ofproto/ofproto-dpif.c

   1 /*
   2  * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18
  19 #include "ofproto/ofproto-dpif.h"
  20 #include "ofproto/ofproto-provider.h"
  21
  22 #include <errno.h>
  23
  24 #include "bfd.h"
  25 #include "bond.h"
  26 #include "bundle.h"
  27 #include "byte-order.h"
  28 #include "connmgr.h"
  29 #include "coverage.h"
  30 #include "cfm.h"
  31 #include "dpif.h"
  32 #include "dynamic-string.h"
  33 #include "fail-open.h"
  34 #include "guarded-list.h"
  35 #include "hmapx.h"
  36 #include "lacp.h"
  37 #include "learn.h"
  38 #include "mac-learning.h"
  39 #include "meta-flow.h"
  40 #include "multipath.h"
  41 #include "netdev-vport.h"
  42 #include "netdev.h"
  43 #include "netlink.h"
  44 #include "nx-match.h"
  45 #include "odp-util.h"
  46 #include "odp-execute.h"
  47 #include "ofp-util.h"
  48 #include "ofpbuf.h"
  49 #include "ofp-actions.h"
  50 #include "ofp-parse.h"
  51 #include "ofp-print.h"
  52 #include "ofproto-dpif-governor.h"
  53 #include "ofproto-dpif-ipfix.h"
  54 #include "ofproto-dpif-mirror.h"
  55 #include "ofproto-dpif-sflow.h"
  56 #include "ofproto-dpif-upcall.h"
  57 #include "ofproto-dpif-xlate.h"
  58 #include "poll-loop.h"
  59 #include "simap.h"
  60 #include "smap.h"
  61 #include "timer.h"
  62 #include "tunnel.h"
  63 #include "unaligned.h"
  64 #include "unixctl.h"
  65 #include "vlan-bitmap.h"
  66 #include "vlog.h"
  67
  68 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
  69
  70 COVERAGE_DEFINE(ofproto_dpif_expired);
  71 COVERAGE_DEFINE(facet_revalidate);
  72 COVERAGE_DEFINE(facet_unexpected);
  73 COVERAGE_DEFINE(facet_create);
  74 COVERAGE_DEFINE(facet_remove);
  75 COVERAGE_DEFINE(subfacet_create);
  76 COVERAGE_DEFINE(subfacet_destroy);
  77 COVERAGE_DEFINE(subfacet_install_fail);
  78 COVERAGE_DEFINE(packet_in_overflow);
  79
  80 /* Number of implemented OpenFlow tables. */
  81 enum { N_TABLES = 255 };
  82 enum { TBL_INTERNAL = N_TABLES - 1 };    /* Used for internal hidden rules. */
  83 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
  84
  85 struct flow_miss;
  86 struct facet;
  87
  88 struct rule_dpif {
  89     struct rule up;
  90
  91     /* These statistics:
  92      *
  93      *   - Do include packets and bytes from facets that have been deleted or
  94      *     whose own statistics have been folded into the rule.
  95      *
  96      *   - Do include packets and bytes sent "by hand" that were accounted to
  97      *     the rule without any facet being involved (this is a rare corner
  98      *     case in rule_execute()).
  99      *
 100      *   - Do not include packet or bytes that can be obtained from any facet's
 101      *     packet_count or byte_count member or that can be obtained from the
 102      *     datapath by, e.g., dpif_flow_get() for any subfacet.
 103      */
 104     struct ovs_mutex stats_mutex;
 105     uint64_t packet_count OVS_GUARDED;  /* Number of packets received. */
 106     uint64_t byte_count OVS_GUARDED;    /* Number of bytes received. */
 107 };
 108
 109 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
 110 static struct rule_dpif *rule_dpif_cast(const struct rule *);
 111
 112 struct ofbundle {
 113     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
 114     struct ofproto_dpif *ofproto; /* Owning ofproto. */
 115     void *aux;                  /* Key supplied by ofproto's client. */
 116     char *name;                 /* Identifier for log messages. */
 117
 118     /* Configuration. */
 119     struct list ports;          /* Contains "struct ofport"s. */
 120     enum port_vlan_mode vlan_mode; /* VLAN mode */
 121     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
 122     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
 123                                  * NULL if all VLANs are trunked. */
 124     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
 125     struct bond *bond;          /* Nonnull iff more than one port. */
 126     bool use_priority_tags;     /* Use 802.1p tag for frames in VLAN 0? */
 127
 128     /* Status. */
 129     bool floodable;          /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
 130 };
 131
 132 static void bundle_remove(struct ofport *);
 133 static void bundle_update(struct ofbundle *);
 134 static void bundle_destroy(struct ofbundle *);
 135 static void bundle_del_port(struct ofport_dpif *);
 136 static void bundle_run(struct ofbundle *);
 137 static void bundle_wait(struct ofbundle *);
 138
 139 static void stp_run(struct ofproto_dpif *ofproto);
 140 static void stp_wait(struct ofproto_dpif *ofproto);
 141 static int set_stp_port(struct ofport *,
 142                         const struct ofproto_port_stp_settings *);
 143
 144 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
 145                               enum slow_path_reason,
 146                               uint64_t *stub, size_t stub_size,
 147                               const struct nlattr **actionsp,
 148                               size_t *actions_lenp);
 149
 150 /* A subfacet (see "struct subfacet" below) has three possible installation
 151  * states:
 152  *
 153  *   - SF_NOT_INSTALLED: Not installed in the datapath.  This will only be the
 154  *     case just after the subfacet is created, just before the subfacet is
 155  *     destroyed, or if the datapath returns an error when we try to install a
 156  *     subfacet.
 157  *
 158  *   - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
 159  *
 160  *   - SF_SLOW_PATH: An action that sends every packet for the subfacet through
 161  *     ofproto_dpif is installed in the datapath.
 162  */
 163 enum subfacet_path {
 164     SF_NOT_INSTALLED,           /* No datapath flow for this subfacet. */
 165     SF_FAST_PATH,               /* Full actions are installed. */
 166     SF_SLOW_PATH,               /* Send-to-userspace action is installed. */
 167 };
 168
 169 /* A dpif flow and actions associated with a facet.
 170  *
 171  * See also the large comment on struct facet. */
 172 struct subfacet {
 173     /* Owners. */
 174     struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
 175     struct list list_node;      /* In struct facet's 'facets' list. */
 176     struct facet *facet;        /* Owning facet. */
 177     struct dpif_backer *backer; /* Owning backer. */
 178
 179     struct nlattr *key;
 180     int key_len;
 181
 182     long long int used;         /* Time last used; time created if not used. */
 183     long long int created;      /* Time created. */
 184
 185     uint64_t dp_packet_count;   /* Last known packet count in the datapath. */
 186     uint64_t dp_byte_count;     /* Last known byte count in the datapath. */
 187
 188     enum subfacet_path path;    /* Installed in datapath? */
 189 };
 190
 191 #define SUBFACET_DESTROY_MAX_BATCH 50
 192
 193 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *);
 194 static struct subfacet *subfacet_find(struct dpif_backer *,
 195                                       const struct nlattr *key, size_t key_len,
 196                                       uint32_t key_hash);
 197 static void subfacet_destroy(struct subfacet *);
 198 static void subfacet_destroy__(struct subfacet *);
 199 static void subfacet_destroy_batch(struct dpif_backer *,
 200                                    struct subfacet **, int n);
 201 static void subfacet_reset_dp_stats(struct subfacet *,
 202                                     struct dpif_flow_stats *);
 203 static void subfacet_update_stats(struct subfacet *,
 204                                   const struct dpif_flow_stats *);
 205 static int subfacet_install(struct subfacet *,
 206                             const struct ofpbuf *odp_actions,
 207                             struct dpif_flow_stats *);
 208 static void subfacet_uninstall(struct subfacet *);
 209
 210 /* A unique, non-overlapping instantiation of an OpenFlow flow.
 211  *
 212  * A facet associates a "struct flow", which represents the Open vSwitch
 213  * userspace idea of an exact-match flow, with one or more subfacets.
 214  * While the facet is created based on an exact-match flow, it is stored
 215  * within the ofproto based on the wildcards that could be expressed
 216  * based on the flow table and other configuration.  (See the 'wc'
 217  * description in "struct xlate_out" for more details.)
 218  *
 219  * Each subfacet tracks the datapath's idea of the flow equivalent to
 220  * the facet.  When the kernel module (or other dpif implementation) and
 221  * Open vSwitch userspace agree on the definition of a flow key, there
 222  * is exactly one subfacet per facet.  If the dpif implementation
 223  * supports more-specific flow matching than userspace, however, a facet
 224  * can have more than one subfacet.  Examples include the dpif
 225  * implementation not supporting the same wildcards as userspace or some
 226  * distinction in flow that userspace simply doesn't understand.
 227  *
 228  * Flow expiration works in terms of subfacets, so a facet must have at
 229  * least one subfacet or it will never expire, leaking memory. */
 230 struct facet {
 231     /* Owner. */
 232     struct ofproto_dpif *ofproto;
 233
 234     /* Owned data. */
 235     struct list subfacets;
 236     long long int used;         /* Time last used; time created if not used. */
 237
 238     /* Key. */
 239     struct flow flow;           /* Flow of the creating subfacet. */
 240     struct cls_rule cr;         /* In 'ofproto_dpif's facets classifier. */
 241
 242     /* These statistics:
 243      *
 244      *   - Do include packets and bytes sent "by hand", e.g. with
 245      *     dpif_execute().
 246      *
 247      *   - Do include packets and bytes that were obtained from the datapath
 248      *     when a subfacet's statistics were reset (e.g. dpif_flow_put() with
 249      *     DPIF_FP_ZERO_STATS).
 250      *
 251      *   - Do not include packets or bytes that can be obtained from the
 252      *     datapath for any existing subfacet.
 253      */
 254     uint64_t packet_count;       /* Number of packets received. */
 255     uint64_t byte_count;         /* Number of bytes received. */
 256
 257     /* Resubmit statistics. */
 258     uint64_t prev_packet_count;  /* Number of packets from last stats push. */
 259     uint64_t prev_byte_count;    /* Number of bytes from last stats push. */
 260     long long int prev_used;     /* Used time from last stats push. */
 261
 262     /* Accounting. */
 263     uint64_t accounted_bytes;    /* Bytes processed by facet_account(). */
 264     struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
 265     uint8_t tcp_flags;           /* TCP flags seen for this 'rule'. */
 266
 267     struct xlate_out xout;
 268
 269     /* Storage for a single subfacet, to reduce malloc() time and space
 270      * overhead.  (A facet always has at least one subfacet and in the common
 271      * case has exactly one subfacet.  However, 'one_subfacet' may not
 272      * always be valid, since it could have been removed after newer
 273      * subfacets were pushed onto the 'subfacets' list.) */
 274     struct subfacet one_subfacet;
 275
 276     long long int learn_rl;      /* Rate limiter for facet_learn(). */
 277 };
 278
 279 static struct facet *facet_create(const struct flow_miss *);
 280 static void facet_remove(struct facet *);
 281 static void facet_free(struct facet *);
 282
 283 static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
 284 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
 285                                         const struct flow *);
 286 static bool facet_revalidate(struct facet *);
 287 static bool facet_check_consistency(struct facet *);
 288
 289 static void facet_flush_stats(struct facet *);
 290
 291 static void facet_reset_counters(struct facet *);
 292 static void flow_push_stats(struct ofproto_dpif *, struct flow *,
 293                             struct dpif_flow_stats *, bool may_learn);
 294 static void facet_push_stats(struct facet *, bool may_learn);
 295 static void facet_learn(struct facet *);
 296 static void facet_account(struct facet *);
 297 static void push_all_stats(void);
 298
 299 static bool facet_is_controller_flow(struct facet *);
 300
 301 struct ofport_dpif {
 302     struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
 303     struct ofport up;
 304
 305     odp_port_t odp_port;
 306     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
 307     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
 308     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
 309     struct bfd *bfd;            /* BFD, if any. */
 310     bool may_enable;            /* May be enabled in bonds. */
 311     bool is_tunnel;             /* This port is a tunnel. */
 312     long long int carrier_seq;  /* Carrier status changes. */
 313     struct ofport_dpif *peer;   /* Peer if patch port. */
 314
 315     /* Spanning tree. */
 316     struct stp_port *stp_port;  /* Spanning Tree Protocol, if any. */
 317     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
 318     long long int stp_state_entered;
 319
 320     /* Queue to DSCP mapping. */
 321     struct ofproto_port_queue *qdscp;
 322     size_t n_qdscp;
 323
 324     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 325      *
 326      * This is deprecated.  It is only for compatibility with broken device
 327      * drivers in old versions of Linux that do not properly support VLANs when
 328      * VLAN devices are not used.  When broken device drivers are no longer in
 329      * widespread use, we will delete these interfaces. */
 330     ofp_port_t realdev_ofp_port;
 331     int vlandev_vid;
 332 };
 333
 334 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
 335  *
 336  * This is deprecated.  It is only for compatibility with broken device drivers
 337  * in old versions of Linux that do not properly support VLANs when VLAN
 338  * devices are not used.  When broken device drivers are no longer in
 339  * widespread use, we will delete these interfaces. */
 340 struct vlan_splinter {
 341     struct hmap_node realdev_vid_node;
 342     struct hmap_node vlandev_node;
 343     ofp_port_t realdev_ofp_port;
 344     ofp_port_t vlandev_ofp_port;
 345     int vid;
 346 };
 347
 348 static void vsp_remove(struct ofport_dpif *);
 349 static void vsp_add(struct ofport_dpif *, ofp_port_t realdev_ofp_port, int vid);
 350
 351 static odp_port_t ofp_port_to_odp_port(const struct ofproto_dpif *,
 352                                        ofp_port_t);
 353
 354 static ofp_port_t odp_port_to_ofp_port(const struct ofproto_dpif *,
 355                                        odp_port_t);
 356
 357 static struct ofport_dpif *
 358 ofport_dpif_cast(const struct ofport *ofport)
 359 {
 360     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
 361 }
 362
 363 static void port_run(struct ofport_dpif *);
 364 static void port_run_fast(struct ofport_dpif *);
 365 static void port_wait(struct ofport_dpif *);
 366 static int set_bfd(struct ofport *, const struct smap *);
 367 static int set_cfm(struct ofport *, const struct cfm_settings *);
 368 static void ofport_update_peer(struct ofport_dpif *);
 369 static void run_fast_rl(void);
 370 static int run_fast(struct ofproto *);
 371
 372 struct dpif_completion {
 373     struct list list_node;
 374     struct ofoperation *op;
 375 };
 376
 377 /* Reasons that we might need to revalidate every facet, and corresponding
 378  * coverage counters.
 379  *
 380  * A value of 0 means that there is no need to revalidate.
 381  *
 382  * It would be nice to have some cleaner way to integrate with coverage
 383  * counters, but with only a few reasons I guess this is good enough for
 384  * now. */
 385 enum revalidate_reason {
 386     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
 387     REV_STP,                   /* Spanning tree protocol port status change. */
 388     REV_BOND,                  /* Bonding changed. */
 389     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
 390     REV_FLOW_TABLE,            /* Flow table changed. */
 391     REV_MAC_LEARNING,          /* Mac learning changed. */
 392     REV_INCONSISTENCY          /* Facet self-check failed. */
 393 };
 394 COVERAGE_DEFINE(rev_reconfigure);
 395 COVERAGE_DEFINE(rev_stp);
 396 COVERAGE_DEFINE(rev_bond);
 397 COVERAGE_DEFINE(rev_port_toggled);
 398 COVERAGE_DEFINE(rev_flow_table);
 399 COVERAGE_DEFINE(rev_mac_learning);
 400 COVERAGE_DEFINE(rev_inconsistency);
 401
 402 struct avg_subfacet_rates {
 403     double add_rate;   /* Moving average of new flows created per minute. */
 404     double del_rate;   /* Moving average of flows deleted per minute. */
 405 };
 406
 407 /* All datapaths of a given type share a single dpif backer instance. */
 408 struct dpif_backer {
 409     char *type;
 410     int refcount;
 411     struct dpif *dpif;
 412     struct udpif *udpif;
 413     struct timer next_expiration;
 414
 415     struct ovs_rwlock odp_to_ofport_lock;
 416     struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */
 417
 418     struct simap tnl_backers;      /* Set of dpif ports backing tunnels. */
 419
 420     /* Facet revalidation flags applying to facets which use this backer. */
 421     enum revalidate_reason need_revalidate; /* Revalidate every facet. */
 422
 423     struct hmap drop_keys; /* Set of dropped odp keys. */
 424     bool recv_set_enable; /* Enables or disables receiving packets. */
 425
 426     struct hmap subfacets;
 427     struct governor *governor;
 428
 429     /* Subfacet statistics.
 430      *
 431      * These keep track of the total number of subfacets added and deleted and
 432      * flow life span.  They are useful for computing the flow rates stats
 433      * exposed via "ovs-appctl dpif/show".  The goal is to learn about
 434      * traffic patterns in ways that we can use later to improve Open vSwitch
 435      * performance in new situations.  */
 436     long long int created;           /* Time when it is created. */
 437     unsigned max_n_subfacet;         /* Maximum number of flows */
 438     unsigned avg_n_subfacet;         /* Average number of flows. */
 439     long long int avg_subfacet_life; /* Average life span of subfacets. */
 440
 441     /* Number of upcall handling threads. */
 442     unsigned int n_handler_threads;
 443 };
 444
 445 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
 446 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
 447
 448 static void drop_key_clear(struct dpif_backer *);
 449
 450 struct ofproto_dpif {
 451     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
 452     struct ofproto up;
 453     struct dpif_backer *backer;
 454
 455     /* Special OpenFlow rules. */
 456     struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
 457     struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
 458     struct rule_dpif *drop_frags_rule; /* Used in OFPC_FRAG_DROP mode. */
 459
 460     /* Bridging. */
 461     struct netflow *netflow;
 462     struct dpif_sflow *sflow;
 463     struct dpif_ipfix *ipfix;
 464     struct hmap bundles;        /* Contains "struct ofbundle"s. */
 465     struct mac_learning *ml;
 466     bool has_bonded_bundles;
 467     struct mbridge *mbridge;
 468
 469     /* Facets. */
 470     struct classifier facets;     /* Contains 'struct facet's. */
 471     long long int consistency_rl;
 472
 473     struct ovs_mutex stats_mutex;
 474     struct netdev_stats stats OVS_GUARDED; /* To account packets generated and
 475                                             * consumed in userspace. */
 476
 477     /* Spanning tree. */
 478     struct stp *stp;
 479     long long int stp_last_tick;
 480
 481     /* VLAN splinters. */
 482     struct ovs_mutex vsp_mutex;
 483     struct hmap realdev_vid_map OVS_GUARDED; /* (realdev,vid) -> vlandev. */
 484     struct hmap vlandev_map OVS_GUARDED;     /* vlandev -> (realdev,vid). */
 485
 486     /* Ports. */
 487     struct sset ports;             /* Set of standard port names. */
 488     struct sset ghost_ports;       /* Ports with no datapath port. */
 489     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
 490     int port_poll_errno;           /* Last errno for port_poll() reply. */
 491
 492     /* Per ofproto's dpif stats. */
 493     uint64_t n_hit;
 494     uint64_t n_missed;
 495
 496     /* Work queues. */
 497     struct guarded_list pins;      /* Contains "struct ofputil_packet_in"s. */
 498 };
 499
 500 /* By default, flows in the datapath are wildcarded (megaflows).  They
 501  * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */
 502 static bool enable_megaflows = true;
 503
 504 /* All existing ofproto_dpif instances, indexed by ->up.name. */
 505 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
 506
 507 static void ofproto_dpif_unixctl_init(void);
 508
 509 static inline struct ofproto_dpif *
 510 ofproto_dpif_cast(const struct ofproto *ofproto)
 511 {
 512     ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
 513     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
 514 }
 515
 516 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto,
 517                                         ofp_port_t ofp_port);
 518 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
 519                           const struct ofpbuf *packet, struct ds *);
 520
 521 /* Upcalls. */
 522 static void handle_upcalls(struct dpif_backer *);
 523
 524 /* Flow expiration. */
 525 static int expire(struct dpif_backer *);
 526
 527 /* NetFlow. */
 528 static void send_netflow_active_timeouts(struct ofproto_dpif *);
 529
 530 /* Global variables. */
 531 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 532
 533 /* Initial mappings of port to bridge mappings. */
 534 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 535
 536 /* Executes 'fm'.  The caller retains ownership of 'fm' and everything in
 537  * it. */
 538 void
 539 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
 540                       struct ofputil_flow_mod *fm)
 541 {
 542     ofproto_flow_mod(&ofproto->up, fm);
 543 }
 544
 545 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
 546  * Takes ownership of 'pin' and pin->packet. */
 547 void
 548 ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
 549                             struct ofputil_packet_in *pin)
 550 {
 551     if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) {
 552         COVERAGE_INC(packet_in_overflow);
 553         free(CONST_CAST(void *, pin->packet));
 554         free(pin);
 555     }
 556 }
 557 \f
 558 /* Factory functions. */
 559
 560 static void
 561 init(const struct shash *iface_hints)
 562 {
 563     struct shash_node *node;
 564
 565     /* Make a local copy, since we don't own 'iface_hints' elements. */
 566     SHASH_FOR_EACH(node, iface_hints) {
 567         const struct iface_hint *orig_hint = node->data;
 568         struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
 569
 570         new_hint->br_name = xstrdup(orig_hint->br_name);
 571         new_hint->br_type = xstrdup(orig_hint->br_type);
 572         new_hint->ofp_port = orig_hint->ofp_port;
 573
 574         shash_add(&init_ofp_ports, node->name, new_hint);
 575     }
 576 }
 577
 578 static void
 579 enumerate_types(struct sset *types)
 580 {
 581     dp_enumerate_types(types);
 582 }
 583
 584 static int
 585 enumerate_names(const char *type, struct sset *names)
 586 {
 587     struct ofproto_dpif *ofproto;
 588
 589     sset_clear(names);
 590     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 591         if (strcmp(type, ofproto->up.type)) {
 592             continue;
 593         }
 594         sset_add(names, ofproto->up.name);
 595     }
 596
 597     return 0;
 598 }
 599
 600 static int
 601 del(const char *type, const char *name)
 602 {
 603     struct dpif *dpif;
 604     int error;
 605
 606     error = dpif_open(name, type, &dpif);
 607     if (!error) {
 608         error = dpif_delete(dpif);
 609         dpif_close(dpif);
 610     }
 611     return error;
 612 }
 613 \f
 614 static const char *
 615 port_open_type(const char *datapath_type, const char *port_type)
 616 {
 617     return dpif_port_open_type(datapath_type, port_type);
 618 }
 619
 620 /* Type functions. */
 621
 622 static void process_dpif_port_changes(struct dpif_backer *);
 623 static void process_dpif_all_ports_changed(struct dpif_backer *);
 624 static void process_dpif_port_change(struct dpif_backer *,
 625                                      const char *devname);
 626 static void process_dpif_port_error(struct dpif_backer *, int error);
 627
 628 static struct ofproto_dpif *
 629 lookup_ofproto_dpif_by_port_name(const char *name)
 630 {
 631     struct ofproto_dpif *ofproto;
 632
 633     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 634         if (sset_contains(&ofproto->ports, name)) {
 635             return ofproto;
 636         }
 637     }
 638
 639     return NULL;
 640 }
 641
 642 static int
 643 type_run(const char *type)
 644 {
 645     static long long int push_timer = LLONG_MIN;
 646     struct dpif_backer *backer;
 647
 648     backer = shash_find_data(&all_dpif_backers, type);
 649     if (!backer) {
 650         /* This is not necessarily a problem, since backers are only
 651          * created on demand. */
 652         return 0;
 653     }
 654
 655     dpif_run(backer->dpif);
 656
 657     /* The most natural place to push facet statistics is when they're pulled
 658      * from the datapath.  However, when there are many flows in the datapath,
 659      * this expensive operation can occur so frequently, that it reduces our
 660      * ability to quickly set up flows.  To reduce the cost, we push statistics
 661      * here instead. */
 662     if (time_msec() > push_timer) {
 663         push_timer = time_msec() + 2000;
 664         push_all_stats();
 665     }
 666
 667     /* If vswitchd started with other_config:flow_restore_wait set as "true",
 668      * and the configuration has now changed to "false", enable receiving
 669      * packets from the datapath. */
 670     if (!backer->recv_set_enable && !ofproto_get_flow_restore_wait()) {
 671         int error;
 672
 673         backer->recv_set_enable = true;
 674
 675         error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
 676         if (error) {
 677             udpif_recv_set(backer->udpif, 0, false);
 678             VLOG_ERR("Failed to enable receiving packets in dpif.");
 679             return error;
 680         }
 681         udpif_recv_set(backer->udpif, n_handler_threads,
 682                        backer->recv_set_enable);
 683         dpif_flow_flush(backer->dpif);
 684         backer->need_revalidate = REV_RECONFIGURE;
 685     }
 686
 687     /* If the n_handler_threads is reconfigured, call udpif_recv_set()
 688      * to reset the handler threads. */
 689     if (backer->n_handler_threads != n_handler_threads) {
 690         udpif_recv_set(backer->udpif, n_handler_threads,
 691                        backer->recv_set_enable);
 692         backer->n_handler_threads = n_handler_threads;
 693     }
 694
 695     if (backer->need_revalidate) {
 696         struct ofproto_dpif *ofproto;
 697         struct simap_node *node;
 698         struct simap tmp_backers;
 699
 700         /* Handle tunnel garbage collection. */
 701         simap_init(&tmp_backers);
 702         simap_swap(&backer->tnl_backers, &tmp_backers);
 703
 704         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 705             struct ofport_dpif *iter;
 706
 707             if (backer != ofproto->backer) {
 708                 continue;
 709             }
 710
 711             HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
 712                 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
 713                 const char *dp_port;
 714
 715                 if (!iter->is_tunnel) {
 716                     continue;
 717                 }
 718
 719                 dp_port = netdev_vport_get_dpif_port(iter->up.netdev,
 720                                                      namebuf, sizeof namebuf);
 721                 node = simap_find(&tmp_backers, dp_port);
 722                 if (node) {
 723                     simap_put(&backer->tnl_backers, dp_port, node->data);
 724                     simap_delete(&tmp_backers, node);
 725                     node = simap_find(&backer->tnl_backers, dp_port);
 726                 } else {
 727                     node = simap_find(&backer->tnl_backers, dp_port);
 728                     if (!node) {
 729                         odp_port_t odp_port = ODPP_NONE;
 730
 731                         if (!dpif_port_add(backer->dpif, iter->up.netdev,
 732                                            &odp_port)) {
 733                             simap_put(&backer->tnl_backers, dp_port,
 734                                       odp_to_u32(odp_port));
 735                             node = simap_find(&backer->tnl_backers, dp_port);
 736                         }
 737                     }
 738                 }
 739
 740                 iter->odp_port = node ? u32_to_odp(node->data) : ODPP_NONE;
 741                 if (tnl_port_reconfigure(iter, iter->up.netdev,
 742                                          iter->odp_port)) {
 743                     backer->need_revalidate = REV_RECONFIGURE;
 744                 }
 745             }
 746         }
 747
 748         SIMAP_FOR_EACH (node, &tmp_backers) {
 749             dpif_port_del(backer->dpif, u32_to_odp(node->data));
 750         }
 751         simap_destroy(&tmp_backers);
 752
 753         switch (backer->need_revalidate) {
 754         case REV_RECONFIGURE:   COVERAGE_INC(rev_reconfigure);   break;
 755         case REV_STP:           COVERAGE_INC(rev_stp);           break;
 756         case REV_BOND:          COVERAGE_INC(rev_bond);          break;
 757         case REV_PORT_TOGGLED:  COVERAGE_INC(rev_port_toggled);  break;
 758         case REV_FLOW_TABLE:    COVERAGE_INC(rev_flow_table);    break;
 759         case REV_MAC_LEARNING:  COVERAGE_INC(rev_mac_learning);  break;
 760         case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
 761         }
 762         backer->need_revalidate = 0;
 763
 764         /* Clear the drop_keys in case we should now be accepting some
 765          * formerly dropped flows. */
 766         drop_key_clear(backer);
 767
 768         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 769             struct facet *facet, *next;
 770             struct ofport_dpif *ofport;
 771             struct cls_cursor cursor;
 772             struct ofbundle *bundle;
 773
 774             if (ofproto->backer != backer) {
 775                 continue;
 776             }
 777
 778             ovs_rwlock_wrlock(&xlate_rwlock);
 779             xlate_ofproto_set(ofproto, ofproto->up.name,
 780                               ofproto->backer->dpif, ofproto->miss_rule,
 781                               ofproto->no_packet_in_rule, ofproto->ml,
 782                               ofproto->stp, ofproto->mbridge,
 783                               ofproto->sflow, ofproto->ipfix,
 784                               ofproto->up.frag_handling,
 785                               ofproto->up.forward_bpdu,
 786                               connmgr_has_in_band(ofproto->up.connmgr),
 787                               ofproto->netflow != NULL);
 788
 789             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
 790                 xlate_bundle_set(ofproto, bundle, bundle->name,
 791                                  bundle->vlan_mode, bundle->vlan,
 792                                  bundle->trunks, bundle->use_priority_tags,
 793                                  bundle->bond, bundle->lacp,
 794                                  bundle->floodable);
 795             }
 796
 797             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
 798                 int stp_port = ofport->stp_port
 799                     ? stp_port_no(ofport->stp_port)
 800                     : -1;
 801                 xlate_ofport_set(ofproto, ofport->bundle, ofport,
 802                                  ofport->up.ofp_port, ofport->odp_port,
 803                                  ofport->up.netdev, ofport->cfm,
 804                                  ofport->bfd, ofport->peer, stp_port,
 805                                  ofport->qdscp, ofport->n_qdscp,
 806                                  ofport->up.pp.config, ofport->is_tunnel,
 807                                  ofport->may_enable);
 808             }
 809             ovs_rwlock_unlock(&xlate_rwlock);
 810
 811             /* Only ofproto-dpif cares about the facet classifier so we just
 812              * lock cls_cursor_init() to appease the thread safety analysis. */
 813             ovs_rwlock_rdlock(&ofproto->facets.rwlock);
 814             cls_cursor_init(&cursor, &ofproto->facets, NULL);
 815             ovs_rwlock_unlock(&ofproto->facets.rwlock);
 816             CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
 817                 facet_revalidate(facet);
 818                 run_fast_rl();
 819             }
 820         }
 821
 822         udpif_revalidate(backer->udpif);
 823     }
 824
 825     if (!backer->recv_set_enable) {
 826         /* Wake up before a max of 1000ms. */
 827         timer_set_duration(&backer->next_expiration, 1000);
 828     } else if (timer_expired(&backer->next_expiration)) {
 829         int delay = expire(backer);
 830         timer_set_duration(&backer->next_expiration, delay);
 831     }
 832
 833     process_dpif_port_changes(backer);
 834
 835     if (backer->governor) {
 836         size_t n_subfacets;
 837
 838         governor_run(backer->governor);
 839
 840         /* If the governor has shrunk to its minimum size and the number of
 841          * subfacets has dwindled, then drop the governor entirely.
