X-Git-Url: http://git.cascardo.eti.br/?a=blobdiff_plain;f=ofproto%2Fofproto.c;h=4fa045ff4aaba4b4c74dc98793e7ec6e2b38be0b;hb=83b03fe05e7a6734b2096dab86937294769987c6;hp=a1f73c07a8ef667865815c8f3ec59de79528f436;hpb=afae68b16f01559df44e3fd62f1fc020faec5731;p=cascardo%2Fovs.git diff --git a/ofproto/ofproto.c b/ofproto/ofproto.c index a1f73c07a..4fa045ff4 100644 --- a/ofproto/ofproto.c +++ b/ofproto/ofproto.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc. + * Copyright (c) 2009-2016 Nicira, Inc. * Copyright (c) 2010 Jean Tourrilhes - HP-Labs. * * Licensed under the Apache License, Version 2.0 (the "License"); @@ -44,6 +44,7 @@ #include "openflow/nicira-ext.h" #include "openflow/openflow.h" #include "ovs-rcu.h" +#include "dp-packet.h" #include "packets.h" #include "pinsched.h" #include "pktbuf.h" @@ -55,9 +56,10 @@ #include "smap.h" #include "sset.h" #include "timeval.h" +#include "tun-metadata.h" #include "unaligned.h" #include "unixctl.h" -#include "vlog.h" +#include "openvswitch/vlog.h" #include "bundles.h" VLOG_DEFINE_THIS_MODULE(ofproto); @@ -80,18 +82,20 @@ static void oftable_destroy(struct oftable *); static void oftable_set_name(struct oftable *, const char *name); -static enum ofperr evict_rules_from_table(struct oftable *, - unsigned int extra_space) +static enum ofperr evict_rules_from_table(struct oftable *) OVS_REQUIRES(ofproto_mutex); -static void oftable_disable_eviction(struct oftable *); -static void oftable_enable_eviction(struct oftable *, - const struct mf_subfield *fields, - size_t n_fields); - -static void oftable_remove_rule(struct rule *rule) OVS_REQUIRES(ofproto_mutex); -static void oftable_remove_rule__(struct ofproto *, struct rule *) +static void oftable_configure_eviction(struct oftable *, + unsigned int eviction, + const struct mf_subfield *fields, + size_t n_fields) OVS_REQUIRES(ofproto_mutex); +/* This is the only combination of OpenFlow eviction flags that OVS supports: a + * combination of OF1.4+ importance, the remaining lifetime of the flow, and + * fairness based on user-specified fields. */ +#define OFPROTO_EVICTION_FLAGS \ + (OFPTMPEF14_OTHER | OFPTMPEF14_IMPORTANCE | OFPTMPEF14_LIFETIME) + /* 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 * needed, is taken from the eviction group that contains the greatest number @@ -116,8 +120,8 @@ struct eviction_group { static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep) OVS_REQUIRES(ofproto_mutex); -static uint32_t rule_eviction_priority(struct ofproto *ofproto, struct rule *) - OVS_REQUIRES(ofproto_mutex);; +static uint64_t rule_eviction_priority(struct ofproto *ofproto, struct rule *) + OVS_REQUIRES(ofproto_mutex); static void eviction_group_add_rule(struct rule *) OVS_REQUIRES(ofproto_mutex); static void eviction_group_remove_rule(struct rule *) @@ -133,6 +137,7 @@ struct rule_criteria { * collect_rules_loose() and "strict" way by collect_rules_strict(), as * defined in the OpenFlow spec. */ struct cls_rule cr; + cls_version_t version; /* Matching criteria for the OpenFlow cookie. Consider a bit B in a rule's * cookie and the corresponding bits C in 'cookie' and M in 'cookie_mask'. @@ -153,8 +158,8 @@ struct rule_criteria { }; static void rule_criteria_init(struct rule_criteria *, uint8_t table_id, - const struct match *match, - unsigned int priority, + const struct match *match, int priority, + cls_version_t version, ovs_be64 cookie, ovs_be64 cookie_mask, ofp_port_t out_port, uint32_t out_group); static void rule_criteria_require_rw(struct rule_criteria *, @@ -170,10 +175,10 @@ static enum ofperr collect_rules_loose(struct ofproto *, * (We can't do this immediately from ofopgroup_complete() because that holds * ofproto_mutex, which rule_execute() needs released.) */ struct rule_execute { - struct list list_node; /* In struct ofproto's "rule_executes" list. */ + struct ovs_list list_node; /* In struct ofproto's "rule_executes" list. */ struct rule *rule; /* Owns a reference to the rule. */ ofp_port_t in_port; - struct ofpbuf *packet; /* Owns the packet. */ + struct dp_packet *packet; /* Owns the packet. */ }; static void run_rule_executes(struct ofproto *) OVS_EXCLUDED(ofproto_mutex); @@ -185,7 +190,7 @@ struct learned_cookie { struct hmap_node hmap_node OVS_GUARDED_BY(ofproto_mutex); /* In 'dead_cookies' list when removed from hmap. */ - struct list list_node; + struct ovs_list list_node; } u; /* Key. */ @@ -205,9 +210,9 @@ static const struct ofpact_learn *next_learn_with_delete( static void learned_cookies_inc(struct ofproto *, const struct rule_actions *) OVS_REQUIRES(ofproto_mutex); static void learned_cookies_dec(struct ofproto *, const struct rule_actions *, - struct list *dead_cookies) + struct ovs_list *dead_cookies) OVS_REQUIRES(ofproto_mutex); -static void learned_cookies_flush(struct ofproto *, struct list *dead_cookies) +static void learned_cookies_flush(struct ofproto *, struct ovs_list *dead_cookies) OVS_REQUIRES(ofproto_mutex); /* ofport. */ @@ -236,8 +241,13 @@ struct ofport_usage { }; /* rule. */ -static void ofproto_rule_send_removed(struct rule *, uint8_t reason); +static void ofproto_rule_send_removed(struct rule *) + OVS_EXCLUDED(ofproto_mutex); static bool rule_is_readonly(const struct rule *); +static void ofproto_rule_insert__(struct ofproto *, struct rule *) + OVS_REQUIRES(ofproto_mutex); +static void ofproto_rule_remove__(struct ofproto *, struct rule *) + OVS_REQUIRES(ofproto_mutex); /* The source of a flow_mod request, in the code that processes flow_mods. * @@ -247,23 +257,38 @@ static bool rule_is_readonly(const struct rule *); * meaningful and thus supplied as NULL. */ struct flow_mod_requester { struct ofconn *ofconn; /* Connection on which flow_mod arrived. */ - ovs_be32 xid; /* OpenFlow xid of flow_mod request. */ + const struct ofp_header *request; }; /* OpenFlow. */ -static enum ofperr add_flow(struct ofproto *, struct ofputil_flow_mod *, - const struct flow_mod_requester *); +static enum ofperr replace_rule_create(struct ofproto *, + struct ofputil_flow_mod *, + struct cls_rule *cr, uint8_t table_id, + struct rule *old_rule, + struct rule **new_rule) + OVS_REQUIRES(ofproto_mutex); + +static void replace_rule_start(struct ofproto *, cls_version_t version, + struct rule *old_rule, struct rule *new_rule, + struct cls_conjunction *, size_t n_conjs) + OVS_REQUIRES(ofproto_mutex); -static enum ofperr modify_flows__(struct ofproto *, struct ofputil_flow_mod *, - const struct rule_collection *, - const struct flow_mod_requester *); -static void delete_flows__(const struct rule_collection *, +static void replace_rule_revert(struct ofproto *, struct rule *old_rule, + struct rule *new_rule) + OVS_REQUIRES(ofproto_mutex); + +static void replace_rule_finish(struct ofproto *, struct ofputil_flow_mod *, + const struct flow_mod_requester *, + struct rule *old_rule, struct rule *new_rule, + struct ovs_list *dead_cookies) + OVS_REQUIRES(ofproto_mutex); +static void delete_flows__(struct rule_collection *, enum ofp_flow_removed_reason, const struct flow_mod_requester *) OVS_REQUIRES(ofproto_mutex); -static enum ofperr send_buffered_packet(struct ofconn *, uint32_t buffer_id, - struct rule *) +static