mpls: Allow l3 and l4 actions to prior to a push_mpls action
[sliver-openvswitch.git] / ofproto / ofproto-dpif.c
index 7035530..6e9023a 100644 (file)
@@ -78,6 +78,7 @@ BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
 struct ofport_dpif;
 struct ofproto_dpif;
 struct flow_miss;
+struct facet;
 
 struct rule_dpif {
     struct rule up;
@@ -118,8 +119,7 @@ static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
 
 static void rule_credit_stats(struct rule_dpif *,
                               const struct dpif_flow_stats *);
-static void flow_push_stats(struct rule_dpif *, const struct flow *,
-                            const struct dpif_flow_stats *);
+static void flow_push_stats(struct facet *, const struct dpif_flow_stats *);
 static tag_type rule_calculate_tag(const struct flow *,
                                    const struct minimask *, uint32_t basis);
 static void rule_invalidate(const struct rule_dpif *);
@@ -216,6 +216,11 @@ struct action_xlate_ctx {
      * this flow when actions change header fields. */
     struct flow flow;
 
+    /* stack for the push and pop actions.
+     * Each stack element is of the type "union mf_subvalue". */
+    struct ofpbuf stack;
+    union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
+
     /* The packet corresponding to 'flow', or a null pointer if we are
      * revalidating without a packet to refer to. */
     const struct ofpbuf *packet;
@@ -285,9 +290,28 @@ struct action_xlate_ctx {
     bool exit;                  /* No further actions should be processed. */
 };
 
+/* Initial values of fields of the packet that may be changed during
+ * flow processing and needed later. */
+struct initial_vals {
+   /* This is the value of vlan_tci in the packet as actually received from
+    * dpif.  This is the same as the facet's flow.vlan_tci unless the packet
+    * was received via a VLAN splinter.  In that case, this value is 0
+    * (because the packet as actually received from the dpif had no 802.1Q
+    * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
+    * represents.
+    *
+    * This member should be removed when the VLAN splinters feature is no
+    * longer needed. */
+    ovs_be16 vlan_tci;
+
+    /* If received on a tunnel, the IP TOS value of the tunnel. */
+    uint8_t tunnel_ip_tos;
+};
+
 static void action_xlate_ctx_init(struct action_xlate_ctx *,
                                   struct ofproto_dpif *, const struct flow *,
-                                  ovs_be16 initial_tci, struct rule_dpif *,
+                                  const struct initial_vals *initial_vals,
+                                  struct rule_dpif *,
                                   uint8_t tcp_flags, const struct ofpbuf *);
 static void xlate_actions(struct action_xlate_ctx *,
                           const struct ofpact *ofpacts, size_t ofpacts_len,
@@ -360,10 +384,8 @@ struct subfacet {
     enum slow_path_reason slow; /* 0 if fast path may be used. */
     enum subfacet_path path;    /* Installed in datapath? */
 
-    /* This value is normally the same as ->facet->flow.vlan_tci.  Only VLAN
-     * splinters can cause it to differ.  This value should be removed when
-     * the VLAN splinters feature is no longer needed.  */
-    ovs_be16 initial_tci;       /* Initial VLAN TCI value. */
+    /* Initial values of the packet that may be needed later. */
+    struct initial_vals initial_vals;
 
     /* Datapath port the packet arrived on.  This is needed to remove
      * flows for ports that are no longer part of the bridge.  Since the
@@ -465,7 +487,9 @@ struct facet {
 
     /* Storage for a single subfacet, to reduce malloc() time and space
      * overhead.  (A facet always has at least one subfacet and in the common
-     * case has exactly one subfacet.) */
+     * case has exactly one subfacet.  However, 'one_subfacet' may not
+     * always be valid, since it could have been removed after newer
+     * subfacets were pushed onto the 'subfacets' list.) */
     struct subfacet one_subfacet;
 };
 
@@ -489,6 +513,8 @@ static void facet_push_stats(struct facet *);
 static void facet_learn(struct facet *);
 static void facet_account(struct facet *);
 
+static struct subfacet *facet_get_subfacet(struct facet *);
+
 static bool facet_is_controller_flow(struct facet *);
 
 struct ofport_dpif {
@@ -639,6 +665,9 @@ static void drop_key_clear(struct dpif_backer *);
 static struct ofport_dpif *
 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
 
