sflow_options *sflow_options)
1949 {
1950     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1951     struct dpif_sflow *ds = ofproto->sflow;
1952
1953     if (sflow_options) {
1954         if (!ds) {
1955             struct ofport_dpif *ofport;
1956
1957             ds = ofproto->sflow = dpif_sflow_create();
1958             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1959                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1960             }
1961             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1962         }
1963         dpif_sflow_set_options(ds, sflow_options);
1964     } else {
1965         if (ds) {
1966             dpif_sflow_unref(ds);
1967             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1968             ofproto->sflow = NULL;
1969         }
1970     }
1971     return 0;
1972 }
1973
1974 static int
1975 set_ipfix(
1976     struct ofproto *ofproto_,
1977     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1978     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1979     size_t n_flow_exporters_options)
1980 {
1981     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1982     struct dpif_ipfix *di = ofproto->ipfix;
1983     bool has_options = bridge_exporter_options || flow_exporters_options;
1984
1985     if (has_options && !di) {
1986         di = ofproto->ipfix = dpif_ipfix_create();
1987     }
1988
1989     if (di) {
1990         /* Call set_options in any case to cleanly flush the flow
1991          * caches in the last exporters that are to be destroyed. */
1992         dpif_ipfix_set_options(
1993             di, bridge_exporter_options, flow_exporters_options,
1994             n_flow_exporters_options);
1995
1996         if (!has_options) {
1997             dpif_ipfix_unref(di);
1998             ofproto->ipfix = NULL;
1999         }
2000     }
2001
2002     return 0;
2003 }
2004
2005 static int
2006 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
2007 {
2008     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2009     int error;
2010
2011     if (!s) {
2012         error = 0;
2013     } else {
2014         if (!ofport->cfm) {
2015             struct ofproto_dpif *ofproto;
2016
2017             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2018             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2019             ofport->cfm = cfm_create(ofport->up.netdev);
2020         }
2021
2022         if (cfm_configure(ofport->cfm, s)) {
2023             return 0;
2024         }
2025
2026         error = EINVAL;
2027     }
2028     cfm_unref(ofport->cfm);
2029     ofport->cfm = NULL;
2030     return error;
2031 }
2032
2033 static bool
2034 get_cfm_status(const struct ofport *ofport_,
2035                struct ofproto_cfm_status *status)
2036 {
2037     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2038
2039     if (ofport->cfm) {
2040         status->faults = cfm_get_fault(ofport->cfm);
2041         status->remote_opstate = cfm_get_opup(ofport->cfm);
2042         status->health = cfm_get_health(ofport->cfm);
2043         cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2044         return true;
2045     } else {
2046         return false;
2047     }
2048 }
2049
2050 static int
2051 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2052 {
2053     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2054     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2055     struct bfd *old;
2056
2057     old = ofport->bfd;
2058     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2059                                 cfg, ofport->up.netdev);
2060     if (ofport->bfd != old) {
2061         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2062     }
2063
2064     return 0;
2065 }
2066
2067 static int
2068 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2069 {
2070     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2071
2072     if (ofport->bfd) {
2073         bfd_get_status(ofport->bfd, smap);
2074         return 0;
2075     } else {
2076         return ENOENT;
2077     }
2078 }
2079 \f
2080 /* Spanning Tree. */
2081
2082 static void
2083 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2084 {
2085     struct ofproto_dpif *ofproto = ofproto_;
2086     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2087     struct ofport_dpif *ofport;
2088
2089     ofport = stp_port_get_aux(sp);
2090     if (!ofport) {
2091         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2092                      ofproto->up.name, port_num);
2093     } else {
2094         struct eth_header *eth = pkt->l2;
2095
2096         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2097         if (eth_addr_is_zero(eth->eth_src)) {
2098             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2099                          "with unknown MAC", ofproto->up.name, port_num);
2100         } else {
2101             send_packet(ofport, pkt);
2102         }
2103     }
2104     ofpbuf_delete(pkt);
2105 }
2106
2107 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2108 static int
2109 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2110 {
2111     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2112
2113     /* Only revalidate flows if the configuration changed. */
2114     if (!s != !ofproto->stp) {
2115         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2116     }
2117
2118     if (s) {
2119         if (!ofproto->stp) {
2120             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2121                                       send_bpdu_cb, ofproto);
2122             ofproto->stp_last_tick = time_msec();
2123         }
2124
2125         stp_set_bridge_id(ofproto->stp, s->system_id);
2126         stp_set_bridge_priority(ofproto->stp, s->priority);
2127         stp_set_hello_time(ofproto->stp, s->hello_time);
2128         stp_set_max_age(ofproto->stp, s->max_age);
2129         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2130     }  else {
2131         struct ofport *ofport;
2132
2133         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2134             set_stp_port(ofport, NULL);
2135         }
2136
2137         stp_unref(ofproto->stp);
2138         ofproto->stp = NULL;
2139     }
2140
2141     return 0;
2142 }
2143
2144 static int
2145 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2146 {
2147     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2148
2149     if (ofproto->stp) {
2150         s->enabled = true;
2151         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2152         s->designated_root = stp_get_designated_root(ofproto->stp);
2153         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2154     } else {
2155         s->enabled = false;
2156     }
2157
2158     return 0;
2159 }
2160
2161 static void
2162 update_stp_port_state(struct ofport_dpif *ofport)
2163 {
2164     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2165     enum stp_state state;
2166
2167     /* Figure out new state. */
2168     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2169                              : STP_DISABLED;
2170
2171     /* Update state. */
2172     if (ofport->stp_state != state) {
2173         enum ofputil_port_state of_state;
2174         bool fwd_change;
2175
2176         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2177                     netdev_get_name(ofport->up.netdev),
2178                     stp_state_name(ofport->stp_state),
2179                     stp_state_name(state));
2180         if (stp_learn_in_state(ofport->stp_state)
2181                 != stp_learn_in_state(state)) {
2182             /* xxx Learning action flows should also be flushed. */
2183             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2184             mac_learning_flush(ofproto->ml);
2185             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2186         }
2187         fwd_change = stp_forward_in_state(ofport->stp_state)
2188                         != stp_forward_in_state(state);
2189
2190         ofproto->backer->need_revalidate = REV_STP;
2191         ofport->stp_state = state;
2192         ofport->stp_state_entered = time_msec();
2193
2194         if (fwd_change && ofport->bundle) {
2195             bundle_update(ofport->bundle);
2196         }
2197
2198         /* Update the STP state bits in the OpenFlow port description. */
2199         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2200         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2201                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2202                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2203                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2204                      : 0);
2205         ofproto_port_set_state(&ofport->up, of_state);
2206     }
2207 }
2208
2209 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2210  * caller is responsible for assigning STP port numbers and ensuring
2211  * there are no duplicates. */
2212 static int
2213 set_stp_port(struct ofport *ofport_,
2214              const struct ofproto_port_stp_settings *s)
2215 {
2216     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2217     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2218     struct stp_port *sp = ofport->stp_port;
2219
2220     if (!s || !s->enable) {
2221         if (sp) {
2222             ofport->stp_port = NULL;
2223             stp_port_disable(sp);
2224             update_stp_port_state(ofport);
2225         }
2226         return 0;
2227     } else if (sp && stp_port_no(sp) != s->port_num
2228             && ofport == stp_port_get_aux(sp)) {
2229         /* The port-id changed, so disable the old one if it's not
2230          * already in use by another port. */
2231         stp_port_disable(sp);
2232     }
2233
2234     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2235     stp_port_enable(sp);
2236
2237     stp_port_set_aux(sp, ofport);
2238     stp_port_set_priority(sp, s->priority);
2239     stp_port_set_path_cost(sp, s->path_cost);
2240
2241     update_stp_port_state(ofport);
2242
2243     return 0;
2244 }
2245
2246 static int
2247 get_stp_port_status(struct ofport *ofport_,
2248                     struct ofproto_port_stp_status *s)
2249 {
2250     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2251     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2252     struct stp_port *sp = ofport->stp_port;
2253
2254     if (!ofproto->stp || !sp) {
2255         s->enabled = false;
2256         return 0;
2257     }
2258
2259     s->enabled = true;
2260     s->port_id = stp_port_get_id(sp);
2261     s->state = stp_port_get_state(sp);
2262     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2263     s->role = stp_port_get_role(sp);
2264     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2265
2266     return 0;
2267 }
2268
2269 static void
2270 stp_run(struct ofproto_dpif *ofproto)
2271 {
2272     if (ofproto->stp) {
2273         long long int now = time_msec();
2274         long long int elapsed = now - ofproto->stp_last_tick;
2275         struct stp_port *sp;
2276
2277         if (elapsed > 0) {
2278             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2279             ofproto->stp_last_tick = now;
2280         }
2281         while (stp_get_changed_port(ofproto->stp, &sp)) {
2282             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2283
2284             if (ofport) {
2285                 update_stp_port_state(ofport);
2286             }
2287         }
2288
2289         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2290             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2291             mac_learning_flush(ofproto->ml);
2292             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2293         }
2294     }
2295 }
2296
2297 static void
2298 stp_wait(struct ofproto_dpif *ofproto)
2299 {
2300     if (ofproto->stp) {
2301         poll_timer_wait(1000);
2302     }
2303 }
2304 \f
2305 static int
2306 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2307            size_t n_qdscp)
2308 {
2309     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2310     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2311
2312     if (ofport->n_qdscp != n_qdscp
2313         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2314                               n_qdscp * sizeof *qdscp))) {
2315         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2316         free(ofport->qdscp);
2317         ofport->qdscp = n_qdscp
2318             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2319             : NULL;
2320         ofport->n_qdscp = n_qdscp;
2321     }
2322
2323     return 0;
2324 }
2325 \f
2326 /* Bundles. */
2327
2328 /* Expires all MAC learning entries associated with 'bundle' and forces its
2329  * ofproto to revalidate every flow.
