ofproto/ofproto-dpif.c

   1 /*
   2  * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18
  19 #include "ofproto/ofproto-dpif.h"
  20 #include "ofproto/ofproto-provider.h"
  21
  22 #include <errno.h>
  23
  24 #include "bfd.h"
  25 #include "bond.h"
  26 #include "bundle.h"
  27 #include "byte-order.h"
  28 #include "connmgr.h"
  29 #include "coverage.h"
  30 #include "cfm.h"
  31 #include "dpif.h"
  32 #include "dynamic-string.h"
  33 #include "fail-open.h"
  34 #include "hmapx.h"
  35 #include "lacp.h"
  36 #include "learn.h"
  37 #include "mac-learning.h"
  38 #include "meta-flow.h"
  39 #include "multipath.h"
  40 #include "netdev-vport.h"
  41 #include "netdev.h"
  42 #include "netlink.h"
  43 #include "nx-match.h"
  44 #include "odp-util.h"
  45 #include "odp-execute.h"
  46 #include "ofp-util.h"
  47 #include "ofpbuf.h"
  48 #include "ofp-actions.h"
  49 #include "ofp-parse.h"
  50 #include "ofp-print.h"
  51 #include "ofproto-dpif-governor.h"
  52 #include "ofproto-dpif-ipfix.h"
  53 #include "ofproto-dpif-mirror.h"
  54 #include "ofproto-dpif-sflow.h"
  55 #include "ofproto-dpif-upcall.h"
  56 #include "ofproto-dpif-xlate.h"
  57 #include "poll-loop.h"
  58 #include "simap.h"
  59 #include "smap.h"
  60 #include "timer.h"
  61 #include "tunnel.h"
  62 #include "unaligned.h"
  63 #include "unixctl.h"
  64 #include "vlan-bitmap.h"
  65 #include "vlog.h"
  66
  67 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  68
  69 COVERAGE_DEFINE(ofproto_dpif_expired);
  70 COVERAGE_DEFINE(facet_changed_rule);
  71 COVERAGE_DEFINE(facet_revalidate);
  72 COVERAGE_DEFINE(facet_unexpected);
  73 COVERAGE_DEFINE(facet_suppress);
  74 COVERAGE_DEFINE(subfacet_install_fail);
  75 COVERAGE_DEFINE(packet_in_overflow);
  76 COVERAGE_DEFINE(flow_mod_overflow);
  77
  78 /* Number of implemented OpenFlow tables. */
  79 enum { N_TABLES = 255 };
  80 enum { TBL_INTERNAL = N_TABLES - 1 };    /* Used for internal hidden rules. */
  81 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
  82
  83 struct flow_miss;
  84 struct facet;
  85
  86 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
  87
  88 struct ofbundle {
  89     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
  90     struct ofproto_dpif *ofproto; /* Owning ofproto. */
  91     void *aux;                  /* Key supplied by ofproto's client. */
  92     char *name;                 /* Identifier for log messages. */
  93
  94     /* Configuration. */
  95     struct list ports;          /* Contains "struct ofport"s. */
  96     enum port_vlan_mode vlan_mode; /* VLAN mode */
  97     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
  98     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
  99                                  * NULL if all VLANs are trunked. */
 100     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 101     struct bond *bond;          /* Nonnull iff more than one port. */
 102     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 103
 104     /* Status. */
 105     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 106 };
 107
 108 static void bundle_remove(struct ofport *);
 109 static void bundle_update(struct ofbundle *);
 110 static void bundle_destroy(struct ofbundle *);
 111 static void bundle_del_port(struct ofport_dpif *);
 112 static void bundle_run(struct ofbundle *);
 113 static void bundle_wait(struct ofbundle *);
 114
 115 static void stp_run(struct ofproto_dpif *ofproto);
 116 static void stp_wait(struct ofproto_dpif *ofproto);
 117 static int set_stp_port(struct ofport *,
 118                         const struct ofproto_port_stp_settings *);
 119
 120 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
 121                               enum slow_path_reason,
 122                               uint64_t *stub, size_t stub_size,
 123                               const struct nlattr **actionsp,
 124                               size_t *actions_lenp);
 125
 126 /* A subfacet (see "struct subfacet" below) has three possible installation
 127  * states:
 128  *
 129  *   - SF_NOT_INSTALLED: Not installed in the datapath.  This will only be the
 130  *     case just after the subfacet is created, just before the subfacet is
 131  *     destroyed, or if the datapath returns an error when we try to install a
 132  *     subfacet.
 133  *
 134  *   - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
 135  *
 136  *   - SF_SLOW_PATH: An action that sends every packet for the subfacet through
 137  *     ofproto_dpif is installed in the datapath.
 138  */
 139 enum subfacet_path {
 140     SF_NOT_INSTALLED,           /* No datapath flow for this subfacet. */
 141     SF_FAST_PATH,               /* Full actions are installed. */
 142     SF_SLOW_PATH,               /* Send-to-userspace action is installed. */
 143 };
 144
 145 /* A dpif flow and actions associated with a facet.
 146  *
 147  * See also the large comment on struct facet. */
 148 struct subfacet {
 149     /* Owners. */
 150     struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
 151     struct list list_node;      /* In struct facet's 'facets' list. */
 152     struct facet *facet;        /* Owning facet. */
 153     struct dpif_backer *backer; /* Owning backer. */
 154
 155     enum odp_key_fitness key_fitness;
 156     struct nlattr *key;
 157     int key_len;
 158
 159     long long int used;         /* Time last used; time created if not used. */
 160     long long int created;      /* Time created. */
 161
 162     uint64_t dp_packet_count;   /* Last known packet count in the datapath. */
 163     uint64_t dp_byte_count;     /* Last known byte count in the datapath. */
 164
 165     enum subfacet_path path;    /* Installed in datapath? */
 166 };
 167
 168 #define SUBFACET_DESTROY_MAX_BATCH 50
 169
 170 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *);
 171 static struct subfacet *subfacet_find(struct dpif_backer *,
 172                                       const struct nlattr *key, size_t key_len,
 173                                       uint32_t key_hash);
 174 static void subfacet_destroy(struct subfacet *);
 175 static void subfacet_destroy__(struct subfacet *);
 176 static void subfacet_destroy_batch(struct dpif_backer *,
 177                                    struct subfacet **, int n);
 178 static void subfacet_reset_dp_stats(struct subfacet *,
 179                                     struct dpif_flow_stats *);
 180 static void subfacet_update_stats(struct subfacet *,
 181                                   const struct dpif_flow_stats *);
 182 static int subfacet_install(struct subfacet *,
 183                             const struct ofpbuf *odp_actions,
 184                             struct dpif_flow_stats *);
 185 static void subfacet_uninstall(struct subfacet *);
 186
 187 /* A unique, non-overlapping instantiation of an OpenFlow flow.
 188  *
 189  * A facet associates a "struct flow", which represents the Open vSwitch
 190  * userspace idea of an exact-match flow, with one or more subfacets.
 191  * While the facet is created based on an exact-match flow, it is stored
 192  * within the ofproto based on the wildcards that could be expressed
 193  * based on the flow table and other configuration.  (See the 'wc'
 194  * description in "struct xlate_out" for more details.)
 195  *
 196  * Each subfacet tracks the datapath's idea of the flow equivalent to
 197  * the facet.  When the kernel module (or other dpif implementation) and
 198  * Open vSwitch userspace agree on the definition of a flow key, there
 199  * is exactly one subfacet per facet.  If the dpif implementation
 200  * supports more-specific flow matching than userspace, however, a facet
 201  * can have more than one subfacet.  Examples include the dpif
 202  * implementation not supporting the same wildcards as userspace or some
 203  * distinction in flow that userspace simply doesn't understand.
 204  *
 205  * Flow expiration works in terms of subfacets, so a facet must have at
 206  * least one subfacet or it will never expire, leaking memory. */
 207 struct facet {
 208     /* Owners. */
 209     struct hmap_node hmap_node;  /* In owning ofproto's 'facets' hmap. */
 210     struct ofproto_dpif *ofproto;
 211
 212     /* Owned data. */
 213     struct list subfacets;
 214     long long int used;         /* Time last used; time created if not used. */
 215
 216     /* Key. */
 217     struct flow flow;           /* Flow of the creating subfacet. */
 218     struct cls_rule cr;         /* In 'ofproto_dpif's facets classifier. */
 219
 220     /* These statistics:
 221      *
 222      *   - Do include packets and bytes sent "by hand", e.g. with
 223      *     dpif_execute().
 224      *
 225      *   - Do include packets and bytes that were obtained from the datapath
 226      *     when a subfacet's statistics were reset (e.g. dpif_flow_put() with
 227      *     DPIF_FP_ZERO_STATS).
 228      *
 229      *   - Do not include packets or bytes that can be obtained from the
 230      *     datapath for any existing subfacet.
 231      */
 232     uint64_t packet_count;       /* Number of packets received. */
 233     uint64_t byte_count;         /* Number of bytes received. */
 234
 235     /* Resubmit statistics. */
 236     uint64_t prev_packet_count;  /* Number of packets from last stats push. */
 237     uint64_t prev_byte_count;    /* Number of bytes from last stats push. */
 238     long long int prev_used;     /* Used time from last stats push. */
 239
 240     /* Accounting. */
 241     uint64_t accounted_bytes;    /* Bytes processed by facet_account(). */
 242     struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
 243     uint8_t tcp_flags;           /* TCP flags seen for this 'rule'. */
 244
 245     struct xlate_out xout;
 246
 247     /* Storage for a single subfacet, to reduce malloc() time and space
 248      * overhead.  (A facet always has at least one subfacet and in the common
 249      * case has exactly one subfacet.  However, 'one_subfacet' may not
 250      * always be valid, since it could have been removed after newer
 251      * subfacets were pushed onto the 'subfacets' list.) */
 252     struct subfacet one_subfacet;
 253
 254     long long int learn_rl;      /* Rate limiter for facet_learn(). */
 255 };
 256
 257 static struct facet *facet_create(const struct flow_miss *);
 258 static void facet_remove(struct facet *);
 259 static void facet_free(struct facet *);
 260
 261 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
 262 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
 263                                         const struct flow *);
 264 static bool facet_revalidate(struct facet *);
 265 static bool facet_check_consistency(struct facet *);
 266
 267 static void facet_flush_stats(struct facet *);
 268
 269 static void facet_reset_counters(struct facet *);
 270 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
 271                             struct dpif_flow_stats *, bool may_learn);
 272 static void facet_push_stats(struct facet *, bool may_learn);
 273 static void facet_learn(struct facet *);
 274 static void facet_account(struct facet *);
 275 static void push_all_stats(void);
 276
 277 static bool facet_is_controller_flow(struct facet *);
 278
 279 struct ofport_dpif {
 280     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 281     struct ofport up;
 282
 283     odp_port_t odp_port;
 284     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 285     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
 286     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 287     struct bfd *bfd;            /* BFD, if any. */
 288     bool may_enable;            /* May be enabled in bonds. */
 289     bool is_tunnel;             /* This port is a tunnel. */
 290     long long int carrier_seq;  /* Carrier status changes. */
 291     struct ofport_dpif *peer;   /* Peer if patch port. */
 292
 293     /* Spanning tree. */
 294     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 295     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 296     long long int stp_state_entered;
 297
 298     /* Queue to DSCP mapping. */
 299     struct ofproto_port_queue *qdscp;
 300     size_t n_qdscp;
 301
 302     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 303      *
 304      * This is deprecated.  It is only for compatibility with broken device
 305      * drivers in old versions of Linux that do not properly support VLANs when
 306      * VLAN devices are not used.  When broken device drivers are no longer in
 307      * widespread use, we will delete these interfaces. */
 308     ofp_port_t realdev_ofp_port;
 309     int vlandev_vid;
 310 };
 311
 312 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 313  *
 314  * This is deprecated.  It is only for compatibility with broken device drivers
 315  * in old versions of Linux that do not properly support VLANs when VLAN
 316  * devices are not used.  When broken device drivers are no longer in
 317  * widespread use, we will delete these interfaces. */
 318 struct vlan_splinter {
 319     struct hmap_node realdev_vid_node;
 320     struct hmap_node vlandev_node;
 321     ofp_port_t realdev_ofp_port;
 322     ofp_port_t vlandev_ofp_port;
 323     int vid;
 324 };
 325
 326 static void vsp_remove(struct ofport_dpif *);
 327 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 328
 329 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 330                                        ofp_port_t);
 331
 332 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 333                                        odp_port_t);
 334
 335 static struct ofport_dpif *
 336 ofport_dpif_cast(const struct ofport *ofport)
 337 {
 338     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 339 }
 340
 341 static void port_run(struct ofport_dpif *);
 342 static void port_run_fast(struct ofport_dpif *);
 343 static void port_wait(struct ofport_dpif *);
 344 static int set_bfd(struct ofport *, const struct smap *);
 345 static int set_cfm(struct ofport *, const struct cfm_settings *);
 346 static void ofport_update_peer(struct ofport_dpif *);
 347 static void run_fast_rl(void);
 348 static int run_fast(struct ofproto *);
 349
 350 struct dpif_completion {
 351     struct list list_node;
 352     struct ofoperation *op;
 353 };
 354
 355 /* Reasons that we might need to revalidate every facet, and corresponding
 356  * coverage counters.
 357  *
 358  * A value of 0 means that there is no need to revalidate.
 359  *
 360  * It would be nice to have some cleaner way to integrate with coverage
 361  * counters, but with only a few reasons I guess this is good enough for
 362  * now. */
 363 enum revalidate_reason {
 364     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 365     REV_STP,                   /* Spanning tree protocol port status change. */
 366     REV_BOND,                  /* Bonding changed. */
 367     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 368     REV_FLOW_TABLE,            /* Flow table changed. */
 369     REV_MAC_LEARNING,          /* Mac learning changed. */
 370     REV_INCONSISTENCY          /* Facet self-check failed. */
 371 };
 372 COVERAGE_DEFINE(rev_reconfigure);
 373 COVERAGE_DEFINE(rev_stp);
 374 COVERAGE_DEFINE(rev_bond);
 375 COVERAGE_DEFINE(rev_port_toggled);
 376 COVERAGE_DEFINE(rev_flow_table);
 377 COVERAGE_DEFINE(rev_mac_learning);
 378 COVERAGE_DEFINE(rev_inconsistency);
 379
 380 struct avg_subfacet_rates {
 381     double add_rate;   /* Moving average of new flows created per minute. */
 382     double del_rate;   /* Moving average of flows deleted per minute. */
 383 };
 384
 385 /* All datapaths of a given type share a single dpif backer instance. */
 386 struct dpif_backer {
 387     char *type;
 388     int refcount;
 389     struct dpif *dpif;
 390     struct udpif *udpif;
 391     struct timer next_expiration;
 392
 393     struct ovs_rwlock odp_to_ofport_lock;
 394     struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
 395
 396     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 397
 398     /* Facet revalidation flags applying to facets which use this backer. */
 399     enum revalidate_reason need_revalidate; /* Revalidate every facet. */
 400
 401     struct hmap drop_keys; /* Set of dropped odp keys. */
 402     bool recv_set_enable; /* Enables or disables receiving packets. */
 403
 404     struct hmap subfacets;
 405     struct governor *governor;
 406
 407     /* Subfacet statistics.
 408      *
 409      * These keep track of the total number of subfacets added and deleted and
 410      * flow life span.  They are useful for computing the flow rates stats
 411      * exposed via "ovs-appctl dpif/show".  The goal is to learn about
 412      * traffic patterns in ways that we can use later to improve Open vSwitch
 413      * performance in new situations.  */
 414     long long int created;           /* Time when it is created. */
 415     unsigned max_n_subfacet;         /* Maximum number of flows */
 416     unsigned avg_n_subfacet;         /* Average number of flows. */
 417     long long int avg_subfacet_life; /* Average life span of subfacets. */
 418
 419     /* The average number of subfacets... */
 420     struct avg_subfacet_rates hourly;   /* ...over the last hour. */
 421     struct avg_subfacet_rates daily;    /* ...over the last day. */
 422     struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */
 423     long long int last_minute;          /* Last time 'hourly' was updated. */
 424
 425     /* Number of subfacets added or deleted since 'last_minute'. */
 426     unsigned subfacet_add_count;
 427     unsigned subfacet_del_count;
 428
 429     /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
 430     unsigned long long int total_subfacet_add_count;
 431     unsigned long long int total_subfacet_del_count;
 432
 433     /* Number of upcall handling threads. */
 434     unsigned int n_handler_threads;
 435 };
 436
 437 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 438 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 439
 440 static void drop_key_clear(struct dpif_backer *);
 441 static void update_moving_averages(struct dpif_backer *backer);
 442
 443 struct ofproto_dpif {
 444     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 445     struct ofproto up;
 446     struct dpif_backer *backer;
 447
 448     /* Special OpenFlow rules. */
 449     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 450     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 451     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 452
 453     /* Bridging. */
 454     struct netflow *netflow;
 455     struct dpif_sflow *sflow;
 456     struct dpif_ipfix *ipfix;
 457     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 458     struct mac_learning *ml;
 459     bool has_bonded_bundles;
 460     struct mbridge *mbridge;
 461
 462     /* Facets. */
 463     struct classifier facets;     /* Contains 'struct facet's. */
 464     long long int consistency_rl;
 465
 466     /* Support for debugging async flow mods. */
 467     struct list completions;
 468
 469     struct netdev_stats stats; /* To account packets generated and consumed in
 470                                 * userspace. */
 471
 472     /* Spanning tree. */
 473     struct stp *stp;
 474     long long int stp_last_tick;
 475
 476     /* VLAN splinters. */
 477     struct ovs_mutex vsp_mutex;
 478     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 479     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 480
 481     /* Ports. */
 482     struct sset ports;             /* Set of standard port names. */
 483     struct sset ghost_ports;       /* Ports with no datapath port. */
 484     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 485     int port_poll_errno;           /* Last errno for port_poll() reply. */
 486
 487     /* Per ofproto's dpif stats. */
 488     uint64_t n_hit;
 489     uint64_t n_missed;
 490
 491     /* Work queues. */
 492     struct ovs_mutex flow_mod_mutex;
 493     struct list flow_mods OVS_GUARDED;
 494     size_t n_flow_mods OVS_GUARDED;
 495
 496     struct ovs_mutex pin_mutex;
 497     struct list pins OVS_GUARDED;
 498     size_t n_pins OVS_GUARDED;
 499 };
 500
 501 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"?  (Useful only
 502  * for debugging the asynchronous flow_mod implementation.) */
 503 static bool clogged;
 504
 505 /* By default, flows in the datapath are wildcarded (megaflows).  They
 506  * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
 507 static bool enable_megaflows = true;
 508
 509 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 510 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 511
 512 static void ofproto_dpif_unixctl_init(void);
 513
 514 static inline struct ofproto_dpif *
 515 ofproto_dpif_cast(const struct ofproto *ofproto)
 516 {
 517     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 518     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 519 }
 520
 521 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
 522                                         ofp_port_t ofp_port);
 523 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
 524                           const struct ofpbuf *packet, struct ds *);
 525
 526 /* Upcalls. */
 527 static void handle_upcalls(struct dpif_backer *);
 528
 529 /* Flow expiration. */
 530 static int expire(struct dpif_backer *);
 531
 532 /* NetFlow. */
 533 static void send_netflow_active_timeouts(struct ofproto_dpif *);
 534
 535 /* Utilities. */
 536 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
 537
 538 /* Global variables. */
 539 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 540
 541 /* Initial mappings of port to bridge mappings. */
 542 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 543
 544 /* Executes and takes ownership of 'fm'. */
 545 void
 546 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 547                       struct ofputil_flow_mod *fm)
 548 {
 549     ovs_mutex_lock(&ofproto->flow_mod_mutex);
 550     if (ofproto->n_flow_mods > 1024) {
 551         ovs_mutex_unlock(&ofproto->flow_mod_mutex);
 552         COVERAGE_INC(flow_mod_overflow);
 553         free(fm->ofpacts);
 554         free(fm);
 555         return;
 556     }
 557
 558     list_push_back(&ofproto->flow_mods, &fm->list_node);
 559     ofproto->n_flow_mods++;
 560     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
 561 }
 562
 563 void
 564 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 565                             struct ofputil_packet_in *pin)
 566 {
 567     ovs_mutex_lock(&ofproto->pin_mutex);
 568     if (ofproto->n_pins > 1024) {
 569         ovs_mutex_unlock(&ofproto->pin_mutex);
 570         COVERAGE_INC(packet_in_overflow);
 571         free(CONST_CAST(void *, pin->packet));
 572         free(pin);
 573         return;
 574     }
 575
 576     list_push_back(&ofproto->pins, &pin->list_node);
 577     ofproto->n_pins++;
 578     ovs_mutex_unlock(&ofproto->pin_mutex);
 579 }
 580 \f
 581 /* Factory functions. */
 582
 583 static void
 584 init(const struct shash *iface_hints)
 585 {
 586     struct shash_node *node;
 587
 588     /* Make a local copy, since we don't own 'iface_hints' elements. */
 589     SHASH_FOR_EACH(node, iface_hints) {
 590         const struct iface_hint *orig_hint = node->data;
 591         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 592
 593         new_hint->br_name = xstrdup(orig_hint->br_name);
 594         new_hint->br_type = xstrdup(orig_hint->br_type);
 595         new_hint->ofp_port = orig_hint->ofp_port;
 596
 597         shash_add(&init_ofp_ports, node->name, new_hint);
 598     }
 599 }
 600
 601 static void
 602 enumerate_types(struct sset *types)
 603 {
 604     dp_enumerate_types(types);
 605 }
 606
 607 static int
 608 enumerate_names(const char *type, struct sset *names)
 609 {
 610     struct ofproto_dpif *ofproto;
 611
 612     sset_clear(names);
 613     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 614         if (strcmp(type, ofproto->up.type)) {
 615             continue;
 616         }
 617         sset_add(names, ofproto->up.name);
 618     }
 619
 620     return 0;
 621 }
 622
 623 static int
 624 del(const char *type, const char *name)
 625 {
 626     struct dpif *dpif;
 627     int error;
 628
 629     error = dpif_open(name, type, &dpif);
 630     if (!error) {
 631         error = dpif_delete(dpif);
 632         dpif_close(dpif);
 633     }
 634     return error;
 635 }
 636 \f
 637 static const char *
 638 port_open_type(const char *datapath_type, const char *port_type)
 639 {
 640     return dpif_port_open_type(datapath_type, port_type);
 641 }
 642
 643 /* Type functions. */
 644
 645 static void process_dpif_port_changes(struct dpif_backer *);
 646 static void process_dpif_all_ports_changed(struct dpif_backer *);
 647 static void process_dpif_port_change(struct dpif_backer *,
 648                                      const char *devname);
 649 static void process_dpif_port_error(struct dpif_backer *, int error);
 650
 651 static struct ofproto_dpif *
 652 lookup_ofproto_dpif_by_port_name(const char *name)
 653 {
 654     struct ofproto_dpif *ofproto;
 655
 656     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 657         if (sset_contains(&ofproto->ports, name)) {
 658             return ofproto;
 659         }
 660     }
 661
 662     return NULL;
 663 }
 664
 665 static int
 666 type_run(const char *type)
 667 {
 668     static long long int push_timer = LLONG_MIN;
 669     struct dpif_backer *backer;
 670
 671     backer = shash_find_data(&all_dpif_backers, type);
 672     if (!backer) {
 673         /* This is not necessarily a problem, since backers are only
 674          * created on demand. */
 675         return 0;
 676     }
 677
 678     dpif_run(backer->dpif);
 679
 680     /* The most natural place to push facet statistics is when they're pulled
 681      * from the datapath.  However, when there are many flows in the datapath,
 682      * this expensive operation can occur so frequently, that it reduces our
 683      * ability to quickly set up flows.  To reduce the cost, we push statistics
 684      * here instead. */
 685     if (time_msec() > push_timer) {
 686         push_timer = time_msec() + 2000;
 687         push_all_stats();
 688     }
 689
 690     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 691      * and the configuration has now changed to "false", enable receiving
 692      * packets from the datapath. */
 693     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 694         int error;
 695
 696         backer->recv_set_enable = true;
 697
 698         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 699         if (error) {
 700             udpif_recv_set(backer->udpif, 0, false);
 701             VLOG_ERR("Failed to enable receiving packets in dpif.");
 702             return error;
 703         }
 704         udpif_recv_set(backer->udpif, n_handler_threads,
 705                        backer->recv_set_enable);
 706         dpif_flow_flush(backer->dpif);
 707         backer->need_revalidate = REV_RECONFIGURE;
 708     }
 709
 710     /* If the n_handler_threads is reconfigured, call udpif_recv_set()
 711      * to reset the handler threads. */
 712     if (backer->n_handler_threads != n_handler_threads) {
 713         udpif_recv_set(backer->udpif, n_handler_threads,
 714                        backer->recv_set_enable);
 715         backer->n_handler_threads = n_handler_threads;
 716     }
 717
 718     if (backer->need_revalidate) {
 719         struct ofproto_dpif *ofproto;
 720         struct simap_node *node;
 721         struct simap tmp_backers;
 722
 723         /* Handle tunnel garbage collection. */
 724         simap_init(&tmp_backers);
 725         simap_swap(&backer->tnl_backers, &tmp_backers);
 726
 727         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 728             struct ofport_dpif *iter;
 729
 730             if (backer != ofproto->backer) {
 731                 continue;
 732             }
 733
 734             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 735                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 736                 const char *dp_port;
 737
 738                 if (!iter->is_tunnel) {
 739                     continue;
 740                 }
 741
 742                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 743                                                      namebuf, sizeof namebuf);
 744                 node = simap_find(&tmp_backers, dp_port);
 745                 if (node) {
 746                     simap_put(&backer->tnl_backers, dp_port, node->data);
 747                     simap_delete(&tmp_backers, node);
 748                     node = simap_find(&backer->tnl_backers, dp_port);
 749                 } else {
 750                     node = simap_find(&backer->tnl_backers, dp_port);
 751                     if (!node) {
 752                         odp_port_t odp_port = ODPP_NONE;
 753
 754                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 755                                            &odp_port)) {
 756                             simap_put(&backer->tnl_backers, dp_port,
 757                                       odp_to_u32(odp_port));
 758                             node = simap_find(&backer->tnl_backers, dp_port);
 759                         }
 760                     }
 761                 }
 762
 763                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 764                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 765                                          iter->odp_port)) {
 766                     backer->need_revalidate = REV_RECONFIGURE;
 767                 }
 768             }
 769         }
 770
 771         SIMAP_FOR_EACH (node, &tmp_backers) {
 772             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 773         }
 774         simap_destroy(&tmp_backers);
 775
 776         switch (backer->need_revalidate) {
 777         case REV_RECONFIGURE:   COVERAGE_INC(rev_reconfigure);   break;
 778         case REV_STP:           COVERAGE_INC(rev_stp);           break;
 779         case REV_BOND:          COVERAGE_INC(rev_bond);          break;
 780         case REV_PORT_TOGGLED:  COVERAGE_INC(rev_port_toggled);  break;
 781         case REV_FLOW_TABLE:    COVERAGE_INC(rev_flow_table);    break;
 782         case REV_MAC_LEARNING:  COVERAGE_INC(rev_mac_learning);  break;
 783         case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
 784         }
 785         backer->need_revalidate = 0;
 786
 787         /* Clear the drop_keys in case we should now be accepting some
 788          * formerly dropped flows. */
 789         drop_key_clear(backer);
 790
 791         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 792             struct facet *facet, *next;
 793             struct ofport_dpif *ofport;
 794             struct cls_cursor cursor;
 795             struct ofbundle *bundle;
 796
 797             if (ofproto->backer != backer) {
 798                 continue;
 799             }
 800
 801             ovs_rwlock_wrlock(&xlate_rwlock);
 802             xlate_ofproto_set(ofproto, ofproto->up.name,
 803                               ofproto->backer->dpif, ofproto->miss_rule,
 804                               ofproto->no_packet_in_rule, ofproto->ml,
 805                               ofproto->stp, ofproto->mbridge,
 806                               ofproto->sflow, ofproto->ipfix,
 807                               ofproto->up.frag_handling,
 808                               ofproto->up.forward_bpdu,
 809                               connmgr_has_in_band(ofproto->up.connmgr),
 810                               ofproto->netflow != NULL);
 811
 812             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 813                 xlate_bundle_set(ofproto, bundle, bundle->name,
 814                                  bundle->vlan_mode, bundle->vlan,
 815                                  bundle->trunks, bundle->use_priority_tags,
 816                                  bundle->bond, bundle->lacp,
 817                                  bundle->floodable);
 818             }
 819
 820             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 821                 int stp_port = ofport->stp_port
 822                     ? stp_port_no(ofport->stp_port)
 823                     : 0;
 824                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 825                                  ofport->up.ofp_port, ofport->odp_port,
 826                                  ofport->up.netdev, ofport->cfm,
 827                                  ofport->bfd, ofport->peer, stp_port,
 828                                  ofport->qdscp, ofport->n_qdscp,
 829                                  ofport->up.pp.config, ofport->is_tunnel,
 830                                  ofport->may_enable);
 831             }
 832             ovs_rwlock_unlock(&xlate_rwlock);
 833
 834             /* Only ofproto-dpif cares about the facet classifier so we just
 835              * lock cls_cursor_init() to appease the thread safety analysis. */
 836             ovs_rwlock_rdlock(&ofproto->facets.rwlock);
 837             cls_cursor_init(&cursor, &ofproto->facets, NULL);
 838             ovs_rwlock_unlock(&ofproto->facets.rwlock);
 839             CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
 840                 facet_revalidate(facet);
 841                 run_fast_rl();
 842             }
 843         }
 844
 845         udpif_revalidate(backer->udpif);
 846     }
 847
 848     if (!backer->recv_set_enable) {
 849         /* Wake up before a max of 1000ms. */
 850         timer_set_duration(&backer->next_expiration, 1000);
 851     } else if (timer_expired(&backer->next_expiration)) {
 852         int delay = expire(backer);
 853         timer_set_duration(&backer->next_expiration, delay);
 854     }
 855
 856     process_dpif_port_changes(backer);
 857
 858     if (backer->governor) {
 859         size_t n_subfacets;
 860
 861         governor_run(backer->governor);
 862
 863         /* If the governor has shrunk to its minimum size and the number of
 864          * subfacets has dwindled, then drop the governor entirely.
