ofproto/ofproto-dpif.c

   1 /*
   2  * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18
  19 #include "ofproto/ofproto-dpif.h"
  20 #include "ofproto/ofproto-provider.h"
  21
  22 #include <errno.h>
  23
  24 #include "bfd.h"
  25 #include "bond.h"
  26 #include "bundle.h"
  27 #include "byte-order.h"
  28 #include "connmgr.h"
  29 #include "coverage.h"
  30 #include "cfm.h"
  31 #include "dpif.h"
  32 #include "dynamic-string.h"
  33 #include "fail-open.h"
  34 #include "hmapx.h"
  35 #include "lacp.h"
  36 #include "learn.h"
  37 #include "mac-learning.h"
  38 #include "meta-flow.h"
  39 #include "multipath.h"
  40 #include "netdev-vport.h"
  41 #include "netdev.h"
  42 #include "netlink.h"
  43 #include "nx-match.h"
  44 #include "odp-util.h"
  45 #include "odp-execute.h"
  46 #include "ofp-util.h"
  47 #include "ofpbuf.h"
  48 #include "ofp-actions.h"
  49 #include "ofp-parse.h"
  50 #include "ofp-print.h"
  51 #include "ofproto-dpif-governor.h"
  52 #include "ofproto-dpif-ipfix.h"
  53 #include "ofproto-dpif-mirror.h"
  54 #include "ofproto-dpif-sflow.h"
  55 #include "ofproto-dpif-upcall.h"
  56 #include "ofproto-dpif-xlate.h"
  57 #include "poll-loop.h"
  58 #include "simap.h"
  59 #include "smap.h"
  60 #include "timer.h"
  61 #include "tunnel.h"
  62 #include "unaligned.h"
  63 #include "unixctl.h"
  64 #include "vlan-bitmap.h"
  65 #include "vlog.h"
  66
  67 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  68
  69 COVERAGE_DEFINE(ofproto_dpif_expired);
  70 COVERAGE_DEFINE(facet_changed_rule);
  71 COVERAGE_DEFINE(facet_revalidate);
  72 COVERAGE_DEFINE(facet_unexpected);
  73 COVERAGE_DEFINE(facet_suppress);
  74 COVERAGE_DEFINE(subfacet_install_fail);
  75 COVERAGE_DEFINE(packet_in_overflow);
  76 COVERAGE_DEFINE(flow_mod_overflow);
  77
  78 /* Number of implemented OpenFlow tables. */
  79 enum { N_TABLES = 255 };
  80 enum { TBL_INTERNAL = N_TABLES - 1 };    /* Used for internal hidden rules. */
  81 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
  82
  83 struct flow_miss;
  84 struct facet;
  85
  86 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
  87
  88 struct ofbundle {
  89     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
  90     struct ofproto_dpif *ofproto; /* Owning ofproto. */
  91     void *aux;                  /* Key supplied by ofproto's client. */
  92     char *name;                 /* Identifier for log messages. */
  93
  94     /* Configuration. */
  95     struct list ports;          /* Contains "struct ofport"s. */
  96     enum port_vlan_mode vlan_mode; /* VLAN mode */
  97     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
  98     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
  99                                  * NULL if all VLANs are trunked. */
 100     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 101     struct bond *bond;          /* Nonnull iff more than one port. */
 102     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 103
 104     /* Status. */
 105     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 106 };
 107
 108 static void bundle_remove(struct ofport *);
 109 static void bundle_update(struct ofbundle *);
 110 static void bundle_destroy(struct ofbundle *);
 111 static void bundle_del_port(struct ofport_dpif *);
 112 static void bundle_run(struct ofbundle *);
 113 static void bundle_wait(struct ofbundle *);
 114
 115 static void stp_run(struct ofproto_dpif *ofproto);
 116 static void stp_wait(struct ofproto_dpif *ofproto);
 117 static int set_stp_port(struct ofport *,
 118                         const struct ofproto_port_stp_settings *);
 119
 120 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
 121                               enum slow_path_reason,
 122                               uint64_t *stub, size_t stub_size,
 123                               const struct nlattr **actionsp,
 124                               size_t *actions_lenp);
 125
 126 /* A subfacet (see "struct subfacet" below) has three possible installation
 127  * states:
 128  *
 129  *   - SF_NOT_INSTALLED: Not installed in the datapath.  This will only be the
 130  *     case just after the subfacet is created, just before the subfacet is
 131  *     destroyed, or if the datapath returns an error when we try to install a
 132  *     subfacet.
 133  *
 134  *   - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
 135  *
 136  *   - SF_SLOW_PATH: An action that sends every packet for the subfacet through
 137  *     ofproto_dpif is installed in the datapath.
 138  */
 139 enum subfacet_path {
 140     SF_NOT_INSTALLED,           /* No datapath flow for this subfacet. */
 141     SF_FAST_PATH,               /* Full actions are installed. */
 142     SF_SLOW_PATH,               /* Send-to-userspace action is installed. */
 143 };
 144
 145 /* A dpif flow and actions associated with a facet.
 146  *
 147  * See also the large comment on struct facet. */
 148 struct subfacet {
 149     /* Owners. */
 150     struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
 151     struct list list_node;      /* In struct facet's 'facets' list. */
 152     struct facet *facet;        /* Owning facet. */
 153     struct dpif_backer *backer; /* Owning backer. */
 154
 155     enum odp_key_fitness key_fitness;
 156     struct nlattr *key;
 157     int key_len;
 158
 159     long long int used;         /* Time last used; time created if not used. */
 160     long long int created;      /* Time created. */
 161
 162     uint64_t dp_packet_count;   /* Last known packet count in the datapath. */
 163     uint64_t dp_byte_count;     /* Last known byte count in the datapath. */
 164
 165     enum subfacet_path path;    /* Installed in datapath? */
 166 };
 167
 168 #define SUBFACET_DESTROY_MAX_BATCH 50
 169
 170 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *);
 171 static struct subfacet *subfacet_find(struct dpif_backer *,
 172                                       const struct nlattr *key, size_t key_len,
 173                                       uint32_t key_hash);
 174 static void subfacet_destroy(struct subfacet *);
 175 static void subfacet_destroy__(struct subfacet *);
 176 static void subfacet_destroy_batch(struct dpif_backer *,
 177                                    struct subfacet **, int n);
 178 static void subfacet_reset_dp_stats(struct subfacet *,
 179                                     struct dpif_flow_stats *);
 180 static void subfacet_update_stats(struct subfacet *,
 181                                   const struct dpif_flow_stats *);
 182 static int subfacet_install(struct subfacet *,
 183                             const struct ofpbuf *odp_actions,
 184                             struct dpif_flow_stats *);
 185 static void subfacet_uninstall(struct subfacet *);
 186
 187 /* A unique, non-overlapping instantiation of an OpenFlow flow.
 188  *
 189  * A facet associates a "struct flow", which represents the Open vSwitch
 190  * userspace idea of an exact-match flow, with one or more subfacets.
 191  * While the facet is created based on an exact-match flow, it is stored
 192  * within the ofproto based on the wildcards that could be expressed
 193  * based on the flow table and other configuration.  (See the 'wc'
 194  * description in "struct xlate_out" for more details.)
 195  *
 196  * Each subfacet tracks the datapath's idea of the flow equivalent to
 197  * the facet.  When the kernel module (or other dpif implementation) and
 198  * Open vSwitch userspace agree on the definition of a flow key, there
 199  * is exactly one subfacet per facet.  If the dpif implementation
 200  * supports more-specific flow matching than userspace, however, a facet
 201  * can have more than one subfacet.  Examples include the dpif
 202  * implementation not supporting the same wildcards as userspace or some
 203  * distinction in flow that userspace simply doesn't understand.
 204  *
 205  * Flow expiration works in terms of subfacets, so a facet must have at
 206  * least one subfacet or it will never expire, leaking memory. */
 207 struct facet {
 208     /* Owners. */
 209     struct hmap_node hmap_node;  /* In owning ofproto's 'facets' hmap. */
 210     struct ofproto_dpif *ofproto;
 211
 212     /* Owned data. */
 213     struct list subfacets;
 214     long long int used;         /* Time last used; time created if not used. */
 215
 216     /* Key. */
 217     struct flow flow;           /* Flow of the creating subfacet. */
 218     struct cls_rule cr;         /* In 'ofproto_dpif's facets classifier. */
 219
 220     /* These statistics:
 221      *
 222      *   - Do include packets and bytes sent "by hand", e.g. with
 223      *     dpif_execute().
 224      *
 225      *   - Do include packets and bytes that were obtained from the datapath
 226      *     when a subfacet's statistics were reset (e.g. dpif_flow_put() with
 227      *     DPIF_FP_ZERO_STATS).
 228      *
 229      *   - Do not include packets or bytes that can be obtained from the
 230      *     datapath for any existing subfacet.
 231      */
 232     uint64_t packet_count;       /* Number of packets received. */
 233     uint64_t byte_count;         /* Number of bytes received. */
 234
 235     /* Resubmit statistics. */
 236     uint64_t prev_packet_count;  /* Number of packets from last stats push. */
 237     uint64_t prev_byte_count;    /* Number of bytes from last stats push. */
 238     long long int prev_used;     /* Used time from last stats push. */
 239
 240     /* Accounting. */
 241     uint64_t accounted_bytes;    /* Bytes processed by facet_account(). */
 242     struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
 243     uint8_t tcp_flags;           /* TCP flags seen for this 'rule'. */
 244
 245     struct xlate_out xout;
 246
 247     /* Storage for a single subfacet, to reduce malloc() time and space
 248      * overhead.  (A facet always has at least one subfacet and in the common
 249      * case has exactly one subfacet.  However, 'one_subfacet' may not
 250      * always be valid, since it could have been removed after newer
 251      * subfacets were pushed onto the 'subfacets' list.) */
 252     struct subfacet one_subfacet;
 253
 254     long long int learn_rl;      /* Rate limiter for facet_learn(). */
 255 };
 256
 257 static struct facet *facet_create(const struct flow_miss *);
 258 static void facet_remove(struct facet *);
 259 static void facet_free(struct facet *);
 260
 261 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
 262 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
 263                                         const struct flow *);
 264 static bool facet_revalidate(struct facet *);
 265 static bool facet_check_consistency(struct facet *);
 266
 267 static void facet_flush_stats(struct facet *);
 268
 269 static void facet_reset_counters(struct facet *);
 270 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
 271                             struct dpif_flow_stats *, bool may_learn);
 272 static void facet_push_stats(struct facet *, bool may_learn);
 273 static void facet_learn(struct facet *);
 274 static void facet_account(struct facet *);
 275 static void push_all_stats(void);
 276
 277 static bool facet_is_controller_flow(struct facet *);
 278
 279 struct ofport_dpif {
 280     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 281     struct ofport up;
 282
 283     odp_port_t odp_port;
 284     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 285     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
 286     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 287     struct bfd *bfd;            /* BFD, if any. */
 288     bool may_enable;            /* May be enabled in bonds. */
 289     bool is_tunnel;             /* This port is a tunnel. */
 290     long long int carrier_seq;  /* Carrier status changes. */
 291     struct ofport_dpif *peer;   /* Peer if patch port. */
 292
 293     /* Spanning tree. */
 294     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 295     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 296     long long int stp_state_entered;
 297
 298     /* Queue to DSCP mapping. */
 299     struct ofproto_port_queue *qdscp;
 300     size_t n_qdscp;
 301
 302     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 303      *
 304      * This is deprecated.  It is only for compatibility with broken device
 305      * drivers in old versions of Linux that do not properly support VLANs when
 306      * VLAN devices are not used.  When broken device drivers are no longer in
 307      * widespread use, we will delete these interfaces. */
 308     ofp_port_t realdev_ofp_port;
 309     int vlandev_vid;
 310 };
 311
 312 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 313  *
 314  * This is deprecated.  It is only for compatibility with broken device drivers
 315  * in old versions of Linux that do not properly support VLANs when VLAN
 316  * devices are not used.  When broken device drivers are no longer in
 317  * widespread use, we will delete these interfaces. */
 318 struct vlan_splinter {
 319     struct hmap_node realdev_vid_node;
 320     struct hmap_node vlandev_node;
 321     ofp_port_t realdev_ofp_port;
 322     ofp_port_t vlandev_ofp_port;
 323     int vid;
 324 };
 325
 326 static void vsp_remove(struct ofport_dpif *);
 327 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 328
 329 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 330                                        ofp_port_t);
 331
 332 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 333                                        odp_port_t);
 334
 335 static struct ofport_dpif *
 336 ofport_dpif_cast(const struct ofport *ofport)
 337 {
 338     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 339 }
 340
 341 static void port_run(struct ofport_dpif *);
 342 static void port_run_fast(struct ofport_dpif *);
 343 static void port_wait(struct ofport_dpif *);
 344 static int set_bfd(struct ofport *, const struct smap *);
 345 static int set_cfm(struct ofport *, const struct cfm_settings *);
 346 static void ofport_update_peer(struct ofport_dpif *);
 347 static void run_fast_rl(void);
 348 static int run_fast(struct ofproto *);
 349
 350 struct dpif_completion {
 351     struct list list_node;
 352     struct ofoperation *op;
 353 };
 354
 355 /* Reasons that we might need to revalidate every facet, and corresponding
 356  * coverage counters.
 357  *
 358  * A value of 0 means that there is no need to revalidate.
 359  *
 360  * It would be nice to have some cleaner way to integrate with coverage
 361  * counters, but with only a few reasons I guess this is good enough for
 362  * now. */
 363 enum revalidate_reason {
 364     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 365     REV_STP,                   /* Spanning tree protocol port status change. */
 366     REV_BOND,                  /* Bonding changed. */
 367     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 368     REV_FLOW_TABLE,            /* Flow table changed. */
 369     REV_MAC_LEARNING,          /* Mac learning changed. */
 370     REV_INCONSISTENCY          /* Facet self-check failed. */
 371 };
 372 COVERAGE_DEFINE(rev_reconfigure);
 373 COVERAGE_DEFINE(rev_stp);
 374 COVERAGE_DEFINE(rev_bond);
 375 COVERAGE_DEFINE(rev_port_toggled);
 376 COVERAGE_DEFINE(rev_flow_table);
 377 COVERAGE_DEFINE(rev_mac_learning);
 378 COVERAGE_DEFINE(rev_inconsistency);
 379
 380 struct avg_subfacet_rates {
 381     double add_rate;   /* Moving average of new flows created per minute. */
 382     double del_rate;   /* Moving average of flows deleted per minute. */
 383 };
 384
 385 /* All datapaths of a given type share a single dpif backer instance. */
 386 struct dpif_backer {
 387     char *type;
 388     int refcount;
 389     struct dpif *dpif;
 390     struct udpif *udpif;
 391     struct timer next_expiration;
 392
 393     struct ovs_rwlock odp_to_ofport_lock;
 394     struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
 395
 396     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 397
 398     /* Facet revalidation flags applying to facets which use this backer. */
 399     enum revalidate_reason need_revalidate; /* Revalidate every facet. */
 400
 401     struct hmap drop_keys; /* Set of dropped odp keys. */
 402     bool recv_set_enable; /* Enables or disables receiving packets. */
 403
 404     struct hmap subfacets;
 405     struct governor *governor;
 406
 407     /* Subfacet statistics.
 408      *
 409      * These keep track of the total number of subfacets added and deleted and
 410      * flow life span.  They are useful for computing the flow rates stats
 411      * exposed via "ovs-appctl dpif/show".  The goal is to learn about
 412      * traffic patterns in ways that we can use later to improve Open vSwitch
 413      * performance in new situations.  */
 414     long long int created;           /* Time when it is created. */
 415     unsigned max_n_subfacet;         /* Maximum number of flows */
 416     unsigned avg_n_subfacet;         /* Average number of flows. */
 417     long long int avg_subfacet_life; /* Average life span of subfacets. */
 418
 419     /* The average number of subfacets... */
 420     struct avg_subfacet_rates hourly;   /* ...over the last hour. */
 421     struct avg_subfacet_rates daily;    /* ...over the last day. */
 422     struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
 423     long long int last_minute;          /* Last time 'hourly' was updated. */
 424
 425     /* Number of subfacets added or deleted since 'last_minute'. */
 426     unsigned subfacet_add_count;
 427     unsigned subfacet_del_count;
 428
 429     /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
 430     unsigned long long int total_subfacet_add_count;
 431     unsigned long long int total_subfacet_del_count;
 432
 433     /* Number of upcall handling threads. */
 434     unsigned int n_handler_threads;
 435 };
 436
 437 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 438 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 439
 440 static void drop_key_clear(struct dpif_backer *);
 441 static void update_moving_averages(struct dpif_backer *backer);
 442
 443 struct ofproto_dpif {
 444     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 445     struct ofproto up;
 446     struct dpif_backer *backer;
 447
 448     /* Special OpenFlow rules. */
 449     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 450     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 451     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 452
 453     /* Bridging. */
 454     struct netflow *netflow;
 455     struct dpif_sflow *sflow;
 456     struct dpif_ipfix *ipfix;
 457     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 458     struct mac_learning *ml;
 459     bool has_bonded_bundles;
 460     struct mbridge *mbridge;
 461
 462     /* Facets. */
 463     struct classifier facets;     /* Contains 'struct facet's. */
 464     long long int consistency_rl;
 465
 466     /* Support for debugging async flow mods. */
 467     struct list completions;
 468
 469     struct netdev_stats stats; /* To account packets generated and consumed in
 470                                 * userspace. */
 471
 472     /* Spanning tree. */
 473     struct stp *stp;
 474     long long int stp_last_tick;
 475
 476     /* VLAN splinters. */
 477     struct ovs_mutex vsp_mutex;
 478     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 479     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 480
 481     /* Ports. */
 482     struct sset ports;             /* Set of standard port names. */
 483     struct sset ghost_ports;       /* Ports with no datapath port. */
 484     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 485     int port_poll_errno;           /* Last errno for port_poll() reply. */
 486
 487     /* Per ofproto's dpif stats. */
 488     uint64_t n_hit;
 489     uint64_t n_missed;
 490
 491     /* Work queues. */
 492     struct ovs_mutex flow_mod_mutex;
 493     struct list flow_mods OVS_GUARDED;
 494     size_t n_flow_mods OVS_GUARDED;
 495
 496     struct ovs_mutex pin_mutex;
 497     struct list pins OVS_GUARDED;
 498     size_t n_pins OVS_GUARDED;
 499 };
 500
 501 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"?  (Useful only
 502  * for debugging the asynchronous flow_mod implementation.) */
 503 static bool clogged;
 504
 505 /* By default, flows in the datapath are wildcarded (megaflows).  They
 506  * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
 507 static bool enable_megaflows = true;
 508
 509 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 510 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 511
 512 static void ofproto_dpif_unixctl_init(void);
 513
 514 static inline struct ofproto_dpif *
 515 ofproto_dpif_cast(const struct ofproto *ofproto)
 516 {
 517     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 518     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 519 }
 520
 521 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
 522                                         ofp_port_t ofp_port);
 523 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
 524                           const struct ofpbuf *packet, struct ds *);
 525
 526 /* Upcalls. */
 527 static void handle_upcalls(struct dpif_backer *);
 528
 529 /* Flow expiration. */
 530 static int expire(struct dpif_backer *);
 531
 532 /* NetFlow. */
 533 static void send_netflow_active_timeouts(struct ofproto_dpif *);
 534
 535 /* Utilities. */
 536 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
 537
 538 /* Global variables. */
 539 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 540
 541 /* Initial mappings of port to bridge mappings. */
 542 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 543
 544 /* Executes and takes ownership of 'fm'. */
 545 void
 546 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 547                       struct ofputil_flow_mod *fm)
 548 {
 549     ovs_mutex_lock(&ofproto->flow_mod_mutex);
 550     if (ofproto->n_flow_mods > 1024) {
 551         ovs_mutex_unlock(&ofproto->flow_mod_mutex);
 552         COVERAGE_INC(flow_mod_overflow);
 553         free(fm->ofpacts);
 554         free(fm);
 555         return;
 556     }
 557
 558     list_push_back(&ofproto->flow_mods, &fm->list_node);
 559     ofproto->n_flow_mods++;
 560     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
 561 }
 562
 563 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 564  * Takes ownership of 'pin' and pin->packet. */
 565 void
 566 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 567                             struct ofputil_packet_in *pin)
 568 {
 569     ovs_mutex_lock(&ofproto->pin_mutex);
 570     if (ofproto->n_pins > 1024) {
 571         ovs_mutex_unlock(&ofproto->pin_mutex);
 572         COVERAGE_INC(packet_in_overflow);
 573         free(CONST_CAST(void *, pin->packet));
 574         free(pin);
 575         return;
 576     }
 577
 578     list_push_back(&ofproto->pins, &pin->list_node);
 579     ofproto->n_pins++;
 580     ovs_mutex_unlock(&ofproto->pin_mutex);
 581 }
 582 \f
 583 /* Factory functions. */
 584
 585 static void
 586 init(const struct shash *iface_hints)
 587 {
 588     struct shash_node *node;
 589
 590     /* Make a local copy, since we don't own 'iface_hints' elements. */
 591     SHASH_FOR_EACH(node, iface_hints) {
 592         const struct iface_hint *orig_hint = node->data;
 593         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 594
 595         new_hint->br_name = xstrdup(orig_hint->br_name);
 596         new_hint->br_type = xstrdup(orig_hint->br_type);
 597         new_hint->ofp_port = orig_hint->ofp_port;
 598
 599         shash_add(&init_ofp_ports, node->name, new_hint);
 600     }
 601 }
 602
 603 static void
 604 enumerate_types(struct sset *types)
 605 {
 606     dp_enumerate_types(types);
 607 }
 608
 609 static int
 610 enumerate_names(const char *type, struct sset *names)
 611 {
 612     struct ofproto_dpif *ofproto;
 613
 614     sset_clear(names);
 615     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 616         if (strcmp(type, ofproto->up.type)) {
 617             continue;
 618         }
 619         sset_add(names, ofproto->up.name);
 620     }
 621
 622     return 0;
 623 }
 624
 625 static int
 626 del(const char *type, const char *name)
 627 {
 628     struct dpif *dpif;
 629     int error;
 630
 631     error = dpif_open(name, type, &dpif);
 632     if (!error) {
 633         error = dpif_delete(dpif);
 634         dpif_close(dpif);
 635     }
 636     return error;
 637 }
 638 \f
 639 static const char *
 640 port_open_type(const char *datapath_type, const char *port_type)
 641 {
 642     return dpif_port_open_type(datapath_type, port_type);
 643 }
 644
 645 /* Type functions. */
 646
 647 static void process_dpif_port_changes(struct dpif_backer *);
 648 static void process_dpif_all_ports_changed(struct dpif_backer *);
 649 static void process_dpif_port_change(struct dpif_backer *,
 650                                      const char *devname);
 651 static void process_dpif_port_error(struct dpif_backer *, int error);
 652
 653 static struct ofproto_dpif *
 654 lookup_ofproto_dpif_by_port_name(const char *name)
 655 {
 656     struct ofproto_dpif *ofproto;
 657
 658     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 659         if (sset_contains(&ofproto->ports, name)) {
 660             return ofproto;
 661         }
 662     }
 663
 664     return NULL;
 665 }
 666
 667 static int
 668 type_run(const char *type)
 669 {
 670     static long long int push_timer = LLONG_MIN;
 671     struct dpif_backer *backer;
 672
 673     backer = shash_find_data(&all_dpif_backers, type);
 674     if (!backer) {
 675         /* This is not necessarily a problem, since backers are only
 676          * created on demand. */
 677         return 0;
 678     }
 679
 680     dpif_run(backer->dpif);
 681
 682     /* The most natural place to push facet statistics is when they're pulled
 683      * from the datapath.  However, when there are many flows in the datapath,
 684      * this expensive operation can occur so frequently, that it reduces our
 685      * ability to quickly set up flows.  To reduce the cost, we push statistics
 686      * here instead. */
 687     if (time_msec() > push_timer) {
 688         push_timer = time_msec() + 2000;
 689         push_all_stats();
 690     }
 691
 692     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 693      * and the configuration has now changed to "false", enable receiving
 694      * packets from the datapath. */
 695     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 696         int error;
 697
 698         backer->recv_set_enable = true;
 699
 700         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 701         if (error) {
 702             udpif_recv_set(backer->udpif, 0, false);
 703             VLOG_ERR("Failed to enable receiving packets in dpif.");
 704             return error;
 705         }
 706         udpif_recv_set(backer->udpif, n_handler_threads,
 707                        backer->recv_set_enable);
 708         dpif_flow_flush(backer->dpif);
 709         backer->need_revalidate = REV_RECONFIGURE;
 710     }
 711
 712     /* If the n_handler_threads is reconfigured, call udpif_recv_set()
 713      * to reset the handler threads. */
 714     if (backer->n_handler_threads != n_handler_threads) {
 715         udpif_recv_set(backer->udpif, n_handler_threads,
 716                        backer->recv_set_enable);
 717         backer->n_handler_threads = n_handler_threads;
 718     }
 719
 720     if (backer->need_revalidate) {
 721         struct ofproto_dpif *ofproto;
 722         struct simap_node *node;
 723         struct simap tmp_backers;
 724
 725         /* Handle tunnel garbage collection. */
 726         simap_init(&tmp_backers);
 727         simap_swap(&backer->tnl_backers, &tmp_backers);
 728
 729         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 730             struct ofport_dpif *iter;
 731
 732             if (backer != ofproto->backer) {
 733                 continue;
 734             }
 735
 736             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 737                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 738                 const char *dp_port;
 739
 740                 if (!iter->is_tunnel) {
 741                     continue;
 742                 }
 743
 744                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 745                                                      namebuf, sizeof namebuf);
 746                 node = simap_find(&tmp_backers, dp_port);
 747                 if (node) {
 748                     simap_put(&backer->tnl_backers, dp_port, node->data);
 749                     simap_delete(&tmp_backers, node);
 750                     node = simap_find(&backer->tnl_backers, dp_port);
 751                 } else {
 752                     node = simap_find(&backer->tnl_backers, dp_port);
 753                     if (!node) {
 754                         odp_port_t odp_port = ODPP_NONE;
 755
 756                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 757                                            &odp_port)) {
 758                             simap_put(&backer->tnl_backers, dp_port,
 759                                       odp_to_u32(odp_port));
 760                             node = simap_find(&backer->tnl_backers, dp_port);
 761                         }
 762                     }
 763                 }
 764
 765                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 766                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 767                                          iter->odp_port)) {
 768                     backer->need_revalidate = REV_RECONFIGURE;
 769                 }
 770             }
 771         }
 772
 773         SIMAP_FOR_EACH (node, &tmp_backers) {
 774             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 775         }
 776         simap_destroy(&tmp_backers);
 777
 778         switch (backer->need_revalidate) {
 779         case REV_RECONFIGURE:   COVERAGE_INC(rev_reconfigure);   break;
 780         case REV_STP:           COVERAGE_INC(rev_stp);           break;
 781         case REV_BOND:          COVERAGE_INC(rev_bond);          break;
 782         case REV_PORT_TOGGLED:  COVERAGE_INC(rev_port_toggled);  break;
 783         case REV_FLOW_TABLE:    COVERAGE_INC(rev_flow_table);    break;
 784         case REV_MAC_LEARNING:  COVERAGE_INC(rev_mac_learning);  break;
 785         case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
 786         }
 787         backer->need_revalidate = 0;
 788
 789         /* Clear the drop_keys in case we should now be accepting some
 790          * formerly dropped flows. */
 791         drop_key_clear(backer);
 792
 793         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 794             struct facet *facet, *next;
 795             struct ofport_dpif *ofport;
 796             struct cls_cursor cursor;
 797             struct ofbundle *bundle;
 798
 799             if (ofproto->backer != backer) {
 800                 continue;
 801             }
 802
 803             ovs_rwlock_wrlock(&xlate_rwlock);
 804             xlate_ofproto_set(ofproto, ofproto->up.name,
 805                               ofproto->backer->dpif, ofproto->miss_rule,
 806                               ofproto->no_packet_in_rule, ofproto->ml,
 807                               ofproto->stp, ofproto->mbridge,
 808                               ofproto->sflow, ofproto->ipfix,
 809                               ofproto->up.frag_handling,
 810                               ofproto->up.forward_bpdu,
 811                               connmgr_has_in_band(ofproto->up.connmgr),
 812                               ofproto->netflow != NULL);
 813
 814             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 815                 xlate_bundle_set(ofproto, bundle, bundle->name,
 816                                  bundle->vlan_mode, bundle->vlan,
 817                                  bundle->trunks, bundle->use_priority_tags,
 818                                  bundle->bond, bundle->lacp,
 819                                  bundle->floodable);
 820             }
 821
 822             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 823                 int stp_port = ofport->stp_port
 824                     ? stp_port_no(ofport->stp_port)
 825                     : 0;
 826                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 827                                  ofport->up.ofp_port, ofport->odp_port,
 828                                  ofport->up.netdev, ofport->cfm,
 829                                  ofport->bfd, ofport->peer, stp_port,
 830                                  ofport->qdscp, ofport->n_qdscp,
 831                                  ofport->up.pp.config, ofport->is_tunnel,
 832                                  ofport->may_enable);
 833             }
 834             ovs_rwlock_unlock(&xlate_rwlock);
 835
 836             /* Only ofproto-dpif cares about the facet classifier so we just
 837              * lock cls_cursor_init() to appease the thread safety analysis. */
 838             ovs_rwlock_rdlock(&ofproto->facets.rwlock);
 839             cls_cursor_init(&cursor, &ofproto->facets, NULL);
 840             ovs_rwlock_unlock(&ofproto->facets.rwlock);
 841             CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
 842                 facet_revalidate(facet);
 843                 run_fast_rl();
 844             }
 845         }
 846
 847         udpif_revalidate(backer->udpif);
 848     }
 849
 850     if (!backer->recv_set_enable) {
 851         /* Wake up before a max of 1000ms. */
 852         timer_set_duration(&backer->next_expiration, 1000);
 853     } else if (timer_expired(&backer->next_expiration)) {
 854         int delay = expire(backer);
 855         timer_set_duration(&backer->next_expiration, delay);
 856     }
 857
 858     process_dpif_port_changes(backer);
 859
 860     if (backer->governor) {
 861         size_t n_subfacets;
 862
 863         governor_run(backer->governor);
 864
 865         /* If the governor has shrunk to its minimum size and the number of
 866          * subfacets has dwindled, then drop the governor entirely.
