X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=ofproto%2Fofproto.c;h=abf14f2952954f3b9c1095df54c746c7bb78b145;hb=cc86b1f84b1e2fe1b322a23a2b3b88c0375f2fa1;hp=a85a6816cdd8a38aa90681e45bff2f9019adde91;hpb=4f9e08a539f26b008f32a33b86ef8c46f5c5bab9;p=sliver-openvswitch.git diff --git a/ofproto/ofproto.c b/ofproto/ofproto.c index a85a6816c..abf14f295 100644 --- a/ofproto/ofproto.c +++ b/ofproto/ofproto.c @@ -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"); @@ -21,6 +21,7 @@ #include #include #include +#include #include "bitmap.h" #include "byte-order.h" #include "classifier.h" @@ -75,7 +76,8 @@ enum ofproto_state { enum ofoperation_type { OFOPERATION_ADD, OFOPERATION_DELETE, - OFOPERATION_MODIFY + OFOPERATION_MODIFY, + OFOPERATION_REPLACE }; /* A single OpenFlow request can execute any number of operations. The @@ -120,17 +122,19 @@ struct ofoperation { struct rule *rule; /* Rule being operated upon. */ enum ofoperation_type type; /* Type of operation. */ - /* OFOPERATION_ADD. */ - struct rule *victim; /* Rule being replaced, if any.. */ - - /* OFOPERATION_MODIFY: The old actions, if the actions are changing. */ + /* OFOPERATION_MODIFY, OFOPERATION_REPLACE: 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. */ ovs_be64 flow_cookie; /* Rule's old flow cookie. */ + uint16_t idle_timeout; /* Rule's old idle timeout. */ + uint16_t hard_timeout; /* Rule's old hard timeout. */ + bool send_flow_removed; /* Rule's old 'send_flow_removed'. */ enum ofperr error; /* 0 if no error. */ }; @@ -151,9 +155,11 @@ static void oftable_enable_eviction(struct oftable *, const struct mf_subfield *fields, size_t n_fields); -static void oftable_remove_rule(struct rule *); -static struct rule *oftable_replace_rule(struct rule *); -static void oftable_substitute_rule(struct rule *old, struct rule *new); +static void oftable_remove_rule(struct rule *rule) OVS_RELEASES(rule->rwlock); +static void oftable_remove_rule__(struct ofproto *ofproto, + struct classifier *cls, struct rule *rule) + OVS_REQ_WRLOCK(cls->rwlock) OVS_RELEASES(rule->rwlock); +static void oftable_insert_rule(struct rule *); /* A set of rules within a single OpenFlow table (oftable) that have the same * values for the oftable's eviction_fields. A rule to be evicted, when one is @@ -177,9 +183,12 @@ 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)->rwlock); static void ofproto_evict(struct ofproto *); static uint32_t rule_eviction_priority(struct rule *); +static void eviction_group_add_rule(struct rule *); +static void eviction_group_remove_rule(struct rule *); /* ofport. */ static void ofport_destroy__(struct ofport *); @@ -190,25 +199,35 @@ static int init_ports(struct ofproto *); static void reinit_ports(struct ofproto *); /* rule. */ +static void ofproto_rule_destroy(struct rule *); static void ofproto_rule_destroy__(struct rule *); static void ofproto_rule_send_removed(struct rule *, uint8_t reason); 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 bool handle_openflow(struct ofconn *, struct ofpbuf *); +static enum ofperr modify_flows__(struct ofproto *, struct ofconn *, + struct ofputil_flow_mod *, + const struct ofp_header *, struct list *); +static void delete_flow__(struct rule *rule, struct ofopgroup *, + enum ofp_flow_removed_reason) + OVS_RELEASES(rule->rwlock); +static enum ofperr add_group(struct ofproto *, struct ofputil_group_mod *); +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); /* ofproto. */ 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); @@ -218,6 +237,10 @@ static const struct ofproto_class **ofproto_classes; 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); @@ -226,6 +249,9 @@ static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports); static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); +/* The default value of true waits for flow restore. */ +static bool flow_restore_wait = true; + /* Must be called to initialize the ofproto library. * * The caller may pass in 'iface_hints', which contains an shash of @@ -403,22 +429,24 @@ ofproto_create(const char *datapath_name, const char *datapath_type, hmap_insert(&all_ofprotos, &ofproto->hmap_node, hash_string(ofproto->name, 0)); ofproto->datapath_id = 0; - ofproto_set_flow_eviction_threshold(ofproto, - 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); simap_init(&ofproto->ofp_requests); - ofproto->max_ports = OFPP_MAX; + ofproto->max_ports = ofp_to_u16(OFPP_MAX); + ofproto->eviction_group_timer = LLONG_MIN; 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); @@ -431,11 +459,13 @@ ofproto_create(const char *datapath_name, const char *datapath_type, ofproto->vlan_bitmap = NULL; ofproto->vlans_changed = false; ofproto->min_mtu = INT_MAX; + ovs_rwlock_init(&ofproto->groups_rwlock); + hmap_init(&ofproto->groups); 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; } @@ -446,17 +476,27 @@ 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); 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; } @@ -469,8 +509,8 @@ 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 +533,7 @@ 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 <= ofp_to_u16(OFPP_MAX)); ofproto->max_ports = max_ports; } @@ -560,13 +600,17 @@ ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id) /* Sets the number of flows at which eviction from the kernel flow table * will occur. */ void -ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold) +ofproto_set_flow_eviction_threshold(unsigned threshold) { - if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) { - ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN; - } else { - ofproto->flow_eviction_threshold = threshold; - } + flow_eviction_threshold = MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN, + 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 @@ -597,56 +641,26 @@ ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time, } } +/* Sets number of upcall handler threads. The default is + * (number of online cores - 2). */ 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_n_handler_threads(unsigned limit) { - 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); + if (limit) { + n_handler_threads = limit; + } else { + int n_proc = sysconf(_SC_NPROCESSORS_ONLN); + n_handler_threads = n_proc > 2 ? n_proc - 2 : 1; } } +void +ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc) +{ + free(p->dp_desc); + p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL; +} + int ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops) { @@ -682,6 +696,32 @@ 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; + } +} + +void +ofproto_set_flow_restore_wait(bool flow_restore_wait_db) +{ + flow_restore_wait = flow_restore_wait_db; +} + +bool +ofproto_get_flow_restore_wait(void) +{ + return flow_restore_wait; +} + /* Spanning Tree Protocol (STP) configuration. */ @@ -720,7 +760,7 @@ ofproto_get_stp_status(struct ofproto *ofproto, * * Returns 0 if successful, otherwise a positive errno value.*/ int -ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port, +ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port, const struct ofproto_port_stp_settings *s) { struct ofport *ofport = ofproto_get_port(ofproto, ofp_port); @@ -741,7 +781,7 @@ ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port, * * Returns 0 if successful, otherwise a positive errno value.*/ int -ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port, +ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port, struct ofproto_port_stp_status *s) { struct ofport *ofport = ofproto_get_port(ofproto, ofp_port); @@ -766,7 +806,7 @@ ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port, * * Returns 0 if successful, otherwise a positive errno value. */ int -ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port, +ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port, const struct ofproto_port_queue *queues, size_t n_queues) { @@ -787,7 +827,7 @@ ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port, /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */ void -ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port) +ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port) { struct ofport *ofport = ofproto_get_port(ofproto, ofp_port); if (ofport && ofproto->ofproto_class->set_cfm) { @@ -802,7 +842,7 @@ ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port) * * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */ void -ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port, +ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port, const struct cfm_settings *s) { struct ofport *ofport; @@ -824,16 +864,56 @@ ofproto_port_set_cfm(struct ofproto *ofproto, uint16_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)); + } +} + +/* 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, ofp_port_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); + return; + } + + 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), + ovs_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, ofp_port_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 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */ int -ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port) +ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port) { struct ofport *ofport = ofproto_get_port(ofproto, ofp_port); return (ofport && ofproto->ofproto_class->port_is_lacp_current @@ -953,7 +1033,7 @@ 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); @@ -969,6 +1049,7 @@ 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 @@ -984,6 +1065,7 @@ ofproto_configure_table(struct ofproto *ofproto, int table_id, break; } } + ovs_rwlock_unlock(&table->cls.rwlock); } bool @@ -998,17 +1080,37 @@ ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops) connmgr_get_snoops(ofproto->connmgr, snoops); } +/* Deletes 'rule' from 'cls' within 'ofproto'. + * + * The 'cls' argument is redundant (it is &ofproto->tables[rule->table_id].cls) + * but it allows Clang to do better checking. */ static void -ofproto_flush__(struct ofproto *ofproto) +ofproto_delete_rule(struct ofproto *ofproto, struct classifier *cls, + struct rule *rule) + OVS_REQ_WRLOCK(cls->rwlock) { struct ofopgroup *group; + + ovs_assert(!rule->pending); + ovs_assert(cls == &ofproto->tables[rule->table_id].cls); + + group = ofopgroup_create_unattached(ofproto); + ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE); + ovs_rwlock_wrlock(&rule->rwlock); + oftable_remove_rule__(ofproto, cls, rule); + ofproto->ofproto_class->rule_delete(rule); + ofopgroup_submit(group); +} + +static void +ofproto_flush__(struct ofproto *ofproto) +{ struct oftable *table; if (ofproto->ofproto_class->flush) { ofproto->ofproto_class->flush(ofproto); } - group = ofopgroup_create_unattached(ofproto); OFPROTO_FOR_EACH_TABLE (table, ofproto) { struct rule *rule, *next_rule; struct cls_cursor cursor; @@ -1017,26 +1119,35 @@ 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); - ofproto->ofproto_class->rule_destruct(rule); + ofproto_delete_rule(ofproto, &table->cls, rule); } } + ovs_rwlock_unlock(&table->cls.rwlock); } - ofopgroup_submit(group); } +static void delete_group(struct ofproto *ofproto, uint32_t group_id); + static void 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); + + if (ofproto->meters) { + meter_delete(ofproto, 1, ofproto->meter_features.max_meters); + free(ofproto->meters); + } + + delete_group(ofproto, OFPG_ALL); + ovs_rwlock_destroy(&ofproto->groups_rwlock); + hmap_destroy(&ofproto->groups); connmgr_destroy(ofproto->connmgr); @@ -1062,6 +1173,7 @@ ofproto_destroy__(struct ofproto *ofproto) free(ofproto->vlan_bitmap); + ovs_mutex_destroy(&ofproto->expirable_mutex); ofproto->ofproto_class->dealloc(ofproto); } @@ -1121,7 +1233,7 @@ 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; } @@ -1138,7 +1250,7 @@ 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; } @@ -1166,7 +1278,39 @@ 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)); + } + + /* Restore the eviction group heap invariant occasionally. */ + if (p->eviction_group_timer < time_msec()) { + size_t i; + + p->eviction_group_timer = time_msec() + 1000; + + for (i = 0; i < p->n_tables; i++) { + struct oftable *table = &p->tables[i]; + struct eviction_group *evg; + struct cls_cursor cursor; + struct rule *rule; + + if (!table->eviction_fields) { + continue; + } + + HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) { + heap_rebuild(&evg->rules); + } + + ovs_rwlock_rdlock(&table->cls.rwlock); + cls_cursor_init(&cursor, &table->cls, NULL); + CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { + if (!rule->eviction_group + && (rule->idle_timeout || rule->hard_timeout)) { + eviction_group_add_rule(rule); + } + } + ovs_rwlock_unlock(&table->cls.rwlock); + } } if (p->ofproto_class->port_poll) { @@ -1276,7 +1420,7 @@ 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; } @@ -1332,7 +1476,9 @@ 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); @@ -1470,16 +1616,17 @@ ofproto_port_open_type(const char *datapath_type, const char *port_type) * 'ofp_portp' is non-null). */ int ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev, - uint16_t *ofp_portp) + ofp_port_t *ofp_portp) { - uint16_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE; + ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE; int error; error = ofproto->ofproto_class->port_add(ofproto, netdev); if (!error) { const char *netdev_name = netdev_get_name(netdev); - simap_put(&ofproto->ofp_requests, netdev_name, ofp_port); + simap_put(&ofproto->ofp_requests, netdev_name, + ofp_to_u16(ofp_port)); update_port(ofproto, netdev_name); } if (ofp_portp) { @@ -1515,7 +1662,7 @@ ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname, /* Deletes port number 'ofp_port' from the datapath for 'ofproto'. * Returns 0 if successful, otherwise a positive errno. */ int -ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port) +ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port) { struct ofport *ofport = ofproto_get_port(ofproto, ofp_port); const char *name = ofport ? netdev_get_name(ofport->netdev) : ""; @@ -1558,8 +1705,10 @@ 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; @@ -1581,7 +1730,7 @@ 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); } @@ -1594,10 +1743,13 @@ bool ofproto_delete_flow(struct ofproto *ofproto, const struct match *target, unsigned int priority) { + struct classifier *cls = &ofproto->tables[0].cls; struct rule *rule; - rule = rule_from_cls_rule(classifier_find_match_exactly( - &ofproto->tables[0].cls, target, priority)); + ovs_rwlock_rdlock(&cls->rwlock); + rule = rule_from_cls_rule(classifier_find_match_exactly(cls, target, + priority)); + ovs_rwlock_unlock(&cls->rwlock); if (!rule) { /* No such rule -> success. */ return true; @@ -1607,11 +1759,10 @@ ofproto_delete_flow(struct ofproto *ofproto, return false; } else { /* Initiate deletion -> success. */ - struct ofopgroup *group = ofopgroup_create_unattached(ofproto); - ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE); - oftable_remove_rule(rule); - ofproto->ofproto_class->rule_destruct(rule); - ofopgroup_submit(group); + ovs_rwlock_wrlock(&cls->rwlock); + ofproto_delete_rule(ofproto, cls, rule); + ovs_rwlock_unlock(&cls->rwlock); + return true; } @@ -1652,50 +1803,45 @@ reinit_ports(struct ofproto *p) sset_destroy(&devnames); } -static uint16_t +static ofp_port_t alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name) { - uint16_t ofp_port; + uint16_t port_idx; - ofp_port = simap_get(&ofproto->ofp_requests, netdev_name); - ofp_port = ofp_port ? ofp_port : OFPP_NONE; + port_idx = simap_get(&ofproto->ofp_requests, netdev_name); + port_idx = port_idx ? port_idx : UINT16_MAX; - if (ofp_port >= ofproto->max_ports - || bitmap_is_set(ofproto->ofp_port_ids, ofp_port)) { - bool retry = ofproto->alloc_port_no ? true : false; + if (port_idx >= ofproto->max_ports + || bitmap_is_set(ofproto->ofp_port_ids, port_idx)) { + 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. */ - 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)) { + port_idx = 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; + bitmap_set1(ofproto->ofp_port_ids, port_idx); + return u16_to_ofp(port_idx); } static void -dealloc_ofp_port(const struct ofproto *ofproto, uint16_t ofp_port) +dealloc_ofp_port(const struct ofproto *ofproto, ofp_port_t ofp_port) { - bitmap_set0(ofproto->ofp_port_ids, ofp_port); + if (ofp_to_u16(ofp_port) < ofproto->max_ports) { + bitmap_set0(ofproto->ofp_port_ids, ofp_to_u16(ofp_port)); + } } /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null @@ -1716,7 +1862,7 @@ 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; } @@ -1736,8 +1882,8 @@ ofport_open(struct ofproto *ofproto, pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN; netdev_get_features(netdev, &pp->curr, &pp->advertised, &pp->supported, &pp->peer); - pp->curr_speed = netdev_features_to_bps(pp->curr, 0); - pp->max_speed = netdev_features_to_bps(pp->supported, 0); + pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000; + pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000; return netdev; } @@ -1782,9 +1928,11 @@ ofport_install(struct ofproto *p, ofport->change_seq = netdev_change_seq(netdev); ofport->pp = *pp; ofport->ofp_port = pp->port_no; + ofport->created = time_msec(); /* Add port to 'p'. */ - hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0)); + hmap_insert(&p->ports, &ofport->hmap_node, + hash_ofp_port(ofport->ofp_port)); shash_add(&p->port_by_name, netdev_name, ofport); update_mtu(p, ofport); @@ -1799,7 +1947,7 @@ 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 { @@ -1860,7 +2008,7 @@ ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state) } void -ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port) +ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port) { struct ofport *port = ofproto_get_port(ofproto, ofp_port); if (port) { @@ -1904,12 +2052,12 @@ ofport_destroy(struct ofport *port) } struct ofport * -ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port) +ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port) { struct ofport *port; - HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, - hash_int(ofp_port, 0), &ofproto->ports) { + HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port), + &ofproto->ports) { if (port->ofp_port == ofp_port) { return port; } @@ -1946,6 +2094,7 @@ update_port(struct ofproto *ofproto, const char *name) netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port) ? ofport_open(ofproto, &ofproto_port, &pp) : NULL); + if (netdev) { port = ofproto_get_port(ofproto, ofproto_port.ofp_port); if (port && !strcmp(netdev_get_name(port->netdev), name)) { @@ -2007,12 +2156,17 @@ init_ports(struct