ofproto/ofproto-dpif.c

   1 /*
   2  * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18
  19 #include "ofproto/ofproto-dpif.h"
  20 #include "ofproto/ofproto-provider.h"
  21
  22 #include <errno.h>
  23
  24 #include "bfd.h"
  25 #include "bond.h"
  26 #include "bundle.h"
  27 #include "byte-order.h"
  28 #include "connmgr.h"
  29 #include "coverage.h"
  30 #include "cfm.h"
  31 #include "dpif.h"
  32 #include "dynamic-string.h"
  33 #include "fail-open.h"
  34 #include "hmapx.h"
  35 #include "lacp.h"
  36 #include "learn.h"
  37 #include "mac-learning.h"
  38 #include "meta-flow.h"
  39 #include "multipath.h"
  40 #include "netdev-vport.h"
  41 #include "netdev.h"
  42 #include "netlink.h"
  43 #include "nx-match.h"
  44 #include "odp-util.h"
  45 #include "odp-execute.h"
  46 #include "ofp-util.h"
  47 #include "ofpbuf.h"
  48 #include "ofp-actions.h"
  49 #include "ofp-parse.h"
  50 #include "ofp-print.h"
  51 #include "ofproto-dpif-governor.h"
  52 #include "ofproto-dpif-ipfix.h"
  53 #include "ofproto-dpif-mirror.h"
  54 #include "ofproto-dpif-sflow.h"
  55 #include "ofproto-dpif-upcall.h"
  56 #include "ofproto-dpif-xlate.h"
  57 #include "poll-loop.h"
  58 #include "simap.h"
  59 #include "smap.h"
  60 #include "timer.h"
  61 #include "tunnel.h"
  62 #include "unaligned.h"
  63 #include "unixctl.h"
  64 #include "vlan-bitmap.h"
  65 #include "vlog.h"
  66
  67 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  68
  69 COVERAGE_DEFINE(ofproto_dpif_expired);
  70 COVERAGE_DEFINE(facet_changed_rule);
  71 COVERAGE_DEFINE(facet_revalidate);
  72 COVERAGE_DEFINE(facet_unexpected);
  73 COVERAGE_DEFINE(facet_suppress);
  74 COVERAGE_DEFINE(subfacet_install_fail);
  75 COVERAGE_DEFINE(packet_in_overflow);
  76 COVERAGE_DEFINE(flow_mod_overflow);
  77
  78 /* Number of implemented OpenFlow tables. */
  79 enum { N_TABLES = 255 };
  80 enum { TBL_INTERNAL = N_TABLES - 1 };    /* Used for internal hidden rules. */
  81 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
  82
  83 struct flow_miss;
  84 struct facet;
  85
  86 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
  87
  88 struct ofbundle {
  89     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
  90     struct ofproto_dpif *ofproto; /* Owning ofproto. */
  91     void *aux;                  /* Key supplied by ofproto's client. */
  92     char *name;                 /* Identifier for log messages. */
  93
  94     /* Configuration. */
  95     struct list ports;          /* Contains "struct ofport"s. */
  96     enum port_vlan_mode vlan_mode; /* VLAN mode */
  97     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
  98     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
  99                                  * NULL if all VLANs are trunked. */
 100     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 101     struct bond *bond;          /* Nonnull iff more than one port. */
 102     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 103
 104     /* Status. */
 105     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 106 };
 107
 108 static void bundle_remove(struct ofport *);
 109 static void bundle_update(struct ofbundle *);
 110 static void bundle_destroy(struct ofbundle *);
 111 static void bundle_del_port(struct ofport_dpif *);
 112 static void bundle_run(struct ofbundle *);
 113 static void bundle_wait(struct ofbundle *);
 114
 115 static void stp_run(struct ofproto_dpif *ofproto);
 116 static void stp_wait(struct ofproto_dpif *ofproto);
 117 static int set_stp_port(struct ofport *,
 118                         const struct ofproto_port_stp_settings *);
 119
 120 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
 121                               enum slow_path_reason,
 122                               uint64_t *stub, size_t stub_size,
 123                               const struct nlattr **actionsp,
 124                               size_t *actions_lenp);
 125
 126 /* A subfacet (see "struct subfacet" below) has three possible installation
 127  * states:
 128  *
 129  *   - SF_NOT_INSTALLED: Not installed in the datapath.  This will only be the
 130  *     case just after the subfacet is created, just before the subfacet is
 131  *     destroyed, or if the datapath returns an error when we try to install a
 132  *     subfacet.
 133  *
 134  *   - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
 135  *
 136  *   - SF_SLOW_PATH: An action that sends every packet for the subfacet through
 137  *     ofproto_dpif is installed in the datapath.
 138  */
 139 enum subfacet_path {
 140     SF_NOT_INSTALLED,           /* No datapath flow for this subfacet. */
 141     SF_FAST_PATH,               /* Full actions are installed. */
 142     SF_SLOW_PATH,               /* Send-to-userspace action is installed. */
 143 };
 144
 145 /* A dpif flow and actions associated with a facet.
 146  *
 147  * See also the large comment on struct facet. */
 148 struct subfacet {
 149     /* Owners. */
 150     struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
 151     struct list list_node;      /* In struct facet's 'facets' list. */
 152     struct facet *facet;        /* Owning facet. */
 153     struct dpif_backer *backer; /* Owning backer. */
 154
 155     enum odp_key_fitness key_fitness;
 156     struct nlattr *key;
 157     int key_len;
 158
 159     long long int used;         /* Time last used; time created if not used. */
 160     long long int created;      /* Time created. */
 161
 162     uint64_t dp_packet_count;   /* Last known packet count in the datapath. */
 163     uint64_t dp_byte_count;     /* Last known byte count in the datapath. */
 164
 165     enum subfacet_path path;    /* Installed in datapath? */
 166 };
 167
 168 #define SUBFACET_DESTROY_MAX_BATCH 50
 169
 170 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *);
 171 static struct subfacet *subfacet_find(struct dpif_backer *,
 172                                       const struct nlattr *key, size_t key_len,
 173                                       uint32_t key_hash);
 174 static void subfacet_destroy(struct subfacet *);
 175 static void subfacet_destroy__(struct subfacet *);
 176 static void subfacet_destroy_batch(struct dpif_backer *,
 177                                    struct subfacet **, int n);
 178 static void subfacet_reset_dp_stats(struct subfacet *,
 179                                     struct dpif_flow_stats *);
 180 static void subfacet_update_stats(struct subfacet *,
 181                                   const struct dpif_flow_stats *);
 182 static int subfacet_install(struct subfacet *,
 183                             const struct ofpbuf *odp_actions,
 184                             struct dpif_flow_stats *);
 185 static void subfacet_uninstall(struct subfacet *);
 186
 187 /* A unique, non-overlapping instantiation of an OpenFlow flow.
 188  *
 189  * A facet associates a "struct flow", which represents the Open vSwitch
 190  * userspace idea of an exact-match flow, with one or more subfacets.
 191  * While the facet is created based on an exact-match flow, it is stored
 192  * within the ofproto based on the wildcards that could be expressed
 193  * based on the flow table and other configuration.  (See the 'wc'
 194  * description in "struct xlate_out" for more details.)
 195  *
 196  * Each subfacet tracks the datapath's idea of the flow equivalent to
 197  * the facet.  When the kernel module (or other dpif implementation) and
 198  * Open vSwitch userspace agree on the definition of a flow key, there
 199  * is exactly one subfacet per facet.  If the dpif implementation
 200  * supports more-specific flow matching than userspace, however, a facet
 201  * can have more than one subfacet.  Examples include the dpif
 202  * implementation not supporting the same wildcards as userspace or some
 203  * distinction in flow that userspace simply doesn't understand.
 204  *
 205  * Flow expiration works in terms of subfacets, so a facet must have at
 206  * least one subfacet or it will never expire, leaking memory. */
 207 struct facet {
 208     /* Owners. */
 209     struct hmap_node hmap_node;  /* In owning ofproto's 'facets' hmap. */
 210     struct ofproto_dpif *ofproto;
 211
 212     /* Owned data. */
 213     struct list subfacets;
 214     long long int used;         /* Time last used; time created if not used. */
 215
 216     /* Key. */
 217     struct flow flow;           /* Flow of the creating subfacet. */
 218     struct cls_rule cr;         /* In 'ofproto_dpif's facets classifier. */
 219
 220     /* These statistics:
 221      *
 222      *   - Do include packets and bytes sent "by hand", e.g. with
 223      *     dpif_execute().
 224      *
 225      *   - Do include packets and bytes that were obtained from the datapath
 226      *     when a subfacet's statistics were reset (e.g. dpif_flow_put() with
 227      *     DPIF_FP_ZERO_STATS).
 228      *
 229      *   - Do not include packets or bytes that can be obtained from the
 230      *     datapath for any existing subfacet.
 231      */
 232     uint64_t packet_count;       /* Number of packets received. */
 233     uint64_t byte_count;         /* Number of bytes received. */
 234
 235     /* Resubmit statistics. */
 236     uint64_t prev_packet_count;  /* Number of packets from last stats push. */
 237     uint64_t prev_byte_count;    /* Number of bytes from last stats push. */
 238     long long int prev_used;     /* Used time from last stats push. */
 239
 240     /* Accounting. */
 241     uint64_t accounted_bytes;    /* Bytes processed by facet_account(). */
 242     struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
 243     uint8_t tcp_flags;           /* TCP flags seen for this 'rule'. */
 244
 245     struct xlate_out xout;
 246
 247     /* Storage for a single subfacet, to reduce malloc() time and space
 248      * overhead.  (A facet always has at least one subfacet and in the common
 249      * case has exactly one subfacet.  However, 'one_subfacet' may not
 250      * always be valid, since it could have been removed after newer
 251      * subfacets were pushed onto the 'subfacets' list.) */
 252     struct subfacet one_subfacet;
 253
 254     long long int learn_rl;      /* Rate limiter for facet_learn(). */
 255 };
 256
 257 static struct facet *facet_create(const struct flow_miss *);
 258 static void facet_remove(struct facet *);
 259 static void facet_free(struct facet *);
 260
 261 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
 262 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
 263                                         const struct flow *);
 264 static bool facet_revalidate(struct facet *);
 265 static bool facet_check_consistency(struct facet *);
 266
 267 static void facet_flush_stats(struct facet *);
 268
 269 static void facet_reset_counters(struct facet *);
 270 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
 271                             struct dpif_flow_stats *, bool may_learn);
 272 static void facet_push_stats(struct facet *, bool may_learn);
 273 static void facet_learn(struct facet *);
 274 static void facet_account(struct facet *);
 275 static void push_all_stats(void);
 276
 277 static bool facet_is_controller_flow(struct facet *);
 278
 279 struct ofport_dpif {
 280     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 281     struct ofport up;
 282
 283     odp_port_t odp_port;
 284     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 285     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
 286     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 287     struct bfd *bfd;            /* BFD, if any. */
 288     bool may_enable;            /* May be enabled in bonds. */
 289     bool is_tunnel;             /* This port is a tunnel. */
 290     long long int carrier_seq;  /* Carrier status changes. */
 291     struct ofport_dpif *peer;   /* Peer if patch port. */
 292
 293     /* Spanning tree. */
 294     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 295     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 296     long long int stp_state_entered;
 297
 298     /* Queue to DSCP mapping. */
 299     struct ofproto_port_queue *qdscp;
 300     size_t n_qdscp;
 301
 302     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 303      *
 304      * This is deprecated.  It is only for compatibility with broken device
 305      * drivers in old versions of Linux that do not properly support VLANs when
 306      * VLAN devices are not used.  When broken device drivers are no longer in
 307      * widespread use, we will delete these interfaces. */
 308     ofp_port_t realdev_ofp_port;
 309     int vlandev_vid;
 310 };
 311
 312 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 313  *
 314  * This is deprecated.  It is only for compatibility with broken device drivers
 315  * in old versions of Linux that do not properly support VLANs when VLAN
 316  * devices are not used.  When broken device drivers are no longer in
 317  * widespread use, we will delete these interfaces. */
 318 struct vlan_splinter {
 319     struct hmap_node realdev_vid_node;
 320     struct hmap_node vlandev_node;
 321     ofp_port_t realdev_ofp_port;
 322     ofp_port_t vlandev_ofp_port;
 323     int vid;
 324 };
 325
 326 static void vsp_remove(struct ofport_dpif *);
 327 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 328
 329 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 330                                        ofp_port_t);
 331
 332 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 333                                        odp_port_t);
 334
 335 static struct ofport_dpif *
 336 ofport_dpif_cast(const struct ofport *ofport)
 337 {
 338     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 339 }
 340
 341 static void port_run(struct ofport_dpif *);
 342 static void port_run_fast(struct ofport_dpif *);
 343 static void port_wait(struct ofport_dpif *);
 344 static int set_bfd(struct ofport *, const struct smap *);
 345 static int set_cfm(struct ofport *, const struct cfm_settings *);
 346 static void ofport_update_peer(struct ofport_dpif *);
 347 static void run_fast_rl(void);
 348 static int run_fast(struct ofproto *);
 349
 350 struct dpif_completion {
 351     struct list list_node;
 352     struct ofoperation *op;
 353 };
 354
 355 /* Reasons that we might need to revalidate every facet, and corresponding
 356  * coverage counters.
 357  *
 358  * A value of 0 means that there is no need to revalidate.
 359  *
 360  * It would be nice to have some cleaner way to integrate with coverage
 361  * counters, but with only a few reasons I guess this is good enough for
 362  * now. */
 363 enum revalidate_reason {
 364     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 365     REV_STP,                   /* Spanning tree protocol port status change. */
 366     REV_BOND,                  /* Bonding changed. */
 367     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 368     REV_FLOW_TABLE,            /* Flow table changed. */
 369     REV_MAC_LEARNING,          /* Mac learning changed. */
 370     REV_INCONSISTENCY          /* Facet self-check failed. */
 371 };
 372 COVERAGE_DEFINE(rev_reconfigure);
 373 COVERAGE_DEFINE(rev_stp);
 374 COVERAGE_DEFINE(rev_bond);
 375 COVERAGE_DEFINE(rev_port_toggled);
 376 COVERAGE_DEFINE(rev_flow_table);
 377 COVERAGE_DEFINE(rev_mac_learning);
 378 COVERAGE_DEFINE(rev_inconsistency);
 379
 380 struct avg_subfacet_rates {
 381     double add_rate;   /* Moving average of new flows created per minute. */
 382     double del_rate;   /* Moving average of flows deleted per minute. */
 383 };
 384
 385 /* All datapaths of a given type share a single dpif backer instance. */
 386 struct dpif_backer {
 387     char *type;
 388     int refcount;
 389     struct dpif *dpif;
 390     struct udpif *udpif;
 391     struct timer next_expiration;
 392
 393     struct ovs_rwlock odp_to_ofport_lock;
 394     struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
 395
 396     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 397
 398     /* Facet revalidation flags applying to facets which use this backer. */
 399     enum revalidate_reason need_revalidate; /* Revalidate every facet. */
 400
 401     struct hmap drop_keys; /* Set of dropped odp keys. */
 402     bool recv_set_enable; /* Enables or disables receiving packets. */
 403
 404     struct hmap subfacets;
 405     struct governor *governor;
 406
 407     /* Subfacet statistics.
 408      *
 409      * These keep track of the total number of subfacets added and deleted and
 410      * flow life span.  They are useful for computing the flow rates stats
 411      * exposed via "ovs-appctl dpif/show".  The goal is to learn about
 412      * traffic patterns in ways that we can use later to improve Open vSwitch
 413      * performance in new situations.  */
 414     long long int created;           /* Time when it is created. */
 415     unsigned max_n_subfacet;         /* Maximum number of flows */
 416     unsigned avg_n_subfacet;         /* Average number of flows. */
 417     long long int avg_subfacet_life; /* Average life span of subfacets. */
 418
 419     /* The average number of subfacets... */
 420     struct avg_subfacet_rates hourly;   /* ...over the last hour. */
 421     struct avg_subfacet_rates daily;    /* ...over the last day. */
 422     struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
 423     long long int last_minute;          /* Last time 'hourly' was updated. */
 424
 425     /* Number of subfacets added or deleted since 'last_minute'. */
 426     unsigned subfacet_add_count;
 427     unsigned subfacet_del_count;
 428
 429     /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
 430     unsigned long long int total_subfacet_add_count;
 431     unsigned long long int total_subfacet_del_count;
 432
 433     /* Number of upcall handling threads. */
 434     unsigned int n_handler_threads;
 435 };
 436
 437 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 438 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 439
 440 static void drop_key_clear(struct dpif_backer *);
 441 static void update_moving_averages(struct dpif_backer *backer);
 442
 443 struct ofproto_dpif {
 444     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 445     struct ofproto up;
 446     struct dpif_backer *backer;
 447
 448     /* Special OpenFlow rules. */
 449     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 450     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 451     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 452
 453     /* Bridging. */
 454     struct netflow *netflow;
 455     struct dpif_sflow *sflow;
 456     struct dpif_ipfix *ipfix;
 457     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 458     struct mac_learning *ml;
 459     bool has_bonded_bundles;
 460     struct mbridge *mbridge;
 461
 462     /* Facets. */
 463     struct classifier facets;     /* Contains 'struct facet's. */
 464     long long int consistency_rl;
 465
 466     /* Support for debugging async flow mods. */
 467     struct list completions;
 468
 469     struct netdev_stats stats; /* To account packets generated and consumed in
 470                                 * userspace. */
 471
 472     /* Spanning tree. */
 473     struct stp *stp;
 474     long long int stp_last_tick;
 475
 476     /* VLAN splinters. */
 477     struct ovs_mutex vsp_mutex;
 478     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 479     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 480
 481     /* Ports. */
 482     struct sset ports;             /* Set of standard port names. */
 483     struct sset ghost_ports;       /* Ports with no datapath port. */
 484     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 485     int port_poll_errno;           /* Last errno for port_poll() reply. */
 486
 487     /* Per ofproto's dpif stats. */
 488     uint64_t n_hit;
 489     uint64_t n_missed;
 490
 491     /* Work queues. */
 492     struct ovs_mutex flow_mod_mutex;
 493     struct list flow_mods OVS_GUARDED;
 494     size_t n_flow_mods OVS_GUARDED;
 495
 496     struct ovs_mutex pin_mutex;
 497     struct list pins OVS_GUARDED;
 498     size_t n_pins OVS_GUARDED;
 499 };
 500
 501 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"?  (Useful only
 502  * for debugging the asynchronous flow_mod implementation.) */
 503 static bool clogged;
 504
 505 /* By default, flows in the datapath are wildcarded (megaflows).  They
 506  * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
 507 static bool enable_megaflows = true;
 508
 509 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 510 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 511
 512 static void ofproto_dpif_unixctl_init(void);
 513
 514 static inline struct ofproto_dpif *
 515 ofproto_dpif_cast(const struct ofproto *ofproto)
 516 {
 517     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 518     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 519 }
 520
 521 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
 522                                         ofp_port_t ofp_port);
 523 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
 524                           const struct ofpbuf *packet, struct ds *);
 525
 526 /* Upcalls. */
 527 static void handle_upcalls(struct dpif_backer *);
 528
 529 /* Flow expiration. */
 530 static int expire(struct dpif_backer *);
 531
 532 /* NetFlow. */
 533 static void send_netflow_active_timeouts(struct ofproto_dpif *);
 534
 535 /* Utilities. */
 536 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
 537
 538 /* Global variables. */
 539 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 540
 541 /* Initial mappings of port to bridge mappings. */
 542 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 543
 544 /* Executes and takes ownership of 'fm'. */
 545 void
 546 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 547                       struct ofputil_flow_mod *fm)
 548 {
 549     ovs_mutex_lock(&ofproto->flow_mod_mutex);
 550     if (ofproto->n_flow_mods > 1024) {
 551         ovs_mutex_unlock(&ofproto->flow_mod_mutex);
 552         COVERAGE_INC(flow_mod_overflow);
 553         free(fm->ofpacts);
 554         free(fm);
 555         return;
 556     }
 557
 558     list_push_back(&ofproto->flow_mods, &fm->list_node);
 559     ofproto->n_flow_mods++;
 560     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
 561 }
 562
 563 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 564  * Takes ownership of 'pin' and pin->packet. */
 565 void
 566 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 567                             struct ofputil_packet_in *pin)
 568 {
 569     ovs_mutex_lock(&ofproto->pin_mutex);
 570     if (ofproto->n_pins > 1024) {
 571         ovs_mutex_unlock(&ofproto->pin_mutex);
 572         COVERAGE_INC(packet_in_overflow);
 573         free(CONST_CAST(void *, pin->packet));
 574         free(pin);
 575         return;
 576     }
 577
 578     list_push_back(&ofproto->pins, &pin->list_node);
 579     ofproto->n_pins++;
 580     ovs_mutex_unlock(&ofproto->pin_mutex);
 581 }
 582 \f
 583 /* Factory functions. */
 584
 585 static void
 586 init(const struct shash *iface_hints)
 587 {
 588     struct shash_node *node;
 589
 590     /* Make a local copy, since we don't own 'iface_hints' elements. */
 591     SHASH_FOR_EACH(node, iface_hints) {
 592         const struct iface_hint *orig_hint = node->data;
 593         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 594
 595         new_hint->br_name = xstrdup(orig_hint->br_name);
 596         new_hint->br_type = xstrdup(orig_hint->br_type);
 597         new_hint->ofp_port = orig_hint->ofp_port;
 598
 599         shash_add(&init_ofp_ports, node->name, new_hint);
 600     }
 601 }
 602
 603 static void
 604 enumerate_types(struct sset *types)
 605 {
 606     dp_enumerate_types(types);
 607 }
 608
 609 static int
 610 enumerate_names(const char *type, struct sset *names)
 611 {
 612     struct ofproto_dpif *ofproto;
 613
 614     sset_clear(names);
 615     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 616         if (strcmp(type, ofproto->up.type)) {
 617             continue;
 618         }
 619         sset_add(names, ofproto->up.name);
 620     }
 621
 622     return 0;
 623 }
 624
 625 static int
 626 del(const char *type, const char *name)
 627 {
 628     struct dpif *dpif;
 629     int error;
 630
 631     error = dpif_open(name, type, &dpif);
 632     if (!error) {
 633         error = dpif_delete(dpif);
 634         dpif_close(dpif);
 635     }
 636     return error;
 637 }
 638 \f
 639 static const char *
 640 port_open_type(const char *datapath_type, const char *port_type)
 641 {
 642     return dpif_port_open_type(datapath_type, port_type);
 643 }
 644
 645 /* Type functions. */
 646
 647 static void process_dpif_port_changes(struct dpif_backer *);
 648 static void process_dpif_all_ports_changed(struct dpif_backer *);
 649 static void process_dpif_port_change(struct dpif_backer *,
 650                                      const char *devname);
 651 static void process_dpif_port_error(struct dpif_backer *, int error);
 652
 653 static struct ofproto_dpif *
 654 lookup_ofproto_dpif_by_port_name(const char *name)
 655 {
 656     struct ofproto_dpif *ofproto;
 657
 658     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 659         if (sset_contains(&ofproto->ports, name)) {
 660             return ofproto;
 661         }
 662     }
 663
 664     return NULL;
 665 }
 666
 667 static int
 668 type_run(const char *type)
 669 {
 670     static long long int push_timer = LLONG_MIN;
 671     struct dpif_backer *backer;
 672
 673     backer = shash_find_data(&all_dpif_backers, type);
 674     if (!backer) {
 675         /* This is not necessarily a problem, since backers are only
 676          * created on demand. */
 677         return 0;
 678     }
 679
 680     dpif_run(backer->dpif);
 681
 682     /* The most natural place to push facet statistics is when they're pulled
 683      * from the datapath.  However, when there are many flows in the datapath,
 684      * this expensive operation can occur so frequently, that it reduces our
 685      * ability to quickly set up flows.  To reduce the cost, we push statistics
 686      * here instead. */
 687     if (time_msec() > push_timer) {
 688         push_timer = time_msec() + 2000;
 689         push_all_stats();
 690     }
 691
 692     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 693      * and the configuration has now changed to "false", enable receiving
 694      * packets from the datapath. */
 695     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 696         int error;
 697
 698         backer->recv_set_enable = true;
 699
 700         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 701         if (error) {
 702             udpif_recv_set(backer->udpif, 0, false);
 703             VLOG_ERR("Failed to enable receiving packets in dpif.");
 704             return error;
 705         }
 706         udpif_recv_set(backer->udpif, n_handler_threads,
 707                        backer->recv_set_enable);
 708         dpif_flow_flush(backer->dpif);
 709         backer->need_revalidate = REV_RECONFIGURE;
 710     }
 711
 712     /* If the n_handler_threads is reconfigured, call udpif_recv_set()
 713      * to reset the handler threads. */
 714     if (backer->n_handler_threads != n_handler_threads) {
 715         udpif_recv_set(backer->udpif, n_handler_threads,
 716                        backer->recv_set_enable);
 717         backer->n_handler_threads = n_handler_threads;
 718     }
 719
 720     if (backer->need_revalidate) {
 721         struct ofproto_dpif *ofproto;
 722         struct simap_node *node;
 723         struct simap tmp_backers;
 724
 725         /* Handle tunnel garbage collection. */
 726         simap_init(&tmp_backers);
 727         simap_swap(&backer->tnl_backers, &tmp_backers);
 728
 729         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 730             struct ofport_dpif *iter;
 731
 732             if (backer != ofproto->backer) {
 733                 continue;
 734             }
 735
 736             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 737                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 738                 const char *dp_port;
 739
 740                 if (!iter->is_tunnel) {
 741                     continue;
 742                 }
 743
 744                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 745                                                      namebuf, sizeof namebuf);
 746                 node = simap_find(&tmp_backers, dp_port);
 747                 if (node) {
 748                     simap_put(&backer->tnl_backers, dp_port, node->data);
 749                     simap_delete(&tmp_backers, node);
 750                     node = simap_find(&backer->tnl_backers, dp_port);
 751                 } else {
 752                     node = simap_find(&backer->tnl_backers, dp_port);
 753                     if (!node) {
 754                         odp_port_t odp_port = ODPP_NONE;
 755
 756                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 757                                            &odp_port)) {
 758                             simap_put(&backer->tnl_backers, dp_port,
 759                                       odp_to_u32(odp_port));
 760                             node = simap_find(&backer->tnl_backers, dp_port);
 761                         }
 762                     }
 763                 }
 764
 765                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 766                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 767                                          iter->odp_port)) {
 768                     backer->need_revalidate = REV_RECONFIGURE;
 769                 }
 770             }
 771         }
 772
 773         SIMAP_FOR_EACH (node, &tmp_backers) {
 774             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 775         }
 776         simap_destroy(&tmp_backers);
 777
 778         switch (backer->need_revalidate) {
 779         case REV_RECONFIGURE:   COVERAGE_INC(rev_reconfigure);   break;
 780         case REV_STP:           COVERAGE_INC(rev_stp);           break;
 781         case REV_BOND:          COVERAGE_INC(rev_bond);          break;
 782         case REV_PORT_TOGGLED:  COVERAGE_INC(rev_port_toggled);  break;
 783         case REV_FLOW_TABLE:    COVERAGE_INC(rev_flow_table);    break;
 784         case REV_MAC_LEARNING:  COVERAGE_INC(rev_mac_learning);  break;
 785         case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
 786         }
 787         backer->need_revalidate = 0;
 788
 789         /* Clear the drop_keys in case we should now be accepting some
 790          * formerly dropped flows. */
 791         drop_key_clear(backer);
 792
 793         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 794             struct facet *facet, *next;
 795             struct ofport_dpif *ofport;
 796             struct cls_cursor cursor;
 797             struct ofbundle *bundle;
 798
 799             if (ofproto->backer != backer) {
 800                 continue;
 801             }
 802
 803             ovs_rwlock_wrlock(&xlate_rwlock);
 804             xlate_ofproto_set(ofproto, ofproto->up.name,
 805                               ofproto->backer->dpif, ofproto->miss_rule,
 806                               ofproto->no_packet_in_rule, ofproto->ml,
 807                               ofproto->stp, ofproto->mbridge,
 808                               ofproto->sflow, ofproto->ipfix,
 809                               ofproto->up.frag_handling,
 810                               ofproto->up.forward_bpdu,
 811                               connmgr_has_in_band(ofproto->up.connmgr),
 812                               ofproto->netflow != NULL);
 813
 814             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 815                 xlate_bundle_set(ofproto, bundle, bundle->name,
 816                                  bundle->vlan_mode, bundle->vlan,
 817                                  bundle->trunks, bundle->use_priority_tags,
 818                                  bundle->bond, bundle->lacp,
 819                                  bundle->floodable);
 820             }
 821
 822             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 823                 int stp_port = ofport->stp_port
 824                     ? stp_port_no(ofport->stp_port)
 825                     : 0;
 826                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 827                                  ofport->up.ofp_port, ofport->odp_port,
 828                                  ofport->up.netdev, ofport->cfm,
 829                                  ofport->bfd, ofport->peer, stp_port,
 830                                  ofport->qdscp, ofport->n_qdscp,
 831                                  ofport->up.pp.config, ofport->is_tunnel,
 832                                  ofport->may_enable);
 833             }
 834             ovs_rwlock_unlock(&xlate_rwlock);
 835
 836             /* Only ofproto-dpif cares about the facet classifier so we just
 837              * lock cls_cursor_init() to appease the thread safety analysis. */
 838             ovs_rwlock_rdlock(&ofproto->facets.rwlock);
 839             cls_cursor_init(&cursor, &ofproto->facets, NULL);
 840             ovs_rwlock_unlock(&ofproto->facets.rwlock);
 841             CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
 842                 facet_revalidate(facet);
 843                 run_fast_rl();
 844             }
 845         }
 846
 847         udpif_revalidate(backer->udpif);
 848     }
 849
 850     if (!backer->recv_set_enable) {
 851         /* Wake up before a max of 1000ms. */
 852         timer_set_duration(&backer->next_expiration, 1000);
 853     } else if (timer_expired(&backer->next_expiration)) {
 854         int delay = expire(backer);
 855         timer_set_duration(&backer->next_expiration, delay);
 856     }
 857
 858     process_dpif_port_changes(backer);
 859
 860     if (backer->governor) {
 861         size_t n_subfacets;
 862
 863         governor_run(backer->governor);
 864
 865         /* If the governor has shrunk to its minimum size and the number of
 866          * subfacets has dwindled, then drop the governor entirely.