 842          *
 843          * For hysteresis, the number of subfacets to drop the governor is
 844          * smaller than the number needed to trigger its creation. */
 845         n_subfacets = hmap_count(&backer->subfacets);
 846         if (n_subfacets * 4 < flow_eviction_threshold
 847             && governor_is_idle(backer->governor)) {
 848             governor_destroy(backer->governor);
 849             backer->governor = NULL;
 850         }
 851     }
 852
 853     return 0;
 854 }
 855
 856 /* Check for and handle port changes in 'backer''s dpif. */
 857 static void
 858 process_dpif_port_changes(struct dpif_backer *backer)
 859 {
 860     for (;;) {
 861         char *devname;
 862         int error;
 863
 864         error = dpif_port_poll(backer->dpif, &devname);
 865         switch (error) {
 866         case EAGAIN:
 867             return;
 868
 869         case ENOBUFS:
 870             process_dpif_all_ports_changed(backer);
 871             break;
 872
 873         case 0:
 874             process_dpif_port_change(backer, devname);
 875             free(devname);
 876             break;
 877
 878         default:
 879             process_dpif_port_error(backer, error);
 880             break;
 881         }
 882     }
 883 }
 884
 885 static void
 886 process_dpif_all_ports_changed(struct dpif_backer *backer)
 887 {
 888     struct ofproto_dpif *ofproto;
 889     struct dpif_port dpif_port;
 890     struct dpif_port_dump dump;
 891     struct sset devnames;
 892     const char *devname;
 893
 894     sset_init(&devnames);
 895     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 896         if (ofproto->backer == backer) {
 897             struct ofport *ofport;
 898
 899             HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
 900                 sset_add(&devnames, netdev_get_name(ofport->netdev));
 901             }
 902         }
 903     }
 904     DPIF_PORT_FOR_EACH (&dpif_port, &dump, backer->dpif) {
 905         sset_add(&devnames, dpif_port.name);
 906     }
 907
 908     SSET_FOR_EACH (devname, &devnames) {
 909         process_dpif_port_change(backer, devname);
 910     }
 911     sset_destroy(&devnames);
 912 }
 913
 914 static void
 915 process_dpif_port_change(struct dpif_backer *backer, const char *devname)
 916 {
 917     struct ofproto_dpif *ofproto;
 918     struct dpif_port port;
 919
 920     /* Don't report on the datapath's device. */
 921     if (!strcmp(devname, dpif_base_name(backer->dpif))) {
 922         return;
 923     }
 924
 925     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
 926                    &all_ofproto_dpifs) {
 927         if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
 928             return;
 929         }
 930     }
 931
 932     ofproto = lookup_ofproto_dpif_by_port_name(devname);
 933     if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
 934         /* The port was removed.  If we know the datapath,
 935          * report it through poll_set().  If we don't, it may be
 936          * notifying us of a removal we initiated, so ignore it.
 937          * If there's a pending ENOBUFS, let it stand, since
 938          * everything will be reevaluated. */
 939         if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
 940             sset_add(&ofproto->port_poll_set, devname);
 941             ofproto->port_poll_errno = 0;
 942         }
 943     } else if (!ofproto) {
 944         /* The port was added, but we don't know with which
 945          * ofproto we should associate it.  Delete it. */
 946         dpif_port_del(backer->dpif, port.port_no);
 947     } else {
 948         struct ofport_dpif *ofport;
 949
 950         ofport = ofport_dpif_cast(shash_find_data(
 951                                       &ofproto->up.port_by_name, devname));
 952         if (ofport
 953             && ofport->odp_port != port.port_no
 954             && !odp_port_to_ofport(backer, port.port_no))
 955         {
 956             /* 'ofport''s datapath port number has changed from
 957              * 'ofport->odp_port' to 'port.port_no'.  Update our internal data
 958              * structures to match. */
 959             ovs_rwlock_wrlock(&backer->odp_to_ofport_lock);
 960             hmap_remove(&backer->odp_to_ofport_map, &ofport->odp_port_node);
 961             ofport->odp_port = port.port_no;
 962             hmap_insert(&backer->odp_to_ofport_map, &ofport->odp_port_node,
 963                         hash_odp_port(port.port_no));
 964             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
 965             backer->need_revalidate = REV_RECONFIGURE;
 966         }
 967     }
 968     dpif_port_destroy(&port);
 969 }
 970
 971 /* Propagate 'error' to all ofprotos based on 'backer'. */
 972 static void
 973 process_dpif_port_error(struct dpif_backer *backer, int error)
 974 {
 975     struct ofproto_dpif *ofproto;
 976
 977     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
 978         if (ofproto->backer == backer) {
 979             sset_clear(&ofproto->port_poll_set);
 980             ofproto->port_poll_errno = error;
 981         }
 982     }
 983 }
 984
 985 static int
 986 dpif_backer_run_fast(struct dpif_backer *backer)
 987 {
 988     handle_upcalls(backer);
 989
 990     return 0;
 991 }
 992
 993 static int
 994 type_run_fast(const char *type)
 995 {
 996     struct dpif_backer *backer;
 997
 998     backer = shash_find_data(&all_dpif_backers, type);
 999     if (!backer) {
1000         /* This is not necessarily a problem, since backers are only
1001          * created on demand. */
1002         return 0;
1003     }
1004
1005     return dpif_backer_run_fast(backer);
1006 }
1007
1008 static void
1009 run_fast_rl(void)
1010 {
1011     static long long int port_rl = LLONG_MIN;
1012
1013     if (time_msec() >= port_rl) {
1014         struct ofproto_dpif *ofproto;
1015
1016         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1017             run_fast(&ofproto->up);
1018         }
1019         port_rl = time_msec() + 200;
1020     }
1021 }
1022
1023 static void
1024 type_wait(const char *type)
1025 {
1026     struct dpif_backer *backer;
1027
1028     backer = shash_find_data(&all_dpif_backers, type);
1029     if (!backer) {
1030         /* This is not necessarily a problem, since backers are only
1031          * created on demand. */
1032         return;
1033     }
1034
1035     if (backer->governor) {
1036         governor_wait(backer->governor);
1037     }
1038
1039     timer_wait(&backer->next_expiration);
1040     dpif_wait(backer->dpif);
1041     udpif_wait(backer->udpif);
1042 }
1043 \f
1044 /* Basic life-cycle. */
1045
1046 static int add_internal_flows(struct ofproto_dpif *);
1047
1048 static struct ofproto *
1049 alloc(void)
1050 {
1051     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1052     return &ofproto->up;
1053 }
1054
1055 static void
1056 dealloc(struct ofproto *ofproto_)
1057 {
1058     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1059     free(ofproto);
1060 }
1061
1062 static void
1063 close_dpif_backer(struct dpif_backer *backer)
1064 {
1065     ovs_assert(backer->refcount > 0);
1066
1067     if (--backer->refcount) {
1068         return;
1069     }
1070
1071     drop_key_clear(backer);
1072     hmap_destroy(&backer->drop_keys);
1073
1074     udpif_destroy(backer->udpif);
1075
1076     simap_destroy(&backer->tnl_backers);
1077     ovs_rwlock_destroy(&backer->odp_to_ofport_lock);
1078     hmap_destroy(&backer->odp_to_ofport_map);
1079     shash_find_and_delete(&all_dpif_backers, backer->type);
1080     free(backer->type);
1081     dpif_close(backer->dpif);
1082
1083     ovs_assert(hmap_is_empty(&backer->subfacets));
1084     hmap_destroy(&backer->subfacets);
1085     governor_destroy(backer->governor);
1086
1087     free(backer);
1088 }
1089
1090 /* Datapath port slated for removal from datapath. */
1091 struct odp_garbage {
1092     struct list list_node;
1093     odp_port_t odp_port;
1094 };
1095
1096 static int
1097 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1098 {
1099     struct dpif_backer *backer;
1100     struct dpif_port_dump port_dump;
1101     struct dpif_port port;
1102     struct shash_node *node;
1103     struct list garbage_list;
1104     struct odp_garbage *garbage, *next;
1105     struct sset names;
1106     char *backer_name;
1107     const char *name;
1108     int error;
1109
1110     backer = shash_find_data(&all_dpif_backers, type);
1111     if (backer) {
1112         backer->refcount++;
1113         *backerp = backer;
1114         return 0;
1115     }
1116
1117     backer_name = xasprintf("ovs-%s", type);
1118
1119     /* Remove any existing datapaths, since we assume we're the only
1120      * userspace controlling the datapath. */
1121     sset_init(&names);
1122     dp_enumerate_names(type, &names);
1123     SSET_FOR_EACH(name, &names) {
1124         struct dpif *old_dpif;
1125
1126         /* Don't remove our backer if it exists. */
1127         if (!strcmp(name, backer_name)) {
1128             continue;
1129         }
1130
1131         if (dpif_open(name, type, &old_dpif)) {
1132             VLOG_WARN("couldn't open old datapath %s to remove it", name);
1133         } else {
1134             dpif_delete(old_dpif);
1135             dpif_close(old_dpif);
1136         }
1137     }
1138     sset_destroy(&names);
1139
1140     backer = xmalloc(sizeof *backer);
1141
1142     error = dpif_create_and_open(backer_name, type, &backer->dpif);
1143     free(backer_name);
1144     if (error) {
1145         VLOG_ERR("failed to open datapath of type %s: %s", type,
1146                  ovs_strerror(error));
1147         free(backer);
1148         return error;
1149     }
1150     backer->udpif = udpif_create(backer, backer->dpif);
1151
1152     backer->type = xstrdup(type);
1153     backer->governor = NULL;
1154     backer->refcount = 1;
1155     hmap_init(&backer->odp_to_ofport_map);
1156     ovs_rwlock_init(&backer->odp_to_ofport_lock);
1157     hmap_init(&backer->drop_keys);
1158     hmap_init(&backer->subfacets);
1159     timer_set_duration(&backer->next_expiration, 1000);
1160     backer->need_revalidate = 0;
1161     simap_init(&backer->tnl_backers);
1162     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
1163     *backerp = backer;
1164
1165     if (backer->recv_set_enable) {
1166         dpif_flow_flush(backer->dpif);
1167     }
1168
1169     /* Loop through the ports already on the datapath and remove any
1170      * that we don't need anymore. */
1171     list_init(&garbage_list);
1172     dpif_port_dump_start(&port_dump, backer->dpif);
1173     while (dpif_port_dump_next(&port_dump, &port)) {
1174         node = shash_find(&init_ofp_ports, port.name);
1175         if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1176             garbage = xmalloc(sizeof *garbage);
1177             garbage->odp_port = port.port_no;
1178             list_push_front(&garbage_list, &garbage->list_node);
1179         }
1180     }
1181     dpif_port_dump_done(&port_dump);
1182
1183     LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1184         dpif_port_del(backer->dpif, garbage->odp_port);
1185         list_remove(&garbage->list_node);
1186         free(garbage);
1187     }
1188
1189     shash_add(&all_dpif_backers, type, backer);
1190
1191     error = dpif_recv_set(backer->dpif, backer->recv_set_enable);
1192     if (error) {
1193         VLOG_ERR("failed to listen on datapath of type %s: %s",
1194                  type, ovs_strerror(error));
1195         close_dpif_backer(backer);
1196         return error;
1197     }
1198     udpif_recv_set(backer->udpif, n_handler_threads,
1199                    backer->recv_set_enable);
1200     backer->n_handler_threads = n_handler_threads;
1201
1202     backer->max_n_subfacet = 0;
1203     backer->created = time_msec();
1204     backer->avg_n_subfacet = 0;
1205     backer->avg_subfacet_life = 0;
1206
1207     return error;
1208 }
1209
1210 static int
1211 construct(struct ofproto *ofproto_)
1212 {
1213     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1214     struct shash_node *node, *next;
1215     uint32_t max_ports;
1216     int error;
1217
1218     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1219     if (error) {
1220         return error;
1221     }
1222
1223     max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1224     ofproto_init_max_ports(ofproto_, MIN(max_ports, ofp_to_u16(OFPP_MAX)));
1225
1226     ofproto->netflow = NULL;
1227     ofproto->sflow = NULL;
1228     ofproto->ipfix = NULL;
1229     ofproto->stp = NULL;
1230     hmap_init(&ofproto->bundles);
1231     ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1232     ofproto->mbridge = mbridge_create();
1233     ofproto->has_bonded_bundles = false;
1234     ovs_mutex_init(&ofproto->stats_mutex);
1235     ovs_mutex_init(&ofproto->vsp_mutex);
1236
1237     classifier_init(&ofproto->facets);
1238     ofproto->consistency_rl = LLONG_MIN;
1239
1240     guarded_list_init(&ofproto->pins);
1241
1242     ofproto_dpif_unixctl_init();
1243
1244     hmap_init(&ofproto->vlandev_map);
1245     hmap_init(&ofproto->realdev_vid_map);
1246
1247     sset_init(&ofproto->ports);
1248     sset_init(&ofproto->ghost_ports);
1249     sset_init(&ofproto->port_poll_set);
1250     ofproto->port_poll_errno = 0;
1251
1252     SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1253         struct iface_hint *iface_hint = node->data;
1254
1255         if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1256             /* Check if the datapath already has this port. */
1257             if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1258                 sset_add(&ofproto->ports, node->name);
1259             }
1260
1261             free(iface_hint->br_name);
1262             free(iface_hint->br_type);
1263             free(iface_hint);
1264             shash_delete(&init_ofp_ports, node);
1265         }
1266     }
1267
1268     hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1269                 hash_string(ofproto->up.name, 0));
1270     memset(&ofproto->stats, 0, sizeof ofproto->stats);
1271
1272     ofproto_init_tables(ofproto_, N_TABLES);
1273     error = add_internal_flows(ofproto);
1274     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1275
1276     ofproto->n_hit = 0;
1277     ofproto->n_missed = 0;
1278
1279     return error;
1280 }
1281
1282 static int
1283 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1284                   const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1285 {
1286     struct ofputil_flow_mod fm;
1287     int error;
1288
1289     match_init_catchall(&fm.match);
1290     fm.priority = 0;
1291     match_set_reg(&fm.match, 0, id);
1292     fm.new_cookie = htonll(0);
1293     fm.cookie = htonll(0);
1294     fm.cookie_mask = htonll(0);
1295     fm.modify_cookie = false;
1296     fm.table_id = TBL_INTERNAL;
1297     fm.command = OFPFC_ADD;
1298     fm.idle_timeout = 0;
1299     fm.hard_timeout = 0;
1300     fm.buffer_id = 0;
1301     fm.out_port = 0;
1302     fm.flags = 0;
1303     fm.ofpacts = ofpacts->data;
1304     fm.ofpacts_len = ofpacts->size;
1305
1306     error = ofproto_flow_mod(&ofproto->up, &fm);
1307     if (error) {
1308         VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1309                     id, ofperr_to_string(error));
1310         return error;
1311     }
1312
1313     if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL,
1314                                   rulep)) {
1315         rule_dpif_unref(*rulep);
1316     } else {
1317         NOT_REACHED();
1318     }
1319
1320     return 0;
1321 }
1322
1323 static int
1324 add_internal_flows(struct ofproto_dpif *ofproto)
1325 {
1326     struct ofpact_controller *controller;
1327     uint64_t ofpacts_stub[128 / 8];
1328     struct ofpbuf ofpacts;
1329     int error;
1330     int id;
1331
1332     ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1333     id = 1;
1334
1335     controller = ofpact_put_CONTROLLER(&ofpacts);
1336     controller->max_len = UINT16_MAX;
1337     controller->controller_id = 0;
1338     controller->reason = OFPR_NO_MATCH;
1339     ofpact_pad(&ofpacts);
1340
1341     error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1342     if (error) {
1343         return error;
1344     }
1345
1346     ofpbuf_clear(&ofpacts);
1347     error = add_internal_flow(ofproto, id++, &ofpacts,
1348                               &ofproto->no_packet_in_rule);
1349     if (error) {
1350         return error;
1351     }
1352
1353     error = add_internal_flow(ofproto, id++, &ofpacts,
1354                               &ofproto->drop_frags_rule);
1355     return error;
1356 }
1357
1358 static void
1359 destruct(struct ofproto *ofproto_)
1360 {
1361     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1362     struct rule_dpif *rule, *next_rule;
1363     struct ofputil_packet_in *pin, *next_pin;
1364     struct facet *facet, *next_facet;
1365     struct cls_cursor cursor;
1366     struct oftable *table;
1367     struct list pins;
1368
1369     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1370     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1371     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1372     CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) {
1373         facet_remove(facet);
1374     }
1375
1376     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1377     ovs_rwlock_wrlock(&xlate_rwlock);
1378     xlate_remove_ofproto(ofproto);
1379     ovs_rwlock_unlock(&xlate_rwlock);
1380
1381     /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a
1382      * use-after-free error. */
1383     udpif_revalidate(ofproto->backer->udpif);
1384
1385     hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1386
1387     OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1388         struct cls_cursor cursor;
1389
1390         ovs_rwlock_rdlock(&table->cls.rwlock);
1391         cls_cursor_init(&cursor, &table->cls, NULL);
1392         ovs_rwlock_unlock(&table->cls.rwlock);
1393         CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1394             ofproto_rule_delete(&ofproto->up, &rule->up);
1395         }
1396     }
1397
1398     guarded_list_pop_all(&ofproto->pins, &pins);
1399     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1400         list_remove(&pin->list_node);
1401         free(CONST_CAST(void *, pin->packet));
1402         free(pin);
1403     }
1404     guarded_list_destroy(&ofproto->pins);
1405
1406     mbridge_unref(ofproto->mbridge);
1407
1408     netflow_destroy(ofproto->netflow);
1409     dpif_sflow_unref(ofproto->sflow);
1410     hmap_destroy(&ofproto->bundles);
1411     mac_learning_unref(ofproto->ml);
1412
1413     classifier_destroy(&ofproto->facets);
1414
1415     hmap_destroy(&ofproto->vlandev_map);
1416     hmap_destroy(&ofproto->realdev_vid_map);
1417
1418     sset_destroy(&ofproto->ports);
1419     sset_destroy(&ofproto->ghost_ports);
1420     sset_destroy(&ofproto->port_poll_set);
1421
1422     ovs_mutex_destroy(&ofproto->stats_mutex);
1423     ovs_mutex_destroy(&ofproto->vsp_mutex);
1424
1425     close_dpif_backer(ofproto->backer);
1426 }
1427
1428 static int
1429 run_fast(struct ofproto *ofproto_)
1430 {
1431     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1432     struct ofputil_packet_in *pin, *next_pin;
1433     struct ofport_dpif *ofport;
1434     struct list pins;
1435
1436     /* Do not perform any periodic activity required by 'ofproto' while
1437      * waiting for flow restore to complete. */
1438     if (ofproto_get_flow_restore_wait()) {
1439         return 0;
1440     }
1441
1442     guarded_list_pop_all(&ofproto->pins, &pins);
1443     LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) {
1444         connmgr_send_packet_in(ofproto->up.connmgr, pin);
1445         list_remove(&pin->list_node);
1446         free(CONST_CAST(void *, pin->packet));
1447         free(pin);
1448     }
1449
1450     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1451         port_run_fast(ofport);
1452     }
1453
1454     return 0;
1455 }
1456
1457 static int
1458 run(struct ofproto *ofproto_)
1459 {
1460     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1461     struct ofport_dpif *ofport;
1462     struct ofbundle *bundle;
1463     int error;
1464
1465     if (mbridge_need_revalidate(ofproto->mbridge)) {
1466         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1467         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1468         mac_learning_flush(ofproto->ml);
1469         ovs_rwlock_unlock(&ofproto->ml->rwlock);
1470     }
1471
1472     /* Do not perform any periodic activity below required by 'ofproto' while
1473      * waiting for flow restore to complete. */
1474     if (ofproto_get_flow_restore_wait()) {
1475         return 0;
1476     }
1477
1478     error = run_fast(ofproto_);
1479     if (error) {
1480         return error;
1481     }
1482
1483     if (ofproto->netflow) {
1484         if (netflow_run(ofproto->netflow)) {
1485             send_netflow_active_timeouts(ofproto);
1486         }
1487     }
1488     if (ofproto->sflow) {
1489         dpif_sflow_run(ofproto->sflow);
1490     }
1491     if (ofproto->ipfix) {
1492         dpif_ipfix_run(ofproto->ipfix);
1493     }
1494
1495     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1496         port_run(ofport);
1497     }
1498     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1499         bundle_run(bundle);
1500     }
1501
1502     stp_run(ofproto);
1503     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
1504     if (mac_learning_run(ofproto->ml)) {
1505         ofproto->backer->need_revalidate = REV_MAC_LEARNING;
1506     }
1507     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1508
1509     /* Check the consistency of a random facet, to aid debugging. */
1510     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1511     if (time_msec() >= ofproto->consistency_rl
1512         && !classifier_is_empty(&ofproto->facets)
1513         && !ofproto->backer->need_revalidate) {
1514         struct cls_table *table;
1515         struct cls_rule *cr;
1516         struct facet *facet;
1517
1518         ofproto->consistency_rl = time_msec() + 250;
1519
1520         table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables),
1521                              struct cls_table, hmap_node);
1522         cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule,
1523                           hmap_node);
1524         facet = CONTAINER_OF(cr, struct facet, cr);
1525
1526         if (!facet_check_consistency(facet)) {
1527             ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1528         }
1529     }
1530     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1531
1532     return 0;
1533 }
1534
1535 static void
1536 wait(struct ofproto *ofproto_)
1537 {
1538     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1539     struct ofport_dpif *ofport;
1540     struct ofbundle *bundle;
1541
1542     if (ofproto_get_flow_restore_wait()) {
1543         return;
1544     }
1545
1546     if (ofproto->sflow) {
1547         dpif_sflow_wait(ofproto->sflow);
1548     }
1549     if (ofproto->ipfix) {
1550         dpif_ipfix_wait(ofproto->ipfix);
1551     }
1552     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1553         port_wait(ofport);
1554     }
1555     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1556         bundle_wait(bundle);
