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 "guarded-list.h"
  35 #include "hmapx.h"
  36 #include "lacp.h"
  37 #include "learn.h"
  38 #include "mac-learning.h"
  39 #include "meta-flow.h"
  40 #include "multipath.h"
  41 #include "netdev-vport.h"
  42 #include "netdev.h"
  43 #include "netlink.h"
  44 #include "nx-match.h"
  45 #include "odp-util.h"
  46 #include "odp-execute.h"
  47 #include "ofp-util.h"
  48 #include "ofpbuf.h"
  49 #include "ofp-actions.h"
  50 #include "ofp-parse.h"
  51 #include "ofp-print.h"
  52 #include "ofproto-dpif-governor.h"
  53 #include "ofproto-dpif-ipfix.h"
  54 #include "ofproto-dpif-mirror.h"
  55 #include "ofproto-dpif-monitor.h"
  56 #include "ofproto-dpif-sflow.h"
  57 #include "ofproto-dpif-upcall.h"
  58 #include "ofproto-dpif-xlate.h"
  59 #include "poll-loop.h"
  60 #include "simap.h"
  61 #include "smap.h"
  62 #include "timer.h"
  63 #include "tunnel.h"
  64 #include "unaligned.h"
  65 #include "unixctl.h"
  66 #include "vlan-bitmap.h"
  67 #include "vlog.h"
  68
  69 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  70
  71 COVERAGE_DEFINE(ofproto_dpif_expired);
  72 COVERAGE_DEFINE(facet_revalidate);
  73 COVERAGE_DEFINE(facet_unexpected);
  74 COVERAGE_DEFINE(facet_create);
  75 COVERAGE_DEFINE(facet_remove);
  76 COVERAGE_DEFINE(subfacet_create);
  77 COVERAGE_DEFINE(subfacet_destroy);
  78 COVERAGE_DEFINE(subfacet_install_fail);
  79 COVERAGE_DEFINE(packet_in_overflow);
  80
  81 /* Number of implemented OpenFlow tables. */
  82 enum { N_TABLES = 255 };
  83 enum { TBL_INTERNAL = N_TABLES - 1 };    /* Used for internal hidden rules. */
  84 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
  85
  86 struct flow_miss;
  87 struct facet;
  88
  89 struct rule_dpif {
  90     struct rule up;
  91
  92     /* These statistics:
  93      *
  94      *   - Do include packets and bytes from facets that have been deleted or
  95      *     whose own statistics have been folded into the rule.
  96      *
  97      *   - Do include packets and bytes sent "by hand" that were accounted to
  98      *     the rule without any facet being involved (this is a rare corner
  99      *     case in rule_execute()).
 100      *
 101      *   - Do not include packet or bytes that can be obtained from any facet's
 102      *     packet_count or byte_count member or that can be obtained from the
 103      *     datapath by, e.g., dpif_flow_get() for any subfacet.
 104      */
 105     struct ovs_mutex stats_mutex;
 106     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 107     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 108 };
 109
 110 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
 111 static struct rule_dpif *rule_dpif_cast(const struct rule *);
 112
 113 struct group_dpif {
 114     struct ofgroup up;
 115
 116     /* These statistics:
 117      *
 118      *   - Do include packets and bytes from facets that have been deleted or
 119      *     whose own statistics have been folded into the rule.
 120      *
 121      *   - Do include packets and bytes sent "by hand" that were accounted to
 122      *     the rule without any facet being involved (this is a rare corner
 123      *     case in rule_execute()).
 124      *
 125      *   - Do not include packet or bytes that can be obtained from any facet's
 126      *     packet_count or byte_count member or that can be obtained from the
 127      *     datapath by, e.g., dpif_flow_get() for any subfacet.
 128      */
 129     struct ovs_mutex stats_mutex;
 130     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 131     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 132     struct bucket_counter *bucket_stats OVS_GUARDED;  /* Bucket statistics. */
 133 };
 134
 135 struct ofbundle {
 136     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
 137     struct ofproto_dpif *ofproto; /* Owning ofproto. */
 138     void *aux;                  /* Key supplied by ofproto's client. */
 139     char *name;                 /* Identifier for log messages. */
 140
 141     /* Configuration. */
 142     struct list ports;          /* Contains "struct ofport"s. */
 143     enum port_vlan_mode vlan_mode; /* VLAN mode */
 144     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
 145     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
 146                                  * NULL if all VLANs are trunked. */
 147     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 148     struct bond *bond;          /* Nonnull iff more than one port. */
 149     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 150
 151     /* Status. */
 152     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 153 };
 154
 155 static void bundle_remove(struct ofport *);
 156 static void bundle_update(struct ofbundle *);
 157 static void bundle_destroy(struct ofbundle *);
 158 static void bundle_del_port(struct ofport_dpif *);
 159 static void bundle_run(struct ofbundle *);
 160 static void bundle_wait(struct ofbundle *);
 161
 162 static void stp_run(struct ofproto_dpif *ofproto);
 163 static void stp_wait(struct ofproto_dpif *ofproto);
 164 static int set_stp_port(struct ofport *,
 165                         const struct ofproto_port_stp_settings *);
 166
 167 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
 168                               enum slow_path_reason,
 169                               uint64_t *stub, size_t stub_size,
 170                               const struct nlattr **actionsp,
 171                               size_t *actions_lenp);
 172
 173 /* A subfacet (see "struct subfacet" below) has three possible installation
 174  * states:
 175  *
 176  *   - SF_NOT_INSTALLED: Not installed in the datapath.  This will only be the
 177  *     case just after the subfacet is created, just before the subfacet is
 178  *     destroyed, or if the datapath returns an error when we try to install a
 179  *     subfacet.
 180  *
 181  *   - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
 182  *
 183  *   - SF_SLOW_PATH: An action that sends every packet for the subfacet through
 184  *     ofproto_dpif is installed in the datapath.
 185  */
 186 enum subfacet_path {
 187     SF_NOT_INSTALLED,           /* No datapath flow for this subfacet. */
 188     SF_FAST_PATH,               /* Full actions are installed. */
 189     SF_SLOW_PATH,               /* Send-to-userspace action is installed. */
 190 };
 191
 192 /* A dpif flow and actions associated with a facet.
 193  *
 194  * See also the large comment on struct facet. */
 195 struct subfacet {
 196     /* Owners. */
 197     struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
 198     struct list list_node;      /* In struct facet's 'facets' list. */
 199     struct facet *facet;        /* Owning facet. */
 200     struct dpif_backer *backer; /* Owning backer. */
 201
 202     struct nlattr *key;
 203     int key_len;
 204
 205     long long int used;         /* Time last used; time created if not used. */
 206     long long int created;      /* Time created. */
 207
 208     uint64_t dp_packet_count;   /* Last known packet count in the datapath. */
 209     uint64_t dp_byte_count;     /* Last known byte count in the datapath. */
 210
 211     enum subfacet_path path;    /* Installed in datapath? */
 212 };
 213
 214 #define SUBFACET_DESTROY_MAX_BATCH 50
 215
 216 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *,
 217                                         uint32_t key_hash);
 218 static struct subfacet *subfacet_find(struct dpif_backer *,
 219                                       const struct nlattr *key, size_t key_len,
 220                                       uint32_t key_hash);
 221 static void subfacet_destroy(struct subfacet *);
 222 static void subfacet_destroy__(struct subfacet *);
 223 static void subfacet_destroy_batch(struct dpif_backer *,
 224                                    struct subfacet **, int n);
 225 static void subfacet_reset_dp_stats(struct subfacet *,
 226                                     struct dpif_flow_stats *);
 227 static void subfacet_update_stats(struct subfacet *,
 228                                   const struct dpif_flow_stats *);
 229 static int subfacet_install(struct subfacet *,
 230                             const struct ofpbuf *odp_actions,
 231                             struct dpif_flow_stats *);
 232 static void subfacet_uninstall(struct subfacet *);
 233
 234 /* A unique, non-overlapping instantiation of an OpenFlow flow.
 235  *
 236  * A facet associates a "struct flow", which represents the Open vSwitch
 237  * userspace idea of an exact-match flow, with one or more subfacets.
 238  * While the facet is created based on an exact-match flow, it is stored
 239  * within the ofproto based on the wildcards that could be expressed
 240  * based on the flow table and other configuration.  (See the 'wc'
 241  * description in "struct xlate_out" for more details.)
 242  *
 243  * Each subfacet tracks the datapath's idea of the flow equivalent to
 244  * the facet.  When the kernel module (or other dpif implementation) and
 245  * Open vSwitch userspace agree on the definition of a flow key, there
 246  * is exactly one subfacet per facet.  If the dpif implementation
 247  * supports more-specific flow matching than userspace, however, a facet
 248  * can have more than one subfacet.  Examples include the dpif
 249  * implementation not supporting the same wildcards as userspace or some
 250  * distinction in flow that userspace simply doesn't understand.
 251  *
 252  * Flow expiration works in terms of subfacets, so a facet must have at
 253  * least one subfacet or it will never expire, leaking memory. */
 254 struct facet {
 255     /* Owner. */
 256     struct ofproto_dpif *ofproto;
 257
 258     /* Owned data. */
 259     struct list subfacets;
 260     long long int used;         /* Time last used; time created if not used. */
 261
 262     /* Key. */
 263     struct flow flow;           /* Flow of the creating subfacet. */
 264     struct cls_rule cr;         /* In 'ofproto_dpif's facets classifier. */
 265
 266     /* These statistics:
 267      *
 268      *   - Do include packets and bytes sent "by hand", e.g. with
 269      *     dpif_execute().
 270      *
 271      *   - Do include packets and bytes that were obtained from the datapath
 272      *     when a subfacet's statistics were reset (e.g. dpif_flow_put() with
 273      *     DPIF_FP_ZERO_STATS).
 274      *
 275      *   - Do not include packets or bytes that can be obtained from the
 276      *     datapath for any existing subfacet.
 277      */
 278     uint64_t packet_count;       /* Number of packets received. */
 279     uint64_t byte_count;         /* Number of bytes received. */
 280
 281     /* Resubmit statistics. */
 282     uint64_t prev_packet_count;  /* Number of packets from last stats push. */
 283     uint64_t prev_byte_count;    /* Number of bytes from last stats push. */
 284     long long int prev_used;     /* Used time from last stats push. */
 285
 286     /* Accounting. */
 287     uint64_t accounted_bytes;    /* Bytes processed by facet_account(). */
 288     struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
 289     uint16_t tcp_flags;          /* TCP flags seen for this 'rule'. */
 290
 291     struct xlate_out xout;
 292
 293     /* Storage for a single subfacet, to reduce malloc() time and space
 294      * overhead.  (A facet always has at least one subfacet and in the common
 295      * case has exactly one subfacet.  However, 'one_subfacet' may not
 296      * always be valid, since it could have been removed after newer
 297      * subfacets were pushed onto the 'subfacets' list.) */
 298     struct subfacet one_subfacet;
 299
 300     long long int learn_rl;      /* Rate limiter for facet_learn(). */
 301 };
 302
 303 static struct facet *facet_create(const struct flow_miss *);
 304 static void facet_remove(struct facet *);
 305 static void facet_free(struct facet *);
 306
 307 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
 308 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
 309                                         const struct flow *);
 310 static bool facet_revalidate(struct facet *);
 311 static bool facet_check_consistency(struct facet *);
 312
 313 static void facet_flush_stats(struct facet *);
 314
 315 static void facet_reset_counters(struct facet *);
 316 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
 317                             struct dpif_flow_stats *, bool may_learn);
 318 static void facet_push_stats(struct facet *, bool may_learn);
 319 static void facet_learn(struct facet *);
 320 static void facet_account(struct facet *);
 321 static void push_all_stats(void);
 322
 323 static bool facet_is_controller_flow(struct facet *);
 324
 325 struct ofport_dpif {
 326     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 327     struct ofport up;
 328
 329     odp_port_t odp_port;
 330     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 331     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
 332     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 333     struct bfd *bfd;            /* BFD, if any. */
 334     bool may_enable;            /* May be enabled in bonds. */
 335     bool is_tunnel;             /* This port is a tunnel. */
 336     bool is_layer3;             /* This is a layer 3 port. */
 337     long long int carrier_seq;  /* Carrier status changes. */
 338     struct ofport_dpif *peer;   /* Peer if patch port. */
 339
 340     /* Spanning tree. */
 341     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 342     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 343     long long int stp_state_entered;
 344
 345     /* Queue to DSCP mapping. */
 346     struct ofproto_port_queue *qdscp;
 347     size_t n_qdscp;
 348
 349     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 350      *
 351      * This is deprecated.  It is only for compatibility with broken device
 352      * drivers in old versions of Linux that do not properly support VLANs when
 353      * VLAN devices are not used.  When broken device drivers are no longer in
 354      * widespread use, we will delete these interfaces. */
 355     ofp_port_t realdev_ofp_port;
 356     int vlandev_vid;
 357 };
 358
 359 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 360  *
 361  * This is deprecated.  It is only for compatibility with broken device drivers
 362  * in old versions of Linux that do not properly support VLANs when VLAN
 363  * devices are not used.  When broken device drivers are no longer in
 364  * widespread use, we will delete these interfaces. */
 365 struct vlan_splinter {
 366     struct hmap_node realdev_vid_node;
 367     struct hmap_node vlandev_node;
 368     ofp_port_t realdev_ofp_port;
 369     ofp_port_t vlandev_ofp_port;
 370     int vid;
 371 };
 372
 373 static void vsp_remove(struct ofport_dpif *);
 374 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 375
 376 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 377                                        ofp_port_t);
 378
 379 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 380                                        odp_port_t);
 381
 382 static struct ofport_dpif *
 383 ofport_dpif_cast(const struct ofport *ofport)
 384 {
 385     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 386 }
 387
 388 static void port_run(struct ofport_dpif *);
 389 static int set_bfd(struct ofport *, const struct smap *);
 390 static int set_cfm(struct ofport *, const struct cfm_settings *);
 391 static void ofport_update_peer(struct ofport_dpif *);
 392 static void run_fast_rl(void);
 393 static int run_fast(struct ofproto *);
 394
 395 struct dpif_completion {
 396     struct list list_node;
 397     struct ofoperation *op;
 398 };
 399
 400 /* Reasons that we might need to revalidate every facet, and corresponding
 401  * coverage counters.
 402  *
 403  * A value of 0 means that there is no need to revalidate.
 404  *
 405  * It would be nice to have some cleaner way to integrate with coverage
 406  * counters, but with only a few reasons I guess this is good enough for
 407  * now. */
 408 enum revalidate_reason {
 409     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 410     REV_STP,                   /* Spanning tree protocol port status change. */
 411     REV_BOND,                  /* Bonding changed. */
 412     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 413     REV_FLOW_TABLE,            /* Flow table changed. */
 414     REV_MAC_LEARNING,          /* Mac learning changed. */
 415     REV_INCONSISTENCY          /* Facet self-check failed. */
 416 };
 417 COVERAGE_DEFINE(rev_reconfigure);
 418 COVERAGE_DEFINE(rev_stp);
 419 COVERAGE_DEFINE(rev_bond);
 420 COVERAGE_DEFINE(rev_port_toggled);
 421 COVERAGE_DEFINE(rev_flow_table);
 422 COVERAGE_DEFINE(rev_mac_learning);
 423 COVERAGE_DEFINE(rev_inconsistency);
 424
 425 /* All datapaths of a given type share a single dpif backer instance. */
 426 struct dpif_backer {
 427     char *type;
 428     int refcount;
 429     struct dpif *dpif;
 430     struct udpif *udpif;
 431     struct timer next_expiration;
 432
 433     struct ovs_rwlock odp_to_ofport_lock;
 434     struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
 435
 436     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 437
 438     /* Facet revalidation flags applying to facets which use this backer. */
 439     enum revalidate_reason need_revalidate; /* Revalidate every facet. */
 440
 441     struct hmap drop_keys; /* Set of dropped odp keys. */
 442     bool recv_set_enable; /* Enables or disables receiving packets. */
 443
 444     struct hmap subfacets;
 445     struct governor *governor;
 446
 447     /* Subfacet statistics.
 448      *
 449      * These keep track of the total number of subfacets added and deleted and
 450      * flow life span.  They are useful for computing the flow rates stats
 451      * exposed via "ovs-appctl dpif/show".  The goal is to learn about
 452      * traffic patterns in ways that we can use later to improve Open vSwitch
 453      * performance in new situations.  */
 454     unsigned max_n_subfacet;         /* Maximum number of flows */
 455     unsigned avg_n_subfacet;         /* Average number of flows. */
 456     long long int avg_subfacet_life; /* Average life span of subfacets. */
 457
 458     /* Number of upcall handling threads. */
 459     unsigned int n_handler_threads;
 460 };
 461
 462 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 463 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 464
 465 static void drop_key_clear(struct dpif_backer *);
 466
 467 struct ofproto_dpif {
 468     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 469     struct ofproto up;
 470     struct dpif_backer *backer;
 471
 472     /* Special OpenFlow rules. */
 473     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 474     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 475     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 476
 477     /* Bridging. */
 478     struct netflow *netflow;
 479     struct dpif_sflow *sflow;
 480     struct dpif_ipfix *ipfix;
 481     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 482     struct mac_learning *ml;
 483     bool has_bonded_bundles;
 484     struct mbridge *mbridge;
 485
 486     /* Facets. */
 487     struct classifier facets;     /* Contains 'struct facet's. */
 488     long long int consistency_rl;
 489
 490     struct ovs_mutex stats_mutex;
 491     struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
 492                                             * consumed in userspace. */
 493
 494     /* Spanning tree. */
 495     struct stp *stp;
 496     long long int stp_last_tick;
 497
 498     /* VLAN splinters. */
 499     struct ovs_mutex vsp_mutex;
 500     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 501     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 502
 503     /* Ports. */
 504     struct sset ports;             /* Set of standard port names. */
 505     struct sset ghost_ports;       /* Ports with no datapath port. */
 506     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 507     int port_poll_errno;           /* Last errno for port_poll() reply. */
 508
 509     /* Per ofproto's dpif stats. */
 510     uint64_t n_hit;
 511     uint64_t n_missed;
 512
 513     /* Work queues. */
 514     struct guarded_list pins;      /* Contains "struct ofputil_packet_in"s. */
 515 };
 516
 517 /* By default, flows in the datapath are wildcarded (megaflows).  They
 518  * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
 519 static bool enable_megaflows = true;
 520
 521 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 522 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 523
 524 static void ofproto_dpif_unixctl_init(void);
 525
 526 static inline struct ofproto_dpif *
 527 ofproto_dpif_cast(const struct ofproto *ofproto)
 528 {
 529     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 530     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 531 }
 532
 533 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
 534                                         ofp_port_t ofp_port);
 535 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
 536                           const struct ofpbuf *packet,
 537                           const struct ofpact[], size_t ofpacts_len,
 538                           struct ds *);
 539
 540 /* Upcalls. */
 541 static void handle_upcalls(struct dpif_backer *);
 542
 543 /* Flow expiration. */
 544 static int expire(struct dpif_backer *);
 545
 546 /* NetFlow. */
 547 static void send_netflow_active_timeouts(struct ofproto_dpif *);
 548
 549 /* Global variables. */
 550 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 551
 552 /* Initial mappings of port to bridge mappings. */
 553 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 554
 555 /* Executes 'fm'.  The caller retains ownership of 'fm' and everything in
 556  * it. */
 557 void
 558 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 559                       struct ofputil_flow_mod *fm)
 560 {
 561     ofproto_flow_mod(&ofproto->up, fm);
 562 }
 563
 564 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 565  * Takes ownership of 'pin' and pin->packet. */
 566 void
 567 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 568                             struct ofproto_packet_in *pin)
 569 {
 570     if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
 571         COVERAGE_INC(packet_in_overflow);
 572         free(CONST_CAST(void *, pin->up.packet));
 573         free(pin);
 574     }
 575 }
 576 \f
 577 /* Factory functions. */
 578
 579 static void
 580 init(const struct shash *iface_hints)
 581 {
 582     struct shash_node *node;
 583
 584     /* Make a local copy, since we don't own 'iface_hints' elements. */
 585     SHASH_FOR_EACH(node, iface_hints) {
 586         const struct iface_hint *orig_hint = node->data;
 587         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 588
 589         new_hint->br_name = xstrdup(orig_hint->br_name);
 590         new_hint->br_type = xstrdup(orig_hint->br_type);
 591         new_hint->ofp_port = orig_hint->ofp_port;
 592
 593         shash_add(&init_ofp_ports, node->name, new_hint);
 594     }
 595 }
 596
 597 static void
 598 enumerate_types(struct sset *types)
 599 {
 600     dp_enumerate_types(types);
 601 }
 602
 603 static int
 604 enumerate_names(const char *type, struct sset *names)
 605 {
 606     struct ofproto_dpif *ofproto;
 607
 608     sset_clear(names);
 609     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 610         if (strcmp(type, ofproto->up.type)) {
 611             continue;
 612         }
 613         sset_add(names, ofproto->up.name);
 614     }
 615
 616     return 0;
 617 }
 618
 619 static int
 620 del(const char *type, const char *name)
 621 {
 622     struct dpif *dpif;
 623     int error;
 624
 625     error = dpif_open(name, type, &dpif);
 626     if (!error) {
 627         error = dpif_delete(dpif);
 628         dpif_close(dpif);
 629     }
 630     return error;
 631 }
 632 \f
 633 static const char *
 634 port_open_type(const char *datapath_type, const char *port_type)
 635 {
 636     return dpif_port_open_type(datapath_type, port_type);
 637 }
 638
 639 /* Type functions. */
 640
 641 static void process_dpif_port_changes(struct dpif_backer *);
 642 static void process_dpif_all_ports_changed(struct dpif_backer *);
 643 static void process_dpif_port_change(struct dpif_backer *,
 644                                      const char *devname);
 645 static void process_dpif_port_error(struct dpif_backer *, int error);
 646
 647 static struct ofproto_dpif *
 648 lookup_ofproto_dpif_by_port_name(const char *name)
 649 {
 650     struct ofproto_dpif *ofproto;
 651
 652     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 653         if (sset_contains(&ofproto->ports, name)) {
 654             return ofproto;
 655         }
 656     }
 657
 658     return NULL;
 659 }
 660
 661 static int
 662 type_run(const char *type)
 663 {
 664     static long long int push_timer = LLONG_MIN;
 665     struct dpif_backer *backer;
 666
 667     backer = shash_find_data(&all_dpif_backers, type);
 668     if (!backer) {
 669         /* This is not necessarily a problem, since backers are only
 670          * created on demand. */
 671         return 0;
 672     }
 673
 674     dpif_run(backer->dpif);
 675
 676     /* The most natural place to push facet statistics is when they're pulled
 677      * from the datapath.  However, when there are many flows in the datapath,
 678      * this expensive operation can occur so frequently, that it reduces our
 679      * ability to quickly set up flows.  To reduce the cost, we push statistics
 680      * here instead. */
 681     if (time_msec() > push_timer) {
 682         push_timer = time_msec() + 2000;
 683         push_all_stats();
 684     }
 685
 686     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 687      * and the configuration has now changed to "false", enable receiving
 688      * packets from the datapath. */
 689     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 690         int error;
 691
 692         backer->recv_set_enable = true;
 693
 694         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 695         if (error) {
 696             udpif_recv_set(backer->udpif, 0, false);
 697             VLOG_ERR("Failed to enable receiving packets in dpif.");
 698             return error;
 699         }
 700         udpif_recv_set(backer->udpif, n_handler_threads,
 701                        backer->recv_set_enable);
 702         dpif_flow_flush(backer->dpif);
 703         backer->need_revalidate = REV_RECONFIGURE;
 704     }
 705
 706     /* If the n_handler_threads is reconfigured, call udpif_recv_set()
 707      * to reset the handler threads. */
 708     if (backer->n_handler_threads != n_handler_threads) {
 709         udpif_recv_set(backer->udpif, n_handler_threads,
 710                        backer->recv_set_enable);
 711         backer->n_handler_threads = n_handler_threads;
 712     }
 713
 714     if (backer->need_revalidate) {
 715         struct ofproto_dpif *ofproto;
 716         struct simap_node *node;
 717         struct simap tmp_backers;
 718
 719         /* Handle tunnel garbage collection. */
 720         simap_init(&tmp_backers);
 721         simap_swap(&backer->tnl_backers, &tmp_backers);
 722
 723         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 724             struct ofport_dpif *iter;
 725
 726             if (backer != ofproto->backer) {
 727                 continue;
 728             }
 729
 730             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 731                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 732                 const char *dp_port;
 733
 734                 if (!iter->is_tunnel) {
 735                     continue;
 736                 }
 737
 738                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 739                                                      namebuf, sizeof namebuf);
 740                 node = simap_find(&tmp_backers, dp_port);
 741                 if (node) {
 742                     simap_put(&backer->tnl_backers, dp_port, node->data);
 743                     simap_delete(&tmp_backers, node);
 744                     node = simap_find(&backer->tnl_backers, dp_port);
 745                 } else {
 746                     node = simap_find(&backer->tnl_backers, dp_port);
 747                     if (!node) {
 748                         odp_port_t odp_port = ODPP_NONE;
 749
 750                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 751                                            &odp_port)) {
 752                             simap_put(&backer->tnl_backers, dp_port,
 753                                       odp_to_u32(odp_port));
 754                             node = simap_find(&backer->tnl_backers, dp_port);
 755                         }
 756                     }
 757                 }
 758
 759                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 760                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 761                                          iter->odp_port)) {
 762                     backer->need_revalidate = REV_RECONFIGURE;
 763                 }
 764             }
 765         }
 766
 767         SIMAP_FOR_EACH (node, &tmp_backers) {
 768             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 769         }
 770         simap_destroy(&tmp_backers);
 771
 772         switch (backer->need_revalidate) {
 773         case REV_RECONFIGURE:   COVERAGE_INC(rev_reconfigure);   break;
 774         case REV_STP:           COVERAGE_INC(rev_stp);           break;
 775         case REV_BOND:          COVERAGE_INC(rev_bond);          break;
 776         case REV_PORT_TOGGLED:  COVERAGE_INC(rev_port_toggled);  break;
 777         case REV_FLOW_TABLE:    COVERAGE_INC(rev_flow_table);    break;
 778         case REV_MAC_LEARNING:  COVERAGE_INC(rev_mac_learning);  break;
 779         case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
 780         }
 781         backer->need_revalidate = 0;
 782
 783         /* Clear the drop_keys in case we should now be accepting some
 784          * formerly dropped flows. */
 785         drop_key_clear(backer);
 786
 787         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 788             struct facet *facet, *next;
 789             struct ofport_dpif *ofport;
 790             struct cls_cursor cursor;
 791             struct ofbundle *bundle;
 792
 793             if (ofproto->backer != backer) {
 794                 continue;
 795             }
 796
 797             ovs_rwlock_wrlock(&xlate_rwlock);
 798             xlate_ofproto_set(ofproto, ofproto->up.name,
 799                               ofproto->backer->dpif, ofproto->miss_rule,
 800                               ofproto->no_packet_in_rule, ofproto->ml,
 801                               ofproto->stp, ofproto->mbridge,
 802                               ofproto->sflow, ofproto->ipfix,
 803                               ofproto->up.frag_handling,
 804                               ofproto->up.forward_bpdu,
 805                               connmgr_has_in_band(ofproto->up.connmgr),
 806                               ofproto->netflow != NULL);
 807
 808             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 809                 xlate_bundle_set(ofproto, bundle, bundle->name,
 810                                  bundle->vlan_mode, bundle->vlan,
 811                                  bundle->trunks, bundle->use_priority_tags,
 812                                  bundle->bond, bundle->lacp,
 813                                  bundle->floodable);
 814             }
 815
 816             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 817                 int stp_port = ofport->stp_port
 818                     ? stp_port_no(ofport->stp_port)
 819                     : -1;
 820                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 821                                  ofport->up.ofp_port, ofport->odp_port,
 822                                  ofport->up.netdev, ofport->cfm,
 823                                  ofport->bfd, ofport->peer, stp_port,
 824                                  ofport->qdscp, ofport->n_qdscp,
 825                                  ofport->up.pp.config, ofport->up.pp.state,
 826                                  ofport->is_tunnel, ofport->may_enable);
 827             }
 828             ovs_rwlock_unlock(&xlate_rwlock);
 829
 830             /* Only ofproto-dpif cares about the facet classifier so we just
 831              * lock cls_cursor_init() to appease the thread safety analysis. */
 832             ovs_rwlock_rdlock(&ofproto->facets.rwlock);
 833             cls_cursor_init(&cursor, &ofproto->facets, NULL);
 834             ovs_rwlock_unlock(&ofproto->facets.rwlock);
 835             CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
 836                 facet_revalidate(facet);
 837                 run_fast_rl();
 838             }
 839         }
 840
 841         udpif_revalidate(backer->udpif);
 842     }
 843
 844     if (!backer->recv_set_enable) {
 845         /* Wake up before a max of 1000ms. */
 846         timer_set_duration(&backer->next_expiration, 1000);
 847     } else if (timer_expired(&backer->next_expiration)) {
 848         int delay = expire(backer);
 849         timer_set_duration(&backer->next_expiration, delay);
 850     }
 851
 852     process_dpif_port_changes(backer);
 853
 854     if (backer->governor) {
 855         size_t n_subfacets;
 856
 857         governor_run(backer->governor);
 858
 859         /* If the governor has shrunk to its minimum size and the number of
 860          * subfacets has dwindled, then drop the governor entirely.
