Merge 'next' into 'master'.
authorBen Pfaff <blp@nicira.com>
Wed, 18 May 2011 21:01:13 +0000 (14:01 -0700)
committerBen Pfaff <blp@nicira.com>
Wed, 18 May 2011 21:01:13 +0000 (14:01 -0700)
I know already that this breaks the statsfixes that were implemented by the
following commits:

827ab71c97f "ofproto: Datapath statistics accounted twice."
6f1435fc8f7 "ofproto: Resubmit statistics improperly account during..."

These were already broken in a previous merge.  I will work on a fix.

19 files changed:
1  2 
lib/bond.c
lib/dpif-linux.c
lib/dpif-netdev.c
lib/flow.c
lib/flow.h
lib/netdev-linux.c
lib/nx-match.c
lib/ofp-print.c
lib/ofp-util.c
lib/ofp-util.h
ofproto/ofproto-dpif.c
ofproto/ofproto.c
tests/ofproto-macros.at
tests/ovs-ofctl.at
utilities/ovs-controller.c
utilities/ovs-ofctl.c
utilities/ovs-openflowd.c
vswitchd/bridge.c
vswitchd/ovs-vswitchd.c

diff --cc lib/bond.c
Simple merge
Simple merge
Simple merge
diff --cc lib/flow.c
Simple merge
diff --cc lib/flow.h
Simple merge
Simple merge
diff --cc lib/nx-match.c
Simple merge
diff --cc lib/ofp-print.c
Simple merge
diff --cc lib/ofp-util.c
index 8a8a7b5cbcfd070e8ad4bc5b36e1a19b3d1604b4,2e12eb5436072042210027835e05213a7c9f6923..4618567e5148b5651bf6a5cf207ec0bc0d7d7af4
@@@ -912,8 -932,9 +946,8 @@@ ofputil_decode_flow_mod(struct flow_mo
          }
  
          /* Translate the message. */
 -        ofputil_cls_rule_from_match(&match, ntohs(ofm->priority), &fm->cr);
          fm->cookie = ofm->cookie;
-         fm->command = ntohs(ofm->command);
+         command = ntohs(ofm->command);
          fm->idle_timeout = ntohs(ofm->idle_timeout);
          fm->hard_timeout = ntohs(ofm->hard_timeout);
          fm->buffer_id = ntohl(ofm->buffer_id);
diff --cc lib/ofp-util.h
Simple merge
index 0000000000000000000000000000000000000000,26a0fbf639abc1d0c4a25c76691db5bb5717946c..36e7213a99a85953f71f44cc068aa722c06a962d
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,3939 +1,3931 @@@
 -static int send_packet(struct ofproto_dpif *,
 -                       uint32_t odp_port, uint16_t vlan_tci,
+ /*
+  * Copyright (c) 2009, 2010, 2011 Nicira Networks.
+  *
+  * Licensed under the Apache License, Version 2.0 (the "License");
+  * you may not use this file except in compliance with the License.
+  * You may obtain a copy of the License at:
+  *
+  *     http://www.apache.org/licenses/LICENSE-2.0
+  *
+  * Unless required by applicable law or agreed to in writing, software
+  * distributed under the License is distributed on an "AS IS" BASIS,
+  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+  * See the License for the specific language governing permissions and
+  * limitations under the License.
+  */
+ #include <config.h>
+ #include "ofproto/private.h"
+ #include <errno.h>
+ #include "autopath.h"
+ #include "bond.h"
+ #include "byte-order.h"
+ #include "connmgr.h"
+ #include "coverage.h"
+ #include "cfm.h"
+ #include "dpif.h"
+ #include "dynamic-string.h"
+ #include "fail-open.h"
+ #include "hmapx.h"
+ #include "lacp.h"
+ #include "mac-learning.h"
+ #include "multipath.h"
+ #include "netdev.h"
+ #include "netlink.h"
+ #include "nx-match.h"
+ #include "odp-util.h"
+ #include "ofp-util.h"
+ #include "ofpbuf.h"
+ #include "ofp-print.h"
+ #include "ofproto-sflow.h"
+ #include "poll-loop.h"
+ #include "timer.h"
+ #include "unaligned.h"
+ #include "unixctl.h"
+ #include "vlan-bitmap.h"
+ #include "vlog.h"
+ VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
+ COVERAGE_DEFINE(ofproto_dpif_ctlr_action);
+ COVERAGE_DEFINE(ofproto_dpif_expired);
+ COVERAGE_DEFINE(ofproto_dpif_no_packet_in);
+ COVERAGE_DEFINE(ofproto_dpif_xlate);
+ COVERAGE_DEFINE(facet_changed_rule);
+ COVERAGE_DEFINE(facet_invalidated);
+ COVERAGE_DEFINE(facet_revalidate);
+ COVERAGE_DEFINE(facet_unexpected);
+ /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
+  * flow translation. */
+ #define MAX_RESUBMIT_RECURSION 16
+ struct ofport_dpif;
+ struct ofproto_dpif;
+ struct rule_dpif {
+     struct rule up;
+     long long int used;         /* Time last used; time created if not used. */
+     /* These statistics:
+      *
+      *   - Do include packets and bytes from facets that have been deleted or
+      *     whose own statistics have been folded into the rule.
+      *
+      *   - Do include packets and bytes sent "by hand" that were accounted to
+      *     the rule without any facet being involved (this is a rare corner
+      *     case in rule_execute()).
+      *
+      *   - Do not include packet or bytes that can be obtained from any facet's
+      *     packet_count or byte_count member or that can be obtained from the
+      *     datapath by, e.g., dpif_flow_get() for any facet.
+      */
+     uint64_t packet_count;       /* Number of packets received. */
+     uint64_t byte_count;         /* Number of bytes received. */
+     struct list facets;          /* List of "struct facet"s. */
+ };
+ static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
+ {
+     return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
+ }
+ static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *ofproto,
+                                           const struct flow *flow);
+ #define MAX_MIRRORS 32
+ typedef uint32_t mirror_mask_t;
+ #define MIRROR_MASK_C(X) UINT32_C(X)
+ BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
+ struct ofmirror {
+     struct ofproto_dpif *ofproto; /* Owning ofproto. */
+     size_t idx;                 /* In ofproto's "mirrors" array. */
+     void *aux;                  /* Key supplied by ofproto's client. */
+     char *name;                 /* Identifier for log messages. */
+     /* Selection criteria. */
+     struct hmapx srcs;          /* Contains "struct ofbundle *"s. */
+     struct hmapx dsts;          /* Contains "struct ofbundle *"s. */
+     unsigned long *vlans;       /* Bitmap of chosen VLANs, NULL selects all. */
+     /* Output (mutually exclusive). */
+     struct ofbundle *out;       /* Output port or NULL. */
+     int out_vlan;               /* Output VLAN or -1. */
+ };
+ static void mirror_destroy(struct ofmirror *);
+ /* A group of one or more OpenFlow ports. */
+ #define OFBUNDLE_FLOOD ((struct ofbundle *) 1)
+ struct ofbundle {
+     struct ofproto_dpif *ofproto; /* Owning ofproto. */
+     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
+     void *aux;                  /* Key supplied by ofproto's client. */
+     char *name;                 /* Identifier for log messages. */
+     /* Configuration. */
+     struct list ports;          /* Contains "struct ofport"s. */
+     int vlan;                   /* -1=trunk port, else a 12-bit VLAN ID. */
+     unsigned long *trunks;      /* Bitmap of trunked VLANs, if 'vlan' == -1.
+                                  * NULL if all VLANs are trunked. */
+     struct lacp *lacp;          /* LACP if LACP is enabled, otherwise NULL. */
+     struct bond *bond;          /* Nonnull iff more than one port. */
+     /* Status. */
+     bool floodable;             /* True if no port has OFPPC_NO_FLOOD set. */
+     /* Port mirroring info. */
+     mirror_mask_t src_mirrors;  /* Mirrors triggered when packet received. */
+     mirror_mask_t dst_mirrors;  /* Mirrors triggered when packet sent. */
+     mirror_mask_t mirror_out;   /* Mirrors that output to this bundle. */
+ };
+ static void bundle_remove(struct ofport *);
+ static void bundle_destroy(struct ofbundle *);
+ static void bundle_del_port(struct ofport_dpif *);
+ static void bundle_run(struct ofbundle *);
+ static void bundle_wait(struct ofbundle *);
+ struct action_xlate_ctx {
+ /* action_xlate_ctx_init() initializes these members. */
+     /* The ofproto. */
+     struct ofproto_dpif *ofproto;
+     /* Flow to which the OpenFlow actions apply.  xlate_actions() will modify
+      * this flow when actions change header fields. */
+     struct flow flow;
+     /* The packet corresponding to 'flow', or a null pointer if we are
+      * revalidating without a packet to refer to. */
+     const struct ofpbuf *packet;
+     /* If nonnull, called just before executing a resubmit action.
+      *
+      * This is normally null so the client has to set it manually after
+      * calling action_xlate_ctx_init(). */
+     void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *);
+ /* xlate_actions() initializes and uses these members.  The client might want
+  * to look at them after it returns. */
+     struct ofpbuf *odp_actions; /* Datapath actions. */
+     tag_type tags;              /* Tags associated with OFPP_NORMAL actions. */
+     bool may_set_up_flow;       /* True ordinarily; false if the actions must
+                                  * be reassessed for every packet. */
+     uint16_t nf_output_iface;   /* Output interface index for NetFlow. */
+ /* xlate_actions() initializes and uses these members, but the client has no
+  * reason to look at them. */
+     int recurse;                /* Recursion level, via xlate_table_action. */
+     int last_pop_priority;      /* Offset in 'odp_actions' just past most
+                                  * recent ODP_ACTION_ATTR_SET_PRIORITY. */
+ };
+ static void action_xlate_ctx_init(struct action_xlate_ctx *,
+                                   struct ofproto_dpif *, const struct flow *,
+                                   const struct ofpbuf *);
+ static struct ofpbuf *xlate_actions(struct action_xlate_ctx *,
+                                     const union ofp_action *in, size_t n_in);
+ /* An exact-match instantiation of an OpenFlow flow. */
+ struct facet {
+     long long int used;         /* Time last used; time created if not used. */
+     /* These statistics:
+      *
+      *   - Do include packets and bytes sent "by hand", e.g. with
+      *     dpif_execute().
+      *
+      *   - Do include packets and bytes that were obtained from the datapath
+      *     when a flow was deleted (e.g. dpif_flow_del()) or when its
+      *     statistics were reset (e.g. dpif_flow_put() with
+      *     DPIF_FP_ZERO_STATS).
+      *
+      *   - Do not include any packets or bytes that can currently be obtained
+      *     from the datapath by, e.g., dpif_flow_get().
+      */
+     uint64_t packet_count;       /* Number of packets received. */
+     uint64_t byte_count;         /* Number of bytes received. */
+     uint64_t dp_packet_count;    /* Last known packet count in the datapath. */
+     uint64_t dp_byte_count;      /* Last known byte count in the datapath. */
+     uint64_t rs_packet_count;    /* Packets pushed to resubmit children. */
+     uint64_t rs_byte_count;      /* Bytes pushed to resubmit children. */
+     long long int rs_used;       /* Used time pushed to resubmit children. */
+     /* Number of bytes passed to account_cb.  This may include bytes that can
+      * currently obtained from the datapath (thus, it can be greater than
+      * byte_count). */
+     uint64_t accounted_bytes;
+     struct hmap_node hmap_node;  /* In owning ofproto's 'facets' hmap. */
+     struct list list_node;       /* In owning rule's 'facets' list. */
+     struct rule_dpif *rule;      /* Owning rule. */
+     struct flow flow;            /* Exact-match flow. */
+     bool installed;              /* Installed in datapath? */
+     bool may_install;            /* True ordinarily; false if actions must
+                                   * be reassessed for every packet. */
+     size_t actions_len;          /* Number of bytes in actions[]. */
+     struct nlattr *actions;      /* Datapath actions. */
+     tag_type tags;               /* Tags. */
+     struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
+ };
+ static struct facet *facet_create(struct rule_dpif *, const struct flow *,
+                                   const struct ofpbuf *packet);
+ static void facet_remove(struct ofproto_dpif *, struct facet *);
+ static void facet_free(struct facet *);
+ static struct facet *facet_find(struct ofproto_dpif *, const struct flow *);
+ static struct facet *facet_lookup_valid(struct ofproto_dpif *,
+                                         const struct flow *);
+ static bool facet_revalidate(struct ofproto_dpif *, struct facet *);
+ static void facet_execute(struct ofproto_dpif *, struct facet *,
+                           struct ofpbuf *packet);
+ static int facet_put__(struct ofproto_dpif *, struct facet *,
+                        const struct nlattr *actions, size_t actions_len,
+                        struct dpif_flow_stats *);
+ static void facet_install(struct ofproto_dpif *, struct facet *,
+                           bool zero_stats);
+ static void facet_uninstall(struct ofproto_dpif *, struct facet *);
+ static void facet_flush_stats(struct ofproto_dpif *, struct facet *);
+ static void facet_make_actions(struct ofproto_dpif *, struct facet *,
+                                const struct ofpbuf *packet);
+ static void facet_update_time(struct ofproto_dpif *, struct facet *,
+                               long long int used);
+ static void facet_update_stats(struct ofproto_dpif *, struct facet *,
+                                const struct dpif_flow_stats *);
+ static void facet_push_stats(struct facet *);
+ static void facet_account(struct ofproto_dpif *, struct facet *,
+                           uint64_t extra_bytes);
+ static bool facet_is_controller_flow(struct facet *);
+ static void flow_push_stats(const struct rule_dpif *,
+                             struct flow *, uint64_t packets, uint64_t bytes,
+                             long long int used);
+ struct ofport_dpif {
+     struct ofport up;
+     uint32_t odp_port;
+     struct ofbundle *bundle;    /* Bundle that contains this port, if any. */
+     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
+     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
+     tag_type tag;               /* Tag associated with this port. */
+     uint32_t bond_stable_id;    /* stable_id to use as bond slave, or 0. */
+ };
+ static struct ofport_dpif *
+ ofport_dpif_cast(const struct ofport *ofport)
+ {
+     assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
+     return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
+ }
+ static void port_run(struct ofport_dpif *);
+ static void port_wait(struct ofport_dpif *);
+ static int set_cfm(struct ofport *, const struct cfm *,
+                    const uint16_t *remote_mps, size_t n_remote_mps);
+ struct ofproto_dpif {
+     struct ofproto up;
+     struct dpif *dpif;
+     int max_ports;
+     /* Statistics. */
+     uint64_t n_matches;
+     /* Bridging. */
+     struct netflow *netflow;
+     struct ofproto_sflow *sflow;
+     struct hmap bundles;        /* Contains "struct ofbundle"s. */
+     struct mac_learning *ml;
+     struct ofmirror *mirrors[MAX_MIRRORS];
+     bool has_bonded_bundles;
+     /* Expiration. */
+     struct timer next_expiration;
+     /* Facets. */
+     struct hmap facets;
+     bool need_revalidate;
+     struct tag_set revalidate_set;
+ };
+ static void ofproto_dpif_unixctl_init(void);
+ static struct ofproto_dpif *
+ ofproto_dpif_cast(const struct ofproto *ofproto)
+ {
+     assert(ofproto->ofproto_class == &ofproto_dpif_class);
+     return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
+ }
+ static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *,
+                                         uint16_t ofp_port);
+ static struct ofport_dpif *get_odp_port(struct ofproto_dpif *,
+                                         uint32_t odp_port);
+ /* Packet processing. */
+ static void update_learning_table(struct ofproto_dpif *,
+                                   const struct flow *, int vlan,
+                                   struct ofbundle *);
+ static bool is_admissible(struct ofproto_dpif *, const struct flow *,
+                           bool have_packet, tag_type *, int *vlanp,
+                           struct ofbundle **in_bundlep);
+ static void handle_upcall(struct ofproto_dpif *, struct dpif_upcall *);
+ /* Flow expiration. */
+ static int expire(struct ofproto_dpif *);
+ /* Utilities. */
 -                        ofport->odp_port, 0, &packet);
++static int send_packet(struct ofproto_dpif *, uint32_t odp_port,