void send_buffered_packet(const struct flow_mod_requester *, + uint32_t buffer_id, struct rule *) OVS_REQUIRES(ofproto_mutex); static bool ofproto_group_exists__(const struct ofproto *ofproto, @@ -274,8 +299,15 @@ static bool ofproto_group_exists(const struct ofproto *ofproto, OVS_EXCLUDED(ofproto->groups_rwlock); static enum ofperr add_group(struct ofproto *, struct ofputil_group_mod *); static void handle_openflow(struct ofconn *, const struct ofpbuf *); +static enum ofperr ofproto_flow_mod_start(struct ofproto *, + struct ofproto_flow_mod *) + OVS_REQUIRES(ofproto_mutex); +static void ofproto_flow_mod_finish(struct ofproto *, + struct ofproto_flow_mod *, + const struct flow_mod_requester *) + OVS_REQUIRES(ofproto_mutex); static enum ofperr handle_flow_mod__(struct ofproto *, - struct ofputil_flow_mod *, + struct ofproto_flow_mod *, const struct flow_mod_requester *) OVS_EXCLUDED(ofproto_mutex); static void calc_duration(long long int start, long long int now, @@ -304,6 +336,8 @@ unsigned ofproto_flow_limit = OFPROTO_FLOW_LIMIT_DEFAULT; unsigned ofproto_max_idle = OFPROTO_MAX_IDLE_DEFAULT; size_t n_handlers, n_revalidators; +size_t n_dpdk_rxqs; +char *pmd_cpu_mask; /* Map from datapath name to struct ofproto, for use by unixctl commands. */ static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos); @@ -349,6 +383,8 @@ ofproto_init(const struct shash *iface_hints) for (i = 0; i < n_ofproto_classes; i++) { ofproto_classes[i]->init(&init_ofp_ports); } + + ofproto_unixctl_init(); } /* 'type' should be a normalized datapath type, as returned by @@ -458,6 +494,14 @@ ofproto_enumerate_names(const char *type, struct sset *names) return class ? class->enumerate_names(type, names) : EAFNOSUPPORT; } +static void +ofproto_bump_tables_version(struct ofproto *ofproto) +{ + ++ofproto->tables_version; + ofproto->ofproto_class->set_tables_version(ofproto, + ofproto->tables_version); +} + int ofproto_create(const char *datapath_name, const char *datapath_type, struct ofproto **ofprotop) @@ -469,8 +513,6 @@ ofproto_create(const char *datapath_name, const char *datapath_type, *ofprotop = NULL; - ofproto_unixctl_init(); - datapath_type = ofproto_normalize_type(datapath_type); class = ofproto_class_find__(datapath_type); if (!class) { @@ -511,6 +553,7 @@ ofproto_create(const char *datapath_name, const char *datapath_type, ofproto->eviction_group_timer = LLONG_MIN; ofproto->tables = NULL; ofproto->n_tables = 0; + ofproto->tables_version = CLS_MIN_VERSION; hindex_init(&ofproto->cookies); hmap_init(&ofproto->learned_cookies); list_init(&ofproto->expirable); @@ -522,35 +565,20 @@ ofproto_create(const char *datapath_name, const char *datapath_type, ovs_rwlock_init(&ofproto->groups_rwlock); hmap_init(&ofproto->groups); ovs_mutex_unlock(&ofproto_mutex); + ofproto->ogf.types = 0xf; ofproto->ogf.capabilities = OFPGFC_CHAINING | OFPGFC_SELECT_LIVENESS | OFPGFC_SELECT_WEIGHT; - ofproto->ogf.max_groups[OFPGT11_ALL] = OFPG_MAX; - ofproto->ogf.max_groups[OFPGT11_SELECT] = OFPG_MAX; - ofproto->ogf.max_groups[OFPGT11_INDIRECT] = OFPG_MAX; - ofproto->ogf.max_groups[OFPGT11_FF] = OFPG_MAX; - ofproto->ogf.actions[0] = - (1 << OFPAT11_OUTPUT) | - (1 << OFPAT11_COPY_TTL_OUT) | - (1 << OFPAT11_COPY_TTL_IN) | - (1 << OFPAT11_SET_MPLS_TTL) | - (1 << OFPAT11_DEC_MPLS_TTL) | - (1 << OFPAT11_PUSH_VLAN) | - (1 << OFPAT11_POP_VLAN) | - (1 << OFPAT11_PUSH_MPLS) | - (1 << OFPAT11_POP_MPLS) | - (1 << OFPAT11_SET_QUEUE) | - (1 << OFPAT11_GROUP) | - (1 << OFPAT11_SET_NW_TTL) | - (1 << OFPAT11_DEC_NW_TTL) | - (1 << OFPAT12_SET_FIELD); -/* not supported: - * (1 << OFPAT13_PUSH_PBB) | - * (1 << OFPAT13_POP_PBB) */ + for (i = 0; i < 4; i++) { + ofproto->ogf.max_groups[i] = OFPG_MAX; + ofproto->ogf.ofpacts[i] = (UINT64_C(1) << N_OFPACTS) - 1; + } + tun_metadata_init(); error = ofproto->ofproto_class->construct(ofproto); if (error) { VLOG_ERR("failed to open datapath %s: %s", datapath_name, ovs_strerror(error)); + connmgr_destroy(ofproto->connmgr); ofproto_destroy__(ofproto); return error; } @@ -577,6 +605,9 @@ ofproto_create(const char *datapath_name, const char *datapath_type, ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1) * sizeof(struct meter *)); + /* Set the initial tables version. */ + ofproto_bump_tables_version(ofproto); + *ofprotop = ofproto; return 0; } @@ -736,18 +767,32 @@ ofproto_set_mcast_snooping(struct ofproto *ofproto, : EOPNOTSUPP); } -/* Configures multicast snooping flood setting on 'ofp_port' of 'ofproto'. +/* Configures multicast snooping flood settings on 'ofp_port' of 'ofproto'. * * Returns 0 if successful, otherwise a positive errno value.*/ int -ofproto_port_set_mcast_snooping(struct ofproto *ofproto, void *aux, bool flood) +ofproto_port_set_mcast_snooping(struct ofproto *ofproto, void *aux, + const struct ofproto_mcast_snooping_port_settings *s) { return (ofproto->ofproto_class->set_mcast_snooping_port - ? ofproto->ofproto_class->set_mcast_snooping_port(ofproto, aux, - flood) + ? ofproto->ofproto_class->set_mcast_snooping_port(ofproto, aux, s) : EOPNOTSUPP); } +void +ofproto_set_n_dpdk_rxqs(int n_rxqs) +{ + n_dpdk_rxqs = MAX(n_rxqs, 0); +} + +void +ofproto_set_cpu_mask(const char *cmask) +{ + free(pmd_cpu_mask); + + pmd_cpu_mask = cmask ? xstrdup(cmask) : NULL; +} + void ofproto_set_threads(int n_handlers_, int n_revalidators_) { @@ -929,6 +974,87 @@ ofproto_port_get_stp_stats(struct ofproto *ofproto, ofp_port_t ofp_port, ? ofproto->ofproto_class->get_stp_port_stats(ofport, s) : EOPNOTSUPP); } + +/* Rapid Spanning Tree Protocol (RSTP) configuration. */ + +/* Configures RSTP on 'ofproto' using the settings defined in 's'. If + * 's' is NULL, disables RSTP. + * + * Returns 0 if successful, otherwise a positive errno value. */ +int +ofproto_set_rstp(struct ofproto *ofproto, + const struct ofproto_rstp_settings *s) +{ + if (!ofproto->ofproto_class->set_rstp) { + return EOPNOTSUPP; + } + ofproto->ofproto_class->set_rstp(ofproto, s); + return 0; +} + +/* Retrieves RSTP status of 'ofproto' and stores it in 's'. If the + * 'enabled' member of 's' is false, then the other members are not + * meaningful. + * + * Returns 0 if successful, otherwise a positive errno value. */ +int +ofproto_get_rstp_status(struct ofproto *ofproto, + struct ofproto_rstp_status *s) +{ + if (!ofproto->ofproto_class->get_rstp_status) { + return EOPNOTSUPP; + } + ofproto->ofproto_class->get_rstp_status(ofproto, s); + return 0; +} + +/* Configures RSTP on 'ofp_port' of 'ofproto' using the settings defined + * in 's'. The caller is responsible for assigning RSTP port numbers + * (using the 'port_num' member in the range of 1 through 255, inclusive) + * and ensuring there are no duplicates. If the 's' is NULL, then RSTP + * is disabled on the port. + * + * Returns 0 if successful, otherwise a positive errno value.*/ +int +ofproto_port_set_rstp(struct ofproto *ofproto, ofp_port_t ofp_port, + const struct ofproto_port_rstp_settings *s) +{ + struct ofport *ofport = ofproto_get_port(ofproto, ofp_port); + if (!ofport) { + VLOG_WARN("%s: cannot configure RSTP on nonexistent port %"PRIu16, + ofproto->name, ofp_port); + return ENODEV; + } + + if (!ofproto->ofproto_class->set_rstp_port) { + return EOPNOTSUPP; + } + ofproto->ofproto_class->set_rstp_port(ofport, s); + return 0; +} + +/* Retrieves RSTP port status of 'ofp_port' on 'ofproto' and stores it in + * 's'. If the 'enabled' member in 's' is false, then the other members + * are not meaningful. + * + * Returns 0 if successful, otherwise a positive errno value.