 861          *
 862          * For hysteresis, the number of subfacets to drop the governor is
 863          * smaller than the number needed to trigger its creation. */
 864         n_subfacets = hmap_count(&backer->subfacets);
 865         if (n_subfacets * 4 < flow_eviction_threshold
 866             && governor_is_idle(backer->governor)) {
 867             governor_destroy(backer->governor);
 868             backer->governor = NULL;
 869         }
 870     }
 871
 872     return 0;
 873 }
 874
 875 /* Check for and handle port changes in 'backer''s dpif. */
 876 static void
 877 process_dpif_port_changes(struct dpif_backer *backer)
 878 {
 879     for (;;) {
 880         char *devname;
 881         int error;
 882
 883         error = dpif_port_poll(backer->dpif, &devname);
 884         switch (error) {
 885         case EAGAIN:
 886             return;
 887
 888         case ENOBUFS:
 889             process_dpif_all_ports_changed(backer);
 890             break;
 891
 892         case 0:
 893             process_dpif_port_change(backer, devname);
 894             free(devname);
 895             break;
 896
 897         default:
 898             process_dpif_port_error(backer, error);
 899             break;
 900         }
 901     }
 902 }
 903
 904 static void
 905 process_dpif_all_ports_changed(struct dpif_backer *backer)
 906 {
 907     struct ofproto_dpif *ofproto;
 908     struct dpif_port dpif_port;
 909     struct dpif_port_dump dump;
 910     struct sset devnames;
 911     const char *devname;
 912
 913     sset_init(&devnames);
 914     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 915         if (ofproto->backer == backer) {
 916             struct ofport *ofport;
 917
 918             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 919                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 920             }
 921         }
 922     }
 923     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 924         sset_add(&devnames, dpif_port.name);
 925     }
 926
 927     SSET_FOR_EACH (devname, &devnames) {
 928         process_dpif_port_change(backer, devname);
 929     }
 930     sset_destroy(&devnames);
 931 }
 932
 933 static void
 934 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 935 {
 936     struct ofproto_dpif *ofproto;
 937     struct dpif_port port;
 938
 939     /* Don't report on the datapath's device. */
 940     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 941         return;
 942     }
 943
 944     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 945                    &all_ofproto_dpifs) {
 946         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 947             return;
 948         }
 949     }
 950
 951     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 952     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 953         /* The port was removed.  If we know the datapath,
 954          * report it through poll_set().  If we don't, it may be
 955          * notifying us of a removal we initiated, so ignore it.
 956          * If there's a pending ENOBUFS, let it stand, since
 957          * everything will be reevaluated. */
 958         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 959             sset_add(&ofproto->port_poll_set, devname);
 960             ofproto->port_poll_errno = 0;
 961         }
 962     } else if (!ofproto) {
 963         /* The port was added, but we don't know with which
 964          * ofproto we should associate it.  Delete it. */
 965         dpif_port_del(backer->dpif, port.port_no);
 966     } else {
 967         struct ofport_dpif *ofport;
 968
 969         ofport = ofport_dpif_cast(shash_find_data(
 970                                       &ofproto->up.port_by_name, devname));
 971         if (ofport
 972             && ofport->odp_port != port.port_no
 973             && !odp_port_to_ofport(backer, port.port_no))
 974         {
 975             /* 'ofport''s datapath port number has changed from
 976              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 977              * structures to match. */
 978             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 979             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 980             ofport->odp_port = port.port_no;
 981             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 982                         hash_odp_port(port.port_no));
 983             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 984             backer->need_revalidate = REV_RECONFIGURE;
 985         }
 986     }
 987     dpif_port_destroy(&port);
 988 }
 989
 990 /* Propagate 'error' to all ofprotos based on 'backer'. */
 991 static void
 992 process_dpif_port_error(struct dpif_backer *backer, int error)
 993 {
 994     struct ofproto_dpif *ofproto;
 995
 996     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 997         if (ofproto->backer == backer) {
 998             sset_clear(&ofproto->port_poll_set);
 999             ofproto->port_poll_errno = error;
1000         }
1001     }
1002 }
1003
1004 static int
1005 dpif_backer_run_fast(struct dpif_backer *backer)
1006 {
1007     handle_upcalls(backer);
1008
1009     return 0;
1010 }
1011
1012 static int
1013 type_run_fast(const char *type)
1014 {
1015     struct dpif_backer *backer;
1016
1017     backer = shash_find_data(&all_dpif_backers, type);
1018     if (!backer) {
1019         /* This is not necessarily a problem, since backers are only
1020          * created on demand. */
1021         return 0;
1022     }
1023
1024     return dpif_backer_run_fast(backer);
1025 }
1026
1027 static void
1028 run_fast_rl(void)
1029 {
1030     static long long int port_rl = LLONG_MIN;
1031
1032     if (time_msec() >= port_rl) {
1033         struct ofproto_dpif *ofproto;
1034
1035         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1036             run_fast(&ofproto->up);
1037         }
1038         port_rl = time_msec() + 200;
1039     }
1040 }
1041
1042 static void
1043 type_wait(const char *type)
1044 {
1045     struct dpif_backer *backer;
1046
1047     backer = shash_find_data(&all_dpif_backers, type);
1048     if (!backer) {
1049         /* This is not necessarily a problem, since backers are only
1050          * created on demand. */
1051         return;
1052     }
1053
1054     if (backer->governor) {
1055         governor_wait(backer->governor);
1056     }
1057
1058     timer_wait(&backer->next_expiration);
1059     dpif_wait(backer->dpif);
1060     udpif_wait(backer->udpif);
1061 }
1062 \f
1063 /* Basic life-cycle. */
1064
1065 static int add_internal_flows(struct ofproto_dpif *);
1066
1067 static struct ofproto *
1068 alloc(void)
1069 {
1070     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1071     return &ofproto->up;
1072 }
1073
1074 static void
1075 dealloc(struct ofproto *ofproto_)
1076 {
1077     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1078     free(ofproto);
1079 }
1080
1081 static void
1082 close_dpif_backer(struct dpif_backer *backer)
1083 {
1084     ovs_assert(backer->refcount > 0);
1085
1086     if (--backer->refcount) {
1087         return;
1088     }
1089
1090     drop_key_clear(backer);
1091     hmap_destroy(&backer->drop_keys);
1092
1093     udpif_destroy(backer->udpif);
1094
1095     simap_destroy(&backer->tnl_backers);
1096     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1097     hmap_destroy(&backer->odp_to_ofport_map);
1098     shash_find_and_delete(&all_dpif_backers, backer->type);
1099     free(backer->type);
1100     dpif_close(backer->dpif);
1101
1102     ovs_assert(hmap_is_empty(&backer->subfacets));
1103     hmap_destroy(&backer->subfacets);
1104     governor_destroy(backer->governor);
1105
1106     free(backer);
1107 }
1108
1109 /* Datapath port slated for removal from datapath. */
1110 struct odp_garbage {
1111     struct list list_node;
1112     odp_port_t odp_port;
1113 };
1114
1115 static int
1116 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1117 {
1118     struct dpif_backer *backer;
1119     struct dpif_port_dump port_dump;
1120     struct dpif_port port;
1121     struct shash_node *node;
1122     struct list garbage_list;
1123     struct odp_garbage *garbage, *next;
1124     struct sset names;
1125     char *backer_name;
1126     const char *name;
1127     int error;
1128
1129     backer = shash_find_data(&all_dpif_backers, type);
1130     if (backer) {
1131         backer->refcount++;
1132         *backerp = backer;
1133         return 0;
1134     }
1135
1136     backer_name = xasprintf("ovs-%s", type);
1137
1138     /* Remove any existing datapaths, since we assume we're the only
1139      * userspace controlling the datapath. */
1140     sset_init(&names);
1141     dp_enumerate_names(type, &names);
1142     SSET_FOR_EACH(name, &names) {
1143         struct dpif *old_dpif;
1144
1145         /* Don't remove our backer if it exists. */
1146         if (!strcmp(name, backer_name)) {
1147             continue;
1148         }
1149
1150         if (dpif_open(name, type, &old_dpif)) {
1151             VLOG_WARN("couldn't open old datapath %s to remove it", name);
1152         } else {
1153             dpif_delete(old_dpif);
1154             dpif_close(old_dpif);
1155         }
1156     }
1157     sset_destroy(&names);
1158
1159     backer = xmalloc(sizeof *backer);
1160
1161     error = dpif_create_and_open(backer_name, type, &backer->dpif);
1162     free(backer_name);
1163     if (error) {
1164         VLOG_ERR("failed to open datapath of type %s: %s", type,
1165                  ovs_strerror(error));
1166         free(backer);
1167         return error;
1168     }
1169     backer->udpif = udpif_create(backer, backer->dpif);
1170
1171     backer->type = xstrdup(type);
1172     backer->governor = NULL;
1173     backer->refcount = 1;
1174     hmap_init(&backer->odp_to_ofport_map);
1175     ovs_rwlock_init(&backer->odp_to_ofport_lock);
1176     hmap_init(&backer->drop_keys);
1177     hmap_init(&backer->subfacets);
1178     timer_set_duration(&backer->next_expiration, 1000);
1179     backer->need_revalidate = 0;
1180     simap_init(&backer->tnl_backers);
1181     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1182     *backerp = backer;
1183
1184     if (backer->recv_set_enable) {
1185         dpif_flow_flush(backer->dpif);
1186     }
1187
1188     /* Loop through the ports already on the datapath and remove any
1189      * that we don't need anymore. */
1190     list_init(&garbage_list);
1191     dpif_port_dump_start(&port_dump, backer->dpif);
1192     while (dpif_port_dump_next(&port_dump, &port)) {
1193         node = shash_find(&init_ofp_ports, port.name);
1194         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1195             garbage = xmalloc(sizeof *garbage);
1196             garbage->odp_port = port.port_no;
1197             list_push_front(&garbage_list, &garbage->list_node);
1198         }
1199     }
1200     dpif_port_dump_done(&port_dump);
1201
1202     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1203         dpif_port_del(backer->dpif, garbage->odp_port);
1204         list_remove(&garbage->list_node);
1205         free(garbage);
1206     }
1207
1208     shash_add(&all_dpif_backers, type, backer);
1209
1210     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1211     if (error) {
1212         VLOG_ERR("failed to listen on datapath of type %s: %s",
1213                  type, ovs_strerror(error));
1214         close_dpif_backer(backer);
1215         return error;
1216     }
1217     udpif_recv_set(backer->udpif, n_handler_threads,
1218                    backer->recv_set_enable);
1219     backer->n_handler_threads = n_handler_threads;
1220
1221     backer->max_n_subfacet = 0;
1222     backer->avg_n_subfacet = 0;
1223     backer->avg_subfacet_life = 0;
1224
1225     return error;
1226 }
1227
1228 static int
1229 construct(struct ofproto *ofproto_)
1230 {
1231     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1232     struct shash_node *node, *next;
1233     int error;
1234
1235     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1236     if (error) {
1237         return error;
1238     }
1239
1240     ofproto->netflow = NULL;
1241     ofproto->sflow = NULL;
1242     ofproto->ipfix = NULL;
1243     ofproto->stp = NULL;
1244     hmap_init(&ofproto->bundles);
1245     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1246     ofproto->mbridge = mbridge_create();
1247     ofproto->has_bonded_bundles = false;
1248     ovs_mutex_init(&ofproto->stats_mutex);
1249     ovs_mutex_init(&ofproto->vsp_mutex);
1250
1251     classifier_init(&ofproto->facets, NULL);
1252     ofproto->consistency_rl = LLONG_MIN;
1253
1254     guarded_list_init(&ofproto->pins);
1255
1256     ofproto_dpif_unixctl_init();
1257
1258     hmap_init(&ofproto->vlandev_map);
1259     hmap_init(&ofproto->realdev_vid_map);
1260
1261     sset_init(&ofproto->ports);
1262     sset_init(&ofproto->ghost_ports);
1263     sset_init(&ofproto->port_poll_set);
1264     ofproto->port_poll_errno = 0;
1265
1266     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1267         struct iface_hint *iface_hint = node->data;
1268
1269         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1270             /* Check if the datapath already has this port. */
1271             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1272                 sset_add(&ofproto->ports, node->name);
1273             }
1274
1275             free(iface_hint->br_name);
1276             free(iface_hint->br_type);
1277             free(iface_hint);
1278             shash_delete(&init_ofp_ports, node);
1279         }
1280     }
1281
1282     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1283                 hash_string(ofproto->up.name, 0));
1284     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1285
1286     ofproto_init_tables(ofproto_, N_TABLES);
1287     error = add_internal_flows(ofproto);
1288     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1289
1290     ofproto->n_hit = 0;
1291     ofproto->n_missed = 0;
1292
1293     return error;
1294 }
1295
1296 static int
1297 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1298                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1299 {
1300     struct ofputil_flow_mod fm;
1301     int error;
1302
1303     match_init_catchall(&fm.match);
1304     fm.priority = 0;
1305     match_set_reg(&fm.match, 0, id);
1306     fm.new_cookie = htonll(0);
1307     fm.cookie = htonll(0);
1308     fm.cookie_mask = htonll(0);
1309     fm.modify_cookie = false;
1310     fm.table_id = TBL_INTERNAL;
1311     fm.command = OFPFC_ADD;
1312     fm.idle_timeout = 0;
1313     fm.hard_timeout = 0;
1314     fm.buffer_id = 0;
1315     fm.out_port = 0;
1316     fm.flags = 0;
1317     fm.ofpacts = ofpacts->data;
1318     fm.ofpacts_len = ofpacts->size;
1319
1320     error = ofproto_flow_mod(&ofproto->up, &fm);
1321     if (error) {
1322         VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1323                     id, ofperr_to_string(error));
1324         return error;
1325     }
1326
1327     if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1328                                   rulep)) {
1329         rule_dpif_unref(*rulep);
1330     } else {
1331         NOT_REACHED();
1332     }
1333
1334     return 0;
1335 }
1336
1337 static int
1338 add_internal_flows(struct ofproto_dpif *ofproto)
1339 {
1340     struct ofpact_controller *controller;
1341     uint64_t ofpacts_stub[128 / 8];
1342     struct ofpbuf ofpacts;
1343     int error;
1344     int id;
1345
1346     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1347     id = 1;
1348
1349     controller = ofpact_put_CONTROLLER(&ofpacts);
1350     controller->max_len = UINT16_MAX;
1351     controller->controller_id = 0;
1352     controller->reason = OFPR_NO_MATCH;
1353     ofpact_pad(&ofpacts);
1354
1355     error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1356     if (error) {
1357         return error;
1358     }
1359
1360     ofpbuf_clear(&ofpacts);
1361     error = add_internal_flow(ofproto, id++, &ofpacts,
1362                               &ofproto->no_packet_in_rule);
1363     if (error) {
1364         return error;
1365     }
1366
1367     error = add_internal_flow(ofproto, id++, &ofpacts,
1368                               &ofproto->drop_frags_rule);
1369     return error;
1370 }
1371
1372 static void
1373 destruct(struct ofproto *ofproto_)
1374 {
1375     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1376     struct rule_dpif *rule, *next_rule;
1377     struct ofproto_packet_in *pin, *next_pin;
1378     struct facet *facet, *next_facet;
1379     struct cls_cursor cursor;
1380     struct oftable *table;
1381     struct list pins;
1382
1383     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1384     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1385     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1386     CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1387         facet_remove(facet);
1388     }
1389
1390     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1391     ovs_rwlock_wrlock(&xlate_rwlock);
1392     xlate_remove_ofproto(ofproto);
1393     ovs_rwlock_unlock(&xlate_rwlock);
1394
1395     /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a
1396      * use-after-free error. */
1397     udpif_revalidate(ofproto->backer->udpif);
1398
1399     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1400
1401     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1402         struct cls_cursor cursor;
1403
1404         ovs_rwlock_rdlock(&table->cls.rwlock);
1405         cls_cursor_init(&cursor, &table->cls, NULL);
1406         ovs_rwlock_unlock(&table->cls.rwlock);
1407         CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1408             ofproto_rule_delete(&ofproto->up, &rule->up);
1409         }
1410     }
1411
1412     guarded_list_pop_all(&ofproto->pins, &pins);
1413     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1414         list_remove(&pin->list_node);
1415         free(CONST_CAST(void *, pin->up.packet));
1416         free(pin);
1417     }
1418     guarded_list_destroy(&ofproto->pins);
1419
1420     mbridge_unref(ofproto->mbridge);
1421
1422     netflow_destroy(ofproto->netflow);
1423     dpif_sflow_unref(ofproto->sflow);
1424     hmap_destroy(&ofproto->bundles);
1425     mac_learning_unref(ofproto->ml);
1426
1427     classifier_destroy(&ofproto->facets);
1428
1429     hmap_destroy(&ofproto->vlandev_map);
1430     hmap_destroy(&ofproto->realdev_vid_map);
1431
1432     sset_destroy(&ofproto->ports);
1433     sset_destroy(&ofproto->ghost_ports);
1434     sset_destroy(&ofproto->port_poll_set);
1435
1436     ovs_mutex_destroy(&ofproto->stats_mutex);
1437     ovs_mutex_destroy(&ofproto->vsp_mutex);
1438
1439     close_dpif_backer(ofproto->backer);
1440 }
1441
1442 static int
1443 run_fast(struct ofproto *ofproto_)
1444 {
1445     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1446     struct ofproto_packet_in *pin, *next_pin;
1447     struct list pins;
1448
1449     /* Do not perform any periodic activity required by 'ofproto' while
1450      * waiting for flow restore to complete. */
1451     if (ofproto_get_flow_restore_wait()) {
1452         return 0;
1453     }
1454
1455     guarded_list_pop_all(&ofproto->pins, &pins);
1456     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1457         connmgr_send_packet_in(ofproto->up.connmgr, pin);
1458         list_remove(&pin->list_node);
1459         free(CONST_CAST(void *, pin->up.packet));
1460         free(pin);
1461     }
1462
1463     return 0;
1464 }
1465
1466 static int
1467 run(struct ofproto *ofproto_)
1468 {
1469     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1470     struct ofport_dpif *ofport;
1471     struct ofbundle *bundle;
1472     int error;
1473
1474     if (mbridge_need_revalidate(ofproto->mbridge)) {
1475         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1476         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1477         mac_learning_flush(ofproto->ml);
1478         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1479     }
1480
1481     /* Do not perform any periodic activity below required by 'ofproto' while
1482      * waiting for flow restore to complete. */
1483     if (ofproto_get_flow_restore_wait()) {
1484         return 0;
1485     }
1486
1487     error = run_fast(ofproto_);
1488     if (error) {
1489         return error;
1490     }
1491
1492     if (ofproto->netflow) {
1493         if (netflow_run(ofproto->netflow)) {
1494             send_netflow_active_timeouts(ofproto);
1495         }
1496     }
1497     if (ofproto->sflow) {
1498         dpif_sflow_run(ofproto->sflow);
1499     }
1500     if (ofproto->ipfix) {
1501         dpif_ipfix_run(ofproto->ipfix);
1502     }
1503
1504     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1505         port_run(ofport);
1506     }
1507     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1508         bundle_run(bundle);
1509     }
1510
1511     stp_run(ofproto);
1512     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1513     if (mac_learning_run(ofproto->ml)) {
1514         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1515     }
1516     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1517
1518     /* Check the consistency of a random facet, to aid debugging. */
1519     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1520     if (time_msec() >= ofproto->consistency_rl
1521         && !classifier_is_empty(&ofproto->facets)
1522         && !ofproto->backer->need_revalidate) {
1523         struct cls_subtable *table;
1524         struct cls_rule *cr;
1525         struct facet *facet;
1526
1527         ofproto->consistency_rl = time_msec() + 250;
1528
1529         table = CONTAINER_OF(hmap_random_node(&ofproto->facets.subtables),
1530                              struct cls_subtable, hmap_node);
1531         cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1532                           hmap_node);
1533         facet = CONTAINER_OF(cr, struct facet, cr);
1534
1535         if (!facet_check_consistency(facet)) {
1536             ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1537         }
1538     }
1539     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1540
1541     return 0;
1542 }
1543
1544 static void
1545 wait(struct ofproto *ofproto_)
1546 {
1547     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1548     struct ofbundle *bundle;
1549
1550     if (ofproto_get_flow_restore_wait()) {
1551         return;
1552     }
1553
1554     if (ofproto->sflow) {
1555         dpif_sflow_wait(ofproto->sflow);
1556     }
1557     if (ofproto->ipfix) {
1558         dpif_ipfix_wait(ofproto->ipfix);
1559     }
1560     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1561         bundle_wait(bundle);
1562     }
1563     if (ofproto->netflow) {
1564         netflow_wait(ofproto->netflow);
1565     }
1566     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1567     mac_learning_wait(ofproto->ml);
1568     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1569     stp_wait(ofproto);
1570     if (ofproto->backer->need_revalidate) {
1571         /* Shouldn't happen, but if it does just go around again. */
1572         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1573         poll_immediate_wake();
1574     }
1575 }
1576
1577 static void
1578 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1579 {
1580     const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1581     struct cls_cursor cursor;
