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 "connectivity.h"
  29 #include "connmgr.h"
  30 #include "coverage.h"
  31 #include "cfm.h"
  32 #include "dpif.h"
  33 #include "dynamic-string.h"
  34 #include "fail-open.h"
  35 #include "guarded-list.h"
  36 #include "hmapx.h"
  37 #include "lacp.h"
  38 #include "learn.h"
  39 #include "mac-learning.h"
  40 #include "meta-flow.h"
  41 #include "multipath.h"
  42 #include "netdev-vport.h"
  43 #include "netdev.h"
  44 #include "netlink.h"
  45 #include "nx-match.h"
  46 #include "odp-util.h"
  47 #include "odp-execute.h"
  48 #include "ofp-util.h"
  49 #include "ofpbuf.h"
  50 #include "ofp-actions.h"
  51 #include "ofp-parse.h"
  52 #include "ofp-print.h"
  53 #include "ofproto-dpif-governor.h"
  54 #include "ofproto-dpif-ipfix.h"
  55 #include "ofproto-dpif-mirror.h"
  56 #include "ofproto-dpif-monitor.h"
  57 #include "ofproto-dpif-sflow.h"
  58 #include "ofproto-dpif-upcall.h"
  59 #include "ofproto-dpif-xlate.h"
  60 #include "poll-loop.h"
  61 #include "seq.h"
  62 #include "simap.h"
  63 #include "smap.h"
  64 #include "timer.h"
  65 #include "tunnel.h"
  66 #include "unaligned.h"
  67 #include "unixctl.h"
  68 #include "vlan-bitmap.h"
  69 #include "vlog.h"
  70
  71 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  72
  73 COVERAGE_DEFINE(ofproto_dpif_expired);
  74 COVERAGE_DEFINE(facet_revalidate);
  75 COVERAGE_DEFINE(facet_unexpected);
  76 COVERAGE_DEFINE(facet_create);
  77 COVERAGE_DEFINE(facet_remove);
  78 COVERAGE_DEFINE(subfacet_create);
  79 COVERAGE_DEFINE(subfacet_destroy);
  80 COVERAGE_DEFINE(subfacet_install_fail);
  81 COVERAGE_DEFINE(packet_in_overflow);
  82
  83 /* Number of implemented OpenFlow tables. */
  84 enum { N_TABLES = 255 };
  85 enum { TBL_INTERNAL = N_TABLES - 1 };    /* Used for internal hidden rules. */
  86 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
  87
  88 struct flow_miss;
  89 struct facet;
  90
  91 struct rule_dpif {
  92     struct rule up;
  93
  94     /* These statistics:
  95      *
  96      *   - Do include packets and bytes from facets that have been deleted or
  97      *     whose own statistics have been folded into the rule.
  98      *
  99      *   - Do include packets and bytes sent "by hand" that were accounted to
 100      *     the rule without any facet being involved (this is a rare corner
 101      *     case in rule_execute()).
 102      *
 103      *   - Do not include packet or bytes that can be obtained from any facet's
 104      *     packet_count or byte_count member or that can be obtained from the
 105      *     datapath by, e.g., dpif_flow_get() for any subfacet.
 106      */
 107     struct ovs_mutex stats_mutex;
 108     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 109     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 110 };
 111
 112 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
 113 static struct rule_dpif *rule_dpif_cast(const struct rule *);
 114
 115 struct group_dpif {
 116     struct ofgroup up;
 117
 118     /* These statistics:
 119      *
 120      *   - Do include packets and bytes from facets that have been deleted or
 121      *     whose own statistics have been folded into the rule.
 122      *
 123      *   - Do include packets and bytes sent "by hand" that were accounted to
 124      *     the rule without any facet being involved (this is a rare corner
 125      *     case in rule_execute()).
 126      *
 127      *   - Do not include packet or bytes that can be obtained from any facet's
 128      *     packet_count or byte_count member or that can be obtained from the
 129      *     datapath by, e.g., dpif_flow_get() for any subfacet.
 130      */
 131     struct ovs_mutex stats_mutex;
 132     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 133     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 134     struct bucket_counter *bucket_stats OVS_GUARDED;  /* Bucket statistics. */
 135 };
 136
 137 struct ofbundle {
 138     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
 139     struct ofproto_dpif *ofproto; /* Owning ofproto. */
 140     void *aux;                  /* Key supplied by ofproto's client. */
 141     char *name;                 /* Identifier for log messages. */
 142
 143     /* Configuration. */
 144     struct list ports;          /* Contains "struct ofport"s. */
 145     enum port_vlan_mode vlan_mode; /* VLAN mode */
 146     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
 147     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
 148                                  * NULL if all VLANs are trunked. */
 149     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 150     struct bond *bond;          /* Nonnull iff more than one port. */
 151     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 152
 153     /* Status. */
 154     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 155 };
 156
 157 static void bundle_remove(struct ofport *);
 158 static void bundle_update(struct ofbundle *);
 159 static void bundle_destroy(struct ofbundle *);
 160 static void bundle_del_port(struct ofport_dpif *);
 161 static void bundle_run(struct ofbundle *);
 162 static void bundle_wait(struct ofbundle *);
 163
 164 static void stp_run(struct ofproto_dpif *ofproto);
 165 static void stp_wait(struct ofproto_dpif *ofproto);
 166 static int set_stp_port(struct ofport *,
 167                         const struct ofproto_port_stp_settings *);
 168
 169 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
 170                               enum slow_path_reason,
 171                               uint64_t *stub, size_t stub_size,
 172                               const struct nlattr **actionsp,
 173                               size_t *actions_lenp);
 174
 175 /* A subfacet (see "struct subfacet" below) has three possible installation
 176  * states:
 177  *
 178  *   - SF_NOT_INSTALLED: Not installed in the datapath.  This will only be the
 179  *     case just after the subfacet is created, just before the subfacet is
 180  *     destroyed, or if the datapath returns an error when we try to install a
 181  *     subfacet.
 182  *
 183  *   - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
 184  *
 185  *   - SF_SLOW_PATH: An action that sends every packet for the subfacet through
 186  *     ofproto_dpif is installed in the datapath.
 187  */
 188 enum subfacet_path {
 189     SF_NOT_INSTALLED,           /* No datapath flow for this subfacet. */
 190     SF_FAST_PATH,               /* Full actions are installed. */
 191     SF_SLOW_PATH,               /* Send-to-userspace action is installed. */
 192 };
 193
 194 /* A dpif flow and actions associated with a facet.
 195  *
 196  * See also the large comment on struct facet. */
 197 struct subfacet {
 198     /* Owners. */
 199     struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
 200     struct list list_node;      /* In struct facet's 'facets' list. */
 201     struct facet *facet;        /* Owning facet. */
 202     struct dpif_backer *backer; /* Owning backer. */
 203
 204     struct nlattr *key;
 205     int key_len;
 206
 207     long long int used;         /* Time last used; time created if not used. */
 208     long long int created;      /* Time created. */
 209
 210     uint64_t dp_packet_count;   /* Last known packet count in the datapath. */
 211     uint64_t dp_byte_count;     /* Last known byte count in the datapath. */
 212
 213     enum subfacet_path path;    /* Installed in datapath? */
 214 };
 215
 216 #define SUBFACET_DESTROY_MAX_BATCH 50
 217
 218 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *,
 219                                         uint32_t key_hash);
 220 static struct subfacet *subfacet_find(struct dpif_backer *,
 221                                       const struct nlattr *key, size_t key_len,
 222                                       uint32_t key_hash);
 223 static void subfacet_destroy(struct subfacet *);
 224 static void subfacet_destroy__(struct subfacet *);
 225 static void subfacet_destroy_batch(struct dpif_backer *,
 226                                    struct subfacet **, int n);
 227 static void subfacet_reset_dp_stats(struct subfacet *,
 228                                     struct dpif_flow_stats *);
 229 static void subfacet_update_stats(struct subfacet *,
 230                                   const struct dpif_flow_stats *);
 231 static int subfacet_install(struct subfacet *,
 232                             const struct ofpbuf *odp_actions,
 233                             struct dpif_flow_stats *);
 234 static void subfacet_uninstall(struct subfacet *);
 235
 236 /* A unique, non-overlapping instantiation of an OpenFlow flow.
 237  *
 238  * A facet associates a "struct flow", which represents the Open vSwitch
 239  * userspace idea of an exact-match flow, with one or more subfacets.
 240  * While the facet is created based on an exact-match flow, it is stored
 241  * within the ofproto based on the wildcards that could be expressed
 242  * based on the flow table and other configuration.  (See the 'wc'
 243  * description in "struct xlate_out" for more details.)
 244  *
 245  * Each subfacet tracks the datapath's idea of the flow equivalent to
 246  * the facet.  When the kernel module (or other dpif implementation) and
 247  * Open vSwitch userspace agree on the definition of a flow key, there
 248  * is exactly one subfacet per facet.  If the dpif implementation
 249  * supports more-specific flow matching than userspace, however, a facet
 250  * can have more than one subfacet.  Examples include the dpif
 251  * implementation not supporting the same wildcards as userspace or some
 252  * distinction in flow that userspace simply doesn't understand.
 253  *
 254  * Flow expiration works in terms of subfacets, so a facet must have at
 255  * least one subfacet or it will never expire, leaking memory. */
 256 struct facet {
 257     /* Owner. */
 258     struct ofproto_dpif *ofproto;
 259
 260     /* Owned data. */
 261     struct list subfacets;
 262     long long int used;         /* Time last used; time created if not used. */
 263
 264     /* Key. */
 265     struct flow flow;           /* Flow of the creating subfacet. */
 266     struct cls_rule cr;         /* In 'ofproto_dpif's facets classifier. */
 267
 268     /* These statistics:
 269      *
 270      *   - Do include packets and bytes sent "by hand", e.g. with
 271      *     dpif_execute().
 272      *
 273      *   - Do include packets and bytes that were obtained from the datapath
 274      *     when a subfacet's statistics were reset (e.g. dpif_flow_put() with
 275      *     DPIF_FP_ZERO_STATS).
 276      *
 277      *   - Do not include packets or bytes that can be obtained from the
 278      *     datapath for any existing subfacet.
 279      */
 280     uint64_t packet_count;       /* Number of packets received. */
 281     uint64_t byte_count;         /* Number of bytes received. */
 282
 283     /* Resubmit statistics. */
 284     uint64_t prev_packet_count;  /* Number of packets from last stats push. */
 285     uint64_t prev_byte_count;    /* Number of bytes from last stats push. */
 286     long long int prev_used;     /* Used time from last stats push. */
 287
 288     /* Accounting. */
 289     uint64_t accounted_bytes;    /* Bytes processed by facet_account(). */
 290     struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
 291     uint16_t tcp_flags;          /* TCP flags seen for this 'rule'. */
 292
 293     struct xlate_out xout;
 294
 295     /* Storage for a single subfacet, to reduce malloc() time and space
 296      * overhead.  (A facet always has at least one subfacet and in the common
 297      * case has exactly one subfacet.  However, 'one_subfacet' may not
 298      * always be valid, since it could have been removed after newer
 299      * subfacets were pushed onto the 'subfacets' list.) */
 300     struct subfacet one_subfacet;
 301
 302     long long int learn_rl;      /* Rate limiter for facet_learn(). */
 303 };
 304
 305 static struct facet *facet_create(const struct flow_miss *);
 306 static void facet_remove(struct facet *);
 307 static void facet_free(struct facet *);
 308
 309 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
 310 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
 311                                         const struct flow *);
 312 static bool facet_revalidate(struct facet *);
 313 static bool facet_check_consistency(struct facet *);
 314
 315 static void facet_flush_stats(struct facet *);
 316
 317 static void facet_reset_counters(struct facet *);
 318 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
 319                             struct dpif_flow_stats *, bool may_learn);
 320 static void facet_push_stats(struct facet *, bool may_learn);
 321 static void facet_learn(struct facet *);
 322 static void facet_account(struct facet *);
 323 static void push_all_stats(void);
 324
 325 static bool facet_is_controller_flow(struct facet *);
 326
 327 struct ofport_dpif {
 328     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 329     struct ofport up;
 330
 331     odp_port_t odp_port;
 332     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 333     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
 334     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 335     struct bfd *bfd;            /* BFD, if any. */
 336     bool may_enable;            /* May be enabled in bonds. */
 337     bool is_tunnel;             /* This port is a tunnel. */
 338     bool is_layer3;             /* This is a layer 3 port. */
 339     long long int carrier_seq;  /* Carrier status changes. */
 340     struct ofport_dpif *peer;   /* Peer if patch port. */
 341
 342     /* Spanning tree. */
 343     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 344     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 345     long long int stp_state_entered;
 346
 347     /* Queue to DSCP mapping. */
 348     struct ofproto_port_queue *qdscp;
 349     size_t n_qdscp;
 350
 351     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 352      *
 353      * This is deprecated.  It is only for compatibility with broken device
 354      * drivers in old versions of Linux that do not properly support VLANs when
 355      * VLAN devices are not used.  When broken device drivers are no longer in
 356      * widespread use, we will delete these interfaces. */
 357     ofp_port_t realdev_ofp_port;
 358     int vlandev_vid;
 359 };
 360
 361 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 362  *
 363  * This is deprecated.  It is only for compatibility with broken device drivers
 364  * in old versions of Linux that do not properly support VLANs when VLAN
 365  * devices are not used.  When broken device drivers are no longer in
 366  * widespread use, we will delete these interfaces. */
 367 struct vlan_splinter {
 368     struct hmap_node realdev_vid_node;
 369     struct hmap_node vlandev_node;
 370     ofp_port_t realdev_ofp_port;
 371     ofp_port_t vlandev_ofp_port;
 372     int vid;
 373 };
 374
 375 static void vsp_remove(struct ofport_dpif *);
 376 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 377
 378 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 379                                        ofp_port_t);
 380
 381 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 382                                        odp_port_t);
 383
 384 static struct ofport_dpif *
 385 ofport_dpif_cast(const struct ofport *ofport)
 386 {
 387     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 388 }
 389
 390 static void port_run(struct ofport_dpif *);
 391 static int set_bfd(struct ofport *, const struct smap *);
 392 static int set_cfm(struct ofport *, const struct cfm_settings *);
 393 static void ofport_update_peer(struct ofport_dpif *);
 394 static void run_fast_rl(void);
 395 static int run_fast(struct ofproto *);
 396
 397 struct dpif_completion {
 398     struct list list_node;
 399     struct ofoperation *op;
 400 };
 401
 402 /* Reasons that we might need to revalidate every facet, and corresponding
 403  * coverage counters.
 404  *
 405  * A value of 0 means that there is no need to revalidate.
 406  *
 407  * It would be nice to have some cleaner way to integrate with coverage
 408  * counters, but with only a few reasons I guess this is good enough for
 409  * now. */
 410 enum revalidate_reason {
 411     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 412     REV_STP,                   /* Spanning tree protocol port status change. */
 413     REV_BOND,                  /* Bonding changed. */
 414     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 415     REV_FLOW_TABLE,            /* Flow table changed. */
 416     REV_MAC_LEARNING,          /* Mac learning changed. */
 417     REV_INCONSISTENCY          /* Facet self-check failed. */
 418 };
 419 COVERAGE_DEFINE(rev_reconfigure);
 420 COVERAGE_DEFINE(rev_stp);
 421 COVERAGE_DEFINE(rev_bond);
 422 COVERAGE_DEFINE(rev_port_toggled);
 423 COVERAGE_DEFINE(rev_flow_table);
 424 COVERAGE_DEFINE(rev_mac_learning);
 425 COVERAGE_DEFINE(rev_inconsistency);
 426
 427 /* All datapaths of a given type share a single dpif backer instance. */
 428 struct dpif_backer {
 429     char *type;
 430     int refcount;
 431     struct dpif *dpif;
 432     struct udpif *udpif;
 433     struct timer next_expiration;
 434
 435     struct ovs_rwlock odp_to_ofport_lock;
 436     struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */
 437
 438     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 439
 440     /* Facet revalidation flags applying to facets which use this backer. */
 441     enum revalidate_reason need_revalidate; /* Revalidate every facet. */
 442
 443     struct hmap drop_keys; /* Set of dropped odp keys. */
 444     bool recv_set_enable; /* Enables or disables receiving packets. */
 445
 446     struct hmap subfacets;
 447     struct governor *governor;
 448
 449     /* Subfacet statistics.
 450      *
 451      * These keep track of the total number of subfacets added and deleted and
 452      * flow life span.  They are useful for computing the flow rates stats
 453      * exposed via "ovs-appctl dpif/show".  The goal is to learn about
 454      * traffic patterns in ways that we can use later to improve Open vSwitch
 455      * performance in new situations.  */
 456     unsigned max_n_subfacet;         /* Maximum number of flows */
 457     unsigned avg_n_subfacet;         /* Average number of flows. */
 458     long long int avg_subfacet_life; /* Average life span of subfacets. */
 459
 460     /* Number of upcall handling threads. */
 461     unsigned int n_handler_threads;
 462 };
 463
 464 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 465 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 466
 467 static void drop_key_clear(struct dpif_backer *);
 468
 469 struct ofproto_dpif {
 470     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 471     struct ofproto up;
 472     struct dpif_backer *backer;
 473
 474     /* Special OpenFlow rules. */
 475     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 476     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 477     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 478
 479     /* Bridging. */
 480     struct netflow *netflow;
 481     struct dpif_sflow *sflow;
 482     struct dpif_ipfix *ipfix;
 483     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 484     struct mac_learning *ml;
 485     bool has_bonded_bundles;
 486     bool lacp_enabled;
 487     struct mbridge *mbridge;
 488
 489     /* Facets. */
 490     struct classifier facets;     /* Contains 'struct facet's. */
 491     long long int consistency_rl;
 492
 493     struct ovs_mutex stats_mutex;
 494     struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
 495                                             * consumed in userspace. */
 496
 497     /* Spanning tree. */
 498     struct stp *stp;
 499     long long int stp_last_tick;
 500
 501     /* VLAN splinters. */
 502     struct ovs_mutex vsp_mutex;
 503     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 504     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 505
 506     /* Ports. */
 507     struct sset ports;             /* Set of standard port names. */
 508     struct sset ghost_ports;       /* Ports with no datapath port. */
 509     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 510     int port_poll_errno;           /* Last errno for port_poll() reply. */
 511     uint64_t change_seq;           /* Connectivity status changes. */
 512
 513     /* Per ofproto's dpif stats. */
 514     uint64_t n_hit;
 515     uint64_t n_missed;
 516
 517     /* Work queues. */
 518     struct guarded_list pins;      /* Contains "struct ofputil_packet_in"s. */
 519 };
 520
 521 /* By default, flows in the datapath are wildcarded (megaflows).  They
 522  * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
 523 static bool enable_megaflows = true;
 524
 525 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 526 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 527
 528 static void ofproto_dpif_unixctl_init(void);
 529
 530 static inline struct ofproto_dpif *
 531 ofproto_dpif_cast(const struct ofproto *ofproto)
 532 {
 533     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 534     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 535 }
 536
 537 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
 538                                         ofp_port_t ofp_port);
 539 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
 540                           const struct ofpbuf *packet,
 541                           const struct ofpact[], size_t ofpacts_len,
 542                           struct ds *);
 543
 544 /* Upcalls. */
 545 static void handle_upcalls(struct dpif_backer *);
 546
 547 /* Flow expiration. */
 548 static int expire(struct dpif_backer *);
 549
 550 /* NetFlow. */
 551 static void send_netflow_active_timeouts(struct ofproto_dpif *);
 552
 553 /* Global variables. */
 554 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 555
 556 /* Initial mappings of port to bridge mappings. */
 557 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 558
 559 /* Executes 'fm'.  The caller retains ownership of 'fm' and everything in
 560  * it. */
 561 void
 562 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 563                       struct ofputil_flow_mod *fm)
 564 {
 565     ofproto_flow_mod(&ofproto->up, fm);
 566 }
 567
 568 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 569  * Takes ownership of 'pin' and pin->packet. */
 570 void
 571 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 572                             struct ofproto_packet_in *pin)
 573 {
 574     if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
 575         COVERAGE_INC(packet_in_overflow);
 576         free(CONST_CAST(void *, pin->up.packet));
 577         free(pin);
 578     }
 579 }
 580 \f
 581 /* Factory functions. */
 582
 583 static void
 584 init(const struct shash *iface_hints)
 585 {
 586     struct shash_node *node;
 587
 588     /* Make a local copy, since we don't own 'iface_hints' elements. */
 589     SHASH_FOR_EACH(node, iface_hints) {
 590         const struct iface_hint *orig_hint = node->data;
 591         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 592
 593         new_hint->br_name = xstrdup(orig_hint->br_name);
 594         new_hint->br_type = xstrdup(orig_hint->br_type);
 595         new_hint->ofp_port = orig_hint->ofp_port;
 596
 597         shash_add(&init_ofp_ports, node->name, new_hint);
 598     }
 599 }
 600
 601 static void
 602 enumerate_types(struct sset *types)
 603 {
 604     dp_enumerate_types(types);
 605 }
 606
 607 static int
 608 enumerate_names(const char *type, struct sset *names)
 609 {
 610     struct ofproto_dpif *ofproto;
 611
 612     sset_clear(names);
 613     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 614         if (strcmp(type, ofproto->up.type)) {
 615             continue;
 616         }
 617         sset_add(names, ofproto->up.name);
 618     }
 619
 620     return 0;
 621 }
 622
 623 static int
 624 del(const char *type, const char *name)
 625 {
 626     struct dpif *dpif;
 627     int error;
 628
 629     error = dpif_open(name, type, &dpif);
 630     if (!error) {
 631         error = dpif_delete(dpif);
 632         dpif_close(dpif);
 633     }
 634     return error;
 635 }
 636 \f
 637 static const char *
 638 port_open_type(const char *datapath_type, const char *port_type)
 639 {
 640     return dpif_port_open_type(datapath_type, port_type);
 641 }
 642
 643 /* Type functions. */
 644
 645 static void process_dpif_port_changes(struct dpif_backer *);
 646 static void process_dpif_all_ports_changed(struct dpif_backer *);
 647 static void process_dpif_port_change(struct dpif_backer *,
 648                                      const char *devname);
 649 static void process_dpif_port_error(struct dpif_backer *, int error);
 650
 651 static struct ofproto_dpif *
 652 lookup_ofproto_dpif_by_port_name(const char *name)
 653 {
 654     struct ofproto_dpif *ofproto;
 655
 656     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 657         if (sset_contains(&ofproto->ports, name)) {
 658             return ofproto;
 659         }
 660     }
 661
 662     return NULL;
 663 }
 664
 665 static int
 666 type_run(const char *type)
 667 {
 668     static long long int push_timer = LLONG_MIN;
 669     struct dpif_backer *backer;
 670
 671     backer = shash_find_data(&all_dpif_backers, type);
 672     if (!backer) {
 673         /* This is not necessarily a problem, since backers are only
 674          * created on demand. */
 675         return 0;
 676     }
 677
 678     dpif_run(backer->dpif);
 679
 680     /* The most natural place to push facet statistics is when they're pulled
 681      * from the datapath.  However, when there are many flows in the datapath,
 682      * this expensive operation can occur so frequently, that it reduces our
 683      * ability to quickly set up flows.  To reduce the cost, we push statistics
 684      * here instead. */
 685     if (time_msec() > push_timer) {
 686         push_timer = time_msec() + 2000;
 687         push_all_stats();
 688     }
 689
 690     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 691      * and the configuration has now changed to "false", enable receiving
 692      * packets from the datapath. */
 693     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 694         int error;
 695
 696         backer->recv_set_enable = true;
 697
 698         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 699         if (error) {
 700             udpif_recv_set(backer->udpif, 0, false);
 701             VLOG_ERR("Failed to enable receiving packets in dpif.");
 702             return error;
 703         }
 704         udpif_recv_set(backer->udpif, n_handler_threads,
 705                        backer->recv_set_enable);
 706         dpif_flow_flush(backer->dpif);
 707         backer->need_revalidate = REV_RECONFIGURE;
 708     }
 709
 710     /* If the n_handler_threads is reconfigured, call udpif_recv_set()
 711      * to reset the handler threads. */
 712     if (backer->n_handler_threads != n_handler_threads) {
 713         udpif_recv_set(backer->udpif, n_handler_threads,
 714                        backer->recv_set_enable);
 715         backer->n_handler_threads = n_handler_threads;
 716     }
 717
 718     if (backer->need_revalidate) {
 719         struct ofproto_dpif *ofproto;
 720         struct simap_node *node;
 721         struct simap tmp_backers;
 722
 723         /* Handle tunnel garbage collection. */
 724         simap_init(&tmp_backers);
 725         simap_swap(&backer->tnl_backers, &tmp_backers);
 726
 727         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 728             struct ofport_dpif *iter;
 729
 730             if (backer != ofproto->backer) {
 731                 continue;
 732             }
 733
 734             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 735                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 736                 const char *dp_port;
 737
 738                 if (!iter->is_tunnel) {
 739                     continue;
 740                 }
 741
 742                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 743                                                      namebuf, sizeof namebuf);
 744                 node = simap_find(&tmp_backers, dp_port);
 745                 if (node) {
 746                     simap_put(&backer->tnl_backers, dp_port, node->data);
 747                     simap_delete(&tmp_backers, node);
 748                     node = simap_find(&backer->tnl_backers, dp_port);
 749                 } else {
 750                     node = simap_find(&backer->tnl_backers, dp_port);
 751                     if (!node) {
 752                         odp_port_t odp_port = ODPP_NONE;
 753
 754                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 755                                            &odp_port)) {
 756                             simap_put(&backer->tnl_backers, dp_port,
 757                                       odp_to_u32(odp_port));
 758                             node = simap_find(&backer->tnl_backers, dp_port);
 759                         }
 760                     }
 761                 }
 762
 763                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 764                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 765                                          iter->odp_port)) {
 766                     backer->need_revalidate = REV_RECONFIGURE;
 767                 }
 768             }
 769         }
 770
 771         SIMAP_FOR_EACH (node, &tmp_backers) {
 772             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 773         }
 774         simap_destroy(&tmp_backers);
 775
 776         switch (backer->need_revalidate) {
 777         case REV_RECONFIGURE:   COVERAGE_INC(rev_reconfigure);   break;
 778         case REV_STP:           COVERAGE_INC(rev_stp);           break;
 779         case REV_BOND:          COVERAGE_INC(rev_bond);          break;
 780         case REV_PORT_TOGGLED:  COVERAGE_INC(rev_port_toggled);  break;
 781         case REV_FLOW_TABLE:    COVERAGE_INC(rev_flow_table);    break;
 782         case REV_MAC_LEARNING:  COVERAGE_INC(rev_mac_learning);  break;
 783         case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
 784         }
 785         backer->need_revalidate = 0;
 786
 787         /* Clear the drop_keys in case we should now be accepting some
 788          * formerly dropped flows. */
 789         drop_key_clear(backer);
 790
 791         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 792             struct facet *facet, *next;
 793             struct ofport_dpif *ofport;
 794             struct cls_cursor cursor;
 795             struct ofbundle *bundle;
 796
 797             if (ofproto->backer != backer) {
 798                 continue;
 799             }
 800
 801             ovs_rwlock_wrlock(&xlate_rwlock);
 802             xlate_ofproto_set(ofproto, ofproto->up.name,
 803                               ofproto->backer->dpif, ofproto->miss_rule,
 804                               ofproto->no_packet_in_rule, ofproto->ml,
 805                               ofproto->stp, ofproto->mbridge,
 806                               ofproto->sflow, ofproto->ipfix,
 807                               ofproto->up.frag_handling,
 808                               ofproto->up.forward_bpdu,
 809                               connmgr_has_in_band(ofproto->up.connmgr),
 810                               ofproto->netflow != NULL);
 811
 812             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 813                 xlate_bundle_set(ofproto, bundle, bundle->name,
 814                                  bundle->vlan_mode, bundle->vlan,
 815                                  bundle->trunks, bundle->use_priority_tags,
 816                                  bundle->bond, bundle->lacp,
 817                                  bundle->floodable);
 818             }
 819
 820             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 821                 int stp_port = ofport->stp_port
 822                     ? stp_port_no(ofport->stp_port)
 823                     : -1;
 824                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 825                                  ofport->up.ofp_port, ofport->odp_port,
 826                                  ofport->up.netdev, ofport->cfm,
 827                                  ofport->bfd, ofport->peer, stp_port,
 828                                  ofport->qdscp, ofport->n_qdscp,
 829                                  ofport->up.pp.config, ofport->up.pp.state,
 830                                  ofport->is_tunnel, ofport->may_enable);
 831             }
 832             ovs_rwlock_unlock(&xlate_rwlock);
 833
 834             /* Only ofproto-dpif cares about the facet classifier so we just
 835              * lock cls_cursor_init() to appease the thread safety analysis. */
 836             ovs_rwlock_rdlock(&ofproto->facets.rwlock);
 837             cls_cursor_init(&cursor, &ofproto->facets, NULL);
 838             ovs_rwlock_unlock(&ofproto->facets.rwlock);
 839             CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
 840                 facet_revalidate(facet);
 841                 run_fast_rl();
 842             }
 843         }
 844
 845         udpif_revalidate(backer->udpif);
 846     }
 847
 848     if (!backer->recv_set_enable) {
 849         /* Wake up before a max of 1000ms. */
 850         timer_set_duration(&backer->next_expiration, 1000);
 851     } else if (timer_expired(&backer->next_expiration)) {
 852         int delay = expire(backer);
 853         timer_set_duration(&backer->next_expiration, delay);
 854     }
 855
 856     process_dpif_port_changes(backer);
 857
 858     if (backer->governor) {
 859         size_t n_subfacets;
 860
 861         governor_run(backer->governor);
 862
 863         /* If the governor has shrunk to its minimum size and the number of
 864          * subfacets has dwindled, then drop the governor entirely.
