X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=ofproto%2Fofproto.c;h=709fd0751e3613135437b2e9d0ce04b72eb9bbf6;hb=69fc54f47bbc35e81bfe2e38e57f5dcfd9858df4;hp=5e6a25258ca8fb8c339ece4fb819804d4362ad5c;hpb=f9c0c3eccdb8e14dc8fbf572e515dbef1eb7b2ee;p=sliver-openvswitch.git diff --git a/ofproto/ofproto.c b/ofproto/ofproto.c index 5e6a25258..709fd0751 100644 --- a/ofproto/ofproto.c +++ b/ofproto/ofproto.c @@ -21,6 +21,7 @@ #include #include #include +#include #include "bitmap.h" #include "byte-order.h" #include "classifier.h" @@ -126,6 +127,7 @@ struct ofoperation { /* OFOPERATION_MODIFY: The old actions, if the actions are changing. */ struct ofpact *ofpacts; size_t ofpacts_len; + uint32_t meter_id; /* OFOPERATION_DELETE. */ enum ofp_flow_removed_reason reason; /* Reason flow was removed. */ @@ -151,7 +153,10 @@ static void oftable_enable_eviction(struct oftable *, const struct mf_subfield *fields, size_t n_fields); -static void oftable_remove_rule(struct rule *); +static void oftable_remove_rule(struct rule *rule) OVS_RELEASES(rule->evict); +static void oftable_remove_rule__(struct ofproto *ofproto, + struct classifier *cls, struct rule *rule) + OVS_REQ_WRLOCK(cls->rwlock) OVS_RELEASES(rule->evict); static struct rule *oftable_replace_rule(struct rule *); static void oftable_substitute_rule(struct rule *old, struct rule *new); @@ -177,7 +182,8 @@ struct eviction_group { struct heap rules; /* Contains "struct rule"s. */ }; -static struct rule *choose_rule_to_evict(struct oftable *); +static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep) + OVS_TRY_WRLOCK(true, (*rulep)->evict); static void ofproto_evict(struct ofproto *); static uint32_t rule_eviction_priority(struct rule *); @@ -196,12 +202,14 @@ static bool rule_is_modifiable(const struct rule *); /* OpenFlow. */ static enum ofperr add_flow(struct ofproto *, struct ofconn *, - const struct ofputil_flow_mod *, + struct ofputil_flow_mod *, const struct ofp_header *); -static void delete_flow__(struct rule *, struct ofopgroup *); +static void delete_flow__(struct rule *rule, struct ofopgroup *, + enum ofp_flow_removed_reason) + OVS_RELEASES(rule->evict); static bool handle_openflow(struct ofconn *, const struct ofpbuf *); static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *, - const struct ofputil_flow_mod *, + struct ofputil_flow_mod *, const struct ofp_header *); static void calc_duration(long long int start, long long int now, uint32_t *sec, uint32_t *nsec); @@ -211,6 +219,7 @@ static uint64_t pick_datapath_id(const struct ofproto *); static uint64_t pick_fallback_dpid(void); static void ofproto_destroy__(struct ofproto *); static void update_mtu(struct ofproto *, struct ofport *); +static void meter_delete(struct ofproto *, uint32_t first, uint32_t last); /* unixctl. */ static void ofproto_unixctl_init(void); @@ -221,6 +230,8 @@ static size_t n_ofproto_classes; static size_t allocated_ofproto_classes; unsigned flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT; +unsigned n_handler_threads; +enum ofproto_flow_miss_model flow_miss_model = OFPROTO_HANDLE_MISS_AUTO; /* Map from datapath name to struct ofproto, for use by unixctl commands. */ static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos); @@ -421,11 +432,12 @@ ofproto_create(const char *datapath_name, const char *datapath_type, hmap_init(&ofproto->ports); shash_init(&ofproto->port_by_name); simap_init(&ofproto->ofp_requests); - ofproto->max_ports = OFPP_MAX; + ofproto->max_ports = ofp_to_u16(OFPP_MAX); ofproto->tables = NULL; ofproto->n_tables = 0; hindex_init(&ofproto->cookies); list_init(&ofproto->expirable); + ovs_mutex_init_recursive(&ofproto->expirable_mutex); ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name); ofproto->state = S_OPENFLOW; list_init(&ofproto->pending); @@ -442,14 +454,14 @@ ofproto_create(const char *datapath_name, const char *datapath_type, error = ofproto->ofproto_class->construct(ofproto); if (error) { VLOG_ERR("failed to open datapath %s: %s", - datapath_name, strerror(error)); + datapath_name, ovs_strerror(error)); ofproto_destroy__(ofproto); return error; } /* The "max_ports" member should have been set by ->construct(ofproto). * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */ - ofproto->ofp_port_ids = bitmap_allocate(ofp_to_u16(ofproto->max_ports)); + ofproto->ofp_port_ids = bitmap_allocate(ofproto->max_ports); bitmap_set1(ofproto->ofp_port_ids, 0); /* Check that hidden tables, if any, are at the end. */ @@ -464,6 +476,16 @@ ofproto_create(const char *datapath_name, const char *datapath_type, ofproto->datapath_id = pick_datapath_id(ofproto); init_ports(ofproto); + /* Initialize meters table. */ + if (ofproto->ofproto_class->meter_get_features) { + ofproto->ofproto_class->meter_get_features(ofproto, + &ofproto->meter_features); + } else { + memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features); + } + ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1) + * sizeof(struct meter *)); + *ofprotop = ofproto; return 0; } @@ -498,9 +520,9 @@ ofproto_init_tables(struct ofproto *ofproto, int n_tables) * Reserved ports numbered OFPP_MAX and higher are special and not subject to * the 'max_ports' restriction. */ void -ofproto_init_max_ports(struct ofproto *ofproto, ofp_port_t max_ports) +ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports) { - ovs_assert(ofp_to_u16(max_ports) <= ofp_to_u16(OFPP_MAX)); + ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX)); ofproto->max_ports = max_ports; } @@ -573,6 +595,13 @@ ofproto_set_flow_eviction_threshold(unsigned threshold) threshold); } +/* Sets the path for handling flow misses. */ +void +ofproto_set_flow_miss_model(unsigned model) +{ + flow_miss_model = model; +} + /* If forward_bpdu is true, the NORMAL action will forward frames with * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false, * the NORMAL action will drop these frames. */ @@ -601,6 +630,18 @@ ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time, } } +/* Sets number of upcall handler threads. The default is + * (number of online cores - 1). */ +void +ofproto_set_n_handler_threads(unsigned