ofproto-dpif: avoid losing track of kernel flows upon reinstallation
[sliver-openvswitch.git] / ofproto / ofproto-dpif.c
index ca0a317..29a93e6 100644 (file)
@@ -71,6 +71,7 @@ COVERAGE_DEFINE(facet_revalidate);
 COVERAGE_DEFINE(facet_unexpected);
 COVERAGE_DEFINE(facet_suppress);
 COVERAGE_DEFINE(subfacet_install_fail);
+COVERAGE_DEFINE(flow_mod_overflow);
 
 /* Number of implemented OpenFlow tables. */
 enum { N_TABLES = 255 };
@@ -85,7 +86,6 @@ static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
                                           struct flow_wildcards *wc);
 
 static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes);
-static void rule_invalidate(const struct rule_dpif *);
 
 struct ofbundle {
     struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
@@ -289,7 +289,6 @@ struct ofport_dpif {
     struct list bundle_node;    /* In struct ofbundle's "ports" list. */
     struct cfm *cfm;            /* Connectivity Fault Management, if any. */
     struct bfd *bfd;            /* BFD, if any. */
-    tag_type tag;               /* Tag associated with this port. */
     bool may_enable;            /* May be enabled in bonds. */
     bool is_tunnel;             /* This port is a tunnel. */
     long long int carrier_seq;  /* Carrier status changes. */
@@ -300,7 +299,9 @@ struct ofport_dpif {
     enum stp_state stp_state;   /* Always STP_DISABLED if STP not in use. */
     long long int stp_state_entered;
 
-    struct hmap priorities;     /* Map of attached 'priority_to_dscp's. */
+    /* Queue to DSCP mapping. */
+    struct ofproto_port_queue *qdscp;
+    size_t n_qdscp;
 
     /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
      *
@@ -312,16 +313,6 @@ struct ofport_dpif {
     int vlandev_vid;
 };
 
-/* Node in 'ofport_dpif''s 'priorities' map.  Used to maintain a map from
- * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
- * traffic egressing the 'ofport' with that priority should be marked with. */
-struct priority_to_dscp {
-    struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
-    uint32_t priority;          /* Priority of this queue (see struct flow). */
-
-    uint8_t dscp;               /* DSCP bits to mark outgoing traffic with. */
-};
-
 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
  *
  * This is deprecated.  It is only for compatibility with broken device drivers
@@ -357,26 +348,15 @@ static void port_run_fast(struct ofport_dpif *);
 static void port_wait(struct ofport_dpif *);
 static int set_bfd(struct ofport *, const struct smap *);
 static int set_cfm(struct ofport *, const struct cfm_settings *);
-static void ofport_clear_priorities(struct ofport_dpif *);
 static void ofport_update_peer(struct ofport_dpif *);
 static void run_fast_rl(void);
+static int run_fast(struct ofproto *);
 
 struct dpif_completion {
     struct list list_node;
     struct ofoperation *op;
 };
 
-/* Extra information about a classifier table.
- * Currently used just for optimized flow revalidation. */
-struct table_dpif {
-    /* If either of these is nonnull, then this table has a form that allows
-     * flows to be tagged to avoid revalidating most flows for the most common
-     * kinds of flow table changes. */
-    struct cls_table *catchall_table; /* Table that wildcards all fields. */
-    struct cls_table *other_table;    /* Table with any other wildcard set. */
-    uint32_t basis;                   /* Keeps each table's tags separate. */
-};
-
 /* Reasons that we might need to revalidate every facet, and corresponding
  * coverage counters.
  *
@@ -388,14 +368,18 @@ struct table_dpif {
 enum revalidate_reason {
     REV_RECONFIGURE = 1,       /* Switch configuration changed. */
     REV_STP,                   /* Spanning tree protocol port status change. */
+    REV_BOND,                  /* Bonding changed. */
     REV_PORT_TOGGLED,          /* Port enabled or disabled by CFM, LACP, ...*/
     REV_FLOW_TABLE,            /* Flow table changed. */
+    REV_MAC_LEARNING,          /* Mac learning changed. */
     REV_INCONSISTENCY          /* Facet self-check failed. */
 };
 COVERAGE_DEFINE(rev_reconfigure);
 COVERAGE_DEFINE(rev_stp);
+COVERAGE_DEFINE(rev_bond);
 COVERAGE_DEFINE(rev_port_toggled);
 COVERAGE_DEFINE(rev_flow_table);
+COVERAGE_DEFINE(rev_mac_learning);
 COVERAGE_DEFINE(rev_inconsistency);
 
 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
@@ -424,7 +408,6 @@ struct dpif_backer {
 