1557     }
1558     if (ofproto->netflow) {
1559         netflow_wait(ofproto->netflow);
1560     }
1561     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
1562     mac_learning_wait(ofproto->ml);
1563     ovs_rwlock_unlock(&ofproto->ml->rwlock);
1564     stp_wait(ofproto);
1565     if (ofproto->backer->need_revalidate) {
1566         /* Shouldn't happen, but if it does just go around again. */
1567         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1568         poll_immediate_wake();
1569     }
1570 }
1571
1572 static void
1573 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1574 {
1575     const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1576     struct cls_cursor cursor;
1577     size_t n_subfacets = 0;
1578     struct facet *facet;
1579
1580     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1581     simap_increase(usage, "facets", classifier_count(&ofproto->facets));
1582     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1583
1584     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
1585     cls_cursor_init(&cursor, &ofproto->facets, NULL);
1586     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
1587         n_subfacets += list_size(&facet->subfacets);
1588     }
1589     ovs_rwlock_unlock(&ofproto->facets.rwlock);
1590     simap_increase(usage, "subfacets", n_subfacets);
1591 }
1592
1593 static void
1594 flush(struct ofproto *ofproto_)
1595 {
1596     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1597     struct subfacet *subfacet, *next_subfacet;
1598     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1599     int n_batch;
1600
1601     n_batch = 0;
1602     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1603                         &ofproto->backer->subfacets) {
1604         if (subfacet->facet->ofproto != ofproto) {
1605             continue;
1606         }
1607
1608         if (subfacet->path != SF_NOT_INSTALLED) {
1609             batch[n_batch++] = subfacet;
1610             if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1611                 subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1612                 n_batch = 0;
1613             }
1614         } else {
1615             subfacet_destroy(subfacet);
1616         }
1617     }
1618
1619     if (n_batch > 0) {
1620         subfacet_destroy_batch(ofproto->backer, batch, n_batch);
1621     }
1622 }
1623
1624 static void
1625 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1626              bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1627 {
1628     *arp_match_ip = true;
1629     *actions = (OFPUTIL_A_OUTPUT |
1630                 OFPUTIL_A_SET_VLAN_VID |
1631                 OFPUTIL_A_SET_VLAN_PCP |
1632                 OFPUTIL_A_STRIP_VLAN |
1633                 OFPUTIL_A_SET_DL_SRC |
1634                 OFPUTIL_A_SET_DL_DST |
1635                 OFPUTIL_A_SET_NW_SRC |
1636                 OFPUTIL_A_SET_NW_DST |
1637                 OFPUTIL_A_SET_NW_TOS |
1638                 OFPUTIL_A_SET_TP_SRC |
1639                 OFPUTIL_A_SET_TP_DST |
1640                 OFPUTIL_A_ENQUEUE);
1641 }
1642
1643 static void
1644 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1645 {
1646     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1647     struct dpif_dp_stats s;
1648     uint64_t n_miss, n_no_pkt_in, n_bytes, n_dropped_frags;
1649     uint64_t n_lookup;
1650
1651     strcpy(ots->name, "classifier");
1652
1653     dpif_get_dp_stats(ofproto->backer->dpif, &s);
1654     rule_get_stats(&ofproto->miss_rule->up, &n_miss, &n_bytes);
1655     rule_get_stats(&ofproto->no_packet_in_rule->up, &n_no_pkt_in, &n_bytes);
1656     rule_get_stats(&ofproto->drop_frags_rule->up, &n_dropped_frags, &n_bytes);
1657
1658     n_lookup = s.n_hit + s.n_missed - n_dropped_frags;
1659     ots->lookup_count = htonll(n_lookup);
1660     ots->matched_count = htonll(n_lookup - n_miss - n_no_pkt_in);
1661 }
1662
1663 static struct ofport *
1664 port_alloc(void)
1665 {
1666     struct ofport_dpif *port = xmalloc(sizeof *port);
1667     return &port->up;
1668 }
1669
1670 static void
1671 port_dealloc(struct ofport *port_)
1672 {
1673     struct ofport_dpif *port = ofport_dpif_cast(port_);
1674     free(port);
1675 }
1676
1677 static int
1678 port_construct(struct ofport *port_)
1679 {
1680     struct ofport_dpif *port = ofport_dpif_cast(port_);
1681     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1682     const struct netdev *netdev = port->up.netdev;
1683     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1684     struct dpif_port dpif_port;
1685     int error;
1686
1687     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1688     port->bundle = NULL;
1689     port->cfm = NULL;
1690     port->bfd = NULL;
1691     port->may_enable = true;
1692     port->stp_port = NULL;
1693     port->stp_state = STP_DISABLED;
1694     port->is_tunnel = false;
1695     port->peer = NULL;
1696     port->qdscp = NULL;
1697     port->n_qdscp = 0;
1698     port->realdev_ofp_port = 0;
1699     port->vlandev_vid = 0;
1700     port->carrier_seq = netdev_get_carrier_resets(netdev);
1701
1702     if (netdev_vport_is_patch(netdev)) {
1703         /* By bailing out here, we don't submit the port to the sFlow module
1704          * to be considered for counter polling export.  This is correct
1705          * because the patch port represents an interface that sFlow considers
1706          * to be "internal" to the switch as a whole, and therefore not an
1707          * candidate for counter polling. */
1708         port->odp_port = ODPP_NONE;
1709         ofport_update_peer(port);
1710         return 0;
1711     }
1712
1713     error = dpif_port_query_by_name(ofproto->backer->dpif,
1714                                     netdev_vport_get_dpif_port(netdev, namebuf,
1715                                                                sizeof namebuf),
1716                                     &dpif_port);
1717     if (error) {
1718         return error;
1719     }
1720
1721     port->odp_port = dpif_port.port_no;
1722
1723     if (netdev_get_tunnel_config(netdev)) {
1724         tnl_port_add(port, port->up.netdev, port->odp_port);
1725         port->is_tunnel = true;
1726     } else {
1727         /* Sanity-check that a mapping doesn't already exist.  This
1728          * shouldn't happen for non-tunnel ports. */
1729         if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1730             VLOG_ERR("port %s already has an OpenFlow port number",
1731                      dpif_port.name);
1732             dpif_port_destroy(&dpif_port);
1733             return EBUSY;
1734         }
1735
1736         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1737         hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1738                     hash_odp_port(port->odp_port));
1739         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1740     }
1741     dpif_port_destroy(&dpif_port);
1742
1743     if (ofproto->sflow) {
1744         dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1745     }
1746
1747     return 0;
1748 }
1749
1750 static void
1751 port_destruct(struct ofport *port_)
1752 {
1753     struct ofport_dpif *port = ofport_dpif_cast(port_);
1754     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1755     const char *devname = netdev_get_name(port->up.netdev);
1756     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
1757     const char *dp_port_name;
1758
1759     ofproto->backer->need_revalidate = REV_RECONFIGURE;
1760     ovs_rwlock_wrlock(&xlate_rwlock);
1761     xlate_ofport_remove(port);
1762     ovs_rwlock_unlock(&xlate_rwlock);
1763
1764     dp_port_name = netdev_vport_get_dpif_port(port->up.netdev, namebuf,
1765                                               sizeof namebuf);
1766     if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1767         /* The underlying device is still there, so delete it.  This
1768          * happens when the ofproto is being destroyed, since the caller
1769          * assumes that removal of attached ports will happen as part of
1770          * destruction. */
1771         if (!port->is_tunnel) {
1772             dpif_port_del(ofproto->backer->dpif, port->odp_port);
1773         }
1774     }
1775
1776     if (port->peer) {
1777         port->peer->peer = NULL;
1778         port->peer = NULL;
1779     }
1780
1781     if (port->odp_port != ODPP_NONE && !port->is_tunnel) {
1782         ovs_rwlock_wrlock(&ofproto->backer->odp_to_ofport_lock);
1783         hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1784         ovs_rwlock_unlock(&ofproto->backer->odp_to_ofport_lock);
1785     }
1786
1787     tnl_port_del(port);
1788     sset_find_and_delete(&ofproto->ports, devname);
1789     sset_find_and_delete(&ofproto->ghost_ports, devname);
1790     bundle_remove(port_);
1791     set_cfm(port_, NULL);
1792     set_bfd(port_, NULL);
1793     if (ofproto->sflow) {
1794         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1795     }
1796
1797     free(port->qdscp);
1798 }
1799
1800 static void
1801 port_modified(struct ofport *port_)
1802 {
1803     struct ofport_dpif *port = ofport_dpif_cast(port_);
1804
1805     if (port->bundle && port->bundle->bond) {
1806         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1807     }
1808
1809     if (port->cfm) {
1810         cfm_set_netdev(port->cfm, port->up.netdev);
1811     }
1812
1813     if (port->bfd) {
1814         bfd_set_netdev(port->bfd, port->up.netdev);
1815     }
1816
1817     if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
1818                                                 port->odp_port)) {
1819         ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
1820             REV_RECONFIGURE;
1821     }
1822
1823     ofport_update_peer(port);
1824 }
1825
1826 static void
1827 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1828 {
1829     struct ofport_dpif *port = ofport_dpif_cast(port_);
1830     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1831     enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1832
1833     if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1834                    OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1835                    OFPUTIL_PC_NO_PACKET_IN)) {
1836         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1837
1838         if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1839             bundle_update(port->bundle);
1840         }
1841     }
1842 }
1843
1844 static int
1845 set_sflow(struct ofproto *ofproto_,
1846           const struct ofproto_sflow_options *sflow_options)
1847 {
1848     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1849     struct dpif_sflow *ds = ofproto->sflow;
1850
1851     if (sflow_options) {
1852         if (!ds) {
1853             struct ofport_dpif *ofport;
1854
1855             ds = ofproto->sflow = dpif_sflow_create();
1856             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1857                 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1858             }
1859             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1860         }
1861         dpif_sflow_set_options(ds, sflow_options);
1862     } else {
1863         if (ds) {
1864             dpif_sflow_unref(ds);
1865             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1866             ofproto->sflow = NULL;
1867         }
1868     }
1869     return 0;
1870 }
1871
1872 static int
1873 set_ipfix(
1874     struct ofproto *ofproto_,
1875     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1876     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1877     size_t n_flow_exporters_options)
1878 {
1879     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1880     struct dpif_ipfix *di = ofproto->ipfix;
1881     bool has_options = bridge_exporter_options || flow_exporters_options;
1882
1883     if (has_options && !di) {
1884         di = ofproto->ipfix = dpif_ipfix_create();
1885     }
1886
1887     if (di) {
1888         /* Call set_options in any case to cleanly flush the flow
1889          * caches in the last exporters that are to be destroyed. */
1890         dpif_ipfix_set_options(
1891             di, bridge_exporter_options, flow_exporters_options,
1892             n_flow_exporters_options);
1893
1894         if (!has_options) {
1895             dpif_ipfix_unref(di);
1896             ofproto->ipfix = NULL;
1897         }
1898     }
1899
1900     return 0;
1901 }
1902
1903 static int
1904 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1905 {
1906     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1907     int error;
1908
1909     if (!s) {
1910         error = 0;
1911     } else {
1912         if (!ofport->cfm) {
1913             struct ofproto_dpif *ofproto;
1914
1915             ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1916             ofproto->backer->need_revalidate = REV_RECONFIGURE;
1917             ofport->cfm = cfm_create(ofport->up.netdev);
1918         }
1919
1920         if (cfm_configure(ofport->cfm, s)) {
1921             return 0;
1922         }
1923
1924         error = EINVAL;
1925     }
1926     cfm_unref(ofport->cfm);
1927     ofport->cfm = NULL;
1928     return error;
1929 }
1930
1931 static bool
1932 get_cfm_status(const struct ofport *ofport_,
1933                struct ofproto_cfm_status *status)
1934 {
1935     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1936
1937     if (ofport->cfm) {
1938         status->faults = cfm_get_fault(ofport->cfm);
1939         status->remote_opstate = cfm_get_opup(ofport->cfm);
1940         status->health = cfm_get_health(ofport->cfm);
1941         cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1942         return true;
1943     } else {
1944         return false;
1945     }
1946 }
1947
1948 static int
1949 set_bfd(struct ofport *ofport_, const struct smap *cfg)
1950 {
1951     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
1952     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1953     struct bfd *old;
1954
1955     old = ofport->bfd;
1956     ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
1957                                 cfg, ofport->up.netdev);
1958     if (ofport->bfd != old) {
1959         ofproto->backer->need_revalidate = REV_RECONFIGURE;
1960     }
1961
1962     return 0;
1963 }
1964
1965 static int
1966 get_bfd_status(struct ofport *ofport_, struct smap *smap)
1967 {
1968     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1969
1970     if (ofport->bfd) {
1971         bfd_get_status(ofport->bfd, smap);
1972         return 0;
1973     } else {
1974         return ENOENT;
1975     }
1976 }
1977 \f
1978 /* Spanning Tree. */
1979
1980 static void
1981 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1982 {
1983     struct ofproto_dpif *ofproto = ofproto_;
1984     struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1985     struct ofport_dpif *ofport;
1986
1987     ofport = stp_port_get_aux(sp);
1988     if (!ofport) {
1989         VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1990                      ofproto->up.name, port_num);
1991     } else {
1992         struct eth_header *eth = pkt->l2;
1993
1994         netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1995         if (eth_addr_is_zero(eth->eth_src)) {
1996             VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1997                          "with unknown MAC", ofproto->up.name, port_num);
1998         } else {
1999             ofproto_dpif_send_packet(ofport, pkt);
2000         }
2001     }
2002     ofpbuf_delete(pkt);
2003 }
2004
2005 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2006 static int
2007 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2008 {
2009     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2010
2011     /* Only revalidate flows if the configuration changed. */
2012     if (!s != !ofproto->stp) {
2013         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2014     }
2015
2016     if (s) {
2017         if (!ofproto->stp) {
2018             ofproto->stp = stp_create(ofproto_->name, s->system_id,
2019                                       send_bpdu_cb, ofproto);
2020             ofproto->stp_last_tick = time_msec();
2021         }
2022
2023         stp_set_bridge_id(ofproto->stp, s->system_id);
2024         stp_set_bridge_priority(ofproto->stp, s->priority);
2025         stp_set_hello_time(ofproto->stp, s->hello_time);
2026         stp_set_max_age(ofproto->stp, s->max_age);
2027         stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2028     }  else {
2029         struct ofport *ofport;
2030
2031         HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2032             set_stp_port(ofport, NULL);
2033         }
2034
2035         stp_unref(ofproto->stp);
2036         ofproto->stp = NULL;
2037     }
2038
2039     return 0;
2040 }
2041
2042 static int
2043 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2044 {
2045     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2046
2047     if (ofproto->stp) {
2048         s->enabled = true;
2049         s->bridge_id = stp_get_bridge_id(ofproto->stp);
2050         s->designated_root = stp_get_designated_root(ofproto->stp);
2051         s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2052     } else {
2053         s->enabled = false;
2054     }
2055
2056     return 0;
2057 }
2058
2059 static void
2060 update_stp_port_state(struct ofport_dpif *ofport)
2061 {
2062     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2063     enum stp_state state;
2064
2065     /* Figure out new state. */
2066     state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2067                              : STP_DISABLED;
2068
2069     /* Update state. */
2070     if (ofport->stp_state != state) {
2071         enum ofputil_port_state of_state;
2072         bool fwd_change;
2073
2074         VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2075                     netdev_get_name(ofport->up.netdev),
2076                     stp_state_name(ofport->stp_state),
2077                     stp_state_name(state));
2078         if (stp_learn_in_state(ofport->stp_state)
2079                 != stp_learn_in_state(state)) {
2080             /* xxx Learning action flows should also be flushed. */
2081             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2082             mac_learning_flush(ofproto->ml);
2083             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2084         }
2085         fwd_change = stp_forward_in_state(ofport->stp_state)
2086                         != stp_forward_in_state(state);
2087
2088         ofproto->backer->need_revalidate = REV_STP;
2089         ofport->stp_state = state;
2090         ofport->stp_state_entered = time_msec();
2091
2092         if (fwd_change && ofport->bundle) {
2093             bundle_update(ofport->bundle);
2094         }
2095
2096         /* Update the STP state bits in the OpenFlow port description. */
2097         of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2098         of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2099                      : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2100                      : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2101                      : state == STP_BLOCKING ?  OFPUTIL_PS_STP_BLOCK
2102                      : 0);
2103         ofproto_port_set_state(&ofport->up, of_state);
2104     }
2105 }
2106
2107 /* Configures STP on 'ofport_' using the settings defined in 's'.  The
2108  * caller is responsible for assigning STP port numbers and ensuring
2109  * there are no duplicates. */
2110 static int
2111 set_stp_port(struct ofport *ofport_,
2112              const struct ofproto_port_stp_settings *s)
2113 {
2114     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2115     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2116     struct stp_port *sp = ofport->stp_port;
2117
2118     if (!s || !s->enable) {
2119         if (sp) {
2120             ofport->stp_port = NULL;
2121             stp_port_disable(sp);
2122             update_stp_port_state(ofport);
2123         }
2124         return 0;
2125     } else if (sp && stp_port_no(sp) != s->port_num
2126             && ofport == stp_port_get_aux(sp)) {
2127         /* The port-id changed, so disable the old one if it's not
2128          * already in use by another port. */
2129         stp_port_disable(sp);
2130     }
2131
2132     sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2133     stp_port_enable(sp);
2134
2135     stp_port_set_aux(sp, ofport);
2136     stp_port_set_priority(sp, s->priority);
2137     stp_port_set_path_cost(sp, s->path_cost);
2138
2139     update_stp_port_state(ofport);
2140
2141     return 0;
2142 }
2143
2144 static int
2145 get_stp_port_status(struct ofport *ofport_,
2146                     struct ofproto_port_stp_status *s)
2147 {
2148     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2149     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2150     struct stp_port *sp = ofport->stp_port;
2151
2152     if (!ofproto->stp || !sp) {
2153         s->enabled = false;
2154         return 0;
2155     }
2156
2157     s->enabled = true;
2158     s->port_id = stp_port_get_id(sp);
2159     s->state = stp_port_get_state(sp);
2160     s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2161     s->role = stp_port_get_role(sp);
2162     stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2163
2164     return 0;
2165 }
2166
2167 static void
2168 stp_run(struct ofproto_dpif *ofproto)
2169 {
2170     if (ofproto->stp) {
2171         long long int now = time_msec();
2172         long long int elapsed = now - ofproto->stp_last_tick;
2173         struct stp_port *sp;
2174
2175         if (elapsed > 0) {
2176             stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2177             ofproto->stp_last_tick = now;
2178         }
2179         while (stp_get_changed_port(ofproto->stp, &sp)) {
2180             struct ofport_dpif *ofport = stp_port_get_aux(sp);
2181
2182             if (ofport) {
2183                 update_stp_port_state(ofport);
2184             }
2185         }
2186
2187         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2188             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2189             mac_learning_flush(ofproto->ml);
2190             ovs_rwlock_unlock(&ofproto->ml->rwlock);
2191         }
2192     }
2193 }
2194
2195 static void
2196 stp_wait(struct ofproto_dpif *ofproto)
2197 {
2198     if (ofproto->stp) {
2199         poll_timer_wait(1000);
2200     }
2201 }
2202 \f
2203 static int
2204 set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
2205            size_t n_qdscp)
2206 {
2207     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2208     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2209
2210     if (ofport->n_qdscp != n_qdscp
2211         || (n_qdscp && memcmp(ofport->qdscp, qdscp,
2212                               n_qdscp * sizeof *qdscp))) {
2213         ofproto->backer->need_revalidate = REV_RECONFIGURE;
2214         free(ofport->qdscp);
2215         ofport->qdscp = n_qdscp
2216             ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
2217             : NULL;
2218         ofport->n_qdscp = n_qdscp;
2219     }
2220
2221     return 0;
2222 }
2223 \f
2224 /* Bundles. */
2225
2226 /* Expires all MAC learning entries associated with 'bundle' and forces its
2227  * ofproto to revalidate every flow.