ofproto *p) node = shash_find(&init_ofp_ports, name); if (node) { const struct iface_hint *iface_hint = node->data; - simap_put(&p->ofp_requests, name, iface_hint->ofp_port); + simap_put(&p->ofp_requests, name, + ofp_to_u16(iface_hint->ofp_port)); } 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)); + } } } } @@ -2103,44 +2257,60 @@ update_mtu(struct ofproto *p, struct ofport *port) } static void -ofproto_rule_destroy__(struct rule *rule) +ofproto_rule_destroy(struct rule *rule) { if (rule) { - cls_rule_destroy(&rule->cr); - free(rule->ofpacts); - rule->ofproto->ofproto_class->rule_dealloc(rule); + rule->ofproto->ofproto_class->rule_destruct(rule); + ofproto_rule_destroy__(rule); } } +static void +ofproto_rule_destroy__(struct rule *rule) +{ + cls_rule_destroy(&rule->cr); + free(rule->ofpacts); + ovs_mutex_destroy(&rule->timeout_mutex); + ovs_rwlock_destroy(&rule->rwlock); + rule->ofproto->ofproto_class->rule_dealloc(rule); +} + /* This function allows an ofproto implementation to destroy any rules that - * remain when its ->destruct() function is called. The caller must have - * already uninitialized any derived members of 'rule' (step 5 described in the - * large comment in ofproto/ofproto-provider.h titled "Life Cycle"). - * This function implements steps 6 and 7. + * remain when its ->destruct() function is called.. This function implements + * steps 4.4 and 4.5 in the section titled "Rule Life Cycle" in + * ofproto-provider.h. * * 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_delete(struct ofproto *ofproto, struct classifier *cls, + struct rule *rule) + OVS_REQ_WRLOCK(cls->rwlock) { - assert(!rule->pending); - oftable_remove_rule(rule); - ofproto_rule_destroy__(rule); + ofproto_delete_rule(ofproto, cls, rule); } /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */ bool -ofproto_rule_has_out_port(const struct rule *rule, uint16_t port) +ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port) { return (port == OFPP_ANY || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port)); } +/* Returns true if 'rule' has group and equals group_id. */ +bool +ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id) +{ + return (group_id == OFPG11_ANY + || ofpacts_output_to_group(rule->ofpacts, rule->ofpacts_len, group_id)); +} + /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or * OFPAT_ENQUEUE action that outputs to 'out_port'. */ bool -ofoperation_has_out_port(const struct ofoperation *op, uint16_t out_port) +ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port) { if (ofproto_rule_has_out_port(op->rule, out_port)) { return true; @@ -2148,12 +2318,11 @@ ofoperation_has_out_port(const struct ofoperation *op, uint16_t out_port) switch (op->type) { case OFOPERATION_ADD: - return op->victim && ofproto_rule_has_out_port(op->victim, out_port); - case OFOPERATION_DELETE: return false; case OFOPERATION_MODIFY: + case OFOPERATION_REPLACE: return ofpacts_output_to_port(op->ofpacts, op->ofpacts_len, out_port); } @@ -2161,21 +2330,20 @@ ofoperation_has_out_port(const struct ofoperation *op, uint16_t out_port) } /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s - * statistics appropriately. 'packet' must have at least sizeof(struct - * ofp10_packet_in) bytes of headroom. + * statistics appropriately. * * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited * with statistics for 'packet' either way. * * Takes ownership of 'packet'. */ static int -rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet) +rule_execute(struct rule *rule, ofp_port_t in_port, struct ofpbuf *packet) { struct flow flow; + union flow_in_port in_port_; - assert(ofpbuf_headroom(packet) >= sizeof(struct ofp10_packet_in)); - - flow_extract(packet, 0, 0, NULL, in_port, &flow); + in_port_.ofp_port = in_port; + flow_extract(packet, 0, 0, NULL, &in_port_, &flow); return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet); } @@ -2222,7 +2390,7 @@ 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 +2454,11 @@ 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; @@ -2320,13 +2488,62 @@ 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; } } +/* 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) { @@ -2336,6 +2553,7 @@ handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh) uint64_t ofpacts_stub[1024 / 8]; struct ofpbuf ofpacts; struct flow flow; + union flow_in_port in_port_; enum ofperr error; COVERAGE_INC(ofproto_packet_out); @@ -2351,7 +2569,8 @@ handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh) if (error) { goto exit_free_ofpacts; } - if (po.in_port >= p->max_ports && po.in_port < OFPP_MAX) { + 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; } @@ -2364,13 +2583,14 @@ handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh) goto exit_free_ofpacts; } } else { - payload = xmalloc(sizeof *payload); - ofpbuf_use_const(payload, po.packet, po.packet_len); + /* Ensure that the L3 header is 32-bit aligned. */ + payload = ofpbuf_clone_data_with_headroom(po.packet, po.packet_len, 2); } /* Verify actions against packet, then send packet if successful. */ - flow_extract(payload, 0, 0, NULL, po.in_port, &flow); - error = ofpacts_check(po.ofpacts, po.ofpacts_len, &flow, p->max_ports); + in_port_.ofp_port = po.in_port; + flow_extract(payload, 0, 0, NULL, &in_port_, &flow); + 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); @@ -2394,9 +2614,9 @@ update_port_config(struct ofport *port, toggle = (config ^ port->pp.config) & mask; if (toggle & OFPUTIL_PC_PORT_DOWN) { if (config & OFPUTIL_PC_PORT_DOWN) { - netdev_turn_flags_off(port->netdev, NETDEV_UP, true); + netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL); } else { - netdev_turn_flags_on(port->netdev, NETDEV_UP, true); + netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL); } toggle &= ~OFPUTIL_PC_PORT_DOWN; } @@ -2443,17 +2663,29 @@ 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; @@ -2490,7 +2722,9 @@ 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); @@ -2527,6 +2761,9 @@ append_port_stat(struct ofport *port, struct list *replies) { struct ofputil_port_stats ops = { .port_no = port->pp.port_no }; + calc_duration(port->created, time_msec(), + &ops.duration_sec, &ops.duration_nsec); + /* Intentionally ignore return value, since errors will set * 'stats' to all-1s, which is correct for OpenFlow, and * netdev_get_stats() will log errors. */ @@ -2542,7 +2779,7 @@ handle_port_stats_request(struct ofconn *ofconn, struct ofproto *p = ofconn_get_ofproto(ofconn); struct ofport *port; struct list replies; - uint16_t port_no; + ofp_port_t port_no; enum ofperr error; error = ofputil_decode_port_stats_request(request, &port_no); @@ -2586,9 +2823,44 @@ handle_port_desc_stats_request(struct ofconn *ofconn, return 0; } +static uint32_t +hash_cookie(ovs_be64 cookie) +{ + return hash_2words((OVS_FORCE uint64_t)cookie >> 32, + (OVS_FORCE uint64_t)cookie); +} + +static void +cookies_insert(struct ofproto *ofproto, struct rule *rule) +{ + hindex_insert(&ofproto->cookies, &rule->cookie_node, + hash_cookie(rule->flow_cookie)); +} + static void -calc_flow_duration__(long long int start, long long int now, - uint32_t *sec, uint32_t *nsec) +cookies_remove(struct ofproto *ofproto, struct rule *rule) +{ + hindex_remove(&ofproto->cookies, &rule->cookie_node); +} + +static void +ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule, + ovs_be64 new_cookie) +{ + if (new_cookie != rule->flow_cookie) { + cookies_remove(ofproto, rule); + + ovs_rwlock_wrlock(&rule->rwlock); + rule->flow_cookie = new_cookie; + ovs_rwlock_unlock(&rule->rwlock); + + cookies_insert(ofproto, rule); + } +} + +static void +calc_duration(long long int start, long long int now, + uint32_t *sec, uint32_t *nsec) { long long int msecs = now - start; *sec = msecs / 1000; @@ -2678,7 +2950,8 @@ static enum ofperr collect_rules_loose(struct ofproto *ofproto, uint8_t table_id, const struct match *match, ovs_be64 cookie, ovs_be64 cookie_mask, - uint16_t out_port, struct list *rules) + ofp_port_t out_port, uint32_t out_group, + struct list *rules) { struct oftable *table; struct cls_rule cr; @@ -2691,22 +2964,53 @@ collect_rules_loose(struct ofproto *ofproto, uint8_t table_id, list_init(rules); cls_rule_init(&cr, match, 0); + + if (cookie_mask == htonll(UINT64_MAX)) { + struct rule *rule; + + HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, hash_cookie(cookie), + &ofproto->cookies) { + if (table_id != rule->table_id && table_id != 0xff) { + continue; + } + if (ofproto_rule_is_hidden(rule)) { + continue; + } + if (cls_rule_is_loose_match(&rule->cr, &cr.match)) { + if (rule->pending) { + error = OFPROTO_POSTPONE; + goto exit; + } + if (rule->flow_cookie == cookie /* Hash collisions possible. */ + && ofproto_rule_has_out_port(rule, out_port) + && ofproto_rule_has_out_group(rule, out_group)) { + list_push_back(rules, &rule->ofproto_node); + } + } + } + goto exit; + } + FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) { 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; } if (!ofproto_rule_is_hidden(rule) && ofproto_rule_has_out_port(rule, out_port) + && ofproto_rule_has_out_group(rule, out_group) && !