+                        const struct ofpbuf *packet);
+ /* Global variables. */
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
\f
+ /* Factory functions. */
+ static void
+ enumerate_types(struct sset *types)
+ {
+     dp_enumerate_types(types);
+ }
+ static int
+ enumerate_names(const char *type, struct sset *names)
+ {
+     return dp_enumerate_names(type, names);
+ }
+ static int
+ del(const char *type, const char *name)
+ {
+     struct dpif *dpif;
+     int error;
+     error = dpif_open(name, type, &dpif);
+     if (!error) {
+         error = dpif_delete(dpif);
+         dpif_close(dpif);
+     }
+     return error;
+ }
\f
+ /* Basic life-cycle. */
+ static struct ofproto *
+ alloc(void)
+ {
+     struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
+     return &ofproto->up;
+ }
+ static void
+ dealloc(struct ofproto *ofproto_)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     free(ofproto);
+ }
+ static int
+ construct(struct ofproto *ofproto_)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     const char *name = ofproto->up.name;
+     int error;
+     int i;
+     error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif);
+     if (error) {
+         VLOG_ERR("failed to open datapath %s: %s", name, strerror(error));
+         return error;
+     }
+     ofproto->max_ports = dpif_get_max_ports(ofproto->dpif);
+     ofproto->n_matches = 0;
+     error = dpif_recv_set_mask(ofproto->dpif,
+                                ((1u << DPIF_UC_MISS) |
+                                 (1u << DPIF_UC_ACTION) |
+                                 (1u << DPIF_UC_SAMPLE)));
+     if (error) {
+         VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error));
+         dpif_close(ofproto->dpif);
+         return error;
+     }
+     dpif_flow_flush(ofproto->dpif);
+     dpif_recv_purge(ofproto->dpif);
+     ofproto->netflow = NULL;
+     ofproto->sflow = NULL;
+     hmap_init(&ofproto->bundles);
+     ofproto->ml = mac_learning_create();
+     for (i = 0; i < MAX_MIRRORS; i++) {
+         ofproto->mirrors[i] = NULL;
+     }
+     ofproto->has_bonded_bundles = false;
+     timer_set_duration(&ofproto->next_expiration, 1000);
+     hmap_init(&ofproto->facets);
+     ofproto->need_revalidate = false;
+     tag_set_init(&ofproto->revalidate_set);
+     ofproto->up.tables = xmalloc(sizeof *ofproto->up.tables);
+     classifier_init(&ofproto->up.tables[0]);
+     ofproto->up.n_tables = 1;
+     ofproto_dpif_unixctl_init();
+     return 0;
+ }
+ static void
+ destruct(struct ofproto *ofproto_)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     int i;
+     for (i = 0; i < MAX_MIRRORS; i++) {
+         mirror_destroy(ofproto->mirrors[i]);
+     }
+     netflow_destroy(ofproto->netflow);
+     ofproto_sflow_destroy(ofproto->sflow);
+     hmap_destroy(&ofproto->bundles);
+     mac_learning_destroy(ofproto->ml);
+     hmap_destroy(&ofproto->facets);
+     dpif_close(ofproto->dpif);
+ }
+ static int
+ run(struct ofproto *ofproto_)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     struct ofport_dpif *ofport;
+     struct ofbundle *bundle;
+     int i;
+     dpif_run(ofproto->dpif);
+     for (i = 0; i < 50; i++) {
+         struct dpif_upcall packet;
+         int error;
+         error = dpif_recv(ofproto->dpif, &packet);
+         if (error) {
+             if (error == ENODEV) {
+                 /* Datapath destroyed. */
+                 return error;
+             }
+             break;
+         }
+         handle_upcall(ofproto, &packet);
+     }
+     if (timer_expired(&ofproto->next_expiration)) {
+         int delay = expire(ofproto);
+         timer_set_duration(&ofproto->next_expiration, delay);
+     }
+     if (ofproto->netflow) {
+         netflow_run(ofproto->netflow);
+     }
+     if (ofproto->sflow) {
+         ofproto_sflow_run(ofproto->sflow);
+     }
+     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
+         port_run(ofport);
+     }
+     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
+         bundle_run(bundle);
+     }
+     /* Now revalidate if there's anything to do. */
+     if (ofproto->need_revalidate
+         || !tag_set_is_empty(&ofproto->revalidate_set)) {
+         struct tag_set revalidate_set = ofproto->revalidate_set;
+         bool revalidate_all = ofproto->need_revalidate;
+         struct facet *facet, *next;
+         /* Clear the revalidation flags. */
+         tag_set_init(&ofproto->revalidate_set);
+         ofproto->need_revalidate = false;
+         HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
+             if (revalidate_all
+                 || tag_set_intersects(&revalidate_set, facet->tags)) {
+                 facet_revalidate(ofproto, facet);
+             }
+         }
+     }
+     return 0;
+ }
+ static void
+ wait(struct ofproto *ofproto_)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     struct ofport_dpif *ofport;
+     struct ofbundle *bundle;
+     dpif_wait(ofproto->dpif);
+     dpif_recv_wait(ofproto->dpif);
+     if (ofproto->sflow) {
+         ofproto_sflow_wait(ofproto->sflow);
+     }
+     if (!tag_set_is_empty(&ofproto->revalidate_set)) {
+         poll_immediate_wake();
+     }
+     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
+         port_wait(ofport);
+     }
+     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
+         bundle_wait(bundle);
+     }
+     if (ofproto->need_revalidate) {
+         /* Shouldn't happen, but if it does just go around again. */
+         VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
+         poll_immediate_wake();
+     } else {
+         timer_wait(&ofproto->next_expiration);
+     }
+ }
+ static void
+ flush(struct ofproto *ofproto_)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     struct facet *facet, *next_facet;
+     HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
+         /* Mark the facet as not installed so that facet_remove() doesn't
+          * bother trying to uninstall it.  There is no point in uninstalling it
+          * individually since we are about to blow away all the facets with
+          * dpif_flow_flush(). */
+         facet->installed = false;
+         facet->dp_packet_count = 0;
+         facet->dp_byte_count = 0;
+         facet_remove(ofproto, facet);
+     }
+     dpif_flow_flush(ofproto->dpif);
+ }
+ static void
+ get_features(struct ofproto *ofproto_ OVS_UNUSED,
+              bool *arp_match_ip, uint32_t *actions)
+ {
+     *arp_match_ip = true;
+     *actions = ((1u << OFPAT_OUTPUT) |
+                 (1u << OFPAT_SET_VLAN_VID) |
+                 (1u << OFPAT_SET_VLAN_PCP) |
+                 (1u << OFPAT_STRIP_VLAN) |
+                 (1u << OFPAT_SET_DL_SRC) |
+                 (1u << OFPAT_SET_DL_DST) |
+                 (1u << OFPAT_SET_NW_SRC) |
+                 (1u << OFPAT_SET_NW_DST) |
+                 (1u << OFPAT_SET_NW_TOS) |
+                 (1u << OFPAT_SET_TP_SRC) |
+                 (1u << OFPAT_SET_TP_DST) |
+                 (1u << OFPAT_ENQUEUE));
+ }
+ static void
+ get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     struct odp_stats s;
+     strcpy(ots->name, "classifier");
+     dpif_get_dp_stats(ofproto->dpif, &s);
+     put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed));
+     put_32aligned_be64(&ots->matched_count,
+                        htonll(s.n_hit + ofproto->n_matches));
+ }
+ static int
+ set_netflow(struct ofproto *ofproto_,
+             const struct netflow_options *netflow_options)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     if (netflow_options) {
+         if (!ofproto->netflow) {
+             ofproto->netflow = netflow_create();
+         }
+         return netflow_set_options(ofproto->netflow, netflow_options);
+     } else {
+         netflow_destroy(ofproto->netflow);
+         ofproto->netflow = NULL;
+         return 0;
+     }
+ }
+ static struct ofport *
+ port_alloc(void)
+ {
+     struct ofport_dpif *port = xmalloc(sizeof *port);
+     return &port->up;
+ }
+ static void
+ port_dealloc(struct ofport *port_)
+ {
+     struct ofport_dpif *port = ofport_dpif_cast(port_);
+     free(port);
+ }
+ static int
+ port_construct(struct ofport *port_)
+ {
+     struct ofport_dpif *port = ofport_dpif_cast(port_);
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
+     port->odp_port = ofp_port_to_odp_port(port->up.ofp_port);
+     port->bundle = NULL;
+     port->cfm = NULL;
+     port->tag = tag_create_random();
+     if (ofproto->sflow) {
+         ofproto_sflow_add_port(ofproto->sflow, port->odp_port,
+                                netdev_get_name(port->up.netdev));
+     }
+     return 0;
+ }
+ static void
+ port_destruct(struct ofport *port_)
+ {
+     struct ofport_dpif *port = ofport_dpif_cast(port_);
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
+     bundle_remove(port_);
+     set_cfm(port_, NULL, NULL, 0);
+     if (ofproto->sflow) {
+         ofproto_sflow_del_port(ofproto->sflow, port->odp_port);
+     }
+ }
+ static void
+ port_modified(struct ofport *port_)
+ {
+     struct ofport_dpif *port = ofport_dpif_cast(port_);
+     if (port->bundle && port->bundle->bond) {
+         bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
+     }
+ }
+ static void
+ port_reconfigured(struct ofport *port_, ovs_be32 old_config)
+ {
+     struct ofport_dpif *port = ofport_dpif_cast(port_);
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
+     ovs_be32 changed = old_config ^ port->up.opp.config;
+     if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
+                         OFPPC_NO_FWD | OFPPC_NO_FLOOD)) {
+         ofproto->need_revalidate = true;
+     }
+ }
+ static int
+ set_sflow(struct ofproto *ofproto_,
+           const struct ofproto_sflow_options *sflow_options)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     struct ofproto_sflow *os = ofproto->sflow;
+     if (sflow_options) {
+         if (!os) {
+             struct ofport_dpif *ofport;
+             os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif);
+             HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
+                 ofproto_sflow_add_port(os, ofport->odp_port,
+                                        netdev_get_name(ofport->up.netdev));
+             }
+         }
+         ofproto_sflow_set_options(os, sflow_options);
+     } else {
+         ofproto_sflow_destroy(os);
+         ofproto->sflow = NULL;
+     }
+     return 0;
+ }
+ static int
+ set_cfm(struct ofport *ofport_, const struct cfm *cfm,
+         const uint16_t *remote_mps, size_t n_remote_mps)
+ {
+     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
+     int error;
+     if (!cfm) {
+         error = 0;
+     } else {
+         if (!ofport->cfm) {
+             ofport->cfm = cfm_create();
+         }
+         ofport->cfm->mpid = cfm->mpid;
+         ofport->cfm->interval = cfm->interval;
+         memcpy(ofport->cfm->maid, cfm->maid, CCM_MAID_LEN);
+         cfm_update_remote_mps(ofport->cfm, remote_mps, n_remote_mps);
+         if (cfm_configure(ofport->cfm)) {
+             return 0;
+         }
+         error = EINVAL;
+     }
+     cfm_destroy(ofport->cfm);
+     ofport->cfm = NULL;
+     return error;
+ }
+ static int
+ get_cfm(struct ofport *ofport_, const struct cfm **cfmp)
+ {
+     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
+     *cfmp = ofport->cfm;
+     return 0;
+ }
\f
+ /* Bundles. */
+ /* Expires all MAC learning entries associated with 'port' and forces ofproto
+  * to revalidate every flow. */
+ static void
+ bundle_flush_macs(struct ofbundle *bundle)
+ {
+     struct ofproto_dpif *ofproto = bundle->ofproto;
+     struct mac_learning *ml = ofproto->ml;
+     struct mac_entry *mac, *next_mac;
+     ofproto->need_revalidate = true;
+     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
+         if (mac->port.p == bundle) {
+             mac_learning_expire(ml, mac);
+         }
+     }
+ }
+ static struct ofbundle *
+ bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
+ {
+     struct ofbundle *bundle;
+     HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
+                              &ofproto->bundles) {
+         if (bundle->aux == aux) {
+             return bundle;
+         }
+     }
+     return NULL;
+ }
+ /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
+  * ones that are found to 'bundles'. */
+ static void
+ bundle_lookup_multiple(struct ofproto_dpif *ofproto,
+                        void **auxes, size_t n_auxes,
+                        struct hmapx *bundles)
+ {
+     size_t i;
+     hmapx_init(bundles);
+     for (i = 0; i < n_auxes; i++) {
+         struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
+         if (bundle) {
+             hmapx_add(bundles, bundle);
+         }
+     }
+ }
+ static void
+ bundle_del_port(struct ofport_dpif *port)
+ {
+     struct ofbundle *bundle = port->bundle;
+     bundle->ofproto->need_revalidate = true;
+     list_remove(&port->bundle_node);
+     port->bundle = NULL;
+     if (bundle->lacp) {
+         lacp_slave_unregister(bundle->lacp, port);
+     }
+     if (bundle->bond) {
+         bond_slave_unregister(bundle->bond, port);
+     }
+     bundle->floodable = true;
+     LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
+         if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
+             bundle->floodable = false;
+         }
+     }
+ }
+ static bool
+ bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
+                 struct lacp_slave_settings *lacp,
+                 uint32_t bond_stable_id)
+ {
+     struct ofport_dpif *port;
+     port = get_ofp_port(bundle->ofproto, ofp_port);
+     if (!port) {
+         return false;
+     }
+     if (port->bundle != bundle) {
+         bundle->ofproto->need_revalidate = true;
+         if (port->bundle) {
+             bundle_del_port(port);
+         }
+         port->bundle = bundle;
+         list_push_back(&bundle->ports, &port->bundle_node);
+         if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) {
+             bundle->floodable = false;
+         }
+     }
+     if (lacp) {
+         lacp_slave_register(bundle->lacp, port, lacp);
+     }
+     port->bond_stable_id = bond_stable_id;
+     return true;
+ }
+ static void
+ bundle_destroy(struct ofbundle *bundle)
+ {
+     struct ofproto_dpif *ofproto;
+     struct ofport_dpif *port, *next_port;
+     int i;
+     if (!bundle) {
+         return;
+     }
+     ofproto = bundle->ofproto;
+     for (i = 0; i < MAX_MIRRORS; i++) {
+         struct ofmirror *m = ofproto->mirrors[i];
+         if (m) {
+             if (m->out == bundle) {
+                 mirror_destroy(m);
+             } else if (hmapx_find_and_delete(&m->srcs, bundle)
+                        || hmapx_find_and_delete(&m->dsts, bundle)) {
+                 ofproto->need_revalidate = true;
+             }
+         }
+     }
+     LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
+         bundle_del_port(port);
+     }
+     bundle_flush_macs(bundle);
+     hmap_remove(&ofproto->bundles, &bundle->hmap_node);
+     free(bundle->name);
+     free(bundle->trunks);
+     lacp_destroy(bundle->lacp);
+     bond_destroy(bundle->bond);
+     free(bundle);
+ }
+ static int
+ bundle_set(struct ofproto *ofproto_, void *aux,
+            const struct ofproto_bundle_settings *s)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     bool need_flush = false;
+     const unsigned long *trunks;
+     struct ofport_dpif *port;
+     struct ofbundle *bundle;
+     size_t i;
+     bool ok;
+     if (!s) {
+         bundle_destroy(bundle_lookup(ofproto, aux));
+         return 0;
+     }
+     assert(s->n_slaves == 1 || s->bond != NULL);
+     assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
+     bundle = bundle_lookup(ofproto, aux);
+     if (!bundle) {
+         bundle = xmalloc(sizeof *bundle);
+         bundle->ofproto = ofproto;
+         hmap_insert(&ofproto->bundles, &bundle->hmap_node,
+                     hash_pointer(aux, 0));
+         bundle->aux = aux;
+         bundle->name = NULL;
+         list_init(&bundle->ports);
+         bundle->vlan = -1;
+         bundle->trunks = NULL;
