bfd: Implement Bidirectional Forwarding Detection.
[sliver-openvswitch.git] / ofproto / ofproto.c
index 42a5baf..262493e 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc.
+ * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
  * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
@@ -407,11 +407,11 @@ ofproto_create(const char *datapath_name, const char *datapath_type,
                                         OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT);
     ofproto->forward_bpdu = false;
     ofproto->fallback_dpid = pick_fallback_dpid();
-    ofproto->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
-    ofproto->hw_desc = xstrdup(DEFAULT_HW_DESC);
-    ofproto->sw_desc = xstrdup(DEFAULT_SW_DESC);
-    ofproto->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
-    ofproto->dp_desc = xstrdup(DEFAULT_DP_DESC);
+    ofproto->mfr_desc = NULL;
+    ofproto->hw_desc = NULL;
+    ofproto->sw_desc = NULL;
+    ofproto->serial_desc = NULL;
+    ofproto->dp_desc = NULL;
     ofproto->frag_handling = OFPC_FRAG_NORMAL;
     hmap_init(&ofproto->ports);
     shash_init(&ofproto->port_by_name);
@@ -419,6 +419,7 @@ ofproto_create(const char *datapath_name, const char *datapath_type,
     ofproto->max_ports = OFPP_MAX;
     ofproto->tables = NULL;
     ofproto->n_tables = 0;
+    list_init(&ofproto->expirable);
     ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
     ofproto->state = S_OPENFLOW;
     list_init(&ofproto->pending);
@@ -446,12 +447,12 @@ ofproto_create(const char *datapath_name, const char *datapath_type,
     bitmap_set1(ofproto->ofp_port_ids, 0);
 
     /* Check that hidden tables, if any, are at the end. */
-    assert(ofproto->n_tables);
+    ovs_assert(ofproto->n_tables);
     for (i = 0; i + 1 < ofproto->n_tables; i++) {
         enum oftable_flags flags = ofproto->tables[i].flags;
         enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
 
-        assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
+        ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
     }
 
     ofproto->datapath_id = pick_datapath_id(ofproto);
@@ -469,8 +470,8 @@ ofproto_init_tables(struct ofproto *ofproto, int n_tables)
 {
     struct oftable *table;
 
-    assert(!ofproto->n_tables);
-    assert(n_tables >= 1 && n_tables <= 255);
+    ovs_assert(!ofproto->n_tables);
+    ovs_assert(n_tables >= 1 && n_tables <= 255);
 
     ofproto->n_tables = n_tables;
     ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
@@ -493,7 +494,7 @@ ofproto_init_tables(struct ofproto *ofproto, int n_tables)
 void
 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
 {
-    assert(max_ports <= OFPP_MAX);
+    ovs_assert(max_ports <= OFPP_MAX);
     ofproto->max_ports = max_ports;
 }
 
@@ -598,53 +599,10 @@ ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
 }
 
 void
-ofproto_set_desc(struct ofproto *p,
-                 const char *mfr_desc, const char *hw_desc,
-                 const char *sw_desc, const char *serial_desc,
-                 const char *dp_desc)
+ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
 {
-    struct ofp_desc_stats *ods;
-
-    if (mfr_desc) {
-        if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
-            VLOG_WARN("%s: truncating mfr_desc, must be less than %zu bytes",
-                      p->name, sizeof ods->mfr_desc);
-        }
-        free(p->mfr_desc);
-        p->mfr_desc = xstrdup(mfr_desc);
-    }
-    if (hw_desc) {
-        if (strlen(hw_desc) >= sizeof ods->hw_desc) {
-            VLOG_WARN("%s: truncating hw_desc, must be less than %zu bytes",
-                      p->name, sizeof ods->hw_desc);
-        }
-        free(p->hw_desc);
-        p->hw_desc = xstrdup(hw_desc);
-    }
-    if (sw_desc) {
-        if (strlen(sw_desc) >= sizeof ods->sw_desc) {
-            VLOG_WARN("%s: truncating sw_desc, must be less than %zu bytes",
-                      p->name, sizeof ods->sw_desc);
-        }
-        free(p->sw_desc);
-        p->sw_desc = xstrdup(sw_desc);
-    }
-    if (serial_desc) {
-        if (strlen(serial_desc) >= sizeof ods->serial_num) {
-            VLOG_WARN("%s: truncating serial_desc, must be less than %zu "
-                      "bytes", p->name, sizeof ods->serial_num);
-        }
-        free(p->serial_desc);
-        p->serial_desc = xstrdup(serial_desc);
-    }
-    if (dp_desc) {
-        if (strlen(dp_desc) >= sizeof ods->dp_desc) {
-            VLOG_WARN("%s: truncating dp_desc, must be less than %zu bytes",
-                      p->name, sizeof ods->dp_desc);
-        }
-        free(p->dp_desc);
-        p->dp_desc = xstrdup(dp_desc);
-    }
+    free(p->dp_desc);
+    p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
 }
 