2330  *
2331  * Normally MAC learning entries are removed only from the ofproto associated
2332  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2333  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2334  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2335  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2336  * with the host from which it migrated. */
2337 static void
2338 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2339 {
2340     struct ofproto_dpif *ofproto = bundle->ofproto;
2341     struct mac_learning *ml = ofproto->ml;
2342     struct mac_entry *mac, *next_mac;
2343
2344     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2345     ovs_rwlock_wrlock(&ml->rwlock);
2346     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2347         if (mac->port.p == bundle) {
2348             if (all_ofprotos) {
2349                 struct ofproto_dpif *o;
2350
2351                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2352                     if (o != ofproto) {
2353                         struct mac_entry *e;
2354
2355                         ovs_rwlock_wrlock(&o->ml->rwlock);
2356                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2357                         if (e) {
2358                             mac_learning_expire(o->ml, e);
2359                         }
2360                         ovs_rwlock_unlock(&o->ml->rwlock);
2361                     }
2362                 }
2363             }
2364
2365             mac_learning_expire(ml, mac);
2366         }
2367     }
2368     ovs_rwlock_unlock(&ml->rwlock);
2369 }
2370
2371 static struct ofbundle *
2372 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2373 {
2374     struct ofbundle *bundle;
2375
2376     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2377                              &ofproto->bundles) {
2378         if (bundle->aux == aux) {
2379             return bundle;
2380         }
2381     }
2382     return NULL;
2383 }
2384
2385 static void
2386 bundle_update(struct ofbundle *bundle)
2387 {
2388     struct ofport_dpif *port;
2389
2390     bundle->floodable = true;
2391     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2392         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2393             || !stp_forward_in_state(port->stp_state)) {
2394             bundle->floodable = false;
2395             break;
2396         }
2397     }
2398 }
2399
2400 static void
2401 bundle_del_port(struct ofport_dpif *port)
2402 {
2403     struct ofbundle *bundle = port->bundle;
2404
2405     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2406
2407     list_remove(&port->bundle_node);
2408     port->bundle = NULL;
2409
2410     if (bundle->lacp) {
2411         lacp_slave_unregister(bundle->lacp, port);
2412     }
2413     if (bundle->bond) {
2414         bond_slave_unregister(bundle->bond, port);
2415     }
2416
2417     bundle_update(bundle);
2418 }
2419
2420 static bool
2421 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2422                 struct lacp_slave_settings *lacp)
2423 {
2424     struct ofport_dpif *port;
2425
2426     port = get_ofp_port(bundle->ofproto, ofp_port);
2427     if (!port) {
2428         return false;
2429     }
2430
2431     if (port->bundle != bundle) {
2432         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2433         if (port->bundle) {
2434             bundle_remove(&port->up);
2435         }
2436
2437         port->bundle = bundle;
2438         list_push_back(&bundle->ports, &port->bundle_node);
2439         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2440             || !stp_forward_in_state(port->stp_state)) {
2441             bundle->floodable = false;
2442         }
2443     }
2444     if (lacp) {
2445         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2446         lacp_slave_register(bundle->lacp, port, lacp);
2447     }
2448
2449     return true;
2450 }
2451
2452 static void
2453 bundle_destroy(struct ofbundle *bundle)
2454 {
2455     struct ofproto_dpif *ofproto;
2456     struct ofport_dpif *port, *next_port;
2457
2458     if (!bundle) {
2459         return;
2460     }
2461
2462     ofproto = bundle->ofproto;
2463     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2464
2465     ovs_rwlock_wrlock(&xlate_rwlock);
2466     xlate_bundle_remove(bundle);
2467     ovs_rwlock_unlock(&xlate_rwlock);
2468
2469     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2470         bundle_del_port(port);
2471     }
2472
2473     bundle_flush_macs(bundle, true);
2474     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2475     free(bundle->name);
2476     free(bundle->trunks);
2477     lacp_unref(bundle->lacp);
2478     bond_unref(bundle->bond);
2479     free(bundle);
2480 }
2481
2482 static int
2483 bundle_set(struct ofproto *ofproto_, void *aux,
2484            const struct ofproto_bundle_settings *s)
2485 {
2486     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2487     bool need_flush = false;
2488     struct ofport_dpif *port;
2489     struct ofbundle *bundle;
2490     unsigned long *trunks;
2491     int vlan;
2492     size_t i;
2493     bool ok;
2494
2495     if (!s) {
2496         bundle_destroy(bundle_lookup(ofproto, aux));
2497         return 0;
2498     }
2499
2500     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2501     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2502
2503     bundle = bundle_lookup(ofproto, aux);
2504     if (!bundle) {
2505         bundle = xmalloc(sizeof *bundle);
2506
2507         bundle->ofproto = ofproto;
2508         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2509                     hash_pointer(aux, 0));
2510         bundle->aux = aux;
2511         bundle->name = NULL;
2512
2513         list_init(&bundle->ports);
2514         bundle->vlan_mode = PORT_VLAN_TRUNK;
2515         bundle->vlan = -1;
2516         bundle->trunks = NULL;
2517         bundle->use_priority_tags = s->use_priority_tags;
2518         bundle->lacp = NULL;
2519         bundle->bond = NULL;
2520
2521         bundle->floodable = true;
2522         mbridge_register_bundle(ofproto->mbridge, bundle);
2523     }
2524
2525     if (!bundle->name || strcmp(s->name, bundle->name)) {
2526         free(bundle->name);
2527         bundle->name = xstrdup(s->name);
2528     }
2529
2530     /* LACP. */
2531     if (s->lacp) {
2532         if (!bundle->lacp) {
2533             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2534             bundle->lacp = lacp_create();
2535         }
2536         lacp_configure(bundle->lacp, s->lacp);
2537     } else {
2538         lacp_unref(bundle->lacp);
2539         bundle->lacp = NULL;
2540     }
2541
2542     /* Update set of ports. */
2543     ok = true;
2544     for (i = 0; i < s->n_slaves; i++) {
2545         if (!bundle_add_port(bundle, s->slaves[i],
2546                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2547             ok = false;
2548         }
2549     }
2550     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2551         struct ofport_dpif *next_port;
2552
2553         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2554             for (i = 0; i < s->n_slaves; i++) {
2555                 if (s->slaves[i] == port->up.ofp_port) {
2556                     goto found;
2557                 }
2558             }
2559
2560             bundle_del_port(port);
2561         found: ;
2562         }
2563     }
2564     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2565
2566     if (list_is_empty(&bundle->ports)) {
2567         bundle_destroy(bundle);
2568         return EINVAL;
2569     }
2570
2571     /* Set VLAN tagging mode */
2572     if (s->vlan_mode != bundle->vlan_mode
2573         || s->use_priority_tags != bundle->use_priority_tags) {
2574         bundle->vlan_mode = s->vlan_mode;
2575         bundle->use_priority_tags = s->use_priority_tags;
2576         need_flush = true;
2577     }
2578
2579     /* Set VLAN tag. */
2580     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2581             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2582             : 0);
2583     if (vlan != bundle->vlan) {
2584         bundle->vlan = vlan;
2585         need_flush = true;
2586     }
2587
2588     /* Get trunked VLANs. */
2589     switch (s->vlan_mode) {
2590     case PORT_VLAN_ACCESS:
2591         trunks = NULL;
2592         break;
2593
2594     case PORT_VLAN_TRUNK:
2595         trunks = CONST_CAST(unsigned long *, s->trunks);
2596         break;
2597
2598     case PORT_VLAN_NATIVE_UNTAGGED:
2599     case PORT_VLAN_NATIVE_TAGGED:
2600         if (vlan != 0 && (!s->trunks
2601                           || !bitmap_is_set(s->trunks, vlan)
2602                           || bitmap_is_set(s->trunks, 0))) {
2603             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2604             if (s->trunks) {
2605                 trunks = bitmap_clone(s->trunks, 4096);
2606             } else {
2607                 trunks = bitmap_allocate1(4096);
2608             }
2609             bitmap_set1(trunks, vlan);
2610             bitmap_set0(trunks, 0);
2611         } else {
2612             trunks = CONST_CAST(unsigned long *, s->trunks);
2613         }
2614         break;
2615
2616     default:
2617         NOT_REACHED();
2618     }
2619     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2620         free(bundle->trunks);
2621         if (trunks == s->trunks) {
2622             bundle->trunks = vlan_bitmap_clone(trunks);
2623         } else {
2624             bundle->trunks = trunks;
2625             trunks = NULL;
2626         }
2627         need_flush = true;
2628     }
2629     if (trunks != s->trunks) {
2630         free(trunks);
2631     }
2632
2633     /* Bonding. */
2634     if (!list_is_short(&bundle->ports)) {
2635         bundle->ofproto->has_bonded_bundles = true;
2636         if (bundle->bond) {
2637             if (bond_reconfigure(bundle->bond, s->bond)) {
2638                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2639             }
2640         } else {
2641             bundle->bond = bond_create(s->bond);
2642             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2643         }
2644
2645         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2646             bond_slave_register(bundle->bond, port, port->up.netdev);
2647         }
2648     } else {
2649         bond_unref(bundle->bond);
2650         bundle->bond = NULL;
2651     }
2652
2653     /* If we changed something that would affect MAC learning, un-learn
2654      * everything on this port and force flow revalidation. */
2655     if (need_flush) {
2656         bundle_flush_macs(bundle, false);
2657     }
2658
2659     return 0;
2660 }
2661
2662 static void
2663 bundle_remove(struct ofport *port_)
2664 {
2665     struct ofport_dpif *port = ofport_dpif_cast(port_);
2666     struct ofbundle *bundle = port->bundle;
2667
2668     if (bundle) {
2669         bundle_del_port(port);
2670         if (list_is_empty(&bundle->ports)) {
2671             bundle_destroy(bundle);
2672         } else if (list_is_short(&bundle->ports)) {
2673             bond_unref(bundle->bond);
2674             bundle->bond = NULL;
2675         }
2676     }
2677 }
2678
2679 static void
2680 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2681 {
2682     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2683     struct ofport_dpif *port = port_;
2684     uint8_t ea[ETH_ADDR_LEN];
2685     int error;
2686
2687     error = netdev_get_etheraddr(port->up.netdev, ea);
2688     if (!error) {
2689         struct ofpbuf packet;
2690         void *packet_pdu;
2691
2692         ofpbuf_init(&packet, 0);
2693         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2694                                  pdu_size);
2695         memcpy(packet_pdu, pdu, pdu_size);
2696
2697         send_packet(port, &packet);
2698         ofpbuf_uninit(&packet);
2699     } else {
2700         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2701                     "%s (%s)", port->bundle->name,
2702                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2703     }
2704 }
2705
2706 static void
2707 bundle_send_learning_packets(struct ofbundle *bundle)
2708 {
2709     struct ofproto_dpif *ofproto = bundle->ofproto;
2710     int error, n_packets, n_errors;
2711     struct mac_entry *e;
2712
2713     error = n_packets = n_errors = 0;
2714     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2715     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2716         if (e->port.p != bundle) {
2717             struct ofpbuf *learning_packet;
2718             struct ofport_dpif *port;
2719             void *port_void;
2720             int ret;
2721
2722             /* The assignment to "port" is unnecessary but makes "grep"ing for
2723              * struct ofport_dpif more effective. */
2724             learning_packet = bond_compose_learning_packet(bundle->bond,
2725                                                            e->mac, e->vlan,
2726                                                            &port_void);
2727             port = port_void;
2728             ret = send_packet(port, learning_packet);
2729             ofpbuf_delete(learning_packet);
2730             if (ret) {
2731                 error = ret;
2732                 n_errors++;
2733             }
2734             n_packets++;
2735         }
2736     }
2737     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2738
2739     if (n_errors) {
2740         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2741         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2742                      "packets, last error was: %s",
2743                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2744     } else {
2745         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2746                  bundle->name, n_packets);
2747     }
2748 }
2749
2750 static void
2751 bundle_run(struct ofbundle *bundle)
2752 {
2753     if (bundle->lacp) {
2754         lacp_run(bundle->lacp, send_pdu_cb);
2755     }
2756     if (bundle->bond) {
2757         struct ofport_dpif *port;
2758
2759         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2760             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2761         }
2762
2763         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2764             bundle->ofproto->backer->need_revalidate = REV_BOND;
2765         }
2766
2767         if (bond_should_send_learning_packets(bundle->bond)) {
2768             bundle_send_learning_packets(bundle);
2769         }
2770     }
2771 }
2772
2773 static void
2774 bundle_wait(struct ofbundle *bundle)
2775 {
2776     if (bundle->lacp) {
2777         lacp_wait(bundle->lacp);
2778     }
2779     if (bundle->bond) {
2780         bond_wait(bundle->bond);
2781     }
2782 }
2783 \f
2784 /* Mirrors. */
2785
2786 static int
2787 mirror_set__(struct ofproto *ofproto_, void *aux,
2788              const struct ofproto_mirror_settings *s)
2789 {
2790     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2791     struct ofbundle **srcs, **dsts;
2792     int error;
2793     size_t i;
2794
2795     if (!s) {
2796         mirror_destroy(ofproto->mbridge, aux);
2797         return 0;
2798     }
2799
2800     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2801     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2802
2803     for (i = 0; i < s->n_srcs; i++) {
2804         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2805     }
2806
2807     for (i = 0; i < s->n_dsts; i++) {
2808         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2809     }
2810
2811     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2812                        s->n_dsts, s->src_vlans,
2813                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2814     free(srcs);
2815     free(dsts);
2816     return error;
2817 }
2818
2819 static int
2820 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2821                    uint64_t *packets, uint64_t *bytes)
2822 {
2823     push_all_stats();
2824     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2825                             bytes);
2826 }
2827
2828 static int
2829 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2830 {
2831     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2832     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2833     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2834         mac_learning_flush(ofproto->ml);
2835     }
2836     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2837     return 0;
2838 }
2839
2840 static bool
2841 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2842 {
2843     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2844     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2845     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2846 }
2847
2848 static void
2849 forward_bpdu_changed(struct ofproto *ofproto_)
2850 {
2851     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2852     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2853 }
2854
2855 static void
2856 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2857                      size_t max_entries)
2858 {
2859     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2860     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2861     mac_learning_set_idle_time(ofproto->ml, idle_time);
2862     mac_learning_set_max_entries(ofproto->ml, max_entries);
2863     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2864 }
2865 \f
2866 /* Ports. */
2867
2868 static struct ofport_dpif *
2869 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2870 {
2871     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2872     return ofport ? ofport_dpif_cast(ofport) : NULL;
2873 }
2874
2875 static struct ofport_dpif *
2876 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2877 {
2878     struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2879     return port && &ofproto->up == port->up.ofproto ? port : NULL;
2880 }
2881
2882 static void
2883 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2884                             struct ofproto_port *ofproto_port,
2885                             struct dpif_port *dpif_port)
2886 {
2887     ofproto_port->name = dpif_port->name;
2888     ofproto_port->type = dpif_port->type;
2889     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2890 }
2891
2892 static void
2893 ofport_update_peer(struct ofport_dpif *ofport)
2894 {
2895     const struct ofproto_dpif *ofproto;
2896     struct dpif_backer *backer;
2897     char *peer_name;
2898
2899     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2900         return;
2901     }
2902
2903     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2904     backer->need_revalidate = REV_RECONFIGURE;
2905
2906     if (ofport->peer) {
2907         ofport->peer->peer = NULL;
2908         ofport->peer = NULL;
2909     }
2910
2911     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2912     if (!peer_name) {
2913         return;
2914     }
2915
2916     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2917         struct ofport *peer_ofport;
2918         struct ofport_dpif *peer;
2919         char *peer_peer;
2920
2921         if (ofproto->backer != backer) {
2922             continue;
2923         }
2924
2925         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2926         if (!peer_ofport) {
2927             continue;
2928         }
2929
2930         peer = ofport_dpif_cast(peer_ofport);
2931         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2932         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2933                                  peer_peer)) {
2934             ofport->peer = peer;
2935             ofport->peer->peer = ofport;
2936         }
2937         free(peer_peer);
2938
2939         break;
2940     }
2941     free(peer_name);
2942 }
2943
2944 static void
2945 port_run_fast(struct ofport_dpif *ofport)
2946 {
2947     if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2948         struct ofpbuf packet;
2949
2950         ofpbuf_init(&packet, 0);
2951         cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2952         send_packet(ofport, &packet);
2953         ofpbuf_uninit(&packet);
2954     }
2955
2956     if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2957         struct ofpbuf packet;
2958
2959         ofpbuf_init(&packet, 0);
2960         bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2961         send_packet(ofport, &packet);
2962         ofpbuf_uninit(&packet);
2963     }
2964 }
2965
2966 static void
2967 port_run(struct ofport_dpif *ofport)
2968 {
2969     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2970     bool carrier_changed = carrier_seq != ofport->carrier_seq;
2971     bool enable = netdev_get_carrier(ofport->up.netdev);
2972     bool cfm_enable = false;
2973     bool bfd_enable = false;
2974
2975     ofport->carrier_seq = carrier_seq;
2976
2977     port_run_fast(ofport);
2978
2979     if (ofport->cfm) {
2980         int cfm_opup = cfm_get_opup(ofport->cfm);
2981
2982         cfm_run(ofport->cfm);
2983         cfm_enable = !cfm_get_fault(ofport->cfm);
2984
2985         if (cfm_opup >= 0) {
2986             cfm_enable = cfm_enable && cfm_opup;
2987         }
2988     }
2989
2990     if (ofport->bfd) {
2991         bfd_run(ofport->bfd);
2992         bfd_enable = bfd_forwarding(ofport->bfd);
2993     }
2994
2995     if (ofport->bfd || ofport->cfm) {
2996         enable = enable && (cfm_enable || bfd_enable);
2997     }
2998
2999     if (ofport->bundle) {
3000         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3001         if (carrier_changed) {
3002             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3003         }
3004     }
3005
3006     if (ofport->may_enable != enable) {
3007         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3008         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3009     }
3010
3011     ofport->may_enable = enable;
3012 }
3013
3014 static void
3015 port_wait(struct ofport_dpif *ofport)
3016 {
3017     if (ofport->cfm) {
3018         cfm_wait(ofport->cfm);
3019     }
3020
3021     if (ofport->bfd) {
3022         bfd_wait(ofport->bfd);
3023     }
3024 }
3025
3026 static int
3027 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3028                    struct ofproto_port *ofproto_port)
3029 {
3030     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3031     struct dpif_port dpif_port;
3032     int error;
3033
3034     if (sset_contains(&ofproto->ghost_ports, devname)) {
3035         const char *type = netdev_get_type_from_name(devname);
3036
3037         /* We may be called before ofproto->up.port_by_name is populated with
3038          * the appropriate ofport.  For this reason, we must get the name and
3039          * type from the netdev layer directly. */
3040         if (type) {
3041             const struct ofport *ofport;
3042
3043             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3044             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3045             ofproto_port->name = xstrdup(devname);
3046             ofproto_port->type = xstrdup(type);
3047             return 0;
3048         }
3049         return ENODEV;
3050     }
3051
3052     if (!sset_contains(&ofproto->ports, devname)) {
3053         return ENODEV;
3054     }
3055     error = dpif_port_query_by_name(ofproto->backer->dpif,
3056                                     devname, &dpif_port);
3057     if (!error) {
3058         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3059     }
3060     return error;
3061 }
3062
3063 static int
3064 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3065 {
3066     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3067     const char *devname = netdev_get_name(netdev);
3068     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3069     const char *dp_port_name;
3070
3071     if (netdev_vport_is_patch(netdev)) {
3072         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3073         return 0;
3074     }
3075
3076     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3077     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3078         odp_port_t port_no = ODPP_NONE;
3079         int error;
3080
3081         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3082         if (error) {
3083             return error;
3084         }
3085         if (netdev_get_tunnel_config(netdev)) {
3086             simap_put(&ofproto->backer->tnl_backers,
3087                       dp_port_name, odp_to_u32(port_no));
3088         }
3089     }
3090
3091     if (netdev_get_tunnel_config(netdev)) {
3092         sset_add(&ofproto->ghost_ports, devname);
3093     } else {
3094         sset_add(&ofproto->ports, devname);
3095     }
3096     return 0;
3097 }
3098
3099 static int
3100 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3101 {
3102     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3103     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3104     int error = 0;
3105
3106     if (!ofport) {
3107         return 0;
3108     }
3109
3110     sset_find_and_delete(&ofproto->ghost_ports,
3111                          netdev_get_name(ofport->up.netdev));
3112     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3113     if (!ofport->is_tunnel) {
3114         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3115         if (!error) {
3116             /* The caller is going to close ofport->up.netdev.  If this is a
3117              * bonded port, then the bond is using that netdev, so remove it
3118              * from the bond.  The client will need to reconfigure everything
3119              * after deleting ports, so then the slave will get re-added. */
3120             bundle_remove(&ofport->up);
3121         }
3122     }
3123     return error;
3124 }
3125
3126 static int
3127 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3128 {
3129     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3130     int error;
3131
3132     push_all_stats();
3133
3134     error = netdev_get_stats(ofport->up.netdev, stats);
3135
3136     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3137         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3138
3139         /* ofproto->stats.tx_packets represents packets that we created
3140          * internally and sent to some port (e.g. packets sent with
3141          * send_packet()).  Account for them as if they had come from
3142          * OFPP_LOCAL and got forwarded. */
3143
3144         if (stats->rx_packets != UINT64_MAX) {
3145             stats->rx_packets += ofproto->stats.tx_packets;
3146         }
3147
3148         if (stats->rx_bytes != UINT64_MAX) {
3149             stats->rx_bytes += ofproto->stats.tx_bytes;
3150         }
3151
3152         /* ofproto->stats.rx_packets represents packets that were received on
3153          * some port and we processed internally and dropped (e.g. STP).