*/ +int +ofproto_port_get_rstp_status(struct ofproto *ofproto, ofp_port_t ofp_port, + struct ofproto_port_rstp_status *s) +{ + struct ofport *ofport = ofproto_get_port(ofproto, ofp_port); + if (!ofport) { + VLOG_WARN_RL(&rl, "%s: cannot get RSTP status on nonexistent " + "port %"PRIu16, ofproto->name, ofp_port); + return ENODEV; + } + + if (!ofproto->ofproto_class->get_rstp_port_status) { + return EOPNOTSUPP; + } + ofproto->ofproto_class->get_rstp_port_status(ofport, s); + return 0; +} /* Queue DSCP configuration. */ @@ -957,6 +1083,83 @@ ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port, : EOPNOTSUPP); } +/* LLDP configuration. */ +void +ofproto_port_set_lldp(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 LLDP on nonexistent port %"PRIu16, + ofproto->name, ofp_port); + return; + } + error = (ofproto->ofproto_class->set_lldp + ? ofproto->ofproto_class->set_lldp(ofport, cfg) + : EOPNOTSUPP); + if (error) { + VLOG_WARN("%s: lldp configuration on port %"PRIu16" (%s) failed (%s)", + ofproto->name, ofp_port, netdev_get_name(ofport->netdev), + ovs_strerror(error)); + } +} + +int +ofproto_set_aa(struct ofproto *ofproto, void *aux OVS_UNUSED, + const struct aa_settings *s) +{ + if (!ofproto->ofproto_class->set_aa) { + return EOPNOTSUPP; + } + ofproto->ofproto_class->set_aa(ofproto, s); + return 0; +} + +int +ofproto_aa_mapping_register(struct ofproto *ofproto, void *aux, + const struct aa_mapping_settings *s) +{ + if (!ofproto->ofproto_class->aa_mapping_set) { + return EOPNOTSUPP; + } + ofproto->ofproto_class->aa_mapping_set(ofproto, aux, s); + return 0; +} + +int +ofproto_aa_mapping_unregister(struct ofproto *ofproto, void *aux) +{ + if (!ofproto->ofproto_class->aa_mapping_unset) { + return EOPNOTSUPP; + } + ofproto->ofproto_class->aa_mapping_unset(ofproto, aux); + return 0; +} + +int +ofproto_aa_vlan_get_queued(struct ofproto *ofproto, + struct ovs_list *list) +{ + if (!ofproto->ofproto_class->aa_vlan_get_queued) { + return EOPNOTSUPP; + } + ofproto->ofproto_class->aa_vlan_get_queued(ofproto, list); + return 0; +} + +unsigned int +ofproto_aa_vlan_get_queue_size(struct ofproto *ofproto) +{ + if (!ofproto->ofproto_class->aa_vlan_get_queue_size) { + return EOPNOTSUPP; + } + return ofproto->ofproto_class->aa_vlan_get_queue_size(ofproto); +} + /* Connectivity Fault Management configuration. */ /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */ @@ -1067,6 +1270,21 @@ ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port) ? ofproto->ofproto_class->port_is_lacp_current(ofport) : -1); } + +int +ofproto_port_get_lacp_stats(const struct ofport *port, struct lacp_slave_stats *stats) +{ + struct ofproto *ofproto = port->ofproto; + int error; + + if (ofproto->ofproto_class->port_get_lacp_stats) { + error = ofproto->ofproto_class->port_get_lacp_stats(port, stats); + } else { + error = EOPNOTSUPP; + } + + return error; +} /* Bundles. */ @@ -1122,7 +1340,8 @@ ofproto_mirror_unregister(struct ofproto *ofproto, void *aux) /* Retrieves statistics from mirror associated with client data pointer * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and * 'bytes', respectively. If a particular counters is not supported, - * the appropriate argument is set to UINT64_MAX. */ + * the appropriate argument is set to UINT64_MAX. + */ int ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux, uint64_t *packets, uint64_t *bytes) @@ -1207,21 +1426,18 @@ ofproto_configure_table(struct ofproto *ofproto, int table_id, return; } - if (s->groups) { - oftable_enable_eviction(table, s->groups, s->n_groups); - } else { - oftable_disable_eviction(table); - } - - table->max_flows = s->max_flows; - if (classifier_set_prefix_fields(&table->cls, s->prefix_fields, s->n_prefix_fields)) { /* XXX: Trigger revalidation. */ } ovs_mutex_lock(&ofproto_mutex); - evict_rules_from_table(table, 0); + unsigned int new_eviction = (s->enable_eviction + ? table->eviction | EVICTION_CLIENT + : table->eviction & ~EVICTION_CLIENT); + oftable_configure_eviction(table, new_eviction, s->groups, s->n_groups); + table->max_flows = s->max_flows; + evict_rules_from_table(table); ovs_mutex_unlock(&ofproto_mutex); } @@ -1237,18 +1453,6 @@ ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops) connmgr_get_snoops(ofproto->connmgr, snoops); } -static void -ofproto_rule_delete__(struct rule *rule, uint8_t reason) - OVS_REQUIRES(ofproto_mutex) -{ - struct rule_collection rules; - - rules.rules = rules.stub; - rules.n = 1; - rules.stub[0] = rule; - delete_flows__(&rules, reason, NULL); -} - /* Deletes 'rule' from 'ofproto'. * * Within an ofproto implementation, this function allows an ofproto @@ -1266,8 +1470,19 @@ ofproto_rule_delete(struct ofproto *ofproto, struct rule *rule) * switch is being deleted and any OpenFlow channels have been or soon will * be killed. */ ovs_mutex_lock(&ofproto_mutex); - oftable_remove_rule__(ofproto, rule); - ofproto->ofproto_class->rule_delete(rule); + + if (!rule->removed) { + /* Make sure there is no postponed removal of the rule. */ + ovs_assert(cls_rule_visible_in_version(&rule->cr, CLS_MAX_VERSION)); + + if (!classifier_remove(&rule->ofproto->tables[rule->table_id].cls, + &rule->cr)) { + OVS_NOT_REACHED(); + } + ofproto_rule_remove__(rule->ofproto, rule); + ofproto->ofproto_class->rule_delete(rule); + ofproto_rule_unref(rule); + } ovs_mutex_unlock(&ofproto_mutex); } @@ -1277,22 +1492,36 @@ ofproto_flush__(struct ofproto *ofproto) { struct oftable *table; + /* This will flush all datapath flows. */ if (ofproto->ofproto_class->flush) { ofproto->ofproto_class->flush(ofproto); } + /* XXX: There is a small race window here, where new datapath flows can be + * created by upcall handlers based on the existing flow table. We can not + * call ofproto class flush while holding 'ofproto_mutex' to prevent this, + * as then we could deadlock on syncing with the handler threads waiting on + * the same mutex. */ + ovs_mutex_lock(&ofproto_mutex); OFPROTO_FOR_EACH_TABLE (table, ofproto) { - struct rule *rule, *next_rule; + struct rule_collection rules; + struct rule *rule; if (table->flags & OFTABLE_HIDDEN) { continue; } - CLS_FOR_EACH_SAFE (rule, next_rule, cr, &table->cls) { - ofproto_rule_delete__(rule, OFPRR_DELETE); + rule_collection_init(&rules); + + CLS_FOR_EACH (rule, cr, &table->cls) { + rule_collection_add(&rules, rule); } + delete_flows__(&rules, OFPRR_DELETE, NULL); } + /* XXX: Concurrent handler threads may insert new learned flows based on + * learn actions of the now deleted flows right after we release + * 'ofproto_mutex'. */ ovs_mutex_unlock(&ofproto_mutex); } @@ -1305,14 +1534,11 @@ ofproto_destroy__(struct ofproto *ofproto) struct oftable *table; destroy_rule_executes(ofproto); - delete_group(ofproto, OFPG_ALL); guarded_list_destroy(&ofproto->rule_executes); ovs_rwlock_destroy(&ofproto->groups_rwlock); hmap_destroy(&ofproto->groups); - connmgr_destroy(ofproto->connmgr); - hmap_remove(&all_ofprotos, &ofproto->hmap_node); free(ofproto->name); free(ofproto->type); @@ -1342,6 +1568,16 @@ ofproto_destroy__(struct ofproto *ofproto) ofproto->ofproto_class->dealloc(ofproto); } +/* Destroying rules is doubly deferred, must have 'ofproto' around for them. + * - 1st we defer the removal of the rules from the classifier + * - 2nd we defer the actual destruction of the rules. */ +static void +ofproto_destroy_defer__(struct ofproto *ofproto) + OVS_EXCLUDED(ofproto_mutex) +{ + ovsrcu_postpone(ofproto_destroy__, ofproto); +} + void ofproto_destroy(struct ofproto *p) OVS_EXCLUDED(ofproto_mutex) @@ -1371,8 +1607,14 @@ ofproto_destroy(struct ofproto *p) } p->ofproto_class->destruct(p); + + /* We should not postpone this because it involves deleting a listening + * socket which we may want to reopen soon. 