 int
@@ -682,6 +640,19 @@ ofproto_set_sflow(struct ofproto *ofproto,
         return oso ? EOPNOTSUPP : 0;
     }
 }
+
+int
+ofproto_set_ipfix(struct ofproto *ofproto,
+                  const struct ofproto_ipfix_bridge_exporter_options *bo,
+                  const struct ofproto_ipfix_flow_exporter_options *fo,
+                  size_t n_fo)
+{
+    if (ofproto->ofproto_class->set_ipfix) {
+        return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
+    } else {
+        return (bo || fo) ? EOPNOTSUPP : 0;
+    }
+}
 \f
 /* Spanning Tree Protocol (STP) configuration. */
 
@@ -828,6 +799,45 @@ ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
     }
 }
 
+/* Configures BFD on 'ofp_port' in 'ofproto'.  This function has no effect if
+ * 'ofproto' does not have a port 'ofp_port'. */
+void
+ofproto_port_set_bfd(struct ofproto *ofproto, uint16_t ofp_port,
+                     const struct smap *cfg)
+{
+    struct ofport *ofport;
+    int error;
+
+    ofport = ofproto_get_port(ofproto, ofp_port);
+    if (!ofport) {
+        VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
+                  ofproto->name, ofp_port);
+    }
+
+    error = (ofproto->ofproto_class->set_bfd
+             ? ofproto->ofproto_class->set_bfd(ofport, cfg)
+             : EOPNOTSUPP);
+    if (error) {
+        VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
+                  ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
+                  strerror(error));
+    }
+}
+
+/* Populates 'status' with key value pairs indicating the status of the BFD
+ * session on 'ofp_port'.  This information is intended to be populated in the
+ * OVS database.  Has no effect if 'ofp_port' is not na OpenFlow port in
+ * 'ofproto'. */
+int
+ofproto_port_get_bfd_status(struct ofproto *ofproto, uint16_t ofp_port,
+                            struct smap *status)
+{
+    struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
+    return (ofport && ofproto->ofproto_class->get_bfd_status
+            ? ofproto->ofproto_class->get_bfd_status(ofport, status)
+            : EOPNOTSUPP);
+}
+
 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
  * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
  * 0 if LACP partner information is not current (generally indicating a
@@ -953,7 +963,7 @@ ofproto_configure_table(struct ofproto *ofproto, int table_id,
 {
     struct oftable *table;
 
-    assert(table_id >= 0 && table_id < ofproto->n_tables);
+    ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
     table = &ofproto->tables[table_id];
 
     oftable_set_name(table, s->name);
@@ -1035,8 +1045,8 @@ ofproto_destroy__(struct ofproto *ofproto)
 {
     struct oftable *table;
 
-    assert(list_is_empty(&ofproto->pending));
-    assert(!ofproto->n_pending);
+    ovs_assert(list_is_empty(&ofproto->pending));
+    ovs_assert(!ofproto->n_pending);
 
     connmgr_destroy(ofproto->connmgr);
 
@@ -1656,38 +1666,30 @@ static uint16_t
 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
 {
     uint16_t ofp_port;
+    uint16_t end_port_no = ofproto->alloc_port_no;
 
     ofp_port = simap_get(&ofproto->ofp_requests, netdev_name);
     ofp_port = ofp_port ? ofp_port : OFPP_NONE;
 