1582     size_t n_subfacets = 0;
1583     struct facet *facet;
1584
1585     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1586     simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1587     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1588
1589     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1590     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1591     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1592         n_subfacets += list_size(&facet->subfacets);
1593     }
1594     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1595     simap_increase(usage, "subfacets", n_subfacets);
1596 }
1597
1598 static void
1599 flush(struct ofproto *ofproto_)
1600 {
1601     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1602     struct subfacet *subfacet, *next_subfacet;
1603     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1604     int n_batch;
1605
1606     n_batch = 0;
1607     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1608                         &ofproto->backer->subfacets) {
1609         if (subfacet->facet->ofproto != ofproto) {
1610             continue;
1611         }
1612
1613         if (subfacet->path != SF_NOT_INSTALLED) {
1614             batch[n_batch++] = subfacet;
1615             if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1616                 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1617                 n_batch = 0;
1618             }
1619         } else {
1620             subfacet_destroy(subfacet);
1621         }
1622     }
1623
1624     if (n_batch > 0) {
1625         subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1626     }
1627 }
1628
1629 static void
1630 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1631              bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1632 {
1633     *arp_match_ip = true;
1634     *actions = (OFPUTIL_A_OUTPUT |
1635                 OFPUTIL_A_SET_VLAN_VID |
1636                 OFPUTIL_A_SET_VLAN_PCP |
1637                 OFPUTIL_A_STRIP_VLAN |
1638                 OFPUTIL_A_SET_DL_SRC |
1639                 OFPUTIL_A_SET_DL_DST |
1640                 OFPUTIL_A_SET_NW_SRC |
1641                 OFPUTIL_A_SET_NW_DST |
1642                 OFPUTIL_A_SET_NW_TOS |
1643                 OFPUTIL_A_SET_TP_SRC |
1644                 OFPUTIL_A_SET_TP_DST |
1645                 OFPUTIL_A_ENQUEUE);
1646 }
1647
1648 static void
1649 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1650 {
1651     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1652     struct dpif_dp_stats s;
1653     uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1654     uint64_t n_lookup;
1655
1656     strcpy(ots->name, "classifier");
1657
1658     dpif_get_dp_stats(ofproto->backer->dpif, &s);
1659     rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1660     rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1661     rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1662
1663     n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1664     ots->lookup_count = htonll(n_lookup);
1665     ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1666 }
1667
1668 static struct ofport *
1669 port_alloc(void)
1670 {
1671     struct ofport_dpif *port = xmalloc(sizeof *port);
1672     return &port->up;
1673 }
1674
1675 static void
1676 port_dealloc(struct ofport *port_)
1677 {
1678     struct ofport_dpif *port = ofport_dpif_cast(port_);
1679     free(port);
1680 }
1681
1682 static int
1683 port_construct(struct ofport *port_)
1684 {
1685     struct ofport_dpif *port = ofport_dpif_cast(port_);
1686     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1687     const struct netdev *netdev = port->up.netdev;
1688     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1689     struct dpif_port dpif_port;
1690     int error;
1691
1692     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1693     port->bundle = NULL;
1694     port->cfm = NULL;
1695     port->bfd = NULL;
1696     port->may_enable = true;
1697     port->stp_port = NULL;
1698     port->stp_state = STP_DISABLED;
1699     port->is_tunnel = false;
1700     port->peer = NULL;
1701     port->qdscp = NULL;
1702     port->n_qdscp = 0;
1703     port->realdev_ofp_port = 0;
1704     port->vlandev_vid = 0;
1705     port->carrier_seq = netdev_get_carrier_resets(netdev);
1706     port->is_layer3 = netdev_vport_is_layer3(netdev);
1707
1708     if (netdev_vport_is_patch(netdev)) {
1709         /* By bailing out here, we don't submit the port to the sFlow module
1710          * to be considered for counter polling export.  This is correct
1711          * because the patch port represents an interface that sFlow considers
1712          * to be "internal" to the switch as a whole, and therefore not an
1713          * candidate for counter polling. */
1714         port->odp_port = ODPP_NONE;
1715         ofport_update_peer(port);
1716         return 0;
1717     }
1718
1719     error = dpif_port_query_by_name(ofproto->backer->dpif,
1720                                     netdev_vport_get_dpif_port(netdev, namebuf,
1721                                                                sizeof namebuf),
1722                                     &dpif_port);
1723     if (error) {
1724         return error;
1725     }
1726
1727     port->odp_port = dpif_port.port_no;
1728
1729     if (netdev_get_tunnel_config(netdev)) {
1730         tnl_port_add(port, port->up.netdev, port->odp_port);
1731         port->is_tunnel = true;
1732     } else {
1733         /* Sanity-check that a mapping doesn't already exist.  This
1734          * shouldn't happen for non-tunnel ports. */
1735         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1736             VLOG_ERR("port %s already has an OpenFlow port number",
1737                      dpif_port.name);
1738             dpif_port_destroy(&dpif_port);
1739             return EBUSY;
1740         }
1741
1742         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1743         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1744                     hash_odp_port(port->odp_port));
1745         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1746     }
1747     dpif_port_destroy(&dpif_port);
1748
1749     if (ofproto->sflow) {
1750         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1751     }
1752
1753     return 0;
1754 }
1755
1756 static void
1757 port_destruct(struct ofport *port_)
1758 {
1759     struct ofport_dpif *port = ofport_dpif_cast(port_);
1760     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1761     const char *devname = netdev_get_name(port->up.netdev);
1762     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1763     const char *dp_port_name;
1764
1765     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1766     ovs_rwlock_wrlock(&xlate_rwlock);
1767     xlate_ofport_remove(port);
1768     ovs_rwlock_unlock(&xlate_rwlock);
1769
1770     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1771                                               sizeof namebuf);
1772     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1773         /* The underlying device is still there, so delete it.  This
1774          * happens when the ofproto is being destroyed, since the caller
1775          * assumes that removal of attached ports will happen as part of
1776          * destruction. */
1777         if (!port->is_tunnel) {
1778             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1779         }
1780     }
1781
1782     if (port->peer) {
1783         port->peer->peer = NULL;
1784         port->peer = NULL;
1785     }
1786
1787     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1788         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1789         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1790         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1791     }
1792
1793     tnl_port_del(port);
1794     sset_find_and_delete(&ofproto->ports, devname);
1795     sset_find_and_delete(&ofproto->ghost_ports, devname);
1796     bundle_remove(port_);
1797     set_cfm(port_, NULL);
1798     set_bfd(port_, NULL);
1799     if (ofproto->sflow) {
1800         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1801     }
1802
1803     free(port->qdscp);
1804 }
1805
1806 static void
1807 port_modified(struct ofport *port_)
1808 {
1809     struct ofport_dpif *port = ofport_dpif_cast(port_);
1810
1811     if (port->bundle && port->bundle->bond) {
1812         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1813     }
1814
1815     if (port->cfm) {
1816         cfm_set_netdev(port->cfm, port->up.netdev);
1817     }
1818
1819     if (port->bfd) {
1820         bfd_set_netdev(port->bfd, port->up.netdev);
1821     }
1822
1823     ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1824                                      port->up.pp.hw_addr);
1825
1826     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1827                                                 port->odp_port)) {
1828         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1829             REV_RECONFIGURE;
1830     }
1831
1832     ofport_update_peer(port);
1833 }
1834
1835 static void
1836 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1837 {
1838     struct ofport_dpif *port = ofport_dpif_cast(port_);
1839     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1840     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1841
1842     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1843                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1844                    OFPUTIL_PC_NO_PACKET_IN)) {
1845         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1846
1847         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1848             bundle_update(port->bundle);
1849         }
1850     }
1851 }
1852
1853 static int
1854 set_sflow(struct ofproto *ofproto_,
1855           const struct ofproto_sflow_options *sflow_options)
1856 {
1857     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1858     struct dpif_sflow *ds = ofproto->sflow;
1859
1860     if (sflow_options) {
1861         if (!ds) {
1862             struct ofport_dpif *ofport;
1863
1864             ds = ofproto->sflow = dpif_sflow_create();
1865             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1866                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1867             }
1868             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1869         }
1870         dpif_sflow_set_options(ds, sflow_options);
1871     } else {
1872         if (ds) {
1873             dpif_sflow_unref(ds);
1874             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1875             ofproto->sflow = NULL;
1876         }
1877     }
1878     return 0;
1879 }
1880
1881 static int
1882 set_ipfix(
1883     struct ofproto *ofproto_,
1884     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1885     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1886     size_t n_flow_exporters_options)
1887 {
1888     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1889     struct dpif_ipfix *di = ofproto->ipfix;
1890     bool has_options = bridge_exporter_options || flow_exporters_options;
1891
1892     if (has_options && !di) {
1893         di = ofproto->ipfix = dpif_ipfix_create();
1894     }
1895
1896     if (di) {
1897         /* Call set_options in any case to cleanly flush the flow
1898          * caches in the last exporters that are to be destroyed. */
1899         dpif_ipfix_set_options(
1900             di, bridge_exporter_options, flow_exporters_options,
1901             n_flow_exporters_options);
1902
1903         if (!has_options) {
1904             dpif_ipfix_unref(di);
1905             ofproto->ipfix = NULL;
1906         }
1907     }
1908
1909     return 0;
1910 }
1911
1912 static int
1913 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1914 {
1915     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1916     int error = 0;
1917
1918     if (s) {
1919         if (!ofport->cfm) {
1920             struct ofproto_dpif *ofproto;
1921
1922             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1923             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1924             ofport->cfm = cfm_create(ofport->up.netdev);
1925         }
1926
1927         if (cfm_configure(ofport->cfm, s)) {
1928             error = 0;
1929             goto out;
1930         }
1931
1932         error = EINVAL;
1933     }
1934     cfm_unref(ofport->cfm);
1935     ofport->cfm = NULL;
1936 out:
1937     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1938                                      ofport->up.pp.hw_addr);
1939     return error;
1940 }
1941
1942 static bool
1943 get_cfm_status(const struct ofport *ofport_,
1944                struct ofproto_cfm_status *status)
1945 {
1946     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1947
1948     if (ofport->cfm) {
1949         status->faults = cfm_get_fault(ofport->cfm);
1950         status->flap_count = cfm_get_flap_count(ofport->cfm);
1951         status->remote_opstate = cfm_get_opup(ofport->cfm);
1952         status->health = cfm_get_health(ofport->cfm);
1953         cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1954         return true;
1955     } else {
1956         return false;
1957     }
1958 }
1959
1960 static int
1961 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1962 {
1963     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1964     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1965     struct bfd *old;
1966
1967     old = ofport->bfd;
1968     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1969                                 cfg, ofport->up.netdev);
1970     if (ofport->bfd != old) {
1971         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1972     }
1973     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1974                                      ofport->up.pp.hw_addr);
1975     return 0;
1976 }
1977
1978 static int
1979 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1980 {
1981     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1982
1983     if (ofport->bfd) {
1984         bfd_get_status(ofport->bfd, smap);
1985         return 0;
1986     } else {
1987         return ENOENT;
1988     }
1989 }
1990 \f
1991 /* Spanning Tree. */
1992
1993 static void
1994 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1995 {
1996     struct ofproto_dpif *ofproto = ofproto_;
1997     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1998     struct ofport_dpif *ofport;
1999
2000     ofport = stp_port_get_aux(sp);
2001     if (!ofport) {
2002         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2003                      ofproto->up.name, port_num);
2004     } else {
2005         struct eth_header *eth = pkt->l2;
2006
2007         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2008         if (eth_addr_is_zero(eth->eth_src)) {
2009             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2010                          "with unknown MAC", ofproto->up.name, port_num);
2011         } else {
2012             ofproto_dpif_send_packet(ofport, pkt);
2013         }
2014     }
2015     ofpbuf_delete(pkt);
2016 }
2017
2018 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2019 static int
2020 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2021 {
2022     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2023
2024     /* Only revalidate flows if the configuration changed. */
2025     if (!s != !ofproto->stp) {
2026         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2027     }
2028
2029     if (s) {
2030         if (!ofproto->stp) {
2031             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2032                                       send_bpdu_cb, ofproto);
2033             ofproto->stp_last_tick = time_msec();
2034         }
2035
2036         stp_set_bridge_id(ofproto->stp, s->system_id);
2037         stp_set_bridge_priority(ofproto->stp, s->priority);
2038         stp_set_hello_time(ofproto->stp, s->hello_time);
2039         stp_set_max_age(ofproto->stp, s->max_age);
2040         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2041     }  else {
2042         struct ofport *ofport;
2043
2044         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2045             set_stp_port(ofport, NULL);
2046         }
2047
2048         stp_unref(ofproto->stp);
2049         ofproto->stp = NULL;
2050     }
2051
2052     return 0;
2053 }
2054
2055 static int
2056 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2057 {
2058     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2059
2060     if (ofproto->stp) {
2061         s->enabled = true;
2062         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2063         s->designated_root = stp_get_designated_root(ofproto->stp);
2064         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2065     } else {
2066         s->enabled = false;
2067     }
2068
2069     return 0;
2070 }
2071
2072 static void
2073 update_stp_port_state(struct ofport_dpif *ofport)
2074 {
2075     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2076     enum stp_state state;
2077
2078     /* Figure out new state. */
2079     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2080                              : STP_DISABLED;
2081
2082     /* Update state. */
2083     if (ofport->stp_state != state) {
2084         enum ofputil_port_state of_state;
2085         bool fwd_change;
2086
2087         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2088                     netdev_get_name(ofport->up.netdev),
2089                     stp_state_name(ofport->stp_state),
2090                     stp_state_name(state));
2091         if (stp_learn_in_state(ofport->stp_state)
2092                 != stp_learn_in_state(state)) {
2093             /* xxx Learning action flows should also be flushed. */
2094             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2095             mac_learning_flush(ofproto->ml);
2096             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2097         }
2098         fwd_change = stp_forward_in_state(ofport->stp_state)
2099                         != stp_forward_in_state(state);
2100
2101         ofproto->backer->need_revalidate = REV_STP;
2102         ofport->stp_state = state;
2103         ofport->stp_state_entered = time_msec();
2104
2105         if (fwd_change && ofport->bundle) {
2106             bundle_update(ofport->bundle);
2107         }
2108
2109         /* Update the STP state bits in the OpenFlow port description. */
2110         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2111         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2112                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2113                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2114                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2115                      : 0);
2116         ofproto_port_set_state(&ofport->up, of_state);
2117     }
2118 }
2119
2120 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2121  * caller is responsible for assigning STP port numbers and ensuring
2122  * there are no duplicates. */
2123 static int
2124 set_stp_port(struct ofport *ofport_,
2125              const struct ofproto_port_stp_settings *s)
2126 {
2127     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2128     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2129     struct stp_port *sp = ofport->stp_port;
2130
2131     if (!s || !s->enable) {
2132         if (sp) {
2133             ofport->stp_port = NULL;
2134             stp_port_disable(sp);
2135             update_stp_port_state(ofport);
2136         }
2137         return 0;
2138     } else if (sp && stp_port_no(sp) != s->port_num
2139             && ofport == stp_port_get_aux(sp)) {
2140         /* The port-id changed, so disable the old one if it's not
2141          * already in use by another port. */
2142         stp_port_disable(sp);
2143     }
2144
2145     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2146     stp_port_enable(sp);
2147
2148     stp_port_set_aux(sp, ofport);
2149     stp_port_set_priority(sp, s->priority);
2150     stp_port_set_path_cost(sp, s->path_cost);
2151
2152     update_stp_port_state(ofport);
2153
2154     return 0;
2155 }
2156
2157 static int
2158 get_stp_port_status(struct ofport *ofport_,
2159                     struct ofproto_port_stp_status *s)
2160 {
2161     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2162     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2163     struct stp_port *sp = ofport->stp_port;
2164
2165     if (!ofproto->stp || !sp) {
2166         s->enabled = false;
2167         return 0;
2168     }
2169
2170     s->enabled = true;
2171     s->port_id = stp_port_get_id(sp);
2172     s->state = stp_port_get_state(sp);
2173     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2174     s->role = stp_port_get_role(sp);
2175
2176     return 0;
2177 }
2178
2179 static int
2180 get_stp_port_stats(struct ofport *ofport_,
2181                    struct ofproto_port_stp_stats *s)
2182 {
2183     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2184     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2185     struct stp_port *sp = ofport->stp_port;
2186
2187     if (!ofproto->stp || !sp) {
2188         s->enabled = false;
2189         return 0;
2190     }
2191
2192     s->enabled = true;
2193     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2194
2195     return 0;
2196 }
2197
2198 static void
2199 stp_run(struct ofproto_dpif *ofproto)
2200 {
2201     if (ofproto->stp) {
2202         long long int now = time_msec();
2203         long long int elapsed = now - ofproto->stp_last_tick;
2204         struct stp_port *sp;
2205
2206         if (elapsed > 0) {
2207             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2208             ofproto->stp_last_tick = now;
2209         }
2210         while (stp_get_changed_port(ofproto->stp, &sp)) {
2211             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2212
2213             if (ofport) {
2214                 update_stp_port_state(ofport);
2215             }
2216         }
2217
2218         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2219             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2220             mac_learning_flush(ofproto->ml);
2221             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2222         }
2223     }
2224 }
2225
2226 static void
2227 stp_wait(struct ofproto_dpif *ofproto)
2228 {
2229     if (ofproto->stp) {
2230         poll_timer_wait(1000);
2231     }
2232 }
2233 \f
2234 static int
2235 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2236            size_t n_qdscp)
2237 {
2238     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2239     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2240
2241     if (ofport->n_qdscp != n_qdscp
2242         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2243                               n_qdscp * sizeof *qdscp))) {
2244         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2245         free(ofport->qdscp);
2246         ofport->qdscp = n_qdscp
2247             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2248             : NULL;
2249         ofport->n_qdscp = n_qdscp;
2250     }
2251
2252     return 0;
2253 }
2254 \f
2255 /* Bundles. */
2256
2257 /* Expires all MAC learning entries associated with 'bundle' and forces its
2258  * ofproto to revalidate every flow.