 867          *
 868          * For hysteresis, the number of subfacets to drop the governor is
 869          * smaller than the number needed to trigger its creation. */
 870         n_subfacets = hmap_count(&backer->subfacets);
 871         if (n_subfacets * 4 < flow_eviction_threshold
 872             && governor_is_idle(backer->governor)) {
 873             governor_destroy(backer->governor);
 874             backer->governor = NULL;
 875         }
 876     }
 877
 878     return 0;
 879 }
 880
 881 /* Check for and handle port changes in 'backer''s dpif. */
 882 static void
 883 process_dpif_port_changes(struct dpif_backer *backer)
 884 {
 885     for (;;) {
 886         char *devname;
 887         int error;
 888
 889         error = dpif_port_poll(backer->dpif, &devname);
 890         switch (error) {
 891         case EAGAIN:
 892             return;
 893
 894         case ENOBUFS:
 895             process_dpif_all_ports_changed(backer);
 896             break;
 897
 898         case 0:
 899             process_dpif_port_change(backer, devname);
 900             free(devname);
 901             break;
 902
 903         default:
 904             process_dpif_port_error(backer, error);
 905             break;
 906         }
 907     }
 908 }
 909
 910 static void
 911 process_dpif_all_ports_changed(struct dpif_backer *backer)
 912 {
 913     struct ofproto_dpif *ofproto;
 914     struct dpif_port dpif_port;
 915     struct dpif_port_dump dump;
 916     struct sset devnames;
 917     const char *devname;
 918
 919     sset_init(&devnames);
 920     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 921         if (ofproto->backer == backer) {
 922             struct ofport *ofport;
 923
 924             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 925                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 926             }
 927         }
 928     }
 929     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 930         sset_add(&devnames, dpif_port.name);
 931     }
 932
 933     SSET_FOR_EACH (devname, &devnames) {
 934         process_dpif_port_change(backer, devname);
 935     }
 936     sset_destroy(&devnames);
 937 }
 938
 939 static void
 940 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 941 {
 942     struct ofproto_dpif *ofproto;
 943     struct dpif_port port;
 944
 945     /* Don't report on the datapath's device. */
 946     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 947         return;
 948     }
 949
 950     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 951                    &all_ofproto_dpifs) {
 952         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 953             return;
 954         }
 955     }
 956
 957     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 958     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 959         /* The port was removed.  If we know the datapath,
 960          * report it through poll_set().  If we don't, it may be
 961          * notifying us of a removal we initiated, so ignore it.
 962          * If there's a pending ENOBUFS, let it stand, since
 963          * everything will be reevaluated. */
 964         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 965             sset_add(&ofproto->port_poll_set, devname);
 966             ofproto->port_poll_errno = 0;
 967         }
 968     } else if (!ofproto) {
 969         /* The port was added, but we don't know with which
 970          * ofproto we should associate it.  Delete it. */
 971         dpif_port_del(backer->dpif, port.port_no);
 972     } else {
 973         struct ofport_dpif *ofport;
 974
 975         ofport = ofport_dpif_cast(shash_find_data(
 976                                       &ofproto->up.port_by_name, devname));
 977         if (ofport
 978             && ofport->odp_port != port.port_no
 979             && !odp_port_to_ofport(backer, port.port_no))
 980         {
 981             /* 'ofport''s datapath port number has changed from
 982              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 983              * structures to match. */
 984             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 985             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 986             ofport->odp_port = port.port_no;
 987             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 988                         hash_odp_port(port.port_no));
 989             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 990             backer->need_revalidate = REV_RECONFIGURE;
 991         }
 992     }
 993     dpif_port_destroy(&port);
 994 }
 995
 996 /* Propagate 'error' to all ofprotos based on 'backer'. */
 997 static void
 998 process_dpif_port_error(struct dpif_backer *backer, int error)
 999 {
1000     struct ofproto_dpif *ofproto;
1001
1002     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1003         if (ofproto->backer == backer) {
1004             sset_clear(&ofproto->port_poll_set);
1005             ofproto->port_poll_errno = error;
1006         }
1007     }
1008 }
1009
1010 static int
1011 dpif_backer_run_fast(struct dpif_backer *backer)
1012 {
1013     udpif_run(backer->udpif);
1014     handle_upcalls(backer);
1015
1016     return 0;
1017 }
1018
1019 static int
1020 type_run_fast(const char *type)
1021 {
1022     struct dpif_backer *backer;
1023
1024     backer = shash_find_data(&all_dpif_backers, type);
1025     if (!backer) {
1026         /* This is not necessarily a problem, since backers are only
1027          * created on demand. */
1028         return 0;
1029     }
1030
1031     return dpif_backer_run_fast(backer);
1032 }
1033
1034 static void
1035 run_fast_rl(void)
1036 {
1037     static long long int port_rl = LLONG_MIN;
1038
1039     if (time_msec() >= port_rl) {
1040         struct ofproto_dpif *ofproto;
1041
1042         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1043             run_fast(&ofproto->up);
1044         }
1045         port_rl = time_msec() + 200;
1046     }
1047 }
1048
1049 static void
1050 type_wait(const char *type)
1051 {
1052     struct dpif_backer *backer;
1053
1054     backer = shash_find_data(&all_dpif_backers, type);
1055     if (!backer) {
1056         /* This is not necessarily a problem, since backers are only
1057          * created on demand. */
1058         return;
1059     }
1060
1061     if (backer->governor) {
1062         governor_wait(backer->governor);
1063     }
1064
1065     timer_wait(&backer->next_expiration);
1066     dpif_wait(backer->dpif);
1067     udpif_wait(backer->udpif);
1068 }
1069 \f
1070 /* Basic life-cycle. */
1071
1072 static int add_internal_flows(struct ofproto_dpif *);
1073
1074 static struct ofproto *
1075 alloc(void)
1076 {
1077     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1078     return &ofproto->up;
1079 }
1080
1081 static void
1082 dealloc(struct ofproto *ofproto_)
1083 {
1084     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1085     free(ofproto);
1086 }
1087
1088 static void
1089 close_dpif_backer(struct dpif_backer *backer)
1090 {
1091     struct shash_node *node;
1092
1093     ovs_assert(backer->refcount > 0);
1094
1095     if (--backer->refcount) {
1096         return;
1097     }
1098
1099     drop_key_clear(backer);
1100     hmap_destroy(&backer->drop_keys);
1101
1102     simap_destroy(&backer->tnl_backers);
1103     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1104     hmap_destroy(&backer->odp_to_ofport_map);
1105     node = shash_find(&all_dpif_backers, backer->type);
1106     free(backer->type);
1107     shash_delete(&all_dpif_backers, node);
1108     udpif_destroy(backer->udpif);
1109     dpif_close(backer->dpif);
1110
1111     ovs_assert(hmap_is_empty(&backer->subfacets));
1112     hmap_destroy(&backer->subfacets);
1113     governor_destroy(backer->governor);
1114
1115     free(backer);
1116 }
1117
1118 /* Datapath port slated for removal from datapath. */
1119 struct odp_garbage {
1120     struct list list_node;
1121     odp_port_t odp_port;
1122 };
1123
1124 static int
1125 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1126 {
1127     struct dpif_backer *backer;
1128     struct dpif_port_dump port_dump;
1129     struct dpif_port port;
1130     struct shash_node *node;
1131     struct list garbage_list;
1132     struct odp_garbage *garbage, *next;
1133     struct sset names;
1134     char *backer_name;
1135     const char *name;
1136     int error;
1137
1138     backer = shash_find_data(&all_dpif_backers, type);
1139     if (backer) {
1140         backer->refcount++;
1141         *backerp = backer;
1142         return 0;
1143     }
1144
1145     backer_name = xasprintf("ovs-%s", type);
1146
1147     /* Remove any existing datapaths, since we assume we're the only
1148      * userspace controlling the datapath. */
1149     sset_init(&names);
1150     dp_enumerate_names(type, &names);
1151     SSET_FOR_EACH(name, &names) {
1152         struct dpif *old_dpif;
1153
1154         /* Don't remove our backer if it exists. */
1155         if (!strcmp(name, backer_name)) {
1156             continue;
1157         }
1158
1159         if (dpif_open(name, type, &old_dpif)) {
1160             VLOG_WARN("couldn't open old datapath %s to remove it", name);
1161         } else {
1162             dpif_delete(old_dpif);
1163             dpif_close(old_dpif);
1164         }
1165     }
1166     sset_destroy(&names);
1167
1168     backer = xmalloc(sizeof *backer);
1169
1170     error = dpif_create_and_open(backer_name, type, &backer->dpif);
1171     free(backer_name);
1172     if (error) {
1173         VLOG_ERR("failed to open datapath of type %s: %s", type,
1174                  ovs_strerror(error));
1175         free(backer);
1176         return error;
1177     }
1178     backer->udpif = udpif_create(backer, backer->dpif);
1179
1180     backer->type = xstrdup(type);
1181     backer->governor = NULL;
1182     backer->refcount = 1;
1183     hmap_init(&backer->odp_to_ofport_map);
1184     ovs_rwlock_init(&backer->odp_to_ofport_lock);
1185     hmap_init(&backer->drop_keys);
1186     hmap_init(&backer->subfacets);
1187     timer_set_duration(&backer->next_expiration, 1000);
1188     backer->need_revalidate = 0;
1189     simap_init(&backer->tnl_backers);
1190     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1191     *backerp = backer;
1192
1193     if (backer->recv_set_enable) {
1194         dpif_flow_flush(backer->dpif);
1195     }
1196
1197     /* Loop through the ports already on the datapath and remove any
1198      * that we don't need anymore. */
1199     list_init(&garbage_list);
1200     dpif_port_dump_start(&port_dump, backer->dpif);
1201     while (dpif_port_dump_next(&port_dump, &port)) {
1202         node = shash_find(&init_ofp_ports, port.name);
1203         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1204             garbage = xmalloc(sizeof *garbage);
1205             garbage->odp_port = port.port_no;
1206             list_push_front(&garbage_list, &garbage->list_node);
1207         }
1208     }
1209     dpif_port_dump_done(&port_dump);
1210
1211     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1212         dpif_port_del(backer->dpif, garbage->odp_port);
1213         list_remove(&garbage->list_node);
1214         free(garbage);
1215     }
1216
1217     shash_add(&all_dpif_backers, type, backer);
1218
1219     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1220     if (error) {
1221         VLOG_ERR("failed to listen on datapath of type %s: %s",
1222                  type, ovs_strerror(error));
1223         close_dpif_backer(backer);
1224         return error;
1225     }
1226     udpif_recv_set(backer->udpif, n_handler_threads,
1227                    backer->recv_set_enable);
1228     backer->n_handler_threads = n_handler_threads;
1229
1230     backer->max_n_subfacet = 0;
1231     backer->created = time_msec();
1232     backer->last_minute = backer->created;
1233     memset(&backer->hourly, 0, sizeof backer->hourly);
1234     memset(&backer->daily, 0, sizeof backer->daily);
1235     memset(&backer->lifetime, 0, sizeof backer->lifetime);
1236     backer->subfacet_add_count = 0;
1237     backer->subfacet_del_count = 0;
1238     backer->total_subfacet_add_count = 0;
1239     backer->total_subfacet_del_count = 0;
1240     backer->avg_n_subfacet = 0;
1241     backer->avg_subfacet_life = 0;
1242
1243     return error;
1244 }
1245
1246 static int
1247 construct(struct ofproto *ofproto_)
1248 {
1249     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1250     struct shash_node *node, *next;
1251     odp_port_t max_ports;
1252     int error;
1253
1254     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1255     if (error) {
1256         return error;
1257     }
1258
1259     max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1260     ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports),
1261                                                     ofp_to_u16(OFPP_MAX))));
1262
1263     ofproto->netflow = NULL;
1264     ofproto->sflow = NULL;
1265     ofproto->ipfix = NULL;
1266     ofproto->stp = NULL;
1267     hmap_init(&ofproto->bundles);
1268     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1269     ofproto->mbridge = mbridge_create();
1270     ofproto->has_bonded_bundles = false;
1271     ovs_mutex_init(&ofproto->vsp_mutex);
1272
1273     classifier_init(&ofproto->facets);
1274     ofproto->consistency_rl = LLONG_MIN;
1275
1276     list_init(&ofproto->completions);
1277
1278     ovs_mutex_init(&ofproto->flow_mod_mutex);
1279     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1280     list_init(&ofproto->flow_mods);
1281     ofproto->n_flow_mods = 0;
1282     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1283
1284     ovs_mutex_init(&ofproto->pin_mutex);
1285     ovs_mutex_lock(&ofproto->pin_mutex);
1286     list_init(&ofproto->pins);
1287     ofproto->n_pins = 0;
1288     ovs_mutex_unlock(&ofproto->pin_mutex);
1289
1290     ofproto_dpif_unixctl_init();
1291
1292     hmap_init(&ofproto->vlandev_map);
1293     hmap_init(&ofproto->realdev_vid_map);
1294
1295     sset_init(&ofproto->ports);
1296     sset_init(&ofproto->ghost_ports);
1297     sset_init(&ofproto->port_poll_set);
1298     ofproto->port_poll_errno = 0;
1299
1300     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1301         struct iface_hint *iface_hint = node->data;
1302
1303         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1304             /* Check if the datapath already has this port. */
1305             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1306                 sset_add(&ofproto->ports, node->name);
1307             }
1308
1309             free(iface_hint->br_name);
1310             free(iface_hint->br_type);
1311             free(iface_hint);
1312             shash_delete(&init_ofp_ports, node);
1313         }
1314     }
1315
1316     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1317                 hash_string(ofproto->up.name, 0));
1318     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1319
1320     ofproto_init_tables(ofproto_, N_TABLES);
1321     error = add_internal_flows(ofproto);
1322     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1323
1324     ofproto->n_hit = 0;
1325     ofproto->n_missed = 0;
1326
1327     return error;
1328 }
1329
1330 static int
1331 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1332                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1333 {
1334     struct ofputil_flow_mod fm;
1335     int error;
1336
1337     match_init_catchall(&fm.match);
1338     fm.priority = 0;
1339     match_set_reg(&fm.match, 0, id);
1340     fm.new_cookie = htonll(0);
1341     fm.cookie = htonll(0);
1342     fm.cookie_mask = htonll(0);
1343     fm.modify_cookie = false;
1344     fm.table_id = TBL_INTERNAL;
1345     fm.command = OFPFC_ADD;
1346     fm.idle_timeout = 0;
1347     fm.hard_timeout = 0;
1348     fm.buffer_id = 0;
1349     fm.out_port = 0;
1350     fm.flags = 0;
1351     fm.ofpacts = ofpacts->data;
1352     fm.ofpacts_len = ofpacts->size;
1353
1354     error = ofproto_flow_mod(&ofproto->up, &fm);
1355     if (error) {
1356         VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1357                     id, ofperr_to_string(error));
1358         return error;
1359     }
1360
1361     if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1362                                   rulep)) {
1363         ovs_rwlock_unlock(&(*rulep)->up.evict);
1364     } else {
1365         NOT_REACHED();
1366     }
1367
1368     return 0;
1369 }
1370
1371 static int
1372 add_internal_flows(struct ofproto_dpif *ofproto)
1373 {
1374     struct ofpact_controller *controller;
1375     uint64_t ofpacts_stub[128 / 8];
1376     struct ofpbuf ofpacts;
1377     int error;
1378     int id;
1379
1380     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1381     id = 1;
1382
1383     controller = ofpact_put_CONTROLLER(&ofpacts);
1384     controller->max_len = UINT16_MAX;
1385     controller->controller_id = 0;
1386     controller->reason = OFPR_NO_MATCH;
1387     ofpact_pad(&ofpacts);
1388
1389     error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1390     if (error) {
1391         return error;
1392     }
1393
1394     ofpbuf_clear(&ofpacts);
1395     error = add_internal_flow(ofproto, id++, &ofpacts,
1396                               &ofproto->no_packet_in_rule);
1397     if (error) {
1398         return error;
1399     }
1400
1401     error = add_internal_flow(ofproto, id++, &ofpacts,
1402                               &ofproto->drop_frags_rule);
1403     return error;
1404 }
1405
1406 static void
1407 complete_operations(struct ofproto_dpif *ofproto)
1408 {
1409     struct dpif_completion *c, *next;
1410
1411     LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1412         ofoperation_complete(c->op, 0);
1413         list_remove(&c->list_node);
1414         free(c);
1415     }
1416 }
1417
1418 static void
1419 destruct(struct ofproto *ofproto_)
1420 {
1421     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1422     struct rule_dpif *rule, *next_rule;
1423     struct ofputil_packet_in *pin, *next_pin;
1424     struct ofputil_flow_mod *fm, *next_fm;
1425     struct oftable *table;
1426
1427     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1428     ovs_rwlock_wrlock(&xlate_rwlock);
1429     xlate_remove_ofproto(ofproto);
1430     ovs_rwlock_unlock(&xlate_rwlock);
1431
1432     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1433     complete_operations(ofproto);
1434
1435     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1436         struct cls_cursor cursor;
1437
1438         ovs_rwlock_wrlock(&table->cls.rwlock);
1439         cls_cursor_init(&cursor, &table->cls, NULL);
1440         CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1441             ofproto_rule_destroy(&ofproto->up, &table->cls, &rule->up);
1442         }
1443         ovs_rwlock_unlock(&table->cls.rwlock);
1444     }
1445
1446     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1447     LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &ofproto->flow_mods) {
1448         list_remove(&fm->list_node);
1449         ofproto->n_flow_mods--;
1450         free(fm->ofpacts);
1451         free(fm);
1452     }
1453     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1454     ovs_mutex_destroy(&ofproto->flow_mod_mutex);
1455
1456     ovs_mutex_lock(&ofproto->pin_mutex);
1457     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &ofproto->pins) {
1458         list_remove(&pin->list_node);
1459         ofproto->n_pins--;
1460         free(CONST_CAST(void *, pin->packet));
1461         free(pin);
1462     }
1463     ovs_mutex_unlock(&ofproto->pin_mutex);
1464     ovs_mutex_destroy(&ofproto->pin_mutex);
1465
1466     mbridge_unref(ofproto->mbridge);
1467
1468     netflow_destroy(ofproto->netflow);
1469     dpif_sflow_unref(ofproto->sflow);
1470     hmap_destroy(&ofproto->bundles);
1471     mac_learning_unref(ofproto->ml);
1472
1473     classifier_destroy(&ofproto->facets);
1474
1475     hmap_destroy(&ofproto->vlandev_map);
1476     hmap_destroy(&ofproto->realdev_vid_map);
1477
1478     sset_destroy(&ofproto->ports);
1479     sset_destroy(&ofproto->ghost_ports);
1480     sset_destroy(&ofproto->port_poll_set);
1481
1482     ovs_mutex_destroy(&ofproto->vsp_mutex);
1483
1484     close_dpif_backer(ofproto->backer);
1485 }
1486
1487 static int
1488 run_fast(struct ofproto *ofproto_)
1489 {
1490     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1491     struct ofputil_packet_in *pin, *next_pin;
1492     struct ofputil_flow_mod *fm, *next_fm;
1493     struct list flow_mods, pins;
1494     struct ofport_dpif *ofport;
1495
1496     /* Do not perform any periodic activity required by 'ofproto' while
1497      * waiting for flow restore to complete. */
1498     if (ofproto_get_flow_restore_wait()) {
1499         return 0;
1500     }
1501
1502     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1503     if (ofproto->n_flow_mods) {
1504         flow_mods = ofproto->flow_mods;
1505         list_moved(&flow_mods);
1506         list_init(&ofproto->flow_mods);
1507         ofproto->n_flow_mods = 0;
1508     } else {
1509         list_init(&flow_mods);
1510     }
1511     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1512
1513     LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &flow_mods) {
1514         int error = ofproto_flow_mod(&ofproto->up, fm);
1515         if (error && !VLOG_DROP_WARN(&rl)) {
1516             VLOG_WARN("learning action failed to modify flow table (%s)",
1517                       ofperr_get_name(error));
1518         }
1519
1520         list_remove(&fm->list_node);
1521         free(fm->ofpacts);
1522         free(fm);
1523     }
1524
1525     ovs_mutex_lock(&ofproto->pin_mutex);
1526     if (ofproto->n_pins) {
1527         pins = ofproto->pins;
1528         list_moved(&pins);
1529         list_init(&ofproto->pins);
1530         ofproto->n_pins = 0;
1531     } else {
1532         list_init(&pins);
1533     }
1534     ovs_mutex_unlock(&ofproto->pin_mutex);
1535
1536     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1537         connmgr_send_packet_in(ofproto->up.connmgr, pin);
1538         list_remove(&pin->list_node);
1539         free(CONST_CAST(void *, pin->packet));
1540         free(pin);
1541     }
1542
1543     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1544         port_run_fast(ofport);
1545     }
1546
1547     return 0;
1548 }
1549
1550 static int
1551 run(struct ofproto *ofproto_)
1552 {
1553     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1554     struct ofport_dpif *ofport;
1555     struct ofbundle *bundle;
1556     int error;
1557
1558     if (!clogged) {
1559         complete_operations(ofproto);
1560     }
1561
1562     if (mbridge_need_revalidate(ofproto->mbridge)) {
1563         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1564         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1565         mac_learning_flush(ofproto->ml);
1566         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1567     }
1568
1569     /* Do not perform any periodic activity below required by 'ofproto' while
1570      * waiting for flow restore to complete. */
1571     if (ofproto_get_flow_restore_wait()) {
1572         return 0;
1573     }
1574
1575     error = run_fast(ofproto_);
1576     if (error) {
1577         return error;
1578     }
1579
1580     if (ofproto->netflow) {
1581         if (netflow_run(ofproto->netflow)) {
1582             send_netflow_active_timeouts(ofproto);
1583         }
1584     }
1585     if (ofproto->sflow) {
1586         dpif_sflow_run(ofproto->sflow);
1587     }
1588
1589     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1590         port_run(ofport);
1591     }
1592     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1593         bundle_run(bundle);
1594     }
1595
1596     stp_run(ofproto);
1597     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1598     if (mac_learning_run(ofproto->ml)) {
1599         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1600     }
1601     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1602
1603     /* Check the consistency of a random facet, to aid debugging. */
1604     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1605     if (time_msec() >= ofproto->consistency_rl
1606         && !classifier_is_empty(&ofproto->facets)
1607         && !ofproto->backer->need_revalidate) {
1608         struct cls_table *table;