     if (ofp_port >= ofproto->max_ports
             || bitmap_is_set(ofproto->ofp_port_ids, ofp_port)) {
-        bool retry = ofproto->alloc_port_no ? true : false;
-
         /* Search for a free OpenFlow port number.  We try not to
          * immediately reuse them to prevent problems due to old
          * flows. */
-        while (ofp_port >= ofproto->max_ports) {
-            for (ofproto->alloc_port_no++;
-                 ofproto->alloc_port_no < ofproto->max_ports;
-                 ofproto->alloc_port_no++) {
-                if (!bitmap_is_set(ofproto->ofp_port_ids,
-                                   ofproto->alloc_port_no)) {
-                    ofp_port = ofproto->alloc_port_no;
-                    break;
-                }
+        for (;;) {
+            if (++ofproto->alloc_port_no >= ofproto->max_ports) {
+                ofproto->alloc_port_no = 0;
             }
-            if (ofproto->alloc_port_no >= ofproto->max_ports) {
-                if (retry) {
-                    ofproto->alloc_port_no = 0;
-                    retry = false;
-                } else {
-                    return OFPP_NONE;
-                }
+            if (!bitmap_is_set(ofproto->ofp_port_ids,
+                               ofproto->alloc_port_no)) {
+                ofp_port = ofproto->alloc_port_no;
+                break;
+            }
+            if (ofproto->alloc_port_no == end_port_no) {
+                return OFPP_NONE;
             }
         }
     }
-
     bitmap_set1(ofproto->ofp_port_ids, ofp_port);
     return ofp_port;
 }
@@ -1695,7 +1697,9 @@ alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
 static void
 dealloc_ofp_port(const struct ofproto *ofproto, uint16_t ofp_port)
 {
-    bitmap_set0(ofproto->ofp_port_ids, ofp_port);
+    if (ofp_port < ofproto->max_ports) {
+        bitmap_set0(ofproto->ofp_port_ids, ofp_port);
+    }
 }
 
 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
@@ -2123,7 +2127,7 @@ ofproto_rule_destroy__(struct rule *rule)
 void
 ofproto_rule_destroy(struct rule *rule)
 {
-    assert(!rule->pending);
+    ovs_assert(!rule->pending);
     oftable_remove_rule(rule);
     ofproto_rule_destroy__(rule);
 }
@@ -2173,7 +2177,7 @@ rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
 {
     struct flow flow;
 
-    assert(ofpbuf_headroom(packet) >= sizeof(struct ofp10_packet_in));
+    ovs_assert(ofpbuf_headroom(packet) >= sizeof(struct ofp10_packet_in));
 
     flow_extract(packet, 0, 0, NULL, in_port, &flow);
     return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
@@ -2222,7 +2226,7 @@ handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
 
     ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
                                          &features.actions);
-    assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
+    ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
 
     /* Count only non-hidden tables in the number of tables.  (Hidden tables,
      * if present, are always at the end.) */
@@ -2286,11 +2290,11 @@ handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
     uint16_t flags = ntohs(osc->flags);
 
     if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
-        || ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
+        || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
         enum ofp_config_flags cur = ofproto->frag_handling;
         enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
 
-        assert((cur & OFPC_FRAG_MASK) == cur);
+        ovs_assert((cur & OFPC_FRAG_MASK) == cur);
         if (cur != next) {
             if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
                 ofproto->frag_handling = next;
@@ -2320,7 +2324,7 @@ static enum ofperr
 reject_slave_controller(struct ofconn *ofconn)
 {
     if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
-        && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
+        && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
         return OFPERR_OFPBRC_EPERM;
     } else {
         return 0;
@@ -2443,17 +2447,29 @@ static enum ofperr
 handle_desc_stats_request(struct ofconn *ofconn,
                           const struct ofp_header *request)
 {
+    static const char *default_mfr_desc = "Nicira, Inc.";
+    static const char *default_hw_desc = "Open vSwitch";
+    static const char *default_sw_desc = VERSION;
+    static const char *default_serial_desc = "None";
+    static const char *default_dp_desc = "None";
+
     struct ofproto *p = ofconn_get_ofproto(ofconn);
     struct ofp_desc_stats *ods;
     struct ofpbuf *msg;
 
     msg = ofpraw_alloc_stats_reply(request, 0);
     ods = ofpbuf_put_zeros(msg, sizeof *ods);
-    ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
-    ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
-    ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
-    ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
-    ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
+    ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
+                sizeof ods->mfr_desc);
+    ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
+                sizeof ods->hw_desc);
+    ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
+                sizeof ods->sw_desc);
+    ovs_strlcpy(ods->serial_num,
+                p->serial_desc ? p->serial_desc : default_serial_desc,
+                sizeof ods->serial_num);
+    ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
+                sizeof ods->dp_desc);
     ofconn_send_reply(ofconn, msg);
 
     return 0;
@@ -2848,11 +2864,7 @@ flow_stats_ds(struct rule *rule, struct ds *results)
     ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
     cls_rule_format(&rule->cr, results);
     ds_put_char(results, ',');
-    if (rule->ofpacts_len > 0) {
-        ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
-    } else {
-        ds_put_cstr(results, "drop");
-    }
+    ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
     ds_put_cstr(results, "\n");
 }
 