 865          *
 866          * For hysteresis, the number of subfacets to drop the governor is
 867          * smaller than the number needed to trigger its creation. */
 868         n_subfacets = hmap_count(&backer->subfacets);
 869         if (n_subfacets * 4 < flow_eviction_threshold
 870             && governor_is_idle(backer->governor)) {
 871             governor_destroy(backer->governor);
 872             backer->governor = NULL;
 873         }
 874     }
 875
 876     return 0;
 877 }
 878
 879 /* Check for and handle port changes in 'backer''s dpif. */
 880 static void
 881 process_dpif_port_changes(struct dpif_backer *backer)
 882 {
 883     for (;;) {
 884         char *devname;
 885         int error;
 886
 887         error = dpif_port_poll(backer->dpif, &devname);
 888         switch (error) {
 889         case EAGAIN:
 890             return;
 891
 892         case ENOBUFS:
 893             process_dpif_all_ports_changed(backer);
 894             break;
 895
 896         case 0:
 897             process_dpif_port_change(backer, devname);
 898             free(devname);
 899             break;
 900
 901         default:
 902             process_dpif_port_error(backer, error);
 903             break;
 904         }
 905     }
 906 }
 907
 908 static void
 909 process_dpif_all_ports_changed(struct dpif_backer *backer)
 910 {
 911     struct ofproto_dpif *ofproto;
 912     struct dpif_port dpif_port;
 913     struct dpif_port_dump dump;
 914     struct sset devnames;
 915     const char *devname;
 916
 917     sset_init(&devnames);
 918     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 919         if (ofproto->backer == backer) {
 920             struct ofport *ofport;
 921
 922             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 923                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 924             }
 925         }
 926     }
 927     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 928         sset_add(&devnames, dpif_port.name);
 929     }
 930
 931     SSET_FOR_EACH (devname, &devnames) {
 932         process_dpif_port_change(backer, devname);
 933     }
 934     sset_destroy(&devnames);
 935 }
 936
 937 static void
 938 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 939 {
 940     struct ofproto_dpif *ofproto;
 941     struct dpif_port port;
 942
 943     /* Don't report on the datapath's device. */
 944     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 945         return;
 946     }
 947
 948     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 949                    &all_ofproto_dpifs) {
 950         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 951             return;
 952         }
 953     }
 954
 955     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 956     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 957         /* The port was removed.  If we know the datapath,
 958          * report it through poll_set().  If we don't, it may be
 959          * notifying us of a removal we initiated, so ignore it.
 960          * If there's a pending ENOBUFS, let it stand, since
 961          * everything will be reevaluated. */
 962         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 963             sset_add(&ofproto->port_poll_set, devname);
 964             ofproto->port_poll_errno = 0;
 965         }
 966     } else if (!ofproto) {
 967         /* The port was added, but we don't know with which
 968          * ofproto we should associate it.  Delete it. */
 969         dpif_port_del(backer->dpif, port.port_no);
 970     } else {
 971         struct ofport_dpif *ofport;
 972
 973         ofport = ofport_dpif_cast(shash_find_data(
 974                                       &ofproto->up.port_by_name, devname));
 975         if (ofport
 976             && ofport->odp_port != port.port_no
 977             && !odp_port_to_ofport(backer, port.port_no))
 978         {
 979             /* 'ofport''s datapath port number has changed from
 980              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 981              * structures to match. */
 982             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 983             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 984             ofport->odp_port = port.port_no;
 985             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 986                         hash_odp_port(port.port_no));
 987             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 988             backer->need_revalidate = REV_RECONFIGURE;
 989         }
 990     }
 991     dpif_port_destroy(&port);
 992 }
 993
 994 /* Propagate 'error' to all ofprotos based on 'backer'. */
 995 static void
 996 process_dpif_port_error(struct dpif_backer *backer, int error)
 997 {
 998     struct ofproto_dpif *ofproto;
 999
1000     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1001         if (ofproto->backer == backer) {
1002             sset_clear(&ofproto->port_poll_set);
1003             ofproto->port_poll_errno = error;
1004         }
1005     }
1006 }
1007
1008 static int
1009 dpif_backer_run_fast(struct dpif_backer *backer)
1010 {
1011     udpif_run(backer->udpif);
1012     handle_upcalls(backer);
1013
1014     return 0;
1015 }
1016
1017 static int
1018 type_run_fast(const char *type)
1019 {
1020     struct dpif_backer *backer;
1021
1022     backer = shash_find_data(&all_dpif_backers, type);
1023     if (!backer) {
1024         /* This is not necessarily a problem, since backers are only
1025          * created on demand. */
1026         return 0;
1027     }
1028
1029     return dpif_backer_run_fast(backer);
1030 }
1031
1032 static void
1033 run_fast_rl(void)
1034 {
1035     static long long int port_rl = LLONG_MIN;
1036
1037     if (time_msec() >= port_rl) {
1038         struct ofproto_dpif *ofproto;
1039
1040         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1041             run_fast(&ofproto->up);
1042         }
1043         port_rl = time_msec() + 200;
1044     }
1045 }
1046
1047 static void
1048 type_wait(const char *type)
1049 {
1050     struct dpif_backer *backer;
1051
1052     backer = shash_find_data(&all_dpif_backers, type);
1053     if (!backer) {
1054         /* This is not necessarily a problem, since backers are only
1055          * created on demand. */
1056         return;
1057     }
1058
1059     if (backer->governor) {
1060         governor_wait(backer->governor);
1061     }
1062
1063     timer_wait(&backer->next_expiration);
1064     dpif_wait(backer->dpif);
1065     udpif_wait(backer->udpif);
1066 }
1067 \f
1068 /* Basic life-cycle. */
1069
1070 static int add_internal_flows(struct ofproto_dpif *);
1071
1072 static struct ofproto *
1073 alloc(void)
1074 {
1075     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1076     return &ofproto->up;
1077 }
1078
1079 static void
1080 dealloc(struct ofproto *ofproto_)
1081 {
1082     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1083     free(ofproto);
1084 }
1085
1086 static void
1087 close_dpif_backer(struct dpif_backer *backer)
1088 {
1089     struct shash_node *node;
1090
1091     ovs_assert(backer->refcount > 0);
1092
1093     if (--backer->refcount) {
1094         return;
1095     }
1096
1097     drop_key_clear(backer);
1098     hmap_destroy(&backer->drop_keys);
1099
1100     simap_destroy(&backer->tnl_backers);
1101     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1102     hmap_destroy(&backer->odp_to_ofport_map);
1103     node = shash_find(&all_dpif_backers, backer->type);
1104     free(backer->type);
1105     shash_delete(&all_dpif_backers, node);
1106     udpif_destroy(backer->udpif);
1107     dpif_close(backer->dpif);
1108
1109     ovs_assert(hmap_is_empty(&backer->subfacets));
1110     hmap_destroy(&backer->subfacets);
1111     governor_destroy(backer->governor);
1112
1113     free(backer);
1114 }
1115
1116 /* Datapath port slated for removal from datapath. */
1117 struct odp_garbage {
1118     struct list list_node;
1119     odp_port_t odp_port;
1120 };
1121
1122 static int
1123 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1124 {
1125     struct dpif_backer *backer;
1126     struct dpif_port_dump port_dump;
1127     struct dpif_port port;
1128     struct shash_node *node;
1129     struct list garbage_list;
1130     struct odp_garbage *garbage, *next;
1131     struct sset names;
1132     char *backer_name;
1133     const char *name;
1134     int error;
1135
1136     backer = shash_find_data(&all_dpif_backers, type);
1137     if (backer) {
1138         backer->refcount++;
1139         *backerp = backer;
1140         return 0;
1141     }
1142
1143     backer_name = xasprintf("ovs-%s", type);
1144
1145     /* Remove any existing datapaths, since we assume we're the only
1146      * userspace controlling the datapath. */
1147     sset_init(&names);
1148     dp_enumerate_names(type, &names);
1149     SSET_FOR_EACH(name, &names) {
1150         struct dpif *old_dpif;
1151
1152         /* Don't remove our backer if it exists. */
1153         if (!strcmp(name, backer_name)) {
1154             continue;
1155         }
1156
1157         if (dpif_open(name, type, &old_dpif)) {
1158             VLOG_WARN("couldn't open old datapath %s to remove it", name);
1159         } else {
1160             dpif_delete(old_dpif);
1161             dpif_close(old_dpif);
1162         }
1163     }
1164     sset_destroy(&names);
1165
1166     backer = xmalloc(sizeof *backer);
1167
1168     error = dpif_create_and_open(backer_name, type, &backer->dpif);
1169     free(backer_name);
1170     if (error) {
1171         VLOG_ERR("failed to open datapath of type %s: %s", type,
1172                  ovs_strerror(error));
1173         free(backer);
1174         return error;
1175     }
1176     backer->udpif = udpif_create(backer, backer->dpif);
1177
1178     backer->type = xstrdup(type);
1179     backer->governor = NULL;
1180     backer->refcount = 1;
1181     hmap_init(&backer->odp_to_ofport_map);
1182     ovs_rwlock_init(&backer->odp_to_ofport_lock);
1183     hmap_init(&backer->drop_keys);
1184     hmap_init(&backer->subfacets);
1185     timer_set_duration(&backer->next_expiration, 1000);
1186     backer->need_revalidate = 0;
1187     simap_init(&backer->tnl_backers);
1188     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1189     *backerp = backer;
1190
1191     if (backer->recv_set_enable) {
1192         dpif_flow_flush(backer->dpif);
1193     }
1194
1195     /* Loop through the ports already on the datapath and remove any
1196      * that we don't need anymore. */
1197     list_init(&garbage_list);
1198     dpif_port_dump_start(&port_dump, backer->dpif);
1199     while (dpif_port_dump_next(&port_dump, &port)) {
1200         node = shash_find(&init_ofp_ports, port.name);
1201         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1202             garbage = xmalloc(sizeof *garbage);
1203             garbage->odp_port = port.port_no;
1204             list_push_front(&garbage_list, &garbage->list_node);
1205         }
1206     }
1207     dpif_port_dump_done(&port_dump);
1208
1209     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1210         dpif_port_del(backer->dpif, garbage->odp_port);
1211         list_remove(&garbage->list_node);
1212         free(garbage);
1213     }
1214
1215     shash_add(&all_dpif_backers, type, backer);
1216
1217     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1218     if (error) {
1219         VLOG_ERR("failed to listen on datapath of type %s: %s",
1220                  type, ovs_strerror(error));
1221         close_dpif_backer(backer);
1222         return error;
1223     }
1224     udpif_recv_set(backer->udpif, n_handler_threads,
1225                    backer->recv_set_enable);
1226     backer->n_handler_threads = n_handler_threads;
1227
1228     backer->max_n_subfacet = 0;
1229     backer->created = time_msec();
1230     backer->last_minute = backer->created;
1231     memset(&backer->hourly, 0, sizeof backer->hourly);
1232     memset(&backer->daily, 0, sizeof backer->daily);
1233     memset(&backer->lifetime, 0, sizeof backer->lifetime);
1234     backer->subfacet_add_count = 0;
1235     backer->subfacet_del_count = 0;
1236     backer->total_subfacet_add_count = 0;
1237     backer->total_subfacet_del_count = 0;
1238     backer->avg_n_subfacet = 0;
1239     backer->avg_subfacet_life = 0;
1240
1241     return error;
1242 }
1243
1244 static int
1245 construct(struct ofproto *ofproto_)
1246 {
1247     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1248     struct shash_node *node, *next;
1249     odp_port_t max_ports;
1250     int error;
1251
1252     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1253     if (error) {
1254         return error;
1255     }
1256
1257     max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1258     ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports),
1259                                                     ofp_to_u16(OFPP_MAX))));
1260
1261     ofproto->netflow = NULL;
1262     ofproto->sflow = NULL;
1263     ofproto->ipfix = NULL;
1264     ofproto->stp = NULL;
1265     hmap_init(&ofproto->bundles);
1266     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1267     ofproto->mbridge = mbridge_create();
1268     ofproto->has_bonded_bundles = false;
1269     ovs_mutex_init(&ofproto->vsp_mutex, PTHREAD_MUTEX_NORMAL);
1270
1271     classifier_init(&ofproto->facets);
1272     ofproto->consistency_rl = LLONG_MIN;
1273
1274     list_init(&ofproto->completions);
1275
1276     ovs_mutex_init(&ofproto->flow_mod_mutex, PTHREAD_MUTEX_NORMAL);
1277     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1278     list_init(&ofproto->flow_mods);
1279     ofproto->n_flow_mods = 0;
1280     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1281
1282     ovs_mutex_init(&ofproto->pin_mutex, PTHREAD_MUTEX_NORMAL);
1283     ovs_mutex_lock(&ofproto->pin_mutex);
1284     list_init(&ofproto->pins);
1285     ofproto->n_pins = 0;
1286     ovs_mutex_unlock(&ofproto->pin_mutex);
1287
1288     ofproto_dpif_unixctl_init();
1289
1290     hmap_init(&ofproto->vlandev_map);
1291     hmap_init(&ofproto->realdev_vid_map);
1292
1293     sset_init(&ofproto->ports);
1294     sset_init(&ofproto->ghost_ports);
1295     sset_init(&ofproto->port_poll_set);
1296     ofproto->port_poll_errno = 0;
1297
1298     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1299         struct iface_hint *iface_hint = node->data;
1300
1301         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1302             /* Check if the datapath already has this port. */
1303             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1304                 sset_add(&ofproto->ports, node->name);
1305             }
1306
1307             free(iface_hint->br_name);
1308             free(iface_hint->br_type);
1309             free(iface_hint);
1310             shash_delete(&init_ofp_ports, node);
1311         }
1312     }
1313
1314     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1315                 hash_string(ofproto->up.name, 0));
1316     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1317
1318     ofproto_init_tables(ofproto_, N_TABLES);
1319     error = add_internal_flows(ofproto);
1320     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1321
1322     ofproto->n_hit = 0;
1323     ofproto->n_missed = 0;
1324
1325     return error;
1326 }
1327
1328 static int
1329 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1330                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1331 {
1332     struct ofputil_flow_mod fm;
1333     int error;
1334
1335     match_init_catchall(&fm.match);
1336     fm.priority = 0;
1337     match_set_reg(&fm.match, 0, id);
1338     fm.new_cookie = htonll(0);
1339     fm.cookie = htonll(0);
1340     fm.cookie_mask = htonll(0);
1341     fm.modify_cookie = false;
1342     fm.table_id = TBL_INTERNAL;
1343     fm.command = OFPFC_ADD;
1344     fm.idle_timeout = 0;
1345     fm.hard_timeout = 0;
1346     fm.buffer_id = 0;
1347     fm.out_port = 0;
1348     fm.flags = 0;
1349     fm.ofpacts = ofpacts->data;
1350     fm.ofpacts_len = ofpacts->size;
1351
1352     error = ofproto_flow_mod(&ofproto->up, &fm);
1353     if (error) {
1354         VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1355                     id, ofperr_to_string(error));
1356         return error;
1357     }
1358
1359     if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1360                                   rulep)) {
1361         ovs_rwlock_unlock(&(*rulep)->up.evict);
1362     } else {
1363         NOT_REACHED();
1364     }
1365
1366     return 0;
1367 }
1368
1369 static int
1370 add_internal_flows(struct ofproto_dpif *ofproto)
1371 {
1372     struct ofpact_controller *controller;
1373     uint64_t ofpacts_stub[128 / 8];
1374     struct ofpbuf ofpacts;
1375     int error;
1376     int id;
1377
1378     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1379     id = 1;
1380
1381     controller = ofpact_put_CONTROLLER(&ofpacts);
1382     controller->max_len = UINT16_MAX;
1383     controller->controller_id = 0;
1384     controller->reason = OFPR_NO_MATCH;
1385     ofpact_pad(&ofpacts);
1386
1387     error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1388     if (error) {
1389         return error;
1390     }
1391
1392     ofpbuf_clear(&ofpacts);
1393     error = add_internal_flow(ofproto, id++, &ofpacts,
1394                               &ofproto->no_packet_in_rule);
1395     if (error) {
1396         return error;
1397     }
1398
1399     error = add_internal_flow(ofproto, id++, &ofpacts,
1400                               &ofproto->drop_frags_rule);
1401     return error;
1402 }
1403
1404 static void
1405 complete_operations(struct ofproto_dpif *ofproto)
1406 {
1407     struct dpif_completion *c, *next;
1408
1409     LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1410         ofoperation_complete(c->op, 0);
1411         list_remove(&c->list_node);
1412         free(c);
1413     }
1414 }
1415
1416 static void
1417 destruct(struct ofproto *ofproto_)
1418 {
1419     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1420     struct rule_dpif *rule, *next_rule;
1421     struct ofputil_flow_mod *pin, *next_pin;
1422     struct ofputil_flow_mod *fm, *next_fm;
1423     struct oftable *table;
1424
1425     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1426     ovs_rwlock_wrlock(&xlate_rwlock);
1427     xlate_remove_ofproto(ofproto);
1428     ovs_rwlock_unlock(&xlate_rwlock);
1429
1430     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1431     complete_operations(ofproto);
1432
1433     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1434         struct cls_cursor cursor;
1435
1436         ovs_rwlock_wrlock(&table->cls.rwlock);
1437         cls_cursor_init(&cursor, &table->cls, NULL);
1438         CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1439             ofproto_rule_destroy(&ofproto->up, &table->cls, &rule->up);
1440         }
1441         ovs_rwlock_unlock(&table->cls.rwlock);
1442     }
1443
1444     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1445     LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &ofproto->flow_mods) {
1446         list_remove(&fm->list_node);
1447         ofproto->n_flow_mods--;
1448         free(fm->ofpacts);
1449         free(fm);
1450     }
1451     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1452     ovs_mutex_destroy(&ofproto->flow_mod_mutex);
1453
1454     ovs_mutex_lock(&ofproto->pin_mutex);
1455     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &ofproto->pins) {
1456         list_remove(&pin->list_node);
1457         ofproto->n_pins--;
1458         free(pin->ofpacts);
1459         free(pin);
1460     }
1461     ovs_mutex_unlock(&ofproto->pin_mutex);
1462     ovs_mutex_destroy(&ofproto->pin_mutex);
1463
1464     mbridge_unref(ofproto->mbridge);
1465
1466     netflow_destroy(ofproto->netflow);
1467     dpif_sflow_unref(ofproto->sflow);
1468     hmap_destroy(&ofproto->bundles);
1469     mac_learning_unref(ofproto->ml);
1470
1471     classifier_destroy(&ofproto->facets);
1472
1473     hmap_destroy(&ofproto->vlandev_map);
1474     hmap_destroy(&ofproto->realdev_vid_map);
1475
1476     sset_destroy(&ofproto->ports);
1477     sset_destroy(&ofproto->ghost_ports);
1478     sset_destroy(&ofproto->port_poll_set);
1479
1480     ovs_mutex_destroy(&ofproto->vsp_mutex);
1481
1482     close_dpif_backer(ofproto->backer);
1483 }
1484
1485 static int
1486 run_fast(struct ofproto *ofproto_)
1487 {
1488     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1489     struct ofputil_packet_in *pin, *next_pin;
1490     struct ofputil_flow_mod *fm, *next_fm;
1491     struct list flow_mods, pins;
1492     struct ofport_dpif *ofport;
1493
1494     /* Do not perform any periodic activity required by 'ofproto' while
1495      * waiting for flow restore to complete. */
1496     if (ofproto_get_flow_restore_wait()) {
1497         return 0;
1498     }
1499
1500     ovs_mutex_lock(&ofproto->flow_mod_mutex);
1501     if (ofproto->n_flow_mods) {
1502         flow_mods = ofproto->flow_mods;
1503         list_moved(&flow_mods);
1504         list_init(&ofproto->flow_mods);
1505         ofproto->n_flow_mods = 0;
1506     } else {
1507         list_init(&flow_mods);
1508     }
1509     ovs_mutex_unlock(&ofproto->flow_mod_mutex);
1510
1511     LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &flow_mods) {
1512         int error = ofproto_flow_mod(&ofproto->up, fm);
1513         if (error && !VLOG_DROP_WARN(&rl)) {
1514             VLOG_WARN("learning action failed to modify flow table (%s)",
1515                       ofperr_get_name(error));
1516         }
1517
1518         list_remove(&fm->list_node);
1519         free(fm->ofpacts);
1520         free(fm);
1521     }
1522
1523     ovs_mutex_lock(&ofproto->pin_mutex);
1524     if (ofproto->n_pins) {
1525         pins = ofproto->pins;
1526         list_moved(&pins);
1527         list_init(&ofproto->pins);
1528         ofproto->n_pins = 0;
1529     } else {
1530         list_init(&pins);
1531     }
1532     ovs_mutex_unlock(&ofproto->pin_mutex);
1533
1534     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1535         connmgr_send_packet_in(ofproto->up.connmgr, pin);
1536         list_remove(&pin->list_node);
1537         free(CONST_CAST(void *, pin->packet));
1538         free(pin);
1539     }
1540
1541     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1542         port_run_fast(ofport);
1543     }
1544
1545     return 0;
1546 }
1547
1548 static int
1549 run(struct ofproto *ofproto_)
1550 {
1551     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1552     struct ofport_dpif *ofport;
1553     struct ofbundle *bundle;
1554     int error;
1555
1556     if (!clogged) {
1557         complete_operations(ofproto);
1558     }
1559
1560     if (mbridge_need_revalidate(ofproto->mbridge)) {
1561         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1562         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1563         mac_learning_flush(ofproto->ml);
1564         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1565     }
1566
1567     /* Do not perform any periodic activity below required by 'ofproto' while
1568      * waiting for flow restore to complete. */
1569     if (ofproto_get_flow_restore_wait()) {
1570         return 0;
1571     }
1572
1573     error = run_fast(ofproto_);
1574     if (error) {
1575         return error;
1576     }
1577
1578     if (ofproto->netflow) {
1579         if (netflow_run(ofproto->netflow)) {
1580             send_netflow_active_timeouts(ofproto);
1581         }
1582     }
1583     if (ofproto->sflow) {
1584         dpif_sflow_run(ofproto->sflow);
1585     }
1586
1587     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1588         port_run(ofport);
1589     }
1590     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1591         bundle_run(bundle);
1592     }
1593
1594     stp_run(ofproto);
1595     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1596     if (mac_learning_run(ofproto->ml)) {
1597         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1598     }
1599     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1600
1601     /* Check the consistency of a random facet, to aid debugging. */
1602     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1603     if (time_msec() >= ofproto->consistency_rl
1604         && !classifier_is_empty(&ofproto->facets)