1609         struct cls_rule *cr;
1610         struct facet *facet;
1611
1612         ofproto->consistency_rl = time_msec() + 250;
1613
1614         table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1615                              struct cls_table, hmap_node);
1616         cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1617                           hmap_node);
1618         facet = CONTAINER_OF(cr, struct facet, cr);
1619
1620         if (!facet_check_consistency(facet)) {
1621             ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1622         }
1623     }
1624     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1625
1626     return 0;
1627 }
1628
1629 static void
1630 wait(struct ofproto *ofproto_)
1631 {
1632     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1633     struct ofport_dpif *ofport;
1634     struct ofbundle *bundle;
1635
1636     if (!clogged && !list_is_empty(&ofproto->completions)) {
1637         poll_immediate_wake();
1638     }
1639
1640     if (ofproto_get_flow_restore_wait()) {
1641         return;
1642     }
1643
1644     if (ofproto->sflow) {
1645         dpif_sflow_wait(ofproto->sflow);
1646     }
1647     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1648         port_wait(ofport);
1649     }
1650     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1651         bundle_wait(bundle);
1652     }
1653     if (ofproto->netflow) {
1654         netflow_wait(ofproto->netflow);
1655     }
1656     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1657     mac_learning_wait(ofproto->ml);
1658     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1659     stp_wait(ofproto);
1660     if (ofproto->backer->need_revalidate) {
1661         /* Shouldn't happen, but if it does just go around again. */
1662         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1663         poll_immediate_wake();
1664     }
1665 }
1666
1667 static void
1668 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1669 {
1670     const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1671     struct cls_cursor cursor;
1672     size_t n_subfacets = 0;
1673     struct facet *facet;
1674
1675     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1676     simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1677     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1678
1679     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1680     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1681     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1682         n_subfacets += list_size(&facet->subfacets);
1683     }
1684     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1685     simap_increase(usage, "subfacets", n_subfacets);
1686 }
1687
1688 static void
1689 flush(struct ofproto *ofproto_)
1690 {
1691     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1692     struct subfacet *subfacet, *next_subfacet;
1693     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1694     int n_batch;
1695
1696     n_batch = 0;
1697     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1698                         &ofproto->backer->subfacets) {
1699         if (subfacet->facet->ofproto != ofproto) {
1700             continue;
1701         }
1702
1703         if (subfacet->path != SF_NOT_INSTALLED) {
1704             batch[n_batch++] = subfacet;
1705             if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1706                 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1707                 n_batch = 0;
1708             }
1709         } else {
1710             subfacet_destroy(subfacet);
1711         }
1712     }
1713
1714     if (n_batch > 0) {
1715         subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1716     }
1717 }
1718
1719 static void
1720 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1721              bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1722 {
1723     *arp_match_ip = true;
1724     *actions = (OFPUTIL_A_OUTPUT |
1725                 OFPUTIL_A_SET_VLAN_VID |
1726                 OFPUTIL_A_SET_VLAN_PCP |
1727                 OFPUTIL_A_STRIP_VLAN |
1728                 OFPUTIL_A_SET_DL_SRC |
1729                 OFPUTIL_A_SET_DL_DST |
1730                 OFPUTIL_A_SET_NW_SRC |
1731                 OFPUTIL_A_SET_NW_DST |
1732                 OFPUTIL_A_SET_NW_TOS |
1733                 OFPUTIL_A_SET_TP_SRC |
1734                 OFPUTIL_A_SET_TP_DST |
1735                 OFPUTIL_A_ENQUEUE);
1736 }
1737
1738 static void
1739 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1740 {
1741     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1742     struct dpif_dp_stats s;
1743     uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1744     uint64_t n_lookup;
1745
1746     strcpy(ots->name, "classifier");
1747
1748     dpif_get_dp_stats(ofproto->backer->dpif, &s);
1749     rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1750     rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1751     rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1752
1753     n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1754     ots->lookup_count = htonll(n_lookup);
1755     ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1756 }
1757
1758 static struct ofport *
1759 port_alloc(void)
1760 {
1761     struct ofport_dpif *port = xmalloc(sizeof *port);
1762     return &port->up;
1763 }
1764
1765 static void
1766 port_dealloc(struct ofport *port_)
1767 {
1768     struct ofport_dpif *port = ofport_dpif_cast(port_);
1769     free(port);
1770 }
1771
1772 static int
1773 port_construct(struct ofport *port_)
1774 {
1775     struct ofport_dpif *port = ofport_dpif_cast(port_);
1776     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1777     const struct netdev *netdev = port->up.netdev;
1778     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1779     struct dpif_port dpif_port;
1780     int error;
1781
1782     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1783     port->bundle = NULL;
1784     port->cfm = NULL;
1785     port->bfd = NULL;
1786     port->may_enable = true;
1787     port->stp_port = NULL;
1788     port->stp_state = STP_DISABLED;
1789     port->is_tunnel = false;
1790     port->peer = NULL;
1791     port->qdscp = NULL;
1792     port->n_qdscp = 0;
1793     port->realdev_ofp_port = 0;
1794     port->vlandev_vid = 0;
1795     port->carrier_seq = netdev_get_carrier_resets(netdev);
1796
1797     if (netdev_vport_is_patch(netdev)) {
1798         /* By bailing out here, we don't submit the port to the sFlow module
1799          * to be considered for counter polling export.  This is correct
1800          * because the patch port represents an interface that sFlow considers
1801          * to be "internal" to the switch as a whole, and therefore not an
1802          * candidate for counter polling. */
1803         port->odp_port = ODPP_NONE;
1804         ofport_update_peer(port);
1805         return 0;
1806     }
1807
1808     error = dpif_port_query_by_name(ofproto->backer->dpif,
1809                                     netdev_vport_get_dpif_port(netdev, namebuf,
1810                                                                sizeof namebuf),
1811                                     &dpif_port);
1812     if (error) {
1813         return error;
1814     }
1815
1816     port->odp_port = dpif_port.port_no;
1817
1818     if (netdev_get_tunnel_config(netdev)) {
1819         tnl_port_add(port, port->up.netdev, port->odp_port);
1820         port->is_tunnel = true;
1821     } else {
1822         /* Sanity-check that a mapping doesn't already exist.  This
1823          * shouldn't happen for non-tunnel ports. */
1824         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1825             VLOG_ERR("port %s already has an OpenFlow port number",
1826                      dpif_port.name);
1827             dpif_port_destroy(&dpif_port);
1828             return EBUSY;
1829         }
1830
1831         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1832         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1833                     hash_odp_port(port->odp_port));
1834         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1835     }
1836     dpif_port_destroy(&dpif_port);
1837
1838     if (ofproto->sflow) {
1839         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1840     }
1841
1842     return 0;
1843 }
1844
1845 static void
1846 port_destruct(struct ofport *port_)
1847 {
1848     struct ofport_dpif *port = ofport_dpif_cast(port_);
1849     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1850     const char *devname = netdev_get_name(port->up.netdev);
1851     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1852     const char *dp_port_name;
1853
1854     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1855     ovs_rwlock_wrlock(&xlate_rwlock);
1856     xlate_ofport_remove(port);
1857     ovs_rwlock_unlock(&xlate_rwlock);
1858
1859     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1860                                               sizeof namebuf);
1861     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1862         /* The underlying device is still there, so delete it.  This
1863          * happens when the ofproto is being destroyed, since the caller
1864          * assumes that removal of attached ports will happen as part of
1865          * destruction. */
1866         if (!port->is_tunnel) {
1867             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1868         }
1869     }
1870
1871     if (port->peer) {
1872         port->peer->peer = NULL;
1873         port->peer = NULL;
1874     }
1875
1876     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1877         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1878         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1879         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1880     }
1881
1882     tnl_port_del(port);
1883     sset_find_and_delete(&ofproto->ports, devname);
1884     sset_find_and_delete(&ofproto->ghost_ports, devname);
1885     bundle_remove(port_);
1886     set_cfm(port_, NULL);
1887     set_bfd(port_, NULL);
1888     if (ofproto->sflow) {
1889         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1890     }
1891
1892     free(port->qdscp);
1893 }
1894
1895 static void
1896 port_modified(struct ofport *port_)
1897 {
1898     struct ofport_dpif *port = ofport_dpif_cast(port_);
1899
1900     if (port->bundle && port->bundle->bond) {
1901         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1902     }
1903
1904     if (port->cfm) {
1905         cfm_set_netdev(port->cfm, port->up.netdev);
1906     }
1907
1908     if (port->bfd) {
1909         bfd_set_netdev(port->bfd, port->up.netdev);
1910     }
1911
1912     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1913                                                 port->odp_port)) {
1914         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1915             REV_RECONFIGURE;
1916     }
1917
1918     ofport_update_peer(port);
1919 }
1920
1921 static void
1922 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1923 {
1924     struct ofport_dpif *port = ofport_dpif_cast(port_);
1925     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1926     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1927
1928     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1929                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1930                    OFPUTIL_PC_NO_PACKET_IN)) {
1931         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1932
1933         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1934             bundle_update(port->bundle);
1935         }
1936     }
1937 }
1938
1939 static int
1940 set_sflow(struct ofproto *ofproto_,
1941           const struct ofproto_sflow_options *sflow_options)
1942 {
1943     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1944     struct dpif_sflow *ds = ofproto->sflow;
1945
1946     if (sflow_options) {
1947         if (!ds) {
1948             struct ofport_dpif *ofport;
1949
1950             ds = ofproto->sflow = dpif_sflow_create();
1951             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1952                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1953             }
1954             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1955         }
1956         dpif_sflow_set_options(ds, sflow_options);
1957     } else {
1958         if (ds) {
1959             dpif_sflow_unref(ds);
1960             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1961             ofproto->sflow = NULL;
1962         }
1963     }
1964     return 0;
1965 }
1966
1967 static int
1968 set_ipfix(
1969     struct ofproto *ofproto_,
1970     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1971     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1972     size_t n_flow_exporters_options)
1973 {
1974     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1975     struct dpif_ipfix *di = ofproto->ipfix;
1976
1977     if (bridge_exporter_options || flow_exporters_options) {
1978         if (!di) {
1979             di = ofproto->ipfix = dpif_ipfix_create();
1980         }
1981         dpif_ipfix_set_options(
1982             di, bridge_exporter_options, flow_exporters_options,
1983             n_flow_exporters_options);
1984     } else {
1985         if (di) {
1986             dpif_ipfix_unref(di);
1987             ofproto->ipfix = NULL;
1988         }
1989     }
1990     return 0;
1991 }
1992
1993 static int
1994 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1995 {
1996     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1997     int error;
1998
1999     if (!s) {
2000         error = 0;
2001     } else {
2002         if (!ofport->cfm) {
2003             struct ofproto_dpif *ofproto;
2004
2005             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2006             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2007             ofport->cfm = cfm_create(ofport->up.netdev);
2008         }
2009
2010         if (cfm_configure(ofport->cfm, s)) {
2011             return 0;
2012         }
2013
2014         error = EINVAL;
2015     }
2016     cfm_unref(ofport->cfm);
2017     ofport->cfm = NULL;
2018     return error;
2019 }
2020
2021 static bool
2022 get_cfm_status(const struct ofport *ofport_,
2023                struct ofproto_cfm_status *status)
2024 {
2025     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2026
2027     if (ofport->cfm) {
2028         status->faults = cfm_get_fault(ofport->cfm);
2029         status->remote_opstate = cfm_get_opup(ofport->cfm);
2030         status->health = cfm_get_health(ofport->cfm);
2031         cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2032         return true;
2033     } else {
2034         return false;
2035     }
2036 }
2037
2038 static int
2039 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2040 {
2041     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2042     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2043     struct bfd *old;
2044
2045     old = ofport->bfd;
2046     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2047                                 cfg, ofport->up.netdev);
2048     if (ofport->bfd != old) {
2049         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2050     }
2051
2052     return 0;
2053 }
2054
2055 static int
2056 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2057 {
2058     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2059
2060     if (ofport->bfd) {
2061         bfd_get_status(ofport->bfd, smap);
2062         return 0;
2063     } else {
2064         return ENOENT;
2065     }
2066 }
2067 \f
2068 /* Spanning Tree. */
2069
2070 static void
2071 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2072 {
2073     struct ofproto_dpif *ofproto = ofproto_;
2074     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2075     struct ofport_dpif *ofport;
2076
2077     ofport = stp_port_get_aux(sp);
2078     if (!ofport) {
2079         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2080                      ofproto->up.name, port_num);
2081     } else {
2082         struct eth_header *eth = pkt->l2;
2083
2084         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2085         if (eth_addr_is_zero(eth->eth_src)) {
2086             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2087                          "with unknown MAC", ofproto->up.name, port_num);
2088         } else {
2089             send_packet(ofport, pkt);
2090         }
2091     }
2092     ofpbuf_delete(pkt);
2093 }
2094
2095 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2096 static int
2097 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2098 {
2099     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2100
2101     /* Only revalidate flows if the configuration changed. */
2102     if (!s != !ofproto->stp) {
2103         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2104     }
2105
2106     if (s) {
2107         if (!ofproto->stp) {
2108             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2109                                       send_bpdu_cb, ofproto);
2110             ofproto->stp_last_tick = time_msec();
2111         }
2112
2113         stp_set_bridge_id(ofproto->stp, s->system_id);
2114         stp_set_bridge_priority(ofproto->stp, s->priority);
2115         stp_set_hello_time(ofproto->stp, s->hello_time);
2116         stp_set_max_age(ofproto->stp, s->max_age);
2117         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2118     }  else {
2119         struct ofport *ofport;
2120
2121         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2122             set_stp_port(ofport, NULL);
2123         }
2124
2125         stp_unref(ofproto->stp);
2126         ofproto->stp = NULL;
2127     }
2128
2129     return 0;
2130 }
2131
2132 static int
2133 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2134 {
2135     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2136
2137     if (ofproto->stp) {
2138         s->enabled = true;
2139         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2140         s->designated_root = stp_get_designated_root(ofproto->stp);
2141         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2142     } else {
2143         s->enabled = false;
2144     }
2145
2146     return 0;
2147 }
2148
2149 static void
2150 update_stp_port_state(struct ofport_dpif *ofport)
2151 {
2152     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2153     enum stp_state state;
2154
2155     /* Figure out new state. */
2156     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2157                              : STP_DISABLED;
2158
2159     /* Update state. */
2160     if (ofport->stp_state != state) {
2161         enum ofputil_port_state of_state;
2162         bool fwd_change;
2163
2164         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2165                     netdev_get_name(ofport->up.netdev),
2166                     stp_state_name(ofport->stp_state),
2167                     stp_state_name(state));
2168         if (stp_learn_in_state(ofport->stp_state)
2169                 != stp_learn_in_state(state)) {
2170             /* xxx Learning action flows should also be flushed. */
2171             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2172             mac_learning_flush(ofproto->ml);
2173             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2174         }
2175         fwd_change = stp_forward_in_state(ofport->stp_state)
2176                         != stp_forward_in_state(state);
2177
2178         ofproto->backer->need_revalidate = REV_STP;
2179         ofport->stp_state = state;
2180         ofport->stp_state_entered = time_msec();
2181
2182         if (fwd_change && ofport->bundle) {
2183             bundle_update(ofport->bundle);
2184         }
2185
2186         /* Update the STP state bits in the OpenFlow port description. */
2187         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2188         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2189                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2190                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2191                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2192                      : 0);
2193         ofproto_port_set_state(&ofport->up, of_state);
2194     }
2195 }
2196
2197 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2198  * caller is responsible for assigning STP port numbers and ensuring
2199  * there are no duplicates. */
2200 static int
2201 set_stp_port(struct ofport *ofport_,
2202              const struct ofproto_port_stp_settings *s)
2203 {
2204     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2205     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2206     struct stp_port *sp = ofport->stp_port;
2207
2208     if (!s || !s->enable) {
2209         if (sp) {
2210             ofport->stp_port = NULL;
2211             stp_port_disable(sp);
2212             update_stp_port_state(ofport);
2213         }
2214         return 0;
2215     } else if (sp && stp_port_no(sp) != s->port_num
2216             && ofport == stp_port_get_aux(sp)) {
2217         /* The port-id changed, so disable the old one if it's not
2218          * already in use by another port. */
2219         stp_port_disable(sp);
2220     }
2221
2222     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2223     stp_port_enable(sp);
2224
2225     stp_port_set_aux(sp, ofport);
2226     stp_port_set_priority(sp, s->priority);
2227     stp_port_set_path_cost(sp, s->path_cost);
2228
2229     update_stp_port_state(ofport);
2230
2231     return 0;
2232 }
2233
2234 static int
2235 get_stp_port_status(struct ofport *ofport_,
2236                     struct ofproto_port_stp_status *s)
2237 {
2238     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2239     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2240     struct stp_port *sp = ofport->stp_port;
2241
2242     if (!ofproto->stp || !sp) {
2243         s->enabled = false;
2244         return 0;
2245     }
2246
2247     s->enabled = true;
2248     s->port_id = stp_port_get_id(sp);
2249     s->state = stp_port_get_state(sp);
2250     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2251     s->role = stp_port_get_role(sp);
2252     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2253
2254     return 0;
2255 }
2256
2257 static void
2258 stp_run(struct ofproto_dpif *ofproto)
2259 {
2260     if (ofproto->stp) {
2261         long long int now = time_msec();
2262         long long int elapsed = now - ofproto->stp_last_tick;
2263         struct stp_port *sp;
2264
2265         if (elapsed > 0) {
2266             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2267             ofproto->stp_last_tick = now;
2268         }
2269         while (stp_get_changed_port(ofproto->stp, &sp)) {
2270             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2271
2272             if (ofport) {
2273                 update_stp_port_state(ofport);
2274             }
2275         }
2276
2277         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2278             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2279             mac_learning_flush(ofproto->ml);
2280             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2281         }
2282     }
2283 }
2284
2285 static void
2286 stp_wait(struct ofproto_dpif *ofproto)
2287 {
2288     if (ofproto->stp) {
2289         poll_timer_wait(1000);
2290     }
2291 }
2292 \f
2293 static int
2294 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2295            size_t n_qdscp)
2296 {
2297     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2298     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2299
2300     if (ofport->n_qdscp != n_qdscp
2301         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2302                               n_qdscp * sizeof *qdscp))) {
2303         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2304         free(ofport->qdscp);
2305         ofport->qdscp = n_qdscp
2306             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2307             : NULL;
2308         ofport->n_qdscp = n_qdscp;
2309     }
2310
2311     return 0;
2312 }
2313 \f
2314 /* Bundles. */
2315
2316 /* Expires all MAC learning entries associated with 'bundle' and forces its
2317  * ofproto to revalidate every flow.