@@ -2883,62 +2895,21 @@ ofproto_get_netflow_ids(const struct ofproto *ofproto,
     ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
 }
 
-/* Checks the fault status of CFM for 'ofp_port' within 'ofproto'.  Returns a
- * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally
- * indicating a connectivity problem).  Returns zero if CFM is not faulted,
- * and -1 if CFM is not enabled on 'ofp_port'. */
-int
-ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
-{
-    struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
-    return (ofport && ofproto->ofproto_class->get_cfm_fault
-            ? ofproto->ofproto_class->get_cfm_fault(ofport)
-            : -1);
-}
-
-/* Checks the operational status reported by the remote CFM endpoint of
- * 'ofp_port'  Returns 1 if operationally up, 0 if operationally down, and -1
- * if CFM is not enabled on 'ofp_port' or does not support operational status.
- */
-int
-ofproto_port_get_cfm_opup(const struct ofproto *ofproto, uint16_t ofp_port)
-{
-    struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
-    return (ofport && ofproto->ofproto_class->get_cfm_opup
-            ? ofproto->ofproto_class->get_cfm_opup(ofport)
-            : -1);
-}
-
-/* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port'
- * within 'ofproto'.  Populates 'rmps' with an array of MPIDs owned by
- * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'.  Returns a
- * number less than 0 if CFM is not enabled on 'ofp_port'. */
-int
-ofproto_port_get_cfm_remote_mpids(const struct ofproto *ofproto,
-                                  uint16_t ofp_port, const uint64_t **rmps,
-                                  size_t *n_rmps)
-{
-    struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
-
-    *rmps = NULL;
-    *n_rmps = 0;
-    return (ofport && ofproto->ofproto_class->get_cfm_remote_mpids
-            ? ofproto->ofproto_class->get_cfm_remote_mpids(ofport, rmps,
-                                                           n_rmps)
-            : -1);
-}
-
-/* Checks the health of the CFM for 'ofp_port' within 'ofproto'.  Returns an
- * integer value between 0 and 100 to indicate the health of the port as a
- * percentage which is the average of cfm health of all the remote_mpids or
- * returns -1 if CFM is not enabled on 'ofport'. */
-int
-ofproto_port_get_cfm_health(const struct ofproto *ofproto, uint16_t ofp_port)
+/* Checks the status of CFM configured on 'ofp_port' within 'ofproto'.  Returns
+ * true if the port's CFM status was successfully stored into '*status'.
+ * Returns false if the port did not have CFM configured, in which case
+ * '*status' is indeterminate.
+ *
+ * The caller must provide and owns '*status', but it does not own and must not
+ * modify or free the array returned in 'status->rmps'. */
+bool
+ofproto_port_get_cfm_status(const struct ofproto *ofproto, uint16_t ofp_port,
+                            struct ofproto_cfm_status *status)
 {
     struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
-    return (ofport && ofproto->ofproto_class->get_cfm_health
-            ? ofproto->ofproto_class->get_cfm_health(ofport)
-            : -1);
+    return (ofport
+            && ofproto->ofproto_class->get_cfm_status
+            && ofproto->ofproto_class->get_cfm_status(ofport, status));
 }
 
 static enum ofperr
@@ -3150,7 +3121,7 @@ add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
             if (error) {
                 return error;
             }
-            assert(table_id < ofproto->n_tables);
+            ovs_assert(table_id < ofproto->n_tables);
             table = &ofproto->tables[table_id];
         } else {
             table = &ofproto->tables[0];
@@ -3205,6 +3176,7 @@ add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
     rule->ofpacts_len = fm->ofpacts_len;
     rule->evictable = true;
     rule->eviction_group = NULL;
+    list_init(&rule->expirable);
     rule->monitor_flags = 0;
     rule->add_seqno = 0;
     rule->modify_seqno = 0;
@@ -3300,10 +3272,6 @@ modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
         new_cookie = (fm->new_cookie != htonll(UINT64_MAX)
                       ? fm->new_cookie
                       : rule->flow_cookie);
-        if (!actions_changed && new_cookie == rule->flow_cookie) {
-            /* No change at all. */
-            continue;
-        }
 
         op = ofoperation_create(group, rule, OFOPERATION_MODIFY, 0);
         rule->flow_cookie = new_cookie;
@@ -3508,7 +3476,7 @@ ofproto_rule_expire(struct rule *rule, uint8_t reason)
     struct ofproto *ofproto = rule->ofproto;
     struct ofopgroup *group;
 