1605         && !ofproto->backer->need_revalidate) {
1606         struct cls_table *table;
1607         struct cls_rule *cr;
1608         struct facet *facet;
1609
1610         ofproto->consistency_rl = time_msec() + 250;
1611
1612         table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1613                              struct cls_table, hmap_node);
1614         cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1615                           hmap_node);
1616         facet = CONTAINER_OF(cr, struct facet, cr);
1617
1618         if (!facet_check_consistency(facet)) {
1619             ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1620         }
1621     }
1622     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1623
1624     return 0;
1625 }
1626
1627 static void
1628 wait(struct ofproto *ofproto_)
1629 {
1630     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1631     struct ofport_dpif *ofport;
1632     struct ofbundle *bundle;
1633
1634     if (!clogged && !list_is_empty(&ofproto->completions)) {
1635         poll_immediate_wake();
1636     }
1637
1638     if (ofproto_get_flow_restore_wait()) {
1639         return;
1640     }
1641
1642     if (ofproto->sflow) {
1643         dpif_sflow_wait(ofproto->sflow);
1644     }
1645     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1646         port_wait(ofport);
1647     }
1648     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1649         bundle_wait(bundle);
1650     }
1651     if (ofproto->netflow) {
1652         netflow_wait(ofproto->netflow);
1653     }
1654     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1655     mac_learning_wait(ofproto->ml);
1656     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1657     stp_wait(ofproto);
1658     if (ofproto->backer->need_revalidate) {
1659         /* Shouldn't happen, but if it does just go around again. */
1660         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1661         poll_immediate_wake();
1662     }
1663 }
1664
1665 static void
1666 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1667 {
1668     const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1669     struct cls_cursor cursor;
1670     size_t n_subfacets = 0;
1671     struct facet *facet;
1672
1673     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1674     simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1675     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1676
1677     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1678     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1679     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1680         n_subfacets += list_size(&facet->subfacets);
1681     }
1682     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1683     simap_increase(usage, "subfacets", n_subfacets);
1684 }
1685
1686 static void
1687 flush(struct ofproto *ofproto_)
1688 {
1689     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1690     struct subfacet *subfacet, *next_subfacet;
1691     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1692     int n_batch;
1693
1694     n_batch = 0;
1695     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1696                         &ofproto->backer->subfacets) {
1697         if (subfacet->facet->ofproto != ofproto) {
1698             continue;
1699         }
1700
1701         if (subfacet->path != SF_NOT_INSTALLED) {
1702             batch[n_batch++] = subfacet;
1703             if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1704                 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1705                 n_batch = 0;
1706             }
1707         } else {
1708             subfacet_destroy(subfacet);
1709         }
1710     }
1711
1712     if (n_batch > 0) {
1713         subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1714     }
1715 }
1716
1717 static void
1718 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1719              bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1720 {
1721     *arp_match_ip = true;
1722     *actions = (OFPUTIL_A_OUTPUT |
1723                 OFPUTIL_A_SET_VLAN_VID |
1724                 OFPUTIL_A_SET_VLAN_PCP |
1725                 OFPUTIL_A_STRIP_VLAN |
1726                 OFPUTIL_A_SET_DL_SRC |
1727                 OFPUTIL_A_SET_DL_DST |
1728                 OFPUTIL_A_SET_NW_SRC |
1729                 OFPUTIL_A_SET_NW_DST |
1730                 OFPUTIL_A_SET_NW_TOS |
1731                 OFPUTIL_A_SET_TP_SRC |
1732                 OFPUTIL_A_SET_TP_DST |
1733                 OFPUTIL_A_ENQUEUE);
1734 }
1735
1736 static void
1737 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1738 {
1739     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1740     struct dpif_dp_stats s;
1741     uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1742     uint64_t n_lookup;
1743
1744     strcpy(ots->name, "classifier");
1745
1746     dpif_get_dp_stats(ofproto->backer->dpif, &s);
1747     rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1748     rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1749     rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1750
1751     n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1752     ots->lookup_count = htonll(n_lookup);
1753     ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1754 }
1755
1756 static struct ofport *
1757 port_alloc(void)
1758 {
1759     struct ofport_dpif *port = xmalloc(sizeof *port);
1760     return &port->up;
1761 }
1762
1763 static void
1764 port_dealloc(struct ofport *port_)
1765 {
1766     struct ofport_dpif *port = ofport_dpif_cast(port_);
1767     free(port);
1768 }
1769
1770 static int
1771 port_construct(struct ofport *port_)
1772 {
1773     struct ofport_dpif *port = ofport_dpif_cast(port_);
1774     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1775     const struct netdev *netdev = port->up.netdev;
1776     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1777     struct dpif_port dpif_port;
1778     int error;
1779
1780     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1781     port->bundle = NULL;
1782     port->cfm = NULL;
1783     port->bfd = NULL;
1784     port->may_enable = true;
1785     port->stp_port = NULL;
1786     port->stp_state = STP_DISABLED;
1787     port->is_tunnel = false;
1788     port->peer = NULL;
1789     port->qdscp = NULL;
1790     port->n_qdscp = 0;
1791     port->realdev_ofp_port = 0;
1792     port->vlandev_vid = 0;
1793     port->carrier_seq = netdev_get_carrier_resets(netdev);
1794
1795     if (netdev_vport_is_patch(netdev)) {
1796         /* By bailing out here, we don't submit the port to the sFlow module
1797          * to be considered for counter polling export.  This is correct
1798          * because the patch port represents an interface that sFlow considers
1799          * to be "internal" to the switch as a whole, and therefore not an
1800          * candidate for counter polling. */
1801         port->odp_port = ODPP_NONE;
1802         ofport_update_peer(port);
1803         return 0;
1804     }
1805
1806     error = dpif_port_query_by_name(ofproto->backer->dpif,
1807                                     netdev_vport_get_dpif_port(netdev, namebuf,
1808                                                                sizeof namebuf),
1809                                     &dpif_port);
1810     if (error) {
1811         return error;
1812     }
1813
1814     port->odp_port = dpif_port.port_no;
1815
1816     if (netdev_get_tunnel_config(netdev)) {
1817         tnl_port_add(port, port->up.netdev, port->odp_port);
1818         port->is_tunnel = true;
1819     } else {
1820         /* Sanity-check that a mapping doesn't already exist.  This
1821          * shouldn't happen for non-tunnel ports. */
1822         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1823             VLOG_ERR("port %s already has an OpenFlow port number",
1824                      dpif_port.name);
1825             dpif_port_destroy(&dpif_port);
1826             return EBUSY;
1827         }
1828
1829         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1830         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1831                     hash_odp_port(port->odp_port));
1832         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1833     }
1834     dpif_port_destroy(&dpif_port);
1835
1836     if (ofproto->sflow) {
1837         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1838     }
1839
1840     return 0;
1841 }
1842
1843 static void
1844 port_destruct(struct ofport *port_)
1845 {
1846     struct ofport_dpif *port = ofport_dpif_cast(port_);
1847     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1848     const char *devname = netdev_get_name(port->up.netdev);
1849     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1850     const char *dp_port_name;
1851
1852     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1853     ovs_rwlock_wrlock(&xlate_rwlock);
1854     xlate_ofport_remove(port);
1855     ovs_rwlock_unlock(&xlate_rwlock);
1856
1857     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1858                                               sizeof namebuf);
1859     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1860         /* The underlying device is still there, so delete it.  This
1861          * happens when the ofproto is being destroyed, since the caller
1862          * assumes that removal of attached ports will happen as part of
1863          * destruction. */
1864         if (!port->is_tunnel) {
1865             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1866         }
1867     }
1868
1869     if (port->peer) {
1870         port->peer->peer = NULL;
1871         port->peer = NULL;
1872     }
1873
1874     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1875         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1876         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1877         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1878     }
1879
1880     tnl_port_del(port);
1881     sset_find_and_delete(&ofproto->ports, devname);
1882     sset_find_and_delete(&ofproto->ghost_ports, devname);
1883     bundle_remove(port_);
1884     set_cfm(port_, NULL);
1885     set_bfd(port_, NULL);
1886     if (ofproto->sflow) {
1887         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1888     }
1889
1890     free(port->qdscp);
1891 }
1892
1893 static void
1894 port_modified(struct ofport *port_)
1895 {
1896     struct ofport_dpif *port = ofport_dpif_cast(port_);
1897
1898     if (port->bundle && port->bundle->bond) {
1899         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1900     }
1901
1902     if (port->cfm) {
1903         cfm_set_netdev(port->cfm, port->up.netdev);
1904     }
1905
1906     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1907                                                 port->odp_port)) {
1908         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1909             REV_RECONFIGURE;
1910     }
1911
1912     ofport_update_peer(port);
1913 }
1914
1915 static void
1916 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1917 {
1918     struct ofport_dpif *port = ofport_dpif_cast(port_);
1919     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1920     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1921
1922     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1923                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1924                    OFPUTIL_PC_NO_PACKET_IN)) {
1925         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1926
1927         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1928             bundle_update(port->bundle);
1929         }
1930     }
1931 }
1932
1933 static int
1934 set_sflow(struct ofproto *ofproto_,
1935           const struct ofproto_sflow_options *sflow_options)
1936 {
1937     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1938     struct dpif_sflow *ds = ofproto->sflow;
1939
1940     if (sflow_options) {
1941         if (!ds) {
1942             struct ofport_dpif *ofport;
1943
1944             ds = ofproto->sflow = dpif_sflow_create();
1945             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1946                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1947             }
1948             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1949         }
1950         dpif_sflow_set_options(ds, sflow_options);
1951     } else {
1952         if (ds) {
1953             dpif_sflow_unref(ds);
1954             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1955             ofproto->sflow = NULL;
1956         }
1957     }
1958     return 0;
1959 }
1960
1961 static int
1962 set_ipfix(
1963     struct ofproto *ofproto_,
1964     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1965     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1966     size_t n_flow_exporters_options)
1967 {
1968     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1969     struct dpif_ipfix *di = ofproto->ipfix;
1970
1971     if (bridge_exporter_options || flow_exporters_options) {
1972         if (!di) {
1973             di = ofproto->ipfix = dpif_ipfix_create();
1974         }
1975         dpif_ipfix_set_options(
1976             di, bridge_exporter_options, flow_exporters_options,
1977             n_flow_exporters_options);
1978     } else {
1979         if (di) {
1980             dpif_ipfix_unref(di);
1981             ofproto->ipfix = NULL;
1982         }
1983     }
1984     return 0;
1985 }
1986
1987 static int
1988 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1989 {
1990     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1991     int error;
1992
1993     if (!s) {
1994         error = 0;
1995     } else {
1996         if (!ofport->cfm) {
1997             struct ofproto_dpif *ofproto;
1998
1999             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2000             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2001             ofport->cfm = cfm_create(ofport->up.netdev);
2002         }
2003
2004         if (cfm_configure(ofport->cfm, s)) {
2005             return 0;
2006         }
2007
2008         error = EINVAL;
2009     }
2010     cfm_unref(ofport->cfm);
2011     ofport->cfm = NULL;
2012     return error;
2013 }
2014
2015 static bool
2016 get_cfm_status(const struct ofport *ofport_,
2017                struct ofproto_cfm_status *status)
2018 {
2019     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2020
2021     if (ofport->cfm) {
2022         status->faults = cfm_get_fault(ofport->cfm);
2023         status->remote_opstate = cfm_get_opup(ofport->cfm);
2024         status->health = cfm_get_health(ofport->cfm);
2025         cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2026         return true;
2027     } else {
2028         return false;
2029     }
2030 }
2031
2032 static int
2033 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2034 {
2035     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2036     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2037     struct bfd *old;
2038
2039     old = ofport->bfd;
2040     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
2041     if (ofport->bfd != old) {
2042         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2043     }
2044
2045     return 0;
2046 }
2047
2048 static int
2049 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2050 {
2051     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2052
2053     if (ofport->bfd) {
2054         bfd_get_status(ofport->bfd, smap);
2055         return 0;
2056     } else {
2057         return ENOENT;
2058     }
2059 }
2060 \f
2061 /* Spanning Tree. */
2062
2063 static void
2064 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2065 {
2066     struct ofproto_dpif *ofproto = ofproto_;
2067     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2068     struct ofport_dpif *ofport;
2069
2070     ofport = stp_port_get_aux(sp);
2071     if (!ofport) {
2072         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2073                      ofproto->up.name, port_num);
2074     } else {
2075         struct eth_header *eth = pkt->l2;
2076
2077         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2078         if (eth_addr_is_zero(eth->eth_src)) {
2079             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2080                          "with unknown MAC", ofproto->up.name, port_num);
2081         } else {
2082             send_packet(ofport, pkt);
2083         }
2084     }
2085     ofpbuf_delete(pkt);
2086 }
2087
2088 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2089 static int
2090 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2091 {
2092     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2093
2094     /* Only revalidate flows if the configuration changed. */
2095     if (!s != !ofproto->stp) {
2096         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2097     }
2098
2099     if (s) {
2100         if (!ofproto->stp) {
2101             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2102                                       send_bpdu_cb, ofproto);
2103             ofproto->stp_last_tick = time_msec();
2104         }
2105
2106         stp_set_bridge_id(ofproto->stp, s->system_id);
2107         stp_set_bridge_priority(ofproto->stp, s->priority);
2108         stp_set_hello_time(ofproto->stp, s->hello_time);
2109         stp_set_max_age(ofproto->stp, s->max_age);
2110         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2111     }  else {
2112         struct ofport *ofport;
2113
2114         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2115             set_stp_port(ofport, NULL);
2116         }
2117
2118         stp_unref(ofproto->stp);
2119         ofproto->stp = NULL;
2120     }
2121
2122     return 0;
2123 }
2124
2125 static int
2126 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2127 {
2128     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2129
2130     if (ofproto->stp) {
2131         s->enabled = true;
2132         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2133         s->designated_root = stp_get_designated_root(ofproto->stp);
2134         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2135     } else {
2136         s->enabled = false;
2137     }
2138
2139     return 0;
2140 }
2141
2142 static void
2143 update_stp_port_state(struct ofport_dpif *ofport)
2144 {
2145     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2146     enum stp_state state;
2147
2148     /* Figure out new state. */
2149     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2150                              : STP_DISABLED;
2151
2152     /* Update state. */
2153     if (ofport->stp_state != state) {
2154         enum ofputil_port_state of_state;
2155         bool fwd_change;
2156
2157         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2158                     netdev_get_name(ofport->up.netdev),
2159                     stp_state_name(ofport->stp_state),
2160                     stp_state_name(state));
2161         if (stp_learn_in_state(ofport->stp_state)
2162                 != stp_learn_in_state(state)) {
2163             /* xxx Learning action flows should also be flushed. */
2164             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2165             mac_learning_flush(ofproto->ml);
2166             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2167         }
2168         fwd_change = stp_forward_in_state(ofport->stp_state)
2169                         != stp_forward_in_state(state);
2170
2171         ofproto->backer->need_revalidate = REV_STP;
2172         ofport->stp_state = state;
2173         ofport->stp_state_entered = time_msec();
2174
2175         if (fwd_change && ofport->bundle) {
2176             bundle_update(ofport->bundle);
2177         }
2178
2179         /* Update the STP state bits in the OpenFlow port description. */
2180         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2181         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2182                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2183                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2184                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2185                      : 0);
2186         ofproto_port_set_state(&ofport->up, of_state);
2187     }
2188 }
2189
2190 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2191  * caller is responsible for assigning STP port numbers and ensuring
2192  * there are no duplicates. */
2193 static int
2194 set_stp_port(struct ofport *ofport_,
2195              const struct ofproto_port_stp_settings *s)
2196 {
2197     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2198     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2199     struct stp_port *sp = ofport->stp_port;
2200
2201     if (!s || !s->enable) {
2202         if (sp) {
2203             ofport->stp_port = NULL;
2204             stp_port_disable(sp);
2205             update_stp_port_state(ofport);
2206         }
2207         return 0;
2208     } else if (sp && stp_port_no(sp) != s->port_num
2209             && ofport == stp_port_get_aux(sp)) {
2210         /* The port-id changed, so disable the old one if it's not
2211          * already in use by another port. */
2212         stp_port_disable(sp);
2213     }
2214
2215     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2216     stp_port_enable(sp);
2217
2218     stp_port_set_aux(sp, ofport);
2219     stp_port_set_priority(sp, s->priority);
2220     stp_port_set_path_cost(sp, s->path_cost);
2221
2222     update_stp_port_state(ofport);
2223
2224     return 0;
2225 }
2226
2227 static int
2228 get_stp_port_status(struct ofport *ofport_,
2229                     struct ofproto_port_stp_status *s)
2230 {
2231     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2232     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2233     struct stp_port *sp = ofport->stp_port;
2234
2235     if (!ofproto->stp || !sp) {
2236         s->enabled = false;
2237         return 0;
2238     }
2239
2240     s->enabled = true;
2241     s->port_id = stp_port_get_id(sp);
2242     s->state = stp_port_get_state(sp);
2243     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2244     s->role = stp_port_get_role(sp);
2245     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2246
2247     return 0;
2248 }
2249
2250 static void
2251 stp_run(struct ofproto_dpif *ofproto)
2252 {
2253     if (ofproto->stp) {
2254         long long int now = time_msec();
2255         long long int elapsed = now - ofproto->stp_last_tick;
2256         struct stp_port *sp;
2257
2258         if (elapsed > 0) {
2259             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2260             ofproto->stp_last_tick = now;
2261         }
2262         while (stp_get_changed_port(ofproto->stp, &sp)) {
2263             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2264
2265             if (ofport) {
2266                 update_stp_port_state(ofport);
2267             }
2268         }
2269
2270         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2271             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2272             mac_learning_flush(ofproto->ml);
2273             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2274         }
2275     }
2276 }
2277
2278 static void
2279 stp_wait(struct ofproto_dpif *ofproto)
2280 {
2281     if (ofproto->stp) {
2282         poll_timer_wait(1000);
2283     }
2284 }
2285 \f
2286 static int
2287 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2288            size_t n_qdscp)
2289 {
2290     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2291     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2292
2293     if (ofport->n_qdscp != n_qdscp
2294         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2295                               n_qdscp * sizeof *qdscp))) {
2296         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2297         free(ofport->qdscp);
2298         ofport->qdscp = n_qdscp
2299             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2300             : NULL;
2301         ofport->n_qdscp = n_qdscp;
2302     }
2303
2304     return 0;
2305 }
2306 \f
2307 /* Bundles. */
2308
2309 /* Expires all MAC learning entries associated with 'bundle' and forces its
2310  * ofproto to revalidate every flow.