2259  *
2260  * Normally MAC learning entries are removed only from the ofproto associated
2261  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2262  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2263  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2264  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2265  * with the host from which it migrated. */
2266 static void
2267 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2268 {
2269     struct ofproto_dpif *ofproto = bundle->ofproto;
2270     struct mac_learning *ml = ofproto->ml;
2271     struct mac_entry *mac, *next_mac;
2272
2273     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2274     ovs_rwlock_wrlock(&ml->rwlock);
2275     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2276         if (mac->port.p == bundle) {
2277             if (all_ofprotos) {
2278                 struct ofproto_dpif *o;
2279
2280                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2281                     if (o != ofproto) {
2282                         struct mac_entry *e;
2283
2284                         ovs_rwlock_wrlock(&o->ml->rwlock);
2285                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2286                         if (e) {
2287                             mac_learning_expire(o->ml, e);
2288                         }
2289                         ovs_rwlock_unlock(&o->ml->rwlock);
2290                     }
2291                 }
2292             }
2293
2294             mac_learning_expire(ml, mac);
2295         }
2296     }
2297     ovs_rwlock_unlock(&ml->rwlock);
2298 }
2299
2300 static struct ofbundle *
2301 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2302 {
2303     struct ofbundle *bundle;
2304
2305     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2306                              &ofproto->bundles) {
2307         if (bundle->aux == aux) {
2308             return bundle;
2309         }
2310     }
2311     return NULL;
2312 }
2313
2314 static void
2315 bundle_update(struct ofbundle *bundle)
2316 {
2317     struct ofport_dpif *port;
2318
2319     bundle->floodable = true;
2320     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2321         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2322             || port->is_layer3
2323             || !stp_forward_in_state(port->stp_state)) {
2324             bundle->floodable = false;
2325             break;
2326         }
2327     }
2328 }
2329
2330 static void
2331 bundle_del_port(struct ofport_dpif *port)
2332 {
2333     struct ofbundle *bundle = port->bundle;
2334
2335     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2336
2337     list_remove(&port->bundle_node);
2338     port->bundle = NULL;
2339
2340     if (bundle->lacp) {
2341         lacp_slave_unregister(bundle->lacp, port);
2342     }
2343     if (bundle->bond) {
2344         bond_slave_unregister(bundle->bond, port);
2345     }
2346
2347     bundle_update(bundle);
2348 }
2349
2350 static bool
2351 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2352                 struct lacp_slave_settings *lacp)
2353 {
2354     struct ofport_dpif *port;
2355
2356     port = get_ofp_port(bundle->ofproto, ofp_port);
2357     if (!port) {
2358         return false;
2359     }
2360
2361     if (port->bundle != bundle) {
2362         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2363         if (port->bundle) {
2364             bundle_remove(&port->up);
2365         }
2366
2367         port->bundle = bundle;
2368         list_push_back(&bundle->ports, &port->bundle_node);
2369         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2370             || port->is_layer3
2371             || !stp_forward_in_state(port->stp_state)) {
2372             bundle->floodable = false;
2373         }
2374     }
2375     if (lacp) {
2376         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2377         lacp_slave_register(bundle->lacp, port, lacp);
2378     }
2379
2380     return true;
2381 }
2382
2383 static void
2384 bundle_destroy(struct ofbundle *bundle)
2385 {
2386     struct ofproto_dpif *ofproto;
2387     struct ofport_dpif *port, *next_port;
2388
2389     if (!bundle) {
2390         return;
2391     }
2392
2393     ofproto = bundle->ofproto;
2394     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2395
2396     ovs_rwlock_wrlock(&xlate_rwlock);
2397     xlate_bundle_remove(bundle);
2398     ovs_rwlock_unlock(&xlate_rwlock);
2399
2400     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2401         bundle_del_port(port);
2402     }
2403
2404     bundle_flush_macs(bundle, true);
2405     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2406     free(bundle->name);
2407     free(bundle->trunks);
2408     lacp_unref(bundle->lacp);
2409     bond_unref(bundle->bond);
2410     free(bundle);
2411 }
2412
2413 static int
2414 bundle_set(struct ofproto *ofproto_, void *aux,
2415            const struct ofproto_bundle_settings *s)
2416 {
2417     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2418     bool need_flush = false;
2419     struct ofport_dpif *port;
2420     struct ofbundle *bundle;
2421     unsigned long *trunks;
2422     int vlan;
2423     size_t i;
2424     bool ok;
2425
2426     if (!s) {
2427         bundle_destroy(bundle_lookup(ofproto, aux));
2428         return 0;
2429     }
2430
2431     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2432     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2433
2434     bundle = bundle_lookup(ofproto, aux);
2435     if (!bundle) {
2436         bundle = xmalloc(sizeof *bundle);
2437
2438         bundle->ofproto = ofproto;
2439         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2440                     hash_pointer(aux, 0));
2441         bundle->aux = aux;
2442         bundle->name = NULL;
2443
2444         list_init(&bundle->ports);
2445         bundle->vlan_mode = PORT_VLAN_TRUNK;
2446         bundle->vlan = -1;
2447         bundle->trunks = NULL;
2448         bundle->use_priority_tags = s->use_priority_tags;
2449         bundle->lacp = NULL;
2450         bundle->bond = NULL;
2451
2452         bundle->floodable = true;
2453         mbridge_register_bundle(ofproto->mbridge, bundle);
2454     }
2455
2456     if (!bundle->name || strcmp(s->name, bundle->name)) {
2457         free(bundle->name);
2458         bundle->name = xstrdup(s->name);
2459     }
2460
2461     /* LACP. */
2462     if (s->lacp) {
2463         if (!bundle->lacp) {
2464             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2465             bundle->lacp = lacp_create();
2466         }
2467         lacp_configure(bundle->lacp, s->lacp);
2468     } else {
2469         lacp_unref(bundle->lacp);
2470         bundle->lacp = NULL;
2471     }
2472
2473     /* Update set of ports. */
2474     ok = true;
2475     for (i = 0; i < s->n_slaves; i++) {
2476         if (!bundle_add_port(bundle, s->slaves[i],
2477                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2478             ok = false;
2479         }
2480     }
2481     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2482         struct ofport_dpif *next_port;
2483
2484         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2485             for (i = 0; i < s->n_slaves; i++) {
2486                 if (s->slaves[i] == port->up.ofp_port) {
2487                     goto found;
2488                 }
2489             }
2490
2491             bundle_del_port(port);
2492         found: ;
2493         }
2494     }
2495     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2496
2497     if (list_is_empty(&bundle->ports)) {
2498         bundle_destroy(bundle);
2499         return EINVAL;
2500     }
2501
2502     /* Set VLAN tagging mode */
2503     if (s->vlan_mode != bundle->vlan_mode
2504         || s->use_priority_tags != bundle->use_priority_tags) {
2505         bundle->vlan_mode = s->vlan_mode;
2506         bundle->use_priority_tags = s->use_priority_tags;
2507         need_flush = true;
2508     }
2509
2510     /* Set VLAN tag. */
2511     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2512             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2513             : 0);
2514     if (vlan != bundle->vlan) {
2515         bundle->vlan = vlan;
2516         need_flush = true;
2517     }
2518
2519     /* Get trunked VLANs. */
2520     switch (s->vlan_mode) {
2521     case PORT_VLAN_ACCESS:
2522         trunks = NULL;
2523         break;
2524
2525     case PORT_VLAN_TRUNK:
2526         trunks = CONST_CAST(unsigned long *, s->trunks);
2527         break;
2528
2529     case PORT_VLAN_NATIVE_UNTAGGED:
2530     case PORT_VLAN_NATIVE_TAGGED:
2531         if (vlan != 0 && (!s->trunks
2532                           || !bitmap_is_set(s->trunks, vlan)
2533                           || bitmap_is_set(s->trunks, 0))) {
2534             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2535             if (s->trunks) {
2536                 trunks = bitmap_clone(s->trunks, 4096);
2537             } else {
2538                 trunks = bitmap_allocate1(4096);
2539             }
2540             bitmap_set1(trunks, vlan);
2541             bitmap_set0(trunks, 0);
2542         } else {
2543             trunks = CONST_CAST(unsigned long *, s->trunks);
2544         }
2545         break;
2546
2547     default:
2548         NOT_REACHED();
2549     }
2550     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2551         free(bundle->trunks);
2552         if (trunks == s->trunks) {
2553             bundle->trunks = vlan_bitmap_clone(trunks);
2554         } else {
2555             bundle->trunks = trunks;
2556             trunks = NULL;
2557         }
2558         need_flush = true;
2559     }
2560     if (trunks != s->trunks) {
2561         free(trunks);
2562     }
2563
2564     /* Bonding. */
2565     if (!list_is_short(&bundle->ports)) {
2566         bundle->ofproto->has_bonded_bundles = true;
2567         if (bundle->bond) {
2568             if (bond_reconfigure(bundle->bond, s->bond)) {
2569                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2570             }
2571         } else {
2572             bundle->bond = bond_create(s->bond);
2573             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2574         }
2575
2576         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2577             bond_slave_register(bundle->bond, port, port->up.netdev);
2578         }
2579     } else {
2580         bond_unref(bundle->bond);
2581         bundle->bond = NULL;
2582     }
2583
2584     /* If we changed something that would affect MAC learning, un-learn
2585      * everything on this port and force flow revalidation. */
2586     if (need_flush) {
2587         bundle_flush_macs(bundle, false);
2588     }
2589
2590     return 0;
2591 }
2592
2593 static void
2594 bundle_remove(struct ofport *port_)
2595 {
2596     struct ofport_dpif *port = ofport_dpif_cast(port_);
2597     struct ofbundle *bundle = port->bundle;
2598
2599     if (bundle) {
2600         bundle_del_port(port);
2601         if (list_is_empty(&bundle->ports)) {
2602             bundle_destroy(bundle);
2603         } else if (list_is_short(&bundle->ports)) {
2604             bond_unref(bundle->bond);
2605             bundle->bond = NULL;
2606         }
2607     }
2608 }
2609
2610 static void
2611 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2612 {
2613     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2614     struct ofport_dpif *port = port_;
2615     uint8_t ea[ETH_ADDR_LEN];
2616     int error;
2617
2618     error = netdev_get_etheraddr(port->up.netdev, ea);
2619     if (!error) {
2620         struct ofpbuf packet;
2621         void *packet_pdu;
2622
2623         ofpbuf_init(&packet, 0);
2624         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2625                                  pdu_size);
2626         memcpy(packet_pdu, pdu, pdu_size);
2627
2628         ofproto_dpif_send_packet(port, &packet);
2629         ofpbuf_uninit(&packet);
2630     } else {
2631         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2632                     "%s (%s)", port->bundle->name,
2633                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2634     }
2635 }
2636
2637 static void
2638 bundle_send_learning_packets(struct ofbundle *bundle)
2639 {
2640     struct ofproto_dpif *ofproto = bundle->ofproto;
2641     struct ofpbuf *learning_packet;
2642     int error, n_packets, n_errors;
2643     struct mac_entry *e;
2644     struct list packets;
2645
2646     list_init(&packets);
2647     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2648     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2649         if (e->port.p != bundle) {
2650             void *port_void;
2651
2652             learning_packet = bond_compose_learning_packet(bundle->bond,
2653                                                            e->mac, e->vlan,
2654                                                            &port_void);
2655             learning_packet->private_p = port_void;
2656             list_push_back(&packets, &learning_packet->list_node);
2657         }
2658     }
2659     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2660
2661     error = n_packets = n_errors = 0;
2662     LIST_FOR_EACH (learning_packet, list_node, &packets) {
2663         int ret;
2664
2665         ret = ofproto_dpif_send_packet(learning_packet->private_p, learning_packet);
2666         if (ret) {
2667             error = ret;
2668             n_errors++;
2669         }
2670         n_packets++;
2671     }
2672     ofpbuf_list_delete(&packets);
2673
2674     if (n_errors) {
2675         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2676         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2677                      "packets, last error was: %s",
2678                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2679     } else {
2680         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2681                  bundle->name, n_packets);
2682     }
2683 }
2684
2685 static void
2686 bundle_run(struct ofbundle *bundle)
2687 {
2688     if (bundle->lacp) {
2689         lacp_run(bundle->lacp, send_pdu_cb);
2690     }
2691     if (bundle->bond) {
2692         struct ofport_dpif *port;
2693
2694         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2695             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2696         }
2697
2698         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2699             bundle->ofproto->backer->need_revalidate = REV_BOND;
2700         }
2701
2702         if (bond_should_send_learning_packets(bundle->bond)) {
2703             bundle_send_learning_packets(bundle);
2704         }
2705     }
2706 }
2707
2708 static void
2709 bundle_wait(struct ofbundle *bundle)
2710 {
2711     if (bundle->lacp) {
2712         lacp_wait(bundle->lacp);
2713     }
2714     if (bundle->bond) {
2715         bond_wait(bundle->bond);
2716     }
2717 }
2718 \f
2719 /* Mirrors. */
2720
2721 static int
2722 mirror_set__(struct ofproto *ofproto_, void *aux,
2723              const struct ofproto_mirror_settings *s)
2724 {
2725     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2726     struct ofbundle **srcs, **dsts;
2727     int error;
2728     size_t i;
2729
2730     if (!s) {
2731         mirror_destroy(ofproto->mbridge, aux);
2732         return 0;
2733     }
2734
2735     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2736     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2737
2738     for (i = 0; i < s->n_srcs; i++) {
2739         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2740     }
2741
2742     for (i = 0; i < s->n_dsts; i++) {
2743         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2744     }
2745
2746     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2747                        s->n_dsts, s->src_vlans,
2748                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2749     free(srcs);
2750     free(dsts);
2751     return error;
2752 }
2753
2754 static int
2755 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2756                    uint64_t *packets, uint64_t *bytes)
2757 {
2758     push_all_stats();
2759     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2760                             bytes);
2761 }
2762
2763 static int
2764 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2765 {
2766     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2767     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2768     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2769         mac_learning_flush(ofproto->ml);
2770     }
2771     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2772     return 0;
2773 }
2774
2775 static bool
2776 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2777 {
2778     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2779     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2780     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2781 }
2782
2783 static void
2784 forward_bpdu_changed(struct ofproto *ofproto_)
2785 {
2786     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2787     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2788 }
2789
2790 static void
2791 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2792                      size_t max_entries)
2793 {
2794     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2795     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2796     mac_learning_set_idle_time(ofproto->ml, idle_time);
2797     mac_learning_set_max_entries(ofproto->ml, max_entries);
2798     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2799 }
2800 \f
2801 /* Ports. */
2802
2803 static struct ofport_dpif *
2804 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2805 {
2806     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2807     return ofport ? ofport_dpif_cast(ofport) : NULL;
2808 }
2809
2810 static struct ofport_dpif *
2811 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2812 {
2813     struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2814     return port && &ofproto->up == port->up.ofproto ? port : NULL;
2815 }
2816
2817 static void
2818 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2819                             struct ofproto_port *ofproto_port,
2820                             struct dpif_port *dpif_port)
2821 {
2822     ofproto_port->name = dpif_port->name;
2823     ofproto_port->type = dpif_port->type;
2824     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2825 }
2826
2827 static void
2828 ofport_update_peer(struct ofport_dpif *ofport)
2829 {
2830     const struct ofproto_dpif *ofproto;
2831     struct dpif_backer *backer;
2832     char *peer_name;
2833
2834     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2835         return;
2836     }
2837
2838     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2839     backer->need_revalidate = REV_RECONFIGURE;
2840
2841     if (ofport->peer) {
2842         ofport->peer->peer = NULL;
2843         ofport->peer = NULL;
2844     }
2845
2846     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2847     if (!peer_name) {
2848         return;
2849     }
2850
2851     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2852         struct ofport *peer_ofport;
2853         struct ofport_dpif *peer;
2854         char *peer_peer;
2855
2856         if (ofproto->backer != backer) {
2857             continue;
2858         }
2859
2860         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2861         if (!peer_ofport) {
2862             continue;
2863         }
2864
2865         peer = ofport_dpif_cast(peer_ofport);
2866         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2867         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2868                                  peer_peer)) {
2869             ofport->peer = peer;
2870             ofport->peer->peer = ofport;
2871         }
2872         free(peer_peer);
2873
2874         break;
2875     }
2876     free(peer_name);
2877 }
2878
2879 static void
2880 port_run(struct ofport_dpif *ofport)
2881 {
2882     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2883     bool carrier_changed = carrier_seq != ofport->carrier_seq;
2884     bool enable = netdev_get_carrier(ofport->up.netdev);
2885     bool cfm_enable = false;
2886     bool bfd_enable = false;
2887
2888     ofport->carrier_seq = carrier_seq;
2889
2890     if (ofport->cfm) {
2891         int cfm_opup = cfm_get_opup(ofport->cfm);
2892
2893         cfm_enable = !cfm_get_fault(ofport->cfm);
2894
2895         if (cfm_opup >= 0) {
2896             cfm_enable = cfm_enable && cfm_opup;
2897         }
2898     }
2899
2900     if (ofport->bfd) {
2901         bfd_enable = bfd_forwarding(ofport->bfd);
2902     }
2903
2904     if (ofport->bfd || ofport->cfm) {
2905         enable = enable && (cfm_enable || bfd_enable);
2906     }
2907
2908     if (ofport->bundle) {
2909         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2910         if (carrier_changed) {
2911             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2912         }
2913     }
2914
2915     if (ofport->may_enable != enable) {
2916         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2917         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2918     }
2919
2920     ofport->may_enable = enable;
2921 }
2922
2923 static int
2924 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2925                    struct ofproto_port *ofproto_port)
2926 {
2927     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2928     struct dpif_port dpif_port;
2929     int error;
2930
2931     if (sset_contains(&ofproto->ghost_ports, devname)) {
2932         const char *type = netdev_get_type_from_name(devname);
2933
2934         /* We may be called before ofproto->up.port_by_name is populated with
2935          * the appropriate ofport.  For this reason, we must get the name and
2936          * type from the netdev layer directly. */
2937         if (type) {
2938             const struct ofport *ofport;
2939
2940             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2941             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2942             ofproto_port->name = xstrdup(devname);
2943             ofproto_port->type = xstrdup(type);
2944             return 0;
2945         }
2946         return ENODEV;
2947     }
2948
2949     if (!sset_contains(&ofproto->ports, devname)) {
2950         return ENODEV;
2951     }
2952     error = dpif_port_query_by_name(ofproto->backer->dpif,
2953                                     devname, &dpif_port);
2954     if (!error) {
2955         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2956     }
2957     return error;
2958 }
2959
2960 static int
2961 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2962 {
2963     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2964     const char *devname = netdev_get_name(netdev);
2965     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2966     const char *dp_port_name;
2967
2968     if (netdev_vport_is_patch(netdev)) {
2969         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2970         return 0;
2971     }
2972
2973     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2974     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2975         odp_port_t port_no = ODPP_NONE;
2976         int error;
2977
2978         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2979         if (error) {
2980             return error;
2981         }
2982         if (netdev_get_tunnel_config(netdev)) {
2983             simap_put(&ofproto->backer->tnl_backers,
2984                       dp_port_name, odp_to_u32(port_no));
2985         }
2986     }
2987
2988     if (netdev_get_tunnel_config(netdev)) {
2989         sset_add(&ofproto->ghost_ports, devname);
2990     } else {
2991         sset_add(&ofproto->ports, devname);
2992     }
2993     return 0;
2994 }
2995
2996 static int
2997 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
2998 {
2999     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3000     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3001     int error = 0;
3002
3003     if (!ofport) {
3004         return 0;
3005     }
3006
3007     sset_find_and_delete(&ofproto->ghost_ports,
3008                          netdev_get_name(ofport->up.netdev));
3009     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3010     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3011         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3012         if (!error) {
3013             /* The caller is going to close ofport->up.netdev.  If this is a
3014              * bonded port, then the bond is using that netdev, so remove it
3015              * from the bond.  The client will need to reconfigure everything
3016              * after deleting ports, so then the slave will get re-added. */
3017             bundle_remove(&ofport->up);
3018         }
3019     }
3020     return error;
3021 }
3022
3023 static int
3024 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3025 {
3026     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3027     int error;
3028
3029     push_all_stats();
3030
3031     error = netdev_get_stats(ofport->up.netdev, stats);
3032
3033     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3034         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3035
3036         ovs_mutex_lock(&ofproto->stats_mutex);
3037         /* ofproto->stats.tx_packets represents packets that we created
3038          * internally and sent to some port (e.g. packets sent with
3039          * ofproto_dpif_send_packet()).  Account for them as if they had
3040          * come from OFPP_LOCAL and got forwarded. */
3041
3042         if (stats->rx_packets != UINT64_MAX) {
3043             stats->rx_packets += ofproto->stats.tx_packets;
3044         }
3045
3046         if (stats->rx_bytes != UINT64_MAX) {
3047             stats->rx_bytes += ofproto->stats.tx_bytes;
3048         }
3049
3050         /* ofproto->stats.rx_packets represents packets that were received on
3051          * some port and we processed internally and dropped (e.g. STP).
3052          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3053
3054         if (stats->tx_packets != UINT64_MAX) {
3055             stats->tx_packets += ofproto->stats.rx_packets;
3056         }
3057
3058         if (stats->tx_bytes != UINT64_MAX) {
3059             stats->tx_bytes += ofproto->stats.rx_bytes;
3060         }
3061         ovs_mutex_unlock(&ofproto->stats_mutex);
3062     }
3063
3064     return error;
3065 }
3066
3067 struct port_dump_state {
3068     uint32_t bucket;
3069     uint32_t offset;
3070     bool ghost;
3071
3072     struct ofproto_port port;
3073     bool has_port;
3074 };
3075
3076 static int
3077 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3078 {
3079     *statep = xzalloc(sizeof(struct port_dump_state));
3080     return 0;
3081 }
3082
3083 static int
3084 port_dump_next(const struct ofproto *ofproto_, void *state_,
3085                struct ofproto_port *port)
3086 {
3087     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3088     struct port_dump_state *state = state_;
3089     const struct sset *sset;
3090     struct sset_node *node;
3091
3092     if (state->has_port) {
3093         ofproto_port_destroy(&state->port);
3094         state->has_port = false;
3095     }
3096     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3097     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3098         int error;
3099
3100         error = port_query_by_name(ofproto_, node->name, &state->port);
3101         if (!error) {
3102             *port = state->port;
3103             state->has_port = true;
3104             return 0;
3105         } else if (error != ENODEV) {
3106             return error;
3107         }
3108     }
3109
3110     if (!state->ghost) {
3111         state->ghost = true;
3112         state->bucket = 0;
3113         state->offset = 0;
3114         return port_dump_next(ofproto_, state_, port);
3115     }
3116
3117     return EOF;
3118 }
3119
3120 static int
3121 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3122 {
3123     struct port_dump_state *state = state_;
3124
3125     if (state->has_port) {
3126         ofproto_port_destroy(&state->port);
3127     }
3128     free(state);
3129     return 0;
3130 }
3131
3132 static int
3133 port_poll(const struct ofproto *ofproto_, char **devnamep)
3134 {
3135     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3136
3137     if (ofproto->port_poll_errno) {
3138         int error = ofproto->port_poll_errno;
3139         ofproto->port_poll_errno = 0;
3140         return error;
3141     }
3142
3143     if (sset_is_empty(&ofproto->port_poll_set)) {
3144         return EAGAIN;
3145     }
3146
3147     *devnamep = sset_pop(&ofproto->port_poll_set);
3148     return 0;
3149 }
3150
3151 static void
3152 port_poll_wait(const struct ofproto *ofproto_)
3153 {
3154     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3155     dpif_port_poll_wait(ofproto->backer->dpif);
3156 }
3157
3158 static int
3159 port_is_lacp_current(const struct ofport *ofport_)
3160 {
3161     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3162     return (ofport->bundle && ofport->bundle->lacp
3163             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3164             : -1);
3165 }
3166 \f
3167 /* Upcall handling. */
3168
3169 struct flow_miss_op {
3170     struct dpif_op dpif_op;
3171
3172     uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3173     struct xlate_out xout;
3174     bool xout_garbage;           /* 'xout' needs to be uninitialized? */
3175
3176     struct ofpbuf mask;          /* Flow mask for "put" ops. */
3177     struct odputil_keybuf maskbuf;
3178
3179     /* If this is a "put" op, then a pointer to the subfacet that should
3180      * be marked as uninstalled if the operation fails. */
3181     struct subfacet *subfacet;
3182 };
3183
3184 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3185  * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3186  * and (usually) installing a datapath flow.  The answer is usually "yes" (a
3187  * return value of true).  However, for short flows the cost of bookkeeping is
3188  * much higher than the benefits, so when the datapath holds a large number of
3189  * flows we impose some heuristics to decide which flows are likely to be worth
3190  * tracking. */
3191 static bool
3192 flow_miss_should_make_facet(struct flow_miss *miss)
3193 {
3194     struct dpif_backer *backer = miss->ofproto->backer;
3195     uint32_t hash;
3196
3197     switch (flow_miss_model) {
3198     case OFPROTO_HANDLE_MISS_AUTO:
3199         break;
3200     case OFPROTO_HANDLE_MISS_WITH_FACETS:
3201         return true;
3202     case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3203         return false;
3204     }
3205
3206     if (!backer->governor) {
3207         size_t n_subfacets;
3208
3209         n_subfacets = hmap_count(&backer->subfacets);
3210         if (n_subfacets * 2 <= flow_eviction_threshold) {
3211             return true;
3212         }
3213
3214         backer->governor = governor_create();
3215     }
3216
3217     hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3218     return governor_should_install_flow(backer->governor, hash,
3219                                         miss->stats.n_packets);
3220 }
3221
3222 /* Handles 'miss', which matches 'facet'.  May add any required datapath
3223  * operations to 'ops', incrementing '*n_ops' for each new op.
3224  *
3225  * All of the packets in 'miss' are considered to have arrived at time
3226  * 'miss->stats.used'.  This is really important only for new facets: if we
3227  * just called time_msec() here, then the new subfacet or its packets could
3228  * look (occasionally) as though it was used some time after the facet was
3229  * used.  That can make a one-packet flow look like it has a nonzero duration,
3230  * which looks odd in e.g. NetFlow statistics. */
3231 static void
3232 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3233                             struct flow_miss_op *ops, size_t *n_ops)
3234 {
3235     enum subfacet_path want_path;
3236     struct subfacet *subfacet;
3237     uint32_t key_hash;
3238
3239     /* Update facet stats. */
3240     facet->packet_count += miss->stats.n_packets;
3241     facet->prev_packet_count += miss->stats.n_packets;
3242     facet->byte_count += miss->stats.n_bytes;
3243     facet->prev_byte_count += miss->stats.n_bytes;
3244
3245     /* Look for an existing subfacet.  If we find one, update its used time. */
3246     key_hash = odp_flow_key_hash(miss->key, miss->key_len);
3247     if (!list_is_empty(&facet->subfacets)) {
3248         subfacet = subfacet_find(miss->ofproto->backer,
3249                                  miss->key, miss->key_len, key_hash);
3250         if (subfacet) {
3251             if (subfacet->facet == facet) {
3252                 subfacet->used = MAX(subfacet->used, miss->stats.used);
3253             } else {
3254                 /* This shouldn't happen. */
3255                 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3256                 subfacet_destroy(subfacet);
3257                 subfacet = NULL;
3258             }
3259         }
3260     } else {
3261         subfacet = NULL;
3262     }
3263
3264     /* Don't install the flow if it's the result of the "userspace"
3265      * action for an already installed facet.  This can occur when a
3266      * datapath flow with wildcards has a "userspace" action and flows
3267      * sent to userspace result in a different subfacet, which will then
3268      * be rejected as overlapping by the datapath. */
3269     if (miss->upcall_type == DPIF_UC_ACTION
3270         && !list_is_empty(&facet->subfacets)) {
3271         return;
3272     }
3273
3274     /* Create a subfacet, if we don't already have one. */
3275     if (!subfacet) {
3276         subfacet = subfacet_create(facet, miss, key_hash);
3277     }
3278
3279     /* Install the subfacet, if it's not already installed. */
3280     want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3281     if (subfacet->path != want_path) {
3282         struct flow_miss_op *op = &ops[(*n_ops)++];
3283         struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3284
3285         subfacet->path = want_path;
3286
3287         ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3288         if (enable_megaflows) {
3289             odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3290                                    &miss->flow, UINT32_MAX);
3291         }
3292
3293         op->xout_garbage = false;
3294         op->dpif_op.type = DPIF_OP_FLOW_PUT;
3295         op->subfacet = subfacet;
3296         put->flags = DPIF_FP_CREATE;
3297         put->key = miss->key;
3298         put->key_len = miss->key_len;
3299         put->mask = op->mask.data;
3300         put->mask_len = op->mask.size;
3301
3302         if (want_path == SF_FAST_PATH) {
3303             put->actions = facet->xout.odp_actions.data;
3304             put->actions_len = facet->xout.odp_actions.size;
3305         } else {
3306             compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3307                               op->slow_stub, sizeof op->slow_stub,
3308                               &put->actions, &put->actions_len);
3309         }
3310         put->stats = NULL;
3311     }
3312 }
3313
3314 /* Handles flow miss 'miss'.  May add any required datapath operations
3315  * to 'ops', incrementing '*n_ops' for each new op. */
3316 static void
3317 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3318                  size_t *n_ops)
3319 {
3320     struct facet *facet;
3321
3322     miss->ofproto->n_missed += miss->stats.n_packets;
3323
3324     facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3325     if (!facet) {
3326         /* There does not exist a bijection between 'struct flow' and datapath
3327          * flow keys with fitness ODP_FIT_TO_LITTLE.  This breaks a fundamental
3328          * assumption used throughout the facet and subfacet handling code.