2318  *
2319  * Normally MAC learning entries are removed only from the ofproto associated
2320  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2321  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2322  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2323  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2324  * with the host from which it migrated. */
2325 static void
2326 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2327 {
2328     struct ofproto_dpif *ofproto = bundle->ofproto;
2329     struct mac_learning *ml = ofproto->ml;
2330     struct mac_entry *mac, *next_mac;
2331
2332     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2333     ovs_rwlock_wrlock(&ml->rwlock);
2334     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2335         if (mac->port.p == bundle) {
2336             if (all_ofprotos) {
2337                 struct ofproto_dpif *o;
2338
2339                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2340                     if (o != ofproto) {
2341                         struct mac_entry *e;
2342
2343                         ovs_rwlock_wrlock(&o->ml->rwlock);
2344                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2345                         if (e) {
2346                             mac_learning_expire(o->ml, e);
2347                         }
2348                         ovs_rwlock_unlock(&o->ml->rwlock);
2349                     }
2350                 }
2351             }
2352
2353             mac_learning_expire(ml, mac);
2354         }
2355     }
2356     ovs_rwlock_unlock(&ml->rwlock);
2357 }
2358
2359 static struct ofbundle *
2360 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2361 {
2362     struct ofbundle *bundle;
2363
2364     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2365                              &ofproto->bundles) {
2366         if (bundle->aux == aux) {
2367             return bundle;
2368         }
2369     }
2370     return NULL;
2371 }
2372
2373 static void
2374 bundle_update(struct ofbundle *bundle)
2375 {
2376     struct ofport_dpif *port;
2377
2378     bundle->floodable = true;
2379     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2380         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2381             || !stp_forward_in_state(port->stp_state)) {
2382             bundle->floodable = false;
2383             break;
2384         }
2385     }
2386 }
2387
2388 static void
2389 bundle_del_port(struct ofport_dpif *port)
2390 {
2391     struct ofbundle *bundle = port->bundle;
2392
2393     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2394
2395     list_remove(&port->bundle_node);
2396     port->bundle = NULL;
2397
2398     if (bundle->lacp) {
2399         lacp_slave_unregister(bundle->lacp, port);
2400     }
2401     if (bundle->bond) {
2402         bond_slave_unregister(bundle->bond, port);
2403     }
2404
2405     bundle_update(bundle);
2406 }
2407
2408 static bool
2409 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2410                 struct lacp_slave_settings *lacp)
2411 {
2412     struct ofport_dpif *port;
2413
2414     port = get_ofp_port(bundle->ofproto, ofp_port);
2415     if (!port) {
2416         return false;
2417     }
2418
2419     if (port->bundle != bundle) {
2420         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2421         if (port->bundle) {
2422             bundle_remove(&port->up);
2423         }
2424
2425         port->bundle = bundle;
2426         list_push_back(&bundle->ports, &port->bundle_node);
2427         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2428             || !stp_forward_in_state(port->stp_state)) {
2429             bundle->floodable = false;
2430         }
2431     }
2432     if (lacp) {
2433         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2434         lacp_slave_register(bundle->lacp, port, lacp);
2435     }
2436
2437     return true;
2438 }
2439
2440 static void
2441 bundle_destroy(struct ofbundle *bundle)
2442 {
2443     struct ofproto_dpif *ofproto;
2444     struct ofport_dpif *port, *next_port;
2445
2446     if (!bundle) {
2447         return;
2448     }
2449
2450     ofproto = bundle->ofproto;
2451     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2452
2453     ovs_rwlock_wrlock(&xlate_rwlock);
2454     xlate_bundle_remove(bundle);
2455     ovs_rwlock_unlock(&xlate_rwlock);
2456
2457     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2458         bundle_del_port(port);
2459     }
2460
2461     bundle_flush_macs(bundle, true);
2462     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2463     free(bundle->name);
2464     free(bundle->trunks);
2465     lacp_unref(bundle->lacp);
2466     bond_unref(bundle->bond);
2467     free(bundle);
2468 }
2469
2470 static int
2471 bundle_set(struct ofproto *ofproto_, void *aux,
2472            const struct ofproto_bundle_settings *s)
2473 {
2474     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2475     bool need_flush = false;
2476     struct ofport_dpif *port;
2477     struct ofbundle *bundle;
2478     unsigned long *trunks;
2479     int vlan;
2480     size_t i;
2481     bool ok;
2482
2483     if (!s) {
2484         bundle_destroy(bundle_lookup(ofproto, aux));
2485         return 0;
2486     }
2487
2488     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2489     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2490
2491     bundle = bundle_lookup(ofproto, aux);
2492     if (!bundle) {
2493         bundle = xmalloc(sizeof *bundle);
2494
2495         bundle->ofproto = ofproto;
2496         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2497                     hash_pointer(aux, 0));
2498         bundle->aux = aux;
2499         bundle->name = NULL;
2500
2501         list_init(&bundle->ports);
2502         bundle->vlan_mode = PORT_VLAN_TRUNK;
2503         bundle->vlan = -1;
2504         bundle->trunks = NULL;
2505         bundle->use_priority_tags = s->use_priority_tags;
2506         bundle->lacp = NULL;
2507         bundle->bond = NULL;
2508
2509         bundle->floodable = true;
2510         mbridge_register_bundle(ofproto->mbridge, bundle);
2511     }
2512
2513     if (!bundle->name || strcmp(s->name, bundle->name)) {
2514         free(bundle->name);
2515         bundle->name = xstrdup(s->name);
2516     }
2517
2518     /* LACP. */
2519     if (s->lacp) {
2520         if (!bundle->lacp) {
2521             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2522             bundle->lacp = lacp_create();
2523         }
2524         lacp_configure(bundle->lacp, s->lacp);
2525     } else {
2526         lacp_unref(bundle->lacp);
2527         bundle->lacp = NULL;
2528     }
2529
2530     /* Update set of ports. */
2531     ok = true;
2532     for (i = 0; i < s->n_slaves; i++) {
2533         if (!bundle_add_port(bundle, s->slaves[i],
2534                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2535             ok = false;
2536         }
2537     }
2538     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2539         struct ofport_dpif *next_port;
2540
2541         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2542             for (i = 0; i < s->n_slaves; i++) {
2543                 if (s->slaves[i] == port->up.ofp_port) {
2544                     goto found;
2545                 }
2546             }
2547
2548             bundle_del_port(port);
2549         found: ;
2550         }
2551     }
2552     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2553
2554     if (list_is_empty(&bundle->ports)) {
2555         bundle_destroy(bundle);
2556         return EINVAL;
2557     }
2558
2559     /* Set VLAN tagging mode */
2560     if (s->vlan_mode != bundle->vlan_mode
2561         || s->use_priority_tags != bundle->use_priority_tags) {
2562         bundle->vlan_mode = s->vlan_mode;
2563         bundle->use_priority_tags = s->use_priority_tags;
2564         need_flush = true;
2565     }
2566
2567     /* Set VLAN tag. */
2568     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2569             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2570             : 0);
2571     if (vlan != bundle->vlan) {
2572         bundle->vlan = vlan;
2573         need_flush = true;
2574     }
2575
2576     /* Get trunked VLANs. */
2577     switch (s->vlan_mode) {
2578     case PORT_VLAN_ACCESS:
2579         trunks = NULL;
2580         break;
2581
2582     case PORT_VLAN_TRUNK:
2583         trunks = CONST_CAST(unsigned long *, s->trunks);
2584         break;
2585
2586     case PORT_VLAN_NATIVE_UNTAGGED:
2587     case PORT_VLAN_NATIVE_TAGGED:
2588         if (vlan != 0 && (!s->trunks
2589                           || !bitmap_is_set(s->trunks, vlan)
2590                           || bitmap_is_set(s->trunks, 0))) {
2591             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2592             if (s->trunks) {
2593                 trunks = bitmap_clone(s->trunks, 4096);
2594             } else {
2595                 trunks = bitmap_allocate1(4096);
2596             }
2597             bitmap_set1(trunks, vlan);
2598             bitmap_set0(trunks, 0);
2599         } else {
2600             trunks = CONST_CAST(unsigned long *, s->trunks);
2601         }
2602         break;
2603
2604     default:
2605         NOT_REACHED();
2606     }
2607     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2608         free(bundle->trunks);
2609         if (trunks == s->trunks) {
2610             bundle->trunks = vlan_bitmap_clone(trunks);
2611         } else {
2612             bundle->trunks = trunks;
2613             trunks = NULL;
2614         }
2615         need_flush = true;
2616     }
2617     if (trunks != s->trunks) {
2618         free(trunks);
2619     }
2620
2621     /* Bonding. */
2622     if (!list_is_short(&bundle->ports)) {
2623         bundle->ofproto->has_bonded_bundles = true;
2624         if (bundle->bond) {
2625             if (bond_reconfigure(bundle->bond, s->bond)) {
2626                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2627             }
2628         } else {
2629             bundle->bond = bond_create(s->bond);
2630             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2631         }
2632
2633         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2634             bond_slave_register(bundle->bond, port, port->up.netdev);
2635         }
2636     } else {
2637         bond_unref(bundle->bond);
2638         bundle->bond = NULL;
2639     }
2640
2641     /* If we changed something that would affect MAC learning, un-learn
2642      * everything on this port and force flow revalidation. */
2643     if (need_flush) {
2644         bundle_flush_macs(bundle, false);
2645     }
2646
2647     return 0;
2648 }
2649
2650 static void
2651 bundle_remove(struct ofport *port_)
2652 {
2653     struct ofport_dpif *port = ofport_dpif_cast(port_);
2654     struct ofbundle *bundle = port->bundle;
2655
2656     if (bundle) {
2657         bundle_del_port(port);
2658         if (list_is_empty(&bundle->ports)) {
2659             bundle_destroy(bundle);
2660         } else if (list_is_short(&bundle->ports)) {
2661             bond_unref(bundle->bond);
2662             bundle->bond = NULL;
2663         }
2664     }
2665 }
2666
2667 static void
2668 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2669 {
2670     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2671     struct ofport_dpif *port = port_;
2672     uint8_t ea[ETH_ADDR_LEN];
2673     int error;
2674
2675     error = netdev_get_etheraddr(port->up.netdev, ea);
2676     if (!error) {
2677         struct ofpbuf packet;
2678         void *packet_pdu;
2679
2680         ofpbuf_init(&packet, 0);
2681         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2682                                  pdu_size);
2683         memcpy(packet_pdu, pdu, pdu_size);
2684
2685         send_packet(port, &packet);
2686         ofpbuf_uninit(&packet);
2687     } else {
2688         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2689                     "%s (%s)", port->bundle->name,
2690                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2691     }
2692 }
2693
2694 static void
2695 bundle_send_learning_packets(struct ofbundle *bundle)
2696 {
2697     struct ofproto_dpif *ofproto = bundle->ofproto;
2698     int error, n_packets, n_errors;
2699     struct mac_entry *e;
2700
2701     error = n_packets = n_errors = 0;
2702     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2703     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2704         if (e->port.p != bundle) {
2705             struct ofpbuf *learning_packet;
2706             struct ofport_dpif *port;
2707             void *port_void;
2708             int ret;
2709
2710             /* The assignment to "port" is unnecessary but makes "grep"ing for
2711              * struct ofport_dpif more effective. */
2712             learning_packet = bond_compose_learning_packet(bundle->bond,
2713                                                            e->mac, e->vlan,
2714                                                            &port_void);
2715             port = port_void;
2716             ret = send_packet(port, learning_packet);
2717             ofpbuf_delete(learning_packet);
2718             if (ret) {
2719                 error = ret;
2720                 n_errors++;
2721             }
2722             n_packets++;
2723         }
2724     }
2725     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2726
2727     if (n_errors) {
2728         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2729         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2730                      "packets, last error was: %s",
2731                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2732     } else {
2733         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2734                  bundle->name, n_packets);
2735     }
2736 }
2737
2738 static void
2739 bundle_run(struct ofbundle *bundle)
2740 {
2741     if (bundle->lacp) {
2742         lacp_run(bundle->lacp, send_pdu_cb);
2743     }
2744     if (bundle->bond) {
2745         struct ofport_dpif *port;
2746
2747         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2748             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2749         }
2750
2751         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2752             bundle->ofproto->backer->need_revalidate = REV_BOND;
2753         }
2754
2755         if (bond_should_send_learning_packets(bundle->bond)) {
2756             bundle_send_learning_packets(bundle);
2757         }
2758     }
2759 }
2760
2761 static void
2762 bundle_wait(struct ofbundle *bundle)
2763 {
2764     if (bundle->lacp) {
2765         lacp_wait(bundle->lacp);
2766     }
2767     if (bundle->bond) {
2768         bond_wait(bundle->bond);
2769     }
2770 }
2771 \f
2772 /* Mirrors. */
2773
2774 static int
2775 mirror_set__(struct ofproto *ofproto_, void *aux,
2776              const struct ofproto_mirror_settings *s)
2777 {
2778     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2779     struct ofbundle **srcs, **dsts;
2780     int error;
2781     size_t i;
2782
2783     if (!s) {
2784         mirror_destroy(ofproto->mbridge, aux);
2785         return 0;
2786     }
2787
2788     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2789     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2790
2791     for (i = 0; i < s->n_srcs; i++) {
2792         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2793     }
2794
2795     for (i = 0; i < s->n_dsts; i++) {
2796         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2797     }
2798
2799     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2800                        s->n_dsts, s->src_vlans,
2801                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2802     free(srcs);
2803     free(dsts);
2804     return error;
2805 }
2806
2807 static int
2808 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2809                    uint64_t *packets, uint64_t *bytes)
2810 {
2811     push_all_stats();
2812     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2813                             bytes);
2814 }
2815
2816 static int
2817 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2818 {
2819     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2820     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2821     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2822         mac_learning_flush(ofproto->ml);
2823     }
2824     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2825     return 0;
2826 }
2827
2828 static bool
2829 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2830 {
2831     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2832     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2833     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2834 }
2835
2836 static void
2837 forward_bpdu_changed(struct ofproto *ofproto_)
2838 {
2839     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2840     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2841 }
2842
2843 static void
2844 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2845                      size_t max_entries)
2846 {
2847     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2848     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2849     mac_learning_set_idle_time(ofproto->ml, idle_time);
2850     mac_learning_set_max_entries(ofproto->ml, max_entries);
2851     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2852 }
2853 \f
2854 /* Ports. */
2855
2856 static struct ofport_dpif *
2857 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2858 {
2859     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2860     return ofport ? ofport_dpif_cast(ofport) : NULL;
2861 }
2862
2863 static struct ofport_dpif *
2864 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2865 {
2866     struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2867     return port && &ofproto->up == port->up.ofproto ? port : NULL;
2868 }
2869
2870 static void
2871 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2872                             struct ofproto_port *ofproto_port,
2873                             struct dpif_port *dpif_port)
2874 {
2875     ofproto_port->name = dpif_port->name;
2876     ofproto_port->type = dpif_port->type;
2877     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2878 }
2879
2880 static void
2881 ofport_update_peer(struct ofport_dpif *ofport)
2882 {
2883     const struct ofproto_dpif *ofproto;
2884     struct dpif_backer *backer;
2885     char *peer_name;
2886
2887     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2888         return;
2889     }
2890
2891     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2892     backer->need_revalidate = REV_RECONFIGURE;
2893
2894     if (ofport->peer) {
2895         ofport->peer->peer = NULL;
2896         ofport->peer = NULL;
2897     }
2898
2899     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2900     if (!peer_name) {
2901         return;
2902     }
2903
2904     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2905         struct ofport *peer_ofport;
2906         struct ofport_dpif *peer;
2907         char *peer_peer;
2908
2909         if (ofproto->backer != backer) {
2910             continue;
2911         }
2912
2913         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2914         if (!peer_ofport) {
2915             continue;
2916         }
2917
2918         peer = ofport_dpif_cast(peer_ofport);
2919         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2920         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2921                                  peer_peer)) {
2922             ofport->peer = peer;
2923             ofport->peer->peer = ofport;
2924         }
2925         free(peer_peer);
2926
2927         break;
2928     }
2929     free(peer_name);
2930 }
2931
2932 static void
2933 port_run_fast(struct ofport_dpif *ofport)
2934 {
2935     if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2936         struct ofpbuf packet;
2937
2938         ofpbuf_init(&packet, 0);
2939         cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2940         send_packet(ofport, &packet);
2941         ofpbuf_uninit(&packet);
2942     }
2943
2944     if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2945         struct ofpbuf packet;
2946
2947         ofpbuf_init(&packet, 0);
2948         bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2949         send_packet(ofport, &packet);
2950         ofpbuf_uninit(&packet);
2951     }
2952 }
2953
2954 static void
2955 port_run(struct ofport_dpif *ofport)
2956 {
2957     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2958     bool carrier_changed = carrier_seq != ofport->carrier_seq;
2959     bool enable = netdev_get_carrier(ofport->up.netdev);
2960     bool cfm_enable = false;
2961     bool bfd_enable = false;
2962
2963     ofport->carrier_seq = carrier_seq;
2964
2965     port_run_fast(ofport);
2966
2967     if (ofport->cfm) {
2968         int cfm_opup = cfm_get_opup(ofport->cfm);
2969
2970         cfm_run(ofport->cfm);
2971         cfm_enable = !cfm_get_fault(ofport->cfm);
2972
2973         if (cfm_opup >= 0) {
2974             cfm_enable = cfm_enable && cfm_opup;
2975         }
2976     }
2977
2978     if (ofport->bfd) {
2979         bfd_run(ofport->bfd);
2980         bfd_enable = bfd_forwarding(ofport->bfd);
2981     }
2982
2983     if (ofport->bfd || ofport->cfm) {
2984         enable = enable && (cfm_enable || bfd_enable);
2985     }
2986
2987     if (ofport->bundle) {
2988         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2989         if (carrier_changed) {
2990             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2991         }
2992     }
2993
2994     if (ofport->may_enable != enable) {
2995         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2996         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2997     }
2998
2999     ofport->may_enable = enable;
3000 }
3001
3002 static void
3003 port_wait(struct ofport_dpif *ofport)
3004 {
3005     if (ofport->cfm) {
3006         cfm_wait(ofport->cfm);
3007     }
3008
3009     if (ofport->bfd) {
3010         bfd_wait(ofport->bfd);
3011     }
3012 }
3013
3014 static int
3015 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3016                    struct ofproto_port *ofproto_port)
3017 {
3018     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3019     struct dpif_port dpif_port;
3020     int error;
3021
3022     if (sset_contains(&ofproto->ghost_ports, devname)) {
3023         const char *type = netdev_get_type_from_name(devname);
3024
3025         /* We may be called before ofproto->up.port_by_name is populated with
3026          * the appropriate ofport.  For this reason, we must get the name and
3027          * type from the netdev layer directly. */
3028         if (type) {
3029             const struct ofport *ofport;
3030
3031             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3032             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3033             ofproto_port->name = xstrdup(devname);
3034             ofproto_port->type = xstrdup(type);
3035             return 0;
3036         }
3037         return ENODEV;
3038     }
3039
3040     if (!sset_contains(&ofproto->ports, devname)) {
3041         return ENODEV;
3042     }
3043     error = dpif_port_query_by_name(ofproto->backer->dpif,
3044                                     devname, &dpif_port);
3045     if (!error) {
3046         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3047     }
3048     return error;
3049 }
3050
3051 static int
3052 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3053 {
3054     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3055     const char *devname = netdev_get_name(netdev);
3056     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3057     const char *dp_port_name;
3058
3059     if (netdev_vport_is_patch(netdev)) {
3060         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3061         return 0;
3062     }
3063
3064     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3065     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3066         odp_port_t port_no = ODPP_NONE;
3067         int error;
3068
3069         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3070         if (error) {
3071             return error;
3072         }
3073         if (netdev_get_tunnel_config(netdev)) {
3074             simap_put(&ofproto->backer->tnl_backers,
3075                       dp_port_name, odp_to_u32(port_no));
3076         }
3077     }
3078
3079     if (netdev_get_tunnel_config(netdev)) {
3080         sset_add(&ofproto->ghost_ports, devname);
3081     } else {
3082         sset_add(&ofproto->ports, devname);
3083     }
3084     return 0;
3085 }
3086
3087 static int
3088 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3089 {
3090     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3091     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3092     int error = 0;
3093
3094     if (!ofport) {
3095         return 0;
3096     }
3097
3098     sset_find_and_delete(&ofproto->ghost_ports,
3099                          netdev_get_name(ofport->up.netdev));
3100     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3101     if (!ofport->is_tunnel) {
3102         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3103         if (!error) {
3104             /* The caller is going to close ofport->up.netdev.  If this is a
3105              * bonded port, then the bond is using that netdev, so remove it
3106              * from the bond.  The client will need to reconfigure everything
3107              * after deleting ports, so then the slave will get re-added. */
3108             bundle_remove(&ofport->up);
3109         }
3110     }
3111     return error;
3112 }
3113
3114 static int
3115 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3116 {
3117     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3118     int error;
3119
3120     push_all_stats();
3121
3122     error = netdev_get_stats(ofport->up.netdev, stats);
3123
3124     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3125         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3126
3127         /* ofproto->stats.tx_packets represents packets that we created
3128          * internally and sent to some port (e.g. packets sent with
3129          * send_packet()).  Account for them as if they had come from
3130          * OFPP_LOCAL and got forwarded. */
3131
3132         if (stats->rx_packets != UINT64_MAX) {
3133             stats->rx_packets += ofproto->stats.tx_packets;
3134         }
3135
3136         if (stats->rx_bytes != UINT64_MAX) {
3137             stats->rx_bytes += ofproto->stats.tx_bytes;
3138         }
3139
3140         /* ofproto->stats.rx_packets represents packets that were received on
3141          * some port and we processed internally and dropped (e.g. STP).
3142          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3143
3144         if (stats->tx_packets != UINT64_MAX) {
3145             stats->tx_packets += ofproto->stats.rx_packets;
3146         }
3147
3148         if (stats->tx_bytes != UINT64_MAX) {
3149             stats->tx_bytes += ofproto->stats.rx_bytes;
3150         }
3151     }
3152
3153     return error;
3154 }
3155
3156 struct port_dump_state {
3157     uint32_t bucket;
3158     uint32_t offset;
3159     bool ghost;
3160
3161     struct ofproto_port port;
3162     bool has_port;
3163 };
3164
3165 static int
3166 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3167 {
3168     *statep = xzalloc(sizeof(struct port_dump_state));
3169     return 0;
3170 }
3171
3172 static int
3173 port_dump_next(const struct ofproto *ofproto_, void *state_,
3174                struct ofproto_port *port)
3175 {
3176     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3177     struct port_dump_state *state = state_;
3178     const struct sset *sset;
3179     struct sset_node *node;
3180
3181     if (state->has_port) {
3182         ofproto_port_destroy(&state->port);
3183         state->has_port = false;
3184     }
3185     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3186     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3187         int error;
3188
3189         error = port_query_by_name(ofproto_, node->name, &state->port);
3190         if (!error) {
3191             *port = state->port;
3192             state->has_port = true;
3193             return 0;
3194         } else if (error != ENODEV) {
3195             return error;
3196         }
3197     }
3198
3199     if (!state->ghost) {
3200         state->ghost = true;
3201         state->bucket = 0;
3202         state->offset = 0;
3203         return port_dump_next(ofproto_, state_, port);
3204     }
3205
3206     return EOF;
3207 }
3208
3209 static int
3210 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3211 {
3212     struct port_dump_state *state = state_;
3213
3214     if (state->has_port) {
3215         ofproto_port_destroy(&state->port);
3216     }
3217     free(state);
3218     return 0;
3219 }
3220
3221 static int
3222 port_poll(const struct ofproto *ofproto_, char **devnamep)
3223 {
3224     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3225
3226     if (ofproto->port_poll_errno) {
3227         int error = ofproto->port_poll_errno;
3228         ofproto->port_poll_errno = 0;
3229         return error;
3230     }
3231
3232     if (sset_is_empty(&ofproto->port_poll_set)) {
3233         return EAGAIN;
3234     }
3235
3236     *devnamep = sset_pop(&ofproto->port_poll_set);
3237     return 0;
3238 }
3239
3240 static void
3241 port_poll_wait(const struct ofproto *ofproto_)
3242 {
3243     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3244     dpif_port_poll_wait(ofproto->backer->dpif);
3245 }
3246
3247 static int
3248 port_is_lacp_current(const struct ofport *ofport_)
3249 {
3250     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3251     return (ofport->bundle && ofport->bundle->lacp
3252             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3253             : -1);
3254 }
3255 \f
3256 /* Upcall handling. */
3257
3258 struct flow_miss_op {
3259     struct dpif_op dpif_op;
3260
3261     uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3262     struct xlate_out xout;
3263     bool xout_garbage;           /* 'xout' needs to be uninitialized? */
3264
3265     struct ofpbuf mask;          /* Flow mask for "put" ops. */
3266     struct odputil_keybuf maskbuf;
3267
3268     /* If this is a "put" op, then a pointer to the subfacet that should
3269      * be marked as uninstalled if the operation fails. */
3270     struct subfacet *subfacet;
3271 };
3272
3273 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3274  * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3275  * and (usually) installing a datapath flow.  The answer is usually "yes" (a
3276  * return value of true).  However, for short flows the cost of bookkeeping is
3277  * much higher than the benefits, so when the datapath holds a large number of
3278  * flows we impose some heuristics to decide which flows are likely to be worth
3279  * tracking. */
3280 static bool
3281 flow_miss_should_make_facet(struct flow_miss *miss)
3282 {
3283     struct dpif_backer *backer = miss->ofproto->backer;
3284     uint32_t hash;
3285
3286     switch (flow_miss_model) {
3287     case OFPROTO_HANDLE_MISS_AUTO:
3288         break;
3289     case OFPROTO_HANDLE_MISS_WITH_FACETS:
3290         return true;
3291     case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3292         return false;
3293     }
3294
3295     if (!backer->governor) {
3296         size_t n_subfacets;
3297
3298         n_subfacets = hmap_count(&backer->subfacets);
3299         if (n_subfacets * 2 <= flow_eviction_threshold) {
3300             return true;
3301         }
3302
3303         backer->governor = governor_create();
3304     }
3305
3306     hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3307     return governor_should_install_flow(backer->governor, hash,
3308                                         list_size(&miss->packets));
3309 }
3310
3311 /* Handles 'miss', which matches 'facet'.  May add any required datapath
3312  * operations to 'ops', incrementing '*n_ops' for each new op.