2311  *
2312  * Normally MAC learning entries are removed only from the ofproto associated
2313  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2314  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2315  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2316  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2317  * with the host from which it migrated. */
2318 static void
2319 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2320 {
2321     struct ofproto_dpif *ofproto = bundle->ofproto;
2322     struct mac_learning *ml = ofproto->ml;
2323     struct mac_entry *mac, *next_mac;
2324
2325     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2326     ovs_rwlock_wrlock(&ml->rwlock);
2327     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2328         if (mac->port.p == bundle) {
2329             if (all_ofprotos) {
2330                 struct ofproto_dpif *o;
2331
2332                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2333                     if (o != ofproto) {
2334                         struct mac_entry *e;
2335
2336                         ovs_rwlock_wrlock(&o->ml->rwlock);
2337                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2338                         if (e) {
2339                             mac_learning_expire(o->ml, e);
2340                         }
2341                         ovs_rwlock_unlock(&o->ml->rwlock);
2342                     }
2343                 }
2344             }
2345
2346             mac_learning_expire(ml, mac);
2347         }
2348     }
2349     ovs_rwlock_unlock(&ml->rwlock);
2350 }
2351
2352 static struct ofbundle *
2353 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2354 {
2355     struct ofbundle *bundle;
2356
2357     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2358                              &ofproto->bundles) {
2359         if (bundle->aux == aux) {
2360             return bundle;
2361         }
2362     }
2363     return NULL;
2364 }
2365
2366 static void
2367 bundle_update(struct ofbundle *bundle)
2368 {
2369     struct ofport_dpif *port;
2370
2371     bundle->floodable = true;
2372     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2373         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2374             || !stp_forward_in_state(port->stp_state)) {
2375             bundle->floodable = false;
2376             break;
2377         }
2378     }
2379 }
2380
2381 static void
2382 bundle_del_port(struct ofport_dpif *port)
2383 {
2384     struct ofbundle *bundle = port->bundle;
2385
2386     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2387
2388     list_remove(&port->bundle_node);
2389     port->bundle = NULL;
2390
2391     if (bundle->lacp) {
2392         lacp_slave_unregister(bundle->lacp, port);
2393     }
2394     if (bundle->bond) {
2395         bond_slave_unregister(bundle->bond, port);
2396     }
2397
2398     bundle_update(bundle);
2399 }
2400
2401 static bool
2402 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2403                 struct lacp_slave_settings *lacp)
2404 {
2405     struct ofport_dpif *port;
2406
2407     port = get_ofp_port(bundle->ofproto, ofp_port);
2408     if (!port) {
2409         return false;
2410     }
2411
2412     if (port->bundle != bundle) {
2413         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2414         if (port->bundle) {
2415             bundle_remove(&port->up);
2416         }
2417
2418         port->bundle = bundle;
2419         list_push_back(&bundle->ports, &port->bundle_node);
2420         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2421             || !stp_forward_in_state(port->stp_state)) {
2422             bundle->floodable = false;
2423         }
2424     }
2425     if (lacp) {
2426         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2427         lacp_slave_register(bundle->lacp, port, lacp);
2428     }
2429
2430     return true;
2431 }
2432
2433 static void
2434 bundle_destroy(struct ofbundle *bundle)
2435 {
2436     struct ofproto_dpif *ofproto;
2437     struct ofport_dpif *port, *next_port;
2438
2439     if (!bundle) {
2440         return;
2441     }
2442
2443     ofproto = bundle->ofproto;
2444     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2445
2446     ovs_rwlock_wrlock(&xlate_rwlock);
2447     xlate_bundle_remove(bundle);
2448     ovs_rwlock_unlock(&xlate_rwlock);
2449
2450     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2451         bundle_del_port(port);
2452     }
2453
2454     bundle_flush_macs(bundle, true);
2455     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2456     free(bundle->name);
2457     free(bundle->trunks);
2458     lacp_unref(bundle->lacp);
2459     bond_unref(bundle->bond);
2460     free(bundle);
2461 }
2462
2463 static int
2464 bundle_set(struct ofproto *ofproto_, void *aux,
2465            const struct ofproto_bundle_settings *s)
2466 {
2467     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2468     bool need_flush = false;
2469     struct ofport_dpif *port;
2470     struct ofbundle *bundle;
2471     unsigned long *trunks;
2472     int vlan;
2473     size_t i;
2474     bool ok;
2475
2476     if (!s) {
2477         bundle_destroy(bundle_lookup(ofproto, aux));
2478         return 0;
2479     }
2480
2481     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2482     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2483
2484     bundle = bundle_lookup(ofproto, aux);
2485     if (!bundle) {
2486         bundle = xmalloc(sizeof *bundle);
2487
2488         bundle->ofproto = ofproto;
2489         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2490                     hash_pointer(aux, 0));
2491         bundle->aux = aux;
2492         bundle->name = NULL;
2493
2494         list_init(&bundle->ports);
2495         bundle->vlan_mode = PORT_VLAN_TRUNK;
2496         bundle->vlan = -1;
2497         bundle->trunks = NULL;
2498         bundle->use_priority_tags = s->use_priority_tags;
2499         bundle->lacp = NULL;
2500         bundle->bond = NULL;
2501
2502         bundle->floodable = true;
2503         mbridge_register_bundle(ofproto->mbridge, bundle);
2504     }
2505
2506     if (!bundle->name || strcmp(s->name, bundle->name)) {
2507         free(bundle->name);
2508         bundle->name = xstrdup(s->name);
2509     }
2510
2511     /* LACP. */
2512     if (s->lacp) {
2513         if (!bundle->lacp) {
2514             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2515             bundle->lacp = lacp_create();
2516         }
2517         lacp_configure(bundle->lacp, s->lacp);
2518     } else {
2519         lacp_unref(bundle->lacp);
2520         bundle->lacp = NULL;
2521     }
2522
2523     /* Update set of ports. */
2524     ok = true;
2525     for (i = 0; i < s->n_slaves; i++) {
2526         if (!bundle_add_port(bundle, s->slaves[i],
2527                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2528             ok = false;
2529         }
2530     }
2531     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2532         struct ofport_dpif *next_port;
2533
2534         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2535             for (i = 0; i < s->n_slaves; i++) {
2536                 if (s->slaves[i] == port->up.ofp_port) {
2537                     goto found;
2538                 }
2539             }
2540
2541             bundle_del_port(port);
2542         found: ;
2543         }
2544     }
2545     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2546
2547     if (list_is_empty(&bundle->ports)) {
2548         bundle_destroy(bundle);
2549         return EINVAL;
2550     }
2551
2552     /* Set VLAN tagging mode */
2553     if (s->vlan_mode != bundle->vlan_mode
2554         || s->use_priority_tags != bundle->use_priority_tags) {
2555         bundle->vlan_mode = s->vlan_mode;
2556         bundle->use_priority_tags = s->use_priority_tags;
2557         need_flush = true;
2558     }
2559
2560     /* Set VLAN tag. */
2561     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2562             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2563             : 0);
2564     if (vlan != bundle->vlan) {
2565         bundle->vlan = vlan;
2566         need_flush = true;
2567     }
2568
2569     /* Get trunked VLANs. */
2570     switch (s->vlan_mode) {
2571     case PORT_VLAN_ACCESS:
2572         trunks = NULL;
2573         break;
2574
2575     case PORT_VLAN_TRUNK:
2576         trunks = CONST_CAST(unsigned long *, s->trunks);
2577         break;
2578
2579     case PORT_VLAN_NATIVE_UNTAGGED:
2580     case PORT_VLAN_NATIVE_TAGGED:
2581         if (vlan != 0 && (!s->trunks
2582                           || !bitmap_is_set(s->trunks, vlan)
2583                           || bitmap_is_set(s->trunks, 0))) {
2584             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2585             if (s->trunks) {
2586                 trunks = bitmap_clone(s->trunks, 4096);
2587             } else {
2588                 trunks = bitmap_allocate1(4096);
2589             }
2590             bitmap_set1(trunks, vlan);
2591             bitmap_set0(trunks, 0);
2592         } else {
2593             trunks = CONST_CAST(unsigned long *, s->trunks);
2594         }
2595         break;
2596
2597     default:
2598         NOT_REACHED();
2599     }
2600     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2601         free(bundle->trunks);
2602         if (trunks == s->trunks) {
2603             bundle->trunks = vlan_bitmap_clone(trunks);
2604         } else {
2605             bundle->trunks = trunks;
2606             trunks = NULL;
2607         }
2608         need_flush = true;
2609     }
2610     if (trunks != s->trunks) {
2611         free(trunks);
2612     }
2613
2614     /* Bonding. */
2615     if (!list_is_short(&bundle->ports)) {
2616         bundle->ofproto->has_bonded_bundles = true;
2617         if (bundle->bond) {
2618             if (bond_reconfigure(bundle->bond, s->bond)) {
2619                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2620             }
2621         } else {
2622             bundle->bond = bond_create(s->bond);
2623             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2624         }
2625
2626         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2627             bond_slave_register(bundle->bond, port, port->up.netdev);
2628         }
2629     } else {
2630         bond_unref(bundle->bond);
2631         bundle->bond = NULL;
2632     }
2633
2634     /* If we changed something that would affect MAC learning, un-learn
2635      * everything on this port and force flow revalidation. */
2636     if (need_flush) {
2637         bundle_flush_macs(bundle, false);
2638     }
2639
2640     return 0;
2641 }
2642
2643 static void
2644 bundle_remove(struct ofport *port_)
2645 {
2646     struct ofport_dpif *port = ofport_dpif_cast(port_);
2647     struct ofbundle *bundle = port->bundle;
2648
2649     if (bundle) {
2650         bundle_del_port(port);
2651         if (list_is_empty(&bundle->ports)) {
2652             bundle_destroy(bundle);
2653         } else if (list_is_short(&bundle->ports)) {
2654             bond_unref(bundle->bond);
2655             bundle->bond = NULL;
2656         }
2657     }
2658 }
2659
2660 static void
2661 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2662 {
2663     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2664     struct ofport_dpif *port = port_;
2665     uint8_t ea[ETH_ADDR_LEN];
2666     int error;
2667
2668     error = netdev_get_etheraddr(port->up.netdev, ea);
2669     if (!error) {
2670         struct ofpbuf packet;
2671         void *packet_pdu;
2672
2673         ofpbuf_init(&packet, 0);
2674         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2675                                  pdu_size);
2676         memcpy(packet_pdu, pdu, pdu_size);
2677
2678         send_packet(port, &packet);
2679         ofpbuf_uninit(&packet);
2680     } else {
2681         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2682                     "%s (%s)", port->bundle->name,
2683                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2684     }
2685 }
2686
2687 static void
2688 bundle_send_learning_packets(struct ofbundle *bundle)
2689 {
2690     struct ofproto_dpif *ofproto = bundle->ofproto;
2691     int error, n_packets, n_errors;
2692     struct mac_entry *e;
2693
2694     error = n_packets = n_errors = 0;
2695     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2696     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2697         if (e->port.p != bundle) {
2698             struct ofpbuf *learning_packet;
2699             struct ofport_dpif *port;
2700             void *port_void;
2701             int ret;
2702
2703             /* The assignment to "port" is unnecessary but makes "grep"ing for
2704              * struct ofport_dpif more effective. */
2705             learning_packet = bond_compose_learning_packet(bundle->bond,
2706                                                            e->mac, e->vlan,
2707                                                            &port_void);
2708             port = port_void;
2709             ret = send_packet(port, learning_packet);
2710             ofpbuf_delete(learning_packet);
2711             if (ret) {
2712                 error = ret;
2713                 n_errors++;
2714             }
2715             n_packets++;
2716         }
2717     }
2718     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2719
2720     if (n_errors) {
2721         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2722         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2723                      "packets, last error was: %s",
2724                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2725     } else {
2726         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2727                  bundle->name, n_packets);
2728     }
2729 }
2730
2731 static void
2732 bundle_run(struct ofbundle *bundle)
2733 {
2734     if (bundle->lacp) {
2735         lacp_run(bundle->lacp, send_pdu_cb);
2736     }
2737     if (bundle->bond) {
2738         struct ofport_dpif *port;
2739
2740         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2741             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2742         }
2743
2744         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2745             bundle->ofproto->backer->need_revalidate = REV_BOND;
2746         }
2747
2748         if (bond_should_send_learning_packets(bundle->bond)) {
2749             bundle_send_learning_packets(bundle);
2750         }
2751     }
2752 }
2753
2754 static void
2755 bundle_wait(struct ofbundle *bundle)
2756 {
2757     if (bundle->lacp) {
2758         lacp_wait(bundle->lacp);
2759     }
2760     if (bundle->bond) {
2761         bond_wait(bundle->bond);
2762     }
2763 }
2764 \f
2765 /* Mirrors. */
2766
2767 static int
2768 mirror_set__(struct ofproto *ofproto_, void *aux,
2769              const struct ofproto_mirror_settings *s)
2770 {
2771     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2772     struct ofbundle **srcs, **dsts;
2773     int error;
2774     size_t i;
2775
2776     if (!s) {
2777         mirror_destroy(ofproto->mbridge, aux);
2778         return 0;
2779     }
2780
2781     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2782     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2783
2784     for (i = 0; i < s->n_srcs; i++) {
2785         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2786     }
2787
2788     for (i = 0; i < s->n_dsts; i++) {
2789         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2790     }
2791
2792     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2793                        s->n_dsts, s->src_vlans,
2794                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2795     free(srcs);
2796     free(dsts);
2797     return error;
2798 }
2799
2800 static int
2801 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2802                    uint64_t *packets, uint64_t *bytes)
2803 {
2804     push_all_stats();
2805     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2806                             bytes);
2807 }
2808
2809 static int
2810 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2811 {
2812     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2813     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2814     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2815         mac_learning_flush(ofproto->ml);
2816     }
2817     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2818     return 0;
2819 }
2820
2821 static bool
2822 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2823 {
2824     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2825     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2826     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2827 }
2828
2829 static void
2830 forward_bpdu_changed(struct ofproto *ofproto_)
2831 {
2832     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2833     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2834 }
2835
2836 static void
2837 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2838                      size_t max_entries)
2839 {
2840     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2841     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2842     mac_learning_set_idle_time(ofproto->ml, idle_time);
2843     mac_learning_set_max_entries(ofproto->ml, max_entries);
2844     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2845 }
2846 \f
2847 /* Ports. */
2848
2849 static struct ofport_dpif *
2850 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2851 {
2852     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2853     return ofport ? ofport_dpif_cast(ofport) : NULL;
2854 }
2855
2856 static struct ofport_dpif *
2857 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2858 {
2859     struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2860     return port && &ofproto->up == port->up.ofproto ? port : NULL;
2861 }
2862
2863 static void
2864 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2865                             struct ofproto_port *ofproto_port,
2866                             struct dpif_port *dpif_port)
2867 {
2868     ofproto_port->name = dpif_port->name;
2869     ofproto_port->type = dpif_port->type;
2870     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2871 }
2872
2873 static void
2874 ofport_update_peer(struct ofport_dpif *ofport)
2875 {
2876     const struct ofproto_dpif *ofproto;
2877     struct dpif_backer *backer;
2878     char *peer_name;
2879
2880     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2881         return;
2882     }
2883
2884     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2885     backer->need_revalidate = REV_RECONFIGURE;
2886
2887     if (ofport->peer) {
2888         ofport->peer->peer = NULL;
2889         ofport->peer = NULL;
2890     }
2891
2892     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2893     if (!peer_name) {
2894         return;
2895     }
2896
2897     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2898         struct ofport *peer_ofport;
2899         struct ofport_dpif *peer;
2900         char *peer_peer;
2901
2902         if (ofproto->backer != backer) {
2903             continue;
2904         }
2905
2906         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2907         if (!peer_ofport) {
2908             continue;
2909         }
2910
2911         peer = ofport_dpif_cast(peer_ofport);
2912         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2913         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2914                                  peer_peer)) {
2915             ofport->peer = peer;
2916             ofport->peer->peer = ofport;
2917         }
2918         free(peer_peer);
2919
2920         break;
2921     }
2922     free(peer_name);
2923 }
2924
2925 static void
2926 port_run_fast(struct ofport_dpif *ofport)
2927 {
2928     if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2929         struct ofpbuf packet;
2930
2931         ofpbuf_init(&packet, 0);
2932         cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2933         send_packet(ofport, &packet);
2934         ofpbuf_uninit(&packet);
2935     }
2936
2937     if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2938         struct ofpbuf packet;
2939
2940         ofpbuf_init(&packet, 0);
2941         bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2942         send_packet(ofport, &packet);
2943         ofpbuf_uninit(&packet);
2944     }
2945 }
2946
2947 static void
2948 port_run(struct ofport_dpif *ofport)
2949 {
2950     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2951     bool carrier_changed = carrier_seq != ofport->carrier_seq;
2952     bool enable = netdev_get_carrier(ofport->up.netdev);
2953
2954     ofport->carrier_seq = carrier_seq;
2955
2956     port_run_fast(ofport);
2957
2958     if (ofport->cfm) {
2959         int cfm_opup = cfm_get_opup(ofport->cfm);
2960
2961         cfm_run(ofport->cfm);
2962         enable = enable && !cfm_get_fault(ofport->cfm);
2963
2964         if (cfm_opup >= 0) {
2965             enable = enable && cfm_opup;
2966         }
2967     }
2968
2969     if (ofport->bfd) {
2970         bfd_run(ofport->bfd);
2971         enable = enable && bfd_forwarding(ofport->bfd);
2972     }
2973
2974     if (ofport->bundle) {
2975         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2976         if (carrier_changed) {
2977             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2978         }
2979     }
2980
2981     if (ofport->may_enable != enable) {
2982         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2983         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2984     }
2985
2986     ofport->may_enable = enable;
2987 }
2988
2989 static void
2990 port_wait(struct ofport_dpif *ofport)
2991 {
2992     if (ofport->cfm) {
2993         cfm_wait(ofport->cfm);
2994     }
2995
2996     if (ofport->bfd) {
2997         bfd_wait(ofport->bfd);
2998     }
2999 }
3000
3001 static int
3002 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3003                    struct ofproto_port *ofproto_port)
3004 {
3005     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3006     struct dpif_port dpif_port;
3007     int error;
3008
3009     if (sset_contains(&ofproto->ghost_ports, devname)) {
3010         const char *type = netdev_get_type_from_name(devname);
3011
3012         /* We may be called before ofproto->up.port_by_name is populated with
3013          * the appropriate ofport.  For this reason, we must get the name and
3014          * type from the netdev layer directly. */
3015         if (type) {
3016             const struct ofport *ofport;
3017
3018             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3019             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3020             ofproto_port->name = xstrdup(devname);
3021             ofproto_port->type = xstrdup(type);
3022             return 0;
3023         }
3024         return ENODEV;
3025     }
3026
3027     if (!sset_contains(&ofproto->ports, devname)) {
3028         return ENODEV;
3029     }
3030     error = dpif_port_query_by_name(ofproto->backer->dpif,
3031                                     devname, &dpif_port);
3032     if (!error) {
3033         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3034     }
3035     return error;
3036 }
3037
3038 static int
3039 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3040 {
3041     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3042     const char *devname = netdev_get_name(netdev);
3043     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
3044     const char *dp_port_name;
3045
3046     if (netdev_vport_is_patch(netdev)) {
3047         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3048         return 0;
3049     }
3050
3051     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
3052     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3053         odp_port_t port_no = ODPP_NONE;
3054         int error;
3055
3056         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3057         if (error) {
3058             return error;
3059         }
3060         if (netdev_get_tunnel_config(netdev)) {
3061             simap_put(&ofproto->backer->tnl_backers,
3062                       dp_port_name, odp_to_u32(port_no));
3063         }
3064     }
3065
3066     if (netdev_get_tunnel_config(netdev)) {
3067         sset_add(&ofproto->ghost_ports, devname);
3068     } else {
3069         sset_add(&ofproto->ports, devname);
3070     }
3071     return 0;
3072 }
3073
3074 static int
3075 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3076 {
3077     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3078     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3079     int error = 0;
3080
3081     if (!ofport) {
3082         return 0;
3083     }
3084
3085     sset_find_and_delete(&ofproto->ghost_ports,
3086                          netdev_get_name(ofport->up.netdev));
3087     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3088     if (!ofport->is_tunnel) {
3089         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3090         if (!error) {
3091             /* The caller is going to close ofport->up.netdev.  If this is a
3092              * bonded port, then the bond is using that netdev, so remove it
3093              * from the bond.  The client will need to reconfigure everything
3094              * after deleting ports, so then the slave will get re-added. */
3095             bundle_remove(&ofport->up);
3096         }
3097     }
3098     return error;
3099 }
3100
3101 static int
3102 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3103 {
3104     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3105     int error;
3106
3107     push_all_stats();
3108
3109     error = netdev_get_stats(ofport->up.netdev, stats);
3110
3111     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3112         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3113
3114         /* ofproto->stats.tx_packets represents packets that we created
3115          * internally and sent to some port (e.g. packets sent with
3116          * send_packet()).  Account for them as if they had come from
3117          * OFPP_LOCAL and got forwarded. */
3118
3119         if (stats->rx_packets != UINT64_MAX) {
3120             stats->rx_packets += ofproto->stats.tx_packets;
3121         }
3122
3123         if (stats->rx_bytes != UINT64_MAX) {
3124             stats->rx_bytes += ofproto->stats.tx_bytes;
3125         }
3126
3127         /* ofproto->stats.rx_packets represents packets that were received on
3128          * some port and we processed internally and dropped (e.g. STP).
3129          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3130
3131         if (stats->tx_packets != UINT64_MAX) {
3132             stats->tx_packets += ofproto->stats.rx_packets;
3133         }
3134
3135         if (stats->tx_bytes != UINT64_MAX) {
3136             stats->tx_bytes += ofproto->stats.rx_bytes;
3137         }
3138     }
3139
3140     return error;
3141 }
3142
3143 struct port_dump_state {
3144     uint32_t bucket;
3145     uint32_t offset;
3146     bool ghost;
3147
3148     struct ofproto_port port;
3149     bool has_port;
3150 };
3151
3152 static int
3153 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3154 {
3155     *statep = xzalloc(sizeof(struct port_dump_state));
3156     return 0;
3157 }
3158
3159 static int
3160 port_dump_next(const struct ofproto *ofproto_, void *state_,
3161                struct ofproto_port *port)
3162 {
3163     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3164     struct port_dump_state *state = state_;
3165     const struct sset *sset;
3166     struct sset_node *node;
3167
3168     if (state->has_port) {
3169         ofproto_port_destroy(&state->port);
3170         state->has_port = false;
3171     }
3172     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3173     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3174         int error;
3175
3176         error = port_query_by_name(ofproto_, node->name, &state->port);
3177         if (!error) {
3178             *port = state->port;
3179             state->has_port = true;
3180             return 0;
3181         } else if (error != ENODEV) {
3182             return error;
3183         }
3184     }
3185
3186     if (!state->ghost) {
3187         state->ghost = true;
3188         state->bucket = 0;
3189         state->offset = 0;
3190         return port_dump_next(ofproto_, state_, port);
3191     }
3192
3193     return EOF;
3194 }
3195
3196 static int
3197 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3198 {
3199     struct port_dump_state *state = state_;
3200
3201     if (state->has_port) {
3202         ofproto_port_destroy(&state->port);
3203     }
3204     free(state);
3205     return 0;
3206 }
3207
3208 static int
3209 port_poll(const struct ofproto *ofproto_, char **devnamep)
3210 {
3211     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3212
3213     if (ofproto->port_poll_errno) {
3214         int error = ofproto->port_poll_errno;
3215         ofproto->port_poll_errno = 0;
3216         return error;
3217     }
3218
3219     if (sset_is_empty(&ofproto->port_poll_set)) {
3220         return EAGAIN;
3221     }
3222
3223     *devnamep = sset_pop(&ofproto->port_poll_set);
3224     return 0;
3225 }
3226
3227 static void
3228 port_poll_wait(const struct ofproto *ofproto_)
3229 {
3230     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3231     dpif_port_poll_wait(ofproto->backer->dpif);
3232 }
3233
3234 static int
3235 port_is_lacp_current(const struct ofport *ofport_)
3236 {
3237     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3238     return (ofport->bundle && ofport->bundle->lacp
3239             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3240             : -1);
3241 }
3242 \f
3243 /* Upcall handling. */
3244
3245 struct flow_miss_op {
3246     struct dpif_op dpif_op;
3247
3248     uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3249     struct xlate_out xout;
3250     bool xout_garbage;           /* 'xout' needs to be uninitialized? */
3251
3252     struct ofpbuf mask;          /* Flow mask for "put" ops. */
3253     struct odputil_keybuf maskbuf;
3254
3255     /* If this is a "put" op, then a pointer to the subfacet that should
3256      * be marked as uninstalled if the operation fails. */
3257     struct subfacet *subfacet;
3258 };
3259
3260 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3261  * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3262  * and (usually) installing a datapath flow.  The answer is usually "yes" (a
3263  * return value of true).  However, for short flows the cost of bookkeeping is
3264  * much higher than the benefits, so when the datapath holds a large number of
3265  * flows we impose some heuristics to decide which flows are likely to be worth
3266  * tracking. */
3267 static bool
3268 flow_miss_should_make_facet(struct flow_miss *miss)
3269 {
3270     struct dpif_backer *backer = miss->ofproto->backer;
3271     uint32_t hash;
3272
3273     switch (flow_miss_model) {
3274     case OFPROTO_HANDLE_MISS_AUTO:
3275         break;
3276     case OFPROTO_HANDLE_MISS_WITH_FACETS:
3277         return true;
3278     case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3279         return false;
3280     }
3281
3282     if (!backer->governor) {
3283         size_t n_subfacets;
3284
3285         n_subfacets = hmap_count(&backer->subfacets);
3286         if (n_subfacets * 2 <= flow_eviction_threshold) {
3287             return true;
3288         }
3289
3290         backer->governor = governor_create();
3291     }
3292
3293     hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3294     return governor_should_install_flow(backer->governor, hash,
3295                                         list_size(&miss->packets));
3296 }
3297
3298 /* Handles 'miss', which matches 'facet'.  May add any required datapath
3299  * operations to 'ops', incrementing '*n_ops' for each new op.