 865          *
 866          * For hysteresis, the number of subfacets to drop the governor is
 867          * smaller than the number needed to trigger its creation. */
 868         n_subfacets = hmap_count(&backer->subfacets);
 869         if (n_subfacets * 4 < flow_eviction_threshold
 870             && governor_is_idle(backer->governor)) {
 871             governor_destroy(backer->governor);
 872             backer->governor = NULL;
 873         }
 874     }
 875
 876     return 0;
 877 }
 878
 879 /* Check for and handle port changes in 'backer''s dpif. */
 880 static void
 881 process_dpif_port_changes(struct dpif_backer *backer)
 882 {
 883     for (;;) {
 884         char *devname;
 885         int error;
 886
 887         error = dpif_port_poll(backer->dpif, &devname);
 888         switch (error) {
 889         case EAGAIN:
 890             return;
 891
 892         case ENOBUFS:
 893             process_dpif_all_ports_changed(backer);
 894             break;
 895
 896         case 0:
 897             process_dpif_port_change(backer, devname);
 898             free(devname);
 899             break;
 900
 901         default:
 902             process_dpif_port_error(backer, error);
 903             break;
 904         }
 905     }
 906 }
 907
 908 static void
 909 process_dpif_all_ports_changed(struct dpif_backer *backer)
 910 {
 911     struct ofproto_dpif *ofproto;
 912     struct dpif_port dpif_port;
 913     struct dpif_port_dump dump;
 914     struct sset devnames;
 915     const char *devname;
 916
 917     sset_init(&devnames);
 918     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 919         if (ofproto->backer == backer) {
 920             struct ofport *ofport;
 921
 922             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 923                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 924             }
 925         }
 926     }
 927     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 928         sset_add(&devnames, dpif_port.name);
 929     }
 930
 931     SSET_FOR_EACH (devname, &devnames) {
 932         process_dpif_port_change(backer, devname);
 933     }
 934     sset_destroy(&devnames);
 935 }
 936
 937 static void
 938 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 939 {
 940     struct ofproto_dpif *ofproto;
 941     struct dpif_port port;
 942
 943     /* Don't report on the datapath's device. */
 944     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 945         return;
 946     }
 947
 948     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 949                    &all_ofproto_dpifs) {
 950         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 951             return;
 952         }
 953     }
 954
 955     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 956     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 957         /* The port was removed.  If we know the datapath,
 958          * report it through poll_set().  If we don't, it may be
 959          * notifying us of a removal we initiated, so ignore it.
 960          * If there's a pending ENOBUFS, let it stand, since
 961          * everything will be reevaluated. */
 962         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 963             sset_add(&ofproto->port_poll_set, devname);
 964             ofproto->port_poll_errno = 0;
 965         }
 966     } else if (!ofproto) {
 967         /* The port was added, but we don't know with which
 968          * ofproto we should associate it.  Delete it. */
 969         dpif_port_del(backer->dpif, port.port_no);
 970     } else {
 971         struct ofport_dpif *ofport;
 972
 973         ofport = ofport_dpif_cast(shash_find_data(
 974                                       &ofproto->up.port_by_name, devname));
 975         if (ofport
 976             && ofport->odp_port != port.port_no
 977             && !odp_port_to_ofport(backer, port.port_no))
 978         {
 979             /* 'ofport''s datapath port number has changed from
 980              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 981              * structures to match. */
 982             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 983             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 984             ofport->odp_port = port.port_no;
 985             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 986                         hash_odp_port(port.port_no));
 987             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 988             backer->need_revalidate = REV_RECONFIGURE;
 989         }
 990     }
 991     dpif_port_destroy(&port);
 992 }
 993
 994 /* Propagate 'error' to all ofprotos based on 'backer'. */
 995 static void
 996 process_dpif_port_error(struct dpif_backer *backer, int error)
 997 {
 998     struct ofproto_dpif *ofproto;
 999
1000     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1001         if (ofproto->backer == backer) {
1002             sset_clear(&ofproto->port_poll_set);
1003             ofproto->port_poll_errno = error;
1004         }
1005     }
1006 }
1007
1008 static int
1009 dpif_backer_run_fast(struct dpif_backer *backer)
1010 {
1011     handle_upcalls(backer);
1012
1013     return 0;
1014 }
1015
1016 static int
1017 type_run_fast(const char *type)
1018 {
1019     struct dpif_backer *backer;
1020
1021     backer = shash_find_data(&all_dpif_backers, type);
1022     if (!backer) {
1023         /* This is not necessarily a problem, since backers are only
1024          * created on demand. */
1025         return 0;
1026     }
1027
1028     return dpif_backer_run_fast(backer);
1029 }
1030
1031 static void
1032 run_fast_rl(void)
1033 {
1034     static long long int port_rl = LLONG_MIN;
1035
1036     if (time_msec() >= port_rl) {
1037         struct ofproto_dpif *ofproto;
1038
1039         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1040             run_fast(&ofproto->up);
1041         }
1042         port_rl = time_msec() + 200;
1043     }
1044 }
1045
1046 static void
1047 type_wait(const char *type)
1048 {
1049     struct dpif_backer *backer;
1050
1051     backer = shash_find_data(&all_dpif_backers, type);
1052     if (!backer) {
1053         /* This is not necessarily a problem, since backers are only
1054          * created on demand. */
1055         return;
1056     }
1057
1058     if (backer->governor) {
1059         governor_wait(backer->governor);
1060     }
1061
1062     timer_wait(&backer->next_expiration);
1063     dpif_wait(backer->dpif);
1064     udpif_wait(backer->udpif);
1065 }
1066 \f
1067 /* Basic life-cycle. */
1068
1069 static int add_internal_flows(struct ofproto_dpif *);
1070
1071 static struct ofproto *
1072 alloc(void)
1073 {
1074     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1075     return &ofproto->up;
1076 }
1077
1078 static void
1079 dealloc(struct ofproto *ofproto_)
1080 {
1081     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1082     free(ofproto);
1083 }
1084
1085 static void
1086 close_dpif_backer(struct dpif_backer *backer)
1087 {
1088     ovs_assert(backer->refcount > 0);
1089
1090     if (--backer->refcount) {
1091         return;
1092     }
1093
1094     drop_key_clear(backer);
1095     hmap_destroy(&backer->drop_keys);
1096
1097     udpif_destroy(backer->udpif);
1098
1099     simap_destroy(&backer->tnl_backers);
1100     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1101     hmap_destroy(&backer->odp_to_ofport_map);
1102     shash_find_and_delete(&all_dpif_backers, backer->type);
1103     free(backer->type);
1104     dpif_close(backer->dpif);
1105
1106     ovs_assert(hmap_is_empty(&backer->subfacets));
1107     hmap_destroy(&backer->subfacets);
1108     governor_destroy(backer->governor);
1109
1110     free(backer);
1111 }
1112
1113 /* Datapath port slated for removal from datapath. */
1114 struct odp_garbage {
1115     struct list list_node;
1116     odp_port_t odp_port;
1117 };
1118
1119 static int
1120 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1121 {
1122     struct dpif_backer *backer;
1123     struct dpif_port_dump port_dump;
1124     struct dpif_port port;
1125     struct shash_node *node;
1126     struct list garbage_list;
1127     struct odp_garbage *garbage, *next;
1128     struct sset names;
1129     char *backer_name;
1130     const char *name;
1131     int error;
1132
1133     backer = shash_find_data(&all_dpif_backers, type);
1134     if (backer) {
1135         backer->refcount++;
1136         *backerp = backer;
1137         return 0;
1138     }
1139
1140     backer_name = xasprintf("ovs-%s", type);
1141
1142     /* Remove any existing datapaths, since we assume we're the only
1143      * userspace controlling the datapath. */
1144     sset_init(&names);
1145     dp_enumerate_names(type, &names);
1146     SSET_FOR_EACH(name, &names) {
1147         struct dpif *old_dpif;
1148
1149         /* Don't remove our backer if it exists. */
1150         if (!strcmp(name, backer_name)) {
1151             continue;
1152         }
1153
1154         if (dpif_open(name, type, &old_dpif)) {
1155             VLOG_WARN("couldn't open old datapath %s to remove it", name);
1156         } else {
1157             dpif_delete(old_dpif);
1158             dpif_close(old_dpif);
1159         }
1160     }
1161     sset_destroy(&names);
1162
1163     backer = xmalloc(sizeof *backer);
1164
1165     error = dpif_create_and_open(backer_name, type, &backer->dpif);
1166     free(backer_name);
1167     if (error) {
1168         VLOG_ERR("failed to open datapath of type %s: %s", type,
1169                  ovs_strerror(error));
1170         free(backer);
1171         return error;
1172     }
1173     backer->udpif = udpif_create(backer, backer->dpif);
1174
1175     backer->type = xstrdup(type);
1176     backer->governor = NULL;
1177     backer->refcount = 1;
1178     hmap_init(&backer->odp_to_ofport_map);
1179     ovs_rwlock_init(&backer->odp_to_ofport_lock);
1180     hmap_init(&backer->drop_keys);
1181     hmap_init(&backer->subfacets);
1182     timer_set_duration(&backer->next_expiration, 1000);
1183     backer->need_revalidate = 0;
1184     simap_init(&backer->tnl_backers);
1185     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1186     *backerp = backer;
1187
1188     if (backer->recv_set_enable) {
1189         dpif_flow_flush(backer->dpif);
1190     }
1191
1192     /* Loop through the ports already on the datapath and remove any
1193      * that we don't need anymore. */
1194     list_init(&garbage_list);
1195     dpif_port_dump_start(&port_dump, backer->dpif);
1196     while (dpif_port_dump_next(&port_dump, &port)) {
1197         node = shash_find(&init_ofp_ports, port.name);
1198         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1199             garbage = xmalloc(sizeof *garbage);
1200             garbage->odp_port = port.port_no;
1201             list_push_front(&garbage_list, &garbage->list_node);
1202         }
1203     }
1204     dpif_port_dump_done(&port_dump);
1205
1206     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1207         dpif_port_del(backer->dpif, garbage->odp_port);
1208         list_remove(&garbage->list_node);
1209         free(garbage);
1210     }
1211
1212     shash_add(&all_dpif_backers, type, backer);
1213
1214     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1215     if (error) {
1216         VLOG_ERR("failed to listen on datapath of type %s: %s",
1217                  type, ovs_strerror(error));
1218         close_dpif_backer(backer);
1219         return error;
1220     }
1221     udpif_recv_set(backer->udpif, n_handler_threads,
1222                    backer->recv_set_enable);
1223     backer->n_handler_threads = n_handler_threads;
1224
1225     backer->max_n_subfacet = 0;
1226     backer->avg_n_subfacet = 0;
1227     backer->avg_subfacet_life = 0;
1228
1229     return error;
1230 }
1231
1232 static int
1233 construct(struct ofproto *ofproto_)
1234 {
1235     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1236     struct shash_node *node, *next;
1237     int error;
1238
1239     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1240     if (error) {
1241         return error;
1242     }
1243
1244     ofproto->netflow = NULL;
1245     ofproto->sflow = NULL;
1246     ofproto->ipfix = NULL;
1247     ofproto->stp = NULL;
1248     hmap_init(&ofproto->bundles);
1249     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1250     ofproto->mbridge = mbridge_create();
1251     ofproto->has_bonded_bundles = false;
1252     ofproto->lacp_enabled = false;
1253     ovs_mutex_init(&ofproto->stats_mutex);
1254     ovs_mutex_init(&ofproto->vsp_mutex);
1255
1256     classifier_init(&ofproto->facets, NULL);
1257     ofproto->consistency_rl = LLONG_MIN;
1258
1259     guarded_list_init(&ofproto->pins);
1260
1261     ofproto_dpif_unixctl_init();
1262
1263     hmap_init(&ofproto->vlandev_map);
1264     hmap_init(&ofproto->realdev_vid_map);
1265
1266     sset_init(&ofproto->ports);
1267     sset_init(&ofproto->ghost_ports);
1268     sset_init(&ofproto->port_poll_set);
1269     ofproto->port_poll_errno = 0;
1270     ofproto->change_seq = 0;
1271
1272     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1273         struct iface_hint *iface_hint = node->data;
1274
1275         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1276             /* Check if the datapath already has this port. */
1277             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1278                 sset_add(&ofproto->ports, node->name);
1279             }
1280
1281             free(iface_hint->br_name);
1282             free(iface_hint->br_type);
1283             free(iface_hint);
1284             shash_delete(&init_ofp_ports, node);
1285         }
1286     }
1287
1288     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1289                 hash_string(ofproto->up.name, 0));
1290     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1291
1292     ofproto_init_tables(ofproto_, N_TABLES);
1293     error = add_internal_flows(ofproto);
1294     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1295
1296     ofproto->n_hit = 0;
1297     ofproto->n_missed = 0;
1298
1299     return error;
1300 }
1301
1302 static int
1303 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1304                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1305 {
1306     struct ofputil_flow_mod fm;
1307     int error;
1308
1309     match_init_catchall(&fm.match);
1310     fm.priority = 0;
1311     match_set_reg(&fm.match, 0, id);
1312     fm.new_cookie = htonll(0);
1313     fm.cookie = htonll(0);
1314     fm.cookie_mask = htonll(0);
1315     fm.modify_cookie = false;
1316     fm.table_id = TBL_INTERNAL;
1317     fm.command = OFPFC_ADD;
1318     fm.idle_timeout = 0;
1319     fm.hard_timeout = 0;
1320     fm.buffer_id = 0;
1321     fm.out_port = 0;
1322     fm.flags = 0;
1323     fm.ofpacts = ofpacts->data;
1324     fm.ofpacts_len = ofpacts->size;
1325
1326     error = ofproto_flow_mod(&ofproto->up, &fm);
1327     if (error) {
1328         VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1329                     id, ofperr_to_string(error));
1330         return error;
1331     }
1332
1333     if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1334                                   rulep)) {
1335         rule_dpif_unref(*rulep);
1336     } else {
1337         NOT_REACHED();
1338     }
1339
1340     return 0;
1341 }
1342
1343 static int
1344 add_internal_flows(struct ofproto_dpif *ofproto)
1345 {
1346     struct ofpact_controller *controller;
1347     uint64_t ofpacts_stub[128 / 8];
1348     struct ofpbuf ofpacts;
1349     int error;
1350     int id;
1351
1352     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1353     id = 1;
1354
1355     controller = ofpact_put_CONTROLLER(&ofpacts);
1356     controller->max_len = UINT16_MAX;
1357     controller->controller_id = 0;
1358     controller->reason = OFPR_NO_MATCH;
1359     ofpact_pad(&ofpacts);
1360
1361     error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1362     if (error) {
1363         return error;
1364     }
1365
1366     ofpbuf_clear(&ofpacts);
1367     error = add_internal_flow(ofproto, id++, &ofpacts,
1368                               &ofproto->no_packet_in_rule);
1369     if (error) {
1370         return error;
1371     }
1372
1373     error = add_internal_flow(ofproto, id++, &ofpacts,
1374                               &ofproto->drop_frags_rule);
1375     return error;
1376 }
1377
1378 static void
1379 destruct(struct ofproto *ofproto_)
1380 {
1381     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1382     struct rule_dpif *rule, *next_rule;
1383     struct ofproto_packet_in *pin, *next_pin;
1384     struct facet *facet, *next_facet;
1385     struct cls_cursor cursor;
1386     struct oftable *table;
1387     struct list pins;
1388
1389     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1390     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1391     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1392     CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1393         facet_remove(facet);
1394     }
1395
1396     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1397     ovs_rwlock_wrlock(&xlate_rwlock);
1398     xlate_remove_ofproto(ofproto);
1399     ovs_rwlock_unlock(&xlate_rwlock);
1400
1401     /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a
1402      * use-after-free error. */
1403     udpif_revalidate(ofproto->backer->udpif);
1404
1405     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1406
1407     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1408         struct cls_cursor cursor;
1409
1410         ovs_rwlock_rdlock(&table->cls.rwlock);
1411         cls_cursor_init(&cursor, &table->cls, NULL);
1412         ovs_rwlock_unlock(&table->cls.rwlock);
1413         CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1414             ofproto_rule_delete(&ofproto->up, &rule->up);
1415         }
1416     }
1417
1418     guarded_list_pop_all(&ofproto->pins, &pins);
1419     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1420         list_remove(&pin->list_node);
1421         free(CONST_CAST(void *, pin->up.packet));
1422         free(pin);
1423     }
1424     guarded_list_destroy(&ofproto->pins);
1425
1426     mbridge_unref(ofproto->mbridge);
1427
1428     netflow_destroy(ofproto->netflow);
1429     dpif_sflow_unref(ofproto->sflow);
1430     hmap_destroy(&ofproto->bundles);
1431     mac_learning_unref(ofproto->ml);
1432
1433     classifier_destroy(&ofproto->facets);
1434
1435     hmap_destroy(&ofproto->vlandev_map);
1436     hmap_destroy(&ofproto->realdev_vid_map);
1437
1438     sset_destroy(&ofproto->ports);
1439     sset_destroy(&ofproto->ghost_ports);
1440     sset_destroy(&ofproto->port_poll_set);
1441
1442     ovs_mutex_destroy(&ofproto->stats_mutex);
1443     ovs_mutex_destroy(&ofproto->vsp_mutex);
1444
1445     close_dpif_backer(ofproto->backer);
1446 }
1447
1448 static int
1449 run_fast(struct ofproto *ofproto_)
1450 {
1451     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1452     struct ofproto_packet_in *pin, *next_pin;
1453     struct list pins;
1454
1455     /* Do not perform any periodic activity required by 'ofproto' while
1456      * waiting for flow restore to complete. */
1457     if (ofproto_get_flow_restore_wait()) {
1458         return 0;
1459     }
1460
1461     guarded_list_pop_all(&ofproto->pins, &pins);
1462     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1463         connmgr_send_packet_in(ofproto->up.connmgr, pin);
1464         list_remove(&pin->list_node);
1465         free(CONST_CAST(void *, pin->up.packet));
1466         free(pin);
1467     }
1468
1469     return 0;
1470 }
1471
1472 static int
1473 run(struct ofproto *ofproto_)
1474 {
1475     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1476     uint64_t new_seq;
1477     int error;
1478
1479     if (mbridge_need_revalidate(ofproto->mbridge)) {
1480         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1481         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1482         mac_learning_flush(ofproto->ml);
1483         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1484     }
1485
1486     /* Do not perform any periodic activity below required by 'ofproto' while
1487      * waiting for flow restore to complete. */
1488     if (ofproto_get_flow_restore_wait()) {
1489         return 0;
1490     }
1491
1492     error = run_fast(ofproto_);
1493     if (error) {
1494         return error;
1495     }
1496
1497     if (ofproto->netflow) {
1498         if (netflow_run(ofproto->netflow)) {
1499             send_netflow_active_timeouts(ofproto);
1500         }
1501     }
1502     if (ofproto->sflow) {
1503         dpif_sflow_run(ofproto->sflow);
1504     }
1505     if (ofproto->ipfix) {
1506         dpif_ipfix_run(ofproto->ipfix);
1507     }
1508
1509     new_seq = seq_read(connectivity_seq_get());
1510     if (ofproto->change_seq != new_seq) {
1511         struct ofport_dpif *ofport;
1512
1513         HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1514             port_run(ofport);
1515         }
1516
1517         ofproto->change_seq = new_seq;
1518     }
1519     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1520         struct ofbundle *bundle;
1521
1522         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1523             bundle_run(bundle);
1524         }
1525     }
1526
1527     stp_run(ofproto);
1528     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1529     if (mac_learning_run(ofproto->ml)) {
1530         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1531     }
1532     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1533
1534     /* Check the consistency of a random facet, to aid debugging. */
1535     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1536     if (time_msec() >= ofproto->consistency_rl
1537         && !classifier_is_empty(&ofproto->facets)
1538         && !ofproto->backer->need_revalidate) {
1539         struct cls_subtable *table;
1540         struct cls_rule *cr;
1541         struct facet *facet;
1542
1543         ofproto->consistency_rl = time_msec() + 250;
1544
1545         table = CONTAINER_OF(hmap_random_node(&ofproto->facets.subtables),
1546                              struct cls_subtable, hmap_node);
1547         cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1548                           hmap_node);
1549         facet = CONTAINER_OF(cr, struct facet, cr);
1550
1551         if (!facet_check_consistency(facet)) {
1552             ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1553         }
1554     }
1555     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1556
1557     return 0;
1558 }
1559
1560 static void
1561 wait(struct ofproto *ofproto_)
1562 {
1563     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1564
1565     if (ofproto_get_flow_restore_wait()) {
1566         return;
1567     }
1568
1569     if (ofproto->sflow) {
1570         dpif_sflow_wait(ofproto->sflow);
1571     }
1572     if (ofproto->ipfix) {
1573         dpif_ipfix_wait(ofproto->ipfix);
1574     }
1575     if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) {
1576         struct ofbundle *bundle;
1577
1578         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1579             bundle_wait(bundle);
1580         }
1581     }
1582     if (ofproto->netflow) {
1583         netflow_wait(ofproto->netflow);
1584     }
1585     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1586     mac_learning_wait(ofproto->ml);
1587     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1588     stp_wait(ofproto);
1589     if (ofproto->backer->need_revalidate) {
1590         /* Shouldn't happen, but if it does just go around again. */
1591         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1592         poll_immediate_wake();
