ofproto/ofproto-dpif.c

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