3329          * Since we have to handle these misses in userspace anyway, we simply
3330          * skip facet creation, avoiding the problem altogether. */
3331         if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3332             || !flow_miss_should_make_facet(miss)) {
3333             return;
3334         }
3335
3336         facet = facet_create(miss);
3337     }
3338     handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3339 }
3340
3341 static struct drop_key *
3342 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3343                 size_t key_len)
3344 {
3345     struct drop_key *drop_key;
3346
3347     HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3348                              &backer->drop_keys) {
3349         if (drop_key->key_len == key_len
3350             && !memcmp(drop_key->key, key, key_len)) {
3351             return drop_key;
3352         }
3353     }
3354     return NULL;
3355 }
3356
3357 static void
3358 drop_key_clear(struct dpif_backer *backer)
3359 {
3360     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3361     struct drop_key *drop_key, *next;
3362
3363     HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3364         int error;
3365
3366         error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3367                               NULL);
3368         if (error && !VLOG_DROP_WARN(&rl)) {
3369             struct ds ds = DS_EMPTY_INITIALIZER;
3370             odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3371             VLOG_WARN("Failed to delete drop key (%s) (%s)",
3372                       ovs_strerror(error), ds_cstr(&ds));
3373             ds_destroy(&ds);
3374         }
3375
3376         hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3377         drop_key_destroy(drop_key);
3378     }
3379
3380     udpif_drop_key_clear(backer->udpif);
3381 }
3382
3383 static void
3384 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3385 {
3386     struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3387     struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3388     struct flow_miss *miss;
3389     size_t n_ops, i;
3390
3391     /* Process each element in the to-do list, constructing the set of
3392      * operations to batch. */
3393     n_ops = 0;
3394     HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3395         handle_flow_miss(miss, flow_miss_ops, &n_ops);
3396     }
3397     ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3398
3399     /* Execute batch. */
3400     for (i = 0; i < n_ops; i++) {
3401         dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3402     }
3403     dpif_operate(backer->dpif, dpif_ops, n_ops);
3404
3405     for (i = 0; i < n_ops; i++) {
3406         if (dpif_ops[i]->error != 0
3407             && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3408             && flow_miss_ops[i].subfacet) {
3409             struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3410
3411             COVERAGE_INC(subfacet_install_fail);
3412
3413             /* Zero-out subfacet counters when installation failed, but
3414              * datapath reported hits.  This should not happen and
3415              * indicates a bug, since if the datapath flow exists, we
3416              * should not be attempting to create a new subfacet.  A
3417              * buggy datapath could trigger this, so just zero out the
3418              * counters and log an error. */
3419             if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3420                 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3421                             "datapath reported hits");
3422                 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3423             }
3424
3425             subfacet->path = SF_NOT_INSTALLED;
3426         }
3427     }
3428 }
3429
3430 static void
3431 handle_upcalls(struct dpif_backer *backer)
3432 {
3433     struct flow_miss_batch *fmb;
3434     int n_processed;
3435
3436     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3437         struct drop_key *drop_key = drop_key_next(backer->udpif);
3438         if (!drop_key) {
3439             break;
3440         }
3441
3442         if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3443             hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3444                         hash_bytes(drop_key->key, drop_key->key_len, 0));
3445             dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3446                           drop_key->key, drop_key->key_len,
3447                           NULL, 0, NULL, 0, NULL);
3448         } else {
3449             drop_key_destroy(drop_key);
3450         }
3451     }
3452
3453     fmb = flow_miss_batch_next(backer->udpif);
3454     if (fmb) {
3455         handle_flow_misses(backer, fmb);
3456         flow_miss_batch_destroy(fmb);
3457     }
3458 }
3459 \f
3460 /* Flow expiration. */
3461
3462 static int subfacet_max_idle(const struct dpif_backer *);
3463 static void update_stats(struct dpif_backer *);
3464 static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex);
3465 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3466
3467 /* This function is called periodically by run().  Its job is to collect
3468  * updates for the flows that have been installed into the datapath, most
3469  * importantly when they last were used, and then use that information to
3470  * expire flows that have not been used recently.
3471  *
3472  * Returns the number of milliseconds after which it should be called again. */
3473 static int
3474 expire(struct dpif_backer *backer)
3475 {
3476     struct ofproto_dpif *ofproto;
3477     size_t n_subfacets;
3478     int max_idle;
3479
3480     /* Periodically clear out the drop keys in an effort to keep them
3481      * relatively few. */
3482     drop_key_clear(backer);
3483
3484     /* Update stats for each flow in the backer. */
3485     update_stats(backer);
3486
3487     n_subfacets = hmap_count(&backer->subfacets);
3488     if (n_subfacets) {
3489         struct subfacet *subfacet;
3490         long long int total, now;
3491
3492         total = 0;
3493         now = time_msec();
3494         HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3495             total += now - subfacet->created;
3496         }
3497         backer->avg_subfacet_life += total / n_subfacets;
3498     }
3499     backer->avg_subfacet_life /= 2;
3500
3501     backer->avg_n_subfacet += n_subfacets;
3502     backer->avg_n_subfacet /= 2;
3503
3504     backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3505
3506     max_idle = subfacet_max_idle(backer);
3507     expire_subfacets(backer, max_idle);
3508
3509     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3510         struct rule *rule, *next_rule;
3511
3512         if (ofproto->backer != backer) {
3513             continue;
3514         }
3515
3516         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3517          * has passed. */
3518         ovs_mutex_lock(&ofproto_mutex);
3519         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3520                             &ofproto->up.expirable) {
3521             rule_expire(rule_dpif_cast(rule));
3522         }
3523         ovs_mutex_unlock(&ofproto_mutex);
3524
3525         /* All outstanding data in existing flows has been accounted, so it's a
3526          * good time to do bond rebalancing. */
3527         if (ofproto->has_bonded_bundles) {
3528             struct ofbundle *bundle;
3529
3530             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3531                 if (bundle->bond) {
3532                     bond_rebalance(bundle->bond);
3533                 }
3534             }
3535         }
3536     }
3537
3538     return MIN(max_idle, 1000);
3539 }
3540
3541 /* Updates flow table statistics given that the datapath just reported 'stats'
3542  * as 'subfacet''s statistics. */
3543 static void
3544 update_subfacet_stats(struct subfacet *subfacet,
3545                       const struct dpif_flow_stats *stats)
3546 {
3547     struct facet *facet = subfacet->facet;
3548     struct dpif_flow_stats diff;
3549
3550     diff.tcp_flags = stats->tcp_flags;
3551     diff.used = stats->used;
3552
3553     if (stats->n_packets >= subfacet->dp_packet_count) {
3554         diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3555     } else {
3556         VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3557         diff.n_packets = 0;
3558     }
3559
3560     if (stats->n_bytes >= subfacet->dp_byte_count) {
3561         diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3562     } else {
3563         VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3564         diff.n_bytes = 0;
3565     }
3566
3567     facet->ofproto->n_hit += diff.n_packets;
3568     subfacet->dp_packet_count = stats->n_packets;
3569     subfacet->dp_byte_count = stats->n_bytes;
3570     subfacet_update_stats(subfacet, &diff);
3571
3572     if (facet->accounted_bytes < facet->byte_count) {
3573         facet_learn(facet);
3574         facet_account(facet);
3575         facet->accounted_bytes = facet->byte_count;
3576     }
3577 }
3578
3579 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3580  * about, or a flow that shouldn't be installed but was anyway.  Delete it. */
3581 static void
3582 delete_unexpected_flow(struct dpif_backer *backer,
3583                        const struct nlattr *key, size_t key_len)
3584 {
3585     if (!VLOG_DROP_WARN(&rl)) {
3586         struct ds s;
3587
3588         ds_init(&s);
3589         odp_flow_key_format(key, key_len, &s);
3590         VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3591         ds_destroy(&s);
3592     }
3593
3594     COVERAGE_INC(facet_unexpected);
3595     dpif_flow_del(backer->dpif, key, key_len, NULL);
3596 }
3597
3598 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3599  *
3600  * This function also pushes statistics updates to rules which each facet
3601  * resubmits into.  Generally these statistics will be accurate.  However, if a
3602  * facet changes the rule it resubmits into at some time in between
3603  * update_stats() runs, it is possible that statistics accrued to the
3604  * old rule will be incorrectly attributed to the new rule.  This could be
3605  * avoided by calling update_stats() whenever rules are created or
3606  * deleted.  However, the performance impact of making so many calls to the
3607  * datapath do not justify the benefit of having perfectly accurate statistics.
3608  *
3609  * In addition, this function maintains per ofproto flow hit counts. The patch
3610  * port is not treated specially. e.g. A packet ingress from br0 patched into
3611  * br1 will increase the hit count of br0 by 1, however, does not affect
3612  * the hit or miss counts of br1.
3613  */
3614 static void
3615 update_stats(struct dpif_backer *backer)
3616 {
3617     const struct dpif_flow_stats *stats;
3618     struct dpif_flow_dump dump;
3619     const struct nlattr *key, *mask;
3620     size_t key_len, mask_len;
3621
3622     dpif_flow_dump_start(&dump, backer->dpif);
3623     while (dpif_flow_dump_next(&dump, &key, &key_len,
3624                                &mask, &mask_len, NULL, NULL, &stats)) {
3625         struct subfacet *subfacet;
3626         uint32_t key_hash;
3627
3628         key_hash = odp_flow_key_hash(key, key_len);
3629         subfacet = subfacet_find(backer, key, key_len, key_hash);
3630         switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3631         case SF_FAST_PATH:
3632             update_subfacet_stats(subfacet, stats);
3633             break;
3634
3635         case SF_SLOW_PATH:
3636             /* Stats are updated per-packet. */
3637             break;
3638
3639         case SF_NOT_INSTALLED:
3640         default:
3641             delete_unexpected_flow(backer, key, key_len);
3642             break;
3643         }
3644         run_fast_rl();
3645     }
3646     dpif_flow_dump_done(&dump);
3647 }
3648
3649 /* Calculates and returns the number of milliseconds of idle time after which
3650  * subfacets should expire from the datapath.  When a subfacet expires, we fold
3651  * its statistics into its facet, and when a facet's last subfacet expires, we
3652  * fold its statistic into its rule. */
3653 static int
3654 subfacet_max_idle(const struct dpif_backer *backer)
3655 {
3656     /*
3657      * Idle time histogram.
3658      *
3659      * Most of the time a switch has a relatively small number of subfacets.
3660      * When this is the case we might as well keep statistics for all of them
3661      * in userspace and to cache them in the kernel datapath for performance as
3662      * well.
3663      *
3664      * As the number of subfacets increases, the memory required to maintain
3665      * statistics about them in userspace and in the kernel becomes
3666      * significant.  However, with a large number of subfacets it is likely
3667      * that only a few of them are "heavy hitters" that consume a large amount
3668      * of bandwidth.  At this point, only heavy hitters are worth caching in
3669      * the kernel and maintaining in userspaces; other subfacets we can
3670      * discard.
3671      *
3672      * The technique used to compute the idle time is to build a histogram with
3673      * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each.  Each subfacet
3674      * that is installed in the kernel gets dropped in the appropriate bucket.
3675      * After the histogram has been built, we compute the cutoff so that only
3676      * the most-recently-used 1% of subfacets (but at least
3677      * flow_eviction_threshold flows) are kept cached.  At least
3678      * the most-recently-used bucket of subfacets is kept, so actually an
3679      * arbitrary number of subfacets can be kept in any given expiration run
3680      * (though the next run will delete most of those unless they receive
3681      * additional data).
3682      *
3683      * This requires a second pass through the subfacets, in addition to the
3684      * pass made by update_stats(), because the former function never looks at
3685      * uninstallable subfacets.
3686      */
3687     enum { BUCKET_WIDTH = 100 };
3688     enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3689     int buckets[N_BUCKETS] = { 0 };
3690     int total, subtotal, bucket;
3691     struct subfacet *subfacet;
3692     long long int now;
3693     int i;
3694
3695     total = hmap_count(&backer->subfacets);
3696     if (total <= flow_eviction_threshold) {
3697         return N_BUCKETS * BUCKET_WIDTH;
3698     }
3699
3700     /* Build histogram. */
3701     now = time_msec();
3702     HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3703         long long int idle = now - subfacet->used;
3704         int bucket = (idle <= 0 ? 0
3705                       : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3706                       : (unsigned int) idle / BUCKET_WIDTH);
3707         buckets[bucket]++;
3708     }
3709
3710     /* Find the first bucket whose flows should be expired. */
3711     subtotal = bucket = 0;
3712     do {
3713         subtotal += buckets[bucket++];
3714     } while (bucket < N_BUCKETS &&
3715              subtotal < MAX(flow_eviction_threshold, total / 100));
3716
3717     if (VLOG_IS_DBG_ENABLED()) {
3718         struct ds s;
3719
3720         ds_init(&s);
3721         ds_put_cstr(&s, "keep");
3722         for (i = 0; i < N_BUCKETS; i++) {
3723             if (i == bucket) {
3724                 ds_put_cstr(&s, ", drop");
3725             }
3726             if (buckets[i]) {
3727                 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3728             }
3729         }
3730         VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3731         ds_destroy(&s);
3732     }
3733
3734     return bucket * BUCKET_WIDTH;
3735 }
3736
3737 static void
3738 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3739 {
3740     /* Cutoff time for most flows. */
3741     long long int normal_cutoff = time_msec() - dp_max_idle;
3742
3743     /* We really want to keep flows for special protocols around, so use a more
3744      * conservative cutoff. */
3745     long long int special_cutoff = time_msec() - 10000;
3746
3747     struct subfacet *subfacet, *next_subfacet;
3748     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3749     int n_batch;
3750
3751     n_batch = 0;
3752     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3753                         &backer->subfacets) {
3754         long long int cutoff;
3755
3756         cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3757                                                 | SLOW_STP)
3758                   ? special_cutoff
3759                   : normal_cutoff);
3760         if (subfacet->used < cutoff) {
3761             if (subfacet->path != SF_NOT_INSTALLED) {
3762                 batch[n_batch++] = subfacet;
3763                 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3764                     subfacet_destroy_batch(backer, batch, n_batch);
3765                     n_batch = 0;
3766                 }
3767             } else {
3768                 subfacet_destroy(subfacet);
3769             }
3770         }
3771     }
3772
3773     if (n_batch > 0) {
3774         subfacet_destroy_batch(backer, batch, n_batch);
3775     }
3776 }
3777
3778 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3779  * then delete it entirely. */
3780 static void
3781 rule_expire(struct rule_dpif *rule)
3782     OVS_REQUIRES(ofproto_mutex)
3783 {
3784     uint16_t idle_timeout, hard_timeout;
3785     long long int now = time_msec();
3786     int reason;
3787
3788     ovs_assert(!rule->up.pending);
3789
3790     /* Has 'rule' expired? */
3791     ovs_mutex_lock(&rule->up.mutex);
3792     hard_timeout = rule->up.hard_timeout;
3793     idle_timeout = rule->up.idle_timeout;
3794     if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3795         reason = OFPRR_HARD_TIMEOUT;
3796     } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3797         reason = OFPRR_IDLE_TIMEOUT;
3798     } else {
3799         reason = -1;
3800     }
3801     ovs_mutex_unlock(&rule->up.mutex);
3802
3803     if (reason >= 0) {
3804         COVERAGE_INC(ofproto_dpif_expired);
3805         ofproto_rule_expire(&rule->up, reason);
3806     }
3807 }
3808 \f
3809 /* Facets. */
3810
3811 /* Creates and returns a new facet based on 'miss'.
3812  *
3813  * The caller must already have determined that no facet with an identical
3814  * 'miss->flow' exists in 'miss->ofproto'.
3815  *
3816  * 'rule' and 'xout' must have been created based on 'miss'.
3817  *
3818  * 'facet'' statistics are initialized based on 'stats'.
3819  *
3820  * The facet will initially have no subfacets.  The caller should create (at
3821  * least) one subfacet with subfacet_create(). */
3822 static struct facet *
3823 facet_create(const struct flow_miss *miss)
3824 {
3825     struct ofproto_dpif *ofproto = miss->ofproto;
3826     struct facet *facet;
3827     struct match match;
3828
3829     COVERAGE_INC(facet_create);
3830     facet = xzalloc(sizeof *facet);
3831     facet->ofproto = miss->ofproto;
3832     facet->used = miss->stats.used;
3833     facet->flow = miss->flow;
3834     facet->learn_rl = time_msec() + 500;
3835
3836     list_init(&facet->subfacets);
3837     netflow_flow_init(&facet->nf_flow);
3838     netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3839
3840     xlate_out_copy(&facet->xout, &miss->xout);
3841
3842     match_init(&match, &facet->flow, &facet->xout.wc);
3843     cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3844     ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3845     classifier_insert(&ofproto->facets, &facet->cr);
3846     ovs_rwlock_unlock(&ofproto->facets.rwlock);
3847
3848     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
3849     return facet;
3850 }
3851
3852 static void
3853 facet_free(struct facet *facet)
3854 {
3855     if (facet) {
3856         xlate_out_uninit(&facet->xout);
3857         free(facet);
3858     }
3859 }
3860
3861 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3862  * 'flow' must reflect the data in 'packet'. */
3863 int
3864 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3865                              const struct flow *flow,
3866                              struct rule_dpif *rule,
3867                              const struct ofpact *ofpacts, size_t ofpacts_len,
3868                              struct ofpbuf *packet)
3869 {
3870     struct odputil_keybuf keybuf;
3871     struct dpif_flow_stats stats;
3872     struct xlate_out xout;
3873     struct xlate_in xin;
3874     ofp_port_t in_port;
3875     struct ofpbuf key;
3876     int error;
3877
3878     ovs_assert((rule != NULL) != (ofpacts != NULL));
3879
3880     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3881     if (rule) {
3882         rule_dpif_credit_stats(rule, &stats);
3883     }
3884
3885     xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
3886     xin.ofpacts = ofpacts;
3887     xin.ofpacts_len = ofpacts_len;
3888     xin.resubmit_stats = &stats;
3889     xlate_actions(&xin, &xout);
3890
3891     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3892     in_port = flow->in_port.ofp_port;
3893     if (in_port == OFPP_NONE) {
3894         in_port = OFPP_LOCAL;
3895     }
3896     odp_flow_key_from_flow(&key, flow, ofp_port_to_odp_port(ofproto, in_port));
3897
3898     error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
3899                          xout.odp_actions.data, xout.odp_actions.size, packet,
3900                          (xout.slow & SLOW_ACTION) != 0);
3901     xlate_out_uninit(&xout);
3902
3903     return error;
3904 }
3905
3906 /* Remove 'facet' from its ofproto and free up the associated memory:
3907  *
3908  *   - If 'facet' was installed in the datapath, uninstalls it and updates its
3909  *     rule's statistics, via subfacet_uninstall().
3910  *
3911  *   - Removes 'facet' from its rule and from ofproto->facets.
3912  */
3913 static void
3914 facet_remove(struct facet *facet)
3915 {
3916     struct subfacet *subfacet, *next_subfacet;
3917
3918     COVERAGE_INC(facet_remove);
3919     ovs_assert(!list_is_empty(&facet->subfacets));
3920
3921     /* First uninstall all of the subfacets to get final statistics. */
3922     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3923         subfacet_uninstall(subfacet);
3924     }
3925
3926     /* Flush the final stats to the rule.
3927      *
3928      * This might require us to have at least one subfacet around so that we
3929      * can use its actions for accounting in facet_account(), which is why we
3930      * have uninstalled but not yet destroyed the subfacets. */
3931     facet_flush_stats(facet);
3932
3933     /* Now we're really all done so destroy everything. */
3934     LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3935                         &facet->subfacets) {
3936         subfacet_destroy__(subfacet);
3937     }
3938     ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
3939     classifier_remove(&facet->ofproto->facets, &facet->cr);
3940     ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
3941     cls_rule_destroy(&facet->cr);
3942     facet_free(facet);
3943 }
3944
3945 /* Feed information from 'facet' back into the learning table to keep it in
3946  * sync with what is actually flowing through the datapath. */
3947 static void
3948 facet_learn(struct facet *facet)
3949 {
3950     long long int now = time_msec();
3951
3952     if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
3953         return;
3954     }
3955
3956     facet->learn_rl = now + 500;
3957
3958     if (!facet->xout.has_learn
3959         && !facet->xout.has_normal
3960         && (!facet->xout.has_fin_timeout
3961             || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3962         return;
3963     }
3964
3965     facet_push_stats(facet, true);
3966 }
3967
3968 static void
3969 facet_account(struct facet *facet)
3970 {
3971     const struct nlattr *a;
3972     unsigned int left;
3973     ovs_be16 vlan_tci;
3974     uint64_t n_bytes;
3975
3976     if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
3977         return;
3978     }
3979     n_bytes = facet->byte_count - facet->accounted_bytes;
3980
3981     /* This loop feeds byte counters to bond_account() for rebalancing to use
3982      * as a basis.  We also need to track the actual VLAN on which the packet
3983      * is going to be sent to ensure that it matches the one passed to
3984      * bond_choose_output_slave().  (Otherwise, we will account to the wrong
3985      * hash bucket.)
3986      *
3987      * We use the actions from an arbitrary subfacet because they should all
3988      * be equally valid for our purpose. */
3989     vlan_tci = facet->flow.vlan_tci;
3990     NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
3991                              facet->xout.odp_actions.size) {
3992         const struct ovs_action_push_vlan *vlan;
3993         struct ofport_dpif *port;
3994
3995         switch (nl_attr_type(a)) {
3996         case OVS_ACTION_ATTR_OUTPUT:
3997             port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
3998             if (port && port->bundle && port->bundle->bond) {
3999                 bond_account(port->bundle->bond, &facet->flow,
4000                              vlan_tci_to_vid(vlan_tci), n_bytes);
4001             }
4002             break;
4003
4004         case OVS_ACTION_ATTR_POP_VLAN:
4005             vlan_tci = htons(0);
4006             break;
4007
4008         case OVS_ACTION_ATTR_PUSH_VLAN:
4009             vlan = nl_attr_get(a);
4010             vlan_tci = vlan->vlan_tci;
4011             break;
4012         }
4013     }
4014 }
4015
4016 /* Returns true if the only action for 'facet' is to send to the controller.