limit) +{ + if (limit) { + n_handler_threads = limit; + } else { + n_handler_threads = MAX(1, sysconf(_SC_NPROCESSORS_ONLN) - 1); + } +} + void ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc) { @@ -811,7 +852,7 @@ ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port, if (error) { VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)", ofproto->name, ofp_port, netdev_get_name(ofport->netdev), - strerror(error)); + ovs_strerror(error)); } } @@ -837,7 +878,7 @@ ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port, if (error) { VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)", ofproto->name, ofp_port, netdev_get_name(ofport->netdev), - strerror(error)); + ovs_strerror(error)); } } @@ -996,6 +1037,7 @@ ofproto_configure_table(struct ofproto *ofproto, int table_id, } table->max_flows = s->max_flows; + ovs_rwlock_rdlock(&table->cls.rwlock); if (classifier_count(&table->cls) > table->max_flows && table->eviction_fields) { /* 'table' contains more flows than allowed. We might not be able to @@ -1011,6 +1053,7 @@ ofproto_configure_table(struct ofproto *ofproto, int table_id, break; } } + ovs_rwlock_unlock(&table->cls.rwlock); } bool @@ -1044,15 +1087,18 @@ ofproto_flush__(struct ofproto *ofproto) continue; } + ovs_rwlock_wrlock(&table->cls.rwlock); cls_cursor_init(&cursor, &table->cls, NULL); CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { if (!rule->pending) { ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE); - oftable_remove_rule(rule); + ovs_rwlock_wrlock(&rule->evict); + oftable_remove_rule__(ofproto, &table->cls, rule); ofproto->ofproto_class->rule_destruct(rule); } } + ovs_rwlock_unlock(&table->cls.rwlock); } ofopgroup_submit(group); } @@ -1065,6 +1111,11 @@ ofproto_destroy__(struct ofproto *ofproto) ovs_assert(list_is_empty(&ofproto->pending)); ovs_assert(!ofproto->n_pending); + if (ofproto->meters) { + meter_delete(ofproto, 1, ofproto->meter_features.max_meters); + free(ofproto->meters); + } + connmgr_destroy(ofproto->connmgr); hmap_remove(&all_ofprotos, &ofproto->hmap_node); @@ -1089,6 +1140,7 @@ ofproto_destroy__(struct ofproto *ofproto) free(ofproto->vlan_bitmap); + ovs_mutex_destroy(&ofproto->expirable_mutex); ofproto->ofproto_class->dealloc(ofproto); } @@ -1148,7 +1200,7 @@ ofproto_type_run(const char *datapath_type) error = class->type_run ? class->type_run(datapath_type) : 0; if (error && error != EAGAIN) { VLOG_ERR_RL(&rl, "%s: type_run failed (%s)", - datapath_type, strerror(error)); + datapath_type, ovs_strerror(error)); } return error; } @@ -1165,7 +1217,7 @@ ofproto_type_run_fast(const char *datapath_type) error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0; if (error && error != EAGAIN) { VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)", - datapath_type, strerror(error)); + datapath_type, ovs_strerror(error)); } return error; } @@ -1193,7 +1245,7 @@ ofproto_run(struct ofproto *p) error = p->ofproto_class->run(p); if (error && error != EAGAIN) { - VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error)); + VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error)); } if (p->ofproto_class->port_poll) { @@ -1303,7 +1355,7 @@ ofproto_run_fast(struct ofproto *p) error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0; if (error && error != EAGAIN) { VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)", - p->name, strerror(error)); + p->name, ovs_strerror(error)); } return error; } @@ -1359,7 +1411,9 @@ ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage) n_rules = 0; OFPROTO_FOR_EACH_TABLE (table, ofproto) { + ovs_rwlock_rdlock(&table->cls.rwlock); n_rules += classifier_count(&table->cls); + ovs_rwlock_unlock(&table->cls.rwlock); } simap_increase(usage, "rules", n_rules); @@ -1586,8 +1640,10 @@ ofproto_add_flow(struct ofproto *ofproto, const struct match *match, { const struct rule *rule; + ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock); rule = rule_from_cls_rule(classifier_find_match_exactly( &ofproto->tables[0].cls, match, priority)); + ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock); if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len, ofpacts, ofpacts_len)) { struct ofputil_flow_mod fm; @@ -1609,7 +1665,7 @@ ofproto_add_flow(struct ofproto *ofproto, const struct match *match, * * This is a helper function for in-band control and fail-open. */ int -ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm) +ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm) { return handle_flow_mod__(ofproto, NULL, fm, NULL); } @@ -1624,8 +1680,10 @@ ofproto_delete_flow(struct ofproto *ofproto, { struct rule *rule; + ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock); rule = rule_from_cls_rule(classifier_find_match_exactly( &ofproto->tables[0].cls, target, priority)); + ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock); if (!rule) { /* No such rule -> success. */ return true; @@ -1637,6 +1695,7 @@ ofproto_delete_flow(struct ofproto *ofproto, /* Initiate deletion -> success. */ struct ofopgroup *group = ofopgroup_create_unattached(ofproto); ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE); + ovs_rwlock_wrlock(&rule->evict); oftable_remove_rule(rule); ofproto->ofproto_class->rule_destruct(rule); ofopgroup_submit(group); @@ -1683,32 +1742,28 @@ reinit_ports(struct ofproto *p) static ofp_port_t alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name) { - uint16_t max_ports = ofp_to_u16(ofproto->max_ports); uint16_t port_idx; port_idx = simap_get(&ofproto->ofp_requests, netdev_name); - if (!port_idx) { - port_idx = UINT16_MAX; - } + port_idx = port_idx ? port_idx : UINT16_MAX; - if (port_idx >= max_ports + if (port_idx >= ofproto->max_ports || bitmap_is_set(ofproto->ofp_port_ids, port_idx)) { - uint16_t end_port_no = ofp_to_u16(ofproto->alloc_port_no); - uint16_t alloc_port_no = end_port_no; + uint16_t end_port_no = ofproto->alloc_port_no; /* Search for a free OpenFlow port number. We try not to * immediately reuse them to prevent problems due to old * flows. */ for (;;) { - if (++alloc_port_no >= max_ports) { - alloc_port_no = 0; + if (++ofproto->alloc_port_no >= ofproto->max_ports) { + ofproto->alloc_port_no = 0; } - if (!bitmap_is_set(ofproto->ofp_port_ids, alloc_port_no)) { - port_idx = alloc_port_no; - ofproto->alloc_port_no = u16_to_ofp(alloc_port_no); + if (!bitmap_is_set(ofproto->ofp_port_ids, + ofproto->alloc_port_no)) { + port_idx = ofproto->alloc_port_no; break; } - if (alloc_port_no == end_port_no) { + if (ofproto->alloc_port_no == end_port_no) { return OFPP_NONE; } } @@ -1720,7 +1775,7 @@ alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name) static void dealloc_ofp_port(const struct ofproto *ofproto, ofp_port_t ofp_port) { - if (ofp_to_u16(ofp_port) < ofp_to_u16(ofproto->max_ports)) { + if (ofp_to_u16(ofp_port) < ofproto->max_ports) { bitmap_set0(ofproto->ofp_port_ids, ofp_to_u16(ofp_port)); } } @@ -1743,7 +1798,7 @@ ofport_open(struct ofproto *ofproto, "cannot be opened (%s)", ofproto->name, ofproto_port->name, ofproto_port->ofp_port, - ofproto_port->name, strerror(error)); + ofproto_port->name, ovs_strerror(error)); return NULL; } @@ -1828,7 +1883,7 @@ ofport_install(struct ofproto *p, error: VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)", - p->name, netdev_name, strerror(error)); + p->name, netdev_name, ovs_strerror(error)); if (ofport) { ofport_destroy__(ofport); } else { @@ -2044,6 +2099,10 @@ init_ports(struct ofproto *p) netdev = ofport_open(p, &ofproto_port, &pp); if (netdev) { ofport_install(p, netdev, &pp); + if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) { + p->alloc_port_no = MAX(p->alloc_port_no, + ofp_to_u16(ofproto_port.ofp_port)); + } } } } @@ -2139,6 +2198,8 @@ ofproto_rule_destroy__(struct rule *rule) if (rule) { cls_rule_destroy(&rule->cr); free(rule->ofpacts); + ovs_mutex_destroy(&rule->timeout_mutex); + ovs_rwlock_destroy(&rule->evict); rule->ofproto->ofproto_class->rule_dealloc(rule); } } @@ -2152,10 +2213,15 @@ ofproto_rule_destroy__(struct rule *rule) * This function should only be called from an ofproto implementation's * ->destruct() function. It is not suitable elsewhere. */ void -ofproto_rule_destroy(struct rule *rule) +ofproto_rule_destroy(struct ofproto *ofproto, struct classifier *cls, + struct rule *rule) OVS_REQ_WRLOCK(cls->rwlock) { ovs_assert(!rule->pending); - oftable_remove_rule(rule); + if (!ovs_rwlock_trywrlock(&rule->evict)) { + oftable_remove_rule__(ofproto, cls, rule); + } else { + NOT_REACHED(); + } ofproto_rule_destroy__(rule); } @@ -2360,6 +2426,55 @@ reject_slave_controller(struct ofconn *ofconn) } } +/* Finds the OFPACT_METER action, if any, in the 'ofpacts_len' bytes of + * 'ofpacts'. If found, returns its meter ID; if not, returns 0. + * + * This function relies on the order of 'ofpacts' being correct (as checked by + * ofpacts_verify()). */ +static uint32_t +find_meter(const struct ofpact ofpacts[], size_t ofpacts_len) +{ + const struct ofpact *a; + + OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) { + enum ovs_instruction_type inst; + + inst = ovs_instruction_type_from_ofpact_type(a->type); + if (a->type == OFPACT_METER) { + return ofpact_get_METER(a)->meter_id; + } else if (inst > OVSINST_OFPIT13_METER) { + break; + } + } + + return 0; +} + +/* Checks that the 'ofpacts_len' bytes of actions in 'ofpacts' are appropriate + * for a packet with the prerequisites satisfied by 'flow' in table 'table_id'. + * 'flow' may be temporarily modified, but is restored at return. + */ +static enum ofperr +ofproto_check_ofpacts(struct ofproto *ofproto, + const struct ofpact ofpacts[], size_t ofpacts_len, + struct flow *flow, uint8_t table_id) +{ + enum ofperr error; + uint32_t mid; + + error = ofpacts_check(ofpacts, ofpacts_len, flow, + u16_to_ofp(ofproto->max_ports), table_id); + if (error) { + return error; + } + + mid = find_meter(ofpacts, ofpacts_len); + if (mid && ofproto_get_provider_meter_id(ofproto, mid) == UINT32_MAX) { + return OFPERR_OFPMMFC_INVALID_METER; + } + return 0; +} + static enum ofperr handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh) { @@ -2385,7 +2500,7 @@ handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh) if (error) { goto exit_free_ofpacts; } - if (ofp_to_u16(po.in_port) >= ofp_to_u16(p->max_ports) + if (ofp_to_u16(po.in_port) >= p->max_ports && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) { error = OFPERR_OFPBRC_BAD_PORT; goto exit_free_ofpacts; @@ -2406,7 +2521,7 @@ handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh) /* Verify actions against packet, then send packet if successful. */ in_port_.ofp_port = po.in_port; flow_extract(payload, 0, 0, NULL, &in_port_, &flow); - error = ofpacts_check(po.ofpacts, po.ofpacts_len, &flow, p->max_ports); + error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len, &flow, 0); if (!error) { error = p->ofproto_class->packet_out(p, payload, &flow, po.ofpacts, po.ofpacts_len); @@ -2538,7 +2653,9 @@ handle_table_stats_request(struct ofconn *ofconn, ots[i].instructions = htonl(OFPIT11_ALL); ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK); ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */ + ovs_rwlock_rdlock(&p->tables[i].cls.rwlock); ots[i].active_count = htonl(classifier_count(&p->tables[i].cls)); + ovs_rwlock_unlock(&p->tables[i].cls.rwlock); } p->ofproto_class->get_tables(p, ots); @@ -2805,9 +2922,11 @@ collect_rules_loose(struct ofproto *ofproto, uint8_t table_id, struct cls_cursor cursor; struct rule *rule; + ovs_rwlock_rdlock(&table->cls.rwlock); cls_cursor_init(&cursor, &table->cls, &cr); CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { if (rule->pending) { + ovs_rwlock_unlock(&table->cls.rwlock); error = OFPROTO_POSTPONE; goto exit; } @@ -2817,6 +2936,7 @@ collect_rules_loose(struct ofproto *ofproto, uint8_t table_id, list_push_back(rules, &rule->ofproto_node); } } + ovs_rwlock_unlock(&table->cls.rwlock); } exit: @@ -2881,8 +3001,10 @@ collect_rules_strict(struct ofproto *ofproto, uint8_t table_id, FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) { struct rule *rule; + ovs_rwlock_rdlock(&table->cls.rwlock); rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr)); + ovs_rwlock_unlock(&table->cls.rwlock); if (rule) { if (rule->pending) { error = OFPROTO_POSTPONE; @@ -2944,8 +3066,6 @@ handle_flow_stats_request(struct ofconn *ofconn, fs.cookie = rule->flow_cookie; fs.table_id = rule->table_id; calc_duration(rule->created, now, &fs.duration_sec, &fs.duration_nsec); - fs.idle_timeout = rule->idle_timeout; - fs.hard_timeout = rule->hard_timeout; fs.idle_age = age_secs(now - rule->used); fs.hard_age = age_secs(now - rule->modified); ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count, @@ -2953,6 +3073,12 @@ handle_flow_stats_request(struct ofconn *ofconn, fs.ofpacts = rule->ofpacts; fs.ofpacts_len = rule->ofpacts_len; fs.flags = 0; + + ovs_mutex_lock(&rule->timeout_mutex); + fs.idle_timeout = rule->idle_timeout; + fs.hard_timeout = rule->hard_timeout; + ovs_mutex_unlock(&rule->timeout_mutex); + if (rule->send_flow_removed) { fs.flags |= OFPFF_SEND_FLOW_REM; /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS @@ -2998,10 +3124,12 @@ ofproto_get_all_flows(struct ofproto *p, struct ds *results) struct cls_cursor cursor; struct rule *rule; + ovs_rwlock_rdlock(&table->cls.rwlock); cls_cursor_init(&cursor, &table->cls, NULL); CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { flow_stats_ds(rule, results); } + ovs_rwlock_unlock(&table->cls.rwlock); } } @@ -3019,8 +3147,7 @@ ofproto_get_netflow_ids(const struct ofproto *ofproto, * 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'. */ + * The caller must provide and owns '*status', and must free 'status->rmps'. */ bool ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port, struct ofproto_cfm_status *status) @@ -3095,18 +3222,26 @@ handle_aggregate_stats_request(struct ofconn *ofconn, struct queue_stats_cbdata { struct ofport *ofport; struct list replies; + long long int now; }; static void put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id, const struct netdev_queue_stats *stats) { + struct ofputil_queue_stats oqs; - struct ofputil_queue_stats oqs = { - .port_no = cbdata->ofport->pp.port_no, - .queue_id = queue_id, - .stats = *stats, - }; + oqs.port_no = cbdata->ofport->pp.port_no; + oqs.queue_id = queue_id; + oqs.tx_bytes = stats->tx_bytes; + oqs.tx_packets = stats->tx_packets; + oqs.tx_errors = stats->tx_errors; + if (stats->created != LLONG_MIN) { + calc_duration(stats->created, cbdata->now, + &oqs.duration_sec, &oqs.duration_nsec); + } else { + oqs.duration_sec = oqs.duration_nsec = UINT32_MAX; + } ofputil_append_queue_stat(&cbdata->replies, &oqs); } @@ -3153,6 +3288,7 @@ handle_queue_stats_request(struct ofconn *ofconn, COVERAGE_INC(ofproto_queue_req); ofpmp_init(&cbdata.replies, rq); + cbdata.now = time_msec(); error = ofputil_decode_queue_stats_request(rq, &oqsr); if (error) { @@ -3216,7 +3352,7 @@ is_flow_deletion_pending(const struct ofproto *ofproto, * if any. */ static enum ofperr add_flow(struct ofproto *ofproto, struct ofconn *ofconn, - const struct ofputil_flow_mod *fm, const struct ofp_header *request) + struct ofputil_flow_mod *fm, const struct ofp_header *request) { struct oftable *table; struct ofopgroup *group; @@ -3255,11 +3391,18 @@ add_flow(struct ofproto *ofproto, struct ofconn *ofconn, return OFPERR_OFPBRC_EPERM; } + /* Verify actions. */ + error = ofproto_check_ofpacts(ofproto, fm->ofpacts, fm->ofpacts_len, + &fm->match.flow, table_id); + if (error) { + return error; + } + /* Allocate new rule and initialize classifier rule. */ rule = ofproto->ofproto_class->rule_alloc(); if (!rule) { VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)", - ofproto->name, strerror(error)); + ofproto->name, ovs_strerror(error)); return ENOMEM; } cls_rule_init(&rule->cr, &fm->match, fm->priority); @@ -3272,11 +3415,18 @@ add_flow(struct ofproto *ofproto, struct ofconn *ofconn, } /* Check for overlap, if requested. */ - if (fm->flags & OFPFF_CHECK_OVERLAP - && classifier_rule_overlaps(&table->cls, &rule->cr)) { - cls_rule_destroy(&rule->cr); - ofproto->ofproto_class->rule_dealloc(rule); - return OFPERR_OFPFMFC_OVERLAP; + if (fm->flags & OFPFF_CHECK_OVERLAP) { + bool overlaps; + + ovs_rwlock_rdlock(&table->cls.rwlock); + overlaps = classifier_rule_overlaps(&table->cls, &rule->cr); + ovs_rwlock_unlock(&table->cls.rwlock); + + if (overlaps) { + cls_rule_destroy(&rule->cr); + ofproto->ofproto_class->rule_dealloc(rule); + return OFPERR_OFPFMFC_OVERLAP; + } } /* FIXME: Implement OFPFF12_RESET_COUNTS */ @@ -3285,20 +3435,27 @@ add_flow(struct ofproto *ofproto, struct ofconn *ofconn, rule->pending = NULL; rule->flow_cookie = fm->new_cookie; rule->created = rule->modified = rule->used = time_msec(); + + ovs_mutex_init(&rule->timeout_mutex); + ovs_mutex_lock(&rule->timeout_mutex); rule->idle_timeout = fm->idle_timeout; rule->hard_timeout = fm->hard_timeout; + ovs_mutex_unlock(&rule->timeout_mutex); + rule->table_id = table - ofproto->tables; rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0; /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS and OFPFF13_NO_BYT_COUNTS */ rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len); rule->ofpacts_len = fm->ofpacts_len; - rule->evictable = true; + rule->meter_id = find_meter(rule->ofpacts, rule->ofpacts_len); + list_init(&rule->meter_list_node); rule->eviction_group = NULL; list_init(&rule->expirable); rule->monitor_flags = 0; rule->add_seqno = 0; rule->modify_seqno = 0; + ovs_rwlock_init(&rule->evict); /* Insert new rule. */ victim = oftable_replace_rule(rule); @@ -3309,21 +3466,24 @@ add_flow(struct ofproto *ofproto, struct ofconn *ofconn, } else { struct ofoperation *op; struct rule *evict; - - if (classifier_count(&table->cls) > table->max_flows) { - bool was_evictable; - - was_evictable = rule->evictable; - rule->evictable = false; - evict = choose_rule_to_evict(table); - rule->evictable = was_evictable; - - if (!evict) { + size_t n_rules; + + ovs_rwlock_rdlock(&table->cls.rwlock); + n_rules = classifier_count(&table->cls); + ovs_rwlock_unlock(&table->cls.rwlock); + if (n_rules > table->max_flows) { + ovs_rwlock_rdlock(&rule->evict); + if (choose_rule_to_evict(table, &evict)) { + ovs_rwlock_unlock(&rule->evict); + ovs_rwlock_unlock(&evict->evict); + if (evict->pending) { + error = OFPROTO_POSTPONE; + goto exit; + } + } else { + ovs_rwlock_unlock(&rule->evict); error = OFPERR_OFPFMFC_TABLE_FULL; goto exit; - } else if (evict->pending) { - error = OFPROTO_POSTPONE; - goto exit; } } else { evict = NULL; @@ -3338,7 +3498,14 @@ add_flow(struct ofproto *ofproto, struct ofconn *ofconn, op->group->n_running--; ofoperation_destroy(rule->pending); } else if (evict) { - delete_flow__(evict, group); + /* It would be better if we maintained the lock we took in + * choose_rule_to_evict() earlier, but that confuses the thread + * safety analysis, and this code is fragile enough that we really + * need it. In the worst case, we'll have to block a little while + * before we perform the eviction, which doesn't seem like a big + * problem. */ + ovs_rwlock_wrlock(&evict->evict); + delete_flow__(evict, group, OFPRR_EVICTION); } ofopgroup_submit(group); } @@ -3363,8 +3530,8 @@ exit: * Returns 0 on success, otherwise an OpenFlow error code. */ static enum ofperr modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn, - const struct ofputil_flow_mod *fm, - const struct ofp_header *request, struct list *rules) + struct ofputil_flow_mod *fm, const struct ofp_header *request, + struct list *rules) { struct ofopgroup *group; struct rule *rule; @@ -3385,19 +3552,28 @@ modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn, continue; } + /* Verify actions. */ + error = ofpacts_check(fm->ofpacts, fm->ofpacts_len, &fm->match.flow, + u16_to_ofp(ofproto->max_ports), rule->table_id); + if (error) { + return error; + } + actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len, rule->ofpacts, rule->ofpacts_len); op = ofoperation_create(group, rule, OFOPERATION_MODIFY, 0); - if (fm->new_cookie != htonll(UINT64_MAX)) { + if (fm->modify_cookie && fm->new_cookie != htonll(UINT64_MAX)) { ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie); } if (actions_changed) { op->ofpacts = rule->ofpacts; op->ofpacts_len = rule->ofpacts_len; + op->meter_id = rule->meter_id; rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len); rule->ofpacts_len = fm->ofpacts_len; + rule->meter_id = find_meter(rule->ofpacts, rule->ofpacts_len); rule->ofproto->ofproto_class->rule_modify_actions(rule); } else { ofoperation_complete(op, 0); @@ -3410,8 +3586,7 @@ modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn, static enum ofperr modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn, - const struct ofputil_flow_mod *fm, - const struct ofp_header *request) + struct ofputil_flow_mod *fm, const struct ofp_header *request) { if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) { return 0; @@ -3426,7 +3601,7 @@ modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn, * if any. */ static enum ofperr modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn, - const struct ofputil_flow_mod *fm, + struct ofputil_flow_mod *fm, const struct ofp_header *request) { struct list rules; @@ -3451,7 +3626,7 @@ modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn, * if any. */ static enum ofperr modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn, - const struct ofputil_flow_mod *fm, + struct ofputil_flow_mod *fm, const struct ofp_header *request) { struct list rules; @@ -3475,13 +3650,14 @@ modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn, /* OFPFC_DELETE implementation. */ static void -delete_flow__(struct rule *rule, struct ofopgroup *group) +delete_flow__(struct rule *rule, struct ofopgroup *group, + enum ofp_flow_removed_reason reason) { struct ofproto *ofproto = rule->ofproto; - ofproto_rule_send_removed(rule, OFPRR_DELETE); + ofproto_rule_send_removed(rule, reason); - ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE); + ofoperation_create(group, rule, OFOPERATION_DELETE, reason); oftable_remove_rule(rule); ofproto->ofproto_class->rule_destruct(rule); } @@ -3491,14 +3667,16 @@ delete_flow__(struct rule *rule, struct ofopgroup *group) * Returns 0 on success, otherwise an OpenFlow error code. */ static enum ofperr delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn, - const struct ofp_header *request, struct list *rules) + const struct ofp_header *request, struct list *rules, + enum ofp_flow_removed_reason reason) { struct rule *rule, *next; struct ofopgroup *group; group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX); LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) { - delete_flow__(rule, group); + ovs_rwlock_wrlock(&rule->evict); + delete_flow__(rule, group, reason); } ofopgroup_submit(group); @@ -3519,7 +3697,7 @@ delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn, fm->out_port, &rules); return (error ? error : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request, - &rules) + &rules, OFPRR_DELETE) : 0); } @@ -3537,7 +3715,8 @@ delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn, fm->out_port, &rules); return (error ? error : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn, - request, &rules) + request, &rules, + OFPRR_DELETE) : 0); } @@ -3557,8 +3736,10 @@ ofproto_rule_send_removed(struct rule *rule, uint8_t reason) fr.table_id = rule->table_id; calc_duration(rule->created, time_msec(), &fr.duration_sec, &fr.duration_nsec); + ovs_mutex_lock(&rule->timeout_mutex); fr.idle_timeout = rule->idle_timeout; fr.hard_timeout = rule->hard_timeout; + ovs_mutex_unlock(&rule->timeout_mutex); rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count, &fr.byte_count); @@ -3623,10 +3804,6 @@ handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh) ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn), &ofpacts); - if (!error) { - error = ofpacts_check(fm.ofpacts, fm.ofpacts_len, - &fm.match.flow, ofproto->max_ports); - } if (!error) { error = handle_flow_mod__(ofproto, ofconn, &fm, oh); } @@ -3668,8 +3845,7 @@ exit: static enum ofperr handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn, - const struct ofputil_flow_mod *fm, - const struct ofp_header *oh) + struct ofputil_flow_mod *fm, const struct ofp_header *oh) { if (ofproto->n_pending >= 50) { ovs_assert(!list_is_empty(&ofproto->pending)); @@ -3869,8 +4045,10 @@ ofproto_compose_flow_refresh_update(const struct rule *rule, fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD) ? NXFME_ADDED : NXFME_MODIFIED); fu.reason = 0; + ovs_mutex_lock(&rule->timeout_mutex); fu.idle_timeout = rule->idle_timeout; fu.hard_timeout = rule->hard_timeout; + ovs_mutex_unlock(&rule->timeout_mutex); fu.table_id = rule->table_id; fu.cookie = rule->flow_cookie; minimatch_expand(&rule->cr.match, &match); @@ -3985,11 +4163,13 @@ ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m, struct cls_cursor cursor; struct rule *rule; + ovs_rwlock_rdlock(&table->cls.rwlock); cls_cursor_init(&cursor, &table->cls, &target); CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { ovs_assert(!rule->pending); /* XXX */ ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules); } + ovs_rwlock_unlock(&table->cls.rwlock); } HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) { @@ -4105,6 +4285,318 @@ handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh) return 0; } +/* Meters implementation. + * + * Meter table entry, indexed by the OpenFlow meter_id. + * These are always dynamically allocated to allocate enough space for + * the bands. + * 'created' is used to compute the duration for meter stats. + * 'list rules' is needed so that we can delete the dependent rules when the + * meter table entry is deleted. + * 'provider_meter_id' is for the provider's private use. + */ +struct meter { + long long int created; /* Time created. */ + struct list rules; /* List of "struct rule_dpif"s. */ + ofproto_meter_id provider_meter_id; + uint16_t flags; /* Meter flags. */ + uint16_t n_bands; /* Number of meter bands. */ + struct ofputil_meter_band *bands; +}; + +/* + * This is used in instruction validation at flow set-up time, + * as flows may not use non-existing meters. + * This is also used by ofproto-providers to translate OpenFlow meter_ids + * in METER instructions to the corresponding provider meter IDs. + * Return value of UINT32_MAX signifies an invalid meter. + */ +uint32_t +ofproto_get_provider_meter_id(const struct ofproto * ofproto, + uint32_t of_meter_id) +{ + if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) { + const struct meter *meter = ofproto->meters[of_meter_id]; + if (meter) { + return meter->provider_meter_id.uint32; + } + } + return UINT32_MAX; +} + +static void +meter_update(struct meter *meter, const struct ofputil_meter_config *config) +{ + free(meter->bands); + + meter->flags = config->flags; + meter->n_bands = config->n_bands; + meter->bands = xmemdup(config->bands, + config->n_bands * sizeof *meter->bands); +} + +static struct meter * +meter_create(const struct ofputil_meter_config *config, + ofproto_meter_id provider_meter_id) +{ + struct meter *meter; + + meter = xzalloc(sizeof *meter); + meter->provider_meter_id = provider_meter_id; + meter->created = time_msec(); + list_init(&meter->rules); + + meter_update(meter, config); + + return meter; +} + +static void +meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last) +{ + uint32_t mid; + for (mid = first; mid <= last; ++mid) { + struct meter *meter = ofproto->meters[mid]; + if (meter) { + ofproto->meters[mid] = NULL; + ofproto->ofproto_class->meter_del(ofproto, + meter->provider_meter_id); + free(meter->bands); + free(meter); + } + } +} + +static enum ofperr +handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm) +{ + ofproto_meter_id provider_meter_id = { UINT32_MAX }; + struct meter **meterp = &ofproto->meters[mm->meter.meter_id]; + enum ofperr error; + + if (*meterp) { + return OFPERR_OFPMMFC_METER_EXISTS; + } + + error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id, + &mm->meter); + if (!error) { + ovs_assert(provider_meter_id.uint32 != UINT32_MAX); + *meterp = meter_create(&mm->meter, provider_meter_id); + } + return 0; +} + +static enum ofperr +handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm) +{ + struct meter *meter = ofproto->meters[mm->meter.meter_id]; + enum ofperr error; + + if (!meter) { + return OFPERR_OFPMMFC_UNKNOWN_METER; + } + + error = ofproto->ofproto_class->meter_set(ofproto, + &meter->provider_meter_id, + &mm->meter); + ovs_assert(meter->provider_meter_id.uint32 != UINT32_MAX); + if (!error) { + meter_update(meter, &mm->meter); + } + return error; +} + +static enum ofperr +handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh, + struct ofputil_meter_mod *mm) +{ + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); + uint32_t meter_id = mm->meter.meter_id; + uint32_t first, last; + struct list rules; + + if (meter_id == OFPM13_ALL) { + first = 1; + last = ofproto->meter_features.max_meters; + } else { + if (!meter_id || meter_id > ofproto->meter_features.max_meters) { + return 0; + } + first = last = meter_id; + } + + /* First delete the rules that use this meter. If any of those rules are + * currently being modified, postpone the whole operation until later. */ + list_init(&rules); + for (meter_id = first; meter_id <= last; ++meter_id) { + struct meter *meter = ofproto->meters[meter_id]; + if (meter && !list_is_empty(&meter->rules)) { + struct rule *rule; + + LIST_FOR_EACH (rule, meter_list_node, &meter->rules) { + if (rule->pending) { + return OFPROTO_POSTPONE; + } + list_push_back(&rules, &rule->ofproto_node); + } + } + } + if (!list_is_empty(&rules)) { + delete_flows__(ofproto, ofconn, oh, &rules, OFPRR_METER_DELETE); + } + + /* Delete the meters. */ + meter_delete(ofproto, first, last); + + return 0; +} + +static enum ofperr +handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh) +{ + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); + struct ofputil_meter_mod mm; + uint64_t bands_stub[256 / 8]; + struct ofpbuf bands; + uint32_t meter_id; + enum ofperr error; + + error = reject_slave_controller(ofconn); + if (error) { + return error; + } + + ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub); + + error = ofputil_decode_meter_mod(oh, &mm, &bands); + if (error) { + goto exit_free_bands; + } + + meter_id = mm.meter.meter_id; + + if (mm.command != OFPMC13_DELETE) { + /* Fails also when meters are not implemented