3313  *
3314  * All of the packets in 'miss' are considered to have arrived at time
3315  * 'miss->stats.used'.  This is really important only for new facets: if we
3316  * just called time_msec() here, then the new subfacet or its packets could
3317  * look (occasionally) as though it was used some time after the facet was
3318  * used.  That can make a one-packet flow look like it has a nonzero duration,
3319  * which looks odd in e.g. NetFlow statistics. */
3320 static void
3321 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3322                             struct flow_miss_op *ops, size_t *n_ops)
3323 {
3324     enum subfacet_path want_path;
3325     struct subfacet *subfacet;
3326
3327     facet->packet_count += miss->stats.n_packets;
3328     facet->prev_packet_count += miss->stats.n_packets;
3329     facet->byte_count += miss->stats.n_bytes;
3330     facet->prev_byte_count += miss->stats.n_bytes;
3331
3332     subfacet = subfacet_create(facet, miss);
3333     want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3334
3335     /* Don't install the flow if it's the result of the "userspace"
3336      * action for an already installed facet.  This can occur when a
3337      * datapath flow with wildcards has a "userspace" action and flows
3338      * sent to userspace result in a different subfacet, which will then
3339      * be rejected as overlapping by the datapath. */
3340     if (miss->upcall_type == DPIF_UC_ACTION
3341         && !list_is_empty(&facet->subfacets)) {
3342         return;
3343     }
3344
3345     subfacet = subfacet_create(facet, miss);
3346     if (subfacet->path != want_path) {
3347         struct flow_miss_op *op = &ops[(*n_ops)++];
3348         struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3349
3350         subfacet->path = want_path;
3351
3352         ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3353         if (enable_megaflows) {
3354             odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3355                                    &miss->flow, UINT32_MAX);
3356         }
3357
3358         op->xout_garbage = false;
3359         op->dpif_op.type = DPIF_OP_FLOW_PUT;
3360         op->subfacet = subfacet;
3361         put->flags = DPIF_FP_CREATE;
3362         put->key = miss->key;
3363         put->key_len = miss->key_len;
3364         put->mask = op->mask.data;
3365         put->mask_len = op->mask.size;
3366
3367         if (want_path == SF_FAST_PATH) {
3368             put->actions = facet->xout.odp_actions.data;
3369             put->actions_len = facet->xout.odp_actions.size;
3370         } else {
3371             compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3372                               op->slow_stub, sizeof op->slow_stub,
3373                               &put->actions, &put->actions_len);
3374         }
3375         put->stats = NULL;
3376     }
3377 }
3378
3379 /* Handles flow miss 'miss'.  May add any required datapath operations
3380  * to 'ops', incrementing '*n_ops' for each new op. */
3381 static void
3382 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3383                  size_t *n_ops)
3384 {
3385     struct facet *facet;
3386
3387     miss->ofproto->n_missed += list_size(&miss->packets);
3388
3389     facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3390     if (!facet) {
3391         /* There does not exist a bijection between 'struct flow' and datapath
3392          * flow keys with fitness ODP_FIT_TO_LITTLE.  This breaks a fundamental
3393          * assumption used throughout the facet and subfacet handling code.
3394          * Since we have to handle these misses in userspace anyway, we simply
3395          * skip facet creation, avoiding the problem altogether. */
3396         if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3397             || !flow_miss_should_make_facet(miss)) {
3398             return;
3399         }
3400
3401         facet = facet_create(miss);
3402     }
3403     handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3404 }
3405
3406 static struct drop_key *
3407 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3408                 size_t key_len)
3409 {
3410     struct drop_key *drop_key;
3411
3412     HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3413                              &backer->drop_keys) {
3414         if (drop_key->key_len == key_len
3415             && !memcmp(drop_key->key, key, key_len)) {
3416             return drop_key;
3417         }
3418     }
3419     return NULL;
3420 }
3421
3422 static void
3423 drop_key_clear(struct dpif_backer *backer)
3424 {
3425     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3426     struct drop_key *drop_key, *next;
3427
3428     HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3429         int error;
3430
3431         error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3432                               NULL);
3433         if (error && !VLOG_DROP_WARN(&rl)) {
3434             struct ds ds = DS_EMPTY_INITIALIZER;
3435             odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3436             VLOG_WARN("Failed to delete drop key (%s) (%s)",
3437                       ovs_strerror(error), ds_cstr(&ds));
3438             ds_destroy(&ds);
3439         }
3440
3441         hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3442         drop_key_destroy(drop_key);
3443     }
3444
3445     udpif_drop_key_clear(backer->udpif);
3446 }
3447
3448 static void
3449 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3450 {
3451     struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3452     struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3453     struct flow_miss *miss;
3454     size_t n_ops, i;
3455
3456     /* Process each element in the to-do list, constructing the set of
3457      * operations to batch. */
3458     n_ops = 0;
3459     HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3460         handle_flow_miss(miss, flow_miss_ops, &n_ops);
3461     }
3462     ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3463
3464     /* Execute batch. */
3465     for (i = 0; i < n_ops; i++) {
3466         dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3467     }
3468     dpif_operate(backer->dpif, dpif_ops, n_ops);
3469
3470     for (i = 0; i < n_ops; i++) {
3471         if (dpif_ops[i]->error != 0
3472             && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3473             && flow_miss_ops[i].subfacet) {
3474             struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3475
3476             COVERAGE_INC(subfacet_install_fail);
3477
3478             /* Zero-out subfacet counters when installation failed, but
3479              * datapath reported hits.  This should not happen and
3480              * indicates a bug, since if the datapath flow exists, we
3481              * should not be attempting to create a new subfacet.  A
3482              * buggy datapath could trigger this, so just zero out the
3483              * counters and log an error. */
3484             if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3485                 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3486                             "datapath reported hits");
3487                 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3488             }
3489
3490             subfacet->path = SF_NOT_INSTALLED;
3491         }
3492     }
3493 }
3494
3495 static void
3496 handle_sflow_upcall(struct dpif_backer *backer,
3497                     const struct dpif_upcall *upcall)
3498 {
3499     struct ofproto_dpif *ofproto;
3500     union user_action_cookie cookie;
3501     struct flow flow;
3502     odp_port_t odp_in_port;
3503
3504     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3505                       &flow, NULL, &ofproto, &odp_in_port)
3506         || !ofproto->sflow) {
3507         return;
3508     }
3509
3510     memset(&cookie, 0, sizeof cookie);
3511     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3512     dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3513                         odp_in_port, &cookie);
3514 }
3515
3516 static void
3517 handle_flow_sample_upcall(struct dpif_backer *backer,
3518                           const struct dpif_upcall *upcall)
3519 {
3520     struct ofproto_dpif *ofproto;
3521     union user_action_cookie cookie;
3522     struct flow flow;
3523
3524     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3525                       &flow, NULL, &ofproto, NULL)
3526         || !ofproto->ipfix) {
3527         return;
3528     }
3529
3530     memset(&cookie, 0, sizeof cookie);
3531     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3532
3533     /* The flow reflects exactly the contents of the packet.  Sample
3534      * the packet using it. */
3535     dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3536                            cookie.flow_sample.collector_set_id,
3537                            cookie.flow_sample.probability,
3538                            cookie.flow_sample.obs_domain_id,
3539                            cookie.flow_sample.obs_point_id);
3540 }
3541
3542 static void
3543 handle_ipfix_upcall(struct dpif_backer *backer,
3544                     const struct dpif_upcall *upcall)
3545 {
3546     struct ofproto_dpif *ofproto;
3547     struct flow flow;
3548
3549     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3550                       &flow, NULL, &ofproto, NULL)
3551         || !ofproto->ipfix) {
3552         return;
3553     }
3554
3555     /* The flow reflects exactly the contents of the packet.  Sample
3556      * the packet using it. */
3557     dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3558 }
3559
3560 static void
3561 handle_upcalls(struct dpif_backer *backer)
3562 {
3563     struct flow_miss_batch *fmb;
3564     int n_processed;
3565
3566     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3567         struct upcall *upcall = upcall_next(backer->udpif);
3568
3569         if (!upcall) {
3570             break;
3571         }
3572
3573         switch (upcall->type) {
3574         case SFLOW_UPCALL:
3575             handle_sflow_upcall(backer, &upcall->dpif_upcall);
3576             break;
3577
3578         case FLOW_SAMPLE_UPCALL:
3579             handle_flow_sample_upcall(backer, &upcall->dpif_upcall);
3580             break;
3581
3582         case IPFIX_UPCALL:
3583             handle_ipfix_upcall(backer, &upcall->dpif_upcall);
3584             break;
3585
3586         case BAD_UPCALL:
3587             break;
3588
3589         case MISS_UPCALL:
3590             NOT_REACHED();
3591         }
3592
3593         upcall_destroy(upcall);
3594     }
3595
3596     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3597         struct drop_key *drop_key = drop_key_next(backer->udpif);
3598         if (!drop_key) {
3599             break;
3600         }
3601
3602         if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3603             hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3604                         hash_bytes(drop_key->key, drop_key->key_len, 0));
3605             dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3606                           drop_key->key, drop_key->key_len,
3607                           NULL, 0, NULL, 0, NULL);
3608         } else {
3609             drop_key_destroy(drop_key);
3610         }
3611     }
3612
3613     fmb = flow_miss_batch_next(backer->udpif);
3614     if (fmb) {
3615         handle_flow_misses(backer, fmb);
3616         flow_miss_batch_destroy(fmb);
3617     }
3618 }
3619 \f
3620 /* Flow expiration. */
3621
3622 static int subfacet_max_idle(const struct dpif_backer *);
3623 static void update_stats(struct dpif_backer *);
3624 static void rule_expire(struct rule_dpif *);
3625 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3626
3627 /* This function is called periodically by run().  Its job is to collect
3628  * updates for the flows that have been installed into the datapath, most
3629  * importantly when they last were used, and then use that information to
3630  * expire flows that have not been used recently.
3631  *
3632  * Returns the number of milliseconds after which it should be called again. */
3633 static int
3634 expire(struct dpif_backer *backer)
3635 {
3636     struct ofproto_dpif *ofproto;
3637     size_t n_subfacets;
3638     int max_idle;
3639
3640     /* Periodically clear out the drop keys in an effort to keep them
3641      * relatively few. */
3642     drop_key_clear(backer);
3643
3644     /* Update stats for each flow in the backer. */
3645     update_stats(backer);
3646
3647     n_subfacets = hmap_count(&backer->subfacets);
3648     if (n_subfacets) {
3649         struct subfacet *subfacet;
3650         long long int total, now;
3651
3652         total = 0;
3653         now = time_msec();
3654         HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3655             total += now - subfacet->created;
3656         }
3657         backer->avg_subfacet_life += total / n_subfacets;
3658     }
3659     backer->avg_subfacet_life /= 2;
3660
3661     backer->avg_n_subfacet += n_subfacets;
3662     backer->avg_n_subfacet /= 2;
3663
3664     backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3665
3666     max_idle = subfacet_max_idle(backer);
3667     expire_subfacets(backer, max_idle);
3668
3669     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3670         struct rule *rule, *next_rule;
3671
3672         if (ofproto->backer != backer) {
3673             continue;
3674         }
3675
3676         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3677          * has passed. */
3678         ovs_mutex_lock(&ofproto->up.expirable_mutex);
3679         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3680                             &ofproto->up.expirable) {
3681             rule_expire(rule_dpif_cast(rule));
3682         }
3683         ovs_mutex_unlock(&ofproto->up.expirable_mutex);
3684
3685         /* All outstanding data in existing flows has been accounted, so it's a
3686          * good time to do bond rebalancing. */
3687         if (ofproto->has_bonded_bundles) {
3688             struct ofbundle *bundle;
3689
3690             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3691                 if (bundle->bond) {
3692                     bond_rebalance(bundle->bond);
3693                 }
3694             }
3695         }
3696     }
3697
3698     return MIN(max_idle, 1000);
3699 }
3700
3701 /* Updates flow table statistics given that the datapath just reported 'stats'
3702  * as 'subfacet''s statistics. */
3703 static void
3704 update_subfacet_stats(struct subfacet *subfacet,
3705                       const struct dpif_flow_stats *stats)
3706 {
3707     struct facet *facet = subfacet->facet;
3708     struct dpif_flow_stats diff;
3709
3710     diff.tcp_flags = stats->tcp_flags;
3711     diff.used = stats->used;
3712
3713     if (stats->n_packets >= subfacet->dp_packet_count) {
3714         diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3715     } else {
3716         VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3717         diff.n_packets = 0;
3718     }
3719
3720     if (stats->n_bytes >= subfacet->dp_byte_count) {
3721         diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3722     } else {
3723         VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3724         diff.n_bytes = 0;
3725     }
3726
3727     facet->ofproto->n_hit += diff.n_packets;
3728     subfacet->dp_packet_count = stats->n_packets;
3729     subfacet->dp_byte_count = stats->n_bytes;
3730     subfacet_update_stats(subfacet, &diff);
3731
3732     if (facet->accounted_bytes < facet->byte_count) {
3733         facet_learn(facet);
3734         facet_account(facet);
3735         facet->accounted_bytes = facet->byte_count;
3736     }
3737 }
3738
3739 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3740  * about, or a flow that shouldn't be installed but was anyway.  Delete it. */
3741 static void
3742 delete_unexpected_flow(struct dpif_backer *backer,
3743                        const struct nlattr *key, size_t key_len)
3744 {
3745     if (!VLOG_DROP_WARN(&rl)) {
3746         struct ds s;
3747
3748         ds_init(&s);
3749         odp_flow_key_format(key, key_len, &s);
3750         VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3751         ds_destroy(&s);
3752     }
3753
3754     COVERAGE_INC(facet_unexpected);
3755     dpif_flow_del(backer->dpif, key, key_len, NULL);
3756 }
3757
3758 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3759  *
3760  * This function also pushes statistics updates to rules which each facet
3761  * resubmits into.  Generally these statistics will be accurate.  However, if a
3762  * facet changes the rule it resubmits into at some time in between
3763  * update_stats() runs, it is possible that statistics accrued to the
3764  * old rule will be incorrectly attributed to the new rule.  This could be
3765  * avoided by calling update_stats() whenever rules are created or
3766  * deleted.  However, the performance impact of making so many calls to the
3767  * datapath do not justify the benefit of having perfectly accurate statistics.
3768  *
3769  * In addition, this function maintains per ofproto flow hit counts. The patch
3770  * port is not treated specially. e.g. A packet ingress from br0 patched into
3771  * br1 will increase the hit count of br0 by 1, however, does not affect
3772  * the hit or miss counts of br1.
3773  */
3774 static void
3775 update_stats(struct dpif_backer *backer)
3776 {
3777     const struct dpif_flow_stats *stats;
3778     struct dpif_flow_dump dump;
3779     const struct nlattr *key, *mask;
3780     size_t key_len, mask_len;
3781
3782     dpif_flow_dump_start(&dump, backer->dpif);
3783     while (dpif_flow_dump_next(&dump, &key, &key_len,
3784                                &mask, &mask_len, NULL, NULL, &stats)) {
3785         struct subfacet *subfacet;
3786         uint32_t key_hash;
3787
3788         key_hash = odp_flow_key_hash(key, key_len);
3789         subfacet = subfacet_find(backer, key, key_len, key_hash);
3790         switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3791         case SF_FAST_PATH:
3792             update_subfacet_stats(subfacet, stats);
3793             break;
3794
3795         case SF_SLOW_PATH:
3796             /* Stats are updated per-packet. */
3797             break;
3798
3799         case SF_NOT_INSTALLED:
3800         default:
3801             delete_unexpected_flow(backer, key, key_len);
3802             break;
3803         }
3804         run_fast_rl();
3805     }
3806     dpif_flow_dump_done(&dump);
3807
3808     update_moving_averages(backer);
3809 }
3810
3811 /* Calculates and returns the number of milliseconds of idle time after which
3812  * subfacets should expire from the datapath.  When a subfacet expires, we fold
3813  * its statistics into its facet, and when a facet's last subfacet expires, we
3814  * fold its statistic into its rule. */
3815 static int
3816 subfacet_max_idle(const struct dpif_backer *backer)
3817 {
3818     /*
3819      * Idle time histogram.
3820      *
3821      * Most of the time a switch has a relatively small number of subfacets.
3822      * When this is the case we might as well keep statistics for all of them
3823      * in userspace and to cache them in the kernel datapath for performance as
3824      * well.
3825      *
3826      * As the number of subfacets increases, the memory required to maintain
3827      * statistics about them in userspace and in the kernel becomes
3828      * significant.  However, with a large number of subfacets it is likely
3829      * that only a few of them are "heavy hitters" that consume a large amount
3830      * of bandwidth.  At this point, only heavy hitters are worth caching in
3831      * the kernel and maintaining in userspaces; other subfacets we can
3832      * discard.
3833      *
3834      * The technique used to compute the idle time is to build a histogram with
3835      * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each.  Each subfacet
3836      * that is installed in the kernel gets dropped in the appropriate bucket.
3837      * After the histogram has been built, we compute the cutoff so that only
3838      * the most-recently-used 1% of subfacets (but at least
3839      * flow_eviction_threshold flows) are kept cached.  At least
3840      * the most-recently-used bucket of subfacets is kept, so actually an
3841      * arbitrary number of subfacets can be kept in any given expiration run
3842      * (though the next run will delete most of those unless they receive
3843      * additional data).
3844      *
3845      * This requires a second pass through the subfacets, in addition to the
3846      * pass made by update_stats(), because the former function never looks at
3847      * uninstallable subfacets.
3848      */
3849     enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3850     enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3851     int buckets[N_BUCKETS] = { 0 };
3852     int total, subtotal, bucket;
3853     struct subfacet *subfacet;
3854     long long int now;
3855     int i;
3856
3857     total = hmap_count(&backer->subfacets);
3858     if (total <= flow_eviction_threshold) {
3859         return N_BUCKETS * BUCKET_WIDTH;
3860     }
3861
3862     /* Build histogram. */
3863     now = time_msec();
3864     HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3865         long long int idle = now - subfacet->used;
3866         int bucket = (idle <= 0 ? 0
3867                       : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3868                       : (unsigned int) idle / BUCKET_WIDTH);
3869         buckets[bucket]++;
3870     }
3871
3872     /* Find the first bucket whose flows should be expired. */
3873     subtotal = bucket = 0;
3874     do {
3875         subtotal += buckets[bucket++];
3876     } while (bucket < N_BUCKETS &&
3877              subtotal < MAX(flow_eviction_threshold, total / 100));
3878
3879     if (VLOG_IS_DBG_ENABLED()) {
3880         struct ds s;
3881
3882         ds_init(&s);
3883         ds_put_cstr(&s, "keep");
3884         for (i = 0; i < N_BUCKETS; i++) {
3885             if (i == bucket) {
3886                 ds_put_cstr(&s, ", drop");
3887             }
3888             if (buckets[i]) {
3889                 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3890             }
3891         }
3892         VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3893         ds_destroy(&s);
3894     }
3895
3896     return bucket * BUCKET_WIDTH;
3897 }
3898
3899 static void
3900 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3901 {
3902     /* Cutoff time for most flows. */
3903     long long int normal_cutoff = time_msec() - dp_max_idle;
3904
3905     /* We really want to keep flows for special protocols around, so use a more
3906      * conservative cutoff. */
3907     long long int special_cutoff = time_msec() - 10000;
3908
3909     struct subfacet *subfacet, *next_subfacet;
3910     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3911     int n_batch;
3912
3913     n_batch = 0;
3914     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3915                         &backer->subfacets) {
3916         long long int cutoff;
3917
3918         cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3919                                                 | SLOW_STP)
3920                   ? special_cutoff
3921                   : normal_cutoff);
3922         if (subfacet->used < cutoff) {
3923             if (subfacet->path != SF_NOT_INSTALLED) {
3924                 batch[n_batch++] = subfacet;
3925                 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3926                     subfacet_destroy_batch(backer, batch, n_batch);
3927                     n_batch = 0;
3928                 }
3929             } else {
3930                 subfacet_destroy(subfacet);
3931             }
3932         }
3933     }
3934
3935     if (n_batch > 0) {
3936         subfacet_destroy_batch(backer, batch, n_batch);
3937     }
3938 }
3939
3940 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3941  * then delete it entirely. */
3942 static void
3943 rule_expire(struct rule_dpif *rule)
3944 {
3945     uint16_t idle_timeout, hard_timeout;
3946     long long int now;
3947     uint8_t reason;
3948
3949     if (rule->up.pending) {
3950         /* We'll have to expire it later. */
3951         return;
3952     }
3953
3954     ovs_mutex_lock(&rule->up.timeout_mutex);
3955     hard_timeout = rule->up.hard_timeout;
3956     idle_timeout = rule->up.idle_timeout;
3957     ovs_mutex_unlock(&rule->up.timeout_mutex);
3958
3959     /* Has 'rule' expired? */
3960     now = time_msec();
3961     if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3962         reason = OFPRR_HARD_TIMEOUT;
3963     } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3964         reason = OFPRR_IDLE_TIMEOUT;
3965     } else {
3966         return;
3967     }
3968
3969     if (!ovs_rwlock_trywrlock(&rule->up.evict)) {
3970         COVERAGE_INC(ofproto_dpif_expired);
3971
3972         /* Get rid of the rule. */
3973         ofproto_rule_expire(&rule->up, reason);
3974     }
3975 }
3976 \f
3977 /* Facets. */
3978
3979 /* Creates and returns a new facet based on 'miss'.
3980  *
3981  * The caller must already have determined that no facet with an identical
3982  * 'miss->flow' exists in 'miss->ofproto'.
3983  *
3984  * 'rule' and 'xout' must have been created based on 'miss'.
3985  *
3986  * 'facet'' statistics are initialized based on 'stats'.