3154          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3155
3156         if (stats->tx_packets != UINT64_MAX) {
3157             stats->tx_packets += ofproto->stats.rx_packets;
3158         }
3159
3160         if (stats->tx_bytes != UINT64_MAX) {
3161             stats->tx_bytes += ofproto->stats.rx_bytes;
3162         }
3163     }
3164
3165     return error;
3166 }
3167
3168 struct port_dump_state {
3169     uint32_t bucket;
3170     uint32_t offset;
3171     bool ghost;
3172
3173     struct ofproto_port port;
3174     bool has_port;
3175 };
3176
3177 static int
3178 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3179 {
3180     *statep = xzalloc(sizeof(struct port_dump_state));
3181     return 0;
3182 }
3183
3184 static int
3185 port_dump_next(const struct ofproto *ofproto_, void *state_,
3186                struct ofproto_port *port)
3187 {
3188     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3189     struct port_dump_state *state = state_;
3190     const struct sset *sset;
3191     struct sset_node *node;
3192
3193     if (state->has_port) {
3194         ofproto_port_destroy(&state->port);
3195         state->has_port = false;
3196     }
3197     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3198     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3199         int error;
3200
3201         error = port_query_by_name(ofproto_, node->name, &state->port);
3202         if (!error) {
3203             *port = state->port;
3204             state->has_port = true;
3205             return 0;
3206         } else if (error != ENODEV) {
3207             return error;
3208         }
3209     }
3210
3211     if (!state->ghost) {
3212         state->ghost = true;
3213         state->bucket = 0;
3214         state->offset = 0;
3215         return port_dump_next(ofproto_, state_, port);
3216     }
3217
3218     return EOF;
3219 }
3220
3221 static int
3222 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3223 {
3224     struct port_dump_state *state = state_;
3225
3226     if (state->has_port) {
3227         ofproto_port_destroy(&state->port);
3228     }
3229     free(state);
3230     return 0;
3231 }
3232
3233 static int
3234 port_poll(const struct ofproto *ofproto_, char **devnamep)
3235 {
3236     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3237
3238     if (ofproto->port_poll_errno) {
3239         int error = ofproto->port_poll_errno;
3240         ofproto->port_poll_errno = 0;
3241         return error;
3242     }
3243
3244     if (sset_is_empty(&ofproto->port_poll_set)) {
3245         return EAGAIN;
3246     }
3247
3248     *devnamep = sset_pop(&ofproto->port_poll_set);
3249     return 0;
3250 }
3251
3252 static void
3253 port_poll_wait(const struct ofproto *ofproto_)
3254 {
3255     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3256     dpif_port_poll_wait(ofproto->backer->dpif);
3257 }
3258
3259 static int
3260 port_is_lacp_current(const struct ofport *ofport_)
3261 {
3262     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3263     return (ofport->bundle && ofport->bundle->lacp
3264             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3265             : -1);
3266 }
3267 \f
3268 /* Upcall handling. */
3269
3270 struct flow_miss_op {
3271     struct dpif_op dpif_op;
3272
3273     uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3274     struct xlate_out xout;
3275     bool xout_garbage;           /* 'xout' needs to be uninitialized? */
3276
3277     struct ofpbuf mask;          /* Flow mask for "put" ops. */
3278     struct odputil_keybuf maskbuf;
3279
3280     /* If this is a "put" op, then a pointer to the subfacet that should
3281      * be marked as uninstalled if the operation fails. */
3282     struct subfacet *subfacet;
3283 };
3284
3285 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3286  * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3287  * and (usually) installing a datapath flow.  The answer is usually "yes" (a
3288  * return value of true).  However, for short flows the cost of bookkeeping is
3289  * much higher than the benefits, so when the datapath holds a large number of
3290  * flows we impose some heuristics to decide which flows are likely to be worth
3291  * tracking. */
3292 static bool
3293 flow_miss_should_make_facet(struct flow_miss *miss)
3294 {
3295     struct dpif_backer *backer = miss->ofproto->backer;
3296     uint32_t hash;
3297
3298     switch (flow_miss_model) {
3299     case OFPROTO_HANDLE_MISS_AUTO:
3300         break;
3301     case OFPROTO_HANDLE_MISS_WITH_FACETS:
3302         return true;
3303     case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3304         return false;
3305     }
3306
3307     if (!backer->governor) {
3308         size_t n_subfacets;
3309
3310         n_subfacets = hmap_count(&backer->subfacets);
3311         if (n_subfacets * 2 <= flow_eviction_threshold) {
3312             return true;
3313         }
3314
3315         backer->governor = governor_create();
3316     }
3317
3318     hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3319     return governor_should_install_flow(backer->governor, hash,
3320                                         list_size(&miss->packets));
3321 }
3322
3323 /* Handles 'miss', which matches 'facet'.  May add any required datapath
3324  * operations to 'ops', incrementing '*n_ops' for each new op.
3325  *
3326  * All of the packets in 'miss' are considered to have arrived at time
3327  * 'miss->stats.used'.  This is really important only for new facets: if we
3328  * just called time_msec() here, then the new subfacet or its packets could
3329  * look (occasionally) as though it was used some time after the facet was
3330  * used.  That can make a one-packet flow look like it has a nonzero duration,
3331  * which looks odd in e.g. NetFlow statistics. */
3332 static void
3333 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3334                             struct flow_miss_op *ops, size_t *n_ops)
3335 {
3336     enum subfacet_path want_path;
3337     struct subfacet *subfacet;
3338
3339     facet->packet_count += miss->stats.n_packets;
3340     facet->prev_packet_count += miss->stats.n_packets;
3341     facet->byte_count += miss->stats.n_bytes;
3342     facet->prev_byte_count += miss->stats.n_bytes;
3343
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     COVERAGE_INC(ofproto_dpif_expired);
3981     ofproto_rule_expire(&rule->up, reason);
3982 }
3983 \f
3984 /* Facets. */
3985
3986 /* Creates and returns a new facet based on 'miss'.
3987  *
3988  * The caller must already have determined that no facet with an identical
3989  * 'miss->flow' exists in 'miss->ofproto'.
3990  *
3991  * 'rule' and 'xout' must have been created based on 'miss'.
3992  *
3993  * 'facet'' statistics are initialized based on 'stats'.
3994  *
3995  * The facet will initially have no subfacets.  The caller should create (at
3996  * least) one subfacet with subfacet_create(). */
3997 static struct facet *
3998 facet_create(const struct flow_miss *miss)
3999 {
4000     struct ofproto_dpif *ofproto = miss->ofproto;
4001     struct facet *facet;
4002     struct match match;
4003
4004     facet = xzalloc(sizeof *facet);
4005     facet->ofproto = miss->ofproto;
4006     facet->used = miss->stats.used;
4007     facet->flow = miss->flow;
4008     facet->learn_rl = time_msec() + 500;
4009
4010     list_init(&facet->subfacets);
4011     netflow_flow_init(&facet->nf_flow);
4012     netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4013
4014     xlate_out_copy(&facet->xout, &miss->xout);
4015
4016     match_init(&match, &facet->flow, &facet->xout.wc);
4017     cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4018     ovs_rwlock_wrlock(&ofproto->facets.rwlock);
4019     classifier_insert(&ofproto->facets, &facet->cr);
4020     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4021
4022     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4023     return facet;
4024 }
4025
4026 static void
4027 facet_free(struct facet *facet)
4028 {
4029     if (facet) {
4030         xlate_out_uninit(&facet->xout);
4031         free(facet);
4032     }
4033 }
4034
4035 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4036  * 'packet', which arrived on 'in_port'. */
4037 static bool
4038 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4039                     const struct nlattr *odp_actions, size_t actions_len,
4040                     struct ofpbuf *packet)
4041 {
4042     struct odputil_keybuf keybuf;
4043     struct ofpbuf key;
4044     int error;
4045
4046     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4047     odp_flow_key_from_flow(&key, flow,
4048                            ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4049
4050     error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4051                          odp_actions, actions_len, packet);
4052     return !error;
4053 }
4054
4055 /* Remove 'facet' from its ofproto and free up the associated memory:
4056  *
4057  *   - If 'facet' was installed in the datapath, uninstalls it and updates its
4058  *     rule's statistics, via subfacet_uninstall().
4059  *
4060  *   - Removes 'facet' from its rule and from ofproto->facets.
4061  */
4062 static void
4063 facet_remove(struct facet *facet)
4064 {
4065     struct subfacet *subfacet, *next_subfacet;
4066
4067     ovs_assert(!list_is_empty(&facet->subfacets));
4068
4069     /* First uninstall all of the subfacets to get final statistics. */
4070     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4071         subfacet_uninstall(subfacet);
4072     }
4073
4074     /* Flush the final stats to the rule.
4075      *
4076      * This might require us to have at least one subfacet around so that we
4077      * can use its actions for accounting in facet_account(), which is why we
4078      * have uninstalled but not yet destroyed the subfacets. */
4079     facet_flush_stats(facet);
4080
4081     /* Now we're really all done so destroy everything. */
4082     LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4083                         &facet->subfacets) {
4084         subfacet_destroy__(subfacet);
4085     }
4086     ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4087     classifier_remove(&facet->ofproto->facets, &facet->cr);
4088     ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4089     cls_rule_destroy(&facet->cr);
4090     facet_free(facet);
4091 }
4092
4093 /* Feed information from 'facet' back into the learning table to keep it in
4094  * sync with what is actually flowing through the datapath. */
4095 static void
4096 facet_learn(struct facet *facet)
4097 {
4098     long long int now = time_msec();
4099
4100     if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4101         return;
4102     }
4103
4104     facet->learn_rl = now + 500;
4105
4106     if (!facet->xout.has_learn
4107         && !facet->xout.has_normal
4108         && (!facet->xout.has_fin_timeout
4109             || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4110         return;
4111     }
4112
4113     facet_push_stats(facet, true);
4114 }
4115
4116 static void
4117 facet_account(struct facet *facet)
4118 {
4119     const struct nlattr *a;
4120     unsigned int left;
4121     ovs_be16 vlan_tci;
4122     uint64_t n_bytes;
4123
4124     if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4125         return;
4126     }
4127     n_bytes = facet->byte_count - facet->accounted_bytes;
4128
4129     /* This loop feeds byte counters to bond_account() for rebalancing to use
4130      * as a basis.  We also need to track the actual VLAN on which the packet
4131      * is going to be sent to ensure that it matches the one passed to
4132      * bond_choose_output_slave().  (Otherwise, we will account to the wrong
4133      * hash bucket.)