by the provider. */ + if (!meter_id || meter_id > ofproto->meter_features.max_meters) { + error = OFPERR_OFPMMFC_INVALID_METER; + goto exit_free_bands; + } + if (mm.meter.n_bands > ofproto->meter_features.max_bands) { + error = OFPERR_OFPMMFC_OUT_OF_BANDS; + goto exit_free_bands; + } + } + + switch (mm.command) { + case OFPMC13_ADD: + error = handle_add_meter(ofproto, &mm); + break; + + case OFPMC13_MODIFY: + error = handle_modify_meter(ofproto, &mm); + break; + + case OFPMC13_DELETE: + error = handle_delete_meter(ofconn, oh, &mm); + break; + + default: + error = OFPERR_OFPMMFC_BAD_COMMAND; + break; + } + +exit_free_bands: + ofpbuf_uninit(&bands); + return error; +} + +static enum ofperr +handle_meter_features_request(struct ofconn *ofconn, + const struct ofp_header *request) +{ + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); + struct ofputil_meter_features features; + struct ofpbuf *b; + + if (ofproto->ofproto_class->meter_get_features) { + ofproto->ofproto_class->meter_get_features(ofproto, &features); + } else { + memset(&features, 0, sizeof features); + } + b = ofputil_encode_meter_features_reply(&features, request); + + ofconn_send_reply(ofconn, b); + return 0; +} + +static enum ofperr +handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request, + enum ofptype type) +{ + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); + struct list replies; + uint64_t bands_stub[256 / 8]; + struct ofpbuf bands; + uint32_t meter_id, first, last; + + ofputil_decode_meter_request(request, &meter_id); + + if (meter_id == OFPM13_ALL) { + first = 1; + last = ofproto->meter_features.max_meters; + } else { + if (!meter_id || meter_id > ofproto->meter_features.max_meters || + !ofproto->meters[meter_id]) { + return OFPERR_OFPMMFC_UNKNOWN_METER; + } + first = last = meter_id; + } + + ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub); + ofpmp_init(&replies, request); + + for (meter_id = first; meter_id <= last; ++meter_id) { + struct meter *meter = ofproto->meters[meter_id]; + if (!meter) { + continue; /* Skip non-existing meters. */ + } + if (type == OFPTYPE_METER_STATS_REQUEST) { + struct ofputil_meter_stats stats; + + stats.meter_id = meter_id; + + /* Provider sets the packet and byte counts, we do the rest. */ + stats.flow_count = list_size(&meter->rules); + calc_duration(meter->created, time_msec(), + &stats.duration_sec, &stats.duration_nsec); + stats.n_bands = meter->n_bands; + ofpbuf_clear(&bands); + stats.bands + = ofpbuf_put_uninit(&bands, + meter->n_bands * sizeof *stats.bands); + + if (!ofproto->ofproto_class->meter_get(ofproto, + meter->provider_meter_id, + &stats)) { + ofputil_append_meter_stats(&replies, &stats); + } + } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */ + struct ofputil_meter_config config; + + config.meter_id = meter_id; + config.flags = meter->flags; + config.n_bands = meter->n_bands; + config.bands = meter->bands; + ofputil_append_meter_config(&replies, &config); + } + } + + ofconn_send_replies(ofconn, &replies); + ofpbuf_uninit(&bands); + return 0; +} + static enum ofperr handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) { @@ -4140,6 +4632,9 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) case OFPTYPE_FLOW_MOD: return handle_flow_mod(ofconn, oh); + case OFPTYPE_METER_MOD: + return handle_meter_mod(ofconn, oh); + case OFPTYPE_BARRIER_REQUEST: return handle_barrier_request(ofconn, oh); @@ -4198,17 +4693,20 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) case OFPTYPE_FLOW_MONITOR_STATS_REQUEST: return handle_flow_monitor_request(ofconn, oh); + case OFPTYPE_METER_STATS_REQUEST: + case OFPTYPE_METER_CONFIG_STATS_REQUEST: + return handle_meter_request(ofconn, oh, type); + + case OFPTYPE_METER_FEATURES_STATS_REQUEST: + return handle_meter_features_request(ofconn, oh); + /* FIXME: Change the following once they are implemented: */ case OFPTYPE_QUEUE_GET_CONFIG_REQUEST: case OFPTYPE_GET_ASYNC_REQUEST: - case OFPTYPE_METER_MOD: - case OFPTYPE_GROUP_REQUEST: - case OFPTYPE_GROUP_DESC_REQUEST: - case OFPTYPE_GROUP_FEATURES_REQUEST: - case OFPTYPE_METER_REQUEST: - case OFPTYPE_METER_CONFIG_REQUEST: - case OFPTYPE_METER_FEATURES_REQUEST: - case OFPTYPE_TABLE_FEATURES_REQUEST: + case OFPTYPE_GROUP_STATS_REQUEST: + case OFPTYPE_GROUP_DESC_STATS_REQUEST: + case OFPTYPE_GROUP_FEATURES_STATS_REQUEST: + case OFPTYPE_TABLE_FEATURES_STATS_REQUEST: return OFPERR_OFPBRC_BAD_TYPE; case OFPTYPE_HELLO: @@ -4232,13 +4730,13 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) case OFPTYPE_FLOW_MONITOR_RESUMED: case OFPTYPE_FLOW_MONITOR_STATS_REPLY: case OFPTYPE_GET_ASYNC_REPLY: - case OFPTYPE_GROUP_REPLY: - case OFPTYPE_GROUP_DESC_REPLY: - case OFPTYPE_GROUP_FEATURES_REPLY: - case OFPTYPE_METER_REPLY: - case OFPTYPE_METER_CONFIG_REPLY: - case OFPTYPE_METER_FEATURES_REPLY: - case OFPTYPE_TABLE_FEATURES_REPLY: + case OFPTYPE_GROUP_STATS_REPLY: + case OFPTYPE_GROUP_DESC_STATS_REPLY: + case OFPTYPE_GROUP_FEATURES_STATS_REPLY: + case OFPTYPE_METER_STATS_REPLY: + case OFPTYPE_METER_CONFIG_STATS_REPLY: + case OFPTYPE_METER_FEATURES_STATS_REPLY: + case OFPTYPE_TABLE_FEATURES_STATS_REPLY: default: return OFPERR_OFPBRC_BAD_TYPE; } @@ -4586,7 +5084,7 @@ pick_datapath_id(const struct ofproto *ofproto) } VLOG_WARN("%s: could not get MAC address for %s (%s)", ofproto->name, netdev_get_name(port->netdev), - strerror(error)); + ovs_strerror(error)); } return ofproto->fallback_dpid; } @@ -4601,17 +5099,18 @@ pick_fallback_dpid(void) /* Table overflow policy. */ -/* Chooses and returns a rule to evict from 'table'. Returns NULL if the table - * is not configured to evict rules or if the table contains no evictable - * rules. (Rules with 'evictable' set to false or with no timeouts are not - * evictable.) */ -static struct rule * -choose_rule_to_evict(struct oftable *table) +/* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep' + * to NULL if the table is not configured to evict rules or if the table + * contains no evictable rules. (Rules with a readlock on their evict rwlock, + * or with no timeouts are not evictable.) */ +static bool +choose_rule_to_evict(struct oftable *table, struct rule **rulep) { struct eviction_group *evg; + *rulep = NULL; if (!table->eviction_fields) { - return NULL; + return false; } /* In the common case, the outer and inner loops here will each be entered @@ -4625,18 +5124,19 @@ choose_rule_to_evict(struct oftable *table) * group has no evictable rules. * * - The outer loop can exit only if table's 'max_flows' is all filled up - * by unevictable rules'. */ + * by unevictable rules. */ HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) { struct rule *rule; HEAP_FOR_EACH (rule, evg_node, &evg->rules) { - if (rule->evictable) { - return rule; + if (!ovs_rwlock_trywrlock(&rule->evict)) { + *rulep = rule; + return true; } } } - return NULL; + return false; } /* Searches 'ofproto' for tables that have more flows than their configured @@ -4653,12 +5153,24 @@ ofproto_evict(struct ofproto *ofproto) group = ofopgroup_create_unattached(ofproto); OFPROTO_FOR_EACH_TABLE (table, ofproto) { - while (classifier_count(&table->cls) > table->max_flows - && table->eviction_fields) { + while (table->eviction_fields) { struct rule *rule; + size_t n_rules; + + ovs_rwlock_rdlock(&table->cls.rwlock); + n_rules = classifier_count(&table->cls); + ovs_rwlock_unlock(&table->cls.rwlock); + + if (n_rules <= table->max_flows) { + break; + } - rule = choose_rule_to_evict(table); - if (!rule || rule->pending) { + if (!choose_rule_to_evict(table, &rule)) { + break; + } + + if (rule->pending) { + ovs_rwlock_unlock(&rule->evict); break; } @@ -4802,6 +5314,7 @@ rule_eviction_priority(struct rule *rule) uint32_t expiration_offset; /* Calculate time of expiration. */ + ovs_mutex_lock(&rule->timeout_mutex); hard_expiration = (rule->hard_timeout ? rule->modified + rule->hard_timeout * 1000 : LLONG_MAX); @@ -4809,6 +5322,7 @@ rule_eviction_priority(struct rule *rule) ? rule->used + rule->idle_timeout * 1000 : LLONG_MAX); expiration = MIN(hard_expiration, idle_expiration); + ovs_mutex_unlock(&rule->timeout_mutex); if (expiration == LLONG_MAX) { return 0; } @@ -4835,9 +5349,13 @@ eviction_group_add_rule(struct rule *rule) { struct ofproto *ofproto = rule->ofproto; struct oftable *table = &ofproto->tables[rule->table_id]; + bool has_timeout; - if (table->eviction_fields - && (rule->hard_timeout || rule->idle_timeout)) { + ovs_mutex_lock(&rule->timeout_mutex); + has_timeout = rule->hard_timeout || rule->idle_timeout; + ovs_mutex_unlock(&rule->timeout_mutex); + + if (table->eviction_fields && has_timeout) { struct eviction_group *evg; evg = eviction_group_find(table, eviction_group_hash_rule(rule)); @@ -4866,7 +5384,9 @@ oftable_init(struct oftable *table) static void oftable_destroy(struct oftable *table) { + ovs_rwlock_rdlock(&table->cls.rwlock); ovs_assert(classifier_is_empty(&table->cls)); + ovs_rwlock_unlock(&table->cls.rwlock); oftable_disable_eviction(table); classifier_destroy(&table->cls); free(table->name); @@ -4946,25 +5466,46 @@ oftable_enable_eviction(struct oftable *table, hmap_init(&table->eviction_groups_by_id); heap_init(&table->eviction_groups_by_size); + ovs_rwlock_rdlock(&table->cls.rwlock); cls_cursor_init(&cursor, &table->cls, NULL); CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { eviction_group_add_rule(rule); } + ovs_rwlock_unlock(&table->cls.rwlock); } /* Removes 'rule' from the oftable that contains it. */ static void -oftable_remove_rule(struct rule *rule) +oftable_remove_rule__(struct ofproto *ofproto, struct classifier *cls, + struct rule *rule) + OVS_REQ_WRLOCK(cls->rwlock) OVS_RELEASES(rule->evict) { - struct ofproto *ofproto = rule->ofproto; - struct oftable *table = &ofproto->tables[rule->table_id]; - - classifier_remove(&table->cls, &rule->cr); + classifier_remove(cls, &rule->cr); + if (rule->meter_id) { + list_remove(&rule->meter_list_node); + } cookies_remove(ofproto, rule); eviction_group_remove_rule(rule); + ovs_mutex_lock(&ofproto->expirable_mutex); if (!list_is_empty(&rule->expirable)) { list_remove(&rule->expirable); } + ovs_mutex_unlock(&ofproto->expirable_mutex); + if (!list_is_empty(&rule->meter_list_node)) { + list_remove(&rule->meter_list_node); + } + ovs_rwlock_unlock(&rule->evict); +} + +static void +oftable_remove_rule(struct rule *rule) +{ + struct ofproto *ofproto = rule->ofproto; + struct oftable *table = &ofproto->tables[rule->table_id]; + + ovs_rwlock_wrlock(&table->cls.rwlock); + oftable_remove_rule__(ofproto, &table->cls, rule); + ovs_rwlock_unlock(&table->cls.rwlock); } /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s @@ -4976,20 +5517,36 @@ 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; + bool may_expire; + + ovs_mutex_lock(&rule->timeout_mutex); + may_expire = rule->hard_timeout || rule->idle_timeout; + ovs_mutex_unlock(&rule->timeout_mutex); if (may_expire) { + ovs_mutex_lock(&ofproto->expirable_mutex); list_insert(&ofproto->expirable, &rule->expirable); + ovs_mutex_unlock(&ofproto->expirable_mutex); } cookies_insert(ofproto, rule); - + if (rule->meter_id) { + struct meter *meter = ofproto->meters[rule->meter_id]; + list_insert(&meter->rules, &rule->meter_list_node); + } + ovs_rwlock_wrlock(&table->cls.rwlock); victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr)); + ovs_rwlock_unlock(&table->cls.rwlock); if (victim) { + if (victim->meter_id) { + list_remove(&victim->meter_list_node); + } cookies_remove(ofproto, victim); + ovs_mutex_lock(&ofproto->expirable_mutex); if (!list_is_empty(&victim->expirable)) { list_remove(&victim->expirable); } + ovs_mutex_unlock(&ofproto->expirable_mutex); eviction_group_remove_rule(victim); } eviction_group_add_rule(rule); @@ -5003,6 +5560,7 @@ oftable_substitute_rule(struct rule *old, struct rule *new) if (new) { oftable_replace_rule(new); } else { + ovs_rwlock_wrlock(&old->evict); oftable_remove_rule(old); } } @@ -5129,7 +5687,7 @@ ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port, if (error) { VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)", ofproto->name, vlandev_ofp_port, - netdev_get_name(ofport->netdev), strerror(error)); + netdev_get_name(ofport->netdev), ovs_strerror(error)); } return error; }