'connmgr' should not be used + * by other threads */ + connmgr_destroy(p->connmgr); + /* Destroying rules is deferred, must have 'ofproto' around for them. */ - ovsrcu_postpone(ofproto_destroy__, p); + ovsrcu_postpone(ofproto_destroy_defer__, p); } /* Destroys the datapath with the respective 'name' and 'type'. With the Linux @@ -1455,10 +1697,17 @@ ofproto_run(struct ofproto *p) struct eviction_group *evg; struct rule *rule; - if (!table->eviction_fields) { + if (!table->eviction) { continue; } + if (table->n_flows > 100000) { + static struct vlog_rate_limit count_rl = + VLOG_RATE_LIMIT_INIT(1, 1); + VLOG_WARN_RL(&count_rl, "Table %"PRIuSIZE" has an excessive" + " number of rules: %d", i, table->n_flows); + } + ovs_mutex_lock(&ofproto_mutex); CLS_FOR_EACH (rule, cr, &table->cls) { if (rule->idle_timeout || rule->hard_timeout) { @@ -1550,7 +1799,7 @@ ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage) n_rules = 0; OFPROTO_FOR_EACH_TABLE (table, ofproto) { - n_rules += classifier_count(&table->cls); + n_rules += table->n_flows; } simap_increase(usage, "rules", n_rules); @@ -1780,40 +2029,35 @@ ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port) static void flow_mod_init(struct ofputil_flow_mod *fm, - const struct match *match, unsigned int priority, + const struct match *match, int priority, const struct ofpact *ofpacts, size_t ofpacts_len, enum ofp_flow_mod_command command) { - memset(fm, 0, sizeof *fm); - fm->match = *match; - fm->priority = priority; - fm->cookie = 0; - fm->new_cookie = 0; - fm->modify_cookie = false; - fm->table_id = 0; - fm->command = command; - fm->idle_timeout = 0; - fm->hard_timeout = 0; - fm->buffer_id = UINT32_MAX; - fm->out_port = OFPP_ANY; - fm->out_group = OFPG_ANY; - fm->flags = 0; - fm->ofpacts = CONST_CAST(struct ofpact *, ofpacts); - fm->ofpacts_len = ofpacts_len; - fm->delete_reason = OFPRR_DELETE; + *fm = (struct ofputil_flow_mod) { + .match = *match, + .priority = priority, + .table_id = 0, + .command = command, + .buffer_id = UINT32_MAX, + .out_port = OFPP_ANY, + .out_group = OFPG_ANY, + .ofpacts = CONST_CAST(struct ofpact *, ofpacts), + .ofpacts_len = ofpacts_len, + .delete_reason = OFPRR_DELETE, + }; } static int simple_flow_mod(struct ofproto *ofproto, - const struct match *match, unsigned int priority, + const struct match *match, int priority, const struct ofpact *ofpacts, size_t ofpacts_len, enum ofp_flow_mod_command command) { - struct ofputil_flow_mod fm; + struct ofproto_flow_mod ofm; - flow_mod_init(&fm, match, priority, ofpacts, ofpacts_len, command); + flow_mod_init(&ofm.fm, match, priority, ofpacts, ofpacts_len, command); - return handle_flow_mod__(ofproto, &fm, NULL); + return handle_flow_mod__(ofproto, &ofm, NULL); } /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and @@ -1829,7 +2073,7 @@ simple_flow_mod(struct ofproto *ofproto, * This is a helper function for in-band control and fail-open. */ void ofproto_add_flow(struct ofproto *ofproto, const struct match *match, - unsigned int priority, + int priority, const struct ofpact *ofpacts, size_t ofpacts_len) OVS_EXCLUDED(ofproto_mutex) { @@ -1839,7 +2083,8 @@ ofproto_add_flow(struct ofproto *ofproto, const struct match *match, /* First do a cheap check whether the rule we're looking for already exists * with the actions that we want. If it does, then we're done. */ rule = rule_from_cls_rule(classifier_find_match_exactly( - &ofproto->tables[0].cls, match, priority)); + &ofproto->tables[0].cls, match, priority, + CLS_MAX_VERSION)); if (rule) { const struct rule_actions *actions = rule_get_actions(rule); must_add = !ofpacts_equal(actions->ofpacts, actions->ofpacts_len, @@ -1858,16 +2103,17 @@ ofproto_add_flow(struct ofproto *ofproto, const struct match *match, } } -/* Executes the flow modification specified in 'fm'. Returns 0 on success, an - * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the - * operation cannot be initiated now but may be retried later. +/* Executes the flow modification specified in 'fm'. Returns 0 on success, or + * an OFPERR_* OpenFlow error code on failure. * * This is a helper function for in-band control and fail-open and the "learn" * action. */ -int -ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm) +enum ofperr +ofproto_flow_mod(struct ofproto *ofproto, struct ofproto_flow_mod *ofm) OVS_EXCLUDED(ofproto_mutex) { + struct ofputil_flow_mod *fm = &ofm->fm; + /* Optimize for the most common case of a repeated learn action. * If an identical flow already exists we only need to update its * 'modified' time. */ @@ -1877,9 +2123,9 @@ ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm) struct rule *rule; bool done = false; - rule = rule_from_cls_rule(classifier_find_match_exactly(&table->cls, - &fm->match, - fm->priority)); + rule = rule_from_cls_rule(classifier_find_match_exactly( + &table->cls, &fm->match, fm->priority, + CLS_MAX_VERSION)); if (rule) { /* Reading many of the rule fields and writing on 'modified' * requires the rule->mutex. Also, rule->actions may change @@ -1890,6 +2136,7 @@ ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm) actions = rule_get_actions(rule); if (rule->idle_timeout == fm->idle_timeout && rule->hard_timeout == fm->hard_timeout + && rule->importance == fm->importance && rule->flags == (fm->flags & OFPUTIL_FF_STATE) && (!fm->modify_cookie || (fm->new_cookie == rule->flow_cookie)) && ofpacts_equal(fm->ofpacts, fm->ofpacts_len, @@ -1907,7 +2154,7 @@ ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm) } } - return handle_flow_mod__(ofproto, fm, NULL); + return handle_flow_mod__(ofproto, ofm, NULL); } /* Searches for a rule with matching criteria exactly equal to 'target' in @@ -1916,7 +2163,7 @@ ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm) * This is a helper function for in-band control and fail-open. */ void ofproto_delete_flow(struct ofproto *ofproto, - const struct match *target, unsigned int priority) + const struct match *target, int priority) OVS_EXCLUDED(ofproto_mutex) { struct classifier *cls = &ofproto->tables[0].cls; @@ -1924,8 +2171,8 @@ ofproto_delete_flow(struct ofproto *ofproto, /* First do a cheap check whether the rule we're looking for has already * been deleted. If so, then we're done. */ - rule = rule_from_cls_rule(classifier_find_match_exactly(cls, target, - priority)); + rule = rule_from_cls_rule(classifier_find_match_exactly( + cls, target, priority, CLS_MAX_VERSION)); if (!rule) { return; } @@ -2064,7 +2311,7 @@ ofport_open(struct ofproto *ofproto, } } pp->port_no = ofproto_port->ofp_port; - netdev_get_etheraddr(netdev, pp->hw_addr); + netdev_get_etheraddr(netdev, &pp->hw_addr); ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name); netdev_get_flags(netdev, &flags); pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN; @@ -2171,7 +2418,7 @@ ofport_remove_with_name(struct ofproto *ofproto, const char *name) static void ofport_modified(struct ofport *port, struct ofputil_phy_port *pp) { - memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN); + port->pp.hw_addr = pp->hw_addr; port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN) | (pp->config & OFPUTIL_PC_PORT_DOWN)); port->pp.state = ((port->pp.state & ~OFPUTIL_PS_LINK_DOWN) @@ -2209,6 +2456,9 @@ ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port) if (port->ofproto->ofproto_class->set_stp_port) { port->ofproto->ofproto_class->set_stp_port(port, NULL); } + if (port->ofproto->ofproto_class->set_rstp_port) { + port->ofproto->ofproto_class->set_rstp_port(port, NULL); + } if (port->ofproto->ofproto_class->set_cfm) { port->ofproto->ofproto_class->set_cfm(port, NULL); } @@ -2508,7 +2758,14 @@ ofproto_rule_destroy__(struct rule *rule) static void rule_destroy_cb(struct rule *rule) + OVS_NO_THREAD_SAFETY_ANALYSIS { + /* Send rule removed if needed. */ + if (rule->flags & OFPUTIL_FF_SEND_FLOW_REM + && rule->removed_reason != OVS_OFPRR_NONE + && !rule_is_hidden(rule)) { + ofproto_rule_send_removed(rule); + } rule->ofproto->ofproto_class->rule_destruct(rule); ofproto_rule_destroy__(rule); } @@ -2544,6 +2801,70 @@ ofproto_rule_unref(struct rule *rule) } } +static void +remove_rule_rcu__(struct rule *rule) + OVS_REQUIRES(ofproto_mutex) +{ + struct ofproto *ofproto = rule->ofproto; + struct oftable *table = &ofproto->tables[rule->table_id]; + + ovs_assert(!cls_rule_visible_in_version(&rule->cr, CLS_MAX_VERSION)); + if (!classifier_remove(&table->cls, &rule->cr)) { + OVS_NOT_REACHED(); + } + ofproto->ofproto_class->rule_delete(rule); + ofproto_rule_unref(rule); +} + +static void +remove_rule_rcu(struct rule *rule) + OVS_EXCLUDED(ofproto_mutex) +{ + ovs_mutex_lock(&ofproto_mutex); + remove_rule_rcu__(rule); + ovs_mutex_unlock(&ofproto_mutex); +} + +/* Removes and deletes rules from a NULL-terminated array of rule pointers. */ +static void +remove_rules_rcu(struct rule **rules) + OVS_EXCLUDED(ofproto_mutex) +{ + struct rule **orig_rules = rules; + + if (*rules) { + struct ofproto *ofproto = rules[0]->ofproto; + unsigned long tables[BITMAP_N_LONGS(256)]; + struct rule *rule; + size_t table_id; + + memset(tables, 0, sizeof tables); + + ovs_mutex_lock(&ofproto_mutex); + while ((rule = *rules++)) { + /* Defer once for each new table. This defers the subtable cleanup + * until later, so that when removing large number of flows the + * operation is faster. */ + if (!bitmap_is_set(tables, rule->table_id)) { + struct classifier *cls = &ofproto->tables[rule->table_id].cls; + + bitmap_set1(tables, rule->table_id); + classifier_defer(cls); + } + remove_rule_rcu__(rule); + } + + BITMAP_FOR_EACH_1(table_id, 256, tables) { + struct classifier *cls = &ofproto->tables[table_id].cls; + + classifier_publish(cls); + } + ovs_mutex_unlock(&ofproto_mutex); + } + + free(orig_rules); +} + void ofproto_group_ref(struct ofgroup *group) { @@ -2636,13 +2957,13 @@ run_rule_executes(struct ofproto *ofproto) OVS_EXCLUDED(ofproto_mutex) { struct rule_execute *e, *next; - struct list executes; + struct ovs_list executes; guarded_list_pop_all(&ofproto->rule_executes, &executes); LIST_FOR_EACH_SAFE (e, next, list_node, &executes) { struct flow flow; - flow_extract(e->packet, NULL, &flow); + flow_extract(e->packet, &flow); flow.in_port.ofp_port = e->in_port; ofproto->ofproto_class->rule_execute(e->rule, &flow, e->packet); @@ -2656,11 +2977,11 @@ static void destroy_rule_executes(struct ofproto *ofproto) { struct rule_execute *e, *next; - struct list executes; + struct ovs_list executes; guarded_list_pop_all(&ofproto->rule_executes, &executes); LIST_FOR_EACH_SAFE (e, next, list_node, &executes) { - ofpbuf_delete(e->packet); + dp_packet_delete(e->packet); rule_execute_destroy(e); } } @@ -2682,7 +3003,7 @@ hash_learned_cookie(ovs_be64 cookie_, uint8_t table_id) static void learned_cookies_update_one__(struct ofproto *ofproto, const struct ofpact_learn *learn, - int delta, struct list *dead_cookies) + int delta, struct ovs_list *dead_cookies) OVS_REQUIRES(ofproto_mutex) { uint32_t hash = hash_learned_cookie(learn->cookie, learn->table_id); @@ -2733,7 +3054,7 @@ next_learn_with_delete(const struct rule_actions *actions, static void learned_cookies_update__(struct ofproto *ofproto, const struct rule_actions *actions, - int delta, struct list *dead_cookies) + int delta, struct ovs_list *dead_cookies) OVS_REQUIRES(ofproto_mutex) { if (actions->has_learn_with_delete) { @@ -2757,33 +3078,31 @@ learned_cookies_inc(struct ofproto *ofproto, static void learned_cookies_dec(struct ofproto *ofproto, const struct rule_actions *actions, - struct list *dead_cookies) + struct ovs_list *dead_cookies) OVS_REQUIRES(ofproto_mutex) { learned_cookies_update__(ofproto, actions, -1, dead_cookies); } static void -learned_cookies_flush(struct ofproto *ofproto, struct list *dead_cookies) +learned_cookies_flush(struct ofproto *ofproto, struct ovs_list *dead_cookies) OVS_REQUIRES(ofproto_mutex) { - struct learned_cookie *c, *next; + struct learned_cookie *c; - LIST_FOR_EACH_SAFE (c, next, u.list_node, dead_cookies) { + LIST_FOR_EACH_POP (c, u.list_node, dead_cookies) { struct rule_criteria criteria; struct rule_collection rules; struct match match; match_init_catchall(&match); - rule_criteria_init(&criteria, c->table_id, &match, 0, + rule_criteria_init(&criteria, c->table_id, &match, 0, CLS_MAX_VERSION, c->cookie, OVS_BE64_MAX, OFPP_ANY, OFPG_ANY); rule_criteria_require_rw(&criteria, false); collect_rules_loose(ofproto, &criteria, &rules); - delete_flows__(&rules, OFPRR_DELETE, NULL); rule_criteria_destroy(&criteria); - rule_collection_destroy(&rules); + delete_flows__(&rules, OFPRR_DELETE, NULL); - list_remove(&c->u.list_node); free(c); } } @@ -2795,33 +3114,132 @@ handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh) return 0; } -static enum ofperr -handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh) +static void +query_tables(struct ofproto *ofproto, + struct ofputil_table_features **featuresp, + struct ofputil_table_stats **statsp) { - struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - struct ofputil_switch_features features; - struct ofport *port; - bool arp_match_ip; - struct ofpbuf *b; + struct mf_bitmap rw_fields = oxm_writable_fields(); + struct mf_bitmap match = oxm_matchable_fields(); + struct mf_bitmap mask = oxm_maskable_fields(); - ofproto->ofproto_class->get_features(ofproto, &arp_match_ip, - &features.actions); - ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */ + struct ofputil_table_features *features; + struct ofputil_table_stats *stats; + int i; - features.datapath_id = ofproto->datapath_id; - features.n_buffers = pktbuf_capacity(); - features.n_tables = ofproto_get_n_visible_tables(ofproto); - features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS | - OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS); - if (arp_match_ip) { - features.capabilities |= OFPUTIL_C_ARP_MATCH_IP; - } - /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */ - features.auxiliary_id = 0; - b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn), - oh->xid); - HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) { - ofputil_put_switch_features_port(&port->pp, b); + features = *featuresp = xcalloc(ofproto->n_tables, sizeof *features); + for (i = 0; i < ofproto->n_tables; i++) { + struct ofputil_table_features *f = &features[i]; + + f->table_id = i; + sprintf(f->name, "table%d", i); + f->metadata_match = OVS_BE64_MAX; + f->metadata_write = OVS_BE64_MAX; + atomic_read_relaxed(&ofproto->tables[i].miss_config, &f->miss_config); + f->max_entries = 1000000; + + bool more_tables = false; + for (int j = i + 1; j < ofproto->n_tables; j++) { + if (!(ofproto->tables[j].flags & OFTABLE_HIDDEN)) { + bitmap_set1(f->nonmiss.next, j); + more_tables = true; + } + } + f->nonmiss.instructions = (1u << N_OVS_INSTRUCTIONS) - 1; + if (!more_tables) { + f->nonmiss.instructions &= ~(1u << OVSINST_OFPIT11_GOTO_TABLE); + } + f->nonmiss.write.ofpacts = (UINT64_C(1) << N_OFPACTS) - 1; + f->nonmiss.write.set_fields = rw_fields; + f->nonmiss.apply = f->nonmiss.write; + f->miss = f->nonmiss; + + f->match = match; + f->mask = mask; + f->wildcard = match; + } + + if (statsp) { + stats = *statsp = xcalloc(ofproto->n_tables, sizeof *stats); + for (i = 0; i < ofproto->n_tables; i++) { + struct ofputil_table_stats *s = &stats[i]; + + s->table_id = i; + s->active_count = ofproto->tables[i].n_flows; + if (i == 0) { + s->active_count -= connmgr_count_hidden_rules( + ofproto->connmgr); + } + } + } else { + stats = NULL; + } + + ofproto->ofproto_class->query_tables(ofproto, features, stats); + + for (i = 0; i < ofproto->n_tables; i++) { + const struct oftable *table = &ofproto->tables[i]; + struct ofputil_table_features *f = &features[i]; + + if (table->name) { + ovs_strzcpy(f->name, table->name, sizeof f->name); + } + + if (table->max_flows < f->max_entries) { + f->max_entries = table->max_flows; + } + } +} + +static void +query_switch_features(struct ofproto *ofproto, + bool *arp_match_ip, uint64_t *ofpacts) +{ + struct ofputil_table_features *features, *f; + + *arp_match_ip = false; + *ofpacts = 0; + + query_tables(ofproto, &features, NULL); + for (f = features; f < &features[ofproto->n_tables]; f++) { + *ofpacts |= f->nonmiss.apply.ofpacts | f->miss.apply.ofpacts; + if (bitmap_is_set(f->match.bm, MFF_ARP_SPA) || + bitmap_is_set(f->match.bm, MFF_ARP_TPA)) { + *arp_match_ip = true; + } + } + free(features); + + /* Sanity check. */ + ovs_assert(*ofpacts & (UINT64_C(1) << OFPACT_OUTPUT)); +} + +static enum ofperr +handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh) +{ + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); + struct ofputil_switch_features features; + struct ofport *port; + bool arp_match_ip; + struct ofpbuf *b; + + query_switch_features(ofproto, &arp_match_ip, &features.ofpacts); + + features.datapath_id = ofproto->datapath_id; + features.n_buffers = pktbuf_capacity(); + features.n_tables = ofproto_get_n_visible_tables(ofproto); + features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS | + OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS | + OFPUTIL_C_GROUP_STATS); + if (arp_match_ip) { + features.capabilities |= OFPUTIL_C_ARP_MATCH_IP; + } + /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */ + features.auxiliary_id = 0; + b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn), + oh->xid); + HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) { + ofputil_put_switch_features_port(&port->pp, b); } ofconn_send_reply(ofconn, b); @@ -2895,7 +3313,7 @@ reject_slave_controller(struct ofconn *ofconn) { if (ofconn_get_type(ofconn) == OFCONN_PRIMARY && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) { - return OFPERR_OFPBRC_EPERM; + return OFPERR_OFPBRC_IS_SLAVE; } else { return 0; } @@ -2909,7 +3327,7 @@ reject_slave_controller(struct ofconn *ofconn) * - If they use any groups, then 'ofproto' has that group configured. * * Returns 0 if successful, otherwise an OpenFlow error. */ -static enum ofperr +enum ofperr ofproto_check_ofpacts(struct ofproto *ofproto, const struct ofpact ofpacts[], size_t ofpacts_len) { @@ -2921,7 +3339,7 @@ ofproto_check_ofpacts(struct ofproto *ofproto, return OFPERR_OFPMMFC_INVALID_METER; } - OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) { + OFPACT_FOR_EACH_FLATTENED (a, ofpacts, ofpacts_len) { if (a->type == OFPACT_GROUP && !ofproto_group_exists(ofproto, ofpact_get_GROUP(a)->group_id)) { return OFPERR_OFPBAC_BAD_OUT_GROUP; @@ -2936,7 +3354,7 @@ handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh) { struct ofproto *p = ofconn_get_ofproto(ofconn); struct ofputil_packet_out po; - struct ofpbuf *payload; + struct dp_packet *payload; uint64_t ofpacts_stub[1024 / 8]; struct ofpbuf ofpacts; struct flow flow; @@ -2969,18 +3387,30 @@ handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh) } } else { /* Ensure that the L3 header is 32-bit aligned. */ - payload = ofpbuf_clone_data_with_headroom(po.packet, po.packet_len, 2); + payload = dp_packet_clone_data_with_headroom(po.packet, po.packet_len, 2); } /* Verify actions against packet, then send packet if successful. */ - flow_extract(payload, NULL, &flow); + flow_extract(payload, &flow); flow.in_port.ofp_port = po.in_port; - error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len); + + /* Check actions like for flow mods. We pass a 'table_id' of 0 to + * ofproto_check_consistency(), which isn't strictly correct because these + * actions aren't in any table. This is OK as 'table_id' is only used to + * check instructions (e.g., goto-table), which can't appear on the action + * list of a packet-out. */ + error = ofpacts_check_consistency(po.ofpacts, po.ofpacts_len, + &flow, u16_to_ofp(p->max_ports), + 0, p->n_tables, + ofconn_get_protocol(ofconn)); if (!error) { - error = p->ofproto_class->packet_out(p, payload, &flow, - po.ofpacts, po.ofpacts_len); + error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len); + if (!error) { + error = p->ofproto_class->packet_out(p, payload, &flow, + po.ofpacts, po.ofpacts_len); + } } - ofpbuf_delete(payload); + dp_packet_delete(payload); exit_free_ofpacts: ofpbuf_uninit(&ofpacts); @@ -3014,9 +3444,34 @@ update_port_config(struct ofconn *ofconn, struct ofport *port, } static enum ofperr -handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh) +port_mod_start(struct ofconn *ofconn, struct ofputil_port_mod *pm, + struct ofport **port) { struct ofproto *p = ofconn_get_ofproto(ofconn); + + *port = ofproto_get_port(p, pm->port_no); + if (!