4134      *
4135      * We use the actions from an arbitrary subfacet because they should all
4136      * be equally valid for our purpose. */
4137     vlan_tci = facet->flow.vlan_tci;
4138     NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4139                              facet->xout.odp_actions.size) {
4140         const struct ovs_action_push_vlan *vlan;
4141         struct ofport_dpif *port;
4142
4143         switch (nl_attr_type(a)) {
4144         case OVS_ACTION_ATTR_OUTPUT:
4145             port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4146             if (port && port->bundle && port->bundle->bond) {
4147                 bond_account(port->bundle->bond, &facet->flow,
4148                              vlan_tci_to_vid(vlan_tci), n_bytes);
4149             }
4150             break;
4151
4152         case OVS_ACTION_ATTR_POP_VLAN:
4153             vlan_tci = htons(0);
4154             break;
4155
4156         case OVS_ACTION_ATTR_PUSH_VLAN:
4157             vlan = nl_attr_get(a);
4158             vlan_tci = vlan->vlan_tci;
4159             break;
4160         }
4161     }
4162 }
4163
4164 /* Returns true if the only action for 'facet' is to send to the controller.
4165  * (We don't report NetFlow expiration messages for such facets because they
4166  * are just part of the control logic for the network, not real traffic). */
4167 static bool
4168 facet_is_controller_flow(struct facet *facet)
4169 {
4170     if (facet) {
4171         struct ofproto_dpif *ofproto = facet->ofproto;
4172         const struct ofpact *ofpacts;
4173         struct rule_dpif *rule;
4174         size_t ofpacts_len;
4175         bool is_controller;
4176
4177         rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4178         ofpacts_len = rule->up.ofpacts_len;
4179         ofpacts = rule->up.ofpacts;
4180         is_controller = ofpacts_len > 0
4181             && ofpacts->type == OFPACT_CONTROLLER
4182             && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4183         rule_release(rule);
4184         return is_controller;
4185     }
4186     return false;
4187 }
4188
4189 /* Folds all of 'facet''s statistics into its rule.  Also updates the
4190  * accounting ofhook and emits a NetFlow expiration if appropriate.  All of
4191  * 'facet''s statistics in the datapath should have been zeroed and folded into
4192  * its packet and byte counts before this function is called. */
4193 static void
4194 facet_flush_stats(struct facet *facet)
4195 {
4196     struct ofproto_dpif *ofproto = facet->ofproto;
4197     struct subfacet *subfacet;
4198
4199     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4200         ovs_assert(!subfacet->dp_byte_count);
4201         ovs_assert(!subfacet->dp_packet_count);
4202     }
4203
4204     facet_push_stats(facet, false);
4205     if (facet->accounted_bytes < facet->byte_count) {
4206         facet_account(facet);
4207         facet->accounted_bytes = facet->byte_count;
4208     }
4209
4210     if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4211         struct ofexpired expired;
4212         expired.flow = facet->flow;
4213         expired.packet_count = facet->packet_count;
4214         expired.byte_count = facet->byte_count;
4215         expired.used = facet->used;
4216         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4217     }
4218
4219     /* Reset counters to prevent double counting if 'facet' ever gets
4220      * reinstalled. */
4221     facet_reset_counters(facet);
4222
4223     netflow_flow_clear(&facet->nf_flow);
4224     facet->tcp_flags = 0;
4225 }
4226
4227 /* Searches 'ofproto''s table of facets for one which would be responsible for
4228  * 'flow'.  Returns it if found, otherwise a null pointer.
4229  *
4230  * The returned facet might need revalidation; use facet_lookup_valid()
4231  * instead if that is important. */
4232 static struct facet *
4233 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4234 {
4235     struct cls_rule *cr;
4236
4237     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4238     cr = classifier_lookup(&ofproto->facets, flow, NULL);
4239     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4240     return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4241 }
4242
4243 /* Searches 'ofproto''s table of facets for one capable that covers
4244  * 'flow'.  Returns it if found, otherwise a null pointer.
4245  *
4246  * The returned facet is guaranteed to be valid. */
4247 static struct facet *
4248 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4249 {
4250     struct facet *facet;
4251
4252     facet = facet_find(ofproto, flow);
4253     if (facet
4254         && ofproto->backer->need_revalidate
4255         && !facet_revalidate(facet)) {
4256         return NULL;
4257     }
4258
4259     return facet;
4260 }
4261
4262 static bool
4263 facet_check_consistency(struct facet *facet)
4264 {
4265     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4266
4267     struct xlate_out xout;
4268     struct xlate_in xin;
4269
4270     struct rule_dpif *rule;
4271     bool ok;
4272
4273     /* Check the datapath actions for consistency. */
4274     rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4275     xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4276     xlate_actions(&xin, &xout);
4277     rule_release(rule);
4278
4279     ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4280         && facet->xout.slow == xout.slow;
4281     if (!ok && !VLOG_DROP_WARN(&rl)) {
4282         struct ds s = DS_EMPTY_INITIALIZER;
4283
4284         flow_format(&s, &facet->flow);
4285         ds_put_cstr(&s, ": inconsistency in facet");
4286
4287         if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4288             ds_put_cstr(&s, " (actions were: ");
4289             format_odp_actions(&s, facet->xout.odp_actions.data,
4290                                facet->xout.odp_actions.size);
4291             ds_put_cstr(&s, ") (correct actions: ");
4292             format_odp_actions(&s, xout.odp_actions.data,
4293                                xout.odp_actions.size);
4294             ds_put_char(&s, ')');
4295         }
4296
4297         if (facet->xout.slow != xout.slow) {
4298             ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4299         }
4300
4301         ds_destroy(&s);
4302     }
4303     xlate_out_uninit(&xout);
4304
4305     return ok;
4306 }
4307
4308 /* Re-searches the classifier for 'facet':
4309  *
4310  *   - If the rule found is different from 'facet''s current rule, moves
4311  *     'facet' to the new rule and recompiles its actions.
4312  *
4313  *   - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4314  *     where it is and recompiles its actions anyway.
4315  *
4316  *   - If any of 'facet''s subfacets correspond to a new flow according to
4317  *     xlate_receive(), 'facet' is removed.
4318  *
4319  *   Returns true if 'facet' is still valid.  False if 'facet' was removed. */
4320 static bool
4321 facet_revalidate(struct facet *facet)
4322 {
4323     struct ofproto_dpif *ofproto = facet->ofproto;
4324     struct rule_dpif *new_rule;
4325     struct subfacet *subfacet;
4326     struct flow_wildcards wc;
4327     struct xlate_out xout;
4328     struct xlate_in xin;
4329
4330     COVERAGE_INC(facet_revalidate);
4331
4332     /* Check that child subfacets still correspond to this facet.  Tunnel
4333      * configuration changes could cause a subfacet's OpenFlow in_port to
4334      * change. */
4335     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4336         struct ofproto_dpif *recv_ofproto;
4337         struct flow recv_flow;
4338         int error;
4339
4340         error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4341                               subfacet->key_len, &recv_flow, NULL,
4342                               &recv_ofproto, NULL);
4343         if (error
4344             || recv_ofproto != ofproto
4345             || facet != facet_find(ofproto, &recv_flow)) {
4346             facet_remove(facet);
4347             return false;
4348         }
4349     }
4350
4351     flow_wildcards_init_catchall(&wc);
4352     rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4353
4354     /* Calculate new datapath actions.
4355      *
4356      * We do not modify any 'facet' state yet, because we might need to, e.g.,
4357      * emit a NetFlow expiration and, if so, we need to have the old state
4358      * around to properly compose it. */
4359     xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4360     xlate_actions(&xin, &xout);
4361     flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4362
4363     /* A facet's slow path reason should only change under dramatic
4364      * circumstances.  Rather than try to update everything, it's simpler to
4365      * remove the facet and start over.
4366      *
4367      * More importantly, if a facet's wildcards change, it will be relatively
4368      * difficult to figure out if its subfacets still belong to it, and if not
4369      * which facet they may belong to.  Again, to avoid the complexity, we
4370      * simply give up instead. */
4371     if (facet->xout.slow != xout.slow
4372         || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4373         facet_remove(facet);
4374         xlate_out_uninit(&xout);
4375         rule_release(new_rule);
4376         return false;
4377     }
4378
4379     if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4380         LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4381             if (subfacet->path == SF_FAST_PATH) {
4382                 struct dpif_flow_stats stats;
4383
4384                 subfacet_install(subfacet, &xout.odp_actions, &stats);
4385                 subfacet_update_stats(subfacet, &stats);
4386             }
4387         }
4388
4389         facet_flush_stats(facet);
4390
4391         ofpbuf_clear(&facet->xout.odp_actions);
4392         ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4393                    xout.odp_actions.size);
4394     }
4395
4396     /* Update 'facet' now that we've taken care of all the old state. */
4397     facet->xout.slow = xout.slow;
4398     facet->xout.has_learn = xout.has_learn;
4399     facet->xout.has_normal = xout.has_normal;
4400     facet->xout.has_fin_timeout = xout.has_fin_timeout;
4401     facet->xout.nf_output_iface = xout.nf_output_iface;
4402     facet->xout.mirrors = xout.mirrors;
4403     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4404     facet->used = MAX(facet->used, new_rule->up.created);
4405
4406     xlate_out_uninit(&xout);
4407     rule_release(new_rule);
4408     return true;
4409 }
4410
4411 static void
4412 facet_reset_counters(struct facet *facet)
4413 {
4414     facet->packet_count = 0;
4415     facet->byte_count = 0;
4416     facet->prev_packet_count = 0;
4417     facet->prev_byte_count = 0;
4418     facet->accounted_bytes = 0;
4419 }
4420
4421 static void
4422 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4423                 struct dpif_flow_stats *stats, bool may_learn)
4424 {
4425     struct ofport_dpif *in_port;
4426     struct rule_dpif *rule;
4427     struct xlate_in xin;
4428
4429     in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4430     if (in_port && in_port->is_tunnel) {
4431         netdev_vport_inc_rx(in_port->up.netdev, stats);
4432     }
4433
4434     rule_dpif_lookup(ofproto, flow, NULL, &rule);
4435     rule_credit_stats(rule, stats);
4436     xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4437     xin.resubmit_stats = stats;
4438     xin.may_learn = may_learn;
4439     xlate_actions_for_side_effects(&xin);
4440     rule_release(rule);
4441 }
4442
4443 static void
4444 facet_push_stats(struct facet *facet, bool may_learn)
4445 {
4446     struct dpif_flow_stats stats;
4447
4448     ovs_assert(facet->packet_count >= facet->prev_packet_count);
4449     ovs_assert(facet->byte_count >= facet->prev_byte_count);
4450     ovs_assert(facet->used >= facet->prev_used);
4451
4452     stats.n_packets = facet->packet_count - facet->prev_packet_count;
4453     stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4454     stats.used = facet->used;
4455     stats.tcp_flags = facet->tcp_flags;
4456
4457     if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4458         facet->prev_packet_count = facet->packet_count;
4459         facet->prev_byte_count = facet->byte_count;
4460         facet->prev_used = facet->used;
4461
4462         netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4463                                  facet->used);
4464         netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4465         mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4466                             stats.n_packets, stats.n_bytes);
4467         flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4468     }
4469 }
4470
4471 static void
4472 push_all_stats__(bool run_fast)
4473 {
4474     static long long int rl = LLONG_MIN;
4475     struct ofproto_dpif *ofproto;
4476
4477     if (time_msec() < rl) {
4478         return;
4479     }
4480
4481     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4482         struct cls_cursor cursor;
4483         struct facet *facet;
4484
4485         ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4486         cls_cursor_init(&cursor, &ofproto->facets, NULL);
4487         CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4488             facet_push_stats(facet, false);
4489             if (run_fast) {
4490                 run_fast_rl();
4491             }
4492         }
4493         ovs_rwlock_unlock(&ofproto->facets.rwlock);
4494     }
4495
4496     rl = time_msec() + 100;
4497 }
4498
4499 static void
4500 push_all_stats(void)
4501 {
4502     push_all_stats__(true);
4503 }
4504
4505 void
4506 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4507 {
4508     ovs_mutex_lock(&rule->stats_mutex);
4509     rule->packet_count += stats->n_packets;
4510     rule->byte_count += stats->n_bytes;
4511     ofproto_rule_update_used(&rule->up, stats->used);
4512     ovs_mutex_unlock(&rule->stats_mutex);
4513 }
4514 \f
4515 /* Subfacets. */
4516
4517 static struct subfacet *
4518 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4519               size_t key_len, uint32_t key_hash)
4520 {
4521     struct subfacet *subfacet;
4522
4523     HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4524                              &backer->subfacets) {
4525         if (subfacet->key_len == key_len
4526             && !memcmp(key, subfacet->key, key_len)) {
4527             return subfacet;
4528         }
4529     }
4530
4531     return NULL;
4532 }
4533
4534 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4535  * 'key_fitness', 'key', and 'key_len' members in 'miss'.  Returns the
4536  * existing subfacet if there is one, otherwise creates and returns a
4537  * new subfacet. */
4538 static struct subfacet *
4539 subfacet_create(struct facet *facet, struct flow_miss *miss)
4540 {
4541     struct dpif_backer *backer = miss->ofproto->backer;
4542     enum odp_key_fitness key_fitness = miss->key_fitness;
4543     const struct nlattr *key = miss->key;
4544     size_t key_len = miss->key_len;
4545     uint32_t key_hash;
4546     struct subfacet *subfacet;
4547
4548     key_hash = odp_flow_key_hash(key, key_len);
4549
4550     if (list_is_empty(&facet->subfacets)) {
4551         subfacet = &facet->one_subfacet;
4552     } else {
4553         subfacet = subfacet_find(backer, key, key_len, key_hash);
4554         if (subfacet) {
4555             if (subfacet->facet == facet) {
4556                 return subfacet;
4557             }
4558
4559             /* This shouldn't happen. */
4560             VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4561             subfacet_destroy(subfacet);
4562         }
4563
4564         subfacet = xmalloc(sizeof *subfacet);
4565     }
4566
4567     hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4568     list_push_back(&facet->subfacets, &subfacet->list_node);
4569     subfacet->facet = facet;
4570     subfacet->key_fitness = key_fitness;
4571     subfacet->key = xmemdup(key, key_len);
4572     subfacet->key_len = key_len;
4573     subfacet->used = miss->stats.used;
4574     subfacet->created = subfacet->used;
4575     subfacet->dp_packet_count = 0;
4576     subfacet->dp_byte_count = 0;
4577     subfacet->path = SF_NOT_INSTALLED;
4578     subfacet->backer = backer;
4579
4580     backer->subfacet_add_count++;
4581     return subfacet;
4582 }
4583
4584 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4585  * its facet within 'ofproto', and frees it. */
4586 static void
4587 subfacet_destroy__(struct subfacet *subfacet)
4588 {
4589     struct facet *facet = subfacet->facet;
4590     struct ofproto_dpif *ofproto = facet->ofproto;
4591
4592     /* Update ofproto stats before uninstall the subfacet. */
4593     ofproto->backer->subfacet_del_count++;
4594
4595     subfacet_uninstall(subfacet);
4596     hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4597     list_remove(&subfacet->list_node);
4598     free(subfacet->key);
4599     if (subfacet != &facet->one_subfacet) {
4600         free(subfacet);
4601     }
4602 }
4603
4604 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4605  * last remaining subfacet in its facet destroys the facet too. */
4606 static void
4607 subfacet_destroy(struct subfacet *subfacet)
4608 {
4609     struct facet *facet = subfacet->facet;
4610
4611     if (list_is_singleton(&facet->subfacets)) {
4612         /* facet_remove() needs at least one subfacet (it will remove it). */
4613         facet_remove(facet);
4614     } else {
4615         subfacet_destroy__(subfacet);
4616     }
4617 }
4618
4619 static void
4620 subfacet_destroy_batch(struct dpif_backer *backer,
4621                        struct subfacet **subfacets, int n)
4622 {
4623     struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4624     struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4625     struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4626     int i;
4627
4628     for (i = 0; i < n; i++) {
4629         ops[i].type = DPIF_OP_FLOW_DEL;
4630         ops[i].u.flow_del.key = subfacets[i]->key;
4631         ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4632         ops[i].u.flow_del.stats = &stats[i];
4633         opsp[i] = &ops[i];
4634     }
4635
4636     dpif_operate(backer->dpif, opsp, n);
4637     for (i = 0; i < n; i++) {
4638         subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4639         subfacets[i]->path = SF_NOT_INSTALLED;
4640         subfacet_destroy(subfacets[i]);
4641         run_fast_rl();
4642     }
4643 }
4644
4645 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4646  * bytes of actions in 'actions'.  If 'stats' is non-null, statistics counters
4647  * in the datapath will be zeroed and 'stats' will be updated with traffic new
4648  * since 'subfacet' was last updated.