+         bundle->lacp = NULL;
+         bundle->bond = NULL;
+         bundle->floodable = true;
+         bundle->src_mirrors = 0;
+         bundle->dst_mirrors = 0;
+         bundle->mirror_out = 0;
+     }
+     if (!bundle->name || strcmp(s->name, bundle->name)) {
+         free(bundle->name);
+         bundle->name = xstrdup(s->name);
+     }
+     /* LACP. */
+     if (s->lacp) {
+         if (!bundle->lacp) {
+             bundle->lacp = lacp_create();
+         }
+         lacp_configure(bundle->lacp, s->lacp);
+     } else {
+         lacp_destroy(bundle->lacp);
+         bundle->lacp = NULL;
+     }
+     /* Update set of ports. */
+     ok = true;
+     for (i = 0; i < s->n_slaves; i++) {
+         if (!bundle_add_port(bundle, s->slaves[i],
+                              s->lacp ? &s->lacp_slaves[i] : NULL,
+                              s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
+             ok = false;
+         }
+     }
+     if (!ok || list_size(&bundle->ports) != s->n_slaves) {
+         struct ofport_dpif *next_port;
+         LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
+             for (i = 0; i < s->n_slaves; i++) {
+                 if (s->slaves[i] == port->up.ofp_port) {
+                     goto found;
+                 }
+             }
+             bundle_del_port(port);
+         found: ;
+         }
+     }
+     assert(list_size(&bundle->ports) <= s->n_slaves);
+     if (list_is_empty(&bundle->ports)) {
+         bundle_destroy(bundle);
+         return EINVAL;
+     }
+     /* Set VLAN tag. */
+     if (s->vlan != bundle->vlan) {
+         bundle->vlan = s->vlan;
+         need_flush = true;
+     }
+     /* Get trunked VLANs. */
+     trunks = s->vlan == -1 ? NULL : s->trunks;
+     if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
+         free(bundle->trunks);
+         bundle->trunks = vlan_bitmap_clone(trunks);
+         need_flush = true;
+     }
+     /* Bonding. */
+     if (!list_is_short(&bundle->ports)) {
+         bundle->ofproto->has_bonded_bundles = true;
+         if (bundle->bond) {
+             if (bond_reconfigure(bundle->bond, s->bond)) {
+                 ofproto->need_revalidate = true;
+             }
+         } else {
+             bundle->bond = bond_create(s->bond);
+             ofproto->need_revalidate = true;
+         }
+         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
+             bond_slave_register(bundle->bond, port, port->bond_stable_id,
+                                 port->up.netdev);
+         }
+     } else {
+         bond_destroy(bundle->bond);
+         bundle->bond = NULL;
+     }
+     /* If we changed something that would affect MAC learning, un-learn
+      * everything on this port and force flow revalidation. */
+     if (need_flush) {
+         bundle_flush_macs(bundle);
+     }
+     return 0;
+ }
+ static void
+ bundle_remove(struct ofport *port_)
+ {
+     struct ofport_dpif *port = ofport_dpif_cast(port_);
+     struct ofbundle *bundle = port->bundle;
+     if (bundle) {
+         bundle_del_port(port);
+         if (list_is_empty(&bundle->ports)) {
+             bundle_destroy(bundle);
+         } else if (list_is_short(&bundle->ports)) {
+             bond_destroy(bundle->bond);
+             bundle->bond = NULL;
+         }
+     }
+ }
+ static void
+ send_pdu_cb(void *port_, const struct lacp_pdu *pdu)
+ {
+     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
+     struct ofport_dpif *port = port_;
+     uint8_t ea[ETH_ADDR_LEN];
+     int error;
+     error = netdev_get_etheraddr(port->up.netdev, ea);
+     if (!error) {
+         struct lacp_pdu *packet_pdu;
+         struct ofpbuf packet;
+         ofpbuf_init(&packet, 0);
+         packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
+                                  sizeof *packet_pdu);
+         *packet_pdu = *pdu;
+         error = netdev_send(port->up.netdev, &packet);
+         if (error) {
+             VLOG_WARN_RL(&rl, "port %s: sending LACP PDU on iface %s failed "
+                          "(%s)", port->bundle->name,
+                          netdev_get_name(port->up.netdev), strerror(error));
+         }
+         ofpbuf_uninit(&packet);
+     } else {
+         VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
+                     "%s (%s)", port->bundle->name,
+                     netdev_get_name(port->up.netdev), strerror(error));
+     }
+ }
+ static void
+ bundle_send_learning_packets(struct ofbundle *bundle)
+ {
+     struct ofproto_dpif *ofproto = bundle->ofproto;
+     int error, n_packets, n_errors;
+     struct mac_entry *e;
+     error = n_packets = n_errors = 0;
+     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
+         if (e->port.p != bundle) {
+             int ret = bond_send_learning_packet(bundle->bond, e->mac, e->vlan);
+             if (ret) {
+                 error = ret;
+                 n_errors++;
+             }
+             n_packets++;
+         }
+     }
+     if (n_errors) {
+         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+         VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
+                      "packets, last error was: %s",
+                      bundle->name, n_errors, n_packets, strerror(error));
+     } else {
+         VLOG_DBG("bond %s: sent %d gratuitous learning packets",
+                  bundle->name, n_packets);
+     }
+ }
+ static void
+ bundle_run(struct ofbundle *bundle)
+ {
+     if (bundle->lacp) {
+         lacp_run(bundle->lacp, send_pdu_cb);
+     }
+     if (bundle->bond) {
+         struct ofport_dpif *port;
+         LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
+             bool may_enable = lacp_slave_may_enable(bundle->lacp, port);
+             bond_slave_set_lacp_may_enable(bundle->bond, port, may_enable);
+         }
+         bond_run(bundle->bond, &bundle->ofproto->revalidate_set,
+                  lacp_negotiated(bundle->lacp));
+         if (bond_should_send_learning_packets(bundle->bond)) {
+             bundle_send_learning_packets(bundle);
+         }
+     }
+ }
+ static void
+ bundle_wait(struct ofbundle *bundle)
+ {
+     if (bundle->lacp) {
+         lacp_wait(bundle->lacp);
+     }
+     if (bundle->bond) {
+         bond_wait(bundle->bond);
+     }
+ }
\f
+ /* Mirrors. */
+ static int
+ mirror_scan(struct ofproto_dpif *ofproto)
+ {
+     int idx;
+     for (idx = 0; idx < MAX_MIRRORS; idx++) {
+         if (!ofproto->mirrors[idx]) {
+             return idx;
+         }
+     }
+     return -1;
+ }
+ static struct ofmirror *
+ mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
+ {
+     int i;
+     for (i = 0; i < MAX_MIRRORS; i++) {
+         struct ofmirror *mirror = ofproto->mirrors[i];
+         if (mirror && mirror->aux == aux) {
+             return mirror;
+         }
+     }
+     return NULL;
+ }
+ static int
+ mirror_set(struct ofproto *ofproto_, void *aux,
+            const struct ofproto_mirror_settings *s)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     mirror_mask_t mirror_bit;
+     struct ofbundle *bundle;
+     struct ofmirror *mirror;
+     struct ofbundle *out;
+     struct hmapx srcs;          /* Contains "struct ofbundle *"s. */
+     struct hmapx dsts;          /* Contains "struct ofbundle *"s. */
+     int out_vlan;
+     mirror = mirror_lookup(ofproto, aux);
+     if (!s) {
+         mirror_destroy(mirror);
+         return 0;
+     }
+     if (!mirror) {
+         int idx;
+         idx = mirror_scan(ofproto);
+         if (idx < 0) {
+             VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
+                       "cannot create %s",
+                       ofproto->up.name, MAX_MIRRORS, s->name);
+             return EFBIG;
+         }
+         mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
+         mirror->ofproto = ofproto;
+         mirror->idx = idx;
+         mirror->out_vlan = -1;
+         mirror->name = NULL;
+     }
+     if (!mirror->name || strcmp(s->name, mirror->name)) {
+         free(mirror->name);
+         mirror->name = xstrdup(s->name);
+     }
+     /* Get the new configuration. */
+     if (s->out_bundle) {
+         out = bundle_lookup(ofproto, s->out_bundle);
+         if (!out) {
+             mirror_destroy(mirror);
+             return EINVAL;
+         }
+         out_vlan = -1;
+     } else {
+         out = NULL;
+         out_vlan = s->out_vlan;
+     }
+     bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
+     bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
+     /* If the configuration has not changed, do nothing. */
+     if (hmapx_equals(&srcs, &mirror->srcs)
+         && hmapx_equals(&dsts, &mirror->dsts)
+         && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
+         && mirror->out == out
+         && mirror->out_vlan == out_vlan)
+     {
+         hmapx_destroy(&srcs);
+         hmapx_destroy(&dsts);
+         return 0;
+     }
+     hmapx_swap(&srcs, &mirror->srcs);
+     hmapx_destroy(&srcs);
+     hmapx_swap(&dsts, &mirror->dsts);
+     hmapx_destroy(&dsts);
+     free(mirror->vlans);
+     mirror->vlans = vlan_bitmap_clone(s->src_vlans);
+     mirror->out = out;
+     mirror->out_vlan = out_vlan;
+     /* Update bundles. */
+     mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
+     HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
+         if (hmapx_contains(&mirror->srcs, bundle)) {
+             bundle->src_mirrors |= mirror_bit;
+         } else {
+             bundle->src_mirrors &= ~mirror_bit;
+         }
+         if (hmapx_contains(&mirror->dsts, bundle)) {
+             bundle->dst_mirrors |= mirror_bit;
+         } else {
+             bundle->dst_mirrors &= ~mirror_bit;
+         }
+         if (mirror->out == bundle) {
+             bundle->mirror_out |= mirror_bit;
+         } else {
+             bundle->mirror_out &= ~mirror_bit;
+         }
+     }
+     ofproto->need_revalidate = true;
+     mac_learning_flush(ofproto->ml);
+     return 0;
+ }
+ static void
+ mirror_destroy(struct ofmirror *mirror)
+ {
+     struct ofproto_dpif *ofproto;
+     mirror_mask_t mirror_bit;
+     struct ofbundle *bundle;
+     if (!mirror) {
+         return;
+     }
+     ofproto = mirror->ofproto;
+     ofproto->need_revalidate = true;
+     mac_learning_flush(ofproto->ml);
+     mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
+     HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
+         bundle->src_mirrors &= ~mirror_bit;
+         bundle->dst_mirrors &= ~mirror_bit;
+         bundle->mirror_out &= ~mirror_bit;
+     }
+     hmapx_destroy(&mirror->srcs);
+     hmapx_destroy(&mirror->dsts);
+     free(mirror->vlans);
+     ofproto->mirrors[mirror->idx] = NULL;
+     free(mirror->name);
+     free(mirror);
+ }
+ static int
+ set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
+         ofproto->need_revalidate = true;
+         mac_learning_flush(ofproto->ml);
+     }
+     return 0;
+ }
+ static bool
+ is_mirror_output_bundle(struct ofproto *ofproto_, void *aux)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     struct ofbundle *bundle = bundle_lookup(ofproto, aux);
+     return bundle && bundle->mirror_out != 0;
+ }
\f
+ /* Ports. */
+ static struct ofport_dpif *
+ get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port)
+ {
+     struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
+     return ofport ? ofport_dpif_cast(ofport) : NULL;
+ }
+ static struct ofport_dpif *
+ get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port)
+ {
+     return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port));
+ }
+ static void
+ ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port,
+                             struct dpif_port *dpif_port)
+ {
+     ofproto_port->name = dpif_port->name;
+     ofproto_port->type = dpif_port->type;
+     ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no);
+ }
+ static void
+ port_run(struct ofport_dpif *ofport)
+ {
+     if (ofport->cfm) {
+         cfm_run(ofport->cfm);
+         if (cfm_should_send_ccm(ofport->cfm)) {
+             struct ofpbuf packet;
+             struct ccm *ccm;
+             ofpbuf_init(&packet, 0);
+             ccm = eth_compose(&packet, eth_addr_ccm, ofport->up.opp.hw_addr,
+                               ETH_TYPE_CFM, sizeof *ccm);
+             cfm_compose_ccm(ofport->cfm, ccm);
+             send_packet(ofproto_dpif_cast(ofport->up.ofproto),
 -        send_packet(ofproto, OFPP_LOCAL, 0, upcall->packet);
++                        ofport->odp_port, &packet);
+             ofpbuf_uninit(&packet);
+         }
+     }
+ }
+ static void
+ port_wait(struct ofport_dpif *ofport)
+ {
+     if (ofport->cfm) {
+         cfm_wait(ofport->cfm);
+     }
+ }
+ static int
+ port_query_by_name(const struct ofproto *ofproto_, const char *devname,
+                    struct ofproto_port *ofproto_port)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     struct dpif_port dpif_port;
+     int error;
+     error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port);
+     if (!error) {
+         ofproto_port_from_dpif_port(ofproto_port, &dpif_port);
+     }
+     return error;
+ }
+ static int
+ port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     uint16_t odp_port;
+     int error;
+     error = dpif_port_add(ofproto->dpif, netdev, &odp_port);
+     if (!error) {
+         *ofp_portp = odp_port_to_ofp_port(odp_port);
+     }
+     return error;
+ }
+ static int
+ port_del(struct ofproto *ofproto_, uint16_t ofp_port)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     int error;
+     error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port));
+     if (!error) {
+         struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
+         if (ofport) {
+             /* The caller is going to close ofport->up.netdev.  If this is a
+              * bonded port, then the bond is using that netdev, so remove it
+              * from the bond.  The client will need to reconfigure everything
+              * after deleting ports, so then the slave will get re-added. */
+             bundle_remove(&ofport->up);
+         }
+     }
+     return error;
+ }
+ struct port_dump_state {
+     struct dpif_port_dump dump;
+     bool done;
+ };
+ static int
+ port_dump_start(const struct ofproto *ofproto_, void **statep)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     struct port_dump_state *state;
+     *statep = state = xmalloc(sizeof *state);
+     dpif_port_dump_start(&state->dump, ofproto->dpif);
+     state->done = false;
+     return 0;
+ }
+ static int
+ port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_,
+                struct ofproto_port *port)
+ {
+     struct port_dump_state *state = state_;
+     struct dpif_port dpif_port;
+     if (dpif_port_dump_next(&state->dump, &dpif_port)) {
+         ofproto_port_from_dpif_port(port, &dpif_port);
+         return 0;
+     } else {
+         int error = dpif_port_dump_done(&state->dump);
+         state->done = true;
+         return error ? error : EOF;
+     }
+ }
+ static int
+ port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
+ {
+     struct port_dump_state *state = state_;
+     if (!state->done) {
+         dpif_port_dump_done(&state->dump);
+     }
+     free(state);
+     return 0;
+ }
+ static int
+ port_poll(const struct ofproto *ofproto_, char **devnamep)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     return dpif_port_poll(ofproto->dpif, devnamep);
+ }
+ static void
+ port_poll_wait(const struct ofproto *ofproto_)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     dpif_port_poll_wait(ofproto->dpif);
+ }
+ static int
+ port_is_lacp_current(const struct ofport *ofport_)
+ {
+     const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
+     return (ofport->bundle && ofport->bundle->lacp
+             ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
+             : -1);
+ }
\f
+ /* Upcall handling. */
+ /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an
+  * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to
+  * their individual configurations.
+  *
+  * If 'clone' is true, the caller retains ownership of 'upcall->packet'.