4017  * (We don't report NetFlow expiration messages for such facets because they
4018  * are just part of the control logic for the network, not real traffic). */
4019 static bool
4020 facet_is_controller_flow(struct facet *facet)
4021 {
4022     if (facet) {
4023         struct ofproto_dpif *ofproto = facet->ofproto;
4024         const struct ofpact *ofpacts;
4025         struct rule_actions *actions;
4026         struct rule_dpif *rule;
4027         size_t ofpacts_len;
4028         bool is_controller;
4029
4030         rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4031         actions = rule_dpif_get_actions(rule);
4032         rule_dpif_unref(rule);
4033
4034         ofpacts_len = actions->ofpacts_len;
4035         ofpacts = actions->ofpacts;
4036         is_controller = ofpacts_len > 0
4037             && ofpacts->type == OFPACT_CONTROLLER
4038             && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4039         rule_actions_unref(actions);
4040
4041         return is_controller;
4042     }
4043     return false;
4044 }
4045
4046 /* Folds all of 'facet''s statistics into its rule.  Also updates the
4047  * accounting ofhook and emits a NetFlow expiration if appropriate.  All of
4048  * 'facet''s statistics in the datapath should have been zeroed and folded into
4049  * its packet and byte counts before this function is called. */
4050 static void
4051 facet_flush_stats(struct facet *facet)
4052 {
4053     struct ofproto_dpif *ofproto = facet->ofproto;
4054     struct subfacet *subfacet;
4055
4056     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4057         ovs_assert(!subfacet->dp_byte_count);
4058         ovs_assert(!subfacet->dp_packet_count);
4059     }
4060
4061     facet_push_stats(facet, false);
4062     if (facet->accounted_bytes < facet->byte_count) {
4063         facet_account(facet);
4064         facet->accounted_bytes = facet->byte_count;
4065     }
4066
4067     if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4068         struct ofexpired expired;
4069         expired.flow = facet->flow;
4070         expired.packet_count = facet->packet_count;
4071         expired.byte_count = facet->byte_count;
4072         expired.used = facet->used;
4073         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4074     }
4075
4076     /* Reset counters to prevent double counting if 'facet' ever gets
4077      * reinstalled. */
4078     facet_reset_counters(facet);
4079
4080     netflow_flow_clear(&facet->nf_flow);
4081     facet->tcp_flags = 0;
4082 }
4083
4084 /* Searches 'ofproto''s table of facets for one which would be responsible for
4085  * 'flow'.  Returns it if found, otherwise a null pointer.
4086  *
4087  * The returned facet might need revalidation; use facet_lookup_valid()
4088  * instead if that is important. */
4089 static struct facet *
4090 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4091 {
4092     struct cls_rule *cr;
4093
4094     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4095     cr = classifier_lookup(&ofproto->facets, flow, NULL);
4096     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4097     return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4098 }
4099
4100 /* Searches 'ofproto''s table of facets for one capable that covers
4101  * 'flow'.  Returns it if found, otherwise a null pointer.
4102  *
4103  * The returned facet is guaranteed to be valid. */
4104 static struct facet *
4105 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4106 {
4107     struct facet *facet;
4108
4109     facet = facet_find(ofproto, flow);
4110     if (facet
4111         && ofproto->backer->need_revalidate
4112         && !facet_revalidate(facet)) {
4113         return NULL;
4114     }
4115
4116     return facet;
4117 }
4118
4119 static bool
4120 facet_check_consistency(struct facet *facet)
4121 {
4122     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4123
4124     struct xlate_out xout;
4125     struct xlate_in xin;
4126     bool ok;
4127
4128     /* Check the datapath actions for consistency. */
4129     xlate_in_init(&xin, facet->ofproto, &facet->flow, NULL, 0, NULL);
4130     xlate_actions(&xin, &xout);
4131
4132     ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4133         && facet->xout.slow == xout.slow;
4134     if (!ok && !VLOG_DROP_WARN(&rl)) {
4135         struct ds s = DS_EMPTY_INITIALIZER;
4136
4137         flow_format(&s, &facet->flow);
4138         ds_put_cstr(&s, ": inconsistency in facet");
4139
4140         if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4141             ds_put_cstr(&s, " (actions were: ");
4142             format_odp_actions(&s, facet->xout.odp_actions.data,
4143                                facet->xout.odp_actions.size);
4144             ds_put_cstr(&s, ") (correct actions: ");
4145             format_odp_actions(&s, xout.odp_actions.data,
4146                                xout.odp_actions.size);
4147             ds_put_char(&s, ')');
4148         }
4149
4150         if (facet->xout.slow != xout.slow) {
4151             ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4152         }
4153
4154         ds_destroy(&s);
4155     }
4156     xlate_out_uninit(&xout);
4157
4158     return ok;
4159 }
4160
4161 /* Re-searches the classifier for 'facet':
4162  *
4163  *   - If the rule found is different from 'facet''s current rule, moves
4164  *     'facet' to the new rule and recompiles its actions.
4165  *
4166  *   - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4167  *     where it is and recompiles its actions anyway.
4168  *
4169  *   - If any of 'facet''s subfacets correspond to a new flow according to
4170  *     xlate_receive(), 'facet' is removed.
4171  *
4172  *   Returns true if 'facet' is still valid.  False if 'facet' was removed. */
4173 static bool
4174 facet_revalidate(struct facet *facet)
4175 {
4176     struct ofproto_dpif *ofproto = facet->ofproto;
4177     struct rule_dpif *new_rule;
4178     struct subfacet *subfacet;
4179     struct flow_wildcards wc;
4180     struct xlate_out xout;
4181     struct xlate_in xin;
4182
4183     COVERAGE_INC(facet_revalidate);
4184
4185     /* Check that child subfacets still correspond to this facet.  Tunnel
4186      * configuration changes could cause a subfacet's OpenFlow in_port to
4187      * change. */
4188     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4189         struct ofproto_dpif *recv_ofproto;
4190         struct flow recv_flow;
4191         int error;
4192
4193         error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4194                               subfacet->key_len, &recv_flow, NULL,
4195                               &recv_ofproto, NULL);
4196         if (error
4197             || recv_ofproto != ofproto
4198             || facet != facet_find(ofproto, &recv_flow)) {
4199             facet_remove(facet);
4200             return false;
4201         }
4202     }
4203
4204     flow_wildcards_init_catchall(&wc);
4205     rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4206
4207     /* Calculate new datapath actions.
4208      *
4209      * We do not modify any 'facet' state yet, because we might need to, e.g.,
4210      * emit a NetFlow expiration and, if so, we need to have the old state
4211      * around to properly compose it. */
4212     xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4213     xlate_actions(&xin, &xout);
4214     flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4215     /* Make sure non -packet fields are not masked. If not cleared,
4216      * the memcmp() below may fail, causing an otherwise valid facet
4217      * to be removed. */
4218     flow_wildcards_clear_non_packet_fields(&xout.wc);
4219
4220     /* A facet's slow path reason should only change under dramatic
4221      * circumstances.  Rather than try to update everything, it's simpler to
4222      * remove the facet and start over.
4223      *
4224      * More importantly, if a facet's wildcards change, it will be relatively
4225      * difficult to figure out if its subfacets still belong to it, and if not
4226      * which facet they may belong to.  Again, to avoid the complexity, we
4227      * simply give up instead. */
4228     if (facet->xout.slow != xout.slow
4229         || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4230         facet_remove(facet);
4231         xlate_out_uninit(&xout);
4232         rule_dpif_unref(new_rule);
4233         return false;
4234     }
4235
4236     if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4237         LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4238             if (subfacet->path == SF_FAST_PATH) {
4239                 struct dpif_flow_stats stats;
4240
4241                 subfacet_install(subfacet, &xout.odp_actions, &stats);
4242                 subfacet_update_stats(subfacet, &stats);
4243             }
4244         }
4245
4246         facet_flush_stats(facet);
4247
4248         ofpbuf_clear(&facet->xout.odp_actions);
4249         ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4250                    xout.odp_actions.size);
4251     }
4252
4253     /* Update 'facet' now that we've taken care of all the old state. */
4254     facet->xout.slow = xout.slow;
4255     facet->xout.has_learn = xout.has_learn;
4256     facet->xout.has_normal = xout.has_normal;
4257     facet->xout.has_fin_timeout = xout.has_fin_timeout;
4258     facet->xout.nf_output_iface = xout.nf_output_iface;
4259     facet->xout.mirrors = xout.mirrors;
4260     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4261
4262     ovs_mutex_lock(&new_rule->up.mutex);
4263     facet->used = MAX(facet->used, new_rule->up.created);
4264     ovs_mutex_unlock(&new_rule->up.mutex);
4265
4266     xlate_out_uninit(&xout);
4267     rule_dpif_unref(new_rule);
4268     return true;
4269 }
4270
4271 static void
4272 facet_reset_counters(struct facet *facet)
4273 {
4274     facet->packet_count = 0;
4275     facet->byte_count = 0;
4276     facet->prev_packet_count = 0;
4277     facet->prev_byte_count = 0;
4278     facet->accounted_bytes = 0;
4279 }
4280
4281 static void
4282 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4283                 struct dpif_flow_stats *stats, bool may_learn)
4284 {
4285     struct ofport_dpif *in_port;
4286     struct xlate_in xin;
4287
4288     in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4289     if (in_port && in_port->is_tunnel) {
4290         netdev_vport_inc_rx(in_port->up.netdev, stats);
4291         if (in_port->bfd) {
4292             bfd_account_rx(in_port->bfd, stats);
4293         }
4294     }
4295
4296     xlate_in_init(&xin, ofproto, flow, NULL, stats->tcp_flags, NULL);
4297     xin.resubmit_stats = stats;
4298     xin.may_learn = may_learn;
4299     xlate_actions_for_side_effects(&xin);
4300 }
4301
4302 static void
4303 facet_push_stats(struct facet *facet, bool may_learn)
4304 {
4305     struct dpif_flow_stats stats;
4306
4307     ovs_assert(facet->packet_count >= facet->prev_packet_count);
4308     ovs_assert(facet->byte_count >= facet->prev_byte_count);
4309     ovs_assert(facet->used >= facet->prev_used);
4310
4311     stats.n_packets = facet->packet_count - facet->prev_packet_count;
4312     stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4313     stats.used = facet->used;
4314     stats.tcp_flags = facet->tcp_flags;
4315
4316     if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4317         facet->prev_packet_count = facet->packet_count;
4318         facet->prev_byte_count = facet->byte_count;
4319         facet->prev_used = facet->used;
4320
4321         netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4322                                  facet->used);
4323         netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4324         mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4325                             stats.n_packets, stats.n_bytes);
4326         flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4327     }
4328 }
4329
4330 static void
4331 push_all_stats__(bool run_fast)
4332 {
4333     static long long int rl = LLONG_MIN;
4334     struct ofproto_dpif *ofproto;
4335
4336     if (time_msec() < rl) {
4337         return;
4338     }
4339
4340     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4341         struct cls_cursor cursor;
4342         struct facet *facet;
4343
4344         ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4345         cls_cursor_init(&cursor, &ofproto->facets, NULL);
4346         CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4347             facet_push_stats(facet, false);
4348             if (run_fast) {
4349                 run_fast_rl();
4350             }
4351         }
4352         ovs_rwlock_unlock(&ofproto->facets.rwlock);
4353     }
4354
4355     rl = time_msec() + 100;
4356 }
4357
4358 static void
4359 push_all_stats(void)
4360 {
4361     push_all_stats__(true);
4362 }
4363
4364 void
4365 rule_dpif_credit_stats(struct rule_dpif *rule,
4366                        const struct dpif_flow_stats *stats)
4367 {
4368     ovs_mutex_lock(&rule->stats_mutex);
4369     rule->packet_count += stats->n_packets;
4370     rule->byte_count += stats->n_bytes;
4371     rule->up.used = MAX(rule->up.used, stats->used);
4372     ovs_mutex_unlock(&rule->stats_mutex);
4373 }
4374
4375 bool
4376 rule_dpif_is_fail_open(const struct rule_dpif *rule)
4377 {
4378     return is_fail_open_rule(&rule->up);
4379 }
4380
4381 bool
4382 rule_dpif_is_table_miss(const struct rule_dpif *rule)
4383 {
4384     return rule_is_table_miss(&rule->up);
4385 }
4386
4387 ovs_be64
4388 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4389     OVS_REQUIRES(rule->up.mutex)
4390 {
4391     return rule->up.flow_cookie;
4392 }
4393
4394 void
4395 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4396                      uint16_t hard_timeout)
4397 {
4398     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4399 }
4400
4401 /* Returns 'rule''s actions.  The caller owns a reference on the returned
4402  * actions and must eventually release it (with rule_actions_unref()) to avoid
4403  * a memory leak. */
4404 struct rule_actions *
4405 rule_dpif_get_actions(const struct rule_dpif *rule)
4406 {
4407     return rule_get_actions(&rule->up);
4408 }
4409 \f
4410 /* Subfacets. */
4411
4412 static struct subfacet *
4413 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4414               size_t key_len, uint32_t key_hash)
4415 {
4416     struct subfacet *subfacet;
4417
4418     HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4419                              &backer->subfacets) {
4420         if (subfacet->key_len == key_len
4421             && !memcmp(key, subfacet->key, key_len)) {
4422             return subfacet;
4423         }
4424     }
4425
4426     return NULL;
4427 }
4428
4429 /* Creates and returns a new subfacet within 'facet' for the flow in 'miss'.
4430  * 'key_hash' must be a hash over miss->key.  The caller must have already
4431  * ensured that no subfacet subfacet already exists. */
4432 static struct subfacet *
4433 subfacet_create(struct facet *facet, struct flow_miss *miss, uint32_t key_hash)
4434 {
4435     struct dpif_backer *backer = miss->ofproto->backer;
4436     const struct nlattr *key = miss->key;
4437     size_t key_len = miss->key_len;
4438     struct subfacet *subfacet;
4439
4440     subfacet = (list_is_empty(&facet->subfacets)
4441                 ? &facet->one_subfacet
4442                 : xmalloc(sizeof *subfacet));
4443
4444     COVERAGE_INC(subfacet_create);
4445     hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4446     list_push_back(&facet->subfacets, &subfacet->list_node);
4447     subfacet->facet = facet;
4448     subfacet->key = xmemdup(key, key_len);
4449     subfacet->key_len = key_len;
4450     subfacet->used = miss->stats.used;
4451     subfacet->created = subfacet->used;
4452     subfacet->dp_packet_count = 0;
4453     subfacet->dp_byte_count = 0;
4454     subfacet->path = SF_NOT_INSTALLED;
4455     subfacet->backer = backer;
4456
4457     return subfacet;
4458 }
4459
4460 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4461  * its facet within 'ofproto', and frees it. */
4462 static void
4463 subfacet_destroy__(struct subfacet *subfacet)
4464 {
4465     struct facet *facet = subfacet->facet;
4466
4467     COVERAGE_INC(subfacet_destroy);
4468     subfacet_uninstall(subfacet);
4469     hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4470     list_remove(&subfacet->list_node);
4471     free(subfacet->key);
4472     if (subfacet != &facet->one_subfacet) {
4473         free(subfacet);
4474     }
4475 }
4476
4477 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4478  * last remaining subfacet in its facet destroys the facet too. */
4479 static void
4480 subfacet_destroy(struct subfacet *subfacet)
4481 {
4482     struct facet *facet = subfacet->facet;
4483
4484     if (list_is_singleton(&facet->subfacets)) {
4485         /* facet_remove() needs at least one subfacet (it will remove it). */
4486         facet_remove(facet);
4487     } else {
4488         subfacet_destroy__(subfacet);
4489     }
4490 }
4491
4492 static void
4493 subfacet_destroy_batch(struct dpif_backer *backer,
4494                        struct subfacet **subfacets, int n)
4495 {
4496     struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4497     struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4498     struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4499     int i;
4500
4501     for (i = 0; i < n; i++) {
4502         ops[i].type = DPIF_OP_FLOW_DEL;
4503         ops[i].u.flow_del.key = subfacets[i]->key;
4504         ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4505         ops[i].u.flow_del.stats = &stats[i];
4506         opsp[i] = &ops[i];
4507     }
4508
4509     dpif_operate(backer->dpif, opsp, n);
4510     for (i = 0; i < n; i++) {
4511         subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4512         subfacets[i]->path = SF_NOT_INSTALLED;
4513         subfacet_destroy(subfacets[i]);
4514         run_fast_rl();
4515     }
4516 }
4517
4518 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4519  * bytes of actions in 'actions'.  If 'stats' is non-null, statistics counters
4520  * in the datapath will be zeroed and 'stats' will be updated with traffic new
4521  * since 'subfacet' was last updated.
4522  *
4523  * Returns 0 if successful, otherwise a positive errno value. */
4524 static int
4525 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4526                  struct dpif_flow_stats *stats)
4527 {
4528     struct facet *facet = subfacet->facet;
4529     enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4530     const struct nlattr *actions = odp_actions->data;
4531     size_t actions_len = odp_actions->size;
4532     struct odputil_keybuf maskbuf;
4533     struct ofpbuf mask;
4534
4535     uint64_t slow_path_stub[128 / 8];
4536     enum dpif_flow_put_flags flags;
4537     int ret;
4538
4539     flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4540                                                : DPIF_FP_MODIFY;
4541     if (stats) {
4542         flags |= DPIF_FP_ZERO_STATS;
4543     }
4544
4545     if (path == SF_SLOW_PATH) {
4546         compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4547                           slow_path_stub, sizeof slow_path_stub,
4548                           &actions, &actions_len);
4549     }
4550
4551     ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4552     if (enable_megaflows) {
4553         odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4554                                &facet->flow, UINT32_MAX);
4555     }
4556
4557     ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4558                         subfacet->key_len,  mask.data, mask.size,
4559                         actions, actions_len, stats);
4560
4561     if (stats) {
4562         subfacet_reset_dp_stats(subfacet, stats);
4563     }
4564
4565     if (ret) {
4566         COVERAGE_INC(subfacet_install_fail);
4567     } else {
4568         subfacet->path = path;
4569     }
4570     return ret;
4571 }
4572
4573 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4574 static void
4575 subfacet_uninstall(struct subfacet *subfacet)
4576 {
4577     if (subfacet->path != SF_NOT_INSTALLED) {
4578         struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4579         struct dpif_flow_stats stats;
4580         int error;
4581
4582         error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4583                               subfacet->key_len, &stats);
4584         subfacet_reset_dp_stats(subfacet, &stats);
4585         if (!error) {
4586             subfacet_update_stats(subfacet, &stats);
4587         }
4588         subfacet->path = SF_NOT_INSTALLED;
4589     } else {
4590         ovs_assert(subfacet->dp_packet_count == 0);
4591         ovs_assert(subfacet->dp_byte_count == 0);
4592     }
4593 }
4594
4595 /* Resets 'subfacet''s datapath statistics counters.  This should be called
4596  * when 'subfacet''s statistics are cleared in the datapath.  If 'stats' is
4597  * non-null, it should contain the statistics returned by dpif when 'subfacet'
4598  * was reset in the datapath.  'stats' will be modified to include only
4599  * statistics new since 'subfacet' was last updated. */
4600 static void
4601 subfacet_reset_dp_stats(struct subfacet *subfacet,
4602                         struct dpif_flow_stats *stats)
4603 {
4604     if (stats
4605         && subfacet->dp_packet_count <= stats->n_packets
4606         && subfacet->dp_byte_count <= stats->n_bytes) {
4607         stats->n_packets -= subfacet->dp_packet_count;
4608         stats->n_bytes -= subfacet->dp_byte_count;
4609     }
4610
4611     subfacet->dp_packet_count = 0;
4612     subfacet->dp_byte_count = 0;
4613 }
4614
4615 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4616  *
4617  * Because of the meaning of a subfacet's counters, it only makes sense to do
4618  * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4619  * represents a packet that was sent by hand or if it represents statistics
4620  * that have been cleared out of the datapath. */
4621 static void
4622 subfacet_update_stats(struct subfacet *subfacet,
4623                       const struct dpif_flow_stats *stats)
4624 {
4625     if (stats->n_packets || stats->used > subfacet->used) {
4626         struct facet *facet = subfacet->facet;
4627
4628         subfacet->used = MAX(subfacet->used, stats->used);
4629         facet->used = MAX(facet->used, stats->used);
4630         facet->packet_count += stats->n_packets;
4631         facet->byte_count += stats->n_bytes;
4632         facet->tcp_flags |= stats->tcp_flags;
4633     }
4634 }
4635 \f
4636 /* Rules. */
4637
4638 /* Lookup 'flow' in 'ofproto''s classifier.  If 'wc' is non-null, sets
4639  * the fields that were relevant as part of the lookup. */
4640 void
4641 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4642                  struct flow_wildcards *wc, struct rule_dpif **rule)
4643 {
4644     struct ofport_dpif *port;
4645
4646     if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4647         return;
4648     }
4649     port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4650     if (!port) {
4651         VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4652                      flow->in_port.ofp_port);
4653     }
4654
4655     choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4656                      ofproto->no_packet_in_rule, rule);
4657 }
4658
4659 bool
4660 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4661                           const struct flow *flow, struct flow_wildcards *wc,
4662                           uint8_t table_id, struct rule_dpif **rule)
4663 {
4664     const struct cls_rule *cls_rule;
4665     struct classifier *cls;
4666     bool frag;
4667
4668     *rule = NULL;
4669     if (table_id >= N_TABLES) {
4670         return false;
4671     }
4672
4673     if (wc) {
4674         memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4675         wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4676     }
4677
4678     cls = &ofproto->up.tables[table_id].cls;
4679     ovs_rwlock_rdlock(&cls->rwlock);
4680     frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4681     if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4682         /* We must pretend that transport ports are unavailable. */
4683         struct flow ofpc_normal_flow = *flow;
4684         ofpc_normal_flow.tp_src = htons(0);
4685         ofpc_normal_flow.tp_dst = htons(0);
4686         cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4687     } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4688         cls_rule = &ofproto->drop_frags_rule->up.cr;
4689         /* Frag mask in wc already set above. */
4690     } else {
4691         cls_rule = classifier_lookup(cls, flow, wc);
4692     }
4693
4694     *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4695     rule_dpif_ref(*rule);
4696     ovs_rwlock_unlock(&cls->rwlock);
4697
4698     return *rule != NULL;
4699 }
4700
4701 /* Given a port configuration (specified as zero if there's no port), chooses
4702  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4703  * flow table miss. */
4704 void
4705 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4706                  struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4707 {
4708     *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4709     rule_dpif_ref(*rule);
4710 }
4711
4712 void
4713 rule_dpif_ref(struct rule_dpif *rule)
4714 {
4715     if (rule) {
4716         ofproto_rule_ref(&rule->up);
4717     }
4718 }
4719
4720 void
4721 rule_dpif_unref(struct rule_dpif *rule)
4722 {
4723     if (rule) {
4724         ofproto_rule_unref(&rule->up);
4725     }
4726 }
4727
4728 static void
4729 complete_operation(struct rule_dpif *rule)
4730     OVS_REQUIRES(ofproto_mutex)
4731 {
4732     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4733
4734     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4735     ofoperation_complete(rule->up.pending, 0);
4736 }
4737
4738 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4739 {
4740     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4741 }
4742
4743 static struct rule *
4744 rule_alloc(void)
4745 {
4746     struct rule_dpif *rule = xmalloc(sizeof *rule);
4747     return &rule->up;
4748 }
4749
4750 static void
4751 rule_dealloc(struct rule *rule_)
4752 {
4753     struct rule_dpif *rule = rule_dpif_cast(rule_);
4754     free(rule);
4755 }
4756
4757 static enum ofperr
4758 rule_construct(struct rule *rule_)
4759 {
4760     struct rule_dpif *rule = rule_dpif_cast(rule_);
4761     ovs_mutex_init(&rule->stats_mutex);
4762     ovs_mutex_lock(&rule->stats_mutex);
4763     rule->packet_count = 0;
4764     rule->byte_count = 0;
4765     ovs_mutex_unlock(&rule->stats_mutex);
4766     return 0;
4767 }
4768
4769 static void
4770 rule_insert(struct rule *rule_)
4771     OVS_REQUIRES(ofproto_mutex)
4772 {
4773     struct rule_dpif *rule = rule_dpif_cast(rule_);
4774     complete_operation(rule);
4775 }
4776
4777 static void
4778 rule_delete(struct rule *rule_)
4779     OVS_REQUIRES(ofproto_mutex)
4780 {
4781     struct rule_dpif *rule = rule_dpif_cast(rule_);
4782     complete_operation(rule);
4783 }
4784
4785 static void
4786 rule_destruct(struct rule *rule_)
4787 {
4788     struct rule_dpif *rule = rule_dpif_cast(rule_);
4789     ovs_mutex_destroy(&rule->stats_mutex);
4790 }
4791
4792 static void
4793 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4794 {
4795     struct rule_dpif *rule = rule_dpif_cast(rule_);
4796
4797     /* push_all_stats() can handle flow misses which, when using the learn
4798      * action, can cause rules to be added and deleted.  This can corrupt our
4799      * caller's datastructures which assume that rule_get_stats() doesn't have
4800      * an impact on the flow table. To be safe, we disable miss handling. */
4801     push_all_stats__(false);
4802
4803     /* Start from historical data for 'rule' itself that are no longer tracked
4804      * in facets.  This counts, for example, facets that have expired. */
4805     ovs_mutex_lock(&rule->stats_mutex);
4806     *packets = rule->packet_count;
4807     *bytes = rule->byte_count;
4808     ovs_mutex_unlock(&rule->stats_mutex);
4809 }
4810
4811 static void
4812 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4813                   struct ofpbuf *packet)
4814 {
4815     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4816
4817     ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
4818 }
4819
4820 static enum ofperr
4821 rule_execute(struct rule *rule, const struct flow *flow,
4822              struct ofpbuf *packet)
4823 {
4824     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4825     ofpbuf_delete(packet);
4826     return 0;
4827 }
4828
4829 static void
4830 rule_modify_actions(struct rule *rule_, bool reset_counters)
4831     OVS_REQUIRES(ofproto_mutex)
4832 {
4833     struct rule_dpif *rule = rule_dpif_cast(rule_);
4834
4835     if (reset_counters) {
4836         ovs_mutex_lock(&rule->stats_mutex);
4837         rule->packet_count = 0;
4838         rule->byte_count = 0;
4839         ovs_mutex_unlock(&rule->stats_mutex);
4840     }
4841
4842     complete_operation(rule);
4843 }
4844
4845 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4846 {
4847     return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4848 }
4849
4850 static struct ofgroup *
4851 group_alloc(void)
4852 {
4853     struct group_dpif *group = xzalloc(sizeof *group);
4854     return &group->up;
4855 }
4856
4857 static void
4858 group_dealloc(struct ofgroup *group_)
4859 {
4860     struct group_dpif *group = group_dpif_cast(group_);
4861     free(group);
4862 }
4863
4864 static void
4865 group_construct_stats(struct group_dpif *group)
4866     OVS_REQUIRES(group->stats_mutex)
4867 {
4868     group->packet_count = 0;
4869     group->byte_count = 0;
4870     if (!group->bucket_stats) {
4871         group->bucket_stats = xcalloc(group->up.n_buckets,
4872                                       sizeof *group->bucket_stats);
4873     } else {
4874         memset(group->bucket_stats, 0, group->up.n_buckets *
4875                sizeof *group->bucket_stats);
4876     }
4877 }
4878
4879 static enum ofperr
4880 group_construct(struct ofgroup *group_)
4881 {
4882     struct group_dpif *group = group_dpif_cast(group_);
4883     ovs_mutex_init(&group->stats_mutex);
4884     ovs_mutex_lock(&group->stats_mutex);
4885     group_construct_stats(group);
4886     ovs_mutex_unlock(&group->stats_mutex);
4887     return 0;
4888 }
4889
4890 static void
4891 group_destruct__(struct group_dpif *group)
4892     OVS_REQUIRES(group->stats_mutex)
4893 {
4894     free(group->bucket_stats);
4895     group->bucket_stats = NULL;
4896 }
4897
4898 static void
4899 group_destruct(struct ofgroup *group_)
4900 {
4901     struct group_dpif *group = group_dpif_cast(group_);
4902     ovs_mutex_lock(&group->stats_mutex);
4903     group_destruct__(group);
4904     ovs_mutex_unlock(&group->stats_mutex);
4905     ovs_mutex_destroy(&group->stats_mutex);
4906 }
4907
4908 static enum ofperr
4909 group_modify(struct ofgroup *group_, struct ofgroup *victim_)
4910 {
4911     struct group_dpif *group = group_dpif_cast(group_);
4912     struct group_dpif *victim = group_dpif_cast(victim_);
4913
4914     ovs_mutex_lock(&group->stats_mutex);
4915     if (victim->up.n_buckets < group->up.n_buckets) {
4916         group_destruct__(group);
4917     }
4918     group_construct_stats(group);
4919     ovs_mutex_unlock(&group->stats_mutex);
4920
4921     return 0;
4922 }
4923
4924 static enum ofperr
4925 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4926 {
4927     struct group_dpif *group = group_dpif_cast(group_);
4928
4929     /* Start from historical data for 'group' itself that are no longer tracked
4930      * in facets.  This counts, for example, facets that have expired. */
4931     ovs_mutex_lock(&group->stats_mutex);
4932     ogs->packet_count = group->packet_count;
4933     ogs->byte_count = group->byte_count;
4934     memcpy(ogs->bucket_stats, group->bucket_stats,
4935            group->up.n_buckets * sizeof *group->bucket_stats);
4936     ovs_mutex_unlock(&group->stats_mutex);
4937
4938     return 0;
4939 }
4940
4941 bool
4942 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4943                   struct group_dpif **group)
4944     OVS_TRY_RDLOCK(true, (*group)->up.rwlock)
4945 {
4946     struct ofgroup *ofgroup;
4947     bool found;
4948
4949     *group = NULL;
4950     found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4951     *group = found ?  group_dpif_cast(ofgroup) : NULL;
4952
4953     return found;
4954 }
4955
4956 void
4957 group_dpif_release(struct group_dpif *group)
4958     OVS_RELEASES(group->up.rwlock)
4959 {
4960     ofproto_group_release(&group->up);
4961 }
4962
4963 void
4964 group_dpif_get_buckets(const struct group_dpif *group,
4965                        const struct list **buckets)
4966 {
4967     *buckets = &group->up.buckets;
4968 }
4969
4970 enum ofp11_group_type
4971 group_dpif_get_type(const struct group_dpif *group)
4972 {
4973     return group->up.type;
4974 }
4975 \f
4976 /* Sends 'packet' out 'ofport'.