3987  *
3988  * The facet will initially have no subfacets.  The caller should create (at
3989  * least) one subfacet with subfacet_create(). */
3990 static struct facet *
3991 facet_create(const struct flow_miss *miss)
3992 {
3993     struct ofproto_dpif *ofproto = miss->ofproto;
3994     struct facet *facet;
3995     struct match match;
3996
3997     facet = xzalloc(sizeof *facet);
3998     facet->ofproto = miss->ofproto;
3999     facet->used = miss->stats.used;
4000     facet->flow = miss->flow;
4001     facet->learn_rl = time_msec() + 500;
4002
4003     list_init(&facet->subfacets);
4004     netflow_flow_init(&facet->nf_flow);
4005     netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4006
4007     xlate_out_copy(&facet->xout, &miss->xout);
4008
4009     match_init(&match, &facet->flow, &facet->xout.wc);
4010     cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4011     ovs_rwlock_wrlock(&ofproto->facets.rwlock);
4012     classifier_insert(&ofproto->facets, &facet->cr);
4013     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4014
4015     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4016     return facet;
4017 }
4018
4019 static void
4020 facet_free(struct facet *facet)
4021 {
4022     if (facet) {
4023         xlate_out_uninit(&facet->xout);
4024         free(facet);
4025     }
4026 }
4027
4028 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4029  * 'packet', which arrived on 'in_port'. */
4030 static bool
4031 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4032                     const struct nlattr *odp_actions, size_t actions_len,
4033                     struct ofpbuf *packet)
4034 {
4035     struct odputil_keybuf keybuf;
4036     struct ofpbuf key;
4037     int error;
4038
4039     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4040     odp_flow_key_from_flow(&key, flow,
4041                            ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4042
4043     error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4044                          odp_actions, actions_len, packet);
4045     return !error;
4046 }
4047
4048 /* Remove 'facet' from its ofproto and free up the associated memory:
4049  *
4050  *   - If 'facet' was installed in the datapath, uninstalls it and updates its
4051  *     rule's statistics, via subfacet_uninstall().
4052  *
4053  *   - Removes 'facet' from its rule and from ofproto->facets.
4054  */
4055 static void
4056 facet_remove(struct facet *facet)
4057 {
4058     struct subfacet *subfacet, *next_subfacet;
4059
4060     ovs_assert(!list_is_empty(&facet->subfacets));
4061
4062     /* First uninstall all of the subfacets to get final statistics. */
4063     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4064         subfacet_uninstall(subfacet);
4065     }
4066
4067     /* Flush the final stats to the rule.
4068      *
4069      * This might require us to have at least one subfacet around so that we
4070      * can use its actions for accounting in facet_account(), which is why we
4071      * have uninstalled but not yet destroyed the subfacets. */
4072     facet_flush_stats(facet);
4073
4074     /* Now we're really all done so destroy everything. */
4075     LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4076                         &facet->subfacets) {
4077         subfacet_destroy__(subfacet);
4078     }
4079     ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4080     classifier_remove(&facet->ofproto->facets, &facet->cr);
4081     ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4082     cls_rule_destroy(&facet->cr);
4083     facet_free(facet);
4084 }
4085
4086 /* Feed information from 'facet' back into the learning table to keep it in
4087  * sync with what is actually flowing through the datapath. */
4088 static void
4089 facet_learn(struct facet *facet)
4090 {
4091     long long int now = time_msec();
4092
4093     if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4094         return;
4095     }
4096
4097     facet->learn_rl = now + 500;
4098
4099     if (!facet->xout.has_learn
4100         && !facet->xout.has_normal
4101         && (!facet->xout.has_fin_timeout
4102             || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4103         return;
4104     }
4105
4106     facet_push_stats(facet, true);
4107 }
4108
4109 static void
4110 facet_account(struct facet *facet)
4111 {
4112     const struct nlattr *a;
4113     unsigned int left;
4114     ovs_be16 vlan_tci;
4115     uint64_t n_bytes;
4116
4117     if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4118         return;
4119     }
4120     n_bytes = facet->byte_count - facet->accounted_bytes;
4121
4122     /* This loop feeds byte counters to bond_account() for rebalancing to use
4123      * as a basis.  We also need to track the actual VLAN on which the packet
4124      * is going to be sent to ensure that it matches the one passed to
4125      * bond_choose_output_slave().  (Otherwise, we will account to the wrong
4126      * hash bucket.)
4127      *
4128      * We use the actions from an arbitrary subfacet because they should all
4129      * be equally valid for our purpose. */
4130     vlan_tci = facet->flow.vlan_tci;
4131     NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4132                              facet->xout.odp_actions.size) {
4133         const struct ovs_action_push_vlan *vlan;
4134         struct ofport_dpif *port;
4135
4136         switch (nl_attr_type(a)) {
4137         case OVS_ACTION_ATTR_OUTPUT:
4138             port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4139             if (port && port->bundle && port->bundle->bond) {
4140                 bond_account(port->bundle->bond, &facet->flow,
4141                              vlan_tci_to_vid(vlan_tci), n_bytes);
4142             }
4143             break;
4144
4145         case OVS_ACTION_ATTR_POP_VLAN:
4146             vlan_tci = htons(0);
4147             break;
4148
4149         case OVS_ACTION_ATTR_PUSH_VLAN:
4150             vlan = nl_attr_get(a);
4151             vlan_tci = vlan->vlan_tci;
4152             break;
4153         }
4154     }
4155 }
4156
4157 /* Returns true if the only action for 'facet' is to send to the controller.
4158  * (We don't report NetFlow expiration messages for such facets because they
4159  * are just part of the control logic for the network, not real traffic). */
4160 static bool
4161 facet_is_controller_flow(struct facet *facet)
4162 {
4163     if (facet) {
4164         struct ofproto_dpif *ofproto = facet->ofproto;
4165         const struct ofpact *ofpacts;
4166         struct rule_dpif *rule;
4167         size_t ofpacts_len;
4168         bool is_controller;
4169
4170         rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4171         ofpacts_len = rule->up.ofpacts_len;
4172         ofpacts = rule->up.ofpacts;
4173         is_controller = ofpacts_len > 0
4174             && ofpacts->type == OFPACT_CONTROLLER
4175             && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4176         rule_release(rule);
4177         return is_controller;
4178     }
4179     return false;
4180 }
4181
4182 /* Folds all of 'facet''s statistics into its rule.  Also updates the
4183  * accounting ofhook and emits a NetFlow expiration if appropriate.  All of
4184  * 'facet''s statistics in the datapath should have been zeroed and folded into
4185  * its packet and byte counts before this function is called. */
4186 static void
4187 facet_flush_stats(struct facet *facet)
4188 {
4189     struct ofproto_dpif *ofproto = facet->ofproto;
4190     struct subfacet *subfacet;
4191
4192     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4193         ovs_assert(!subfacet->dp_byte_count);
4194         ovs_assert(!subfacet->dp_packet_count);
4195     }
4196
4197     facet_push_stats(facet, false);
4198     if (facet->accounted_bytes < facet->byte_count) {
4199         facet_account(facet);
4200         facet->accounted_bytes = facet->byte_count;
4201     }
4202
4203     if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4204         struct ofexpired expired;
4205         expired.flow = facet->flow;
4206         expired.packet_count = facet->packet_count;
4207         expired.byte_count = facet->byte_count;
4208         expired.used = facet->used;
4209         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4210     }
4211
4212     /* Reset counters to prevent double counting if 'facet' ever gets
4213      * reinstalled. */
4214     facet_reset_counters(facet);
4215
4216     netflow_flow_clear(&facet->nf_flow);
4217     facet->tcp_flags = 0;
4218 }
4219
4220 /* Searches 'ofproto''s table of facets for one which would be responsible for
4221  * 'flow'.  Returns it if found, otherwise a null pointer.
4222  *
4223  * The returned facet might need revalidation; use facet_lookup_valid()
4224  * instead if that is important. */
4225 static struct facet *
4226 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4227 {
4228     struct cls_rule *cr;
4229
4230     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4231     cr = classifier_lookup(&ofproto->facets, flow, NULL);
4232     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4233     return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4234 }
4235
4236 /* Searches 'ofproto''s table of facets for one capable that covers
4237  * 'flow'.  Returns it if found, otherwise a null pointer.
4238  *
4239  * The returned facet is guaranteed to be valid. */
4240 static struct facet *
4241 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4242 {
4243     struct facet *facet;
4244
4245     facet = facet_find(ofproto, flow);
4246     if (facet
4247         && ofproto->backer->need_revalidate
4248         && !facet_revalidate(facet)) {
4249         return NULL;
4250     }
4251
4252     return facet;
4253 }
4254
4255 static bool
4256 facet_check_consistency(struct facet *facet)
4257 {
4258     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4259
4260     struct xlate_out xout;
4261     struct xlate_in xin;
4262
4263     struct rule_dpif *rule;
4264     bool ok;
4265
4266     /* Check the datapath actions for consistency. */
4267     rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4268     xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4269     xlate_actions(&xin, &xout);
4270     rule_release(rule);
4271
4272     ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4273         && facet->xout.slow == xout.slow;
4274     if (!ok && !VLOG_DROP_WARN(&rl)) {
4275         struct ds s = DS_EMPTY_INITIALIZER;
4276
4277         flow_format(&s, &facet->flow);
4278         ds_put_cstr(&s, ": inconsistency in facet");
4279
4280         if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4281             ds_put_cstr(&s, " (actions were: ");
4282             format_odp_actions(&s, facet->xout.odp_actions.data,
4283                                facet->xout.odp_actions.size);
4284             ds_put_cstr(&s, ") (correct actions: ");
4285             format_odp_actions(&s, xout.odp_actions.data,
4286                                xout.odp_actions.size);
4287             ds_put_char(&s, ')');
4288         }
4289
4290         if (facet->xout.slow != xout.slow) {
4291             ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4292         }
4293
4294         ds_destroy(&s);
4295     }
4296     xlate_out_uninit(&xout);
4297
4298     return ok;
4299 }
4300
4301 /* Re-searches the classifier for 'facet':
4302  *
4303  *   - If the rule found is different from 'facet''s current rule, moves
4304  *     'facet' to the new rule and recompiles its actions.
4305  *
4306  *   - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4307  *     where it is and recompiles its actions anyway.
4308  *
4309  *   - If any of 'facet''s subfacets correspond to a new flow according to
4310  *     xlate_receive(), 'facet' is removed.
4311  *
4312  *   Returns true if 'facet' is still valid.  False if 'facet' was removed. */
4313 static bool
4314 facet_revalidate(struct facet *facet)
4315 {
4316     struct ofproto_dpif *ofproto = facet->ofproto;
4317     struct rule_dpif *new_rule;
4318     struct subfacet *subfacet;
4319     struct flow_wildcards wc;
4320     struct xlate_out xout;
4321     struct xlate_in xin;
4322
4323     COVERAGE_INC(facet_revalidate);
4324
4325     /* Check that child subfacets still correspond to this facet.  Tunnel
4326      * configuration changes could cause a subfacet's OpenFlow in_port to
4327      * change. */
4328     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4329         struct ofproto_dpif *recv_ofproto;
4330         struct flow recv_flow;
4331         int error;
4332
4333         error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4334                               subfacet->key_len, &recv_flow, NULL,
4335                               &recv_ofproto, NULL);
4336         if (error
4337             || recv_ofproto != ofproto
4338             || facet != facet_find(ofproto, &recv_flow)) {
4339             facet_remove(facet);
4340             return false;
4341         }
4342     }
4343
4344     flow_wildcards_init_catchall(&wc);
4345     rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4346
4347     /* Calculate new datapath actions.
4348      *
4349      * We do not modify any 'facet' state yet, because we might need to, e.g.,
4350      * emit a NetFlow expiration and, if so, we need to have the old state
4351      * around to properly compose it. */
4352     xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4353     xlate_actions(&xin, &xout);
4354     flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4355
4356     /* A facet's slow path reason should only change under dramatic
4357      * circumstances.  Rather than try to update everything, it's simpler to
4358      * remove the facet and start over.
4359      *
4360      * More importantly, if a facet's wildcards change, it will be relatively
4361      * difficult to figure out if its subfacets still belong to it, and if not
4362      * which facet they may belong to.  Again, to avoid the complexity, we
4363      * simply give up instead. */
4364     if (facet->xout.slow != xout.slow
4365         || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4366         facet_remove(facet);
4367         xlate_out_uninit(&xout);
4368         rule_release(new_rule);
4369         return false;
4370     }
4371
4372     if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4373         LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4374             if (subfacet->path == SF_FAST_PATH) {
4375                 struct dpif_flow_stats stats;
4376
4377                 subfacet_install(subfacet, &xout.odp_actions, &stats);
4378                 subfacet_update_stats(subfacet, &stats);
4379             }
4380         }
4381
4382         facet_flush_stats(facet);
4383
4384         ofpbuf_clear(&facet->xout.odp_actions);
4385         ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4386                    xout.odp_actions.size);
4387     }
4388
4389     /* Update 'facet' now that we've taken care of all the old state. */
4390     facet->xout.slow = xout.slow;
4391     facet->xout.has_learn = xout.has_learn;
4392     facet->xout.has_normal = xout.has_normal;
4393     facet->xout.has_fin_timeout = xout.has_fin_timeout;
4394     facet->xout.nf_output_iface = xout.nf_output_iface;
4395     facet->xout.mirrors = xout.mirrors;
4396     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4397     facet->used = MAX(facet->used, new_rule->up.created);
4398
4399     xlate_out_uninit(&xout);
4400     rule_release(new_rule);
4401     return true;
4402 }
4403
4404 static void
4405 facet_reset_counters(struct facet *facet)
4406 {
4407     facet->packet_count = 0;
4408     facet->byte_count = 0;
4409     facet->prev_packet_count = 0;
4410     facet->prev_byte_count = 0;
4411     facet->accounted_bytes = 0;
4412 }
4413
4414 static void
4415 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4416                 struct dpif_flow_stats *stats, bool may_learn)
4417 {
4418     struct ofport_dpif *in_port;
4419     struct rule_dpif *rule;
4420     struct xlate_in xin;
4421
4422     in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4423     if (in_port && in_port->is_tunnel) {
4424         netdev_vport_inc_rx(in_port->up.netdev, stats);
4425     }
4426
4427     rule_dpif_lookup(ofproto, flow, NULL, &rule);
4428     rule_credit_stats(rule, stats);
4429     xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4430     xin.resubmit_stats = stats;
4431     xin.may_learn = may_learn;
4432     xlate_actions_for_side_effects(&xin);
4433     rule_release(rule);
4434 }
4435
4436 static void
4437 facet_push_stats(struct facet *facet, bool may_learn)
4438 {
4439     struct dpif_flow_stats stats;
4440
4441     ovs_assert(facet->packet_count >= facet->prev_packet_count);
4442     ovs_assert(facet->byte_count >= facet->prev_byte_count);
4443     ovs_assert(facet->used >= facet->prev_used);
4444
4445     stats.n_packets = facet->packet_count - facet->prev_packet_count;
4446     stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4447     stats.used = facet->used;
4448     stats.tcp_flags = facet->tcp_flags;
4449
4450     if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4451         facet->prev_packet_count = facet->packet_count;
4452         facet->prev_byte_count = facet->byte_count;
4453         facet->prev_used = facet->used;
4454
4455         netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4456                                  facet->used);
4457         netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4458         mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4459                             stats.n_packets, stats.n_bytes);
4460         flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4461     }
4462 }
4463
4464 static void
4465 push_all_stats__(bool run_fast)
4466 {
4467     static long long int rl = LLONG_MIN;
4468     struct ofproto_dpif *ofproto;
4469
4470     if (time_msec() < rl) {
4471         return;
4472     }
4473
4474     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4475         struct cls_cursor cursor;
4476         struct facet *facet;
4477
4478         ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4479         cls_cursor_init(&cursor, &ofproto->facets, NULL);
4480         CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4481             facet_push_stats(facet, false);
4482             if (run_fast) {
4483                 run_fast_rl();
4484             }
4485         }
4486         ovs_rwlock_unlock(&ofproto->facets.rwlock);
4487     }
4488
4489     rl = time_msec() + 100;
4490 }
4491
4492 static void
4493 push_all_stats(void)
4494 {
4495     push_all_stats__(true);
4496 }
4497
4498 void
4499 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4500 {
4501     ovs_mutex_lock(&rule->stats_mutex);
4502     rule->packet_count += stats->n_packets;
4503     rule->byte_count += stats->n_bytes;
4504     ofproto_rule_update_used(&rule->up, stats->used);
4505     ovs_mutex_unlock(&rule->stats_mutex);
4506 }
4507 \f
4508 /* Subfacets. */
4509
4510 static struct subfacet *
4511 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4512               size_t key_len, uint32_t key_hash)
4513 {
4514     struct subfacet *subfacet;
4515
4516     HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4517                              &backer->subfacets) {
4518         if (subfacet->key_len == key_len
4519             && !memcmp(key, subfacet->key, key_len)) {
4520             return subfacet;
4521         }
4522     }
4523
4524     return NULL;
4525 }
4526
4527 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4528  * 'key_fitness', 'key', and 'key_len' members in 'miss'.  Returns the
4529  * existing subfacet if there is one, otherwise creates and returns a
4530  * new subfacet. */
4531 static struct subfacet *
4532 subfacet_create(struct facet *facet, struct flow_miss *miss)
4533 {
4534     struct dpif_backer *backer = miss->ofproto->backer;
4535     enum odp_key_fitness key_fitness = miss->key_fitness;
4536     const struct nlattr *key = miss->key;
4537     size_t key_len = miss->key_len;
4538     uint32_t key_hash;
4539     struct subfacet *subfacet;
4540
4541     key_hash = odp_flow_key_hash(key, key_len);
4542
4543     if (list_is_empty(&facet->subfacets)) {
4544         subfacet = &facet->one_subfacet;
4545     } else {
4546         subfacet = subfacet_find(backer, key, key_len, key_hash);
4547         if (subfacet) {
4548             if (subfacet->facet == facet) {
4549                 return subfacet;
4550             }
4551
4552             /* This shouldn't happen. */
4553             VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4554             subfacet_destroy(subfacet);
4555         }
4556
4557         subfacet = xmalloc(sizeof *subfacet);
4558     }
4559
4560     hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4561     list_push_back(&facet->subfacets, &subfacet->list_node);
4562     subfacet->facet = facet;
4563     subfacet->key_fitness = key_fitness;
4564     subfacet->key = xmemdup(key, key_len);
4565     subfacet->key_len = key_len;
4566     subfacet->used = miss->stats.used;
4567     subfacet->created = subfacet->used;
4568     subfacet->dp_packet_count = 0;
4569     subfacet->dp_byte_count = 0;
4570     subfacet->path = SF_NOT_INSTALLED;
4571     subfacet->backer = backer;
4572
4573     backer->subfacet_add_count++;
4574     return subfacet;
4575 }
4576
4577 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4578  * its facet within 'ofproto', and frees it. */
4579 static void
4580 subfacet_destroy__(struct subfacet *subfacet)
4581 {
4582     struct facet *facet = subfacet->facet;
4583     struct ofproto_dpif *ofproto = facet->ofproto;
4584
4585     /* Update ofproto stats before uninstall the subfacet. */
4586     ofproto->backer->subfacet_del_count++;
4587
4588     subfacet_uninstall(subfacet);
4589     hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4590     list_remove(&subfacet->list_node);
4591     free(subfacet->key);
4592     if (subfacet != &facet->one_subfacet) {
4593         free(subfacet);
4594     }
4595 }
4596
4597 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4598  * last remaining subfacet in its facet destroys the facet too. */
4599 static void
4600 subfacet_destroy(struct subfacet *subfacet)
4601 {
4602     struct facet *facet = subfacet->facet;
4603
4604     if (list_is_singleton(&facet->subfacets)) {
4605         /* facet_remove() needs at least one subfacet (it will remove it). */
4606         facet_remove(facet);
4607     } else {
4608         subfacet_destroy__(subfacet);
4609     }
4610 }
4611
4612 static void
4613 subfacet_destroy_batch(struct dpif_backer *backer,
4614                        struct subfacet **subfacets, int n)
4615 {
4616     struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4617     struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4618     struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4619     int i;
4620
4621     for (i = 0; i < n; i++) {
4622         ops[i].type = DPIF_OP_FLOW_DEL;
4623         ops[i].u.flow_del.key = subfacets[i]->key;
4624         ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4625         ops[i].u.flow_del.stats = &stats[i];
4626         opsp[i] = &ops[i];
4627     }
4628
4629     dpif_operate(backer->dpif, opsp, n);
4630     for (i = 0; i < n; i++) {
4631         subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4632         subfacets[i]->path = SF_NOT_INSTALLED;
4633         subfacet_destroy(subfacets[i]);
4634         run_fast_rl();
4635     }
4636 }
4637
4638 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4639  * bytes of actions in 'actions'.  If 'stats' is non-null, statistics counters
4640  * in the datapath will be zeroed and 'stats' will be updated with traffic new
4641  * since 'subfacet' was last updated.