     /* Facet revalidation flags applying to facets which use this backer. */
     enum revalidate_reason need_revalidate; /* Revalidate every facet. */
-    struct tag_set revalidate_set; /* Revalidate only matching facets. */
 
     struct hmap drop_keys; /* Set of dropped odp keys. */
     bool recv_set_enable; /* Enables or disables receiving packets. */
@@ -490,9 +473,6 @@ struct ofproto_dpif {
     struct classifier facets;     /* Contains 'struct facet's. */
     long long int consistency_rl;
 
-    /* Revalidation. */
-    struct table_dpif tables[N_TABLES];
-
     /* Support for debugging async flow mods. */
     struct list completions;
 
@@ -516,6 +496,11 @@ struct ofproto_dpif {
     /* Per ofproto's dpif stats. */
     uint64_t n_hit;
     uint64_t n_missed;
+
+    /* Work queues. */
+    struct ovs_mutex flow_mod_mutex;
+    struct list flow_mods OVS_GUARDED;
+    size_t n_flow_mods OVS_GUARDED;
 };
 
 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"?  (Useful only
@@ -560,11 +545,23 @@ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 /* Initial mappings of port to bridge mappings. */
 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
 
-int
+/* Executes and takes ownership of 'fm'. */
+void
 ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto,
                       struct ofputil_flow_mod *fm)
 {
-    return ofproto_flow_mod(&ofproto->up, fm);
+    ovs_mutex_lock(&ofproto->flow_mod_mutex);
+    if (ofproto->n_flow_mods > 1024) {
+        ovs_mutex_unlock(&ofproto->flow_mod_mutex);
+        COVERAGE_INC(flow_mod_overflow);
+        free(fm->ofpacts);
+        free(fm);
+        return;
+    }
+
+    list_push_back(&ofproto->flow_mods, &fm->list_node);
+    ofproto->n_flow_mods++;
+    ovs_mutex_unlock(&ofproto->flow_mod_mutex);
 }
 
 void
@@ -700,10 +697,7 @@ type_run(const char *type)
         backer->need_revalidate = REV_RECONFIGURE;
     }
 
-    if (backer->need_revalidate
-        || !tag_set_is_empty(&backer->revalidate_set)) {
-        struct tag_set revalidate_set = backer->revalidate_set;
-        bool need_revalidate = backer->need_revalidate;
+    if (backer->need_revalidate) {
         struct ofproto_dpif *ofproto;
         struct simap_node *node;
         struct simap tmp_backers;
@@ -764,66 +758,61 @@ type_run(const char *type)
         switch (backer->need_revalidate) {
         case REV_RECONFIGURE:   COVERAGE_INC(rev_reconfigure);   break;
         case REV_STP:           COVERAGE_INC(rev_stp);           break;
+        case REV_BOND:          COVERAGE_INC(rev_bond);          break;
         case REV_PORT_TOGGLED:  COVERAGE_INC(rev_port_toggled);  break;
         case REV_FLOW_TABLE:    COVERAGE_INC(rev_flow_table);    break;
+        case REV_MAC_LEARNING:  COVERAGE_INC(rev_mac_learning);  break;
         case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
         }
-
-        if (backer->need_revalidate) {
-            /* Clear the drop_keys in case we should now be accepting some
-             * formerly dropped flows. */
-            drop_key_clear(backer);
-        }
-
-        /* Clear the revalidation flags. */
-        tag_set_init(&backer->revalidate_set);
         backer->need_revalidate = 0;
 
+        /* Clear the drop_keys in case we should now be accepting some
+         * formerly dropped flows. */
+        drop_key_clear(backer);
+
         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
             struct facet *facet, *next;
+            struct ofport_dpif *ofport;
             struct cls_cursor cursor;
+            struct ofbundle *bundle;
 
             if (ofproto->backer != backer) {
                 continue;
             }
 
-            if (need_revalidate) {
-                struct ofport_dpif *ofport;
-                struct ofbundle *bundle;
-
-                xlate_ofproto_set(ofproto, ofproto->up.name, ofproto->ml,
-                                  ofproto->mbridge, ofproto->sflow,
-                                  ofproto->ipfix, ofproto->up.frag_handling,
-                                  ofproto->up.forward_bpdu,
-                                  connmgr_has_in_band(ofproto->up.connmgr),
-                                  ofproto->netflow != NULL,
-                                  ofproto->stp != NULL);
-
-                HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
-                    xlate_bundle_set(ofproto, bundle, bundle->name,
-                                     bundle->vlan_mode, bundle->vlan,
-                                     bundle->trunks, bundle->use_priority_tags,
-                                     bundle->bond, bundle->lacp,
-                                     bundle->floodable);
-                }
+            xlate_ofproto_set(ofproto, ofproto->up.name, ofproto->ml,
+                              ofproto->stp, ofproto->mbridge,
+                              ofproto->sflow, ofproto->ipfix,
+                              ofproto->up.frag_handling,
+                              ofproto->up.forward_bpdu,
+                              connmgr_has_in_band(ofproto->up.connmgr),
+                              ofproto->netflow != NULL);
 