+static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
+                                        uint64_t delta);
+
 struct ofproto_dpif {
     struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
     struct ofproto up;
@@ -688,6 +717,10 @@ struct ofproto_dpif {
     struct sset ghost_ports;       /* Ports with no datapath port. */
     struct sset port_poll_set;     /* Queued names for port_poll() reply. */
     int port_poll_errno;           /* Last errno for port_poll() reply. */
+
+    /* Per ofproto's dpif stats. */
+    uint64_t n_hit;
+    uint64_t n_missed;
 };
 
 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"?  (Useful only
@@ -711,8 +744,8 @@ static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
                                         uint32_t odp_port);
 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
-                          const struct ofpbuf *, ovs_be16 initial_tci,
-                          struct ds *);
+                          const struct ofpbuf *,
+                          const struct initial_vals *, struct ds *);
 
 /* Packet processing. */
 static void update_learning_table(struct ofproto_dpif *,
@@ -1265,6 +1298,9 @@ construct(struct ofproto *ofproto_)
     error = add_internal_flows(ofproto);
     ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
 
+    ofproto->n_hit = 0;
+    ofproto->n_missed = 0;
+
     return error;
 }
 
@@ -1778,43 +1814,22 @@ set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
     return error;
 }
 
-static int
-get_cfm_fault(const struct ofport *ofport_)
-{
-    struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
-
-    return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
-}
-
-static int
-get_cfm_opup(const struct ofport *ofport_)
-{
-    struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
-
-    return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
-}
-
-static int
-get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
-                     size_t *n_rmps)
+static bool
+get_cfm_status(const struct ofport *ofport_,
+               struct ofproto_cfm_status *status)
 {
     struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
 
     if (ofport->cfm) {
-        cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
-        return 0;
+        status->faults = cfm_get_fault(ofport->cfm);
+        status->remote_opstate = cfm_get_opup(ofport->cfm);
+        status->health = cfm_get_health(ofport->cfm);
+        cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
+        return true;
     } else {
-        return -1;
+        return false;
     }
 }
-
-static int
-get_cfm_health(const struct ofport *ofport_)
-{
-    struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
-
-    return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
-}
 \f
 /* Spanning Tree. */
 
@@ -3280,7 +3295,7 @@ struct flow_miss {
     enum odp_key_fitness key_fitness;
     const struct nlattr *key;
     size_t key_len;
-    ovs_be16 initial_tci;
+    struct initial_vals initial_vals;
     struct list packets;
     enum dpif_upcall_type upcall_type;
     uint32_t odp_in_port;
@@ -3367,7 +3382,7 @@ static void
 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
                           struct flow_miss_op *op)
 {
-    if (miss->flow.vlan_tci != miss->initial_tci) {
+    if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
         /* This packet was received on a VLAN splinter port.  We
          * added a VLAN to the packet to make the packet resemble
          * the flow, but the actions were composed assuming that
@@ -3459,8 +3474,8 @@ handle_flow_miss_without_facet(struct flow_miss *miss,
         dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
         rule_credit_stats(rule, &stats);
 
-        action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
-                              rule, 0, packet);
+        action_xlate_ctx_init(&ctx, ofproto, &miss->flow,
+                              &miss->initial_vals, rule, 0, packet);
         ctx.resubmit_stats = &stats;
         xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
                       &odp_actions);
@@ -3650,10 +3665,12 @@ drop_key_clear(struct dpif_backer *backer)
  * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
  * a VLAN header onto 'packet' (if it is nonnull).
  *
- * Optionally, if nonnull, sets '*initial_tci' to the VLAN TCI with which the
- * packet was really received, that is, the actual VLAN TCI extracted by
- * odp_flow_key_to_flow().  (This differs from the value returned in
- * flow->vlan_tci only for packets received on VLAN splinters.)
+ * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
+ * to the VLAN TCI with which the packet was really received, that is, the
+ * actual VLAN TCI extracted by odp_flow_key_to_flow().  (This differs from
+ * the value returned in flow->vlan_tci only for packets received on
+ * VLAN splinters.)  Also, if received on an IP tunnel, sets
+ * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
  *
  * Similarly, this function also includes some logic to help with tunnels.  It
  * may modify 'flow' as necessary to make the tunneling implementation
@@ -3666,7 +3683,7 @@ ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
                 const struct nlattr *key, size_t key_len,
                 struct flow *flow, enum odp_key_fitness *fitnessp,
                 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
-                ovs_be16 *initial_tci)
+                struct initial_vals *initial_vals)
 {
     const struct ofport_dpif *port;
     enum odp_key_fitness fitness;
@@ -3678,8 +3695,9 @@ ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
         goto exit;
     }
 
