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