1593     }
1594 }
1595
1596 static void
1597 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1598 {
1599     const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1600     struct cls_cursor cursor;
1601     size_t n_subfacets = 0;
1602     struct facet *facet;
1603
1604     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1605     simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1606     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1607
1608     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1609     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1610     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1611         n_subfacets += list_size(&facet->subfacets);
1612     }
1613     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1614     simap_increase(usage, "subfacets", n_subfacets);
1615 }
1616
1617 static void
1618 flush(struct ofproto *ofproto_)
1619 {
1620     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1621     struct subfacet *subfacet, *next_subfacet;
1622     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1623     int n_batch;
1624
1625     n_batch = 0;
1626     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1627                         &ofproto->backer->subfacets) {
1628         if (subfacet->facet->ofproto != ofproto) {
1629             continue;
1630         }
1631
1632         if (subfacet->path != SF_NOT_INSTALLED) {
1633             batch[n_batch++] = subfacet;
1634             if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1635                 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1636                 n_batch = 0;
1637             }
1638         } else {
1639             subfacet_destroy(subfacet);
1640         }
1641     }
1642
1643     if (n_batch > 0) {
1644         subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1645     }
1646 }
1647
1648 static void
1649 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1650              bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1651 {
1652     *arp_match_ip = true;
1653     *actions = (OFPUTIL_A_OUTPUT |
1654                 OFPUTIL_A_SET_VLAN_VID |
1655                 OFPUTIL_A_SET_VLAN_PCP |
1656                 OFPUTIL_A_STRIP_VLAN |
1657                 OFPUTIL_A_SET_DL_SRC |
1658                 OFPUTIL_A_SET_DL_DST |
1659                 OFPUTIL_A_SET_NW_SRC |
1660                 OFPUTIL_A_SET_NW_DST |
1661                 OFPUTIL_A_SET_NW_TOS |
1662                 OFPUTIL_A_SET_TP_SRC |
1663                 OFPUTIL_A_SET_TP_DST |
1664                 OFPUTIL_A_ENQUEUE);
1665 }
1666
1667 static void
1668 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1669 {
1670     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1671     struct dpif_dp_stats s;
1672     uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1673     uint64_t n_lookup;
1674
1675     strcpy(ots->name, "classifier");
1676
1677     dpif_get_dp_stats(ofproto->backer->dpif, &s);
1678     rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1679     rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1680     rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1681
1682     n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1683     ots->lookup_count = htonll(n_lookup);
1684     ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1685 }
1686
1687 static struct ofport *
1688 port_alloc(void)
1689 {
1690     struct ofport_dpif *port = xmalloc(sizeof *port);
1691     return &port->up;
1692 }
1693
1694 static void
1695 port_dealloc(struct ofport *port_)
1696 {
1697     struct ofport_dpif *port = ofport_dpif_cast(port_);
1698     free(port);
1699 }
1700
1701 static int
1702 port_construct(struct ofport *port_)
1703 {
1704     struct ofport_dpif *port = ofport_dpif_cast(port_);
1705     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1706     const struct netdev *netdev = port->up.netdev;
1707     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1708     struct dpif_port dpif_port;
1709     int error;
1710
1711     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1712     port->bundle = NULL;
1713     port->cfm = NULL;
1714     port->bfd = NULL;
1715     port->may_enable = true;
1716     port->stp_port = NULL;
1717     port->stp_state = STP_DISABLED;
1718     port->is_tunnel = false;
1719     port->peer = NULL;
1720     port->qdscp = NULL;
1721     port->n_qdscp = 0;
1722     port->realdev_ofp_port = 0;
1723     port->vlandev_vid = 0;
1724     port->carrier_seq = netdev_get_carrier_resets(netdev);
1725     port->is_layer3 = netdev_vport_is_layer3(netdev);
1726
1727     if (netdev_vport_is_patch(netdev)) {
1728         /* By bailing out here, we don't submit the port to the sFlow module
1729          * to be considered for counter polling export.  This is correct
1730          * because the patch port represents an interface that sFlow considers
1731          * to be "internal" to the switch as a whole, and therefore not an
1732          * candidate for counter polling. */
1733         port->odp_port = ODPP_NONE;
1734         ofport_update_peer(port);
1735         return 0;
1736     }
1737
1738     error = dpif_port_query_by_name(ofproto->backer->dpif,
1739                                     netdev_vport_get_dpif_port(netdev, namebuf,
1740                                                                sizeof namebuf),
1741                                     &dpif_port);
1742     if (error) {
1743         return error;
1744     }
1745
1746     port->odp_port = dpif_port.port_no;
1747
1748     if (netdev_get_tunnel_config(netdev)) {
1749         tnl_port_add(port, port->up.netdev, port->odp_port);
1750         port->is_tunnel = true;
1751     } else {
1752         /* Sanity-check that a mapping doesn't already exist.  This
1753          * shouldn't happen for non-tunnel ports. */
1754         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1755             VLOG_ERR("port %s already has an OpenFlow port number",
1756                      dpif_port.name);
1757             dpif_port_destroy(&dpif_port);
1758             return EBUSY;
1759         }
1760
1761         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1762         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1763                     hash_odp_port(port->odp_port));
1764         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1765     }
1766     dpif_port_destroy(&dpif_port);
1767
1768     if (ofproto->sflow) {
1769         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1770     }
1771
1772     return 0;
1773 }
1774
1775 static void
1776 port_destruct(struct ofport *port_)
1777 {
1778     struct ofport_dpif *port = ofport_dpif_cast(port_);
1779     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1780     const char *devname = netdev_get_name(port->up.netdev);
1781     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1782     const char *dp_port_name;
1783
1784     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1785     ovs_rwlock_wrlock(&xlate_rwlock);
1786     xlate_ofport_remove(port);
1787     ovs_rwlock_unlock(&xlate_rwlock);
1788
1789     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1790                                               sizeof namebuf);
1791     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1792         /* The underlying device is still there, so delete it.  This
1793          * happens when the ofproto is being destroyed, since the caller
1794          * assumes that removal of attached ports will happen as part of
1795          * destruction. */
1796         if (!port->is_tunnel) {
1797             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1798         }
1799     }
1800
1801     if (port->peer) {
1802         port->peer->peer = NULL;
1803         port->peer = NULL;
1804     }
1805
1806     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1807         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1808         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1809         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1810     }
1811
1812     tnl_port_del(port);
1813     sset_find_and_delete(&ofproto->ports, devname);
1814     sset_find_and_delete(&ofproto->ghost_ports, devname);
1815     bundle_remove(port_);
1816     set_cfm(port_, NULL);
1817     set_bfd(port_, NULL);
1818     if (ofproto->sflow) {
1819         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1820     }
1821
1822     free(port->qdscp);
1823 }
1824
1825 static void
1826 port_modified(struct ofport *port_)
1827 {
1828     struct ofport_dpif *port = ofport_dpif_cast(port_);
1829
1830     if (port->bundle && port->bundle->bond) {
1831         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1832     }
1833
1834     if (port->cfm) {
1835         cfm_set_netdev(port->cfm, port->up.netdev);
1836     }
1837
1838     if (port->bfd) {
1839         bfd_set_netdev(port->bfd, port->up.netdev);
1840     }
1841
1842     ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm,
1843                                      port->up.pp.hw_addr);
1844
1845     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1846                                                 port->odp_port)) {
1847         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1848             REV_RECONFIGURE;
1849     }
1850
1851     ofport_update_peer(port);
1852 }
1853
1854 static void
1855 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1856 {
1857     struct ofport_dpif *port = ofport_dpif_cast(port_);
1858     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1859     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1860
1861     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1862                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1863                    OFPUTIL_PC_NO_PACKET_IN)) {
1864         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1865
1866         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1867             bundle_update(port->bundle);
1868         }
1869     }
1870 }
1871
1872 static int
1873 set_sflow(struct ofproto *ofproto_,
1874           const struct ofproto_sflow_options *sflow_options)
1875 {
1876     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1877     struct dpif_sflow *ds = ofproto->sflow;
1878
1879     if (sflow_options) {
1880         if (!ds) {
1881             struct ofport_dpif *ofport;
1882
1883             ds = ofproto->sflow = dpif_sflow_create();
1884             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1885                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1886             }
1887             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1888         }
1889         dpif_sflow_set_options(ds, sflow_options);
1890     } else {
1891         if (ds) {
1892             dpif_sflow_unref(ds);
1893             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1894             ofproto->sflow = NULL;
1895         }
1896     }
1897     return 0;
1898 }
1899
1900 static int
1901 set_ipfix(
1902     struct ofproto *ofproto_,
1903     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1904     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1905     size_t n_flow_exporters_options)
1906 {
1907     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1908     struct dpif_ipfix *di = ofproto->ipfix;
1909     bool has_options = bridge_exporter_options || flow_exporters_options;
1910
1911     if (has_options && !di) {
1912         di = ofproto->ipfix = dpif_ipfix_create();
1913     }
1914
1915     if (di) {
1916         /* Call set_options in any case to cleanly flush the flow
1917          * caches in the last exporters that are to be destroyed. */
1918         dpif_ipfix_set_options(
1919             di, bridge_exporter_options, flow_exporters_options,
1920             n_flow_exporters_options);
1921
1922         if (!has_options) {
1923             dpif_ipfix_unref(di);
1924             ofproto->ipfix = NULL;
1925         }
1926     }
1927
1928     return 0;
1929 }
1930
1931 static int
1932 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1933 {
1934     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1935     int error = 0;
1936
1937     if (s) {
1938         if (!ofport->cfm) {
1939             struct ofproto_dpif *ofproto;
1940
1941             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1942             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1943             ofport->cfm = cfm_create(ofport->up.netdev);
1944         }
1945
1946         if (cfm_configure(ofport->cfm, s)) {
1947             error = 0;
1948             goto out;
1949         }
1950
1951         error = EINVAL;
1952     }
1953     cfm_unref(ofport->cfm);
1954     ofport->cfm = NULL;
1955 out:
1956     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1957                                      ofport->up.pp.hw_addr);
1958     return error;
1959 }
1960
1961 static bool
1962 get_cfm_status(const struct ofport *ofport_,
1963                struct ofproto_cfm_status *status)
1964 {
1965     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1966
1967     if (ofport->cfm) {
1968         status->faults = cfm_get_fault(ofport->cfm);
1969         status->flap_count = cfm_get_flap_count(ofport->cfm);
1970         status->remote_opstate = cfm_get_opup(ofport->cfm);
1971         status->health = cfm_get_health(ofport->cfm);
1972         cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1973         return true;
1974     } else {
1975         return false;
1976     }
1977 }
1978
1979 static int
1980 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1981 {
1982     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1983     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1984     struct bfd *old;
1985
1986     old = ofport->bfd;
1987     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1988                                 cfg, ofport->up.netdev);
1989     if (ofport->bfd != old) {
1990         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1991     }
1992     ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm,
1993                                      ofport->up.pp.hw_addr);
1994     return 0;
1995 }
1996
1997 static int
1998 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1999 {
2000     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2001
2002     if (ofport->bfd) {
2003         bfd_get_status(ofport->bfd, smap);
2004         return 0;
2005     } else {
2006         return ENOENT;
2007     }
2008 }
2009 \f
2010 /* Spanning Tree. */
2011
2012 static void
2013 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2014 {
2015     struct ofproto_dpif *ofproto = ofproto_;
2016     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2017     struct ofport_dpif *ofport;
2018
2019     ofport = stp_port_get_aux(sp);
2020     if (!ofport) {
2021         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2022                      ofproto->up.name, port_num);
2023     } else {
2024         struct eth_header *eth = pkt->l2;
2025
2026         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2027         if (eth_addr_is_zero(eth->eth_src)) {
2028             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2029                          "with unknown MAC", ofproto->up.name, port_num);
2030         } else {
2031             ofproto_dpif_send_packet(ofport, pkt);
2032         }
2033     }
2034     ofpbuf_delete(pkt);
2035 }
2036
2037 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2038 static int
2039 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2040 {
2041     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2042
2043     /* Only revalidate flows if the configuration changed. */
2044     if (!s != !ofproto->stp) {
2045         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2046     }
2047
2048     if (s) {
2049         if (!ofproto->stp) {
2050             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2051                                       send_bpdu_cb, ofproto);
2052             ofproto->stp_last_tick = time_msec();
2053         }
2054
2055         stp_set_bridge_id(ofproto->stp, s->system_id);
2056         stp_set_bridge_priority(ofproto->stp, s->priority);
2057         stp_set_hello_time(ofproto->stp, s->hello_time);
2058         stp_set_max_age(ofproto->stp, s->max_age);
2059         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2060     }  else {
2061         struct ofport *ofport;
2062
2063         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2064             set_stp_port(ofport, NULL);
2065         }
2066
2067         stp_unref(ofproto->stp);
2068         ofproto->stp = NULL;
2069     }
2070
2071     return 0;
2072 }
2073
2074 static int
2075 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2076 {
2077     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2078
2079     if (ofproto->stp) {
2080         s->enabled = true;
2081         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2082         s->designated_root = stp_get_designated_root(ofproto->stp);
2083         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2084     } else {
2085         s->enabled = false;
2086     }
2087
2088     return 0;
2089 }
2090
2091 static void
2092 update_stp_port_state(struct ofport_dpif *ofport)
2093 {
2094     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2095     enum stp_state state;
2096
2097     /* Figure out new state. */
2098     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2099                              : STP_DISABLED;
2100
2101     /* Update state. */
2102     if (ofport->stp_state != state) {
2103         enum ofputil_port_state of_state;
2104         bool fwd_change;
2105
2106         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2107                     netdev_get_name(ofport->up.netdev),
2108                     stp_state_name(ofport->stp_state),
2109                     stp_state_name(state));
2110         if (stp_learn_in_state(ofport->stp_state)
2111                 != stp_learn_in_state(state)) {
2112             /* xxx Learning action flows should also be flushed. */
2113             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2114             mac_learning_flush(ofproto->ml);
2115             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2116         }
2117         fwd_change = stp_forward_in_state(ofport->stp_state)
2118                         != stp_forward_in_state(state);
2119
2120         ofproto->backer->need_revalidate = REV_STP;
2121         ofport->stp_state = state;
2122         ofport->stp_state_entered = time_msec();
2123
2124         if (fwd_change && ofport->bundle) {
2125             bundle_update(ofport->bundle);
2126         }
2127
2128         /* Update the STP state bits in the OpenFlow port description. */
2129         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2130         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2131                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2132                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2133                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2134                      : 0);
2135         ofproto_port_set_state(&ofport->up, of_state);
2136     }
2137 }
2138
2139 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2140  * caller is responsible for assigning STP port numbers and ensuring
2141  * there are no duplicates. */
2142 static int
2143 set_stp_port(struct ofport *ofport_,
2144              const struct ofproto_port_stp_settings *s)
2145 {
2146     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2147     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2148     struct stp_port *sp = ofport->stp_port;
2149
2150     if (!s || !s->enable) {
2151         if (sp) {
2152             ofport->stp_port = NULL;
2153             stp_port_disable(sp);
2154             update_stp_port_state(ofport);
2155         }
2156         return 0;
2157     } else if (sp && stp_port_no(sp) != s->port_num
2158             && ofport == stp_port_get_aux(sp)) {
2159         /* The port-id changed, so disable the old one if it's not
2160          * already in use by another port. */
2161         stp_port_disable(sp);
2162     }
2163
2164     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2165     stp_port_enable(sp);
2166
2167     stp_port_set_aux(sp, ofport);
2168     stp_port_set_priority(sp, s->priority);
2169     stp_port_set_path_cost(sp, s->path_cost);
2170
2171     update_stp_port_state(ofport);
2172
2173     return 0;
2174 }
2175
2176 static int
2177 get_stp_port_status(struct ofport *ofport_,
2178                     struct ofproto_port_stp_status *s)
2179 {
2180     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2181     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2182     struct stp_port *sp = ofport->stp_port;
2183
2184     if (!ofproto->stp || !sp) {
2185         s->enabled = false;
2186         return 0;
2187     }
2188
2189     s->enabled = true;
2190     s->port_id = stp_port_get_id(sp);
2191     s->state = stp_port_get_state(sp);
2192     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2193     s->role = stp_port_get_role(sp);
2194
2195     return 0;
2196 }
2197
2198 static int
2199 get_stp_port_stats(struct ofport *ofport_,
2200                    struct ofproto_port_stp_stats *s)
2201 {
2202     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2203     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2204     struct stp_port *sp = ofport->stp_port;
2205
2206     if (!ofproto->stp || !sp) {
2207         s->enabled = false;
2208         return 0;
2209     }
2210
2211     s->enabled = true;
2212     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2213
2214     return 0;
2215 }
2216
2217 static void
2218 stp_run(struct ofproto_dpif *ofproto)
2219 {
2220     if (ofproto->stp) {
2221         long long int now = time_msec();
2222         long long int elapsed = now - ofproto->stp_last_tick;
2223         struct stp_port *sp;
2224
2225         if (elapsed > 0) {
2226             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2227             ofproto->stp_last_tick = now;
2228         }
2229         while (stp_get_changed_port(ofproto->stp, &sp)) {
2230             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2231
2232             if (ofport) {
2233                 update_stp_port_state(ofport);
2234             }
2235         }
2236
2237         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2238             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2239             mac_learning_flush(ofproto->ml);
2240             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2241         }
2242     }
2243 }
2244
2245 static void
2246 stp_wait(struct ofproto_dpif *ofproto)
2247 {
2248     if (ofproto->stp) {
2249         poll_timer_wait(1000);
2250     }
2251 }
2252 \f
2253 static int
2254 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2255            size_t n_qdscp)
2256 {
2257     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2258     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2259
2260     if (ofport->n_qdscp != n_qdscp
2261         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2262                               n_qdscp * sizeof *qdscp))) {
2263         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2264         free(ofport->qdscp);
2265         ofport->qdscp = n_qdscp
2266             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2267             : NULL;
2268         ofport->n_qdscp = n_qdscp;
2269     }
2270
2271     return 0;
2272 }
2273 \f
2274 /* Bundles. */
2275
2276 /* Expires all MAC learning entries associated with 'bundle' and forces its
2277  * ofproto to revalidate every flow.