-    if (initial_tci) {
-        *initial_tci = flow->vlan_tci;
+    if (initial_vals) {
+        initial_vals->vlan_tci = flow->vlan_tci;
+        initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
     }
 
     if (odp_in_port) {
@@ -3783,7 +3801,7 @@ handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
 
         error = ofproto_receive(backer, upcall->packet, upcall->key,
                                 upcall->key_len, &flow, &miss->key_fitness,
-                                &ofproto, &odp_in_port, &miss->initial_tci);
+                                &ofproto, &odp_in_port, &miss->initial_vals);
         if (error == ENODEV) {
             struct drop_key *drop_key;
 
@@ -3812,6 +3830,8 @@ handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
         if (error) {
             continue;
         }
+
+        ofproto->n_missed++;
         flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
                      &flow.tunnel, flow.in_port, &miss->flow);
 
@@ -4099,6 +4119,11 @@ delete_unexpected_flow(struct ofproto_dpif *ofproto,
  * 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.
+ *
+ * In addition, this function maintains per ofproto flow hit counts. The patch
+ * port is not treated specially. e.g. A packet ingress from br0 patched into
+ * br1 will increase the hit count of br0 by 1, however, does not affect
+ * the hit or miss counts of br1.
  */
 static void
 update_stats(struct dpif_backer *backer)
@@ -4130,6 +4155,12 @@ update_stats(struct dpif_backer *backer)
         subfacet = subfacet_find(ofproto, key, key_len, key_hash);
         switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
         case SF_FAST_PATH:
+            /* Update ofproto_dpif's hit count. */
+            if (stats->n_packets > subfacet->dp_packet_count) {
+                uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
+                dpif_stats_update_hit_count(ofproto, delta);
+            }
+
             update_subfacet_stats(subfacet, stats);
             break;
 
@@ -4412,6 +4443,8 @@ static void
 facet_learn(struct facet *facet)
 {
     struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
+    struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets),
+                                            struct subfacet, list_node);
     struct action_xlate_ctx ctx;
 
     if (!facet->has_learn
@@ -4422,7 +4455,7 @@ facet_learn(struct facet *facet)
     }
 
     action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
-                          facet->flow.vlan_tci,
+                          &subfacet->initial_vals,
                           facet->rule, facet->tcp_flags, NULL);
     ctx.may_learn = true;
     xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
@@ -4433,7 +4466,7 @@ static void
 facet_account(struct facet *facet)
 {
     struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
-    struct subfacet *subfacet;
+    struct subfacet *subfacet = facet_get_subfacet(facet);
     const struct nlattr *a;
     unsigned int left;
     ovs_be16 vlan_tci;
@@ -4452,8 +4485,6 @@ facet_account(struct facet *facet)
      *
      * We use the actions from an arbitrary subfacet because they should all
      * be equally valid for our purpose. */
-    subfacet = CONTAINER_OF(list_front(&facet->subfacets),
-                            struct subfacet, list_node);
     vlan_tci = facet->flow.vlan_tci;
     NL_ATTR_FOR_EACH_UNSAFE (a, left,
                              subfacet->actions, subfacet->actions_len) {
@@ -4590,6 +4621,14 @@ facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
     return facet;
 }
 