+  * Otherwise, ownership is transferred to this function. */
+ static void
+ send_packet_in(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall,
+                const struct flow *flow, bool clone)
+ {
+     struct ofputil_packet_in pin;
+     pin.packet = upcall->packet;
+     pin.in_port = flow->in_port;
+     pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION;
+     pin.buffer_id = 0;          /* not yet known */
+     pin.send_len = upcall->userdata;
+     connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow,
+                            clone ? NULL : upcall->packet);
+ }
+ static bool
+ process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
+                 const struct ofpbuf *packet)
+ {
+     if (cfm_should_process_flow(flow)) {
+         struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port);
+         if (ofport && ofport->cfm) {
+             cfm_process_heartbeat(ofport->cfm, packet);
+         }
+         return true;
+     } else if (flow->dl_type == htons(ETH_TYPE_LACP)) {
+         struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port);
+         if (port && port->bundle && port->bundle->lacp) {
+             const struct lacp_pdu *pdu = parse_lacp_packet(packet);
+             if (pdu) {
+                 lacp_process_pdu(port->bundle->lacp, port, pdu);
+             }
+             return true;
+         }
+     }
+     return false;
+ }
+ static void
+ handle_miss_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
+ {
+     struct facet *facet;
+     struct flow flow;
+     /* Obtain in_port and tun_id, at least. */
+     odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
+     /* Set header pointers in 'flow'. */
+     flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow);
+     /* Handle 802.1ag and LACP. */
+     if (process_special(ofproto, &flow, upcall->packet)) {
+         ofpbuf_delete(upcall->packet);
+         ofproto->n_matches++;
+         return;
+     }
+     /* Check with in-band control to see if this packet should be sent
+      * to the local port regardless of the flow table. */
+     if (connmgr_msg_in_hook(ofproto->up.connmgr, &flow, upcall->packet)) {
 -/* Sends 'packet' out of port 'odp_port' within 'ofproto'.  If 'vlan_tci' is
 - * zero the packet will not have any 802.1Q hader; if it is nonzero, then the
 - * packet will be sent with the VLAN TCI specified by 'vlan_tci & ~VLAN_CFI'.
 - *
++        send_packet(ofproto, OFPP_LOCAL, upcall->packet);
+     }
+     facet = facet_lookup_valid(ofproto, &flow);
+     if (!facet) {
+         struct rule_dpif *rule = rule_dpif_lookup(ofproto, &flow);
+         if (!rule) {
+             /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
+             struct ofport_dpif *port = get_ofp_port(ofproto, flow.in_port);
+             if (port) {
+                 if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) {
+                     COVERAGE_INC(ofproto_dpif_no_packet_in);
+                     /* XXX install 'drop' flow entry */
+                     ofpbuf_delete(upcall->packet);
+                     return;
+                 }
+             } else {
+                 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16,
+                              flow.in_port);
+             }
+             send_packet_in(ofproto, upcall, &flow, false);
+             return;
+         }
+         facet = facet_create(rule, &flow, upcall->packet);
+     } else if (!facet->may_install) {
+         /* The facet is not installable, that is, we need to process every
+          * packet, so process the current packet's actions into 'facet'. */
+         facet_make_actions(ofproto, facet, upcall->packet);
+     }
+     if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
+         /*
+          * Extra-special case for fail-open mode.
+          *
+          * We are in fail-open mode and the packet matched the fail-open rule,
+          * but we are connected to a controller too.  We should send the packet
+          * up to the controller in the hope that it will try to set up a flow
+          * and thereby allow us to exit fail-open.
+          *
+          * See the top-level comment in fail-open.c for more information.
+          */
+         send_packet_in(ofproto, upcall, &flow, true);
+     }
+     facet_execute(ofproto, facet, upcall->packet);
+     facet_install(ofproto, facet, false);
+     ofproto->n_matches++;
+ }
+ static void
+ handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall)
+ {
+     struct flow flow;
+     switch (upcall->type) {
+     case DPIF_UC_ACTION:
+         COVERAGE_INC(ofproto_dpif_ctlr_action);
+         odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
+         send_packet_in(ofproto, upcall, &flow, false);
+         break;
+     case DPIF_UC_SAMPLE:
+         if (ofproto->sflow) {
+             odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
+             ofproto_sflow_received(ofproto->sflow, upcall, &flow);
+         }
+         ofpbuf_delete(upcall->packet);
+         break;
+     case DPIF_UC_MISS:
+         handle_miss_upcall(ofproto, upcall);
+         break;
+     case DPIF_N_UC_TYPES:
+     default:
+         VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
+         break;
+     }
+ }
\f
+ /* Flow expiration. */
+ static int facet_max_idle(const struct ofproto_dpif *);
+ static void update_stats(struct ofproto_dpif *);
+ static void rule_expire(struct rule_dpif *);
+ static void expire_facets(struct ofproto_dpif *, int dp_max_idle);
+ /* This function is called periodically by run().  Its job is to collect
+  * updates for the flows that have been installed into the datapath, most
+  * importantly when they last were used, and then use that information to
+  * expire flows that have not been used recently.
+  *
+  * Returns the number of milliseconds after which it should be called again. */
+ static int
+ expire(struct ofproto_dpif *ofproto)
+ {
+     struct rule_dpif *rule, *next_rule;
+     struct cls_cursor cursor;
+     int dp_max_idle;
+     /* Update stats for each flow in the datapath. */
+     update_stats(ofproto);
+     /* Expire facets that have been idle too long. */
+     dp_max_idle = facet_max_idle(ofproto);
+     expire_facets(ofproto, dp_max_idle);
+     /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
+     cls_cursor_init(&cursor, &ofproto->up.tables[0], NULL);
+     CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
+         rule_expire(rule);
+     }
+     /* All outstanding data in existing flows has been accounted, so it's a
+      * good time to do bond rebalancing. */
+     if (ofproto->has_bonded_bundles) {
+         struct ofbundle *bundle;
+         HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
+             if (bundle->bond) {
+                 bond_rebalance(bundle->bond, &ofproto->revalidate_set);
+             }
+         }
+     }
+     return MIN(dp_max_idle, 1000);
+ }
+ /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
+  *
+  * This function also pushes statistics updates to rules which each facet
+  * resubmits into.  Generally these statistics will be accurate.  However, if a
+  * facet changes the rule it resubmits into at some time in between
+  * update_stats() runs, it is possible that statistics accrued to the
+  * old rule will be incorrectly attributed to the new rule.  This could be
+  * avoided by calling update_stats() whenever rules are created or
+  * deleted.  However, the performance impact of making so many calls to the
+  * datapath do not justify the benefit of having perfectly accurate statistics.
+  */
+ static void
+ update_stats(struct ofproto_dpif *p)
+ {
+     const struct dpif_flow_stats *stats;
+     struct dpif_flow_dump dump;
+     const struct nlattr *key;
+     size_t key_len;
+     dpif_flow_dump_start(&dump, p->dpif);
+     while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
+         struct facet *facet;
+         struct flow flow;
+         if (odp_flow_key_to_flow(key, key_len, &flow)) {
+             struct ds s;
+             ds_init(&s);
+             odp_flow_key_format(key, key_len, &s);
+             VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s",
+                          ds_cstr(&s));
+             ds_destroy(&s);
+             continue;
+         }
+         facet = facet_find(p, &flow);
+         if (facet && facet->installed) {
+             if (stats->n_packets >= facet->dp_packet_count) {
+                 uint64_t extra = stats->n_packets - facet->dp_packet_count;
+                 facet->packet_count += extra;
+             } else {
+                 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
+             }
+             if (stats->n_bytes >= facet->dp_byte_count) {
+                 facet->byte_count += stats->n_bytes - facet->dp_byte_count;
+             } else {
+                 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
+             }
+             facet->dp_packet_count = stats->n_packets;
+             facet->dp_byte_count = stats->n_bytes;
+             facet_update_time(p, facet, stats->used);
+             facet_account(p, facet, stats->n_bytes);
+             facet_push_stats(facet);
+         } else {
+             /* There's a flow in the datapath that we know nothing about.
+              * Delete it. */
+             COVERAGE_INC(facet_unexpected);
+             dpif_flow_del(p->dpif, key, key_len, NULL);
+         }
+     }
+     dpif_flow_dump_done(&dump);
+ }
+ /* Calculates and returns the number of milliseconds of idle time after which
+  * facets should expire from the datapath and we should fold their statistics
+  * into their parent rules in userspace. */
+ static int
+ facet_max_idle(const struct ofproto_dpif *ofproto)
+ {
+     /*
+      * Idle time histogram.
+      *
+      * Most of the time a switch has a relatively small number of facets.  When
+      * this is the case we might as well keep statistics for all of them in
+      * userspace and to cache them in the kernel datapath for performance as
+      * well.
+      *
+      * As the number of facets increases, the memory required to maintain
+      * statistics about them in userspace and in the kernel becomes
+      * significant.  However, with a large number of facets it is likely that
+      * only a few of them are "heavy hitters" that consume a large amount of
+      * bandwidth.  At this point, only heavy hitters are worth caching in the
+      * kernel and maintaining in userspaces; other facets we can discard.
+      *
+      * The technique used to compute the idle time is to build a histogram with
+      * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each.  Each facet
+      * that is installed in the kernel gets dropped in the appropriate bucket.
+      * After the histogram has been built, we compute the cutoff so that only
+      * the most-recently-used 1% of facets (but at least 1000 flows) are kept
+      * cached.  At least the most-recently-used bucket of facets is kept, so
+      * actually an arbitrary number of facets can be kept in any given
+      * expiration run (though the next run will delete most of those unless
+      * they receive additional data).
+      *
+      * This requires a second pass through the facets, in addition to the pass
+      * made by update_stats(), because the former function never looks
+      * at uninstallable facets.
+      */
+     enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
+     enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
+     int buckets[N_BUCKETS] = { 0 };
+     struct facet *facet;
+     int total, bucket;
+     long long int now;
+     int i;
+     total = hmap_count(&ofproto->facets);
+     if (total <= 1000) {
+         return N_BUCKETS * BUCKET_WIDTH;
+     }
+     /* Build histogram. */
+     now = time_msec();
+     HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
+         long long int idle = now - facet->used;
+         int bucket = (idle <= 0 ? 0
+                       : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
+                       : (unsigned int) idle / BUCKET_WIDTH);
+         buckets[bucket]++;
+     }
+     /* Find the first bucket whose flows should be expired. */
+     for (bucket = 0; bucket < N_BUCKETS; bucket++) {
+         if (buckets[bucket]) {
+             int subtotal = 0;
+             do {
+                 subtotal += buckets[bucket++];
+             } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100));
+             break;
+         }
+     }
+     if (VLOG_IS_DBG_ENABLED()) {
+         struct ds s;
+         ds_init(&s);
+         ds_put_cstr(&s, "keep");
+         for (i = 0; i < N_BUCKETS; i++) {
+             if (i == bucket) {
+                 ds_put_cstr(&s, ", drop");
+             }
+             if (buckets[i]) {
+                 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
+             }
+         }
+         VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
+         ds_destroy(&s);
+     }
+     return bucket * BUCKET_WIDTH;
+ }
+ static void
+ facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
+ {
+     if (ofproto->netflow && !facet_is_controller_flow(facet) &&
+         netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
+         struct ofexpired expired;
+         if (facet->installed) {
+             struct dpif_flow_stats stats;
+             facet_put__(ofproto, facet, facet->actions, facet->actions_len,
+                         &stats);
+             facet_update_stats(ofproto, facet, &stats);
+         }
+         expired.flow = facet->flow;
+         expired.packet_count = facet->packet_count;
+         expired.byte_count = facet->byte_count;
+         expired.used = facet->used;
+         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
+     }
+ }
+ static void
+ expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle)
+ {
+     long long int cutoff = time_msec() - dp_max_idle;
+     struct facet *facet, *next_facet;
+     HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) {
+         facet_active_timeout(ofproto, facet);
+         if (facet->used < cutoff) {
+             facet_remove(ofproto, facet);
+         }
+     }
+ }
+ /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
+  * then delete it entirely. */
+ static void
+ rule_expire(struct rule_dpif *rule)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
+     struct facet *facet, *next_facet;
+     long long int now;
+     uint8_t reason;
+     /* Has 'rule' expired? */
+     now = time_msec();
+     if (rule->up.hard_timeout
+         && now > rule->up.created + rule->up.hard_timeout * 1000) {
+         reason = OFPRR_HARD_TIMEOUT;
+     } else if (rule->up.idle_timeout && list_is_empty(&rule->facets)
+                && now > rule->used + rule->up.idle_timeout * 1000) {
+         reason = OFPRR_IDLE_TIMEOUT;
+     } else {
+         return;
+     }
+     COVERAGE_INC(ofproto_dpif_expired);
+     /* Update stats.  (This is a no-op if the rule expired due to an idle
+      * timeout, because that only happens when the rule has no facets left.) */
+     LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
+         facet_remove(ofproto, facet);
+     }
+     /* Get rid of the rule. */
+     ofproto_rule_expire(&rule->up, reason);
+ }
\f
+ /* Facets. */
+ /* Creates and returns a new facet owned by 'rule', given a 'flow' and an
+  * example 'packet' within that flow.
+  *
+  * The caller must already have determined that no facet with an identical
+  * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
+  * the ofproto's classifier table. */
+ static struct facet *
+ facet_create(struct rule_dpif *rule, const struct flow *flow,
+              const struct ofpbuf *packet)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
+     struct facet *facet;
+     facet = xzalloc(sizeof *facet);
+     facet->used = time_msec();
+     hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0));
+     list_push_back(&rule->facets, &facet->list_node);
+     facet->rule = rule;
+     facet->flow = *flow;
+     netflow_flow_init(&facet->nf_flow);
+     netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
+     facet_make_actions(ofproto, facet, packet);
+     return facet;
+ }
+ static void
+ facet_free(struct facet *facet)
+ {
+     free(facet->actions);
+     free(facet);
+ }
+ /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
+  * 'packet', which arrived on 'in_port'.
+  *
+  * Takes ownership of 'packet'. */
+ static bool
+ execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
+                     const struct nlattr *odp_actions, size_t actions_len,
+                     struct ofpbuf *packet)
+ {
+     if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t))
+         && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) {
+         /* As an optimization, avoid a round-trip from userspace to kernel to
+          * userspace.  This also avoids possibly filling up kernel packet
+          * buffers along the way. */
+         struct dpif_upcall upcall;
+         upcall.type = DPIF_UC_ACTION;
+         upcall.packet = packet;
+         upcall.key = NULL;
+         upcall.key_len = 0;
+         upcall.userdata = nl_attr_get_u64(odp_actions);
+         upcall.sample_pool = 0;
+         upcall.actions = NULL;
+         upcall.actions_len = 0;
+         send_packet_in(ofproto, &upcall, flow, false);
+         return true;
+     } else {
+         int error;
+         error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet);
+         ofpbuf_delete(packet);
+         return !error;
+     }
+ }
+ /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
+  * statistics appropriately.  'packet' must have at least sizeof(struct
+  * ofp_packet_in) bytes of headroom.
+  *
+  * For correct results, 'packet' must actually be in 'facet''s flow; that is,
+  * applying flow_extract() to 'packet' would yield the same flow as
+  * 'facet->flow'.
+  *
+  * 'facet' must have accurately composed ODP actions; that is, it must not be
+  * in need of revalidation.
+  *
+  * Takes ownership of 'packet'. */
+ static void
+ facet_execute(struct ofproto_dpif *ofproto, struct facet *facet,
+               struct ofpbuf *packet)
+ {
+     struct dpif_flow_stats stats;
+     assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
+     flow_extract_stats(&facet->flow, packet, &stats);
+     stats.used = time_msec();
+     if (execute_odp_actions(ofproto, &facet->flow,
+                             facet->actions, facet->actions_len, packet)) {
+         facet_update_stats(ofproto, facet, &stats);
+     }
+ }
+ /* Remove 'facet' from 'ofproto' and free up the associated memory:
+  *
+  *   - If 'facet' was installed in the datapath, uninstalls it and updates its
+  *     rule's statistics, via facet_uninstall().
+  *
+  *   - Removes 'facet' from its rule and from ofproto->facets.