4977  * May modify 'packet'.
4978  * Returns 0 if successful, otherwise a positive errno value. */
4979 int
4980 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4981 {
4982     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4983     int error;
4984
4985     error = xlate_send_packet(ofport, packet);
4986
4987     ovs_mutex_lock(&ofproto->stats_mutex);
4988     ofproto->stats.tx_packets++;
4989     ofproto->stats.tx_bytes += packet->size;
4990     ovs_mutex_unlock(&ofproto->stats_mutex);
4991     return error;
4992 }
4993
4994 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4995  * The action will state 'slow' as the reason that the action is in the slow
4996  * path.  (This is purely informational: it allows a human viewing "ovs-dpctl
4997  * dump-flows" output to see why a flow is in the slow path.)
4998  *
4999  * The 'stub_size' bytes in 'stub' will be used to store the action.
5000  * 'stub_size' must be large enough for the action.
5001  *
5002  * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5003  * respectively. */
5004 static void
5005 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5006                   enum slow_path_reason slow,
5007                   uint64_t *stub, size_t stub_size,
5008                   const struct nlattr **actionsp, size_t *actions_lenp)
5009 {
5010     union user_action_cookie cookie;
5011     struct ofpbuf buf;
5012
5013     cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5014     cookie.slow_path.unused = 0;
5015     cookie.slow_path.reason = slow;
5016
5017     ofpbuf_use_stack(&buf, stub, stub_size);
5018     if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5019         uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5020                                          ODPP_NONE);
5021         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5022     } else {
5023         odp_port_t odp_port;
5024         uint32_t pid;
5025
5026         odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5027         pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5028         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5029     }
5030     *actionsp = buf.data;
5031     *actions_lenp = buf.size;
5032 }
5033 \f
5034 static bool
5035 set_frag_handling(struct ofproto *ofproto_,
5036                   enum ofp_config_flags frag_handling)
5037 {
5038     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5039     if (frag_handling != OFPC_FRAG_REASM) {
5040         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5041         return true;
5042     } else {
5043         return false;
5044     }
5045 }
5046
5047 static enum ofperr
5048 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5049            const struct flow *flow,
5050            const struct ofpact *ofpacts, size_t ofpacts_len)
5051 {
5052     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5053
5054     ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
5055                                  ofpacts_len, packet);
5056     return 0;
5057 }
5058 \f
5059 /* NetFlow. */
5060
5061 static int
5062 set_netflow(struct ofproto *ofproto_,
5063             const struct netflow_options *netflow_options)
5064 {
5065     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5066
5067     if (netflow_options) {
5068         if (!ofproto->netflow) {
5069             ofproto->netflow = netflow_create();
5070             ofproto->backer->need_revalidate = REV_RECONFIGURE;
5071         }
5072         return netflow_set_options(ofproto->netflow, netflow_options);
5073     } else if (ofproto->netflow) {
5074         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5075         netflow_destroy(ofproto->netflow);
5076         ofproto->netflow = NULL;
5077     }
5078
5079     return 0;
5080 }
5081
5082 static void
5083 get_netflow_ids(const struct ofproto *ofproto_,
5084                 uint8_t *engine_type, uint8_t *engine_id)
5085 {
5086     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5087
5088     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5089 }
5090
5091 static void
5092 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5093 {
5094     if (!facet_is_controller_flow(facet) &&
5095         netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5096         struct subfacet *subfacet;
5097         struct ofexpired expired;
5098
5099         LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5100             if (subfacet->path == SF_FAST_PATH) {
5101                 struct dpif_flow_stats stats;
5102
5103                 subfacet_install(subfacet, &facet->xout.odp_actions,
5104                                  &stats);
5105                 subfacet_update_stats(subfacet, &stats);
5106             }
5107         }
5108
5109         expired.flow = facet->flow;
5110         expired.packet_count = facet->packet_count;
5111         expired.byte_count = facet->byte_count;
5112         expired.used = facet->used;
5113         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5114     }
5115 }
5116
5117 static void
5118 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5119 {
5120     struct cls_cursor cursor;
5121     struct facet *facet;
5122
5123     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5124     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5125     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5126         send_active_timeout(ofproto, facet);
5127     }
5128     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5129 }
5130 \f
5131 static struct ofproto_dpif *
5132 ofproto_dpif_lookup(const char *name)
5133 {
5134     struct ofproto_dpif *ofproto;
5135
5136     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5137                              hash_string(name, 0), &all_ofproto_dpifs) {
5138         if (!strcmp(ofproto->up.name, name)) {
5139             return ofproto;
5140         }
5141     }
5142     return NULL;
5143 }
5144
5145 static void
5146 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5147                           const char *argv[], void *aux OVS_UNUSED)
5148 {
5149     struct ofproto_dpif *ofproto;
5150
5151     if (argc > 1) {
5152         ofproto = ofproto_dpif_lookup(argv[1]);
5153         if (!ofproto) {
5154             unixctl_command_reply_error(conn, "no such bridge");
5155             return;
5156         }
5157         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5158         mac_learning_flush(ofproto->ml);
5159         ovs_rwlock_unlock(&ofproto->ml->rwlock);
5160     } else {
5161         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5162             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5163             mac_learning_flush(ofproto->ml);
5164             ovs_rwlock_unlock(&ofproto->ml->rwlock);
5165         }
5166     }
5167
5168     unixctl_command_reply(conn, "table successfully flushed");
5169 }
5170
5171 static struct ofport_dpif *
5172 ofbundle_get_a_port(const struct ofbundle *bundle)
5173 {
5174     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5175                         bundle_node);
5176 }
5177
5178 static void
5179 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5180                          const char *argv[], void *aux OVS_UNUSED)
5181 {
5182     struct ds ds = DS_EMPTY_INITIALIZER;
5183     const struct ofproto_dpif *ofproto;
5184     const struct mac_entry *e;
5185
5186     ofproto = ofproto_dpif_lookup(argv[1]);
5187     if (!ofproto) {
5188         unixctl_command_reply_error(conn, "no such bridge");
5189         return;
5190     }
5191
5192     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
5193     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5194     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5195         struct ofbundle *bundle = e->port.p;
5196         char name[OFP_MAX_PORT_NAME_LEN];
5197
5198         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5199                                name, sizeof name);
5200         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
5201                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
5202                       mac_entry_age(ofproto->ml, e));
5203     }
5204     ovs_rwlock_unlock(&ofproto->ml->rwlock);
5205     unixctl_command_reply(conn, ds_cstr(&ds));
5206     ds_destroy(&ds);
5207 }
5208
5209 struct trace_ctx {
5210     struct xlate_out xout;
5211     struct xlate_in xin;
5212     struct flow flow;
5213     struct ds *result;
5214 };
5215
5216 static void
5217 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5218 {
5219     struct rule_actions *actions;
5220     ovs_be64 cookie;
5221
5222     ds_put_char_multiple(result, '\t', level);
5223     if (!rule) {
5224         ds_put_cstr(result, "No match\n");
5225         return;
5226     }
5227
5228     ovs_mutex_lock(&rule->up.mutex);
5229     cookie = rule->up.flow_cookie;
5230     ovs_mutex_unlock(&rule->up.mutex);
5231
5232     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5233                   rule ? rule->up.table_id : 0, ntohll(cookie));
5234     cls_rule_format(&rule->up.cr, result);
5235     ds_put_char(result, '\n');
5236
5237     actions = rule_dpif_get_actions(rule);
5238
5239     ds_put_char_multiple(result, '\t', level);
5240     ds_put_cstr(result, "OpenFlow actions=");
5241     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
5242     ds_put_char(result, '\n');
5243
5244     rule_actions_unref(actions);
5245 }
5246
5247 static void
5248 trace_format_flow(struct ds *result, int level, const char *title,
5249                   struct trace_ctx *trace)
5250 {
5251     ds_put_char_multiple(result, '\t', level);
5252     ds_put_format(result, "%s: ", title);
5253     if (flow_equal(&trace->xin.flow, &trace->flow)) {
5254         ds_put_cstr(result, "unchanged");
5255     } else {
5256         flow_format(result, &trace->xin.flow);
5257         trace->flow = trace->xin.flow;
5258     }
5259     ds_put_char(result, '\n');
5260 }
5261
5262 static void
5263 trace_format_regs(struct ds *result, int level, const char *title,
5264                   struct trace_ctx *trace)
5265 {
5266     size_t i;
5267
5268     ds_put_char_multiple(result, '\t', level);
5269     ds_put_format(result, "%s:", title);
5270     for (i = 0; i < FLOW_N_REGS; i++) {
5271         ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
5272     }
5273     ds_put_char(result, '\n');
5274 }
5275
5276 static void
5277 trace_format_odp(struct ds *result, int level, const char *title,
5278                  struct trace_ctx *trace)
5279 {
5280     struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5281
5282     ds_put_char_multiple(result, '\t', level);
5283     ds_put_format(result, "%s: ", title);
5284     format_odp_actions(result, odp_actions->data, odp_actions->size);
5285     ds_put_char(result, '\n');
5286 }
5287
5288 static void
5289 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5290 {
5291     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5292     struct ds *result = trace->result;
5293
5294     ds_put_char(result, '\n');
5295     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5296     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5297     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
5298     trace_format_rule(result, recurse + 1, rule);
5299 }
5300
5301 static void
5302 trace_report(struct xlate_in *xin, const char *s, int recurse)
5303 {
5304     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5305     struct ds *result = trace->result;
5306
5307     ds_put_char_multiple(result, '\t', recurse);
5308     ds_put_cstr(result, s);
5309     ds_put_char(result, '\n');
5310 }
5311
5312 /* Parses the 'argc' elements of 'argv', ignoring argv[0].  The following
5313  * forms are supported:
5314  *
5315  *     - [dpname] odp_flow [-generate | packet]
5316  *     - bridge br_flow [-generate | packet]
5317  *
5318  * On success, initializes '*ofprotop' and 'flow' and returns NULL.  On failure
5319  * returns a nonnull error message. */
5320 static const char *
5321 parse_flow_and_packet(int argc, const char *argv[],
5322                       struct ofproto_dpif **ofprotop, struct flow *flow,
5323                       struct ofpbuf **packetp)
5324 {
5325     const struct dpif_backer *backer = NULL;
5326     const char *error = NULL;
5327     struct simap port_names = SIMAP_INITIALIZER(&port_names);
5328     struct ofpbuf *packet;
5329     struct ofpbuf odp_key;
5330     struct ofpbuf odp_mask;
5331
5332     ofpbuf_init(&odp_key, 0);
5333     ofpbuf_init(&odp_mask, 0);
5334
5335     /* Handle "-generate" or a hex string as the last argument. */
5336     if (!strcmp(argv[argc - 1], "-generate")) {
5337         packet = ofpbuf_new(0);
5338         argc--;
5339     } else {
5340         error = eth_from_hex(argv[argc - 1], &packet);
5341         if (!error) {
5342             argc--;
5343         } else if (argc == 4) {
5344             /* The 3-argument form must end in "-generate' or a hex string. */
5345             goto exit;
5346         }
5347     }
5348
5349     /* odp_flow can have its in_port specified as a name instead of port no.
5350      * We do not yet know whether a given flow is a odp_flow or a br_flow.
5351      * But, to know whether a flow is odp_flow through odp_flow_from_string(),
5352      * we need to create a simap of name to port no. */
5353     if (argc == 3) {
5354         const char *dp_type;
5355         if (!strncmp(argv[1], "ovs-", 4)) {
5356             dp_type = argv[1] + 4;
5357         } else {
5358             dp_type = argv[1];
5359         }
5360         backer = shash_find_data(&all_dpif_backers, dp_type);
5361     } else if (argc == 2) {
5362         struct shash_node *node;
5363         if (shash_count(&all_dpif_backers) == 1) {
5364             node = shash_first(&all_dpif_backers);
5365             backer = node->data;
5366         }
5367     } else {
5368         error = "Syntax error";
5369         goto exit;
5370     }
5371     if (backer && backer->dpif) {
5372         struct dpif_port dpif_port;
5373         struct dpif_port_dump port_dump;
5374         DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
5375             simap_put(&port_names, dpif_port.name,
5376                       odp_to_u32(dpif_port.port_no));
5377         }
5378     }
5379
5380     /* Parse the flow and determine whether a datapath or
5381      * bridge is specified. If function odp_flow_key_from_string()
5382      * returns 0, the flow is a odp_flow. If function
5383      * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5384     if (!odp_flow_from_string(argv[argc - 1], &port_names,
5385                               &odp_key, &odp_mask)) {
5386         if (!backer) {
5387             error = "Cannot find the datapath";
5388             goto exit;
5389         }
5390
5391         if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, flow,
5392                           NULL, ofprotop, NULL)) {
5393             error = "Invalid datapath flow";
5394             goto exit;
5395         }
5396     } else if (!parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL)) {
5397         if (argc != 3) {
5398             error = "Must specify bridge name";
5399             goto exit;
5400         }
5401
5402         *ofprotop = ofproto_dpif_lookup(argv[1]);
5403         if (!*ofprotop) {
5404             error = "Unknown bridge name";
5405             goto exit;
5406         }
5407     } else {
5408         error = "Bad flow syntax";
5409         goto exit;
5410     }
5411
5412     /* Generate a packet, if requested. */
5413     if (packet) {
5414         if (!packet->size) {
5415             flow_compose(packet, flow);
5416         } else {
5417             union flow_in_port in_port = flow->in_port;
5418
5419             /* Use the metadata from the flow and the packet argument
5420              * to reconstruct the flow. */
5421             flow_extract(packet, flow->skb_priority, flow->pkt_mark, NULL,
5422                          &in_port, flow);
5423         }
5424     }
5425
5426     error = NULL;
5427
5428 exit:
5429     if (error) {
5430         ofpbuf_delete(packet);
5431         packet = NULL;
5432     }
5433     *packetp = packet;
5434     ofpbuf_uninit(&odp_key);
5435     ofpbuf_uninit(&odp_mask);
5436     simap_destroy(&port_names);
5437     return error;
5438 }
5439
5440 static void
5441 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5442                       void *aux OVS_UNUSED)
5443 {
5444     struct ofproto_dpif *ofproto;
5445     struct ofpbuf *packet;
5446     const char *error;
5447     struct flow flow;
5448
5449     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
5450     if (!error) {
5451         struct ds result;
5452
5453         ds_init(&result);
5454         ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
5455         unixctl_command_reply(conn, ds_cstr(&result));
5456         ds_destroy(&result);
5457         ofpbuf_delete(packet);
5458     } else {
5459         unixctl_command_reply_error(conn, error);
5460     }
5461 }
5462
5463 static void
5464 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
5465                               const char *argv[], void *aux OVS_UNUSED)
5466 {
5467     enum ofputil_protocol usable_protocols;
5468     struct ofproto_dpif *ofproto;
5469     bool enforce_consistency;
5470     struct ofpbuf ofpacts;
5471     struct ofpbuf *packet;
5472     struct ds result;
5473     struct flow flow;
5474     uint16_t in_port;
5475
5476     /* Three kinds of error return values! */
5477     enum ofperr retval;
5478     const char *error;
5479     char *rw_error;
5480
5481     packet = NULL;
5482     ds_init(&result);
5483     ofpbuf_init(&ofpacts, 0);
5484
5485     /* Parse actions. */
5486     rw_error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols);
5487     if (rw_error) {
5488         unixctl_command_reply_error(conn, rw_error);
5489         free(rw_error);
5490         goto exit;
5491     }
5492
5493     /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
5494      * consistency between the flow and the actions.  With -consistent, we
5495      * enforce consistency even for a flow supported in OpenFlow 1.0. */
5496     if (!strcmp(argv[1], "-consistent")) {
5497         enforce_consistency = true;
5498         argv++;
5499         argc--;
5500     } else {
5501         enforce_consistency = false;
5502     }
5503
5504     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
5505     if (error) {
5506         unixctl_command_reply_error(conn, error);
5507         goto exit;
5508     }
5509
5510     /* Do the same checks as handle_packet_out() in ofproto.c.
5511      *
5512      * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
5513      * strictly correct because these actions aren't in any table, but it's OK
5514      * because it 'table_id' is used only to check goto_table instructions, but
5515      * packet-outs take a list of actions and therefore it can't include
5516      * instructions.
5517      *
5518      * We skip the "meter" check here because meter is an instruction, not an
5519      * action, and thus cannot appear in ofpacts. */
5520     in_port = ofp_to_u16(flow.in_port.ofp_port);
5521     if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
5522         unixctl_command_reply_error(conn, "invalid in_port");
5523         goto exit;
5524     }
5525     if (enforce_consistency) {
5526         retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size, &flow,
5527                                            u16_to_ofp(ofproto->up.max_ports),
5528                                            0, 0, usable_protocols);
5529     } else {
5530         retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow,
5531                                u16_to_ofp(ofproto->up.max_ports), 0, 0,
5532                                &usable_protocols);
5533     }
5534
5535     if (retval) {
5536         ds_clear(&result);
5537         ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
5538         unixctl_command_reply_error(conn, ds_cstr(&result));
5539         goto exit;
5540     }
5541
5542     ofproto_trace(ofproto, &flow, packet, ofpacts.data, ofpacts.size, &result);
5543     unixctl_command_reply(conn, ds_cstr(&result));
5544
5545 exit:
5546     ds_destroy(&result);
5547     ofpbuf_delete(packet);
5548     ofpbuf_uninit(&ofpacts);
5549 }
5550
5551 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
5552  * description of the results to 'ds'.
5553  *
5554  * The trace follows a packet with the specified 'flow' through the flow
5555  * table.  'packet' may be nonnull to trace an actual packet, with consequent
5556  * side effects (if it is nonnull then its flow must be 'flow').