+/* Return a subfacet from 'facet'.  A facet consists of one or more
+ * subfacets, and this function returns one of them. */
+static struct subfacet *facet_get_subfacet(struct facet *facet)
+{
+    return CONTAINER_OF(list_front(&facet->subfacets), struct subfacet,
+                        list_node);
+}
+
 static const char *
 subfacet_path_to_string(enum subfacet_path path)
 {
@@ -4674,7 +4713,7 @@ facet_check_consistency(struct facet *facet)
         struct ds s;
 
         action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
-                              subfacet->initial_tci, rule, 0, NULL);
+                              &subfacet->initial_vals, rule, 0, NULL);
         xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
                       &odp_actions);
 
@@ -4805,7 +4844,7 @@ facet_revalidate(struct facet *facet)
         enum slow_path_reason slow;
 
         action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
-                              subfacet->initial_tci, new_rule, 0, NULL);
+                              &subfacet->initial_vals, new_rule, 0, NULL);
         xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
                       &odp_actions);
 
@@ -4907,7 +4946,7 @@ facet_push_stats(struct facet *facet)
         facet->prev_byte_count = facet->byte_count;
         facet->prev_used = facet->used;
 
-        flow_push_stats(facet->rule, &facet->flow, &stats);
+        flow_push_stats(facet, &stats);
 
         update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
                             facet->mirrors, stats.n_packets, stats.n_bytes);
@@ -4922,19 +4961,20 @@ rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
     ofproto_rule_update_used(&rule->up, stats->used);
 }
 
-/* Pushes flow statistics to the rules which 'flow' resubmits into given
- * 'rule''s actions and mirrors. */
+/* Pushes flow statistics to the rules which 'facet->flow' resubmits
+ * into given 'facet->rule''s actions and mirrors. */
 static void
-flow_push_stats(struct rule_dpif *rule,
-                const struct flow *flow, const struct dpif_flow_stats *stats)
+flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
 {
+    struct rule_dpif *rule = facet->rule;
     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
+    struct subfacet *subfacet = facet_get_subfacet(facet);
     struct action_xlate_ctx ctx;
 
     ofproto_rule_update_used(&rule->up, stats->used);
 
-    action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
-                          0, NULL);
+    action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
+                          &subfacet->initial_vals, rule, 0, NULL);
     ctx.resubmit_stats = stats;
     xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
                                    rule->up.ofpacts_len);
@@ -5012,7 +5052,7 @@ subfacet_create(struct facet *facet, struct flow_miss *miss,
                       ? SLOW_MATCH
                       : 0);
     subfacet->path = SF_NOT_INSTALLED;
-    subfacet->initial_tci = miss->initial_tci;
+    subfacet->initial_vals = miss->initial_vals;
     subfacet->odp_in_port = miss->odp_in_port;
 
     return subfacet;
@@ -5089,8 +5129,8 @@ subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
 
     struct action_xlate_ctx ctx;
 
-    action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
-                          rule, 0, packet);
+    action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
+                          &subfacet->initial_vals, rule, 0, packet);
     xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
     facet->tags = ctx.tags;
     facet->has_learn = ctx.has_learn;
@@ -5408,9 +5448,8 @@ rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
                   struct ofpbuf *packet)
 {
     struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
-
+    struct initial_vals initial_vals;
     struct dpif_flow_stats stats;
-
     struct action_xlate_ctx ctx;
     uint64_t odp_actions_stub[1024 / 8];
     struct ofpbuf odp_actions;
@@ -5418,8 +5457,10 @@ rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
     rule_credit_stats(rule, &stats);
 
+    initial_vals.vlan_tci = flow->vlan_tci;
+    initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
     ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
-    action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
+    action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
                           rule, stats.tcp_flags, packet);
     ctx.resubmit_stats = &stats;
     xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
@@ -5695,7 +5736,7 @@ compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
 
     /* If 'struct flow' gets additional metadata, we'll need to zero it out
      * before traversing a patch port. */
-    BUILD_ASSERT_DECL(FLOW_WC_SEQ == 19);
+    BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
 
     if (!ofport) {
         xlate_report(ctx, "Nonexistent output port");
@@ -5768,7 +5809,6 @@ compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
         ctx->flow.nw_tos |= pdscp->dscp;
     }
 