3300  *
3301  * All of the packets in 'miss' are considered to have arrived at time
3302  * 'miss->stats.used'.  This is really important only for new facets: if we
3303  * just called time_msec() here, then the new subfacet or its packets could
3304  * look (occasionally) as though it was used some time after the facet was
3305  * used.  That can make a one-packet flow look like it has a nonzero duration,
3306  * which looks odd in e.g. NetFlow statistics. */
3307 static void
3308 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3309                             struct flow_miss_op *ops, size_t *n_ops)
3310 {
3311     enum subfacet_path want_path;
3312     struct subfacet *subfacet;
3313
3314     facet->packet_count += miss->stats.n_packets;
3315     facet->prev_packet_count += miss->stats.n_packets;
3316     facet->byte_count += miss->stats.n_bytes;
3317     facet->prev_byte_count += miss->stats.n_bytes;
3318
3319     subfacet = subfacet_create(facet, miss);
3320     want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3321
3322     /* Don't install the flow if it's the result of the "userspace"
3323      * action for an already installed facet.  This can occur when a
3324      * datapath flow with wildcards has a "userspace" action and flows
3325      * sent to userspace result in a different subfacet, which will then
3326      * be rejected as overlapping by the datapath. */
3327     if (miss->upcall_type == DPIF_UC_ACTION
3328         && !list_is_empty(&facet->subfacets)) {
3329         return;
3330     }
3331
3332     subfacet = subfacet_create(facet, miss);
3333     if (subfacet->path != want_path) {
3334         struct flow_miss_op *op = &ops[(*n_ops)++];
3335         struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3336
3337         subfacet->path = want_path;
3338
3339         ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3340         if (enable_megaflows) {
3341             odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3342                                    &miss->flow, UINT32_MAX);
3343         }
3344
3345         op->xout_garbage = false;
3346         op->dpif_op.type = DPIF_OP_FLOW_PUT;
3347         op->subfacet = subfacet;
3348         put->flags = DPIF_FP_CREATE;
3349         put->key = miss->key;
3350         put->key_len = miss->key_len;
3351         put->mask = op->mask.data;
3352         put->mask_len = op->mask.size;
3353
3354         if (want_path == SF_FAST_PATH) {
3355             put->actions = facet->xout.odp_actions.data;
3356             put->actions_len = facet->xout.odp_actions.size;
3357         } else {
3358             compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3359                               op->slow_stub, sizeof op->slow_stub,
3360                               &put->actions, &put->actions_len);
3361         }
3362         put->stats = NULL;
3363     }
3364 }
3365
3366 /* Handles flow miss 'miss'.  May add any required datapath operations
3367  * to 'ops', incrementing '*n_ops' for each new op. */
3368 static void
3369 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3370                  size_t *n_ops)
3371 {
3372     struct facet *facet;
3373
3374     miss->ofproto->n_missed += list_size(&miss->packets);
3375
3376     facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3377     if (!facet) {
3378         /* There does not exist a bijection between 'struct flow' and datapath
3379          * flow keys with fitness ODP_FIT_TO_LITTLE.  This breaks a fundamental
3380          * assumption used throughout the facet and subfacet handling code.
3381          * Since we have to handle these misses in userspace anyway, we simply
3382          * skip facet creation, avoiding the problem altogether. */
3383         if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3384             || !flow_miss_should_make_facet(miss)) {
3385             return;
3386         }
3387
3388         facet = facet_create(miss);
3389     }
3390     handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3391 }
3392
3393 static struct drop_key *
3394 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3395                 size_t key_len)
3396 {
3397     struct drop_key *drop_key;
3398
3399     HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3400                              &backer->drop_keys) {
3401         if (drop_key->key_len == key_len
3402             && !memcmp(drop_key->key, key, key_len)) {
3403             return drop_key;
3404         }
3405     }
3406     return NULL;
3407 }
3408
3409 static void
3410 drop_key_clear(struct dpif_backer *backer)
3411 {
3412     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3413     struct drop_key *drop_key, *next;
3414
3415     HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3416         int error;
3417
3418         error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3419                               NULL);
3420         if (error && !VLOG_DROP_WARN(&rl)) {
3421             struct ds ds = DS_EMPTY_INITIALIZER;
3422             odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3423             VLOG_WARN("Failed to delete drop key (%s) (%s)",
3424                       ovs_strerror(error), ds_cstr(&ds));
3425             ds_destroy(&ds);
3426         }
3427
3428         hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3429         drop_key_destroy(drop_key);
3430     }
3431
3432     udpif_drop_key_clear(backer->udpif);
3433 }
3434
3435 static void
3436 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3437 {
3438     struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3439     struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3440     struct flow_miss *miss;
3441     size_t n_ops, i;
3442
3443     /* Process each element in the to-do list, constructing the set of
3444      * operations to batch. */
3445     n_ops = 0;
3446     HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3447         handle_flow_miss(miss, flow_miss_ops, &n_ops);
3448     }
3449     ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3450
3451     /* Execute batch. */
3452     for (i = 0; i < n_ops; i++) {
3453         dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3454     }
3455     dpif_operate(backer->dpif, dpif_ops, n_ops);
3456
3457     for (i = 0; i < n_ops; i++) {
3458         if (dpif_ops[i]->error != 0
3459             && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3460             && flow_miss_ops[i].subfacet) {
3461             struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3462
3463             COVERAGE_INC(subfacet_install_fail);
3464
3465             /* Zero-out subfacet counters when installation failed, but
3466              * datapath reported hits.  This should not happen and
3467              * indicates a bug, since if the datapath flow exists, we
3468              * should not be attempting to create a new subfacet.  A
3469              * buggy datapath could trigger this, so just zero out the
3470              * counters and log an error. */
3471             if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3472                 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3473                             "datapath reported hits");
3474                 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3475             }
3476
3477             subfacet->path = SF_NOT_INSTALLED;
3478         }
3479     }
3480 }
3481
3482 static void
3483 handle_sflow_upcall(struct dpif_backer *backer,
3484                     const struct dpif_upcall *upcall)
3485 {
3486     struct ofproto_dpif *ofproto;
3487     union user_action_cookie cookie;
3488     struct flow flow;
3489     odp_port_t odp_in_port;
3490
3491     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3492                       &flow, NULL, &ofproto, &odp_in_port)
3493         || !ofproto->sflow) {
3494         return;
3495     }
3496
3497     memset(&cookie, 0, sizeof cookie);
3498     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
3499     dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3500                         odp_in_port, &cookie);
3501 }
3502
3503 static void
3504 handle_flow_sample_upcall(struct dpif_backer *backer,
3505                           const struct dpif_upcall *upcall)
3506 {
3507     struct ofproto_dpif *ofproto;
3508     union user_action_cookie cookie;
3509     struct flow flow;
3510
3511     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3512                       &flow, NULL, &ofproto, NULL)
3513         || !ofproto->ipfix) {
3514         return;
3515     }
3516
3517     memset(&cookie, 0, sizeof cookie);
3518     memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
3519
3520     /* The flow reflects exactly the contents of the packet.  Sample
3521      * the packet using it. */
3522     dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
3523                            cookie.flow_sample.collector_set_id,
3524                            cookie.flow_sample.probability,
3525                            cookie.flow_sample.obs_domain_id,
3526                            cookie.flow_sample.obs_point_id);
3527 }
3528
3529 static void
3530 handle_ipfix_upcall(struct dpif_backer *backer,
3531                     const struct dpif_upcall *upcall)
3532 {
3533     struct ofproto_dpif *ofproto;
3534     struct flow flow;
3535
3536     if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3537                       &flow, NULL, &ofproto, NULL)
3538         || !ofproto->ipfix) {
3539         return;
3540     }
3541
3542     /* The flow reflects exactly the contents of the packet.  Sample
3543      * the packet using it. */
3544     dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
3545 }
3546
3547 static void
3548 handle_upcalls(struct dpif_backer *backer)
3549 {
3550     struct flow_miss_batch *fmb;
3551     int n_processed;
3552
3553     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3554         struct upcall *upcall = upcall_next(backer->udpif);
3555
3556         if (!upcall) {
3557             break;
3558         }
3559
3560         switch (upcall->type) {
3561         case SFLOW_UPCALL:
3562             handle_sflow_upcall(backer, &upcall->dpif_upcall);
3563             break;
3564
3565         case FLOW_SAMPLE_UPCALL:
3566             handle_flow_sample_upcall(backer, &upcall->dpif_upcall);
3567             break;
3568
3569         case IPFIX_UPCALL:
3570             handle_ipfix_upcall(backer, &upcall->dpif_upcall);
3571             break;
3572
3573         case BAD_UPCALL:
3574             break;
3575
3576         case MISS_UPCALL:
3577             NOT_REACHED();
3578         }
3579
3580         upcall_destroy(upcall);
3581     }
3582
3583     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3584         struct drop_key *drop_key = drop_key_next(backer->udpif);
3585         if (!drop_key) {
3586             break;
3587         }
3588
3589         if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3590             hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3591                         hash_bytes(drop_key->key, drop_key->key_len, 0));
3592             dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3593                           drop_key->key, drop_key->key_len,
3594                           NULL, 0, NULL, 0, NULL);
3595         } else {
3596             drop_key_destroy(drop_key);
3597         }
3598     }
3599
3600     fmb = flow_miss_batch_next(backer->udpif);
3601     if (fmb) {
3602         handle_flow_misses(backer, fmb);
3603         flow_miss_batch_destroy(fmb);
3604     }
3605 }
3606 \f
3607 /* Flow expiration. */
3608
3609 static int subfacet_max_idle(const struct dpif_backer *);
3610 static void update_stats(struct dpif_backer *);
3611 static void rule_expire(struct rule_dpif *);
3612 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3613
3614 /* This function is called periodically by run().  Its job is to collect
3615  * updates for the flows that have been installed into the datapath, most
3616  * importantly when they last were used, and then use that information to
3617  * expire flows that have not been used recently.
3618  *
3619  * Returns the number of milliseconds after which it should be called again. */
3620 static int
3621 expire(struct dpif_backer *backer)
3622 {
3623     struct ofproto_dpif *ofproto;
3624     size_t n_subfacets;
3625     int max_idle;
3626
3627     /* Periodically clear out the drop keys in an effort to keep them
3628      * relatively few. */
3629     drop_key_clear(backer);
3630
3631     /* Update stats for each flow in the backer. */
3632     update_stats(backer);
3633
3634     n_subfacets = hmap_count(&backer->subfacets);
3635     if (n_subfacets) {
3636         struct subfacet *subfacet;
3637         long long int total, now;
3638
3639         total = 0;
3640         now = time_msec();
3641         HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3642             total += now - subfacet->created;
3643         }
3644         backer->avg_subfacet_life += total / n_subfacets;
3645     }
3646     backer->avg_subfacet_life /= 2;
3647
3648     backer->avg_n_subfacet += n_subfacets;
3649     backer->avg_n_subfacet /= 2;
3650
3651     backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3652
3653     max_idle = subfacet_max_idle(backer);
3654     expire_subfacets(backer, max_idle);
3655
3656     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3657         struct rule *rule, *next_rule;
3658
3659         if (ofproto->backer != backer) {
3660             continue;
3661         }
3662
3663         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3664          * has passed. */
3665         ovs_mutex_lock(&ofproto->up.expirable_mutex);
3666         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3667                             &ofproto->up.expirable) {
3668             rule_expire(rule_dpif_cast(rule));
3669         }
3670         ovs_mutex_unlock(&ofproto->up.expirable_mutex);
3671
3672         /* All outstanding data in existing flows has been accounted, so it's a
3673          * good time to do bond rebalancing. */
3674         if (ofproto->has_bonded_bundles) {
3675             struct ofbundle *bundle;
3676
3677             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3678                 if (bundle->bond) {
3679                     bond_rebalance(bundle->bond);
3680                 }
3681             }
3682         }
3683     }
3684
3685     return MIN(max_idle, 1000);
3686 }
3687
3688 /* Updates flow table statistics given that the datapath just reported 'stats'
3689  * as 'subfacet''s statistics. */
3690 static void
3691 update_subfacet_stats(struct subfacet *subfacet,
3692                       const struct dpif_flow_stats *stats)
3693 {
3694     struct facet *facet = subfacet->facet;
3695     struct dpif_flow_stats diff;
3696
3697     diff.tcp_flags = stats->tcp_flags;
3698     diff.used = stats->used;
3699
3700     if (stats->n_packets >= subfacet->dp_packet_count) {
3701         diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3702     } else {
3703         VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3704         diff.n_packets = 0;
3705     }
3706
3707     if (stats->n_bytes >= subfacet->dp_byte_count) {
3708         diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3709     } else {
3710         VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3711         diff.n_bytes = 0;
3712     }
3713
3714     facet->ofproto->n_hit += diff.n_packets;
3715     subfacet->dp_packet_count = stats->n_packets;
3716     subfacet->dp_byte_count = stats->n_bytes;
3717     subfacet_update_stats(subfacet, &diff);
3718
3719     if (facet->accounted_bytes < facet->byte_count) {
3720         facet_learn(facet);
3721         facet_account(facet);
3722         facet->accounted_bytes = facet->byte_count;
3723     }
3724 }
3725
3726 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3727  * about, or a flow that shouldn't be installed but was anyway.  Delete it. */
3728 static void
3729 delete_unexpected_flow(struct dpif_backer *backer,
3730                        const struct nlattr *key, size_t key_len)
3731 {
3732     if (!VLOG_DROP_WARN(&rl)) {
3733         struct ds s;
3734
3735         ds_init(&s);
3736         odp_flow_key_format(key, key_len, &s);
3737         VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3738         ds_destroy(&s);
3739     }
3740
3741     COVERAGE_INC(facet_unexpected);
3742     dpif_flow_del(backer->dpif, key, key_len, NULL);
3743 }
3744
3745 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3746  *
3747  * This function also pushes statistics updates to rules which each facet
3748  * resubmits into.  Generally these statistics will be accurate.  However, if a
3749  * facet changes the rule it resubmits into at some time in between
3750  * update_stats() runs, it is possible that statistics accrued to the
3751  * old rule will be incorrectly attributed to the new rule.  This could be
3752  * avoided by calling update_stats() whenever rules are created or
3753  * deleted.  However, the performance impact of making so many calls to the
3754  * datapath do not justify the benefit of having perfectly accurate statistics.
3755  *
3756  * In addition, this function maintains per ofproto flow hit counts. The patch
3757  * port is not treated specially. e.g. A packet ingress from br0 patched into
3758  * br1 will increase the hit count of br0 by 1, however, does not affect
3759  * the hit or miss counts of br1.
3760  */
3761 static void
3762 update_stats(struct dpif_backer *backer)
3763 {
3764     const struct dpif_flow_stats *stats;
3765     struct dpif_flow_dump dump;
3766     const struct nlattr *key, *mask;
3767     size_t key_len, mask_len;
3768
3769     dpif_flow_dump_start(&dump, backer->dpif);
3770     while (dpif_flow_dump_next(&dump, &key, &key_len,
3771                                &mask, &mask_len, NULL, NULL, &stats)) {
3772         struct subfacet *subfacet;
3773         uint32_t key_hash;
3774
3775         key_hash = odp_flow_key_hash(key, key_len);
3776         subfacet = subfacet_find(backer, key, key_len, key_hash);
3777         switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3778         case SF_FAST_PATH:
3779             update_subfacet_stats(subfacet, stats);
3780             break;
3781
3782         case SF_SLOW_PATH:
3783             /* Stats are updated per-packet. */
3784             break;
3785
3786         case SF_NOT_INSTALLED:
3787         default:
3788             delete_unexpected_flow(backer, key, key_len);
3789             break;
3790         }
3791         run_fast_rl();
3792     }
3793     dpif_flow_dump_done(&dump);
3794
3795     update_moving_averages(backer);
3796 }
3797
3798 /* Calculates and returns the number of milliseconds of idle time after which
3799  * subfacets should expire from the datapath.  When a subfacet expires, we fold
3800  * its statistics into its facet, and when a facet's last subfacet expires, we
3801  * fold its statistic into its rule. */
3802 static int
3803 subfacet_max_idle(const struct dpif_backer *backer)
3804 {
3805     /*
3806      * Idle time histogram.
3807      *
3808      * Most of the time a switch has a relatively small number of subfacets.
3809      * When this is the case we might as well keep statistics for all of them
3810      * in userspace and to cache them in the kernel datapath for performance as
3811      * well.
3812      *
3813      * As the number of subfacets increases, the memory required to maintain
3814      * statistics about them in userspace and in the kernel becomes
3815      * significant.  However, with a large number of subfacets it is likely
3816      * that only a few of them are "heavy hitters" that consume a large amount
3817      * of bandwidth.  At this point, only heavy hitters are worth caching in
3818      * the kernel and maintaining in userspaces; other subfacets we can
3819      * discard.
3820      *
3821      * The technique used to compute the idle time is to build a histogram with
3822      * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each.  Each subfacet
3823      * that is installed in the kernel gets dropped in the appropriate bucket.
3824      * After the histogram has been built, we compute the cutoff so that only
3825      * the most-recently-used 1% of subfacets (but at least
3826      * flow_eviction_threshold flows) are kept cached.  At least
3827      * the most-recently-used bucket of subfacets is kept, so actually an
3828      * arbitrary number of subfacets can be kept in any given expiration run
3829      * (though the next run will delete most of those unless they receive
3830      * additional data).
3831      *
3832      * This requires a second pass through the subfacets, in addition to the
3833      * pass made by update_stats(), because the former function never looks at
3834      * uninstallable subfacets.
3835      */
3836     enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
3837     enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3838     int buckets[N_BUCKETS] = { 0 };
3839     int total, subtotal, bucket;
3840     struct subfacet *subfacet;
3841     long long int now;
3842     int i;
3843
3844     total = hmap_count(&backer->subfacets);
3845     if (total <= flow_eviction_threshold) {
3846         return N_BUCKETS * BUCKET_WIDTH;
3847     }
3848
3849     /* Build histogram. */
3850     now = time_msec();
3851     HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3852         long long int idle = now - subfacet->used;
3853         int bucket = (idle <= 0 ? 0
3854                       : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3855                       : (unsigned int) idle / BUCKET_WIDTH);
3856         buckets[bucket]++;
3857     }
3858
3859     /* Find the first bucket whose flows should be expired. */
3860     subtotal = bucket = 0;
3861     do {
3862         subtotal += buckets[bucket++];
3863     } while (bucket < N_BUCKETS &&
3864              subtotal < MAX(flow_eviction_threshold, total / 100));
3865
3866     if (VLOG_IS_DBG_ENABLED()) {
3867         struct ds s;
3868
3869         ds_init(&s);
3870         ds_put_cstr(&s, "keep");
3871         for (i = 0; i < N_BUCKETS; i++) {
3872             if (i == bucket) {
3873                 ds_put_cstr(&s, ", drop");
3874             }
3875             if (buckets[i]) {
3876                 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3877             }
3878         }
3879         VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3880         ds_destroy(&s);
3881     }
3882
3883     return bucket * BUCKET_WIDTH;
3884 }
3885
3886 static void
3887 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3888 {
3889     /* Cutoff time for most flows. */
3890     long long int normal_cutoff = time_msec() - dp_max_idle;
3891
3892     /* We really want to keep flows for special protocols around, so use a more
3893      * conservative cutoff. */
3894     long long int special_cutoff = time_msec() - 10000;
3895
3896     struct subfacet *subfacet, *next_subfacet;
3897     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3898     int n_batch;
3899
3900     n_batch = 0;
3901     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3902                         &backer->subfacets) {
3903         long long int cutoff;
3904
3905         cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3906                                                 | SLOW_STP)
3907                   ? special_cutoff
3908                   : normal_cutoff);
3909         if (subfacet->used < cutoff) {
3910             if (subfacet->path != SF_NOT_INSTALLED) {
3911                 batch[n_batch++] = subfacet;
3912                 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3913                     subfacet_destroy_batch(backer, batch, n_batch);
3914                     n_batch = 0;
3915                 }
3916             } else {
3917                 subfacet_destroy(subfacet);
3918             }
3919         }
3920     }
3921
3922     if (n_batch > 0) {
3923         subfacet_destroy_batch(backer, batch, n_batch);
3924     }
3925 }
3926
3927 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3928  * then delete it entirely. */
3929 static void
3930 rule_expire(struct rule_dpif *rule)
3931 {
3932     uint16_t idle_timeout, hard_timeout;
3933     long long int now;
3934     uint8_t reason;
3935
3936     if (rule->up.pending) {
3937         /* We'll have to expire it later. */
3938         return;
3939     }
3940
3941     ovs_mutex_lock(&rule->up.timeout_mutex);
3942     hard_timeout = rule->up.hard_timeout;
3943     idle_timeout = rule->up.idle_timeout;
3944     ovs_mutex_unlock(&rule->up.timeout_mutex);
3945
3946     /* Has 'rule' expired? */
3947     now = time_msec();
3948     if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3949         reason = OFPRR_HARD_TIMEOUT;
3950     } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3951         reason = OFPRR_IDLE_TIMEOUT;
3952     } else {
3953         return;
3954     }
3955
3956     if (!ovs_rwlock_trywrlock(&rule->up.evict)) {
3957         COVERAGE_INC(ofproto_dpif_expired);
3958
3959         /* Get rid of the rule. */
3960         ofproto_rule_expire(&rule->up, reason);
3961     }
3962 }
3963 \f
3964 /* Facets. */
3965
3966 /* Creates and returns a new facet based on 'miss'.
3967  *
3968  * The caller must already have determined that no facet with an identical
3969  * 'miss->flow' exists in 'miss->ofproto'.
3970  *
3971  * 'rule' and 'xout' must have been created based on 'miss'.
3972  *
3973  * 'facet'' statistics are initialized based on 'stats'.
3974  *
3975  * The facet will initially have no subfacets.  The caller should create (at
3976  * least) one subfacet with subfacet_create(). */
3977 static struct facet *
3978 facet_create(const struct flow_miss *miss)
3979 {
3980     struct ofproto_dpif *ofproto = miss->ofproto;
3981     struct facet *facet;
3982     struct match match;
3983
3984     facet = xzalloc(sizeof *facet);
3985     facet->ofproto = miss->ofproto;
3986     facet->used = miss->stats.used;
3987     facet->flow = miss->flow;
3988     facet->learn_rl = time_msec() + 500;
3989
3990     list_init(&facet->subfacets);
3991     netflow_flow_init(&facet->nf_flow);
3992     netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3993
3994     xlate_out_copy(&facet->xout, &miss->xout);
3995
3996     match_init(&match, &facet->flow, &facet->xout.wc);
3997     cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3998     ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3999     classifier_insert(&ofproto->facets, &facet->cr);
4000     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4001
4002     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4003     return facet;
4004 }
4005
4006 static void
4007 facet_free(struct facet *facet)
4008 {
4009     if (facet) {
4010         xlate_out_uninit(&facet->xout);
4011         free(facet);
4012     }
4013 }
4014
4015 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4016  * 'packet', which arrived on 'in_port'. */
4017 static bool
4018 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4019                     const struct nlattr *odp_actions, size_t actions_len,
4020                     struct ofpbuf *packet)
4021 {
4022     struct odputil_keybuf keybuf;
4023     struct ofpbuf key;
4024     int error;
4025
4026     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4027     odp_flow_key_from_flow(&key, flow,
4028                            ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
4029
4030     error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4031                          odp_actions, actions_len, packet);
4032     return !error;
4033 }
4034
4035 /* Remove 'facet' from its ofproto and free up the associated memory:
4036  *
4037  *   - If 'facet' was installed in the datapath, uninstalls it and updates its
4038  *     rule's statistics, via subfacet_uninstall().