2228  *
2229  * Normally MAC learning entries are removed only from the ofproto associated
2230  * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2231  * are removed from every ofproto.  When patch ports and SLB bonds are in use
2232  * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2233  * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2234  * with the host from which it migrated. */
2235 static void
2236 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2237 {
2238     struct ofproto_dpif *ofproto = bundle->ofproto;
2239     struct mac_learning *ml = ofproto->ml;
2240     struct mac_entry *mac, *next_mac;
2241
2242     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2243     ovs_rwlock_wrlock(&ml->rwlock);
2244     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2245         if (mac->port.p == bundle) {
2246             if (all_ofprotos) {
2247                 struct ofproto_dpif *o;
2248
2249                 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2250                     if (o != ofproto) {
2251                         struct mac_entry *e;
2252
2253                         ovs_rwlock_wrlock(&o->ml->rwlock);
2254                         e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
2255                         if (e) {
2256                             mac_learning_expire(o->ml, e);
2257                         }
2258                         ovs_rwlock_unlock(&o->ml->rwlock);
2259                     }
2260                 }
2261             }
2262
2263             mac_learning_expire(ml, mac);
2264         }
2265     }
2266     ovs_rwlock_unlock(&ml->rwlock);
2267 }
2268
2269 static struct ofbundle *
2270 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2271 {
2272     struct ofbundle *bundle;
2273
2274     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2275                              &ofproto->bundles) {
2276         if (bundle->aux == aux) {
2277             return bundle;
2278         }
2279     }
2280     return NULL;
2281 }
2282
2283 static void
2284 bundle_update(struct ofbundle *bundle)
2285 {
2286     struct ofport_dpif *port;
2287
2288     bundle->floodable = true;
2289     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2290         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2291             || !stp_forward_in_state(port->stp_state)) {
2292             bundle->floodable = false;
2293             break;
2294         }
2295     }
2296 }
2297
2298 static void
2299 bundle_del_port(struct ofport_dpif *port)
2300 {
2301     struct ofbundle *bundle = port->bundle;
2302
2303     bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2304
2305     list_remove(&port->bundle_node);
2306     port->bundle = NULL;
2307
2308     if (bundle->lacp) {
2309         lacp_slave_unregister(bundle->lacp, port);
2310     }
2311     if (bundle->bond) {
2312         bond_slave_unregister(bundle->bond, port);
2313     }
2314
2315     bundle_update(bundle);
2316 }
2317
2318 static bool
2319 bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port,
2320                 struct lacp_slave_settings *lacp)
2321 {
2322     struct ofport_dpif *port;
2323
2324     port = get_ofp_port(bundle->ofproto, ofp_port);
2325     if (!port) {
2326         return false;
2327     }
2328
2329     if (port->bundle != bundle) {
2330         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2331         if (port->bundle) {
2332             bundle_remove(&port->up);
2333         }
2334
2335         port->bundle = bundle;
2336         list_push_back(&bundle->ports, &port->bundle_node);
2337         if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2338             || !stp_forward_in_state(port->stp_state)) {
2339             bundle->floodable = false;
2340         }
2341     }
2342     if (lacp) {
2343         bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2344         lacp_slave_register(bundle->lacp, port, lacp);
2345     }
2346
2347     return true;
2348 }
2349
2350 static void
2351 bundle_destroy(struct ofbundle *bundle)
2352 {
2353     struct ofproto_dpif *ofproto;
2354     struct ofport_dpif *port, *next_port;
2355
2356     if (!bundle) {
2357         return;
2358     }
2359
2360     ofproto = bundle->ofproto;
2361     mbridge_unregister_bundle(ofproto->mbridge, bundle->aux);
2362
2363     ovs_rwlock_wrlock(&xlate_rwlock);
2364     xlate_bundle_remove(bundle);
2365     ovs_rwlock_unlock(&xlate_rwlock);
2366
2367     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2368         bundle_del_port(port);
2369     }
2370
2371     bundle_flush_macs(bundle, true);
2372     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2373     free(bundle->name);
2374     free(bundle->trunks);
2375     lacp_unref(bundle->lacp);
2376     bond_unref(bundle->bond);
2377     free(bundle);
2378 }
2379
2380 static int
2381 bundle_set(struct ofproto *ofproto_, void *aux,
2382            const struct ofproto_bundle_settings *s)
2383 {
2384     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2385     bool need_flush = false;
2386     struct ofport_dpif *port;
2387     struct ofbundle *bundle;
2388     unsigned long *trunks;
2389     int vlan;
2390     size_t i;
2391     bool ok;
2392
2393     if (!s) {
2394         bundle_destroy(bundle_lookup(ofproto, aux));
2395         return 0;
2396     }
2397
2398     ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2399     ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2400
2401     bundle = bundle_lookup(ofproto, aux);
2402     if (!bundle) {
2403         bundle = xmalloc(sizeof *bundle);
2404
2405         bundle->ofproto = ofproto;
2406         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2407                     hash_pointer(aux, 0));
2408         bundle->aux = aux;
2409         bundle->name = NULL;
2410
2411         list_init(&bundle->ports);
2412         bundle->vlan_mode = PORT_VLAN_TRUNK;
2413         bundle->vlan = -1;
2414         bundle->trunks = NULL;
2415         bundle->use_priority_tags = s->use_priority_tags;
2416         bundle->lacp = NULL;
2417         bundle->bond = NULL;
2418
2419         bundle->floodable = true;
2420         mbridge_register_bundle(ofproto->mbridge, bundle);
2421     }
2422
2423     if (!bundle->name || strcmp(s->name, bundle->name)) {
2424         free(bundle->name);
2425         bundle->name = xstrdup(s->name);
2426     }
2427
2428     /* LACP. */
2429     if (s->lacp) {
2430         if (!bundle->lacp) {
2431             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2432             bundle->lacp = lacp_create();
2433         }
2434         lacp_configure(bundle->lacp, s->lacp);
2435     } else {
2436         lacp_unref(bundle->lacp);
2437         bundle->lacp = NULL;
2438     }
2439
2440     /* Update set of ports. */
2441     ok = true;
2442     for (i = 0; i < s->n_slaves; i++) {
2443         if (!bundle_add_port(bundle, s->slaves[i],
2444                              s->lacp ? &s->lacp_slaves[i] : NULL)) {
2445             ok = false;
2446         }
2447     }
2448     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2449         struct ofport_dpif *next_port;
2450
2451         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2452             for (i = 0; i < s->n_slaves; i++) {
2453                 if (s->slaves[i] == port->up.ofp_port) {
2454                     goto found;
2455                 }
2456             }
2457
2458             bundle_del_port(port);
2459         found: ;
2460         }
2461     }
2462     ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2463
2464     if (list_is_empty(&bundle->ports)) {
2465         bundle_destroy(bundle);
2466         return EINVAL;
2467     }
2468
2469     /* Set VLAN tagging mode */
2470     if (s->vlan_mode != bundle->vlan_mode
2471         || s->use_priority_tags != bundle->use_priority_tags) {
2472         bundle->vlan_mode = s->vlan_mode;
2473         bundle->use_priority_tags = s->use_priority_tags;
2474         need_flush = true;
2475     }
2476
2477     /* Set VLAN tag. */
2478     vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2479             : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2480             : 0);
2481     if (vlan != bundle->vlan) {
2482         bundle->vlan = vlan;
2483         need_flush = true;
2484     }
2485
2486     /* Get trunked VLANs. */
2487     switch (s->vlan_mode) {
2488     case PORT_VLAN_ACCESS:
2489         trunks = NULL;
2490         break;
2491
2492     case PORT_VLAN_TRUNK:
2493         trunks = CONST_CAST(unsigned long *, s->trunks);
2494         break;
2495
2496     case PORT_VLAN_NATIVE_UNTAGGED:
2497     case PORT_VLAN_NATIVE_TAGGED:
2498         if (vlan != 0 && (!s->trunks
2499                           || !bitmap_is_set(s->trunks, vlan)
2500                           || bitmap_is_set(s->trunks, 0))) {
2501             /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2502             if (s->trunks) {
2503                 trunks = bitmap_clone(s->trunks, 4096);
2504             } else {
2505                 trunks = bitmap_allocate1(4096);
2506             }
2507             bitmap_set1(trunks, vlan);
2508             bitmap_set0(trunks, 0);
2509         } else {
2510             trunks = CONST_CAST(unsigned long *, s->trunks);
2511         }
2512         break;
2513
2514     default:
2515         NOT_REACHED();
2516     }
2517     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2518         free(bundle->trunks);
2519         if (trunks == s->trunks) {
2520             bundle->trunks = vlan_bitmap_clone(trunks);
2521         } else {
2522             bundle->trunks = trunks;
2523             trunks = NULL;
2524         }
2525         need_flush = true;
2526     }
2527     if (trunks != s->trunks) {
2528         free(trunks);
2529     }
2530
2531     /* Bonding. */
2532     if (!list_is_short(&bundle->ports)) {
2533         bundle->ofproto->has_bonded_bundles = true;
2534         if (bundle->bond) {
2535             if (bond_reconfigure(bundle->bond, s->bond)) {
2536                 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2537             }
2538         } else {
2539             bundle->bond = bond_create(s->bond);
2540             ofproto->backer->need_revalidate = REV_RECONFIGURE;
2541         }
2542
2543         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2544             bond_slave_register(bundle->bond, port, port->up.netdev);
2545         }
2546     } else {
2547         bond_unref(bundle->bond);
2548         bundle->bond = NULL;
2549     }
2550
2551     /* If we changed something that would affect MAC learning, un-learn
2552      * everything on this port and force flow revalidation. */
2553     if (need_flush) {
2554         bundle_flush_macs(bundle, false);
2555     }
2556
2557     return 0;
2558 }
2559
2560 static void
2561 bundle_remove(struct ofport *port_)
2562 {
2563     struct ofport_dpif *port = ofport_dpif_cast(port_);
2564     struct ofbundle *bundle = port->bundle;
2565
2566     if (bundle) {
2567         bundle_del_port(port);
2568         if (list_is_empty(&bundle->ports)) {
2569             bundle_destroy(bundle);
2570         } else if (list_is_short(&bundle->ports)) {
2571             bond_unref(bundle->bond);
2572             bundle->bond = NULL;
2573         }
2574     }
2575 }
2576
2577 static void
2578 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2579 {
2580     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2581     struct ofport_dpif *port = port_;
2582     uint8_t ea[ETH_ADDR_LEN];
2583     int error;
2584
2585     error = netdev_get_etheraddr(port->up.netdev, ea);
2586     if (!error) {
2587         struct ofpbuf packet;
2588         void *packet_pdu;
2589
2590         ofpbuf_init(&packet, 0);
2591         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2592                                  pdu_size);
2593         memcpy(packet_pdu, pdu, pdu_size);
2594
2595         ofproto_dpif_send_packet(port, &packet);
2596         ofpbuf_uninit(&packet);
2597     } else {
2598         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2599                     "%s (%s)", port->bundle->name,
2600                     netdev_get_name(port->up.netdev), ovs_strerror(error));
2601     }
2602 }
2603
2604 static void
2605 bundle_send_learning_packets(struct ofbundle *bundle)
2606 {
2607     struct ofproto_dpif *ofproto = bundle->ofproto;
2608     struct ofpbuf *learning_packet;
2609     int error, n_packets, n_errors;
2610     struct mac_entry *e;
2611     struct list packets;
2612
2613     list_init(&packets);
2614     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
2615     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2616         if (e->port.p != bundle) {
2617             void *port_void;
2618
2619             learning_packet = bond_compose_learning_packet(bundle->bond,
2620                                                            e->mac, e->vlan,
2621                                                            &port_void);
2622             learning_packet->private_p = port_void;
2623             list_push_back(&packets, &learning_packet->list_node);
2624         }
2625     }
2626     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2627
2628     error = n_packets = n_errors = 0;
2629     LIST_FOR_EACH (learning_packet, list_node, &packets) {
2630         int ret;
2631
2632         ret = ofproto_dpif_send_packet(learning_packet->private_p, learning_packet);
2633         if (ret) {
2634             error = ret;
2635             n_errors++;
2636         }
2637         n_packets++;
2638     }
2639     ofpbuf_list_delete(&packets);
2640
2641     if (n_errors) {
2642         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2643         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2644                      "packets, last error was: %s",
2645                      bundle->name, n_errors, n_packets, ovs_strerror(error));
2646     } else {
2647         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2648                  bundle->name, n_packets);
2649     }
2650 }
2651
2652 static void
2653 bundle_run(struct ofbundle *bundle)
2654 {
2655     if (bundle->lacp) {
2656         lacp_run(bundle->lacp, send_pdu_cb);
2657     }
2658     if (bundle->bond) {
2659         struct ofport_dpif *port;
2660
2661         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2662             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2663         }
2664
2665         if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
2666             bundle->ofproto->backer->need_revalidate = REV_BOND;
2667         }
2668
2669         if (bond_should_send_learning_packets(bundle->bond)) {
2670             bundle_send_learning_packets(bundle);
2671         }
2672     }
2673 }
2674
2675 static void
2676 bundle_wait(struct ofbundle *bundle)
2677 {
2678     if (bundle->lacp) {
2679         lacp_wait(bundle->lacp);
2680     }
2681     if (bundle->bond) {
2682         bond_wait(bundle->bond);
2683     }
2684 }
2685 \f
2686 /* Mirrors. */
2687
2688 static int
2689 mirror_set__(struct ofproto *ofproto_, void *aux,
2690              const struct ofproto_mirror_settings *s)
2691 {
2692     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2693     struct ofbundle **srcs, **dsts;
2694     int error;
2695     size_t i;
2696
2697     if (!s) {
2698         mirror_destroy(ofproto->mbridge, aux);
2699         return 0;
2700     }
2701
2702     srcs = xmalloc(s->n_srcs * sizeof *srcs);
2703     dsts = xmalloc(s->n_dsts * sizeof *dsts);
2704
2705     for (i = 0; i < s->n_srcs; i++) {
2706         srcs[i] = bundle_lookup(ofproto, s->srcs[i]);
2707     }
2708
2709     for (i = 0; i < s->n_dsts; i++) {
2710         dsts[i] = bundle_lookup(ofproto, s->dsts[i]);
2711     }
2712
2713     error = mirror_set(ofproto->mbridge, aux, s->name, srcs, s->n_srcs, dsts,
2714                        s->n_dsts, s->src_vlans,
2715                        bundle_lookup(ofproto, s->out_bundle), s->out_vlan);
2716     free(srcs);
2717     free(dsts);
2718     return error;
2719 }
2720
2721 static int
2722 mirror_get_stats__(struct ofproto *ofproto, void *aux,
2723                    uint64_t *packets, uint64_t *bytes)
2724 {
2725     push_all_stats();
2726     return mirror_get_stats(ofproto_dpif_cast(ofproto)->mbridge, aux, packets,
2727                             bytes);
2728 }
2729
2730 static int
2731 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2732 {
2733     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2734     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2735     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2736         mac_learning_flush(ofproto->ml);
2737     }
2738     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2739     return 0;
2740 }
2741
2742 static bool
2743 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2744 {
2745     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2746     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2747     return bundle && mirror_bundle_out(ofproto->mbridge, bundle) != 0;
2748 }
2749
2750 static void
2751 forward_bpdu_changed(struct ofproto *ofproto_)
2752 {
2753     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2754     ofproto->backer->need_revalidate = REV_RECONFIGURE;
2755 }
2756
2757 static void
2758 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2759                      size_t max_entries)
2760 {
2761     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2762     ovs_rwlock_wrlock(&ofproto->ml->rwlock);
2763     mac_learning_set_idle_time(ofproto->ml, idle_time);
2764     mac_learning_set_max_entries(ofproto->ml, max_entries);
2765     ovs_rwlock_unlock(&ofproto->ml->rwlock);
2766 }
2767 \f
2768 /* Ports. */
2769
2770 static struct ofport_dpif *
2771 get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
2772 {
2773     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2774     return ofport ? ofport_dpif_cast(ofport) : NULL;
2775 }
2776
2777 static struct ofport_dpif *
2778 get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
2779 {
2780     struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2781     return port && &ofproto->up == port->up.ofproto ? port : NULL;
2782 }
2783
2784 static void
2785 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2786                             struct ofproto_port *ofproto_port,
2787                             struct dpif_port *dpif_port)
2788 {
2789     ofproto_port->name = dpif_port->name;
2790     ofproto_port->type = dpif_port->type;
2791     ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2792 }
2793
2794 static void
2795 ofport_update_peer(struct ofport_dpif *ofport)
2796 {
2797     const struct ofproto_dpif *ofproto;
2798     struct dpif_backer *backer;
2799     char *peer_name;
2800
2801     if (!netdev_vport_is_patch(ofport->up.netdev)) {
2802         return;
2803     }
2804
2805     backer = ofproto_dpif_cast(ofport->up.ofproto)->backer;
2806     backer->need_revalidate = REV_RECONFIGURE;
2807
2808     if (ofport->peer) {
2809         ofport->peer->peer = NULL;
2810         ofport->peer = NULL;
2811     }
2812
2813     peer_name = netdev_vport_patch_peer(ofport->up.netdev);
2814     if (!peer_name) {
2815         return;
2816     }
2817
2818     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2819         struct ofport *peer_ofport;
2820         struct ofport_dpif *peer;
2821         char *peer_peer;
2822
2823         if (ofproto->backer != backer) {
2824             continue;
2825         }
2826
2827         peer_ofport = shash_find_data(&ofproto->up.port_by_name, peer_name);
2828         if (!peer_ofport) {
2829             continue;
2830         }
2831
2832         peer = ofport_dpif_cast(peer_ofport);
2833         peer_peer = netdev_vport_patch_peer(peer->up.netdev);
2834         if (peer_peer && !strcmp(netdev_get_name(ofport->up.netdev),
2835                                  peer_peer)) {
2836             ofport->peer = peer;
2837             ofport->peer->peer = ofport;
2838         }
2839         free(peer_peer);
2840
2841         break;
2842     }
2843     free(peer_name);
2844 }
2845
2846 static void
2847 port_run_fast(struct ofport_dpif *ofport)
2848 {
2849     if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2850         struct ofpbuf packet;
2851
2852         ofpbuf_init(&packet, 0);
2853         cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2854         ofproto_dpif_send_packet(ofport, &packet);
2855         ofpbuf_uninit(&packet);
2856     }
2857
2858     if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
2859         struct ofpbuf packet;
2860
2861         ofpbuf_init(&packet, 0);
2862         bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
2863         ofproto_dpif_send_packet(ofport, &packet);
2864         ofpbuf_uninit(&packet);
2865     }
2866 }
2867
2868 static void
2869 port_run(struct ofport_dpif *ofport)
2870 {
2871     long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2872     bool carrier_changed = carrier_seq != ofport->carrier_seq;
2873     bool enable = netdev_get_carrier(ofport->up.netdev);
2874     bool cfm_enable = false;
2875     bool bfd_enable = false;
2876
2877     ofport->carrier_seq = carrier_seq;
2878
2879     port_run_fast(ofport);
2880
2881     if (ofport->cfm) {
2882         int cfm_opup = cfm_get_opup(ofport->cfm);
2883
2884         cfm_run(ofport->cfm);
2885         cfm_enable = !cfm_get_fault(ofport->cfm);
2886
2887         if (cfm_opup >= 0) {
2888             cfm_enable = cfm_enable && cfm_opup;
2889         }
2890     }
2891
2892     if (ofport->bfd) {
2893         bfd_run(ofport->bfd);
2894         bfd_enable = bfd_forwarding(ofport->bfd);
2895     }
2896
2897     if (ofport->bfd || ofport->cfm) {
2898         enable = enable && (cfm_enable || bfd_enable);
2899     }
2900
2901     if (ofport->bundle) {
2902         enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2903         if (carrier_changed) {
2904             lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2905         }
2906     }
2907
2908     if (ofport->may_enable != enable) {
2909         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2910         ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
2911     }
2912
2913     ofport->may_enable = enable;
2914 }
2915
2916 static void
2917 port_wait(struct ofport_dpif *ofport)
2918 {
2919     if (ofport->cfm) {
2920         cfm_wait(ofport->cfm);
2921     }
2922
2923     if (ofport->bfd) {
2924         bfd_wait(ofport->bfd);
2925     }
2926 }
2927
2928 static int
2929 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
2930                    struct ofproto_port *ofproto_port)
2931 {
2932     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2933     struct dpif_port dpif_port;
2934     int error;
2935
2936     if (sset_contains(&ofproto->ghost_ports, devname)) {
2937         const char *type = netdev_get_type_from_name(devname);
2938
2939         /* We may be called before ofproto->up.port_by_name is populated with
2940          * the appropriate ofport.  For this reason, we must get the name and
2941          * type from the netdev layer directly. */
2942         if (type) {
2943             const struct ofport *ofport;
2944
2945             ofport = shash_find_data(&ofproto->up.port_by_name, devname);
2946             ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
2947             ofproto_port->name = xstrdup(devname);
2948             ofproto_port->type = xstrdup(type);
2949             return 0;
2950         }
2951         return ENODEV;
2952     }
2953
2954     if (!sset_contains(&ofproto->ports, devname)) {
2955         return ENODEV;
2956     }
2957     error = dpif_port_query_by_name(ofproto->backer->dpif,
2958                                     devname, &dpif_port);
2959     if (!error) {
2960         ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
2961     }
2962     return error;
2963 }
2964
2965 static int
2966 port_add(struct ofproto *ofproto_, struct netdev *netdev)
2967 {
2968     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2969     const char *devname = netdev_get_name(netdev);
2970     char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
2971     const char *dp_port_name;
2972
2973     if (netdev_vport_is_patch(netdev)) {
2974         sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
2975         return 0;
2976     }
2977
2978     dp_port_name = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
2979     if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
2980         odp_port_t port_no = ODPP_NONE;
2981         int error;
2982
2983         error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
2984         if (error) {
2985             return error;
2986         }
2987         if (netdev_get_tunnel_config(netdev)) {
2988             simap_put(&ofproto->backer->tnl_backers,
2989                       dp_port_name, odp_to_u32(port_no));
2990         }
2991     }
2992
2993     if (netdev_get_tunnel_config(netdev)) {
2994         sset_add(&ofproto->ghost_ports, devname);
2995     } else {
2996         sset_add(&ofproto->ports, devname);
2997     }
2998     return 0;
2999 }
3000
3001 static int
3002 port_del(struct ofproto *ofproto_, ofp_port_t ofp_port)
3003 {
3004     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3005     struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3006     int error = 0;
3007
3008     if (!ofport) {
3009         return 0;
3010     }
3011
3012     sset_find_and_delete(&ofproto->ghost_ports,
3013                          netdev_get_name(ofport->up.netdev));
3014     ofproto->backer->need_revalidate = REV_RECONFIGURE;
3015     if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) {
3016         error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3017         if (!error) {
3018             /* The caller is going to close ofport->up.netdev.  If this is a
3019              * bonded port, then the bond is using that netdev, so remove it
3020              * from the bond.  The client will need to reconfigure everything
3021              * after deleting ports, so then the slave will get re-added. */
3022             bundle_remove(&ofport->up);
3023         }
3024     }
3025     return error;
3026 }
3027
3028 static int
3029 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3030 {
3031     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3032     int error;
3033
3034     push_all_stats();
3035
3036     error = netdev_get_stats(ofport->up.netdev, stats);
3037
3038     if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3039         struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3040
3041         ovs_mutex_lock(&ofproto->stats_mutex);
3042         /* ofproto->stats.tx_packets represents packets that we created
3043          * internally and sent to some port (e.g. packets sent with
3044          * ofproto_dpif_send_packet()).  Account for them as if they had
3045          * come from OFPP_LOCAL and got forwarded. */
3046
3047         if (stats->rx_packets != UINT64_MAX) {
3048             stats->rx_packets += ofproto->stats.tx_packets;
3049         }
3050
3051         if (stats->rx_bytes != UINT64_MAX) {
3052             stats->rx_bytes += ofproto->stats.tx_bytes;
3053         }
3054
3055         /* ofproto->stats.rx_packets represents packets that were received on
3056          * some port and we processed internally and dropped (e.g. STP).
3057          * Account for them as if they had been forwarded to OFPP_LOCAL. */
3058
3059         if (stats->tx_packets != UINT64_MAX) {
3060             stats->tx_packets += ofproto->stats.rx_packets;
3061         }
3062
3063         if (stats->tx_bytes != UINT64_MAX) {
3064             stats->tx_bytes += ofproto->stats.rx_bytes;
3065         }
3066         ovs_mutex_unlock(&ofproto->stats_mutex);
3067     }
3068
3069     return error;
3070 }
3071
3072 struct port_dump_state {
3073     uint32_t bucket;
3074     uint32_t offset;
3075     bool ghost;
3076
3077     struct ofproto_port port;
3078     bool has_port;
3079 };
3080
3081 static int
3082 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3083 {
3084     *statep = xzalloc(sizeof(struct port_dump_state));
3085     return 0;
3086 }
3087
3088 static int
3089 port_dump_next(const struct ofproto *ofproto_, void *state_,
3090                struct ofproto_port *port)
3091 {
3092     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3093     struct port_dump_state *state = state_;
3094     const struct sset *sset;
3095     struct sset_node *node;
3096
3097     if (state->has_port) {
3098         ofproto_port_destroy(&state->port);
3099         state->has_port = false;
3100     }
3101     sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3102     while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3103         int error;
3104
3105         error = port_query_by_name(ofproto_, node->name, &state->port);
3106         if (!error) {
3107             *port = state->port;
3108             state->has_port = true;
3109             return 0;
3110         } else if (error != ENODEV) {
3111             return error;
3112         }
3113     }
3114
3115     if (!state->ghost) {
3116         state->ghost = true;
3117         state->bucket = 0;
3118         state->offset = 0;
3119         return port_dump_next(ofproto_, state_, port);
3120     }
3121
3122     return EOF;
3123 }
3124
3125 static int
3126 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3127 {
3128     struct port_dump_state *state = state_;
3129
3130     if (state->has_port) {
3131         ofproto_port_destroy(&state->port);
3132     }
3133     free(state);
3134     return 0;
3135 }
3136
3137 static int
3138 port_poll(const struct ofproto *ofproto_, char **devnamep)
3139 {
3140     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3141
3142     if (ofproto->port_poll_errno) {
3143         int error = ofproto->port_poll_errno;
3144         ofproto->port_poll_errno = 0;
3145         return error;
3146     }
3147
3148     if (sset_is_empty(&ofproto->port_poll_set)) {
3149         return EAGAIN;
3150     }
3151
3152     *devnamep = sset_pop(&ofproto->port_poll_set);
3153     return 0;
3154 }
3155
3156 static void
3157 port_poll_wait(const struct ofproto *ofproto_)
3158 {
3159     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3160     dpif_port_poll_wait(ofproto->backer->dpif);
3161 }
3162
3163 static int
3164 port_is_lacp_current(const struct ofport *ofport_)
3165 {
3166     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3167     return (ofport->bundle && ofport->bundle->lacp
3168             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3169             : -1);
3170 }
3171 \f
3172 /* Upcall handling. */
3173
3174 struct flow_miss_op {
3175     struct dpif_op dpif_op;
3176
3177     uint64_t slow_stub[128 / 8]; /* Buffer for compose_slow_path() */
3178     struct xlate_out xout;
3179     bool xout_garbage;           /* 'xout' needs to be uninitialized? */
3180
3181     struct ofpbuf mask;          /* Flow mask for "put" ops. */
3182     struct odputil_keybuf maskbuf;
3183
3184     /* If this is a "put" op, then a pointer to the subfacet that should
3185      * be marked as uninstalled if the operation fails. */
3186     struct subfacet *subfacet;
3187 };
3188
3189 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3190  * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace
3191  * and (usually) installing a datapath flow.  The answer is usually "yes" (a
3192  * return value of true).  However, for short flows the cost of bookkeeping is
3193  * much higher than the benefits, so when the datapath holds a large number of
3194  * flows we impose some heuristics to decide which flows are likely to be worth
3195  * tracking. */
3196 static bool
3197 flow_miss_should_make_facet(struct flow_miss *miss)
3198 {
3199     struct dpif_backer *backer = miss->ofproto->backer;
3200     uint32_t hash;
3201
3202     switch (flow_miss_model) {
3203     case OFPROTO_HANDLE_MISS_AUTO:
3204         break;
3205     case OFPROTO_HANDLE_MISS_WITH_FACETS:
3206         return true;
3207     case OFPROTO_HANDLE_MISS_WITHOUT_FACETS:
3208         return false;
3209     }
3210
3211     if (!backer->governor) {
3212         size_t n_subfacets;
3213
3214         n_subfacets = hmap_count(&backer->subfacets);
3215         if (n_subfacets * 2 <= flow_eviction_threshold) {
3216             return true;
3217         }
3218
3219         backer->governor = governor_create();
3220     }
3221
3222     hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0);
3223     return governor_should_install_flow(backer->governor, hash,
3224                                         miss->stats.n_packets);
3225 }
3226
3227 /* Handles 'miss', which matches 'facet'.  May add any required datapath
3228  * operations to 'ops', incrementing '*n_ops' for each new op.