-    odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
     if (ofport->tnl_port) {
         odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
         if (odp_port == OVSP_NONE) {
@@ -5783,11 +5823,13 @@ compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
         commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
                                  ctx->odp_actions);
     } else {
+        odp_port = ofport->odp_port;
         out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
                                           ctx->flow.vlan_tci);
         if (out_port != odp_port) {
             ctx->flow.vlan_tci = htons(0);
         }
+        ctx->flow.skb_mark &= ~IPSEC_MARK;
     }
     commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
     nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
@@ -5806,62 +5848,74 @@ compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
     compose_output_action__(ctx, ofp_port, true);
 }
 
+static void
+tag_the_flow(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
+{
+    struct ofproto_dpif *ofproto = ctx->ofproto;
+    uint8_t table_id = ctx->table_id;
+
+    if (table_id > 0 && table_id < N_TABLES) {
+        struct table_dpif *table = &ofproto->tables[table_id];
+        if (table->other_table) {
+            ctx->tags |= (rule && rule->tag
+                          ? rule->tag
+                          : rule_calculate_tag(&ctx->flow,
+                                               &table->other_table->mask,
+                                               table->basis));
+        }
+    }
+}
+
+/* Common rule processing in one place to avoid duplicating code. */
+static struct rule_dpif *
+ctx_rule_hooks(struct action_xlate_ctx *ctx, struct rule_dpif *rule,
+               bool may_packet_in)
+{
+    if (ctx->resubmit_hook) {
+        ctx->resubmit_hook(ctx, rule);
+    }
+    if (rule == NULL && may_packet_in) {
+        /* XXX
+         * check if table configuration flags
+         * OFPTC_TABLE_MISS_CONTROLLER, default.
+         * OFPTC_TABLE_MISS_CONTINUE,
+         * OFPTC_TABLE_MISS_DROP
+         * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
+         */
+        rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->flow);
+    }
+    if (rule && ctx->resubmit_stats) {
+        rule_credit_stats(rule, ctx->resubmit_stats);
+    }
+    return rule;
+}
+
 static void
 xlate_table_action(struct action_xlate_ctx *ctx,
                    uint16_t in_port, uint8_t table_id, bool may_packet_in)
 {
     if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
-        struct ofproto_dpif *ofproto = ctx->ofproto;
         struct rule_dpif *rule;
-        uint16_t old_in_port;
-        uint8_t old_table_id;
+        uint16_t old_in_port = ctx->flow.in_port;
+        uint8_t old_table_id = ctx->table_id;
 
-        old_table_id = ctx->table_id;
         ctx->table_id = table_id;
 
         /* Look up a flow with 'in_port' as the input port. */
-        old_in_port = ctx->flow.in_port;
         ctx->flow.in_port = in_port;
-        rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
-
-        /* Tag the flow. */
-        if (table_id > 0 && table_id < N_TABLES) {
-            struct table_dpif *table = &ofproto->tables[table_id];
-            if (table->other_table) {
-                ctx->tags |= (rule && rule->tag
-                              ? rule->tag
-                              : rule_calculate_tag(&ctx->flow,
-                                                   &table->other_table->mask,
-                                                   table->basis));
-            }
-        }
+        rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, table_id);
+
+        tag_the_flow(ctx, rule);
 
         /* Restore the original input port.  Otherwise OFPP_NORMAL and
          * OFPP_IN_PORT will have surprising behavior. */
         ctx->flow.in_port = old_in_port;
 
-        if (ctx->resubmit_hook) {
-            ctx->resubmit_hook(ctx, rule);
-        }
-
-        if (rule == NULL && may_packet_in) {
-            /* XXX
-             * check if table configuration flags
-             * OFPTC_TABLE_MISS_CONTROLLER, default.
-             * OFPTC_TABLE_MISS_CONTINUE,
-             * OFPTC_TABLE_MISS_DROP
-             * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
-             */
-            rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
-        }
+        rule = ctx_rule_hooks(ctx, rule, may_packet_in);
 
         if (rule) {
             struct rule_dpif *old_rule = ctx->rule;
 