+  */
+ static void
+ facet_remove(struct ofproto_dpif *ofproto, struct facet *facet)
+ {
+     facet_uninstall(ofproto, facet);
+     facet_flush_stats(ofproto, facet);
+     hmap_remove(&ofproto->facets, &facet->hmap_node);
+     list_remove(&facet->list_node);
+     facet_free(facet);
+ }
+ /* Composes the ODP actions for 'facet' based on its rule's actions. */
+ static void
+ facet_make_actions(struct ofproto_dpif *p, struct facet *facet,
+                    const struct ofpbuf *packet)
+ {
+     const struct rule_dpif *rule = facet->rule;
+     struct ofpbuf *odp_actions;
+     struct action_xlate_ctx ctx;
+     action_xlate_ctx_init(&ctx, p, &facet->flow, packet);
+     odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
+     facet->tags = ctx.tags;
+     facet->may_install = ctx.may_set_up_flow;
+     facet->nf_flow.output_iface = ctx.nf_output_iface;
+     if (facet->actions_len != odp_actions->size
+         || memcmp(facet->actions, odp_actions->data, odp_actions->size)) {
+         free(facet->actions);
+         facet->actions_len = odp_actions->size;
+         facet->actions = xmemdup(odp_actions->data, odp_actions->size);
+     }
+     ofpbuf_delete(odp_actions);
+ }
+ static int
+ facet_put__(struct ofproto_dpif *ofproto, struct facet *facet,
+             const struct nlattr *actions, size_t actions_len,
+             struct dpif_flow_stats *stats)
+ {
+     struct odputil_keybuf keybuf;
+     enum dpif_flow_put_flags flags;
+     struct ofpbuf key;
+     flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
+     if (stats) {
+         flags |= DPIF_FP_ZERO_STATS;
+         facet->dp_packet_count = 0;
+         facet->dp_byte_count = 0;
+     }
+     ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
+     odp_flow_key_from_flow(&key, &facet->flow);
+     return dpif_flow_put(ofproto->dpif, flags, key.data, key.size,
+                          actions, actions_len, stats);
+ }
+ /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath.  If
+  * 'zero_stats' is true, clears any existing statistics from the datapath for
+  * 'facet'. */
+ static void
+ facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats)
+ {
+     struct dpif_flow_stats stats;
+     if (facet->may_install
+         && !facet_put__(p, facet, facet->actions, facet->actions_len,
+                         zero_stats ? &stats : NULL)) {
+         facet->installed = true;
+     }
+ }
+ static void
+ facet_account(struct ofproto_dpif *ofproto,
+               struct facet *facet, uint64_t extra_bytes)
+ {
+     uint64_t total_bytes, n_bytes;
+     struct ofbundle *in_bundle;
+     const struct nlattr *a;
+     tag_type dummy = 0;
+     unsigned int left;
+     int vlan;
+     total_bytes = facet->byte_count + extra_bytes;
+     if (total_bytes <= facet->accounted_bytes) {
+         return;
+     }
+     n_bytes = total_bytes - facet->accounted_bytes;
+     facet->accounted_bytes = total_bytes;
+     /* Test that 'tags' is nonzero to ensure that only flows that include an
+      * OFPP_NORMAL action are used for learning and bond slave rebalancing.
+      * This works because OFPP_NORMAL always sets a nonzero tag value.
+      *
+      * Feed information from the active flows back into the learning table to
+      * ensure that table is always in sync with what is actually flowing
+      * through the datapath. */
+     if (!facet->tags
+         || !is_admissible(ofproto, &facet->flow, false, &dummy,
+                           &vlan, &in_bundle)) {
+         return;
+     }
+     update_learning_table(ofproto, &facet->flow, vlan, in_bundle);
+     if (!ofproto->has_bonded_bundles) {
+         return;
+     }
+     NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) {
+         if (nl_attr_type(a) == ODP_ACTION_ATTR_OUTPUT) {
+             struct ofport_dpif *port;
+             port = get_odp_port(ofproto, nl_attr_get_u32(a));
+             if (port && port->bundle && port->bundle->bond) {
+                 bond_account(port->bundle->bond, &facet->flow, vlan, n_bytes);
+             }
+         }
+     }
+ }
+ /* If 'rule' is installed in the datapath, uninstalls it. */
+ static void
+ facet_uninstall(struct ofproto_dpif *p, struct facet *facet)
+ {
+     if (facet->installed) {
+         struct odputil_keybuf keybuf;
+         struct dpif_flow_stats stats;
+         struct ofpbuf key;
+         ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
+         odp_flow_key_from_flow(&key, &facet->flow);
+         if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) {
+             facet_update_stats(p, facet, &stats);
+         }
+         facet->installed = false;
+         facet->dp_packet_count = 0;
+         facet->dp_byte_count = 0;
+     } else {
+         assert(facet->dp_packet_count == 0);
+         assert(facet->dp_byte_count == 0);
+     }
+ }
+ /* Returns true if the only action for 'facet' is to send to the controller.
+  * (We don't report NetFlow expiration messages for such facets because they
+  * are just part of the control logic for the network, not real traffic). */
+ static bool
+ facet_is_controller_flow(struct facet *facet)
+ {
+     return (facet
+             && facet->rule->up.n_actions == 1
+             && action_outputs_to_port(&facet->rule->up.actions[0],
+                                       htons(OFPP_CONTROLLER)));
+ }
+ /* Folds all of 'facet''s statistics into its rule.  Also updates the
+  * accounting ofhook and emits a NetFlow expiration if appropriate.  All of
+  * 'facet''s statistics in the datapath should have been zeroed and folded into
+  * its packet and byte counts before this function is called. */
+ static void
+ facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet)
+ {
+     assert(!facet->dp_byte_count);
+     assert(!facet->dp_packet_count);
+     facet_push_stats(facet);
+     facet_account(ofproto, facet, 0);
+     if (ofproto->netflow && !facet_is_controller_flow(facet)) {
+         struct ofexpired expired;
+         expired.flow = facet->flow;
+         expired.packet_count = facet->packet_count;
+         expired.byte_count = facet->byte_count;
+         expired.used = facet->used;
+         netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
+     }
+     facet->rule->packet_count += facet->packet_count;
+     facet->rule->byte_count += facet->byte_count;
+     /* Reset counters to prevent double counting if 'facet' ever gets
+      * reinstalled. */
+     facet->packet_count = 0;
+     facet->byte_count = 0;
+     facet->rs_packet_count = 0;
+     facet->rs_byte_count = 0;
+     facet->accounted_bytes = 0;
+     netflow_flow_clear(&facet->nf_flow);
+ }
+ /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
+  * Returns it if found, otherwise a null pointer.
+  *
+  * The returned facet might need revalidation; use facet_lookup_valid()
+  * instead if that is important. */
+ static struct facet *
+ facet_find(struct ofproto_dpif *ofproto, const struct flow *flow)
+ {
+     struct facet *facet;
+     HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0),
+                              &ofproto->facets) {
+         if (flow_equal(flow, &facet->flow)) {
+             return facet;
+         }
+     }
+     return NULL;
+ }
+ /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
+  * Returns it if found, otherwise a null pointer.
+  *
+  * The returned facet is guaranteed to be valid. */
+ static struct facet *
+ facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
+ {
+     struct facet *facet = facet_find(ofproto, flow);
+     /* The facet we found might not be valid, since we could be in need of
+      * revalidation.  If it is not valid, don't return it. */
+     if (facet
+         && ofproto->need_revalidate
+         && !facet_revalidate(ofproto, facet)) {
+         COVERAGE_INC(facet_invalidated);
+         return NULL;
+     }
+     return facet;
+ }
+ /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
+  *
+  *   - If the rule found is different from 'facet''s current rule, moves
+  *     'facet' to the new rule and recompiles its actions.
+  *
+  *   - If the rule found is the same as 'facet''s current rule, leaves 'facet'
+  *     where it is and recompiles its actions anyway.
+  *
+  *   - If there is none, destroys 'facet'.
+  *
+  * Returns true if 'facet' still exists, false if it has been destroyed. */
+ static bool
+ facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet)
+ {
+     struct action_xlate_ctx ctx;
+     struct ofpbuf *odp_actions;
+     struct rule_dpif *new_rule;
+     bool actions_changed;
+     COVERAGE_INC(facet_revalidate);
+     /* Determine the new rule. */
+     new_rule = rule_dpif_lookup(ofproto, &facet->flow);
+     if (!new_rule) {
+         /* No new rule, so delete the facet. */
+         facet_remove(ofproto, facet);
+         return false;
+     }
+     /* Calculate new ODP actions.
+      *
+      * We do not modify any 'facet' state yet, because we might need to, e.g.,
+      * emit a NetFlow expiration and, if so, we need to have the old state
+      * around to properly compose it. */
+     action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL);
+     odp_actions = xlate_actions(&ctx,
+                                 new_rule->up.actions, new_rule->up.n_actions);
+     actions_changed = (facet->actions_len != odp_actions->size
+                        || memcmp(facet->actions, odp_actions->data,
+                                  facet->actions_len));
+     /* If the ODP actions changed or the installability changed, then we need
+      * to talk to the datapath. */
+     if (actions_changed || ctx.may_set_up_flow != facet->installed) {
+         if (ctx.may_set_up_flow) {
+             struct dpif_flow_stats stats;
+             facet_put__(ofproto, facet,
+                         odp_actions->data, odp_actions->size, &stats);
+             facet_update_stats(ofproto, facet, &stats);
+         } else {
+             facet_uninstall(ofproto, facet);
+         }
+         /* The datapath flow is gone or has zeroed stats, so push stats out of
+          * 'facet' into 'rule'. */
+         facet_flush_stats(ofproto, facet);
+     }
+     /* Update 'facet' now that we've taken care of all the old state. */
+     facet->tags = ctx.tags;
+     facet->nf_flow.output_iface = ctx.nf_output_iface;
+     facet->may_install = ctx.may_set_up_flow;
+     if (actions_changed) {
+         free(facet->actions);
+         facet->actions_len = odp_actions->size;
+         facet->actions = xmemdup(odp_actions->data, odp_actions->size);
+     }
+     if (facet->rule != new_rule) {
+         COVERAGE_INC(facet_changed_rule);
+         list_remove(&facet->list_node);
+         list_push_back(&new_rule->facets, &facet->list_node);
+         facet->rule = new_rule;
+         facet->used = new_rule->up.created;
+         facet->rs_used = facet->used;
+     }
+     ofpbuf_delete(odp_actions);
+     return true;
+ }
+ /* Updates 'facet''s used time.  Caller is responsible for calling
+  * facet_push_stats() to update the flows which 'facet' resubmits into. */
+ static void
+ facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet,
+                   long long int used)
+ {
+     if (used > facet->used) {
+         facet->used = used;
+         if (used > facet->rule->used) {
+             facet->rule->used = used;
+         }
+         netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
+     }
+ }
+ /* Folds the statistics from 'stats' into the counters in 'facet'.
+  *
+  * Because of the meaning of a facet's counters, it only makes sense to do this
+  * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
+  * packet that was sent by hand or if it represents statistics that have been
+  * cleared out of the datapath. */
+ static void
+ facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet,
+                    const struct dpif_flow_stats *stats)
+ {
+     if (stats->n_packets || stats->used > facet->used) {
+         facet_update_time(ofproto, facet, stats->used);
+         facet->packet_count += stats->n_packets;
+         facet->byte_count += stats->n_bytes;
+         facet_push_stats(facet);
+         netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
+     }
+ }
+ static void
+ facet_push_stats(struct facet *facet)
+ {
+     uint64_t rs_packets, rs_bytes;
+     assert(facet->packet_count >= facet->rs_packet_count);
+     assert(facet->byte_count >= facet->rs_byte_count);
+     assert(facet->used >= facet->rs_used);
+     rs_packets = facet->packet_count - facet->rs_packet_count;
+     rs_bytes = facet->byte_count - facet->rs_byte_count;
+     if (rs_packets || rs_bytes || facet->used > facet->rs_used) {
+         facet->rs_packet_count = facet->packet_count;
+         facet->rs_byte_count = facet->byte_count;
+         facet->rs_used = facet->used;
+         flow_push_stats(facet->rule, &facet->flow,
+                         rs_packets, rs_bytes, facet->used);
+     }
+ }
+ struct ofproto_push {
+     struct action_xlate_ctx ctx;
+     uint64_t packets;
+     uint64_t bytes;
+     long long int used;
+ };
+ static void
+ push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
+ {
+     struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx);
+     if (rule) {
+         rule->packet_count += push->packets;
+         rule->byte_count += push->bytes;
+         rule->used = MAX(push->used, rule->used);
+     }
+ }
+ /* Pushes flow statistics to the rules which 'flow' resubmits into given
+  * 'rule''s actions. */
+ static void
+ flow_push_stats(const struct rule_dpif *rule,
+                 struct flow *flow, uint64_t packets, uint64_t bytes,
+                 long long int used)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
+     struct ofproto_push push;
+     push.packets = packets;
+     push.bytes = bytes;
+     push.used = used;
+     action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL);
+     push.ctx.resubmit_hook = push_resubmit;
+     ofpbuf_delete(xlate_actions(&push.ctx,
+                                 rule->up.actions, rule->up.n_actions));
+ }
\f
+ /* Rules. */
+ static struct rule_dpif *
+ rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
+ {
+     return rule_dpif_cast(rule_from_cls_rule(
+                               classifier_lookup(&ofproto->up.tables[0],
+                                                 flow)));
+ }
+ static struct rule *
+ rule_alloc(void)
+ {
+     struct rule_dpif *rule = xmalloc(sizeof *rule);
+     return &rule->up;
+ }
+ static void
+ rule_dealloc(struct rule *rule_)
+ {
+     struct rule_dpif *rule = rule_dpif_cast(rule_);
+     free(rule);
+ }
+ static int
+ rule_construct(struct rule *rule_)
+ {
+     struct rule_dpif *rule = rule_dpif_cast(rule_);
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
+     struct rule_dpif *old_rule;
+     int error;
+     error = validate_actions(rule->up.actions, rule->up.n_actions,
+                              &rule->up.cr.flow, ofproto->max_ports);
+     if (error) {
+         return error;
+     }
+     old_rule = rule_dpif_cast(rule_from_cls_rule(classifier_find_rule_exactly(
+                                                      &ofproto->up.tables[0],
+                                                      &rule->up.cr)));
+     if (old_rule) {
+         ofproto_rule_destroy(&old_rule->up);
+     }
+     rule->used = rule->up.created;
+     rule->packet_count = 0;
+     rule->byte_count = 0;
+     list_init(&rule->facets);
+     classifier_insert(&ofproto->up.tables[0], &rule->up.cr);
+     ofproto->need_revalidate = true;
+     return 0;
+ }
+ static void
+ rule_destruct(struct rule *rule_)
+ {
+     struct rule_dpif *rule = rule_dpif_cast(rule_);
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
+     struct facet *facet, *next_facet;
+     classifier_remove(&ofproto->up.tables[0], &rule->up.cr);
+     LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
+         facet_revalidate(ofproto, facet);
+     }
+     ofproto->need_revalidate = true;
+ }
+ static void
+ rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
+ {
+     struct rule_dpif *rule = rule_dpif_cast(rule_);
+     struct facet *facet;
+     /* Start from historical data for 'rule' itself that are no longer tracked
+      * in facets.  This counts, for example, facets that have expired. */
+     *packets = rule->packet_count;
+     *bytes = rule->byte_count;
+     /* Add any statistics that are tracked by facets.  This includes
+      * statistical data recently updated by ofproto_update_stats() as well as
+      * stats for packets that were executed "by hand" via dpif_execute(). */
+     LIST_FOR_EACH (facet, list_node, &rule->facets) {
+         *packets += facet->packet_count;
+         *bytes += facet->byte_count;
+     }
+ }
+ static int
+ rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet)
+ {
+     struct rule_dpif *rule = rule_dpif_cast(rule_);
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
+     struct action_xlate_ctx ctx;
+     struct ofpbuf *odp_actions;
+     struct facet *facet;
+     size_t size;
+     /* First look for a related facet.  If we find one, account it to that. */
+     facet = facet_lookup_valid(ofproto, flow);
+     if (facet && facet->rule == rule) {
+         facet_execute(ofproto, facet, packet);
+         return 0;
+     }
+     /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
+      * create a new facet for it and use that. */
+     if (rule_dpif_lookup(ofproto, flow) == rule) {
+         facet = facet_create(rule, flow, packet);
+         facet_execute(ofproto, facet, packet);
+         facet_install(ofproto, facet, true);
+         return 0;
+     }
+     /* We can't account anything to a facet.  If we were to try, then that
+      * facet would have a non-matching rule, busting our invariants. */
+     action_xlate_ctx_init(&ctx, ofproto, flow, packet);
+     odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions);
+     size = packet->size;
+     if (execute_odp_actions(ofproto, flow, odp_actions->data,
+                             odp_actions->size, packet)) {
+         rule->used = time_msec();
+         rule->packet_count++;
+         rule->byte_count += size;
+         flow_push_stats(rule, flow, 1, size, rule->used);
+     }
+     ofpbuf_delete(odp_actions);
+     return 0;
+ }
+ static int
+ rule_modify_actions(struct rule *rule_,
+                     const union ofp_action *actions, size_t n_actions)
+ {
+     struct rule_dpif *rule = rule_dpif_cast(rule_);
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
+     int error;
+     error = validate_actions(actions, n_actions, &rule->up.cr.flow,
+                              ofproto->max_ports);
+     if (!error) {
+         ofproto->need_revalidate = true;
+     }
+     return error;
+ }
\f
 -send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port, uint16_t vlan_tci,
++/* Sends 'packet' out of port 'odp_port' within 'p'.