3229  *
3230  * All of the packets in 'miss' are considered to have arrived at time
3231  * 'miss->stats.used'.  This is really important only for new facets: if we
3232  * just called time_msec() here, then the new subfacet or its packets could
3233  * look (occasionally) as though it was used some time after the facet was
3234  * used.  That can make a one-packet flow look like it has a nonzero duration,
3235  * which looks odd in e.g. NetFlow statistics. */
3236 static void
3237 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3238                             struct flow_miss_op *ops, size_t *n_ops)
3239 {
3240     enum subfacet_path want_path;
3241     struct subfacet *subfacet;
3242
3243     facet->packet_count += miss->stats.n_packets;
3244     facet->prev_packet_count += miss->stats.n_packets;
3245     facet->byte_count += miss->stats.n_bytes;
3246     facet->prev_byte_count += miss->stats.n_bytes;
3247
3248     want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
3249
3250     /* Don't install the flow if it's the result of the "userspace"
3251      * action for an already installed facet.  This can occur when a
3252      * datapath flow with wildcards has a "userspace" action and flows
3253      * sent to userspace result in a different subfacet, which will then
3254      * be rejected as overlapping by the datapath. */
3255     if (miss->upcall_type == DPIF_UC_ACTION
3256         && !list_is_empty(&facet->subfacets)) {
3257         return;
3258     }
3259
3260     subfacet = subfacet_create(facet, miss);
3261     if (subfacet->path != want_path) {
3262         struct flow_miss_op *op = &ops[(*n_ops)++];
3263         struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3264
3265         subfacet->path = want_path;
3266
3267         ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf);
3268         if (enable_megaflows) {
3269             odp_flow_key_from_mask(&op->mask, &facet->xout.wc.masks,
3270                                    &miss->flow, UINT32_MAX);
3271         }
3272
3273         op->xout_garbage = false;
3274         op->dpif_op.type = DPIF_OP_FLOW_PUT;
3275         op->subfacet = subfacet;
3276         put->flags = DPIF_FP_CREATE;
3277         put->key = miss->key;
3278         put->key_len = miss->key_len;
3279         put->mask = op->mask.data;
3280         put->mask_len = op->mask.size;
3281
3282         if (want_path == SF_FAST_PATH) {
3283             put->actions = facet->xout.odp_actions.data;
3284             put->actions_len = facet->xout.odp_actions.size;
3285         } else {
3286             compose_slow_path(facet->ofproto, &miss->flow, facet->xout.slow,
3287                               op->slow_stub, sizeof op->slow_stub,
3288                               &put->actions, &put->actions_len);
3289         }
3290         put->stats = NULL;
3291     }
3292 }
3293
3294 /* Handles flow miss 'miss'.  May add any required datapath operations
3295  * to 'ops', incrementing '*n_ops' for each new op. */
3296 static void
3297 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3298                  size_t *n_ops)
3299 {
3300     struct facet *facet;
3301
3302     miss->ofproto->n_missed += miss->stats.n_packets;
3303
3304     facet = facet_lookup_valid(miss->ofproto, &miss->flow);
3305     if (!facet) {
3306         /* There does not exist a bijection between 'struct flow' and datapath
3307          * flow keys with fitness ODP_FIT_TO_LITTLE.  This breaks a fundamental
3308          * assumption used throughout the facet and subfacet handling code.
3309          * Since we have to handle these misses in userspace anyway, we simply
3310          * skip facet creation, avoiding the problem altogether. */
3311         if (miss->key_fitness == ODP_FIT_TOO_LITTLE
3312             || !flow_miss_should_make_facet(miss)) {
3313             return;
3314         }
3315
3316         facet = facet_create(miss);
3317     }
3318     handle_flow_miss_with_facet(miss, facet, ops, n_ops);
3319 }
3320
3321 static struct drop_key *
3322 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3323                 size_t key_len)
3324 {
3325     struct drop_key *drop_key;
3326
3327     HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3328                              &backer->drop_keys) {
3329         if (drop_key->key_len == key_len
3330             && !memcmp(drop_key->key, key, key_len)) {
3331             return drop_key;
3332         }
3333     }
3334     return NULL;
3335 }
3336
3337 static void
3338 drop_key_clear(struct dpif_backer *backer)
3339 {
3340     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3341     struct drop_key *drop_key, *next;
3342
3343     HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3344         int error;
3345
3346         error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3347                               NULL);
3348         if (error && !VLOG_DROP_WARN(&rl)) {
3349             struct ds ds = DS_EMPTY_INITIALIZER;
3350             odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3351             VLOG_WARN("Failed to delete drop key (%s) (%s)",
3352                       ovs_strerror(error), ds_cstr(&ds));
3353             ds_destroy(&ds);
3354         }
3355
3356         hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3357         drop_key_destroy(drop_key);
3358     }
3359
3360     udpif_drop_key_clear(backer->udpif);
3361 }
3362
3363 static void
3364 handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb)
3365 {
3366     struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH];
3367     struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH];
3368     struct flow_miss *miss;
3369     size_t n_ops, i;
3370
3371     /* Process each element in the to-do list, constructing the set of
3372      * operations to batch. */
3373     n_ops = 0;
3374     HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) {
3375         handle_flow_miss(miss, flow_miss_ops, &n_ops);
3376     }
3377     ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3378
3379     /* Execute batch. */
3380     for (i = 0; i < n_ops; i++) {
3381         dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3382     }
3383     dpif_operate(backer->dpif, dpif_ops, n_ops);
3384
3385     for (i = 0; i < n_ops; i++) {
3386         if (dpif_ops[i]->error != 0
3387             && flow_miss_ops[i].dpif_op.type == DPIF_OP_FLOW_PUT
3388             && flow_miss_ops[i].subfacet) {
3389             struct subfacet *subfacet = flow_miss_ops[i].subfacet;
3390
3391             COVERAGE_INC(subfacet_install_fail);
3392
3393             /* Zero-out subfacet counters when installation failed, but
3394              * datapath reported hits.  This should not happen and
3395              * indicates a bug, since if the datapath flow exists, we
3396              * should not be attempting to create a new subfacet.  A
3397              * buggy datapath could trigger this, so just zero out the
3398              * counters and log an error. */
3399             if (subfacet->dp_packet_count || subfacet->dp_byte_count) {
3400                 VLOG_ERR_RL(&rl, "failed to install subfacet for which "
3401                             "datapath reported hits");
3402                 subfacet->dp_packet_count = subfacet->dp_byte_count = 0;
3403             }
3404
3405             subfacet->path = SF_NOT_INSTALLED;
3406         }
3407     }
3408 }
3409
3410 static void
3411 handle_upcalls(struct dpif_backer *backer)
3412 {
3413     struct flow_miss_batch *fmb;
3414     int n_processed;
3415
3416     for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) {
3417         struct drop_key *drop_key = drop_key_next(backer->udpif);
3418         if (!drop_key) {
3419             break;
3420         }
3421
3422         if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) {
3423             hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3424                         hash_bytes(drop_key->key, drop_key->key_len, 0));
3425             dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3426                           drop_key->key, drop_key->key_len,
3427                           NULL, 0, NULL, 0, NULL);
3428         } else {
3429             drop_key_destroy(drop_key);
3430         }
3431     }
3432
3433     fmb = flow_miss_batch_next(backer->udpif);
3434     if (fmb) {
3435         handle_flow_misses(backer, fmb);
3436         flow_miss_batch_destroy(fmb);
3437     }
3438 }
3439 \f
3440 /* Flow expiration. */
3441
3442 static int subfacet_max_idle(const struct dpif_backer *);
3443 static void update_stats(struct dpif_backer *);
3444 static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex);
3445 static void expire_subfacets(struct dpif_backer *, int dp_max_idle);
3446
3447 /* This function is called periodically by run().  Its job is to collect
3448  * updates for the flows that have been installed into the datapath, most
3449  * importantly when they last were used, and then use that information to
3450  * expire flows that have not been used recently.
3451  *
3452  * Returns the number of milliseconds after which it should be called again. */
3453 static int
3454 expire(struct dpif_backer *backer)
3455 {
3456     struct ofproto_dpif *ofproto;
3457     size_t n_subfacets;
3458     int max_idle;
3459
3460     /* Periodically clear out the drop keys in an effort to keep them
3461      * relatively few. */
3462     drop_key_clear(backer);
3463
3464     /* Update stats for each flow in the backer. */
3465     update_stats(backer);
3466
3467     n_subfacets = hmap_count(&backer->subfacets);
3468     if (n_subfacets) {
3469         struct subfacet *subfacet;
3470         long long int total, now;
3471
3472         total = 0;
3473         now = time_msec();
3474         HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3475             total += now - subfacet->created;
3476         }
3477         backer->avg_subfacet_life += total / n_subfacets;
3478     }
3479     backer->avg_subfacet_life /= 2;
3480
3481     backer->avg_n_subfacet += n_subfacets;
3482     backer->avg_n_subfacet /= 2;
3483
3484     backer->max_n_subfacet = MAX(backer->max_n_subfacet, n_subfacets);
3485
3486     max_idle = subfacet_max_idle(backer);
3487     expire_subfacets(backer, max_idle);
3488
3489     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3490         struct rule *rule, *next_rule;
3491
3492         if (ofproto->backer != backer) {
3493             continue;
3494         }
3495
3496         /* Expire OpenFlow flows whose idle_timeout or hard_timeout
3497          * has passed. */
3498         ovs_mutex_lock(&ofproto_mutex);
3499         LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
3500                             &ofproto->up.expirable) {
3501             rule_expire(rule_dpif_cast(rule));
3502         }
3503         ovs_mutex_unlock(&ofproto_mutex);
3504
3505         /* All outstanding data in existing flows has been accounted, so it's a
3506          * good time to do bond rebalancing. */
3507         if (ofproto->has_bonded_bundles) {
3508             struct ofbundle *bundle;
3509
3510             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3511                 if (bundle->bond) {
3512                     bond_rebalance(bundle->bond);
3513                 }
3514             }
3515         }
3516     }
3517
3518     return MIN(max_idle, 1000);
3519 }
3520
3521 /* Updates flow table statistics given that the datapath just reported 'stats'
3522  * as 'subfacet''s statistics. */
3523 static void
3524 update_subfacet_stats(struct subfacet *subfacet,
3525                       const struct dpif_flow_stats *stats)
3526 {
3527     struct facet *facet = subfacet->facet;
3528     struct dpif_flow_stats diff;
3529
3530     diff.tcp_flags = stats->tcp_flags;
3531     diff.used = stats->used;
3532
3533     if (stats->n_packets >= subfacet->dp_packet_count) {
3534         diff.n_packets = stats->n_packets - subfacet->dp_packet_count;
3535     } else {
3536         VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
3537         diff.n_packets = 0;
3538     }
3539
3540     if (stats->n_bytes >= subfacet->dp_byte_count) {
3541         diff.n_bytes = stats->n_bytes - subfacet->dp_byte_count;
3542     } else {
3543         VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
3544         diff.n_bytes = 0;
3545     }
3546
3547     facet->ofproto->n_hit += diff.n_packets;
3548     subfacet->dp_packet_count = stats->n_packets;
3549     subfacet->dp_byte_count = stats->n_bytes;
3550     subfacet_update_stats(subfacet, &diff);
3551
3552     if (facet->accounted_bytes < facet->byte_count) {
3553         facet_learn(facet);
3554         facet_account(facet);
3555         facet->accounted_bytes = facet->byte_count;
3556     }
3557 }
3558
3559 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
3560  * about, or a flow that shouldn't be installed but was anyway.  Delete it. */
3561 static void
3562 delete_unexpected_flow(struct dpif_backer *backer,
3563                        const struct nlattr *key, size_t key_len)
3564 {
3565     if (!VLOG_DROP_WARN(&rl)) {
3566         struct ds s;
3567
3568         ds_init(&s);
3569         odp_flow_key_format(key, key_len, &s);
3570         VLOG_WARN("unexpected flow: %s", ds_cstr(&s));
3571         ds_destroy(&s);
3572     }
3573
3574     COVERAGE_INC(facet_unexpected);
3575     dpif_flow_del(backer->dpif, key, key_len, NULL);
3576 }
3577
3578 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
3579  *
3580  * This function also pushes statistics updates to rules which each facet
3581  * resubmits into.  Generally these statistics will be accurate.  However, if a
3582  * facet changes the rule it resubmits into at some time in between
3583  * update_stats() runs, it is possible that statistics accrued to the
3584  * old rule will be incorrectly attributed to the new rule.  This could be
3585  * avoided by calling update_stats() whenever rules are created or
3586  * deleted.  However, the performance impact of making so many calls to the
3587  * datapath do not justify the benefit of having perfectly accurate statistics.
3588  *
3589  * In addition, this function maintains per ofproto flow hit counts. The patch
3590  * port is not treated specially. e.g. A packet ingress from br0 patched into
3591  * br1 will increase the hit count of br0 by 1, however, does not affect
3592  * the hit or miss counts of br1.
3593  */
3594 static void
3595 update_stats(struct dpif_backer *backer)
3596 {
3597     const struct dpif_flow_stats *stats;
3598     struct dpif_flow_dump dump;
3599     const struct nlattr *key, *mask;
3600     size_t key_len, mask_len;
3601
3602     dpif_flow_dump_start(&dump, backer->dpif);
3603     while (dpif_flow_dump_next(&dump, &key, &key_len,
3604                                &mask, &mask_len, NULL, NULL, &stats)) {
3605         struct subfacet *subfacet;
3606         uint32_t key_hash;
3607
3608         key_hash = odp_flow_key_hash(key, key_len);
3609         subfacet = subfacet_find(backer, key, key_len, key_hash);
3610         switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
3611         case SF_FAST_PATH:
3612             update_subfacet_stats(subfacet, stats);
3613             break;
3614
3615         case SF_SLOW_PATH:
3616             /* Stats are updated per-packet. */
3617             break;
3618
3619         case SF_NOT_INSTALLED:
3620         default:
3621             delete_unexpected_flow(backer, key, key_len);
3622             break;
3623         }
3624         run_fast_rl();
3625     }
3626     dpif_flow_dump_done(&dump);
3627 }
3628
3629 /* Calculates and returns the number of milliseconds of idle time after which
3630  * subfacets should expire from the datapath.  When a subfacet expires, we fold
3631  * its statistics into its facet, and when a facet's last subfacet expires, we
3632  * fold its statistic into its rule. */
3633 static int
3634 subfacet_max_idle(const struct dpif_backer *backer)
3635 {
3636     /*
3637      * Idle time histogram.
3638      *
3639      * Most of the time a switch has a relatively small number of subfacets.
3640      * When this is the case we might as well keep statistics for all of them
3641      * in userspace and to cache them in the kernel datapath for performance as
3642      * well.
3643      *
3644      * As the number of subfacets increases, the memory required to maintain
3645      * statistics about them in userspace and in the kernel becomes
3646      * significant.  However, with a large number of subfacets it is likely
3647      * that only a few of them are "heavy hitters" that consume a large amount
3648      * of bandwidth.  At this point, only heavy hitters are worth caching in
3649      * the kernel and maintaining in userspaces; other subfacets we can
3650      * discard.
3651      *
3652      * The technique used to compute the idle time is to build a histogram with
3653      * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each.  Each subfacet
3654      * that is installed in the kernel gets dropped in the appropriate bucket.
3655      * After the histogram has been built, we compute the cutoff so that only
3656      * the most-recently-used 1% of subfacets (but at least
3657      * flow_eviction_threshold flows) are kept cached.  At least
3658      * the most-recently-used bucket of subfacets is kept, so actually an
3659      * arbitrary number of subfacets can be kept in any given expiration run
3660      * (though the next run will delete most of those unless they receive
3661      * additional data).
3662      *
3663      * This requires a second pass through the subfacets, in addition to the
3664      * pass made by update_stats(), because the former function never looks at
3665      * uninstallable subfacets.
3666      */
3667     enum { BUCKET_WIDTH = 100 };
3668     enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
3669     int buckets[N_BUCKETS] = { 0 };
3670     int total, subtotal, bucket;
3671     struct subfacet *subfacet;
3672     long long int now;
3673     int i;
3674
3675     total = hmap_count(&backer->subfacets);
3676     if (total <= flow_eviction_threshold) {
3677         return N_BUCKETS * BUCKET_WIDTH;
3678     }
3679
3680     /* Build histogram. */
3681     now = time_msec();
3682     HMAP_FOR_EACH (subfacet, hmap_node, &backer->subfacets) {
3683         long long int idle = now - subfacet->used;
3684         int bucket = (idle <= 0 ? 0
3685                       : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
3686                       : (unsigned int) idle / BUCKET_WIDTH);
3687         buckets[bucket]++;
3688     }
3689
3690     /* Find the first bucket whose flows should be expired. */
3691     subtotal = bucket = 0;
3692     do {
3693         subtotal += buckets[bucket++];
3694     } while (bucket < N_BUCKETS &&
3695              subtotal < MAX(flow_eviction_threshold, total / 100));
3696
3697     if (VLOG_IS_DBG_ENABLED()) {
3698         struct ds s;
3699
3700         ds_init(&s);
3701         ds_put_cstr(&s, "keep");
3702         for (i = 0; i < N_BUCKETS; i++) {
3703             if (i == bucket) {
3704                 ds_put_cstr(&s, ", drop");
3705             }
3706             if (buckets[i]) {
3707                 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
3708             }
3709         }
3710         VLOG_INFO("%s (msec:count)", ds_cstr(&s));
3711         ds_destroy(&s);
3712     }
3713
3714     return bucket * BUCKET_WIDTH;
3715 }
3716
3717 static void
3718 expire_subfacets(struct dpif_backer *backer, int dp_max_idle)
3719 {
3720     /* Cutoff time for most flows. */
3721     long long int normal_cutoff = time_msec() - dp_max_idle;
3722
3723     /* We really want to keep flows for special protocols around, so use a more
3724      * conservative cutoff. */
3725     long long int special_cutoff = time_msec() - 10000;
3726
3727     struct subfacet *subfacet, *next_subfacet;
3728     struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
3729     int n_batch;
3730
3731     n_batch = 0;
3732     HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
3733                         &backer->subfacets) {
3734         long long int cutoff;
3735
3736         cutoff = (subfacet->facet->xout.slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP
3737                                                 | SLOW_STP)
3738                   ? special_cutoff
3739                   : normal_cutoff);
3740         if (subfacet->used < cutoff) {
3741             if (subfacet->path != SF_NOT_INSTALLED) {
3742                 batch[n_batch++] = subfacet;
3743                 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
3744                     subfacet_destroy_batch(backer, batch, n_batch);
3745                     n_batch = 0;
3746                 }
3747             } else {
3748                 subfacet_destroy(subfacet);
3749             }
3750         }
3751     }
3752
3753     if (n_batch > 0) {
3754         subfacet_destroy_batch(backer, batch, n_batch);
3755     }
3756 }
3757
3758 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3759  * then delete it entirely. */
3760 static void
3761 rule_expire(struct rule_dpif *rule)
3762     OVS_REQUIRES(ofproto_mutex)
3763 {
3764     uint16_t idle_timeout, hard_timeout;
3765     long long int now = time_msec();
3766     int reason;
3767
3768     ovs_assert(!rule->up.pending);
3769
3770     /* Has 'rule' expired? */
3771     ovs_mutex_lock(&rule->up.mutex);
3772     hard_timeout = rule->up.hard_timeout;
3773     idle_timeout = rule->up.idle_timeout;
3774     if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) {
3775         reason = OFPRR_HARD_TIMEOUT;
3776     } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) {
3777         reason = OFPRR_IDLE_TIMEOUT;
3778     } else {
3779         reason = -1;
3780     }
3781     ovs_mutex_unlock(&rule->up.mutex);
3782
3783     if (reason >= 0) {
3784         COVERAGE_INC(ofproto_dpif_expired);
3785         ofproto_rule_expire(&rule->up, reason);
3786     }
3787 }
3788 \f
3789 /* Facets. */
3790
3791 /* Creates and returns a new facet based on 'miss'.
3792  *
3793  * The caller must already have determined that no facet with an identical
3794  * 'miss->flow' exists in 'miss->ofproto'.
3795  *
3796  * 'rule' and 'xout' must have been created based on 'miss'.
3797  *
3798  * 'facet'' statistics are initialized based on 'stats'.
3799  *
3800  * The facet will initially have no subfacets.  The caller should create (at
3801  * least) one subfacet with subfacet_create(). */
3802 static struct facet *
3803 facet_create(const struct flow_miss *miss)
3804 {
3805     struct ofproto_dpif *ofproto = miss->ofproto;
3806     struct facet *facet;
3807     struct match match;
3808
3809     COVERAGE_INC(facet_create);
3810     facet = xzalloc(sizeof *facet);
3811     facet->ofproto = miss->ofproto;
3812     facet->used = miss->stats.used;
3813     facet->flow = miss->flow;
3814     facet->learn_rl = time_msec() + 500;
3815
3816     list_init(&facet->subfacets);
3817     netflow_flow_init(&facet->nf_flow);
3818     netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
3819
3820     xlate_out_copy(&facet->xout, &miss->xout);
3821
3822     match_init(&match, &facet->flow, &facet->xout.wc);
3823     cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY);
3824     ovs_rwlock_wrlock(&ofproto->facets.rwlock);
3825     classifier_insert(&ofproto->facets, &facet->cr);
3826     ovs_rwlock_unlock(&ofproto->facets.rwlock);
3827
3828     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
3829     return facet;
3830 }
3831
3832 static void
3833 facet_free(struct facet *facet)
3834 {
3835     if (facet) {
3836         xlate_out_uninit(&facet->xout);
3837         free(facet);
3838     }
3839 }
3840
3841 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
3842  * 'packet', which arrived on 'in_port'. */
3843 static bool
3844 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
3845                     const struct nlattr *odp_actions, size_t actions_len,
3846                     struct ofpbuf *packet)
3847 {
3848     struct odputil_keybuf keybuf;
3849     struct ofpbuf key;
3850     int error;
3851
3852     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
3853     odp_flow_key_from_flow(&key, flow,
3854                            ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port));
3855
3856     error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
3857                          odp_actions, actions_len, packet);
3858     return !error;
3859 }
3860
3861 /* Remove 'facet' from its ofproto and free up the associated memory:
3862  *
3863  *   - If 'facet' was installed in the datapath, uninstalls it and updates its
3864  *     rule's statistics, via subfacet_uninstall().
3865  *
3866  *   - Removes 'facet' from its rule and from ofproto->facets.
3867  */
3868 static void
3869 facet_remove(struct facet *facet)
3870 {
3871     struct subfacet *subfacet, *next_subfacet;
3872
3873     COVERAGE_INC(facet_remove);
3874     ovs_assert(!list_is_empty(&facet->subfacets));
3875
3876     /* First uninstall all of the subfacets to get final statistics. */
3877     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
3878         subfacet_uninstall(subfacet);
3879     }
3880
3881     /* Flush the final stats to the rule.
3882      *
3883      * This might require us to have at least one subfacet around so that we
3884      * can use its actions for accounting in facet_account(), which is why we
3885      * have uninstalled but not yet destroyed the subfacets. */
3886     facet_flush_stats(facet);
3887
3888     /* Now we're really all done so destroy everything. */
3889     LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
3890                         &facet->subfacets) {
3891         subfacet_destroy__(subfacet);
3892     }
3893     ovs_rwlock_wrlock(&facet->ofproto->facets.rwlock);
3894     classifier_remove(&facet->ofproto->facets, &facet->cr);
3895     ovs_rwlock_unlock(&facet->ofproto->facets.rwlock);
3896     cls_rule_destroy(&facet->cr);
3897     facet_free(facet);
3898 }
3899
3900 /* Feed information from 'facet' back into the learning table to keep it in
3901  * sync with what is actually flowing through the datapath. */
3902 static void
3903 facet_learn(struct facet *facet)
3904 {
3905     long long int now = time_msec();
3906
3907     if (!facet->xout.has_fin_timeout && now < facet->learn_rl) {
3908         return;
3909     }
3910
3911     facet->learn_rl = now + 500;
3912
3913     if (!facet->xout.has_learn
3914         && !facet->xout.has_normal
3915         && (!facet->xout.has_fin_timeout
3916             || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
3917         return;
3918     }
3919
3920     facet_push_stats(facet, true);
3921 }
3922
3923 static void
3924 facet_account(struct facet *facet)
3925 {
3926     const struct nlattr *a;
3927     unsigned int left;
3928     ovs_be16 vlan_tci;
3929     uint64_t n_bytes;
3930
3931     if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) {
3932         return;
3933     }
3934     n_bytes = facet->byte_count - facet->accounted_bytes;
3935
3936     /* This loop feeds byte counters to bond_account() for rebalancing to use
3937      * as a basis.  We also need to track the actual VLAN on which the packet
3938      * is going to be sent to ensure that it matches the one passed to
3939      * bond_choose_output_slave().  (Otherwise, we will account to the wrong
3940      * hash bucket.)