2278  *
2279  * Normally MAC learning entries are removed only from the ofproto associated
2280  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2281  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2282  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2283  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2284  * with the host from which it migrated. */
2285 static void
2286 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2287 {
2288     struct ofproto_dpif *ofproto = bundle->ofproto;
2289     struct mac_learning *ml = ofproto->ml;
2290     struct mac_entry *mac, *next_mac;
2291
2292     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2293     ovs_rwlock_wrlock(&ml->rwlock);
2294     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2295         if (mac->port.p == bundle) {
2296             if (all_ofprotos) {
2297                 struct ofproto_dpif *o;
2298
2299                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2300                     if (o != ofproto) {
2301                         struct mac_entry *e;
2302
2303                         ovs_rwlock_wrlock(&o->ml->rwlock);
2304                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2305                         if (e) {
2306                             mac_learning_expire(o->ml, e);
2307                         }
2308                         ovs_rwlock_unlock(&o->ml->rwlock);
2309                     }
2310                 }
2311             }
2312
2313             mac_learning_expire(ml, mac);
2314         }
2315     }
2316     ovs_rwlock_unlock(&ml->rwlock);
2317 }
2318
2319 static struct ofbundle *
2320 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2321 {
2322     struct ofbundle *bundle;
2323
2324     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2325                              &ofproto->bundles) {
2326         if (bundle->aux == aux) {
2327             return bundle;
2328         }
2329     }
2330     return NULL;
2331 }
2332
2333 static void
2334 bundle_update(struct ofbundle *bundle)
2335 {
2336     struct ofport_dpif *port;
2337
2338     bundle->floodable = true;
2339     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2340         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2341             || port->is_layer3
2342             || !stp_forward_in_state(port->stp_state)) {
2343             bundle->floodable = false;
2344             break;
2345         }
2346     }
2347 }
2348
2349 static void
2350 bundle_del_port(struct ofport_dpif *port)
2351 {
2352     struct ofbundle *bundle = port->bundle;
2353
2354     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2355
2356     list_remove(&port->bundle_node);
2357     port->bundle = NULL;
2358
2359     if (bundle->lacp) {
2360         lacp_slave_unregister(bundle->lacp, port);
2361     }
2362     if (bundle->bond) {
2363         bond_slave_unregister(bundle->bond, port);
2364     }
2365
2366     bundle_update(bundle);
2367 }
2368
2369 static bool
2370 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2371                 struct lacp_slave_settings *lacp)
2372 {
2373     struct ofport_dpif *port;
2374
2375     port = get_ofp_port(bundle->ofproto, ofp_port);
2376     if (!port) {
2377         return false;
2378     }
2379
2380     if (port->bundle != bundle) {
2381         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2382         if (port->bundle) {
2383             bundle_remove(&port->up);
2384         }
2385
2386         port->bundle = bundle;
2387         list_push_back(&bundle->ports, &port->bundle_node);
2388         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2389             || port->is_layer3
2390             || !stp_forward_in_state(port->stp_state)) {
2391             bundle->floodable = false;
2392         }
2393     }
2394     if (lacp) {
2395         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2396         lacp_slave_register(bundle->lacp, port, lacp);
2397     }
2398
2399     return true;
2400 }
2401
2402 static void
2403 bundle_destroy(struct ofbundle *bundle)
2404 {
2405     struct ofproto_dpif *ofproto;
2406     struct ofport_dpif *port, *next_port;
2407
2408     if (!bundle) {
2409         return;
2410     }
2411
2412     ofproto = bundle->ofproto;
2413     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2414
2415     ovs_rwlock_wrlock(&xlate_rwlock);
2416     xlate_bundle_remove(bundle);
2417     ovs_rwlock_unlock(&xlate_rwlock);
2418
2419     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2420         bundle_del_port(port);
2421     }
2422
2423     bundle_flush_macs(bundle, true);
2424     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2425     free(bundle->name);
2426     free(bundle->trunks);
2427     lacp_unref(bundle->lacp);
2428     bond_unref(bundle->bond);
2429     free(bundle);
2430 }
2431
2432 static int
2433 bundle_set(struct ofproto *ofproto_, void *aux,
2434            const struct ofproto_bundle_settings *s)
2435 {
2436     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2437     bool need_flush = false;
2438     struct ofport_dpif *port;
2439     struct ofbundle *bundle;
2440     unsigned long *trunks;
2441     int vlan;
2442     size_t i;
2443     bool ok;
2444
2445     if (!s) {
2446         bundle_destroy(bundle_lookup(ofproto, aux));
2447         return 0;
2448     }
2449
2450     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2451     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2452
2453     bundle = bundle_lookup(ofproto, aux);
2454     if (!bundle) {
2455         bundle = xmalloc(sizeof *bundle);
2456
2457         bundle->ofproto = ofproto;
2458         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2459                     hash_pointer(aux, 0));
2460         bundle->aux = aux;
2461         bundle->name = NULL;
2462
2463         list_init(&bundle->ports);
2464         bundle->vlan_mode = PORT_VLAN_TRUNK;
2465         bundle->vlan = -1;
2466         bundle->trunks = NULL;
2467         bundle->use_priority_tags = s->use_priority_tags;
2468         bundle->lacp = NULL;
2469         bundle->bond = NULL;
2470
2471         bundle->floodable = true;
2472         mbridge_register_bundle(ofproto->mbridge, bundle);
2473     }
2474
2475     if (!bundle->name || strcmp(s->name, bundle->name)) {
2476         free(bundle->name);
2477         bundle->name = xstrdup(s->name);
2478     }
2479
2480     /* LACP. */
2481     if (s->lacp) {
2482         ofproto->lacp_enabled = true;
2483         if (!bundle->lacp) {
2484             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2485             bundle->lacp = lacp_create();
2486         }
2487         lacp_configure(bundle->lacp, s->lacp);
2488     } else {
2489         lacp_unref(bundle->lacp);
2490         bundle->lacp = NULL;
2491     }
2492
2493     /* Update set of ports. */
2494     ok = true;
2495     for (i = 0; i < s->n_slaves; i++) {
2496         if (!bundle_add_port(bundle, s->slaves[i],
2497                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2498             ok = false;
2499         }
2500     }
2501     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2502         struct ofport_dpif *next_port;
2503
2504         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2505             for (i = 0; i < s->n_slaves; i++) {
2506                 if (s->slaves[i] == port->up.ofp_port) {
2507                     goto found;
2508                 }
2509             }
2510
2511             bundle_del_port(port);
2512         found: ;
2513         }
2514     }
2515     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2516
2517     if (list_is_empty(&bundle->ports)) {
2518         bundle_destroy(bundle);
2519         return EINVAL;
2520     }
2521
2522     /* Set VLAN tagging mode */
2523     if (s->vlan_mode != bundle->vlan_mode
2524         || s->use_priority_tags != bundle->use_priority_tags) {
2525         bundle->vlan_mode = s->vlan_mode;
2526         bundle->use_priority_tags = s->use_priority_tags;
2527         need_flush = true;
2528     }
2529
2530     /* Set VLAN tag. */
2531     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2532             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2533             : 0);
2534     if (vlan != bundle->vlan) {
2535         bundle->vlan = vlan;
2536         need_flush = true;
2537     }
2538
2539     /* Get trunked VLANs. */
2540     switch (s->vlan_mode) {
2541     case PORT_VLAN_ACCESS:
2542         trunks = NULL;
2543         break;
2544
2545     case PORT_VLAN_TRUNK:
2546         trunks = CONST_CAST(unsigned long *, s->trunks);
2547         break;
2548
2549     case PORT_VLAN_NATIVE_UNTAGGED:
2550     case PORT_VLAN_NATIVE_TAGGED:
2551         if (vlan != 0 && (!s->trunks
2552                           || !bitmap_is_set(s->trunks, vlan)
2553                           || bitmap_is_set(s->trunks, 0))) {
2554             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2555             if (s->trunks) {
2556                 trunks = bitmap_clone(s->trunks, 4096);
2557             } else {
2558                 trunks = bitmap_allocate1(4096);
2559             }
2560             bitmap_set1(trunks, vlan);
2561             bitmap_set0(trunks, 0);
2562         } else {
2563             trunks = CONST_CAST(unsigned long *, s->trunks);
2564         }
2565         break;
2566
2567     default:
2568         NOT_REACHED();
2569     }
2570     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2571         free(bundle->trunks);
2572         if (trunks == s->trunks) {
2573             bundle->trunks = vlan_bitmap_clone(trunks);
2574         } else {
2575             bundle->trunks = trunks;
2576             trunks = NULL;
2577         }
2578         need_flush = true;
2579     }
2580     if (trunks != s->trunks) {
2581         free(trunks);
2582     }
2583
2584     /* Bonding. */
2585     if (!list_is_short(&bundle->ports)) {
2586         bundle->ofproto->has_bonded_bundles = true;
2587         if (bundle->bond) {
2588             if (bond_reconfigure(bundle->bond, s->bond)) {
2589                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2590             }
2591         } else {
2592             bundle->bond = bond_create(s->bond);
2593             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2594         }
2595
2596         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2597             bond_slave_register(bundle->bond, port, port->up.netdev);
2598         }
2599     } else {
2600         bond_unref(bundle->bond);
2601         bundle->bond = NULL;
2602     }
2603
2604     /* If we changed something that would affect MAC learning, un-learn
2605      * everything on this port and force flow revalidation. */
2606     if (need_flush) {
2607         bundle_flush_macs(bundle, false);
2608     }
2609
2610     return 0;
2611 }
2612
2613 static void
2614 bundle_remove(struct ofport *port_)
2615 {
2616     struct ofport_dpif *port = ofport_dpif_cast(port_);
2617     struct ofbundle *bundle = port->bundle;
2618
2619     if (bundle) {
2620         bundle_del_port(port);
2621         if (list_is_empty(&bundle->ports)) {
2622             bundle_destroy(bundle);
2623         } else if (list_is_short(&bundle->ports)) {
2624             bond_unref(bundle->bond);
2625             bundle->bond = NULL;
2626         }
2627     }
2628 }
2629
2630 static void
2631 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2632 {
2633     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2634     struct ofport_dpif *port = port_;
2635     uint8_t ea[ETH_ADDR_LEN];
2636     int error;
2637
2638     error = netdev_get_etheraddr(port->up.netdev, ea);
2639     if (!error) {
2640         struct ofpbuf packet;
2641         void *packet_pdu;
2642
2643         ofpbuf_init(&packet, 0);
2644         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2645                                  pdu_size);
2646         memcpy(packet_pdu, pdu, pdu_size);
2647
2648         ofproto_dpif_send_packet(port, &packet);
2649         ofpbuf_uninit(&packet);
2650     } else {
2651         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2652                     "%s (%s)", port->bundle->name,
2653                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2654     }
2655 }
2656
2657 static void
2658 bundle_send_learning_packets(struct ofbundle *bundle)
2659 {
2660     struct ofproto_dpif *ofproto = bundle->ofproto;
2661     struct ofpbuf *learning_packet;
2662     int error, n_packets, n_errors;
2663     struct mac_entry *e;
2664     struct list packets;
2665
2666     list_init(&packets);
2667     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2668     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2669         if (e->port.p != bundle) {
2670             void *port_void;
2671
2672             learning_packet = bond_compose_learning_packet(bundle->bond,
2673                                                            e->mac, e->vlan,
2674                                                            &port_void);
2675             learning_packet->private_p = port_void;
2676             list_push_back(&packets, &learning_packet->list_node);
2677         }
2678     }
2679     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2680
2681     error = n_packets = n_errors = 0;
2682     LIST_FOR_EACH (learning_packet, list_node, &packets) {
2683         int ret;
2684
2685         ret = ofproto_dpif_send_packet(learning_packet->private_p, learning_packet);
2686         if (ret) {
2687             error = ret;
2688             n_errors++;
2689         }
2690         n_packets++;
2691     }
2692     ofpbuf_list_delete(&packets);
2693
2694     if (n_errors) {
2695         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2696         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2697                      "packets, last error was: %s",
2698                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2699     } else {
2700         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2701                  bundle->name, n_packets);
2702     }
2703 }
2704
2705 static void
2706 bundle_run(struct ofbundle *bundle)
2707 {
2708     if (bundle->lacp) {
2709         lacp_run(bundle->lacp, send_pdu_cb);
2710     }
2711     if (bundle->bond) {
2712         struct ofport_dpif *port;
2713
2714         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2715             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2716         }
2717
2718         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2719             bundle->ofproto->backer->need_revalidate = REV_BOND;
2720         }
2721
2722         if (bond_should_send_learning_packets(bundle->bond)) {
2723             bundle_send_learning_packets(bundle);
2724         }
2725     }
2726 }
2727
2728 static void
2729 bundle_wait(struct ofbundle *bundle)
2730 {
2731     if (bundle->lacp) {
2732         lacp_wait(bundle->lacp);
2733     }
2734     if (bundle->bond) {
2735         bond_wait(bundle->bond);
2736     }
2737 }
2738 \f
2739 /* Mirrors. */
2740
2741 static int
2742 mirror_set__(struct ofproto *ofproto_, void *aux,
2743              const struct ofproto_mirror_settings *s)
2744 {
2745     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2746     struct ofbundle **srcs, **dsts;
2747     int error;
2748     size_t i;
2749
2750     if (!s) {
2751         mirror_destroy(ofproto->mbridge, aux);
2752         return 0;
2753     }
2754
2755     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2756     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2757
2758     for (i = 0; i < s->n_srcs; i++) {
2759         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2760     }
2761
2762     for (i = 0; i < s->n_dsts; i++) {
2763         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2764     }
2765
2766     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2767                        s->n_dsts, s->src_vlans,
2768                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2769     free(srcs);
2770     free(dsts);
2771     return error;
2772 }
2773
2774 static int
2775 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2776                    uint64_t *packets, uint64_t *bytes)
2777 {
2778     push_all_stats();
2779     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2780                             bytes);
2781 }
2782
2783 static int
2784 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2785 {
2786     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2787     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2788     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2789         mac_learning_flush(ofproto->ml);
2790     }
2791     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2792     return 0;
2793 }
2794
2795 static bool
2796 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2797 {
2798     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2799     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2800     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2801 }
2802
2803 static void
2804 forward_bpdu_changed(struct ofproto *ofproto_)
2805 {
2806     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2807     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2808 }
2809
2810 static void
2811 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2812                      size_t max_entries)
2813 {
2814     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2815     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2816     mac_learning_set_idle_time(ofproto->ml, idle_time);
2817     mac_learning_set_max_entries(ofproto->ml, max_entries);
2818     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2819 }
2820 \f
2821 /* Ports. */
2822
2823 static struct ofport_dpif *
2824 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2825 {
2826     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2827     return ofport ? ofport_dpif_cast(ofport) : NULL;
2828 }
2829
2830 static struct ofport_dpif *
2831 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2832 {
2833     struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2834     return port && &ofproto->up == port->up.ofproto ? port : NULL;
2835 }
2836
2837 static void
2838 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2839                             struct ofproto_port *ofproto_port,
2840                             struct dpif_port *dpif_port)
2841 {
2842     ofproto_port->name = dpif_port->name;
2843     ofproto_port->type = dpif_port->type;
2844     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2845 }
2846
2847 static void
2848 ofport_update_peer(struct ofport_dpif *ofport)
2849 {
2850     const struct ofproto_dpif *ofproto;
2851     struct dpif_backer *backer;
2852     char *peer_name;
2853
2854     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2855         return;
2856     }
2857
2858     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2859     backer->need_revalidate = REV_RECONFIGURE;
2860
2861     if (ofport->peer) {
2862         ofport->peer->peer = NULL;
2863         ofport->peer = NULL;
2864     }
2865
2866     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2867     if (!peer_name) {
2868         return;
2869     }
2870
2871     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2872         struct ofport *peer_ofport;
2873         struct ofport_dpif *peer;
2874         char *peer_peer;
2875
2876         if (ofproto->backer != backer) {
2877             continue;
2878         }
2879
2880         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2881         if (!peer_ofport) {
2882             continue;
2883         }
2884
2885         peer = ofport_dpif_cast(peer_ofport);
2886         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2887         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2888                                  peer_peer)) {
2889             ofport->peer = peer;
2890             ofport->peer->peer = ofport;
2891         }
2892         free(peer_peer);
2893
2894         break;
2895     }
2896     free(peer_name);
2897 }
2898
2899 static void
2900 port_run(struct ofport_dpif *ofport)
2901 {
2902     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2903     bool carrier_changed = carrier_seq != ofport->carrier_seq;
2904     bool enable = netdev_get_carrier(ofport->up.netdev);
2905     bool cfm_enable = false;
2906     bool bfd_enable = false;
2907
2908     ofport->carrier_seq = carrier_seq;
2909
2910     if (ofport->cfm) {
2911         int cfm_opup = cfm_get_opup(ofport->cfm);
2912
2913         cfm_enable = !cfm_get_fault(ofport->cfm);
2914
2915         if (cfm_opup >= 0) {
2916             cfm_enable = cfm_enable && cfm_opup;
2917         }
2918     }
2919
2920     if (ofport->bfd) {
2921         bfd_enable = bfd_forwarding(ofport->bfd);
2922     }
2923
2924     if (ofport->bfd || ofport->cfm) {
2925         enable = enable && (cfm_enable || bfd_enable);
2926     }
2927
2928     if (ofport->bundle) {
2929         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2930         if (carrier_changed) {
2931             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2932         }
2933     }
2934
2935     if (ofport->may_enable != enable) {
2936         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2937         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2938     }
2939
2940     ofport->may_enable = enable;
2941 }
2942
2943 static int
2944 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2945                    struct ofproto_port *ofproto_port)
2946 {
2947     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2948     struct dpif_port dpif_port;
2949     int error;
2950
2951     if (sset_contains(&ofproto->ghost_ports, devname)) {
2952         const char *type = netdev_get_type_from_name(devname);
2953
2954         /* We may be called before ofproto->up.port_by_name is populated with
2955          * the appropriate ofport.  For this reason, we must get the name and
2956          * type from the netdev layer directly. */
2957         if (type) {
2958             const struct ofport *ofport;
2959
2960             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2961             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2962             ofproto_port->name = xstrdup(devname);
2963             ofproto_port->type = xstrdup(type);
2964             return 0;
2965         }
2966         return ENODEV;
2967     }
2968
2969     if (!sset_contains(&ofproto->ports, devname)) {
2970         return ENODEV;
2971     }
2972     error = dpif_port_query_by_name(ofproto->backer->dpif,
2973                                     devname, &dpif_port);
2974     if (!error) {
2975         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2976     }
2977     return error;
2978 }
2979
2980 static int
2981 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2982 {
2983     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2984     const char *devname = netdev_get_name(netdev);
2985     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2986     const char *dp_port_name;
2987
2988     if (netdev_vport_is_patch(netdev)) {
2989         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2990         return 0;
2991     }
2992
2993     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2994     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2995         odp_port_t port_no = ODPP_NONE;
2996         int error;
2997
2998         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2999         if (error) {
3000             return error;
3001         }
3002         if (netdev_get_tunnel_config(netdev)) {
3003             simap_put(&ofproto->backer->tnl_backers,
3004                       dp_port_name, odp_to_u32(port_no));
3005         }
3006     }
3007
3008     if (netdev_get_tunnel_config(netdev)) {
3009         sset_add(&ofproto->ghost_ports, devname);
3010     } else {
3011         sset_add(&ofproto->ports, devname);
3012     }
3013     return 0;
3014 }
3015
3016 static int
3017 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3018 {
3019     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3020     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3021     int error = 0;
3022
3023     if (!ofport) {
3024         return 0;
3025     }
3026
3027     sset_find_and_delete(&ofproto->ghost_ports,
3028                          netdev_get_name(ofport->up.netdev));
3029     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3030     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3031         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3032         if (!error) {
3033             /* The caller is going to close ofport->up.netdev.  If this is a
3034              * bonded port, then the bond is using that netdev, so remove it
3035              * from the bond.  The client will need to reconfigure everything
3036              * after deleting ports, so then the slave will get re-added. */
3037             bundle_remove(&ofport->up);
3038         }
3039     }
3040     return error;
3041 }
3042
3043 static int
3044 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3045 {
3046     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3047     int error;
3048
3049     push_all_stats();
3050
3051     error = netdev_get_stats(ofport->up.netdev, stats);
3052
3053     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3054         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3055
3056         ovs_mutex_lock(&ofproto->stats_mutex);
3057         /* ofproto->stats.tx_packets represents packets that we created
3058          * internally and sent to some port (e.g. packets sent with
3059          * ofproto_dpif_send_packet()).  Account for them as if they had
3060          * come from OFPP_LOCAL and got forwarded. */
3061
3062         if (stats->rx_packets != UINT64_MAX) {
3063             stats->rx_packets += ofproto->stats.tx_packets;
3064         }
3065
3066         if (stats->rx_bytes != UINT64_MAX) {
3067             stats->rx_bytes += ofproto->stats.tx_bytes;
3068         }
3069
3070         /* ofproto->stats.rx_packets represents packets that were received on
3071          * some port and we processed internally and dropped (e.g. STP).
3072          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3073
3074         if (stats->tx_packets != UINT64_MAX) {
3075             stats->tx_packets += ofproto->stats.rx_packets;
3076         }
3077
3078         if (stats->tx_bytes != UINT64_MAX) {
3079             stats->tx_bytes += ofproto->stats.rx_bytes;
3080         }
3081         ovs_mutex_unlock(&ofproto->stats_mutex);
3082     }
3083
3084     return error;
3085 }
3086
3087 struct port_dump_state {
3088     uint32_t bucket;
3089     uint32_t offset;
3090     bool ghost;
3091
3092     struct ofproto_port port;
3093     bool has_port;
3094 };
3095
3096 static int
3097 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3098 {
3099     *statep = xzalloc(sizeof(struct port_dump_state));
3100     return 0;
3101 }
3102
3103 static int
3104 port_dump_next(const struct ofproto *ofproto_, void *state_,
3105                struct ofproto_port *port)
3106 {
3107     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3108     struct port_dump_state *state = state_;
3109     const struct sset *sset;
3110     struct sset_node *node;
3111
3112     if (state->has_port) {
3113         ofproto_port_destroy(&state->port);
3114         state->has_port = false;
3115     }
3116     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3117     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3118         int error;
3119
3120         error = port_query_by_name(ofproto_, node->name, &state->port);
3121         if (!error) {
3122             *port = state->port;
3123             state->has_port = true;
3124             return 0;
3125         } else if (error != ENODEV) {
3126             return error;
3127         }
3128     }
3129
3130     if (!state->ghost) {
3131         state->ghost = true;
3132         state->bucket = 0;
3133         state->offset = 0;
3134         return port_dump_next(ofproto_, state_, port);
3135     }
3136
3137     return EOF;
3138 }
3139
3140 static int
3141 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3142 {
3143     struct port_dump_state *state = state_;
3144
3145     if (state->has_port) {
3146         ofproto_port_destroy(&state->port);
3147     }
3148     free(state);
3149     return 0;
3150 }
3151
3152 static int
3153 port_poll(const struct ofproto *ofproto_, char **devnamep)
3154 {
3155     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3156
3157     if (ofproto->port_poll_errno) {
3158         int error = ofproto->port_poll_errno;
3159         ofproto->port_poll_errno = 0;
3160         return error;
3161     }
3162
3163     if (sset_is_empty(&ofproto->port_poll_set)) {
3164         return EAGAIN;
3165     }
3166
3167     *devnamep = sset_pop(&ofproto->port_poll_set);
3168     return 0;
3169 }
3170
3171 static void
3172 port_poll_wait(const struct ofproto *ofproto_)
3173 {
3174     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3175     dpif_port_poll_wait(ofproto->backer->dpif);
3176 }
3177
3178 static int
3179 port_is_lacp_current(const struct ofport *ofport_)
3180 {
3181     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3182     return (ofport->bundle && ofport->bundle->lacp
3183             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3184             : -1);
3185 }
3186 \f
3187 /* Upcall handling. */
3188
3189 struct flow_miss_op {
3190     struct dpif_op dpif_op;
3191
3192     uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3193     struct xlate_out xout;
3194     bool xout_garbage;           /* 'xout' needs to be uninitialized? */
3195
3196     struct ofpbuf mask;          /* Flow mask for "put" ops. */
3197     struct odputil_keybuf maskbuf;
3198
3199     /* If this is a "put" op, then a pointer to the subfacet that should
3200      * be marked as uninstalled if the operation fails. */
3201     struct subfacet *subfacet;
3202 };
3203
3204 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3205  * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3206  * and (usually) installing a datapath flow.  The answer is usually "yes" (a
3207  * return value of true).  However, for short flows the cost of bookkeeping is
3208  * much higher than the benefits, so when the datapath holds a large number of
3209  * flows we impose some heuristics to decide which flows are likely to be worth
3210  * tracking. */
3211 static bool
3212 flow_miss_should_make_facet(struct flow_miss *miss)
3213 {
3214     struct dpif_backer *backer = miss->ofproto->backer;
3215     uint32_t hash;
3216
3217     switch (flow_miss_model) {
3218     case OFPROTO_HANDLE_MISS_AUTO:
3219         break;
3220     case OFPROTO_HANDLE_MISS_WITH_FACETS:
3221         return true;
3222     case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3223         return false;
3224     }
3225
3226     if (!backer->governor) {
3227         size_t n_subfacets;
3228
3229         n_subfacets = hmap_count(&backer->subfacets);
3230         if (n_subfacets * 2 <= flow_eviction_threshold) {
3231             return true;
3232         }
3233
3234         backer->governor = governor_create();
3235     }
3236
3237     hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3238     return governor_should_install_flow(backer->governor, hash,
3239                                         miss->stats.n_packets);
3240 }
3241
3242 /* Handles 'miss', which matches 'facet'.  May add any required datapath
3243  * operations to 'ops', incrementing '*n_ops' for each new op.
3244  *
3245  * All of the packets in 'miss' are considered to have arrived at time
3246  * 'miss->stats.used'.  This is really important only for new facets: if we
3247  * just called time_msec() here, then the new subfacet or its packets could
3248  * look (occasionally) as though it was used some time after the facet was
3249  * used.  That can make a one-packet flow look like it has a nonzero duration,
3250  * which looks odd in e.g. NetFlow statistics. */
3251 static void
3252 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3253                             struct flow_miss_op *ops, size_t *n_ops)
3254 {
3255     enum subfacet_path want_path;
3256     struct subfacet *subfacet;
3257     uint32_t key_hash;
3258
3259     /* Update facet stats. */
3260     facet->packet_count += miss->stats.n_packets;
3261     facet->prev_packet_count += miss->stats.n_packets;
3262     facet->byte_count += miss->stats.n_bytes;
3263     facet->prev_byte_count += miss->stats.n_bytes;
3264
3265     /* Look for an existing subfacet.  If we find one, update its used time. */
3266     key_hash = odp_flow_key_hash(miss->key, miss->key_len);
3267     if (!list_is_empty(&facet->subfacets)) {
3268         subfacet = subfacet_find(miss->ofproto->backer,
3269                                  miss->key, miss->key_len, key_hash);
3270         if (subfacet) {
3271             if (subfacet->facet == facet) {
3272                 subfacet->used = MAX(subfacet->used, miss->stats.used);
3273             } else {
3274                 /* This shouldn't happen. */
3275                 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
3276                 subfacet_destroy(subfacet);
3277                 subfacet = NULL;
3278             }
3279         }
3280     } else {
3281         subfacet = NULL;
3282     }
3283
3284     /* Don't install the flow if it's the result of the "userspace"
3285      * action for an already installed facet.  This can occur when a
3286      * datapath flow with wildcards has a "userspace" action and flows
3287      * sent to userspace result in a different subfacet, which will then
3288      * be rejected as overlapping by the datapath. */
3289     if (miss->upcall_type == DPIF_UC_ACTION
3290         && !list_is_empty(&facet->subfacets)) {
3291         return;
3292     }
3293
3294     /* Create a subfacet, if we don't already have one. */
3295     if (!subfacet) {
3296         subfacet = subfacet_create(facet, miss, key_hash);
3297     }
3298
3299     /* Install the subfacet, if it's not already installed. */
3300     want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3301     if (subfacet->path != want_path) {
3302         struct flow_miss_op *op = &ops[(*n_ops)++];
3303         struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3304
3305         subfacet->path = want_path;
3306
3307         ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3308         if (enable_megaflows) {
3309             odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3310                                    &miss->flow, UINT32_MAX);
3311         }
3312
3313         op->xout_garbage = false;
3314         op->dpif_op.type = DPIF_OP_FLOW_PUT;
3315         op->subfacet = subfacet;
3316         put->flags = DPIF_FP_CREATE;
3317         put->key = miss->key;
3318         put->key_len = miss->key_len;
3319         put->mask = op->mask.data;
3320         put->mask_len = op->mask.size;
3321
3322         if (want_path == SF_FAST_PATH) {
3323             put->actions = facet->xout.odp_actions.data;
3324             put->actions_len = facet->xout.odp_actions.size;
3325         } else {
3326             compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3327                               op->slow_stub, sizeof op->slow_stub,
3328                               &put->actions, &put->actions_len);
3329         }
3330         put->stats = NULL;
3331     }
3332 }
3333
3334 /* Handles flow miss 'miss'.  May add any required datapath operations
3335  * to 'ops', incrementing '*n_ops' for each new op. */
3336 static void
3337 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3338                  size_t *n_ops)
3339 {
3340     struct facet *facet;
3341
3342     miss->ofproto->n_missed += miss->stats.n_packets;
3343
3344     facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3345     if (!facet) {
3346         /* There does not exist a bijection between 'struct flow' and datapath
3347          * flow keys with fitness ODP_FIT_TO_LITTLE.  This breaks a fundamental
3348          * assumption used throughout the facet and subfacet handling code.