 867          *
 868          * For hysteresis, the number of subfacets to drop the governor is
 869          * smaller than the number needed to trigger its creation. */
 870         n_subfacets = hmap_count(&backer->subfacets);
 871         if (n_subfacets * 4 < flow_eviction_threshold
 872             && governor_is_idle(backer->governor)) {
 873             governor_destroy(backer->governor);
 874             backer->governor = NULL;
 875         }
 876     }
 877
 878     return 0;
 879 }
 880
 881 /* Check for and handle port changes in 'backer''s dpif. */
 882 static void
 883 process_dpif_port_changes(struct dpif_backer *backer)
 884 {
 885     for (;;) {
 886         char *devname;
 887         int error;
 888
 889         error = dpif_port_poll(backer->dpif, &devname);
 890         switch (error) {
 891         case EAGAIN:
 892             return;
 893
 894         case ENOBUFS:
 895             process_dpif_all_ports_changed(backer);
 896             break;
 897
 898         case 0:
 899             process_dpif_port_change(backer, devname);
 900             free(devname);
 901             break;
 902
 903         default:
 904             process_dpif_port_error(backer, error);
 905             break;
 906         }
 907     }
 908 }
 909
 910 static void
 911 process_dpif_all_ports_changed(struct dpif_backer *backer)
 912 {
 913     struct ofproto_dpif *ofproto;
 914     struct dpif_port dpif_port;
 915     struct dpif_port_dump dump;
 916     struct sset devnames;
 917     const char *devname;
 918
 919     sset_init(&devnames);
 920     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 921         if (ofproto->backer == backer) {
 922             struct ofport *ofport;
 923
 924             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 925                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 926             }
 927         }
 928     }
 929     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 930         sset_add(&devnames, dpif_port.name);
 931     }
 932
 933     SSET_FOR_EACH (devname, &devnames) {
 934         process_dpif_port_change(backer, devname);
 935     }
 936     sset_destroy(&devnames);
 937 }
 938
 939 static void
 940 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 941 {
 942     struct ofproto_dpif *ofproto;
 943     struct dpif_port port;
 944
 945     /* Don't report on the datapath's device. */
 946     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 947         return;
 948     }
 949
 950     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 951                    &all_ofproto_dpifs) {
 952         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 953             return;
 954         }
 955     }
 956
 957     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 958     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 959         /* The port was removed.  If we know the datapath,
 960          * report it through poll_set().  If we don't, it may be
 961          * notifying us of a removal we initiated, so ignore it.
 962          * If there's a pending ENOBUFS, let it stand, since
 963          * everything will be reevaluated. */
 964         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 965             sset_add(&ofproto->port_poll_set, devname);
 966             ofproto->port_poll_errno = 0;
 967         }
 968     } else if (!ofproto) {
 969         /* The port was added, but we don't know with which
 970          * ofproto we should associate it.  Delete it. */
 971         dpif_port_del(backer->dpif, port.port_no);
 972     } else {
 973         struct ofport_dpif *ofport;
 974
 975         ofport = ofport_dpif_cast(shash_find_data(
 976                                       &ofproto->up.port_by_name, devname));
 977         if (ofport
 978             && ofport->odp_port != port.port_no
 979             && !odp_port_to_ofport(backer, port.port_no))
 980         {
 981             /* 'ofport''s datapath port number has changed from
 982              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 983              * structures to match. */
 984             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 985             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 986             ofport->odp_port = port.port_no;
 987             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 988                         hash_odp_port(port.port_no));
 989             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 990             backer->need_revalidate = REV_RECONFIGURE;
 991         }
 992     }
 993     dpif_port_destroy(&port);
 994 }
 995
 996 /* Propagate 'error' to all ofprotos based on 'backer'. */
 997 static void
 998 process_dpif_port_error(struct dpif_backer *backer, int error)
 999 {
1000     struct ofproto_dpif *ofproto;
1001
1002     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1003         if (ofproto->backer == backer) {
1004             sset_clear(&ofproto->port_poll_set);
1005             ofproto->port_poll_errno = error;
1006         }
1007     }
1008 }
1009
1010 static int
1011 dpif_backer_run_fast(struct dpif_backer *backer)
1012 {
1013     udpif_run(backer->udpif);
1014     handle_upcalls(backer);
1015
1016     return 0;
1017 }
1018
1019 static int
1020 type_run_fast(const char *type)
1021 {
1022     struct dpif_backer *backer;
1023
1024     backer = shash_find_data(&all_dpif_backers, type);
1025     if (!backer) {
1026         /* This is not necessarily a problem, since backers are only
1027          * created on demand. */
1028         return 0;
1029     }
1030
1031     return dpif_backer_run_fast(backer);
1032 }
1033
1034 static void
1035 run_fast_rl(void)
1036 {
1037     static long long int port_rl = LLONG_MIN;
1038
1039     if (time_msec() >= port_rl) {
1040         struct ofproto_dpif *ofproto;
1041
1042         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1043             run_fast(&ofproto->up);
1044         }
1045         port_rl = time_msec() + 200;
1046     }
1047 }
1048
1049 static void
1050 type_wait(const char *type)
1051 {
1052     struct dpif_backer *backer;
1053
1054     backer = shash_find_data(&all_dpif_backers, type);
1055     if (!backer) {
1056         /* This is not necessarily a problem, since backers are only
1057          * created on demand. */
1058         return;
1059     }
1060
1061     if (backer->governor) {
1062         governor_wait(backer->governor);
1063     }
1064
1065     timer_wait(&backer->next_expiration);
1066     dpif_wait(backer->dpif);
1067     udpif_wait(backer->udpif);
1068 }
1069 \f
1070 /* Basic life-cycle. */
1071
1072 static int add_internal_flows(struct ofproto_dpif *);
1073
1074 static struct ofproto *
1075 alloc(void)
1076 {
1077     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1078     return &ofproto->up;
1079 }
1080
1081 static void
1082 dealloc(struct ofproto *ofproto_)
1083 {
1084     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1085     free(ofproto);
1086 }
1087
1088 static void
1089 close_dpif_backer(struct dpif_backer *backer)
1090 {
1091     struct shash_node *node;
1092
1093     ovs_assert(backer->refcount > 0);
1094
1095     if (--backer->refcount) {
1096         return;
1097     }
1098
1099     drop_key_clear(backer);
1100     hmap_destroy(&backer->drop_keys);
1101
1102     simap_destroy(&backer->tnl_backers);
1103     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1104     hmap_destroy(&backer->odp_to_ofport_map);
1105     node = shash_find(&all_dpif_backers, backer->type);
1106     free(backer->type);
1107     shash_delete(&all_dpif_backers, node);
1108     udpif_destroy(backer->udpif);
1109     dpif_close(backer->dpif);
1110
1111     ovs_assert(hmap_is_empty(&backer->subfacets));
1112     hmap_destroy(&backer->subfacets);
1113     governor_destroy(backer->governor);
1114
1115     free(backer);
1116 }
1117
1118 /* Datapath port slated for removal from datapath. */
1119 struct odp_garbage {
1120     struct list list_node;
1121     odp_port_t odp_port;
1122 };
1123
1124 static int
1125 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1126 {
1127     struct dpif_backer *backer;
1128     struct dpif_port_dump port_dump;
1129     struct dpif_port port;
1130     struct shash_node *node;
1131     struct list garbage_list;
1132     struct odp_garbage *garbage, *next;
1133     struct sset names;
1134     char *backer_name;
1135     const char *name;
1136     int error;
1137
1138     backer = shash_find_data(&all_dpif_backers, type);
1139     if (backer) {
1140         backer->refcount++;
1141         *backerp = backer;
1142         return 0;
1143     }
1144
1145     backer_name = xasprintf("ovs-%s", type);
1146
1147     /* Remove any existing datapaths, since we assume we're the only
1148      * userspace controlling the datapath. */
1149     sset_init(&names);
1150     dp_enumerate_names(type, &names);
1151     SSET_FOR_EACH(name, &names) {
1152         struct dpif *old_dpif;
1153
1154         /* Don't remove our backer if it exists. */
1155         if (!strcmp(name, backer_name)) {
1156             continue;
1157         }
1158
1159         if (dpif_open(name, type, &old_dpif)) {
1160             VLOG_WARN("couldn't open old datapath %s to remove it", name);
1161         } else {
1162             dpif_delete(old_dpif);
1163             dpif_close(old_dpif);
1164         }
1165     }
1166     sset_destroy(&names);
1167
1168     backer = xmalloc(sizeof *backer);
1169
1170     error = dpif_create_and_open(backer_name, type, &backer->dpif);
1171     free(backer_name);
1172     if (error) {
1173         VLOG_ERR("failed to open datapath of type %s: %s", type,
1174                  ovs_strerror(error));
1175         free(backer);
1176         return error;
1177     }
1178     backer->udpif = udpif_create(backer, backer->dpif);
1179
1180     backer->type = xstrdup(type);
1181     backer->governor = NULL;
1182     backer->refcount = 1;
1183     hmap_init(&backer->odp_to_ofport_map);
1184     ovs_rwlock_init(&backer->odp_to_ofport_lock);
1185     hmap_init(&backer->drop_keys);
1186     hmap_init(&backer->subfacets);
1187     timer_set_duration(&backer->next_expiration, 1000);
1188     backer->need_revalidate = 0;
1189     simap_init(&backer->tnl_backers);
1190     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1191     *backerp = backer;
1192
1193     if (backer->recv_set_enable) {
1194         dpif_flow_flush(backer->dpif);
1195     }
1196
1197     /* Loop through the ports already on the datapath and remove any
1198      * that we don't need anymore. */
1199     list_init(&garbage_list);
1200     dpif_port_dump_start(&port_dump, backer->dpif);
1201     while (dpif_port_dump_next(&port_dump, &port)) {
1202         node = shash_find(&init_ofp_ports, port.name);
1203         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1204             garbage = xmalloc(sizeof *garbage);
1205             garbage->odp_port = port.port_no;
1206             list_push_front(&garbage_list, &garbage->list_node);
1207         }
1208     }
1209     dpif_port_dump_done(&port_dump);
1210
1211     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1212         dpif_port_del(backer->dpif, garbage->odp_port);
1213         list_remove(&garbage->list_node);
1214         free(garbage);
1215     }
1216
1217     shash_add(&all_dpif_backers, type, backer);
1218
1219     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1220     if (error) {
1221         VLOG_ERR("failed to listen on datapath of type %s: %s",
1222                  type, ovs_strerror(error));
1223         close_dpif_backer(backer);
1224         return error;
1225     }
1226     udpif_recv_set(backer->udpif, n_handler_threads,
1227                    backer->recv_set_enable);
1228     backer->n_handler_threads = n_handler_threads;
1229
1230     backer->max_n_subfacet = 0;
1231     backer->created = time_msec();
1232     backer->last_minute = backer->created;
1233     memset(&backer->hourly, 0, sizeof backer->hourly);
1234     memset(&backer->daily, 0, sizeof backer->daily);
1235     memset(&backer->lifetime, 0, sizeof backer->lifetime);
1236     backer->subfacet_add_count = 0;
1237     backer->subfacet_del_count = 0;
1238     backer->total_subfacet_add_count = 0;
1239     backer->total_subfacet_del_count = 0;
1240     backer->avg_n_subfacet = 0;
1241     backer->avg_subfacet_life = 0;
1242
1243     return error;
1244 }
1245
1246 static int
1247 construct(struct ofproto *ofproto_)
1248 {
1249     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1250     struct shash_node *node, *next;
1251     uint32_t max_ports;
1252     int error;
1253
1254     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1255     if (error) {
1256         return error;
1257     }
1258
1259     max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1260     ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX)));
1261
1262     ofproto->netflow = NULL;
1263     ofproto->sflow = NULL;
1264     ofproto->ipfix = NULL;
1265     ofproto->stp = NULL;
1266     hmap_init(&ofproto->bundles);
1267     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1268     ofproto->mbridge = mbridge_create();
1269     ofproto->has_bonded_bundles = false;
1270     ovs_mutex_init(&ofproto->vsp_mutex);
1271
1272     classifier_init(&ofproto->facets);
1273     ofproto->consistency_rl = LLONG_MIN;
1274
1275     list_init(&ofproto->completions);
1276
1277     ovs_mutex_init(&ofproto->flow_mod_mutex);
1278     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1279     list_init(&ofproto->flow_mods);
1280     ofproto->n_flow_mods = 0;
1281     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1282
1283     ovs_mutex_init(&ofproto->pin_mutex);
1284     ovs_mutex_lock(&ofproto->pin_mutex);
1285     list_init(&ofproto->pins);
1286     ofproto->n_pins = 0;
1287     ovs_mutex_unlock(&ofproto->pin_mutex);
1288
1289     ofproto_dpif_unixctl_init();
1290
1291     hmap_init(&ofproto->vlandev_map);
1292     hmap_init(&ofproto->realdev_vid_map);
1293
1294     sset_init(&ofproto->ports);
1295     sset_init(&ofproto->ghost_ports);
1296     sset_init(&ofproto->port_poll_set);
1297     ofproto->port_poll_errno = 0;
1298
1299     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1300         struct iface_hint *iface_hint = node->data;
1301
1302         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1303             /* Check if the datapath already has this port. */
1304             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1305                 sset_add(&ofproto->ports, node->name);
1306             }
1307
1308             free(iface_hint->br_name);
1309             free(iface_hint->br_type);
1310             free(iface_hint);
1311             shash_delete(&init_ofp_ports, node);
1312         }
1313     }
1314
1315     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1316                 hash_string(ofproto->up.name, 0));
1317     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1318
1319     ofproto_init_tables(ofproto_, N_TABLES);
1320     error = add_internal_flows(ofproto);
1321     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1322
1323     ofproto->n_hit = 0;
1324     ofproto->n_missed = 0;
1325
1326     return error;
1327 }
1328
1329 static int
1330 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1331                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1332 {
1333     struct ofputil_flow_mod fm;
1334     int error;
1335
1336     match_init_catchall(&fm.match);
1337     fm.priority = 0;
1338     match_set_reg(&fm.match, 0, id);
1339     fm.new_cookie = htonll(0);
1340     fm.cookie = htonll(0);
1341     fm.cookie_mask = htonll(0);
1342     fm.modify_cookie = false;
1343     fm.table_id = TBL_INTERNAL;
1344     fm.command = OFPFC_ADD;
1345     fm.idle_timeout = 0;
1346     fm.hard_timeout = 0;
1347     fm.buffer_id = 0;
1348     fm.out_port = 0;
1349     fm.flags = 0;
1350     fm.ofpacts = ofpacts->data;
1351     fm.ofpacts_len = ofpacts->size;
1352
1353     error = ofproto_flow_mod(&ofproto->up, &fm);
1354     if (error) {
1355         VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1356                     id, ofperr_to_string(error));
1357         return error;
1358     }
1359
1360     if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1361                                   rulep)) {
1362         ovs_rwlock_unlock(&(*rulep)->up.evict);
1363     } else {
1364         NOT_REACHED();
1365     }
1366
1367     return 0;
1368 }
1369
1370 static int
1371 add_internal_flows(struct ofproto_dpif *ofproto)
1372 {
1373     struct ofpact_controller *controller;
1374     uint64_t ofpacts_stub[128 / 8];
1375     struct ofpbuf ofpacts;
1376     int error;
1377     int id;
1378
1379     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1380     id = 1;
1381
1382     controller = ofpact_put_CONTROLLER(&ofpacts);
1383     controller->max_len = UINT16_MAX;
1384     controller->controller_id = 0;
1385     controller->reason = OFPR_NO_MATCH;
1386     ofpact_pad(&ofpacts);
1387
1388     error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1389     if (error) {
1390         return error;
1391     }
1392
1393     ofpbuf_clear(&ofpacts);
1394     error = add_internal_flow(ofproto, id++, &ofpacts,
1395                               &ofproto->no_packet_in_rule);
1396     if (error) {
1397         return error;
1398     }
1399
1400     error = add_internal_flow(ofproto, id++, &ofpacts,
1401                               &ofproto->drop_frags_rule);
1402     return error;
1403 }
1404
1405 static void
1406 complete_operations(struct ofproto_dpif *ofproto)
1407 {
1408     struct dpif_completion *c, *next;
1409
1410     LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1411         ofoperation_complete(c->op, 0);
1412         list_remove(&c->list_node);
1413         free(c);
1414     }
1415 }
1416
1417 static void
1418 destruct(struct ofproto *ofproto_)
1419 {
1420     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1421     struct rule_dpif *rule, *next_rule;
1422     struct ofputil_packet_in *pin, *next_pin;
1423     struct ofputil_flow_mod *fm, *next_fm;
1424     struct oftable *table;
1425
1426     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1427     ovs_rwlock_wrlock(&xlate_rwlock);
1428     xlate_remove_ofproto(ofproto);
1429     ovs_rwlock_unlock(&xlate_rwlock);
1430
1431     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1432     complete_operations(ofproto);
1433
1434     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1435         struct cls_cursor cursor;
1436
1437         ovs_rwlock_wrlock(&table->cls.rwlock);
1438         cls_cursor_init(&cursor, &table->cls, NULL);
1439         CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1440             ofproto_rule_delete(&ofproto->up, &table->cls, &rule->up);
1441         }
1442         ovs_rwlock_unlock(&table->cls.rwlock);
1443     }
1444     complete_operations(ofproto);
1445
1446     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1447     LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &ofproto->flow_mods) {
1448         list_remove(&fm->list_node);
1449         ofproto->n_flow_mods--;
1450         free(fm->ofpacts);
1451         free(fm);
1452     }
1453     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1454     ovs_mutex_destroy(&ofproto->flow_mod_mutex);
1455
1456     ovs_mutex_lock(&ofproto->pin_mutex);
1457     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &ofproto->pins) {
1458         list_remove(&pin->list_node);
1459         ofproto->n_pins--;
1460         free(CONST_CAST(void *, pin->packet));
1461         free(pin);
1462     }
1463     ovs_mutex_unlock(&ofproto->pin_mutex);
1464     ovs_mutex_destroy(&ofproto->pin_mutex);
1465
1466     mbridge_unref(ofproto->mbridge);
1467
1468     netflow_destroy(ofproto->netflow);
1469     dpif_sflow_unref(ofproto->sflow);
1470     hmap_destroy(&ofproto->bundles);
1471     mac_learning_unref(ofproto->ml);
1472
1473     classifier_destroy(&ofproto->facets);
1474
1475     hmap_destroy(&ofproto->vlandev_map);
1476     hmap_destroy(&ofproto->realdev_vid_map);
1477
1478     sset_destroy(&ofproto->ports);
1479     sset_destroy(&ofproto->ghost_ports);
1480     sset_destroy(&ofproto->port_poll_set);
1481
1482     ovs_mutex_destroy(&ofproto->vsp_mutex);
1483
1484     close_dpif_backer(ofproto->backer);
1485 }
1486
1487 static int
1488 run_fast(struct ofproto *ofproto_)
1489 {
1490     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1491     struct ofputil_packet_in *pin, *next_pin;
1492     struct ofputil_flow_mod *fm, *next_fm;
1493     struct list flow_mods, pins;
1494     struct ofport_dpif *ofport;
1495
1496     /* Do not perform any periodic activity required by 'ofproto' while
1497      * waiting for flow restore to complete. */
1498     if (ofproto_get_flow_restore_wait()) {
1499         return 0;
1500     }
1501
1502     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1503     if (ofproto->n_flow_mods) {
1504         flow_mods = ofproto->flow_mods;
1505         list_moved(&flow_mods);
1506         list_init(&ofproto->flow_mods);
1507         ofproto->n_flow_mods = 0;
1508     } else {
1509         list_init(&flow_mods);
1510     }
1511     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1512
1513     LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &flow_mods) {
1514         int error = ofproto_flow_mod(&ofproto->up, fm);
1515         if (error && !VLOG_DROP_WARN(&rl)) {
1516             VLOG_WARN("learning action failed to modify flow table (%s)",
1517                       ofperr_get_name(error));
1518         }
1519
1520         list_remove(&fm->list_node);
1521         free(fm->ofpacts);
1522         free(fm);
1523     }
1524
1525     ovs_mutex_lock(&ofproto->pin_mutex);
1526     if (ofproto->n_pins) {
1527         pins = ofproto->pins;
1528         list_moved(&pins);
1529         list_init(&ofproto->pins);
1530         ofproto->n_pins = 0;
1531     } else {
1532         list_init(&pins);
1533     }
1534     ovs_mutex_unlock(&ofproto->pin_mutex);
1535
1536     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1537         connmgr_send_packet_in(ofproto->up.connmgr, pin);
1538         list_remove(&pin->list_node);
1539         free(CONST_CAST(void *, pin->packet));
1540         free(pin);
1541     }
1542
1543     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1544         port_run_fast(ofport);
1545     }
1546
1547     return 0;
1548 }
1549
1550 static int
1551 run(struct ofproto *ofproto_)
1552 {
1553     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1554     struct ofport_dpif *ofport;
1555     struct ofbundle *bundle;
1556     int error;
1557
1558     if (!clogged) {
1559         complete_operations(ofproto);
1560     }
1561
1562     if (mbridge_need_revalidate(ofproto->mbridge)) {
1563         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1564         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1565         mac_learning_flush(ofproto->ml);
1566         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1567     }
1568
1569     /* Do not perform any periodic activity below required by 'ofproto' while
1570      * waiting for flow restore to complete. */
1571     if (ofproto_get_flow_restore_wait()) {
1572         return 0;
1573     }
1574
1575     error = run_fast(ofproto_);
1576     if (error) {
1577         return error;
1578     }
1579
1580     if (ofproto->netflow) {
1581         if (netflow_run(ofproto->netflow)) {
1582             send_netflow_active_timeouts(ofproto);
1583         }
1584     }
1585     if (ofproto->sflow) {
1586         dpif_sflow_run(ofproto->sflow);
1587     }
1588     if (ofproto->ipfix) {
1589         dpif_ipfix_run(ofproto->ipfix);
1590     }
1591
1592     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1593         port_run(ofport);
1594     }
1595     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1596         bundle_run(bundle);
1597     }
1598
1599     stp_run(ofproto);
1600     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1601     if (mac_learning_run(ofproto->ml)) {
1602         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1603     }
1604     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1605
1606     /* Check the consistency of a random facet, to aid debugging. */
1607     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1608     if (time_msec() >= ofproto->consistency_rl
1609         && !classifier_is_empty(&ofproto->facets)
1610         && !ofproto->backer->need_revalidate) {
1611         struct cls_table *table;
1612         struct cls_rule *cr;
1613         struct facet *facet;
1614
1615         ofproto->consistency_rl = time_msec() + 250;