-                HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
-                    xlate_ofport_set(ofproto, ofport->bundle, ofport,
-                                     ofport->up.ofp_port, ofport->odp_port,
-                                     ofport->up.netdev, ofport->cfm,
-                                     ofport->bfd, ofport->peer,
-                                     ofport->up.pp.config, ofport->stp_state,
-                                     ofport->is_tunnel, ofport->may_enable);
-                }
+            HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
+                xlate_bundle_set(ofproto, bundle, bundle->name,
+                                 bundle->vlan_mode, bundle->vlan,
+                                 bundle->trunks, bundle->use_priority_tags,
+                                 bundle->bond, bundle->lacp,
+                                 bundle->floodable);
+            }
+
+            HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
+                int stp_port = ofport->stp_port
+                    ? stp_port_no(ofport->stp_port)
+                    : 0;
+                xlate_ofport_set(ofproto, ofport->bundle, ofport,
+                                 ofport->up.ofp_port, ofport->odp_port,
+                                 ofport->up.netdev, ofport->cfm,
+                                 ofport->bfd, ofport->peer, stp_port,
+                                 ofport->qdscp, ofport->n_qdscp,
+                                 ofport->up.pp.config, ofport->is_tunnel,
+                                 ofport->may_enable);
             }
 
             cls_cursor_init(&cursor, &ofproto->facets, NULL);
             CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) {
-                if (need_revalidate
-                    || tag_set_intersects(&revalidate_set, facet->xout.tags)) {
-                    facet_revalidate(facet);
-                    run_fast_rl();
-                }
+                facet_revalidate(facet);
+                run_fast_rl();
             }
         }
     }
@@ -1040,13 +1029,9 @@ run_fast_rl(void)
 
     if (time_msec() >= port_rl) {
         struct ofproto_dpif *ofproto;
-        struct ofport_dpif *ofport;
 
         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
-
-            HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
-                port_run_fast(ofport);
-            }
+            run_fast(&ofproto->up);
         }
         port_rl = time_msec() + 200;
     }
@@ -1204,7 +1189,6 @@ open_dpif_backer(const char *type, struct dpif_backer **backerp)
     timer_set_duration(&backer->next_expiration, 1000);
     backer->need_revalidate = 0;
     simap_init(&backer->tnl_backers);
-    tag_set_init(&backer->revalidate_set);
     backer->recv_set_enable = !ofproto_get_flow_restore_wait();
     *backerp = backer;
 
@@ -1265,7 +1249,6 @@ construct(struct ofproto *ofproto_)
     struct shash_node *node, *next;
     odp_port_t max_ports;
     int error;
-    int i;
 
     error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
     if (error) {
@@ -1288,16 +1271,14 @@ construct(struct ofproto *ofproto_)
     classifier_init(&ofproto->facets);
     ofproto->consistency_rl = LLONG_MIN;
 
-    for (i = 0; i < N_TABLES; i++) {
-        struct table_dpif *table = &ofproto->tables[i];
-
-        table->catchall_table = NULL;
-        table->other_table = NULL;
-        table->basis = random_uint32();
-    }
-
     list_init(&ofproto->completions);
 
+    ovs_mutex_init(&ofproto->flow_mod_mutex, PTHREAD_MUTEX_NORMAL);
+    ovs_mutex_lock(&ofproto->flow_mod_mutex);
+    list_init(&ofproto->flow_mods);
+    ofproto->n_flow_mods = 0;
+    ovs_mutex_unlock(&ofproto->flow_mod_mutex);
+
     ofproto_dpif_unixctl_init();
 
     hmap_init(&ofproto->vlandev_map);
@@ -1428,6 +1409,7 @@ destruct(struct ofproto *ofproto_)
 {
     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
     struct rule_dpif *rule, *next_rule;
+    struct ofputil_flow_mod *fm, *next_fm;
     struct oftable *table;
 
     ofproto->backer->need_revalidate = REV_RECONFIGURE;
@@ -1445,6 +1427,16 @@ destruct(struct ofproto *ofproto_)
         }
     }
 
+    ovs_mutex_lock(&ofproto->flow_mod_mutex);
+    LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &ofproto->flow_mods) {
+        list_remove(&fm->list_node);
+        ofproto->n_flow_mods--;
+        free(fm->ofpacts);
+        free(fm);
+    }
+    ovs_mutex_unlock(&ofproto->flow_mod_mutex);
+    ovs_mutex_destroy(&ofproto->flow_mod_mutex);
+
     mbridge_unref(ofproto->mbridge);
 
     netflow_destroy(ofproto->netflow);
@@ -1468,7 +1460,9 @@ static int
 run_fast(struct ofproto *ofproto_)
 {
     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+    struct ofputil_flow_mod *fm, *next;
     struct ofport_dpif *ofport;
+    struct list flow_mods;
 