+  * Returns 0 if successful, otherwise a positive errno value. */
+ static int
 -    if (vlan_tci != 0) {
 -        nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
 -                       ntohs(vlan_tci & ~VLAN_CFI));
 -    }
++send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port,
+             const struct ofpbuf *packet)
+ {
+     struct ofpbuf odp_actions;
+     int error;
+     ofpbuf_init(&odp_actions, 32);
+     nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
+     error = dpif_execute(ofproto->dpif, odp_actions.data, odp_actions.size,
+                          packet);
+     ofpbuf_uninit(&odp_actions);
+     if (error) {
+         VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
+                      ofproto->up.name, odp_port, strerror(error));
+     }
+     return error;
+ }
\f
+ /* OpenFlow to ODP action translation. */
+ static void do_xlate_actions(const union ofp_action *in, size_t n_in,
+                              struct action_xlate_ctx *ctx);
+ static bool xlate_normal(struct action_xlate_ctx *);
+ static void
+ add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
+ {
+     const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
+     uint16_t odp_port = ofp_port_to_odp_port(ofp_port);
+     if (ofport) {
+         if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)) {
+             /* Forwarding disabled on port. */
+             return;
+         }
+     } else {
+         /*
+          * We don't have an ofport record for this port, but it doesn't hurt to
+          * allow forwarding to it anyhow.  Maybe such a port will appear later
+          * and we're pre-populating the flow table.
+          */
+     }
+     nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port);
+     ctx->nf_output_iface = ofp_port;
+ }
+ static void
+ xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port)
+ {
+     if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
+         struct rule_dpif *rule;
+         uint16_t old_in_port;
+         /* Look up a flow with 'in_port' as the input port.  Then restore the
+          * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
+          * have surprising behavior). */
+         old_in_port = ctx->flow.in_port;
+         ctx->flow.in_port = in_port;
+         rule = rule_dpif_lookup(ctx->ofproto, &ctx->flow);
+         ctx->flow.in_port = old_in_port;
+         if (ctx->resubmit_hook) {
+             ctx->resubmit_hook(ctx, rule);
+         }
+         if (rule) {
+             ctx->recurse++;
+             do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx);
+             ctx->recurse--;
+         }
+     } else {
+         static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
+         VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times",
+                     MAX_RESUBMIT_RECURSION);
+     }
+ }
+ static void
+ flood_packets(struct ofproto_dpif *ofproto,
+               uint16_t ofp_in_port, ovs_be32 mask,
+               uint16_t *nf_output_iface, struct ofpbuf *odp_actions)
+ {
+     struct ofport_dpif *ofport;
+     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
+         uint16_t ofp_port = ofport->up.ofp_port;
+         if (ofp_port != ofp_in_port && !(ofport->up.opp.config & mask)) {
+             nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT,
+                            ofport->odp_port);
+         }
+     }
+     *nf_output_iface = NF_OUT_FLOOD;
+ }
+ static void
+ xlate_output_action__(struct action_xlate_ctx *ctx,
+                       uint16_t port, uint16_t max_len)
+ {
+     uint16_t prev_nf_output_iface = ctx->nf_output_iface;
+     ctx->nf_output_iface = NF_OUT_DROP;
+     switch (port) {
+     case OFPP_IN_PORT:
+         add_output_action(ctx, ctx->flow.in_port);
+         break;
+     case OFPP_TABLE:
+         xlate_table_action(ctx, ctx->flow.in_port);
+         break;
+     case OFPP_NORMAL:
+         xlate_normal(ctx);
+         break;
+     case OFPP_FLOOD:
+         flood_packets(ctx->ofproto, ctx->flow.in_port, htonl(OFPPC_NO_FLOOD),
+                       &ctx->nf_output_iface, ctx->odp_actions);
+         break;
+     case OFPP_ALL:
+         flood_packets(ctx->ofproto, ctx->flow.in_port, htonl(0),
+                       &ctx->nf_output_iface, ctx->odp_actions);
+         break;
+     case OFPP_CONTROLLER:
+         nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len);
+         break;
+     case OFPP_LOCAL:
+         add_output_action(ctx, OFPP_LOCAL);
+         break;
+     default:
+         if (port != ctx->flow.in_port) {
+             add_output_action(ctx, port);
+         }
+         break;
+     }
+     if (prev_nf_output_iface == NF_OUT_FLOOD) {
+         ctx->nf_output_iface = NF_OUT_FLOOD;
+     } else if (ctx->nf_output_iface == NF_OUT_DROP) {
+         ctx->nf_output_iface = prev_nf_output_iface;
+     } else if (prev_nf_output_iface != NF_OUT_DROP &&
+                ctx->nf_output_iface != NF_OUT_FLOOD) {
+         ctx->nf_output_iface = NF_OUT_MULTI;
+     }
+ }
+ static void
+ xlate_output_action(struct action_xlate_ctx *ctx,
+                     const struct ofp_action_output *oao)
+ {
+     xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len));
+ }
+ /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
+  * optimization, because we're going to add another action that sets the
+  * priority immediately after, or because there are no actions following the
+  * pop.  */
+ static void
+ remove_pop_action(struct action_xlate_ctx *ctx)
+ {
+     if (ctx->odp_actions->size == ctx->last_pop_priority) {
+         ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN);
+         ctx->last_pop_priority = -1;
+     }
+ }
+ static void
+ add_pop_action(struct action_xlate_ctx *ctx)
+ {
+     if (ctx->odp_actions->size != ctx->last_pop_priority) {
+         nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY);
+         ctx->last_pop_priority = ctx->odp_actions->size;
+     }
+ }
+ static void
+ xlate_enqueue_action(struct action_xlate_ctx *ctx,
+                      const struct ofp_action_enqueue *oae)
+ {
+     uint16_t ofp_port, odp_port;
+     uint32_t priority;
+     int error;
+     error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id),
+                                    &priority);
+     if (error) {
+         /* Fall back to ordinary output action. */
+         xlate_output_action__(ctx, ntohs(oae->port), 0);
+         return;
+     }
+     /* Figure out ODP output port. */
+     ofp_port = ntohs(oae->port);
+     if (ofp_port == OFPP_IN_PORT) {
+         ofp_port = ctx->flow.in_port;
+     }
+     odp_port = ofp_port_to_odp_port(ofp_port);
+     /* Add ODP actions. */
+     remove_pop_action(ctx);
+     nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
+     add_output_action(ctx, odp_port);
+     add_pop_action(ctx);
+     /* Update NetFlow output port. */
+     if (ctx->nf_output_iface == NF_OUT_DROP) {
+         ctx->nf_output_iface = odp_port;
+     } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
+         ctx->nf_output_iface = NF_OUT_MULTI;
+     }
+ }
+ static void
+ xlate_set_queue_action(struct action_xlate_ctx *ctx,
+                        const struct nx_action_set_queue *nasq)
+ {
+     uint32_t priority;
+     int error;
+     error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id),
+                                    &priority);
+     if (error) {
+         /* Couldn't translate queue to a priority, so ignore.  A warning
+          * has already been logged. */
+         return;
+     }
+     remove_pop_action(ctx);
+     nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority);
+ }
+ static void
+ xlate_set_dl_tci(struct action_xlate_ctx *ctx)
+ {
+     ovs_be16 tci = ctx->flow.vlan_tci;
+     if (!(tci & htons(VLAN_CFI))) {
+         nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
+     } else {
+         nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI,
+                         tci & ~htons(VLAN_CFI));
+     }
+ }
+ struct xlate_reg_state {
+     ovs_be16 vlan_tci;
+     ovs_be64 tun_id;
+ };
+ static void
+ save_reg_state(const struct action_xlate_ctx *ctx,
+                struct xlate_reg_state *state)
+ {
+     state->vlan_tci = ctx->flow.vlan_tci;
+     state->tun_id = ctx->flow.tun_id;
+ }
+ static void
+ update_reg_state(struct action_xlate_ctx *ctx,
+                  const struct xlate_reg_state *state)
+ {
+     if (ctx->flow.vlan_tci != state->vlan_tci) {
+         xlate_set_dl_tci(ctx);
+     }
+     if (ctx->flow.tun_id != state->tun_id) {
+         nl_msg_put_be64(ctx->odp_actions,
+                         ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id);
+     }
+ }
+ static void
+ xlate_autopath(struct action_xlate_ctx *ctx,
+                const struct nx_action_autopath *naa)
+ {
+     uint16_t ofp_port = ntohl(naa->id);
+     struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
+     if (!port || !port->bundle) {
+         ofp_port = OFPP_NONE;
+     } else if (port->bundle->bond) {
+         /* Autopath does not support VLAN hashing. */
+         struct ofport_dpif *slave = bond_choose_output_slave(
+             port->bundle->bond, &ctx->flow, OFP_VLAN_NONE, &ctx->tags);
+         if (slave) {
+             ofp_port = slave->up.ofp_port;
+         }
+     }
+     autopath_execute(naa, &ctx->flow, ofp_port);
+ }
+ static void
+ xlate_nicira_action(struct action_xlate_ctx *ctx,
+                     const struct nx_action_header *nah)
+ {
+     const struct nx_action_resubmit *nar;
+     const struct nx_action_set_tunnel *nast;
+     const struct nx_action_set_queue *nasq;
+     const struct nx_action_multipath *nam;
+     const struct nx_action_autopath *naa;
+     enum nx_action_subtype subtype = ntohs(nah->subtype);
+     struct xlate_reg_state state;
+     ovs_be64 tun_id;
+     assert(nah->vendor == htonl(NX_VENDOR_ID));
+     switch (subtype) {
+     case NXAST_RESUBMIT:
+         nar = (const struct nx_action_resubmit *) nah;
+         xlate_table_action(ctx, ntohs(nar->in_port));
+         break;
+     case NXAST_SET_TUNNEL:
+         nast = (const struct nx_action_set_tunnel *) nah;
+         tun_id = htonll(ntohl(nast->tun_id));
+         nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
+         ctx->flow.tun_id = tun_id;
+         break;
+     case NXAST_DROP_SPOOFED_ARP:
+         if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) {
+             nl_msg_put_flag(ctx->odp_actions,
+                             ODP_ACTION_ATTR_DROP_SPOOFED_ARP);
+         }
+         break;
+     case NXAST_SET_QUEUE:
+         nasq = (const struct nx_action_set_queue *) nah;
+         xlate_set_queue_action(ctx, nasq);
+         break;
+     case NXAST_POP_QUEUE:
+         add_pop_action(ctx);
+         break;
+     case NXAST_REG_MOVE:
+         save_reg_state(ctx, &state);
+         nxm_execute_reg_move((const struct nx_action_reg_move *) nah,
+                              &ctx->flow);
+         update_reg_state(ctx, &state);
+         break;
+     case NXAST_REG_LOAD:
+         save_reg_state(ctx, &state);
+         nxm_execute_reg_load((const struct nx_action_reg_load *) nah,
+                              &ctx->flow);
+         update_reg_state(ctx, &state);
+         break;
+     case NXAST_NOTE:
+         /* Nothing to do. */
+         break;
+     case NXAST_SET_TUNNEL64:
+         tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id;
+         nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id);
+         ctx->flow.tun_id = tun_id;
+         break;
+     case NXAST_MULTIPATH:
+         nam = (const struct nx_action_multipath *) nah;
+         multipath_execute(nam, &ctx->flow);
+         break;
+     case NXAST_AUTOPATH:
+         naa = (const struct nx_action_autopath *) nah;
+         xlate_autopath(ctx, naa);
+         break;
+     /* If you add a new action here that modifies flow data, don't forget to
+      * update the flow key in ctx->flow at the same time. */
+     case NXAST_SNAT__OBSOLETE:
+     default:
+         VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype);
+         break;
+     }
+ }
+ static void
+ do_xlate_actions(const union ofp_action *in, size_t n_in,
+                  struct action_xlate_ctx *ctx)
+ {
+     const struct ofport_dpif *port;
+     struct actions_iterator iter;
+     const union ofp_action *ia;
+     port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
+     if (port
+         && port->up.opp.config & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
+         port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
+                                ? htonl(OFPPC_NO_RECV_STP)
+                                : htonl(OFPPC_NO_RECV))) {
+         /* Drop this flow. */
+         return;
+     }
+     for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) {
+         enum ofp_action_type type = ntohs(ia->type);
+         const struct ofp_action_dl_addr *oada;
+         switch (type) {
+         case OFPAT_OUTPUT:
+             xlate_output_action(ctx, &ia->output);
+             break;
+         case OFPAT_SET_VLAN_VID:
+             ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
+             ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI);
+             xlate_set_dl_tci(ctx);
+             break;
+         case OFPAT_SET_VLAN_PCP:
+             ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
+             ctx->flow.vlan_tci |= htons(
+                 (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI);
+             xlate_set_dl_tci(ctx);
+             break;
+         case OFPAT_STRIP_VLAN:
+             ctx->flow.vlan_tci = htons(0);
+             xlate_set_dl_tci(ctx);
+             break;
+         case OFPAT_SET_DL_SRC:
+             oada = ((struct ofp_action_dl_addr *) ia);
+             nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC,
+                               oada->dl_addr, ETH_ADDR_LEN);
+             memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN);
+             break;
+         case OFPAT_SET_DL_DST:
+             oada = ((struct ofp_action_dl_addr *) ia);
+             nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST,
+                               oada->dl_addr, ETH_ADDR_LEN);
+             memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN);
+             break;
+         case OFPAT_SET_NW_SRC:
+             nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC,
+                             ia->nw_addr.nw_addr);
+             ctx->flow.nw_src = ia->nw_addr.nw_addr;
+             break;
+         case OFPAT_SET_NW_DST:
+             nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST,
+                             ia->nw_addr.nw_addr);
+             ctx->flow.nw_dst = ia->nw_addr.nw_addr;
+             break;
+         case OFPAT_SET_NW_TOS:
+             nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS,
+                           ia->nw_tos.nw_tos);
+             ctx->flow.nw_tos = ia->nw_tos.nw_tos;
+             break;
+         case OFPAT_SET_TP_SRC:
+             nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC,
+                             ia->tp_port.tp_port);
+             ctx->flow.tp_src = ia->tp_port.tp_port;
+             break;
+         case OFPAT_SET_TP_DST:
+             nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST,