1616
1617         table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1618                              struct cls_table, hmap_node);
1619         cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1620                           hmap_node);
1621         facet = CONTAINER_OF(cr, struct facet, cr);
1622
1623         if (!facet_check_consistency(facet)) {
1624             ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1625         }
1626     }
1627     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1628
1629     return 0;
1630 }
1631
1632 static void
1633 wait(struct ofproto *ofproto_)
1634 {
1635     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1636     struct ofport_dpif *ofport;
1637     struct ofbundle *bundle;
1638
1639     if (!clogged && !list_is_empty(&ofproto->completions)) {
1640         poll_immediate_wake();
1641     }
1642
1643     if (ofproto_get_flow_restore_wait()) {
1644         return;
1645     }
1646
1647     if (ofproto->sflow) {
1648         dpif_sflow_wait(ofproto->sflow);
1649     }
1650     if (ofproto->ipfix) {
1651         dpif_ipfix_wait(ofproto->ipfix);
1652     }
1653     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1654         port_wait(ofport);
1655     }
1656     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1657         bundle_wait(bundle);
1658     }
1659     if (ofproto->netflow) {
1660         netflow_wait(ofproto->netflow);
1661     }
1662     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1663     mac_learning_wait(ofproto->ml);
1664     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1665     stp_wait(ofproto);
1666     if (ofproto->backer->need_revalidate) {
1667         /* Shouldn't happen, but if it does just go around again. */
1668         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1669         poll_immediate_wake();
1670     }
1671 }
1672
1673 static void
1674 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1675 {
1676     const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1677     struct cls_cursor cursor;
1678     size_t n_subfacets = 0;
1679     struct facet *facet;
1680
1681     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1682     simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1683     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1684
1685     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1686     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1687     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1688         n_subfacets += list_size(&facet->subfacets);
1689     }
1690     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1691     simap_increase(usage, "subfacets", n_subfacets);
1692 }
1693
1694 static void
1695 flush(struct ofproto *ofproto_)
1696 {
1697     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1698     struct subfacet *subfacet, *next_subfacet;
1699     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1700     int n_batch;
1701
1702     n_batch = 0;
1703     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1704                         &ofproto->backer->subfacets) {
1705         if (subfacet->facet->ofproto != ofproto) {
1706             continue;
1707         }
1708
1709         if (subfacet->path != SF_NOT_INSTALLED) {
1710             batch[n_batch++] = subfacet;
1711             if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1712                 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1713                 n_batch = 0;
1714             }
1715         } else {
1716             subfacet_destroy(subfacet);
1717         }
1718     }
1719
1720     if (n_batch > 0) {
1721         subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1722     }
1723 }
1724
1725 static void
1726 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1727              bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1728 {
1729     *arp_match_ip = true;
1730     *actions = (OFPUTIL_A_OUTPUT |
1731                 OFPUTIL_A_SET_VLAN_VID |
1732                 OFPUTIL_A_SET_VLAN_PCP |
1733                 OFPUTIL_A_STRIP_VLAN |
1734                 OFPUTIL_A_SET_DL_SRC |
1735                 OFPUTIL_A_SET_DL_DST |
1736                 OFPUTIL_A_SET_NW_SRC |
1737                 OFPUTIL_A_SET_NW_DST |
1738                 OFPUTIL_A_SET_NW_TOS |
1739                 OFPUTIL_A_SET_TP_SRC |
1740                 OFPUTIL_A_SET_TP_DST |
1741                 OFPUTIL_A_ENQUEUE);
1742 }
1743
1744 static void
1745 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1746 {
1747     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1748     struct dpif_dp_stats s;
1749     uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1750     uint64_t n_lookup;
1751
1752     strcpy(ots->name, "classifier");
1753
1754     dpif_get_dp_stats(ofproto->backer->dpif, &s);
1755     rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1756     rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1757     rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1758
1759     n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1760     ots->lookup_count = htonll(n_lookup);
1761     ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1762 }
1763
1764 static struct ofport *
1765 port_alloc(void)
1766 {
1767     struct ofport_dpif *port = xmalloc(sizeof *port);
1768     return &port->up;
1769 }
1770
1771 static void
1772 port_dealloc(struct ofport *port_)
1773 {
1774     struct ofport_dpif *port = ofport_dpif_cast(port_);
1775     free(port);
1776 }
1777
1778 static int
1779 port_construct(struct ofport *port_)
1780 {
1781     struct ofport_dpif *port = ofport_dpif_cast(port_);
1782     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1783     const struct netdev *netdev = port->up.netdev;
1784     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1785     struct dpif_port dpif_port;
1786     int error;
1787
1788     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1789     port->bundle = NULL;
1790     port->cfm = NULL;
1791     port->bfd = NULL;
1792     port->may_enable = true;
1793     port->stp_port = NULL;
1794     port->stp_state = STP_DISABLED;
1795     port->is_tunnel = false;
1796     port->peer = NULL;
1797     port->qdscp = NULL;
1798     port->n_qdscp = 0;
1799     port->realdev_ofp_port = 0;
1800     port->vlandev_vid = 0;
1801     port->carrier_seq = netdev_get_carrier_resets(netdev);
1802
1803     if (netdev_vport_is_patch(netdev)) {
1804         /* By bailing out here, we don't submit the port to the sFlow module
1805          * to be considered for counter polling export.  This is correct
1806          * because the patch port represents an interface that sFlow considers
1807          * to be "internal" to the switch as a whole, and therefore not an
1808          * candidate for counter polling. */
1809         port->odp_port = ODPP_NONE;
1810         ofport_update_peer(port);
1811         return 0;
1812     }
1813
1814     error = dpif_port_query_by_name(ofproto->backer->dpif,
1815                                     netdev_vport_get_dpif_port(netdev, namebuf,
1816                                                                sizeof namebuf),
1817                                     &dpif_port);
1818     if (error) {
1819         return error;
1820     }
1821
1822     port->odp_port = dpif_port.port_no;
1823
1824     if (netdev_get_tunnel_config(netdev)) {
1825         tnl_port_add(port, port->up.netdev, port->odp_port);
1826         port->is_tunnel = true;
1827     } else {
1828         /* Sanity-check that a mapping doesn't already exist.  This
1829          * shouldn't happen for non-tunnel ports. */
1830         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1831             VLOG_ERR("port %s already has an OpenFlow port number",
1832                      dpif_port.name);
1833             dpif_port_destroy(&dpif_port);
1834             return EBUSY;
1835         }
1836
1837         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1838         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1839                     hash_odp_port(port->odp_port));
1840         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1841     }
1842     dpif_port_destroy(&dpif_port);
1843
1844     if (ofproto->sflow) {
1845         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1846     }
1847
1848     return 0;
1849 }
1850
1851 static void
1852 port_destruct(struct ofport *port_)
1853 {
1854     struct ofport_dpif *port = ofport_dpif_cast(port_);
1855     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1856     const char *devname = netdev_get_name(port->up.netdev);
1857     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1858     const char *dp_port_name;
1859
1860     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1861     ovs_rwlock_wrlock(&xlate_rwlock);
1862     xlate_ofport_remove(port);
1863     ovs_rwlock_unlock(&xlate_rwlock);
1864
1865     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1866                                               sizeof namebuf);
1867     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1868         /* The underlying device is still there, so delete it.  This
1869          * happens when the ofproto is being destroyed, since the caller
1870          * assumes that removal of attached ports will happen as part of
1871          * destruction. */
1872         if (!port->is_tunnel) {
1873             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1874         }
1875     }
1876
1877     if (port->peer) {
1878         port->peer->peer = NULL;
1879         port->peer = NULL;
1880     }
1881
1882     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1883         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1884         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1885         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1886     }
1887
1888     tnl_port_del(port);
1889     sset_find_and_delete(&ofproto->ports, devname);
1890     sset_find_and_delete(&ofproto->ghost_ports, devname);
1891     bundle_remove(port_);
1892     set_cfm(port_, NULL);
1893     set_bfd(port_, NULL);
1894     if (ofproto->sflow) {
1895         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1896     }
1897
1898     free(port->qdscp);
1899 }
1900
1901 static void
1902 port_modified(struct ofport *port_)
1903 {
1904     struct ofport_dpif *port = ofport_dpif_cast(port_);
1905
1906     if (port->bundle && port->bundle->bond) {
1907         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1908     }
1909
1910     if (port->cfm) {
1911         cfm_set_netdev(port->cfm, port->up.netdev);
1912     }
1913
1914     if (port->bfd) {
1915         bfd_set_netdev(port->bfd, port->up.netdev);
1916     }
1917
1918     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1919                                                 port->odp_port)) {
1920         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1921             REV_RECONFIGURE;
1922     }
1923
1924     ofport_update_peer(port);
1925 }
1926
1927 static void
1928 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1929 {
1930     struct ofport_dpif *port = ofport_dpif_cast(port_);
1931     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1932     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1933
1934     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1935                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1936                    OFPUTIL_PC_NO_PACKET_IN)) {
1937         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1938
1939         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1940             bundle_update(port->bundle);
1941         }
1942     }
1943 }
1944
1945 static int
1946 set_sflow(struct ofproto *ofproto_,
1947           const struct ofproto_sflow_options *sflow_options)
1948 {
1949     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1950     struct dpif_sflow *ds = ofproto->sflow;
1951
1952     if (sflow_options) {
1953         if (!ds) {
1954             struct ofport_dpif *ofport;
1955
1956             ds = ofproto->sflow = dpif_sflow_create();
1957             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1958                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1959             }
1960             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1961         }
1962         dpif_sflow_set_options(ds, sflow_options);
1963     } else {
1964         if (ds) {
1965             dpif_sflow_unref(ds);
1966             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1967             ofproto->sflow = NULL;
1968         }
1969     }
1970     return 0;
1971 }
1972
1973 static int
1974 set_ipfix(
1975     struct ofproto *ofproto_,
1976     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1977     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1978     size_t n_flow_exporters_options)
1979 {
1980     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1981     struct dpif_ipfix *di = ofproto->ipfix;
1982     bool has_options = bridge_exporter_options || flow_exporters_options;
1983
1984     if (has_options && !di) {
1985         di = ofproto->ipfix = dpif_ipfix_create();
1986     }
1987
1988     if (di) {
1989         /* Call set_options in any case to cleanly flush the flow
1990          * caches in the last exporters that are to be destroyed. */
1991         dpif_ipfix_set_options(
1992             di, bridge_exporter_options, flow_exporters_options,
1993             n_flow_exporters_options);
1994
1995         if (!has_options) {
1996             dpif_ipfix_unref(di);
1997             ofproto->ipfix = NULL;
1998         }
1999     }
2000
2001     return 0;
2002 }
2003
2004 static int
2005 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
2006 {
2007     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2008     int error;
2009
2010     if (!s) {
2011         error = 0;
2012     } else {
2013         if (!ofport->cfm) {
2014             struct ofproto_dpif *ofproto;
2015
2016             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2017             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2018             ofport->cfm = cfm_create(ofport->up.netdev);
2019         }
2020
2021         if (cfm_configure(ofport->cfm, s)) {
2022             return 0;
2023         }
2024
2025         error = EINVAL;
2026     }
2027     cfm_unref(ofport->cfm);
2028     ofport->cfm = NULL;
2029     return error;
2030 }
2031
2032 static bool
2033 get_cfm_status(const struct ofport *ofport_,
2034                struct ofproto_cfm_status *status)
2035 {
2036     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2037
2038     if (ofport->cfm) {
2039         status->faults = cfm_get_fault(ofport->cfm);
2040         status->remote_opstate = cfm_get_opup(ofport->cfm);
2041         status->health = cfm_get_health(ofport->cfm);
2042         cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2043         return true;
2044     } else {
2045         return false;
2046     }
2047 }
2048
2049 static int
2050 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2051 {
2052     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2053     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2054     struct bfd *old;
2055
2056     old = ofport->bfd;
2057     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
2058                                 cfg, ofport->up.netdev);
2059     if (ofport->bfd != old) {
2060         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2061     }
2062
2063     return 0;
2064 }
2065
2066 static int
2067 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2068 {
2069     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2070
2071     if (ofport->bfd) {
2072         bfd_get_status(ofport->bfd, smap);
2073         return 0;
2074     } else {
2075         return ENOENT;
2076     }
2077 }
2078 \f
2079 /* Spanning Tree. */
2080
2081 static void
2082 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2083 {
2084     struct ofproto_dpif *ofproto = ofproto_;
2085     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2086     struct ofport_dpif *ofport;
2087
2088     ofport = stp_port_get_aux(sp);
2089     if (!ofport) {
2090         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2091                      ofproto->up.name, port_num);
2092     } else {
2093         struct eth_header *eth = pkt->l2;
2094
2095         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2096         if (eth_addr_is_zero(eth->eth_src)) {
2097             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2098                          "with unknown MAC", ofproto->up.name, port_num);
2099         } else {
2100             send_packet(ofport, pkt);
2101         }
2102     }
2103     ofpbuf_delete(pkt);
2104 }
2105
2106 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2107 static int
2108 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2109 {
2110     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2111
2112     /* Only revalidate flows if the configuration changed. */
2113     if (!s != !ofproto->stp) {
2114         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2115     }
2116
2117     if (s) {
2118         if (!ofproto->stp) {
2119             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2120                                       send_bpdu_cb, ofproto);
2121             ofproto->stp_last_tick = time_msec();
2122         }
2123
2124         stp_set_bridge_id(ofproto->stp, s->system_id);
2125         stp_set_bridge_priority(ofproto->stp, s->priority);
2126         stp_set_hello_time(ofproto->stp, s->hello_time);
2127         stp_set_max_age(ofproto->stp, s->max_age);
2128         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2129     }  else {
2130         struct ofport *ofport;
2131
2132         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2133             set_stp_port(ofport, NULL);
2134         }
2135
2136         stp_unref(ofproto->stp);
2137         ofproto->stp = NULL;
2138     }
2139
2140     return 0;
2141 }
2142
2143 static int
2144 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2145 {
2146     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2147
2148     if (ofproto->stp) {
2149         s->enabled = true;
2150         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2151         s->designated_root = stp_get_designated_root(ofproto->stp);
2152         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2153     } else {
2154         s->enabled = false;
2155     }
2156
2157     return 0;
2158 }
2159
2160 static void
2161 update_stp_port_state(struct ofport_dpif *ofport)
2162 {
2163     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2164     enum stp_state state;
2165
2166     /* Figure out new state. */
2167     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2168                              : STP_DISABLED;
2169
2170     /* Update state. */
2171     if (ofport->stp_state != state) {
2172         enum ofputil_port_state of_state;
2173         bool fwd_change;
2174
2175         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2176                     netdev_get_name(ofport->up.netdev),
2177                     stp_state_name(ofport->stp_state),
2178                     stp_state_name(state));
2179         if (stp_learn_in_state(ofport->stp_state)
2180                 != stp_learn_in_state(state)) {
2181             /* xxx Learning action flows should also be flushed. */
2182             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2183             mac_learning_flush(ofproto->ml);
2184             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2185         }
2186         fwd_change = stp_forward_in_state(ofport->stp_state)
2187                         != stp_forward_in_state(state);
2188
2189         ofproto->backer->need_revalidate = REV_STP;
2190         ofport->stp_state = state;
2191         ofport->stp_state_entered = time_msec();
2192
2193         if (fwd_change && ofport->bundle) {
2194             bundle_update(ofport->bundle);
2195         }
2196
2197         /* Update the STP state bits in the OpenFlow port description. */
2198         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2199         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2200                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2201                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2202                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2203                      : 0);
2204         ofproto_port_set_state(&ofport->up, of_state);
2205     }
2206 }
2207
2208 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2209  * caller is responsible for assigning STP port numbers and ensuring
2210  * there are no duplicates. */
2211 static int
2212 set_stp_port(struct ofport *ofport_,
2213              const struct ofproto_port_stp_settings *s)
2214 {
2215     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2216     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2217     struct stp_port *sp = ofport->stp_port;
2218
2219     if (!s || !s->enable) {
2220         if (sp) {
2221             ofport->stp_port = NULL;
2222             stp_port_disable(sp);
2223             update_stp_port_state(ofport);
2224         }
2225         return 0;
2226     } else if (sp && stp_port_no(sp) != s->port_num
2227             && ofport == stp_port_get_aux(sp)) {
2228         /* The port-id changed, so disable the old one if it's not
2229          * already in use by another port. */
2230         stp_port_disable(sp);
2231     }
2232
2233     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2234     stp_port_enable(sp);
2235
2236     stp_port_set_aux(sp, ofport);
2237     stp_port_set_priority(sp, s->priority);
2238     stp_port_set_path_cost(sp, s->path_cost);
2239
2240     update_stp_port_state(ofport);
2241
2242     return 0;
2243 }
2244
2245 static int
2246 get_stp_port_status(struct ofport *ofport_,
2247                     struct ofproto_port_stp_status *s)
2248 {
2249     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2250     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2251     struct stp_port *sp = ofport->stp_port;
2252
2253     if (!ofproto->stp || !sp) {
2254         s->enabled = false;
2255         return 0;
2256     }
2257
2258     s->enabled = true;
2259     s->port_id = stp_port_get_id(sp);
2260     s->state = stp_port_get_state(sp);
2261     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2262     s->role = stp_port_get_role(sp);
2263     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2264
2265     return 0;
2266 }
2267
2268 static void
2269 stp_run(struct ofproto_dpif *ofproto)
2270 {
2271     if (ofproto->stp) {
2272         long long int now = time_msec();
2273         long long int elapsed = now - ofproto->stp_last_tick;
2274         struct stp_port *sp;
2275
2276         if (elapsed > 0) {
2277             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2278             ofproto->stp_last_tick = now;
2279         }
2280         while (stp_get_changed_port(ofproto->stp, &sp)) {
2281             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2282
2283             if (ofport) {
2284                 update_stp_port_state(ofport);
2285             }
2286         }
2287
2288         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2289             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2290             mac_learning_flush(ofproto->ml);
2291             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2292         }
2293     }
2294 }
2295
2296 static void
2297 stp_wait(struct ofproto_dpif *ofproto)
2298 {
2299     if (ofproto->stp) {
2300         poll_timer_wait(1000);
2301     }
2302 }
2303 \f
2304 static int
2305 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2306            size_t n_qdscp)
2307 {
2308     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2309     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2310
2311     if (ofport->n_qdscp != n_qdscp
2312         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2313                               n_qdscp * sizeof *qdscp))) {
2314         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2315         free(ofport->qdscp);
2316         ofport->qdscp = n_qdscp
2317             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2318             : NULL;
2319         ofport->n_qdscp = n_qdscp;
2320     }
2321
2322     return 0;
2323 }
2324 \f
2325 /* Bundles. */
2326
2327 /* Expires all MAC learning entries associated with 'bundle' and forces its
2328  * ofproto to revalidate every flow.
