ofproto/ofproto-dpif.c

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