+                             ia->tp_port.tp_port);
+             ctx->flow.tp_dst = ia->tp_port.tp_port;
+             break;
+         case OFPAT_VENDOR:
+             xlate_nicira_action(ctx, (const struct nx_action_header *) ia);
+             break;
+         case OFPAT_ENQUEUE:
+             xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia);
+             break;
+         default:
+             VLOG_DBG_RL(&rl, "unknown action type %d", (int) type);
+             break;
+         }
+     }
+ }
+ static void
+ action_xlate_ctx_init(struct action_xlate_ctx *ctx,
+                       struct ofproto_dpif *ofproto, const struct flow *flow,
+                       const struct ofpbuf *packet)
+ {
+     ctx->ofproto = ofproto;
+     ctx->flow = *flow;
+     ctx->packet = packet;
+     ctx->resubmit_hook = NULL;
+ }
+ static struct ofpbuf *
+ xlate_actions(struct action_xlate_ctx *ctx,
+               const union ofp_action *in, size_t n_in)
+ {
+     COVERAGE_INC(ofproto_dpif_xlate);
+     ctx->odp_actions = ofpbuf_new(512);
+     ctx->tags = 0;
+     ctx->may_set_up_flow = true;
+     ctx->nf_output_iface = NF_OUT_DROP;
+     ctx->recurse = 0;
+     ctx->last_pop_priority = -1;
+     if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) {
+         ctx->may_set_up_flow = false;
+     } else {
+         do_xlate_actions(in, n_in, ctx);
+     }
+     remove_pop_action(ctx);
+     /* Check with in-band control to see if we're allowed to set up this
+      * flow. */
+     if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
+                                  ctx->odp_actions->data,
+                                  ctx->odp_actions->size)) {
+         ctx->may_set_up_flow = false;
+     }
+     return ctx->odp_actions;
+ }
\f
+ /* OFPP_NORMAL implementation. */
+ struct dst {
+     struct ofport_dpif *port;
+     uint16_t vlan;
+ };
+ struct dst_set {
+     struct dst builtin[32];
+     struct dst *dsts;
+     size_t n, allocated;
+ };
+ static void dst_set_init(struct dst_set *);
+ static void dst_set_add(struct dst_set *, const struct dst *);
+ static void dst_set_free(struct dst_set *);
+ static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
+ static bool
+ set_dst(struct action_xlate_ctx *ctx, struct dst *dst,
+         const struct ofbundle *in_bundle, const struct ofbundle *out_bundle)
+ {
+     dst->vlan = (out_bundle->vlan >= 0 ? OFP_VLAN_NONE
+                  : in_bundle->vlan >= 0 ? in_bundle->vlan
+                  : ctx->flow.vlan_tci == 0 ? OFP_VLAN_NONE
+                  : vlan_tci_to_vid(ctx->flow.vlan_tci));
+     dst->port = (!out_bundle->bond
+                  ? ofbundle_get_a_port(out_bundle)
+                  : bond_choose_output_slave(out_bundle->bond, &ctx->flow,
+                                             dst->vlan, &ctx->tags));
+     return dst->port != NULL;
+ }
+ static int
+ mirror_mask_ffs(mirror_mask_t mask)
+ {
+     BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
+     return ffs(mask);
+ }
+ static void
+ dst_set_init(struct dst_set *set)
+ {
+     set->dsts = set->builtin;
+     set->n = 0;
+     set->allocated = ARRAY_SIZE(set->builtin);
+ }
+ static void
+ dst_set_add(struct dst_set *set, const struct dst *dst)
+ {
+     if (set->n >= set->allocated) {
+         size_t new_allocated;
+         struct dst *new_dsts;
+         new_allocated = set->allocated * 2;
+         new_dsts = xmalloc(new_allocated * sizeof *new_dsts);
+         memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts);
+         dst_set_free(set);
+         set->dsts = new_dsts;
+         set->allocated = new_allocated;
+     }
+     set->dsts[set->n++] = *dst;
+ }
+ static void
+ dst_set_free(struct dst_set *set)
+ {
+     if (set->dsts != set->builtin) {
+         free(set->dsts);
+     }
+ }
+ static bool
+ dst_is_duplicate(const struct dst_set *set, const struct dst *test)
+ {
+     size_t i;
+     for (i = 0; i < set->n; i++) {
+         if (set->dsts[i].vlan == test->vlan
+             && set->dsts[i].port == test->port) {
+             return true;
+         }
+     }
+     return false;
+ }
+ static bool
+ ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
+ {
+     return bundle->vlan < 0 && vlan_bitmap_contains(bundle->trunks, vlan);
+ }
+ static bool
+ ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
+ {
+     return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
+ }
+ /* Returns an arbitrary interface within 'bundle'. */
+ static struct ofport_dpif *
+ ofbundle_get_a_port(const struct ofbundle *bundle)
+ {
+     return CONTAINER_OF(list_front(&bundle->ports),
+                         struct ofport_dpif, bundle_node);
+ }
+ static void
+ compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan,
+              const struct ofbundle *in_bundle,
+              const struct ofbundle *out_bundle, struct dst_set *set)
+ {
+     struct dst dst;
+     if (out_bundle == OFBUNDLE_FLOOD) {
+         struct ofbundle *bundle;
+         HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
+             if (bundle != in_bundle
+                 && ofbundle_includes_vlan(bundle, vlan)
+                 && bundle->floodable
+                 && !bundle->mirror_out
+                 && set_dst(ctx, &dst, in_bundle, bundle)) {
+                 dst_set_add(set, &dst);
+             }
+         }
+         ctx->nf_output_iface = NF_OUT_FLOOD;
+     } else if (out_bundle && set_dst(ctx, &dst, in_bundle, out_bundle)) {
+         dst_set_add(set, &dst);
+         ctx->nf_output_iface = dst.port->odp_port;
+     }
+ }
+ static bool
+ vlan_is_mirrored(const struct ofmirror *m, int vlan)
+ {
+     return vlan_bitmap_contains(m->vlans, vlan);
+ }
+ static void
+ compose_mirror_dsts(struct action_xlate_ctx *ctx,
+                     uint16_t vlan, const struct ofbundle *in_bundle,
+                     struct dst_set *set)
+ {
+     struct ofproto_dpif *ofproto = ctx->ofproto;
+     mirror_mask_t mirrors;
+     int flow_vlan;
+     size_t i;
+     mirrors = in_bundle->src_mirrors;
+     for (i = 0; i < set->n; i++) {
+         mirrors |= set->dsts[i].port->bundle->dst_mirrors;
+     }
+     if (!mirrors) {
+         return;
+     }
+     flow_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci);
+     if (flow_vlan == 0) {
+         flow_vlan = OFP_VLAN_NONE;
+     }
+     while (mirrors) {
+         struct ofmirror *m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
+         if (vlan_is_mirrored(m, vlan)) {
+             struct dst dst;
+             if (m->out) {
+                 if (set_dst(ctx, &dst, in_bundle, m->out)
+                     && !dst_is_duplicate(set, &dst)) {
+                     dst_set_add(set, &dst);
+                 }
+             } else {
+                 struct ofbundle *bundle;
+                 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
+                     if (ofbundle_includes_vlan(bundle, m->out_vlan)
+                         && set_dst(ctx, &dst, in_bundle, bundle))
+                     {
+                         if (bundle->vlan < 0) {
+                             dst.vlan = m->out_vlan;
+                         }
+                         if (dst_is_duplicate(set, &dst)) {
+                             continue;
+                         }
+                         /* Use the vlan tag on the original flow instead of
+                          * the one passed in the vlan parameter.  This ensures
+                          * that we compare the vlan from before any implicit
+                          * tagging tags place. This is necessary because
+                          * dst->vlan is the final vlan, after removing implicit
+                          * tags. */
+                         if (bundle == in_bundle && dst.vlan == flow_vlan) {
+                             /* Don't send out input port on same VLAN. */
+                             continue;
+                         }
+                         dst_set_add(set, &dst);
+                     }
+                 }
+             }
+         }
+         mirrors &= mirrors - 1;
+     }
+ }
+ static void
+ compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan,
+                 const struct ofbundle *in_bundle,
+                 const struct ofbundle *out_bundle)
+ {
+     uint16_t initial_vlan, cur_vlan;
+     const struct dst *dst;
+     struct dst_set set;
+     dst_set_init(&set);
+     compose_dsts(ctx, vlan, in_bundle, out_bundle, &set);
+     compose_mirror_dsts(ctx, vlan, in_bundle, &set);
+     /* Output all the packets we can without having to change the VLAN. */
+     initial_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci);
+     if (initial_vlan == 0) {
+         initial_vlan = OFP_VLAN_NONE;
+     }
+     for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
+         if (dst->vlan != initial_vlan) {
+             continue;
+         }
+         nl_msg_put_u32(ctx->odp_actions,
+                        ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port);
+     }
+     /* Then output the rest. */
+     cur_vlan = initial_vlan;
+     for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) {
+         if (dst->vlan == initial_vlan) {
+             continue;
+         }
+         if (dst->vlan != cur_vlan) {
+             if (dst->vlan == OFP_VLAN_NONE) {
+                 nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN);
+             } else {
+                 ovs_be16 tci;
+                 tci = htons(dst->vlan & VLAN_VID_MASK);
+                 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
+                 nl_msg_put_be16(ctx->odp_actions,
+                                 ODP_ACTION_ATTR_SET_DL_TCI, tci);
+             }
+             cur_vlan = dst->vlan;
+         }
+         nl_msg_put_u32(ctx->odp_actions,
+                        ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port);
+     }
+     dst_set_free(&set);
+ }
+ /* Returns the effective vlan of a packet, taking into account both the
+  * 802.1Q header and implicitly tagged ports.  A value of 0 indicates that
+  * the packet is untagged and -1 indicates it has an invalid header and
+  * should be dropped. */
+ static int
+ flow_get_vlan(struct ofproto_dpif *ofproto, const struct flow *flow,
+               struct ofbundle *in_bundle, bool have_packet)
+ {
+     int vlan = vlan_tci_to_vid(flow->vlan_tci);
+     if (in_bundle->vlan >= 0) {
+         if (vlan) {
+             if (have_packet) {
+                 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+                 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
+                              "packet received on port %s configured with "
+                              "implicit VLAN %"PRIu16,
+                              ofproto->up.name, vlan,
+                              in_bundle->name, in_bundle->vlan);
+             }
+             return -1;
+         }
+         vlan = in_bundle->vlan;
+     } else {
+         if (!ofbundle_includes_vlan(in_bundle, vlan)) {
+             if (have_packet) {
+                 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+                 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
+                              "packet received on port %s not configured for "
+                              "trunking VLAN %d",
+                              ofproto->up.name, vlan, in_bundle->name, vlan);
+             }
+             return -1;
+         }
+     }
+     return vlan;
+ }
+ /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
+  * migration.  Older Citrix-patched Linux DomU used gratuitous ARP replies to
+  * indicate this; newer upstream kernels use gratuitous ARP requests. */
+ static bool
+ is_gratuitous_arp(const struct flow *flow)
+ {
+     return (flow->dl_type == htons(ETH_TYPE_ARP)
+             && eth_addr_is_broadcast(flow->dl_dst)
+             && (flow->nw_proto == ARP_OP_REPLY
+                 || (flow->nw_proto == ARP_OP_REQUEST
+                     && flow->nw_src == flow->nw_dst)));
+ }
+ static void
+ update_learning_table(struct ofproto_dpif *ofproto,
+                       const struct flow *flow, int vlan,
+                       struct ofbundle *in_bundle)
+ {
+     struct mac_entry *mac;
+     if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
+         return;
+     }
+     mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
+     if (is_gratuitous_arp(flow)) {
+         /* We don't want to learn from gratuitous ARP packets that are
+          * reflected back over bond slaves so we lock the learning table. */
+         if (!in_bundle->bond) {
+             mac_entry_set_grat_arp_lock(mac);
+         } else if (mac_entry_is_grat_arp_locked(mac)) {
+             return;
+         }
+     }
+     if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
+         /* The log messages here could actually be useful in debugging,
+          * so keep the rate limit relatively high. */
+         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
+         VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
+                     "on port %s in VLAN %d",
+                     ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
+                     in_bundle->name, vlan);
+         mac->port.p = in_bundle;
+         tag_set_add(&ofproto->revalidate_set,
+                     mac_learning_changed(ofproto->ml, mac));
+     }
+ }
+ /* Determines whether packets in 'flow' within 'br' should be forwarded or
+  * dropped.  Returns true if they may be forwarded, false if they should be
+  * dropped.
+  *
+  * If 'have_packet' is true, it indicates that the caller is processing a
+  * received packet.  If 'have_packet' is false, then the caller is just
+  * revalidating an existing flow because configuration has changed.  Either
+  * way, 'have_packet' only affects logging (there is no point in logging errors
+  * during revalidation).
+  *
+  * Sets '*in_portp' to the input port.  This will be a null pointer if
+  * flow->in_port does not designate a known input port (in which case
+  * is_admissible() returns false).
+  *
+  * When returning true, sets '*vlanp' to the effective VLAN of the input
+  * packet, as returned by flow_get_vlan().
+  *
+  * May also add tags to '*tags', although the current implementation only does
+  * so in one special case.
+  */
+ static bool
+ is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow,
+               bool have_packet,
+               tag_type *tags, int *vlanp, struct ofbundle **in_bundlep)
+ {
+     struct ofport_dpif *in_port;
+     struct ofbundle *in_bundle;
+     int vlan;
+     /* Find the port and bundle for the received packet. */
+     in_port = get_ofp_port(ofproto, flow->in_port);
+     *in_bundlep = in_bundle = in_port ? in_port->bundle : NULL;
+     if (!in_port || !in_bundle) {
+         /* No interface?  Something fishy... */
+         if (have_packet) {
+             /* Odd.  A few possible reasons here:
+              *
+              * - We deleted a port but there are still a few packets queued up
+              *   from it.
+              *
+              * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
+              *   we don't know about.
+              *
+              * - Packet arrived on the local port but the local port is not
+              *   part of a bundle.