2329  *
2330  * Normally MAC learning entries are removed only from the ofproto associated
2331  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2332  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2333  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2334  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2335  * with the host from which it migrated. */
2336 static void
2337 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2338 {
2339     struct ofproto_dpif *ofproto = bundle->ofproto;
2340     struct mac_learning *ml = ofproto->ml;
2341     struct mac_entry *mac, *next_mac;
2342
2343     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2344     ovs_rwlock_wrlock(&ml->rwlock);
2345     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2346         if (mac->port.p == bundle) {
2347             if (all_ofprotos) {
2348                 struct ofproto_dpif *o;
2349
2350                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2351                     if (o != ofproto) {
2352                         struct mac_entry *e;
2353
2354                         ovs_rwlock_wrlock(&o->ml->rwlock);
2355                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2356                         if (e) {
2357                             mac_learning_expire(o->ml, e);
2358                         }
2359                         ovs_rwlock_unlock(&o->ml->rwlock);
2360                     }
2361                 }
2362             }
2363
2364             mac_learning_expire(ml, mac);
2365         }
2366     }
2367     ovs_rwlock_unlock(&ml->rwlock);
2368 }
2369
2370 static struct ofbundle *
2371 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2372 {
2373     struct ofbundle *bundle;
2374
2375     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2376                              &ofproto->bundles) {
2377         if (bundle->aux == aux) {
2378             return bundle;
2379         }
2380     }
2381     return NULL;
2382 }
2383
2384 static void
2385 bundle_update(struct ofbundle *bundle)
2386 {
2387     struct ofport_dpif *port;
2388
2389     bundle->floodable = true;
2390     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2391         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2392             || !stp_forward_in_state(port->stp_state)) {
2393             bundle->floodable = false;
2394             break;
2395         }
2396     }
2397 }
2398
2399 static void
2400 bundle_del_port(struct ofport_dpif *port)
2401 {
2402     struct ofbundle *bundle = port->bundle;
2403
2404     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2405
2406     list_remove(&port->bundle_node);
2407     port->bundle = NULL;
2408
2409     if (bundle->lacp) {
2410         lacp_slave_unregister(bundle->lacp, port);
2411     }
2412     if (bundle->bond) {
2413         bond_slave_unregister(bundle->bond, port);
2414     }
2415
2416     bundle_update(bundle);
2417 }
2418
2419 static bool
2420 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2421                 struct lacp_slave_settings *lacp)
2422 {
2423     struct ofport_dpif *port;
2424
2425     port = get_ofp_port(bundle->ofproto, ofp_port);
2426     if (!port) {
2427         return false;
2428     }
2429
2430     if (port->bundle != bundle) {
2431         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2432         if (port->bundle) {
2433             bundle_remove(&port->up);
2434         }
2435
2436         port->bundle = bundle;
2437         list_push_back(&bundle->ports, &port->bundle_node);
2438         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2439             || !stp_forward_in_state(port->stp_state)) {
2440             bundle->floodable = false;
2441         }
2442     }
2443     if (lacp) {
2444         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2445         lacp_slave_register(bundle->lacp, port, lacp);
2446     }
2447
2448     return true;
2449 }
2450
2451 static void
2452 bundle_destroy(struct ofbundle *bundle)
2453 {
2454     struct ofproto_dpif *ofproto;
2455     struct ofport_dpif *port, *next_port;
2456
2457     if (!bundle) {
2458         return;
2459     }
2460
2461     ofproto = bundle->ofproto;
2462     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2463
2464     ovs_rwlock_wrlock(&xlate_rwlock);
2465     xlate_bundle_remove(bundle);
2466     ovs_rwlock_unlock(&xlate_rwlock);
2467
2468     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2469         bundle_del_port(port);
2470     }
2471
2472     bundle_flush_macs(bundle, true);
2473     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2474     free(bundle->name);
2475     free(bundle->trunks);
2476     lacp_unref(bundle->lacp);
2477     bond_unref(bundle->bond);
2478     free(bundle);
2479 }
2480
2481 static int
2482 bundle_set(struct ofproto *ofproto_, void *aux,
2483            const struct ofproto_bundle_settings *s)
2484 {
2485     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2486     bool need_flush = false;
2487     struct ofport_dpif *port;
2488     struct ofbundle *bundle;
2489     unsigned long *trunks;
2490     int vlan;
2491     size_t i;
2492     bool ok;
2493
2494     if (!s) {
2495         bundle_destroy(bundle_lookup(ofproto, aux));
2496         return 0;
2497     }
2498
2499     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2500     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2501
2502     bundle = bundle_lookup(ofproto, aux);
2503     if (!bundle) {
2504         bundle = xmalloc(sizeof *bundle);
2505
2506         bundle->ofproto = ofproto;
2507         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2508                     hash_pointer(aux, 0));
2509         bundle->aux = aux;
2510         bundle->name = NULL;
2511
2512         list_init(&bundle->ports);
2513         bundle->vlan_mode = PORT_VLAN_TRUNK;
2514         bundle->vlan = -1;
2515         bundle->trunks = NULL;
2516         bundle->use_priority_tags = s->use_priority_tags;
2517         bundle->lacp = NULL;
2518         bundle->bond = NULL;
2519
2520         bundle->floodable = true;
2521         mbridge_register_bundle(ofproto->mbridge, bundle);
2522     }
2523
2524     if (!bundle->name || strcmp(s->name, bundle->name)) {
2525         free(bundle->name);
2526         bundle->name = xstrdup(s->name);
2527     }
2528
2529     /* LACP. */
2530     if (s->lacp) {
2531         if (!bundle->lacp) {
2532             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2533             bundle->lacp = lacp_create();
2534         }
2535         lacp_configure(bundle->lacp, s->lacp);
2536     } else {
2537         lacp_unref(bundle->lacp);
2538         bundle->lacp = NULL;
2539     }
2540
2541     /* Update set of ports. */
2542     ok = true;
2543     for (i = 0; i < s->n_slaves; i++) {
2544         if (!bundle_add_port(bundle, s->slaves[i],
2545                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2546             ok = false;
2547         }
2548     }
2549     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2550         struct ofport_dpif *next_port;
2551
2552         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2553             for (i = 0; i < s->n_slaves; i++) {
2554                 if (s->slaves[i] == port->up.ofp_port) {
2555                     goto found;
2556                 }
2557             }
2558
2559             bundle_del_port(port);
2560         found: ;
2561         }
2562     }
2563     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2564
2565     if (list_is_empty(&bundle->ports)) {
2566         bundle_destroy(bundle);
2567         return EINVAL;
2568     }
2569
2570     /* Set VLAN tagging mode */
2571     if (s->vlan_mode != bundle->vlan_mode
2572         || s->use_priority_tags != bundle->use_priority_tags) {
2573         bundle->vlan_mode = s->vlan_mode;
2574         bundle->use_priority_tags = s->use_priority_tags;
2575         need_flush = true;
2576     }
2577
2578     /* Set VLAN tag. */
2579     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2580             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2581             : 0);
2582     if (vlan != bundle->vlan) {
2583         bundle->vlan = vlan;
2584         need_flush = true;
2585     }
2586
2587     /* Get trunked VLANs. */
2588     switch (s->vlan_mode) {
2589     case PORT_VLAN_ACCESS:
2590         trunks = NULL;
2591         break;
2592
2593     case PORT_VLAN_TRUNK:
2594         trunks = CONST_CAST(unsigned long *, s->trunks);
2595         break;
2596
2597     case PORT_VLAN_NATIVE_UNTAGGED:
2598     case PORT_VLAN_NATIVE_TAGGED:
2599         if (vlan != 0 && (!s->trunks
2600                           || !bitmap_is_set(s->trunks, vlan)
2601                           || bitmap_is_set(s->trunks, 0))) {
2602             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2603             if (s->trunks) {
2604                 trunks = bitmap_clone(s->trunks, 4096);
2605             } else {
2606                 trunks = bitmap_allocate1(4096);
2607             }
2608             bitmap_set1(trunks, vlan);
2609             bitmap_set0(trunks, 0);
2610         } else {
2611             trunks = CONST_CAST(unsigned long *, s->trunks);
2612         }
2613         break;
2614
2615     default:
2616         NOT_REACHED();
2617     }
2618     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2619         free(bundle->trunks);
2620         if (trunks == s->trunks) {
2621             bundle->trunks = vlan_bitmap_clone(trunks);
2622         } else {
2623             bundle->trunks = trunks;
2624             trunks = NULL;
2625         }
2626         need_flush = true;
2627     }
2628     if (trunks != s->trunks) {
2629         free(trunks);
2630     }
2631
2632     /* Bonding. */
2633     if (!list_is_short(&bundle->ports)) {
2634         bundle->ofproto->has_bonded_bundles = true;
2635         if (bundle->bond) {
2636             if (bond_reconfigure(bundle->bond, s->bond)) {
2637                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2638             }
2639         } else {
2640             bundle->bond = bond_create(s->bond);
2641             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2642         }
2643
2644         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2645             bond_slave_register(bundle->bond, port, port->up.netdev);
2646         }
2647     } else {
2648         bond_unref(bundle->bond);
2649         bundle->bond = NULL;
2650     }
2651
2652     /* If we changed something that would affect MAC learning, un-learn
2653      * everything on this port and force flow revalidation. */
2654     if (need_flush) {
2655         bundle_flush_macs(bundle, false);
2656     }
2657
2658     return 0;
2659 }
2660
2661 static void
2662 bundle_remove(struct ofport *port_)
2663 {
2664     struct ofport_dpif *port = ofport_dpif_cast(port_);
2665     struct ofbundle *bundle = port->bundle;
2666
2667     if (bundle) {
2668         bundle_del_port(port);
2669         if (list_is_empty(&bundle->ports)) {
2670             bundle_destroy(bundle);
2671         } else if (list_is_short(&bundle->ports)) {
2672             bond_unref(bundle->bond);
2673             bundle->bond = NULL;
2674         }
2675     }
2676 }
2677
2678 static void
2679 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2680 {
2681     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2682     struct ofport_dpif *port = port_;
2683     uint8_t ea[ETH_ADDR_LEN];
2684     int error;
2685
2686     error = netdev_get_etheraddr(port->up.netdev, ea);
2687     if (!error) {
2688         struct ofpbuf packet;
2689         void *packet_pdu;
2690
2691         ofpbuf_init(&packet, 0);
2692         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2693                                  pdu_size);
2694         memcpy(packet_pdu, pdu, pdu_size);
2695
2696         send_packet(port, &packet);
2697         ofpbuf_uninit(&packet);
2698     } else {
2699         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2700                     "%s (%s)", port->bundle->name,
2701                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2702     }
2703 }
2704
2705 static void
2706 bundle_send_learning_packets(struct ofbundle *bundle)
2707 {
2708     struct ofproto_dpif *ofproto = bundle->ofproto;
2709     int error, n_packets, n_errors;
2710     struct mac_entry *e;
2711
2712     error = n_packets = n_errors = 0;
2713     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2714     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2715         if (e->port.p != bundle) {
2716             struct ofpbuf *learning_packet;
2717             struct ofport_dpif *port;
2718             void *port_void;
2719             int ret;
2720
2721             /* The assignment to "port" is unnecessary but makes "grep"ing for
2722              * struct ofport_dpif more effective. */
2723             learning_packet = bond_compose_learning_packet(bundle->bond,
2724                                                            e->mac, e->vlan,
2725                                                            &port_void);
2726             port = port_void;
2727             ret = send_packet(port, learning_packet);
2728             ofpbuf_delete(learning_packet);
2729             if (ret) {
2730                 error = ret;
2731                 n_errors++;
2732             }
2733             n_packets++;
2734         }
2735     }
2736     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2737
2738     if (n_errors) {
2739         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2740         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2741                      "packets, last error was: %s",
2742                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2743     } else {
2744         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2745                  bundle->name, n_packets);
2746     }
2747 }
2748
2749 static void
2750 bundle_run(struct ofbundle *bundle)
2751 {
2752     if (bundle->lacp) {
2753         lacp_run(bundle->lacp, send_pdu_cb);
2754     }
2755     if (bundle->bond) {
2756         struct ofport_dpif *port;
2757
2758         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2759             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2760         }
2761
2762         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2763             bundle->ofproto->backer->need_revalidate = REV_BOND;
2764         }
2765
2766         if (bond_should_send_learning_packets(bundle->bond)) {
2767             bundle_send_learning_packets(bundle);
2768         }
2769     }
2770 }
2771
2772 static void
2773 bundle_wait(struct ofbundle *bundle)
2774 {
2775     if (bundle->lacp) {
2776         lacp_wait(bundle->lacp);
2777     }
2778     if (bundle->bond) {
2779         bond_wait(bundle->bond);
2780     }
2781 }
2782 \f
2783 /* Mirrors. */
2784
2785 static int
2786 mirror_set__(struct ofproto *ofproto_, void *aux,
2787              const struct ofproto_mirror_settings *s)
2788 {
2789     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2790     struct ofbundle **srcs, **dsts;
2791     int error;
2792     size_t i;
2793
2794     if (!s) {
2795         mirror_destroy(ofproto->mbridge, aux);
2796         return 0;
2797     }
2798
2799     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2800     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2801
2802     for (i = 0; i < s->n_srcs; i++) {
2803         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2804     }
2805
2806     for (i = 0; i < s->n_dsts; i++) {
2807         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2808     }
2809
2810     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2811                        s->n_dsts, s->src_vlans,
2812                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2813     free(srcs);
2814     free(dsts);
2815     return error;
2816 }
2817
2818 static int
2819 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2820                    uint64_t *packets, uint64_t *bytes)
2821 {
2822     push_all_stats();
2823     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2824                             bytes);
2825 }
2826
2827 static int
2828 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2829 {
2830     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2831     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2832     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2833         mac_learning_flush(ofproto->ml);
2834     }
2835     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2836     return 0;
2837 }
2838
2839 static bool
2840 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2841 {
2842     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2843     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2844     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2845 }
2846
2847 static void
2848 forward_bpdu_changed(struct ofproto *ofproto_)
2849 {
2850     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2851     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2852 }
2853
2854 static void
2855 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2856                      size_t max_entries)
2857 {
2858     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2859     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2860     mac_learning_set_idle_time(ofproto->ml, idle_time);
2861     mac_learning_set_max_entries(ofproto->ml, max_entries);
2862     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2863 }
2864 \f
2865 /* Ports. */
2866
2867 static struct ofport_dpif *
2868 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2869 {
2870     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2871     return ofport ? ofport_dpif_cast(ofport) : NULL;
2872 }
2873
2874 static struct ofport_dpif *
2875 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2876 {
2877     struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2878     return port && &ofproto->up == port->up.ofproto ? port : NULL;
2879 }
2880
2881 static void
2882 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2883                             struct ofproto_port *ofproto_port,
2884                             struct dpif_port *dpif_port)
2885 {
2886     ofproto_port->name = dpif_port->name;
2887     ofproto_port->type = dpif_port->type;
2888     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2889 }
2890
2891 static void
2892 ofport_update_peer(struct ofport_dpif *ofport)
2893 {
2894     const struct ofproto_dpif *ofproto;
2895     struct dpif_backer *backer;
2896     char *peer_name;
2897
2898     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2899         return;
2900     }
2901
2902     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2903     backer->need_revalidate = REV_RECONFIGURE;
2904
2905     if (ofport->peer) {
2906         ofport->peer->peer = NULL;
2907         ofport->peer = NULL;
2908     }
2909
2910     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2911     if (!peer_name) {
2912         return;
2913     }
2914
2915     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2916         struct ofport *peer_ofport;
2917         struct ofport_dpif *peer;
2918         char *peer_peer;
2919
2920         if (ofproto->backer != backer) {
2921             continue;
2922         }
2923
2924         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2925         if (!peer_ofport) {
2926             continue;
2927         }
2928
2929         peer = ofport_dpif_cast(peer_ofport);
2930         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2931         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2932                                  peer_peer)) {
2933             ofport->peer = peer;
2934             ofport->peer->peer = ofport;
2935         }
2936         free(peer_peer);
2937
2938         break;
2939     }
2940     free(peer_name);
2941 }
2942
2943 static void
2944 port_run_fast(struct ofport_dpif *ofport)
2945 {
2946     if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2947         struct ofpbuf packet;
2948
2949         ofpbuf_init(&packet, 0);
2950         cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2951         send_packet(ofport, &packet);
2952         ofpbuf_uninit(&packet);
2953     }
2954
2955     if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2956         struct ofpbuf packet;
2957
2958         ofpbuf_init(&packet, 0);
2959         bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2960         send_packet(ofport, &packet);
2961         ofpbuf_uninit(&packet);
2962     }
2963 }
2964
2965 static void
2966 port_run(struct ofport_dpif *ofport)
2967 {
2968     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2969     bool carrier_changed = carrier_seq != ofport->carrier_seq;
2970     bool enable = netdev_get_carrier(ofport->up.netdev);
2971     bool cfm_enable = false;
2972     bool bfd_enable = false;
2973
2974     ofport->carrier_seq = carrier_seq;
2975
2976     port_run_fast(ofport);
2977
2978     if (ofport->cfm) {
2979         int cfm_opup = cfm_get_opup(ofport->cfm);
2980
2981         cfm_run(ofport->cfm);
2982         cfm_enable = !cfm_get_fault(ofport->cfm);
2983
2984         if (cfm_opup >= 0) {
2985             cfm_enable = cfm_enable && cfm_opup;
2986         }
2987     }
2988
2989     if (ofport->bfd) {
2990         bfd_run(ofport->bfd);
2991         bfd_enable = bfd_forwarding(ofport->bfd);
2992     }
2993
2994     if (ofport->bfd || ofport->cfm) {
2995         enable = enable && (cfm_enable || bfd_enable);
2996     }
2997
2998     if (ofport->bundle) {
2999         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3000         if (carrier_changed) {
3001             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3002         }
3003     }
3004
3005     if (ofport->may_enable != enable) {
3006         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3007         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3008     }
3009
3010     ofport->may_enable = enable;
3011 }
3012
3013 static void
3014 port_wait(struct ofport_dpif *ofport)
3015 {
3016     if (ofport->cfm) {
3017         cfm_wait(ofport->cfm);
3018     }
3019
3020     if (ofport->bfd) {
3021         bfd_wait(ofport->bfd);
3022     }
3023 }
3024
3025 static int
3026 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3027                    struct ofproto_port *ofproto_port)
3028 {
3029     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3030     struct dpif_port dpif_port;
3031     int error;
3032
3033     if (sset_contains(&ofproto->ghost_ports, devname)) {
3034         const char *type = netdev_get_type_from_name(devname);
3035
3036         /* We may be called before ofproto->up.port_by_name is populated with
3037          * the appropriate ofport.  For this reason, we must get the name and
3038          * type from the netdev layer directly. */
3039         if (type) {
3040             const struct ofport *ofport;
3041
3042             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3043             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3044             ofproto_port->name = xstrdup(devname);
3045             ofproto_port->type = xstrdup(type);
3046             return 0;
3047         }
3048         return ENODEV;
3049     }
3050
3051     if (!sset_contains(&ofproto->ports, devname)) {
3052         return ENODEV;
3053     }
3054     error = dpif_port_query_by_name(ofproto->backer->dpif,
3055                                     devname, &dpif_port);
3056     if (!error) {
3057         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3058     }
3059     return error;
3060 }
3061
3062 static int
3063 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3064 {
3065     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3066     const char *devname = netdev_get_name(netdev);
3067     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3068     const char *dp_port_name;
3069
3070     if (netdev_vport_is_patch(netdev)) {
3071         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3072         return 0;
3073     }
3074
3075     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3076     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3077         odp_port_t port_no = ODPP_NONE;
3078         int error;
3079
3080         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3081         if (error) {
3082             return error;
3083         }
3084         if (netdev_get_tunnel_config(netdev)) {
3085             simap_put(&ofproto->backer->tnl_backers,
3086                       dp_port_name, odp_to_u32(port_no));
3087         }
3088     }
3089
3090     if (netdev_get_tunnel_config(netdev)) {
3091         sset_add(&ofproto->ghost_ports, devname);
3092     } else {
3093         sset_add(&ofproto->ports, devname);
3094     }
3095     return 0;
3096 }
3097
3098 static int
3099 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3100 {
3101     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3102     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3103     int error = 0;
3104
3105     if (!ofport) {
3106         return 0;
3107     }
3108
3109     sset_find_and_delete(&ofproto->ghost_ports,
3110                          netdev_get_name(ofport->up.netdev));
3111     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3112     if (!ofport->is_tunnel) {
3113         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3114         if (!error) {
3115             /* The caller is going to close ofport->up.netdev.  If this is a
3116              * bonded port, then the bond is using that netdev, so remove it
3117              * from the bond.  The client will need to reconfigure everything
3118              * after deleting ports, so then the slave will get re-added. */
3119             bundle_remove(&ofport->up);
3120         }
3121     }
3122     return error;
3123 }
3124
3125 static int
3126 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3127 {
3128     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3129     int error;
3130
3131     push_all_stats();
3132
3133     error = netdev_get_stats(ofport->up.netdev, stats);
3134
3135     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3136         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3137
3138         /* ofproto->stats.tx_packets represents packets that we created
3139          * internally and sent to some port (e.g. packets sent with
3140          * send_packet()).  Account for them as if they had come from
3141          * OFPP_LOCAL and got forwarded. */
3142
3143         if (stats->rx_packets != UINT64_MAX) {
3144             stats->rx_packets += ofproto->stats.tx_packets;
3145         }
3146
3147         if (stats->rx_bytes != UINT64_MAX) {
3148             stats->rx_bytes += ofproto->stats.tx_bytes;
3149         }
3150
3151         /* ofproto->stats.rx_packets represents packets that were received on
3152          * some port and we processed internally and dropped (e.g. STP).
3153          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3154
3155         if (stats->tx_packets != UINT64_MAX) {
3156             stats->tx_packets += ofproto->stats.rx_packets;
3157         }
3158
3159         if (stats->tx_bytes != UINT64_MAX) {
3160             stats->tx_bytes += ofproto->stats.rx_bytes;
3161         }
3162     }
3163
3164     return error;
3165 }
3166
3167 struct port_dump_state {
3168     uint32_t bucket;
3169     uint32_t offset;
3170     bool ghost;
3171
3172     struct ofproto_port port;
3173     bool has_port;
3174 };
3175
3176 static int
3177 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3178 {
3179     *statep = xzalloc(sizeof(struct port_dump_state));
3180     return 0;
3181 }
3182
3183 static int
3184 port_dump_next(const struct ofproto *ofproto_, void *state_,
3185                struct ofproto_port *port)
3186 {
3187     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3188     struct port_dump_state *state = state_;
3189     const struct sset *sset;
3190     struct sset_node *node;
3191
3192     if (state->has_port) {
3193         ofproto_port_destroy(&state->port);
3194         state->has_port = false;
3195     }
3196     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3197     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3198         int error;
3199
3200         error = port_query_by_name(ofproto_, node->name, &state->port);
3201         if (!error) {
3202             *port = state->port;
3203             state->has_port = true;
3204             return 0;
3205         } else if (error != ENODEV) {
3206             return error;
3207         }
3208     }
3209
3210     if (!state->ghost) {
3211         state->ghost = true;
3212         state->bucket = 0;
3213         state->offset = 0;
3214         return port_dump_next(ofproto_, state_, port);
3215     }
3216
3217     return EOF;
3218 }
3219
3220 static int
3221 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3222 {
3223     struct port_dump_state *state = state_;
3224
3225     if (state->has_port) {
3226         ofproto_port_destroy(&state->port);
3227     }
3228     free(state);
3229     return 0;
3230 }
3231
3232 static int
3233 port_poll(const struct ofproto *ofproto_, char **devnamep)
3234 {
3235     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3236
3237     if (ofproto->port_poll_errno) {
3238         int error = ofproto->port_poll_errno;
3239         ofproto->port_poll_errno = 0;
3240         return error;
3241     }
3242
3243     if (sset_is_empty(&ofproto->port_poll_set)) {
3244         return EAGAIN;
3245     }
3246
3247     *devnamep = sset_pop(&ofproto->port_poll_set);
3248     return 0;
3249 }
3250
3251 static void
3252 port_poll_wait(const struct ofproto *ofproto_)
3253 {
3254     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3255     dpif_port_poll_wait(ofproto->backer->dpif);
3256 }
3257
3258 static int
3259 port_is_lacp_current(const struct ofport *ofport_)
3260 {
3261     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3262     return (ofport->bundle && ofport->bundle->lacp
3263             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3264             : -1);
3265 }
3266 \f
3267 /* Upcall handling. */
3268
3269 struct flow_miss_op {
3270     struct dpif_op dpif_op;
3271
3272     uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3273     struct xlate_out xout;
3274     bool xout_garbage;           /* 'xout' needs to be uninitialized? */
3275
3276     struct ofpbuf mask;          /* Flow mask for "put" ops. */
3277     struct odputil_keybuf maskbuf;
3278
3279     /* If this is a "put" op, then a pointer to the subfacet that should
3280      * be marked as uninstalled if the operation fails. */
3281     struct subfacet *subfacet;
3282 };
3283
3284 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3285  * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3286  * and (usually) installing a datapath flow.  The answer is usually "yes" (a
3287  * return value of true).  However, for short flows the cost of bookkeeping is
3288  * much higher than the benefits, so when the datapath holds a large number of
3289  * flows we impose some heuristics to decide which flows are likely to be worth
3290  * tracking. */
3291 static bool
3292 flow_miss_should_make_facet(struct flow_miss *miss)
3293 {
3294     struct dpif_backer *backer = miss->ofproto->backer;
3295     uint32_t hash;
3296
3297     switch (flow_miss_model) {
3298     case OFPROTO_HANDLE_MISS_AUTO:
3299         break;
3300     case OFPROTO_HANDLE_MISS_WITH_FACETS:
3301         return true;
3302     case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3303         return false;
3304     }
3305
3306     if (!backer->governor) {
3307         size_t n_subfacets;
3308
3309         n_subfacets = hmap_count(&backer->subfacets);
3310         if (n_subfacets * 2 <= flow_eviction_threshold) {
3311             return true;
3312         }
3313
3314         backer->governor = governor_create();
3315     }
3316
3317     hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3318     return governor_should_install_flow(backer->governor, hash,
3319                                         list_size(&miss->packets));
3320 }
3321
3322 /* Handles 'miss', which matches 'facet'.  May add any required datapath
3323  * operations to 'ops', incrementing '*n_ops' for each new op.