4642  *
4643  * Returns 0 if successful, otherwise a positive errno value. */
4644 static int
4645 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4646                  struct dpif_flow_stats *stats)
4647 {
4648     struct facet *facet = subfacet->facet;
4649     enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4650     const struct nlattr *actions = odp_actions->data;
4651     size_t actions_len = odp_actions->size;
4652     struct odputil_keybuf maskbuf;
4653     struct ofpbuf mask;
4654
4655     uint64_t slow_path_stub[128 / 8];
4656     enum dpif_flow_put_flags flags;
4657     int ret;
4658
4659     flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4660                                                : DPIF_FP_MODIFY;
4661     if (stats) {
4662         flags |= DPIF_FP_ZERO_STATS;
4663     }
4664
4665     if (path == SF_SLOW_PATH) {
4666         compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4667                           slow_path_stub, sizeof slow_path_stub,
4668                           &actions, &actions_len);
4669     }
4670
4671     ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4672     if (enable_megaflows) {
4673         odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4674                                &facet->flow, UINT32_MAX);
4675     }
4676
4677     ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4678                         subfacet->key_len,  mask.data, mask.size,
4679                         actions, actions_len, stats);
4680
4681     if (stats) {
4682         subfacet_reset_dp_stats(subfacet, stats);
4683     }
4684
4685     if (ret) {
4686         COVERAGE_INC(subfacet_install_fail);
4687     } else {
4688         subfacet->path = path;
4689     }
4690     return ret;
4691 }
4692
4693 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4694 static void
4695 subfacet_uninstall(struct subfacet *subfacet)
4696 {
4697     if (subfacet->path != SF_NOT_INSTALLED) {
4698         struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4699         struct dpif_flow_stats stats;
4700         int error;
4701
4702         error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4703                               subfacet->key_len, &stats);
4704         subfacet_reset_dp_stats(subfacet, &stats);
4705         if (!error) {
4706             subfacet_update_stats(subfacet, &stats);
4707         }
4708         subfacet->path = SF_NOT_INSTALLED;
4709     } else {
4710         ovs_assert(subfacet->dp_packet_count == 0);
4711         ovs_assert(subfacet->dp_byte_count == 0);
4712     }
4713 }
4714
4715 /* Resets 'subfacet''s datapath statistics counters.  This should be called
4716  * when 'subfacet''s statistics are cleared in the datapath.  If 'stats' is
4717  * non-null, it should contain the statistics returned by dpif when 'subfacet'
4718  * was reset in the datapath.  'stats' will be modified to include only
4719  * statistics new since 'subfacet' was last updated. */
4720 static void
4721 subfacet_reset_dp_stats(struct subfacet *subfacet,
4722                         struct dpif_flow_stats *stats)
4723 {
4724     if (stats
4725         && subfacet->dp_packet_count <= stats->n_packets
4726         && subfacet->dp_byte_count <= stats->n_bytes) {
4727         stats->n_packets -= subfacet->dp_packet_count;
4728         stats->n_bytes -= subfacet->dp_byte_count;
4729     }
4730
4731     subfacet->dp_packet_count = 0;
4732     subfacet->dp_byte_count = 0;
4733 }
4734
4735 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4736  *
4737  * Because of the meaning of a subfacet's counters, it only makes sense to do
4738  * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4739  * represents a packet that was sent by hand or if it represents statistics
4740  * that have been cleared out of the datapath. */
4741 static void
4742 subfacet_update_stats(struct subfacet *subfacet,
4743                       const struct dpif_flow_stats *stats)
4744 {
4745     if (stats->n_packets || stats->used > subfacet->used) {
4746         struct facet *facet = subfacet->facet;
4747
4748         subfacet->used = MAX(subfacet->used, stats->used);
4749         facet->used = MAX(facet->used, stats->used);
4750         facet->packet_count += stats->n_packets;
4751         facet->byte_count += stats->n_bytes;
4752         facet->tcp_flags |= stats->tcp_flags;
4753     }
4754 }
4755 \f
4756 /* Rules. */
4757
4758 /* Lookup 'flow' in 'ofproto''s classifier.  If 'wc' is non-null, sets
4759  * the fields that were relevant as part of the lookup. */
4760 void
4761 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4762                  struct flow_wildcards *wc, struct rule_dpif **rule)
4763 {
4764     struct ofport_dpif *port;
4765
4766     if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4767         return;
4768     }
4769     port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4770     if (!port) {
4771         VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4772                      flow->in_port.ofp_port);
4773     }
4774
4775     *rule = choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4776                              ofproto->no_packet_in_rule);
4777     ovs_rwlock_rdlock(&(*rule)->up.evict);
4778 }
4779
4780 bool
4781 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4782                           const struct flow *flow, struct flow_wildcards *wc,
4783                           uint8_t table_id, struct rule_dpif **rule)
4784     OVS_ACQ_RDLOCK((*rule)->up.evict)
4785 {
4786     struct cls_rule *cls_rule;
4787     struct classifier *cls;
4788     bool frag;
4789
4790     *rule = NULL;
4791     if (table_id >= N_TABLES) {
4792         return false;
4793     }
4794
4795     if (wc) {
4796         memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4797         wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4798     }
4799
4800     cls = &ofproto->up.tables[table_id].cls;
4801     ovs_rwlock_rdlock(&cls->rwlock);
4802     frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4803     if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4804         /* We must pretend that transport ports are unavailable. */
4805         struct flow ofpc_normal_flow = *flow;
4806         ofpc_normal_flow.tp_src = htons(0);
4807         ofpc_normal_flow.tp_dst = htons(0);
4808         cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4809     } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4810         cls_rule = &ofproto->drop_frags_rule->up.cr;
4811         if (wc) {
4812             flow_wildcards_init_exact(wc);
4813         }
4814     } else {
4815         cls_rule = classifier_lookup(cls, flow, wc);
4816     }
4817
4818     *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4819     if (*rule && ovs_rwlock_tryrdlock(&(*rule)->up.evict)) {
4820         /* The rule is in the process of being removed.  Best we can do is
4821          * pretend it isn't there. */
4822         *rule = NULL;
4823     }
4824     ovs_rwlock_unlock(&cls->rwlock);
4825
4826     return *rule != NULL;
4827 }
4828
4829 /* Given a port configuration (specified as zero if there's no port), chooses
4830  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4831  * flow table miss. */
4832 struct rule_dpif *
4833 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4834                  struct rule_dpif *no_packet_in_rule)
4835 {
4836     return config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4837 }
4838
4839 void
4840 rule_release(struct rule_dpif *rule)
4841 {
4842     if (rule) {
4843         ovs_rwlock_unlock(&rule->up.evict);
4844     }
4845 }
4846
4847 static void
4848 complete_operation(struct rule_dpif *rule)
4849 {
4850     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4851
4852     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4853     if (clogged) {
4854         struct dpif_completion *c = xmalloc(sizeof *c);
4855         c->op = rule->up.pending;
4856         list_push_back(&ofproto->completions, &c->list_node);
4857     } else {
4858         ofoperation_complete(rule->up.pending, 0);
4859     }
4860 }
4861
4862 static struct rule *
4863 rule_alloc(void)
4864 {
4865     struct rule_dpif *rule = xmalloc(sizeof *rule);
4866     return &rule->up;
4867 }
4868
4869 static void
4870 rule_dealloc(struct rule *rule_)
4871 {
4872     struct rule_dpif *rule = rule_dpif_cast(rule_);
4873     free(rule);
4874 }
4875
4876 static enum ofperr
4877 rule_construct(struct rule *rule_)
4878 {
4879     struct rule_dpif *rule = rule_dpif_cast(rule_);
4880     ovs_mutex_init(&rule->stats_mutex);
4881     ovs_mutex_lock(&rule->stats_mutex);
4882     rule->packet_count = 0;
4883     rule->byte_count = 0;
4884     ovs_mutex_unlock(&rule->stats_mutex);
4885     complete_operation(rule);
4886     return 0;
4887 }
4888
4889 static void
4890 rule_destruct(struct rule *rule_)
4891 {
4892     struct rule_dpif *rule = rule_dpif_cast(rule_);
4893     complete_operation(rule);
4894     ovs_mutex_destroy(&rule->stats_mutex);
4895 }
4896
4897 static void
4898 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4899 {
4900     struct rule_dpif *rule = rule_dpif_cast(rule_);
4901
4902     /* push_all_stats() can handle flow misses which, when using the learn
4903      * action, can cause rules to be added and deleted.  This can corrupt our
4904      * caller's datastructures which assume that rule_get_stats() doesn't have
4905      * an impact on the flow table. To be safe, we disable miss handling. */
4906     push_all_stats__(false);
4907
4908     /* Start from historical data for 'rule' itself that are no longer tracked
4909      * in facets.  This counts, for example, facets that have expired. */
4910     ovs_mutex_lock(&rule->stats_mutex);
4911     *packets = rule->packet_count;
4912     *bytes = rule->byte_count;
4913     ovs_mutex_unlock(&rule->stats_mutex);
4914 }
4915
4916 static void
4917 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4918                   struct ofpbuf *packet)
4919 {
4920     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4921     struct dpif_flow_stats stats;
4922     struct xlate_out xout;
4923     struct xlate_in xin;
4924
4925     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4926     rule_credit_stats(rule, &stats);
4927
4928     xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4929     xin.resubmit_stats = &stats;
4930     xlate_actions(&xin, &xout);
4931
4932     execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4933                         xout.odp_actions.size, packet);
4934
4935     xlate_out_uninit(&xout);
4936 }
4937
4938 static enum ofperr
4939 rule_execute(struct rule *rule, const struct flow *flow,
4940              struct ofpbuf *packet)
4941 {
4942     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4943     ofpbuf_delete(packet);
4944     return 0;
4945 }
4946
4947 static void
4948 rule_modify_actions(struct rule *rule_)
4949 {
4950     struct rule_dpif *rule = rule_dpif_cast(rule_);
4951
4952     complete_operation(rule);
4953 }
4954 \f
4955 /* Sends 'packet' out 'ofport'.
4956  * May modify 'packet'.
4957  * Returns 0 if successful, otherwise a positive errno value. */
4958 static int
4959 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4960 {
4961     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4962     uint64_t odp_actions_stub[1024 / 8];
4963     struct ofpbuf key, odp_actions;
4964     struct dpif_flow_stats stats;
4965     struct odputil_keybuf keybuf;
4966     struct ofpact_output output;
4967     struct xlate_out xout;
4968     struct xlate_in xin;
4969     struct flow flow;
4970     union flow_in_port in_port_;
4971     int error;
4972
4973     ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4974     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4975
4976     /* Use OFPP_NONE as the in_port to avoid special packet processing. */
4977     in_port_.ofp_port = OFPP_NONE;
4978     flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
4979     odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
4980                                                              OFPP_LOCAL));
4981     dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
4982
4983     ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
4984     output.port = ofport->up.ofp_port;
4985     output.max_len = 0;
4986
4987     xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
4988     xin.ofpacts_len = sizeof output;
4989     xin.ofpacts = &output.ofpact;
4990     xin.resubmit_stats = &stats;
4991     xlate_actions(&xin, &xout);
4992
4993     error = dpif_execute(ofproto->backer->dpif,
4994                          key.data, key.size,
4995                          xout.odp_actions.data, xout.odp_actions.size,
4996                          packet);
4997     xlate_out_uninit(&xout);
4998
4999     if (error) {
5000         VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5001                      ofproto->up.name, netdev_get_name(ofport->up.netdev),
5002                      ovs_strerror(error));
5003     }
5004
5005     ofproto->stats.tx_packets++;
5006     ofproto->stats.tx_bytes += packet->size;
5007     return error;
5008 }
5009
5010 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5011  * The action will state 'slow' as the reason that the action is in the slow
5012  * path.  (This is purely informational: it allows a human viewing "ovs-dpctl
5013  * dump-flows" output to see why a flow is in the slow path.)
5014  *
5015  * The 'stub_size' bytes in 'stub' will be used to store the action.
5016  * 'stub_size' must be large enough for the action.
5017  *
5018  * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5019  * respectively. */
5020 static void
5021 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5022                   enum slow_path_reason slow,
5023                   uint64_t *stub, size_t stub_size,
5024                   const struct nlattr **actionsp, size_t *actions_lenp)
5025 {
5026     union user_action_cookie cookie;
5027     struct ofpbuf buf;
5028
5029     cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5030     cookie.slow_path.unused = 0;
5031     cookie.slow_path.reason = slow;
5032
5033     ofpbuf_use_stack(&buf, stub, stub_size);
5034     if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5035         uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5036                                          ODPP_NONE);
5037         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5038     } else {
5039         odp_port_t odp_port;
5040         uint32_t pid;
5041
5042         odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5043         pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5044         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5045     }
5046     *actionsp = buf.data;
5047     *actions_lenp = buf.size;
5048 }
5049 \f
5050 static bool
5051 set_frag_handling(struct ofproto *ofproto_,
5052                   enum ofp_config_flags frag_handling)
5053 {
5054     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5055     if (frag_handling != OFPC_FRAG_REASM) {
5056         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5057         return true;
5058     } else {
5059         return false;
5060     }
5061 }
5062
5063 static enum ofperr
5064 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5065            const struct flow *flow,
5066            const struct ofpact *ofpacts, size_t ofpacts_len)
5067 {
5068     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5069     struct odputil_keybuf keybuf;
5070     struct dpif_flow_stats stats;
5071     struct xlate_out xout;
5072     struct xlate_in xin;
5073     struct ofpbuf key;
5074
5075
5076     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5077     odp_flow_key_from_flow(&key, flow,
5078                            ofp_port_to_odp_port(ofproto,
5079                                       flow->in_port.ofp_port));
5080
5081     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5082
5083     xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5084     xin.resubmit_stats = &stats;
5085     xin.ofpacts_len = ofpacts_len;
5086     xin.ofpacts = ofpacts;
5087
5088     xlate_actions(&xin, &xout);
5089     dpif_execute(ofproto->backer->dpif, key.data, key.size,
5090                  xout.odp_actions.data, xout.odp_actions.size, packet);
5091     xlate_out_uninit(&xout);
5092
5093     return 0;
5094 }
5095 \f
5096 /* NetFlow. */
5097
5098 static int
5099 set_netflow(struct ofproto *ofproto_,
5100             const struct netflow_options *netflow_options)
5101 {
5102     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5103
5104     if (netflow_options) {
5105         if (!ofproto->netflow) {
5106             ofproto->netflow = netflow_create();
5107             ofproto->backer->need_revalidate = REV_RECONFIGURE;
5108         }
5109         return netflow_set_options(ofproto->netflow, netflow_options);
5110     } else if (ofproto->netflow) {
5111         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5112         netflow_destroy(ofproto->netflow);
5113         ofproto->netflow = NULL;
5114     }
5115
5116     return 0;
5117 }
5118
5119 static void
5120 get_netflow_ids(const struct ofproto *ofproto_,
5121                 uint8_t *engine_type, uint8_t *engine_id)
5122 {
5123     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5124
5125     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5126 }
5127
5128 static void
5129 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5130 {
5131     if (!facet_is_controller_flow(facet) &&
5132         netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5133         struct subfacet *subfacet;
5134         struct ofexpired expired;
5135
5136         LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5137             if (subfacet->path == SF_FAST_PATH) {
5138                 struct dpif_flow_stats stats;
5139
5140                 subfacet_install(subfacet, &facet->xout.odp_actions,
5141                                  &stats);
5142                 subfacet_update_stats(subfacet, &stats);
5143             }
5144         }
5145
5146         expired.flow = facet->flow;
5147         expired.packet_count = facet->packet_count;
5148         expired.byte_count = facet->byte_count;
5149         expired.used = facet->used;
5150         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5151     }
5152 }
5153
5154 static void
5155 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5156 {
5157     struct cls_cursor cursor;
5158     struct facet *facet;
5159
5160     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5161     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5162     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5163         send_active_timeout(ofproto, facet);
5164     }
5165     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5166 }
5167 \f
5168 static struct ofproto_dpif *
5169 ofproto_dpif_lookup(const char *name)
5170 {
5171     struct ofproto_dpif *ofproto;
5172
5173     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5174                              hash_string(name, 0), &all_ofproto_dpifs) {
5175         if (!strcmp(ofproto->up.name, name)) {
5176             return ofproto;
5177         }
5178     }
5179     return NULL;
5180 }
5181
5182 static void
5183 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5184                           const char *argv[], void *aux OVS_UNUSED)
5185 {
5186     struct ofproto_dpif *ofproto;
5187
5188     if (argc > 1) {
5189         ofproto = ofproto_dpif_lookup(argv[1]);
5190         if (!ofproto) {
5191             unixctl_command_reply_error(conn, "no such bridge");
5192             return;
5193         }
5194         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5195         mac_learning_flush(ofproto->ml);
5196         ovs_rwlock_unlock(&ofproto->ml->rwlock);
5197     } else {
5198         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5199             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5200             mac_learning_flush(ofproto->ml);
5201             ovs_rwlock_unlock(&ofproto->ml->rwlock);
5202         }
5203     }
5204
5205     unixctl_command_reply(conn, "table successfully flushed");
5206 }
5207
5208 static struct ofport_dpif *
5209 ofbundle_get_a_port(const struct ofbundle *bundle)
5210 {
5211     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5212                         bundle_node);
5213 }
5214
5215 static void
5216 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5217                          const char *argv[], void *aux OVS_UNUSED)
5218 {
5219     struct ds ds = DS_EMPTY_INITIALIZER;
5220     const struct ofproto_dpif *ofproto;
5221     const struct mac_entry *e;
5222
5223     ofproto = ofproto_dpif_lookup(argv[1]);
5224     if (!ofproto) {
5225         unixctl_command_reply_error(conn, "no such bridge");
5226         return;
5227     }
5228
5229     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
5230     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5231     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5232         struct ofbundle *bundle = e->port.p;
5233         char name[OFP_MAX_PORT_NAME_LEN];
5234
5235         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5236                                name, sizeof name);
5237         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
5238                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
5239                       mac_entry_age(ofproto->ml, e));
5240     }
5241     ovs_rwlock_unlock(&ofproto->ml->rwlock);
5242     unixctl_command_reply(conn, ds_cstr(&ds));
5243     ds_destroy(&ds);
5244 }
5245
5246 struct trace_ctx {
5247     struct xlate_out xout;
5248     struct xlate_in xin;
5249     struct flow flow;
5250     struct ds *result;
5251 };
5252
5253 static void
5254 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5255 {
5256     ds_put_char_multiple(result, '\t', level);
5257     if (!rule) {
5258         ds_put_cstr(result, "No match\n");
5259         return;
5260     }
5261
5262     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5263                   rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5264     cls_rule_format(&rule->up.cr, result);
5265     ds_put_char(result, '\n');
5266
5267     ds_put_char_multiple(result, '\t', level);
5268     ds_put_cstr(result, "OpenFlow ");
5269     ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5270     ds_put_char(result, '\n');
5271 }
5272
5273 static void
5274 trace_format_flow(struct ds *result, int level, const char *title,
5275                   struct trace_ctx *trace)
5276 {
5277     ds_put_char_multiple(result, '\t', level);
5278     ds_put_format(result, "%s: ", title);
5279     if (flow_equal(&trace->xin.flow, &trace->flow)) {
5280         ds_put_cstr(result, "unchanged");
5281     } else {
5282         flow_format(result, &trace->xin.flow);
5283         trace->flow = trace->xin.flow;
5284     }
5285     ds_put_char(result, '\n');
5286 }
5287
5288 static void
5289 trace_format_regs(struct ds *result, int level, const char *title,
5290                   struct trace_ctx *trace)
5291 {
5292     size_t i;
5293
5294     ds_put_char_multiple(result, '\t', level);
5295     ds_put_format(result, "%s:", title);
5296     for (i = 0; i < FLOW_N_REGS; i++) {
5297         ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5298     }
5299     ds_put_char(result, '\n');
5300 }
5301
5302 static void
5303 trace_format_odp(struct ds *result, int level, const char *title,
5304                  struct trace_ctx *trace)
5305 {
5306     struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5307
5308     ds_put_char_multiple(result, '\t', level);
5309     ds_put_format(result, "%s: ", title);
5310     format_odp_actions(result, odp_actions->data, odp_actions->size);
5311     ds_put_char(result, '\n');
5312 }
5313
5314 static void
5315 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5316 {
5317     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5318     struct ds *result = trace->result;
5319
5320     ds_put_char(result, '\n');
5321     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5322     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5323     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
5324     trace_format_rule(result, recurse + 1, rule);
5325 }
5326
5327 static void
5328 trace_report(struct xlate_in *xin, const char *s, int recurse)
5329 {
5330     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5331     struct ds *result = trace->result;
5332
5333     ds_put_char_multiple(result, '\t', recurse);
5334     ds_put_cstr(result, s);
5335     ds_put_char(result, '\n');
5336 }
5337
5338 static void
5339 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5340                       void *aux OVS_UNUSED)
5341 {
5342     const struct dpif_backer *backer;
5343     struct ofproto_dpif *ofproto;
5344     struct ofpbuf odp_key, odp_mask;
5345     struct ofpbuf *packet;
5346     struct ds result;
5347     struct flow flow;
5348     char *s;
5349
5350     packet = NULL;
5351     backer = NULL;
5352     ds_init(&result);
5353     ofpbuf_init(&odp_key, 0);
5354     ofpbuf_init(&odp_mask, 0);
5355
5356     /* Handle "-generate" or a hex string as the last argument. */
5357     if (!strcmp(argv[argc - 1], "-generate")) {
5358         packet = ofpbuf_new(0);
5359         argc--;
5360     } else {
5361         const char *error = eth_from_hex(argv[argc - 1], &packet);
5362         if (!error) {
5363             argc--;
5364         } else if (argc == 4) {
5365             /* The 3-argument form must end in "-generate' or a hex string. */
5366             unixctl_command_reply_error(conn, error);
5367             goto exit;
5368         }
5369     }
5370
5371     /* Parse the flow and determine whether a datapath or
5372      * bridge is specified. If function odp_flow_key_from_string()
5373      * returns 0, the flow is a odp_flow. If function
5374      * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5375     if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5376         /* If the odp_flow is the second argument,
5377          * the datapath name is the first argument. */
5378         if (argc == 3) {
5379             const char *dp_type;
5380             if (!strncmp(argv[1], "ovs-", 4)) {
5381                 dp_type = argv[1] + 4;
5382             } else {
5383                 dp_type = argv[1];
5384             }
5385             backer = shash_find_data(&all_dpif_backers, dp_type);
5386             if (!backer) {
5387                 unixctl_command_reply_error(conn, "Cannot find datapath "
5388                                "of this name");
5389                 goto exit;
5390             }
5391         } else {
5392             /* No datapath name specified, so there should be only one
5393              * datapath. */
5394             struct shash_node *node;
5395             if (shash_count(&all_dpif_backers) != 1) {
5396                 unixctl_command_reply_error(conn, "Must specify datapath "
5397                          "name, there is more than one type of datapath");
5398                 goto exit;
5399             }
5400             node = shash_first(&all_dpif_backers);
5401             backer = node->data;
5402         }
5403
5404         if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5405                           NULL, &ofproto, NULL)) {
5406             unixctl_command_reply_error(conn, "Invalid datapath flow");
5407             goto exit;
5408         }
5409         ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5410     } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5411         if (argc != 3) {
5412             unixctl_command_reply_error(conn, "Must specify bridge name");
5413             goto exit;
5414         }
5415
5416         ofproto = ofproto_dpif_lookup(argv[1]);
5417         if (!ofproto) {
5418             unixctl_command_reply_error(conn, "Unknown bridge name");
5419             goto exit;
5420         }
5421     } else {
5422         unixctl_command_reply_error(conn, "Bad flow syntax");
5423         goto exit;
5424     }
5425
5426     /* Generate a packet, if requested. */
5427     if (packet) {
5428         if (!packet->size) {
5429             flow_compose(packet, &flow);
5430         } else {
5431             union flow_in_port in_port_;
5432
5433             in_port_ = flow.in_port;
5434             ds_put_cstr(&result, "Packet: ");
5435             s = ofp_packet_to_string(packet->data, packet->size);
5436             ds_put_cstr(&result, s);
5437             free(s);
5438
5439             /* Use the metadata from the flow and the packet argument
5440              * to reconstruct the flow. */
5441             flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5442                          &in_port_, &flow);
5443         }
5444     }
5445
5446     ofproto_trace(ofproto, &flow, packet, &result);
5447     unixctl_command_reply(conn, ds_cstr(&result));
5448
5449 exit:
5450     ds_destroy(&result);
5451     ofpbuf_delete(packet);
5452     ofpbuf_uninit(&odp_key);
5453     ofpbuf_uninit(&odp_mask);
5454 }
5455
5456 static void
5457 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5458               const struct ofpbuf *packet, struct ds *ds)
5459 {
5460     struct rule_dpif *rule;
5461     struct flow_wildcards wc;
5462
5463     ds_put_cstr(ds, "Flow: ");
5464     flow_format(ds, flow);
5465     ds_put_char(ds, '\n');
5466
5467     flow_wildcards_init_catchall(&wc);
5468     rule_dpif_lookup(ofproto, flow, &wc, &rule);
5469
5470     trace_format_rule(ds, 0, rule);
5471     if (rule == ofproto->miss_rule) {
5472         ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5473     } else if (rule == ofproto->no_packet_in_rule) {
5474         ds_put_cstr(ds, "\nNo match, packets dropped because "
5475                     "OFPPC_NO_PACKET_IN is set on in_port.\n");
5476     } else if (rule == ofproto->drop_frags_rule) {
5477         ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5478                     "and the fragment handling mode is \"drop\".\n");
5479     }
5480
5481     if (rule) {
5482         uint64_t odp_actions_stub[1024 / 8];
5483         struct ofpbuf odp_actions;
5484         struct trace_ctx trace;
5485         struct match match;
5486         uint8_t tcp_flags;
5487
5488         tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5489         trace.result = ds;
5490         trace.flow = *flow;
5491         ofpbuf_use_stub(&odp_actions,
5492                         odp_actions_stub, sizeof odp_actions_stub);
5493         xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5494         trace.xin.resubmit_hook = trace_resubmit;
5495         trace.xin.report_hook = trace_report;
5496
5497         xlate_actions(&trace.xin, &trace.xout);
5498         flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5499
5500         ds_put_char(ds, '\n');
5501         trace_format_flow(ds, 0, "Final flow", &trace);
5502
5503         match_init(&match, flow, &trace.xout.wc);
5504         ds_put_cstr(ds, "Relevant fields: ");
5505         match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5506         ds_put_char(ds, '\n');
5507
5508         ds_put_cstr(ds, "Datapath actions: ");
5509         format_odp_actions(ds, trace.xout.odp_actions.data,
5510                            trace.xout.odp_actions.size);
5511
5512         if (trace.xout.slow) {
5513             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5514                         "slow path because it:");
5515             switch (trace.xout.slow) {
5516             case SLOW_CFM:
5517                 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5518                 break;
5519             case SLOW_LACP:
5520                 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5521                 break;
5522             case SLOW_STP:
5523                 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5524                 break;
5525             case SLOW_BFD:
5526                 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5527                 break;
5528             case SLOW_CONTROLLER:
5529                 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5530                             "to the OpenFlow controller.");
5531                 break;
5532             case __SLOW_MAX:
5533                 NOT_REACHED();
5534             }
5535         }
5536
5537         xlate_out_uninit(&trace.xout);
5538     }
5539
5540     rule_release(rule);
5541 }
5542
5543 static void
5544 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5545                   const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5546 {
5547     clogged = true;
5548     unixctl_command_reply(conn, NULL);
5549 }
5550
5551 static void
5552 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5553                     const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5554 {
5555     clogged = false;
5556     unixctl_command_reply(conn, NULL);
5557 }
5558
5559 /* Runs a self-check of flow translations in 'ofproto'.  Appends a message to
5560  * 'reply' describing the results. */
5561 static void
5562 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5563 {
5564     struct cls_cursor cursor;
5565     struct facet *facet;
5566     int errors;
5567
5568     errors = 0;
5569     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5570     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5571     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5572         if (!facet_check_consistency(facet)) {
5573             errors++;
5574         }
5575     }
5576     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5577     if (errors) {
5578         ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5579     }
5580
5581     if (errors) {
5582         ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5583                       ofproto->up.name, errors);
5584     } else {
5585         ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5586     }
5587 }
5588
5589 static void
5590 ofproto_dpif_self_check(struct unixctl_conn *conn,
5591                         int argc, const char *argv[], void *aux OVS_UNUSED)
5592 {
5593     struct ds reply = DS_EMPTY_INITIALIZER;
5594     struct ofproto_dpif *ofproto;
5595
5596     if (argc > 1) {
5597         ofproto = ofproto_dpif_lookup(argv[1]);
5598         if (!ofproto) {
5599             unixctl_command_reply_error(conn, "Unknown ofproto (use "
5600                                         "ofproto/list for help)");
5601             return;
5602         }
5603         ofproto_dpif_self_check__(ofproto, &reply);
5604     } else {
5605         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5606             ofproto_dpif_self_check__(ofproto, &reply);
5607         }
5608     }
5609
5610     unixctl_command_reply(conn, ds_cstr(&reply));
5611     ds_destroy(&reply);
5612 }
5613
5614 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
5615  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
5616  * to destroy 'ofproto_shash' and free the returned value. */
5617 static const struct shash_node **
5618 get_ofprotos(struct shash *ofproto_shash)
5619 {
5620     const struct ofproto_dpif *ofproto;
5621
5622     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5623         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5624         shash_add_nocopy(ofproto_shash, name, ofproto);
5625     }
5626
5627     return shash_sort(ofproto_shash);
5628 }
5629
5630 static void
5631 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5632                               const char *argv[] OVS_UNUSED,
5633                               void *aux OVS_UNUSED)
5634 {
5635     struct ds ds = DS_EMPTY_INITIALIZER;
5636     struct shash ofproto_shash;
5637     const struct shash_node **sorted_ofprotos;
5638     int i;
5639
5640     shash_init(&ofproto_shash);
5641     sorted_ofprotos = get_ofprotos(&ofproto_shash);
5642     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5643         const struct shash_node *node = sorted_ofprotos[i];
5644         ds_put_format(&ds, "%s\n", node->name);
5645     }
5646
5647     shash_destroy(&ofproto_shash);
5648     free(sorted_ofprotos);
5649
5650     unixctl_command_reply(conn, ds_cstr(&ds));
5651     ds_destroy(&ds);
5652 }
5653
5654 static void
5655 show_dp_rates(struct ds *ds, const char *heading,
5656               const struct avg_subfacet_rates *rates)
5657 {
5658     ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5659                   heading, rates->add_rate, rates->del_rate);
5660 }
5661
5662 static void
5663 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5664 {
5665     const struct shash_node **ofprotos;
5666     struct ofproto_dpif *ofproto;
5667     struct shash ofproto_shash;
5668     uint64_t n_hit, n_missed;
5669     long long int minutes;
5670     size_t i;
5671
5672     n_hit = n_missed = 0;
5673     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5674         if (ofproto->backer == backer) {
5675             n_missed += ofproto->n_missed;
5676             n_hit += ofproto->n_hit;
5677         }
5678     }
5679
5680     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5681                   dpif_name(backer->dpif), n_hit, n_missed);
5682     ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5683                   " life span: %lldms\n", hmap_count(&backer->subfacets),
5684                   backer->avg_n_subfacet, backer->max_n_subfacet,
5685                   backer->avg_subfacet_life);
5686
5687     minutes = (time_msec() - backer->created) / (1000 * 60);
5688     if (minutes >= 60) {
5689         show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5690     }
5691     if (minutes >= 60 * 24) {
5692         show_dp_rates(ds, "\tdaily avg:",  &backer->daily);
5693     }
5694     show_dp_rates(ds, "\toverall avg:",  &backer->lifetime);
5695
5696     shash_init(&ofproto_shash);
5697     ofprotos = get_ofprotos(&ofproto_shash);
5698     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5699         struct ofproto_dpif *ofproto = ofprotos[i]->data;
5700         const struct shash_node **ports;
5701         size_t j;
5702
5703         if (ofproto->backer != backer) {
5704             continue;
5705         }
5706
5707         ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5708                       ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5709
5710         ports = shash_sort(&ofproto->up.port_by_name);
5711         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5712             const struct shash_node *node = ports[j];
5713             struct ofport *ofport = node->data;
5714             struct smap config;
5715             odp_port_t odp_port;
5716
5717             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5718                           ofport->ofp_port);
5719
5720             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5721             if (odp_port != ODPP_NONE) {
5722                 ds_put_format(ds, "%"PRIu32":", odp_port);
5723             } else {
5724                 ds_put_cstr(ds, "none:");
5725             }
5726
5727             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5728
5729             smap_init(&config);
5730             if (!netdev_get_config(ofport->netdev, &config)) {
5731                 const struct smap_node **nodes;
5732                 size_t i;
5733
5734                 nodes = smap_sort(&config);
5735                 for (i = 0; i < smap_count(&config); i++) {
5736                     const struct smap_node *node = nodes[i];
5737                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5738                                   node->key, node->value);
5739                 }
5740                 free(nodes);
5741             }
5742             smap_destroy(&config);
5743
5744             ds_put_char(ds, ')');
5745             ds_put_char(ds, '\n');
5746         }
5747         free(ports);
5748     }
5749     shash_destroy(&ofproto_shash);
5750     free(ofprotos);
5751 }
5752
5753 static void
5754 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5755                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5756 {
5757     struct ds ds = DS_EMPTY_INITIALIZER;
5758     const struct shash_node **backers;
5759     int i;
5760
5761     backers = shash_sort(&all_dpif_backers);
5762     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5763         dpif_show_backer(backers[i]->data, &ds);
5764     }
5765     free(backers);
5766
5767     unixctl_command_reply(conn, ds_cstr(&ds));
5768     ds_destroy(&ds);
5769 }
5770
5771 /* Dump the megaflow (facet) cache.  This is useful to check the
5772  * correctness of flow wildcarding, since the same mechanism is used for
5773  * both xlate caching and kernel wildcarding.