((rule->flow_cookie ^ cookie) & cookie_mask)) { list_push_back(rules, &rule->ofproto_node); } } + ovs_rwlock_unlock(&table->cls.rwlock); } exit: @@ -2729,7 +3033,8 @@ static enum ofperr collect_rules_strict(struct ofproto *ofproto, uint8_t table_id, const struct match *match, unsigned int priority, ovs_be64 cookie, ovs_be64 cookie_mask, - uint16_t out_port, struct list *rules) + ofp_port_t out_port, uint32_t out_group, + struct list *rules) { struct oftable *table; struct cls_rule cr; @@ -2742,11 +3047,40 @@ collect_rules_strict(struct ofproto *ofproto, uint8_t table_id, list_init(rules); cls_rule_init(&cr, match, priority); + + if (cookie_mask == htonll(UINT64_MAX)) { + struct rule *rule; + + HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, hash_cookie(cookie), + &ofproto->cookies) { + if (table_id != rule->table_id && table_id != 0xff) { + continue; + } + if (ofproto_rule_is_hidden(rule)) { + continue; + } + if (cls_rule_equal(&rule->cr, &cr)) { + if (rule->pending) { + error = OFPROTO_POSTPONE; + goto exit; + } + if (rule->flow_cookie == cookie /* Hash collisions possible. */ + && ofproto_rule_has_out_port(rule, out_port) + && ofproto_rule_has_out_group(rule, out_group)) { + list_push_back(rules, &rule->ofproto_node); + } + } + } + goto exit; + } + 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; @@ -2754,6 +3088,7 @@ collect_rules_strict(struct ofproto *ofproto, uint8_t table_id, } if (!ofproto_rule_is_hidden(rule) && ofproto_rule_has_out_port(rule, out_port) + && ofproto_rule_has_out_group(rule, out_group) && !((rule->flow_cookie ^ cookie) & cookie_mask)) { list_push_back(rules, &rule->ofproto_node); } @@ -2793,7 +3128,7 @@ handle_flow_stats_request(struct ofconn *ofconn, error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match, fsr.cookie, fsr.cookie_mask, - fsr.out_port, &rules); + fsr.out_port, fsr.out_group, &rules); if (error) { return error; } @@ -2807,21 +3142,24 @@ handle_flow_stats_request(struct ofconn *ofconn, fs.priority = rule->cr.priority; fs.cookie = rule->flow_cookie; fs.table_id = rule->table_id; - calc_flow_duration__(rule->created, now, &fs.duration_sec, - &fs.duration_nsec); - fs.idle_timeout = rule->idle_timeout; - fs.hard_timeout = rule->hard_timeout; + calc_duration(rule->created, now, &fs.duration_sec, &fs.duration_nsec); 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, &fs.byte_count); fs.ofpacts = rule->ofpacts; fs.ofpacts_len = rule->ofpacts_len; + + ovs_mutex_lock(&rule->timeout_mutex); + fs.idle_timeout = rule->idle_timeout; + fs.hard_timeout = rule->hard_timeout; + ovs_mutex_unlock(&rule->timeout_mutex); + fs.flags = 0; if (rule->send_flow_removed) { - fs.flags |= OFPFF_SEND_FLOW_REM; - /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS - and OFPFF13_NO_BYT_COUNTS */ + fs.flags |= OFPUTIL_FF_SEND_FLOW_REM; + /* FIXME: Implement OFPUTIL_FF_NO_PKT_COUNTS and + OFPUTIL_FF_NO_BYT_COUNTS. */ } ofputil_append_flow_stats_reply(&fs, &replies); } @@ -2848,11 +3186,7 @@ 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"); } @@ -2867,10 +3201,12 @@ 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); } } @@ -2883,62 +3219,20 @@ 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', 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) { 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 @@ -2961,7 +3255,7 @@ handle_aggregate_stats_request(struct ofconn *ofconn, error = collect_rules_loose(ofproto, request.table_id, &request.match, request.cookie, request.cookie_mask, - request.out_port, &rules); + request.out_port, request.out_group, &rules); if (error) { return error; } @@ -3005,18 +3299,26 @@ 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); } @@ -3063,6 +3365,7 @@ 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) { @@ -3111,6 +3414,34 @@ is_flow_deletion_pending(const struct ofproto *ofproto, return false; } +static enum ofperr +evict_rule_from_table(struct ofproto *ofproto, struct oftable *table) +{ + 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) { + return 0; + } else if (!choose_rule_to_evict(table, &rule)) { + return OFPERR_OFPFMFC_TABLE_FULL; + } else if (rule->pending) { + ovs_rwlock_unlock(&rule->rwlock); + return OFPROTO_POSTPONE; + } else { + struct ofopgroup *group; + + group = ofopgroup_create_unattached(ofproto); + delete_flow__(rule, group, OFPRR_EVICTION); + ofopgroup_submit(group); + + return 0; + } +} + /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT * in which no matching flow already exists in the flow table. * @@ -3126,13 +3457,13 @@ 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; - struct rule *victim; struct cls_rule cr; struct rule *rule; + uint8_t table_id; int error; error = check_table_id(ofproto, fm->table_id); @@ -3142,7 +3473,6 @@ add_flow(struct ofproto *ofproto, struct ofconn *ofconn, /* Pick table. */ if (fm->table_id == 0xff) { - uint8_t table_id; if (ofproto->ofproto_class->rule_choose_table) { error = ofproto->ofproto_class->rule_choose_table(ofproto, &fm->match, @@ -3150,114 +3480,129 @@ add_flow(struct ofproto *ofproto, struct ofconn *ofconn, if (error) { return error; } - assert(table_id < ofproto->n_tables); - table = &ofproto->tables[table_id]; + ovs_assert(table_id < ofproto->n_tables); } else { - table = &ofproto->tables[0]; + table_id = 0; } } else if (fm->table_id < ofproto->n_tables) { - table = &ofproto->tables[fm->table_id]; + table_id = fm->table_id; } else { return OFPERR_OFPBRC_BAD_TABLE_ID; } + table = &ofproto->tables[table_id]; + if (table->flags & OFTABLE_READONLY) { return OFPERR_OFPBRC_EPERM; } - /* 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)); - return ENOMEM; - } - cls_rule_init(&rule->cr, &fm->match, fm->priority); + cls_rule_init(&cr, &fm->match, fm->priority); - /* Serialize against pending deletion. */ - if (is_flow_deletion_pending(ofproto, &cr, table - ofproto->tables)) { - cls_rule_destroy(&rule->cr); - ofproto->ofproto_class->rule_dealloc(rule); - return OFPROTO_POSTPONE; + /* Transform "add" into "modify" if there's an existing identical flow. */ + 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) { + cls_rule_destroy(&cr); + if (!rule_is_modifiable(rule)) { + return OFPERR_OFPBRC_EPERM; + } else if (rule->pending) { + return OFPROTO_POSTPONE; + } else { + struct list rules; + + list_init(&rules); + list_push_back(&rules, &rule->ofproto_node); + fm->modify_cookie = true; + return modify_flows__(ofproto, ofconn, fm, request, &rules); + } + } + + /* Verify actions. */ + error = ofproto_check_ofpacts(ofproto, fm->ofpacts, fm->ofpacts_len, + &fm->match.flow, table_id); + if (error) { + cls_rule_destroy(&cr); + return error; + } + + /* Serialize against pending deletion. */ + if (is_flow_deletion_pending(ofproto, &cr, table_id)) { + cls_rule_destroy(&cr); + return OFPROTO_POSTPONE; } /* 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 & OFPUTIL_FF_CHECK_OVERLAP) { + bool overlaps; + + ovs_rwlock_rdlock(&table->cls.rwlock); + overlaps = classifier_rule_overlaps(&table->cls, &cr); + ovs_rwlock_unlock(&table->cls.rwlock); + + if (overlaps) { + cls_rule_destroy(&cr); + return OFPERR_OFPFMFC_OVERLAP; + } + } + + /* If necessary, evict an existing rule to clear out space. */ + error = evict_rule_from_table(ofproto, table); + if (error) { + cls_rule_destroy(&cr); + return error; } - /* FIXME: Implement OFPFF12_RESET_COUNTS */ + /* Allocate new rule. */ + rule = ofproto->ofproto_class->rule_alloc(); + if (!rule) { + cls_rule_destroy(&cr); + VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)", + ofproto->name, ovs_strerror(error)); + return ENOMEM; + } + /* Initialize base state. */ rule->ofproto = ofproto; + cls_rule_move(&rule->cr, &cr); 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->send_flow_removed = (fm->flags & OFPUTIL_FF_SEND_FLOW_REM) != 0; 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->rwlock); - /* Insert new rule. */ - victim = oftable_replace_rule(rule); - if (victim && !rule_is_modifiable(victim)) { - error = OFPERR_OFPBRC_EPERM; - } else if (victim && victim->pending) { - error = OFPROTO_POSTPONE; - } 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) { - error = OFPERR_OFPFMFC_TABLE_FULL; - goto exit; - } else if (evict->pending) { - error = OFPROTO_POSTPONE; - goto exit; - } - } else { - evict = NULL; - } - - group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id); - op = ofoperation_create(group, rule, OFOPERATION_ADD, 0); - op->victim = victim; - - error = ofproto->ofproto_class->rule_construct(rule); - if (error) { - op->group->n_running--; - ofoperation_destroy(rule->pending); - } else if (evict) { - delete_flow__(evict, group); - } - ofopgroup_submit(group); - } - -exit: - /* Back out if an error occurred. */ + /* Construct rule, initializing derived state. */ + error = ofproto->ofproto_class->rule_construct(rule); if (error) { - oftable_substitute_rule(rule, victim); ofproto_rule_destroy__(rule); + return error; } + + /* Insert rule. */ + oftable_insert_rule(rule); + + group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id); + ofoperation_create(group, rule, OFOPERATION_ADD, 0); + ofproto->ofproto_class->rule_insert(rule); + ofopgroup_submit(group); + return error; } @@ -3272,21 +3617,23 @@ 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) { + enum ofoperation_type type; struct ofopgroup *group; struct rule *rule; enum ofperr error; + type = fm->command == OFPFC_ADD ? OFOPERATION_REPLACE : OFOPERATION_MODIFY; group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id); error = OFPERR_OFPBRC_EPERM; LIST_FOR_EACH (rule, ofproto_node, rules) { struct ofoperation *op; bool actions_changed; - ovs_be64 new_cookie; + bool reset_counters; - /* FIXME: Implement OFPFF12_RESET_COUNTS */ + /* FIXME: Implement OFPFUTIL_FF_RESET_COUNTS */ if (rule_is_modifiable(rule)) { /* At least one rule is modifiable, don't report EPERM error. */ @@ -3295,24 +3642,53 @@ 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); - 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, type, 0); + + if (fm->modify_cookie && fm->new_cookie != htonll(UINT64_MAX)) { + ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie); + } + if (type == OFOPERATION_REPLACE) { + 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->send_flow_removed = (fm->flags + & OFPUTIL_FF_SEND_FLOW_REM) != 0; + + if (fm->idle_timeout || fm->hard_timeout) { + if (!rule->eviction_group) { + eviction_group_add_rule(rule); + } + } else { + eviction_group_remove_rule(rule); + } } - op = ofoperation_create(group, rule, OFOPERATION_MODIFY, 0); - rule->flow_cookie = new_cookie; - if (actions_changed) { + reset_counters = (fm->flags & OFPUTIL_FF_RESET_COUNTS) != 0; + if (actions_changed || reset_counters) { op->ofpacts = rule->ofpacts; op->ofpacts_len = rule->ofpacts_len; + op->meter_id = rule->meter_id; + + ovs_rwlock_wrlock(&rule->rwlock); rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len); rule->ofpacts_len = fm->ofpacts_len; - rule->ofproto->ofproto_class->rule_modify_actions(rule); + ovs_rwlock_unlock(&rule->rwlock); + + rule->meter_id = find_meter(rule->ofpacts, rule->ofpacts_len); + rule->ofproto->ofproto_class->rule_modify_actions(rule, + reset_counters); } else { ofoperation_complete(op, 0); } @@ -3324,8 +3700,7 @@ 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; @@ -3340,7 +3715,7 @@ 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; @@ -3348,7 +3723,7 @@ modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn, error = collect_rules_loose(ofproto, fm->table_id, &fm->match, fm->cookie, fm->cookie_mask, - OFPP_ANY, &rules); + OFPP_ANY, OFPG11_ANY, &rules); if (error) { return error; } else if (list_is_empty(&rules)) { @@ -3365,7 +3740,7 @@ 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; @@ -3373,8 +3748,7 @@ modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn, error = collect_rules_strict(ofproto, fm->table_id, &fm->match, fm->priority, fm->cookie, fm->cookie_mask, - OFPP_ANY, &rules); - + OFPP_ANY, OFPG11_ANY, &rules); if (error) { return error; } else if (list_is_empty(&rules)) { @@ -3389,15 +3763,16 @@ 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); + ofproto->ofproto_class->rule_delete(rule); } /* Deletes the rules listed in 'rules'. @@ -3405,14 +3780,16 @@ 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->rwlock); + delete_flow__(rule, group, reason); } ofopgroup_submit(group); @@ -3430,10 +3807,10 @@ delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn, error = collect_rules_loose(ofproto, fm->table_id, &fm->match, fm->cookie, fm->cookie_mask, - fm->out_port, &rules); + fm->out_port, fm->out_group, &rules); return (error ? error : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request, - &rules) + &rules, OFPRR_DELETE) : 0); } @@ -3448,10 +3825,11 @@ delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn, error = collect_rules_strict(ofproto, fm->table_id, &fm->match, fm->priority, fm->cookie, fm->cookie_mask, - fm->out_port, &rules); + fm->out_port, fm->out_group, &rules); return (error ? error : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn, - request, &rules) + request, &rules, + OFPRR_DELETE) : 0); } @@ -3469,30 +3847,18 @@ ofproto_rule_send_removed(struct rule *rule, uint8_t reason) fr.cookie = rule->flow_cookie; fr.reason = reason; fr.table_id = rule->table_id; - calc_flow_duration__(rule->created, time_msec(), - &fr.duration_sec, &fr.duration_nsec); + 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); connmgr_send_flow_removed(rule->ofproto->connmgr, &fr); } -void -ofproto_rule_update_used(struct rule *rule, long long int used) -{ - if (used > rule->used) { - struct eviction_group *evg = rule->eviction_group; - - rule->used = used; - if (evg) { - heap_change(&evg->rules, &rule->evg_node, - rule_eviction_priority(rule)); - } - } -} - /* Sends an OpenFlow "flow removed" message with the given 'reason' (either * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its * ofproto. @@ -3506,17 +3872,51 @@ void 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); + struct classifier *cls = &ofproto->tables[rule->table_id].cls; + ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT + || reason == OFPRR_DELETE || reason == OFPRR_GROUP_DELETE); ofproto_rule_send_removed(rule, reason); - group = ofopgroup_create_unattached(ofproto); - ofoperation_create(group, rule, OFOPERATION_DELETE, reason); - oftable_remove_rule(rule); - ofproto->ofproto_class->rule_destruct(rule); - ofopgroup_submit(group); + ovs_rwlock_wrlock(&cls->rwlock); + ofproto_delete_rule(ofproto, cls, rule); + ovs_rwlock_unlock(&cls->rwlock); +} + +/* Reduces '*timeout' to no more than 'max'. A value of zero in either case + * means "infinite". */ +static void +reduce_timeout(uint16_t max, uint16_t *timeout) +{ + if (max && (!*timeout || *timeout > max)) { + *timeout = max; + } +} + +/* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle + * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to + * 'idle_timeout' seconds. Similarly for 'hard_timeout'. + * + * Suitable for implementing OFPACT_FIN_TIMEOUT. */ +void +ofproto_rule_reduce_timeouts(struct rule *rule, + uint16_t idle_timeout, uint16_t hard_timeout) + OVS_EXCLUDED(rule->ofproto->expirable_mutex, rule->timeout_mutex) +{ + if (!idle_timeout && !hard_timeout) { + return; + } + + ovs_mutex_lock(&rule->ofproto->expirable_mutex); + if (list_is_empty(&rule->expirable)) { + list_insert(&rule->ofproto->expirable, &rule->expirable); + } + ovs_mutex_unlock(&rule->ofproto->expirable_mutex); + + ovs_mutex_lock(&rule->timeout_mutex); + reduce_timeout(idle_timeout, &rule->idle_timeout); + reduce_timeout(hard_timeout, &rule->hard_timeout); + ovs_mutex_unlock(&rule->timeout_mutex); } static enum ofperr @@ -3538,16 +3938,12 @@ handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh) 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__(ofconn_get_ofproto(ofconn), ofconn, &fm, oh); + error = handle_flow_mod__(ofproto, ofconn, &fm, oh); } if (error) { goto exit_free_ofpacts; } - + /* Record the operation for logging a summary report. */ switch (fm.command) { case OFPFC_ADD: @@ -3582,11 +3978,10 @@ 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) { - assert(!list_is_empty(&ofproto->pending)); + ovs_assert(!list_is_empty(&ofproto->pending)); return OFPROTO_POSTPONE; } @@ -3619,27 +4014,34 @@ 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; @@ -3728,6 +4130,31 @@ handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh) return 0; } +static enum ofperr +handle_nxt_get_async_request(struct ofconn *ofconn, const struct ofp_header *oh) +{ + struct ofpbuf *buf; + uint32_t master[OAM_N_TYPES]; + uint32_t slave[OAM_N_TYPES]; + struct nx_async_config *msg; + + ofconn_get_async_config(ofconn, master, slave); + buf = ofpraw_alloc_reply(OFPRAW_OFPT13_GET_ASYNC_REPLY, oh, 0); + msg = ofpbuf_put_zeros(buf, sizeof *msg); + + msg->packet_in_mask[0] = htonl(master[OAM_PACKET_IN]); + msg->port_status_mask[0] = htonl(master[OAM_PORT_STATUS]); + msg->flow_removed_mask[0] = htonl(master[OAM_FLOW_REMOVED]); + + msg->packet_in_mask[1] = htonl(slave[OAM_PACKET_IN]); + msg->port_status_mask[1] = htonl(slave[OAM_PORT_STATUS]); + msg->flow_removed_mask[1] = htonl(slave[OAM_FLOW_REMOVED]); + + ofconn_send_reply(ofconn, buf); + + return 0; +} + static enum ofperr handle_nxt_set_controller_id(struct ofconn *ofconn, const struct ofp_header *oh) @@ -3767,7 +4194,7 @@ ofproto_compose_flow_refresh_update(const struct rule *rule, struct ofputil_flow_update fu; struct match match; - if (op && op->type == OFOPERATION_ADD && !op->victim) { + if (op && op->type == OFOPERATION_ADD) { /* We'll report the final flow when the operation completes. Reporting * it now would cause a duplicate report later. */ return; @@ -3776,8 +4203,10 @@ 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); @@ -3794,12 +4223,10 @@ ofproto_compose_flow_refresh_update(const struct rule *rule, * actions, so that when the operation commits we report the change. */ switch (op->type) { case OFOPERATION_ADD: - /* We already verified that there was a victim. */ - fu.ofpacts = op->victim->ofpacts; - fu.ofpacts_len = op->victim->ofpacts_len; - break; + NOT_REACHED(); case OFOPERATION_MODIFY: + case OFOPERATION_REPLACE: if (op->ofpacts) { fu.ofpacts = op->ofpacts; fu.ofpacts_len = op->ofpacts_len; @@ -3825,190 +4252,924 @@ ofproto_compose_flow_refresh_update(const struct rule *rule, ofputil_append_flow_update(&fu, msgs); } -void -ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs) -{ - struct rule *rule; +void +ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs) +{ + struct rule *rule; + + LIST_FOR_EACH (rule, ofproto_node, rules) { + enum nx_flow_monitor_flags flags = rule->monitor_flags; + rule->monitor_flags = 0; + + ofproto_compose_flow_refresh_update(rule, flags, msgs); + } +} + +static void +ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m, + struct rule *rule, uint64_t seqno, + struct list *rules) +{ + enum nx_flow_monitor_flags update; + + if (ofproto_rule_is_hidden(rule)) { + return; + } + + if (!