3324  *
3325  * All of the packets in 'miss' are considered to have arrived at time
3326  * 'miss->stats.used'.  This is really important only for new facets: if we
3327  * just called time_msec() here, then the new subfacet or its packets could
3328  * look (occasionally) as though it was used some time after the facet was
3329  * used.  That can make a one-packet flow look like it has a nonzero duration,
3330  * which looks odd in e.g. NetFlow statistics. */
3331 static void
3332 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3333                             struct flow_miss_op *ops, size_t *n_ops)
3334 {
3335     enum subfacet_path want_path;
3336     struct subfacet *subfacet;
3337
3338     facet->packet_count += miss->stats.n_packets;
3339     facet->prev_packet_count += miss->stats.n_packets;
3340     facet->byte_count += miss->stats.n_bytes;
3341     facet->prev_byte_count += miss->stats.n_bytes;
3342
3343     want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3344
3345     /* Don't install the flow if it's the result of the "userspace"
3346      * action for an already installed facet.  This can occur when a
3347      * datapath flow with wildcards has a "userspace" action and flows
3348      * sent to userspace result in a different subfacet, which will then
3349      * be rejected as overlapping by the datapath. */
3350     if (miss->upcall_type == DPIF_UC_ACTION
3351         && !list_is_empty(&facet->subfacets)) {
3352         return;
3353     }
3354
3355     subfacet = subfacet_create(facet, miss);
3356     if (subfacet->path != want_path) {
3357         struct flow_miss_op *op = &ops[(*n_ops)++];
3358         struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3359
3360         subfacet->path = want_path;
3361
3362         ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3363         if (enable_megaflows) {
3364             odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3365                                    &miss->flow, UINT32_MAX);
3366         }
3367
3368         op->xout_garbage = false;
3369         op->dpif_op.type = DPIF_OP_FLOW_PUT;
3370         op->subfacet = subfacet;
3371         put->flags = DPIF_FP_CREATE;
3372         put->key = miss->key;
3373         put->key_len = miss->key_len;
3374         put->mask = op->mask.data;
3375         put->mask_len = op->mask.size;
3376
3377         if (want_path == SF_FAST_PATH) {
3378             put->actions = facet->xout.odp_actions.data;
3379             put->actions_len = facet->xout.odp_actions.size;
3380         } else {
3381             compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3382                               op->slow_stub, sizeof op->slow_stub,
3383                               &put->actions, &put->actions_len);
3384         }
3385         put->stats = NULL;
3386     }
3387 }
3388
3389 /* Handles flow miss 'miss'.  May add any required datapath operations
3390  * to 'ops', incrementing '*n_ops' for each new op. */
3391 static void
3392 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3393                  size_t *n_ops)
3394 {
3395     struct facet *facet;
3396
3397     miss->ofproto->n_missed += list_size(&miss->packets);
3398
3399     facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3400     if (!facet) {
3401         /* There does not exist a bijection between 'struct flow' and datapath
3402          * flow keys with fitness ODP_FIT_TO_LITTLE.  This breaks a fundamental
3403          * assumption used throughout the facet and subfacet handling code.
3404          * Since we have to handle these misses in userspace anyway, we simply
3405          * skip facet creation, avoiding the problem altogether. */
3406         if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3407             || !flow_miss_should_make_facet(miss)) {
3408             return;
3409         }
3410
3411         facet = facet_create(miss);
3412     }
3413     handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3414 }
3415
3416 static struct drop_key *
3417 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3418                 size_t key_len)
3419 {
3420     struct drop_key *drop_key;
3421
3422     HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3423                              &backer->drop_keys) {
3424         if (drop_key->key_len == key_len
3425             && !memcmp(drop_key->key, key, key_len)) {
3426             return drop_key;
3427         }
3428     }
3429     return NULL;
3430 }
3431
3432 static void
3433 drop_key_clear(struct dpif_backer *backer)
3434 {
3435     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3436     struct drop_key *drop_key, *next;
3437
3438     HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3439         int error;
3440
3441         error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3442                               NULL);
3443         if (error && !VLOG_DROP_WARN(&rl)) {
3444             struct ds ds = DS_EMPTY_INITIALIZER;
3445             odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3446             VLOG_WARN("Failed to delete drop key (%s) (%s)",
3447                       ovs_strerror(error), ds_cstr(&ds));
3448             ds_destroy(&ds);
3449         }
3450
3451         hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3452         drop_key_destroy(drop_key);
3453     }
3454
3455     udpif_drop_key_clear(backer->udpif);
3456 }
3457
3458 static void
3459 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3460 {
3461     struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3462     struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3463     struct flow_miss *miss;
3464     size_t n_ops, i;
3465
3466     /* Process each element in the to-do list, constructing the set of
3467      * operations to batch. */
3468     n_ops = 0;
3469     HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3470         handle_flow_miss(miss, flow_miss_ops, &n_ops);
3471     }
3472     ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3473
3474     /* Execute batch. */
3475     for (i = 0; i < n_ops; i++) {
3476         dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3477     }
3478     dpif_operate(backer->dpif, dpif_ops, n_ops);
3479
3480     for (i = 0; i < n_ops; i++) {
3481         if (dpif_ops[i]->error != 0
3482             && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3483             && flow_miss_ops[i].subfacet) {
3484             struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3485
3486             COVERAGE_INC(subfacet_install_fail);
3487
3488             /* Zero-out subfacet counters when installation failed, but
3489              * datapath reported hits.  This should not happen and
3490              * indicates a bug, since if the datapath flow exists, we
3491              * should not be attempting to create a new subfacet.  A
3492              * buggy datapath could trigger this, so just zero out the
3493              * counters and log an error. */
3494             if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3495                 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3496                             "datapath reported hits");
3497                 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3498             }
3499
3500             subfacet->path = SF_NOT_INSTALLED;
3501         }
3502     }
3503 }
3504
3505 static void
3506 handle_sflow_upcall(struct dpif_backer *backer,
3507                     const struct dpif_upcall *upcall)
3508 {
3509     struct ofproto_dpif *ofproto;
3510     union user_action_cookie cookie;
3511     struct flow flow;
3512     odp_port_t odp_in_port;
3513
3514     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3515                       &flow, NULL, &ofproto, &odp_in_port)
3516         || !ofproto->sflow) {
3517         return;
3518     }
3519
3520     memset(&cookie, 0, sizeof cookie);
3521     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3522     dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3523                         odp_in_port, &cookie);
3524 }
3525
3526 static void
3527 handle_flow_sample_upcall(struct dpif_backer *backer,
3528                           const struct dpif_upcall *upcall)
3529 {
3530     struct ofproto_dpif *ofproto;
3531     union user_action_cookie cookie;
3532     struct flow flow;
3533
3534     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3535                       &flow, NULL, &ofproto, NULL)
3536         || !ofproto->ipfix) {
3537         return;
3538     }
3539
3540     memset(&cookie, 0, sizeof cookie);
3541     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3542
3543     /* The flow reflects exactly the contents of the packet.  Sample
3544      * the packet using it. */
3545     dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3546                            cookie.flow_sample.collector_set_id,
3547                            cookie.flow_sample.probability,
3548                            cookie.flow_sample.obs_domain_id,
3549                            cookie.flow_sample.obs_point_id);
3550 }
3551
3552 static void
3553 handle_ipfix_upcall(struct dpif_backer *backer,
3554                     const struct dpif_upcall *upcall)
3555 {
3556     struct ofproto_dpif *ofproto;
3557     struct flow flow;
3558
3559     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3560                       &flow, NULL, &ofproto, NULL)
3561         || !ofproto->ipfix) {
3562         return;
3563     }
3564
3565     /* The flow reflects exactly the contents of the packet.  Sample
3566      * the packet using it. */
3567     dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3568 }
3569
3570 static void
3571 handle_upcalls(struct dpif_backer *backer)
3572 {
3573     struct flow_miss_batch *fmb;
3574     int n_processed;
3575
3576     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3577         struct upcall *upcall = upcall_next(backer->udpif);
3578
3579         if (!upcall) {
3580             break;
3581         }
3582
3583         switch (upcall->type) {
3584         case SFLOW_UPCALL:
3585             handle_sflow_upcall(backer, &upcall->dpif_upcall);
3586             break;
3587
3588         case FLOW_SAMPLE_UPCALL:
3589             handle_flow_sample_upcall(backer, &upcall->dpif_upcall);
3590             break;
3591
3592         case IPFIX_UPCALL:
3593             handle_ipfix_upcall(backer, &upcall->dpif_upcall);
3594             break;
3595
3596         case BAD_UPCALL:
3597             break;
3598
3599         case MISS_UPCALL:
3600             NOT_REACHED();
3601         }
3602
3603         upcall_destroy(upcall);
3604     }
3605
3606     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3607         struct drop_key *drop_key = drop_key_next(backer->udpif);
3608         if (!drop_key) {
3609             break;
3610         }
3611
3612         if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3613             hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3614                         hash_bytes(drop_key->key, drop_key->key_len, 0));
3615             dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3616                           drop_key->key, drop_key->key_len,
3617                           NULL, 0, NULL, 0, NULL);
3618         } else {
3619             drop_key_destroy(drop_key);
3620         }
3621     }
3622
3623     fmb = flow_miss_batch_next(backer->udpif);
3624     if (fmb) {
3625         handle_flow_misses(backer, fmb);
3626         flow_miss_batch_destroy(fmb);
3627     }
3628 }
3629 \f
3630 /* Flow expiration. */
3631
3632 static int subfacet_max_idle(const struct dpif_backer *);
3633 static void update_stats(struct dpif_backer *);
3634 static void rule_expire(struct rule_dpif *);
3635 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3636
3637 /* This function is called periodically by run().  Its job is to collect
3638  * updates for the flows that have been installed into the datapath, most
3639  * importantly when they last were used, and then use that information to
3640  * expire flows that have not been used recently.
3641  *
3642  * Returns the number of milliseconds after which it should be called again. */
3643 static int
3644 expire(struct dpif_backer *backer)
3645 {
3646     struct ofproto_dpif *ofproto;
3647     size_t n_subfacets;
3648     int max_idle;
3649
3650     /* Periodically clear out the drop keys in an effort to keep them
3651      * relatively few. */
3652     drop_key_clear(backer);
3653
3654     /* Update stats for each flow in the backer. */
3655     update_stats(backer);
3656
3657     n_subfacets = hmap_count(&backer->subfacets);
3658     if (n_subfacets) {
3659         struct subfacet *subfacet;
3660         long long int total, now;
3661
3662         total = 0;
3663         now = time_msec();
3664         HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3665             total += now - subfacet->created;
3666         }
3667         backer->avg_subfacet_life += total / n_subfacets;
3668     }
3669     backer->avg_subfacet_life /= 2;
3670
3671     backer->avg_n_subfacet += n_subfacets;
3672     backer->avg_n_subfacet /= 2;
3673
3674     backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3675
3676     max_idle = subfacet_max_idle(backer);
3677     expire_subfacets(backer, max_idle);
3678
3679     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3680         struct rule *rule, *next_rule;
3681
3682         if (ofproto->backer != backer) {
3683             continue;
3684         }
3685
3686         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3687          * has passed. */
3688         ovs_mutex_lock(&ofproto->up.expirable_mutex);
3689         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3690                             &ofproto->up.expirable) {
3691             rule_expire(rule_dpif_cast(rule));
3692         }
3693         ovs_mutex_unlock(&ofproto->up.expirable_mutex);
3694
3695         /* All outstanding data in existing flows has been accounted, so it's a
3696          * good time to do bond rebalancing. */
3697         if (ofproto->has_bonded_bundles) {
3698             struct ofbundle *bundle;
3699
3700             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3701                 if (bundle->bond) {
3702                     bond_rebalance(bundle->bond);
3703                 }
3704             }
3705         }
3706     }
3707
3708     return MIN(max_idle, 1000);
3709 }
3710
3711 /* Updates flow table statistics given that the datapath just reported 'stats'
3712  * as 'subfacet''s statistics. */
3713 static void
3714 update_subfacet_stats(struct subfacet *subfacet,
3715                       const struct dpif_flow_stats *stats)
3716 {
3717     struct facet *facet = subfacet->facet;
3718     struct dpif_flow_stats diff;
3719
3720     diff.tcp_flags = stats->tcp_flags;
3721     diff.used = stats->used;
3722
3723     if (stats->n_packets >= subfacet->dp_packet_count) {
3724         diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3725     } else {
3726         VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3727         diff.n_packets = 0;
3728     }
3729
3730     if (stats->n_bytes >= subfacet->dp_byte_count) {
3731         diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3732     } else {
3733         VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3734         diff.n_bytes = 0;
3735     }
3736
3737     facet->ofproto->n_hit += diff.n_packets;
3738     subfacet->dp_packet_count = stats->n_packets;
3739     subfacet->dp_byte_count = stats->n_bytes;
3740     subfacet_update_stats(subfacet, &diff);
3741
3742     if (facet->accounted_bytes < facet->byte_count) {
3743         facet_learn(facet);
3744         facet_account(facet);
3745         facet->accounted_bytes = facet->byte_count;
3746     }
3747 }
3748
3749 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3750  * about, or a flow that shouldn't be installed but was anyway.  Delete it. */
3751 static void
3752 delete_unexpected_flow(struct dpif_backer *backer,
3753                        const struct nlattr *key, size_t key_len)
3754 {
3755     if (!VLOG_DROP_WARN(&rl)) {
3756         struct ds s;
3757
3758         ds_init(&s);
3759         odp_flow_key_format(key, key_len, &s);
3760         VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3761         ds_destroy(&s);
3762     }
3763
3764     COVERAGE_INC(facet_unexpected);
3765     dpif_flow_del(backer->dpif, key, key_len, NULL);
3766 }
3767
3768 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3769  *
3770  * This function also pushes statistics updates to rules which each facet
3771  * resubmits into.  Generally these statistics will be accurate.  However, if a
3772  * facet changes the rule it resubmits into at some time in between
3773  * update_stats() runs, it is possible that statistics accrued to the
3774  * old rule will be incorrectly attributed to the new rule.  This could be
3775  * avoided by calling update_stats() whenever rules are created or
3776  * deleted.  However, the performance impact of making so many calls to the
3777  * datapath do not justify the benefit of having perfectly accurate statistics.
3778  *
3779  * In addition, this function maintains per ofproto flow hit counts. The patch
3780  * port is not treated specially. e.g. A packet ingress from br0 patched into
3781  * br1 will increase the hit count of br0 by 1, however, does not affect
3782  * the hit or miss counts of br1.
3783  */
3784 static void
3785 update_stats(struct dpif_backer *backer)
3786 {
3787     const struct dpif_flow_stats *stats;
3788     struct dpif_flow_dump dump;
3789     const struct nlattr *key, *mask;
3790     size_t key_len, mask_len;
3791
3792     dpif_flow_dump_start(&dump, backer->dpif);
3793     while (dpif_flow_dump_next(&dump, &key, &key_len,
3794                                &mask, &mask_len, NULL, NULL, &stats)) {
3795         struct subfacet *subfacet;
3796         uint32_t key_hash;
3797
3798         key_hash = odp_flow_key_hash(key, key_len);
3799         subfacet = subfacet_find(backer, key, key_len, key_hash);
3800         switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3801         case SF_FAST_PATH:
3802             update_subfacet_stats(subfacet, stats);
3803             break;
3804
3805         case SF_SLOW_PATH:
3806             /* Stats are updated per-packet. */
3807             break;
3808
3809         case SF_NOT_INSTALLED:
3810         default:
3811             delete_unexpected_flow(backer, key, key_len);
3812             break;
3813         }
3814         run_fast_rl();
3815     }
3816     dpif_flow_dump_done(&dump);
3817
3818     update_moving_averages(backer);
3819 }
3820
3821 /* Calculates and returns the number of milliseconds of idle time after which
3822  * subfacets should expire from the datapath.  When a subfacet expires, we fold
3823  * its statistics into its facet, and when a facet's last subfacet expires, we
3824  * fold its statistic into its rule. */
3825 static int
3826 subfacet_max_idle(const struct dpif_backer *backer)
3827 {
3828     /*
3829      * Idle time histogram.
3830      *
3831      * Most of the time a switch has a relatively small number of subfacets.
3832      * When this is the case we might as well keep statistics for all of them
3833      * in userspace and to cache them in the kernel datapath for performance as
3834      * well.
3835      *
3836      * As the number of subfacets increases, the memory required to maintain
3837      * statistics about them in userspace and in the kernel becomes
3838      * significant.  However, with a large number of subfacets it is likely
3839      * that only a few of them are "heavy hitters" that consume a large amount
3840      * of bandwidth.  At this point, only heavy hitters are worth caching in
3841      * the kernel and maintaining in userspaces; other subfacets we can
3842      * discard.
3843      *
3844      * The technique used to compute the idle time is to build a histogram with
3845      * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each.  Each subfacet
3846      * that is installed in the kernel gets dropped in the appropriate bucket.
3847      * After the histogram has been built, we compute the cutoff so that only
3848      * the most-recently-used 1% of subfacets (but at least
3849      * flow_eviction_threshold flows) are kept cached.  At least
3850      * the most-recently-used bucket of subfacets is kept, so actually an
3851      * arbitrary number of subfacets can be kept in any given expiration run
3852      * (though the next run will delete most of those unless they receive
3853      * additional data).
3854      *
3855      * This requires a second pass through the subfacets, in addition to the
3856      * pass made by update_stats(), because the former function never looks at
3857      * uninstallable subfacets.
3858      */
3859     enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3860     enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3861     int buckets[N_BUCKETS] = { 0 };
3862     int total, subtotal, bucket;
3863     struct subfacet *subfacet;
3864     long long int now;
3865     int i;
3866
3867     total = hmap_count(&backer->subfacets);
3868     if (total <= flow_eviction_threshold) {
3869         return N_BUCKETS * BUCKET_WIDTH;
3870     }
3871
3872     /* Build histogram. */
3873     now = time_msec();
3874     HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3875         long long int idle = now - subfacet->used;
3876         int bucket = (idle <= 0 ? 0
3877                       : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3878                       : (unsigned int) idle / BUCKET_WIDTH);
3879         buckets[bucket]++;
3880     }
3881
3882     /* Find the first bucket whose flows should be expired. */
3883     subtotal = bucket = 0;
3884     do {
3885         subtotal += buckets[bucket++];
3886     } while (bucket < N_BUCKETS &&
3887              subtotal < MAX(flow_eviction_threshold, total / 100));
3888
3889     if (VLOG_IS_DBG_ENABLED()) {
3890         struct ds s;
3891
3892         ds_init(&s);
3893         ds_put_cstr(&s, "keep");
3894         for (i = 0; i < N_BUCKETS; i++) {
3895             if (i == bucket) {
3896                 ds_put_cstr(&s, ", drop");
3897             }
3898             if (buckets[i]) {
3899                 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3900             }
3901         }
3902         VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3903         ds_destroy(&s);
3904     }
3905
3906     return bucket * BUCKET_WIDTH;
3907 }
3908
3909 static void
3910 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3911 {
3912     /* Cutoff time for most flows. */
3913     long long int normal_cutoff = time_msec() - dp_max_idle;
3914
3915     /* We really want to keep flows for special protocols around, so use a more
3916      * conservative cutoff. */
3917     long long int special_cutoff = time_msec() - 10000;
3918
3919     struct subfacet *subfacet, *next_subfacet;
3920     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3921     int n_batch;
3922
3923     n_batch = 0;
3924     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3925                         &backer->subfacets) {
3926         long long int cutoff;
3927
3928         cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3929                                                 | SLOW_STP)
3930                   ? special_cutoff
3931                   : normal_cutoff);
3932         if (subfacet->used < cutoff) {
3933             if (subfacet->path != SF_NOT_INSTALLED) {
3934                 batch[n_batch++] = subfacet;
3935                 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3936                     subfacet_destroy_batch(backer, batch, n_batch);
3937                     n_batch = 0;
3938                 }
3939             } else {
3940                 subfacet_destroy(subfacet);
3941             }
3942         }
3943     }
3944
3945     if (n_batch > 0) {
3946         subfacet_destroy_batch(backer, batch, n_batch);
3947     }
3948 }
3949
3950 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3951  * then delete it entirely. */
3952 static void
3953 rule_expire(struct rule_dpif *rule)
3954 {
3955     uint16_t idle_timeout, hard_timeout;
3956     long long int now;
3957     uint8_t reason;
3958
3959     if (rule->up.pending) {
3960         /* We'll have to expire it later. */
3961         return;
3962     }
3963
3964     ovs_mutex_lock(&rule->up.timeout_mutex);
3965     hard_timeout = rule->up.hard_timeout;
3966     idle_timeout = rule->up.idle_timeout;
3967     ovs_mutex_unlock(&rule->up.timeout_mutex);
3968
3969     /* Has 'rule' expired? */
3970     now = time_msec();
3971     if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3972         reason = OFPRR_HARD_TIMEOUT;
3973     } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3974         reason = OFPRR_IDLE_TIMEOUT;
3975     } else {
3976         return;
3977     }
3978
3979     COVERAGE_INC(ofproto_dpif_expired);
3980     ofproto_rule_expire(&rule->up, reason);
3981 }
3982 \f
3983 /* Facets. */
3984
3985 /* Creates and returns a new facet based on 'miss'.
3986  *
3987  * The caller must already have determined that no facet with an identical
3988  * 'miss->flow' exists in 'miss->ofproto'.
3989  *
3990  * 'rule' and 'xout' must have been created based on 'miss'.
3991  *
3992  * 'facet'' statistics are initialized based on 'stats'.
