ofproto/ofproto-dpif-xlate.c

   1 /* Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
   2  *
   3  * Licensed under the Apache License, Version 2.0 (the "License");
   4  * you may not use this file except in compliance with the License.
   5  * You may obtain a copy of the License at:
   6  *
   7  *     http://www.apache.org/licenses/LICENSE-2.0
   8  *
   9  * Unless required by applicable law or agreed to in writing, software
  10  * distributed under the License is distributed on an "AS IS" BASIS,
  11  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  12  * See the License for the specific language governing permissions and
  13  * limitations under the License. */
  14
  15 #include <config.h>
  16
  17 #include "ofproto/ofproto-dpif-xlate.h"
  18
  19 #include <errno.h>
  20
  21 #include "bfd.h"
  22 #include "bitmap.h"
  23 #include "bond.h"
  24 #include "bundle.h"
  25 #include "byte-order.h"
  26 #include "cfm.h"
  27 #include "connmgr.h"
  28 #include "coverage.h"
  29 #include "dpif.h"
  30 #include "dynamic-string.h"
  31 #include "in-band.h"
  32 #include "lacp.h"
  33 #include "learn.h"
  34 #include "list.h"
  35 #include "mac-learning.h"
  36 #include "meta-flow.h"
  37 #include "multipath.h"
  38 #include "netdev-vport.h"
  39 #include "netlink.h"
  40 #include "nx-match.h"
  41 #include "odp-execute.h"
  42 #include "ofp-actions.h"
  43 #include "ofproto/ofproto-dpif-ipfix.h"
  44 #include "ofproto/ofproto-dpif-mirror.h"
  45 #include "ofproto/ofproto-dpif-monitor.h"
  46 #include "ofproto/ofproto-dpif-sflow.h"
  47 #include "ofproto/ofproto-dpif.h"
  48 #include "ofproto/ofproto-provider.h"
  49 #include "tunnel.h"
  50 #include "vlog.h"
  51
  52 COVERAGE_DEFINE(xlate_actions);
  53 COVERAGE_DEFINE(xlate_actions_oversize);
  54 COVERAGE_DEFINE(xlate_actions_mpls_overflow);
  55
  56 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate);
  57
  58 /* Maximum depth of flow table recursion (due to resubmit actions) in a
  59  * flow translation. */
  60 #define MAX_RESUBMIT_RECURSION 64
  61
  62 /* Maximum number of resubmit actions in a flow translation, whether they are
  63  * recursive or not. */
  64 #define MAX_RESUBMITS (MAX_RESUBMIT_RECURSION * MAX_RESUBMIT_RECURSION)
  65
  66 struct ovs_rwlock xlate_rwlock = OVS_RWLOCK_INITIALIZER;
  67
  68 struct xbridge {
  69     struct hmap_node hmap_node;   /* Node in global 'xbridges' map. */
  70     struct ofproto_dpif *ofproto; /* Key in global 'xbridges' map. */
  71
  72     struct list xbundles;         /* Owned xbundles. */
  73     struct hmap xports;           /* Indexed by ofp_port. */
  74
  75     char *name;                   /* Name used in log messages. */
  76     struct dpif *dpif;            /* Datapath interface. */
  77     struct mac_learning *ml;      /* Mac learning handle. */
  78     struct mbridge *mbridge;      /* Mirroring. */
  79     struct dpif_sflow *sflow;     /* SFlow handle, or null. */
  80     struct dpif_ipfix *ipfix;     /* Ipfix handle, or null. */
  81     struct netflow *netflow;      /* Netflow handle, or null. */
  82     struct stp *stp;              /* STP or null if disabled. */
  83
  84     /* Special rules installed by ofproto-dpif. */
  85     struct rule_dpif *miss_rule;
  86     struct rule_dpif *no_packet_in_rule;
  87
  88     enum ofp_config_flags frag;   /* Fragmentation handling. */
  89     bool has_in_band;             /* Bridge has in band control? */
  90     bool forward_bpdu;            /* Bridge forwards STP BPDUs? */
  91
  92     /* True if the datapath supports variable-length
  93      * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
  94      * False if the datapath supports only 8-byte (or shorter) userdata. */
  95     bool variable_length_userdata;
  96
  97     /* Number of MPLS label stack entries that the datapath supports
  98      * in matches. */
  99     size_t max_mpls_depth;
 100 };
 101
 102 struct xbundle {
 103     struct hmap_node hmap_node;    /* In global 'xbundles' map. */
 104     struct ofbundle *ofbundle;     /* Key in global 'xbundles' map. */
 105
 106     struct list list_node;         /* In parent 'xbridges' list. */
 107     struct xbridge *xbridge;       /* Parent xbridge. */
 108
 109     struct list xports;            /* Contains "struct xport"s. */
 110
 111     char *name;                    /* Name used in log messages. */
 112     struct bond *bond;             /* Nonnull iff more than one port. */
 113     struct lacp *lacp;             /* LACP handle or null. */
 114
 115     enum port_vlan_mode vlan_mode; /* VLAN mode. */
 116     int vlan;                      /* -1=trunk port, else a 12-bit VLAN ID. */
 117     unsigned long *trunks;         /* Bitmap of trunked VLANs, if 'vlan' == -1.
 118                                     * NULL if all VLANs are trunked. */
 119     bool use_priority_tags;        /* Use 802.1p tag for frames in VLAN 0? */
 120     bool floodable;                /* No port has OFPUTIL_PC_NO_FLOOD set? */
 121 };
 122
 123 struct xport {
 124     struct hmap_node hmap_node;      /* Node in global 'xports' map. */
 125     struct ofport_dpif *ofport;      /* Key in global 'xports map. */
 126
 127     struct hmap_node ofp_node;       /* Node in parent xbridge 'xports' map. */
 128     ofp_port_t ofp_port;             /* Key in parent xbridge 'xports' map. */
 129
 130     odp_port_t odp_port;             /* Datapath port number or ODPP_NONE. */
 131
 132     struct list bundle_node;         /* In parent xbundle (if it exists). */
 133     struct xbundle *xbundle;         /* Parent xbundle or null. */
 134
 135     struct netdev *netdev;           /* 'ofport''s netdev. */
 136
 137     struct xbridge *xbridge;         /* Parent bridge. */
 138     struct xport *peer;              /* Patch port peer or null. */
 139
 140     enum ofputil_port_config config; /* OpenFlow port configuration. */
 141     enum ofputil_port_state state;   /* OpenFlow port state. */
 142     int stp_port_no;                 /* STP port number or -1 if not in use. */
 143
 144     struct hmap skb_priorities;      /* Map of 'skb_priority_to_dscp's. */
 145
 146     bool may_enable;                 /* May be enabled in bonds. */
 147     bool is_tunnel;                  /* Is a tunnel port. */
 148
 149     struct cfm *cfm;                 /* CFM handle or null. */
 150     struct bfd *bfd;                 /* BFD handle or null. */
 151 };
 152
 153 struct xlate_ctx {
 154     struct xlate_in *xin;
 155     struct xlate_out *xout;
 156
 157     const struct xbridge *xbridge;
 158
 159     /* Flow at the last commit. */
 160     struct flow base_flow;
 161
 162     /* Tunnel IP destination address as received.  This is stored separately
 163      * as the base_flow.tunnel is cleared on init to reflect the datapath
 164      * behavior.  Used to make sure not to send tunneled output to ourselves,
 165      * which might lead to an infinite loop.  This could happen easily
 166      * if a tunnel is marked as 'ip_remote=flow', and the flow does not
 167      * actually set the tun_dst field. */
 168     ovs_be32 orig_tunnel_ip_dst;
 169
 170     /* Stack for the push and pop actions.  Each stack element is of type
 171      * "union mf_subvalue". */
 172     union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
 173     struct ofpbuf stack;
 174
 175     /* The rule that we are currently translating, or NULL. */
 176     struct rule_dpif *rule;
 177
 178     /* Resubmit statistics, via xlate_table_action(). */
 179     int recurse;                /* Current resubmit nesting depth. */
 180     int resubmits;              /* Total number of resubmits. */
 181     bool in_group;              /* Currently translating ofgroup, if true. */
 182
 183     uint32_t orig_skb_priority; /* Priority when packet arrived. */
 184     uint8_t table_id;           /* OpenFlow table ID where flow was found. */
 185     uint32_t sflow_n_outputs;   /* Number of output ports. */
 186     odp_port_t sflow_odp_port;  /* Output port for composing sFlow action. */
 187     uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
 188     bool exit;                  /* No further actions should be processed. */
 189
 190     /* OpenFlow 1.1+ action set.
 191      *
 192      * 'action_set' accumulates "struct ofpact"s added by OFPACT_WRITE_ACTIONS.
 193      * When translation is otherwise complete, ofpacts_execute_action_set()
 194      * converts it to a set of "struct ofpact"s that can be translated into
 195      * datapath actions.   */
 196     struct ofpbuf action_set;   /* Action set. */
 197     uint64_t action_set_stub[1024 / 8];
 198 };
 199
 200 /* A controller may use OFPP_NONE as the ingress port to indicate that
 201  * it did not arrive on a "real" port.  'ofpp_none_bundle' exists for
 202  * when an input bundle is needed for validation (e.g., mirroring or
 203  * OFPP_NORMAL processing).  It is not connected to an 'ofproto' or have
 204  * any 'port' structs, so care must be taken when dealing with it.
 205  * The bundle's name and vlan mode are initialized in lookup_input_bundle() */
 206 static struct xbundle ofpp_none_bundle;
 207
 208 /* Node in 'xport''s 'skb_priorities' map.  Used to maintain a map from
 209  * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
 210  * traffic egressing the 'ofport' with that priority should be marked with. */
 211 struct skb_priority_to_dscp {
 212     struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'skb_priorities'. */
 213     uint32_t skb_priority;      /* Priority of this queue (see struct flow). */
 214
 215     uint8_t dscp;               /* DSCP bits to mark outgoing traffic with. */
 216 };
 217
 218 static struct hmap xbridges = HMAP_INITIALIZER(&xbridges);
 219 static struct hmap xbundles = HMAP_INITIALIZER(&xbundles);
 220 static struct hmap xports = HMAP_INITIALIZER(&xports);
 221
 222 static bool may_receive(const struct xport *, struct xlate_ctx *);
 223 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
 224                              struct xlate_ctx *);
 225 static void xlate_actions__(struct xlate_in *, struct xlate_out *)
 226     OVS_REQ_RDLOCK(xlate_rwlock);
 227     static void xlate_normal(struct xlate_ctx *);
 228     static void xlate_report(struct xlate_ctx *, const char *);
 229     static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
 230                                    uint8_t table_id, bool may_packet_in);
 231 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
 232 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
 233 static void output_normal(struct xlate_ctx *, const struct xbundle *,
 234                           uint16_t vlan);
 235 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port);
 236
 237 static struct xbridge *xbridge_lookup(const struct ofproto_dpif *);
 238 static struct xbundle *xbundle_lookup(const struct ofbundle *);
 239 static struct xport *xport_lookup(const struct ofport_dpif *);
 240 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
 241 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
 242                                                      uint32_t skb_priority);
 243 static void clear_skb_priorities(struct xport *);
 244 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
 245                                    uint8_t *dscp);
 246
 247 void
 248 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
 249                   struct dpif *dpif, struct rule_dpif *miss_rule,
 250                   struct rule_dpif *no_packet_in_rule,
 251                   const struct mac_learning *ml, struct stp *stp,
 252                   const struct mbridge *mbridge,
 253                   const struct dpif_sflow *sflow,
 254                   const struct dpif_ipfix *ipfix,
 255                   const struct netflow *netflow, enum ofp_config_flags frag,
 256                   bool forward_bpdu, bool has_in_band,
 257                   bool variable_length_userdata,
 258                   size_t max_mpls_depth)
 259 {
 260     struct xbridge *xbridge = xbridge_lookup(ofproto);
 261
 262     if (!xbridge) {
 263         xbridge = xzalloc(sizeof *xbridge);
 264         xbridge->ofproto = ofproto;
 265
 266         hmap_insert(&xbridges, &xbridge->hmap_node, hash_pointer(ofproto, 0));
 267         hmap_init(&xbridge->xports);
 268         list_init(&xbridge->xbundles);
 269     }
 270
 271     if (xbridge->ml != ml) {
 272         mac_learning_unref(xbridge->ml);
 273         xbridge->ml = mac_learning_ref(ml);
 274     }
 275
 276     if (xbridge->mbridge != mbridge) {
 277         mbridge_unref(xbridge->mbridge);
 278         xbridge->mbridge = mbridge_ref(mbridge);
 279     }
 280
 281     if (xbridge->sflow != sflow) {
 282         dpif_sflow_unref(xbridge->sflow);
 283         xbridge->sflow = dpif_sflow_ref(sflow);
 284     }
 285
 286     if (xbridge->ipfix != ipfix) {
 287         dpif_ipfix_unref(xbridge->ipfix);
 288         xbridge->ipfix = dpif_ipfix_ref(ipfix);
 289     }
 290
 291     if (xbridge->stp != stp) {
 292         stp_unref(xbridge->stp);
 293         xbridge->stp = stp_ref(stp);
 294     }
 295
 296     if (xbridge->netflow != netflow) {
 297         netflow_unref(xbridge->netflow);
 298         xbridge->netflow = netflow_ref(netflow);
 299     }
 300
 301     free(xbridge->name);
 302     xbridge->name = xstrdup(name);
 303
 304     xbridge->dpif = dpif;
 305     xbridge->forward_bpdu = forward_bpdu;
 306     xbridge->has_in_band = has_in_band;
 307     xbridge->frag = frag;
 308     xbridge->miss_rule = miss_rule;
 309     xbridge->no_packet_in_rule = no_packet_in_rule;
 310     xbridge->variable_length_userdata = variable_length_userdata;
 311     xbridge->max_mpls_depth = max_mpls_depth;
 312 }
 313
 314 void
 315 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
 316 {
 317     struct xbridge *xbridge = xbridge_lookup(ofproto);
 318     struct xbundle *xbundle, *next_xbundle;
 319     struct xport *xport, *next_xport;
 320
 321     if (!xbridge) {
 322         return;
 323     }
 324
 325     HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
 326         xlate_ofport_remove(xport->ofport);
 327     }
 328
 329     LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
 330         xlate_bundle_remove(xbundle->ofbundle);
 331     }
 332
 333     hmap_remove(&xbridges, &xbridge->hmap_node);
 334     mac_learning_unref(xbridge->ml);