(rule->pending + ? ofoperation_has_out_port(rule->pending, m->out_port) + : ofproto_rule_has_out_port(rule, m->out_port))) { + return; + } + + if (seqno) { + if (rule->add_seqno > seqno) { + update = NXFMF_ADD | NXFMF_MODIFY; + } else if (rule->modify_seqno > seqno) { + update = NXFMF_MODIFY; + } else { + return; + } + + if (!(m->flags & update)) { + return; + } + } else { + update = NXFMF_INITIAL; + } + + if (!rule->monitor_flags) { + list_push_back(rules, &rule->ofproto_node); + } + rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS); +} + +static void +ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m, + uint64_t seqno, + struct list *rules) +{ + const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn); + const struct ofoperation *op; + const struct oftable *table; + struct cls_rule target; + + cls_rule_init_from_minimatch(&target, &m->match, 0); + FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) { + 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) { + struct rule *rule = op->rule; + + if (((m->table_id == 0xff + ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN) + : m->table_id == rule->table_id)) + && cls_rule_is_loose_match(&rule->cr, &target.match)) { + ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules); + } + } + cls_rule_destroy(&target); +} + +static void +ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m, + struct list *rules) +{ + if (m->flags & NXFMF_INITIAL) { + ofproto_collect_ofmonitor_refresh_rules(m, 0, rules); + } +} + +void +ofmonitor_collect_resume_rules(struct ofmonitor *m, + uint64_t seqno, struct list *rules) +{ + ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules); +} + +static enum ofperr +handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh) +{ + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); + struct ofmonitor **monitors; + size_t n_monitors, allocated_monitors; + struct list replies; + enum ofperr error; + struct list rules; + struct ofpbuf b; + size_t i; + + error = 0; + ofpbuf_use_const(&b, oh, ntohs(oh->length)); + monitors = NULL; + n_monitors = allocated_monitors = 0; + for (;;) { + struct ofputil_flow_monitor_request request; + struct ofmonitor *m; + int retval; + + retval = ofputil_decode_flow_monitor_request(&request, &b); + if (retval == EOF) { + break; + } else if (retval) { + error = retval; + goto error; + } + + if (request.table_id != 0xff + && request.table_id >= ofproto->n_tables) { + error = OFPERR_OFPBRC_BAD_TABLE_ID; + goto error; + } + + error = ofmonitor_create(&request, ofconn, &m); + if (error) { + goto error; + } + + if (n_monitors >= allocated_monitors) { + monitors = x2nrealloc(monitors, &allocated_monitors, + sizeof *monitors); + } + monitors[n_monitors++] = m; + } + + list_init(&rules); + for (i = 0; i < n_monitors; i++) { + ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules); + } + + ofpmp_init(&replies, oh); + ofmonitor_compose_refresh_updates(&rules, &replies); + ofconn_send_replies(ofconn, &replies); + + free(monitors); + + return 0; + +error: + for (i = 0; i < n_monitors; i++) { + ofmonitor_destroy(monitors[i]); + } + free(monitors); + return error; +} + +static enum ofperr +handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh) +{ + struct ofmonitor *m; + uint32_t id; + + id = ofputil_decode_flow_monitor_cancel(oh); + m = ofmonitor_lookup(ofconn, id); + if (!m) { + return OFPERR_NXBRC_FM_BAD_ID; + } + + ofmonitor_destroy(m); + 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; +} + +bool +ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id, + struct ofgroup **group) + OVS_TRY_RDLOCK(true, (*group)->rwlock) +{ + ovs_rwlock_rdlock(&ofproto->groups_rwlock); + HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node, + hash_int(group_id, 0), &ofproto->groups) { + if ((*group)->group_id == group_id) { + ovs_rwlock_rdlock(&(*group)->rwlock); + ovs_rwlock_unlock(&ofproto->groups_rwlock); + return true; + } + } + ovs_rwlock_unlock(&ofproto->groups_rwlock); + return false; +} + +void +ofproto_group_release(struct ofgroup *group) + OVS_RELEASES(group->rwlock) +{ + ovs_rwlock_unlock(&group->rwlock); +} + +static bool +ofproto_group_write_lookup(const struct ofproto *ofproto, uint32_t group_id, + struct ofgroup **group) + OVS_TRY_WRLOCK(true, ofproto->groups_rwlock) + OVS_TRY_WRLOCK(true, (*group)->rwlock) +{ + ovs_rwlock_wrlock(&ofproto->groups_rwlock); + HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node, + hash_int(group_id, 0), &ofproto->groups) { + if ((*group)->group_id == group_id) { + ovs_rwlock_wrlock(&(*group)->rwlock); + return true; + } + } + ovs_rwlock_unlock(&ofproto->groups_rwlock); + return false; +} + +static bool +ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id) + OVS_REQ_RDLOCK(ofproto->groups_rwlock) +{ + struct ofgroup *grp; + + HMAP_FOR_EACH_IN_BUCKET (grp, hmap_node, + hash_int(group_id, 0), &ofproto->groups) { + if (grp->group_id == group_id) { + return true; + } + } + return false; +} + +static void +append_group_stats(struct ofgroup *group, struct list *replies) + OVS_REQ_RDLOCK(group->rwlock) +{ + struct ofputil_group_stats ogs; + struct ofproto *ofproto = group->ofproto; + long long int now = time_msec(); + int error; + + ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats); + + error = (ofproto->ofproto_class->group_get_stats + ? ofproto->ofproto_class->group_get_stats(group, &ogs) + : EOPNOTSUPP); + if (error) { + ogs.ref_count = UINT32_MAX; + ogs.packet_count = UINT64_MAX; + ogs.byte_count = UINT64_MAX; + ogs.n_buckets = group->n_buckets; + memset(ogs.bucket_stats, 0xff, + ogs.n_buckets * sizeof *ogs.bucket_stats); + } + + ogs.group_id = group->group_id; + calc_duration(group->created, now, &ogs.duration_sec, &ogs.duration_nsec); + + ofputil_append_group_stats(replies, &ogs); + + free(ogs.bucket_stats); +} + +static enum ofperr +handle_group_stats_request(struct ofconn *ofconn, + const struct ofp_header *request) +{ + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); + struct list replies; + enum ofperr error; + struct ofgroup *group; + uint32_t group_id; + + error = ofputil_decode_group_stats_request(request, &group_id); + if (error) { + return error; + } + + ofpmp_init(&replies, request); + + if (group_id == OFPG_ALL) { + ovs_rwlock_rdlock(&ofproto->groups_rwlock); + HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) { + ovs_rwlock_rdlock(&group->rwlock); + append_group_stats(group, &replies); + ovs_rwlock_unlock(&group->rwlock); + } + ovs_rwlock_unlock(&ofproto->groups_rwlock); + } else { + if (ofproto_group_lookup(ofproto, group_id, &group)) { + append_group_stats(group, &replies); + ofproto_group_release(group); + } + } + + ofconn_send_replies(ofconn, &replies); + + return 0; +} + +static enum ofperr +handle_group_desc_stats_request(struct ofconn *ofconn, + const struct ofp_header *request) +{ + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); + struct list replies; + struct ofputil_group_desc gds; + struct ofgroup *group; + + ofpmp_init(&replies, request); + + ovs_rwlock_rdlock(&ofproto->groups_rwlock); + HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) { + gds.group_id = group->group_id; + gds.type = group->type; + ofputil_append_group_desc_reply(&gds, &group->buckets, &replies); + } + ovs_rwlock_unlock(&ofproto->groups_rwlock); + + ofconn_send_replies(ofconn, &replies); + + return 0; +} + +static enum ofperr +handle_group_features_stats_request(struct ofconn *ofconn, + const struct ofp_header *request) +{ + struct ofproto *p = ofconn_get_ofproto(ofconn); + struct ofpbuf *msg; + + msg = ofputil_encode_group_features_reply(&p->ogf, request); + if (msg) { + ofconn_send_reply(ofconn, msg); + } + + return 0; +} + +/* Implements OFPGC11_ADD + * in which no matching flow already exists in the flow table. + * + * Adds the flow specified by 'ofm', which is followed by 'n_actions' + * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow + * error code on failure, or OFPROTO_POSTPONE if the operation cannot be + * initiated now but may be retried later. + * + * Upon successful return, takes ownership of 'fm->ofpacts'. On failure, + * ownership remains with the caller. + * + * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, + * if any. */ +static enum ofperr +add_group(struct ofproto *ofproto, struct ofputil_group_mod *gm) +{ + struct ofgroup *ofgroup; + enum ofperr error; + + if (gm->group_id > OFPG_MAX) { + return OFPERR_OFPGMFC_INVALID_GROUP; + } + if (gm->type > OFPGT11_FF) { + return OFPERR_OFPGMFC_BAD_TYPE; + } + + /* Allocate new group and initialize it. */ + ofgroup = ofproto->ofproto_class->group_alloc(); + if (!ofgroup) { + VLOG_WARN_RL(&rl, "%s: failed to create group", ofproto->name); + return OFPERR_OFPGMFC_OUT_OF_GROUPS; + } + + ovs_rwlock_init(&ofgroup->rwlock); + ofgroup->ofproto = ofproto; + ofgroup->group_id = gm->group_id; + ofgroup->type = gm->type; + ofgroup->created = ofgroup->modified = time_msec(); - LIST_FOR_EACH (rule, ofproto_node, rules) { - enum nx_flow_monitor_flags flags = rule->monitor_flags; - rule->monitor_flags = 0; + list_move(&ofgroup->buckets, &gm->buckets); + ofgroup->n_buckets = list_size(&ofgroup->buckets); - ofproto_compose_flow_refresh_update(rule, flags, msgs); + /* Construct called BEFORE any locks are held. */ + error = ofproto->ofproto_class->group_construct(ofgroup); + if (error) { + goto free_out; } -} -static void -ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m, - struct rule *rule, uint64_t seqno, - struct list *rules) -{ - enum nx_flow_monitor_flags update; + /* We wrlock as late as possible to minimize the time we jam any other + * threads: No visible state changes before acquiring the lock. */ + ovs_rwlock_wrlock(&ofproto->groups_rwlock); - if (ofproto_rule_is_hidden(rule)) { - return; + if (ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) { + error = OFPERR_OFPGMFC_OUT_OF_GROUPS; + goto unlock_out; } - if (!