     /* Do not perform any periodic activity required by 'ofproto' while
      * waiting for flow restore to complete. */
@@ -1476,6 +1470,29 @@ run_fast(struct ofproto *ofproto_)
         return 0;
     }
 
+    ovs_mutex_lock(&ofproto->flow_mod_mutex);
+    if (ofproto->n_flow_mods) {
+        flow_mods = ofproto->flow_mods;
+        list_moved(&flow_mods);
+        list_init(&ofproto->flow_mods);
+        ofproto->n_flow_mods = 0;
+    } else {
+        list_init(&flow_mods);
+    }
+    ovs_mutex_unlock(&ofproto->flow_mod_mutex);
+
+    LIST_FOR_EACH_SAFE (fm, next, list_node, &flow_mods) {
+        int error = ofproto_flow_mod(&ofproto->up, fm);
+        if (error && !VLOG_DROP_WARN(&rl)) {
+            VLOG_WARN("learning action failed to modify flow table (%s)",
+                      ofperr_get_name(error));
+        }
+
+        list_remove(&fm->list_node);
+        free(fm->ofpacts);
+        free(fm);
+    }
+
     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
         port_run_fast(ofport);
     }
@@ -1497,7 +1514,9 @@ run(struct ofproto *ofproto_)
 
     if (mbridge_need_revalidate(ofproto->mbridge)) {
         ofproto->backer->need_revalidate = REV_RECONFIGURE;
-        mac_learning_flush(ofproto->ml, NULL);
+        ovs_rwlock_wrlock(&ofproto->ml->rwlock);
+        mac_learning_flush(ofproto->ml);
+        ovs_rwlock_unlock(&ofproto->ml->rwlock);
     }
 
     /* Do not perform any periodic activity below required by 'ofproto' while
@@ -1528,7 +1547,11 @@ run(struct ofproto *ofproto_)
     }
 
     stp_run(ofproto);
-    mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
+    ovs_rwlock_wrlock(&ofproto->ml->rwlock);
+    if (mac_learning_run(ofproto->ml)) {
+        ofproto->backer->need_revalidate = REV_MAC_LEARNING;
+    }
+    ovs_rwlock_unlock(&ofproto->ml->rwlock);
 
     /* Check the consistency of a random facet, to aid debugging. */
     if (time_msec() >= ofproto->consistency_rl
@@ -1546,11 +1569,8 @@ run(struct ofproto *ofproto_)
                           hmap_node);
         facet = CONTAINER_OF(cr, struct facet, cr);
 
-        if (!tag_set_intersects(&ofproto->backer->revalidate_set,
-                                facet->xout.tags)) {
-            if (!facet_check_consistency(facet)) {
-                ofproto->backer->need_revalidate = REV_INCONSISTENCY;
-            }
+        if (!facet_check_consistency(facet)) {
+            ofproto->backer->need_revalidate = REV_INCONSISTENCY;
         }
     }
 
@@ -1577,9 +1597,6 @@ wait(struct ofproto *ofproto_)
     if (ofproto->sflow) {
         dpif_sflow_wait(ofproto->sflow);
     }
-    if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
-        poll_immediate_wake();
-    }
     HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
         port_wait(ofport);
     }
@@ -1589,7 +1606,9 @@ wait(struct ofproto *ofproto_)
     if (ofproto->netflow) {
         netflow_wait(ofproto->netflow);
     }
+    ovs_rwlock_rdlock(&ofproto->ml->rwlock);
     mac_learning_wait(ofproto->ml);
+    ovs_rwlock_unlock(&ofproto->ml->rwlock);
     stp_wait(ofproto);
     if (ofproto->backer->need_revalidate) {
         /* Shouldn't happen, but if it does just go around again. */
@@ -1713,13 +1732,13 @@ port_construct(struct ofport *port_)
     port->bundle = NULL;
     port->cfm = NULL;
     port->bfd = NULL;
-    port->tag = tag_create_random();
     port->may_enable = true;
     port->stp_port = NULL;
     port->stp_state = STP_DISABLED;
     port->is_tunnel = false;
     port->peer = NULL;
-    hmap_init(&port->priorities);
+    port->qdscp = NULL;
+    port->n_qdscp = 0;
     port->realdev_ofp_port = 0;
     port->vlandev_vid = 0;
     port->carrier_seq = netdev_get_carrier_resets(netdev);
@@ -1813,8 +1832,7 @@ port_destruct(struct ofport *port_)
         dpif_sflow_del_port(ofproto->sflow, port->odp_port);
     }
 