4649  *
4650  * Returns 0 if successful, otherwise a positive errno value. */
4651 static int
4652 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4653                  struct dpif_flow_stats *stats)
4654 {
4655     struct facet *facet = subfacet->facet;
4656     enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4657     const struct nlattr *actions = odp_actions->data;
4658     size_t actions_len = odp_actions->size;
4659     struct odputil_keybuf maskbuf;
4660     struct ofpbuf mask;
4661
4662     uint64_t slow_path_stub[128 / 8];
4663     enum dpif_flow_put_flags flags;
4664     int ret;
4665
4666     flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4667                                                : DPIF_FP_MODIFY;
4668     if (stats) {
4669         flags |= DPIF_FP_ZERO_STATS;
4670     }
4671
4672     if (path == SF_SLOW_PATH) {
4673         compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4674                           slow_path_stub, sizeof slow_path_stub,
4675                           &actions, &actions_len);
4676     }
4677
4678     ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4679     if (enable_megaflows) {
4680         odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4681                                &facet->flow, UINT32_MAX);
4682     }
4683
4684     ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4685                         subfacet->key_len,  mask.data, mask.size,
4686                         actions, actions_len, stats);
4687
4688     if (stats) {
4689         subfacet_reset_dp_stats(subfacet, stats);
4690     }
4691
4692     if (ret) {
4693         COVERAGE_INC(subfacet_install_fail);
4694     } else {
4695         subfacet->path = path;
4696     }
4697     return ret;
4698 }
4699
4700 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4701 static void
4702 subfacet_uninstall(struct subfacet *subfacet)
4703 {
4704     if (subfacet->path != SF_NOT_INSTALLED) {
4705         struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4706         struct dpif_flow_stats stats;
4707         int error;
4708
4709         error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4710                               subfacet->key_len, &stats);
4711         subfacet_reset_dp_stats(subfacet, &stats);
4712         if (!error) {
4713             subfacet_update_stats(subfacet, &stats);
4714         }
4715         subfacet->path = SF_NOT_INSTALLED;
4716     } else {
4717         ovs_assert(subfacet->dp_packet_count == 0);
4718         ovs_assert(subfacet->dp_byte_count == 0);
4719     }
4720 }
4721
4722 /* Resets 'subfacet''s datapath statistics counters.  This should be called
4723  * when 'subfacet''s statistics are cleared in the datapath.  If 'stats' is
4724  * non-null, it should contain the statistics returned by dpif when 'subfacet'
4725  * was reset in the datapath.  'stats' will be modified to include only
4726  * statistics new since 'subfacet' was last updated. */
4727 static void
4728 subfacet_reset_dp_stats(struct subfacet *subfacet,
4729                         struct dpif_flow_stats *stats)
4730 {
4731     if (stats
4732         && subfacet->dp_packet_count <= stats->n_packets
4733         && subfacet->dp_byte_count <= stats->n_bytes) {
4734         stats->n_packets -= subfacet->dp_packet_count;
4735         stats->n_bytes -= subfacet->dp_byte_count;
4736     }
4737
4738     subfacet->dp_packet_count = 0;
4739     subfacet->dp_byte_count = 0;
4740 }
4741
4742 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4743  *
4744  * Because of the meaning of a subfacet's counters, it only makes sense to do
4745  * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4746  * represents a packet that was sent by hand or if it represents statistics
4747  * that have been cleared out of the datapath. */
4748 static void
4749 subfacet_update_stats(struct subfacet *subfacet,
4750                       const struct dpif_flow_stats *stats)
4751 {
4752     if (stats->n_packets || stats->used > subfacet->used) {
4753         struct facet *facet = subfacet->facet;
4754
4755         subfacet->used = MAX(subfacet->used, stats->used);
4756         facet->used = MAX(facet->used, stats->used);
4757         facet->packet_count += stats->n_packets;
4758         facet->byte_count += stats->n_bytes;
4759         facet->tcp_flags |= stats->tcp_flags;
4760     }
4761 }
4762 \f
4763 /* Rules. */
4764
4765 /* Lookup 'flow' in 'ofproto''s classifier.  If 'wc' is non-null, sets
4766  * the fields that were relevant as part of the lookup. */
4767 void
4768 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4769                  struct flow_wildcards *wc, struct rule_dpif **rule)
4770 {
4771     struct ofport_dpif *port;
4772
4773     if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4774         return;
4775     }
4776     port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4777     if (!port) {
4778         VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4779                      flow->in_port.ofp_port);
4780     }
4781
4782     *rule = choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4783                              ofproto->no_packet_in_rule);
4784     ovs_rwlock_rdlock(&(*rule)->up.evict);
4785 }
4786
4787 bool
4788 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4789                           const struct flow *flow, struct flow_wildcards *wc,
4790                           uint8_t table_id, struct rule_dpif **rule)
4791     OVS_TRY_RDLOCK(true, (*rule)->up.evict)
4792 {
4793     struct cls_rule *cls_rule;
4794     struct classifier *cls;
4795     bool frag;
4796
4797     *rule = NULL;
4798     if (table_id >= N_TABLES) {
4799         return false;
4800     }
4801
4802     if (wc) {
4803         memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4804         wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4805     }
4806
4807     cls = &ofproto->up.tables[table_id].cls;
4808     ovs_rwlock_rdlock(&cls->rwlock);
4809     frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4810     if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4811         /* We must pretend that transport ports are unavailable. */
4812         struct flow ofpc_normal_flow = *flow;
4813         ofpc_normal_flow.tp_src = htons(0);
4814         ofpc_normal_flow.tp_dst = htons(0);
4815         cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4816     } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4817         cls_rule = &ofproto->drop_frags_rule->up.cr;
4818         if (wc) {
4819             flow_wildcards_init_exact(wc);
4820         }
4821     } else {
4822         cls_rule = classifier_lookup(cls, flow, wc);
4823     }
4824
4825     *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4826     if (*rule && ovs_rwlock_tryrdlock(&(*rule)->up.evict)) {
4827         /* The rule is in the process of being removed.  Best we can do is
4828          * pretend it isn't there. */
4829         *rule = NULL;
4830     }
4831     ovs_rwlock_unlock(&cls->rwlock);
4832
4833     return *rule != NULL;
4834 }
4835
4836 /* Given a port configuration (specified as zero if there's no port), chooses
4837  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4838  * flow table miss. */
4839 struct rule_dpif *
4840 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4841                  struct rule_dpif *no_packet_in_rule)
4842 {
4843     return config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4844 }
4845
4846 void
4847 rule_release(struct rule_dpif *rule)
4848     OVS_NO_THREAD_SAFETY_ANALYSIS
4849 {
4850     if (rule) {
4851         ovs_rwlock_unlock(&rule->up.evict);
4852     }
4853 }
4854
4855 static void
4856 complete_operation(struct rule_dpif *rule)
4857 {
4858     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4859
4860     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4861     if (clogged) {
4862         struct dpif_completion *c = xmalloc(sizeof *c);
4863         c->op = rule->up.pending;
4864         list_push_back(&ofproto->completions, &c->list_node);
4865     } else {
4866         ofoperation_complete(rule->up.pending, 0);
4867     }
4868 }
4869
4870 static struct rule *
4871 rule_alloc(void)
4872 {
4873     struct rule_dpif *rule = xmalloc(sizeof *rule);
4874     return &rule->up;
4875 }
4876
4877 static void
4878 rule_dealloc(struct rule *rule_)
4879 {
4880     struct rule_dpif *rule = rule_dpif_cast(rule_);
4881     free(rule);
4882 }
4883
4884 static enum ofperr
4885 rule_construct(struct rule *rule_)
4886 {
4887     struct rule_dpif *rule = rule_dpif_cast(rule_);
4888     ovs_mutex_init(&rule->stats_mutex);
4889     ovs_mutex_lock(&rule->stats_mutex);
4890     rule->packet_count = 0;
4891     rule->byte_count = 0;
4892     ovs_mutex_unlock(&rule->stats_mutex);
4893     return 0;
4894 }
4895
4896 static void
4897 rule_insert(struct rule *rule_)
4898 {
4899     struct rule_dpif *rule = rule_dpif_cast(rule_);
4900     complete_operation(rule);
4901 }
4902
4903 static void
4904 rule_delete(struct rule *rule_)
4905 {
4906     struct rule_dpif *rule = rule_dpif_cast(rule_);
4907     complete_operation(rule);
4908 }
4909
4910 static void
4911 rule_destruct(struct rule *rule_)
4912 {
4913     struct rule_dpif *rule = rule_dpif_cast(rule_);
4914     ovs_mutex_destroy(&rule->stats_mutex);
4915 }
4916
4917 static void
4918 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4919 {
4920     struct rule_dpif *rule = rule_dpif_cast(rule_);
4921
4922     /* push_all_stats() can handle flow misses which, when using the learn
4923      * action, can cause rules to be added and deleted.  This can corrupt our
4924      * caller's datastructures which assume that rule_get_stats() doesn't have
4925      * an impact on the flow table. To be safe, we disable miss handling. */
4926     push_all_stats__(false);
4927
4928     /* Start from historical data for 'rule' itself that are no longer tracked
4929      * in facets.  This counts, for example, facets that have expired. */
4930     ovs_mutex_lock(&rule->stats_mutex);
4931     *packets = rule->packet_count;
4932     *bytes = rule->byte_count;
4933     ovs_mutex_unlock(&rule->stats_mutex);
4934 }
4935
4936 static void
4937 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4938                   struct ofpbuf *packet)
4939 {
4940     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4941     struct dpif_flow_stats stats;
4942     struct xlate_out xout;
4943     struct xlate_in xin;
4944
4945     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4946     rule_credit_stats(rule, &stats);
4947
4948     xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4949     xin.resubmit_stats = &stats;
4950     xlate_actions(&xin, &xout);
4951
4952     execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4953                         xout.odp_actions.size, packet);
4954
4955     xlate_out_uninit(&xout);
4956 }
4957
4958 static enum ofperr
4959 rule_execute(struct rule *rule, const struct flow *flow,
4960              struct ofpbuf *packet)
4961 {
4962     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4963     ofpbuf_delete(packet);
4964     return 0;
4965 }
4966
4967 static void
4968 rule_modify_actions(struct rule *rule_, bool reset_counters)
4969 {
4970     struct rule_dpif *rule = rule_dpif_cast(rule_);
4971
4972     if (reset_counters) {
4973         ovs_mutex_lock(&rule->stats_mutex);
4974         rule->packet_count = 0;
4975         rule->byte_count = 0;
4976         ovs_mutex_unlock(&rule->stats_mutex);
4977     }
4978
4979     complete_operation(rule);
4980 }
4981 \f
4982 /* Sends 'packet' out 'ofport'.
4983  * May modify 'packet'.