-            if (ctx->resubmit_stats) {
-                rule_credit_stats(rule, ctx->resubmit_stats);
-            }
-
             ctx->recurse++;
             ctx->rule = rule;
             do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
@@ -6012,7 +6066,6 @@ execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
         tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
         ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
         ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
-        ctx->flow.encap_dl_type = ctx->flow.dl_type;
         ctx->flow.mpls_depth = 1;
     }
     ctx->flow.dl_type = eth_type;
@@ -6029,7 +6082,6 @@ execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
         ctx->flow.mpls_lse = htonl(0);
         if (!ctx->flow.mpls_depth) {
             ctx->flow.dl_type = eth_type;
-            ctx->flow.encap_dl_type = htons(0);
         }
     }
 }
@@ -6058,6 +6110,38 @@ compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
     }
 }
 
+static bool
+execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl)
+{
+    if (!eth_type_mpls(ctx->flow.dl_type)) {
+        return true;
+    }
+
+    set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
+    return false;
+}
+
+static bool
+execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx)
+{
+    uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse);
+
+    if (!eth_type_mpls(ctx->flow.dl_type)) {
+        return false;
+    }
+
+    if (ttl > 1) {
+        ttl--;
+        set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
+        return false;
+    } else {
+        execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
+
+        /* Stop processing for current table. */
+        return true;
+    }
+}
+
 static void
 xlate_output_action(struct action_xlate_ctx *ctx,
                     uint16_t port, uint16_t max_len, bool may_packet_in)
@@ -6276,6 +6360,24 @@ may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
     return true;
 }
 
+static bool
+tunnel_ecn_ok(struct action_xlate_ctx *ctx)
+{
+    if (is_ip_any(&ctx->base_flow)
+        && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
+        if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
+            VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
+                         " but is not ECN capable");
+            return false;
+        } else {
+            /* Set the ECN CE value in the tunneled packet. */
+            ctx->flow.nw_tos |= IP_ECN_CE;
+        }
+    }
+
+    return true;
+}
+
 static void
 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
                  struct action_xlate_ctx *ctx)
@@ -6288,6 +6390,8 @@ do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
         was_evictable = ctx->rule->up.evictable;
         ctx->rule->up.evictable = false;
     }
+
+ do_xlate_actions_again:
     OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
         struct ofpact_controller *controller;
         const struct ofpact_metadata *metadata;
@@ -6346,11 +6450,15 @@ do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
             break;
 
         case OFPACT_SET_IPV4_SRC:
-            ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
+            if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
+                ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
+            }
             break;
 
         case OFPACT_SET_IPV4_DST:
-            ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
+            if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
+                ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
+            }
             break;
 
         case OFPACT_SET_IPV4_DSCP:
@@ -6362,11 +6470,15 @@ do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
             break;
 
         case OFPACT_SET_L4_SRC_PORT:
-            ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
+            if (is_ip_any(&ctx->flow)) {
+                ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
+            }
             break;
 
         case OFPACT_SET_L4_DST_PORT:
-            ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
+            if (is_ip_any(&ctx->flow)) {
+                ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
+            }
             break;
 
         case OFPACT_RESUBMIT:
@@ -6393,6 +6505,16 @@ do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
             nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
             break;
 
+        case OFPACT_STACK_PUSH:
+            nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow,
+                                   &ctx->stack);
+            break;
+
+        case OFPACT_STACK_POP:
+            nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow,
+                                  &ctx->stack);
+            break;
+
         case OFPACT_PUSH_MPLS:
             execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
             break;
@@ -6401,6 +6523,18 @@ do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
             execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
             break;
 
+        case OFPACT_SET_MPLS_TTL:
+            if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) {
+                goto out;
+            }
+            break;
+
+        case OFPACT_DEC_MPLS_TTL:
+            if (execute_dec_mpls_ttl_action(ctx)) {
+                goto out;
+            }
+            break;
+
         case OFPACT_DEC_TTL:
             if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
                 goto out;
@@ -6455,11 +6589,34 @@ do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
             break;
 