3349          * Since we have to handle these misses in userspace anyway, we simply
3350          * skip facet creation, avoiding the problem altogether. */
3351         if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3352             || !flow_miss_should_make_facet(miss)) {
3353             return;
3354         }
3355
3356         facet = facet_create(miss);
3357     }
3358     handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3359 }
3360
3361 static struct drop_key *
3362 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3363                 size_t key_len)
3364 {
3365     struct drop_key *drop_key;
3366
3367     HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3368                              &backer->drop_keys) {
3369         if (drop_key->key_len == key_len
3370             && !memcmp(drop_key->key, key, key_len)) {
3371             return drop_key;
3372         }
3373     }
3374     return NULL;
3375 }
3376
3377 static void
3378 drop_key_clear(struct dpif_backer *backer)
3379 {
3380     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3381     struct drop_key *drop_key, *next;
3382
3383     HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3384         int error;
3385
3386         error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3387                               NULL);
3388         if (error && !VLOG_DROP_WARN(&rl)) {
3389             struct ds ds = DS_EMPTY_INITIALIZER;
3390             odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3391             VLOG_WARN("Failed to delete drop key (%s) (%s)",
3392                       ovs_strerror(error), ds_cstr(&ds));
3393             ds_destroy(&ds);
3394         }
3395
3396         hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3397         drop_key_destroy(drop_key);
3398     }
3399
3400     udpif_drop_key_clear(backer->udpif);
3401 }
3402
3403 static void
3404 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3405 {
3406     struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3407     struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3408     struct flow_miss *miss;
3409     size_t n_ops, i;
3410
3411     /* Process each element in the to-do list, constructing the set of
3412      * operations to batch. */
3413     n_ops = 0;
3414     HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3415         handle_flow_miss(miss, flow_miss_ops, &n_ops);
3416     }
3417     ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3418
3419     /* Execute batch. */
3420     for (i = 0; i < n_ops; i++) {
3421         dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3422     }
3423     dpif_operate(backer->dpif, dpif_ops, n_ops);
3424
3425     for (i = 0; i < n_ops; i++) {
3426         if (dpif_ops[i]->error != 0
3427             && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3428             && flow_miss_ops[i].subfacet) {
3429             struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3430
3431             COVERAGE_INC(subfacet_install_fail);
3432
3433             /* Zero-out subfacet counters when installation failed, but
3434              * datapath reported hits.  This should not happen and
3435              * indicates a bug, since if the datapath flow exists, we
3436              * should not be attempting to create a new subfacet.  A
3437              * buggy datapath could trigger this, so just zero out the
3438              * counters and log an error. */
3439             if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3440                 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3441                             "datapath reported hits");
3442                 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3443             }
3444
3445             subfacet->path = SF_NOT_INSTALLED;
3446         }
3447     }
3448 }
3449
3450 static void
3451 handle_upcalls(struct dpif_backer *backer)
3452 {
3453     struct flow_miss_batch *fmb;
3454     int n_processed;
3455
3456     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3457         struct drop_key *drop_key = drop_key_next(backer->udpif);
3458         if (!drop_key) {
3459             break;
3460         }
3461
3462         if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3463             hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3464                         hash_bytes(drop_key->key, drop_key->key_len, 0));
3465             dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3466                           drop_key->key, drop_key->key_len,
3467                           NULL, 0, NULL, 0, NULL);
3468         } else {
3469             drop_key_destroy(drop_key);
3470         }
3471     }
3472
3473     fmb = flow_miss_batch_next(backer->udpif);
3474     if (fmb) {
3475         handle_flow_misses(backer, fmb);
3476         flow_miss_batch_destroy(fmb);
3477     }
3478 }
3479 \f
3480 /* Flow expiration. */
3481
3482 static int subfacet_max_idle(const struct dpif_backer *);
3483 static void update_stats(struct dpif_backer *);
3484 static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex);
3485 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3486
3487 /* This function is called periodically by run().  Its job is to collect
3488  * updates for the flows that have been installed into the datapath, most
3489  * importantly when they last were used, and then use that information to
3490  * expire flows that have not been used recently.
3491  *
3492  * Returns the number of milliseconds after which it should be called again. */
3493 static int
3494 expire(struct dpif_backer *backer)
3495 {
3496     struct ofproto_dpif *ofproto;
3497     size_t n_subfacets;
3498     int max_idle;
3499
3500     /* Periodically clear out the drop keys in an effort to keep them
3501      * relatively few. */
3502     drop_key_clear(backer);
3503
3504     /* Update stats for each flow in the backer. */
3505     update_stats(backer);
3506
3507     n_subfacets = hmap_count(&backer->subfacets);
3508     if (n_subfacets) {
3509         struct subfacet *subfacet;
3510         long long int total, now;
3511
3512         total = 0;
3513         now = time_msec();
3514         HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3515             total += now - subfacet->created;
3516         }
3517         backer->avg_subfacet_life += total / n_subfacets;
3518     }
3519     backer->avg_subfacet_life /= 2;
3520
3521     backer->avg_n_subfacet += n_subfacets;
3522     backer->avg_n_subfacet /= 2;
3523
3524     backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3525
3526     max_idle = subfacet_max_idle(backer);
3527     expire_subfacets(backer, max_idle);
3528
3529     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3530         struct rule *rule, *next_rule;
3531
3532         if (ofproto->backer != backer) {
3533             continue;
3534         }
3535
3536         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3537          * has passed. */
3538         ovs_mutex_lock(&ofproto_mutex);
3539         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3540                             &ofproto->up.expirable) {
3541             rule_expire(rule_dpif_cast(rule));
3542         }
3543         ovs_mutex_unlock(&ofproto_mutex);
3544
3545         /* All outstanding data in existing flows has been accounted, so it's a
3546          * good time to do bond rebalancing. */
3547         if (ofproto->has_bonded_bundles) {
3548             struct ofbundle *bundle;
3549
3550             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3551                 if (bundle->bond) {
3552                     bond_rebalance(bundle->bond);
3553                 }
3554             }
3555         }
3556     }
3557
3558     return MIN(max_idle, 1000);
3559 }
3560
3561 /* Updates flow table statistics given that the datapath just reported 'stats'
3562  * as 'subfacet''s statistics. */
3563 static void
3564 update_subfacet_stats(struct subfacet *subfacet,
3565                       const struct dpif_flow_stats *stats)
3566 {
3567     struct facet *facet = subfacet->facet;
3568     struct dpif_flow_stats diff;
3569
3570     diff.tcp_flags = stats->tcp_flags;
3571     diff.used = stats->used;
3572
3573     if (stats->n_packets >= subfacet->dp_packet_count) {
3574         diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3575     } else {
3576         VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3577         diff.n_packets = 0;
3578     }
3579
3580     if (stats->n_bytes >= subfacet->dp_byte_count) {
3581         diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3582     } else {
3583         VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3584         diff.n_bytes = 0;
3585     }
3586
3587     facet->ofproto->n_hit += diff.n_packets;
3588     subfacet->dp_packet_count = stats->n_packets;
3589     subfacet->dp_byte_count = stats->n_bytes;
3590     subfacet_update_stats(subfacet, &diff);
3591
3592     if (facet->accounted_bytes < facet->byte_count) {
3593         facet_learn(facet);
3594         facet_account(facet);
3595         facet->accounted_bytes = facet->byte_count;
3596     }
3597 }
3598
3599 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3600  * about, or a flow that shouldn't be installed but was anyway.  Delete it. */
3601 static void
3602 delete_unexpected_flow(struct dpif_backer *backer,
3603                        const struct nlattr *key, size_t key_len)
3604 {
3605     if (!VLOG_DROP_WARN(&rl)) {
3606         struct ds s;
3607
3608         ds_init(&s);
3609         odp_flow_key_format(key, key_len, &s);
3610         VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3611         ds_destroy(&s);
3612     }
3613
3614     COVERAGE_INC(facet_unexpected);
3615     dpif_flow_del(backer->dpif, key, key_len, NULL);
3616 }
3617
3618 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3619  *
3620  * This function also pushes statistics updates to rules which each facet
3621  * resubmits into.  Generally these statistics will be accurate.  However, if a
3622  * facet changes the rule it resubmits into at some time in between
3623  * update_stats() runs, it is possible that statistics accrued to the
3624  * old rule will be incorrectly attributed to the new rule.  This could be
3625  * avoided by calling update_stats() whenever rules are created or
3626  * deleted.  However, the performance impact of making so many calls to the
3627  * datapath do not justify the benefit of having perfectly accurate statistics.
3628  *
3629  * In addition, this function maintains per ofproto flow hit counts. The patch
3630  * port is not treated specially. e.g. A packet ingress from br0 patched into
3631  * br1 will increase the hit count of br0 by 1, however, does not affect
3632  * the hit or miss counts of br1.
3633  */
3634 static void
3635 update_stats(struct dpif_backer *backer)
3636 {
3637     const struct dpif_flow_stats *stats;
3638     struct dpif_flow_dump dump;
3639     const struct nlattr *key, *mask;
3640     size_t key_len, mask_len;
3641
3642     dpif_flow_dump_start(&dump, backer->dpif);
3643     while (dpif_flow_dump_next(&dump, &key, &key_len,
3644                                &mask, &mask_len, NULL, NULL, &stats)) {
3645         struct subfacet *subfacet;
3646         uint32_t key_hash;
3647
3648         key_hash = odp_flow_key_hash(key, key_len);
3649         subfacet = subfacet_find(backer, key, key_len, key_hash);
3650         switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3651         case SF_FAST_PATH:
3652             update_subfacet_stats(subfacet, stats);
3653             break;
3654
3655         case SF_SLOW_PATH:
3656             /* Stats are updated per-packet. */
3657             break;
3658
3659         case SF_NOT_INSTALLED:
3660         default:
3661             delete_unexpected_flow(backer, key, key_len);
3662             break;
3663         }
3664         run_fast_rl();
3665     }
3666     dpif_flow_dump_done(&dump);
3667 }
3668
3669 /* Calculates and returns the number of milliseconds of idle time after which
3670  * subfacets should expire from the datapath.  When a subfacet expires, we fold
3671  * its statistics into its facet, and when a facet's last subfacet expires, we
3672  * fold its statistic into its rule. */
3673 static int
3674 subfacet_max_idle(const struct dpif_backer *backer)
3675 {
3676     /*
3677      * Idle time histogram.
3678      *
3679      * Most of the time a switch has a relatively small number of subfacets.
3680      * When this is the case we might as well keep statistics for all of them
3681      * in userspace and to cache them in the kernel datapath for performance as
3682      * well.
3683      *
3684      * As the number of subfacets increases, the memory required to maintain
3685      * statistics about them in userspace and in the kernel becomes
3686      * significant.  However, with a large number of subfacets it is likely
3687      * that only a few of them are "heavy hitters" that consume a large amount
3688      * of bandwidth.  At this point, only heavy hitters are worth caching in
3689      * the kernel and maintaining in userspaces; other subfacets we can
3690      * discard.
3691      *
3692      * The technique used to compute the idle time is to build a histogram with
3693      * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each.  Each subfacet
3694      * that is installed in the kernel gets dropped in the appropriate bucket.
3695      * After the histogram has been built, we compute the cutoff so that only
3696      * the most-recently-used 1% of subfacets (but at least
3697      * flow_eviction_threshold flows) are kept cached.  At least
3698      * the most-recently-used bucket of subfacets is kept, so actually an
3699      * arbitrary number of subfacets can be kept in any given expiration run
3700      * (though the next run will delete most of those unless they receive
3701      * additional data).
3702      *
3703      * This requires a second pass through the subfacets, in addition to the
3704      * pass made by update_stats(), because the former function never looks at
3705      * uninstallable subfacets.
3706      */
3707     enum { BUCKET_WIDTH = 100 };
3708     enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3709     int buckets[N_BUCKETS] = { 0 };
3710     int total, subtotal, bucket;
3711     struct subfacet *subfacet;
3712     long long int now;
3713     int i;
3714
3715     total = hmap_count(&backer->subfacets);
3716     if (total <= flow_eviction_threshold) {
3717         return N_BUCKETS * BUCKET_WIDTH;
3718     }
3719
3720     /* Build histogram. */
3721     now = time_msec();
3722     HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3723         long long int idle = now - subfacet->used;
3724         int bucket = (idle <= 0 ? 0
3725                       : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3726                       : (unsigned int) idle / BUCKET_WIDTH);
3727         buckets[bucket]++;
3728     }
3729
3730     /* Find the first bucket whose flows should be expired. */
3731     subtotal = bucket = 0;
3732     do {
3733         subtotal += buckets[bucket++];
3734     } while (bucket < N_BUCKETS &&
3735              subtotal < MAX(flow_eviction_threshold, total / 100));
3736
3737     if (VLOG_IS_DBG_ENABLED()) {
3738         struct ds s;
3739
3740         ds_init(&s);
3741         ds_put_cstr(&s, "keep");
3742         for (i = 0; i < N_BUCKETS; i++) {
3743             if (i == bucket) {
3744                 ds_put_cstr(&s, ", drop");
3745             }
3746             if (buckets[i]) {
3747                 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3748             }
3749         }
3750         VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3751         ds_destroy(&s);
3752     }
3753
3754     return bucket * BUCKET_WIDTH;
3755 }
3756
3757 static void
3758 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3759 {
3760     /* Cutoff time for most flows. */
3761     long long int normal_cutoff = time_msec() - dp_max_idle;
3762
3763     /* We really want to keep flows for special protocols around, so use a more
3764      * conservative cutoff. */
3765     long long int special_cutoff = time_msec() - 10000;
3766
3767     struct subfacet *subfacet, *next_subfacet;
3768     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3769     int n_batch;
3770
3771     n_batch = 0;
3772     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3773                         &backer->subfacets) {
3774         long long int cutoff;
3775
3776         cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3777                                                 | SLOW_STP)
3778                   ? special_cutoff
3779                   : normal_cutoff);
3780         if (subfacet->used < cutoff) {
3781             if (subfacet->path != SF_NOT_INSTALLED) {
3782                 batch[n_batch++] = subfacet;
3783                 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3784                     subfacet_destroy_batch(backer, batch, n_batch);
3785                     n_batch = 0;
3786                 }
3787             } else {
3788                 subfacet_destroy(subfacet);
3789             }
3790         }
3791     }
3792
3793     if (n_batch > 0) {
3794         subfacet_destroy_batch(backer, batch, n_batch);
3795     }
3796 }
3797
3798 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3799  * then delete it entirely. */
3800 static void
3801 rule_expire(struct rule_dpif *rule)
3802     OVS_REQUIRES(ofproto_mutex)
3803 {
3804     uint16_t idle_timeout, hard_timeout;
3805     long long int now = time_msec();
3806     int reason;
3807
3808     ovs_assert(!rule->up.pending);
3809
3810     /* Has 'rule' expired? */
3811     ovs_mutex_lock(&rule->up.mutex);
3812     hard_timeout = rule->up.hard_timeout;
3813     idle_timeout = rule->up.idle_timeout;
3814     if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3815         reason = OFPRR_HARD_TIMEOUT;
3816     } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3817         reason = OFPRR_IDLE_TIMEOUT;
3818     } else {
3819         reason = -1;
3820     }
3821     ovs_mutex_unlock(&rule->up.mutex);
3822
3823     if (reason >= 0) {
3824         COVERAGE_INC(ofproto_dpif_expired);
3825         ofproto_rule_expire(&rule->up, reason);
3826     }
3827 }
3828 \f
3829 /* Facets. */
3830
3831 /* Creates and returns a new facet based on 'miss'.
3832  *
3833  * The caller must already have determined that no facet with an identical
3834  * 'miss->flow' exists in 'miss->ofproto'.
3835  *
3836  * 'rule' and 'xout' must have been created based on 'miss'.
3837  *
3838  * 'facet'' statistics are initialized based on 'stats'.
3839  *
3840  * The facet will initially have no subfacets.  The caller should create (at
3841  * least) one subfacet with subfacet_create(). */
3842 static struct facet *
3843 facet_create(const struct flow_miss *miss)
3844 {
3845     struct ofproto_dpif *ofproto = miss->ofproto;
3846     struct facet *facet;
3847     struct match match;
3848
3849     COVERAGE_INC(facet_create);
3850     facet = xzalloc(sizeof *facet);
3851     facet->ofproto = miss->ofproto;
3852     facet->used = miss->stats.used;
3853     facet->flow = miss->flow;
3854     facet->learn_rl = time_msec() + 500;
3855
3856     list_init(&facet->subfacets);
3857     netflow_flow_init(&facet->nf_flow);
3858     netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3859
3860     xlate_out_copy(&facet->xout, &miss->xout);
3861
3862     match_init(&match, &facet->flow, &facet->xout.wc);
3863     cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3864     ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3865     classifier_insert(&ofproto->facets, &facet->cr);
3866     ovs_rwlock_unlock(&ofproto->facets.rwlock);
3867
3868     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
3869     return facet;
3870 }
3871
3872 static void
3873 facet_free(struct facet *facet)
3874 {
3875     if (facet) {
3876         xlate_out_uninit(&facet->xout);
3877         free(facet);
3878     }
3879 }
3880
3881 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3882  * 'flow' must reflect the data in 'packet'. */
3883 int
3884 ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto,
3885                              const struct flow *flow,
3886                              struct rule_dpif *rule,
3887                              const struct ofpact *ofpacts, size_t ofpacts_len,
3888                              struct ofpbuf *packet)
3889 {
3890     struct odputil_keybuf keybuf;
3891     struct dpif_flow_stats stats;
3892     struct xlate_out xout;
3893     struct xlate_in xin;
3894     ofp_port_t in_port;
3895     struct ofpbuf key;
3896     int error;
3897
3898     ovs_assert((rule != NULL) != (ofpacts != NULL));
3899
3900     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
3901     if (rule) {
3902         rule_dpif_credit_stats(rule, &stats);
3903     }
3904
3905     xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
3906     xin.ofpacts = ofpacts;
3907     xin.ofpacts_len = ofpacts_len;
3908     xin.resubmit_stats = &stats;
3909     xlate_actions(&xin, &xout);
3910
3911     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3912     in_port = flow->in_port.ofp_port;
3913     if (in_port == OFPP_NONE) {
3914         in_port = OFPP_LOCAL;
3915     }
3916     odp_flow_key_from_flow(&key, flow, ofp_port_to_odp_port(ofproto, in_port));
3917
3918     error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
3919                          xout.odp_actions.data, xout.odp_actions.size, packet,
3920                          (xout.slow & SLOW_ACTION) != 0);
3921     xlate_out_uninit(&xout);
3922
3923     return error;
3924 }
3925
3926 /* Remove 'facet' from its ofproto and free up the associated memory:
3927  *
3928  *   - If 'facet' was installed in the datapath, uninstalls it and updates its
3929  *     rule's statistics, via subfacet_uninstall().
3930  *
3931  *   - Removes 'facet' from its rule and from ofproto->facets.
3932  */
3933 static void
3934 facet_remove(struct facet *facet)
3935 {
3936     struct subfacet *subfacet, *next_subfacet;
3937
3938     COVERAGE_INC(facet_remove);
3939     ovs_assert(!list_is_empty(&facet->subfacets));
3940
3941     /* First uninstall all of the subfacets to get final statistics. */
3942     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3943         subfacet_uninstall(subfacet);
3944     }
3945
3946     /* Flush the final stats to the rule.
3947      *
3948      * This might require us to have at least one subfacet around so that we
3949      * can use its actions for accounting in facet_account(), which is why we
3950      * have uninstalled but not yet destroyed the subfacets. */
3951     facet_flush_stats(facet);
3952
3953     /* Now we're really all done so destroy everything. */
3954     LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3955                         &facet->subfacets) {
3956         subfacet_destroy__(subfacet);
3957     }
3958     ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
3959     classifier_remove(&facet->ofproto->facets, &facet->cr);
3960     ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
3961     cls_rule_destroy(&facet->cr);
3962     facet_free(facet);
3963 }
3964
3965 /* Feed information from 'facet' back into the learning table to keep it in
3966  * sync with what is actually flowing through the datapath. */
3967 static void
3968 facet_learn(struct facet *facet)
3969 {
3970     long long int now = time_msec();
3971
3972     if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
3973         return;
3974     }
3975
3976     facet->learn_rl = now + 500;
3977
3978     if (!facet->xout.has_learn
3979         && !facet->xout.has_normal
3980         && (!facet->xout.has_fin_timeout
3981             || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3982         return;
3983     }
3984
3985     facet_push_stats(facet, true);
3986 }
3987
3988 static void
3989 facet_account(struct facet *facet)
3990 {
3991     const struct nlattr *a;
3992     unsigned int left;
3993     ovs_be16 vlan_tci;
3994     uint64_t n_bytes;
3995
3996     if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
3997         return;
3998     }
3999     n_bytes = facet->byte_count - facet->accounted_bytes;
4000
4001     /* This loop feeds byte counters to bond_account() for rebalancing to use
4002      * as a basis.  We also need to track the actual VLAN on which the packet
4003      * is going to be sent to ensure that it matches the one passed to
4004      * bond_choose_output_slave().  (Otherwise, we will account to the wrong
4005      * hash bucket.)
4006      *
4007      * We use the actions from an arbitrary subfacet because they should all
4008      * be equally valid for our purpose. */
4009     vlan_tci = facet->flow.vlan_tci;
4010     NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4011                              facet->xout.odp_actions.size) {
4012         const struct ovs_action_push_vlan *vlan;
4013         struct ofport_dpif *port;
4014
4015         switch (nl_attr_type(a)) {
4016         case OVS_ACTION_ATTR_OUTPUT:
4017             port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4018             if (port && port->bundle && port->bundle->bond) {
4019                 bond_account(port->bundle->bond, &facet->flow,
4020                              vlan_tci_to_vid(vlan_tci), n_bytes);
4021             }
4022             break;
4023
4024         case OVS_ACTION_ATTR_POP_VLAN:
4025             vlan_tci = htons(0);
4026             break;
4027
4028         case OVS_ACTION_ATTR_PUSH_VLAN:
4029             vlan = nl_attr_get(a);
4030             vlan_tci = vlan->vlan_tci;
4031             break;
4032         }
4033     }
4034 }
4035
4036 /* Returns true if the only action for 'facet' is to send to the controller.