4039  *
4040  *   - Removes 'facet' from its rule and from ofproto->facets.
4041  */
4042 static void
4043 facet_remove(struct facet *facet)
4044 {
4045     struct subfacet *subfacet, *next_subfacet;
4046
4047     ovs_assert(!list_is_empty(&facet->subfacets));
4048
4049     /* First uninstall all of the subfacets to get final statistics. */
4050     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4051         subfacet_uninstall(subfacet);
4052     }
4053
4054     /* Flush the final stats to the rule.
4055      *
4056      * This might require us to have at least one subfacet around so that we
4057      * can use its actions for accounting in facet_account(), which is why we
4058      * have uninstalled but not yet destroyed the subfacets. */
4059     facet_flush_stats(facet);
4060
4061     /* Now we're really all done so destroy everything. */
4062     LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4063                         &facet->subfacets) {
4064         subfacet_destroy__(subfacet);
4065     }
4066     ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
4067     classifier_remove(&facet->ofproto->facets, &facet->cr);
4068     ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
4069     cls_rule_destroy(&facet->cr);
4070     facet_free(facet);
4071 }
4072
4073 /* Feed information from 'facet' back into the learning table to keep it in
4074  * sync with what is actually flowing through the datapath. */
4075 static void
4076 facet_learn(struct facet *facet)
4077 {
4078     long long int now = time_msec();
4079
4080     if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
4081         return;
4082     }
4083
4084     facet->learn_rl = now + 500;
4085
4086     if (!facet->xout.has_learn
4087         && !facet->xout.has_normal
4088         && (!facet->xout.has_fin_timeout
4089             || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4090         return;
4091     }
4092
4093     facet_push_stats(facet, true);
4094 }
4095
4096 static void
4097 facet_account(struct facet *facet)
4098 {
4099     const struct nlattr *a;
4100     unsigned int left;
4101     ovs_be16 vlan_tci;
4102     uint64_t n_bytes;
4103
4104     if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
4105         return;
4106     }
4107     n_bytes = facet->byte_count - facet->accounted_bytes;
4108
4109     /* This loop feeds byte counters to bond_account() for rebalancing to use
4110      * as a basis.  We also need to track the actual VLAN on which the packet
4111      * is going to be sent to ensure that it matches the one passed to
4112      * bond_choose_output_slave().  (Otherwise, we will account to the wrong
4113      * hash bucket.)
4114      *
4115      * We use the actions from an arbitrary subfacet because they should all
4116      * be equally valid for our purpose. */
4117     vlan_tci = facet->flow.vlan_tci;
4118     NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
4119                              facet->xout.odp_actions.size) {
4120         const struct ovs_action_push_vlan *vlan;
4121         struct ofport_dpif *port;
4122
4123         switch (nl_attr_type(a)) {
4124         case OVS_ACTION_ATTR_OUTPUT:
4125             port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
4126             if (port && port->bundle && port->bundle->bond) {
4127                 bond_account(port->bundle->bond, &facet->flow,
4128                              vlan_tci_to_vid(vlan_tci), n_bytes);
4129             }
4130             break;
4131
4132         case OVS_ACTION_ATTR_POP_VLAN:
4133             vlan_tci = htons(0);
4134             break;
4135
4136         case OVS_ACTION_ATTR_PUSH_VLAN:
4137             vlan = nl_attr_get(a);
4138             vlan_tci = vlan->vlan_tci;
4139             break;
4140         }
4141     }
4142 }
4143
4144 /* Returns true if the only action for 'facet' is to send to the controller.
4145  * (We don't report NetFlow expiration messages for such facets because they
4146  * are just part of the control logic for the network, not real traffic). */
4147 static bool
4148 facet_is_controller_flow(struct facet *facet)
4149 {
4150     if (facet) {
4151         struct ofproto_dpif *ofproto = facet->ofproto;
4152         const struct ofpact *ofpacts;
4153         struct rule_dpif *rule;
4154         size_t ofpacts_len;
4155         bool is_controller;
4156
4157         rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
4158         ofpacts_len = rule->up.ofpacts_len;
4159         ofpacts = rule->up.ofpacts;
4160         is_controller = ofpacts_len > 0
4161             && ofpacts->type == OFPACT_CONTROLLER
4162             && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
4163         rule_release(rule);
4164         return is_controller;
4165     }
4166     return false;
4167 }
4168
4169 /* Folds all of 'facet''s statistics into its rule.  Also updates the
4170  * accounting ofhook and emits a NetFlow expiration if appropriate.  All of
4171  * 'facet''s statistics in the datapath should have been zeroed and folded into
4172  * its packet and byte counts before this function is called. */
4173 static void
4174 facet_flush_stats(struct facet *facet)
4175 {
4176     struct ofproto_dpif *ofproto = facet->ofproto;
4177     struct subfacet *subfacet;
4178
4179     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4180         ovs_assert(!subfacet->dp_byte_count);
4181         ovs_assert(!subfacet->dp_packet_count);
4182     }
4183
4184     facet_push_stats(facet, false);
4185     if (facet->accounted_bytes < facet->byte_count) {
4186         facet_account(facet);
4187         facet->accounted_bytes = facet->byte_count;
4188     }
4189
4190     if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4191         struct ofexpired expired;
4192         expired.flow = facet->flow;
4193         expired.packet_count = facet->packet_count;
4194         expired.byte_count = facet->byte_count;
4195         expired.used = facet->used;
4196         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4197     }
4198
4199     /* Reset counters to prevent double counting if 'facet' ever gets
4200      * reinstalled. */
4201     facet_reset_counters(facet);
4202
4203     netflow_flow_clear(&facet->nf_flow);
4204     facet->tcp_flags = 0;
4205 }
4206
4207 /* Searches 'ofproto''s table of facets for one which would be responsible for
4208  * 'flow'.  Returns it if found, otherwise a null pointer.
4209  *
4210  * The returned facet might need revalidation; use facet_lookup_valid()
4211  * instead if that is important. */
4212 static struct facet *
4213 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4214 {
4215     struct cls_rule *cr;
4216
4217     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4218     cr = classifier_lookup(&ofproto->facets, flow, NULL);
4219     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4220     return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4221 }
4222
4223 /* Searches 'ofproto''s table of facets for one capable that covers
4224  * 'flow'.  Returns it if found, otherwise a null pointer.
4225  *
4226  * The returned facet is guaranteed to be valid. */
4227 static struct facet *
4228 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4229 {
4230     struct facet *facet;
4231
4232     facet = facet_find(ofproto, flow);
4233     if (facet
4234         && ofproto->backer->need_revalidate
4235         && !facet_revalidate(facet)) {
4236         return NULL;
4237     }
4238
4239     return facet;
4240 }
4241
4242 static bool
4243 facet_check_consistency(struct facet *facet)
4244 {
4245     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4246
4247     struct xlate_out xout;
4248     struct xlate_in xin;
4249
4250     struct rule_dpif *rule;
4251     bool ok;
4252
4253     /* Check the datapath actions for consistency. */
4254     rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4255     xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4256     xlate_actions(&xin, &xout);
4257     rule_release(rule);
4258
4259     ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4260         && facet->xout.slow == xout.slow;
4261     if (!ok && !VLOG_DROP_WARN(&rl)) {
4262         struct ds s = DS_EMPTY_INITIALIZER;
4263
4264         flow_format(&s, &facet->flow);
4265         ds_put_cstr(&s, ": inconsistency in facet");
4266
4267         if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4268             ds_put_cstr(&s, " (actions were: ");
4269             format_odp_actions(&s, facet->xout.odp_actions.data,
4270                                facet->xout.odp_actions.size);
4271             ds_put_cstr(&s, ") (correct actions: ");
4272             format_odp_actions(&s, xout.odp_actions.data,
4273                                xout.odp_actions.size);
4274             ds_put_char(&s, ')');
4275         }
4276
4277         if (facet->xout.slow != xout.slow) {
4278             ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4279         }
4280
4281         ds_destroy(&s);
4282     }
4283     xlate_out_uninit(&xout);
4284
4285     return ok;
4286 }
4287
4288 /* Re-searches the classifier for 'facet':
4289  *
4290  *   - If the rule found is different from 'facet''s current rule, moves
4291  *     'facet' to the new rule and recompiles its actions.
4292  *
4293  *   - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4294  *     where it is and recompiles its actions anyway.
4295  *
4296  *   - If any of 'facet''s subfacets correspond to a new flow according to
4297  *     xlate_receive(), 'facet' is removed.
4298  *
4299  *   Returns true if 'facet' is still valid.  False if 'facet' was removed. */
4300 static bool
4301 facet_revalidate(struct facet *facet)
4302 {
4303     struct ofproto_dpif *ofproto = facet->ofproto;
4304     struct rule_dpif *new_rule;
4305     struct subfacet *subfacet;
4306     struct flow_wildcards wc;
4307     struct xlate_out xout;
4308     struct xlate_in xin;
4309
4310     COVERAGE_INC(facet_revalidate);
4311
4312     /* Check that child subfacets still correspond to this facet.  Tunnel
4313      * configuration changes could cause a subfacet's OpenFlow in_port to
4314      * change. */
4315     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4316         struct ofproto_dpif *recv_ofproto;
4317         struct flow recv_flow;
4318         int error;
4319
4320         error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4321                               subfacet->key_len, &recv_flow, NULL,
4322                               &recv_ofproto, NULL);
4323         if (error
4324             || recv_ofproto != ofproto
4325             || facet != facet_find(ofproto, &recv_flow)) {
4326             facet_remove(facet);
4327             return false;
4328         }
4329     }
4330
4331     flow_wildcards_init_catchall(&wc);
4332     rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4333
4334     /* Calculate new datapath actions.
4335      *
4336      * We do not modify any 'facet' state yet, because we might need to, e.g.,
4337      * emit a NetFlow expiration and, if so, we need to have the old state
4338      * around to properly compose it. */
4339     xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4340     xlate_actions(&xin, &xout);
4341     flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4342
4343     /* A facet's slow path reason should only change under dramatic
4344      * circumstances.  Rather than try to update everything, it's simpler to
4345      * remove the facet and start over.
4346      *
4347      * More importantly, if a facet's wildcards change, it will be relatively
4348      * difficult to figure out if its subfacets still belong to it, and if not
4349      * which facet they may belong to.  Again, to avoid the complexity, we
4350      * simply give up instead. */
4351     if (facet->xout.slow != xout.slow
4352         || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4353         facet_remove(facet);
4354         xlate_out_uninit(&xout);
4355         rule_release(new_rule);
4356         return false;
4357     }
4358
4359     if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4360         LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4361             if (subfacet->path == SF_FAST_PATH) {
4362                 struct dpif_flow_stats stats;
4363
4364                 subfacet_install(subfacet, &xout.odp_actions, &stats);
4365                 subfacet_update_stats(subfacet, &stats);
4366             }
4367         }
4368
4369         facet_flush_stats(facet);
4370
4371         ofpbuf_clear(&facet->xout.odp_actions);
4372         ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4373                    xout.odp_actions.size);
4374     }
4375
4376     /* Update 'facet' now that we've taken care of all the old state. */
4377     facet->xout.slow = xout.slow;
4378     facet->xout.has_learn = xout.has_learn;
4379     facet->xout.has_normal = xout.has_normal;
4380     facet->xout.has_fin_timeout = xout.has_fin_timeout;
4381     facet->xout.nf_output_iface = xout.nf_output_iface;
4382     facet->xout.mirrors = xout.mirrors;
4383     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4384     facet->used = MAX(facet->used, new_rule->up.created);
4385
4386     xlate_out_uninit(&xout);
4387     rule_release(new_rule);
4388     return true;
4389 }
4390
4391 static void
4392 facet_reset_counters(struct facet *facet)
4393 {
4394     facet->packet_count = 0;
4395     facet->byte_count = 0;
4396     facet->prev_packet_count = 0;
4397     facet->prev_byte_count = 0;
4398     facet->accounted_bytes = 0;
4399 }
4400
4401 static void
4402 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4403                 struct dpif_flow_stats *stats, bool may_learn)
4404 {
4405     struct ofport_dpif *in_port;
4406     struct rule_dpif *rule;
4407     struct xlate_in xin;
4408
4409     in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4410     if (in_port && in_port->is_tunnel) {
4411         netdev_vport_inc_rx(in_port->up.netdev, stats);
4412     }
4413
4414     rule_dpif_lookup(ofproto, flow, NULL, &rule);
4415     rule_credit_stats(rule, stats);
4416     xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4417     xin.resubmit_stats = stats;
4418     xin.may_learn = may_learn;
4419     xlate_actions_for_side_effects(&xin);
4420     rule_release(rule);
4421 }
4422
4423 static void
4424 facet_push_stats(struct facet *facet, bool may_learn)
4425 {
4426     struct dpif_flow_stats stats;
4427
4428     ovs_assert(facet->packet_count >= facet->prev_packet_count);
4429     ovs_assert(facet->byte_count >= facet->prev_byte_count);
4430     ovs_assert(facet->used >= facet->prev_used);
4431
4432     stats.n_packets = facet->packet_count - facet->prev_packet_count;
4433     stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4434     stats.used = facet->used;
4435     stats.tcp_flags = facet->tcp_flags;
4436
4437     if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4438         facet->prev_packet_count = facet->packet_count;
4439         facet->prev_byte_count = facet->byte_count;
4440         facet->prev_used = facet->used;
4441
4442         netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4443                                  facet->used);
4444         netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4445         mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4446                             stats.n_packets, stats.n_bytes);
4447         flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4448     }
4449 }
4450
4451 static void
4452 push_all_stats__(bool run_fast)
4453 {
4454     static long long int rl = LLONG_MIN;
4455     struct ofproto_dpif *ofproto;
4456
4457     if (time_msec() < rl) {
4458         return;
4459     }
4460
4461     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4462         struct cls_cursor cursor;
4463         struct facet *facet;
4464
4465         ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4466         cls_cursor_init(&cursor, &ofproto->facets, NULL);
4467         CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4468             facet_push_stats(facet, false);
4469             if (run_fast) {
4470                 run_fast_rl();
4471             }
4472         }
4473         ovs_rwlock_unlock(&ofproto->facets.rwlock);
4474     }
4475
4476     rl = time_msec() + 100;
4477 }
4478
4479 static void
4480 push_all_stats(void)
4481 {
4482     push_all_stats__(true);
4483 }
4484
4485 void
4486 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4487 {
4488     ovs_mutex_lock(&rule->stats_mutex);
4489     rule->packet_count += stats->n_packets;
4490     rule->byte_count += stats->n_bytes;
4491     ofproto_rule_update_used(&rule->up, stats->used);
4492     ovs_mutex_unlock(&rule->stats_mutex);
4493 }
4494 \f
4495 /* Subfacets. */
4496
4497 static struct subfacet *
4498 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4499               size_t key_len, uint32_t key_hash)
4500 {
4501     struct subfacet *subfacet;
4502
4503     HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4504                              &backer->subfacets) {
4505         if (subfacet->key_len == key_len
4506             && !memcmp(key, subfacet->key, key_len)) {
4507             return subfacet;
4508         }
4509     }
4510
4511     return NULL;
4512 }
4513
4514 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4515  * 'key_fitness', 'key', and 'key_len' members in 'miss'.  Returns the
4516  * existing subfacet if there is one, otherwise creates and returns a
4517  * new subfacet. */
4518 static struct subfacet *
4519 subfacet_create(struct facet *facet, struct flow_miss *miss)
4520 {
4521     struct dpif_backer *backer = miss->ofproto->backer;
4522     enum odp_key_fitness key_fitness = miss->key_fitness;
4523     const struct nlattr *key = miss->key;
4524     size_t key_len = miss->key_len;
4525     uint32_t key_hash;
4526     struct subfacet *subfacet;
4527
4528     key_hash = odp_flow_key_hash(key, key_len);
4529
4530     if (list_is_empty(&facet->subfacets)) {
4531         subfacet = &facet->one_subfacet;
4532     } else {
4533         subfacet = subfacet_find(backer, key, key_len, key_hash);
4534         if (subfacet) {
4535             if (subfacet->facet == facet) {
4536                 return subfacet;
4537             }
4538
4539             /* This shouldn't happen. */
4540             VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4541             subfacet_destroy(subfacet);
4542         }
4543
4544         subfacet = xmalloc(sizeof *subfacet);
4545     }
4546
4547     hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4548     list_push_back(&facet->subfacets, &subfacet->list_node);
4549     subfacet->facet = facet;
4550     subfacet->key_fitness = key_fitness;
4551     subfacet->key = xmemdup(key, key_len);
4552     subfacet->key_len = key_len;
4553     subfacet->used = miss->stats.used;
4554     subfacet->created = subfacet->used;
4555     subfacet->dp_packet_count = 0;
4556     subfacet->dp_byte_count = 0;
4557     subfacet->path = SF_NOT_INSTALLED;
4558     subfacet->backer = backer;
4559
4560     backer->subfacet_add_count++;
4561     return subfacet;
4562 }
4563
4564 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4565  * its facet within 'ofproto', and frees it. */
4566 static void
4567 subfacet_destroy__(struct subfacet *subfacet)
4568 {
4569     struct facet *facet = subfacet->facet;
4570     struct ofproto_dpif *ofproto = facet->ofproto;
4571
4572     /* Update ofproto stats before uninstall the subfacet. */
4573     ofproto->backer->subfacet_del_count++;
4574
4575     subfacet_uninstall(subfacet);
4576     hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4577     list_remove(&subfacet->list_node);
4578     free(subfacet->key);
4579     if (subfacet != &facet->one_subfacet) {
4580         free(subfacet);
4581     }
4582 }
4583
4584 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4585  * last remaining subfacet in its facet destroys the facet too. */
4586 static void
4587 subfacet_destroy(struct subfacet *subfacet)
4588 {
4589     struct facet *facet = subfacet->facet;
4590
4591     if (list_is_singleton(&facet->subfacets)) {
4592         /* facet_remove() needs at least one subfacet (it will remove it). */
4593         facet_remove(facet);
4594     } else {
4595         subfacet_destroy__(subfacet);
4596     }
4597 }
4598
4599 static void
4600 subfacet_destroy_batch(struct dpif_backer *backer,
4601                        struct subfacet **subfacets, int n)
4602 {
4603     struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4604     struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4605     struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4606     int i;
4607
4608     for (i = 0; i < n; i++) {
4609         ops[i].type = DPIF_OP_FLOW_DEL;
4610         ops[i].u.flow_del.key = subfacets[i]->key;
4611         ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4612         ops[i].u.flow_del.stats = &stats[i];
4613         opsp[i] = &ops[i];
4614     }
4615
4616     dpif_operate(backer->dpif, opsp, n);
4617     for (i = 0; i < n; i++) {
4618         subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4619         subfacets[i]->path = SF_NOT_INSTALLED;
4620         subfacet_destroy(subfacets[i]);
4621         run_fast_rl();
4622     }
4623 }
4624
4625 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4626  * bytes of actions in 'actions'.  If 'stats' is non-null, statistics counters
4627  * in the datapath will be zeroed and 'stats' will be updated with traffic new
4628  * since 'subfacet' was last updated.