4984  * Returns 0 if successful, otherwise a positive errno value. */
4985 static int
4986 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4987 {
4988     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4989     uint64_t odp_actions_stub[1024 / 8];
4990     struct ofpbuf key, odp_actions;
4991     struct dpif_flow_stats stats;
4992     struct odputil_keybuf keybuf;
4993     struct ofpact_output output;
4994     struct xlate_out xout;
4995     struct xlate_in xin;
4996     struct flow flow;
4997     union flow_in_port in_port_;
4998     int error;
4999
5000     ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5001     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5002
5003     /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5004     in_port_.ofp_port = OFPP_NONE;
5005     flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5006     odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5007                                                              OFPP_LOCAL));
5008     dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5009
5010     ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5011     output.port = ofport->up.ofp_port;
5012     output.max_len = 0;
5013
5014     xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5015     xin.ofpacts_len = sizeof output;
5016     xin.ofpacts = &output.ofpact;
5017     xin.resubmit_stats = &stats;
5018     xlate_actions(&xin, &xout);
5019
5020     error = dpif_execute(ofproto->backer->dpif,
5021                          key.data, key.size,
5022                          xout.odp_actions.data, xout.odp_actions.size,
5023                          packet);
5024     xlate_out_uninit(&xout);
5025
5026     if (error) {
5027         VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5028                      ofproto->up.name, netdev_get_name(ofport->up.netdev),
5029                      ovs_strerror(error));
5030     }
5031
5032     ofproto->stats.tx_packets++;
5033     ofproto->stats.tx_bytes += packet->size;
5034     return error;
5035 }
5036
5037 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5038  * The action will state 'slow' as the reason that the action is in the slow
5039  * path.  (This is purely informational: it allows a human viewing "ovs-dpctl
5040  * dump-flows" output to see why a flow is in the slow path.)
5041  *
5042  * The 'stub_size' bytes in 'stub' will be used to store the action.
5043  * 'stub_size' must be large enough for the action.
5044  *
5045  * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5046  * respectively. */
5047 static void
5048 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5049                   enum slow_path_reason slow,
5050                   uint64_t *stub, size_t stub_size,
5051                   const struct nlattr **actionsp, size_t *actions_lenp)
5052 {
5053     union user_action_cookie cookie;
5054     struct ofpbuf buf;
5055
5056     cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5057     cookie.slow_path.unused = 0;
5058     cookie.slow_path.reason = slow;
5059
5060     ofpbuf_use_stack(&buf, stub, stub_size);
5061     if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5062         uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5063                                          ODPP_NONE);
5064         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5065     } else {
5066         odp_port_t odp_port;
5067         uint32_t pid;
5068
5069         odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5070         pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5071         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5072     }
5073     *actionsp = buf.data;
5074     *actions_lenp = buf.size;
5075 }
5076 \f
5077 static bool
5078 set_frag_handling(struct ofproto *ofproto_,
5079                   enum ofp_config_flags frag_handling)
5080 {
5081     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5082     if (frag_handling != OFPC_FRAG_REASM) {
5083         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5084         return true;
5085     } else {
5086         return false;
5087     }
5088 }
5089
5090 static enum ofperr
5091 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5092            const struct flow *flow,
5093            const struct ofpact *ofpacts, size_t ofpacts_len)
5094 {
5095     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5096     struct odputil_keybuf keybuf;
5097     struct dpif_flow_stats stats;
5098     struct xlate_out xout;
5099     struct xlate_in xin;
5100     struct ofpbuf key;
5101
5102
5103     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5104     odp_flow_key_from_flow(&key, flow,
5105                            ofp_port_to_odp_port(ofproto,
5106                                       flow->in_port.ofp_port));
5107
5108     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5109
5110     xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5111     xin.resubmit_stats = &stats;
5112     xin.ofpacts_len = ofpacts_len;
5113     xin.ofpacts = ofpacts;
5114
5115     xlate_actions(&xin, &xout);
5116     dpif_execute(ofproto->backer->dpif, key.data, key.size,
5117                  xout.odp_actions.data, xout.odp_actions.size, packet);
5118     xlate_out_uninit(&xout);
5119
5120     return 0;
5121 }
5122 \f
5123 /* NetFlow. */
5124
5125 static int
5126 set_netflow(struct ofproto *ofproto_,
5127             const struct netflow_options *netflow_options)
5128 {
5129     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5130
5131     if (netflow_options) {
5132         if (!ofproto->netflow) {
5133             ofproto->netflow = netflow_create();
5134             ofproto->backer->need_revalidate = REV_RECONFIGURE;
5135         }
5136         return netflow_set_options(ofproto->netflow, netflow_options);
5137     } else if (ofproto->netflow) {
5138         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5139         netflow_destroy(ofproto->netflow);
5140         ofproto->netflow = NULL;
5141     }
5142
5143     return 0;
5144 }
5145
5146 static void
5147 get_netflow_ids(const struct ofproto *ofproto_,
5148                 uint8_t *engine_type, uint8_t *engine_id)
5149 {
5150     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5151
5152     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5153 }
5154
5155 static void
5156 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5157 {
5158     if (!facet_is_controller_flow(facet) &&
5159         netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5160         struct subfacet *subfacet;
5161         struct ofexpired expired;
5162
5163         LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5164             if (subfacet->path == SF_FAST_PATH) {
5165                 struct dpif_flow_stats stats;
5166
5167                 subfacet_install(subfacet, &facet->xout.odp_actions,
5168                                  &stats);
5169                 subfacet_update_stats(subfacet, &stats);
5170             }
5171         }
5172
5173         expired.flow = facet->flow;
5174         expired.packet_count = facet->packet_count;
5175         expired.byte_count = facet->byte_count;
5176         expired.used = facet->used;
5177         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5178     }
5179 }
5180
5181 static void
5182 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5183 {
5184     struct cls_cursor cursor;
5185     struct facet *facet;
5186
5187     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5188     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5189     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5190         send_active_timeout(ofproto, facet);
5191     }
5192     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5193 }
5194 \f
5195 static struct ofproto_dpif *
5196 ofproto_dpif_lookup(const char *name)
5197 {
5198     struct ofproto_dpif *ofproto;
5199
5200     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5201                              hash_string(name, 0), &all_ofproto_dpifs) {
5202         if (!strcmp(ofproto->up.name, name)) {
5203             return ofproto;
5204         }
5205     }
5206     return NULL;
5207 }
5208
5209 static void
5210 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5211                           const char *argv[], void *aux OVS_UNUSED)
5212 {
5213     struct ofproto_dpif *ofproto;
5214
5215     if (argc > 1) {
5216         ofproto = ofproto_dpif_lookup(argv[1]);
5217         if (!ofproto) {
5218             unixctl_command_reply_error(conn, "no such bridge");
5219             return;
5220         }
5221         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5222         mac_learning_flush(ofproto->ml);
5223         ovs_rwlock_unlock(&ofproto->ml->rwlock);
5224     } else {
5225         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5226             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5227             mac_learning_flush(ofproto->ml);
5228             ovs_rwlock_unlock(&ofproto->ml->rwlock);
5229         }
5230     }
5231
5232     unixctl_command_reply(conn, "table successfully flushed");
5233 }
5234
5235 static struct ofport_dpif *
5236 ofbundle_get_a_port(const struct ofbundle *bundle)
5237 {
5238     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5239                         bundle_node);
5240 }
5241
5242 static void
5243 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5244                          const char *argv[], void *aux OVS_UNUSED)
5245 {
5246     struct ds ds = DS_EMPTY_INITIALIZER;
5247     const struct ofproto_dpif *ofproto;
5248     const struct mac_entry *e;
5249
5250     ofproto = ofproto_dpif_lookup(argv[1]);
5251     if (!ofproto) {
5252         unixctl_command_reply_error(conn, "no such bridge");
5253         return;
5254     }
5255
5256     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
5257     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5258     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5259         struct ofbundle *bundle = e->port.p;
5260         char name[OFP_MAX_PORT_NAME_LEN];
5261
5262         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5263                                name, sizeof name);
5264         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
5265                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
5266                       mac_entry_age(ofproto->ml, e));
5267     }
5268     ovs_rwlock_unlock(&ofproto->ml->rwlock);
5269     unixctl_command_reply(conn, ds_cstr(&ds));
5270     ds_destroy(&ds);
5271 }
5272
5273 struct trace_ctx {
5274     struct xlate_out xout;
5275     struct xlate_in xin;
5276     struct flow flow;
5277     struct ds *result;
5278 };
5279
5280 static void
5281 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5282 {
5283     ds_put_char_multiple(result, '\t', level);
5284     if (!rule) {
5285         ds_put_cstr(result, "No match\n");
5286         return;
5287     }
5288
5289     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5290                   rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5291     cls_rule_format(&rule->up.cr, result);
5292     ds_put_char(result, '\n');
5293
5294     ds_put_char_multiple(result, '\t', level);
5295     ds_put_cstr(result, "OpenFlow ");
5296     ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5297     ds_put_char(result, '\n');
5298 }
5299
5300 static void
5301 trace_format_flow(struct ds *result, int level, const char *title,
5302                   struct trace_ctx *trace)
5303 {
5304     ds_put_char_multiple(result, '\t', level);
5305     ds_put_format(result, "%s: ", title);
5306     if (flow_equal(&trace->xin.flow, &trace->flow)) {
5307         ds_put_cstr(result, "unchanged");
5308     } else {
5309         flow_format(result, &trace->xin.flow);
5310         trace->flow = trace->xin.flow;
5311     }
5312     ds_put_char(result, '\n');
5313 }
5314
5315 static void
5316 trace_format_regs(struct ds *result, int level, const char *title,
5317                   struct trace_ctx *trace)
5318 {
5319     size_t i;
5320
5321     ds_put_char_multiple(result, '\t', level);
5322     ds_put_format(result, "%s:", title);
5323     for (i = 0; i < FLOW_N_REGS; i++) {
5324         ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5325     }
5326     ds_put_char(result, '\n');
5327 }
5328
5329 static void
5330 trace_format_odp(struct ds *result, int level, const char *title,
5331                  struct trace_ctx *trace)
5332 {
5333     struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5334
5335     ds_put_char_multiple(result, '\t', level);
5336     ds_put_format(result, "%s: ", title);
5337     format_odp_actions(result, odp_actions->data, odp_actions->size);
5338     ds_put_char(result, '\n');
5339 }
5340
5341 static void
5342 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5343 {
5344     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5345     struct ds *result = trace->result;
5346
5347     ds_put_char(result, '\n');
5348     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5349     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5350     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
5351     trace_format_rule(result, recurse + 1, rule);
5352 }
5353
5354 static void
5355 trace_report(struct xlate_in *xin, const char *s, int recurse)
5356 {
5357     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5358     struct ds *result = trace->result;
5359
5360     ds_put_char_multiple(result, '\t', recurse);
5361     ds_put_cstr(result, s);
5362     ds_put_char(result, '\n');
5363 }
5364
5365 static void
5366 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5367                       void *aux OVS_UNUSED)
5368 {
5369     const struct dpif_backer *backer;
5370     struct ofproto_dpif *ofproto;
5371     struct ofpbuf odp_key, odp_mask;
5372     struct ofpbuf *packet;
5373     struct ds result;
5374     struct flow flow;
5375     char *s;
5376
5377     packet = NULL;
5378     backer = NULL;
5379     ds_init(&result);
5380     ofpbuf_init(&odp_key, 0);
5381     ofpbuf_init(&odp_mask, 0);
5382
5383     /* Handle "-generate" or a hex string as the last argument. */
5384     if (!strcmp(argv[argc - 1], "-generate")) {
5385         packet = ofpbuf_new(0);
5386         argc--;
5387     } else {
5388         const char *error = eth_from_hex(argv[argc - 1], &packet);
5389         if (!error) {
5390             argc--;
5391         } else if (argc == 4) {
5392             /* The 3-argument form must end in "-generate' or a hex string. */
5393             unixctl_command_reply_error(conn, error);
5394             goto exit;
5395         }
5396     }
5397
5398     /* Parse the flow and determine whether a datapath or
5399      * bridge is specified. If function odp_flow_key_from_string()
5400      * returns 0, the flow is a odp_flow. If function
5401      * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5402     if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5403         /* If the odp_flow is the second argument,
5404          * the datapath name is the first argument. */
5405         if (argc == 3) {
5406             const char *dp_type;
5407             if (!strncmp(argv[1], "ovs-", 4)) {
5408                 dp_type = argv[1] + 4;
5409             } else {
5410                 dp_type = argv[1];
5411             }
5412             backer = shash_find_data(&all_dpif_backers, dp_type);
5413             if (!backer) {
5414                 unixctl_command_reply_error(conn, "Cannot find datapath "
5415                                "of this name");
5416                 goto exit;
5417             }
5418         } else {
5419             /* No datapath name specified, so there should be only one
5420              * datapath. */
5421             struct shash_node *node;
5422             if (shash_count(&all_dpif_backers) != 1) {
5423                 unixctl_command_reply_error(conn, "Must specify datapath "
5424                          "name, there is more than one type of datapath");
5425                 goto exit;
5426             }
5427             node = shash_first(&all_dpif_backers);
5428             backer = node->data;
5429         }
5430
5431         if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5432                           NULL, &ofproto, NULL)) {