*port) { + return OFPERR_OFPPMFC_BAD_PORT; + } + if (!eth_addr_equals((*port)->pp.hw_addr, pm->hw_addr)) { + return OFPERR_OFPPMFC_BAD_HW_ADDR; + } + return 0; +} + +static void +port_mod_finish(struct ofconn *ofconn, struct ofputil_port_mod *pm, + struct ofport *port) +{ + update_port_config(ofconn, port, pm->config, pm->mask); + if (pm->advertise) { + netdev_set_advertisements(port->netdev, pm->advertise); + } +} + +static enum ofperr +handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh) +{ struct ofputil_port_mod pm; struct ofport *port; enum ofperr error; @@ -3031,18 +3486,11 @@ handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh) return error; } - port = ofproto_get_port(p, pm.port_no); - if (!port) { - return OFPERR_OFPPMFC_BAD_PORT; - } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) { - return OFPERR_OFPPMFC_BAD_HW_ADDR; - } else { - update_port_config(ofconn, port, pm.config, pm.mask); - if (pm.advertise) { - netdev_set_advertisements(port->netdev, pm.advertise); - } + error = port_mod_start(ofconn, &pm, &port); + if (!error) { + port_mod_finish(ofconn, &pm, port); } - return 0; + return error; } static enum ofperr @@ -3081,67 +3529,118 @@ static enum ofperr handle_table_stats_request(struct ofconn *ofconn, const struct ofp_header *request) { - struct ofproto *p = ofconn_get_ofproto(ofconn); - struct ofp12_table_stats *ots; - struct ofpbuf *msg; - int n_tables; + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); + struct ofputil_table_features *features; + struct ofputil_table_stats *stats; + struct ofpbuf *reply; size_t i; - /* Set up default values. - * - * ofp12_table_stats is used as a generic structure as - * it is able to hold all the fields for ofp10_table_stats - * and ofp11_table_stats (and of course itself). - */ - ots = xcalloc(p->n_tables, sizeof *ots); - for (i = 0; i < p->n_tables; i++) { - ots[i].table_id = i; - sprintf(ots[i].name, "table%"PRIuSIZE, i); - ots[i].match = htonll(OFPXMT13_MASK); - ots[i].wildcards = htonll(OFPXMT13_MASK); - ots[i].write_actions = htonl(OFPAT11_OUTPUT); - ots[i].apply_actions = htonl(OFPAT11_OUTPUT); - ots[i].write_setfields = htonll(OFPXMT13_MASK); - ots[i].apply_setfields = htonll(OFPXMT13_MASK); - ots[i].metadata_match = OVS_BE64_MAX; - ots[i].metadata_write = OVS_BE64_MAX; - ots[i].instructions = htonl(OFPIT11_ALL); - ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK); - ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */ - ots[i].active_count = htonl(classifier_count(&p->tables[i].cls)); - } - - p->ofproto_class->get_tables(p, ots); - - /* Post-process the tables, dropping hidden tables. */ - n_tables = p->n_tables; - for (i = 0; i < p->n_tables; i++) { - const struct oftable *table = &p->tables[i]; + query_tables(ofproto, &features, &stats); - if (table->flags & OFTABLE_HIDDEN) { - n_tables = i; - break; + reply = ofputil_encode_table_stats_reply(request); + for (i = 0; i < ofproto->n_tables; i++) { + if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) { + ofputil_append_table_stats_reply(reply, &stats[i], &features[i]); } + } + ofconn_send_reply(ofconn, reply); - if (table->name) { - ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name); - } + free(features); + free(stats); + + return 0; +} + +static enum ofperr +handle_table_features_request(struct ofconn *ofconn, + const struct ofp_header *request) +{ + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); + struct ofputil_table_features *features; + struct ovs_list replies; + struct ofpbuf msg; + size_t i; + + ofpbuf_use_const(&msg, request, ntohs(request->length)); + ofpraw_pull_assert(&msg); + if (msg.size || ofpmp_more(request)) { + return OFPERR_OFPTFFC_EPERM; + } + + query_tables(ofproto, &features, NULL); - if (table->max_flows < ntohl(ots[i].max_entries)) { - ots[i].max_entries = htonl(table->max_flows); + ofpmp_init(&replies, request); + for (i = 0; i < ofproto->n_tables; i++) { + if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) { + ofputil_append_table_features_reply(&features[i], &replies); } } + ofconn_send_replies(ofconn, &replies); - msg = ofputil_encode_table_stats_reply(ots, n_tables, request); - ofconn_send_reply(ofconn, msg); + free(features); + + return 0; +} + +static void +query_table_desc__(struct ofputil_table_desc *td, + struct ofproto *ofproto, uint8_t table_id) +{ + unsigned int count = ofproto->tables[table_id].n_flows; + unsigned int max_flows = ofproto->tables[table_id].max_flows; + + td->table_id = table_id; + td->eviction = (ofproto->tables[table_id].eviction & EVICTION_OPENFLOW + ? OFPUTIL_TABLE_EVICTION_ON + : OFPUTIL_TABLE_EVICTION_OFF); + td->eviction_flags = OFPROTO_EVICTION_FLAGS; + td->vacancy = (ofproto->tables[table_id].vacancy_enabled + ? OFPUTIL_TABLE_VACANCY_ON + : OFPUTIL_TABLE_VACANCY_OFF); + td->table_vacancy.vacancy_down = ofproto->tables[table_id].vacancy_down; + td->table_vacancy.vacancy_up = ofproto->tables[table_id].vacancy_up; + td->table_vacancy.vacancy = max_flows ? (count * 100) / max_flows : 0; +} + +/* This function queries the database for dumping table-desc. */ +static void +query_tables_desc(struct ofproto *ofproto, struct ofputil_table_desc **descp) +{ + struct ofputil_table_desc *table_desc; + size_t i; + + table_desc = *descp = xcalloc(ofproto->n_tables, sizeof *table_desc); + for (i = 0; i < ofproto->n_tables; i++) { + struct ofputil_table_desc *td = &table_desc[i]; + query_table_desc__(td, ofproto, i); + } +} - free(ots); +/* Function to handle dump-table-desc request. */ +static enum ofperr +handle_table_desc_request(struct ofconn *ofconn, + const struct ofp_header *request) +{ + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); + struct ofputil_table_desc *table_desc; + struct ovs_list replies; + size_t i; + query_tables_desc(ofproto, &table_desc); + ofpmp_init(&replies, request); + for (i = 0; i < ofproto->n_tables; i++) { + if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) { + ofputil_append_table_desc_reply(&table_desc[i], &replies, + request->version); + } + } + ofconn_send_replies(ofconn, &replies); + free(table_desc); return 0; } static void -append_port_stat(struct ofport *port, struct list *replies) +append_port_stat(struct ofport *port, struct ovs_list *replies) { struct ofputil_port_stats ops = { .port_no = port->pp.port_no }; @@ -3159,11 +3658,11 @@ append_port_stat(struct ofport *port, struct list *replies) static void handle_port_request(struct ofconn *ofconn, const struct ofp_header *request, ofp_port_t port_no, - void (*cb)(struct ofport *, struct list *replies)) + void (*cb)(struct ofport *, struct ovs_list *replies)) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); struct ofport *port; - struct list replies; + struct ovs_list replies; ofpmp_init(&replies, request); if (port_no != OFPP_ANY) { @@ -3195,7 +3694,7 @@ handle_port_stats_request(struct ofconn *ofconn, } static void -append_port_desc(struct ofport *port, struct list *replies) +append_port_desc(struct ofport *port, struct ovs_list *replies) { ofputil_append_port_desc_stats_reply(&port->pp, replies); } @@ -3320,12 +3819,14 @@ next_matching_table(const struct ofproto *ofproto, * supplied as 0. */ static void rule_criteria_init(struct rule_criteria *criteria, uint8_t table_id, - const struct match *match, unsigned int