3941      *
3942      * We use the actions from an arbitrary subfacet because they should all
3943      * be equally valid for our purpose. */
3944     vlan_tci = facet->flow.vlan_tci;
3945     NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data,
3946                              facet->xout.odp_actions.size) {
3947         const struct ovs_action_push_vlan *vlan;
3948         struct ofport_dpif *port;
3949
3950         switch (nl_attr_type(a)) {
3951         case OVS_ACTION_ATTR_OUTPUT:
3952             port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a));
3953             if (port && port->bundle && port->bundle->bond) {
3954                 bond_account(port->bundle->bond, &facet->flow,
3955                              vlan_tci_to_vid(vlan_tci), n_bytes);
3956             }
3957             break;
3958
3959         case OVS_ACTION_ATTR_POP_VLAN:
3960             vlan_tci = htons(0);
3961             break;
3962
3963         case OVS_ACTION_ATTR_PUSH_VLAN:
3964             vlan = nl_attr_get(a);
3965             vlan_tci = vlan->vlan_tci;
3966             break;
3967         }
3968     }
3969 }
3970
3971 /* Returns true if the only action for 'facet' is to send to the controller.
3972  * (We don't report NetFlow expiration messages for such facets because they
3973  * are just part of the control logic for the network, not real traffic). */
3974 static bool
3975 facet_is_controller_flow(struct facet *facet)
3976 {
3977     if (facet) {
3978         struct ofproto_dpif *ofproto = facet->ofproto;
3979         const struct ofpact *ofpacts;
3980         struct rule_actions *actions;
3981         struct rule_dpif *rule;
3982         size_t ofpacts_len;
3983         bool is_controller;
3984
3985         rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule);
3986         actions = rule_dpif_get_actions(rule);
3987         rule_dpif_unref(rule);
3988
3989         ofpacts_len = actions->ofpacts_len;
3990         ofpacts = actions->ofpacts;
3991         is_controller = ofpacts_len > 0
3992             && ofpacts->type == OFPACT_CONTROLLER
3993             && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len);
3994         rule_actions_unref(actions);
3995
3996         return is_controller;
3997     }
3998     return false;
3999 }
4000
4001 /* Folds all of 'facet''s statistics into its rule.  Also updates the
4002  * accounting ofhook and emits a NetFlow expiration if appropriate.  All of
4003  * 'facet''s statistics in the datapath should have been zeroed and folded into
4004  * its packet and byte counts before this function is called. */
4005 static void
4006 facet_flush_stats(struct facet *facet)
4007 {
4008     struct ofproto_dpif *ofproto = facet->ofproto;
4009     struct subfacet *subfacet;
4010
4011     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4012         ovs_assert(!subfacet->dp_byte_count);
4013         ovs_assert(!subfacet->dp_packet_count);
4014     }
4015
4016     facet_push_stats(facet, false);
4017     if (facet->accounted_bytes < facet->byte_count) {
4018         facet_account(facet);
4019         facet->accounted_bytes = facet->byte_count;
4020     }
4021
4022     if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4023         struct ofexpired expired;
4024         expired.flow = facet->flow;
4025         expired.packet_count = facet->packet_count;
4026         expired.byte_count = facet->byte_count;
4027         expired.used = facet->used;
4028         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4029     }
4030
4031     /* Reset counters to prevent double counting if 'facet' ever gets
4032      * reinstalled. */
4033     facet_reset_counters(facet);
4034
4035     netflow_flow_clear(&facet->nf_flow);
4036     facet->tcp_flags = 0;
4037 }
4038
4039 /* Searches 'ofproto''s table of facets for one which would be responsible for
4040  * 'flow'.  Returns it if found, otherwise a null pointer.
4041  *
4042  * The returned facet might need revalidation; use facet_lookup_valid()
4043  * instead if that is important. */
4044 static struct facet *
4045 facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
4046 {
4047     struct cls_rule *cr;
4048
4049     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4050     cr = classifier_lookup(&ofproto->facets, flow, NULL);
4051     ovs_rwlock_unlock(&ofproto->facets.rwlock);
4052     return cr ? CONTAINER_OF(cr, struct facet, cr) : NULL;
4053 }
4054
4055 /* Searches 'ofproto''s table of facets for one capable that covers
4056  * 'flow'.  Returns it if found, otherwise a null pointer.
4057  *
4058  * The returned facet is guaranteed to be valid. */
4059 static struct facet *
4060 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
4061 {
4062     struct facet *facet;
4063
4064     facet = facet_find(ofproto, flow);
4065     if (facet
4066         && ofproto->backer->need_revalidate
4067         && !facet_revalidate(facet)) {
4068         return NULL;
4069     }
4070
4071     return facet;
4072 }
4073
4074 static bool
4075 facet_check_consistency(struct facet *facet)
4076 {
4077     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4078
4079     struct xlate_out xout;
4080     struct xlate_in xin;
4081
4082     struct rule_dpif *rule;
4083     bool ok;
4084
4085     /* Check the datapath actions for consistency. */
4086     rule_dpif_lookup(facet->ofproto, &facet->flow, NULL, &rule);
4087     xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL);
4088     xlate_actions(&xin, &xout);
4089     rule_dpif_unref(rule);
4090
4091     ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)
4092         && facet->xout.slow == xout.slow;
4093     if (!ok && !VLOG_DROP_WARN(&rl)) {
4094         struct ds s = DS_EMPTY_INITIALIZER;
4095
4096         flow_format(&s, &facet->flow);
4097         ds_put_cstr(&s, ": inconsistency in facet");
4098
4099         if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4100             ds_put_cstr(&s, " (actions were: ");
4101             format_odp_actions(&s, facet->xout.odp_actions.data,
4102                                facet->xout.odp_actions.size);
4103             ds_put_cstr(&s, ") (correct actions: ");
4104             format_odp_actions(&s, xout.odp_actions.data,
4105                                xout.odp_actions.size);
4106             ds_put_char(&s, ')');
4107         }
4108
4109         if (facet->xout.slow != xout.slow) {
4110             ds_put_format(&s, " slow path incorrect. should be %d", xout.slow);
4111         }
4112
4113         ds_destroy(&s);
4114     }
4115     xlate_out_uninit(&xout);
4116
4117     return ok;
4118 }
4119
4120 /* Re-searches the classifier for 'facet':
4121  *
4122  *   - If the rule found is different from 'facet''s current rule, moves
4123  *     'facet' to the new rule and recompiles its actions.
4124  *
4125  *   - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4126  *     where it is and recompiles its actions anyway.
4127  *
4128  *   - If any of 'facet''s subfacets correspond to a new flow according to
4129  *     xlate_receive(), 'facet' is removed.
4130  *
4131  *   Returns true if 'facet' is still valid.  False if 'facet' was removed. */
4132 static bool
4133 facet_revalidate(struct facet *facet)
4134 {
4135     struct ofproto_dpif *ofproto = facet->ofproto;
4136     struct rule_dpif *new_rule;
4137     struct subfacet *subfacet;
4138     struct flow_wildcards wc;
4139     struct xlate_out xout;
4140     struct xlate_in xin;
4141
4142     COVERAGE_INC(facet_revalidate);
4143
4144     /* Check that child subfacets still correspond to this facet.  Tunnel
4145      * configuration changes could cause a subfacet's OpenFlow in_port to
4146      * change. */
4147     LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4148         struct ofproto_dpif *recv_ofproto;
4149         struct flow recv_flow;
4150         int error;
4151
4152         error = xlate_receive(ofproto->backer, NULL, subfacet->key,
4153                               subfacet->key_len, &recv_flow, NULL,
4154                               &recv_ofproto, NULL);
4155         if (error
4156             || recv_ofproto != ofproto
4157             || facet != facet_find(ofproto, &recv_flow)) {
4158             facet_remove(facet);
4159             return false;
4160         }
4161     }
4162
4163     flow_wildcards_init_catchall(&wc);
4164     rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule);
4165
4166     /* Calculate new datapath actions.
4167      *
4168      * We do not modify any 'facet' state yet, because we might need to, e.g.,
4169      * emit a NetFlow expiration and, if so, we need to have the old state
4170      * around to properly compose it. */
4171     xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL);
4172     xlate_actions(&xin, &xout);
4173     flow_wildcards_or(&xout.wc, &xout.wc, &wc);
4174
4175     /* A facet's slow path reason should only change under dramatic
4176      * circumstances.  Rather than try to update everything, it's simpler to
4177      * remove the facet and start over.
4178      *
4179      * More importantly, if a facet's wildcards change, it will be relatively
4180      * difficult to figure out if its subfacets still belong to it, and if not
4181      * which facet they may belong to.  Again, to avoid the complexity, we
4182      * simply give up instead. */
4183     if (facet->xout.slow != xout.slow
4184         || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) {
4185         facet_remove(facet);
4186         xlate_out_uninit(&xout);
4187         rule_dpif_unref(new_rule);
4188         return false;
4189     }
4190
4191     if (!ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions)) {
4192         LIST_FOR_EACH(subfacet, list_node, &facet->subfacets) {
4193             if (subfacet->path == SF_FAST_PATH) {
4194                 struct dpif_flow_stats stats;
4195
4196                 subfacet_install(subfacet, &xout.odp_actions, &stats);
4197                 subfacet_update_stats(subfacet, &stats);
4198             }
4199         }
4200
4201         facet_flush_stats(facet);
4202
4203         ofpbuf_clear(&facet->xout.odp_actions);
4204         ofpbuf_put(&facet->xout.odp_actions, xout.odp_actions.data,
4205                    xout.odp_actions.size);
4206     }
4207
4208     /* Update 'facet' now that we've taken care of all the old state. */
4209     facet->xout.slow = xout.slow;
4210     facet->xout.has_learn = xout.has_learn;
4211     facet->xout.has_normal = xout.has_normal;
4212     facet->xout.has_fin_timeout = xout.has_fin_timeout;
4213     facet->xout.nf_output_iface = xout.nf_output_iface;
4214     facet->xout.mirrors = xout.mirrors;
4215     facet->nf_flow.output_iface = facet->xout.nf_output_iface;
4216
4217     ovs_mutex_lock(&new_rule->up.mutex);
4218     facet->used = MAX(facet->used, new_rule->up.created);
4219     ovs_mutex_unlock(&new_rule->up.mutex);
4220
4221     xlate_out_uninit(&xout);
4222     rule_dpif_unref(new_rule);
4223     return true;
4224 }
4225
4226 static void
4227 facet_reset_counters(struct facet *facet)
4228 {
4229     facet->packet_count = 0;
4230     facet->byte_count = 0;
4231     facet->prev_packet_count = 0;
4232     facet->prev_byte_count = 0;
4233     facet->accounted_bytes = 0;
4234 }
4235
4236 static void
4237 flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow,
4238                 struct dpif_flow_stats *stats, bool may_learn)
4239 {
4240     struct ofport_dpif *in_port;
4241     struct rule_dpif *rule;
4242     struct xlate_in xin;
4243
4244     in_port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4245     if (in_port && in_port->is_tunnel) {
4246         netdev_vport_inc_rx(in_port->up.netdev, stats);
4247     }
4248
4249     rule_dpif_lookup(ofproto, flow, NULL, &rule);
4250     rule_dpif_credit_stats(rule, stats);
4251     xlate_in_init(&xin, ofproto, flow, rule, stats->tcp_flags, NULL);
4252     xin.resubmit_stats = stats;
4253     xin.may_learn = may_learn;
4254     xlate_actions_for_side_effects(&xin);
4255     rule_dpif_unref(rule);
4256 }
4257
4258 static void
4259 facet_push_stats(struct facet *facet, bool may_learn)
4260 {
4261     struct dpif_flow_stats stats;
4262
4263     ovs_assert(facet->packet_count >= facet->prev_packet_count);
4264     ovs_assert(facet->byte_count >= facet->prev_byte_count);
4265     ovs_assert(facet->used >= facet->prev_used);
4266
4267     stats.n_packets = facet->packet_count - facet->prev_packet_count;
4268     stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4269     stats.used = facet->used;
4270     stats.tcp_flags = facet->tcp_flags;
4271
4272     if (may_learn || stats.n_packets || facet->used > facet->prev_used) {
4273         facet->prev_packet_count = facet->packet_count;
4274         facet->prev_byte_count = facet->byte_count;
4275         facet->prev_used = facet->used;
4276
4277         netflow_flow_update_time(facet->ofproto->netflow, &facet->nf_flow,
4278                                  facet->used);
4279         netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags);
4280         mirror_update_stats(facet->ofproto->mbridge, facet->xout.mirrors,
4281                             stats.n_packets, stats.n_bytes);
4282         flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn);
4283     }
4284 }
4285
4286 static void
4287 push_all_stats__(bool run_fast)
4288 {
4289     static long long int rl = LLONG_MIN;
4290     struct ofproto_dpif *ofproto;
4291
4292     if (time_msec() < rl) {
4293         return;
4294     }
4295
4296     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4297         struct cls_cursor cursor;
4298         struct facet *facet;
4299
4300         ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4301         cls_cursor_init(&cursor, &ofproto->facets, NULL);
4302         CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4303             facet_push_stats(facet, false);
4304             if (run_fast) {
4305                 run_fast_rl();
4306             }
4307         }
4308         ovs_rwlock_unlock(&ofproto->facets.rwlock);
4309     }
4310
4311     rl = time_msec() + 100;
4312 }
4313
4314 static void
4315 push_all_stats(void)
4316 {
4317     push_all_stats__(true);
4318 }
4319
4320 void
4321 rule_dpif_credit_stats(struct rule_dpif *rule,
4322                        const struct dpif_flow_stats *stats)
4323 {
4324     ovs_mutex_lock(&rule->stats_mutex);
4325     rule->packet_count += stats->n_packets;
4326     rule->byte_count += stats->n_bytes;
4327     rule->up.used = MAX(rule->up.used, stats->used);
4328     ovs_mutex_unlock(&rule->stats_mutex);
4329 }
4330
4331 bool
4332 rule_dpif_fail_open(const struct rule_dpif *rule)
4333 {
4334     return rule->up.cr.priority == FAIL_OPEN_PRIORITY;
4335 }
4336
4337 ovs_be64
4338 rule_dpif_get_flow_cookie(const struct rule_dpif *rule)
4339     OVS_REQUIRES(rule->up.mutex)
4340 {
4341     return rule->up.flow_cookie;
4342 }
4343
4344 void
4345 rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout,
4346                      uint16_t hard_timeout)
4347 {
4348     ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout);
4349 }
4350
4351 /* Returns 'rule''s actions.  The caller owns a reference on the returned
4352  * actions and must eventually release it (with rule_actions_unref()) to avoid
4353  * a memory leak. */
4354 struct rule_actions *
4355 rule_dpif_get_actions(const struct rule_dpif *rule)
4356 {
4357     return rule_get_actions(&rule->up);
4358 }
4359 \f
4360 /* Subfacets. */
4361
4362 static struct subfacet *
4363 subfacet_find(struct dpif_backer *backer, const struct nlattr *key,
4364               size_t key_len, uint32_t key_hash)
4365 {
4366     struct subfacet *subfacet;
4367
4368     HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4369                              &backer->subfacets) {
4370         if (subfacet->key_len == key_len
4371             && !memcmp(key, subfacet->key, key_len)) {
4372             return subfacet;
4373         }
4374     }
4375
4376     return NULL;
4377 }
4378
4379 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4380  * 'key_fitness', 'key', and 'key_len' members in 'miss'.  Returns the
4381  * existing subfacet if there is one, otherwise creates and returns a
4382  * new subfacet. */
4383 static struct subfacet *
4384 subfacet_create(struct facet *facet, struct flow_miss *miss)
4385 {
4386     struct dpif_backer *backer = miss->ofproto->backer;
4387     const struct nlattr *key = miss->key;
4388     size_t key_len = miss->key_len;
4389     uint32_t key_hash;
4390     struct subfacet *subfacet;
4391
4392     key_hash = odp_flow_key_hash(key, key_len);
4393
4394     if (list_is_empty(&facet->subfacets)) {
4395         subfacet = &facet->one_subfacet;
4396     } else {
4397         subfacet = subfacet_find(backer, key, key_len, key_hash);
4398         if (subfacet) {
4399             if (subfacet->facet == facet) {
4400                 return subfacet;
4401             }
4402
4403             /* This shouldn't happen. */
4404             VLOG_ERR_RL(&rl, "subfacet with wrong facet");
4405             subfacet_destroy(subfacet);
4406         }
4407
4408         subfacet = xmalloc(sizeof *subfacet);
4409     }
4410
4411     COVERAGE_INC(subfacet_create);
4412     hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash);
4413     list_push_back(&facet->subfacets, &subfacet->list_node);
4414     subfacet->facet = facet;
4415     subfacet->key = xmemdup(key, key_len);
4416     subfacet->key_len = key_len;
4417     subfacet->used = miss->stats.used;
4418     subfacet->created = subfacet->used;
4419     subfacet->dp_packet_count = 0;
4420     subfacet->dp_byte_count = 0;
4421     subfacet->path = SF_NOT_INSTALLED;
4422     subfacet->backer = backer;
4423
4424     return subfacet;
4425 }
4426
4427 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
4428  * its facet within 'ofproto', and frees it. */
4429 static void
4430 subfacet_destroy__(struct subfacet *subfacet)
4431 {
4432     struct facet *facet = subfacet->facet;
4433
4434     COVERAGE_INC(subfacet_destroy);
4435     subfacet_uninstall(subfacet);
4436     hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node);
4437     list_remove(&subfacet->list_node);
4438     free(subfacet->key);
4439     if (subfacet != &facet->one_subfacet) {
4440         free(subfacet);
4441     }
4442 }
4443
4444 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
4445  * last remaining subfacet in its facet destroys the facet too. */
4446 static void
4447 subfacet_destroy(struct subfacet *subfacet)
4448 {
4449     struct facet *facet = subfacet->facet;
4450
4451     if (list_is_singleton(&facet->subfacets)) {
4452         /* facet_remove() needs at least one subfacet (it will remove it). */
4453         facet_remove(facet);
4454     } else {
4455         subfacet_destroy__(subfacet);
4456     }
4457 }
4458
4459 static void
4460 subfacet_destroy_batch(struct dpif_backer *backer,
4461                        struct subfacet **subfacets, int n)
4462 {
4463     struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
4464     struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
4465     struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
4466     int i;
4467
4468     for (i = 0; i < n; i++) {
4469         ops[i].type = DPIF_OP_FLOW_DEL;
4470         ops[i].u.flow_del.key = subfacets[i]->key;
4471         ops[i].u.flow_del.key_len = subfacets[i]->key_len;
4472         ops[i].u.flow_del.stats = &stats[i];
4473         opsp[i] = &ops[i];
4474     }
4475
4476     dpif_operate(backer->dpif, opsp, n);
4477     for (i = 0; i < n; i++) {
4478         subfacet_reset_dp_stats(subfacets[i], &stats[i]);
4479         subfacets[i]->path = SF_NOT_INSTALLED;
4480         subfacet_destroy(subfacets[i]);
4481         run_fast_rl();
4482     }
4483 }
4484
4485 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
4486  * bytes of actions in 'actions'.  If 'stats' is non-null, statistics counters
4487  * in the datapath will be zeroed and 'stats' will be updated with traffic new
4488  * since 'subfacet' was last updated.