 335     mbridge_unref(xbridge->mbridge);
 336     dpif_sflow_unref(xbridge->sflow);
 337     dpif_ipfix_unref(xbridge->ipfix);
 338     stp_unref(xbridge->stp);
 339     hmap_destroy(&xbridge->xports);
 340     free(xbridge->name);
 341     free(xbridge);
 342 }
 343
 344 void
 345 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
 346                  const char *name, enum port_vlan_mode vlan_mode, int vlan,
 347                  unsigned long *trunks, bool use_priority_tags,
 348                  const struct bond *bond, const struct lacp *lacp,
 349                  bool floodable)
 350 {
 351     struct xbundle *xbundle = xbundle_lookup(ofbundle);
 352
 353     if (!xbundle) {
 354         xbundle = xzalloc(sizeof *xbundle);
 355         xbundle->ofbundle = ofbundle;
 356         xbundle->xbridge = xbridge_lookup(ofproto);
 357
 358         hmap_insert(&xbundles, &xbundle->hmap_node, hash_pointer(ofbundle, 0));
 359         list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
 360         list_init(&xbundle->xports);
 361     }
 362
 363     ovs_assert(xbundle->xbridge);
 364
 365     free(xbundle->name);
 366     xbundle->name = xstrdup(name);
 367
 368     xbundle->vlan_mode = vlan_mode;
 369     xbundle->vlan = vlan;
 370     xbundle->trunks = trunks;
 371     xbundle->use_priority_tags = use_priority_tags;
 372     xbundle->floodable = floodable;
 373
 374     if (xbundle->bond != bond) {
 375         bond_unref(xbundle->bond);
 376         xbundle->bond = bond_ref(bond);
 377     }
 378
 379     if (xbundle->lacp != lacp) {
 380         lacp_unref(xbundle->lacp);
 381         xbundle->lacp = lacp_ref(lacp);
 382     }
 383 }
 384
 385 void
 386 xlate_bundle_remove(struct ofbundle *ofbundle)
 387 {
 388     struct xbundle *xbundle = xbundle_lookup(ofbundle);
 389     struct xport *xport, *next;
 390
 391     if (!xbundle) {
 392         return;
 393     }
 394
 395     LIST_FOR_EACH_SAFE (xport, next, bundle_node, &xbundle->xports) {
 396         list_remove(&xport->bundle_node);
 397         xport->xbundle = NULL;
 398     }
 399
 400     hmap_remove(&xbundles, &xbundle->hmap_node);
 401     list_remove(&xbundle->list_node);
 402     bond_unref(xbundle->bond);
 403     lacp_unref(xbundle->lacp);
 404     free(xbundle->name);
 405     free(xbundle);
 406 }
 407
 408 void
 409 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
 410                  struct ofport_dpif *ofport, ofp_port_t ofp_port,
 411                  odp_port_t odp_port, const struct netdev *netdev,
 412                  const struct cfm *cfm, const struct bfd *bfd,
 413                  struct ofport_dpif *peer, int stp_port_no,
 414                  const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
 415                  enum ofputil_port_config config,
 416                  enum ofputil_port_state state, bool is_tunnel,
 417                  bool may_enable)
 418 {
 419     struct xport *xport = xport_lookup(ofport);
 420     size_t i;
 421
 422     if (!xport) {
 423         xport = xzalloc(sizeof *xport);
 424         xport->ofport = ofport;
 425         xport->xbridge = xbridge_lookup(ofproto);
 426         xport->ofp_port = ofp_port;
 427
 428         hmap_init(&xport->skb_priorities);
 429         hmap_insert(&xports, &xport->hmap_node, hash_pointer(ofport, 0));
 430         hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
 431                     hash_ofp_port(xport->ofp_port));
 432     }
 433
 434     ovs_assert(xport->ofp_port == ofp_port);
 435
 436     xport->config = config;
 437     xport->state = state;
 438     xport->stp_port_no = stp_port_no;
 439     xport->is_tunnel = is_tunnel;
 440     xport->may_enable = may_enable;
 441     xport->odp_port = odp_port;
 442
 443     if (xport->netdev != netdev) {
 444         netdev_close(xport->netdev);
 445         xport->netdev = netdev_ref(netdev);
 446     }
 447
 448     if (xport->cfm != cfm) {
 449         cfm_unref(xport->cfm);
 450         xport->cfm = cfm_ref(cfm);
 451     }
 452
 453     if (xport->bfd != bfd) {
 454         bfd_unref(xport->bfd);
 455         xport->bfd = bfd_ref(bfd);
 456     }
 457
 458     if (xport->peer) {
 459         xport->peer->peer = NULL;
 460     }
 461     xport->peer = xport_lookup(peer);
 462     if (xport->peer) {
 463         xport->peer->peer = xport;
 464     }
 465
 466     if (xport->xbundle) {
 467         list_remove(&xport->bundle_node);
 468     }
 469     xport->xbundle = xbundle_lookup(ofbundle);
 470     if (xport->xbundle) {
 471         list_insert(&xport->xbundle->xports, &xport->bundle_node);
 472     }
 473
 474     clear_skb_priorities(xport);
 475     for (i = 0; i < n_qdscp; i++) {
 476         struct skb_priority_to_dscp *pdscp;
 477         uint32_t skb_priority;
 478
 479         if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
 480                                    &skb_priority)) {
 481             continue;
 482         }
 483
 484         pdscp = xmalloc(sizeof *pdscp);
 485         pdscp->skb_priority = skb_priority;
 486         pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
 487         hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
 488                     hash_int(pdscp->skb_priority, 0));
 489     }
 490 }
 491
 492 void
 493 xlate_ofport_remove(struct ofport_dpif *ofport)
 494 {
 495     struct xport *xport = xport_lookup(ofport);
 496
 497     if (!xport) {
 498         return;
 499     }
 500
 501     if (xport->peer) {
 502         xport->peer->peer = NULL;
 503         xport->peer = NULL;
 504     }
 505
 506     if (xport->xbundle) {
 507         list_remove(&xport->bundle_node);
 508     }
 509
 510     clear_skb_priorities(xport);
 511     hmap_destroy(&xport->skb_priorities);
 512
 513     hmap_remove(&xports, &xport->hmap_node);
 514     hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
 515
 516     netdev_close(xport->netdev);
 517     cfm_unref(xport->cfm);
 518     bfd_unref(xport->bfd);
 519     free(xport);
 520 }
 521
 522 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
 523  * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
 524  * Optionally populates 'ofproto' with the ofproto_dpif, 'odp_in_port' with
 525  * the datapath in_port, that 'packet' ingressed, and 'ipfix', 'sflow', and
 526  * 'netflow' with the appropriate handles for those protocols if they're
 527  * enabled.  Caller is responsible for unrefing them.
 528  *
 529  * If 'ofproto' is nonnull, requires 'flow''s in_port to exist.  Otherwise sets
 530  * 'flow''s in_port to OFPP_NONE.
 531  *
 532  * This function does post-processing on data returned from
 533  * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
 534  * of the upcall processing logic.  In particular, if the extracted in_port is
 535  * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
 536  * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
 537  * a VLAN header onto 'packet' (if it is nonnull).
 538  *
 539  * Similarly, this function also includes some logic to help with tunnels.  It
 540  * may modify 'flow' as necessary to make the tunneling implementation
 541  * transparent to the upcall processing logic.
 542  *
 543  * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
 544  * or some other positive errno if there are other problems. */
 545 int
 546 xlate_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
 547               const struct nlattr *key, size_t key_len, struct flow *flow,
 548               struct ofproto_dpif **ofproto, struct dpif_ipfix **ipfix,
 549               struct dpif_sflow **sflow, struct netflow **netflow,
 550               odp_port_t *odp_in_port)
 551 {
 552     const struct xport *xport;
 553     int error = ENODEV;
 554
 555     ovs_rwlock_rdlock(&xlate_rwlock);
 556     if (odp_flow_key_to_flow(key, key_len, flow) == ODP_FIT_ERROR) {
 557         error = EINVAL;
 558         goto exit;
 559     }
 560
 561     if (odp_in_port) {
 562         *odp_in_port = flow->in_port.odp_port;
 563     }
 564
 565     xport = xport_lookup(tnl_port_should_receive(flow)
 566                          ? tnl_port_receive(flow)
 567                          : odp_port_to_ofport(backer, flow->in_port.odp_port));
 568
 569     flow->in_port.ofp_port = xport ? xport->ofp_port : OFPP_NONE;
 570     if (!xport) {
 571         goto exit;
 572     }
 573
 574     if (vsp_adjust_flow(xport->xbridge->ofproto, flow)) {
 575         if (packet) {
 576             /* Make the packet resemble the flow, so that it gets sent to
 577              * an OpenFlow controller properly, so that it looks correct
 578              * for sFlow, and so that flow_extract() will get the correct
 579              * vlan_tci if it is called on 'packet'. */
 580             eth_push_vlan(packet, htons(ETH_TYPE_VLAN), flow->vlan_tci);
 581         }
 582     }
 583     error = 0;
 584
 585     if (ofproto) {
 586         *ofproto = xport->xbridge->ofproto;
 587     }
 588
 589     if (ipfix) {
 590         *ipfix = dpif_ipfix_ref(xport->xbridge->ipfix);
 591     }
 592
 593     if (sflow) {
 594         *sflow = dpif_sflow_ref(xport->xbridge->sflow);
 595     }
 596
 597     if (netflow) {
 598         *netflow = netflow_ref(xport->xbridge->netflow);
 599     }
 600
 601 exit:
 602     ovs_rwlock_unlock(&xlate_rwlock);
 603     return error;
 604 }
 605
 606 static struct xbridge *
 607 xbridge_lookup(const struct ofproto_dpif *ofproto)
 608 {
 609     struct xbridge *xbridge;
 610
 611     if (!ofproto) {
 612         return NULL;
 613     }
 614
 615     HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
 616                              &xbridges) {
 617         if (xbridge->ofproto == ofproto) {
 618             return xbridge;
 619         }
 620     }
 621     return NULL;
 622 }
 623
 624 static struct xbundle *
 625 xbundle_lookup(const struct ofbundle *ofbundle)
 626 {
 627     struct xbundle *xbundle;
 628
 629     if (!ofbundle) {
 630         return NULL;
 631     }
 632
 633     HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
 634                              &xbundles) {
 635         if (xbundle->ofbundle == ofbundle) {
 636             return xbundle;
 637         }
 638     }
 639     return NULL;
 640 }
 641
 642 static struct xport *
 643 xport_lookup(const struct ofport_dpif *ofport)
 644 {
 645     struct xport *xport;
 646
 647     if (!ofport) {
 648         return NULL;
 649     }
 650
 651     HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
 652                              &xports) {
 653         if (xport->ofport == ofport) {
 654             return xport;
 655         }
 656     }
 657     return NULL;
 658 }
 659
 660 static struct stp_port *
 661 xport_get_stp_port(const struct xport *xport)
 662 {
 663     return xport->xbridge->stp && xport->stp_port_no != -1
 664         ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
 665         : NULL;
 666 }
 667
 668 static bool
 669 xport_stp_learn_state(const struct xport *xport)
 670 {
 671     struct stp_port *sp = xport_get_stp_port(xport);
 672     return stp_learn_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
 673 }
 674
 675 static bool
 676 xport_stp_forward_state(const struct xport *xport)
 677 {
 678     struct stp_port *sp = xport_get_stp_port(xport);
 679     return stp_forward_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
 680 }
 681
 682 static bool
 683 xport_stp_listen_state(const struct xport *xport)
 684 {
 685     struct stp_port *sp = xport_get_stp_port(xport);
 686     return stp_listen_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
 687 }
 688
 689 /* Returns true if STP should process 'flow'.  Sets fields in 'wc' that
 690  * were used to make the determination.*/
 691 static bool
 692 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
 693 {
 694     memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
 695     return eth_addr_equals(flow->dl_dst, eth_addr_stp);
 696 }
 697
 698 static void
 699 stp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
 700 {
 701     struct stp_port *sp = xport_get_stp_port(xport);
 702     struct ofpbuf payload = *packet;
 703     struct eth_header *eth = payload.data;
 704
 705     /* Sink packets on ports that have STP disabled when the bridge has
 706      * STP enabled. */
 707     if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
 708         return;
 709     }
 710
 711     /* Trim off padding on payload. */
 712     if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
 713         payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
 714     }
 715
 716     if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
 717         stp_received_bpdu(sp, payload.data, payload.size);
 718     }
 719 }
 720
 721 static struct xport *
 722 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
 723 {
 724     struct xport *xport;
 725
 726     HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
 727                              &xbridge->xports) {
 728         if (xport->ofp_port == ofp_port) {
 729             return xport;
 730         }
 731     }
 732     return NULL;
 733 }
 734
 735 static odp_port_t
 736 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
 737 {
 738     const struct xport *xport = get_ofp_port(xbridge, ofp_port);
 739     return xport ? xport->odp_port : ODPP_NONE;
 740 }
 741
 742 static bool
 743 odp_port_is_alive(const struct xlate_ctx *ctx, ofp_port_t ofp_port)
 744 {
 745     struct xport *xport;
 746
 747     xport = get_ofp_port(ctx->xbridge, ofp_port);
 748     if (!xport || xport->config & OFPUTIL_PC_PORT_DOWN ||
 749         xport->state & OFPUTIL_PS_LINK_DOWN) {
 750         return false;
 751     }
 752
 753     return true;
 754 }
 755
 756 static const struct ofputil_bucket *
 757 group_first_live_bucket(const struct xlate_ctx *, const struct group_dpif *,
 758                         int depth);
 759
 760 static bool
 761 group_is_alive(const struct xlate_ctx *ctx, uint32_t group_id, int depth)
 762 {
 763     struct group_dpif *group;
 764     bool hit;
 765
 766     hit = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
 767     if (!hit) {
 768         return false;
 769     }
 770
 771     hit = group_first_live_bucket(ctx, group, depth) != NULL;
 772
 773     group_dpif_release(group);
 774     return hit;
 775 }
 776
 777 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
 778
 779 static bool
 780 bucket_is_alive(const struct xlate_ctx *ctx,