5774  *
5775  * It's important to note that in the output the flow description uses
5776  * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5777  *
5778  * This command is only needed for advanced debugging, so it's not
5779  * documented in the man page. */
5780 static void
5781 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5782                                     int argc OVS_UNUSED, const char *argv[],
5783                                     void *aux OVS_UNUSED)
5784 {
5785     struct ds ds = DS_EMPTY_INITIALIZER;
5786     const struct ofproto_dpif *ofproto;
5787     long long int now = time_msec();
5788     struct cls_cursor cursor;
5789     struct facet *facet;
5790
5791     ofproto = ofproto_dpif_lookup(argv[1]);
5792     if (!ofproto) {
5793         unixctl_command_reply_error(conn, "no such bridge");
5794         return;
5795     }
5796
5797     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5798     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5799     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5800         cls_rule_format(&facet->cr, &ds);
5801         ds_put_cstr(&ds, ", ");
5802         ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5803         ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5804         ds_put_cstr(&ds, "Datapath actions: ");
5805         if (facet->xout.slow) {
5806             uint64_t slow_path_stub[128 / 8];
5807             const struct nlattr *actions;
5808             size_t actions_len;
5809
5810             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5811                               slow_path_stub, sizeof slow_path_stub,
5812                               &actions, &actions_len);
5813             format_odp_actions(&ds, actions, actions_len);
5814         } else {
5815             format_odp_actions(&ds, facet->xout.odp_actions.data,
5816                                facet->xout.odp_actions.size);
5817         }
5818         ds_put_cstr(&ds, "\n");
5819     }
5820     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5821
5822     ds_chomp(&ds, '\n');
5823     unixctl_command_reply(conn, ds_cstr(&ds));
5824     ds_destroy(&ds);
5825 }
5826
5827 /* Disable using the megaflows.
5828  *
5829  * This command is only needed for advanced debugging, so it's not
5830  * documented in the man page. */
5831 static void
5832 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5833                                        int argc OVS_UNUSED,
5834                                        const char *argv[] OVS_UNUSED,
5835                                        void *aux OVS_UNUSED)
5836 {
5837     struct ofproto_dpif *ofproto;
5838
5839     enable_megaflows = false;
5840
5841     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5842         flush(&ofproto->up);
5843     }
5844
5845     unixctl_command_reply(conn, "megaflows disabled");
5846 }
5847
5848 /* Re-enable using megaflows.
5849  *
5850  * This command is only needed for advanced debugging, so it's not
5851  * documented in the man page. */
5852 static void
5853 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5854                                       int argc OVS_UNUSED,
5855                                       const char *argv[] OVS_UNUSED,
5856                                       void *aux OVS_UNUSED)
5857 {
5858     struct ofproto_dpif *ofproto;
5859
5860     enable_megaflows = true;
5861
5862     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5863         flush(&ofproto->up);
5864     }
5865
5866     unixctl_command_reply(conn, "megaflows enabled");
5867 }
5868
5869 static void
5870 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5871                                 int argc OVS_UNUSED, const char *argv[],
5872                                 void *aux OVS_UNUSED)
5873 {
5874     struct ds ds = DS_EMPTY_INITIALIZER;
5875     const struct ofproto_dpif *ofproto;
5876     struct subfacet *subfacet;
5877
5878     ofproto = ofproto_dpif_lookup(argv[1]);
5879     if (!ofproto) {
5880         unixctl_command_reply_error(conn, "no such bridge");
5881         return;
5882     }
5883
5884     update_stats(ofproto->backer);
5885
5886     HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5887         struct facet *facet = subfacet->facet;
5888         struct odputil_keybuf maskbuf;
5889         struct ofpbuf mask;
5890
5891         if (facet->ofproto != ofproto) {
5892             continue;
5893         }
5894
5895         ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5896         if (enable_megaflows) {
5897             odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5898                                    &facet->flow, UINT32_MAX);
5899         }
5900
5901         odp_flow_format(subfacet->key, subfacet->key_len,
5902                         mask.data, mask.size, &ds, false);
5903
5904         ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5905                       subfacet->dp_packet_count, subfacet->dp_byte_count);
5906         if (subfacet->used) {
5907             ds_put_format(&ds, "%.3fs",
5908                           (time_msec() - subfacet->used) / 1000.0);
5909         } else {
5910             ds_put_format(&ds, "never");
5911         }
5912         if (subfacet->facet->tcp_flags) {
5913             ds_put_cstr(&ds, ", flags:");
5914             packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5915         }
5916
5917         ds_put_cstr(&ds, ", actions:");
5918         if (facet->xout.slow) {
5919             uint64_t slow_path_stub[128 / 8];
5920             const struct nlattr *actions;
5921             size_t actions_len;
5922
5923             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5924                               slow_path_stub, sizeof slow_path_stub,
5925                               &actions, &actions_len);
5926             format_odp_actions(&ds, actions, actions_len);
5927         } else {
5928             format_odp_actions(&ds, facet->xout.odp_actions.data,
5929                                facet->xout.odp_actions.size);
5930         }
5931         ds_put_char(&ds, '\n');
5932     }
5933
5934     unixctl_command_reply(conn, ds_cstr(&ds));
5935     ds_destroy(&ds);
5936 }
5937
5938 static void
5939 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5940                                int argc OVS_UNUSED, const char *argv[],
5941                                void *aux OVS_UNUSED)
5942 {
5943     struct ds ds = DS_EMPTY_INITIALIZER;
5944     struct ofproto_dpif *ofproto;
5945
5946     ofproto = ofproto_dpif_lookup(argv[1]);
5947     if (!ofproto) {
5948         unixctl_command_reply_error(conn, "no such bridge");
5949         return;
5950     }
5951
5952     flush(&ofproto->up);
5953
5954     unixctl_command_reply(conn, ds_cstr(&ds));
5955     ds_destroy(&ds);
5956 }
5957
5958 static void
5959 ofproto_dpif_unixctl_init(void)
5960 {
5961     static bool registered;
5962     if (registered) {
5963         return;
5964     }
5965     registered = true;
5966
5967     unixctl_command_register(
5968         "ofproto/trace",
5969         "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
5970         1, 3, ofproto_unixctl_trace, NULL);
5971     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5972                              ofproto_unixctl_fdb_flush, NULL);
5973     unixctl_command_register("fdb/show", "bridge", 1, 1,
5974                              ofproto_unixctl_fdb_show, NULL);
5975     unixctl_command_register("ofproto/clog", "", 0, 0,
5976                              ofproto_dpif_clog, NULL);
5977     unixctl_command_register("ofproto/unclog", "", 0, 0,
5978                              ofproto_dpif_unclog, NULL);
5979     unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
5980                              ofproto_dpif_self_check, NULL);
5981     unixctl_command_register("dpif/dump-dps", "", 0, 0,
5982                              ofproto_unixctl_dpif_dump_dps, NULL);
5983     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5984                              NULL);
5985     unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
5986                              ofproto_unixctl_dpif_dump_flows, NULL);
5987     unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
5988                              ofproto_unixctl_dpif_del_flows, NULL);
5989     unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
5990                              ofproto_unixctl_dpif_dump_megaflows, NULL);
5991     unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
5992                              ofproto_unixctl_dpif_disable_megaflows, NULL);
5993     unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
5994                              ofproto_unixctl_dpif_enable_megaflows, NULL);
5995 }
5996 \f
5997 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5998  *
5999  * This is deprecated.  It is only for compatibility with broken device drivers
6000  * in old versions of Linux that do not properly support VLANs when VLAN
6001  * devices are not used.  When broken device drivers are no longer in
6002  * widespread use, we will delete these interfaces. */
6003
6004 static int
6005 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6006 {
6007     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6008     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6009
6010     if (realdev_ofp_port == ofport->realdev_ofp_port
6011         && vid == ofport->vlandev_vid) {
6012         return 0;
6013     }
6014
6015     ofproto->backer->need_revalidate = REV_RECONFIGURE;
6016
6017     if (ofport->realdev_ofp_port) {
6018         vsp_remove(ofport);
6019     }
6020     if (realdev_ofp_port && ofport->bundle) {
6021         /* vlandevs are enslaved to their realdevs, so they are not allowed to
6022          * themselves be part of a bundle. */
6023         bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6024     }
6025
6026     ofport->realdev_ofp_port = realdev_ofp_port;
6027     ofport->vlandev_vid = vid;
6028
6029     if (realdev_ofp_port) {
6030         vsp_add(ofport, realdev_ofp_port, vid);
6031     }
6032
6033     return 0;
6034 }
6035
6036 static uint32_t
6037 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6038 {
6039     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6040 }
6041
6042 bool
6043 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6044     OVS_EXCLUDED(ofproto->vsp_mutex)
6045 {
6046     bool ret;
6047
6048     ovs_mutex_lock(&ofproto->vsp_mutex);
6049     ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6050     ovs_mutex_unlock(&ofproto->vsp_mutex);
6051     return ret;
6052 }
6053
6054 static ofp_port_t
6055 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6056                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6057     OVS_REQUIRES(ofproto->vsp_mutex)
6058 {
6059     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6060         int vid = vlan_tci_to_vid(vlan_tci);
6061         const struct vlan_splinter *vsp;
6062
6063         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6064                                  hash_realdev_vid(realdev_ofp_port, vid),
6065                                  &ofproto->realdev_vid_map) {
6066             if (vsp->realdev_ofp_port == realdev_ofp_port
6067                 && vsp->vid == vid) {
6068                 return vsp->vlandev_ofp_port;
6069             }
6070         }
6071     }
6072     return realdev_ofp_port;
6073 }
6074
6075 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6076  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6077  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
6078  * 'vlan_tci' 9, it would return the port number of eth0.9.
6079  *
6080  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6081  * function just returns its 'realdev_ofp_port' argument. */
6082 ofp_port_t
6083 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6084                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6085     OVS_EXCLUDED(ofproto->vsp_mutex)
6086 {
6087     ofp_port_t ret;
6088
6089     ovs_mutex_lock(&ofproto->vsp_mutex);
6090     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6091     ovs_mutex_unlock(&ofproto->vsp_mutex);
6092     return ret;
6093 }
6094
6095 static struct vlan_splinter *
6096 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6097 {
6098     struct vlan_splinter *vsp;
6099
6100     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6101                              hash_ofp_port(vlandev_ofp_port),
6102                              &ofproto->vlandev_map) {
6103         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6104             return vsp;
6105         }
6106     }
6107
6108     return NULL;
6109 }
6110
6111 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6112  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6113  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
6114  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6115  * eth0 and store 9 in '*vid'.
6116  *
6117  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6118  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
6119  * always does.*/
6120 static ofp_port_t
6121 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6122                        ofp_port_t vlandev_ofp_port, int *vid)
6123     OVS_REQUIRES(ofproto->vsp_mutex)
6124 {
6125     if (!hmap_is_empty(&ofproto->vlandev_map)) {
6126         const struct vlan_splinter *vsp;
6127
6128         vsp = vlandev_find(ofproto, vlandev_ofp_port);
6129         if (vsp) {
6130             if (vid) {
6131                 *vid = vsp->vid;
6132             }
6133             return vsp->realdev_ofp_port;
6134         }
6135     }
6136     return 0;
6137 }
6138
6139 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6140  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
6141  * 'flow->in_port' to the "real" device backing the VLAN device, sets
6142  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Otherwise (which is
6143  * always the case unless VLAN splinters are enabled), returns false without
6144  * making any changes. */
6145 bool
6146 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6147     OVS_EXCLUDED(ofproto->vsp_mutex)
6148 {
6149     ofp_port_t realdev;
6150     int vid;
6151
6152     ovs_mutex_lock(&ofproto->vsp_mutex);
6153     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6154     ovs_mutex_unlock(&ofproto->vsp_mutex);
6155     if (!realdev) {
6156         return false;
6157     }
6158
6159     /* Cause the flow to be processed as if it came in on the real device with
6160      * the VLAN device's VLAN ID. */
6161     flow->in_port.ofp_port = realdev;
6162     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6163     return true;
6164 }
6165
6166 static void
6167 vsp_remove(struct ofport_dpif *port)
6168 {
6169     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6170     struct vlan_splinter *vsp;
6171
6172     ovs_mutex_lock(&ofproto->vsp_mutex);
6173     vsp = vlandev_find(ofproto, port->up.ofp_port);
6174     if (vsp) {
6175         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6176         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6177         free(vsp);
6178
6179         port->realdev_ofp_port = 0;
6180     } else {
6181         VLOG_ERR("missing vlan device record");
6182     }
6183     ovs_mutex_unlock(&ofproto->vsp_mutex);
6184 }
6185
6186 static void
6187 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6188 {
6189     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6190
6191     ovs_mutex_lock(&ofproto->vsp_mutex);
6192     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6193         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6194             == realdev_ofp_port)) {
6195         struct vlan_splinter *vsp;
6196
6197         vsp = xmalloc(sizeof *vsp);
6198         vsp->realdev_ofp_port = realdev_ofp_port;
6199         vsp->vlandev_ofp_port = port->up.ofp_port;
6200         vsp->vid = vid;
6201
6202         port->realdev_ofp_port = realdev_ofp_port;
6203
6204         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6205                     hash_ofp_port(port->up.ofp_port));
6206         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6207                     hash_realdev_vid(realdev_ofp_port, vid));
6208     } else {
6209         VLOG_ERR("duplicate vlan device record");
6210     }
6211     ovs_mutex_unlock(&ofproto->vsp_mutex);
6212 }
6213
6214 static odp_port_t
6215 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6216 {
6217     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6218     return ofport ? ofport->odp_port : ODPP_NONE;
6219 }
6220
6221 struct ofport_dpif *
6222 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6223 {
6224     struct ofport_dpif *port;
6225
6226     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6227     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6228                              &backer->odp_to_ofport_map) {
6229         if (port->odp_port == odp_port) {
6230             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6231             return port;
6232         }
6233     }
6234
6235     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6236     return NULL;
6237 }
6238
6239 static ofp_port_t
6240 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6241 {
6242     struct ofport_dpif *port;
6243
6244     port = odp_port_to_ofport(ofproto->backer, odp_port);
6245     if (port && &ofproto->up == port->up.ofproto) {
6246         return port->up.ofp_port;
6247     } else {
6248         return OFPP_NONE;
6249     }
6250 }
6251
6252 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6253  * most heavily weighted element.  'base' designates the rate of decay: after
6254  * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6255  * (about .37). */
6256 static void
6257 exp_mavg(double *avg, int base, double new)
6258 {
6259     *avg = (*avg * (base - 1) + new) / base;
6260 }
6261
6262 static void
6263 update_moving_averages(struct dpif_backer *backer)
6264 {
6265     const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6266     long long int minutes = (time_msec() - backer->created) / min_ms;
6267
6268     if (minutes > 0) {
6269         backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6270             / minutes;
6271         backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6272             / minutes;
6273     } else {
6274         backer->lifetime.add_rate = 0.0;
6275         backer->lifetime.del_rate = 0.0;
6276     }
6277
6278     /* Update hourly averages on the minute boundaries. */
6279     if (time_msec() - backer->last_minute >= min_ms) {
6280         exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6281         exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6282
6283         /* Update daily averages on the hour boundaries. */
6284         if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6285             exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6286             exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6287         }
6288
6289         backer->total_subfacet_add_count += backer->subfacet_add_count;
6290         backer->total_subfacet_del_count += backer->subfacet_del_count;
6291         backer->subfacet_add_count = 0;
6292         backer->subfacet_del_count = 0;
6293         backer->last_minute += min_ms;
6294     }
6295 }
6296
6297 const struct ofproto_class ofproto_dpif_class = {
6298     init,
6299     enumerate_types,
6300     enumerate_names,
6301     del,
6302     port_open_type,
6303     type_run,
6304     type_run_fast,
6305     type_wait,
6306     alloc,
6307     construct,
6308     destruct,
6309     dealloc,
6310     run,
6311     run_fast,
6312     wait,
6313     get_memory_usage,
6314     flush,
6315     get_features,
6316     get_tables,
6317     port_alloc,
6318     port_construct,
6319     port_destruct,
6320     port_dealloc,
6321     port_modified,
6322     port_reconfigured,
6323     port_query_by_name,
6324     port_add,
6325     port_del,
6326     port_get_stats,
6327     port_dump_start,
6328     port_dump_next,
6329     port_dump_done,
6330     port_poll,
6331     port_poll_wait,
6332     port_is_lacp_current,
6333     NULL,                       /* rule_choose_table */
6334     rule_alloc,
6335     rule_construct,
6336     rule_destruct,
6337     rule_dealloc,
6338     rule_get_stats,
6339     rule_execute,
6340     rule_modify_actions,
6341     set_frag_handling,
6342     packet_out,
6343     set_netflow,
6344     get_netflow_ids,
6345     set_sflow,
6346     set_ipfix,
6347     set_cfm,
6348     get_cfm_status,
6349     set_bfd,
6350     get_bfd_status,
6351     set_stp,
6352     get_stp_status,
6353     set_stp_port,
6354     get_stp_port_status,
6355     set_queues,
6356     bundle_set,
6357     bundle_remove,
6358     mirror_set__,
6359     mirror_get_stats__,
6360     set_flood_vlans,
6361     is_mirror_output_bundle,
6362     forward_bpdu_changed,
6363     set_mac_table_config,
6364     set_realdev,
6365     NULL,                       /* meter_get_features */
6366     NULL,                       /* meter_set */
6367     NULL,                       /* meter_get */
6368     NULL,                       /* meter_del */
6369 };