5433             unixctl_command_reply_error(conn, "Invalid datapath flow");
5434             goto exit;
5435         }
5436         ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5437     } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5438         if (argc != 3) {
5439             unixctl_command_reply_error(conn, "Must specify bridge name");
5440             goto exit;
5441         }
5442
5443         ofproto = ofproto_dpif_lookup(argv[1]);
5444         if (!ofproto) {
5445             unixctl_command_reply_error(conn, "Unknown bridge name");
5446             goto exit;
5447         }
5448     } else {
5449         unixctl_command_reply_error(conn, "Bad flow syntax");
5450         goto exit;
5451     }
5452
5453     /* Generate a packet, if requested. */
5454     if (packet) {
5455         if (!packet->size) {
5456             flow_compose(packet, &flow);
5457         } else {
5458             union flow_in_port in_port_;
5459
5460             in_port_ = flow.in_port;
5461             ds_put_cstr(&result, "Packet: ");
5462             s = ofp_packet_to_string(packet->data, packet->size);
5463             ds_put_cstr(&result, s);
5464             free(s);
5465
5466             /* Use the metadata from the flow and the packet argument
5467              * to reconstruct the flow. */
5468             flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5469                          &in_port_, &flow);
5470         }
5471     }
5472
5473     ofproto_trace(ofproto, &flow, packet, &result);
5474     unixctl_command_reply(conn, ds_cstr(&result));
5475
5476 exit:
5477     ds_destroy(&result);
5478     ofpbuf_delete(packet);
5479     ofpbuf_uninit(&odp_key);
5480     ofpbuf_uninit(&odp_mask);
5481 }
5482
5483 static void
5484 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5485               const struct ofpbuf *packet, struct ds *ds)
5486 {
5487     struct rule_dpif *rule;
5488     struct flow_wildcards wc;
5489
5490     ds_put_cstr(ds, "Flow: ");
5491     flow_format(ds, flow);
5492     ds_put_char(ds, '\n');
5493
5494     flow_wildcards_init_catchall(&wc);
5495     rule_dpif_lookup(ofproto, flow, &wc, &rule);
5496
5497     trace_format_rule(ds, 0, rule);
5498     if (rule == ofproto->miss_rule) {
5499         ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5500     } else if (rule == ofproto->no_packet_in_rule) {
5501         ds_put_cstr(ds, "\nNo match, packets dropped because "
5502                     "OFPPC_NO_PACKET_IN is set on in_port.\n");
5503     } else if (rule == ofproto->drop_frags_rule) {
5504         ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5505                     "and the fragment handling mode is \"drop\".\n");
5506     }
5507
5508     if (rule) {
5509         uint64_t odp_actions_stub[1024 / 8];
5510         struct ofpbuf odp_actions;
5511         struct trace_ctx trace;
5512         struct match match;
5513         uint8_t tcp_flags;
5514
5515         tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5516         trace.result = ds;
5517         trace.flow = *flow;
5518         ofpbuf_use_stub(&odp_actions,
5519                         odp_actions_stub, sizeof odp_actions_stub);
5520         xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5521         trace.xin.resubmit_hook = trace_resubmit;
5522         trace.xin.report_hook = trace_report;
5523
5524         xlate_actions(&trace.xin, &trace.xout);
5525         flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5526
5527         ds_put_char(ds, '\n');
5528         trace_format_flow(ds, 0, "Final flow", &trace);
5529
5530         match_init(&match, flow, &trace.xout.wc);
5531         ds_put_cstr(ds, "Relevant fields: ");
5532         match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5533         ds_put_char(ds, '\n');
5534
5535         ds_put_cstr(ds, "Datapath actions: ");
5536         format_odp_actions(ds, trace.xout.odp_actions.data,
5537                            trace.xout.odp_actions.size);
5538
5539         if (trace.xout.slow) {
5540             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5541                         "slow path because it:");
5542             switch (trace.xout.slow) {
5543             case SLOW_CFM:
5544                 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5545                 break;
5546             case SLOW_LACP:
5547                 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5548                 break;
5549             case SLOW_STP:
5550                 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5551                 break;
5552             case SLOW_BFD:
5553                 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5554                 break;
5555             case SLOW_CONTROLLER:
5556                 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5557                             "to the OpenFlow controller.");
5558                 break;
5559             case __SLOW_MAX:
5560                 NOT_REACHED();
5561             }
5562         }
5563
5564         xlate_out_uninit(&trace.xout);
5565     }
5566
5567     rule_release(rule);
5568 }
5569
5570 static void
5571 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5572                   const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5573 {
5574     clogged = true;
5575     unixctl_command_reply(conn, NULL);
5576 }
5577
5578 static void
5579 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5580                     const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5581 {
5582     clogged = false;
5583     unixctl_command_reply(conn, NULL);
5584 }
5585
5586 /* Runs a self-check of flow translations in 'ofproto'.  Appends a message to
5587  * 'reply' describing the results. */
5588 static void
5589 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5590 {
5591     struct cls_cursor cursor;
5592     struct facet *facet;
5593     int errors;
5594
5595     errors = 0;
5596     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5597     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5598     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5599         if (!facet_check_consistency(facet)) {
5600             errors++;
5601         }
5602     }
5603     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5604     if (errors) {
5605         ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5606     }
5607
5608     if (errors) {
5609         ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5610                       ofproto->up.name, errors);
5611     } else {
5612         ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5613     }
5614 }
5615
5616 static void
5617 ofproto_dpif_self_check(struct unixctl_conn *conn,
5618                         int argc, const char *argv[], void *aux OVS_UNUSED)
5619 {
5620     struct ds reply = DS_EMPTY_INITIALIZER;
5621     struct ofproto_dpif *ofproto;
5622
5623     if (argc > 1) {
5624         ofproto = ofproto_dpif_lookup(argv[1]);
5625         if (!ofproto) {
5626             unixctl_command_reply_error(conn, "Unknown ofproto (use "
5627                                         "ofproto/list for help)");
5628             return;
5629         }
5630         ofproto_dpif_self_check__(ofproto, &reply);
5631     } else {
5632         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5633             ofproto_dpif_self_check__(ofproto, &reply);
5634         }
5635     }
5636
5637     unixctl_command_reply(conn, ds_cstr(&reply));
5638     ds_destroy(&reply);
5639 }
5640
5641 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
5642  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
5643  * to destroy 'ofproto_shash' and free the returned value. */
5644 static const struct shash_node **
5645 get_ofprotos(struct shash *ofproto_shash)
5646 {
5647     const struct ofproto_dpif *ofproto;
5648
5649     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5650         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5651         shash_add_nocopy(ofproto_shash, name, ofproto);
5652     }
5653
5654     return shash_sort(ofproto_shash);
5655 }
5656
5657 static void
5658 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5659                               const char *argv[] OVS_UNUSED,
5660                               void *aux OVS_UNUSED)
5661 {
5662     struct ds ds = DS_EMPTY_INITIALIZER;
5663     struct shash ofproto_shash;
5664     const struct shash_node **sorted_ofprotos;
5665     int i;
5666
5667     shash_init(&ofproto_shash);
5668     sorted_ofprotos = get_ofprotos(&ofproto_shash);
5669     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5670         const struct shash_node *node = sorted_ofprotos[i];
5671         ds_put_format(&ds, "%s\n", node->name);
5672     }
5673
5674     shash_destroy(&ofproto_shash);
5675     free(sorted_ofprotos);
5676
5677     unixctl_command_reply(conn, ds_cstr(&ds));
5678     ds_destroy(&ds);
5679 }
5680
5681 static void
5682 show_dp_rates(struct ds *ds, const char *heading,
5683               const struct avg_subfacet_rates *rates)
5684 {
5685     ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5686                   heading, rates->add_rate, rates->del_rate);
5687 }
5688
5689 static void
5690 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5691 {
5692     const struct shash_node **ofprotos;
5693     struct ofproto_dpif *ofproto;
5694     struct shash ofproto_shash;
5695     uint64_t n_hit, n_missed;
5696     long long int minutes;
5697     size_t i;
5698
5699     n_hit = n_missed = 0;
5700     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5701         if (ofproto->backer == backer) {
5702             n_missed += ofproto->n_missed;
5703             n_hit += ofproto->n_hit;
5704         }
5705     }
5706
5707     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5708                   dpif_name(backer->dpif), n_hit, n_missed);
5709     ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5710                   " life span: %lldms\n", hmap_count(&backer->subfacets),
5711                   backer->avg_n_subfacet, backer->max_n_subfacet,
5712                   backer->avg_subfacet_life);
5713
5714     minutes = (time_msec() - backer->created) / (1000 * 60);
5715     if (minutes >= 60) {
5716         show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5717     }
5718     if (minutes >= 60 * 24) {
5719         show_dp_rates(ds, "\tdaily avg:",  &backer->daily);
5720     }
5721     show_dp_rates(ds, "\toverall avg:",  &backer->lifetime);
5722
5723     shash_init(&ofproto_shash);
5724     ofprotos = get_ofprotos(&ofproto_shash);
5725     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5726         struct ofproto_dpif *ofproto = ofprotos[i]->data;
5727         const struct shash_node **ports;
5728         size_t j;
5729
5730         if (ofproto->backer != backer) {
5731             continue;
5732         }
5733
5734         ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5735                       ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5736
5737         ports = shash_sort(&ofproto->up.port_by_name);
5738         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5739             const struct shash_node *node = ports[j];
5740             struct ofport *ofport = node->data;
5741             struct smap config;
5742             odp_port_t odp_port;
5743
5744             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5745                           ofport->ofp_port);
5746
5747             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5748             if (odp_port != ODPP_NONE) {
5749                 ds_put_format(ds, "%"PRIu32":", odp_port);
5750             } else {
5751                 ds_put_cstr(ds, "none:");
5752             }
5753
5754             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5755
5756             smap_init(&config);
5757             if (!netdev_get_config(ofport->netdev, &config)) {
5758                 const struct smap_node **nodes;
5759                 size_t i;
5760
5761                 nodes = smap_sort(&config);
5762                 for (i = 0; i < smap_count(&config); i++) {
5763                     const struct smap_node *node = nodes[i];
5764                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5765                                   node->key, node->value);
5766                 }
5767                 free(nodes);
5768             }
5769             smap_destroy(&config);
5770
5771             ds_put_char(ds, ')');
5772             ds_put_char(ds, '\n');
5773         }
5774         free(ports);
5775     }
5776     shash_destroy(&ofproto_shash);
5777     free(ofprotos);
5778 }
5779
5780 static void
5781 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5782                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5783 {
5784     struct ds ds = DS_EMPTY_INITIALIZER;
5785     const struct shash_node **backers;
5786     int i;
5787
5788     backers = shash_sort(&all_dpif_backers);
5789     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5790         dpif_show_backer(backers[i]->data, &ds);
5791     }
5792     free(backers);
5793
5794     unixctl_command_reply(conn, ds_cstr(&ds));
5795     ds_destroy(&ds);
5796 }
5797
5798 /* Dump the megaflow (facet) cache.  This is useful to check the
5799  * correctness of flow wildcarding, since the same mechanism is used for
5800  * both xlate caching and kernel wildcarding.
5801  *
5802  * It's important to note that in the output the flow description uses
5803  * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5804  *
5805  * This command is only needed for advanced debugging, so it's not
5806  * documented in the man page. */
5807 static void
5808 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5809                                     int argc OVS_UNUSED, const char *argv[],
5810                                     void *aux OVS_UNUSED)
5811 {
5812     struct ds ds = DS_EMPTY_INITIALIZER;
5813     const struct ofproto_dpif *ofproto;
5814     long long int now = time_msec();
5815     struct cls_cursor cursor;
5816     struct facet *facet;
5817
5818     ofproto = ofproto_dpif_lookup(argv[1]);
5819     if (!ofproto) {
5820         unixctl_command_reply_error(conn, "no such bridge");
5821         return;
5822     }
5823
5824     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5825     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5826     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5827         cls_rule_format(&facet->cr, &ds);
5828         ds_put_cstr(&ds, ", ");
5829         ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5830         ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5831         ds_put_cstr(&ds, "Datapath actions: ");
5832         if (facet->xout.slow) {
5833             uint64_t slow_path_stub[128 / 8];
5834             const struct nlattr *actions;
5835             size_t actions_len;
5836
5837             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5838                               slow_path_stub, sizeof slow_path_stub,
5839                               &actions, &actions_len);
5840             format_odp_actions(&ds, actions, actions_len);
5841         } else {
5842             format_odp_actions(&ds, facet->xout.odp_actions.data,
5843                                facet->xout.odp_actions.size);
5844         }
5845         ds_put_cstr(&ds, "\n");
5846     }
5847     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5848
5849     ds_chomp(&ds, '\n');
5850     unixctl_command_reply(conn, ds_cstr(&ds));
5851     ds_destroy(&ds);
5852 }
5853
5854 /* Disable using the megaflows.
5855  *
5856  * This command is only needed for advanced debugging, so it's not
5857  * documented in the man page. */
5858 static void
5859 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5860                                        int argc OVS_UNUSED,
5861                                        const char *argv[] OVS_UNUSED,
5862                                        void *aux OVS_UNUSED)
5863 {
5864     struct ofproto_dpif *ofproto;
5865
5866     enable_megaflows = false;
5867
5868     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5869         flush(&ofproto->up);
5870     }
5871
5872     unixctl_command_reply(conn, "megaflows disabled");
5873 }
5874
5875 /* Re-enable using megaflows.