-    ofport_clear_priorities(port);
-    hmap_destroy(&port->priorities);
+    free(port->qdscp);
 }
 
 static void
@@ -2042,7 +2060,7 @@ set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
             set_stp_port(ofport, NULL);
         }
 
-        stp_destroy(ofproto->stp);
+        stp_unref(ofproto->stp);
         ofproto->stp = NULL;
     }
 
@@ -2088,8 +2106,9 @@ update_stp_port_state(struct ofport_dpif *ofport)
         if (stp_learn_in_state(ofport->stp_state)
                 != stp_learn_in_state(state)) {
             /* xxx Learning action flows should also be flushed. */
-            mac_learning_flush(ofproto->ml,
-                               &ofproto->backer->revalidate_set);
+            ovs_rwlock_wrlock(&ofproto->ml->rwlock);
+            mac_learning_flush(ofproto->ml);
+            ovs_rwlock_unlock(&ofproto->ml->rwlock);
         }
         fwd_change = stp_forward_in_state(ofport->stp_state)
                         != stp_forward_in_state(state);
@@ -2194,7 +2213,9 @@ stp_run(struct ofproto_dpif *ofproto)
         }
 
         if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
-            mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
+            ovs_rwlock_wrlock(&ofproto->ml->rwlock);
+            mac_learning_flush(ofproto->ml);
+            ovs_rwlock_unlock(&ofproto->ml->rwlock);
         }
     }
 }
@@ -2206,39 +2227,6 @@ stp_wait(struct ofproto_dpif *ofproto)
         poll_timer_wait(1000);
     }
 }
-
-/* Returns true if STP should process 'flow'.  Sets fields in 'wc' that
- * were used to make the determination.*/
-bool
-stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
-{
-    memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
-    return eth_addr_equals(flow->dl_dst, eth_addr_stp);
-}
-
-void
-stp_process_packet(const struct ofport_dpif *ofport,
-                   const struct ofpbuf *packet)
-{
-    struct ofpbuf payload = *packet;
-    struct eth_header *eth = payload.data;
-    struct stp_port *sp = ofport->stp_port;
-
-    /* Sink packets on ports that have STP disabled when the bridge has
-     * STP enabled. */
-    if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
-        return;
-    }
-
-    /* Trim off padding on payload. */
-    if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
-        payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
-    }
-
-    if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
-        stp_received_bpdu(sp, payload.data, payload.size);
-    }
-}
 \f
 int
 ofproto_dpif_queue_to_priority(const struct ofproto_dpif *ofproto,
@@ -2247,88 +2235,24 @@ ofproto_dpif_queue_to_priority(const struct ofproto_dpif *ofproto,
     return dpif_queue_to_priority(ofproto->backer->dpif, queue_id, priority);
 }
 
-static struct priority_to_dscp *
-get_priority(const struct ofport_dpif *ofport, uint32_t priority)
-{
-    struct priority_to_dscp *pdscp;
-    uint32_t hash;
-
-    hash = hash_int(priority, 0);
-    HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
-        if (pdscp->priority == priority) {
-            return pdscp;
-        }
-    }
-    return NULL;
-}
-
-bool
-ofproto_dpif_dscp_from_priority(const struct ofport_dpif *ofport,
-                                uint32_t priority, uint8_t *dscp)
-{
-    struct priority_to_dscp *pdscp = get_priority(ofport, priority);
-    *dscp = pdscp ? pdscp->dscp : 0;
-    return pdscp != NULL;
-}
-
-static void
-ofport_clear_priorities(struct ofport_dpif *ofport)
-{
-    struct priority_to_dscp *pdscp, *next;
-
-    HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
-        hmap_remove(&ofport->priorities, &pdscp->hmap_node);
-        free(pdscp);
-    }
-}
-
 static int
-set_queues(struct ofport *ofport_,
-           const struct ofproto_port_queue *qdscp_list,
+set_queues(struct ofport *ofport_, const struct ofproto_port_queue *qdscp,
            size_t n_qdscp)
 {
     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
-    struct hmap new = HMAP_INITIALIZER(&new);
-    size_t i;
-
-    for (i = 0; i < n_qdscp; i++) {
-        struct priority_to_dscp *pdscp;
-        uint32_t priority;
-        uint8_t dscp;
-
-        dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
-        if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
-                                   &priority)) {
-            continue;
-        }
-
-        pdscp = get_priority(ofport, priority);
-        if (pdscp) {
-            hmap_remove(&ofport->priorities, &pdscp->hmap_node);
-        } else {
-            pdscp = xmalloc(sizeof *pdscp);
-            pdscp->priority = priority;
-            pdscp->dscp = dscp;
-            ofproto->backer->need_revalidate = REV_RECONFIGURE;
-        }
-
-        if (pdscp->dscp != dscp) {
-            pdscp->dscp = dscp;
-            ofproto->backer->need_revalidate = REV_RECONFIGURE;
-        }
-
-        hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
-    }
 