5557  *
5558  * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
5559  * trace, otherwise the actions are determined by a flow table lookup. */
5560 static void
5561 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5562               const struct ofpbuf *packet,
5563               const struct ofpact ofpacts[], size_t ofpacts_len,
5564               struct ds *ds)
5565 {
5566     struct rule_dpif *rule;
5567     struct flow_wildcards wc;
5568
5569     ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
5570     ds_put_cstr(ds, "Flow: ");
5571     flow_format(ds, flow);
5572     ds_put_char(ds, '\n');
5573
5574     flow_wildcards_init_catchall(&wc);
5575     if (ofpacts) {
5576         rule = NULL;
5577     } else {
5578         rule_dpif_lookup(ofproto, flow, &wc, &rule);
5579
5580         trace_format_rule(ds, 0, rule);
5581         if (rule == ofproto->miss_rule) {
5582             ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5583         } else if (rule == ofproto->no_packet_in_rule) {
5584             ds_put_cstr(ds, "\nNo match, packets dropped because "
5585                         "OFPPC_NO_PACKET_IN is set on in_port.\n");
5586         } else if (rule == ofproto->drop_frags_rule) {
5587             ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5588                         "and the fragment handling mode is \"drop\".\n");
5589         }
5590     }
5591
5592     if (rule || ofpacts) {
5593         uint64_t odp_actions_stub[1024 / 8];
5594         struct ofpbuf odp_actions;
5595         struct trace_ctx trace;
5596         struct match match;
5597         uint16_t tcp_flags;
5598
5599         tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5600         trace.result = ds;
5601         trace.flow = *flow;
5602         ofpbuf_use_stub(&odp_actions,
5603                         odp_actions_stub, sizeof odp_actions_stub);
5604         xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5605         if (ofpacts) {
5606             trace.xin.ofpacts = ofpacts;
5607             trace.xin.ofpacts_len = ofpacts_len;
5608         }
5609         trace.xin.resubmit_hook = trace_resubmit;
5610         trace.xin.report_hook = trace_report;
5611
5612         xlate_actions(&trace.xin, &trace.xout);
5613         flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5614
5615         ds_put_char(ds, '\n');
5616         trace_format_flow(ds, 0, "Final flow", &trace);
5617
5618         match_init(&match, flow, &trace.xout.wc);
5619         ds_put_cstr(ds, "Relevant fields: ");
5620         match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5621         ds_put_char(ds, '\n');
5622
5623         ds_put_cstr(ds, "Datapath actions: ");
5624         format_odp_actions(ds, trace.xout.odp_actions.data,
5625                            trace.xout.odp_actions.size);
5626
5627         if (trace.xout.slow) {
5628             enum slow_path_reason slow;
5629
5630             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5631                         "slow path because it:");
5632
5633             slow = trace.xout.slow;
5634             while (slow) {
5635                 enum slow_path_reason bit = rightmost_1bit(slow);
5636
5637                 ds_put_format(ds, "\n\t- %s.",
5638                               slow_path_reason_to_explanation(bit));
5639
5640                 slow &= ~bit;
5641             }
5642         }
5643
5644         xlate_out_uninit(&trace.xout);
5645     }
5646
5647     rule_dpif_unref(rule);
5648 }
5649
5650 /* Runs a self-check of flow translations in 'ofproto'.  Appends a message to
5651  * 'reply' describing the results. */
5652 static void
5653 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5654 {
5655     struct cls_cursor cursor;
5656     struct facet *facet;
5657     int errors;
5658
5659     errors = 0;
5660     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5661     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5662     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5663         if (!facet_check_consistency(facet)) {
5664             errors++;
5665         }
5666     }
5667     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5668     if (errors) {
5669         ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5670     }
5671
5672     if (errors) {
5673         ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5674                       ofproto->up.name, errors);
5675     } else {
5676         ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5677     }
5678 }
5679
5680 static void
5681 ofproto_dpif_self_check(struct unixctl_conn *conn,
5682                         int argc, const char *argv[], void *aux OVS_UNUSED)
5683 {
5684     struct ds reply = DS_EMPTY_INITIALIZER;
5685     struct ofproto_dpif *ofproto;
5686
5687     if (argc > 1) {
5688         ofproto = ofproto_dpif_lookup(argv[1]);
5689         if (!ofproto) {
5690             unixctl_command_reply_error(conn, "Unknown ofproto (use "
5691                                         "ofproto/list for help)");
5692             return;
5693         }
5694         ofproto_dpif_self_check__(ofproto, &reply);
5695     } else {
5696         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5697             ofproto_dpif_self_check__(ofproto, &reply);
5698         }
5699     }
5700
5701     unixctl_command_reply(conn, ds_cstr(&reply));
5702     ds_destroy(&reply);
5703 }
5704
5705 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
5706  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
5707  * to destroy 'ofproto_shash' and free the returned value. */
5708 static const struct shash_node **
5709 get_ofprotos(struct shash *ofproto_shash)
5710 {
5711     const struct ofproto_dpif *ofproto;
5712
5713     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5714         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5715         shash_add_nocopy(ofproto_shash, name, ofproto);
5716     }
5717
5718     return shash_sort(ofproto_shash);
5719 }
5720
5721 static void
5722 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5723                               const char *argv[] OVS_UNUSED,
5724                               void *aux OVS_UNUSED)
5725 {
5726     struct ds ds = DS_EMPTY_INITIALIZER;
5727     struct shash ofproto_shash;
5728     const struct shash_node **sorted_ofprotos;
5729     int i;
5730
5731     shash_init(&ofproto_shash);
5732     sorted_ofprotos = get_ofprotos(&ofproto_shash);
5733     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5734         const struct shash_node *node = sorted_ofprotos[i];
5735         ds_put_format(&ds, "%s\n", node->name);
5736     }
5737
5738     shash_destroy(&ofproto_shash);
5739     free(sorted_ofprotos);
5740
5741     unixctl_command_reply(conn, ds_cstr(&ds));
5742     ds_destroy(&ds);
5743 }
5744
5745 static void
5746 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5747 {
5748     const struct shash_node **ofprotos;
5749     struct ofproto_dpif *ofproto;
5750     struct shash ofproto_shash;
5751     uint64_t n_hit, n_missed;
5752     size_t i;
5753
5754     n_hit = n_missed = 0;
5755     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5756         if (ofproto->backer == backer) {
5757             n_missed += ofproto->n_missed;
5758             n_hit += ofproto->n_hit;
5759         }
5760     }
5761
5762     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5763                   dpif_name(backer->dpif), n_hit, n_missed);
5764     ds_put_format(ds, "\tflows: cur: %"PRIuSIZE", avg: %u, max: %u,"
5765                   " life span: %lldms\n", hmap_count(&backer->subfacets),
5766                   backer->avg_n_subfacet, backer->max_n_subfacet,
5767                   backer->avg_subfacet_life);
5768
5769     shash_init(&ofproto_shash);
5770     ofprotos = get_ofprotos(&ofproto_shash);
5771     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5772         struct ofproto_dpif *ofproto = ofprotos[i]->data;
5773         const struct shash_node **ports;
5774         size_t j;
5775
5776         if (ofproto->backer != backer) {
5777             continue;
5778         }
5779
5780         ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5781                       ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5782
5783         ports = shash_sort(&ofproto->up.port_by_name);
5784         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5785             const struct shash_node *node = ports[j];
5786             struct ofport *ofport = node->data;
5787             struct smap config;
5788             odp_port_t odp_port;
5789
5790             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5791                           ofport->ofp_port);
5792
5793             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5794             if (odp_port != ODPP_NONE) {
5795                 ds_put_format(ds, "%"PRIu32":", odp_port);
5796             } else {
5797                 ds_put_cstr(ds, "none:");
5798             }
5799
5800             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5801
5802             smap_init(&config);
5803             if (!netdev_get_config(ofport->netdev, &config)) {
5804                 const struct smap_node **nodes;
5805                 size_t i;
5806
5807                 nodes = smap_sort(&config);
5808                 for (i = 0; i < smap_count(&config); i++) {
5809                     const struct smap_node *node = nodes[i];
5810                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5811                                   node->key, node->value);
5812                 }
5813                 free(nodes);
5814             }
5815             smap_destroy(&config);
5816
5817             ds_put_char(ds, ')');
5818             ds_put_char(ds, '\n');
5819         }
5820         free(ports);
5821     }
5822     shash_destroy(&ofproto_shash);
5823     free(ofprotos);
5824 }
5825
5826 static void
5827 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5828                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5829 {
5830     struct ds ds = DS_EMPTY_INITIALIZER;
5831     const struct shash_node **backers;
5832     int i;
5833
5834     backers = shash_sort(&all_dpif_backers);
5835     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5836         dpif_show_backer(backers[i]->data, &ds);
5837     }
5838     free(backers);
5839
5840     unixctl_command_reply(conn, ds_cstr(&ds));
5841     ds_destroy(&ds);
5842 }
5843
5844 /* Dump the megaflow (facet) cache.  This is useful to check the
5845  * correctness of flow wildcarding, since the same mechanism is used for
5846  * both xlate caching and kernel wildcarding.
5847  *
5848  * It's important to note that in the output the flow description uses
5849  * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5850  *
5851  * This command is only needed for advanced debugging, so it's not
5852  * documented in the man page. */
5853 static void
5854 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5855                                     int argc OVS_UNUSED, const char *argv[],
5856                                     void *aux OVS_UNUSED)
5857 {
5858     struct ds ds = DS_EMPTY_INITIALIZER;
5859     const struct ofproto_dpif *ofproto;
5860     long long int now = time_msec();
5861     struct cls_cursor cursor;
5862     struct facet *facet;
5863
5864     ofproto = ofproto_dpif_lookup(argv[1]);
5865     if (!ofproto) {
5866         unixctl_command_reply_error(conn, "no such bridge");
5867         return;
5868     }
5869
5870     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5871     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5872     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5873         cls_rule_format(&facet->cr, &ds);
5874         ds_put_cstr(&ds, ", ");
5875         ds_put_format(&ds, "n_subfacets:%"PRIuSIZE", ", list_size(&facet->subfacets));
5876         ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5877         ds_put_cstr(&ds, "Datapath actions: ");
5878         if (facet->xout.slow) {
5879             uint64_t slow_path_stub[128 / 8];
5880             const struct nlattr *actions;
5881             size_t actions_len;
5882
5883             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5884                               slow_path_stub, sizeof slow_path_stub,
5885                               &actions, &actions_len);
5886             format_odp_actions(&ds, actions, actions_len);
5887         } else {
5888             format_odp_actions(&ds, facet->xout.odp_actions.data,
5889                                facet->xout.odp_actions.size);
5890         }
5891         ds_put_cstr(&ds, "\n");
5892     }
5893     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5894
5895     ds_chomp(&ds, '\n');
5896     unixctl_command_reply(conn, ds_cstr(&ds));
5897     ds_destroy(&ds);
5898 }
5899
5900 /* Disable using the megaflows.
5901  *
5902  * This command is only needed for advanced debugging, so it's not
5903  * documented in the man page. */
5904 static void
5905 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5906                                        int argc OVS_UNUSED,
5907                                        const char *argv[] OVS_UNUSED,
5908                                        void *aux OVS_UNUSED)
5909 {
5910     struct ofproto_dpif *ofproto;
5911
5912     enable_megaflows = false;
5913
5914     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5915         flush(&ofproto->up);
5916     }
5917
5918     unixctl_command_reply(conn, "megaflows disabled");
5919 }
5920
5921 /* Re-enable using megaflows.
5922  *
5923  * This command is only needed for advanced debugging, so it's not
5924  * documented in the man page. */
5925 static void
5926 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5927                                       int argc OVS_UNUSED,
5928                                       const char *argv[] OVS_UNUSED,
5929                                       void *aux OVS_UNUSED)
5930 {
5931     struct ofproto_dpif *ofproto;
5932
5933     enable_megaflows = true;
5934
5935     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5936         flush(&ofproto->up);
5937     }
5938
5939     unixctl_command_reply(conn, "megaflows enabled");
5940 }
5941
5942 static void
5943 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5944                                 int argc OVS_UNUSED, const char *argv[],
5945                                 void *aux OVS_UNUSED)
5946 {
5947     struct ds ds = DS_EMPTY_INITIALIZER;
5948     const struct ofproto_dpif *ofproto;
5949     struct subfacet *subfacet;
5950
5951     ofproto = ofproto_dpif_lookup(argv[1]);
5952     if (!ofproto) {
5953         unixctl_command_reply_error(conn, "no such bridge");
5954         return;
5955     }
5956
5957     update_stats(ofproto->backer);
5958
5959     HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5960         struct facet *facet = subfacet->facet;
5961         struct odputil_keybuf maskbuf;
5962         struct ofpbuf mask;
5963
5964         if (facet->ofproto != ofproto) {
5965             continue;
5966         }
5967
5968         ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5969         if (enable_megaflows) {
5970             odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5971                                    &facet->flow, UINT32_MAX);
5972         }
5973
5974         odp_flow_format(subfacet->key, subfacet->key_len,
5975                         mask.data, mask.size, NULL, &ds, false);
5976
5977         ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5978                       subfacet->dp_packet_count, subfacet->dp_byte_count);
5979         if (subfacet->used) {
5980             ds_put_format(&ds, "%.3fs",
5981                           (time_msec() - subfacet->used) / 1000.0);
5982         } else {
5983             ds_put_format(&ds, "never");
5984         }
5985         if (subfacet->facet->tcp_flags) {
5986             ds_put_cstr(&ds, ", flags:");
5987             packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5988         }
5989
5990         ds_put_cstr(&ds, ", actions:");
5991         if (facet->xout.slow) {
5992             uint64_t slow_path_stub[128 / 8];
5993             const struct nlattr *actions;
5994             size_t actions_len;
5995
5996             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5997                               slow_path_stub, sizeof slow_path_stub,
5998                               &actions, &actions_len);
5999             format_odp_actions(&ds, actions, actions_len);
6000         } else {
6001             format_odp_actions(&ds, facet->xout.odp_actions.data,
6002                                facet->xout.odp_actions.size);
6003         }
6004         ds_put_char(&ds, '\n');
6005     }
6006
6007     unixctl_command_reply(conn, ds_cstr(&ds));
6008     ds_destroy(&ds);
6009 }
6010
6011 static void
6012 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6013                                int argc OVS_UNUSED, const char *argv[],
6014                                void *aux OVS_UNUSED)
6015 {
6016     struct ds ds = DS_EMPTY_INITIALIZER;
6017     struct ofproto_dpif *ofproto;
6018
6019     ofproto = ofproto_dpif_lookup(argv[1]);
6020     if (!ofproto) {
6021         unixctl_command_reply_error(conn, "no such bridge");
6022         return;
6023     }
6024
6025     flush(&ofproto->up);
6026
6027     unixctl_command_reply(conn, ds_cstr(&ds));
6028     ds_destroy(&ds);
6029 }
6030
6031 static void
6032 ofproto_dpif_unixctl_init(void)
6033 {
6034     static bool registered;
6035     if (registered) {
6036         return;
6037     }
6038     registered = true;
6039
6040     unixctl_command_register(
6041         "ofproto/trace",
6042         "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
6043         1, 3, ofproto_unixctl_trace, NULL);
6044     unixctl_command_register(
6045         "ofproto/trace-packet-out",
6046         "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
6047         2, 6, ofproto_unixctl_trace_actions, NULL);
6048     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6049                              ofproto_unixctl_fdb_flush, NULL);
6050     unixctl_command_register("fdb/show", "bridge", 1, 1,
6051                              ofproto_unixctl_fdb_show, NULL);
6052     unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6053                              ofproto_dpif_self_check, NULL);
6054     unixctl_command_register("dpif/dump-dps", "", 0, 0,
6055                              ofproto_unixctl_dpif_dump_dps, NULL);
6056     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6057                              NULL);
6058     unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6059                              ofproto_unixctl_dpif_dump_flows, NULL);
6060     unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6061                              ofproto_unixctl_dpif_del_flows, NULL);
6062     unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6063                              ofproto_unixctl_dpif_dump_megaflows, NULL);
6064     unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6065                              ofproto_unixctl_dpif_disable_megaflows, NULL);
6066     unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6067                              ofproto_unixctl_dpif_enable_megaflows, NULL);
6068 }
6069 \f
6070 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6071  *
6072  * This is deprecated.  It is only for compatibility with broken device drivers
6073  * in old versions of Linux that do not properly support VLANs when VLAN
6074  * devices are not used.  When broken device drivers are no longer in
6075  * widespread use, we will delete these interfaces. */
6076
6077 static int
6078 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6079 {
6080     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6081     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6082
6083     if (realdev_ofp_port == ofport->realdev_ofp_port
6084         && vid == ofport->vlandev_vid) {
6085         return 0;
6086     }
6087
6088     ofproto->backer->need_revalidate = REV_RECONFIGURE;
6089
6090     if (ofport->realdev_ofp_port) {
6091         vsp_remove(ofport);
6092     }
6093     if (realdev_ofp_port && ofport->bundle) {
6094         /* vlandevs are enslaved to their realdevs, so they are not allowed to
6095          * themselves be part of a bundle. */
6096         bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6097     }
6098
6099     ofport->realdev_ofp_port = realdev_ofp_port;
6100     ofport->vlandev_vid = vid;
6101
6102     if (realdev_ofp_port) {
6103         vsp_add(ofport, realdev_ofp_port, vid);
6104     }
6105
6106     return 0;
6107 }
6108
6109 static uint32_t
6110 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6111 {
6112     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6113 }
6114
6115 bool
6116 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6117     OVS_EXCLUDED(ofproto->vsp_mutex)
6118 {
6119     bool ret;
6120
6121     ovs_mutex_lock(&ofproto->vsp_mutex);
6122     ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6123     ovs_mutex_unlock(&ofproto->vsp_mutex);
6124     return ret;
6125 }
6126
6127 static ofp_port_t
6128 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6129                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6130     OVS_REQUIRES(ofproto->vsp_mutex)
6131 {
6132     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6133         int vid = vlan_tci_to_vid(vlan_tci);
6134         const struct vlan_splinter *vsp;
6135
6136         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6137                                  hash_realdev_vid(realdev_ofp_port, vid),
6138                                  &ofproto->realdev_vid_map) {
6139             if (vsp->realdev_ofp_port == realdev_ofp_port
6140                 && vsp->vid == vid) {
6141                 return vsp->vlandev_ofp_port;
6142             }
6143         }
6144     }
6145     return realdev_ofp_port;
6146 }
6147
6148 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6149  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6150  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
6151  * 'vlan_tci' 9, it would return the port number of eth0.9.
6152  *
6153  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6154  * function just returns its 'realdev_ofp_port' argument. */
6155 ofp_port_t
6156 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6157                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6158     OVS_EXCLUDED(ofproto->vsp_mutex)
6159 {
6160     ofp_port_t ret;
6161
6162     ovs_mutex_lock(&ofproto->vsp_mutex);
6163     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6164     ovs_mutex_unlock(&ofproto->vsp_mutex);
6165     return ret;
6166 }
6167
6168 static struct vlan_splinter *
6169 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6170 {
6171     struct vlan_splinter *vsp;
6172
6173     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6174                              hash_ofp_port(vlandev_ofp_port),
6175                              &ofproto->vlandev_map) {
6176         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6177             return vsp;
6178         }
6179     }
6180
6181     return NULL;
6182 }
6183
6184 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6185  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6186  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
6187  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6188  * eth0 and store 9 in '*vid'.
6189  *
6190  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6191  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
6192  * always does.*/
6193 static ofp_port_t
6194 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6195                        ofp_port_t vlandev_ofp_port, int *vid)
6196     OVS_REQUIRES(ofproto->vsp_mutex)
6197 {
6198     if (!hmap_is_empty(&ofproto->vlandev_map)) {
6199         const struct vlan_splinter *vsp;
6200
6201         vsp = vlandev_find(ofproto, vlandev_ofp_port);
6202         if (vsp) {
6203             if (vid) {
6204                 *vid = vsp->vid;
6205             }
6206             return vsp->realdev_ofp_port;
6207         }
6208     }
6209     return 0;
6210 }
6211
6212 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6213  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
6214  * 'flow->in_port' to the "real" device backing the VLAN device, sets
6215  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Otherwise (which is
6216  * always the case unless VLAN splinters are enabled), returns false without
6217  * making any changes. */
6218 bool
6219 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6220     OVS_EXCLUDED(ofproto->vsp_mutex)
6221 {
6222     ofp_port_t realdev;
6223     int vid;
6224
6225     ovs_mutex_lock(&ofproto->vsp_mutex);
6226     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6227     ovs_mutex_unlock(&ofproto->vsp_mutex);
6228     if (!realdev) {
6229         return false;
6230     }
6231
6232     /* Cause the flow to be processed as if it came in on the real device with
6233      * the VLAN device's VLAN ID. */
6234     flow->in_port.ofp_port = realdev;
6235     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6236     return true;
6237 }
6238
6239 static void
6240 vsp_remove(struct ofport_dpif *port)
6241 {
6242     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6243     struct vlan_splinter *vsp;
6244
6245     ovs_mutex_lock(&ofproto->vsp_mutex);
6246     vsp = vlandev_find(ofproto, port->up.ofp_port);
6247     if (vsp) {
6248         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6249         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6250         free(vsp);
6251
6252         port->realdev_ofp_port = 0;
6253     } else {
6254         VLOG_ERR("missing vlan device record");
6255     }
6256     ovs_mutex_unlock(&ofproto->vsp_mutex);
6257 }
6258
6259 static void
6260 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6261 {
6262     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6263
6264     ovs_mutex_lock(&ofproto->vsp_mutex);
6265     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6266         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6267             == realdev_ofp_port)) {
6268         struct vlan_splinter *vsp;
6269
6270         vsp = xmalloc(sizeof *vsp);
6271         vsp->realdev_ofp_port = realdev_ofp_port;
6272         vsp->vlandev_ofp_port = port->up.ofp_port;
6273         vsp->vid = vid;
6274
6275         port->realdev_ofp_port = realdev_ofp_port;
6276
6277         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6278                     hash_ofp_port(port->up.ofp_port));
6279         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6280                     hash_realdev_vid(realdev_ofp_port, vid));
6281     } else {
6282         VLOG_ERR("duplicate vlan device record");
6283     }
6284     ovs_mutex_unlock(&ofproto->vsp_mutex);
6285 }
6286
6287 static odp_port_t
6288 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6289 {
6290     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6291     return ofport ? ofport->odp_port : ODPP_NONE;
6292 }
6293
6294 struct ofport_dpif *
6295 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6296 {
6297     struct ofport_dpif *port;
6298
6299     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6300     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6301                              &backer->odp_to_ofport_map) {
6302         if (port->odp_port == odp_port) {
6303             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6304             return port;
6305         }
6306     }
6307
6308     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6309     return NULL;
6310 }
6311
6312 static ofp_port_t
6313 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6314 {
6315     struct ofport_dpif *port;
6316
6317     port = odp_port_to_ofport(ofproto->backer, odp_port);
6318     if (port && &ofproto->up == port->up.ofproto) {
6319         return port->up.ofp_port;
6320     } else {
6321         return OFPP_NONE;
6322     }
6323 }
6324
6325 const struct ofproto_class ofproto_dpif_class = {
6326     init,
6327     enumerate_types,
6328     enumerate_names,
6329     del,
6330     port_open_type,
6331     type_run,
6332     type_run_fast,
6333     type_wait,
6334     alloc,
6335     construct,
6336     destruct,
6337     dealloc,
6338     run,
6339     run_fast,
6340     wait,
6341     get_memory_usage,
6342     flush,
6343     get_features,
6344     get_tables,
6345     port_alloc,
6346     port_construct,
6347     port_destruct,
6348     port_dealloc,
6349     port_modified,
6350     port_reconfigured,
6351     port_query_by_name,
6352     port_add,
6353     port_del,
6354     port_get_stats,
6355     port_dump_start,
6356     port_dump_next,
6357     port_dump_done,
6358     port_poll,
6359     port_poll_wait,
6360     port_is_lacp_current,
6361     NULL,                       /* rule_choose_table */
6362     rule_alloc,
6363     rule_construct,
6364     rule_insert,
6365     rule_delete,
6366     rule_destruct,
6367     rule_dealloc,
6368     rule_get_stats,
6369     rule_execute,
6370     rule_modify_actions,
6371     set_frag_handling,
6372     packet_out,
6373     set_netflow,
6374     get_netflow_ids,
6375     set_sflow,
6376     set_ipfix,
6377     set_cfm,
6378     get_cfm_status,
6379     set_bfd,
6380     get_bfd_status,
6381     set_stp,
6382     get_stp_status,
6383     set_stp_port,
6384     get_stp_port_status,
6385     get_stp_port_stats,
6386     set_queues,
6387     bundle_set,
6388     bundle_remove,
6389     mirror_set__,
6390     mirror_get_stats__,
6391     set_flood_vlans,
6392     is_mirror_output_bundle,
6393     forward_bpdu_changed,
6394     set_mac_table_config,
6395     set_realdev,
6396     NULL,                       /* meter_get_features */
6397     NULL,                       /* meter_set */
6398     NULL,                       /* meter_get */
6399     NULL,                       /* meter_del */
6400     group_alloc,                /* group_alloc */
6401     group_construct,            /* group_construct */
6402     group_destruct,             /* group_destruct */
6403     group_dealloc,              /* group_dealloc */
6404     group_modify,               /* group_modify */
6405     group_get_stats,            /* group_get_stats */
6406 };