3993  *
3994  * The facet will initially have no subfacets.  The caller should create (at
3995  * least) one subfacet with subfacet_create(). */
3996 static struct facet *
3997 facet_create(const struct flow_miss *miss)
3998 {
3999     struct ofproto_dpif *ofproto = miss->ofproto;
4000     struct facet *facet;
4001     struct match match;
4002
4003     facet = xzalloc(sizeof *facet);
4004     facet->ofproto = miss->ofproto;
4005     facet->used = miss->stats.used;
4006     facet->flow = miss->flow;
4007     facet->learn_rl = time_msec() + 500;
4008
4009     list_init(&facet->subfacets);
4010     netflow_flow_init(&facet->nf_flow);
4011     netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4012
4013     xlate_out_copy(&facet->xout, &miss->xout);
4014
4015     match_init(&match, &facet->flow, &facet->xout.wc);
4016     cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
4017     ovs_rwlock_wrlock(&ofproto->facets.rwlock);
4018     classifier_insert(&ofproto->facets, &facet->cr);
4019     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4020
4021     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4022     return facet;
4023 }
4024
4025 static void
4026 facet_free(struct facet *facet)
4027 {
4028     if (facet) {
4029         xlate_out_uninit(&facet->xout);
4030         free(facet);
4031     }
4032 }
4033
4034 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4035  * 'packet', which arrived on 'in_port'. */
4036 static bool
4037 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4038                     const struct nlattr *odp_actions, size_t actions_len,
4039                     struct ofpbuf *packet)
4040 {
4041     struct odputil_keybuf keybuf;
4042     struct ofpbuf key;
4043     int error;
4044
4045     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4046     odp_flow_key_from_flow(&key, flow,
4047                            ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4048
4049     error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4050                          odp_actions, actions_len, packet);
4051     return !error;
4052 }
4053
4054 /* Remove 'facet' from its ofproto and free up the associated memory:
4055  *
4056  *   - If 'facet' was installed in the datapath, uninstalls it and updates its
4057  *     rule's statistics, via subfacet_uninstall().
4058  *
4059  *   - Removes 'facet' from its rule and from ofproto->facets.
4060  */
4061 static void
4062 facet_remove(struct facet *facet)
4063 {
4064     struct subfacet *subfacet, *next_subfacet;
4065
4066     ovs_assert(!list_is_empty(&facet->subfacets));
4067
4068     /* First uninstall all of the subfacets to get final statistics. */
4069     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4070         subfacet_uninstall(subfacet);
4071     }
4072
4073     /* Flush the final stats to the rule.
4074      *
4075      * This might require us to have at least one subfacet around so that we
4076      * can use its actions for accounting in facet_account(), which is why we
4077      * have uninstalled but not yet destroyed the subfacets. */
4078     facet_flush_stats(facet);
4079
4080     /* Now we're really all done so destroy everything. */
4081     LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4082                         &facet->subfacets) {
4083         subfacet_destroy__(subfacet);
4084     }
4085     ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4086     classifier_remove(&facet->ofproto->facets, &facet->cr);
4087     ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4088     cls_rule_destroy(&facet->cr);
4089     facet_free(facet);
4090 }
4091
4092 /* Feed information from 'facet' back into the learning table to keep it in
4093  * sync with what is actually flowing through the datapath. */
4094 static void
4095 facet_learn(struct facet *facet)
4096 {
4097     long long int now = time_msec();
4098
4099     if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4100         return;
4101     }
4102
4103     facet->learn_rl = now + 500;
4104
4105     if (!facet->xout.has_learn
4106         && !facet->xout.has_normal
4107         && (!facet->xout.has_fin_timeout
4108             || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4109         return;
4110     }
4111
4112     facet_push_stats(facet, true);
4113 }
4114
4115 static void
4116 facet_account(struct facet *facet)
4117 {
4118     const struct nlattr *a;
4119     unsigned int left;
4120     ovs_be16 vlan_tci;
4121     uint64_t n_bytes;
4122
4123     if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4124         return;
4125     }
4126     n_bytes = facet->byte_count - facet->accounted_bytes;
4127
4128     /* This loop feeds byte counters to bond_account() for rebalancing to use
4129      * as a basis.  We also need to track the actual VLAN on which the packet
4130      * is going to be sent to ensure that it matches the one passed to
4131      * bond_choose_output_slave().  (Otherwise, we will account to the wrong
4132      * hash bucket.)
4133      *
4134      * We use the actions from an arbitrary subfacet because they should all
4135      * be equally valid for our purpose. */
4136     vlan_tci = facet->flow.vlan_tci;
4137     NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4138                              facet->xout.odp_actions.size) {
4139         const struct ovs_action_push_vlan *vlan;
4140         struct ofport_dpif *port;
4141
4142         switch (nl_attr_type(a)) {
4143         case OVS_ACTION_ATTR_OUTPUT:
4144             port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4145             if (port && port->bundle && port->bundle->bond) {
4146                 bond_account(port->bundle->bond, &facet->flow,
4147                              vlan_tci_to_vid(vlan_tci), n_bytes);
4148             }
4149             break;
4150
4151         case OVS_ACTION_ATTR_POP_VLAN:
4152             vlan_tci = htons(0);
4153             break;
4154
4155         case OVS_ACTION_ATTR_PUSH_VLAN:
4156             vlan = nl_attr_get(a);
4157             vlan_tci = vlan->vlan_tci;
4158             break;
4159         }
4160     }
4161 }
4162
4163 /* Returns true if the only action for 'facet' is to send to the controller.
4164  * (We don't report NetFlow expiration messages for such facets because they
4165  * are just part of the control logic for the network, not real traffic). */
4166 static bool
4167 facet_is_controller_flow(struct facet *facet)
4168 {
4169     if (facet) {
4170         struct ofproto_dpif *ofproto = facet->ofproto;
4171         const struct ofpact *ofpacts;
4172         struct rule_dpif *rule;
4173         size_t ofpacts_len;
4174         bool is_controller;
4175
4176         rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4177         ofpacts_len = rule->up.ofpacts_len;
4178         ofpacts = rule->up.ofpacts;
4179         is_controller = ofpacts_len > 0
4180             && ofpacts->type == OFPACT_CONTROLLER
4181             && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4182         rule_release(rule);
4183         return is_controller;
4184     }
4185     return false;
4186 }
4187
4188 /* Folds all of 'facet''s statistics into its rule.  Also updates the
4189  * accounting ofhook and emits a NetFlow expiration if appropriate.  All of
4190  * 'facet''s statistics in the datapath should have been zeroed and folded into
4191  * its packet and byte counts before this function is called. */
4192 static void
4193 facet_flush_stats(struct facet *facet)
4194 {
4195     struct ofproto_dpif *ofproto = facet->ofproto;
4196     struct subfacet *subfacet;
4197
4198     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4199         ovs_assert(!subfacet->dp_byte_count);
4200         ovs_assert(!subfacet->dp_packet_count);
4201     }
4202
4203     facet_push_stats(facet, false);
4204     if (facet->accounted_bytes < facet->byte_count) {
4205         facet_account(facet);
4206         facet->accounted_bytes = facet->byte_count;
4207     }
4208
4209     if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4210         struct ofexpired expired;
4211         expired.flow = facet->flow;
4212         expired.packet_count = facet->packet_count;
4213         expired.byte_count = facet->byte_count;
4214         expired.used = facet->used;
4215         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4216     }
4217
4218     /* Reset counters to prevent double counting if 'facet' ever gets
4219      * reinstalled. */
4220     facet_reset_counters(facet);
4221
4222     netflow_flow_clear(&facet->nf_flow);
4223     facet->tcp_flags = 0;
4224 }
4225
4226 /* Searches 'ofproto''s table of facets for one which would be responsible for
4227  * 'flow'.  Returns it if found, otherwise a null pointer.
4228  *
4229  * The returned facet might need revalidation; use facet_lookup_valid()
4230  * instead if that is important. */
4231 static struct facet *
4232 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4233 {
4234     struct cls_rule *cr;
4235
4236     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4237     cr = classifier_lookup(&ofproto->facets, flow, NULL);
4238     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4239     return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4240 }
4241
4242 /* Searches 'ofproto''s table of facets for one capable that covers
4243  * 'flow'.  Returns it if found, otherwise a null pointer.
4244  *
4245  * The returned facet is guaranteed to be valid. */
4246 static struct facet *
4247 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4248 {
4249     struct facet *facet;
4250
4251     facet = facet_find(ofproto, flow);
4252     if (facet
4253         && ofproto->backer->need_revalidate
4254         && !facet_revalidate(facet)) {
4255         return NULL;
4256     }
4257
4258     return facet;
4259 }
4260
4261 static bool
4262 facet_check_consistency(struct facet *facet)
4263 {
4264     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4265
4266     struct xlate_out xout;
4267     struct xlate_in xin;
4268
4269     struct rule_dpif *rule;
4270     bool ok;
4271
4272     /* Check the datapath actions for consistency. */
4273     rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4274     xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4275     xlate_actions(&xin, &xout);
4276     rule_release(rule);
4277
4278     ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4279         && facet->xout.slow == xout.slow;
4280     if (!ok && !VLOG_DROP_WARN(&rl)) {
4281         struct ds s = DS_EMPTY_INITIALIZER;
4282
4283         flow_format(&s, &facet->flow);
4284         ds_put_cstr(&s, ": inconsistency in facet");
4285
4286         if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4287             ds_put_cstr(&s, " (actions were: ");
4288             format_odp_actions(&s, facet->xout.odp_actions.data,
4289                                facet->xout.odp_actions.size);
4290             ds_put_cstr(&s, ") (correct actions: ");
4291             format_odp_actions(&s, xout.odp_actions.data,
4292                                xout.odp_actions.size);
4293             ds_put_char(&s, ')');
4294         }
4295
4296         if (facet->xout.slow != xout.slow) {
4297             ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4298         }
4299
4300         ds_destroy(&s);
4301     }
4302     xlate_out_uninit(&xout);
4303
4304     return ok;
4305 }
4306
4307 /* Re-searches the classifier for 'facet':
4308  *
4309  *   - If the rule found is different from 'facet''s current rule, moves
4310  *     'facet' to the new rule and recompiles its actions.
4311  *
4312  *   - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4313  *     where it is and recompiles its actions anyway.
4314  *
4315  *   - If any of 'facet''s subfacets correspond to a new flow according to
4316  *     xlate_receive(), 'facet' is removed.
4317  *
4318  *   Returns true if 'facet' is still valid.  False if 'facet' was removed. */
4319 static bool
4320 facet_revalidate(struct facet *facet)
4321 {
4322     struct ofproto_dpif *ofproto = facet->ofproto;
4323     struct rule_dpif *new_rule;
4324     struct subfacet *subfacet;
4325     struct flow_wildcards wc;
4326     struct xlate_out xout;
4327     struct xlate_in xin;
4328
4329     COVERAGE_INC(facet_revalidate);
4330
4331     /* Check that child subfacets still correspond to this facet.  Tunnel
4332      * configuration changes could cause a subfacet's OpenFlow in_port to
4333      * change. */
4334     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4335         struct ofproto_dpif *recv_ofproto;
4336         struct flow recv_flow;
4337         int error;
4338
4339         error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4340                               subfacet->key_len, &recv_flow, NULL,
4341                               &recv_ofproto, NULL);
4342         if (error
4343             || recv_ofproto != ofproto
4344             || facet != facet_find(ofproto, &recv_flow)) {
4345             facet_remove(facet);
4346             return false;
4347         }
4348     }
4349
4350     flow_wildcards_init_catchall(&wc);
4351     rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4352
4353     /* Calculate new datapath actions.
4354      *
4355      * We do not modify any 'facet' state yet, because we might need to, e.g.,
4356      * emit a NetFlow expiration and, if so, we need to have the old state
4357      * around to properly compose it. */
4358     xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4359     xlate_actions(&xin, &xout);
4360     flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4361
4362     /* A facet's slow path reason should only change under dramatic
4363      * circumstances.  Rather than try to update everything, it's simpler to
4364      * remove the facet and start over.
4365      *
4366      * More importantly, if a facet's wildcards change, it will be relatively
4367      * difficult to figure out if its subfacets still belong to it, and if not
4368      * which facet they may belong to.  Again, to avoid the complexity, we
4369      * simply give up instead. */
4370     if (facet->xout.slow != xout.slow
4371         || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4372         facet_remove(facet);
4373         xlate_out_uninit(&xout);
4374         rule_release(new_rule);
4375         return false;
4376     }
4377
4378     if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4379         LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4380             if (subfacet->path == SF_FAST_PATH) {
4381                 struct dpif_flow_stats stats;
4382
4383                 subfacet_install(subfacet, &xout.odp_actions, &stats);
4384                 subfacet_update_stats(subfacet, &stats);
4385             }
4386         }
4387
4388         facet_flush_stats(facet);
4389
4390         ofpbuf_clear(&facet->xout.odp_actions);
4391         ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4392                    xout.odp_actions.size);
4393     }
4394
4395     /* Update 'facet' now that we've taken care of all the old state. */
4396     facet->xout.slow = xout.slow;
4397     facet->xout.has_learn = xout.has_learn;
4398     facet->xout.has_normal = xout.has_normal;
4399     facet->xout.has_fin_timeout = xout.has_fin_timeout;
4400     facet->xout.nf_output_iface = xout.nf_output_iface;
4401     facet->xout.mirrors = xout.mirrors;
4402     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4403     facet->used = MAX(facet->used, new_rule->up.created);
4404
4405     xlate_out_uninit(&xout);
4406     rule_release(new_rule);
4407     return true;
4408 }
4409
4410 static void
4411 facet_reset_counters(struct facet *facet)
4412 {
4413     facet->packet_count = 0;
4414     facet->byte_count = 0;
4415     facet->prev_packet_count = 0;
4416     facet->prev_byte_count = 0;
4417     facet->accounted_bytes = 0;
4418 }
4419
4420 static void
4421 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4422                 struct dpif_flow_stats *stats, bool may_learn)
4423 {
4424     struct ofport_dpif *in_port;
4425     struct rule_dpif *rule;
4426     struct xlate_in xin;
4427
4428     in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4429     if (in_port && in_port->is_tunnel) {
4430         netdev_vport_inc_rx(in_port->up.netdev, stats);
4431     }
4432
4433     rule_dpif_lookup(ofproto, flow, NULL, &rule);
4434     rule_credit_stats(rule, stats);
4435     xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4436     xin.resubmit_stats = stats;
4437     xin.may_learn = may_learn;
4438     xlate_actions_for_side_effects(&xin);
4439     rule_release(rule);
4440 }
4441
4442 static void
4443 facet_push_stats(struct facet *facet, bool may_learn)
4444 {
4445     struct dpif_flow_stats stats;
4446
4447     ovs_assert(facet->packet_count >= facet->prev_packet_count);
4448     ovs_assert(facet->byte_count >= facet->prev_byte_count);
4449     ovs_assert(facet->used >= facet->prev_used);
4450
4451     stats.n_packets = facet->packet_count - facet->prev_packet_count;
4452     stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4453     stats.used = facet->used;
4454     stats.tcp_flags = facet->tcp_flags;
4455
4456     if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4457         facet->prev_packet_count = facet->packet_count;
4458         facet->prev_byte_count = facet->byte_count;
4459         facet->prev_used = facet->used;
4460
4461         netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4462                                  facet->used);
4463         netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4464         mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4465                             stats.n_packets, stats.n_bytes);
4466         flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4467     }
4468 }
4469
4470 static void
4471 push_all_stats__(bool run_fast)
4472 {
4473     static long long int rl = LLONG_MIN;
4474     struct ofproto_dpif *ofproto;
4475
4476     if (time_msec() < rl) {
4477         return;
4478     }
4479
4480     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4481         struct cls_cursor cursor;
4482         struct facet *facet;
4483
4484         ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4485         cls_cursor_init(&cursor, &ofproto->facets, NULL);
4486         CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4487             facet_push_stats(facet, false);
4488             if (run_fast) {
4489                 run_fast_rl();
4490             }
4491         }
4492         ovs_rwlock_unlock(&ofproto->facets.rwlock);
4493     }
4494
4495     rl = time_msec() + 100;
4496 }
4497
4498 static void
4499 push_all_stats(void)
4500 {
4501     push_all_stats__(true);
4502 }
4503
4504 void
4505 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4506 {
4507     ovs_mutex_lock(&rule->stats_mutex);
4508     rule->packet_count += stats->n_packets;
4509     rule->byte_count += stats->n_bytes;
4510     ofproto_rule_update_used(&rule->up, stats->used);
4511     ovs_mutex_unlock(&rule->stats_mutex);
4512 }
4513 \f
4514 /* Subfacets. */
4515
4516 static struct subfacet *
4517 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4518               size_t key_len, uint32_t key_hash)
4519 {
4520     struct subfacet *subfacet;
4521
4522     HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4523                              &backer->subfacets) {
4524         if (subfacet->key_len == key_len
4525             && !memcmp(key, subfacet->key, key_len)) {
4526             return subfacet;
4527         }
4528     }
4529
4530     return NULL;
4531 }
4532
4533 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4534  * 'key_fitness', 'key', and 'key_len' members in 'miss'.  Returns the
4535  * existing subfacet if there is one, otherwise creates and returns a
4536  * new subfacet. */
4537 static struct subfacet *
4538 subfacet_create(struct facet *facet, struct flow_miss *miss)
4539 {
4540     struct dpif_backer *backer = miss->ofproto->backer;
4541     enum odp_key_fitness key_fitness = miss->key_fitness;
4542     const struct nlattr *key = miss->key;
4543     size_t key_len = miss->key_len;
4544     uint32_t key_hash;
4545     struct subfacet *subfacet;
4546
4547     key_hash = odp_flow_key_hash(key, key_len);
4548
4549     if (list_is_empty(&facet->subfacets)) {
4550         subfacet = &facet->one_subfacet;
4551     } else {
4552         subfacet = subfacet_find(backer, key, key_len, key_hash);
4553         if (subfacet) {
4554             if (subfacet->facet == facet) {
4555                 return subfacet;
4556             }
4557
4558             /* This shouldn't happen. */
4559             VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4560             subfacet_destroy(subfacet);
4561         }
4562
4563         subfacet = xmalloc(sizeof *subfacet);
4564     }
4565
4566     hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4567     list_push_back(&facet->subfacets, &subfacet->list_node);
4568     subfacet->facet = facet;
4569     subfacet->key_fitness = key_fitness;
4570     subfacet->key = xmemdup(key, key_len);
4571     subfacet->key_len = key_len;
4572     subfacet->used = miss->stats.used;
4573     subfacet->created = subfacet->used;
4574     subfacet->dp_packet_count = 0;
4575     subfacet->dp_byte_count = 0;
4576     subfacet->path = SF_NOT_INSTALLED;
4577     subfacet->backer = backer;
4578
4579     backer->subfacet_add_count++;
4580     return subfacet;
4581 }
4582
4583 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4584  * its facet within 'ofproto', and frees it. */
4585 static void
4586 subfacet_destroy__(struct subfacet *subfacet)
4587 {
4588     struct facet *facet = subfacet->facet;
4589     struct ofproto_dpif *ofproto = facet->ofproto;
4590
4591     /* Update ofproto stats before uninstall the subfacet. */
4592     ofproto->backer->subfacet_del_count++;
4593
4594     subfacet_uninstall(subfacet);
4595     hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4596     list_remove(&subfacet->list_node);
4597     free(subfacet->key);
4598     if (subfacet != &facet->one_subfacet) {
4599         free(subfacet);
4600     }
4601 }
4602
4603 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4604  * last remaining subfacet in its facet destroys the facet too. */
4605 static void
4606 subfacet_destroy(struct subfacet *subfacet)
4607 {
4608     struct facet *facet = subfacet->facet;
4609
4610     if (list_is_singleton(&facet->subfacets)) {
4611         /* facet_remove() needs at least one subfacet (it will remove it). */
4612         facet_remove(facet);
4613     } else {
4614         subfacet_destroy__(subfacet);
4615     }
4616 }
4617
4618 static void
4619 subfacet_destroy_batch(struct dpif_backer *backer,
4620                        struct subfacet **subfacets, int n)
4621 {
4622     struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4623     struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4624     struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4625     int i;
4626
4627     for (i = 0; i < n; i++) {
4628         ops[i].type = DPIF_OP_FLOW_DEL;
4629         ops[i].u.flow_del.key = subfacets[i]->key;
4630         ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4631         ops[i].u.flow_del.stats = &stats[i];
4632         opsp[i] = &ops[i];
4633     }
4634
4635     dpif_operate(backer->dpif, opsp, n);
4636     for (i = 0; i < n; i++) {
4637         subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4638         subfacets[i]->path = SF_NOT_INSTALLED;
4639         subfacet_destroy(subfacets[i]);
4640         run_fast_rl();
4641     }
4642 }
4643
4644 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4645  * bytes of actions in 'actions'.  If 'stats' is non-null, statistics counters
4646  * in the datapath will be zeroed and 'stats' will be updated with traffic new
4647  * since 'subfacet' was last updated.