-    if (!hmap_is_empty(&ofport->priorities)) {
-        ofport_clear_priorities(ofport);
+    if (ofport->n_qdscp != n_qdscp
+        || (n_qdscp && memcmp(ofport->qdscp, qdscp,
+                              n_qdscp * sizeof *qdscp))) {
         ofproto->backer->need_revalidate = REV_RECONFIGURE;
+        free(ofport->qdscp);
+        ofport->qdscp = n_qdscp
+            ? xmemdup(qdscp, n_qdscp * sizeof *qdscp)
+            : NULL;
+        ofport->n_qdscp = n_qdscp;
     }
 
-    hmap_swap(&new, &ofport->priorities);
-    hmap_destroy(&new);
-
     return 0;
 }
 \f
@@ -2351,6 +2275,7 @@ bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
     struct mac_entry *mac, *next_mac;
 
     ofproto->backer->need_revalidate = REV_RECONFIGURE;
+    ovs_rwlock_wrlock(&ml->rwlock);
     LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
         if (mac->port.p == bundle) {
             if (all_ofprotos) {
@@ -2360,11 +2285,12 @@ bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
                     if (o != ofproto) {
                         struct mac_entry *e;
 
-                        e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
-                                                NULL);
+                        ovs_rwlock_wrlock(&o->ml->rwlock);
+                        e = mac_learning_lookup(o->ml, mac->mac, mac->vlan);
                         if (e) {
                             mac_learning_expire(o->ml, e);
                         }
+                        ovs_rwlock_unlock(&o->ml->rwlock);
                     }
                 }
             }
@@ -2372,6 +2298,7 @@ bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
             mac_learning_expire(ml, mac);
         }
     }
+    ovs_rwlock_unlock(&ml->rwlock);
 }
 
 static struct ofbundle *
@@ -2715,6 +2642,7 @@ bundle_send_learning_packets(struct ofbundle *bundle)
     struct mac_entry *e;
 
     error = n_packets = n_errors = 0;
+    ovs_rwlock_rdlock(&ofproto->ml->rwlock);
     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
         if (e->port.p != bundle) {
             struct ofpbuf *learning_packet;
@@ -2737,6 +2665,7 @@ bundle_send_learning_packets(struct ofbundle *bundle)
             n_packets++;
         }
     }
+    ovs_rwlock_unlock(&ofproto->ml->rwlock);
 
     if (n_errors) {
         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
@@ -2762,8 +2691,10 @@ bundle_run(struct ofbundle *bundle)
             bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
         }
 
-        bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
-                 lacp_status(bundle->lacp));
+        if (bond_run(bundle->bond, lacp_status(bundle->lacp))) {
+            bundle->ofproto->backer->need_revalidate = REV_BOND;
+        }
+
         if (bond_should_send_learning_packets(bundle->bond)) {
             bundle_send_learning_packets(bundle);
         }
@@ -2829,9 +2760,11 @@ static int
 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
 {
     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+    ovs_rwlock_wrlock(&ofproto->ml->rwlock);
     if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
-        mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
+        mac_learning_flush(ofproto->ml);
     }
+    ovs_rwlock_unlock(&ofproto->ml->rwlock);
     return 0;
 }
 
@@ -2855,8 +2788,10 @@ set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
                      size_t max_entries)
 {
     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+    ovs_rwlock_wrlock(&ofproto->ml->rwlock);
     mac_learning_set_idle_time(ofproto->ml, idle_time);
     mac_learning_set_max_entries(ofproto->ml, max_entries);
+    ovs_rwlock_unlock(&ofproto->ml->rwlock);
 }
 \f
 /* Ports. */
@@ -3525,7 +3460,7 @@ handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
         subfacet_update_stats(subfacet, stats);
     }
 
-    if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
+    if (subfacet->path != want_path) {
         struct flow_miss_op *op = &ops[(*n_ops)++];
         struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
 