 781                 const struct ofputil_bucket *bucket, int depth)
 782 {
 783     if (depth >= MAX_LIVENESS_RECURSION) {
 784         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
 785
 786         VLOG_WARN_RL(&rl, "bucket chaining exceeded %d links",
 787                      MAX_LIVENESS_RECURSION);
 788         return false;
 789     }
 790
 791     return !ofputil_bucket_has_liveness(bucket) ||
 792         (bucket->watch_port != OFPP_ANY &&
 793          odp_port_is_alive(ctx, bucket->watch_port)) ||
 794         (bucket->watch_group != OFPG_ANY &&
 795          group_is_alive(ctx, bucket->watch_group, depth + 1));
 796 }
 797
 798 static const struct ofputil_bucket *
 799 group_first_live_bucket(const struct xlate_ctx *ctx,
 800                         const struct group_dpif *group, int depth)
 801 {
 802     struct ofputil_bucket *bucket;
 803     const struct list *buckets;
 804
 805     group_dpif_get_buckets(group, &buckets);
 806     LIST_FOR_EACH (bucket, list_node, buckets) {
 807         if (bucket_is_alive(ctx, bucket, depth)) {
 808             return bucket;
 809         }
 810     }
 811
 812     return NULL;
 813 }
 814
 815 static const struct ofputil_bucket *
 816 group_best_live_bucket(const struct xlate_ctx *ctx,
 817                        const struct group_dpif *group,
 818                        uint32_t basis)
 819 {
 820     const struct ofputil_bucket *best_bucket = NULL;
 821     uint32_t best_score = 0;
 822     int i = 0;
 823
 824     const struct ofputil_bucket *bucket;
 825     const struct list *buckets;
 826
 827     group_dpif_get_buckets(group, &buckets);
 828     LIST_FOR_EACH (bucket, list_node, buckets) {
 829         if (bucket_is_alive(ctx, bucket, 0)) {
 830             uint32_t score = (hash_int(i, basis) & 0xffff) * bucket->weight;
 831             if (score >= best_score) {
 832                 best_bucket = bucket;
 833                 best_score = score;
 834             }
 835         }
 836         i++;
 837     }
 838
 839     return best_bucket;
 840 }
 841
 842 static bool
 843 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
 844 {
 845     return (bundle->vlan_mode != PORT_VLAN_ACCESS
 846             && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
 847 }
 848
 849 static bool
 850 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
 851 {
 852     return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
 853 }
 854
 855 static mirror_mask_t
 856 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
 857 {
 858     return xbundle != &ofpp_none_bundle
 859         ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
 860         : 0;
 861 }
 862
 863 static mirror_mask_t
 864 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
 865 {
 866     return xbundle != &ofpp_none_bundle
 867         ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
 868         : 0;
 869 }
 870
 871 static mirror_mask_t
 872 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
 873 {
 874     return xbundle != &ofpp_none_bundle
 875         ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
 876         : 0;
 877 }
 878
 879 static struct xbundle *
 880 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
 881                     bool warn, struct xport **in_xportp)
 882 {
 883     struct xport *xport;
 884
 885     /* Find the port and bundle for the received packet. */
 886     xport = get_ofp_port(xbridge, in_port);
 887     if (in_xportp) {
 888         *in_xportp = xport;
 889     }
 890     if (xport && xport->xbundle) {
 891         return xport->xbundle;
 892     }
 893
 894     /* Special-case OFPP_NONE, which a controller may use as the ingress
 895      * port for traffic that it is sourcing. */
 896     if (in_port == OFPP_NONE) {
 897         ofpp_none_bundle.name = "OFPP_NONE";
 898         ofpp_none_bundle.vlan_mode = PORT_VLAN_TRUNK;
 899         return &ofpp_none_bundle;
 900     }
 901
 902     /* Odd.  A few possible reasons here:
 903      *
 904      * - We deleted a port but there are still a few packets queued up
 905      *   from it.
 906      *
 907      * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
 908      *   we don't know about.
 909      *
 910      * - The ofproto client didn't configure the port as part of a bundle.
 911      *   This is particularly likely to happen if a packet was received on the
 912      *   port after it was created, but before the client had a chance to
 913      *   configure its bundle.
 914      */
 915     if (warn) {
 916         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 917
 918         VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
 919                      "port %"PRIu16, xbridge->name, in_port);
 920     }
 921     return NULL;
 922 }
 923
 924 static void
 925 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
 926 {
 927     const struct xbridge *xbridge = ctx->xbridge;
 928     mirror_mask_t mirrors;
 929     struct xbundle *in_xbundle;
 930     uint16_t vlan;
 931     uint16_t vid;
 932
 933     mirrors = ctx->xout->mirrors;
 934     ctx->xout->mirrors = 0;
 935
 936     in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
 937                                      ctx->xin->packet != NULL, NULL);
 938     if (!in_xbundle) {
 939         return;
 940     }
 941     mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
 942
 943     /* Drop frames on bundles reserved for mirroring. */
 944     if (xbundle_mirror_out(xbridge, in_xbundle)) {
 945         if (ctx->xin->packet != NULL) {
 946             static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 947             VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
 948                          "%s, which is reserved exclusively for mirroring",
 949                          ctx->xbridge->name, in_xbundle->name);
 950         }
 951         ofpbuf_clear(&ctx->xout->odp_actions);
 952         return;
 953     }
 954
 955     /* Check VLAN. */
 956     vid = vlan_tci_to_vid(orig_flow->vlan_tci);
 957     if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
 958         return;
 959     }
 960     vlan = input_vid_to_vlan(in_xbundle, vid);
 961
 962     if (!mirrors) {
 963         return;
 964     }
 965
 966     /* Restore the original packet before adding the mirror actions. */
 967     ctx->xin->flow = *orig_flow;
 968
 969     while (mirrors) {
 970         mirror_mask_t dup_mirrors;
 971         struct ofbundle *out;
 972         unsigned long *vlans;
 973         bool vlan_mirrored;
 974         bool has_mirror;
 975         int out_vlan;
 976
 977         has_mirror = mirror_get(xbridge->mbridge, raw_ctz(mirrors),
 978                                 &vlans, &dup_mirrors, &out, &out_vlan);
 979         ovs_assert(has_mirror);
 980
 981         if (vlans) {
 982             ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
 983         }
 984         vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
 985         free(vlans);
 986
 987         if (!vlan_mirrored) {
 988             mirrors = zero_rightmost_1bit(mirrors);
 989             continue;
 990         }
 991
 992         mirrors &= ~dup_mirrors;
 993         ctx->xout->mirrors |= dup_mirrors;
 994         if (out) {
 995             struct xbundle *out_xbundle = xbundle_lookup(out);
 996             if (out_xbundle) {
 997                 output_normal(ctx, out_xbundle, vlan);
 998             }
 999         } else if (vlan != out_vlan
1000                    && !eth_addr_is_reserved(orig_flow->dl_dst)) {
1001             struct xbundle *xbundle;
1002
1003             LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
1004                 if (xbundle_includes_vlan(xbundle, out_vlan)
1005                     && !xbundle_mirror_out(xbridge, xbundle)) {
1006                     output_normal(ctx, xbundle, out_vlan);
1007                 }
1008             }
1009         }
1010     }
1011 }
1012
1013 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1014  * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1015  * the bundle on which the packet was received, returns the VLAN to which the
1016  * packet belongs.
1017  *
1018  * Both 'vid' and the return value are in the range 0...4095. */
1019 static uint16_t
1020 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
1021 {
1022     switch (in_xbundle->vlan_mode) {
1023     case PORT_VLAN_ACCESS:
1024         return in_xbundle->vlan;
1025         break;
1026
1027     case PORT_VLAN_TRUNK:
1028         return vid;
1029
1030     case PORT_VLAN_NATIVE_UNTAGGED:
1031     case PORT_VLAN_NATIVE_TAGGED:
1032         return vid ? vid : in_xbundle->vlan;
1033
1034     default:
1035         OVS_NOT_REACHED();
1036     }
1037 }
1038
1039 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1040  * If so, returns true.  Otherwise, returns false and, if 'warn' is true, logs
1041  * a warning.
1042  *
1043  * 'vid' should be the VID obtained from the 802.1Q header that was received as
1044  * part of a packet (specify 0 if there was no 802.1Q header), in the range
1045  * 0...4095. */
1046 static bool
1047 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
1048 {
1049     /* Allow any VID on the OFPP_NONE port. */
1050     if (in_xbundle == &ofpp_none_bundle) {
1051         return true;
1052     }
1053
1054     switch (in_xbundle->vlan_mode) {
1055     case PORT_VLAN_ACCESS:
1056         if (vid) {
1057             if (warn) {
1058                 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1059                 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
1060                              "packet received on port %s configured as VLAN "
1061                              "%"PRIu16" access port", vid, in_xbundle->name,
1062                              in_xbundle->vlan);
1063             }
1064             return false;
1065         }
1066         return true;
1067
1068     case PORT_VLAN_NATIVE_UNTAGGED:
1069     case PORT_VLAN_NATIVE_TAGGED:
1070         if (!vid) {
1071             /* Port must always carry its native VLAN. */
1072             return true;
1073         }
1074         /* Fall through. */
1075     case PORT_VLAN_TRUNK:
1076         if (!xbundle_includes_vlan(in_xbundle, vid)) {
1077             if (warn) {
1078                 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1079                 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
1080                              "received on port %s not configured for trunking "
1081                              "VLAN %"PRIu16, vid, in_xbundle->name, vid);
1082             }
1083             return false;
1084         }
1085         return true;
1086
1087     default:
1088         OVS_NOT_REACHED();
1089     }
1090
1091 }
1092
1093 /* Given 'vlan', the VLAN that a packet belongs to, and
1094  * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1095  * that should be included in the 802.1Q header.  (If the return value is 0,
1096  * then the 802.1Q header should only be included in the packet if there is a
1097  * nonzero PCP.)
1098  *
1099  * Both 'vlan' and the return value are in the range 0...4095. */
1100 static uint16_t
1101 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
1102 {
1103     switch (out_xbundle->vlan_mode) {
1104     case PORT_VLAN_ACCESS:
1105         return 0;
1106
1107     case PORT_VLAN_TRUNK:
1108     case PORT_VLAN_NATIVE_TAGGED:
1109         return vlan;
1110
1111     case PORT_VLAN_NATIVE_UNTAGGED:
1112         return vlan == out_xbundle->vlan ? 0 : vlan;
1113
1114     default:
1115         OVS_NOT_REACHED();
1116     }
1117 }
1118
1119 static void
1120 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
1121               uint16_t vlan)
1122 {
1123     ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
1124     uint16_t vid;
1125     ovs_be16 tci, old_tci;
1126     struct xport *xport;
1127
1128     vid = output_vlan_to_vid(out_xbundle, vlan);
1129     if (list_is_empty(&out_xbundle->xports)) {
1130         /* Partially configured bundle with no slaves.  Drop the packet. */
1131         return;
1132     } else if (!out_xbundle->bond) {
1133         xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
1134                              bundle_node);
1135     } else {
1136         struct ofport_dpif *ofport;
1137
1138         ofport = bond_choose_output_slave(out_xbundle->bond, &ctx->xin->flow,
1139                                           &ctx->xout->wc, vid);
1140         xport = xport_lookup(ofport);
1141
1142         if (!xport) {
1143             /* No slaves enabled, so drop packet. */
1144             return;
1145         }
1146
1147         if (ctx->xin->resubmit_stats) {
1148             bond_account(out_xbundle->bond, &ctx->xin->flow, vid,
1149                          ctx->xin->resubmit_stats->n_bytes);
1150         }
1151     }
1152
1153     old_tci = *flow_tci;
1154     tci = htons(vid);
1155     if (tci || out_xbundle->use_priority_tags) {
1156         tci |= *flow_tci & htons(VLAN_PCP_MASK);
1157         if (tci) {
1158             tci |= htons(VLAN_CFI);
1159         }
1160     }
1161     *flow_tci = tci;
1162
1163     compose_output_action(ctx, xport->ofp_port);
1164     *flow_tci = old_tci;
1165 }
1166
1167 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1168  * migration.  Older Citrix-patched Linux DomU used gratuitous ARP replies to
1169  * indicate this; newer upstream kernels use gratuitous ARP requests. */
1170 static bool
1171 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1172 {
1173     if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1174         return false;
1175     }
1176
1177     memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1178     if (!eth_addr_is_broadcast(flow->dl_dst)) {
1179         return false;
1180     }
1181
1182     memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1183     if (flow->nw_proto == ARP_OP_REPLY) {
1184         return true;
1185     } else if (flow->nw_proto == ARP_OP_REQUEST) {
1186         memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1187         memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1188
1189         return flow->nw_src == flow->nw_dst;
1190     } else {
1191         return false;
1192     }
1193 }
1194
1195 /* Checks whether a MAC learning update is necessary for MAC learning table
1196  * 'ml' given that a packet matching 'flow' was received  on 'in_xbundle' in
1197  * 'vlan'.