4648  *
4649  * Returns 0 if successful, otherwise a positive errno value. */
4650 static int
4651 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4652                  struct dpif_flow_stats *stats)
4653 {
4654     struct facet *facet = subfacet->facet;
4655     enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4656     const struct nlattr *actions = odp_actions->data;
4657     size_t actions_len = odp_actions->size;
4658     struct odputil_keybuf maskbuf;
4659     struct ofpbuf mask;
4660
4661     uint64_t slow_path_stub[128 / 8];
4662     enum dpif_flow_put_flags flags;
4663     int ret;
4664
4665     flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4666                                                : DPIF_FP_MODIFY;
4667     if (stats) {
4668         flags |= DPIF_FP_ZERO_STATS;
4669     }
4670
4671     if (path == SF_SLOW_PATH) {
4672         compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4673                           slow_path_stub, sizeof slow_path_stub,
4674                           &actions, &actions_len);
4675     }
4676
4677     ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4678     if (enable_megaflows) {
4679         odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4680                                &facet->flow, UINT32_MAX);
4681     }
4682
4683     ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4684                         subfacet->key_len,  mask.data, mask.size,
4685                         actions, actions_len, stats);
4686
4687     if (stats) {
4688         subfacet_reset_dp_stats(subfacet, stats);
4689     }
4690
4691     if (ret) {
4692         COVERAGE_INC(subfacet_install_fail);
4693     } else {
4694         subfacet->path = path;
4695     }
4696     return ret;
4697 }
4698
4699 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4700 static void
4701 subfacet_uninstall(struct subfacet *subfacet)
4702 {
4703     if (subfacet->path != SF_NOT_INSTALLED) {
4704         struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4705         struct dpif_flow_stats stats;
4706         int error;
4707
4708         error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4709                               subfacet->key_len, &stats);
4710         subfacet_reset_dp_stats(subfacet, &stats);
4711         if (!error) {
4712             subfacet_update_stats(subfacet, &stats);
4713         }
4714         subfacet->path = SF_NOT_INSTALLED;
4715     } else {
4716         ovs_assert(subfacet->dp_packet_count == 0);
4717         ovs_assert(subfacet->dp_byte_count == 0);
4718     }
4719 }
4720
4721 /* Resets 'subfacet''s datapath statistics counters.  This should be called
4722  * when 'subfacet''s statistics are cleared in the datapath.  If 'stats' is
4723  * non-null, it should contain the statistics returned by dpif when 'subfacet'
4724  * was reset in the datapath.  'stats' will be modified to include only
4725  * statistics new since 'subfacet' was last updated. */
4726 static void
4727 subfacet_reset_dp_stats(struct subfacet *subfacet,
4728                         struct dpif_flow_stats *stats)
4729 {
4730     if (stats
4731         && subfacet->dp_packet_count <= stats->n_packets
4732         && subfacet->dp_byte_count <= stats->n_bytes) {
4733         stats->n_packets -= subfacet->dp_packet_count;
4734         stats->n_bytes -= subfacet->dp_byte_count;
4735     }
4736
4737     subfacet->dp_packet_count = 0;
4738     subfacet->dp_byte_count = 0;
4739 }
4740
4741 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4742  *
4743  * Because of the meaning of a subfacet's counters, it only makes sense to do
4744  * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4745  * represents a packet that was sent by hand or if it represents statistics
4746  * that have been cleared out of the datapath. */
4747 static void
4748 subfacet_update_stats(struct subfacet *subfacet,
4749                       const struct dpif_flow_stats *stats)
4750 {
4751     if (stats->n_packets || stats->used > subfacet->used) {
4752         struct facet *facet = subfacet->facet;
4753
4754         subfacet->used = MAX(subfacet->used, stats->used);
4755         facet->used = MAX(facet->used, stats->used);
4756         facet->packet_count += stats->n_packets;
4757         facet->byte_count += stats->n_bytes;
4758         facet->tcp_flags |= stats->tcp_flags;
4759     }
4760 }
4761 \f
4762 /* Rules. */
4763
4764 /* Lookup 'flow' in 'ofproto''s classifier.  If 'wc' is non-null, sets
4765  * the fields that were relevant as part of the lookup. */
4766 void
4767 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4768                  struct flow_wildcards *wc, struct rule_dpif **rule)
4769 {
4770     struct ofport_dpif *port;
4771
4772     if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4773         return;
4774     }
4775     port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4776     if (!port) {
4777         VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4778                      flow->in_port.ofp_port);
4779     }
4780
4781     *rule = choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4782                              ofproto->no_packet_in_rule);
4783     ovs_rwlock_rdlock(&(*rule)->up.evict);
4784 }
4785
4786 bool
4787 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4788                           const struct flow *flow, struct flow_wildcards *wc,
4789                           uint8_t table_id, struct rule_dpif **rule)
4790     OVS_TRY_RDLOCK(true, (*rule)->up.evict)
4791 {
4792     struct cls_rule *cls_rule;
4793     struct classifier *cls;
4794     bool frag;
4795
4796     *rule = NULL;
4797     if (table_id >= N_TABLES) {
4798         return false;
4799     }
4800
4801     if (wc) {
4802         memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4803         wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4804     }
4805
4806     cls = &ofproto->up.tables[table_id].cls;
4807     ovs_rwlock_rdlock(&cls->rwlock);
4808     frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4809     if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4810         /* We must pretend that transport ports are unavailable. */
4811         struct flow ofpc_normal_flow = *flow;
4812         ofpc_normal_flow.tp_src = htons(0);
4813         ofpc_normal_flow.tp_dst = htons(0);
4814         cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4815     } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4816         cls_rule = &ofproto->drop_frags_rule->up.cr;
4817         if (wc) {
4818             flow_wildcards_init_exact(wc);
4819         }
4820     } else {
4821         cls_rule = classifier_lookup(cls, flow, wc);
4822     }
4823
4824     *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4825     if (*rule && ovs_rwlock_tryrdlock(&(*rule)->up.evict)) {
4826         /* The rule is in the process of being removed.  Best we can do is
4827          * pretend it isn't there. */
4828         *rule = NULL;
4829     }
4830     ovs_rwlock_unlock(&cls->rwlock);
4831
4832     return *rule != NULL;
4833 }
4834
4835 /* Given a port configuration (specified as zero if there's no port), chooses
4836  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4837  * flow table miss. */
4838 struct rule_dpif *
4839 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4840                  struct rule_dpif *no_packet_in_rule)
4841 {
4842     return config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4843 }
4844
4845 void
4846 rule_release(struct rule_dpif *rule)
4847     OVS_NO_THREAD_SAFETY_ANALYSIS
4848 {
4849     if (rule) {
4850         ovs_rwlock_unlock(&rule->up.evict);
4851     }
4852 }
4853
4854 static void
4855 complete_operation(struct rule_dpif *rule)
4856 {
4857     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4858
4859     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4860     if (clogged) {
4861         struct dpif_completion *c = xmalloc(sizeof *c);
4862         c->op = rule->up.pending;
4863         list_push_back(&ofproto->completions, &c->list_node);
4864     } else {
4865         ofoperation_complete(rule->up.pending, 0);
4866     }
4867 }
4868
4869 static struct rule *
4870 rule_alloc(void)
4871 {
4872     struct rule_dpif *rule = xmalloc(sizeof *rule);
4873     return &rule->up;
4874 }
4875
4876 static void
4877 rule_dealloc(struct rule *rule_)
4878 {
4879     struct rule_dpif *rule = rule_dpif_cast(rule_);
4880     free(rule);
4881 }
4882
4883 static enum ofperr
4884 rule_construct(struct rule *rule_)
4885 {
4886     struct rule_dpif *rule = rule_dpif_cast(rule_);
4887     ovs_mutex_init(&rule->stats_mutex);
4888     ovs_mutex_lock(&rule->stats_mutex);
4889     rule->packet_count = 0;
4890     rule->byte_count = 0;
4891     ovs_mutex_unlock(&rule->stats_mutex);
4892     return 0;
4893 }
4894
4895 static void
4896 rule_insert(struct rule *rule_)
4897 {
4898     struct rule_dpif *rule = rule_dpif_cast(rule_);
4899     complete_operation(rule);
4900 }
4901
4902 static void
4903 rule_delete(struct rule *rule_)
4904 {
4905     struct rule_dpif *rule = rule_dpif_cast(rule_);
4906     complete_operation(rule);
4907 }
4908
4909 static void
4910 rule_destruct(struct rule *rule_)
4911 {
4912     struct rule_dpif *rule = rule_dpif_cast(rule_);
4913     ovs_mutex_destroy(&rule->stats_mutex);
4914 }
4915
4916 static void
4917 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4918 {
4919     struct rule_dpif *rule = rule_dpif_cast(rule_);
4920
4921     /* push_all_stats() can handle flow misses which, when using the learn
4922      * action, can cause rules to be added and deleted.  This can corrupt our
4923      * caller's datastructures which assume that rule_get_stats() doesn't have
4924      * an impact on the flow table. To be safe, we disable miss handling. */
4925     push_all_stats__(false);
4926
4927     /* Start from historical data for 'rule' itself that are no longer tracked
4928      * in facets.  This counts, for example, facets that have expired. */
4929     ovs_mutex_lock(&rule->stats_mutex);
4930     *packets = rule->packet_count;
4931     *bytes = rule->byte_count;
4932     ovs_mutex_unlock(&rule->stats_mutex);
4933 }
4934
4935 static void
4936 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4937                   struct ofpbuf *packet)
4938 {
4939     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4940     struct dpif_flow_stats stats;
4941     struct xlate_out xout;
4942     struct xlate_in xin;
4943
4944     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4945     rule_credit_stats(rule, &stats);
4946
4947     xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4948     xin.resubmit_stats = &stats;
4949     xlate_actions(&xin, &xout);
4950
4951     execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4952                         xout.odp_actions.size, packet);
4953
4954     xlate_out_uninit(&xout);
4955 }
4956
4957 static enum ofperr
4958 rule_execute(struct rule *rule, const struct flow *flow,
4959              struct ofpbuf *packet)
4960 {
4961     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4962     ofpbuf_delete(packet);
4963     return 0;
4964 }
4965
4966 static void
4967 rule_modify_actions(struct rule *rule_, bool reset_counters)
4968 {
4969     struct rule_dpif *rule = rule_dpif_cast(rule_);
4970
4971     if (reset_counters) {
4972         ovs_mutex_lock(&rule->stats_mutex);
4973         rule->packet_count = 0;
4974         rule->byte_count = 0;
4975         ovs_mutex_unlock(&rule->stats_mutex);
4976     }
4977
4978     complete_operation(rule);
4979 }
4980 \f
4981 /* Sends 'packet' out 'ofport'.
4982  * May modify 'packet'.
4983  * Returns 0 if successful, otherwise a positive errno value. */
4984 static int
4985 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4986 {
4987     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4988     uint64_t odp_actions_stub[1024 / 8];
4989     struct ofpbuf key, odp_actions;
4990     struct dpif_flow_stats stats;
4991     struct odputil_keybuf keybuf;
4992     struct ofpact_output output;
4993     struct xlate_out xout;
4994     struct xlate_in xin;
4995     struct flow flow;
4996     union flow_in_port in_port_;
4997     int error;
4998
4999     ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5000     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5001
5002     /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5003     in_port_.ofp_port = OFPP_NONE;
5004     flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
5005     odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
5006                                                              OFPP_LOCAL));
5007     dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5008
5009     ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5010     output.port = ofport->up.ofp_port;
5011     output.max_len = 0;
5012
5013     xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
5014     xin.ofpacts_len = sizeof output;
5015     xin.ofpacts = &output.ofpact;
5016     xin.resubmit_stats = &stats;
5017     xlate_actions(&xin, &xout);
5018
5019     error = dpif_execute(ofproto->backer->dpif,
5020                          key.data, key.size,
5021                          xout.odp_actions.data, xout.odp_actions.size,
5022                          packet);
5023     xlate_out_uninit(&xout);
5024
5025     if (error) {
5026         VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
5027                      ofproto->up.name, netdev_get_name(ofport->up.netdev),
5028                      ovs_strerror(error));
5029     }
5030
5031     ofproto->stats.tx_packets++;
5032     ofproto->stats.tx_bytes += packet->size;
5033     return error;
5034 }
5035
5036 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5037  * The action will state 'slow' as the reason that the action is in the slow
5038  * path.  (This is purely informational: it allows a human viewing "ovs-dpctl
5039  * dump-flows" output to see why a flow is in the slow path.)
5040  *
5041  * The 'stub_size' bytes in 'stub' will be used to store the action.
5042  * 'stub_size' must be large enough for the action.
5043  *
5044  * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5045  * respectively. */
5046 static void
5047 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5048                   enum slow_path_reason slow,
5049                   uint64_t *stub, size_t stub_size,
5050                   const struct nlattr **actionsp, size_t *actions_lenp)
5051 {
5052     union user_action_cookie cookie;
5053     struct ofpbuf buf;
5054
5055     cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5056     cookie.slow_path.unused = 0;
5057     cookie.slow_path.reason = slow;
5058
5059     ofpbuf_use_stack(&buf, stub, stub_size);
5060     if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5061         uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5062                                          ODPP_NONE);
5063         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5064     } else {
5065         odp_port_t odp_port;
5066         uint32_t pid;
5067
5068         odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5069         pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5070         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5071     }
5072     *actionsp = buf.data;
5073     *actions_lenp = buf.size;
5074 }
5075 \f
5076 static bool
5077 set_frag_handling(struct ofproto *ofproto_,
5078                   enum ofp_config_flags frag_handling)
5079 {
5080     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5081     if (frag_handling != OFPC_FRAG_REASM) {
5082         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5083         return true;
5084     } else {
5085         return false;
5086     }
5087 }
5088
5089 static enum ofperr
5090 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5091            const struct flow *flow,
5092            const struct ofpact *ofpacts, size_t ofpacts_len)
5093 {
5094     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5095     struct odputil_keybuf keybuf;
5096     struct dpif_flow_stats stats;
5097     struct xlate_out xout;
5098     struct xlate_in xin;
5099     struct ofpbuf key;
5100
5101
5102     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5103     odp_flow_key_from_flow(&key, flow,
5104                            ofp_port_to_odp_port(ofproto,
5105                                       flow->in_port.ofp_port));
5106
5107     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5108
5109     xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5110     xin.resubmit_stats = &stats;
5111     xin.ofpacts_len = ofpacts_len;
5112     xin.ofpacts = ofpacts;
5113
5114     xlate_actions(&xin, &xout);
5115     dpif_execute(ofproto->backer->dpif, key.data, key.size,
5116                  xout.odp_actions.data, xout.odp_actions.size, packet);
5117     xlate_out_uninit(&xout);
5118
5119     return 0;
5120 }
5121 \f
5122 /* NetFlow. */
5123
5124 static int
5125 set_netflow(struct ofproto *ofproto_,
5126             const struct netflow_options *netflow_options)
5127 {
5128     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5129
5130     if (netflow_options) {
5131         if (!ofproto->netflow) {
5132             ofproto->netflow = netflow_create();
5133             ofproto->backer->need_revalidate = REV_RECONFIGURE;
5134         }
5135         return netflow_set_options(ofproto->netflow, netflow_options);
5136     } else if (ofproto->netflow) {
5137         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5138         netflow_destroy(ofproto->netflow);
5139         ofproto->netflow = NULL;
5140     }
5141
5142     return 0;
5143 }
5144
5145 static void
5146 get_netflow_ids(const struct ofproto *ofproto_,
5147                 uint8_t *engine_type, uint8_t *engine_id)
5148 {
5149     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5150
5151     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5152 }
5153
5154 static void
5155 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5156 {
5157     if (!facet_is_controller_flow(facet) &&
5158         netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5159         struct subfacet *subfacet;
5160         struct ofexpired expired;
5161
5162         LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5163             if (subfacet->path == SF_FAST_PATH) {
5164                 struct dpif_flow_stats stats;
5165
5166                 subfacet_install(subfacet, &facet->xout.odp_actions,
5167                                  &stats);
5168                 subfacet_update_stats(subfacet, &stats);
5169             }
5170         }
5171
5172         expired.flow = facet->flow;
5173         expired.packet_count = facet->packet_count;
5174         expired.byte_count = facet->byte_count;
5175         expired.used = facet->used;
5176         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5177     }
5178 }
5179
5180 static void
5181 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5182 {
5183     struct cls_cursor cursor;
5184     struct facet *facet;
5185
5186     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5187     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5188     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5189         send_active_timeout(ofproto, facet);
5190     }
5191     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5192 }
5193 \f
5194 static struct ofproto_dpif *
5195 ofproto_dpif_lookup(const char *name)
5196 {
5197     struct ofproto_dpif *ofproto;
5198
5199     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5200                              hash_string(name, 0), &all_ofproto_dpifs) {
5201         if (!strcmp(ofproto->up.name, name)) {
5202             return ofproto;
5203         }
5204     }
5205     return NULL;
5206 }
5207
5208 static void
5209 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5210                           const char *argv[], void *aux OVS_UNUSED)
5211 {
5212     struct ofproto_dpif *ofproto;
5213
5214     if (argc > 1) {
5215         ofproto = ofproto_dpif_lookup(argv[1]);
5216         if (!ofproto) {
5217             unixctl_command_reply_error(conn, "no such bridge");
5218             return;
5219         }
5220         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5221         mac_learning_flush(ofproto->ml);
5222         ovs_rwlock_unlock(&ofproto->ml->rwlock);
5223     } else {
5224         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5225             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5226             mac_learning_flush(ofproto->ml);
5227             ovs_rwlock_unlock(&ofproto->ml->rwlock);
5228         }
5229     }
5230
5231     unixctl_command_reply(conn, "table successfully flushed");
5232 }
5233
5234 static struct ofport_dpif *
5235 ofbundle_get_a_port(const struct ofbundle *bundle)
5236 {
5237     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5238                         bundle_node);
5239 }
5240
5241 static void
5242 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5243                          const char *argv[], void *aux OVS_UNUSED)
5244 {
5245     struct ds ds = DS_EMPTY_INITIALIZER;
5246     const struct ofproto_dpif *ofproto;
5247     const struct mac_entry *e;
5248
5249     ofproto = ofproto_dpif_lookup(argv[1]);
5250     if (!ofproto) {
5251         unixctl_command_reply_error(conn, "no such bridge");
5252         return;
5253     }
5254
5255     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
5256     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5257     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5258         struct ofbundle *bundle = e->port.p;
5259         char name[OFP_MAX_PORT_NAME_LEN];
5260
5261         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5262                                name, sizeof name);
5263         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
5264                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
5265                       mac_entry_age(ofproto->ml, e));
5266     }
5267     ovs_rwlock_unlock(&ofproto->ml->rwlock);
5268     unixctl_command_reply(conn, ds_cstr(&ds));
5269     ds_destroy(&ds);
5270 }
5271
5272 struct trace_ctx {
5273     struct xlate_out xout;
5274     struct xlate_in xin;
5275     struct flow flow;
5276     struct ds *result;
5277 };
5278
5279 static void
5280 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5281 {
5282     ds_put_char_multiple(result, '\t', level);
5283     if (!rule) {
5284         ds_put_cstr(result, "No match\n");
5285         return;
5286     }
5287
5288     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5289                   rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5290     cls_rule_format(&rule->up.cr, result);
5291     ds_put_char(result, '\n');
5292
5293     ds_put_char_multiple(result, '\t', level);
5294     ds_put_cstr(result, "OpenFlow ");
5295     ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5296     ds_put_char(result, '\n');
5297 }
5298
5299 static void
5300 trace_format_flow(struct ds *result, int level, const char *title,
5301                   struct trace_ctx *trace)
5302 {
5303     ds_put_char_multiple(result, '\t', level);
5304     ds_put_format(result, "%s: ", title);
5305     if (flow_equal(&trace->xin.flow, &trace->flow)) {
5306         ds_put_cstr(result, "unchanged");
5307     } else {
5308         flow_format(result, &trace->xin.flow);
5309         trace->flow = trace->xin.flow;
5310     }
5311     ds_put_char(result, '\n');
5312 }
5313
5314 static void
5315 trace_format_regs(struct ds *result, int level, const char *title,
5316                   struct trace_ctx *trace)
5317 {
5318     size_t i;
5319
5320     ds_put_char_multiple(result, '\t', level);
5321     ds_put_format(result, "%s:", title);
5322     for (i = 0; i < FLOW_N_REGS; i++) {
5323         ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5324     }
5325     ds_put_char(result, '\n');
5326 }
5327
5328 static void
5329 trace_format_odp(struct ds *result, int level, const char *title,
5330                  struct trace_ctx *trace)
5331 {
5332     struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5333
5334     ds_put_char_multiple(result, '\t', level);
5335     ds_put_format(result, "%s: ", title);
5336     format_odp_actions(result, odp_actions->data, odp_actions->size);
5337     ds_put_char(result, '\n');
5338 }
5339
5340 static void
5341 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5342 {
5343     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5344     struct ds *result = trace->result;
5345
5346     ds_put_char(result, '\n');
5347     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5348     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5349     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
5350     trace_format_rule(result, recurse + 1, rule);
5351 }
5352
5353 static void
5354 trace_report(struct xlate_in *xin, const char *s, int recurse)
5355 {
5356     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5357     struct ds *result = trace->result;
5358
5359     ds_put_char_multiple(result, '\t', recurse);
5360     ds_put_cstr(result, s);
5361     ds_put_char(result, '\n');
5362 }
5363
5364 static void
5365 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5366                       void *aux OVS_UNUSED)
5367 {
5368     const struct dpif_backer *backer;
5369     struct ofproto_dpif *ofproto;
5370     struct ofpbuf odp_key, odp_mask;
5371     struct ofpbuf *packet;
5372     struct ds result;
5373     struct flow flow;
5374     char *s;
5375
5376     packet = NULL;
5377     backer = NULL;
5378     ds_init(&result);
5379     ofpbuf_init(&odp_key, 0);
5380     ofpbuf_init(&odp_mask, 0);
5381
5382     /* Handle "-generate" or a hex string as the last argument. */
5383     if (!strcmp(argv[argc - 1], "-generate")) {
5384         packet = ofpbuf_new(0);
5385         argc--;
5386     } else {
5387         const char *error = eth_from_hex(argv[argc - 1], &packet);
5388         if (!error) {
5389             argc--;
5390         } else if (argc == 4) {
5391             /* The 3-argument form must end in "-generate' or a hex string. */
5392             unixctl_command_reply_error(conn, error);
5393             goto exit;
5394         }
5395     }
5396
5397     /* Parse the flow and determine whether a datapath or
5398      * bridge is specified. If function odp_flow_key_from_string()
5399      * returns 0, the flow is a odp_flow. If function
5400      * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5401     if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5402         /* If the odp_flow is the second argument,
5403          * the datapath name is the first argument. */
5404         if (argc == 3) {
5405             const char *dp_type;
5406             if (!strncmp(argv[1], "ovs-", 4)) {
5407                 dp_type = argv[1] + 4;
5408             } else {
5409                 dp_type = argv[1];
5410             }
5411             backer = shash_find_data(&all_dpif_backers, dp_type);
5412             if (!backer) {
5413                 unixctl_command_reply_error(conn, "Cannot find datapath "
5414                                "of this name");
5415                 goto exit;
5416             }
5417         } else {
5418             /* No datapath name specified, so there should be only one
5419              * datapath. */
5420             struct shash_node *node;
5421             if (shash_count(&all_dpif_backers) != 1) {
5422                 unixctl_command_reply_error(conn, "Must specify datapath "
5423                          "name, there is more than one type of datapath");
5424                 goto exit;
5425             }
5426             node = shash_first(&all_dpif_backers);
5427             backer = node->data;
5428         }
5429
5430         if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5431                           NULL, &ofproto, NULL)) {
5432             unixctl_command_reply_error(conn, "Invalid datapath flow");
5433             goto exit;
5434         }
5435         ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5436     } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5437         if (argc != 3) {
5438             unixctl_command_reply_error(conn, "Must specify bridge name");
5439             goto exit;
5440         }
5441
5442         ofproto = ofproto_dpif_lookup(argv[1]);
5443         if (!ofproto) {
5444             unixctl_command_reply_error(conn, "Unknown bridge name");
5445             goto exit;
5446         }
5447     } else {
5448         unixctl_command_reply_error(conn, "Bad flow syntax");
5449         goto exit;
5450     }
5451
5452     /* Generate a packet, if requested. */
5453     if (packet) {
5454         if (!packet->size) {
5455             flow_compose(packet, &flow);
5456         } else {
5457             union flow_in_port in_port_;
5458
5459             in_port_ = flow.in_port;
5460             ds_put_cstr(&result, "Packet: ");
5461             s = ofp_packet_to_string(packet->data, packet->size);
5462             ds_put_cstr(&result, s);
5463             free(s);
5464
5465             /* Use the metadata from the flow and the packet argument
5466              * to reconstruct the flow. */
5467             flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5468                          &in_port_, &flow);
5469         }
5470     }
5471
5472     ofproto_trace(ofproto, &flow, packet, &result);
5473     unixctl_command_reply(conn, ds_cstr(&result));
5474
5475 exit:
5476     ds_destroy(&result);
5477     ofpbuf_delete(packet);
5478     ofpbuf_uninit(&odp_key);
5479     ofpbuf_uninit(&odp_mask);
5480 }
5481
5482 static void
5483 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5484               const struct ofpbuf *packet, struct ds *ds)
5485 {
5486     struct rule_dpif *rule;
5487     struct flow_wildcards wc;
5488
5489     ds_put_cstr(ds, "Flow: ");
5490     flow_format(ds, flow);
5491     ds_put_char(ds, '\n');
5492
5493     flow_wildcards_init_catchall(&wc);
5494     rule_dpif_lookup(ofproto, flow, &wc, &rule);
5495
5496     trace_format_rule(ds, 0, rule);
5497     if (rule == ofproto->miss_rule) {
5498         ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5499     } else if (rule == ofproto->no_packet_in_rule) {
5500         ds_put_cstr(ds, "\nNo match, packets dropped because "
5501                     "OFPPC_NO_PACKET_IN is set on in_port.\n");
5502     } else if (rule == ofproto->drop_frags_rule) {
5503         ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5504                     "and the fragment handling mode is \"drop\".\n");
5505     }
5506
5507     if (rule) {
5508         uint64_t odp_actions_stub[1024 / 8];
5509         struct ofpbuf odp_actions;
5510         struct trace_ctx trace;
5511         struct match match;
5512         uint8_t tcp_flags;
5513
5514         tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5515         trace.result = ds;
5516         trace.flow = *flow;
5517         ofpbuf_use_stub(&odp_actions,
5518                         odp_actions_stub, sizeof odp_actions_stub);
5519         xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5520         trace.xin.resubmit_hook = trace_resubmit;
5521         trace.xin.report_hook = trace_report;
5522
5523         xlate_actions(&trace.xin, &trace.xout);
5524         flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5525
5526         ds_put_char(ds, '\n');
5527         trace_format_flow(ds, 0, "Final flow", &trace);
5528
5529         match_init(&match, flow, &trace.xout.wc);
5530         ds_put_cstr(ds, "Relevant fields: ");
5531         match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5532         ds_put_char(ds, '\n');
5533
5534         ds_put_cstr(ds, "Datapath actions: ");
5535         format_odp_actions(ds, trace.xout.odp_actions.data,
5536                            trace.xout.odp_actions.size);
5537
5538         if (trace.xout.slow) {
5539             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5540                         "slow path because it:");
5541             switch (trace.xout.slow) {
5542             case SLOW_CFM:
5543                 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5544                 break;
5545             case SLOW_LACP:
5546                 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5547                 break;
5548             case SLOW_STP:
5549                 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5550                 break;
5551             case SLOW_BFD:
5552                 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5553                 break;
5554             case SLOW_CONTROLLER:
5555                 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5556                             "to the OpenFlow controller.");
5557                 break;
5558             case __SLOW_MAX:
5559                 NOT_REACHED();
5560             }
5561         }
5562
5563         xlate_out_uninit(&trace.xout);
5564     }
5565
5566     rule_release(rule);
5567 }
5568
5569 static void
5570 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5571                   const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5572 {
5573     clogged = true;
5574     unixctl_command_reply(conn, NULL);
5575 }
5576
5577 static void
5578 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5579                     const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5580 {
5581     clogged = false;
5582     unixctl_command_reply(conn, NULL);
5583 }
5584
5585 /* Runs a self-check of flow translations in 'ofproto'.  Appends a message to
5586  * 'reply' describing the results. */
5587 static void
5588 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5589 {
5590     struct cls_cursor cursor;
5591     struct facet *facet;
5592     int errors;
5593
5594     errors = 0;
5595     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5596     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5597     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5598         if (!facet_check_consistency(facet)) {
5599             errors++;
5600         }
5601     }
5602     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5603     if (errors) {
5604         ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5605     }
5606
5607     if (errors) {
5608         ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5609                       ofproto->up.name, errors);
5610     } else {
5611         ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5612     }
5613 }
5614
5615 static void
5616 ofproto_dpif_self_check(struct unixctl_conn *conn,
5617                         int argc, const char *argv[], void *aux OVS_UNUSED)
5618 {
5619     struct ds reply = DS_EMPTY_INITIALIZER;
5620     struct ofproto_dpif *ofproto;
5621
5622     if (argc > 1) {
5623         ofproto = ofproto_dpif_lookup(argv[1]);
5624         if (!ofproto) {
5625             unixctl_command_reply_error(conn, "Unknown ofproto (use "
5626                                         "ofproto/list for help)");
5627             return;
5628         }
5629         ofproto_dpif_self_check__(ofproto, &reply);
5630     } else {
5631         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5632             ofproto_dpif_self_check__(ofproto, &reply);
5633         }
5634     }
5635
5636     unixctl_command_reply(conn, ds_cstr(&reply));
5637     ds_destroy(&reply);
5638 }
5639
5640 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
5641  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
5642  * to destroy 'ofproto_shash' and free the returned value. */
5643 static const struct shash_node **
5644 get_ofprotos(struct shash *ofproto_shash)
5645 {
5646     const struct ofproto_dpif *ofproto;
5647
5648     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5649         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5650         shash_add_nocopy(ofproto_shash, name, ofproto);
5651     }
5652
5653     return shash_sort(ofproto_shash);
5654 }
5655
5656 static void
5657 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5658                               const char *argv[] OVS_UNUSED,
5659                               void *aux OVS_UNUSED)
5660 {
5661     struct ds ds = DS_EMPTY_INITIALIZER;
5662     struct shash ofproto_shash;
5663     const struct shash_node **sorted_ofprotos;
5664     int i;
5665
5666     shash_init(&ofproto_shash);
5667     sorted_ofprotos = get_ofprotos(&ofproto_shash);
5668     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5669         const struct shash_node *node = sorted_ofprotos[i];
5670         ds_put_format(&ds, "%s\n", node->name);
5671     }
5672
5673     shash_destroy(&ofproto_shash);
5674     free(sorted_ofprotos);
5675
5676     unixctl_command_reply(conn, ds_cstr(&ds));
5677     ds_destroy(&ds);
5678 }
5679
5680 static void
5681 show_dp_rates(struct ds *ds, const char *heading,
5682               const struct avg_subfacet_rates *rates)
5683 {
5684     ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5685                   heading, rates->add_rate, rates->del_rate);
5686 }
5687
5688 static void
5689 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5690 {
5691     const struct shash_node **ofprotos;
5692     struct ofproto_dpif *ofproto;
5693     struct shash ofproto_shash;
5694     uint64_t n_hit, n_missed;
5695     long long int minutes;
5696     size_t i;
5697
5698     n_hit = n_missed = 0;
5699     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5700         if (ofproto->backer == backer) {
5701             n_missed += ofproto->n_missed;
5702             n_hit += ofproto->n_hit;
5703         }
5704     }
5705
5706     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5707                   dpif_name(backer->dpif), n_hit, n_missed);
5708     ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5709                   " life span: %lldms\n", hmap_count(&backer->subfacets),
5710                   backer->avg_n_subfacet, backer->max_n_subfacet,
5711                   backer->avg_subfacet_life);
5712
5713     minutes = (time_msec() - backer->created) / (1000 * 60);
5714     if (minutes >= 60) {
5715         show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5716     }
5717     if (minutes >= 60 * 24) {
5718         show_dp_rates(ds, "\tdaily avg:",  &backer->daily);
5719     }
5720     show_dp_rates(ds, "\toverall avg:",  &backer->lifetime);
5721
5722     shash_init(&ofproto_shash);
5723     ofprotos = get_ofprotos(&ofproto_shash);
5724     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5725         struct ofproto_dpif *ofproto = ofprotos[i]->data;
5726         const struct shash_node **ports;
5727         size_t j;
5728
5729         if (ofproto->backer != backer) {
5730             continue;
5731         }
5732
5733         ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5734                       ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5735
5736         ports = shash_sort(&ofproto->up.port_by_name);
5737         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5738             const struct shash_node *node = ports[j];
5739             struct ofport *ofport = node->data;
5740             struct smap config;
5741             odp_port_t odp_port;
5742
5743             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5744                           ofport->ofp_port);
5745
5746             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5747             if (odp_port != ODPP_NONE) {
5748                 ds_put_format(ds, "%"PRIu32":", odp_port);
5749             } else {
5750                 ds_put_cstr(ds, "none:");
5751             }
5752
5753             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5754
5755             smap_init(&config);
5756             if (!netdev_get_config(ofport->netdev, &config)) {
5757                 const struct smap_node **nodes;
5758                 size_t i;
5759
5760                 nodes = smap_sort(&config);
5761                 for (i = 0; i < smap_count(&config); i++) {
5762                     const struct smap_node *node = nodes[i];
5763                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5764                                   node->key, node->value);
5765                 }
5766                 free(nodes);
5767             }
5768             smap_destroy(&config);
5769
5770             ds_put_char(ds, ')');
5771             ds_put_char(ds, '\n');
5772         }
5773         free(ports);
5774     }
5775     shash_destroy(&ofproto_shash);
5776     free(ofprotos);
5777 }
5778
5779 static void
5780 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5781                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5782 {
5783     struct ds ds = DS_EMPTY_INITIALIZER;
5784     const struct shash_node **backers;
5785     int i;
5786
5787     backers = shash_sort(&all_dpif_backers);
5788     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5789         dpif_show_backer(backers[i]->data, &ds);
5790     }
5791     free(backers);
5792
5793     unixctl_command_reply(conn, ds_cstr(&ds));
5794     ds_destroy(&ds);
5795 }
5796
5797 /* Dump the megaflow (facet) cache.  This is useful to check the
5798  * correctness of flow wildcarding, since the same mechanism is used for
5799  * both xlate caching and kernel wildcarding.