@@ -3540,7 +3475,7 @@ handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
         op->xout_garbage = false;
         op->dpif_op.type = DPIF_OP_FLOW_PUT;
         op->subfacet = subfacet;
-        put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
+        put->flags = DPIF_FP_CREATE;
         put->key = miss->key;
         put->key_len = miss->key_len;
         put->mask = op->mask.data;
@@ -3799,15 +3734,20 @@ handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
 
             drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
             if (!drop_key) {
-                drop_key = xmalloc(sizeof *drop_key);
-                drop_key->key = xmemdup(upcall->key, upcall->key_len);
-                drop_key->key_len = upcall->key_len;
-
-                hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
-                            hash_bytes(drop_key->key, drop_key->key_len, 0));
-                dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
-                              drop_key->key, drop_key->key_len,
-                              NULL, 0, NULL, 0, NULL);
+                int ret;
+                ret = dpif_flow_put(backer->dpif,
+                                    DPIF_FP_CREATE | DPIF_FP_MODIFY,
+                                    upcall->key, upcall->key_len,
+                                    NULL, 0, NULL, 0, NULL);
+
+                if (!ret) {
+                    drop_key = xmalloc(sizeof *drop_key);
+                    drop_key->key = xmemdup(upcall->key, upcall->key_len);
+                    drop_key->key_len = upcall->key_len;
+
+                    hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
+                                hash_bytes(drop_key->key, drop_key->key_len, 0));
+                }
             }
             continue;
         }
@@ -4133,7 +4073,7 @@ expire(struct dpif_backer *backer)
 
             HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
                 if (bundle->bond) {
-                    bond_rebalance(bundle->bond, &backer->revalidate_set);
+                    bond_rebalance(bundle->bond);
                 }
             }
         }
@@ -4677,9 +4617,7 @@ facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow)
 
     facet = facet_find(ofproto, flow);
     if (facet
-        && (ofproto->backer->need_revalidate
-            || tag_set_intersects(&ofproto->backer->revalidate_set,
-                                  facet->xout.tags))
+        && ofproto->backer->need_revalidate
         && !facet_revalidate(facet)) {
         return NULL;
     }
@@ -4826,7 +4764,6 @@ facet_revalidate(struct facet *facet)
     }
 
     /* Update 'facet' now that we've taken care of all the old state. */
-    facet->xout.tags = xout.tags;
     facet->xout.slow = xout.slow;
     facet->xout.has_learn = xout.has_learn;
     facet->xout.has_normal = xout.has_normal;
@@ -5088,7 +5025,8 @@ subfacet_install(struct subfacet *subfacet, const struct ofpbuf *odp_actions,
     enum dpif_flow_put_flags flags;
     int ret;
 
-    flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
+    flags = subfacet->path == SF_NOT_INSTALLED ? DPIF_FP_CREATE
+                                               : DPIF_FP_MODIFY;
     if (stats) {
         flags |= DPIF_FP_ZERO_STATS;
     }
@@ -5262,7 +5200,7 @@ complete_operation(struct rule_dpif *rule)
 {
     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
 
-    rule_invalidate(rule);
+    ofproto->backer->need_revalidate = REV_FLOW_TABLE;
     if (clogged) {
         struct dpif_completion *c = xmalloc(sizeof *c);
         c->op = rule->up.pending;
@@ -5290,27 +5228,8 @@ static enum ofperr
 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 *victim;
-    uint8_t table_id;
-
     rule->packet_count = 0;
     rule->byte_count = 0;
-
-    table_id = rule->up.table_id;
-    victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
-    if (victim) {
-        rule->tag = victim->tag;
-    } else if (table_id == 0) {
-        rule->tag = 0;
-    } else {
-        struct flow flow;
-
-        miniflow_expand(&rule->up.cr.match.flow, &flow);
-        rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
-                                       ofproto->tables[table_id].basis);
-    }
-
     complete_operation(rule);
     return 0;
 }
@@ -5483,136 +5402,6 @@ put_userspace_action(const struct ofproto_dpif *ofproto,
 