+              */
+             static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+             VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
+                          "port %"PRIu16,
+                          ofproto->up.name, flow->in_port);
+         }
+         return false;
+     }
+     *vlanp = vlan = flow_get_vlan(ofproto, flow, in_bundle, have_packet);
+     if (vlan < 0) {
+         return false;
+     }
+     /* Drop frames for reserved multicast addresses. */
+     if (eth_addr_is_reserved(flow->dl_dst)) {
+         return false;
+     }
+     /* Drop frames on bundles reserved for mirroring. */
+     if (in_bundle->mirror_out) {
+         if (have_packet) {
+             static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+             VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
+                          "%s, which is reserved exclusively for mirroring",
+                          ofproto->up.name, in_bundle->name);
+         }
+         return false;
+     }
+     if (in_bundle->bond) {
+         struct mac_entry *mac;
+         switch (bond_check_admissibility(in_bundle->bond, in_port,
+                                          flow->dl_dst, tags)) {
+         case BV_ACCEPT:
+             break;
+         case BV_DROP:
+             return false;
+         case BV_DROP_IF_MOVED:
+             mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
+             if (mac && mac->port.p != in_bundle &&
+                 (!is_gratuitous_arp(flow)
+                  || mac_entry_is_grat_arp_locked(mac))) {
+                 return false;
+             }
+             break;
+         }
+     }
+     return true;
+ }
+ /* If the composed actions may be applied to any packet in the given 'flow',
+  * returns true.  Otherwise, the actions should only be applied to 'packet', or
+  * not at all, if 'packet' was NULL. */
+ static bool
+ xlate_normal(struct action_xlate_ctx *ctx)
+ {
+     struct ofbundle *in_bundle;
+     struct ofbundle *out_bundle;
+     struct mac_entry *mac;
+     int vlan;
+     /* Check whether we should drop packets in this flow. */
+     if (!is_admissible(ctx->ofproto, &ctx->flow, ctx->packet != NULL,
+                        &ctx->tags, &vlan, &in_bundle)) {
+         out_bundle = NULL;
+         goto done;
+     }
+     /* Learn source MAC (but don't try to learn from revalidation). */
+     if (ctx->packet) {
+         update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
+     }
+     /* Determine output bundle. */
+     mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
+                               &ctx->tags);
+     if (mac) {
+         out_bundle = mac->port.p;
+     } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) {
+         /* If we are revalidating but don't have a learning entry then eject
+          * the flow.  Installing a flow that floods packets opens up a window
+          * of time where we could learn from a packet reflected on a bond and
+          * blackhole packets before the learning table is updated to reflect
+          * the correct port. */
+         return false;
+     } else {
+         out_bundle = OFBUNDLE_FLOOD;
+     }
+     /* Don't send packets out their input bundles. */
+     if (in_bundle == out_bundle) {
+         out_bundle = NULL;
+     }
+ done:
+     if (in_bundle) {
+         compose_actions(ctx, vlan, in_bundle, out_bundle);
+     }
+     return true;
+ }
\f
+ static bool
+ get_drop_frags(struct ofproto *ofproto_)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     bool drop_frags;
+     dpif_get_drop_frags(ofproto->dpif, &drop_frags);
+     return drop_frags;
+ }
+ static void
+ set_drop_frags(struct ofproto *ofproto_, bool drop_frags)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     dpif_set_drop_frags(ofproto->dpif, drop_frags);
+ }
+ static int
+ packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
+            const struct flow *flow,
+            const union ofp_action *ofp_actions, size_t n_ofp_actions)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     int error;
+     error = validate_actions(ofp_actions, n_ofp_actions, flow,
+                              ofproto->max_ports);
+     if (!error) {
+         struct action_xlate_ctx ctx;
+         struct ofpbuf *odp_actions;
+         action_xlate_ctx_init(&ctx, ofproto, flow, packet);
+         odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions);
+         dpif_execute(ofproto->dpif, odp_actions->data, odp_actions->size,
+                      packet);
+         ofpbuf_delete(odp_actions);
+     }
+     return error;
+ }
+ static void
+ get_netflow_ids(const struct ofproto *ofproto_,
+                 uint8_t *engine_type, uint8_t *engine_id)
+ {
+     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+     dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id);
+ }
\f
+ static struct ofproto_dpif *
+ ofproto_dpif_lookup(const char *name)
+ {
+     struct ofproto *ofproto = ofproto_lookup(name);
+     return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class
+             ? ofproto_dpif_cast(ofproto)
+             : NULL);
+ }
+ static void
+ ofproto_unixctl_fdb_show(struct unixctl_conn *conn,
+                          const char *args, void *aux OVS_UNUSED)
+ {
+     struct ds ds = DS_EMPTY_INITIALIZER;
+     const struct ofproto_dpif *ofproto;
+     const struct mac_entry *e;
+     ofproto = ofproto_dpif_lookup(args);
+     if (!ofproto) {
+         unixctl_command_reply(conn, 501, "no such bridge");
+         return;
+     }
+     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
+     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
+         struct ofbundle *bundle = e->port.p;
+         ds_put_format(&ds, "%5d  %4d  "ETH_ADDR_FMT"  %3d\n",
+                       ofbundle_get_a_port(bundle)->odp_port,
+                       e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
+     }
+     unixctl_command_reply(conn, 200, ds_cstr(&ds));
+     ds_destroy(&ds);
+ }
+ struct ofproto_trace {
+     struct action_xlate_ctx ctx;
+     struct flow flow;
+     struct ds *result;
+ };
+ static void
+ trace_format_rule(struct ds *result, int level, const struct rule *rule)
+ {
+     ds_put_char_multiple(result, '\t', level);
+     if (!rule) {
+         ds_put_cstr(result, "No match\n");
+         return;
+     }
+     ds_put_format(result, "Rule: cookie=%#"PRIx64" ",
+                   ntohll(rule->flow_cookie));
+     cls_rule_format(&rule->cr, result);
+     ds_put_char(result, '\n');
+     ds_put_char_multiple(result, '\t', level);
+     ds_put_cstr(result, "OpenFlow ");
+     ofp_print_actions(result, (const struct ofp_action_header *) rule->actions,
+                       rule->n_actions * sizeof *rule->actions);
+     ds_put_char(result, '\n');
+ }
+ static void
+ trace_format_flow(struct ds *result, int level, const char *title,
+                  struct ofproto_trace *trace)
+ {
+     ds_put_char_multiple(result, '\t', level);
+     ds_put_format(result, "%s: ", title);
+     if (flow_equal(&trace->ctx.flow, &trace->flow)) {
+         ds_put_cstr(result, "unchanged");
+     } else {
+         flow_format(result, &trace->ctx.flow);
+         trace->flow = trace->ctx.flow;
+     }
+     ds_put_char(result, '\n');
+ }
+ static void
+ trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
+ {
+     struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx);
+     struct ds *result = trace->result;
+     ds_put_char(result, '\n');
+     trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
+     trace_format_rule(result, ctx->recurse + 1, &rule->up);
+ }
+ static void
+ ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_,
+                       void *aux OVS_UNUSED)
+ {
+     char *dpname, *in_port_s, *tun_id_s, *packet_s;
+     char *args = xstrdup(args_);
+     char *save_ptr = NULL;
+     struct ofproto_dpif *ofproto;
+     struct ofpbuf packet;
+     struct rule_dpif *rule;
+     struct ds result;
+     struct flow flow;
+     uint16_t in_port;
+     ovs_be64 tun_id;
+     char *s;
+     ofpbuf_init(&packet, strlen(args) / 2);
+     ds_init(&result);
+     dpname = strtok_r(args, " ", &save_ptr);
+     tun_id_s = strtok_r(NULL, " ", &save_ptr);
+     in_port_s = strtok_r(NULL, " ", &save_ptr);
+     packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */
+     if (!dpname || !in_port_s || !packet_s) {
+         unixctl_command_reply(conn, 501, "Bad command syntax");
+         goto exit;
+     }
+     ofproto = ofproto_dpif_lookup(dpname);
+     if (!ofproto) {
+         unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list "
+                               "for help)");
+         goto exit;
+     }
+     tun_id = htonll(strtoull(tun_id_s, NULL, 0));
+     in_port = ofp_port_to_odp_port(atoi(in_port_s));
+     packet_s = ofpbuf_put_hex(&packet, packet_s, NULL);
+     packet_s += strspn(packet_s, " ");
+     if (*packet_s != '\0') {
+         unixctl_command_reply(conn, 501, "Trailing garbage in command");
+         goto exit;
+     }
+     if (packet.size < ETH_HEADER_LEN) {
+         unixctl_command_reply(conn, 501, "Packet data too short for Ethernet");
+         goto exit;
+     }
+     ds_put_cstr(&result, "Packet: ");
+     s = ofp_packet_to_string(packet.data, packet.size, packet.size);
+     ds_put_cstr(&result, s);
+     free(s);
+     flow_extract(&packet, tun_id, in_port, &flow);
+     ds_put_cstr(&result, "Flow: ");
+     flow_format(&result, &flow);
+     ds_put_char(&result, '\n');
+     rule = rule_dpif_lookup(ofproto, &flow);
+     trace_format_rule(&result, 0, &rule->up);
+     if (rule) {
+         struct ofproto_trace trace;
+         struct ofpbuf *odp_actions;
+         trace.result = &result;
+         trace.flow = flow;
+         action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet);
+         trace.ctx.resubmit_hook = trace_resubmit;
+         odp_actions = xlate_actions(&trace.ctx,
+                                     rule->up.actions, rule->up.n_actions);
+         ds_put_char(&result, '\n');
+         trace_format_flow(&result, 0, "Final flow", &trace);
+         ds_put_cstr(&result, "Datapath actions: ");
+         format_odp_actions(&result, odp_actions->data, odp_actions->size);
+         ofpbuf_delete(odp_actions);
+     }
+     unixctl_command_reply(conn, 200, ds_cstr(&result));
+ exit:
+     ds_destroy(&result);
+     ofpbuf_uninit(&packet);
+     free(args);
+ }
+ static void
+ ofproto_dpif_unixctl_init(void)
+ {
+     static bool registered;
+     if (registered) {
+         return;
+     }
+     registered = true;
+     unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL);
+     unixctl_command_register("fdb/show", ofproto_unixctl_fdb_show, NULL);
+ }
\f
+ const struct ofproto_class ofproto_dpif_class = {
+     enumerate_types,
+     enumerate_names,
+     del,
+     alloc,
+     construct,
+     destruct,
+     dealloc,
+     run,
+     wait,
+     flush,
+     get_features,
+     get_tables,
+     port_alloc,
+     port_construct,
+     port_destruct,
+     port_dealloc,
+     port_modified,
+     port_reconfigured,
+     port_query_by_name,
+     port_add,
+     port_del,
+     port_dump_start,
+     port_dump_next,
+     port_dump_done,
+     port_poll,
+     port_poll_wait,
+     port_is_lacp_current,
+     rule_alloc,
+     rule_construct,
+     rule_destruct,
+     rule_dealloc,
+     rule_get_stats,
+     rule_execute,
+     rule_modify_actions,
+     get_drop_frags,
+     set_drop_frags,
+     packet_out,
+     set_netflow,
+     get_netflow_ids,
+     set_sflow,
+     set_cfm,
+     get_cfm,
+     bundle_set,
+     bundle_remove,
+     mirror_set,
+     set_flood_vlans,
+     is_mirror_output_bundle,
+ };
Simple merge
index fabdb849ba64131610ce922d6a6014f1c656af82,d5ff2ad6a0f89a3b32f6b4b11a3dab8da393b6e8..3927550684e2d8e6e1377e3975c27fd86199b2c4
@@@ -7,14 -7,8 +7,14 @@@ m4_define([OFPROTO_START]
     trap 'kill `cat ovs-openflowd.pid`' 0
     AT_CAPTURE_FILE([ovs-openflowd.log])
     AT_CHECK(
-      [ovs-openflowd --detach --pidfile --enable-dummy --log-file dummy@br0 none --datapath-id=fedcba9876543210 $1],
+      [ovs-openflowd --detach --pidfile --enable-dummy --log-file --fail=closed dummy@br0 none --datapath-id=fedcba9876543210 $1],
 -     [0], [ignore], [ignore])
 +     [0], [], [stderr])
 +   AT_CHECK([[sed < stderr '
 +/vlog|INFO|opened log file/d
 +/openflowd|INFO|Open vSwitch version/d
 +/openflowd|INFO|OpenFlow protocol version/d
 +/ofproto|INFO|using datapath ID/d
 +/ofproto|INFO|datapath ID changed to fedcba9876543210/d']])
  ])
  
  m4_define([OFPROTO_STOP],
index c23945f570a0297af7712a50396d04501ca66dfd,3a8f0b6da019ace5789340d1ec02d8a74ec720ca..997cc306298d00b6e119158aa6d69de2e5d0fece
@@@ -32,9 -32,21 +32,10 @@@ NXT_SET_FLOW_FORMAT: format=nx
  NXT_FLOW_MOD: ADD tun_id=0x1234 cookie:0x5678 actions=FLOOD
  NXT_FLOW_MOD: ADD actions=set_tunnel:0x1234,set_tunnel64:0x9876,set_tunnel64:0x123456789
  NXT_FLOW_MOD: ADD actions=multipath(eth_src,50,hrw,12,0,NXM_NX_REG0[0..3]),multipath(symmetric_l4,1024,iter_hash,5000,5050,NXM_NX_REG0[0..12])
- NXT_FLOW_MOD: ADD actions=drop
- NXT_FLOW_MOD: ADD tun_id=0x1234000056780000/0xffff0000ffff0000 actions=drop
+ NXT_FLOW_MOD_TABLE_ID: enable
+ NXT_FLOW_MOD: ADD table_id:1 actions=drop
+ NXT_FLOW_MOD: ADD table_id:255 tun_id=0x1234000056780000/0xffff0000ffff0000 actions=drop
  ]])
 -AT_CHECK([sed 's/.*|//' stderr], [0], [dnl
 -normalization changed ofp_match, details:
 - pre: wildcards=  0x3820f8  in_port=65534  dl_src=00:0a:e4:25:6b:b0  dl_dst=00:00:00:00:00:00  dl_vlan=    9  dl_vlan_pcp=  0  dl_type=     0  nw_tos=   0  nw_proto=   0  nw_src=         0  nw_dst=         0  tp_src=    0  tp_dst=    0
 -post: wildcards=  0x3ffff8  in_port=65534  dl_src=00:0a:e4:25:6b:b0  dl_dst=00:00:00:00:00:00  dl_vlan=    9  dl_vlan_pcp=  0  dl_type=     0  nw_tos=   0  nw_proto=   0  nw_src=         0  nw_dst=         0  tp_src=    0  tp_dst=    0
 -normalization changed ofp_match, details:
 - pre: wildcards=  0x3820ff  in_port=    0  dl_src=00:00:00:00:00:00  dl_dst=00:00:00:00:00:00  dl_vlan=    0  dl_vlan_pcp=  0  dl_type=     0  nw_tos=   0  nw_proto=   0  nw_src=         0  nw_dst=         0  tp_src=    0  tp_dst=    0
 -post: wildcards=  0x3fffff  in_port=    0  dl_src=00:00:00:00:00:00  dl_dst=00:00:00:00:00:00  dl_vlan=    0  dl_vlan_pcp=  0  dl_type=     0  nw_tos=   0  nw_proto=   0  nw_src=         0  nw_dst=         0  tp_src=    0  tp_dst=    0
 -normalization changed ofp_match, details:
 - pre: wildcards=  0x3820ff  in_port=    0  dl_src=00:00:00:00:00:00  dl_dst=00:00:00:00:00:00  dl_vlan=    0  dl_vlan_pcp=  0  dl_type=     0  nw_tos=   0  nw_proto=   0  nw_src=         0  nw_dst=         0  tp_src=    0  tp_dst=    0
 -post: wildcards=  0x3fffff  in_port=    0  dl_src=00:00:00:00:00:00  dl_dst=00:00:00:00:00:00  dl_vlan=    0  dl_vlan_pcp=  0  dl_type=     0  nw_tos=   0  nw_proto=   0  nw_src=         0  nw_dst=         0  tp_src=    0  tp_dst=    0
 -])
  AT_CLEANUP
  
  AT_SETUP([ovs-ofctl -F nxm parse-flows])
Simple merge
Simple merge
Simple merge
index 9accbddd9714a5c9f8499d39c5affd9a517672ea,dd0d1b2a4bac255f92b8008705bcf644d666471a..3faeee1e897869faacaa26f1f3d8df1e8f904f8c
@@@ -1675,32 -1626,10 +1626,14 @@@ bridge_create(const struct ovsrec_bridg
      assert(!bridge_lookup(br_cfg->name));
      br = xzalloc(sizeof *br);
  
-     error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type,
-                                  &br->dpif);
-     if (error) {
-         free(br);
-         return NULL;
-     }
-     error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks,
-                            br, &br->ofproto);
-     if (error) {
-         VLOG_ERR("failed to create switch %s: %s", br_cfg->name,
-                  strerror(error));
-         dpif_delete(br->dpif);
-         dpif_close(br->dpif);
-         free(br);
-         return NULL;
-     }
      br->name = xstrdup(br_cfg->name);
+     br->type = xstrdup(ofproto_normalize_type(br_cfg->datapath_type));
      br->cfg = br_cfg;
-     br->ml = mac_learning_create();
 -    eth_addr_nicira_random(br->default_ea);
 +
 +    /* Derive the default Ethernet address from the bridge's UUID.  This should
 +     * be unique and it will be stable between ovs-vswitchd runs.  */
 +    memcpy(br->default_ea, &br_cfg->header_.uuid, ETH_ADDR_LEN);
 +    eth_addr_mark_random(br->default_ea);
  
      hmap_init(&br->ports);
      hmap_init(&br->ifaces);
Simple merge