         case OFPACT_GOTO_TABLE: {
-            /* XXX remove recursion */
-            /* It is assumed that goto-table is last action */
+            /* It is assumed that goto-table is the last action. */
             struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
+            struct rule_dpif *rule;
+
             ovs_assert(ctx->table_id < ogt->table_id);
-            xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
+
+            ctx->table_id = ogt->table_id;
+
+            /* Look up a flow from the new table. */
+            rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, ctx->table_id);
+
+            tag_the_flow(ctx, rule);
+
+            rule = ctx_rule_hooks(ctx, rule, true);
+
+            if (rule) {
+                if (ctx->rule) {
+                    ctx->rule->up.evictable = was_evictable;
+                }
+                ctx->rule = rule;
+                was_evictable = rule->up.evictable;
+                rule->up.evictable = false;
+
+                /* Tail recursion removal. */
+                ofpacts = rule->up.ofpacts;
+                ofpacts_len = rule->up.ofpacts_len;
+                goto do_xlate_actions_again;
+            }
             break;
         }
         }
@@ -6474,7 +6631,8 @@ out:
 static void
 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
                       struct ofproto_dpif *ofproto, const struct flow *flow,
-                      ovs_be16 initial_tci, struct rule_dpif *rule,
+                      const struct initial_vals *initial_vals,
+                      struct rule_dpif *rule,
                       uint8_t tcp_flags, const struct ofpbuf *packet)
 {
     ovs_be64 initial_tun_id = flow->tunnel.tun_id;
@@ -6504,7 +6662,8 @@ action_xlate_ctx_init(struct action_xlate_ctx *ctx,
     ctx->flow = *flow;
     memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
     ctx->base_flow = ctx->flow;
-    ctx->base_flow.vlan_tci = initial_tci;
+    ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
+    ctx->base_flow.tunnel.ip_tos = initial_vals->tunnel_ip_tos;
     ctx->flow.tunnel.tun_id = initial_tun_id;
     ctx->rule = rule;
     ctx->packet = packet;
@@ -6550,6 +6709,8 @@ xlate_actions(struct action_xlate_ctx *ctx,
     ctx->table_id = 0;
     ctx->exit = false;
 
+    ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack);
+
     if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
         /* Do this conditionally because the copy is expensive enough that it
          * shows up in profiles. */
@@ -6585,12 +6746,15 @@ xlate_actions(struct action_xlate_ctx *ctx,
         ctx->slow |= special;
     } else {
         static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
-        ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
+        struct initial_vals initial_vals;
         uint32_t local_odp_port;
 
+        initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
+        initial_vals.tunnel_ip_tos = ctx->base_flow.tunnel.ip_tos;
+
         add_sflow_action(ctx);
 
-        if (!in_port || may_receive(in_port, ctx)) {
+        if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) {
             do_xlate_actions(ofpacts, ofpacts_len, ctx);
 
             /* We've let OFPP_NORMAL and the learning action look at the
@@ -6610,7 +6774,7 @@ xlate_actions(struct action_xlate_ctx *ctx,
                 struct ds ds = DS_EMPTY_INITIALIZER;
 
                 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
-                              initial_tci, &ds);
+                              &initial_vals, &ds);
                 VLOG_ERR("Trace triggered by excessive resubmit "
                          "recursion:\n%s", ds_cstr(&ds));
                 ds_destroy(&ds);
@@ -6634,6 +6798,8 @@ xlate_actions(struct action_xlate_ctx *ctx,
         }
         fix_sflow_action(ctx);
     }
+
+    ofpbuf_uninit(&ctx->stack);
 }
 
 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
@@ -7330,6 +7496,7 @@ packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
            const struct ofpact *ofpacts, size_t ofpacts_len)
 {
     struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
+    struct initial_vals initial_vals;
     struct odputil_keybuf keybuf;
     struct dpif_flow_stats stats;
 
@@ -7345,7 +7512,9 @@ packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
 
     dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
 
-    action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
+    initial_vals.vlan_tci = flow->vlan_tci;
+    initial_vals.tunnel_ip_tos = 0;
+    action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
                           packet_get_tcp_flags(packet, flow), packet);
     ctx.resubmit_stats = &stats;
 
@@ -7585,7 +7754,7 @@ ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
     struct ofproto_dpif *ofproto;
     struct ofpbuf odp_key;
     struct ofpbuf *packet;
-    ovs_be16 initial_tci;
+    struct initial_vals initial_vals;
     struct ds result;
     struct flow flow;
     char *s;
@@ -7626,16 +7795,16 @@ ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
                 goto exit;
             }
 