4037  * (We don't report NetFlow expiration messages for such facets because they
4038  * are just part of the control logic for the network, not real traffic). */
4039 static bool
4040 facet_is_controller_flow(struct facet *facet)
4041 {
4042     if (facet) {
4043         struct ofproto_dpif *ofproto = facet->ofproto;
4044         const struct ofpact *ofpacts;
4045         struct rule_actions *actions;
4046         struct rule_dpif *rule;
4047         size_t ofpacts_len;
4048         bool is_controller;
4049
4050         rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4051         actions = rule_dpif_get_actions(rule);
4052         rule_dpif_unref(rule);
4053
4054         ofpacts_len = actions->ofpacts_len;
4055         ofpacts = actions->ofpacts;
4056         is_controller = ofpacts_len > 0
4057             && ofpacts->type == OFPACT_CONTROLLER
4058             && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4059         rule_actions_unref(actions);
4060
4061         return is_controller;
4062     }
4063     return false;
4064 }
4065
4066 /* Folds all of 'facet''s statistics into its rule.  Also updates the
4067  * accounting ofhook and emits a NetFlow expiration if appropriate.  All of
4068  * 'facet''s statistics in the datapath should have been zeroed and folded into
4069  * its packet and byte counts before this function is called. */
4070 static void
4071 facet_flush_stats(struct facet *facet)
4072 {
4073     struct ofproto_dpif *ofproto = facet->ofproto;
4074     struct subfacet *subfacet;
4075
4076     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4077         ovs_assert(!subfacet->dp_byte_count);
4078         ovs_assert(!subfacet->dp_packet_count);
4079     }
4080
4081     facet_push_stats(facet, false);
4082     if (facet->accounted_bytes < facet->byte_count) {
4083         facet_account(facet);
4084         facet->accounted_bytes = facet->byte_count;
4085     }
4086
4087     if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4088         struct ofexpired expired;
4089         expired.flow = facet->flow;
4090         expired.packet_count = facet->packet_count;
4091         expired.byte_count = facet->byte_count;
4092         expired.used = facet->used;
4093         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4094     }
4095
4096     /* Reset counters to prevent double counting if 'facet' ever gets
4097      * reinstalled. */
4098     facet_reset_counters(facet);
4099
4100     netflow_flow_clear(&facet->nf_flow);
4101     facet->tcp_flags = 0;
4102 }
4103
4104 /* Searches 'ofproto''s table of facets for one which would be responsible for
4105  * 'flow'.  Returns it if found, otherwise a null pointer.
4106  *
4107  * The returned facet might need revalidation; use facet_lookup_valid()
4108  * instead if that is important. */
4109 static struct facet *
4110 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4111 {
4112     struct cls_rule *cr;
4113
4114     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4115     cr = classifier_lookup(&ofproto->facets, flow, NULL);
4116     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4117     return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4118 }
4119
4120 /* Searches 'ofproto''s table of facets for one capable that covers
4121  * 'flow'.  Returns it if found, otherwise a null pointer.
4122  *
4123  * The returned facet is guaranteed to be valid. */
4124 static struct facet *
4125 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4126 {
4127     struct facet *facet;
4128
4129     facet = facet_find(ofproto, flow);
4130     if (facet
4131         && ofproto->backer->need_revalidate
4132         && !facet_revalidate(facet)) {
4133         return NULL;
4134     }
4135
4136     return facet;
4137 }
4138
4139 static bool
4140 facet_check_consistency(struct facet *facet)
4141 {
4142     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4143
4144     struct xlate_out xout;
4145     struct xlate_in xin;
4146     bool ok;
4147
4148     /* Check the datapath actions for consistency. */
4149     xlate_in_init(&xin, facet->ofproto, &facet->flow, NULL, 0, NULL);
4150     xlate_actions(&xin, &xout);
4151
4152     ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4153         && facet->xout.slow == xout.slow;
4154     if (!ok && !VLOG_DROP_WARN(&rl)) {
4155         struct ds s = DS_EMPTY_INITIALIZER;
4156
4157         flow_format(&s, &facet->flow);
4158         ds_put_cstr(&s, ": inconsistency in facet");
4159
4160         if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4161             ds_put_cstr(&s, " (actions were: ");
4162             format_odp_actions(&s, facet->xout.odp_actions.data,
4163                                facet->xout.odp_actions.size);
4164             ds_put_cstr(&s, ") (correct actions: ");
4165             format_odp_actions(&s, xout.odp_actions.data,
4166                                xout.odp_actions.size);
4167             ds_put_char(&s, ')');
4168         }
4169
4170         if (facet->xout.slow != xout.slow) {
4171             ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4172         }
4173
4174         ds_destroy(&s);
4175     }
4176     xlate_out_uninit(&xout);
4177
4178     return ok;
4179 }
4180
4181 /* Re-searches the classifier for 'facet':
4182  *
4183  *   - If the rule found is different from 'facet''s current rule, moves
4184  *     'facet' to the new rule and recompiles its actions.
4185  *
4186  *   - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4187  *     where it is and recompiles its actions anyway.
4188  *
4189  *   - If any of 'facet''s subfacets correspond to a new flow according to
4190  *     xlate_receive(), 'facet' is removed.
4191  *
4192  *   Returns true if 'facet' is still valid.  False if 'facet' was removed. */
4193 static bool
4194 facet_revalidate(struct facet *facet)
4195 {
4196     struct ofproto_dpif *ofproto = facet->ofproto;
4197     struct rule_dpif *new_rule;
4198     struct subfacet *subfacet;
4199     struct flow_wildcards wc;
4200     struct xlate_out xout;
4201     struct xlate_in xin;
4202
4203     COVERAGE_INC(facet_revalidate);
4204
4205     /* Check that child subfacets still correspond to this facet.  Tunnel
4206      * configuration changes could cause a subfacet's OpenFlow in_port to
4207      * change. */
4208     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4209         struct ofproto_dpif *recv_ofproto;
4210         struct flow recv_flow;
4211         int error;
4212
4213         error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4214                               subfacet->key_len, &recv_flow, NULL,
4215                               &recv_ofproto, NULL);
4216         if (error
4217             || recv_ofproto != ofproto
4218             || facet != facet_find(ofproto, &recv_flow)) {
4219             facet_remove(facet);
4220             return false;
4221         }
4222     }
4223
4224     flow_wildcards_init_catchall(&wc);
4225     rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4226
4227     /* Calculate new datapath actions.
4228      *
4229      * We do not modify any 'facet' state yet, because we might need to, e.g.,
4230      * emit a NetFlow expiration and, if so, we need to have the old state
4231      * around to properly compose it. */
4232     xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4233     xlate_actions(&xin, &xout);
4234     flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4235     /* Make sure non -packet fields are not masked. If not cleared,
4236      * the memcmp() below may fail, causing an otherwise valid facet
4237      * to be removed. */
4238     flow_wildcards_clear_non_packet_fields(&xout.wc);
4239
4240     /* A facet's slow path reason should only change under dramatic
4241      * circumstances.  Rather than try to update everything, it's simpler to
4242      * remove the facet and start over.
4243      *
4244      * More importantly, if a facet's wildcards change, it will be relatively
4245      * difficult to figure out if its subfacets still belong to it, and if not
4246      * which facet they may belong to.  Again, to avoid the complexity, we
4247      * simply give up instead. */
4248     if (facet->xout.slow != xout.slow
4249         || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4250         facet_remove(facet);
4251         xlate_out_uninit(&xout);
4252         rule_dpif_unref(new_rule);
4253         return false;
4254     }
4255
4256     if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4257         LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4258             if (subfacet->path == SF_FAST_PATH) {
4259                 struct dpif_flow_stats stats;
4260
4261                 subfacet_install(subfacet, &xout.odp_actions, &stats);
4262                 subfacet_update_stats(subfacet, &stats);
4263             }
4264         }
4265
4266         facet_flush_stats(facet);
4267
4268         ofpbuf_clear(&facet->xout.odp_actions);
4269         ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4270                    xout.odp_actions.size);
4271     }
4272
4273     /* Update 'facet' now that we've taken care of all the old state. */
4274     facet->xout.slow = xout.slow;
4275     facet->xout.has_learn = xout.has_learn;
4276     facet->xout.has_normal = xout.has_normal;
4277     facet->xout.has_fin_timeout = xout.has_fin_timeout;
4278     facet->xout.nf_output_iface = xout.nf_output_iface;
4279     facet->xout.mirrors = xout.mirrors;
4280     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4281
4282     ovs_mutex_lock(&new_rule->up.mutex);
4283     facet->used = MAX(facet->used, new_rule->up.created);
4284     ovs_mutex_unlock(&new_rule->up.mutex);
4285
4286     xlate_out_uninit(&xout);
4287     rule_dpif_unref(new_rule);
4288     return true;
4289 }
4290
4291 static void
4292 facet_reset_counters(struct facet *facet)
4293 {
4294     facet->packet_count = 0;
4295     facet->byte_count = 0;
4296     facet->prev_packet_count = 0;
4297     facet->prev_byte_count = 0;
4298     facet->accounted_bytes = 0;
4299 }
4300
4301 static void
4302 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4303                 struct dpif_flow_stats *stats, bool may_learn)
4304 {
4305     struct ofport_dpif *in_port;
4306     struct xlate_in xin;
4307
4308     in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4309     if (in_port && in_port->is_tunnel) {
4310         netdev_vport_inc_rx(in_port->up.netdev, stats);
4311         if (in_port->bfd) {
4312             bfd_account_rx(in_port->bfd, stats);
4313         }
4314     }
4315
4316     xlate_in_init(&xin, ofproto, flow, NULL, stats->tcp_flags, NULL);
4317     xin.resubmit_stats = stats;
4318     xin.may_learn = may_learn;
4319     xlate_actions_for_side_effects(&xin);
4320 }
4321
4322 static void
4323 facet_push_stats(struct facet *facet, bool may_learn)
4324 {
4325     struct dpif_flow_stats stats;
4326
4327     ovs_assert(facet->packet_count >= facet->prev_packet_count);
4328     ovs_assert(facet->byte_count >= facet->prev_byte_count);
4329     ovs_assert(facet->used >= facet->prev_used);
4330
4331     stats.n_packets = facet->packet_count - facet->prev_packet_count;
4332     stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4333     stats.used = facet->used;
4334     stats.tcp_flags = facet->tcp_flags;
4335
4336     if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4337         facet->prev_packet_count = facet->packet_count;
4338         facet->prev_byte_count = facet->byte_count;
4339         facet->prev_used = facet->used;
4340
4341         netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4342                                  facet->used);
4343         netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4344         mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4345                             stats.n_packets, stats.n_bytes);
4346         flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4347     }
4348 }
4349
4350 static void
4351 push_all_stats__(bool run_fast)
4352 {
4353     static long long int rl = LLONG_MIN;
4354     struct ofproto_dpif *ofproto;
4355
4356     if (time_msec() < rl) {
4357         return;
4358     }
4359
4360     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4361         struct cls_cursor cursor;
4362         struct facet *facet;
4363
4364         ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4365         cls_cursor_init(&cursor, &ofproto->facets, NULL);
4366         CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4367             facet_push_stats(facet, false);
4368             if (run_fast) {
4369                 run_fast_rl();
4370             }
4371         }
4372         ovs_rwlock_unlock(&ofproto->facets.rwlock);
4373     }
4374
4375     rl = time_msec() + 100;
4376 }
4377
4378 static void
4379 push_all_stats(void)
4380 {
4381     push_all_stats__(true);
4382 }
4383
4384 void
4385 rule_dpif_credit_stats(struct rule_dpif *rule,
4386                        const struct dpif_flow_stats *stats)
4387 {
4388     ovs_mutex_lock(&rule->stats_mutex);
4389     rule->packet_count += stats->n_packets;
4390     rule->byte_count += stats->n_bytes;
4391     rule->up.used = MAX(rule->up.used, stats->used);
4392     ovs_mutex_unlock(&rule->stats_mutex);
4393 }
4394
4395 bool
4396 rule_dpif_is_fail_open(const struct rule_dpif *rule)
4397 {
4398     return is_fail_open_rule(&rule->up);
4399 }
4400
4401 bool
4402 rule_dpif_is_table_miss(const struct rule_dpif *rule)
4403 {
4404     return rule_is_table_miss(&rule->up);
4405 }
4406
4407 ovs_be64
4408 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4409     OVS_REQUIRES(rule->up.mutex)
4410 {
4411     return rule->up.flow_cookie;
4412 }
4413
4414 void
4415 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4416                      uint16_t hard_timeout)
4417 {
4418     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4419 }
4420
4421 /* Returns 'rule''s actions.  The caller owns a reference on the returned
4422  * actions and must eventually release it (with rule_actions_unref()) to avoid
4423  * a memory leak. */
4424 struct rule_actions *
4425 rule_dpif_get_actions(const struct rule_dpif *rule)
4426 {
4427     return rule_get_actions(&rule->up);
4428 }
4429 \f
4430 /* Subfacets. */
4431
4432 static struct subfacet *
4433 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4434               size_t key_len, uint32_t key_hash)
4435 {
4436     struct subfacet *subfacet;
4437
4438     HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4439                              &backer->subfacets) {
4440         if (subfacet->key_len == key_len
4441             && !memcmp(key, subfacet->key, key_len)) {
4442             return subfacet;
4443         }
4444     }
4445
4446     return NULL;
4447 }
4448
4449 /* Creates and returns a new subfacet within 'facet' for the flow in 'miss'.
4450  * 'key_hash' must be a hash over miss->key.  The caller must have already
4451  * ensured that no subfacet subfacet already exists. */
4452 static struct subfacet *
4453 subfacet_create(struct facet *facet, struct flow_miss *miss, uint32_t key_hash)
4454 {
4455     struct dpif_backer *backer = miss->ofproto->backer;
4456     const struct nlattr *key = miss->key;
4457     size_t key_len = miss->key_len;
4458     struct subfacet *subfacet;
4459
4460     subfacet = (list_is_empty(&facet->subfacets)
4461                 ? &facet->one_subfacet
4462                 : xmalloc(sizeof *subfacet));
4463
4464     COVERAGE_INC(subfacet_create);
4465     hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4466     list_push_back(&facet->subfacets, &subfacet->list_node);
4467     subfacet->facet = facet;
4468     subfacet->key = xmemdup(key, key_len);
4469     subfacet->key_len = key_len;
4470     subfacet->used = miss->stats.used;
4471     subfacet->created = subfacet->used;
4472     subfacet->dp_packet_count = 0;
4473     subfacet->dp_byte_count = 0;
4474     subfacet->path = SF_NOT_INSTALLED;
4475     subfacet->backer = backer;
4476
4477     return subfacet;
4478 }
4479
4480 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4481  * its facet within 'ofproto', and frees it. */
4482 static void
4483 subfacet_destroy__(struct subfacet *subfacet)
4484 {
4485     struct facet *facet = subfacet->facet;
4486
4487     COVERAGE_INC(subfacet_destroy);
4488     subfacet_uninstall(subfacet);
4489     hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4490     list_remove(&subfacet->list_node);
4491     free(subfacet->key);
4492     if (subfacet != &facet->one_subfacet) {
4493         free(subfacet);
4494     }
4495 }
4496
4497 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4498  * last remaining subfacet in its facet destroys the facet too. */
4499 static void
4500 subfacet_destroy(struct subfacet *subfacet)
4501 {
4502     struct facet *facet = subfacet->facet;
4503
4504     if (list_is_singleton(&facet->subfacets)) {
4505         /* facet_remove() needs at least one subfacet (it will remove it). */
4506         facet_remove(facet);
4507     } else {
4508         subfacet_destroy__(subfacet);
4509     }
4510 }
4511
4512 static void
4513 subfacet_destroy_batch(struct dpif_backer *backer,
4514                        struct subfacet **subfacets, int n)
4515 {
4516     struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4517     struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4518     struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4519     int i;
4520
4521     for (i = 0; i < n; i++) {
4522         ops[i].type = DPIF_OP_FLOW_DEL;
4523         ops[i].u.flow_del.key = subfacets[i]->key;
4524         ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4525         ops[i].u.flow_del.stats = &stats[i];
4526         opsp[i] = &ops[i];
4527     }
4528
4529     dpif_operate(backer->dpif, opsp, n);
4530     for (i = 0; i < n; i++) {
4531         subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4532         subfacets[i]->path = SF_NOT_INSTALLED;
4533         subfacet_destroy(subfacets[i]);
4534         run_fast_rl();
4535     }
4536 }
4537
4538 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4539  * bytes of actions in 'actions'.  If 'stats' is non-null, statistics counters
4540  * in the datapath will be zeroed and 'stats' will be updated with traffic new
4541  * since 'subfacet' was last updated.
4542  *
4543  * Returns 0 if successful, otherwise a positive errno value. */
4544 static int
4545 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4546                  struct dpif_flow_stats *stats)
4547 {
4548     struct facet *facet = subfacet->facet;
4549     enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4550     const struct nlattr *actions = odp_actions->data;
4551     size_t actions_len = odp_actions->size;
4552     struct odputil_keybuf maskbuf;
4553     struct ofpbuf mask;
4554
4555     uint64_t slow_path_stub[128 / 8];
4556     enum dpif_flow_put_flags flags;
4557     int ret;
4558
4559     flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4560                                                : DPIF_FP_MODIFY;
4561     if (stats) {
4562         flags |= DPIF_FP_ZERO_STATS;
4563     }
4564
4565     if (path == SF_SLOW_PATH) {
4566         compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4567                           slow_path_stub, sizeof slow_path_stub,
4568                           &actions, &actions_len);
4569     }
4570
4571     ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4572     if (enable_megaflows) {
4573         odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4574                                &facet->flow, UINT32_MAX);
4575     }
4576
4577     ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4578                         subfacet->key_len,  mask.data, mask.size,
4579                         actions, actions_len, stats);
4580
4581     if (stats) {
4582         subfacet_reset_dp_stats(subfacet, stats);
4583     }
4584
4585     if (ret) {
4586         COVERAGE_INC(subfacet_install_fail);
4587     } else {
4588         subfacet->path = path;
4589     }
4590     return ret;
4591 }
4592
4593 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4594 static void
4595 subfacet_uninstall(struct subfacet *subfacet)
4596 {
4597     if (subfacet->path != SF_NOT_INSTALLED) {
4598         struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4599         struct dpif_flow_stats stats;
4600         int error;
4601
4602         error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4603                               subfacet->key_len, &stats);
4604         subfacet_reset_dp_stats(subfacet, &stats);
4605         if (!error) {
4606             subfacet_update_stats(subfacet, &stats);
4607         }
4608         subfacet->path = SF_NOT_INSTALLED;
4609     } else {
4610         ovs_assert(subfacet->dp_packet_count == 0);
4611         ovs_assert(subfacet->dp_byte_count == 0);
4612     }
4613 }
4614
4615 /* Resets 'subfacet''s datapath statistics counters.  This should be called
4616  * when 'subfacet''s statistics are cleared in the datapath.  If 'stats' is
4617  * non-null, it should contain the statistics returned by dpif when 'subfacet'
4618  * was reset in the datapath.  'stats' will be modified to include only
4619  * statistics new since 'subfacet' was last updated. */
4620 static void
4621 subfacet_reset_dp_stats(struct subfacet *subfacet,
4622                         struct dpif_flow_stats *stats)
4623 {
4624     if (stats
4625         && subfacet->dp_packet_count <= stats->n_packets
4626         && subfacet->dp_byte_count <= stats->n_bytes) {
4627         stats->n_packets -= subfacet->dp_packet_count;
4628         stats->n_bytes -= subfacet->dp_byte_count;
4629     }
4630
4631     subfacet->dp_packet_count = 0;
4632     subfacet->dp_byte_count = 0;
4633 }
4634
4635 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4636  *
4637  * Because of the meaning of a subfacet's counters, it only makes sense to do
4638  * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4639  * represents a packet that was sent by hand or if it represents statistics
4640  * that have been cleared out of the datapath. */
4641 static void
4642 subfacet_update_stats(struct subfacet *subfacet,
4643                       const struct dpif_flow_stats *stats)
4644 {
4645     if (stats->n_packets || stats->used > subfacet->used) {
4646         struct facet *facet = subfacet->facet;
4647
4648         subfacet->used = MAX(subfacet->used, stats->used);
4649         facet->used = MAX(facet->used, stats->used);
4650         facet->packet_count += stats->n_packets;
4651         facet->byte_count += stats->n_bytes;
4652         facet->tcp_flags |= stats->tcp_flags;
4653     }
4654 }
4655 \f
4656 /* Rules. */
4657
4658 /* Lookup 'flow' in 'ofproto''s classifier.  If 'wc' is non-null, sets
4659  * the fields that were relevant as part of the lookup. */
4660 void
4661 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4662                  struct flow_wildcards *wc, struct rule_dpif **rule)
4663 {
4664     struct ofport_dpif *port;
4665
4666     if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4667         return;
4668     }
4669     port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4670     if (!port) {
4671         VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4672                      flow->in_port.ofp_port);
4673     }
4674
4675     choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4676                      ofproto->no_packet_in_rule, rule);
4677 }
4678
4679 bool
4680 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4681                           const struct flow *flow, struct flow_wildcards *wc,
4682                           uint8_t table_id, struct rule_dpif **rule)
4683 {
4684     const struct cls_rule *cls_rule;
4685     struct classifier *cls;
4686     bool frag;
4687
4688     *rule = NULL;
4689     if (table_id >= N_TABLES) {
4690         return false;
4691     }
4692
4693     if (wc) {
4694         memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4695         wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4696     }
4697
4698     cls = &ofproto->up.tables[table_id].cls;
4699     ovs_rwlock_rdlock(&cls->rwlock);
4700     frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4701     if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4702         /* We must pretend that transport ports are unavailable. */
4703         struct flow ofpc_normal_flow = *flow;
4704         ofpc_normal_flow.tp_src = htons(0);
4705         ofpc_normal_flow.tp_dst = htons(0);
4706         cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4707     } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4708         cls_rule = &ofproto->drop_frags_rule->up.cr;
4709         /* Frag mask in wc already set above. */
4710     } else {
4711         cls_rule = classifier_lookup(cls, flow, wc);
4712     }
4713
4714     *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4715     rule_dpif_ref(*rule);
4716     ovs_rwlock_unlock(&cls->rwlock);
4717
4718     return *rule != NULL;
4719 }
4720
4721 /* Given a port configuration (specified as zero if there's no port), chooses
4722  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4723  * flow table miss. */
4724 void
4725 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4726                  struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4727 {
4728     *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4729     rule_dpif_ref(*rule);
4730 }
4731
4732 void
4733 rule_dpif_ref(struct rule_dpif *rule)
4734 {
4735     if (rule) {
4736         ofproto_rule_ref(&rule->up);
4737     }
4738 }
4739
4740 void
4741 rule_dpif_unref(struct rule_dpif *rule)
4742 {
4743     if (rule) {
4744         ofproto_rule_unref(&rule->up);
4745     }
4746 }
4747
4748 static void
4749 complete_operation(struct rule_dpif *rule)
4750     OVS_REQUIRES(ofproto_mutex)
4751 {
4752     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4753
4754     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4755     ofoperation_complete(rule->up.pending, 0);
4756 }
4757
4758 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4759 {
4760     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4761 }
4762
4763 static struct rule *
4764 rule_alloc(void)
4765 {
4766     struct rule_dpif *rule = xmalloc(sizeof *rule);
4767     return &rule->up;
4768 }
4769
4770 static void
4771 rule_dealloc(struct rule *rule_)
4772 {
4773     struct rule_dpif *rule = rule_dpif_cast(rule_);
4774     free(rule);
4775 }
4776
4777 static enum ofperr
4778 rule_construct(struct rule *rule_)
4779 {
4780     struct rule_dpif *rule = rule_dpif_cast(rule_);
4781     ovs_mutex_init(&rule->stats_mutex);
4782     ovs_mutex_lock(&rule->stats_mutex);
4783     rule->packet_count = 0;
4784     rule->byte_count = 0;
4785     ovs_mutex_unlock(&rule->stats_mutex);
4786     return 0;
4787 }
4788
4789 static void
4790 rule_insert(struct rule *rule_)
4791     OVS_REQUIRES(ofproto_mutex)
4792 {
4793     struct rule_dpif *rule = rule_dpif_cast(rule_);
4794     complete_operation(rule);
4795 }
4796
4797 static void
4798 rule_delete(struct rule *rule_)
4799     OVS_REQUIRES(ofproto_mutex)
4800 {
4801     struct rule_dpif *rule = rule_dpif_cast(rule_);
4802     complete_operation(rule);
4803 }
4804
4805 static void
4806 rule_destruct(struct rule *rule_)
4807 {
4808     struct rule_dpif *rule = rule_dpif_cast(rule_);
4809     ovs_mutex_destroy(&rule->stats_mutex);
4810 }
4811
4812 static void
4813 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4814 {
4815     struct rule_dpif *rule = rule_dpif_cast(rule_);
4816
4817     /* push_all_stats() can handle flow misses which, when using the learn
4818      * action, can cause rules to be added and deleted.  This can corrupt our
4819      * caller's datastructures which assume that rule_get_stats() doesn't have
4820      * an impact on the flow table. To be safe, we disable miss handling. */
4821     push_all_stats__(false);
4822
4823     /* Start from historical data for 'rule' itself that are no longer tracked
4824      * in facets.  This counts, for example, facets that have expired. */
4825     ovs_mutex_lock(&rule->stats_mutex);
4826     *packets = rule->packet_count;
4827     *bytes = rule->byte_count;
4828     ovs_mutex_unlock(&rule->stats_mutex);
4829 }
4830
4831 static void
4832 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4833                   struct ofpbuf *packet)
4834 {
4835     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4836
4837     ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet);
4838 }
4839
4840 static enum ofperr
4841 rule_execute(struct rule *rule, const struct flow *flow,
4842              struct ofpbuf *packet)
4843 {
4844     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4845     ofpbuf_delete(packet);
4846     return 0;
4847 }
4848
4849 static void
4850 rule_modify_actions(struct rule *rule_, bool reset_counters)
4851     OVS_REQUIRES(ofproto_mutex)
4852 {
4853     struct rule_dpif *rule = rule_dpif_cast(rule_);
4854
4855     if (reset_counters) {
4856         ovs_mutex_lock(&rule->stats_mutex);
4857         rule->packet_count = 0;
4858         rule->byte_count = 0;
4859         ovs_mutex_unlock(&rule->stats_mutex);
4860     }
4861
4862     complete_operation(rule);
4863 }
4864
4865 static struct group_dpif *group_dpif_cast(const struct ofgroup *group)
4866 {
4867     return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL;
4868 }
4869
4870 static struct ofgroup *
4871 group_alloc(void)
4872 {
4873     struct group_dpif *group = xzalloc(sizeof *group);
4874     return &group->up;
4875 }
4876
4877 static void
4878 group_dealloc(struct ofgroup *group_)
4879 {
4880     struct group_dpif *group = group_dpif_cast(group_);
4881     free(group);
4882 }
4883
4884 static void
4885 group_construct_stats(struct group_dpif *group)
4886     OVS_REQUIRES(group->stats_mutex)
4887 {
4888     group->packet_count = 0;
4889     group->byte_count = 0;
4890     if (!group->bucket_stats) {
4891         group->bucket_stats = xcalloc(group->up.n_buckets,
4892                                       sizeof *group->bucket_stats);
4893     } else {
4894         memset(group->bucket_stats, 0, group->up.n_buckets *
4895                sizeof *group->bucket_stats);
4896     }
4897 }
4898
4899 static enum ofperr
4900 group_construct(struct ofgroup *group_)
4901 {
4902     struct group_dpif *group = group_dpif_cast(group_);
4903     ovs_mutex_init(&group->stats_mutex);
4904     ovs_mutex_lock(&group->stats_mutex);
4905     group_construct_stats(group);
4906     ovs_mutex_unlock(&group->stats_mutex);
4907     return 0;
4908 }
4909
4910 static void
4911 group_destruct__(struct group_dpif *group)
4912     OVS_REQUIRES(group->stats_mutex)
4913 {
4914     free(group->bucket_stats);
4915     group->bucket_stats = NULL;
4916 }
4917
4918 static void
4919 group_destruct(struct ofgroup *group_)
4920 {
4921     struct group_dpif *group = group_dpif_cast(group_);
4922     ovs_mutex_lock(&group->stats_mutex);
4923     group_destruct__(group);
4924     ovs_mutex_unlock(&group->stats_mutex);
4925     ovs_mutex_destroy(&group->stats_mutex);
4926 }
4927
4928 static enum ofperr
4929 group_modify(struct ofgroup *group_, struct ofgroup *victim_)
4930 {
4931     struct group_dpif *group = group_dpif_cast(group_);
4932     struct group_dpif *victim = group_dpif_cast(victim_);
4933
4934     ovs_mutex_lock(&group->stats_mutex);
4935     if (victim->up.n_buckets < group->up.n_buckets) {
4936         group_destruct__(group);
4937     }
4938     group_construct_stats(group);
4939     ovs_mutex_unlock(&group->stats_mutex);
4940
4941     return 0;
4942 }
4943
4944 static enum ofperr
4945 group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs)
4946 {
4947     struct group_dpif *group = group_dpif_cast(group_);
4948
4949     /* Start from historical data for 'group' itself that are no longer tracked
4950      * in facets.  This counts, for example, facets that have expired. */
4951     ovs_mutex_lock(&group->stats_mutex);
4952     ogs->packet_count = group->packet_count;
4953     ogs->byte_count = group->byte_count;
4954     memcpy(ogs->bucket_stats, group->bucket_stats,
4955            group->up.n_buckets * sizeof *group->bucket_stats);
4956     ovs_mutex_unlock(&group->stats_mutex);
4957
4958     return 0;
4959 }
4960
4961 bool
4962 group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id,
4963                   struct group_dpif **group)
4964     OVS_TRY_RDLOCK(true, (*group)->up.rwlock)
4965 {
4966     struct ofgroup *ofgroup;
4967     bool found;
4968
4969     *group = NULL;
4970     found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup);
4971     *group = found ?  group_dpif_cast(ofgroup) : NULL;
4972
4973     return found;
4974 }
4975
4976 void
4977 group_dpif_release(struct group_dpif *group)
4978     OVS_RELEASES(group->up.rwlock)
4979 {
4980     ofproto_group_release(&group->up);
4981 }
4982
4983 void
4984 group_dpif_get_buckets(const struct group_dpif *group,
4985                        const struct list **buckets)
4986 {
4987     *buckets = &group->up.buckets;
4988 }
4989
4990 enum ofp11_group_type
4991 group_dpif_get_type(const struct group_dpif *group)
4992 {
4993     return group->up.type;
4994 }
4995 \f
4996 /* Sends 'packet' out 'ofport'.