1198  *
1199  * Most packets processed through the MAC learning table do not actually
1200  * change it in any way.  This function requires only a read lock on the MAC
1201  * learning table, so it is much cheaper in this common case.
1202  *
1203  * Keep the code here synchronized with that in update_learning_table__()
1204  * below. */
1205 static bool
1206 is_mac_learning_update_needed(const struct mac_learning *ml,
1207                               const struct flow *flow,
1208                               struct flow_wildcards *wc,
1209                               int vlan, struct xbundle *in_xbundle)
1210 OVS_REQ_RDLOCK(ml->rwlock)
1211 {
1212     struct mac_entry *mac;
1213
1214     if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1215         return false;
1216     }
1217
1218     mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1219     if (!mac || mac_entry_age(ml, mac)) {
1220         return true;
1221     }
1222
1223     if (is_gratuitous_arp(flow, wc)) {
1224         /* We don't want to learn from gratuitous ARP packets that are
1225          * reflected back over bond slaves so we lock the learning table. */
1226         if (!in_xbundle->bond) {
1227             return true;
1228         } else if (mac_entry_is_grat_arp_locked(mac)) {
1229             return false;
1230         }
1231     }
1232
1233     return mac->port.p != in_xbundle->ofbundle;
1234 }
1235
1236
1237 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1238  * received on 'in_xbundle' in 'vlan'.
1239  *
1240  * This code repeats all the checks in is_mac_learning_update_needed() because
1241  * the lock was released between there and here and thus the MAC learning state
1242  * could have changed.
1243  *
1244  * Keep the code here synchronized with that in is_mac_learning_update_needed()
1245  * above. */
1246 static void
1247 update_learning_table__(const struct xbridge *xbridge,
1248                         const struct flow *flow, struct flow_wildcards *wc,
1249                         int vlan, struct xbundle *in_xbundle)
1250 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1251 {
1252     struct mac_entry *mac;
1253
1254     if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1255         return;
1256     }
1257
1258     mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1259     if (is_gratuitous_arp(flow, wc)) {
1260         /* We don't want to learn from gratuitous ARP packets that are
1261          * reflected back over bond slaves so we lock the learning table. */
1262         if (!in_xbundle->bond) {
1263             mac_entry_set_grat_arp_lock(mac);
1264         } else if (mac_entry_is_grat_arp_locked(mac)) {
1265             return;
1266         }
1267     }
1268
1269     if (mac->port.p != in_xbundle->ofbundle) {
1270         /* The log messages here could actually be useful in debugging,
1271          * so keep the rate limit relatively high. */
1272         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1273
1274         VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1275                     "on port %s in VLAN %d",
1276                     xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1277                     in_xbundle->name, vlan);
1278
1279         mac->port.p = in_xbundle->ofbundle;
1280         mac_learning_changed(xbridge->ml);
1281     }
1282 }
1283
1284 static void
1285 update_learning_table(const struct xbridge *xbridge,
1286                       const struct flow *flow, struct flow_wildcards *wc,
1287                       int vlan, struct xbundle *in_xbundle)
1288 {
1289     bool need_update;
1290
1291     /* Don't learn the OFPP_NONE port. */
1292     if (in_xbundle == &ofpp_none_bundle) {
1293         return;
1294     }
1295
1296     /* First try the common case: no change to MAC learning table. */
1297     ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1298     need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1299                                                 in_xbundle);
1300     ovs_rwlock_unlock(&xbridge->ml->rwlock);
1301
1302     if (need_update) {
1303         /* Slow path: MAC learning table might need an update. */
1304         ovs_rwlock_wrlock(&xbridge->ml->rwlock);
1305         update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
1306         ovs_rwlock_unlock(&xbridge->ml->rwlock);
1307     }
1308 }
1309
1310 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1311  * dropped.  Returns true if they may be forwarded, false if they should be
1312  * dropped.
1313  *
1314  * 'in_port' must be the xport that corresponds to flow->in_port.
1315  * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1316  *
1317  * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1318  * returned by input_vid_to_vlan().  It must be a valid VLAN for 'in_port', as
1319  * checked by input_vid_is_valid().
1320  *
1321  * May also add tags to '*tags', although the current implementation only does
1322  * so in one special case.
1323  */
1324 static bool
1325 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1326               uint16_t vlan)
1327 {
1328     struct xbundle *in_xbundle = in_port->xbundle;
1329     const struct xbridge *xbridge = ctx->xbridge;
1330     struct flow *flow = &ctx->xin->flow;
1331
1332     /* Drop frames for reserved multicast addresses
1333      * only if forward_bpdu option is absent. */
1334     if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1335         xlate_report(ctx, "packet has reserved destination MAC, dropping");
1336         return false;
1337     }
1338
1339     if (in_xbundle->bond) {
1340         struct mac_entry *mac;
1341
1342         switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1343                                          flow->dl_dst)) {
1344         case BV_ACCEPT:
1345             break;
1346
1347         case BV_DROP:
1348             xlate_report(ctx, "bonding refused admissibility, dropping");
1349             return false;
1350
1351         case BV_DROP_IF_MOVED:
1352             ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1353             mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1354             if (mac && mac->port.p != in_xbundle->ofbundle &&
1355                 (!is_gratuitous_arp(flow, &ctx->xout->wc)
1356                  || mac_entry_is_grat_arp_locked(mac))) {
1357                 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1358                 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1359                              "dropping");
1360                 return false;
1361             }
1362             ovs_rwlock_unlock(&xbridge->ml->rwlock);
1363             break;
1364         }
1365     }
1366
1367     return true;
1368 }
1369
1370 static void
1371 xlate_normal(struct xlate_ctx *ctx)
1372 {
1373     struct flow_wildcards *wc = &ctx->xout->wc;
1374     struct flow *flow = &ctx->xin->flow;
1375     struct xbundle *in_xbundle;
1376     struct xport *in_port;
1377     struct mac_entry *mac;
1378     void *mac_port;
1379     uint16_t vlan;
1380     uint16_t vid;
1381
1382     ctx->xout->has_normal = true;
1383
1384     memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1385     memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1386     wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1387
1388     in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
1389                                      ctx->xin->packet != NULL, &in_port);
1390     if (!in_xbundle) {
1391         xlate_report(ctx, "no input bundle, dropping");
1392         return;
1393     }
1394
1395     /* Drop malformed frames. */
1396     if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
1397         !(flow->vlan_tci & htons(VLAN_CFI))) {
1398         if (ctx->xin->packet != NULL) {
1399             static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1400             VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
1401                          "VLAN tag received on port %s",
1402                          ctx->xbridge->name, in_xbundle->name);
1403         }
1404         xlate_report(ctx, "partial VLAN tag, dropping");
1405         return;
1406     }
1407
1408     /* Drop frames on bundles reserved for mirroring. */
1409     if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
1410         if (ctx->xin->packet != NULL) {
1411             static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1412             VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1413                          "%s, which is reserved exclusively for mirroring",
1414                          ctx->xbridge->name, in_xbundle->name);
1415         }
1416         xlate_report(ctx, "input port is mirror output port, dropping");
1417         return;
1418     }
1419
1420     /* Check VLAN. */
1421     vid = vlan_tci_to_vid(flow->vlan_tci);
1422     if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1423         xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
1424         return;
1425     }
1426     vlan = input_vid_to_vlan(in_xbundle, vid);
1427
1428     /* Check other admissibility requirements. */
1429     if (in_port && !is_admissible(ctx, in_port, vlan)) {
1430         return;
1431     }
1432
1433     /* Learn source MAC. */
1434     if (ctx->xin->may_learn) {
1435         update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
1436     }
1437
1438     /* Determine output bundle. */
1439     ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
1440     mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
1441     mac_port = mac ? mac->port.p : NULL;
1442     ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
1443
1444     if (mac_port) {
1445         struct xbundle *mac_xbundle = xbundle_lookup(mac_port);
1446         if (mac_xbundle && mac_xbundle != in_xbundle) {
1447             xlate_report(ctx, "forwarding to learned port");
1448             output_normal(ctx, mac_xbundle, vlan);
1449         } else if (!mac_xbundle) {
1450             xlate_report(ctx, "learned port is unknown, dropping");
1451         } else {
1452             xlate_report(ctx, "learned port is input port, dropping");
1453         }
1454     } else {
1455         struct xbundle *xbundle;
1456
1457         xlate_report(ctx, "no learned MAC for destination, flooding");
1458         LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
1459             if (xbundle != in_xbundle
1460                 && xbundle_includes_vlan(xbundle, vlan)
1461                 && xbundle->floodable
1462                 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
1463                 output_normal(ctx, xbundle, vlan);
1464             }
1465         }
1466         ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1467     }
1468 }
1469
1470 /* Compose SAMPLE action for sFlow or IPFIX.  The given probability is
1471  * the number of packets out of UINT32_MAX to sample.  The given
1472  * cookie is passed back in the callback for each sampled packet.
1473  */
1474 static size_t
1475 compose_sample_action(const struct xbridge *xbridge,
1476                       struct ofpbuf *odp_actions,
1477                       const struct flow *flow,
1478                       const uint32_t probability,
1479                       const union user_action_cookie *cookie,
1480                       const size_t cookie_size)
1481 {
1482     size_t sample_offset, actions_offset;
1483     odp_port_t odp_port;
1484     int cookie_offset;
1485     uint32_t pid;
1486
1487     sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
1488
1489     nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
1490
1491     actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
1492
1493     odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
1494     pid = dpif_port_get_pid(xbridge->dpif, odp_port);
1495     cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
1496
1497     nl_msg_end_nested(odp_actions, actions_offset);
1498     nl_msg_end_nested(odp_actions, sample_offset);
1499     return cookie_offset;
1500 }
1501
1502 static void
1503 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
1504                      odp_port_t odp_port, unsigned int n_outputs,
1505                      union user_action_cookie *cookie)
1506 {
1507     int ifindex;
1508
1509     cookie->type = USER_ACTION_COOKIE_SFLOW;
1510     cookie->sflow.vlan_tci = vlan_tci;
1511
1512     /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
1513      * port information") for the interpretation of cookie->output. */
1514     switch (n_outputs) {
1515     case 0:
1516         /* 0x40000000 | 256 means "packet dropped for unknown reason". */
1517         cookie->sflow.output = 0x40000000 | 256;
1518         break;
1519
1520     case 1:
1521         ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
1522         if (ifindex) {
1523             cookie->sflow.output = ifindex;
1524             break;
1525         }
1526         /* Fall through. */
1527     default:
1528         /* 0x80000000 means "multiple output ports. */
1529         cookie->sflow.output = 0x80000000 | n_outputs;
1530         break;
1531     }
1532 }
1533
1534 /* Compose SAMPLE action for sFlow bridge sampling. */
1535 static size_t
1536 compose_sflow_action(const struct xbridge *xbridge,
1537                      struct ofpbuf *odp_actions,
1538                      const struct flow *flow,
1539                      odp_port_t odp_port)
1540 {
1541     uint32_t probability;
1542     union user_action_cookie cookie;
1543
1544     if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
1545         return 0;
1546     }
1547
1548     probability = dpif_sflow_get_probability(xbridge->sflow);
1549     compose_sflow_cookie(xbridge, htons(0), odp_port,
1550                          odp_port == ODPP_NONE ? 0 : 1, &cookie);
1551
1552     return compose_sample_action(xbridge, odp_actions, flow,  probability,
1553                                  &cookie, sizeof cookie.sflow);
1554 }
1555
1556 static void
1557 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
1558                            uint32_t obs_domain_id, uint32_t obs_point_id,
1559                            union user_action_cookie *cookie)
1560 {
1561     cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
1562     cookie->flow_sample.probability = probability;
1563     cookie->flow_sample.collector_set_id = collector_set_id;
1564     cookie->flow_sample.obs_domain_id = obs_domain_id;
1565     cookie->flow_sample.obs_point_id = obs_point_id;
1566 }
1567
1568 static void
1569 compose_ipfix_cookie(union user_action_cookie *cookie)
1570 {
1571     cookie->type = USER_ACTION_COOKIE_IPFIX;
1572 }
1573
1574 /* Compose SAMPLE action for IPFIX bridge sampling. */
1575 static void
1576 compose_ipfix_action(const struct xbridge *xbridge,
1577                      struct ofpbuf *odp_actions,
1578                      const struct flow *flow)
1579 {
1580     uint32_t probability;
1581     union user_action_cookie cookie;
1582
1583     if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
1584         return;
1585     }
1586
1587     probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
1588     compose_ipfix_cookie(&cookie);
1589
1590     compose_sample_action(xbridge, odp_actions, flow,  probability,
1591                           &cookie, sizeof cookie.ipfix);
1592 }
1593
1594 /* SAMPLE action for sFlow must be first action in any given list of
1595  * actions.  At this point we do not have all information required to
1596  * build it. So try to build sample action as complete as possible. */
1597 static void
1598 add_sflow_action(struct xlate_ctx *ctx)
1599 {
1600     ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
1601                                                    &ctx->xout->odp_actions,
1602                                                    &ctx->xin->flow, ODPP_NONE);
1603     ctx->sflow_odp_port = 0;
1604     ctx->sflow_n_outputs = 0;
1605 }
1606
1607 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
1608  * of actions, eventually after the SAMPLE action for sFlow. */
1609 static void
1610 add_ipfix_action(struct xlate_ctx *ctx)
1611 {
1612     compose_ipfix_action(ctx->xbridge, &ctx->xout->odp_actions,
1613                          &ctx->xin->flow);
1614 }
1615
1616 /* Fix SAMPLE action according to data collected while composing ODP actions.