5800  *
5801  * It's important to note that in the output the flow description uses
5802  * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5803  *
5804  * This command is only needed for advanced debugging, so it's not
5805  * documented in the man page. */
5806 static void
5807 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5808                                     int argc OVS_UNUSED, const char *argv[],
5809                                     void *aux OVS_UNUSED)
5810 {
5811     struct ds ds = DS_EMPTY_INITIALIZER;
5812     const struct ofproto_dpif *ofproto;
5813     long long int now = time_msec();
5814     struct cls_cursor cursor;
5815     struct facet *facet;
5816
5817     ofproto = ofproto_dpif_lookup(argv[1]);
5818     if (!ofproto) {
5819         unixctl_command_reply_error(conn, "no such bridge");
5820         return;
5821     }
5822
5823     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5824     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5825     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5826         cls_rule_format(&facet->cr, &ds);
5827         ds_put_cstr(&ds, ", ");
5828         ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5829         ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5830         ds_put_cstr(&ds, "Datapath actions: ");
5831         if (facet->xout.slow) {
5832             uint64_t slow_path_stub[128 / 8];
5833             const struct nlattr *actions;
5834             size_t actions_len;
5835
5836             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5837                               slow_path_stub, sizeof slow_path_stub,
5838                               &actions, &actions_len);
5839             format_odp_actions(&ds, actions, actions_len);
5840         } else {
5841             format_odp_actions(&ds, facet->xout.odp_actions.data,
5842                                facet->xout.odp_actions.size);
5843         }
5844         ds_put_cstr(&ds, "\n");
5845     }
5846     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5847
5848     ds_chomp(&ds, '\n');
5849     unixctl_command_reply(conn, ds_cstr(&ds));
5850     ds_destroy(&ds);
5851 }
5852
5853 /* Disable using the megaflows.
5854  *
5855  * This command is only needed for advanced debugging, so it's not
5856  * documented in the man page. */
5857 static void
5858 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5859                                        int argc OVS_UNUSED,
5860                                        const char *argv[] OVS_UNUSED,
5861                                        void *aux OVS_UNUSED)
5862 {
5863     struct ofproto_dpif *ofproto;
5864
5865     enable_megaflows = false;
5866
5867     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5868         flush(&ofproto->up);
5869     }
5870
5871     unixctl_command_reply(conn, "megaflows disabled");
5872 }
5873
5874 /* Re-enable using megaflows.
5875  *
5876  * This command is only needed for advanced debugging, so it's not
5877  * documented in the man page. */
5878 static void
5879 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5880                                       int argc OVS_UNUSED,
5881                                       const char *argv[] OVS_UNUSED,
5882                                       void *aux OVS_UNUSED)
5883 {
5884     struct ofproto_dpif *ofproto;
5885
5886     enable_megaflows = true;
5887
5888     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5889         flush(&ofproto->up);
5890     }
5891
5892     unixctl_command_reply(conn, "megaflows enabled");
5893 }
5894
5895 static void
5896 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5897                                 int argc OVS_UNUSED, const char *argv[],
5898                                 void *aux OVS_UNUSED)
5899 {
5900     struct ds ds = DS_EMPTY_INITIALIZER;
5901     const struct ofproto_dpif *ofproto;
5902     struct subfacet *subfacet;
5903
5904     ofproto = ofproto_dpif_lookup(argv[1]);
5905     if (!ofproto) {
5906         unixctl_command_reply_error(conn, "no such bridge");
5907         return;
5908     }
5909
5910     update_stats(ofproto->backer);
5911
5912     HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5913         struct facet *facet = subfacet->facet;
5914         struct odputil_keybuf maskbuf;
5915         struct ofpbuf mask;
5916
5917         if (facet->ofproto != ofproto) {
5918             continue;
5919         }
5920
5921         ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5922         if (enable_megaflows) {
5923             odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5924                                    &facet->flow, UINT32_MAX);
5925         }
5926
5927         odp_flow_format(subfacet->key, subfacet->key_len,
5928                         mask.data, mask.size, &ds, false);
5929
5930         ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5931                       subfacet->dp_packet_count, subfacet->dp_byte_count);
5932         if (subfacet->used) {
5933             ds_put_format(&ds, "%.3fs",
5934                           (time_msec() - subfacet->used) / 1000.0);
5935         } else {
5936             ds_put_format(&ds, "never");
5937         }
5938         if (subfacet->facet->tcp_flags) {
5939             ds_put_cstr(&ds, ", flags:");
5940             packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5941         }
5942
5943         ds_put_cstr(&ds, ", actions:");
5944         if (facet->xout.slow) {
5945             uint64_t slow_path_stub[128 / 8];
5946             const struct nlattr *actions;
5947             size_t actions_len;
5948
5949             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5950                               slow_path_stub, sizeof slow_path_stub,
5951                               &actions, &actions_len);
5952             format_odp_actions(&ds, actions, actions_len);
5953         } else {
5954             format_odp_actions(&ds, facet->xout.odp_actions.data,
5955                                facet->xout.odp_actions.size);
5956         }
5957         ds_put_char(&ds, '\n');
5958     }
5959
5960     unixctl_command_reply(conn, ds_cstr(&ds));
5961     ds_destroy(&ds);
5962 }
5963
5964 static void
5965 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5966                                int argc OVS_UNUSED, const char *argv[],
5967                                void *aux OVS_UNUSED)
5968 {
5969     struct ds ds = DS_EMPTY_INITIALIZER;
5970     struct ofproto_dpif *ofproto;
5971
5972     ofproto = ofproto_dpif_lookup(argv[1]);
5973     if (!ofproto) {
5974         unixctl_command_reply_error(conn, "no such bridge");
5975         return;
5976     }
5977
5978     flush(&ofproto->up);
5979
5980     unixctl_command_reply(conn, ds_cstr(&ds));
5981     ds_destroy(&ds);
5982 }
5983
5984 static void
5985 ofproto_dpif_unixctl_init(void)
5986 {
5987     static bool registered;
5988     if (registered) {
5989         return;
5990     }
5991     registered = true;
5992
5993     unixctl_command_register(
5994         "ofproto/trace",
5995         "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
5996         1, 3, ofproto_unixctl_trace, NULL);
5997     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5998                              ofproto_unixctl_fdb_flush, NULL);
5999     unixctl_command_register("fdb/show", "bridge", 1, 1,
6000                              ofproto_unixctl_fdb_show, NULL);
6001     unixctl_command_register("ofproto/clog", "", 0, 0,
6002                              ofproto_dpif_clog, NULL);
6003     unixctl_command_register("ofproto/unclog", "", 0, 0,
6004                              ofproto_dpif_unclog, NULL);
6005     unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6006                              ofproto_dpif_self_check, NULL);
6007     unixctl_command_register("dpif/dump-dps", "", 0, 0,
6008                              ofproto_unixctl_dpif_dump_dps, NULL);
6009     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6010                              NULL);
6011     unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6012                              ofproto_unixctl_dpif_dump_flows, NULL);
6013     unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6014                              ofproto_unixctl_dpif_del_flows, NULL);
6015     unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6016                              ofproto_unixctl_dpif_dump_megaflows, NULL);
6017     unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6018                              ofproto_unixctl_dpif_disable_megaflows, NULL);
6019     unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6020                              ofproto_unixctl_dpif_enable_megaflows, NULL);
6021 }
6022 \f
6023 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6024  *
6025  * This is deprecated.  It is only for compatibility with broken device drivers
6026  * in old versions of Linux that do not properly support VLANs when VLAN
6027  * devices are not used.  When broken device drivers are no longer in
6028  * widespread use, we will delete these interfaces. */
6029
6030 static int
6031 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6032 {
6033     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6034     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6035
6036     if (realdev_ofp_port == ofport->realdev_ofp_port
6037         && vid == ofport->vlandev_vid) {
6038         return 0;
6039     }
6040
6041     ofproto->backer->need_revalidate = REV_RECONFIGURE;
6042
6043     if (ofport->realdev_ofp_port) {
6044         vsp_remove(ofport);
6045     }
6046     if (realdev_ofp_port && ofport->bundle) {
6047         /* vlandevs are enslaved to their realdevs, so they are not allowed to
6048          * themselves be part of a bundle. */
6049         bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6050     }
6051
6052     ofport->realdev_ofp_port = realdev_ofp_port;
6053     ofport->vlandev_vid = vid;
6054
6055     if (realdev_ofp_port) {
6056         vsp_add(ofport, realdev_ofp_port, vid);
6057     }
6058
6059     return 0;
6060 }
6061
6062 static uint32_t
6063 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6064 {
6065     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6066 }
6067
6068 bool
6069 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6070     OVS_EXCLUDED(ofproto->vsp_mutex)
6071 {
6072     bool ret;
6073
6074     ovs_mutex_lock(&ofproto->vsp_mutex);
6075     ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6076     ovs_mutex_unlock(&ofproto->vsp_mutex);
6077     return ret;
6078 }
6079
6080 static ofp_port_t
6081 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6082                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6083     OVS_REQUIRES(ofproto->vsp_mutex)
6084 {
6085     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6086         int vid = vlan_tci_to_vid(vlan_tci);
6087         const struct vlan_splinter *vsp;
6088
6089         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6090                                  hash_realdev_vid(realdev_ofp_port, vid),
6091                                  &ofproto->realdev_vid_map) {
6092             if (vsp->realdev_ofp_port == realdev_ofp_port
6093                 && vsp->vid == vid) {
6094                 return vsp->vlandev_ofp_port;
6095             }
6096         }
6097     }
6098     return realdev_ofp_port;
6099 }
6100
6101 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6102  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6103  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
6104  * 'vlan_tci' 9, it would return the port number of eth0.9.
6105  *
6106  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6107  * function just returns its 'realdev_ofp_port' argument. */
6108 ofp_port_t
6109 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6110                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6111     OVS_EXCLUDED(ofproto->vsp_mutex)
6112 {
6113     ofp_port_t ret;
6114
6115     ovs_mutex_lock(&ofproto->vsp_mutex);
6116     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6117     ovs_mutex_unlock(&ofproto->vsp_mutex);
6118     return ret;
6119 }
6120
6121 static struct vlan_splinter *
6122 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6123 {
6124     struct vlan_splinter *vsp;
6125
6126     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6127                              hash_ofp_port(vlandev_ofp_port),
6128                              &ofproto->vlandev_map) {
6129         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6130             return vsp;
6131         }
6132     }
6133
6134     return NULL;
6135 }
6136
6137 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6138  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6139  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
6140  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6141  * eth0 and store 9 in '*vid'.
6142  *
6143  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6144  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
6145  * always does.*/
6146 static ofp_port_t
6147 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6148                        ofp_port_t vlandev_ofp_port, int *vid)
6149     OVS_REQUIRES(ofproto->vsp_mutex)
6150 {
6151     if (!hmap_is_empty(&ofproto->vlandev_map)) {
6152         const struct vlan_splinter *vsp;
6153
6154         vsp = vlandev_find(ofproto, vlandev_ofp_port);
6155         if (vsp) {
6156             if (vid) {
6157                 *vid = vsp->vid;
6158             }
6159             return vsp->realdev_ofp_port;
6160         }
6161     }
6162     return 0;
6163 }
6164
6165 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6166  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
6167  * 'flow->in_port' to the "real" device backing the VLAN device, sets
6168  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Otherwise (which is
6169  * always the case unless VLAN splinters are enabled), returns false without
6170  * making any changes. */
6171 bool
6172 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6173     OVS_EXCLUDED(ofproto->vsp_mutex)
6174 {
6175     ofp_port_t realdev;
6176     int vid;
6177
6178     ovs_mutex_lock(&ofproto->vsp_mutex);
6179     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6180     ovs_mutex_unlock(&ofproto->vsp_mutex);
6181     if (!realdev) {
6182         return false;
6183     }
6184
6185     /* Cause the flow to be processed as if it came in on the real device with
6186      * the VLAN device's VLAN ID. */
6187     flow->in_port.ofp_port = realdev;
6188     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6189     return true;
6190 }
6191
6192 static void
6193 vsp_remove(struct ofport_dpif *port)
6194 {
6195     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6196     struct vlan_splinter *vsp;
6197
6198     ovs_mutex_lock(&ofproto->vsp_mutex);
6199     vsp = vlandev_find(ofproto, port->up.ofp_port);
6200     if (vsp) {
6201         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6202         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6203         free(vsp);
6204
6205         port->realdev_ofp_port = 0;
6206     } else {
6207         VLOG_ERR("missing vlan device record");
6208     }
6209     ovs_mutex_unlock(&ofproto->vsp_mutex);
6210 }
6211
6212 static void
6213 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6214 {
6215     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6216
6217     ovs_mutex_lock(&ofproto->vsp_mutex);
6218     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6219         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6220             == realdev_ofp_port)) {
6221         struct vlan_splinter *vsp;
6222
6223         vsp = xmalloc(sizeof *vsp);
6224         vsp->realdev_ofp_port = realdev_ofp_port;
6225         vsp->vlandev_ofp_port = port->up.ofp_port;
6226         vsp->vid = vid;
6227
6228         port->realdev_ofp_port = realdev_ofp_port;
6229
6230         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6231                     hash_ofp_port(port->up.ofp_port));
6232         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6233                     hash_realdev_vid(realdev_ofp_port, vid));
6234     } else {
6235         VLOG_ERR("duplicate vlan device record");
6236     }
6237     ovs_mutex_unlock(&ofproto->vsp_mutex);
6238 }
6239
6240 static odp_port_t
6241 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6242 {
6243     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6244     return ofport ? ofport->odp_port : ODPP_NONE;
6245 }
6246
6247 struct ofport_dpif *
6248 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6249 {
6250     struct ofport_dpif *port;
6251
6252     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6253     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6254                              &backer->odp_to_ofport_map) {
6255         if (port->odp_port == odp_port) {
6256             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6257             return port;
6258         }
6259     }
6260
6261     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6262     return NULL;
6263 }
6264
6265 static ofp_port_t
6266 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6267 {
6268     struct ofport_dpif *port;
6269
6270     port = odp_port_to_ofport(ofproto->backer, odp_port);
6271     if (port && &ofproto->up == port->up.ofproto) {
6272         return port->up.ofp_port;
6273     } else {
6274         return OFPP_NONE;
6275     }
6276 }
6277
6278 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6279  * most heavily weighted element.  'base' designates the rate of decay: after
6280  * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6281  * (about .37). */
6282 static void
6283 exp_mavg(double *avg, int base, double new)
6284 {
6285     *avg = (*avg * (base - 1) + new) / base;
6286 }
6287
6288 static void
6289 update_moving_averages(struct dpif_backer *backer)
6290 {
6291     const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6292     long long int minutes = (time_msec() - backer->created) / min_ms;
6293
6294     if (minutes > 0) {
6295         backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6296             / minutes;
6297         backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6298             / minutes;
6299     } else {
6300         backer->lifetime.add_rate = 0.0;
6301         backer->lifetime.del_rate = 0.0;
6302     }
6303
6304     /* Update hourly averages on the minute boundaries. */
6305     if (time_msec() - backer->last_minute >= min_ms) {
6306         exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6307         exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6308
6309         /* Update daily averages on the hour boundaries. */
6310         if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6311             exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6312             exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6313         }
6314
6315         backer->total_subfacet_add_count += backer->subfacet_add_count;
6316         backer->total_subfacet_del_count += backer->subfacet_del_count;
6317         backer->subfacet_add_count = 0;
6318         backer->subfacet_del_count = 0;
6319         backer->last_minute += min_ms;
6320     }
6321 }
6322
6323 const struct ofproto_class ofproto_dpif_class = {
6324     init,
6325     enumerate_types,
6326     enumerate_names,
6327     del,
6328     port_open_type,
6329     type_run,
6330     type_run_fast,
6331     type_wait,
6332     alloc,
6333     construct,
6334     destruct,
6335     dealloc,
6336     run,
6337     run_fast,
6338     wait,
6339     get_memory_usage,
6340     flush,
6341     get_features,
6342     get_tables,
6343     port_alloc,
6344     port_construct,
6345     port_destruct,
6346     port_dealloc,
6347     port_modified,
6348     port_reconfigured,
6349     port_query_by_name,
6350     port_add,
6351     port_del,
6352     port_get_stats,
6353     port_dump_start,
6354     port_dump_next,
6355     port_dump_done,
6356     port_poll,
6357     port_poll_wait,
6358     port_is_lacp_current,
6359     NULL,                       /* rule_choose_table */
6360     rule_alloc,
6361     rule_construct,
6362     rule_insert,
6363     rule_delete,
6364     rule_destruct,
6365     rule_dealloc,
6366     rule_get_stats,
6367     rule_execute,
6368     rule_modify_actions,
6369     set_frag_handling,
6370     packet_out,
6371     set_netflow,
6372     get_netflow_ids,
6373     set_sflow,
6374     set_ipfix,
6375     set_cfm,
6376     get_cfm_status,
6377     set_bfd,
6378     get_bfd_status,
6379     set_stp,
6380     get_stp_status,
6381     set_stp_port,
6382     get_stp_port_status,
6383     set_queues,
6384     bundle_set,
6385     bundle_remove,
6386     mirror_set__,
6387     mirror_get_stats__,
6388     set_flood_vlans,
6389     is_mirror_output_bundle,
6390     forward_bpdu_changed,
6391     set_mac_table_config,
6392     set_realdev,
6393     NULL,                       /* meter_get_features */
6394     NULL,                       /* meter_set */
6395     NULL,                       /* meter_get */
6396     NULL,                       /* meter_del */
6397 };