-    assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
+    ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
 
     ofproto_rule_send_removed(rule, reason);
 
@@ -3542,7 +3510,7 @@ handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
                               &fm.match.flow, ofproto->max_ports);
     }
     if (!error) {
-        error = handle_flow_mod__(ofconn_get_ofproto(ofconn), ofconn, &fm, oh);
+        error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
     }
     if (error) {
         goto exit_free_ofpacts;
@@ -3586,7 +3554,7 @@ handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
                   const struct ofp_header *oh)
 {
     if (ofproto->n_pending >= 50) {
-        assert(!list_is_empty(&ofproto->pending));
+        ovs_assert(!list_is_empty(&ofproto->pending));
         return OFPROTO_POSTPONE;
     }
 
@@ -3619,27 +3587,34 @@ handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
 static enum ofperr
 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
 {
-    const struct nx_role_request *nrr = ofpmsg_body(oh);
-    struct nx_role_request *reply;
+    struct ofputil_role_request request;
+    struct ofputil_role_request reply;
     struct ofpbuf *buf;
-    uint32_t role;
+    enum ofperr error;
 
-    role = ntohl(nrr->role);
-    if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
-        && role != NX_ROLE_SLAVE) {
-        return OFPERR_OFPRRFC_BAD_ROLE;
+    error = ofputil_decode_role_message(oh, &request);
+    if (error) {
+        return error;
     }
 
-    if (ofconn_get_role(ofconn) != role
-        && ofconn_has_pending_opgroups(ofconn)) {
-        return OFPROTO_POSTPONE;
-    }
+    if (request.role != OFPCR12_ROLE_NOCHANGE) {
+        if (ofconn_get_role(ofconn) != request.role
+            && ofconn_has_pending_opgroups(ofconn)) {
+            return OFPROTO_POSTPONE;
+        }
+
+        if (request.have_generation_id
+            && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
+                return OFPERR_OFPRRFC_STALE;
+        }
 
-    ofconn_set_role(ofconn, role);
+        ofconn_set_role(ofconn, request.role);
+    }
 
-    buf = ofpraw_alloc_reply(OFPRAW_NXT_ROLE_REPLY, oh, 0);
-    reply = ofpbuf_put_zeros(buf, sizeof *reply);
-    reply->role = htonl(role);
+    reply.role = ofconn_get_role(ofconn);
+    reply.have_generation_id = ofconn_get_master_election_id(
+        ofconn, &reply.generation_id);
+    buf = ofputil_encode_role_reply(oh, &reply);
     ofconn_send_reply(ofconn, buf);
 
     return 0;
@@ -3894,7 +3869,7 @@ ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
 
         cls_cursor_init(&cursor, &table->cls, &target);
         CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
-            assert(!rule->pending); /* XXX */
+            ovs_assert(!rule->pending); /* XXX */
             ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
         }
     }
@@ -4050,14 +4025,14 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
     case OFPTYPE_BARRIER_REQUEST:
         return handle_barrier_request(ofconn, oh);
 
+    case OFPTYPE_ROLE_REQUEST:
+        return handle_role_request(ofconn, oh);
+
         /* OpenFlow replies. */
     case OFPTYPE_ECHO_REPLY:
         return 0;
 
         /* Nicira extension requests. */
-    case OFPTYPE_ROLE_REQUEST:
-        return handle_role_request(ofconn, oh);
-
     case OFPTYPE_FLOW_MOD_TABLE_ID:
         return handle_nxt_flow_mod_table_id(ofconn, oh);
 
@@ -4200,7 +4175,7 @@ ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
     if (ofconn) {
         size_t request_len = ntohs(request->length);
 
-        assert(ofconn_get_ofproto(ofconn) == ofproto);
+        ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
 
         ofconn_add_opgroup(ofconn, &group->ofconn_node);
         group->ofconn = ofconn;
@@ -4238,7 +4213,7 @@ ofopgroup_complete(struct ofopgroup *group)
     struct ofoperation *op, *next_op;
     int error;
 
-    assert(!group->n_running);
+    ovs_assert(!group->n_running);
 
     error = 0;
     LIST_FOR_EACH (op, group_node, &group->ops) {
@@ -4257,7 +4232,7 @@ ofopgroup_complete(struct ofopgroup *group)
                 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
                                                &packet, &in_port);
                 if (packet) {
-                    assert(!error);
+                    ovs_assert(!error);
                     error = rule_execute(op->rule, in_port, packet);
                 }
                 break;
@@ -4275,7 +4250,19 @@ ofopgroup_complete(struct ofopgroup *group)
     LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
         struct rule *rule = op->rule;
 