1617  * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
1618  * USERSPACE action's user-cookie which is required for sflow. */
1619 static void
1620 fix_sflow_action(struct xlate_ctx *ctx)
1621 {
1622     const struct flow *base = &ctx->base_flow;
1623     union user_action_cookie *cookie;
1624
1625     if (!ctx->user_cookie_offset) {
1626         return;
1627     }
1628
1629     cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
1630                        sizeof cookie->sflow);
1631     ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
1632
1633     compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
1634                          ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
1635 }
1636
1637 static enum slow_path_reason
1638 process_special(struct xlate_ctx *ctx, const struct flow *flow,
1639                 const struct xport *xport, const struct ofpbuf *packet)
1640 {
1641     struct flow_wildcards *wc = &ctx->xout->wc;
1642     const struct xbridge *xbridge = ctx->xbridge;
1643
1644     if (!xport) {
1645         return 0;
1646     } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
1647         if (packet) {
1648             cfm_process_heartbeat(xport->cfm, packet);
1649         }
1650         return SLOW_CFM;
1651     } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
1652         if (packet) {
1653             bfd_process_packet(xport->bfd, flow, packet);
1654             /* If POLL received, immediately sends FINAL back. */
1655             if (bfd_should_send_packet(xport->bfd)) {
1656                 if (xport->peer) {
1657                     ofproto_dpif_monitor_port_send_soon(xport->ofport);
1658                 } else {
1659                     ofproto_dpif_monitor_port_send_soon_safe(xport->ofport);
1660                 }
1661             }
1662         }
1663         return SLOW_BFD;
1664     } else if (xport->xbundle && xport->xbundle->lacp
1665                && flow->dl_type == htons(ETH_TYPE_LACP)) {
1666         if (packet) {
1667             lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
1668         }
1669         return SLOW_LACP;
1670     } else if (xbridge->stp && stp_should_process_flow(flow, wc)) {
1671         if (packet) {
1672             stp_process_packet(xport, packet);
1673         }
1674         return SLOW_STP;
1675     } else {
1676         return 0;
1677     }
1678 }
1679
1680 static void
1681 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
1682                         bool check_stp)
1683 {
1684     const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1685     struct flow_wildcards *wc = &ctx->xout->wc;
1686     struct flow *flow = &ctx->xin->flow;
1687     ovs_be16 flow_vlan_tci;
1688     uint32_t flow_pkt_mark;
1689     uint8_t flow_nw_tos;
1690     odp_port_t out_port, odp_port;
1691     uint8_t dscp;
1692
1693     /* If 'struct flow' gets additional metadata, we'll need to zero it out
1694      * before traversing a patch port. */
1695     BUILD_ASSERT_DECL(FLOW_WC_SEQ == 24);
1696
1697     if (!xport) {
1698         xlate_report(ctx, "Nonexistent output port");
1699         return;
1700     } else if (xport->config & OFPUTIL_PC_NO_FWD) {
1701         xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
1702         return;
1703     } else if (check_stp) {
1704         if (eth_addr_equals(ctx->base_flow.dl_dst, eth_addr_stp)) {
1705             if (!xport_stp_listen_state(xport)) {
1706                 xlate_report(ctx, "STP not in listening state, "
1707                              "skipping bpdu output");
1708                 return;
1709             }
1710         } else if (!xport_stp_forward_state(xport)) {
1711             xlate_report(ctx, "STP not in forwarding state, "
1712                          "skipping output");
1713             return;
1714         }
1715     }
1716
1717     if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
1718         ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
1719                                                  xport->xbundle);
1720     }
1721
1722     if (xport->peer) {
1723         const struct xport *peer = xport->peer;
1724         struct flow old_flow = ctx->xin->flow;
1725         enum slow_path_reason special;
1726
1727         ctx->xbridge = peer->xbridge;
1728         flow->in_port.ofp_port = peer->ofp_port;
1729         flow->metadata = htonll(0);
1730         memset(&flow->tunnel, 0, sizeof flow->tunnel);
1731         memset(flow->regs, 0, sizeof flow->regs);
1732
1733         special = process_special(ctx, &ctx->xin->flow, peer,
1734                                   ctx->xin->packet);
1735         if (special) {
1736             ctx->xout->slow |= special;
1737         } else if (may_receive(peer, ctx)) {
1738             if (xport_stp_forward_state(peer)) {
1739                 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1740             } else {
1741                 /* Forwarding is disabled by STP.  Let OFPP_NORMAL and the
1742                  * learning action look at the packet, then drop it. */
1743                 struct flow old_base_flow = ctx->base_flow;
1744                 size_t old_size = ctx->xout->odp_actions.size;
1745                 mirror_mask_t old_mirrors = ctx->xout->mirrors;
1746                 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1747                 ctx->xout->mirrors = old_mirrors;
1748                 ctx->base_flow = old_base_flow;
1749                 ctx->xout->odp_actions.size = old_size;
1750             }
1751         }
1752
1753         ctx->xin->flow = old_flow;
1754         ctx->xbridge = xport->xbridge;
1755
1756         if (ctx->xin->resubmit_stats) {
1757             netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1758             netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
1759             if (peer->bfd) {
1760                 bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
1761             }
1762         }
1763
1764         return;
1765     }
1766
1767     flow_vlan_tci = flow->vlan_tci;
1768     flow_pkt_mark = flow->pkt_mark;
1769     flow_nw_tos = flow->nw_tos;
1770
1771     if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
1772         wc->masks.nw_tos |= IP_ECN_MASK;
1773         flow->nw_tos &= ~IP_DSCP_MASK;
1774         flow->nw_tos |= dscp;
1775     }
1776
1777     if (xport->is_tunnel) {
1778          /* Save tunnel metadata so that changes made due to
1779           * the Logical (tunnel) Port are not visible for any further
1780           * matches, while explicit set actions on tunnel metadata are.
1781           */
1782         struct flow_tnl flow_tnl = flow->tunnel;
1783         odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
1784         if (odp_port == ODPP_NONE) {
1785             xlate_report(ctx, "Tunneling decided against output");
1786             goto out; /* restore flow_nw_tos */
1787         }
1788         if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
1789             xlate_report(ctx, "Not tunneling to our own address");
1790             goto out; /* restore flow_nw_tos */
1791         }
1792         if (ctx->xin->resubmit_stats) {
1793             netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1794         }
1795         out_port = odp_port;
1796         commit_odp_tunnel_action(flow, &ctx->base_flow,
1797                                  &ctx->xout->odp_actions);
1798         flow->tunnel = flow_tnl; /* Restore tunnel metadata */
1799     } else {
1800         odp_port = xport->odp_port;
1801         out_port = odp_port;
1802         if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
1803             ofp_port_t vlandev_port;
1804
1805             wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1806             vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto,
1807                                                   ofp_port, flow->vlan_tci);
1808             if (vlandev_port != ofp_port) {
1809                 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
1810                 flow->vlan_tci = htons(0);
1811             }
1812         }
1813     }
1814
1815     if (out_port != ODPP_NONE) {
1816         ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
1817                                               &ctx->xout->odp_actions,
1818                                               &ctx->xout->wc);
1819         nl_msg_put_odp_port(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
1820                             out_port);
1821
1822         ctx->sflow_odp_port = odp_port;
1823         ctx->sflow_n_outputs++;
1824         ctx->xout->nf_output_iface = ofp_port;
1825     }
1826
1827  out:
1828     /* Restore flow */
1829     flow->vlan_tci = flow_vlan_tci;
1830     flow->pkt_mark = flow_pkt_mark;
1831     flow->nw_tos = flow_nw_tos;
1832 }
1833
1834 static void
1835 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
1836 {
1837     compose_output_action__(ctx, ofp_port, true);
1838 }
1839
1840 static void
1841 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
1842 {
1843     struct rule_dpif *old_rule = ctx->rule;
1844     struct rule_actions *actions;
1845
1846     if (ctx->xin->resubmit_stats) {
1847         rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
1848     }
1849
1850     ctx->resubmits++;
1851     ctx->recurse++;
1852     ctx->rule = rule;
1853     actions = rule_dpif_get_actions(rule);
1854     do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
1855     ctx->rule = old_rule;
1856     ctx->recurse--;
1857 }
1858
1859 static bool
1860 xlate_resubmit_resource_check(struct xlate_ctx *ctx)
1861 {
1862     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1863
1864     if (ctx->recurse >= MAX_RESUBMIT_RECURSION) {
1865         VLOG_ERR_RL(&rl, "resubmit actions recursed over %d times",
1866                     MAX_RESUBMIT_RECURSION);
1867     } else if (ctx->resubmits >= MAX_RESUBMITS) {
1868         VLOG_ERR_RL(&rl, "over %d resubmit actions", MAX_RESUBMITS);
1869     } else if (ctx->xout->odp_actions.size > UINT16_MAX) {
1870         VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of actions");
1871     } else if (ctx->stack.size >= 65536) {
1872         VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of stack");
1873     } else {
1874         return true;
1875     }
1876
1877     return false;
1878 }
1879
1880 static void
1881 xlate_table_action(struct xlate_ctx *ctx,
1882                    ofp_port_t in_port, uint8_t table_id, bool may_packet_in)
1883 {
1884     if (xlate_resubmit_resource_check(ctx)) {
1885         ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
1886         bool skip_wildcards = ctx->xin->skip_wildcards;
1887         uint8_t old_table_id = ctx->table_id;
1888         struct rule_dpif *rule;
1889
1890         ctx->table_id = table_id;
1891
1892         /* Look up a flow with 'in_port' as the input port.  Then restore the
1893          * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
1894          * have surprising behavior). */
1895         ctx->xin->flow.in_port.ofp_port = in_port;
1896         rule_dpif_lookup_in_table(ctx->xbridge->ofproto, &ctx->xin->flow,
1897                                   !skip_wildcards ? &ctx->xout->wc : NULL,
1898                                   table_id, &rule);
1899         ctx->xin->flow.in_port.ofp_port = old_in_port;
1900
1901         if (ctx->xin->resubmit_hook) {
1902             ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
1903         }
1904
1905         if (!rule && may_packet_in) {
1906             struct xport *xport;
1907
1908             /* XXX
1909              * check if table configuration flags
1910              * OFPTC11_TABLE_MISS_CONTROLLER, default.
1911              * OFPTC11_TABLE_MISS_CONTINUE,
1912              * OFPTC11_TABLE_MISS_DROP
1913              * When OF1.0, OFPTC11_TABLE_MISS_CONTINUE is used. What to do? */
1914             xport = get_ofp_port(ctx->xbridge, ctx->xin->flow.in_port.ofp_port);
1915             choose_miss_rule(xport ? xport->config : 0,
1916                              ctx->xbridge->miss_rule,
1917                              ctx->xbridge->no_packet_in_rule, &rule);
1918         }
1919         if (rule) {
1920             xlate_recursively(ctx, rule);
1921             rule_dpif_unref(rule);
1922         }
1923
1924         ctx->table_id = old_table_id;
1925         return;
1926     }
1927
1928     ctx->exit = true;
1929 }
1930
1931 static void
1932 xlate_group_bucket(struct xlate_ctx *ctx, const struct ofputil_bucket *bucket)
1933 {
1934     uint64_t action_list_stub[1024 / 8];
1935     struct ofpbuf action_list, action_set;
1936
1937     ofpbuf_use_const(&action_set, bucket->ofpacts, bucket->ofpacts_len);
1938     ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
1939
1940     ofpacts_execute_action_set(&action_list, &action_set);
1941     ctx->recurse++;
1942     do_xlate_actions(action_list.data, action_list.size, ctx);
1943     ctx->recurse--;
1944
1945     ofpbuf_uninit(&action_set);
1946     ofpbuf_uninit(&action_list);
1947 }
1948
1949 static void
1950 xlate_all_group(struct xlate_ctx *ctx, struct group_dpif *group)
1951 {
1952     const struct ofputil_bucket *bucket;
1953     const struct list *buckets;
1954     struct flow old_flow = ctx->xin->flow;
1955
1956     group_dpif_get_buckets(group, &buckets);
1957
1958     LIST_FOR_EACH (bucket, list_node, buckets) {
1959         xlate_group_bucket(ctx, bucket);
1960         /* Roll back flow to previous state.
1961          * This is equivalent to cloning the packet for each bucket.