5876  *
5877  * This command is only needed for advanced debugging, so it's not
5878  * documented in the man page. */
5879 static void
5880 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5881                                       int argc OVS_UNUSED,
5882                                       const char *argv[] OVS_UNUSED,
5883                                       void *aux OVS_UNUSED)
5884 {
5885     struct ofproto_dpif *ofproto;
5886
5887     enable_megaflows = true;
5888
5889     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5890         flush(&ofproto->up);
5891     }
5892
5893     unixctl_command_reply(conn, "megaflows enabled");
5894 }
5895
5896 static void
5897 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5898                                 int argc OVS_UNUSED, const char *argv[],
5899                                 void *aux OVS_UNUSED)
5900 {
5901     struct ds ds = DS_EMPTY_INITIALIZER;
5902     const struct ofproto_dpif *ofproto;
5903     struct subfacet *subfacet;
5904
5905     ofproto = ofproto_dpif_lookup(argv[1]);
5906     if (!ofproto) {
5907         unixctl_command_reply_error(conn, "no such bridge");
5908         return;
5909     }
5910
5911     update_stats(ofproto->backer);
5912
5913     HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5914         struct facet *facet = subfacet->facet;
5915         struct odputil_keybuf maskbuf;
5916         struct ofpbuf mask;
5917
5918         if (facet->ofproto != ofproto) {
5919             continue;
5920         }
5921
5922         ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5923         if (enable_megaflows) {
5924             odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5925                                    &facet->flow, UINT32_MAX);
5926         }
5927
5928         odp_flow_format(subfacet->key, subfacet->key_len,
5929                         mask.data, mask.size, &ds, false);
5930
5931         ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5932                       subfacet->dp_packet_count, subfacet->dp_byte_count);
5933         if (subfacet->used) {
5934             ds_put_format(&ds, "%.3fs",
5935                           (time_msec() - subfacet->used) / 1000.0);
5936         } else {
5937             ds_put_format(&ds, "never");
5938         }
5939         if (subfacet->facet->tcp_flags) {
5940             ds_put_cstr(&ds, ", flags:");
5941             packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5942         }
5943
5944         ds_put_cstr(&ds, ", actions:");
5945         if (facet->xout.slow) {
5946             uint64_t slow_path_stub[128 / 8];
5947             const struct nlattr *actions;
5948             size_t actions_len;
5949
5950             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5951                               slow_path_stub, sizeof slow_path_stub,
5952                               &actions, &actions_len);
5953             format_odp_actions(&ds, actions, actions_len);
5954         } else {
5955             format_odp_actions(&ds, facet->xout.odp_actions.data,
5956                                facet->xout.odp_actions.size);
5957         }
5958         ds_put_char(&ds, '\n');
5959     }
5960
5961     unixctl_command_reply(conn, ds_cstr(&ds));
5962     ds_destroy(&ds);
5963 }
5964
5965 static void
5966 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5967                                int argc OVS_UNUSED, const char *argv[],
5968                                void *aux OVS_UNUSED)
5969 {
5970     struct ds ds = DS_EMPTY_INITIALIZER;
5971     struct ofproto_dpif *ofproto;
5972
5973     ofproto = ofproto_dpif_lookup(argv[1]);
5974     if (!ofproto) {
5975         unixctl_command_reply_error(conn, "no such bridge");
5976         return;
5977     }
5978
5979     flush(&ofproto->up);
5980
5981     unixctl_command_reply(conn, ds_cstr(&ds));
5982     ds_destroy(&ds);
5983 }
5984
5985 static void
5986 ofproto_dpif_unixctl_init(void)
5987 {
5988     static bool registered;
5989     if (registered) {
5990         return;
5991     }
5992     registered = true;
5993
5994     unixctl_command_register(
5995         "ofproto/trace",
5996         "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
5997         1, 3, ofproto_unixctl_trace, NULL);
5998     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5999                              ofproto_unixctl_fdb_flush, NULL);
6000     unixctl_command_register("fdb/show", "bridge", 1, 1,
6001                              ofproto_unixctl_fdb_show, NULL);
6002     unixctl_command_register("ofproto/clog", "", 0, 0,
6003                              ofproto_dpif_clog, NULL);
6004     unixctl_command_register("ofproto/unclog", "", 0, 0,
6005                              ofproto_dpif_unclog, NULL);
6006     unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6007                              ofproto_dpif_self_check, NULL);
6008     unixctl_command_register("dpif/dump-dps", "", 0, 0,
6009                              ofproto_unixctl_dpif_dump_dps, NULL);
6010     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6011                              NULL);
6012     unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6013                              ofproto_unixctl_dpif_dump_flows, NULL);
6014     unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6015                              ofproto_unixctl_dpif_del_flows, NULL);
6016     unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6017                              ofproto_unixctl_dpif_dump_megaflows, NULL);
6018     unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6019                              ofproto_unixctl_dpif_disable_megaflows, NULL);
6020     unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6021                              ofproto_unixctl_dpif_enable_megaflows, NULL);
6022 }
6023 \f
6024 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6025  *
6026  * This is deprecated.  It is only for compatibility with broken device drivers
6027  * in old versions of Linux that do not properly support VLANs when VLAN
6028  * devices are not used.  When broken device drivers are no longer in
6029  * widespread use, we will delete these interfaces. */
6030
6031 static int
6032 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6033 {
6034     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6035     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6036
6037     if (realdev_ofp_port == ofport->realdev_ofp_port
6038         && vid == ofport->vlandev_vid) {
6039         return 0;
6040     }
6041
6042     ofproto->backer->need_revalidate = REV_RECONFIGURE;
6043
6044     if (ofport->realdev_ofp_port) {
6045         vsp_remove(ofport);
6046     }
6047     if (realdev_ofp_port && ofport->bundle) {
6048         /* vlandevs are enslaved to their realdevs, so they are not allowed to
6049          * themselves be part of a bundle. */
6050         bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6051     }
6052
6053     ofport->realdev_ofp_port = realdev_ofp_port;
6054     ofport->vlandev_vid = vid;
6055
6056     if (realdev_ofp_port) {
6057         vsp_add(ofport, realdev_ofp_port, vid);
6058     }
6059
6060     return 0;
6061 }
6062
6063 static uint32_t
6064 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6065 {
6066     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6067 }
6068
6069 bool
6070 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6071     OVS_EXCLUDED(ofproto->vsp_mutex)
6072 {
6073     bool ret;
6074
6075     ovs_mutex_lock(&ofproto->vsp_mutex);
6076     ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6077     ovs_mutex_unlock(&ofproto->vsp_mutex);
6078     return ret;
6079 }
6080
6081 static ofp_port_t
6082 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6083                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6084     OVS_REQUIRES(ofproto->vsp_mutex)
6085 {
6086     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6087         int vid = vlan_tci_to_vid(vlan_tci);
6088         const struct vlan_splinter *vsp;
6089
6090         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6091                                  hash_realdev_vid(realdev_ofp_port, vid),
6092                                  &ofproto->realdev_vid_map) {
6093             if (vsp->realdev_ofp_port == realdev_ofp_port
6094                 && vsp->vid == vid) {
6095                 return vsp->vlandev_ofp_port;
6096             }
6097         }
6098     }
6099     return realdev_ofp_port;
6100 }
6101
6102 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6103  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6104  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
6105  * 'vlan_tci' 9, it would return the port number of eth0.9.
6106  *
6107  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6108  * function just returns its 'realdev_ofp_port' argument. */
6109 ofp_port_t
6110 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6111                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6112     OVS_EXCLUDED(ofproto->vsp_mutex)
6113 {
6114     ofp_port_t ret;
6115
6116     ovs_mutex_lock(&ofproto->vsp_mutex);
6117     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6118     ovs_mutex_unlock(&ofproto->vsp_mutex);
6119     return ret;
6120 }
6121
6122 static struct vlan_splinter *
6123 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6124 {
6125     struct vlan_splinter *vsp;
6126
6127     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6128                              hash_ofp_port(vlandev_ofp_port),
6129                              &ofproto->vlandev_map) {
6130         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6131             return vsp;
6132         }
6133     }
6134
6135     return NULL;
6136 }
6137
6138 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6139  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6140  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
6141  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6142  * eth0 and store 9 in '*vid'.
6143  *
6144  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6145  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
6146  * always does.*/
6147 static ofp_port_t
6148 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6149                        ofp_port_t vlandev_ofp_port, int *vid)
6150     OVS_REQUIRES(ofproto->vsp_mutex)
6151 {
6152     if (!hmap_is_empty(&ofproto->vlandev_map)) {
6153         const struct vlan_splinter *vsp;
6154
6155         vsp = vlandev_find(ofproto, vlandev_ofp_port);
6156         if (vsp) {
6157             if (vid) {
6158                 *vid = vsp->vid;
6159             }
6160             return vsp->realdev_ofp_port;
6161         }
6162     }
6163     return 0;
6164 }
6165
6166 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6167  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
6168  * 'flow->in_port' to the "real" device backing the VLAN device, sets
6169  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Otherwise (which is
6170  * always the case unless VLAN splinters are enabled), returns false without
6171  * making any changes. */
6172 bool
6173 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6174     OVS_EXCLUDED(ofproto->vsp_mutex)
6175 {
6176     ofp_port_t realdev;
6177     int vid;
6178
6179     ovs_mutex_lock(&ofproto->vsp_mutex);
6180     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6181     ovs_mutex_unlock(&ofproto->vsp_mutex);
6182     if (!realdev) {
6183         return false;
6184     }
6185
6186     /* Cause the flow to be processed as if it came in on the real device with
6187      * the VLAN device's VLAN ID. */
6188     flow->in_port.ofp_port = realdev;
6189     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6190     return true;
6191 }
6192
6193 static void
6194 vsp_remove(struct ofport_dpif *port)
6195 {
6196     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6197     struct vlan_splinter *vsp;
6198
6199     ovs_mutex_lock(&ofproto->vsp_mutex);
6200     vsp = vlandev_find(ofproto, port->up.ofp_port);
6201     if (vsp) {
6202         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6203         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6204         free(vsp);
6205
6206         port->realdev_ofp_port = 0;
6207     } else {
6208         VLOG_ERR("missing vlan device record");
6209     }
6210     ovs_mutex_unlock(&ofproto->vsp_mutex);
6211 }
6212
6213 static void
6214 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6215 {
6216     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6217
6218     ovs_mutex_lock(&ofproto->vsp_mutex);
6219     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6220         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6221             == realdev_ofp_port)) {
6222         struct vlan_splinter *vsp;
6223
6224         vsp = xmalloc(sizeof *vsp);
6225         vsp->realdev_ofp_port = realdev_ofp_port;
6226         vsp->vlandev_ofp_port = port->up.ofp_port;
6227         vsp->vid = vid;
6228
6229         port->realdev_ofp_port = realdev_ofp_port;
6230
6231         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6232                     hash_ofp_port(port->up.ofp_port));
6233         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6234                     hash_realdev_vid(realdev_ofp_port, vid));
6235     } else {
6236         VLOG_ERR("duplicate vlan device record");
6237     }
6238     ovs_mutex_unlock(&ofproto->vsp_mutex);
6239 }
6240
6241 static odp_port_t
6242 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6243 {
6244     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6245     return ofport ? ofport->odp_port : ODPP_NONE;
6246 }
6247
6248 struct ofport_dpif *
6249 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6250 {
6251     struct ofport_dpif *port;
6252
6253     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6254     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6255                              &backer->odp_to_ofport_map) {
6256         if (port->odp_port == odp_port) {
6257             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6258             return port;
6259         }
6260     }
6261
6262     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6263     return NULL;
6264 }
6265
6266 static ofp_port_t
6267 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6268 {
6269     struct ofport_dpif *port;
6270
6271     port = odp_port_to_ofport(ofproto->backer, odp_port);
6272     if (port && &ofproto->up == port->up.ofproto) {
6273         return port->up.ofp_port;
6274     } else {
6275         return OFPP_NONE;
6276     }
6277 }
6278
6279 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6280  * most heavily weighted element.  'base' designates the rate of decay: after
6281  * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6282  * (about .37). */
6283 static void
6284 exp_mavg(double *avg, int base, double new)
6285 {
6286     *avg = (*avg * (base - 1) + new) / base;
6287 }
6288
6289 static void
6290 update_moving_averages(struct dpif_backer *backer)
6291 {
6292     const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6293     long long int minutes = (time_msec() - backer->created) / min_ms;
6294
6295     if (minutes > 0) {
6296         backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6297             / minutes;
6298         backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6299             / minutes;
6300     } else {
6301         backer->lifetime.add_rate = 0.0;
6302         backer->lifetime.del_rate = 0.0;
6303     }
6304
6305     /* Update hourly averages on the minute boundaries. */
6306     if (time_msec() - backer->last_minute >= min_ms) {
6307         exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6308         exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6309
6310         /* Update daily averages on the hour boundaries. */
6311         if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6312             exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6313             exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6314         }
6315
6316         backer->total_subfacet_add_count += backer->subfacet_add_count;
6317         backer->total_subfacet_del_count += backer->subfacet_del_count;
6318         backer->subfacet_add_count = 0;
6319         backer->subfacet_del_count = 0;
6320         backer->last_minute += min_ms;
6321     }
6322 }
6323
6324 const struct ofproto_class ofproto_dpif_class = {
6325     init,
6326     enumerate_types,
6327     enumerate_names,
6328     del,
6329     port_open_type,
6330     type_run,
6331     type_run_fast,
6332     type_wait,
6333     alloc,
6334     construct,
6335     destruct,
6336     dealloc,
6337     run,
6338     run_fast,
6339     wait,
6340     get_memory_usage,
6341     flush,
6342     get_features,
6343     get_tables,
6344     port_alloc,
6345     port_construct,
6346     port_destruct,
6347     port_dealloc,
6348     port_modified,
6349     port_reconfigured,
6350     port_query_by_name,
6351     port_add,
6352     port_del,
6353     port_get_stats,
6354     port_dump_start,
6355     port_dump_next,
6356     port_dump_done,
6357     port_poll,
6358     port_poll_wait,
6359     port_is_lacp_current,
6360     NULL,                       /* rule_choose_table */
6361     rule_alloc,
6362     rule_construct,
6363     rule_insert,
6364     rule_delete,
6365     rule_destruct,
6366     rule_dealloc,
6367     rule_get_stats,
6368     rule_execute,
6369     rule_modify_actions,
6370     set_frag_handling,
6371     packet_out,
6372     set_netflow,
6373     get_netflow_ids,
6374     set_sflow,
6375     set_ipfix,
6376     set_cfm,
6377     get_cfm_status,
6378     set_bfd,
6379     get_bfd_status,
6380     set_stp,
6381     get_stp_status,
6382     set_stp_port,
6383     get_stp_port_status,
6384     set_queues,
6385     bundle_set,
6386     bundle_remove,
6387     mirror_set__,
6388     mirror_get_stats__,
6389     set_flood_vlans,
6390     is_mirror_output_bundle,
6391     forward_bpdu_changed,
6392     set_mac_table_config,
6393     set_realdev,
6394     NULL,                       /* meter_get_features */
6395     NULL,                       /* meter_set */
6396     NULL,                       /* meter_get */
6397     NULL,                       /* meter_del */
6398     NULL,                       /* group_alloc */
6399     NULL,                       /* group_construct */
6400     NULL,                       /* group_destruct */
6401     NULL,                       /* group_dealloc */
6402     NULL,                       /* group_modify */
6403     NULL,                       /* group_get_stats */
6404 };