4489  *
4490  * Returns 0 if successful, otherwise a positive errno value. */
4491 static int
4492 subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
4493                  struct dpif_flow_stats *stats)
4494 {
4495     struct facet *facet = subfacet->facet;
4496     enum subfacet_path path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH;
4497     const struct nlattr *actions = odp_actions->data;
4498     size_t actions_len = odp_actions->size;
4499     struct odputil_keybuf maskbuf;
4500     struct ofpbuf mask;
4501
4502     uint64_t slow_path_stub[128 / 8];
4503     enum dpif_flow_put_flags flags;
4504     int ret;
4505
4506     flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
4507                                                : DPIF_FP_MODIFY;
4508     if (stats) {
4509         flags |= DPIF_FP_ZERO_STATS;
4510     }
4511
4512     if (path == SF_SLOW_PATH) {
4513         compose_slow_path(facet->ofproto, &facet->flow, facet->xout.slow,
4514                           slow_path_stub, sizeof slow_path_stub,
4515                           &actions, &actions_len);
4516     }
4517
4518     ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
4519     if (enable_megaflows) {
4520         odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
4521                                &facet->flow, UINT32_MAX);
4522     }
4523
4524     ret = dpif_flow_put(subfacet->backer->dpif, flags, subfacet->key,
4525                         subfacet->key_len,  mask.data, mask.size,
4526                         actions, actions_len, stats);
4527
4528     if (stats) {
4529         subfacet_reset_dp_stats(subfacet, stats);
4530     }
4531
4532     if (ret) {
4533         COVERAGE_INC(subfacet_install_fail);
4534     } else {
4535         subfacet->path = path;
4536     }
4537     return ret;
4538 }
4539
4540 /* If 'subfacet' is installed in the datapath, uninstalls it. */
4541 static void
4542 subfacet_uninstall(struct subfacet *subfacet)
4543 {
4544     if (subfacet->path != SF_NOT_INSTALLED) {
4545         struct ofproto_dpif *ofproto = subfacet->facet->ofproto;
4546         struct dpif_flow_stats stats;
4547         int error;
4548
4549         error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
4550                               subfacet->key_len, &stats);
4551         subfacet_reset_dp_stats(subfacet, &stats);
4552         if (!error) {
4553             subfacet_update_stats(subfacet, &stats);
4554         }
4555         subfacet->path = SF_NOT_INSTALLED;
4556     } else {
4557         ovs_assert(subfacet->dp_packet_count == 0);
4558         ovs_assert(subfacet->dp_byte_count == 0);
4559     }
4560 }
4561
4562 /* Resets 'subfacet''s datapath statistics counters.  This should be called
4563  * when 'subfacet''s statistics are cleared in the datapath.  If 'stats' is
4564  * non-null, it should contain the statistics returned by dpif when 'subfacet'
4565  * was reset in the datapath.  'stats' will be modified to include only
4566  * statistics new since 'subfacet' was last updated. */
4567 static void
4568 subfacet_reset_dp_stats(struct subfacet *subfacet,
4569                         struct dpif_flow_stats *stats)
4570 {
4571     if (stats
4572         && subfacet->dp_packet_count <= stats->n_packets
4573         && subfacet->dp_byte_count <= stats->n_bytes) {
4574         stats->n_packets -= subfacet->dp_packet_count;
4575         stats->n_bytes -= subfacet->dp_byte_count;
4576     }
4577
4578     subfacet->dp_packet_count = 0;
4579     subfacet->dp_byte_count = 0;
4580 }
4581
4582 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
4583  *
4584  * Because of the meaning of a subfacet's counters, it only makes sense to do
4585  * this if 'stats' are not tracked in the datapath, that is, if 'stats'
4586  * represents a packet that was sent by hand or if it represents statistics
4587  * that have been cleared out of the datapath. */
4588 static void
4589 subfacet_update_stats(struct subfacet *subfacet,
4590                       const struct dpif_flow_stats *stats)
4591 {
4592     if (stats->n_packets || stats->used > subfacet->used) {
4593         struct facet *facet = subfacet->facet;
4594
4595         subfacet->used = MAX(subfacet->used, stats->used);
4596         facet->used = MAX(facet->used, stats->used);
4597         facet->packet_count += stats->n_packets;
4598         facet->byte_count += stats->n_bytes;
4599         facet->tcp_flags |= stats->tcp_flags;
4600     }
4601 }
4602 \f
4603 /* Rules. */
4604
4605 /* Lookup 'flow' in 'ofproto''s classifier.  If 'wc' is non-null, sets
4606  * the fields that were relevant as part of the lookup. */
4607 void
4608 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow,
4609                  struct flow_wildcards *wc, struct rule_dpif **rule)
4610 {
4611     struct ofport_dpif *port;
4612
4613     if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) {
4614         return;
4615     }
4616     port = get_ofp_port(ofproto, flow->in_port.ofp_port);
4617     if (!port) {
4618         VLOG_WARN_RL(&rl, "packet-in on unknown OpenFlow port %"PRIu16,
4619                      flow->in_port.ofp_port);
4620     }
4621
4622     choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule,
4623                      ofproto->no_packet_in_rule, rule);
4624 }
4625
4626 bool
4627 rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto,
4628                           const struct flow *flow, struct flow_wildcards *wc,
4629                           uint8_t table_id, struct rule_dpif **rule)
4630 {
4631     const struct cls_rule *cls_rule;
4632     struct classifier *cls;
4633     bool frag;
4634
4635     *rule = NULL;
4636     if (table_id >= N_TABLES) {
4637         return false;
4638     }
4639
4640     if (wc) {
4641         memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4642         wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4643     }
4644
4645     cls = &ofproto->up.tables[table_id].cls;
4646     ovs_rwlock_rdlock(&cls->rwlock);
4647     frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0;
4648     if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
4649         /* We must pretend that transport ports are unavailable. */
4650         struct flow ofpc_normal_flow = *flow;
4651         ofpc_normal_flow.tp_src = htons(0);
4652         ofpc_normal_flow.tp_dst = htons(0);
4653         cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc);
4654     } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) {
4655         cls_rule = &ofproto->drop_frags_rule->up.cr;
4656         if (wc) {
4657             flow_wildcards_init_exact(wc);
4658         }
4659     } else {
4660         cls_rule = classifier_lookup(cls, flow, wc);
4661     }
4662
4663     *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule));
4664     rule_dpif_ref(*rule);
4665     ovs_rwlock_unlock(&cls->rwlock);
4666
4667     return *rule != NULL;
4668 }
4669
4670 /* Given a port configuration (specified as zero if there's no port), chooses
4671  * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
4672  * flow table miss. */
4673 void
4674 choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule,
4675                  struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule)
4676 {
4677     *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule;
4678     rule_dpif_ref(*rule);
4679 }
4680
4681 void
4682 rule_dpif_ref(struct rule_dpif *rule)
4683 {
4684     if (rule) {
4685         ofproto_rule_ref(&rule->up);
4686     }
4687 }
4688
4689 void
4690 rule_dpif_unref(struct rule_dpif *rule)
4691 {
4692     if (rule) {
4693         ofproto_rule_unref(&rule->up);
4694     }
4695 }
4696
4697 static void
4698 complete_operation(struct rule_dpif *rule)
4699     OVS_REQUIRES(ofproto_mutex)
4700 {
4701     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4702
4703     ofproto->backer->need_revalidate = REV_FLOW_TABLE;
4704     ofoperation_complete(rule->up.pending, 0);
4705 }
4706
4707 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
4708 {
4709     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
4710 }
4711
4712 static struct rule *
4713 rule_alloc(void)
4714 {
4715     struct rule_dpif *rule = xmalloc(sizeof *rule);
4716     return &rule->up;
4717 }
4718
4719 static void
4720 rule_dealloc(struct rule *rule_)
4721 {
4722     struct rule_dpif *rule = rule_dpif_cast(rule_);
4723     free(rule);
4724 }
4725
4726 static enum ofperr
4727 rule_construct(struct rule *rule_)
4728 {
4729     struct rule_dpif *rule = rule_dpif_cast(rule_);
4730     ovs_mutex_init(&rule->stats_mutex);
4731     ovs_mutex_lock(&rule->stats_mutex);
4732     rule->packet_count = 0;
4733     rule->byte_count = 0;
4734     ovs_mutex_unlock(&rule->stats_mutex);
4735     return 0;
4736 }
4737
4738 static void
4739 rule_insert(struct rule *rule_)
4740     OVS_REQUIRES(ofproto_mutex)
4741 {
4742     struct rule_dpif *rule = rule_dpif_cast(rule_);
4743     complete_operation(rule);
4744 }
4745
4746 static void
4747 rule_delete(struct rule *rule_)
4748     OVS_REQUIRES(ofproto_mutex)
4749 {
4750     struct rule_dpif *rule = rule_dpif_cast(rule_);
4751     complete_operation(rule);
4752 }
4753
4754 static void
4755 rule_destruct(struct rule *rule_)
4756 {
4757     struct rule_dpif *rule = rule_dpif_cast(rule_);
4758     ovs_mutex_destroy(&rule->stats_mutex);
4759 }
4760
4761 static void
4762 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
4763 {
4764     struct rule_dpif *rule = rule_dpif_cast(rule_);
4765
4766     /* push_all_stats() can handle flow misses which, when using the learn
4767      * action, can cause rules to be added and deleted.  This can corrupt our
4768      * caller's datastructures which assume that rule_get_stats() doesn't have
4769      * an impact on the flow table. To be safe, we disable miss handling. */
4770     push_all_stats__(false);
4771
4772     /* Start from historical data for 'rule' itself that are no longer tracked
4773      * in facets.  This counts, for example, facets that have expired. */
4774     ovs_mutex_lock(&rule->stats_mutex);
4775     *packets = rule->packet_count;
4776     *bytes = rule->byte_count;
4777     ovs_mutex_unlock(&rule->stats_mutex);
4778 }
4779
4780 static void
4781 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
4782                   struct ofpbuf *packet)
4783 {
4784     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4785     struct dpif_flow_stats stats;
4786     struct xlate_out xout;
4787     struct xlate_in xin;
4788
4789     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4790     rule_dpif_credit_stats(rule, &stats);
4791
4792     xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet);
4793     xin.resubmit_stats = &stats;
4794     xlate_actions(&xin, &xout);
4795
4796     execute_odp_actions(ofproto, flow, xout.odp_actions.data,
4797                         xout.odp_actions.size, packet);
4798
4799     xlate_out_uninit(&xout);
4800 }
4801
4802 static enum ofperr
4803 rule_execute(struct rule *rule, const struct flow *flow,
4804              struct ofpbuf *packet)
4805 {
4806     rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
4807     ofpbuf_delete(packet);
4808     return 0;
4809 }
4810
4811 static void
4812 rule_modify_actions(struct rule *rule_, bool reset_counters)
4813     OVS_REQUIRES(ofproto_mutex)
4814 {
4815     struct rule_dpif *rule = rule_dpif_cast(rule_);
4816
4817     if (reset_counters) {
4818         ovs_mutex_lock(&rule->stats_mutex);
4819         rule->packet_count = 0;
4820         rule->byte_count = 0;
4821         ovs_mutex_unlock(&rule->stats_mutex);
4822     }
4823
4824     complete_operation(rule);
4825 }
4826 \f
4827 /* Sends 'packet' out 'ofport'.
4828  * May modify 'packet'.
4829  * Returns 0 if successful, otherwise a positive errno value. */
4830 int
4831 ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4832 {
4833     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
4834     int error;
4835
4836     error = xlate_send_packet(ofport, packet);
4837
4838     ovs_mutex_lock(&ofproto->stats_mutex);
4839     ofproto->stats.tx_packets++;
4840     ofproto->stats.tx_bytes += packet->size;
4841     ovs_mutex_unlock(&ofproto->stats_mutex);
4842     return error;
4843 }
4844
4845 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
4846  * The action will state 'slow' as the reason that the action is in the slow
4847  * path.  (This is purely informational: it allows a human viewing "ovs-dpctl
4848  * dump-flows" output to see why a flow is in the slow path.)
4849  *
4850  * The 'stub_size' bytes in 'stub' will be used to store the action.
4851  * 'stub_size' must be large enough for the action.
4852  *
4853  * The action and its size will be stored in '*actionsp' and '*actions_lenp',
4854  * respectively. */
4855 static void
4856 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
4857                   enum slow_path_reason slow,
4858                   uint64_t *stub, size_t stub_size,
4859                   const struct nlattr **actionsp, size_t *actions_lenp)
4860 {
4861     union user_action_cookie cookie;
4862     struct ofpbuf buf;
4863
4864     cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
4865     cookie.slow_path.unused = 0;
4866     cookie.slow_path.reason = slow;
4867
4868     ofpbuf_use_stack(&buf, stub, stub_size);
4869     if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
4870         uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif,
4871                                          ODPP_NONE);
4872         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
4873     } else {
4874         odp_port_t odp_port;
4875         uint32_t pid;
4876
4877         odp_port = ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port);
4878         pid = dpif_port_get_pid(ofproto->backer->dpif, odp_port);
4879         odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
4880     }
4881     *actionsp = buf.data;
4882     *actions_lenp = buf.size;
4883 }
4884 \f
4885 static bool
4886 set_frag_handling(struct ofproto *ofproto_,
4887                   enum ofp_config_flags frag_handling)
4888 {
4889     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4890     if (frag_handling != OFPC_FRAG_REASM) {
4891         ofproto->backer->need_revalidate = REV_RECONFIGURE;
4892         return true;
4893     } else {
4894         return false;
4895     }
4896 }
4897
4898 static enum ofperr
4899 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
4900            const struct flow *flow,
4901            const struct ofpact *ofpacts, size_t ofpacts_len)
4902 {
4903     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4904     struct odputil_keybuf keybuf;
4905     struct dpif_flow_stats stats;
4906     struct xlate_out xout;
4907     struct xlate_in xin;
4908     struct ofpbuf key;
4909
4910
4911     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4912     odp_flow_key_from_flow(&key, flow,
4913                            ofp_port_to_odp_port(ofproto,
4914                                       flow->in_port.ofp_port));
4915
4916     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
4917
4918     xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet);
4919     xin.resubmit_stats = &stats;
4920     xin.ofpacts_len = ofpacts_len;
4921     xin.ofpacts = ofpacts;
4922
4923     xlate_actions(&xin, &xout);
4924     dpif_execute(ofproto->backer->dpif, key.data, key.size,
4925                  xout.odp_actions.data, xout.odp_actions.size, packet);
4926     xlate_out_uninit(&xout);
4927
4928     return 0;
4929 }
4930 \f
4931 /* NetFlow. */
4932
4933 static int
4934 set_netflow(struct ofproto *ofproto_,
4935             const struct netflow_options *netflow_options)
4936 {
4937     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4938
4939     if (netflow_options) {
4940         if (!ofproto->netflow) {
4941             ofproto->netflow = netflow_create();
4942             ofproto->backer->need_revalidate = REV_RECONFIGURE;
4943         }
4944         return netflow_set_options(ofproto->netflow, netflow_options);
4945     } else if (ofproto->netflow) {
4946         ofproto->backer->need_revalidate = REV_RECONFIGURE;
4947         netflow_destroy(ofproto->netflow);
4948         ofproto->netflow = NULL;
4949     }
4950
4951     return 0;
4952 }
4953
4954 static void
4955 get_netflow_ids(const struct ofproto *ofproto_,
4956                 uint8_t *engine_type, uint8_t *engine_id)
4957 {
4958     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
4959
4960     dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
4961 }
4962
4963 static void
4964 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
4965 {
4966     if (!facet_is_controller_flow(facet) &&
4967         netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
4968         struct subfacet *subfacet;
4969         struct ofexpired expired;
4970
4971         LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4972             if (subfacet->path == SF_FAST_PATH) {
4973                 struct dpif_flow_stats stats;
4974
4975                 subfacet_install(subfacet, &facet->xout.odp_actions,
4976                                  &stats);
4977                 subfacet_update_stats(subfacet, &stats);
4978             }
4979         }
4980
4981         expired.flow = facet->flow;
4982         expired.packet_count = facet->packet_count;
4983         expired.byte_count = facet->byte_count;
4984         expired.used = facet->used;
4985         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4986     }
4987 }
4988
4989 static void
4990 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
4991 {
4992     struct cls_cursor cursor;
4993     struct facet *facet;
4994
4995     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
4996     cls_cursor_init(&cursor, &ofproto->facets, NULL);
4997     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
4998         send_active_timeout(ofproto, facet);
4999     }
5000     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5001 }
5002 \f
5003 static struct ofproto_dpif *
5004 ofproto_dpif_lookup(const char *name)
5005 {
5006     struct ofproto_dpif *ofproto;
5007
5008     HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
5009                              hash_string(name, 0), &all_ofproto_dpifs) {
5010         if (!strcmp(ofproto->up.name, name)) {
5011             return ofproto;
5012         }
5013     }
5014     return NULL;
5015 }
5016
5017 static void
5018 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
5019                           const char *argv[], void *aux OVS_UNUSED)
5020 {
5021     struct ofproto_dpif *ofproto;
5022
5023     if (argc > 1) {
5024         ofproto = ofproto_dpif_lookup(argv[1]);
5025         if (!ofproto) {
5026             unixctl_command_reply_error(conn, "no such bridge");
5027             return;
5028         }
5029         ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5030         mac_learning_flush(ofproto->ml);
5031         ovs_rwlock_unlock(&ofproto->ml->rwlock);
5032     } else {
5033         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5034             ovs_rwlock_wrlock(&ofproto->ml->rwlock);
5035             mac_learning_flush(ofproto->ml);
5036             ovs_rwlock_unlock(&ofproto->ml->rwlock);
5037         }
5038     }
5039
5040     unixctl_command_reply(conn, "table successfully flushed");
5041 }
5042
5043 static struct ofport_dpif *
5044 ofbundle_get_a_port(const struct ofbundle *bundle)
5045 {
5046     return CONTAINER_OF(list_front(&bundle->ports), struct ofport_dpif,
5047                         bundle_node);
5048 }
5049
5050 static void
5051 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5052                          const char *argv[], void *aux OVS_UNUSED)
5053 {
5054     struct ds ds = DS_EMPTY_INITIALIZER;
5055     const struct ofproto_dpif *ofproto;
5056     const struct mac_entry *e;
5057
5058     ofproto = ofproto_dpif_lookup(argv[1]);
5059     if (!ofproto) {
5060         unixctl_command_reply_error(conn, "no such bridge");
5061         return;
5062     }
5063
5064     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
5065     ovs_rwlock_rdlock(&ofproto->ml->rwlock);
5066     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
5067         struct ofbundle *bundle = e->port.p;
5068         char name[OFP_MAX_PORT_NAME_LEN];
5069
5070         ofputil_port_to_string(ofbundle_get_a_port(bundle)->up.ofp_port,
5071                                name, sizeof name);
5072         ds_put_format(&ds, "%5s  %4d  "ETH_ADDR_FMT"  %3d\n",
5073                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
5074                       mac_entry_age(ofproto->ml, e));
5075     }
5076     ovs_rwlock_unlock(&ofproto->ml->rwlock);
5077     unixctl_command_reply(conn, ds_cstr(&ds));
5078     ds_destroy(&ds);
5079 }
5080
5081 struct trace_ctx {
5082     struct xlate_out xout;
5083     struct xlate_in xin;
5084     struct flow flow;
5085     struct ds *result;
5086 };
5087
5088 static void
5089 trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule)
5090 {
5091     struct rule_actions *actions;
5092     ovs_be64 cookie;
5093
5094     ds_put_char_multiple(result, '\t', level);
5095     if (!rule) {
5096         ds_put_cstr(result, "No match\n");
5097         return;
5098     }
5099
5100     ovs_mutex_lock(&rule->up.mutex);
5101     cookie = rule->up.flow_cookie;
5102     ovs_mutex_unlock(&rule->up.mutex);
5103
5104     ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
5105                   rule ? rule->up.table_id : 0, ntohll(cookie));
5106     cls_rule_format(&rule->up.cr, result);
5107     ds_put_char(result, '\n');
5108
5109     actions = rule_dpif_get_actions(rule);
5110
5111     ds_put_char_multiple(result, '\t', level);
5112     ds_put_cstr(result, "OpenFlow ");
5113     ofpacts_format(actions->ofpacts, actions->ofpacts_len, result);
5114     ds_put_char(result, '\n');
5115
5116     rule_actions_unref(actions);
5117 }
5118
5119 static void
5120 trace_format_flow(struct ds *result, int level, const char *title,
5121                   struct trace_ctx *trace)
5122 {
5123     ds_put_char_multiple(result, '\t', level);
5124     ds_put_format(result, "%s: ", title);
5125     if (flow_equal(&trace->xin.flow, &trace->flow)) {
5126         ds_put_cstr(result, "unchanged");
5127     } else {
5128         flow_format(result, &trace->xin.flow);
5129         trace->flow = trace->xin.flow;
5130     }
5131     ds_put_char(result, '\n');
5132 }
5133
5134 static void
5135 trace_format_regs(struct ds *result, int level, const char *title,
5136                   struct trace_ctx *trace)
5137 {
5138     size_t i;
5139
5140     ds_put_char_multiple(result, '\t', level);
5141     ds_put_format(result, "%s:", title);
5142     for (i = 0; i < FLOW_N_REGS; i++) {
5143         ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
5144     }
5145     ds_put_char(result, '\n');
5146 }
5147
5148 static void
5149 trace_format_odp(struct ds *result, int level, const char *title,
5150                  struct trace_ctx *trace)
5151 {
5152     struct ofpbuf *odp_actions = &trace->xout.odp_actions;
5153
5154     ds_put_char_multiple(result, '\t', level);
5155     ds_put_format(result, "%s: ", title);
5156     format_odp_actions(result, odp_actions->data, odp_actions->size);
5157     ds_put_char(result, '\n');
5158 }
5159
5160 static void
5161 trace_resubmit(struct xlate_in *xin, struct rule_dpif *rule, int recurse)
5162 {
5163     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5164     struct ds *result = trace->result;
5165
5166     ds_put_char(result, '\n');
5167     trace_format_flow(result, recurse + 1, "Resubmitted flow", trace);
5168     trace_format_regs(result, recurse + 1, "Resubmitted regs", trace);
5169     trace_format_odp(result,  recurse + 1, "Resubmitted  odp", trace);
5170     trace_format_rule(result, recurse + 1, rule);
5171 }
5172
5173 static void
5174 trace_report(struct xlate_in *xin, const char *s, int recurse)
5175 {
5176     struct trace_ctx *trace = CONTAINER_OF(xin, struct trace_ctx, xin);
5177     struct ds *result = trace->result;
5178
5179     ds_put_char_multiple(result, '\t', recurse);
5180     ds_put_cstr(result, s);
5181     ds_put_char(result, '\n');
5182 }
5183
5184 static void
5185 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
5186                       void *aux OVS_UNUSED)
5187 {
5188     const struct dpif_backer *backer = NULL;
5189     struct ofproto_dpif *ofproto;
5190     struct ofpbuf odp_key, odp_mask;
5191     struct ofpbuf *packet;
5192     struct ds result;
5193     struct flow flow;
5194     struct simap port_names;
5195     char *s;
5196
5197     packet = NULL;
5198     backer = NULL;
5199     ds_init(&result);
5200     ofpbuf_init(&odp_key, 0);
5201     ofpbuf_init(&odp_mask, 0);
5202     simap_init(&port_names);
5203
5204     /* Handle "-generate" or a hex string as the last argument. */
5205     if (!strcmp(argv[argc - 1], "-generate")) {
5206         packet = ofpbuf_new(0);
5207         argc--;
5208     } else {
5209         const char *error = eth_from_hex(argv[argc - 1], &packet);
5210         if (!error) {
5211             argc--;
5212         } else if (argc == 4) {
5213             /* The 3-argument form must end in "-generate' or a hex string. */
5214             unixctl_command_reply_error(conn, error);
5215             goto exit;
5216         }
5217     }
5218
5219     /* odp_flow can have its in_port specified as a name instead of port no.
5220      * We do not yet know whether a given flow is a odp_flow or a br_flow.
5221      * But, to know whether a flow is odp_flow through odp_flow_from_string(),
5222      * we need to create a simap of name to port no. */
5223     if (argc == 3) {
5224         const char *dp_type;
5225         if (!strncmp(argv[1], "ovs-", 4)) {
5226             dp_type = argv[1] + 4;
5227         } else {
5228             dp_type = argv[1];
5229         }
5230         backer = shash_find_data(&all_dpif_backers, dp_type);
5231     } else {
5232         struct shash_node *node;
5233         if (shash_count(&all_dpif_backers) == 1) {
5234             node = shash_first(&all_dpif_backers);
5235             backer = node->data;
5236         }
5237     }
5238     if (backer && backer->dpif) {
5239         struct dpif_port dpif_port;
5240         struct dpif_port_dump port_dump;
5241         DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) {
5242             simap_put(&port_names, dpif_port.name,
5243                       odp_to_u32(dpif_port.port_no));
5244         }
5245     }
5246
5247     /* Parse the flow and determine whether a datapath or
5248      * bridge is specified. If function odp_flow_key_from_string()
5249      * returns 0, the flow is a odp_flow. If function
5250      * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */
5251     if (!odp_flow_from_string(argv[argc - 1], &port_names, &odp_key, &odp_mask)) {
5252         if (!backer) {
5253             unixctl_command_reply_error(conn, "Cannot find the datapath");
5254             goto exit;
5255         }
5256
5257         if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow,
5258                           NULL, &ofproto, NULL)) {
5259             unixctl_command_reply_error(conn, "Invalid datapath flow");
5260             goto exit;
5261         }
5262         ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
5263     } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) {
5264         if (argc != 3) {
5265             unixctl_command_reply_error(conn, "Must specify bridge name");
5266             goto exit;
5267         }
5268
5269         ofproto = ofproto_dpif_lookup(argv[1]);
5270         if (!ofproto) {
5271             unixctl_command_reply_error(conn, "Unknown bridge name");
5272             goto exit;
5273         }
5274     } else {
5275         unixctl_command_reply_error(conn, "Bad flow syntax");
5276         goto exit;
5277     }
5278
5279     /* Generate a packet, if requested. */
5280     if (packet) {
5281         if (!packet->size) {
5282             flow_compose(packet, &flow);
5283         } else {
5284             union flow_in_port in_port_;
5285
5286             in_port_ = flow.in_port;
5287             ds_put_cstr(&result, "Packet: ");
5288             s = ofp_packet_to_string(packet->data, packet->size);
5289             ds_put_cstr(&result, s);
5290             free(s);
5291
5292             /* Use the metadata from the flow and the packet argument
5293              * to reconstruct the flow. */
5294             flow_extract(packet, flow.skb_priority, flow.pkt_mark, NULL,
5295                          &in_port_, &flow);
5296         }
5297     }
5298
5299     ofproto_trace(ofproto, &flow, packet, &result);
5300     unixctl_command_reply(conn, ds_cstr(&result));
5301
5302 exit:
5303     ds_destroy(&result);
5304     ofpbuf_delete(packet);
5305     ofpbuf_uninit(&odp_key);
5306     ofpbuf_uninit(&odp_mask);
5307     simap_destroy(&port_names);
5308 }
5309
5310 static void
5311 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
5312               const struct ofpbuf *packet, struct ds *ds)
5313 {
5314     struct rule_dpif *rule;
5315     struct flow_wildcards wc;
5316
5317     ds_put_cstr(ds, "Flow: ");
5318     flow_format(ds, flow);
5319     ds_put_char(ds, '\n');
5320
5321     flow_wildcards_init_catchall(&wc);
5322     rule_dpif_lookup(ofproto, flow, &wc, &rule);
5323
5324     trace_format_rule(ds, 0, rule);
5325     if (rule == ofproto->miss_rule) {
5326         ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
5327     } else if (rule == ofproto->no_packet_in_rule) {
5328         ds_put_cstr(ds, "\nNo match, packets dropped because "
5329                     "OFPPC_NO_PACKET_IN is set on in_port.\n");
5330     } else if (rule == ofproto->drop_frags_rule) {
5331         ds_put_cstr(ds, "\nPackets dropped because they are IP fragments "
5332                     "and the fragment handling mode is \"drop\".\n");
5333     }
5334
5335     if (rule) {
5336         uint64_t odp_actions_stub[1024 / 8];
5337         struct ofpbuf odp_actions;
5338         struct trace_ctx trace;
5339         struct match match;
5340         uint8_t tcp_flags;
5341
5342         tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
5343         trace.result = ds;
5344         trace.flow = *flow;
5345         ofpbuf_use_stub(&odp_actions,
5346                         odp_actions_stub, sizeof odp_actions_stub);
5347         xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet);
5348         trace.xin.resubmit_hook = trace_resubmit;
5349         trace.xin.report_hook = trace_report;
5350
5351         xlate_actions(&trace.xin, &trace.xout);
5352         flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc);
5353
5354         ds_put_char(ds, '\n');
5355         trace_format_flow(ds, 0, "Final flow", &trace);
5356
5357         match_init(&match, flow, &trace.xout.wc);
5358         ds_put_cstr(ds, "Relevant fields: ");
5359         match_format(&match, ds, OFP_DEFAULT_PRIORITY);
5360         ds_put_char(ds, '\n');
5361
5362         ds_put_cstr(ds, "Datapath actions: ");
5363         format_odp_actions(ds, trace.xout.odp_actions.data,
5364                            trace.xout.odp_actions.size);
5365
5366         if (trace.xout.slow) {
5367             ds_put_cstr(ds, "\nThis flow is handled by the userspace "
5368                         "slow path because it:");
5369             switch (trace.xout.slow) {
5370             case SLOW_CFM:
5371                 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
5372                 break;
5373             case SLOW_LACP:
5374                 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
5375                 break;
5376             case SLOW_STP:
5377                 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
5378                 break;
5379             case SLOW_BFD:
5380                 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
5381                 break;
5382             case SLOW_CONTROLLER:
5383                 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
5384                             "to the OpenFlow controller.");
5385                 break;
5386             case __SLOW_MAX:
5387                 NOT_REACHED();
5388             }
5389         }
5390
5391         xlate_out_uninit(&trace.xout);
5392     }
5393
5394     rule_dpif_unref(rule);
5395 }
5396
5397 /* Runs a self-check of flow translations in 'ofproto'.  Appends a message to
5398  * 'reply' describing the results. */
5399 static void
5400 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
5401 {
5402     struct cls_cursor cursor;
5403     struct facet *facet;
5404     int errors;
5405
5406     errors = 0;
5407     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5408     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5409     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5410         if (!facet_check_consistency(facet)) {
5411             errors++;
5412         }
5413     }
5414     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5415     if (errors) {
5416         ofproto->backer->need_revalidate = REV_INCONSISTENCY;
5417     }
5418
5419     if (errors) {
5420         ds_put_format(reply, "%s: self-check failed (%d errors)\n",
5421                       ofproto->up.name, errors);
5422     } else {
5423         ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
5424     }
5425 }
5426
5427 static void
5428 ofproto_dpif_self_check(struct unixctl_conn *conn,
5429                         int argc, const char *argv[], void *aux OVS_UNUSED)
5430 {
5431     struct ds reply = DS_EMPTY_INITIALIZER;
5432     struct ofproto_dpif *ofproto;
5433
5434     if (argc > 1) {
5435         ofproto = ofproto_dpif_lookup(argv[1]);
5436         if (!ofproto) {
5437             unixctl_command_reply_error(conn, "Unknown ofproto (use "
5438                                         "ofproto/list for help)");
5439             return;
5440         }
5441         ofproto_dpif_self_check__(ofproto, &reply);
5442     } else {
5443         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5444             ofproto_dpif_self_check__(ofproto, &reply);
5445         }
5446     }
5447
5448     unixctl_command_reply(conn, ds_cstr(&reply));
5449     ds_destroy(&reply);
5450 }
5451
5452 /* Store the current ofprotos in 'ofproto_shash'.  Returns a sorted list
5453  * of the 'ofproto_shash' nodes.  It is the responsibility of the caller
5454  * to destroy 'ofproto_shash' and free the returned value. */
5455 static const struct shash_node **
5456 get_ofprotos(struct shash *ofproto_shash)
5457 {
5458     const struct ofproto_dpif *ofproto;
5459
5460     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5461         char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
5462         shash_add_nocopy(ofproto_shash, name, ofproto);
5463     }
5464
5465     return shash_sort(ofproto_shash);
5466 }
5467
5468 static void
5469 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
5470                               const char *argv[] OVS_UNUSED,
5471                               void *aux OVS_UNUSED)
5472 {
5473     struct ds ds = DS_EMPTY_INITIALIZER;
5474     struct shash ofproto_shash;
5475     const struct shash_node **sorted_ofprotos;
5476     int i;
5477
5478     shash_init(&ofproto_shash);
5479     sorted_ofprotos = get_ofprotos(&ofproto_shash);
5480     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5481         const struct shash_node *node = sorted_ofprotos[i];
5482         ds_put_format(&ds, "%s\n", node->name);
5483     }
5484
5485     shash_destroy(&ofproto_shash);
5486     free(sorted_ofprotos);
5487
5488     unixctl_command_reply(conn, ds_cstr(&ds));
5489     ds_destroy(&ds);
5490 }
5491
5492 static void
5493 dpif_show_backer(const struct dpif_backer *backer, struct ds *ds)
5494 {
5495     const struct shash_node **ofprotos;
5496     struct ofproto_dpif *ofproto;
5497     struct shash ofproto_shash;
5498     uint64_t n_hit, n_missed;
5499     size_t i;
5500
5501     n_hit = n_missed = 0;
5502     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5503         if (ofproto->backer == backer) {
5504             n_missed += ofproto->n_missed;
5505             n_hit += ofproto->n_hit;
5506         }
5507     }
5508
5509     ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5510                   dpif_name(backer->dpif), n_hit, n_missed);
5511     ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u,"
5512                   " life span: %lldms\n", hmap_count(&backer->subfacets),
5513                   backer->avg_n_subfacet, backer->max_n_subfacet,
5514                   backer->avg_subfacet_life);
5515
5516     shash_init(&ofproto_shash);
5517     ofprotos = get_ofprotos(&ofproto_shash);
5518     for (i = 0; i < shash_count(&ofproto_shash); i++) {
5519         struct ofproto_dpif *ofproto = ofprotos[i]->data;
5520         const struct shash_node **ports;
5521         size_t j;
5522
5523         if (ofproto->backer != backer) {
5524             continue;
5525         }
5526
5527         ds_put_format(ds, "\t%s: hit:%"PRIu64" missed:%"PRIu64"\n",
5528                       ofproto->up.name, ofproto->n_hit, ofproto->n_missed);
5529
5530         ports = shash_sort(&ofproto->up.port_by_name);
5531         for (j = 0; j < shash_count(&ofproto->up.port_by_name); j++) {
5532             const struct shash_node *node = ports[j];
5533             struct ofport *ofport = node->data;
5534             struct smap config;
5535             odp_port_t odp_port;
5536
5537             ds_put_format(ds, "\t\t%s %u/", netdev_get_name(ofport->netdev),
5538                           ofport->ofp_port);
5539
5540             odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
5541             if (odp_port != ODPP_NONE) {
5542                 ds_put_format(ds, "%"PRIu32":", odp_port);
5543             } else {
5544                 ds_put_cstr(ds, "none:");
5545             }
5546
5547             ds_put_format(ds, " (%s", netdev_get_type(ofport->netdev));
5548
5549             smap_init(&config);
5550             if (!netdev_get_config(ofport->netdev, &config)) {
5551                 const struct smap_node **nodes;
5552                 size_t i;
5553
5554                 nodes = smap_sort(&config);
5555                 for (i = 0; i < smap_count(&config); i++) {
5556                     const struct smap_node *node = nodes[i];
5557                     ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
5558                                   node->key, node->value);
5559                 }
5560                 free(nodes);
5561             }
5562             smap_destroy(&config);
5563
5564             ds_put_char(ds, ')');
5565             ds_put_char(ds, '\n');
5566         }
5567         free(ports);
5568     }
5569     shash_destroy(&ofproto_shash);
5570     free(ofprotos);
5571 }
5572
5573 static void
5574 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
5575                           const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5576 {
5577     struct ds ds = DS_EMPTY_INITIALIZER;
5578     const struct shash_node **backers;
5579     int i;
5580
5581     backers = shash_sort(&all_dpif_backers);
5582     for (i = 0; i < shash_count(&all_dpif_backers); i++) {
5583         dpif_show_backer(backers[i]->data, &ds);
5584     }
5585     free(backers);
5586
5587     unixctl_command_reply(conn, ds_cstr(&ds));
5588     ds_destroy(&ds);
5589 }
5590
5591 /* Dump the megaflow (facet) cache.  This is useful to check the
5592  * correctness of flow wildcarding, since the same mechanism is used for
5593  * both xlate caching and kernel wildcarding.