4997  * May modify 'packet'.
4998  * Returns 0 if successful, otherwise a positive errno value. */
4999 int
5000 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5001 {
5002     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5003     int error;
5004
5005     error = xlate_send_packet(ofport, packet);
5006
5007     ovs_mutex_lock(&ofproto->stats_mutex);
5008     ofproto->stats.tx_packets++;
5009     ofproto->stats.tx_bytes += packet->size;
5010     ovs_mutex_unlock(&ofproto->stats_mutex);
5011     return error;
5012 }
5013
5014 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5015  * The action will state 'slow' as the reason that the action is in the slow
5016  * path.  (This is purely informational: it allows a human viewing "ovs-dpctl
5017  * dump-flows" output to see why a flow is in the slow path.)
5018  *
5019  * The 'stub_size' bytes in 'stub' will be used to store the action.
5020  * 'stub_size' must be large enough for the action.
5021  *
5022  * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5023  * respectively. */
5024 static void
5025 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5026                   enum slow_path_reason slow,
5027                   uint64_t *stub, size_t stub_size,
5028                   const struct nlattr **actionsp, size_t *actions_lenp)
5029 {
5030     union user_action_cookie cookie;
5031     struct ofpbuf buf;
5032
5033     cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5034     cookie.slow_path.unused = 0;
5035     cookie.slow_path.reason = slow;
5036
5037     ofpbuf_use_stack(&buf, stub, stub_size);
5038     if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5039         uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5040                                          ODPP_NONE);
5041         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5042     } else {
5043         odp_port_t odp_port;
5044         uint32_t pid;
5045
5046         odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5047         pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5048         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5049     }
5050     *actionsp = buf.data;
5051     *actions_lenp = buf.size;
5052 }
5053 \f
5054 static bool
5055 set_frag_handling(struct ofproto *ofproto_,
5056                   enum ofp_config_flags frag_handling)
5057 {
5058     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5059     if (frag_handling != OFPC_FRAG_REASM) {
5060         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5061         return true;
5062     } else {
5063         return false;
5064     }
5065 }
5066
5067 static enum ofperr
5068 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5069            const struct flow *flow,
5070            const struct ofpact *ofpacts, size_t ofpacts_len)
5071 {
5072     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5073
5074     ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts,
5075                                  ofpacts_len, packet);
5076     return 0;
5077 }
5078 \f
5079 /* NetFlow. */
5080
5081 static int
5082 set_netflow(struct ofproto *ofproto_,
5083             const struct netflow_options *netflow_options)
5084 {
5085     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5086
5087     if (netflow_options) {
5088         if (!ofproto->netflow) {
5089             ofproto->netflow = netflow_create();
5090             ofproto->backer->need_revalidate = REV_RECONFIGURE;
5091         }
5092         return netflow_set_options(ofproto->netflow, netflow_options);
5093     } else if (ofproto->netflow) {
5094         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5095         netflow_destroy(ofproto->netflow);
5096         ofproto->netflow = NULL;
5097     }
5098
5099     return 0;
5100 }
5101
5102 static void
5103 get_netflow_ids(const struct ofproto *ofproto_,
5104                 uint8_t *engine_type, uint8_t *engine_id)
5105 {
5106     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5107
5108     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5109 }
5110
5111 static void
5112 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5113 {
5114     if (!facet_is_controller_flow(facet) &&
5115         netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5116         struct subfacet *subfacet;
5117         struct ofexpired expired;
5118
5119         LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5120             if (subfacet->path == SF_FAST_PATH) {
5121                 struct dpif_flow_stats stats;
5122
5123                 subfacet_install(subfacet, &facet->xout.odp_actions,
5124                                  &stats);
5125                 subfacet_update_stats(subfacet, &stats);
5126             }
5127         }
5128
5129         expired.flow = facet->flow;
5130         expired.packet_count = facet->packet_count;
5131         expired.byte_count = facet->byte_count;
5132         expired.used = facet->used;
5133         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5134     }
5135 }
5136
5137 static void
5138 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5139 {
5140     struct cls_cursor cursor;
5141     struct facet *facet;
5142
5143     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5144     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5145     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5146         send_active_timeout(ofproto, facet);
5147     }
5148     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5149 }
5150 \f
5151 static struct ofproto_dpif *
5152 ofproto_dpif_lookup(const char *name)
5153 {
5154     struct ofproto_dpif *ofproto;
5155
5156     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5157                              hash_string(name, 0), &all_ofproto_dpifs) {
5158         if (!strcmp(ofproto->up.name, name)) {
5159             return ofproto;
5160         }
5161     }
5162     return NULL;
5163 }
5164
5165 static void
5166 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5167                           const char *argv[], void *aux OVS_UNUSED)
5168 {
5169     struct ofproto_dpif *ofproto;
5170
5171     if (argc > 1) {
5172         ofproto = ofproto_dpif_lookup(argv[1]);
5173         if (!ofproto) {
5174             unixctl_command_reply_error(conn, "no such bridge");
5175             return;
5176         }
5177         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5178         mac_learning_flush(ofproto->ml);
5179         ovs_rwlock_unlock(&ofproto->ml->rwlock);
5180     } else {
5181         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5182             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5183             mac_learning_flush(ofproto->ml);
5184             ovs_rwlock_unlock(&ofproto->ml->rwlock);
5185         }
5186     }
5187
5188     unixctl_command_reply(conn, "table successfully flushed");
5189 }
5190
5191 static struct ofport_dpif *
5192 ofbundle_get_a_port(const struct ofbundle *bundle)
5193 {
5194     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5195                         bundle_node);
5196 }
5197
5198 static void
5199 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5200                          const char *argv[], void *aux OVS_UNUSED)
5201 {
5202     struct ds ds = DS_EMPTY_INITIALIZER;
5203     const struct ofproto_dpif *ofproto;
5204     const struct mac_entry *e;
5205
5206     ofproto = ofproto_dpif_lookup(argv[1]);
5207     if (!ofproto) {
5208         unixctl_command_reply_error(conn, "no such bridge");
5209         return;
5210     }
5211
5212     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
5213     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5214     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5215         struct ofbundle *bundle = e->port.p;
5216         char name[OFP_MAX_PORT_NAME_LEN];
5217
5218         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5219                                name, sizeof name);
5220         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
5221                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
5222                       mac_entry_age(ofproto->ml, e));
5223     }
5224     ovs_rwlock_unlock(&ofproto->ml->rwlock);
5225     unixctl_command_reply(conn, ds_cstr(&ds));
5226     ds_destroy(&ds);
5227 }
5228
5229 struct trace_ctx {
5230     struct xlate_out xout;
5231     struct xlate_in xin;
5232     struct flow flow;
5233     struct ds *result;
5234 };
5235
5236 static void
5237 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5238 {
5239     struct rule_actions *actions;
5240     ovs_be64 cookie;
5241
5242     ds_put_char_multiple(result, '\t', level);
5243     if (!rule) {
5244         ds_put_cstr(result, "No match\n");
5245         return;
5246     }
5247
5248     ovs_mutex_lock(&rule->up.mutex);
5249     cookie = rule->up.flow_cookie;
5250     ovs_mutex_unlock(&rule->up.mutex);
5251
5252     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5253                   rule ? rule->up.table_id : 0, ntohll(cookie));
5254     cls_rule_format(&rule->up.cr, result);
5255     ds_put_char(result, '\n');
5256
5257     actions = rule_dpif_get_actions(rule);
5258
5259     ds_put_char_multiple(result, '\t', level);
5260     ds_put_cstr(result, "OpenFlow actions=");
5261     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
5262     ds_put_char(result, '\n');
5263
5264     rule_actions_unref(actions);
5265 }
5266
5267 static void
5268 trace_format_flow(struct ds *result, int level, const char *title,
5269                   struct trace_ctx *trace)
5270 {
5271     ds_put_char_multiple(result, '\t', level);
5272     ds_put_format(result, "%s: ", title);
5273     if (flow_equal(&trace->xin.flow, &trace->flow)) {
5274         ds_put_cstr(result, "unchanged");
5275     } else {
5276         flow_format(result, &trace->xin.flow);
5277         trace->flow = trace->xin.flow;
5278     }
5279     ds_put_char(result, '\n');
5280 }
5281
5282 static void
5283 trace_format_regs(struct ds *result, int level, const char *title,
5284                   struct trace_ctx *trace)
5285 {
5286     size_t i;
5287
5288     ds_put_char_multiple(result, '\t', level);
5289     ds_put_format(result, "%s:", title);
5290     for (i = 0; i < FLOW_N_REGS; i++) {
5291         ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]);
5292     }
5293     ds_put_char(result, '\n');
5294 }
5295
5296 static void
5297 trace_format_odp(struct ds *result, int level, const char *title,
5298                  struct trace_ctx *trace)
5299 {
5300     struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5301
5302     ds_put_char_multiple(result, '\t', level);
5303     ds_put_format(result, "%s: ", title);
5304     format_odp_actions(result, odp_actions->data, odp_actions->size);
5305     ds_put_char(result, '\n');
5306 }
5307
5308 static void
5309 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5310 {
5311     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5312     struct ds *result = trace->result;
5313
5314     ds_put_char(result, '\n');
5315     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5316     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5317     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
5318     trace_format_rule(result, recurse + 1, rule);
5319 }
5320
5321 static void
5322 trace_report(struct xlate_in *xin, const char *s, int recurse)
5323 {
5324     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5325     struct ds *result = trace->result;
5326
5327     ds_put_char_multiple(result, '\t', recurse);
5328     ds_put_cstr(result, s);
5329     ds_put_char(result, '\n');
5330 }
5331
5332 /* Parses the 'argc' elements of 'argv', ignoring argv[0].  The following
5333  * forms are supported:
5334  *
5335  *     - [dpname] odp_flow [-generate | packet]
5336  *     - bridge br_flow [-generate | packet]
5337  *
5338  * On success, initializes '*ofprotop' and 'flow' and returns NULL.  On failure
5339  * returns a nonnull error message. */
5340 static const char *
5341 parse_flow_and_packet(int argc, const char *argv[],
5342                       struct ofproto_dpif **ofprotop, struct flow *flow,
5343                       struct ofpbuf **packetp)
5344 {
5345     const struct dpif_backer *backer = NULL;
5346     const char *error = NULL;
5347     struct simap port_names = SIMAP_INITIALIZER(&port_names);
5348     struct ofpbuf *packet;
5349     struct ofpbuf odp_key;
5350     struct ofpbuf odp_mask;
5351
5352     ofpbuf_init(&odp_key, 0);
5353     ofpbuf_init(&odp_mask, 0);
5354
5355     /* Handle "-generate" or a hex string as the last argument. */
5356     if (!strcmp(argv[argc - 1], "-generate")) {
5357         packet = ofpbuf_new(0);
5358         argc--;
5359     } else {
5360         error = eth_from_hex(argv[argc - 1], &packet);
5361         if (!error) {
5362             argc--;
5363         } else if (argc == 4) {
5364             /* The 3-argument form must end in "-generate' or a hex string. */
5365             goto exit;
5366         }
5367     }
5368
5369     /* odp_flow can have its in_port specified as a name instead of port no.
5370      * We do not yet know whether a given flow is a odp_flow or a br_flow.
5371      * But, to know whether a flow is odp_flow through odp_flow_from_string(),
5372      * we need to create a simap of name to port no. */
5373     if (argc == 3) {
5374         const char *dp_type;
5375         if (!strncmp(argv[1], "ovs-", 4)) {
5376             dp_type = argv[1] + 4;
5377         } else {
5378             dp_type = argv[1];
5379         }
5380         backer = shash_find_data(&all_dpif_backers, dp_type);
5381     } else if (argc == 2) {
5382         struct shash_node *node;
5383         if (shash_count(&all_dpif_backers) == 1) {
5384             node = shash_first(&all_dpif_backers);
5385             backer = node->data;
5386         }
5387     } else {
5388         error = "Syntax error";
5389         goto exit;
5390     }
5391     if (backer && backer->dpif) {
5392         struct dpif_port dpif_port;
5393         struct dpif_port_dump port_dump;
5394         DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
5395             simap_put(&port_names, dpif_port.name,
5396                       odp_to_u32(dpif_port.port_no));
5397         }
5398     }
5399
5400     /* Parse the flow and determine whether a datapath or
5401      * bridge is specified. If function odp_flow_key_from_string()
5402      * returns 0, the flow is a odp_flow. If function
5403      * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5404     if (!odp_flow_from_string(argv[argc - 1], &port_names,
5405                               &odp_key, &odp_mask)) {
5406         if (!backer) {
5407             error = "Cannot find the datapath";
5408             goto exit;
5409         }
5410
5411         if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, flow,
5412                           NULL, ofprotop, NULL)) {
5413             error = "Invalid datapath flow";
5414             goto exit;
5415         }
5416     } else if (!parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL)) {
5417         if (argc != 3) {
5418             error = "Must specify bridge name";
5419             goto exit;
5420         }
5421
5422         *ofprotop = ofproto_dpif_lookup(argv[1]);
5423         if (!*ofprotop) {
5424             error = "Unknown bridge name";
5425             goto exit;
5426         }
5427     } else {
5428         error = "Bad flow syntax";
5429         goto exit;
5430     }
5431
5432     /* Generate a packet, if requested. */
5433     if (packet) {
5434         if (!packet->size) {
5435             flow_compose(packet, flow);
5436         } else {
5437             union flow_in_port in_port = flow->in_port;
5438
5439             /* Use the metadata from the flow and the packet argument
5440              * to reconstruct the flow. */
5441             flow_extract(packet, flow->skb_priority, flow->pkt_mark, NULL,
5442                          &in_port, flow);
5443         }
5444     }
5445
5446     error = NULL;
5447
5448 exit:
5449     if (error) {
5450         ofpbuf_delete(packet);
5451         packet = NULL;
5452     }
5453     *packetp = packet;
5454     ofpbuf_uninit(&odp_key);
5455     ofpbuf_uninit(&odp_mask);
5456     simap_destroy(&port_names);
5457     return error;
5458 }
5459
5460 static void
5461 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5462                       void *aux OVS_UNUSED)
5463 {
5464     struct ofproto_dpif *ofproto;
5465     struct ofpbuf *packet;
5466     const char *error;
5467     struct flow flow;
5468
5469     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
5470     if (!error) {
5471         struct ds result;
5472
5473         ds_init(&result);
5474         ofproto_trace(ofproto, &flow, packet, NULL, 0, &result);
5475         unixctl_command_reply(conn, ds_cstr(&result));
5476         ds_destroy(&result);
5477         ofpbuf_delete(packet);
5478     } else {
5479         unixctl_command_reply_error(conn, error);
5480     }
5481 }
5482
5483 static void
5484 ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc,
5485                               const char *argv[], void *aux OVS_UNUSED)
5486 {
5487     enum ofputil_protocol usable_protocols;
5488     struct ofproto_dpif *ofproto;
5489     bool enforce_consistency;
5490     struct ofpbuf ofpacts;
5491     struct ofpbuf *packet;
5492     struct ds result;
5493     struct flow flow;
5494     uint16_t in_port;
5495
5496     /* Three kinds of error return values! */
5497     enum ofperr retval;
5498     const char *error;
5499     char *rw_error;
5500
5501     packet = NULL;
5502     ds_init(&result);
5503     ofpbuf_init(&ofpacts, 0);
5504
5505     /* Parse actions. */
5506     rw_error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols);
5507     if (rw_error) {
5508         unixctl_command_reply_error(conn, rw_error);
5509         free(rw_error);
5510         goto exit;
5511     }
5512
5513     /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
5514      * consistency between the flow and the actions.  With -consistent, we
5515      * enforce consistency even for a flow supported in OpenFlow 1.0. */
5516     if (!strcmp(argv[1], "-consistent")) {
5517         enforce_consistency = true;
5518         argv++;
5519         argc--;
5520     } else {
5521         enforce_consistency = false;
5522     }
5523
5524     error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet);
5525     if (error) {
5526         unixctl_command_reply_error(conn, error);
5527         goto exit;
5528     }
5529
5530     /* Do the same checks as handle_packet_out() in ofproto.c.
5531      *
5532      * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
5533      * strictly correct because these actions aren't in any table, but it's OK
5534      * because it 'table_id' is used only to check goto_table instructions, but
5535      * packet-outs take a list of actions and therefore it can't include
5536      * instructions.
5537      *
5538      * We skip the "meter" check here because meter is an instruction, not an
5539      * action, and thus cannot appear in ofpacts. */
5540     in_port = ofp_to_u16(flow.in_port.ofp_port);
5541     if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) {
5542         unixctl_command_reply_error(conn, "invalid in_port");
5543         goto exit;
5544     }
5545     if (enforce_consistency) {
5546         retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size, &flow,
5547                                            u16_to_ofp(ofproto->up.max_ports),
5548                                            0, 0, usable_protocols);
5549     } else {
5550         retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow,
5551                                u16_to_ofp(ofproto->up.max_ports), 0, 0,
5552                                &usable_protocols);
5553     }
5554
5555     if (retval) {
5556         ds_clear(&result);
5557         ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval));
5558         unixctl_command_reply_error(conn, ds_cstr(&result));
5559         goto exit;
5560     }
5561
5562     ofproto_trace(ofproto, &flow, packet, ofpacts.data, ofpacts.size, &result);
5563     unixctl_command_reply(conn, ds_cstr(&result));
5564
5565 exit:
5566     ds_destroy(&result);
5567     ofpbuf_delete(packet);
5568     ofpbuf_uninit(&ofpacts);
5569 }
5570
5571 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
5572  * description of the results to 'ds'.
5573  *
5574  * The trace follows a packet with the specified 'flow' through the flow
5575  * table.  'packet' may be nonnull to trace an actual packet, with consequent
5576  * side effects (if it is nonnull then its flow must be 'flow').