4629  *
4630  * Returns 0 if successful, otherwise a positive errno value. */
4631 static int
4632 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4633                  struct dpif_flow_stats *stats)
4634 {
4635     struct facet *facet = subfacet->facet;
4636     enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4637     const struct nlattr *actions = odp_actions->data;
4638     size_t actions_len = odp_actions->size;
4639     struct odputil_keybuf maskbuf;
4640     struct ofpbuf mask;
4641
4642     uint64_t slow_path_stub[128 / 8];
4643     enum dpif_flow_put_flags flags;
4644     int ret;
4645
4646     flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4647                                                : DPIF_FP_MODIFY;
4648     if (stats) {
4649         flags |= DPIF_FP_ZERO_STATS;
4650     }
4651
4652     if (path == SF_SLOW_PATH) {
4653         compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4654                           slow_path_stub, sizeof slow_path_stub,
4655                           &actions, &actions_len);
4656     }
4657
4658     ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4659     if (enable_megaflows) {
4660         odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4661                                &facet->flow, UINT32_MAX);
4662     }
4663
4664     ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4665                         subfacet->key_len,  mask.data, mask.size,
4666                         actions, actions_len, stats);
4667
4668     if (stats) {
4669         subfacet_reset_dp_stats(subfacet, stats);
4670     }
4671
4672     if (ret) {
4673         COVERAGE_INC(subfacet_install_fail);
4674     } else {
4675         subfacet->path = path;
4676     }
4677     return ret;
4678 }
4679
4680 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4681 static void
4682 subfacet_uninstall(struct subfacet *subfacet)
4683 {
4684     if (subfacet->path != SF_NOT_INSTALLED) {
4685         struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4686         struct dpif_flow_stats stats;
4687         int error;
4688
4689         error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4690                               subfacet->key_len, &stats);
4691         subfacet_reset_dp_stats(subfacet, &stats);
4692         if (!error) {
4693             subfacet_update_stats(subfacet, &stats);
4694         }
4695         subfacet->path = SF_NOT_INSTALLED;
4696     } else {
4697         ovs_assert(subfacet->dp_packet_count == 0);
4698         ovs_assert(subfacet->dp_byte_count == 0);
4699     }
4700 }
4701
4702 /* Resets 'subfacet''s datapath statistics counters.  This should be called
4703  * when 'subfacet''s statistics are cleared in the datapath.  If 'stats' is
4704  * non-null, it should contain the statistics returned by dpif when 'subfacet'
4705  * was reset in the datapath.  'stats' will be modified to include only
4706  * statistics new since 'subfacet' was last updated. */
4707 static void
4708 subfacet_reset_dp_stats(struct subfacet *subfacet,
4709                         struct dpif_flow_stats *stats)
4710 {
4711     if (stats
4712         && subfacet->dp_packet_count <= stats->n_packets
4713         && subfacet->dp_byte_count <= stats->n_bytes) {
4714         stats->n_packets -= subfacet->dp_packet_count;
4715         stats->n_bytes -= subfacet->dp_byte_count;
4716     }
4717
4718     subfacet->dp_packet_count = 0;
4719     subfacet->dp_byte_count = 0;
4720 }
4721
4722 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4723  *
4724  * Because of the meaning of a subfacet's counters, it only makes sense to do
4725  * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4726  * represents a packet that was sent by hand or if it represents statistics
4727  * that have been cleared out of the datapath. */
4728 static void
4729 subfacet_update_stats(struct subfacet *subfacet,
4730                       const struct dpif_flow_stats *stats)
4731 {
4732     if (stats->n_packets || stats->used > subfacet->used) {
4733         struct facet *facet = subfacet->facet;
4734
4735         subfacet->used = MAX(subfacet->used, stats->used);
4736         facet->used = MAX(facet->used, stats->used);
4737         facet->packet_count += stats->n_packets;
4738         facet->byte_count += stats->n_bytes;
4739         facet->tcp_flags |= stats->tcp_flags;
4740     }
4741 }
4742 \f
4743 /* Rules. */
4744
4745 /* Lookup 'flow' in 'ofproto''s classifier.  If 'wc' is non-null, sets
4746  * the fields that were relevant as part of the lookup. */
4747 void
4748 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4749                  struct flow_wildcards *wc, struct rule_dpif **rule)
4750 {
4751     struct ofport_dpif *port;
4752
4753     if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4754         return;
4755     }
4756     port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4757     if (!port) {
4758         VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4759                      flow->in_port.ofp_port);
4760     }
4761
4762     *rule = choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4763                              ofproto->no_packet_in_rule);
4764     ovs_rwlock_rdlock(&(*rule)->up.evict);
4765 }
4766
4767 bool
4768 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4769                           const struct flow *flow, struct flow_wildcards *wc,
4770                           uint8_t table_id, struct rule_dpif **rule)
4771     OVS_ACQ_RDLOCK((*rule)->up.evict)
4772 {
4773     struct cls_rule *cls_rule;
4774     struct classifier *cls;
4775     bool frag;
4776
4777     *rule = NULL;
4778     if (table_id >= N_TABLES) {
4779         return false;
4780     }
4781
4782     if (wc) {
4783         memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4784         wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4785     }
4786
4787     cls = &ofproto->up.tables[table_id].cls;
4788     ovs_rwlock_rdlock(&cls->rwlock);
4789     frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4790     if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4791         /* We must pretend that transport ports are unavailable. */
4792         struct flow ofpc_normal_flow = *flow;
4793         ofpc_normal_flow.tp_src = htons(0);
4794         ofpc_normal_flow.tp_dst = htons(0);
4795         cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4796     } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4797         cls_rule = &ofproto->drop_frags_rule->up.cr;
4798         if (wc) {
4799             flow_wildcards_init_exact(wc);
4800         }
4801     } else {
4802         cls_rule = classifier_lookup(cls, flow, wc);
4803     }
4804
4805     *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4806     if (*rule && ovs_rwlock_tryrdlock(&(*rule)->up.evict)) {
4807         /* The rule is in the process of being removed.  Best we can do is
4808          * pretend it isn't there. */
4809         *rule = NULL;
4810     }
4811     ovs_rwlock_unlock(&cls->rwlock);
4812
4813     return *rule != NULL;
4814 }
4815
4816 /* Given a port configuration (specified as zero if there's no port), chooses
4817  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4818  * flow table miss. */
4819 struct rule_dpif *
4820 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4821                  struct rule_dpif *no_packet_in_rule)
4822 {
4823     return config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4824 }
4825
4826 void
4827 rule_release(struct rule_dpif *rule)
4828 {
4829     if (rule) {
4830         ovs_rwlock_unlock(&rule->up.evict);
4831     }
4832 }
4833
4834 static void
4835 complete_operation(struct rule_dpif *rule)
4836 {
4837     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4838
4839     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4840     if (clogged) {
4841         struct dpif_completion *c = xmalloc(sizeof *c);
4842         c->op = rule->up.pending;
4843         list_push_back(&ofproto->completions, &c->list_node);
4844     } else {
4845         ofoperation_complete(rule->up.pending, 0);
4846     }
4847 }
4848
4849 static struct rule *
4850 rule_alloc(void)
4851 {
4852     struct rule_dpif *rule = xmalloc(sizeof *rule);
4853     return &rule->up;
4854 }
4855
4856 static void
4857 rule_dealloc(struct rule *rule_)
4858 {
4859     struct rule_dpif *rule = rule_dpif_cast(rule_);
4860     free(rule);
4861 }
4862
4863 static enum ofperr
4864 rule_construct(struct rule *rule_)
4865 {
4866     struct rule_dpif *rule = rule_dpif_cast(rule_);
4867     ovs_mutex_init(&rule->stats_mutex, PTHREAD_MUTEX_NORMAL);
4868     ovs_mutex_lock(&rule->stats_mutex);
4869     rule->packet_count = 0;
4870     rule->byte_count = 0;
4871     ovs_mutex_unlock(&rule->stats_mutex);
4872     complete_operation(rule);
4873     return 0;
4874 }
4875
4876 static void
4877 rule_destruct(struct rule *rule_)
4878 {
4879     struct rule_dpif *rule = rule_dpif_cast(rule_);
4880     complete_operation(rule);
4881     ovs_mutex_destroy(&rule->stats_mutex);
4882 }
4883
4884 static void
4885 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4886 {
4887     struct rule_dpif *rule = rule_dpif_cast(rule_);
4888
4889     /* push_all_stats() can handle flow misses which, when using the learn
4890      * action, can cause rules to be added and deleted.  This can corrupt our
4891      * caller's datastructures which assume that rule_get_stats() doesn't have
4892      * an impact on the flow table. To be safe, we disable miss handling. */
4893     push_all_stats__(false);
4894
4895     /* Start from historical data for 'rule' itself that are no longer tracked
4896      * in facets.  This counts, for example, facets that have expired. */
4897     ovs_mutex_lock(&rule->stats_mutex);
4898     *packets = rule->packet_count;
4899     *bytes = rule->byte_count;
4900     ovs_mutex_unlock(&rule->stats_mutex);
4901 }
4902
4903 static void
4904 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4905                   struct ofpbuf *packet)
4906 {
4907     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4908     struct dpif_flow_stats stats;
4909     struct xlate_out xout;
4910     struct xlate_in xin;
4911
4912     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4913     rule_credit_stats(rule, &stats);
4914
4915     xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4916     xin.resubmit_stats = &stats;
4917     xlate_actions(&xin, &xout);
4918
4919     execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4920                         xout.odp_actions.size, packet);
4921
4922     xlate_out_uninit(&xout);
4923 }
4924
4925 static enum ofperr
4926 rule_execute(struct rule *rule, const struct flow *flow,
4927              struct ofpbuf *packet)
4928 {
4929     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4930     ofpbuf_delete(packet);
4931     return 0;
4932 }
4933
4934 static void
4935 rule_modify_actions(struct rule *rule_)
4936 {
4937     struct rule_dpif *rule = rule_dpif_cast(rule_);
4938
4939     complete_operation(rule);
4940 }
4941 \f
4942 /* Sends 'packet' out 'ofport'.
4943  * May modify 'packet'.
4944  * Returns 0 if successful, otherwise a positive errno value. */
4945 static int
4946 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4947 {
4948     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4949     uint64_t odp_actions_stub[1024 / 8];
4950     struct ofpbuf key, odp_actions;
4951     struct dpif_flow_stats stats;
4952     struct odputil_keybuf keybuf;
4953     struct ofpact_output output;
4954     struct xlate_out xout;
4955     struct xlate_in xin;
4956     struct flow flow;
4957     union flow_in_port in_port_;
4958     int error;
4959
4960     ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4961     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4962
4963     /* Use OFPP_NONE as the in_port to avoid special packet processing. */
4964     in_port_.ofp_port = OFPP_NONE;
4965     flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
4966     odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto,
4967                                                              OFPP_LOCAL));
4968     dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
4969
4970     ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
4971     output.port = ofport->up.ofp_port;
4972     output.max_len = 0;
4973
4974     xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet);
4975     xin.ofpacts_len = sizeof output;
4976     xin.ofpacts = &output.ofpact;
4977     xin.resubmit_stats = &stats;
4978     xlate_actions(&xin, &xout);
4979
4980     error = dpif_execute(ofproto->backer->dpif,
4981                          key.data, key.size,
4982                          xout.odp_actions.data, xout.odp_actions.size,
4983                          packet);
4984     xlate_out_uninit(&xout);
4985
4986     if (error) {
4987         VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)",
4988                      ofproto->up.name, netdev_get_name(ofport->up.netdev),
4989                      ovs_strerror(error));
4990     }
4991
4992     ofproto->stats.tx_packets++;
4993     ofproto->stats.tx_bytes += packet->size;
4994     return error;
4995 }
4996
4997 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4998  * The action will state 'slow' as the reason that the action is in the slow
4999  * path.  (This is purely informational: it allows a human viewing "ovs-dpctl
5000  * dump-flows" output to see why a flow is in the slow path.)
5001  *
5002  * The 'stub_size' bytes in 'stub' will be used to store the action.
5003  * 'stub_size' must be large enough for the action.
5004  *
5005  * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5006  * respectively. */
5007 static void
5008 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5009                   enum slow_path_reason slow,
5010                   uint64_t *stub, size_t stub_size,
5011                   const struct nlattr **actionsp, size_t *actions_lenp)
5012 {
5013     union user_action_cookie cookie;
5014     struct ofpbuf buf;
5015
5016     cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5017     cookie.slow_path.unused = 0;
5018     cookie.slow_path.reason = slow;
5019
5020     ofpbuf_use_stack(&buf, stub, stub_size);
5021     if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5022         uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
5023                                          ODPP_NONE);
5024         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5025     } else {
5026         odp_port_t odp_port;
5027         uint32_t pid;
5028
5029         odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
5030         pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
5031         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5032     }
5033     *actionsp = buf.data;
5034     *actions_lenp = buf.size;
5035 }
5036 \f
5037 static bool
5038 set_frag_handling(struct ofproto *ofproto_,
5039                   enum ofp_config_flags frag_handling)
5040 {
5041     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5042     if (frag_handling != OFPC_FRAG_REASM) {
5043         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5044         return true;
5045     } else {
5046         return false;
5047     }
5048 }
5049
5050 static enum ofperr
5051 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
5052            const struct flow *flow,
5053            const struct ofpact *ofpacts, size_t ofpacts_len)
5054 {
5055     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5056     struct odputil_keybuf keybuf;
5057     struct dpif_flow_stats stats;
5058     struct xlate_out xout;
5059     struct xlate_in xin;
5060     struct ofpbuf key;
5061
5062
5063     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5064     odp_flow_key_from_flow(&key, flow,
5065                            ofp_port_to_odp_port(ofproto,
5066                                       flow->in_port.ofp_port));
5067
5068     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5069
5070     xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
5071     xin.resubmit_stats = &stats;
5072     xin.ofpacts_len = ofpacts_len;
5073     xin.ofpacts = ofpacts;
5074
5075     xlate_actions(&xin, &xout);
5076     dpif_execute(ofproto->backer->dpif, key.data, key.size,
5077                  xout.odp_actions.data, xout.odp_actions.size, packet);
5078     xlate_out_uninit(&xout);
5079
5080     return 0;
5081 }
5082 \f
5083 /* NetFlow. */
5084
5085 static int
5086 set_netflow(struct ofproto *ofproto_,
5087             const struct netflow_options *netflow_options)
5088 {
5089     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5090
5091     if (netflow_options) {
5092         if (!ofproto->netflow) {
5093             ofproto->netflow = netflow_create();
5094             ofproto->backer->need_revalidate = REV_RECONFIGURE;
5095         }
5096         return netflow_set_options(ofproto->netflow, netflow_options);
5097     } else if (ofproto->netflow) {
5098         ofproto->backer->need_revalidate = REV_RECONFIGURE;
5099         netflow_destroy(ofproto->netflow);
5100         ofproto->netflow = NULL;
5101     }
5102
5103     return 0;
5104 }
5105
5106 static void
5107 get_netflow_ids(const struct ofproto *ofproto_,
5108                 uint8_t *engine_type, uint8_t *engine_id)
5109 {
5110     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
5111
5112     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
5113 }
5114
5115 static void
5116 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
5117 {
5118     if (!facet_is_controller_flow(facet) &&
5119         netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
5120         struct subfacet *subfacet;
5121         struct ofexpired expired;
5122
5123         LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5124             if (subfacet->path == SF_FAST_PATH) {
5125                 struct dpif_flow_stats stats;
5126
5127                 subfacet_install(subfacet, &facet->xout.odp_actions,
5128                                  &stats);
5129                 subfacet_update_stats(subfacet, &stats);
5130             }
5131         }
5132
5133         expired.flow = facet->flow;
5134         expired.packet_count = facet->packet_count;
5135         expired.byte_count = facet->byte_count;
5136         expired.used = facet->used;
5137         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
5138     }
5139 }
5140
5141 static void
5142 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
5143 {
5144     struct cls_cursor cursor;
5145     struct facet *facet;
5146
5147     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5148     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5149     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5150         send_active_timeout(ofproto, facet);
5151     }
5152     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5153 }
5154 \f
5155 static struct ofproto_dpif *
5156 ofproto_dpif_lookup(const char *name)
5157 {
5158     struct ofproto_dpif *ofproto;
5159
5160     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5161                              hash_string(name, 0), &all_ofproto_dpifs) {
5162         if (!strcmp(ofproto->up.name, name)) {
5163             return ofproto;
5164         }
5165     }
5166     return NULL;
5167 }
5168
5169 static void
5170 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5171                           const char *argv[], void *aux OVS_UNUSED)
5172 {
5173     struct ofproto_dpif *ofproto;
5174
5175     if (argc > 1) {
5176         ofproto = ofproto_dpif_lookup(argv[1]);
5177         if (!ofproto) {
5178             unixctl_command_reply_error(conn, "no such bridge");
5179             return;
5180         }
5181         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5182         mac_learning_flush(ofproto->ml);
5183         ovs_rwlock_unlock(&ofproto->ml->rwlock);
5184     } else {
5185         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5186             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5187             mac_learning_flush(ofproto->ml);
5188             ovs_rwlock_unlock(&ofproto->ml->rwlock);
5189         }
5190     }
5191
5192     unixctl_command_reply(conn, "table successfully flushed");
5193 }
5194
5195 static struct ofport_dpif *
5196 ofbundle_get_a_port(const struct ofbundle *bundle)
5197 {
5198     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5199                         bundle_node);
5200 }
5201
5202 static void
5203 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5204                          const char *argv[], void *aux OVS_UNUSED)
5205 {
5206     struct ds ds = DS_EMPTY_INITIALIZER;
5207     const struct ofproto_dpif *ofproto;
5208     const struct mac_entry *e;
5209
5210     ofproto = ofproto_dpif_lookup(argv[1]);
5211     if (!ofproto) {
5212         unixctl_command_reply_error(conn, "no such bridge");
5213         return;
5214     }
5215
5216     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
5217     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5218     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5219         struct ofbundle *bundle = e->port.p;
5220         char name[OFP_MAX_PORT_NAME_LEN];
5221
5222         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5223                                name, sizeof name);
5224         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
5225                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
5226                       mac_entry_age(ofproto->ml, e));
5227     }
5228     ovs_rwlock_unlock(&ofproto->ml->rwlock);
5229     unixctl_command_reply(conn, ds_cstr(&ds));
5230     ds_destroy(&ds);
5231 }
5232
5233 struct trace_ctx {
5234     struct xlate_out xout;
5235     struct xlate_in xin;
5236     struct flow flow;
5237     struct ds *result;
5238 };
5239
5240 static void
5241 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5242 {
5243     ds_put_char_multiple(result, '\t', level);
5244     if (!rule) {
5245         ds_put_cstr(result, "No match\n");
5246         return;
5247     }
5248
5249     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5250                   rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie));
5251     cls_rule_format(&rule->up.cr, result);
5252     ds_put_char(result, '\n');
5253
5254     ds_put_char_multiple(result, '\t', level);
5255     ds_put_cstr(result, "OpenFlow ");
5256     ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
5257     ds_put_char(result, '\n');
5258 }
5259
5260 static void
5261 trace_format_flow(struct ds *result, int level, const char *title,
5262                   struct trace_ctx *trace)
5263 {
5264     ds_put_char_multiple(result, '\t', level);
5265     ds_put_format(result, "%s: ", title);
5266     if (flow_equal(&trace->xin.flow, &trace->flow)) {
5267         ds_put_cstr(result, "unchanged");
5268     } else {
5269         flow_format(result, &trace->xin.flow);
5270         trace->flow = trace->xin.flow;
5271     }
5272     ds_put_char(result, '\n');
5273 }
5274
5275 static void
5276 trace_format_regs(struct ds *result, int level, const char *title,
5277                   struct trace_ctx *trace)
5278 {
5279     size_t i;
5280
5281     ds_put_char_multiple(result, '\t', level);
5282     ds_put_format(result, "%s:", title);
5283     for (i = 0; i < FLOW_N_REGS; i++) {
5284         ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5285     }
5286     ds_put_char(result, '\n');
5287 }
5288
5289 static void
5290 trace_format_odp(struct ds *result, int level, const char *title,
5291                  struct trace_ctx *trace)
5292 {
5293     struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5294
5295     ds_put_char_multiple(result, '\t', level);
5296     ds_put_format(result, "%s: ", title);
5297     format_odp_actions(result, odp_actions->data, odp_actions->size);
5298     ds_put_char(result, '\n');
5299 }
5300
5301 static void
5302 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, 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(result, '\n');
5308     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5309     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5310     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
5311     trace_format_rule(result, recurse + 1, rule);
5312 }
5313
5314 static void
5315 trace_report(struct xlate_in *xin, const char *s, int recurse)
5316 {
5317     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5318     struct ds *result = trace->result;
5319
5320     ds_put_char_multiple(result, '\t', recurse);
5321     ds_put_cstr(result, s);
5322     ds_put_char(result, '\n');
5323 }
5324
5325 static void
5326 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5327                       void *aux OVS_UNUSED)
5328 {
5329     const struct dpif_backer *backer;
5330     struct ofproto_dpif *ofproto;
5331     struct ofpbuf odp_key, odp_mask;
5332     struct ofpbuf *packet;
5333     struct ds result;
5334     struct flow flow;
5335     char *s;
5336
5337     packet = NULL;
5338     backer = NULL;
5339     ds_init(&result);
5340     ofpbuf_init(&odp_key, 0);
5341     ofpbuf_init(&odp_mask, 0);
5342
5343     /* Handle "-generate" or a hex string as the last argument. */
5344     if (!strcmp(argv[argc - 1], "-generate")) {
5345         packet = ofpbuf_new(0);
5346         argc--;
5347     } else {
5348         const char *error = eth_from_hex(argv[argc - 1], &packet);
5349         if (!error) {
5350             argc--;
5351         } else if (argc == 4) {
5352             /* The 3-argument form must end in "-generate' or a hex string. */
5353             unixctl_command_reply_error(conn, error);
5354             goto exit;
5355         }
5356     }
5357
5358     /* Parse the flow and determine whether a datapath or
5359      * bridge is specified. If function odp_flow_key_from_string()
5360      * returns 0, the flow is a odp_flow. If function
5361      * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5362     if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) {
5363         /* If the odp_flow is the second argument,
5364          * the datapath name is the first argument. */
5365         if (argc == 3) {
5366             const char *dp_type;
5367             if (!strncmp(argv[1], "ovs-", 4)) {
5368                 dp_type = argv[1] + 4;
5369             } else {
5370                 dp_type = argv[1];
5371             }
5372             backer = shash_find_data(&all_dpif_backers, dp_type);
5373             if (!backer) {
5374                 unixctl_command_reply_error(conn, "Cannot find datapath "
5375                                "of this name");
5376                 goto exit;
5377             }
5378         } else {
5379             /* No datapath name specified, so there should be only one
5380              * datapath. */
5381             struct shash_node *node;
5382             if (shash_count(&all_dpif_backers) != 1) {
5383                 unixctl_command_reply_error(conn, "Must specify datapath "
5384                          "name, there is more than one type of datapath");
5385                 goto exit;
5386             }
5387             node = shash_first(&all_dpif_backers);
5388             backer = node->data;
5389         }
5390
5391         if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5392                           NULL, &ofproto, NULL)) {
5393             unixctl_command_reply_error(conn, "Invalid datapath flow");
5394             goto exit;
5395         }
5396         ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5397     } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5398         if (argc != 3) {
5399             unixctl_command_reply_error(conn, "Must specify bridge name");
5400             goto exit;
5401         }
5402
5403         ofproto = ofproto_dpif_lookup(argv[1]);
5404         if (!ofproto) {
5405             unixctl_command_reply_error(conn, "Unknown bridge name");
5406             goto exit;
5407         }
5408     } else {
5409         unixctl_command_reply_error(conn, "Bad flow syntax");
5410         goto exit;
5411     }
5412
5413     /* Generate a packet, if requested. */
5414     if (packet) {
5415         if (!packet->size) {
5416             flow_compose(packet, &flow);
5417         } else {
5418             union flow_in_port in_port_;
5419
5420             in_port_ = flow.in_port;
5421             ds_put_cstr(&result, "Packet: ");
5422             s = ofp_packet_to_string(packet->data, packet->size);
5423             ds_put_cstr(&result, s);
5424             free(s);
5425
5426             /* Use the metadata from the flow and the packet argument
5427              * to reconstruct the flow. */
5428             flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5429                          &in_port_, &flow);
5430         }
5431     }
5432
5433     ofproto_trace(ofproto, &flow, packet, &result);
5434     unixctl_command_reply(conn, ds_cstr(&result));
5435
5436 exit:
5437     ds_destroy(&result);
5438     ofpbuf_delete(packet);
5439     ofpbuf_uninit(&odp_key);
5440     ofpbuf_uninit(&odp_mask);
5441 }
5442
5443 static void
5444 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5445               const struct ofpbuf *packet, struct ds *ds)
5446 {
5447     struct rule_dpif *rule;
5448     struct flow_wildcards wc;
5449
5450     ds_put_cstr(ds, "Flow: ");
5451     flow_format(ds, flow);
5452     ds_put_char(ds, '\n');
5453
5454     flow_wildcards_init_catchall(&wc);
5455     rule_dpif_lookup(ofproto, flow, &wc, &rule);
5456
5457     trace_format_rule(ds, 0, rule);
5458     if (rule == ofproto->miss_rule) {
5459         ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5460     } else if (rule == ofproto->no_packet_in_rule) {
5461         ds_put_cstr(ds, "\nNo match, packets dropped because "
5462                     "OFPPC_NO_PACKET_IN is set on in_port.\n");
5463     } else if (rule == ofproto->drop_frags_rule) {
5464         ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5465                     "and the fragment handling mode is \"drop\".\n");
5466     }
5467
5468     if (rule) {
5469         uint64_t odp_actions_stub[1024 / 8];
5470         struct ofpbuf odp_actions;
5471         struct trace_ctx trace;
5472         struct match match;
5473         uint8_t tcp_flags;
5474
5475         tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5476         trace.result = ds;
5477         trace.flow = *flow;
5478         ofpbuf_use_stub(&odp_actions,
5479                         odp_actions_stub, sizeof odp_actions_stub);
5480         xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5481         trace.xin.resubmit_hook = trace_resubmit;
5482         trace.xin.report_hook = trace_report;
5483
5484         xlate_actions(&trace.xin, &trace.xout);
5485         flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5486
5487         ds_put_char(ds, '\n');
5488         trace_format_flow(ds, 0, "Final flow", &trace);
5489
5490         match_init(&match, flow, &trace.xout.wc);
5491         ds_put_cstr(ds, "Relevant fields: ");
5492         match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5493         ds_put_char(ds, '\n');
5494
5495         ds_put_cstr(ds, "Datapath actions: ");
5496         format_odp_actions(ds, trace.xout.odp_actions.data,
5497                            trace.xout.odp_actions.size);
5498
5499         if (trace.xout.slow) {
5500             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5501                         "slow path because it:");
5502             switch (trace.xout.slow) {
5503             case SLOW_CFM:
5504                 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5505                 break;
5506             case SLOW_LACP:
5507                 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5508                 break;
5509             case SLOW_STP:
5510                 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5511                 break;
5512             case SLOW_BFD:
5513                 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5514                 break;
5515             case SLOW_CONTROLLER:
5516                 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5517                             "to the OpenFlow controller.");
5518                 break;
5519             case __SLOW_MAX:
5520                 NOT_REACHED();
5521             }
5522         }
5523
5524         xlate_out_uninit(&trace.xout);
5525     }
5526
5527     rule_release(rule);
5528 }
5529
5530 static void
5531 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5532                   const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5533 {
5534     clogged = true;
5535     unixctl_command_reply(conn, NULL);
5536 }
5537
5538 static void
5539 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
5540                     const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5541 {
5542     clogged = false;
5543     unixctl_command_reply(conn, NULL);
5544 }
5545
5546 /* Runs a self-check of flow translations in 'ofproto'.  Appends a message to
5547  * 'reply' describing the results. */
5548 static void
5549 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5550 {
5551     struct cls_cursor cursor;
5552     struct facet *facet;
5553     int errors;
5554
5555     errors = 0;
5556     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5557     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5558     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5559         if (!facet_check_consistency(facet)) {
5560             errors++;
5561         }
5562     }
5563     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5564     if (errors) {
5565         ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5566     }
5567
5568     if (errors) {
5569         ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5570                       ofproto->up.name, errors);
5571     } else {
5572         ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5573     }
5574 }
5575
5576 static void
5577 ofproto_dpif_self_check(struct unixctl_conn *conn,
5578                         int argc, const char *argv[], void *aux OVS_UNUSED)
5579 {
5580     struct ds reply = DS_EMPTY_INITIALIZER;
5581     struct ofproto_dpif *ofproto;
5582
5583     if (argc > 1) {
5584         ofproto = ofproto_dpif_lookup(argv[1]);
5585         if (!ofproto) {
5586             unixctl_command_reply_error(conn, "Unknown ofproto (use "
5587                                         "ofproto/list for help)");
5588             return;
5589         }
5590         ofproto_dpif_self_check__(ofproto, &reply);
5591     } else {
5592         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5593             ofproto_dpif_self_check__(ofproto, &reply);
5594         }
5595     }
5596
5597     unixctl_command_reply(conn, ds_cstr(&reply));
5598     ds_destroy(&reply);
5599 }
5600
5601 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
5602  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
5603  * to destroy 'ofproto_shash' and free the returned value. */
5604 static const struct shash_node **
5605 get_ofprotos(struct shash *ofproto_shash)
5606 {
5607     const struct ofproto_dpif *ofproto;
5608
5609     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5610         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5611         shash_add_nocopy(ofproto_shash, name, ofproto);
5612     }
5613
5614     return shash_sort(ofproto_shash);
5615 }
5616
5617 static void
5618 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5619                               const char *argv[] OVS_UNUSED,
5620                               void *aux OVS_UNUSED)
5621 {
5622     struct ds ds = DS_EMPTY_INITIALIZER;
5623     struct shash ofproto_shash;
5624     const struct shash_node **sorted_ofprotos;
5625     int i;
5626
5627     shash_init(&ofproto_shash);
5628     sorted_ofprotos = get_ofprotos(&ofproto_shash);
5629     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5630         const struct shash_node *node = sorted_ofprotos[i];
5631         ds_put_format(&ds, "%s\n", node->name);
5632     }
5633
5634     shash_destroy(&ofproto_shash);
5635     free(sorted_ofprotos);
5636
5637     unixctl_command_reply(conn, ds_cstr(&ds));
5638     ds_destroy(&ds);
5639 }
5640
5641 static void
5642 show_dp_rates(struct ds *ds, const char *heading,
5643               const struct avg_subfacet_rates *rates)
5644 {
5645     ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
5646                   heading, rates->add_rate, rates->del_rate);
5647 }
5648
5649 static void
5650 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5651 {
5652     const struct shash_node **ofprotos;
5653     struct ofproto_dpif *ofproto;
5654     struct shash ofproto_shash;
5655     uint64_t n_hit, n_missed;
5656     long long int minutes;
5657     size_t i;
5658
5659     n_hit = n_missed = 0;
5660     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5661         if (ofproto->backer == backer) {
5662             n_missed += ofproto->n_missed;
5663             n_hit += ofproto->n_hit;
5664         }
5665     }
5666
5667     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5668                   dpif_name(backer->dpif), n_hit, n_missed);
5669     ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5670                   " life span: %lldms\n", hmap_count(&backer->subfacets),
5671                   backer->avg_n_subfacet, backer->max_n_subfacet,
5672                   backer->avg_subfacet_life);
5673
5674     minutes = (time_msec() - backer->created) / (1000 * 60);
5675     if (minutes >= 60) {
5676         show_dp_rates(ds, "\thourly avg:", &backer->hourly);
5677     }
5678     if (minutes >= 60 * 24) {
5679         show_dp_rates(ds, "\tdaily avg:",  &backer->daily);
5680     }
5681     show_dp_rates(ds, "\toverall avg:",  &backer->lifetime);
5682
5683     shash_init(&ofproto_shash);
5684     ofprotos = get_ofprotos(&ofproto_shash);
5685     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5686         struct ofproto_dpif *ofproto = ofprotos[i]->data;
5687         const struct shash_node **ports;
5688         size_t j;
5689
5690         if (ofproto->backer != backer) {
5691             continue;
5692         }
5693
5694         ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5695                       ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5696
5697         ports = shash_sort(&ofproto->up.port_by_name);
5698         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5699             const struct shash_node *node = ports[j];
5700             struct ofport *ofport = node->data;
5701             struct smap config;
5702             odp_port_t odp_port;
5703
5704             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5705                           ofport->ofp_port);
5706
5707             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5708             if (odp_port != ODPP_NONE) {
5709                 ds_put_format(ds, "%"PRIu32":", odp_port);
5710             } else {
5711                 ds_put_cstr(ds, "none:");
5712             }
5713
5714             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5715
5716             smap_init(&config);
5717             if (!netdev_get_config(ofport->netdev, &config)) {
5718                 const struct smap_node **nodes;
5719                 size_t i;
5720
5721                 nodes = smap_sort(&config);
5722                 for (i = 0; i < smap_count(&config); i++) {
5723                     const struct smap_node *node = nodes[i];
5724                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5725                                   node->key, node->value);
5726                 }
5727                 free(nodes);
5728             }
5729             smap_destroy(&config);
5730
5731             ds_put_char(ds, ')');
5732             ds_put_char(ds, '\n');
5733         }
5734         free(ports);
5735     }
5736     shash_destroy(&ofproto_shash);
5737     free(ofprotos);
5738 }
5739
5740 static void
5741 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5742                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5743 {
5744     struct ds ds = DS_EMPTY_INITIALIZER;
5745     const struct shash_node **backers;
5746     int i;
5747
5748     backers = shash_sort(&all_dpif_backers);
5749     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5750         dpif_show_backer(backers[i]->data, &ds);
5751     }
5752     free(backers);
5753
5754     unixctl_command_reply(conn, ds_cstr(&ds));
5755     ds_destroy(&ds);
5756 }
5757
5758 /* Dump the megaflow (facet) cache.  This is useful to check the
5759  * correctness of flow wildcarding, since the same mechanism is used for
5760  * both xlate caching and kernel wildcarding.