1962          *
1963          * As a side effect any subsequently applied actions will
1964          * also effectively be applied to a clone of the packet taken
1965          * just before applying the all or indirect group. */
1966         ctx->xin->flow = old_flow;
1967     }
1968 }
1969
1970 static void
1971 xlate_ff_group(struct xlate_ctx *ctx, struct group_dpif *group)
1972 {
1973     const struct ofputil_bucket *bucket;
1974
1975     bucket = group_first_live_bucket(ctx, group, 0);
1976     if (bucket) {
1977         xlate_group_bucket(ctx, bucket);
1978     }
1979 }
1980
1981 static void
1982 xlate_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
1983 {
1984     struct flow_wildcards *wc = &ctx->xout->wc;
1985     const struct ofputil_bucket *bucket;
1986     uint32_t basis;
1987
1988     basis = hash_bytes(ctx->xin->flow.dl_dst, sizeof ctx->xin->flow.dl_dst, 0);
1989     bucket = group_best_live_bucket(ctx, group, basis);
1990     if (bucket) {
1991         memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1992         xlate_group_bucket(ctx, bucket);
1993     }
1994 }
1995
1996 static void
1997 xlate_group_action__(struct xlate_ctx *ctx, struct group_dpif *group)
1998 {
1999     ctx->in_group = true;
2000
2001     switch (group_dpif_get_type(group)) {
2002     case OFPGT11_ALL:
2003     case OFPGT11_INDIRECT:
2004         xlate_all_group(ctx, group);
2005         break;
2006     case OFPGT11_SELECT:
2007         xlate_select_group(ctx, group);
2008         break;
2009     case OFPGT11_FF:
2010         xlate_ff_group(ctx, group);
2011         break;
2012     default:
2013         OVS_NOT_REACHED();
2014     }
2015     group_dpif_release(group);
2016
2017     ctx->in_group = false;
2018 }
2019
2020 static bool
2021 xlate_group_resource_check(struct xlate_ctx *ctx)
2022 {
2023     if (!xlate_resubmit_resource_check(ctx)) {
2024         return false;
2025     } else if (ctx->in_group) {
2026         /* Prevent nested translation of OpenFlow groups.
2027          *
2028          * OpenFlow allows this restriction.  We enforce this restriction only
2029          * because, with the current architecture, we would otherwise have to
2030          * take a possibly recursive read lock on the ofgroup rwlock, which is
2031          * unsafe given that POSIX allows taking a read lock to block if there
2032          * is a thread blocked on taking the write lock.  Other solutions
2033          * without this restriction are also possible, but seem unwarranted
2034          * given the current limited use of groups. */
2035         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2036
2037         VLOG_ERR_RL(&rl, "cannot recursively translate OpenFlow group");
2038         return false;
2039     } else {
2040         return true;
2041     }
2042 }
2043
2044 static bool
2045 xlate_group_action(struct xlate_ctx *ctx, uint32_t group_id)
2046 {
2047     if (xlate_group_resource_check(ctx)) {
2048         struct group_dpif *group;
2049         bool got_group;
2050
2051         got_group = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
2052         if (got_group) {
2053             xlate_group_action__(ctx, group);
2054         } else {
2055             return true;
2056         }
2057     }
2058
2059     return false;
2060 }
2061
2062 static void
2063 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
2064                       const struct ofpact_resubmit *resubmit)
2065 {
2066     ofp_port_t in_port;
2067     uint8_t table_id;
2068
2069     in_port = resubmit->in_port;
2070     if (in_port == OFPP_IN_PORT) {
2071         in_port = ctx->xin->flow.in_port.ofp_port;
2072     }
2073
2074     table_id = resubmit->table_id;
2075     if (table_id == 255) {
2076         table_id = ctx->table_id;
2077     }
2078
2079     xlate_table_action(ctx, in_port, table_id, false);
2080 }
2081
2082 static void
2083 flood_packets(struct xlate_ctx *ctx, bool all)
2084 {
2085     const struct xport *xport;
2086
2087     HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
2088         if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
2089             continue;
2090         }
2091
2092         if (all) {
2093             compose_output_action__(ctx, xport->ofp_port, false);
2094         } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
2095             compose_output_action(ctx, xport->ofp_port);
2096         }
2097     }
2098
2099     ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2100 }
2101
2102 static void
2103 execute_controller_action(struct xlate_ctx *ctx, int len,
2104                           enum ofp_packet_in_reason reason,
2105                           uint16_t controller_id)
2106 {
2107     struct ofproto_packet_in *pin;
2108     struct ofpbuf *packet;
2109     struct pkt_metadata md = PKT_METADATA_INITIALIZER(0);
2110
2111     ctx->xout->slow |= SLOW_CONTROLLER;
2112     if (!ctx->xin->packet) {
2113         return;
2114     }
2115
2116     packet = ofpbuf_clone(ctx->xin->packet);
2117
2118     ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2119                                           &ctx->xout->odp_actions,
2120                                           &ctx->xout->wc);
2121
2122     odp_execute_actions(NULL, packet, &md, ctx->xout->odp_actions.data,
2123                         ctx->xout->odp_actions.size, NULL);
2124
2125     pin = xmalloc(sizeof *pin);
2126     pin->up.packet_len = packet->size;
2127     pin->up.packet = ofpbuf_steal_data(packet);
2128     pin->up.reason = reason;
2129     pin->up.table_id = ctx->table_id;
2130     pin->up.cookie = (ctx->rule
2131                       ? rule_dpif_get_flow_cookie(ctx->rule)
2132                       : OVS_BE64_MAX);
2133
2134     flow_get_metadata(&ctx->xin->flow, &pin->up.fmd);
2135
2136     pin->controller_id = controller_id;
2137     pin->send_len = len;
2138     pin->generated_by_table_miss = (ctx->rule
2139                                     && rule_dpif_is_table_miss(ctx->rule));
2140     ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
2141     ofpbuf_delete(packet);
2142 }
2143
2144 static void
2145 compose_mpls_push_action(struct xlate_ctx *ctx, struct ofpact_push_mpls *mpls)
2146 {
2147     struct flow_wildcards *wc = &ctx->xout->wc;
2148     struct flow *flow = &ctx->xin->flow;
2149     int n;
2150
2151     ovs_assert(eth_type_mpls(mpls->ethertype));
2152
2153     n = flow_count_mpls_labels(flow, wc);
2154     if (!n) {
2155         ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
2156                                               &ctx->xout->odp_actions,
2157                                               &ctx->xout->wc);
2158     } else if (n >= FLOW_MAX_MPLS_LABELS) {
2159         if (ctx->xin->packet != NULL) {
2160             static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2161             VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
2162                          "MPLS push action can't be performed as it would "
2163                          "have more MPLS LSEs than the %d supported.",
2164                          ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
2165         }
2166         ctx->exit = true;
2167         return;
2168     } else if (n >= ctx->xbridge->max_mpls_depth) {
2169         COVERAGE_INC(xlate_actions_mpls_overflow);
2170         ctx->xout->slow |= SLOW_ACTION;
2171     }
2172
2173     flow_push_mpls(flow, n, mpls->ethertype, wc);
2174 }
2175
2176 static void
2177 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
2178 {
2179     struct flow_wildcards *wc = &ctx->xout->wc;
2180     struct flow *flow = &ctx->xin->flow;
2181     int n = flow_count_mpls_labels(flow, wc);
2182
2183     if (!flow_pop_mpls(flow, n, eth_type, wc) && n >= FLOW_MAX_MPLS_LABELS) {
2184         if (ctx->xin->packet != NULL) {
2185             static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2186             VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
2187                          "MPLS pop action can't be performed as it has "
2188                          "more MPLS LSEs than the %d supported.",
2189                          ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
2190         }
2191         ctx->exit = true;
2192         ofpbuf_clear(&ctx->xout->odp_actions);
2193     }
2194 }
2195
2196 static bool
2197 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
2198 {
2199     struct flow *flow = &ctx->xin->flow;
2200
2201     if (!is_ip_any(flow)) {
2202         return false;
2203     }
2204
2205     ctx->xout->wc.masks.nw_ttl = 0xff;
2206     if (flow->nw_ttl > 1) {
2207         flow->nw_ttl--;
2208         return false;
2209     } else {
2210         size_t i;
2211
2212         for (i = 0; i < ids->n_controllers; i++) {
2213             execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
2214                                       ids->cnt_ids[i]);
2215         }
2216
2217         /* Stop processing for current table. */
2218         return true;
2219     }
2220 }
2221
2222 static void
2223 compose_set_mpls_label_action(struct xlate_ctx *ctx, ovs_be32 label)
2224 {
2225     if (eth_type_mpls(ctx->xin->flow.dl_type)) {
2226         ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_LABEL_MASK);
2227         set_mpls_lse_label(&ctx->xin->flow.mpls_lse[0], label);
2228     }
2229 }
2230
2231 static void
2232 compose_set_mpls_tc_action(struct xlate_ctx *ctx, uint8_t tc)
2233 {
2234     if (eth_type_mpls(ctx->xin->flow.dl_type)) {
2235         ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TC_MASK);
2236         set_mpls_lse_tc(&ctx->xin->flow.mpls_lse[0], tc);
2237     }
2238 }
2239
2240 static void
2241 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
2242 {
2243     if (eth_type_mpls(ctx->xin->flow.dl_type)) {
2244         ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
2245         set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse[0], ttl);
2246     }
2247 }
2248
2249 static bool
2250 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
2251 {
2252     struct flow *flow = &ctx->xin->flow;
2253     uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse[0]);
2254     struct flow_wildcards *wc = &ctx->xout->wc;
2255
2256     memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
2257     if (eth_type_mpls(flow->dl_type)) {
2258         if (ttl > 1) {
2259             ttl--;
2260             set_mpls_lse_ttl(&flow->mpls_lse[0], ttl);
2261             return false;
2262         } else {
2263             execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
2264
2265             /* Stop processing for current table. */
2266             return true;
2267         }
2268     } else {
2269         return true;
2270     }
2271 }
2272
2273 static void
2274 xlate_output_action(struct xlate_ctx *ctx,
2275                     ofp_port_t port, uint16_t max_len, bool may_packet_in)
2276 {
2277     ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
2278
2279     ctx->xout->nf_output_iface = NF_OUT_DROP;
2280
2281     switch (port) {
2282     case OFPP_IN_PORT:
2283         compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
2284         break;
2285     case OFPP_TABLE:
2286         xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2287                            0, may_packet_in);
2288         break;
2289     case OFPP_NORMAL:
2290         xlate_normal(ctx);
2291         break;
2292     case OFPP_FLOOD:
2293         flood_packets(ctx,  false);
2294         break;
2295     case OFPP_ALL:
2296         flood_packets(ctx, true);
2297         break;
2298     case OFPP_CONTROLLER:
2299         execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
2300         break;
2301     case OFPP_NONE:
2302         break;
2303     case OFPP_LOCAL:
2304     default:
2305         if (port != ctx->xin->flow.in_port.ofp_port) {
2306             compose_output_action(ctx, port);
2307         } else {
2308             xlate_report(ctx, "skipping output to input port");
2309         }
2310         break;
2311     }
2312
2313     if (prev_nf_output_iface == NF_OUT_FLOOD) {
2314         ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2315     } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2316         ctx->xout->nf_output_iface = prev_nf_output_iface;
2317     } else if (prev_nf_output_iface != NF_OUT_DROP &&
2318                ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2319         ctx->xout->nf_output_iface = NF_OUT_MULTI;
2320     }
2321 }
2322
2323 static void
2324 xlate_output_reg_action(struct xlate_ctx *ctx,
2325                         const struct ofpact_output_reg *or)
2326 {
2327     uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
2328     if (port <= UINT16_MAX) {
2329         union mf_subvalue value;
2330
2331         memset(&value, 0xff, sizeof value);
2332         mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
2333         xlate_output_action(ctx, u16_to_ofp(port),
2334                             or->max_len, false);
2335     }
2336 }
2337
2338 static void
2339 xlate_enqueue_action(struct xlate_ctx *ctx,
2340                      const struct ofpact_enqueue *enqueue)
2341 {
2342     ofp_port_t ofp_port = enqueue->port;
2343     uint32_t queue_id = enqueue->queue;
2344     uint32_t flow_priority, priority;
2345     int error;
2346
2347     /* Translate queue to priority. */
2348     error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
2349     if (error) {
2350         /* Fall back to ordinary output action. */
2351         xlate_output_action(ctx, enqueue->port, 0, false);
2352         return;
2353     }
2354
2355     /* Check output port. */
2356     if (ofp_port == OFPP_IN_PORT) {
2357         ofp_port = ctx->xin->flow.in_port.ofp_port;
2358     } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
2359         return;
2360     }
2361
2362     /* Add datapath actions. */
2363     flow_priority = ctx->xin->flow.skb_priority;
2364     ctx->xin->flow.skb_priority = priority;
2365     compose_output_action(ctx, ofp_port);
2366     ctx->xin->flow.skb_priority = flow_priority;
2367
2368     /* Update NetFlow output port. */
2369     if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2370         ctx->xout->nf_output_iface = ofp_port;
2371     } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2372         ctx->xout->nf_output_iface = NF_OUT_MULTI;
2373     }
2374 }
2375
2376 static void
2377 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
2378 {
2379     uint32_t skb_priority;
2380
2381     if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
2382         ctx->xin->flow.skb_priority = skb_priority;
2383     } else {
2384         /* Couldn't translate queue to a priority.  Nothing to do.  A warning
2385          * has already been logged. */
2386     }
2387 }
2388
2389 static bool
2390 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
2391 {
2392     const struct xbridge *xbridge = xbridge_;
2393     struct xport *port;
2394
2395     switch (ofp_port) {
2396     case OFPP_IN_PORT:
2397     case OFPP_TABLE:
2398     case OFPP_NORMAL:
2399     case OFPP_FLOOD:
2400     case OFPP_ALL:
2401     case OFPP_NONE:
2402         return true;
2403     case OFPP_CONTROLLER: /* Not supported by the bundle action. */
2404         return false;
2405     default:
2406         port = get_ofp_port(xbridge, ofp_port);
2407         return port ? port->may_enable : false;
2408     }
2409 }
2410
2411 static void
2412 xlate_bundle_action(struct xlate_ctx *ctx,
2413                     const struct ofpact_bundle *bundle)
2414 {
2415     ofp_port_t port;
2416
2417     port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
2418                           slave_enabled_cb,
2419                           CONST_CAST(struct xbridge *, ctx->xbridge));
2420     if (bundle->dst.field) {
2421         nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
2422                      &ctx->xout->wc);
2423     } else {
2424         xlate_output_action(ctx, port, 0, false);
2425     }
2426 }
2427
2428 static void
2429 xlate_learn_action(struct xlate_ctx *ctx,
2430                    const struct ofpact_learn *learn)
2431 {
2432     uint64_t ofpacts_stub[1024 / 8];
2433     struct ofputil_flow_mod fm;
2434     struct ofpbuf ofpacts;
2435
2436     ctx->xout->has_learn = true;
2437
2438     learn_mask(learn, &ctx->xout->wc);
2439
2440     if (!ctx->xin->may_learn) {
2441         return;
2442     }
2443
2444     ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2445     learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
2446     ofproto_dpif_flow_mod(ctx->xbridge->ofproto, &fm);
2447     ofpbuf_uninit(&ofpacts);
2448 }
2449
2450 static void
2451 xlate_fin_timeout(struct xlate_ctx *ctx,
2452                   const struct ofpact_fin_timeout *oft)
2453 {
2454     if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
2455         rule_dpif_reduce_timeouts(ctx->rule, oft->fin_idle_timeout,
2456                                   oft->fin_hard_timeout);
2457     }
2458 }
2459
2460 static void
2461 xlate_sample_action(struct xlate_ctx *ctx,
2462                     const struct ofpact_sample *os)
2463 {
2464   union user_action_cookie cookie;