5594  *
5595  * It's important to note that in the output the flow description uses
5596  * OpenFlow (OFP) ports, but the actions use datapath (ODP) ports.
5597  *
5598  * This command is only needed for advanced debugging, so it's not
5599  * documented in the man page. */
5600 static void
5601 ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn,
5602                                     int argc OVS_UNUSED, const char *argv[],
5603                                     void *aux OVS_UNUSED)
5604 {
5605     struct ds ds = DS_EMPTY_INITIALIZER;
5606     const struct ofproto_dpif *ofproto;
5607     long long int now = time_msec();
5608     struct cls_cursor cursor;
5609     struct facet *facet;
5610
5611     ofproto = ofproto_dpif_lookup(argv[1]);
5612     if (!ofproto) {
5613         unixctl_command_reply_error(conn, "no such bridge");
5614         return;
5615     }
5616
5617     ovs_rwlock_rdlock(&ofproto->facets.rwlock);
5618     cls_cursor_init(&cursor, &ofproto->facets, NULL);
5619     CLS_CURSOR_FOR_EACH (facet, cr, &cursor) {
5620         cls_rule_format(&facet->cr, &ds);
5621         ds_put_cstr(&ds, ", ");
5622         ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets));
5623         ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0);
5624         ds_put_cstr(&ds, "Datapath actions: ");
5625         if (facet->xout.slow) {
5626             uint64_t slow_path_stub[128 / 8];
5627             const struct nlattr *actions;
5628             size_t actions_len;
5629
5630             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5631                               slow_path_stub, sizeof slow_path_stub,
5632                               &actions, &actions_len);
5633             format_odp_actions(&ds, actions, actions_len);
5634         } else {
5635             format_odp_actions(&ds, facet->xout.odp_actions.data,
5636                                facet->xout.odp_actions.size);
5637         }
5638         ds_put_cstr(&ds, "\n");
5639     }
5640     ovs_rwlock_unlock(&ofproto->facets.rwlock);
5641
5642     ds_chomp(&ds, '\n');
5643     unixctl_command_reply(conn, ds_cstr(&ds));
5644     ds_destroy(&ds);
5645 }
5646
5647 /* Disable using the megaflows.
5648  *
5649  * This command is only needed for advanced debugging, so it's not
5650  * documented in the man page. */
5651 static void
5652 ofproto_unixctl_dpif_disable_megaflows(struct unixctl_conn *conn,
5653                                        int argc OVS_UNUSED,
5654                                        const char *argv[] OVS_UNUSED,
5655                                        void *aux OVS_UNUSED)
5656 {
5657     struct ofproto_dpif *ofproto;
5658
5659     enable_megaflows = false;
5660
5661     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5662         flush(&ofproto->up);
5663     }
5664
5665     unixctl_command_reply(conn, "megaflows disabled");
5666 }
5667
5668 /* Re-enable using megaflows.
5669  *
5670  * This command is only needed for advanced debugging, so it's not
5671  * documented in the man page. */
5672 static void
5673 ofproto_unixctl_dpif_enable_megaflows(struct unixctl_conn *conn,
5674                                       int argc OVS_UNUSED,
5675                                       const char *argv[] OVS_UNUSED,
5676                                       void *aux OVS_UNUSED)
5677 {
5678     struct ofproto_dpif *ofproto;
5679
5680     enable_megaflows = true;
5681
5682     HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5683         flush(&ofproto->up);
5684     }
5685
5686     unixctl_command_reply(conn, "megaflows enabled");
5687 }
5688
5689 static void
5690 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
5691                                 int argc OVS_UNUSED, const char *argv[],
5692                                 void *aux OVS_UNUSED)
5693 {
5694     struct ds ds = DS_EMPTY_INITIALIZER;
5695     const struct ofproto_dpif *ofproto;
5696     struct subfacet *subfacet;
5697
5698     ofproto = ofproto_dpif_lookup(argv[1]);
5699     if (!ofproto) {
5700         unixctl_command_reply_error(conn, "no such bridge");
5701         return;
5702     }
5703
5704     update_stats(ofproto->backer);
5705
5706     HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->backer->subfacets) {
5707         struct facet *facet = subfacet->facet;
5708         struct odputil_keybuf maskbuf;
5709         struct ofpbuf mask;
5710
5711         if (facet->ofproto != ofproto) {
5712             continue;
5713         }
5714
5715         ofpbuf_use_stack(&mask, &maskbuf, sizeof maskbuf);
5716         if (enable_megaflows) {
5717             odp_flow_key_from_mask(&mask, &facet->xout.wc.masks,
5718                                    &facet->flow, UINT32_MAX);
5719         }
5720
5721         odp_flow_format(subfacet->key, subfacet->key_len,
5722                         mask.data, mask.size, NULL, &ds, false);
5723
5724         ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
5725                       subfacet->dp_packet_count, subfacet->dp_byte_count);
5726         if (subfacet->used) {
5727             ds_put_format(&ds, "%.3fs",
5728                           (time_msec() - subfacet->used) / 1000.0);
5729         } else {
5730             ds_put_format(&ds, "never");
5731         }
5732         if (subfacet->facet->tcp_flags) {
5733             ds_put_cstr(&ds, ", flags:");
5734             packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
5735         }
5736
5737         ds_put_cstr(&ds, ", actions:");
5738         if (facet->xout.slow) {
5739             uint64_t slow_path_stub[128 / 8];
5740             const struct nlattr *actions;
5741             size_t actions_len;
5742
5743             compose_slow_path(ofproto, &facet->flow, facet->xout.slow,
5744                               slow_path_stub, sizeof slow_path_stub,
5745                               &actions, &actions_len);
5746             format_odp_actions(&ds, actions, actions_len);
5747         } else {
5748             format_odp_actions(&ds, facet->xout.odp_actions.data,
5749                                facet->xout.odp_actions.size);
5750         }
5751         ds_put_char(&ds, '\n');
5752     }
5753
5754     unixctl_command_reply(conn, ds_cstr(&ds));
5755     ds_destroy(&ds);
5756 }
5757
5758 static void
5759 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
5760                                int argc OVS_UNUSED, const char *argv[],
5761                                void *aux OVS_UNUSED)
5762 {
5763     struct ds ds = DS_EMPTY_INITIALIZER;
5764     struct ofproto_dpif *ofproto;
5765
5766     ofproto = ofproto_dpif_lookup(argv[1]);
5767     if (!ofproto) {
5768         unixctl_command_reply_error(conn, "no such bridge");
5769         return;
5770     }
5771
5772     flush(&ofproto->up);
5773
5774     unixctl_command_reply(conn, ds_cstr(&ds));
5775     ds_destroy(&ds);
5776 }
5777
5778 static void
5779 ofproto_dpif_unixctl_init(void)
5780 {
5781     static bool registered;
5782     if (registered) {
5783         return;
5784     }
5785     registered = true;
5786
5787     unixctl_command_register(
5788         "ofproto/trace",
5789         "[dp_name]|bridge odp_flow|br_flow [-generate|packet]",
5790         1, 3, ofproto_unixctl_trace, NULL);
5791     unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
5792                              ofproto_unixctl_fdb_flush, NULL);
5793     unixctl_command_register("fdb/show", "bridge", 1, 1,
5794                              ofproto_unixctl_fdb_show, NULL);
5795     unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
5796                              ofproto_dpif_self_check, NULL);
5797     unixctl_command_register("dpif/dump-dps", "", 0, 0,
5798                              ofproto_unixctl_dpif_dump_dps, NULL);
5799     unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show,
5800                              NULL);
5801     unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
5802                              ofproto_unixctl_dpif_dump_flows, NULL);
5803     unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
5804                              ofproto_unixctl_dpif_del_flows, NULL);
5805     unixctl_command_register("dpif/dump-megaflows", "bridge", 1, 1,
5806                              ofproto_unixctl_dpif_dump_megaflows, NULL);
5807     unixctl_command_register("dpif/disable-megaflows", "", 0, 0,
5808                              ofproto_unixctl_dpif_disable_megaflows, NULL);
5809     unixctl_command_register("dpif/enable-megaflows", "", 0, 0,
5810                              ofproto_unixctl_dpif_enable_megaflows, NULL);
5811 }
5812 \f
5813 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5814  *
5815  * This is deprecated.  It is only for compatibility with broken device drivers
5816  * in old versions of Linux that do not properly support VLANs when VLAN
5817  * devices are not used.  When broken device drivers are no longer in
5818  * widespread use, we will delete these interfaces. */
5819
5820 static int
5821 set_realdev(struct ofport *ofport_, ofp_port_t realdev_ofp_port, int vid)
5822 {
5823     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
5824     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
5825
5826     if (realdev_ofp_port == ofport->realdev_ofp_port
5827         && vid == ofport->vlandev_vid) {
5828         return 0;
5829     }
5830
5831     ofproto->backer->need_revalidate = REV_RECONFIGURE;
5832
5833     if (ofport->realdev_ofp_port) {
5834         vsp_remove(ofport);
5835     }
5836     if (realdev_ofp_port && ofport->bundle) {
5837         /* vlandevs are enslaved to their realdevs, so they are not allowed to
5838          * themselves be part of a bundle. */
5839         bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
5840     }
5841
5842     ofport->realdev_ofp_port = realdev_ofp_port;
5843     ofport->vlandev_vid = vid;
5844
5845     if (realdev_ofp_port) {
5846         vsp_add(ofport, realdev_ofp_port, vid);
5847     }
5848
5849     return 0;
5850 }
5851
5852 static uint32_t
5853 hash_realdev_vid(ofp_port_t realdev_ofp_port, int vid)
5854 {
5855     return hash_2words(ofp_to_u16(realdev_ofp_port), vid);
5856 }
5857
5858 bool
5859 ofproto_has_vlan_splinters(const struct ofproto_dpif *ofproto)
5860     OVS_EXCLUDED(ofproto->vsp_mutex)
5861 {
5862     bool ret;
5863
5864     ovs_mutex_lock(&ofproto->vsp_mutex);
5865     ret = !hmap_is_empty(&ofproto->realdev_vid_map);
5866     ovs_mutex_unlock(&ofproto->vsp_mutex);
5867     return ret;
5868 }
5869
5870 static ofp_port_t
5871 vsp_realdev_to_vlandev__(const struct ofproto_dpif *ofproto,
5872                          ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5873     OVS_REQUIRES(ofproto->vsp_mutex)
5874 {
5875     if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
5876         int vid = vlan_tci_to_vid(vlan_tci);
5877         const struct vlan_splinter *vsp;
5878
5879         HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
5880                                  hash_realdev_vid(realdev_ofp_port, vid),
5881                                  &ofproto->realdev_vid_map) {
5882             if (vsp->realdev_ofp_port == realdev_ofp_port
5883                 && vsp->vid == vid) {
5884                 return vsp->vlandev_ofp_port;
5885             }
5886         }
5887     }
5888     return realdev_ofp_port;
5889 }
5890
5891 /* Returns the OFP port number of the Linux VLAN device that corresponds to
5892  * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
5893  * 'struct ofport_dpif'.  For example, given 'realdev_ofp_port' of eth0 and
5894  * 'vlan_tci' 9, it would return the port number of eth0.9.
5895  *
5896  * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
5897  * function just returns its 'realdev_ofp_port' argument. */
5898 ofp_port_t
5899 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
5900                        ofp_port_t realdev_ofp_port, ovs_be16 vlan_tci)
5901     OVS_EXCLUDED(ofproto->vsp_mutex)
5902 {
5903     ofp_port_t ret;
5904
5905     ovs_mutex_lock(&ofproto->vsp_mutex);
5906     ret = vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, vlan_tci);
5907     ovs_mutex_unlock(&ofproto->vsp_mutex);
5908     return ret;
5909 }
5910
5911 static struct vlan_splinter *
5912 vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port)
5913 {
5914     struct vlan_splinter *vsp;
5915
5916     HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node,
5917                              hash_ofp_port(vlandev_ofp_port),
5918                              &ofproto->vlandev_map) {
5919         if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
5920             return vsp;
5921         }
5922     }
5923
5924     return NULL;
5925 }
5926
5927 /* Returns the OpenFlow port number of the "real" device underlying the Linux
5928  * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
5929  * VLAN VID of the Linux VLAN device in '*vid'.  For example, given
5930  * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
5931  * eth0 and store 9 in '*vid'.
5932  *
5933  * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
5934  * VLAN device.  Unless VLAN splinters are enabled, this is what this function
5935  * always does.*/
5936 static ofp_port_t
5937 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
5938                        ofp_port_t vlandev_ofp_port, int *vid)
5939     OVS_REQUIRES(ofproto->vsp_mutex)
5940 {
5941     if (!hmap_is_empty(&ofproto->vlandev_map)) {
5942         const struct vlan_splinter *vsp;
5943
5944         vsp = vlandev_find(ofproto, vlandev_ofp_port);
5945         if (vsp) {
5946             if (vid) {
5947                 *vid = vsp->vid;
5948             }
5949             return vsp->realdev_ofp_port;
5950         }
5951     }
5952     return 0;
5953 }
5954
5955 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
5956  * whether 'flow->in_port' represents a Linux VLAN device.  If so, changes
5957  * 'flow->in_port' to the "real" device backing the VLAN device, sets
5958  * 'flow->vlan_tci' to the VLAN VID, and returns true.  Otherwise (which is
5959  * always the case unless VLAN splinters are enabled), returns false without
5960  * making any changes. */
5961 bool
5962 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
5963     OVS_EXCLUDED(ofproto->vsp_mutex)
5964 {
5965     ofp_port_t realdev;
5966     int vid;
5967
5968     ovs_mutex_lock(&ofproto->vsp_mutex);
5969     realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port.ofp_port, &vid);
5970     ovs_mutex_unlock(&ofproto->vsp_mutex);
5971     if (!realdev) {
5972         return false;
5973     }
5974
5975     /* Cause the flow to be processed as if it came in on the real device with
5976      * the VLAN device's VLAN ID. */
5977     flow->in_port.ofp_port = realdev;
5978     flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
5979     return true;
5980 }
5981
5982 static void
5983 vsp_remove(struct ofport_dpif *port)
5984 {
5985     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
5986     struct vlan_splinter *vsp;
5987
5988     ovs_mutex_lock(&ofproto->vsp_mutex);
5989     vsp = vlandev_find(ofproto, port->up.ofp_port);
5990     if (vsp) {
5991         hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
5992         hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
5993         free(vsp);
5994
5995         port->realdev_ofp_port = 0;
5996     } else {
5997         VLOG_ERR("missing vlan device record");
5998     }
5999     ovs_mutex_unlock(&ofproto->vsp_mutex);
6000 }
6001
6002 static void
6003 vsp_add(struct ofport_dpif *port, ofp_port_t realdev_ofp_port, int vid)
6004 {
6005     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
6006
6007     ovs_mutex_lock(&ofproto->vsp_mutex);
6008     if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
6009         && (vsp_realdev_to_vlandev__(ofproto, realdev_ofp_port, htons(vid))
6010             == realdev_ofp_port)) {
6011         struct vlan_splinter *vsp;
6012
6013         vsp = xmalloc(sizeof *vsp);
6014         vsp->realdev_ofp_port = realdev_ofp_port;
6015         vsp->vlandev_ofp_port = port->up.ofp_port;
6016         vsp->vid = vid;
6017
6018         port->realdev_ofp_port = realdev_ofp_port;
6019
6020         hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
6021                     hash_ofp_port(port->up.ofp_port));
6022         hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
6023                     hash_realdev_vid(realdev_ofp_port, vid));
6024     } else {
6025         VLOG_ERR("duplicate vlan device record");
6026     }
6027     ovs_mutex_unlock(&ofproto->vsp_mutex);
6028 }
6029
6030 static odp_port_t
6031 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port)
6032 {
6033     const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
6034     return ofport ? ofport->odp_port : ODPP_NONE;
6035 }
6036
6037 struct ofport_dpif *
6038 odp_port_to_ofport(const struct dpif_backer *backer, odp_port_t odp_port)
6039 {
6040     struct ofport_dpif *port;
6041
6042     ovs_rwlock_rdlock(&backer->odp_to_ofport_lock);
6043     HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, hash_odp_port(odp_port),
6044                              &backer->odp_to_ofport_map) {
6045         if (port->odp_port == odp_port) {
6046             ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6047             return port;
6048         }
6049     }
6050
6051     ovs_rwlock_unlock(&backer->odp_to_ofport_lock);
6052     return NULL;
6053 }
6054
6055 static ofp_port_t
6056 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port)
6057 {
6058     struct ofport_dpif *port;
6059
6060     port = odp_port_to_ofport(ofproto->backer, odp_port);
6061     if (port && &ofproto->up == port->up.ofproto) {
6062         return port->up.ofp_port;
6063     } else {
6064         return OFPP_NONE;
6065     }
6066 }
6067
6068 const struct ofproto_class ofproto_dpif_class = {
6069     init,
6070     enumerate_types,
6071     enumerate_names,
6072     del,
6073     port_open_type,
6074     type_run,
6075     type_run_fast,
6076     type_wait,
6077     alloc,
6078     construct,
6079     destruct,
6080     dealloc,
6081     run,
6082     run_fast,
6083     wait,
6084     get_memory_usage,
6085     flush,
6086     get_features,
6087     get_tables,
6088     port_alloc,
6089     port_construct,
6090     port_destruct,
6091     port_dealloc,
6092     port_modified,
6093     port_reconfigured,
6094     port_query_by_name,
6095     port_add,
6096     port_del,
6097     port_get_stats,
6098     port_dump_start,
6099     port_dump_next,
6100     port_dump_done,
6101     port_poll,
6102     port_poll_wait,
6103     port_is_lacp_current,
6104     NULL,                       /* rule_choose_table */
6105     rule_alloc,
6106     rule_construct,
6107     rule_insert,
6108     rule_delete,
6109     rule_destruct,
6110     rule_dealloc,
6111     rule_get_stats,
6112     rule_execute,
6113     rule_modify_actions,
6114     set_frag_handling,
6115     packet_out,
6116     set_netflow,
6117     get_netflow_ids,
6118     set_sflow,
6119     set_ipfix,
6120     set_cfm,
6121     get_cfm_status,
6122     set_bfd,
6123     get_bfd_status,
6124     set_stp,
6125     get_stp_status,
6126     set_stp_port,
6127     get_stp_port_status,
6128     set_queues,
6129     bundle_set,
6130     bundle_remove,
6131     mirror_set__,
6132     mirror_get_stats__,
6133     set_flood_vlans,
6134     is_mirror_output_bundle,
6135     forward_bpdu_changed,
6136     set_mac_table_config,
6137     set_realdev,
6138     NULL,                       /* meter_get_features */
6139     NULL,                       /* meter_set */
6140     NULL,                       /* meter_get */
6141     NULL,                       /* meter_del */
6142     NULL,                       /* group_alloc */
6143     NULL,                       /* group_construct */
6144     NULL,                       /* group_destruct */
6145     NULL,                       /* group_dealloc */
6146     NULL,                       /* group_modify */
6147     NULL,                       /* group_get_stats */
6148 };