-        if (!op->error && !ofproto_rule_is_hidden(rule)) {
+        /* We generally want to report the change to active OpenFlow flow
+           monitors (e.g. NXST_FLOW_MONITOR).  There are three exceptions:
+
+              - The operation failed.
+
+              - The affected rule is not visible to controllers.
+
+              - The operation's only effect was to update rule->modified. */
+        if (!(op->error
+              || ofproto_rule_is_hidden(rule)
+              || (op->type == OFOPERATION_MODIFY
+                  && op->ofpacts
+                  && rule->flow_cookie == op->flow_cookie))) {
             /* Check that we can just cast from ofoperation_type to
              * nx_flow_update_event. */
             BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_ADD
@@ -4317,7 +4304,7 @@ ofopgroup_complete(struct ofopgroup *group)
             break;
 
         case OFOPERATION_DELETE:
-            assert(!op->error);
+            ovs_assert(!op->error);
             ofproto_rule_destroy__(rule);
             op->rule = NULL;
             break;
@@ -4347,7 +4334,7 @@ ofopgroup_complete(struct ofopgroup *group)
     ofmonitor_flush(ofproto->connmgr);
 
     if (!list_is_empty(&group->ofproto_node)) {
-        assert(ofproto->n_pending > 0);
+        ovs_assert(ofproto->n_pending > 0);
         ofproto->n_pending--;
         list_remove(&group->ofproto_node);
     }
@@ -4378,7 +4365,7 @@ ofoperation_create(struct ofopgroup *group, struct rule *rule,
     struct ofproto *ofproto = group->ofproto;
     struct ofoperation *op;
 
-    assert(!rule->pending);
+    ovs_assert(!rule->pending);
 
     op = rule->pending = xzalloc(sizeof *op);
     op->group = group;
@@ -4448,9 +4435,9 @@ ofoperation_complete(struct ofoperation *op, enum ofperr error)
 {
     struct ofopgroup *group = op->group;
 
-    assert(op->rule->pending == op);
-    assert(group->n_running > 0);
-    assert(!error || op->type != OFOPERATION_DELETE);
+    ovs_assert(op->rule->pending == op);
+    ovs_assert(group->n_running > 0);
+    ovs_assert(!error || op->type != OFOPERATION_DELETE);
 
     op->error = error;
     if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
@@ -4461,7 +4448,7 @@ ofoperation_complete(struct ofoperation *op, enum ofperr error)
 struct rule *
 ofoperation_get_victim(struct ofoperation *op)
 {
-    assert(op->type == OFOPERATION_ADD);
+    ovs_assert(op->type == OFOPERATION_ADD);
     return op->victim;
 }
 \f
@@ -4761,7 +4748,7 @@ oftable_init(struct oftable *table)
 static void
 oftable_destroy(struct oftable *table)
 {
-    assert(classifier_is_empty(&table->cls));
+    ovs_assert(classifier_is_empty(&table->cls));
     oftable_disable_eviction(table);
     classifier_destroy(&table->cls);
     free(table->name);
@@ -4856,6 +4843,9 @@ oftable_remove_rule(struct rule *rule)
 
     classifier_remove(&table->cls, &rule->cr);
     eviction_group_remove_rule(rule);
+    if (!list_is_empty(&rule->expirable)) {
+        list_remove(&rule->expirable);
+    }
 }
 
 /* Inserts 'rule' into its oftable.  Removes any existing rule from 'rule''s
@@ -4867,9 +4857,17 @@ oftable_replace_rule(struct rule *rule)
     struct ofproto *ofproto = rule->ofproto;
     struct oftable *table = &ofproto->tables[rule->table_id];
     struct rule *victim;
+    bool may_expire = rule->hard_timeout || rule->idle_timeout;
+
+    if (may_expire) {
+        list_insert(&ofproto->expirable, &rule->expirable);
+    }
 
     victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
     if (victim) {
+        if (!list_is_empty(&victim->expirable)) {
+            list_remove(&victim->expirable);
+        }
         eviction_group_remove_rule(victim);
     }
     eviction_group_add_rule(rule);
@@ -4988,7 +4986,7 @@ ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port,
     struct ofport *ofport;
     int error;
 
-    assert(vlandev_ofp_port != realdev_ofp_port);
+    ovs_assert(vlandev_ofp_port != realdev_ofp_port);
 
     ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
     if (!ofport) {