2465   /* Scale the probability from 16-bit to 32-bit while representing
2466    * the same percentage. */
2467   uint32_t probability = (os->probability << 16) | os->probability;
2468
2469   if (!ctx->xbridge->variable_length_userdata) {
2470       static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2471
2472       VLOG_ERR_RL(&rl, "ignoring NXAST_SAMPLE action because datapath "
2473                   "lacks support (needs Linux 3.10+ or kernel module from "
2474                   "OVS 1.11+)");
2475       return;
2476   }
2477
2478   ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2479                                         &ctx->xout->odp_actions,
2480                                         &ctx->xout->wc);
2481
2482   compose_flow_sample_cookie(os->probability, os->collector_set_id,
2483                              os->obs_domain_id, os->obs_point_id, &cookie);
2484   compose_sample_action(ctx->xbridge, &ctx->xout->odp_actions, &ctx->xin->flow,
2485                         probability, &cookie, sizeof cookie.flow_sample);
2486 }
2487
2488 static bool
2489 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
2490 {
2491     if (xport->config & (eth_addr_equals(ctx->xin->flow.dl_dst, eth_addr_stp)
2492                          ? OFPUTIL_PC_NO_RECV_STP
2493                          : OFPUTIL_PC_NO_RECV)) {
2494         return false;
2495     }
2496
2497     /* Only drop packets here if both forwarding and learning are
2498      * disabled.  If just learning is enabled, we need to have
2499      * OFPP_NORMAL and the learning action have a look at the packet
2500      * before we can drop it. */
2501     if (!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) {
2502         return false;
2503     }
2504
2505     return true;
2506 }
2507
2508 static void
2509 xlate_write_actions(struct xlate_ctx *ctx, const struct ofpact *a)
2510 {
2511     struct ofpact_nest *on = ofpact_get_WRITE_ACTIONS(a);
2512     ofpbuf_put(&ctx->action_set, on->actions, ofpact_nest_get_action_len(on));
2513     ofpact_pad(&ctx->action_set);
2514 }
2515
2516 static void
2517 xlate_action_set(struct xlate_ctx *ctx)
2518 {
2519     uint64_t action_list_stub[1024 / 64];
2520     struct ofpbuf action_list;
2521
2522     ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
2523     ofpacts_execute_action_set(&action_list, &ctx->action_set);
2524     do_xlate_actions(action_list.data, action_list.size, ctx);
2525     ofpbuf_uninit(&action_list);
2526 }
2527
2528 static void
2529 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
2530                  struct xlate_ctx *ctx)
2531 {
2532     struct flow_wildcards *wc = &ctx->xout->wc;
2533     struct flow *flow = &ctx->xin->flow;
2534     const struct ofpact *a;
2535
2536     /* dl_type already in the mask, not set below. */
2537
2538     OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2539         struct ofpact_controller *controller;
2540         const struct ofpact_metadata *metadata;
2541         const struct ofpact_set_field *set_field;
2542         const struct mf_field *mf;
2543
2544         if (ctx->exit) {
2545             break;
2546         }
2547
2548         switch (a->type) {
2549         case OFPACT_OUTPUT:
2550             xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
2551                                 ofpact_get_OUTPUT(a)->max_len, true);
2552             break;
2553
2554         case OFPACT_GROUP:
2555             if (xlate_group_action(ctx, ofpact_get_GROUP(a)->group_id)) {
2556                 return;
2557             }
2558             break;
2559
2560         case OFPACT_CONTROLLER:
2561             controller = ofpact_get_CONTROLLER(a);
2562             execute_controller_action(ctx, controller->max_len,
2563                                       controller->reason,
2564                                       controller->controller_id);
2565             break;
2566
2567         case OFPACT_ENQUEUE:
2568             xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
2569             break;
2570
2571         case OFPACT_SET_VLAN_VID:
2572             wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2573             if (flow->vlan_tci & htons(VLAN_CFI) ||
2574                 ofpact_get_SET_VLAN_VID(a)->push_vlan_if_needed) {
2575                 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
2576                 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
2577                                    | htons(VLAN_CFI));
2578             }
2579             break;
2580
2581         case OFPACT_SET_VLAN_PCP:
2582             wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
2583             if (flow->vlan_tci & htons(VLAN_CFI) ||
2584                 ofpact_get_SET_VLAN_PCP(a)->push_vlan_if_needed) {
2585                 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
2586                 flow->vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
2587                                          << VLAN_PCP_SHIFT) | VLAN_CFI);
2588             }
2589             break;
2590
2591         case OFPACT_STRIP_VLAN:
2592             memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2593             flow->vlan_tci = htons(0);
2594             break;
2595
2596         case OFPACT_PUSH_VLAN:
2597             /* XXX 802.1AD(QinQ) */
2598             memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2599             flow->vlan_tci = htons(VLAN_CFI);
2600             break;
2601
2602         case OFPACT_SET_ETH_SRC:
2603             memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2604             memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
2605             break;
2606
2607         case OFPACT_SET_ETH_DST:
2608             memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2609             memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
2610             break;
2611
2612         case OFPACT_SET_IPV4_SRC:
2613             if (flow->dl_type == htons(ETH_TYPE_IP)) {
2614                 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
2615                 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
2616             }
2617             break;
2618
2619         case OFPACT_SET_IPV4_DST:
2620             if (flow->dl_type == htons(ETH_TYPE_IP)) {
2621                 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
2622                 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
2623             }
2624             break;
2625
2626         case OFPACT_SET_IP_DSCP:
2627             if (is_ip_any(flow)) {
2628                 wc->masks.nw_tos |= IP_DSCP_MASK;
2629                 flow->nw_tos &= ~IP_DSCP_MASK;
2630                 flow->nw_tos |= ofpact_get_SET_IP_DSCP(a)->dscp;
2631             }
2632             break;
2633
2634         case OFPACT_SET_IP_ECN:
2635             if (is_ip_any(flow)) {
2636                 wc->masks.nw_tos |= IP_ECN_MASK;
2637                 flow->nw_tos &= ~IP_ECN_MASK;
2638                 flow->nw_tos |= ofpact_get_SET_IP_ECN(a)->ecn;
2639             }
2640             break;
2641
2642         case OFPACT_SET_IP_TTL:
2643             if (is_ip_any(flow)) {
2644                 wc->masks.nw_ttl = 0xff;
2645                 flow->nw_ttl = ofpact_get_SET_IP_TTL(a)->ttl;
2646             }
2647             break;
2648
2649         case OFPACT_SET_L4_SRC_PORT:
2650             if (is_ip_any(flow)) {
2651                 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2652                 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
2653                 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
2654             }
2655             break;
2656
2657         case OFPACT_SET_L4_DST_PORT:
2658             if (is_ip_any(flow)) {
2659                 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2660                 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
2661                 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
2662             }
2663             break;
2664
2665         case OFPACT_RESUBMIT:
2666             xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
2667             break;
2668
2669         case OFPACT_SET_TUNNEL:
2670             flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
2671             break;
2672
2673         case OFPACT_SET_QUEUE:
2674             xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
2675             break;
2676
2677         case OFPACT_POP_QUEUE:
2678             flow->skb_priority = ctx->orig_skb_priority;
2679             break;
2680
2681         case OFPACT_REG_MOVE:
2682             nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
2683             break;
2684
2685         case OFPACT_REG_LOAD:
2686             nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow, wc);
2687             break;
2688
2689         case OFPACT_SET_FIELD:
2690             set_field = ofpact_get_SET_FIELD(a);
2691             mf = set_field->field;
2692             mf_mask_field_and_prereqs(mf, &wc->masks);
2693
2694             /* Set field action only ever overwrites packet's outermost
2695              * applicable header fields.  Do nothing if no header exists. */
2696             if ((mf->id != MFF_VLAN_VID || flow->vlan_tci & htons(VLAN_CFI))
2697                 && ((mf->id != MFF_MPLS_LABEL && mf->id != MFF_MPLS_TC)
2698                     || eth_type_mpls(flow->dl_type))) {
2699                 mf_set_flow_value(mf, &set_field->value, flow);
2700             }
2701             break;
2702
2703         case OFPACT_STACK_PUSH:
2704             nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
2705                                    &ctx->stack);
2706             break;
2707
2708         case OFPACT_STACK_POP:
2709             nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
2710                                   &ctx->stack);
2711             break;
2712
2713         case OFPACT_PUSH_MPLS:
2714             compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a));
2715             break;
2716
2717         case OFPACT_POP_MPLS:
2718             compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
2719             break;
2720
2721         case OFPACT_SET_MPLS_LABEL:
2722             compose_set_mpls_label_action(
2723                 ctx, ofpact_get_SET_MPLS_LABEL(a)->label);
2724         break;
2725
2726         case OFPACT_SET_MPLS_TC:
2727             compose_set_mpls_tc_action(ctx, ofpact_get_SET_MPLS_TC(a)->tc);
2728             break;
2729
2730         case OFPACT_SET_MPLS_TTL:
2731             compose_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl);
2732             break;
2733
2734         case OFPACT_DEC_MPLS_TTL:
2735             if (compose_dec_mpls_ttl_action(ctx)) {
2736                 return;
2737             }
2738             break;
2739
2740         case OFPACT_DEC_TTL:
2741             wc->masks.nw_ttl = 0xff;
2742             if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
2743                 return;
2744             }
2745             break;
2746
2747         case OFPACT_NOTE:
2748             /* Nothing to do. */
2749             break;
2750
2751         case OFPACT_MULTIPATH:
2752             multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
2753             break;
2754
2755         case OFPACT_BUNDLE:
2756             xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
2757             break;
2758
2759         case OFPACT_OUTPUT_REG:
2760             xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
2761             break;
2762
2763         case OFPACT_LEARN:
2764             xlate_learn_action(ctx, ofpact_get_LEARN(a));
2765             break;
2766
2767         case OFPACT_EXIT:
2768             ctx->exit = true;
2769             break;
2770
2771         case OFPACT_FIN_TIMEOUT:
2772             memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2773             ctx->xout->has_fin_timeout = true;
2774             xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
2775             break;
2776
2777         case OFPACT_CLEAR_ACTIONS:
2778             ofpbuf_clear(&ctx->action_set);
2779             break;
2780
2781         case OFPACT_WRITE_ACTIONS:
2782             xlate_write_actions(ctx, a);
2783             break;
2784
2785         case OFPACT_WRITE_METADATA:
2786             metadata = ofpact_get_WRITE_METADATA(a);
2787             flow->metadata &= ~metadata->mask;
2788             flow->metadata |= metadata->metadata & metadata->mask;
2789             break;
2790
2791         case OFPACT_METER:
2792             /* Not implemented yet. */
2793             break;
2794
2795         case OFPACT_GOTO_TABLE: {
2796             struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
2797
2798             ovs_assert(ctx->table_id < ogt->table_id);
2799             xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2800                                ogt->table_id, true);
2801             break;
2802         }
2803
2804         case OFPACT_SAMPLE:
2805             xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
2806             break;
2807         }
2808     }
2809 }
2810
2811 void
2812 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
2813               const struct flow *flow, struct rule_dpif *rule,
2814               uint16_t tcp_flags, const struct ofpbuf *packet)
2815 {
2816     xin->ofproto = ofproto;
2817     xin->flow = *flow;
2818     xin->packet = packet;
2819     xin->may_learn = packet != NULL;
2820     xin->rule = rule;
2821     xin->ofpacts = NULL;
2822     xin->ofpacts_len = 0;
2823     xin->tcp_flags = tcp_flags;
2824     xin->resubmit_hook = NULL;
2825     xin->report_hook = NULL;
2826     xin->resubmit_stats = NULL;
2827     xin->skip_wildcards = false;
2828 }
2829
2830 void
2831 xlate_out_uninit(struct xlate_out *xout)
2832 {
2833     if (xout) {
2834         ofpbuf_uninit(&xout->odp_actions);
2835     }
2836 }
2837
2838 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
2839  * into datapath actions, using 'ctx', and discards the datapath actions. */
2840 void
2841 xlate_actions_for_side_effects(struct xlate_in *xin)
2842 {
2843     struct xlate_out xout;
2844
2845     xlate_actions(xin, &xout);
2846     xlate_out_uninit(&xout);
2847 }
2848
2849 static void
2850 xlate_report(struct xlate_ctx *ctx, const char *s)
2851 {
2852     if (ctx->xin->report_hook) {
2853         ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
2854     }
2855 }
2856
2857 void
2858 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
2859 {
2860     dst->wc = src->wc;
2861     dst->slow = src->slow;
2862     dst->has_learn = src->has_learn;
2863     dst->has_normal = src->has_normal;
2864     dst->has_fin_timeout = src->has_fin_timeout;
2865     dst->nf_output_iface = src->nf_output_iface;
2866     dst->mirrors = src->mirrors;
2867
2868     ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
2869                     sizeof dst->odp_actions_stub);
2870     ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
2871                src->odp_actions.size);
2872 }
2873 \f
2874 static struct skb_priority_to_dscp *
2875 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
2876 {
2877     struct skb_priority_to_dscp *pdscp;
2878     uint32_t hash;
2879
2880     hash = hash_int(skb_priority, 0);
2881     HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
2882         if (pdscp->skb_priority == skb_priority) {
2883             return pdscp;
2884         }
2885     }
2886     return NULL;
2887 }
2888
2889 static bool
2890 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
2891                        uint8_t *dscp)
2892 {
2893     struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
2894     *dscp = pdscp ? pdscp->dscp : 0;
2895     return pdscp != NULL;
2896 }
2897
2898 static void
2899 clear_skb_priorities(struct xport *xport)
2900 {
2901     struct skb_priority_to_dscp *pdscp, *next;
2902
2903     HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
2904         hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
2905         free(pdscp);
2906     }
2907 }
2908
2909 static bool
2910 actions_output_to_local_port(const struct xlate_ctx *ctx)
2911 {
2912     odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
2913     const struct nlattr *a;
2914     unsigned int left;
2915
2916     NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->xout->odp_actions.data,
2917                              ctx->xout->odp_actions.size) {
2918         if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
2919             && nl_attr_get_odp_port(a) == local_odp_port) {
2920             return true;
2921         }
2922     }
2923     return false;
2924 }
2925
2926 /* Thread safe call to xlate_actions__(). */
2927 void
2928 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
2929     OVS_EXCLUDED(xlate_rwlock)
2930 {
2931     ovs_rwlock_rdlock(&xlate_rwlock);
2932     xlate_actions__(xin, xout);
2933     ovs_rwlock_unlock(&xlate_rwlock);
2934 }
2935
2936 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
2937  * into datapath actions in 'odp_actions', using 'ctx'.