5577  *
5578  * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
5579  * trace, otherwise the actions are determined by a flow table lookup. */
5580 static void
5581 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5582               const struct ofpbuf *packet,
5583               const struct ofpact ofpacts[], size_t ofpacts_len,
5584               struct ds *ds)
5585 {
5586     struct rule_dpif *rule;
5587     struct flow_wildcards wc;
5588
5589     ds_put_format(ds, "Bridge: %s\n", ofproto->up.name);
5590     ds_put_cstr(ds, "Flow: ");
5591     flow_format(ds, flow);
5592     ds_put_char(ds, '\n');
5593
5594     flow_wildcards_init_catchall(&wc);
5595     if (ofpacts) {
5596         rule = NULL;
5597     } else {
5598         rule_dpif_lookup(ofproto, flow, &wc, &rule);
5599
5600         trace_format_rule(ds, 0, rule);
5601         if (rule == ofproto->miss_rule) {
5602             ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5603         } else if (rule == ofproto->no_packet_in_rule) {
5604             ds_put_cstr(ds, "\nNo match, packets dropped because "
5605                         "OFPPC_NO_PACKET_IN is set on in_port.\n");
5606         } else if (rule == ofproto->drop_frags_rule) {
5607             ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5608                         "and the fragment handling mode is \"drop\".\n");
5609         }
5610     }
5611
5612     if (rule || ofpacts) {
5613         uint64_t odp_actions_stub[1024 / 8];
5614         struct ofpbuf odp_actions;
5615         struct trace_ctx trace;
5616         struct match match;
5617         uint16_t tcp_flags;
5618
5619         tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5620         trace.result = ds;
5621         trace.flow = *flow;
5622         ofpbuf_use_stub(&odp_actions,
5623                         odp_actions_stub, sizeof odp_actions_stub);
5624         xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5625         if (ofpacts) {
5626             trace.xin.ofpacts = ofpacts;
5627             trace.xin.ofpacts_len = ofpacts_len;
5628         }
5629         trace.xin.resubmit_hook = trace_resubmit;
5630         trace.xin.report_hook = trace_report;
5631
5632         xlate_actions(&trace.xin, &trace.xout);
5633         flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5634
5635         ds_put_char(ds, '\n');
5636         trace_format_flow(ds, 0, "Final flow", &trace);
5637
5638         match_init(&match, flow, &trace.xout.wc);
5639         ds_put_cstr(ds, "Relevant fields: ");
5640         match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5641         ds_put_char(ds, '\n');
5642
5643         ds_put_cstr(ds, "Datapath actions: ");
5644         format_odp_actions(ds, trace.xout.odp_actions.data,
5645                            trace.xout.odp_actions.size);
5646
5647         if (trace.xout.slow) {
5648             enum slow_path_reason slow;
5649
5650             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5651                         "slow path because it:");
5652
5653             slow = trace.xout.slow;
5654             while (slow) {
5655                 enum slow_path_reason bit = rightmost_1bit(slow);
5656
5657                 ds_put_format(ds, "\n\t- %s.",
5658                               slow_path_reason_to_explanation(bit));
5659
5660                 slow &= ~bit;
5661             }
5662         }
5663
5664         xlate_out_uninit(&trace.xout);
5665     }
5666
5667     rule_dpif_unref(rule);
5668 }
5669
5670 /* Runs a self-check of flow translations in 'ofproto'.  Appends a message to
5671  * 'reply' describing the results. */
5672 static void
5673 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5674 {
5675     struct cls_cursor cursor;
5676     struct facet *facet;
5677     int errors;
5678
5679     errors = 0;
5680     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5681     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5682     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5683         if (!facet_check_consistency(facet)) {
5684             errors++;
5685         }
5686     }
5687     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5688     if (errors) {
5689         ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5690     }
5691
5692     if (errors) {
5693         ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5694                       ofproto->up.name, errors);
5695     } else {
5696         ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5697     }
5698 }
5699
5700 static void
5701 ofproto_dpif_self_check(struct unixctl_conn *conn,
5702                         int argc, const char *argv[], void *aux OVS_UNUSED)
5703 {
5704     struct ds reply = DS_EMPTY_INITIALIZER;
5705     struct ofproto_dpif *ofproto;
5706
5707     if (argc > 1) {
5708         ofproto = ofproto_dpif_lookup(argv[1]);
5709         if (!ofproto) {
5710             unixctl_command_reply_error(conn, "Unknown ofproto (use "
5711                                         "ofproto/list for help)");
5712             return;
5713         }
5714         ofproto_dpif_self_check__(ofproto, &reply);
5715     } else {
5716         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5717             ofproto_dpif_self_check__(ofproto, &reply);
5718         }
5719     }
5720
5721     unixctl_command_reply(conn, ds_cstr(&reply));
5722     ds_destroy(&reply);
5723 }
5724
5725 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
5726  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
5727  * to destroy 'ofproto_shash' and free the returned value. */
5728 static const struct shash_node **
5729 get_ofprotos(struct shash *ofproto_shash)
5730 {
5731     const struct ofproto_dpif *ofproto;
5732
5733     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5734         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5735         shash_add_nocopy(ofproto_shash, name, ofproto);
5736     }
5737
5738     return shash_sort(ofproto_shash);
5739 }
5740
5741 static void
5742 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5743                               const char *argv[] OVS_UNUSED,
5744                               void *aux OVS_UNUSED)
5745 {
5746     struct ds ds = DS_EMPTY_INITIALIZER;
5747     struct shash ofproto_shash;
5748     const struct shash_node **sorted_ofprotos;
5749     int i;
5750
5751     shash_init(&ofproto_shash);
5752     sorted_ofprotos = get_ofprotos(&ofproto_shash);
5753     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5754         const struct shash_node *node = sorted_ofprotos[i];
5755         ds_put_format(&ds, "%s\n", node->name);
5756     }
5757
5758     shash_destroy(&ofproto_shash);
5759     free(sorted_ofprotos);
5760
5761     unixctl_command_reply(conn, ds_cstr(&ds));
5762     ds_destroy(&ds);
5763 }
5764
5765 static void
5766 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5767 {
5768     const struct shash_node **ofprotos;
5769     struct ofproto_dpif *ofproto;
5770     struct shash ofproto_shash;
5771     uint64_t n_hit, n_missed;
5772     size_t i;
5773
5774     n_hit = n_missed = 0;
5775     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5776         if (ofproto->backer == backer) {
5777             n_missed += ofproto->n_missed;
5778             n_hit += ofproto->n_hit;
5779         }
5780     }
5781
5782     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5783                   dpif_name(backer->dpif), n_hit, n_missed);
5784     ds_put_format(ds, "\tflows: cur: %"PRIuSIZE", avg: %u, max: %u,"
5785                   " life span: %lldms\n", hmap_count(&backer->subfacets),
5786                   backer->avg_n_subfacet, backer->max_n_subfacet,
5787                   backer->avg_subfacet_life);
5788
5789     shash_init(&ofproto_shash);
5790     ofprotos = get_ofprotos(&ofproto_shash);
5791     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5792         struct ofproto_dpif *ofproto = ofprotos[i]->data;
5793         const struct shash_node **ports;
5794         size_t j;
5795
5796         if (ofproto->backer != backer) {
5797             continue;
5798         }
5799
5800         ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5801                       ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5802
5803         ports = shash_sort(&ofproto->up.port_by_name);
5804         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5805             const struct shash_node *node = ports[j];
5806             struct ofport *ofport = node->data;
5807             struct smap config;
5808             odp_port_t odp_port;
5809
5810             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5811                           ofport->ofp_port);
5812
5813             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5814             if (odp_port != ODPP_NONE) {
5815                 ds_put_format(ds, "%"PRIu32":", odp_port);
5816             } else {
5817                 ds_put_cstr(ds, "none:");
5818             }
5819
5820             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5821
5822             smap_init(&config);
5823             if (!netdev_get_config(ofport->netdev, &config)) {
5824                 const struct smap_node **nodes;
5825                 size_t i;
5826
5827                 nodes = smap_sort(&config);
5828                 for (i = 0; i < smap_count(&config); i++) {
5829                     const struct smap_node *node = nodes[i];
5830                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5831                                   node->key, node->value);
5832                 }
5833                 free(nodes);
5834             }
5835             smap_destroy(&config);
5836
5837             ds_put_char(ds, ')');
5838             ds_put_char(ds, '\n');
5839         }
5840         free(ports);
5841     }
5842     shash_destroy(&ofproto_shash);
5843     free(ofprotos);
5844 }
5845
5846 static void
5847 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5848                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5849 {
5850     struct ds ds = DS_EMPTY_INITIALIZER;
5851     const struct shash_node **backers;
5852     int i;
5853
5854     backers = shash_sort(&all_dpif_backers);
5855     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5856         dpif_show_backer(backers[i]->data, &ds);
5857     }
5858     free(backers);
5859
5860     unixctl_command_reply(conn, ds_cstr(&ds));
5861     ds_destroy(&ds);
5862 }
5863
5864 /* Dump the megaflow (facet) cache.  This is useful to check the
5865  * correctness of flow wildcarding, since the same mechanism is used for
5866  * both xlate caching and kernel wildcarding.
5867  *
5868  * It's important to note that in the output the flow description uses
5869  * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5870  *
5871  * This command is only needed for advanced debugging, so it's not
5872  * documented in the man page. */
5873 static void
5874 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5875                                     int argc OVS_UNUSED, const char *argv[],
5876                                     void *aux OVS_UNUSED)
5877 {
5878     struct ds ds = DS_EMPTY_INITIALIZER;
5879     const struct ofproto_dpif *ofproto;
5880     long long int now = time_msec();
5881     struct cls_cursor cursor;
5882     struct facet *facet;
5883
5884     ofproto = ofproto_dpif_lookup(argv[1]);
5885     if (!ofproto) {
5886         unixctl_command_reply_error(conn, "no such bridge");
5887         return;
5888     }
5889
5890     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5891     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5892     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5893         cls_rule_format(&facet->cr, &ds);
5894         ds_put_cstr(&ds, ", ");
5895         ds_put_format(&ds, "n_subfacets:%"PRIuSIZE", ", list_size(&facet->subfacets));
5896         ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5897         ds_put_cstr(&ds, "Datapath actions: ");
5898         if (facet->xout.slow) {
5899             uint64_t slow_path_stub[128 / 8];
5900             const struct nlattr *actions;
5901             size_t actions_len;
5902
5903             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5904                               slow_path_stub, sizeof slow_path_stub,
5905                               &actions, &actions_len);
5906             format_odp_actions(&ds, actions, actions_len);
5907         } else {
5908             format_odp_actions(&ds, facet->xout.odp_actions.data,
5909                                facet->xout.odp_actions.size);
5910         }
5911         ds_put_cstr(&ds, "\n");
5912     }
5913     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5914
5915     ds_chomp(&ds, '\n');
5916     unixctl_command_reply(conn, ds_cstr(&ds));
5917     ds_destroy(&ds);
5918 }
5919
5920 /* Disable using the megaflows.
5921  *
5922  * This command is only needed for advanced debugging, so it's not
5923  * documented in the man page. */
5924 static void
5925 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5926                                        int argc OVS_UNUSED,
5927                                        const char *argv[] OVS_UNUSED,
5928                                        void *aux OVS_UNUSED)
5929 {
5930     struct ofproto_dpif *ofproto;
5931
5932     enable_megaflows = false;
5933
5934     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5935         flush(&ofproto->up);
5936     }
5937
5938     unixctl_command_reply(conn, "megaflows disabled");
5939 }
5940
5941 /* Re-enable using megaflows.
5942  *
5943  * This command is only needed for advanced debugging, so it's not
5944  * documented in the man page. */
5945 static void
5946 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5947                                       int argc OVS_UNUSED,
5948                                       const char *argv[] OVS_UNUSED,
5949                                       void *aux OVS_UNUSED)
5950 {
5951     struct ofproto_dpif *ofproto;
5952
5953     enable_megaflows = true;
5954
5955     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5956         flush(&ofproto->up);
5957     }
5958
5959     unixctl_command_reply(conn, "megaflows enabled");
5960 }
5961
5962 static void
5963 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5964                                 int argc OVS_UNUSED, const char *argv[],
5965                                 void *aux OVS_UNUSED)
5966 {
5967     struct ds ds = DS_EMPTY_INITIALIZER;
5968     const struct ofproto_dpif *ofproto;
5969     struct subfacet *subfacet;
5970
5971     ofproto = ofproto_dpif_lookup(argv[1]);
5972     if (!ofproto) {
5973         unixctl_command_reply_error(conn, "no such bridge");
5974         return;
5975     }
5976
5977     update_stats(ofproto->backer);
5978
5979     HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5980         struct facet *facet = subfacet->facet;
5981         struct odputil_keybuf maskbuf;
5982         struct ofpbuf mask;
5983
5984         if (facet->ofproto != ofproto) {
5985             continue;
5986         }
5987
5988         ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5989         if (enable_megaflows) {
5990             odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5991                                    &facet->flow, UINT32_MAX);
5992         }
5993
5994         odp_flow_format(subfacet->key, subfacet->key_len,
5995                         mask.data, mask.size, NULL, &ds, false);
5996
5997         ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5998                       subfacet->dp_packet_count, subfacet->dp_byte_count);
5999         if (subfacet->used) {
6000             ds_put_format(&ds, "%.3fs",
6001                           (time_msec() - subfacet->used) / 1000.0);
6002         } else {
6003             ds_put_format(&ds, "never");
6004         }
6005         if (subfacet->facet->tcp_flags) {
6006             ds_put_cstr(&ds, ", flags:");
6007             packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
6008         }
6009
6010         ds_put_cstr(&ds, ", actions:");
6011         if (facet->xout.slow) {
6012             uint64_t slow_path_stub[128 / 8];
6013             const struct nlattr *actions;
6014             size_t actions_len;
6015
6016             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
6017                               slow_path_stub, sizeof slow_path_stub,
6018                               &actions, &actions_len);
6019             format_odp_actions(&ds, actions, actions_len);
6020         } else {
6021             format_odp_actions(&ds, facet->xout.odp_actions.data,
6022                                facet->xout.odp_actions.size);
6023         }
6024         ds_put_char(&ds, '\n');
6025     }
6026
6027     unixctl_command_reply(conn, ds_cstr(&ds));
6028     ds_destroy(&ds);
6029 }
6030
6031 static void
6032 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
6033                                int argc OVS_UNUSED, const char *argv[],
6034                                void *aux OVS_UNUSED)
6035 {
6036     struct ds ds = DS_EMPTY_INITIALIZER;
6037     struct ofproto_dpif *ofproto;
6038
6039     ofproto = ofproto_dpif_lookup(argv[1]);
6040     if (!ofproto) {
6041         unixctl_command_reply_error(conn, "no such bridge");
6042         return;
6043     }
6044
6045     flush(&ofproto->up);
6046
6047     unixctl_command_reply(conn, ds_cstr(&ds));
6048     ds_destroy(&ds);
6049 }
6050
6051 static void
6052 ofproto_dpif_unixctl_init(void)
6053 {
6054     static bool registered;
6055     if (registered) {
6056         return;
6057     }
6058     registered = true;
6059
6060     unixctl_command_register(
6061         "ofproto/trace",
6062         "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
6063         1, 3, ofproto_unixctl_trace, NULL);
6064     unixctl_command_register(
6065         "ofproto/trace-packet-out",
6066         "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
6067         2, 6, ofproto_unixctl_trace_actions, NULL);
6068     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
6069                              ofproto_unixctl_fdb_flush, NULL);
6070     unixctl_command_register("fdb/show", "bridge", 1, 1,
6071                              ofproto_unixctl_fdb_show, NULL);
6072     unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
6073                              ofproto_dpif_self_check, NULL);
6074     unixctl_command_register("dpif/dump-dps", "", 0, 0,
6075                              ofproto_unixctl_dpif_dump_dps, NULL);
6076     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
6077                              NULL);
6078     unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
6079                              ofproto_unixctl_dpif_dump_flows, NULL);
6080     unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
6081                              ofproto_unixctl_dpif_del_flows, NULL);
6082     unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
6083                              ofproto_unixctl_dpif_dump_megaflows, NULL);
6084     unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
6085                              ofproto_unixctl_dpif_disable_megaflows, NULL);
6086     unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
6087                              ofproto_unixctl_dpif_enable_megaflows, NULL);
6088 }
6089 \f
6090 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6091  *
6092  * This is deprecated.  It is only for compatibility with broken device drivers
6093  * in old versions of Linux that do not properly support VLANs when VLAN
6094  * devices are not used.  When broken device drivers are no longer in
6095  * widespread use, we will delete these interfaces. */
6096
6097 static int
6098 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
6099 {
6100     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
6101     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
6102
6103     if (realdev_ofp_port == ofport->realdev_ofp_port
6104         && vid == ofport->vlandev_vid) {
6105         return 0;
6106     }
6107
6108     ofproto->backer->need_revalidate = REV_RECONFIGURE;
6109
6110     if (ofport->realdev_ofp_port) {
6111         vsp_remove(ofport);
6112     }
6113     if (realdev_ofp_port && ofport->bundle) {
6114         /* vlandevs are enslaved to their realdevs, so they are not allowed to
6115          * themselves be part of a bundle. */
6116         bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6117     }
6118
6119     ofport->realdev_ofp_port = realdev_ofp_port;
6120     ofport->vlandev_vid = vid;
6121
6122     if (realdev_ofp_port) {
6123         vsp_add(ofport, realdev_ofp_port, vid);
6124     }
6125
6126     return 0;
6127 }
6128
6129 static uint32_t
6130 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6131 {
6132     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6133 }
6134
6135 bool
6136 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6137     OVS_EXCLUDED(ofproto->vsp_mutex)
6138 {
6139     bool ret;
6140
6141     ovs_mutex_lock(&ofproto->vsp_mutex);
6142     ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6143     ovs_mutex_unlock(&ofproto->vsp_mutex);
6144     return ret;
6145 }
6146
6147 static ofp_port_t
6148 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6149                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6150     OVS_REQUIRES(ofproto->vsp_mutex)
6151 {
6152     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6153         int vid = vlan_tci_to_vid(vlan_tci);
6154         const struct vlan_splinter *vsp;
6155
6156         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6157                                  hash_realdev_vid(realdev_ofp_port, vid),
6158                                  &ofproto->realdev_vid_map) {
6159             if (vsp->realdev_ofp_port == realdev_ofp_port
6160                 && vsp->vid == vid) {
6161                 return vsp->vlandev_ofp_port;
6162             }
6163         }
6164     }
6165     return realdev_ofp_port;
6166 }
6167
6168 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6169  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6170  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
6171  * 'vlan_tci' 9, it would return the port number of eth0.9.
6172  *
6173  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6174  * function just returns its 'realdev_ofp_port' argument. */
6175 ofp_port_t
6176 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6177                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6178     OVS_EXCLUDED(ofproto->vsp_mutex)
6179 {
6180     ofp_port_t ret;
6181
6182     ovs_mutex_lock(&ofproto->vsp_mutex);
6183     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6184     ovs_mutex_unlock(&ofproto->vsp_mutex);
6185     return ret;
6186 }
6187
6188 static struct vlan_splinter *
6189 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6190 {
6191     struct vlan_splinter *vsp;
6192
6193     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6194                              hash_ofp_port(vlandev_ofp_port),
6195                              &ofproto->vlandev_map) {
6196         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6197             return vsp;
6198         }
6199     }
6200
6201     return NULL;
6202 }
6203
6204 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6205  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6206  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
6207  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6208  * eth0 and store 9 in '*vid'.
6209  *
6210  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6211  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
6212  * always does.*/
6213 static ofp_port_t
6214 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6215                        ofp_port_t vlandev_ofp_port, int *vid)
6216     OVS_REQUIRES(ofproto->vsp_mutex)
6217 {
6218     if (!hmap_is_empty(&ofproto->vlandev_map)) {
6219         const struct vlan_splinter *vsp;
6220
6221         vsp = vlandev_find(ofproto, vlandev_ofp_port);
6222         if (vsp) {
6223             if (vid) {
6224                 *vid = vsp->vid;
6225             }
6226             return vsp->realdev_ofp_port;
6227         }
6228     }
6229     return 0;
6230 }
6231
6232 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6233  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
6234  * 'flow->in_port' to the "real" device backing the VLAN device, sets
6235  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Otherwise (which is
6236  * always the case unless VLAN splinters are enabled), returns false without
6237  * making any changes. */
6238 bool
6239 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6240     OVS_EXCLUDED(ofproto->vsp_mutex)
6241 {
6242     ofp_port_t realdev;
6243     int vid;
6244
6245     ovs_mutex_lock(&ofproto->vsp_mutex);
6246     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6247     ovs_mutex_unlock(&ofproto->vsp_mutex);
6248     if (!realdev) {
6249         return false;
6250     }
6251
6252     /* Cause the flow to be processed as if it came in on the real device with
6253      * the VLAN device's VLAN ID. */
6254     flow->in_port.ofp_port = realdev;
6255     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6256     return true;
6257 }
6258
6259 static void
6260 vsp_remove(struct ofport_dpif *port)
6261 {
6262     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6263     struct vlan_splinter *vsp;
6264
6265     ovs_mutex_lock(&ofproto->vsp_mutex);
6266     vsp = vlandev_find(ofproto, port->up.ofp_port);
6267     if (vsp) {
6268         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6269         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6270         free(vsp);
6271
6272         port->realdev_ofp_port = 0;
6273     } else {
6274         VLOG_ERR("missing vlan device record");
6275     }
6276     ovs_mutex_unlock(&ofproto->vsp_mutex);
6277 }
6278
6279 static void
6280 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6281 {
6282     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6283
6284     ovs_mutex_lock(&ofproto->vsp_mutex);
6285     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6286         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6287             == realdev_ofp_port)) {
6288         struct vlan_splinter *vsp;
6289
6290         vsp = xmalloc(sizeof *vsp);
6291         vsp->realdev_ofp_port = realdev_ofp_port;
6292         vsp->vlandev_ofp_port = port->up.ofp_port;
6293         vsp->vid = vid;
6294
6295         port->realdev_ofp_port = realdev_ofp_port;
6296
6297         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6298                     hash_ofp_port(port->up.ofp_port));
6299         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6300                     hash_realdev_vid(realdev_ofp_port, vid));
6301     } else {
6302         VLOG_ERR("duplicate vlan device record");
6303     }
6304     ovs_mutex_unlock(&ofproto->vsp_mutex);
6305 }
6306
6307 static odp_port_t
6308 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6309 {
6310     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6311     return ofport ? ofport->odp_port : ODPP_NONE;
6312 }
6313
6314 struct ofport_dpif *
6315 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6316 {
6317     struct ofport_dpif *port;
6318
6319     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6320     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6321                              &backer->odp_to_ofport_map) {
6322         if (port->odp_port == odp_port) {
6323             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6324             return port;
6325         }
6326     }
6327
6328     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6329     return NULL;
6330 }
6331
6332 static ofp_port_t
6333 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6334 {
6335     struct ofport_dpif *port;
6336
6337     port = odp_port_to_ofport(ofproto->backer, odp_port);
6338     if (port && &ofproto->up == port->up.ofproto) {
6339         return port->up.ofp_port;
6340     } else {
6341         return OFPP_NONE;
6342     }
6343 }
6344
6345 const struct ofproto_class ofproto_dpif_class = {
6346     init,
6347     enumerate_types,
6348     enumerate_names,
6349     del,
6350     port_open_type,
6351     type_run,
6352     type_run_fast,
6353     type_wait,
6354     alloc,
6355     construct,
6356     destruct,
6357     dealloc,
6358     run,
6359     run_fast,
6360     wait,
6361     get_memory_usage,
6362     flush,
6363     get_features,
6364     get_tables,
6365     port_alloc,
6366     port_construct,
6367     port_destruct,
6368     port_dealloc,
6369     port_modified,
6370     port_reconfigured,
6371     port_query_by_name,
6372     port_add,
6373     port_del,
6374     port_get_stats,
6375     port_dump_start,
6376     port_dump_next,
6377     port_dump_done,
6378     port_poll,
6379     port_poll_wait,
6380     port_is_lacp_current,
6381     NULL,                       /* rule_choose_table */
6382     rule_alloc,
6383     rule_construct,
6384     rule_insert,
6385     rule_delete,
6386     rule_destruct,
6387     rule_dealloc,
6388     rule_get_stats,
6389     rule_execute,
6390     rule_modify_actions,
6391     set_frag_handling,
6392     packet_out,
6393     set_netflow,
6394     get_netflow_ids,
6395     set_sflow,
6396     set_ipfix,
6397     set_cfm,
6398     get_cfm_status,
6399     set_bfd,
6400     get_bfd_status,
6401     set_stp,
6402     get_stp_status,
6403     set_stp_port,
6404     get_stp_port_status,
6405     get_stp_port_stats,
6406     set_queues,
6407     bundle_set,
6408     bundle_remove,
6409     mirror_set__,
6410     mirror_get_stats__,
6411     set_flood_vlans,
6412     is_mirror_output_bundle,
6413     forward_bpdu_changed,
6414     set_mac_table_config,
6415     set_realdev,
6416     NULL,                       /* meter_get_features */
6417     NULL,                       /* meter_set */
6418     NULL,                       /* meter_get */
6419     NULL,                       /* meter_del */
6420     group_alloc,                /* group_alloc */
6421     group_construct,            /* group_construct */
6422     group_destruct,             /* group_destruct */
6423     group_dealloc,              /* group_dealloc */
6424     group_modify,               /* group_modify */
6425     group_get_stats,            /* group_get_stats */
6426 };