     return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
 }
-
-tag_type
-calculate_flow_tag(struct ofproto_dpif *ofproto, const struct flow *flow,
-                   uint8_t table_id, struct rule_dpif *rule)
-{
-    if (table_id > 0 && table_id < N_TABLES) {
-        struct table_dpif *table = &ofproto->tables[table_id];
-        if (table->other_table) {
-            return (rule && rule->tag
-                    ? rule->tag
-                    : rule_calculate_tag(flow, &table->other_table->mask,
-                                         table->basis));
-        }
-    }
-
-    return 0;
-}
-\f
-/* Optimized flow revalidation.
- *
- * It's a difficult problem, in general, to tell which facets need to have
- * their actions recalculated whenever the OpenFlow flow table changes.  We
- * don't try to solve that general problem: for most kinds of OpenFlow flow
- * table changes, we recalculate the actions for every facet.  This is
- * relatively expensive, but it's good enough if the OpenFlow flow table
- * doesn't change very often.
- *
- * However, we can expect one particular kind of OpenFlow flow table change to
- * happen frequently: changes caused by MAC learning.  To avoid wasting a lot
- * of CPU on revalidating every facet whenever MAC learning modifies the flow
- * table, we add a special case that applies to flow tables in which every rule
- * has the same form (that is, the same wildcards), except that the table is
- * also allowed to have a single "catch-all" flow that matches all packets.  We
- * optimize this case by tagging all of the facets that resubmit into the table
- * and invalidating the same tag whenever a flow changes in that table.  The
- * end result is that we revalidate just the facets that need it (and sometimes
- * a few more, but not all of the facets or even all of the facets that
- * resubmit to the table modified by MAC learning). */
-
-/* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
- * into an OpenFlow table with the given 'basis'. */
-tag_type
-rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
-                   uint32_t secret)
-{
-    if (minimask_is_catchall(mask)) {
-        return 0;
-    } else {
-        uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
-        return tag_create_deterministic(hash);
-    }
-}
-
-/* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
- * taggability of that table.
- *
- * This function must be called after *each* change to a flow table.  If you
- * skip calling it on some changes then the pointer comparisons at the end can
- * be invalid if you get unlucky.  For example, if a flow removal causes a
- * cls_table to be destroyed and then a flow insertion causes a cls_table with
- * different wildcards to be created with the same address, then this function
- * will incorrectly skip revalidation. */
-static void
-table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
-{
-    struct table_dpif *table = &ofproto->tables[table_id];
-    const struct oftable *oftable = &ofproto->up.tables[table_id];
-    struct cls_table *catchall, *other;
-    struct cls_table *t;
-
-    catchall = other = NULL;
-
-    switch (hmap_count(&oftable->cls.tables)) {
-    case 0:
-        /* We could tag this OpenFlow table but it would make the logic a
-         * little harder and it's a corner case that doesn't seem worth it
-         * yet. */
-        break;
-
-    case 1:
-    case 2:
-        HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
-            if (cls_table_is_catchall(t)) {
-                catchall = t;
-            } else if (!other) {
-                other = t;
-            } else {
-                /* Indicate that we can't tag this by setting both tables to
-                 * NULL.  (We know that 'catchall' is already NULL.) */
-                other = NULL;
-            }
-        }
-        break;
-
-    default:
-        /* Can't tag this table. */
-        break;
-    }
-
-    if (table->catchall_table != catchall || table->other_table != other) {
-        table->catchall_table = catchall;
-        table->other_table = other;
-        ofproto->backer->need_revalidate = REV_FLOW_TABLE;
-    }
-}
-
-/* Given 'rule' that has changed in some way (either it is a rule being
- * inserted, a rule being deleted, or a rule whose actions are being
- * modified), marks facets for revalidation to ensure that packets will be
- * forwarded correctly according to the new state of the flow table.
- *
- * This function must be called after *each* change to a flow table.  See
- * the comment on table_update_taggable() for more information. */
-static void
-rule_invalidate(const struct rule_dpif *rule)
-{
-    struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
-
-    table_update_taggable(ofproto, rule->up.table_id);
-
-    if (!ofproto->backer->need_revalidate) {
-        struct table_dpif *table = &ofproto->tables[rule->up.table_id];
-
-        if (table->other_table && rule->tag) {
-            tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
-        } else {
-            ofproto->backer->need_revalidate = REV_FLOW_TABLE;
-        }
-    }
-}
 \f
 static bool
 set_frag_handling(struct ofproto *ofproto_,
@@ -5756,10 +5545,14 @@ ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
             unixctl_command_reply_error(conn, "no such bridge");
             return;
         }
-        mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
+        ovs_rwlock_wrlock(&ofproto->ml->rwlock);
+        mac_learning_flush(ofproto->ml);
+        ovs_rwlock_unlock(&ofproto->ml->rwlock);
     } else {
         HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
-            mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
+            ovs_rwlock_wrlock(&ofproto->ml->rwlock);
+            mac_learning_flush(ofproto->ml);
+            ovs_rwlock_unlock(&ofproto->ml->rwlock);
         }
     }
 
@@ -5788,6 +5581,7 @@ ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
     }
 
     ds_put_cstr(&ds, " port  VLAN  MAC                Age\n");
+    ovs_rwlock_rdlock(&ofproto->ml->rwlock);
     LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
         struct ofbundle *bundle = e->port.p;
         char name[OFP_MAX_PORT_NAME_LEN];
@@ -5798,6 +5592,7 @@ ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
                       name, e->vlan, ETH_ADDR_ARGS(e->mac),
                       mac_entry_age(ofproto->ml, e));
     }
+    ovs_rwlock_unlock(&ofproto->ml->rwlock);
     unixctl_command_reply(conn, ds_cstr(&ds));
     ds_destroy(&ds);
 }