2938  *
2939  * The caller must take responsibility for eventually freeing 'xout', with
2940  * xlate_out_uninit(). */
2941 static void
2942 xlate_actions__(struct xlate_in *xin, struct xlate_out *xout)
2943     OVS_REQ_RDLOCK(xlate_rwlock)
2944 {
2945     struct flow_wildcards *wc = &xout->wc;
2946     struct flow *flow = &xin->flow;
2947     struct rule_dpif *rule = NULL;
2948
2949     struct rule_actions *actions = NULL;
2950     enum slow_path_reason special;
2951     const struct ofpact *ofpacts;
2952     struct xport *in_port;
2953     struct flow orig_flow;
2954     struct xlate_ctx ctx;
2955     size_t ofpacts_len;
2956     bool tnl_may_send;
2957     bool is_icmp;
2958
2959     COVERAGE_INC(xlate_actions);
2960
2961     /* Flow initialization rules:
2962      * - 'base_flow' must match the kernel's view of the packet at the
2963      *   time that action processing starts.  'flow' represents any
2964      *   transformations we wish to make through actions.
2965      * - By default 'base_flow' and 'flow' are the same since the input
2966      *   packet matches the output before any actions are applied.
2967      * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
2968      *   of the received packet as seen by the kernel.  If we later output
2969      *   to another device without any modifications this will cause us to
2970      *   insert a new tag since the original one was stripped off by the
2971      *   VLAN device.
2972      * - Tunnel metadata as received is retained in 'flow'. This allows
2973      *   tunnel metadata matching also in later tables.
2974      *   Since a kernel action for setting the tunnel metadata will only be
2975      *   generated with actual tunnel output, changing the tunnel metadata
2976      *   values in 'flow' (such as tun_id) will only have effect with a later
2977      *   tunnel output action.
2978      * - Tunnel 'base_flow' is completely cleared since that is what the
2979      *   kernel does.  If we wish to maintain the original values an action
2980      *   needs to be generated. */
2981
2982     ctx.xin = xin;
2983     ctx.xout = xout;
2984     ctx.xout->slow = 0;
2985     ctx.xout->has_learn = false;
2986     ctx.xout->has_normal = false;
2987     ctx.xout->has_fin_timeout = false;
2988     ctx.xout->nf_output_iface = NF_OUT_DROP;
2989     ctx.xout->mirrors = 0;
2990     ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
2991                     sizeof ctx.xout->odp_actions_stub);
2992     ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
2993
2994     ctx.xbridge = xbridge_lookup(xin->ofproto);
2995     if (!ctx.xbridge) {
2996         goto out;
2997     }
2998
2999     ctx.rule = xin->rule;
3000
3001     ctx.base_flow = *flow;
3002     memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
3003     ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
3004
3005     flow_wildcards_init_catchall(wc);
3006     memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
3007     memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
3008     memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
3009     if (is_ip_any(flow)) {
3010         wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
3011     }
3012     is_icmp = is_icmpv4(flow) || is_icmpv6(flow);
3013
3014     tnl_may_send = tnl_xlate_init(&ctx.base_flow, flow, wc);
3015     if (ctx.xbridge->netflow) {
3016         netflow_mask_wc(flow, wc);
3017     }
3018
3019     ctx.recurse = 0;
3020     ctx.resubmits = 0;
3021     ctx.in_group = false;
3022     ctx.orig_skb_priority = flow->skb_priority;
3023     ctx.table_id = 0;
3024     ctx.exit = false;
3025
3026     if (!xin->ofpacts && !ctx.rule) {
3027         rule_dpif_lookup(ctx.xbridge->ofproto, flow,
3028                          !xin->skip_wildcards ? wc : NULL, &rule);
3029         if (ctx.xin->resubmit_stats) {
3030             rule_dpif_credit_stats(rule, ctx.xin->resubmit_stats);
3031         }
3032         ctx.rule = rule;
3033     }
3034     xout->fail_open = ctx.rule && rule_dpif_is_fail_open(ctx.rule);
3035
3036     if (xin->ofpacts) {
3037         ofpacts = xin->ofpacts;
3038         ofpacts_len = xin->ofpacts_len;
3039     } else if (ctx.rule) {
3040         actions = rule_dpif_get_actions(ctx.rule);
3041         ofpacts = actions->ofpacts;
3042         ofpacts_len = actions->ofpacts_len;
3043     } else {
3044         OVS_NOT_REACHED();
3045     }
3046
3047     ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
3048     ofpbuf_use_stub(&ctx.action_set,
3049                     ctx.action_set_stub, sizeof ctx.action_set_stub);
3050
3051     if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
3052         /* Do this conditionally because the copy is expensive enough that it
3053          * shows up in profiles. */
3054         orig_flow = *flow;
3055     }
3056
3057     if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
3058         switch (ctx.xbridge->frag) {
3059         case OFPC_FRAG_NORMAL:
3060             /* We must pretend that transport ports are unavailable. */
3061             flow->tp_src = ctx.base_flow.tp_src = htons(0);
3062             flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
3063             break;
3064
3065         case OFPC_FRAG_DROP:
3066             goto out;
3067
3068         case OFPC_FRAG_REASM:
3069             OVS_NOT_REACHED();
3070
3071         case OFPC_FRAG_NX_MATCH:
3072             /* Nothing to do. */
3073             break;
3074
3075         case OFPC_INVALID_TTL_TO_CONTROLLER:
3076             OVS_NOT_REACHED();
3077         }
3078     }
3079
3080     in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
3081     if (in_port && in_port->is_tunnel && ctx.xin->resubmit_stats) {
3082         netdev_vport_inc_rx(in_port->netdev, ctx.xin->resubmit_stats);
3083         if (in_port->bfd) {
3084             bfd_account_rx(in_port->bfd, ctx.xin->resubmit_stats);
3085         }
3086     }
3087
3088     special = process_special(&ctx, flow, in_port, ctx.xin->packet);
3089     if (special) {
3090         ctx.xout->slow |= special;
3091     } else {
3092         size_t sample_actions_len;
3093
3094         if (flow->in_port.ofp_port
3095             != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
3096                                       flow->in_port.ofp_port,
3097                                       flow->vlan_tci)) {
3098             ctx.base_flow.vlan_tci = 0;
3099         }
3100
3101         add_sflow_action(&ctx);
3102         add_ipfix_action(&ctx);
3103         sample_actions_len = ctx.xout->odp_actions.size;
3104
3105         if (tnl_may_send && (!in_port || may_receive(in_port, &ctx))) {
3106             do_xlate_actions(ofpacts, ofpacts_len, &ctx);
3107
3108             /* We've let OFPP_NORMAL and the learning action look at the
3109              * packet, so drop it now if forwarding is disabled. */
3110             if (in_port && !xport_stp_forward_state(in_port)) {
3111                 ctx.xout->odp_actions.size = sample_actions_len;
3112             }
3113         }
3114
3115         if (ctx.action_set.size) {
3116             xlate_action_set(&ctx);
3117         }
3118
3119         if (ctx.xbridge->has_in_band
3120             && in_band_must_output_to_local_port(flow)
3121             && !actions_output_to_local_port(&ctx)) {
3122             compose_output_action(&ctx, OFPP_LOCAL);
3123         }
3124
3125         fix_sflow_action(&ctx);
3126
3127         if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
3128             add_mirror_actions(&ctx, &orig_flow);
3129         }
3130     }
3131
3132     if (nl_attr_oversized(ctx.xout->odp_actions.size)) {
3133         /* These datapath actions are too big for a Netlink attribute, so we
3134          * can't hand them to the kernel directly.  dpif_execute() can execute
3135          * them one by one with help, so just mark the result as SLOW_ACTION to
3136          * prevent the flow from being installed. */
3137         COVERAGE_INC(xlate_actions_oversize);
3138         ctx.xout->slow |= SLOW_ACTION;
3139     }
3140
3141     if (ctx.xin->resubmit_stats) {
3142         mirror_update_stats(ctx.xbridge->mbridge, xout->mirrors,
3143                             ctx.xin->resubmit_stats->n_packets,
3144                             ctx.xin->resubmit_stats->n_bytes);
3145
3146         if (ctx.xbridge->netflow) {
3147             const struct ofpact *ofpacts;
3148             size_t ofpacts_len;
3149
3150             ofpacts_len = actions->ofpacts_len;
3151             ofpacts = actions->ofpacts;
3152             if (ofpacts_len == 0
3153                 || ofpacts->type != OFPACT_CONTROLLER
3154                 || ofpact_next(ofpacts) < ofpact_end(ofpacts, ofpacts_len)) {
3155                 /* Only update netflow if we don't have controller flow.  We don't
3156                  * report NetFlow expiration messages for such facets because they
3157                  * are just part of the control logic for the network, not real
3158                  * traffic. */
3159                 netflow_flow_update(ctx.xbridge->netflow, flow,
3160                                     xout->nf_output_iface,
3161                                     ctx.xin->resubmit_stats);
3162             }
3163         }
3164     }
3165
3166     ofpbuf_uninit(&ctx.stack);
3167     ofpbuf_uninit(&ctx.action_set);
3168
3169     /* Clear the metadata and register wildcard masks, because we won't
3170      * use non-header fields as part of the cache. */
3171     flow_wildcards_clear_non_packet_fields(wc);
3172
3173     /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields.  struct flow uses
3174      * the low 8 bits of the 16-bit tp_src and tp_dst members to represent
3175      * these fields.  The datapath interface, on the other hand, represents
3176      * them with just 8 bits each.  This means that if the high 8 bits of the
3177      * masks for these fields somehow become set, then they will get chopped
3178      * off by a round trip through the datapath, and revalidation will spot
3179      * that as an inconsistency and delete the flow.  Avoid the problem here by
3180      * making sure that only the low 8 bits of either field can be unwildcarded
3181      * for ICMP.
3182      */
3183     if (is_icmp) {
3184         wc->masks.tp_src &= htons(UINT8_MAX);
3185         wc->masks.tp_dst &= htons(UINT8_MAX);
3186     }
3187
3188 out:
3189     rule_dpif_unref(rule);
3190 }
3191
3192 /* Sends 'packet' out 'ofport'.
3193  * May modify 'packet'.
3194  * Returns 0 if successful, otherwise a positive errno value. */
3195 int
3196 xlate_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3197 {
3198     struct xport *xport;
3199     struct ofpact_output output;
3200     struct flow flow;
3201
3202     ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
3203     /* Use OFPP_NONE as the in_port to avoid special packet processing. */
3204     flow_extract(packet, NULL, &flow);
3205     flow.in_port.ofp_port = OFPP_NONE;
3206
3207     ovs_rwlock_rdlock(&xlate_rwlock);
3208     xport = xport_lookup(ofport);
3209     if (!xport) {
3210         ovs_rwlock_unlock(&xlate_rwlock);
3211         return EINVAL;
3212     }
3213     output.port = xport->ofp_port;
3214     output.max_len = 0;
3215     ovs_rwlock_unlock(&xlate_rwlock);
3216
3217     return ofproto_dpif_execute_actions(xport->xbridge->ofproto, &flow, NULL,
3218                                         &output.ofpact, sizeof output,
3219                                         packet);
3220 }