-            /* XXX: Since we allow the user to specify an ofproto, it's
-             * possible they will specify a different ofproto than the one the
-             * port actually belongs too.  Ideally we should simply remove the
-             * ability to specify the ofproto. */
+            /* The user might have specified the wrong ofproto but within the
+             * same backer.  That's OK, ofproto_receive() can find the right
+             * one for us. */
             if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
-                                odp_key.size, &flow, NULL, NULL, NULL,
-                                &initial_tci)) {
+                                odp_key.size, &flow, NULL, &ofproto, NULL,
+                                &initial_vals)) {
                 unixctl_command_reply_error(conn, "Invalid flow");
                 goto exit;
             }
+            ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
         } else {
             char *error_s;
 
@@ -7646,7 +7815,8 @@ ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
                 goto exit;
             }
 
-            initial_tci = flow.vlan_tci;
+            initial_vals.vlan_tci = flow.vlan_tci;
+            initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
         }
 
         /* Generate a packet, if requested. */
@@ -7680,13 +7850,14 @@ ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
 
         flow_extract(packet, priority, mark, NULL, in_port, &flow);
         flow.tunnel.tun_id = tun_id;
-        initial_tci = flow.vlan_tci;
+        initial_vals.vlan_tci = flow.vlan_tci;
+        initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
     } else {
         unixctl_command_reply_error(conn, "Bad command syntax");
         goto exit;
     }
 
-    ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
+    ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
     unixctl_command_reply(conn, ds_cstr(&result));
 
 exit:
@@ -7697,8 +7868,8 @@ exit:
 
 static void
 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
-              const struct ofpbuf *packet, ovs_be16 initial_tci,
-              struct ds *ds)
+              const struct ofpbuf *packet,
+              const struct initial_vals *initial_vals, struct ds *ds)
 {
     struct rule_dpif *rule;
 
@@ -7728,7 +7899,7 @@ ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
         trace.flow = *flow;
         ofpbuf_use_stub(&odp_actions,
                         odp_actions_stub, sizeof odp_actions_stub);
-        action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
+        action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
                               rule, tcp_flags, packet);
         trace.ctx.resubmit_hook = trace_resubmit;
         trace.ctx.report_hook = trace_report;
@@ -7899,19 +8070,14 @@ ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
 static void
 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
 {
-    struct dpif_dp_stats s;
     const struct shash_node **ports;
     int i;
 
-    dpif_get_dp_stats(ofproto->backer->dpif, &s);
-
     ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
                   dpif_name(ofproto->backer->dpif));
-    /* xxx It would be better to show bridge-specific stats instead
-     * xxx of dp ones. */
     ds_put_format(ds,
-                  "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
-                  s.n_hit, s.n_missed, s.n_lost);
+                  "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
+                  ofproto->n_hit, ofproto->n_missed);
     ds_put_format(ds, "\tflows: %zu\n",
                   hmap_count(&ofproto->subfacets));
 
@@ -8040,7 +8206,18 @@ ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
         }
 
         ds_put_cstr(&ds, ", actions:");
-        format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
+        if (subfacet->slow) {
+            uint64_t slow_path_stub[128 / 8];
+            const struct nlattr *actions;
+            size_t actions_len;
+
+            compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow,
+                              slow_path_stub, sizeof slow_path_stub,
+                              &actions, &actions_len);
+            format_odp_actions(&ds, actions, actions_len);
+        } else {
+            format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
+        }
         ds_put_char(&ds, '\n');
     }
 
@@ -8320,6 +8497,12 @@ odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
     }
 }
 
+static void
+dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
+{
+    ofproto->n_hit += delta;
+}
+
 const struct ofproto_class ofproto_dpif_class = {
     init,
     enumerate_types,
@@ -8370,10 +8553,7 @@ const struct ofproto_class ofproto_dpif_class = {
     get_netflow_ids,
     set_sflow,
     set_cfm,
-    get_cfm_fault,
-    get_cfm_opup,
-    get_cfm_remote_mpids,
-    get_cfm_health,
+    get_cfm_status,
     set_stp,
     get_stp_status,
     set_stp_port,