5761  *
5762  * It's important to note that in the output the flow description uses
5763  * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5764  *
5765  * This command is only needed for advanced debugging, so it's not
5766  * documented in the man page. */
5767 static void
5768 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5769                                     int argc OVS_UNUSED, const char *argv[],
5770                                     void *aux OVS_UNUSED)
5771 {
5772     struct ds ds = DS_EMPTY_INITIALIZER;
5773     const struct ofproto_dpif *ofproto;
5774     long long int now = time_msec();
5775     struct cls_cursor cursor;
5776     struct facet *facet;
5777
5778     ofproto = ofproto_dpif_lookup(argv[1]);
5779     if (!ofproto) {
5780         unixctl_command_reply_error(conn, "no such bridge");
5781         return;
5782     }
5783
5784     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5785     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5786     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5787         cls_rule_format(&facet->cr, &ds);
5788         ds_put_cstr(&ds, ", ");
5789         ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5790         ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5791         ds_put_cstr(&ds, "Datapath actions: ");
5792         if (facet->xout.slow) {
5793             uint64_t slow_path_stub[128 / 8];
5794             const struct nlattr *actions;
5795             size_t actions_len;
5796
5797             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5798                               slow_path_stub, sizeof slow_path_stub,
5799                               &actions, &actions_len);
5800             format_odp_actions(&ds, actions, actions_len);
5801         } else {
5802             format_odp_actions(&ds, facet->xout.odp_actions.data,
5803                                facet->xout.odp_actions.size);
5804         }
5805         ds_put_cstr(&ds, "\n");
5806     }
5807     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5808
5809     ds_chomp(&ds, '\n');
5810     unixctl_command_reply(conn, ds_cstr(&ds));
5811     ds_destroy(&ds);
5812 }
5813
5814 /* Disable using the megaflows.
5815  *
5816  * This command is only needed for advanced debugging, so it's not
5817  * documented in the man page. */
5818 static void
5819 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5820                                        int argc OVS_UNUSED,
5821                                        const char *argv[] OVS_UNUSED,
5822                                        void *aux OVS_UNUSED)
5823 {
5824     struct ofproto_dpif *ofproto;
5825
5826     enable_megaflows = false;
5827
5828     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5829         flush(&ofproto->up);
5830     }
5831
5832     unixctl_command_reply(conn, "megaflows disabled");
5833 }
5834
5835 /* Re-enable using megaflows.
5836  *
5837  * This command is only needed for advanced debugging, so it's not
5838  * documented in the man page. */
5839 static void
5840 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5841                                       int argc OVS_UNUSED,
5842                                       const char *argv[] OVS_UNUSED,
5843                                       void *aux OVS_UNUSED)
5844 {
5845     struct ofproto_dpif *ofproto;
5846
5847     enable_megaflows = true;
5848
5849     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5850         flush(&ofproto->up);
5851     }
5852
5853     unixctl_command_reply(conn, "megaflows enabled");
5854 }
5855
5856 static void
5857 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5858                                 int argc OVS_UNUSED, const char *argv[],
5859                                 void *aux OVS_UNUSED)
5860 {
5861     struct ds ds = DS_EMPTY_INITIALIZER;
5862     const struct ofproto_dpif *ofproto;
5863     struct subfacet *subfacet;
5864
5865     ofproto = ofproto_dpif_lookup(argv[1]);
5866     if (!ofproto) {
5867         unixctl_command_reply_error(conn, "no such bridge");
5868         return;
5869     }
5870
5871     update_stats(ofproto->backer);
5872
5873     HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5874         struct facet *facet = subfacet->facet;
5875         struct odputil_keybuf maskbuf;
5876         struct ofpbuf mask;
5877
5878         if (facet->ofproto != ofproto) {
5879             continue;
5880         }
5881
5882         ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5883         if (enable_megaflows) {
5884             odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5885                                    &facet->flow, UINT32_MAX);
5886         }
5887
5888         odp_flow_format(subfacet->key, subfacet->key_len,
5889                         mask.data, mask.size, &ds, false);
5890
5891         ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5892                       subfacet->dp_packet_count, subfacet->dp_byte_count);
5893         if (subfacet->used) {
5894             ds_put_format(&ds, "%.3fs",
5895                           (time_msec() - subfacet->used) / 1000.0);
5896         } else {
5897             ds_put_format(&ds, "never");
5898         }
5899         if (subfacet->facet->tcp_flags) {
5900             ds_put_cstr(&ds, ", flags:");
5901             packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5902         }
5903
5904         ds_put_cstr(&ds, ", actions:");
5905         if (facet->xout.slow) {
5906             uint64_t slow_path_stub[128 / 8];
5907             const struct nlattr *actions;
5908             size_t actions_len;
5909
5910             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5911                               slow_path_stub, sizeof slow_path_stub,
5912                               &actions, &actions_len);
5913             format_odp_actions(&ds, actions, actions_len);
5914         } else {
5915             format_odp_actions(&ds, facet->xout.odp_actions.data,
5916                                facet->xout.odp_actions.size);
5917         }
5918         ds_put_char(&ds, '\n');
5919     }
5920
5921     unixctl_command_reply(conn, ds_cstr(&ds));
5922     ds_destroy(&ds);
5923 }
5924
5925 static void
5926 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5927                                int argc OVS_UNUSED, const char *argv[],
5928                                void *aux OVS_UNUSED)
5929 {
5930     struct ds ds = DS_EMPTY_INITIALIZER;
5931     struct ofproto_dpif *ofproto;
5932
5933     ofproto = ofproto_dpif_lookup(argv[1]);
5934     if (!ofproto) {
5935         unixctl_command_reply_error(conn, "no such bridge");
5936         return;
5937     }
5938
5939     flush(&ofproto->up);
5940
5941     unixctl_command_reply(conn, ds_cstr(&ds));
5942     ds_destroy(&ds);
5943 }
5944
5945 static void
5946 ofproto_dpif_unixctl_init(void)
5947 {
5948     static bool registered;
5949     if (registered) {
5950         return;
5951     }
5952     registered = true;
5953
5954     unixctl_command_register(
5955         "ofproto/trace",
5956         "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
5957         1, 3, ofproto_unixctl_trace, NULL);
5958     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5959                              ofproto_unixctl_fdb_flush, NULL);
5960     unixctl_command_register("fdb/show", "bridge", 1, 1,
5961                              ofproto_unixctl_fdb_show, NULL);
5962     unixctl_command_register("ofproto/clog", "", 0, 0,
5963                              ofproto_dpif_clog, NULL);
5964     unixctl_command_register("ofproto/unclog", "", 0, 0,
5965                              ofproto_dpif_unclog, NULL);
5966     unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
5967                              ofproto_dpif_self_check, NULL);
5968     unixctl_command_register("dpif/dump-dps", "", 0, 0,
5969                              ofproto_unixctl_dpif_dump_dps, NULL);
5970     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5971                              NULL);
5972     unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
5973                              ofproto_unixctl_dpif_dump_flows, NULL);
5974     unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
5975                              ofproto_unixctl_dpif_del_flows, NULL);
5976     unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
5977                              ofproto_unixctl_dpif_dump_megaflows, NULL);
5978     unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
5979                              ofproto_unixctl_dpif_disable_megaflows, NULL);
5980     unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
5981                              ofproto_unixctl_dpif_enable_megaflows, NULL);
5982 }
5983 \f
5984 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5985  *
5986  * This is deprecated.  It is only for compatibility with broken device drivers
5987  * in old versions of Linux that do not properly support VLANs when VLAN
5988  * devices are not used.  When broken device drivers are no longer in
5989  * widespread use, we will delete these interfaces. */
5990
5991 static int
5992 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5993 {
5994     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5995     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5996
5997     if (realdev_ofp_port == ofport->realdev_ofp_port
5998         && vid == ofport->vlandev_vid) {
5999         return 0;
6000     }
6001
6002     ofproto->backer->need_revalidate = REV_RECONFIGURE;
6003
6004     if (ofport->realdev_ofp_port) {
6005         vsp_remove(ofport);
6006     }
6007     if (realdev_ofp_port && ofport->bundle) {
6008         /* vlandevs are enslaved to their realdevs, so they are not allowed to
6009          * themselves be part of a bundle. */
6010         bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
6011     }
6012
6013     ofport->realdev_ofp_port = realdev_ofp_port;
6014     ofport->vlandev_vid = vid;
6015
6016     if (realdev_ofp_port) {
6017         vsp_add(ofport, realdev_ofp_port, vid);
6018     }
6019
6020     return 0;
6021 }
6022
6023 static uint32_t
6024 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
6025 {
6026     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
6027 }
6028
6029 bool
6030 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
6031     OVS_EXCLUDED(ofproto->vsp_mutex)
6032 {
6033     bool ret;
6034
6035     ovs_mutex_lock(&ofproto->vsp_mutex);
6036     ret = !hmap_is_empty(&ofproto->realdev_vid_map);
6037     ovs_mutex_unlock(&ofproto->vsp_mutex);
6038     return ret;
6039 }
6040
6041 static ofp_port_t
6042 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
6043                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6044     OVS_REQUIRES(ofproto->vsp_mutex)
6045 {
6046     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
6047         int vid = vlan_tci_to_vid(vlan_tci);
6048         const struct vlan_splinter *vsp;
6049
6050         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
6051                                  hash_realdev_vid(realdev_ofp_port, vid),
6052                                  &ofproto->realdev_vid_map) {
6053             if (vsp->realdev_ofp_port == realdev_ofp_port
6054                 && vsp->vid == vid) {
6055                 return vsp->vlandev_ofp_port;
6056             }
6057         }
6058     }
6059     return realdev_ofp_port;
6060 }
6061
6062 /* Returns the OFP port number of the Linux VLAN device that corresponds to
6063  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
6064  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
6065  * 'vlan_tci' 9, it would return the port number of eth0.9.
6066  *
6067  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
6068  * function just returns its 'realdev_ofp_port' argument. */
6069 ofp_port_t
6070 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
6071                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
6072     OVS_EXCLUDED(ofproto->vsp_mutex)
6073 {
6074     ofp_port_t ret;
6075
6076     ovs_mutex_lock(&ofproto->vsp_mutex);
6077     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
6078     ovs_mutex_unlock(&ofproto->vsp_mutex);
6079     return ret;
6080 }
6081
6082 static struct vlan_splinter *
6083 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
6084 {
6085     struct vlan_splinter *vsp;
6086
6087     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
6088                              hash_ofp_port(vlandev_ofp_port),
6089                              &ofproto->vlandev_map) {
6090         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
6091             return vsp;
6092         }
6093     }
6094
6095     return NULL;
6096 }
6097
6098 /* Returns the OpenFlow port number of the "real" device underlying the Linux
6099  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
6100  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
6101  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
6102  * eth0 and store 9 in '*vid'.
6103  *
6104  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
6105  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
6106  * always does.*/
6107 static ofp_port_t
6108 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
6109                        ofp_port_t vlandev_ofp_port, int *vid)
6110     OVS_REQUIRES(ofproto->vsp_mutex)
6111 {
6112     if (!hmap_is_empty(&ofproto->vlandev_map)) {
6113         const struct vlan_splinter *vsp;
6114
6115         vsp = vlandev_find(ofproto, vlandev_ofp_port);
6116         if (vsp) {
6117             if (vid) {
6118                 *vid = vsp->vid;
6119             }
6120             return vsp->realdev_ofp_port;
6121         }
6122     }
6123     return 0;
6124 }
6125
6126 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
6127  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
6128  * 'flow->in_port' to the "real" device backing the VLAN device, sets
6129  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Otherwise (which is
6130  * always the case unless VLAN splinters are enabled), returns false without
6131  * making any changes. */
6132 bool
6133 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
6134     OVS_EXCLUDED(ofproto->vsp_mutex)
6135 {
6136     ofp_port_t realdev;
6137     int vid;
6138
6139     ovs_mutex_lock(&ofproto->vsp_mutex);
6140     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
6141     ovs_mutex_unlock(&ofproto->vsp_mutex);
6142     if (!realdev) {
6143         return false;
6144     }
6145
6146     /* Cause the flow to be processed as if it came in on the real device with
6147      * the VLAN device's VLAN ID. */
6148     flow->in_port.ofp_port = realdev;
6149     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
6150     return true;
6151 }
6152
6153 static void
6154 vsp_remove(struct ofport_dpif *port)
6155 {
6156     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6157     struct vlan_splinter *vsp;
6158
6159     ovs_mutex_lock(&ofproto->vsp_mutex);
6160     vsp = vlandev_find(ofproto, port->up.ofp_port);
6161     if (vsp) {
6162         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
6163         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
6164         free(vsp);
6165
6166         port->realdev_ofp_port = 0;
6167     } else {
6168         VLOG_ERR("missing vlan device record");
6169     }
6170     ovs_mutex_unlock(&ofproto->vsp_mutex);
6171 }
6172
6173 static void
6174 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6175 {
6176     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6177
6178     ovs_mutex_lock(&ofproto->vsp_mutex);
6179     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6180         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6181             == realdev_ofp_port)) {
6182         struct vlan_splinter *vsp;
6183
6184         vsp = xmalloc(sizeof *vsp);
6185         vsp->realdev_ofp_port = realdev_ofp_port;
6186         vsp->vlandev_ofp_port = port->up.ofp_port;
6187         vsp->vid = vid;
6188
6189         port->realdev_ofp_port = realdev_ofp_port;
6190
6191         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6192                     hash_ofp_port(port->up.ofp_port));
6193         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6194                     hash_realdev_vid(realdev_ofp_port, vid));
6195     } else {
6196         VLOG_ERR("duplicate vlan device record");
6197     }
6198     ovs_mutex_unlock(&ofproto->vsp_mutex);
6199 }
6200
6201 static odp_port_t
6202 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6203 {
6204     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6205     return ofport ? ofport->odp_port : ODPP_NONE;
6206 }
6207
6208 struct ofport_dpif *
6209 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6210 {
6211     struct ofport_dpif *port;
6212
6213     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6214     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6215                              &backer->odp_to_ofport_map) {
6216         if (port->odp_port == odp_port) {
6217             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6218             return port;
6219         }
6220     }
6221
6222     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6223     return NULL;
6224 }
6225
6226 static ofp_port_t
6227 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6228 {
6229     struct ofport_dpif *port;
6230
6231     port = odp_port_to_ofport(ofproto->backer, odp_port);
6232     if (port && &ofproto->up == port->up.ofproto) {
6233         return port->up.ofp_port;
6234     } else {
6235         return OFPP_NONE;
6236     }
6237 }
6238
6239 /* Compute exponentially weighted moving average, adding 'new' as the newest,
6240  * most heavily weighted element.  'base' designates the rate of decay: after
6241  * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
6242  * (about .37). */
6243 static void
6244 exp_mavg(double *avg, int base, double new)
6245 {
6246     *avg = (*avg * (base - 1) + new) / base;
6247 }
6248
6249 static void
6250 update_moving_averages(struct dpif_backer *backer)
6251 {
6252     const int min_ms = 60 * 1000; /* milliseconds in one minute. */
6253     long long int minutes = (time_msec() - backer->created) / min_ms;
6254
6255     if (minutes > 0) {
6256         backer->lifetime.add_rate = (double) backer->total_subfacet_add_count
6257             / minutes;
6258         backer->lifetime.del_rate = (double) backer->total_subfacet_del_count
6259             / minutes;
6260     } else {
6261         backer->lifetime.add_rate = 0.0;
6262         backer->lifetime.del_rate = 0.0;
6263     }
6264
6265     /* Update hourly averages on the minute boundaries. */
6266     if (time_msec() - backer->last_minute >= min_ms) {
6267         exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count);
6268         exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count);
6269
6270         /* Update daily averages on the hour boundaries. */
6271         if ((backer->last_minute - backer->created) / min_ms % 60 == 59) {
6272             exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate);
6273             exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate);
6274         }
6275
6276         backer->total_subfacet_add_count += backer->subfacet_add_count;
6277         backer->total_subfacet_del_count += backer->subfacet_del_count;
6278         backer->subfacet_add_count = 0;
6279         backer->subfacet_del_count = 0;
6280         backer->last_minute += min_ms;
6281     }
6282 }
6283
6284 const struct ofproto_class ofproto_dpif_class = {
6285     init,
6286     enumerate_types,
6287     enumerate_names,
6288     del,
6289     port_open_type,
6290     type_run,
6291     type_run_fast,
6292     type_wait,
6293     alloc,
6294     construct,
6295     destruct,
6296     dealloc,
6297     run,
6298     run_fast,
6299     wait,
6300     get_memory_usage,
6301     flush,
6302     get_features,
6303     get_tables,
6304     port_alloc,
6305     port_construct,
6306     port_destruct,
6307     port_dealloc,
6308     port_modified,
6309     port_reconfigured,
6310     port_query_by_name,
6311     port_add,
6312     port_del,
6313     port_get_stats,
6314     port_dump_start,
6315     port_dump_next,
6316     port_dump_done,
6317     port_poll,
6318     port_poll_wait,
6319     port_is_lacp_current,
6320     NULL,                       /* rule_choose_table */
6321     rule_alloc,
6322     rule_construct,
6323     rule_destruct,
6324     rule_dealloc,
6325     rule_get_stats,
6326     rule_execute,
6327     rule_modify_actions,
6328     set_frag_handling,
6329     packet_out,
6330     set_netflow,
6331     get_netflow_ids,
6332     set_sflow,
6333     set_ipfix,
6334     set_cfm,
6335     get_cfm_status,
6336     set_bfd,
6337     get_bfd_status,
6338     set_stp,
6339     get_stp_status,
6340     set_stp_port,
6341     get_stp_port_status,
6342     set_queues,
6343     bundle_set,
6344     bundle_remove,
6345     mirror_set__,
6346     mirror_get_stats__,
6347     set_flood_vlans,
6348     is_mirror_output_bundle,
6349     forward_bpdu_changed,
6350     set_mac_table_config,
6351     set_realdev,
6352     NULL,                       /* meter_get_features */
6353     NULL,                       /* meter_set */
6354     NULL,                       /* meter_get */
6355     NULL,                       /* meter_del */
6356 };