(rule->pending - ? ofoperation_has_out_port(rule->pending, m->out_port) - : ofproto_rule_has_out_port(rule, m->out_port))) { - return; + if (ofproto_group_exists(ofproto, gm->group_id)) { + error = OFPERR_OFPGMFC_GROUP_EXISTS; + goto unlock_out; } - if (seqno) { - if (rule->add_seqno > seqno) { - update = NXFMF_ADD | NXFMF_MODIFY; - } else if (rule->modify_seqno > seqno) { - update = NXFMF_MODIFY; - } else { - return; - } + if (!error) { + /* Insert new group. */ + hmap_insert(&ofproto->groups, &ofgroup->hmap_node, + hash_int(ofgroup->group_id, 0)); + ofproto->n_groups[ofgroup->type]++; - if (!(m->flags & update)) { - return; - } - } else { - update = NXFMF_INITIAL; + ovs_rwlock_unlock(&ofproto->groups_rwlock); + return error; } - if (!rule->monitor_flags) { - list_push_back(rules, &rule->ofproto_node); - } - rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS); + unlock_out: + ovs_rwlock_unlock(&ofproto->groups_rwlock); + ofproto->ofproto_class->group_destruct(ofgroup); + free_out: + ofputil_bucket_list_destroy(&ofgroup->buckets); + ofproto->ofproto_class->group_dealloc(ofgroup); + + return error; } -static void -ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m, - uint64_t seqno, - struct list *rules) +/* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on + * failure. + * + * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id, + * if any. */ +static enum ofperr +modify_group(struct ofproto *ofproto, struct ofputil_group_mod *gm) { - const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn); - const struct ofoperation *op; - const struct oftable *table; - struct cls_rule target; + struct ofgroup *ofgroup; + struct ofgroup *victim; + enum ofperr error; - cls_rule_init_from_minimatch(&target, &m->match, 0); - FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) { - struct cls_cursor cursor; - struct rule *rule; + if (gm->group_id > OFPG_MAX) { + return OFPERR_OFPGMFC_INVALID_GROUP; + } - cls_cursor_init(&cursor, &table->cls, &target); - CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { - assert(!rule->pending); /* XXX */ - ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules); - } + if (gm->type > OFPGT11_FF) { + return OFPERR_OFPGMFC_BAD_TYPE; } - HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) { - struct rule *rule = op->rule; + victim = ofproto->ofproto_class->group_alloc(); + if (!victim) { + VLOG_WARN_RL(&rl, "%s: failed to allocate group", ofproto->name); + return OFPERR_OFPGMFC_OUT_OF_GROUPS; + } - if (((m->table_id == 0xff - ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN) - : m->table_id == rule->table_id)) - && cls_rule_is_loose_match(&rule->cr, &target.match)) { - ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules); - } + if (!ofproto_group_write_lookup(ofproto, gm->group_id, &ofgroup)) { + error = OFPERR_OFPGMFC_UNKNOWN_GROUP; + goto free_out; } - cls_rule_destroy(&target); + /* Both group's and its container's write locks held now. + * Also, n_groups[] is protected by ofproto->groups_rwlock. */ + if (ofgroup->type != gm->type + && ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) { + error = OFPERR_OFPGMFC_OUT_OF_GROUPS; + goto unlock_out; + } + + *victim = *ofgroup; + list_move(&victim->buckets, &ofgroup->buckets); + + ofgroup->type = gm->type; + list_move(&ofgroup->buckets, &gm->buckets); + ofgroup->n_buckets = list_size(&ofgroup->buckets); + + error = ofproto->ofproto_class->group_modify(ofgroup, victim); + if (!error) { + ofputil_bucket_list_destroy(&victim->buckets); + ofproto->n_groups[victim->type]--; + ofproto->n_groups[ofgroup->type]++; + ofgroup->modified = time_msec(); + } else { + ofputil_bucket_list_destroy(&ofgroup->buckets); + + *ofgroup = *victim; + list_move(&ofgroup->buckets, &victim->buckets); + } + + unlock_out: + ovs_rwlock_unlock(&ofgroup->rwlock); + ovs_rwlock_unlock(&ofproto->groups_rwlock); + free_out: + ofproto->ofproto_class->group_dealloc(victim); + return error; } static void -ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m, - struct list *rules) +delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup) + OVS_RELEASES(ofproto->groups_rwlock) { - if (m->flags & NXFMF_INITIAL) { - ofproto_collect_ofmonitor_refresh_rules(m, 0, rules); - } + /* Must wait until existing readers are done, + * while holding the container's write lock at the same time. */ + ovs_rwlock_wrlock(&ofgroup->rwlock); + hmap_remove(&ofproto->groups, &ofgroup->hmap_node); + /* No-one can find this group any more. */ + ofproto->n_groups[ofgroup->type]--; + ovs_rwlock_unlock(&ofproto->groups_rwlock); + + ofproto->ofproto_class->group_destruct(ofgroup); + ofputil_bucket_list_destroy(&ofgroup->buckets); + ovs_rwlock_unlock(&ofgroup->rwlock); + ovs_rwlock_destroy(&ofgroup->rwlock); + ofproto->ofproto_class->group_dealloc(ofgroup); } -void -ofmonitor_collect_resume_rules(struct ofmonitor *m, - uint64_t seqno, struct list *rules) +/* Implements OFPGC_DELETE. */ +static void +delete_group(struct ofproto *ofproto, uint32_t group_id) { - ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules); + struct ofgroup *ofgroup; + + ovs_rwlock_wrlock(&ofproto->groups_rwlock); + if (group_id == OFPG_ALL) { + for (;;) { + struct hmap_node *node = hmap_first(&ofproto->groups); + if (!node) { + break; + } + ofgroup = CONTAINER_OF(node, struct ofgroup, hmap_node); + delete_group__(ofproto, ofgroup); + /* Lock for each node separately, so that we will not jam the + * other threads for too long time. */ + ovs_rwlock_wrlock(&ofproto->groups_rwlock); + } + } else { + HMAP_FOR_EACH_IN_BUCKET (ofgroup, hmap_node, + hash_int(group_id, 0), &ofproto->groups) { + if (ofgroup->group_id == group_id) { + delete_group__(ofproto, ofgroup); + return; + } + } + } + ovs_rwlock_unlock(&ofproto->groups_rwlock); } static enum ofperr -handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh) +handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - struct ofmonitor **monitors; - size_t n_monitors, allocated_monitors; - struct list replies; + struct ofputil_group_mod gm; enum ofperr error; - struct list rules; - struct ofpbuf b; - size_t i; - - error = 0; - ofpbuf_use_const(&b, oh, ntohs(oh->length)); - monitors = NULL; - n_monitors = allocated_monitors = 0; - for (;;) { - struct ofputil_flow_monitor_request request; - struct ofmonitor *m; - int retval; - - retval = ofputil_decode_flow_monitor_request(&request, &b); - if (retval == EOF) { - break; - } else if (retval) { - error = retval; - goto error; - } - - if (request.table_id != 0xff - && request.table_id >= ofproto->n_tables) { - error = OFPERR_OFPBRC_BAD_TABLE_ID; - goto error; - } - error = ofmonitor_create(&request, ofconn, &m); - if (error) { - goto error; - } - - if (n_monitors >= allocated_monitors) { - monitors = x2nrealloc(monitors, &allocated_monitors, - sizeof *monitors); - } - monitors[n_monitors++] = m; + error = reject_slave_controller(ofconn); + if (error) { + return error; } - list_init(&rules); - for (i = 0; i < n_monitors; i++) { - ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules); + error = ofputil_decode_group_mod(oh, &gm); + if (error) { + return error; } - ofpmp_init(&replies, oh); - ofmonitor_compose_refresh_updates(&rules, &replies); - ofconn_send_replies(ofconn, &replies); + switch (gm.command) { + case OFPGC11_ADD: + return add_group(ofproto, &gm); - free(monitors); + case OFPGC11_MODIFY: + return modify_group(ofproto, &gm); - return 0; + case OFPGC11_DELETE: + delete_group(ofproto, gm.group_id); + return 0; -error: - for (i = 0; i < n_monitors; i++) { - ofmonitor_destroy(monitors[i]); + default: + if (gm.command > OFPGC11_DELETE) { + VLOG_WARN_RL(&rl, "%s: Invalid group_mod command type %d", + ofproto->name, gm.command); + } + return OFPERR_OFPGMFC_BAD_COMMAND; } - free(monitors); - return error; } static enum ofperr -handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh) +handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh) { - struct ofmonitor *m; - uint32_t id; + struct ofputil_table_mod tm; + enum ofperr error; - id = ofputil_decode_flow_monitor_cancel(oh); - m = ofmonitor_lookup(ofconn, id); - if (!m) { - return OFPERR_NXBRC_FM_BAD_ID; + error = reject_slave_controller(ofconn); + if (error) { + return error; } - ofmonitor_destroy(m); + error = ofputil_decode_table_mod(oh, &tm); + if (error) { + return error; + } + + /* XXX Actual table mod support is not implemented yet. */ return 0; } @@ -4047,17 +5208,26 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) case OFPTYPE_FLOW_MOD: return handle_flow_mod(ofconn, oh); + case OFPTYPE_GROUP_MOD: + return handle_group_mod(ofconn, oh); + + case OFPTYPE_TABLE_MOD: + return handle_table_mod(ofconn, oh); + + case OFPTYPE_METER_MOD: + return handle_meter_mod(ofconn, oh); + 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); @@ -4080,6 +5250,9 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) case OFPTYPE_SET_ASYNC_CONFIG: return handle_nxt_set_async_config(ofconn, oh); + case OFPTYPE_GET_ASYNC_REQUEST: + return handle_nxt_get_async_request(ofconn, oh); + /* Statistics requests. */ case OFPTYPE_DESC_STATS_REQUEST: return handle_desc_stats_request(ofconn, oh); @@ -4105,17 +5278,25 @@ 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); + + case OFPTYPE_GROUP_STATS_REQUEST: + return handle_group_stats_request(ofconn, oh); + + case OFPTYPE_GROUP_DESC_STATS_REQUEST: + return handle_group_desc_stats_request(ofconn, oh); + + case OFPTYPE_GROUP_FEATURES_STATS_REQUEST: + return handle_group_features_stats_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_TABLE_FEATURES_STATS_REQUEST: return OFPERR_OFPBRC_BAD_TYPE; case OFPTYPE_HELLO: @@ -4139,20 +5320,20 @@ 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; } } static bool -handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg) +handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg) { int error = handle_openflow__(ofconn, ofp_msg); if (error && error != OFPROTO_POSTPONE) { @@ -4200,7 +5381,7 @@ 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 +5419,7 @@ 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) { @@ -4252,12 +5433,12 @@ ofopgroup_complete(struct ofopgroup *group) LIST_FOR_EACH (op, group_node, &group->ops) { if (op->type != OFOPERATION_DELETE) { struct ofpbuf *packet; - uint16_t in_port; + ofp_port_t in_port; 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; @@ -4275,18 +5456,42 @@ 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 - == NXFME_ADDED); - BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_DELETE - == NXFME_DELETED); - BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_MODIFY - == NXFME_MODIFIED); - - ofmonitor_report(ofproto->connmgr, rule, - (enum nx_flow_update_event) op->type, + enum nx_flow_update_event event_type; + + switch (op->type) { + case OFOPERATION_ADD: + case OFOPERATION_REPLACE: + event_type = NXFME_ADDED; + break; + + case OFOPERATION_DELETE: + event_type = NXFME_DELETED; + break; + + case OFOPERATION_MODIFY: + event_type = NXFME_MODIFIED; + break; + + default: + NOT_REACHED(); + } + + ofmonitor_report(ofproto->connmgr, rule, event_type, op->reason, abbrev_ofconn, abbrev_xid); } @@ -4297,7 +5502,6 @@ ofopgroup_complete(struct ofopgroup *group) if (!op->error) { uint16_t vid_mask; - ofproto_rule_destroy__(op->victim); vid_mask = minimask_get_vid_mask(&rule->cr.match.mask); if (vid_mask == VLAN_VID_MASK) { if (ofproto->vlan_bitmap) { @@ -4311,29 +5515,45 @@ ofopgroup_complete(struct ofopgroup *group) } } } else { - oftable_substitute_rule(rule, op->victim); - ofproto_rule_destroy__(rule); + ovs_rwlock_wrlock(&rule->rwlock); + oftable_remove_rule(rule); + ofproto_rule_destroy(rule); } break; case OFOPERATION_DELETE: - assert(!op->error); - ofproto_rule_destroy__(rule); + ovs_assert(!op->error); + ofproto_rule_destroy(rule); op->rule = NULL; break; case OFOPERATION_MODIFY: + case OFOPERATION_REPLACE: if (!op->error) { - rule->modified = time_msec(); + long long int now = time_msec(); + + rule->modified = now; + if (op->type == OFOPERATION_REPLACE) { + rule->created = rule->used = now; + } } else { - rule->flow_cookie = op->flow_cookie; + ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie); + ovs_mutex_lock(&rule->timeout_mutex); + rule->idle_timeout = op->idle_timeout; + rule->hard_timeout = op->hard_timeout; + ovs_mutex_unlock(&rule->timeout_mutex); if (op->ofpacts) { free(rule->ofpacts); + + ovs_rwlock_wrlock(&rule->rwlock); rule->ofpacts = op->ofpacts; rule->ofpacts_len = op->ofpacts_len; + ovs_rwlock_unlock(&rule->rwlock); + op->ofpacts = NULL; op->ofpacts_len = 0; } + rule->send_flow_removed = op->send_flow_removed; } break; @@ -4347,7 +5567,7 @@ 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 +5598,7 @@ 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; @@ -4387,6 +5607,11 @@ ofoperation_create(struct ofopgroup *group, struct rule *rule, op->type = type; op->reason = reason; op->flow_cookie = rule->flow_cookie; + ovs_mutex_lock(&rule->timeout_mutex); + op->idle_timeout = rule->idle_timeout; + op->hard_timeout = rule->hard_timeout; + ovs_mutex_unlock(&rule->timeout_mutex); + op->send_flow_removed = rule->send_flow_removed; group->n_running++; @@ -4418,14 +5643,7 @@ ofoperation_destroy(struct ofoperation *op) * indicate success or an OpenFlow error code on failure. * * If 'error' is 0, indicating success, the operation will be committed - * permanently to the flow table. There is one interesting subcase: - * - * - If 'op' is an "add flow" operation that is replacing an existing rule in - * the flow table (the "victim" rule) by a new one, then the caller must - * have uninitialized any derived state in the victim rule, as in step 5 in - * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete() - * performs steps 6 and 7 for the victim rule, most notably by calling its - * ->rule_dealloc() function. + * permanently to the flow table. * * If 'error' is nonzero, then generally the operation will be rolled back: * @@ -4448,22 +5666,15 @@ 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)) { ofopgroup_complete(group); } } - -struct rule * -ofoperation_get_victim(struct ofoperation *op) -{ - assert(op->type == OFOPERATION_ADD); - return op->victim; -} static uint64_t pick_datapath_id(const struct ofproto *ofproto) @@ -4481,7 +5692,7 @@ 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; } @@ -4496,17 +5707,18 @@ 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 @@ -4520,18 +5732,19 @@ 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->rwlock)) { + *rulep = rule; + return true; } } } - return NULL; + return false; } /* Searches 'ofproto' for tables that have more flows than their configured @@ -4548,19 +5761,31 @@ 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; + } + + if (!choose_rule_to_evict(table, &rule)) { + break; + } - rule = choose_rule_to_evict(table); - if (!rule || rule->pending) { + if (rule->pending) { + ovs_rwlock_unlock(&rule->rwlock); break; } ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_EVICTION); oftable_remove_rule(rule); - ofproto->ofproto_class->rule_destruct(rule); + ofproto->ofproto_class->rule_delete(rule); } } ofopgroup_submit(group); @@ -4697,6 +5922,7 @@ 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); @@ -4704,6 +5930,7 @@ 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; } @@ -4730,9 +5957,13 @@ 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)); @@ -4761,7 +5992,9 @@ oftable_init(struct oftable *table) static void oftable_destroy(struct oftable *table) { - assert(classifier_is_empty(&table->cls)); + 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); @@ -4841,50 +6074,73 @@ 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->rwlock) { - struct ofproto *ofproto = rule->ofproto; - struct oftable *table = &ofproto->tables[rule->table_id]; - - classifier_remove(&table->cls, &rule->cr); + classifier_remove(cls, &rule->cr); + 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); + list_init(&rule->meter_list_node); + } + ovs_rwlock_unlock(&rule->rwlock); } -/* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s - * oftable that has an identical cls_rule. Returns the rule that was removed, - * if any, and otherwise NULL. */ -static struct rule * -oftable_replace_rule(struct rule *rule) +static void +oftable_remove_rule(struct rule *rule) { struct ofproto *ofproto = rule->ofproto; struct oftable *table = &ofproto->tables[rule->table_id]; - struct rule *victim; - victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr)); - if (victim) { - eviction_group_remove_rule(victim); - } - eviction_group_add_rule(rule); - return victim; + ovs_rwlock_wrlock(&table->cls.rwlock); + oftable_remove_rule__(ofproto, &table->cls, rule); + ovs_rwlock_unlock(&table->cls.rwlock); } -/* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */ +/* Inserts 'rule' into its oftable, which must not already contain any rule for + * the same cls_rule. */ static void -oftable_substitute_rule(struct rule *old, struct rule *new) +oftable_insert_rule(struct rule *rule) { - if (new) { - oftable_replace_rule(new); - } else { - oftable_remove_rule(old); + struct ofproto *ofproto = rule->ofproto; + struct oftable *table = &ofproto->tables[rule->table_id]; + 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); + classifier_insert(&table->cls, &rule->cr); + ovs_rwlock_unlock(&table->cls.rwlock); + eviction_group_add_rule(rule); } /* unixctl commands. */ @@ -4982,13 +6238,13 @@ ofproto_has_vlan_usage_changed(const struct ofproto *ofproto) * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero, * then the VLAN device is un-enslaved. */ int -ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port, - uint16_t realdev_ofp_port, int vid) +ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port, + ofp_port_t realdev_ofp_port, int vid) { 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) { @@ -5009,7 +6265,7 @@ ofproto_port_set_realdev(struct ofproto *ofproto, uint16_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; }