priority, - ovs_be64 cookie, ovs_be64 cookie_mask, - ofp_port_t out_port, uint32_t out_group) + const struct match *match, int priority, + cls_version_t version, ovs_be64 cookie, + ovs_be64 cookie_mask, ofp_port_t out_port, + uint32_t out_group) { criteria->table_id = table_id; cls_rule_init(&criteria->cr, match, priority); + criteria->version = version; criteria->cookie = cookie; criteria->cookie_mask = cookie_mask; criteria->out_port = out_port; @@ -3414,18 +3915,56 @@ rule_collection_unref(struct rule_collection *rules) } } +/* Returns a NULL-terminated array of rule pointers, + * destroys 'rules'. */ +static struct rule ** +rule_collection_detach(struct rule_collection *rules) +{ + struct rule **rule_array; + + rule_collection_add(rules, NULL); + + if (rules->rules == rules->stub) { + rules->rules = xmemdup(rules->rules, rules->n * sizeof *rules->rules); + } + + rule_array = rules->rules; + rule_collection_init(rules); + + return rule_array; +} + void rule_collection_destroy(struct rule_collection *rules) { if (rules->rules != rules->stub) { free(rules->rules); } + + /* Make repeated destruction harmless. */ + rule_collection_init(rules); +} + +/* Schedules postponed removal of rules, destroys 'rules'. */ +static void +rule_collection_remove_postponed(struct rule_collection *rules) + OVS_REQUIRES(ofproto_mutex) +{ + if (rules->n > 0) { + if (rules->n == 1) { + ovsrcu_postpone(remove_rule_rcu, rules->rules[0]); + } else { + ovsrcu_postpone(remove_rules_rcu, rule_collection_detach(rules)); + } + } } /* Checks whether 'rule' matches 'c' and, if so, adds it to 'rules'. This * function verifies most of the criteria in 'c' itself, but the caller must * check 'c->cr' itself. * + * Rules that have already been marked for removal are not collected. + * * Increments '*n_readonly' if 'rule' wasn't added because it's read-only (and * 'c' only includes modifiable rules). */ static void @@ -3437,7 +3976,8 @@ collect_rule(struct rule *rule, const struct rule_criteria *c, && ofproto_rule_has_out_port(rule, c->out_port) && ofproto_rule_has_out_group(rule, c->out_group) && !((rule->flow_cookie ^ c->cookie) & c->cookie_mask) - && (!rule_is_hidden(rule) || c->include_hidden)) { + && (!rule_is_hidden(rule) || c->include_hidden) + && cls_rule_visible_in_version(&rule->cr, c->version)) { /* Rule matches all the criteria... */ if (!rule_is_readonly(rule) || c->include_readonly) { /* ...add it. */ @@ -3486,7 +4026,8 @@ collect_rules_loose(struct ofproto *ofproto, FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) { struct rule *rule; - CLS_FOR_EACH_TARGET (rule, cr, &table->cls, &criteria->cr) { + CLS_FOR_EACH_TARGET (rule, cr, &table->cls, &criteria->cr, + criteria->version) { collect_rule(rule, criteria, rules, &n_readonly); } } @@ -3518,7 +4059,7 @@ collect_rules_strict(struct ofproto *ofproto, { struct oftable *table; size_t n_readonly = 0; - int error = 0; + enum ofperr error = 0; rule_collection_init(rules); @@ -3542,7 +4083,8 @@ collect_rules_strict(struct ofproto *ofproto, struct rule *rule; rule = rule_from_cls_rule(classifier_find_rule_exactly( - &table->cls, &criteria->cr)); + &table->cls, &criteria->cr, + criteria->version)); if (rule) { collect_rule(rule, criteria, rules, &n_readonly); } @@ -3580,7 +4122,7 @@ handle_flow_stats_request(struct ofconn *ofconn, struct ofputil_flow_stats_request fsr; struct rule_criteria criteria; struct rule_collection rules; - struct list replies; + struct ovs_list replies; enum ofperr error; size_t i; @@ -3589,8 +4131,9 @@ handle_flow_stats_request(struct ofconn *ofconn, return error; } - rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, fsr.cookie, - fsr.cookie_mask, fsr.out_port, fsr.out_group); + rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, CLS_MAX_VERSION, + fsr.cookie, fsr.cookie_mask, fsr.out_port, + fsr.out_group); ovs_mutex_lock(&ofproto_mutex); error = collect_rules_loose(ofproto, &criteria, &rules); @@ -3617,6 +4160,7 @@ handle_flow_stats_request(struct ofconn *ofconn, fs.cookie = rule->flow_cookie; fs.idle_timeout = rule->idle_timeout; fs.hard_timeout = rule->hard_timeout; + fs.importance = rule->importance; created = rule->created; modified = rule->modified; actions = rule_get_actions(rule); @@ -3752,7 +4296,7 @@ handle_aggregate_stats_request(struct ofconn *ofconn, } rule_criteria_init(&criteria, request.table_id, &request.match, 0, - request.cookie, request.cookie_mask, + CLS_MAX_VERSION, request.cookie, request.cookie_mask, request.out_port, request.out_group); ovs_mutex_lock(&ofproto_mutex); @@ -3810,7 +4354,7 @@ handle_aggregate_stats_request(struct ofconn *ofconn, struct queue_stats_cbdata { struct ofport *ofport; - struct list replies; + struct ovs_list replies; long long int now; }; @@ -3906,54 +4450,96 @@ handle_queue_stats_request(struct ofconn *ofconn, return error; } -static bool -should_evict_a_rule(struct oftable *table, unsigned int extra_space) - OVS_REQUIRES(ofproto_mutex) - OVS_NO_THREAD_SAFETY_ANALYSIS -{ - return classifier_count(&table->cls) + extra_space > table->max_flows; -} - static enum ofperr -evict_rules_from_table(struct oftable *table, unsigned int extra_space) +evict_rules_from_table(struct oftable *table) OVS_REQUIRES(ofproto_mutex) { - while (should_evict_a_rule(table, extra_space)) { + enum ofperr error = 0; + struct rule_collection rules; + unsigned int count = table->n_flows; + unsigned int max_flows = table->max_flows; + + rule_collection_init(&rules); + + while (count-- > max_flows) { struct rule *rule; if (!choose_rule_to_evict(table, &rule)) { - return OFPERR_OFPFMFC_TABLE_FULL; + error = OFPERR_OFPFMFC_TABLE_FULL; + break; } else { - ofproto_rule_delete__(rule, OFPRR_EVICTION); + eviction_group_remove_rule(rule); + rule_collection_add(&rules, rule); } } + delete_flows__(&rules, OFPRR_EVICTION, NULL); - return 0; + return error; +} + +static void +get_conjunctions(const struct ofputil_flow_mod *fm, + struct cls_conjunction **conjsp, size_t *n_conjsp) + OVS_REQUIRES(ofproto_mutex) +{ + struct cls_conjunction *conjs = NULL; + int n_conjs = 0; + + const struct ofpact *ofpact; + OFPACT_FOR_EACH (ofpact, fm->ofpacts, fm->ofpacts_len) { + if (ofpact->type == OFPACT_CONJUNCTION) { + n_conjs++; + } else if (ofpact->type != OFPACT_NOTE) { + /* "conjunction" may appear with "note" actions but not with any + * other type of actions. */ + ovs_assert(!n_conjs); + break; + } + } + if (n_conjs) { + int i = 0; + + conjs = xzalloc(n_conjs * sizeof *conjs); + OFPACT_FOR_EACH (ofpact, fm->ofpacts, fm->ofpacts_len) { + if (ofpact->type == OFPACT_CONJUNCTION) { + struct ofpact_conjunction *oc = ofpact_get_CONJUNCTION(ofpact); + conjs[i].clause = oc->clause; + conjs[i].n_clauses = oc->n_clauses; + conjs[i].id = oc->id; + i++; + } + } + } + + *conjsp = conjs; + *n_conjsp = n_conjs; } /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT * 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. + * Adds the flow specified by 'fm', to the ofproto's flow table. Returns 0 on + * success, or an OpenFlow error code on failure. * - * The caller retains ownership of 'fm->ofpacts'. + * On successful return the caller must complete the operation either by + * calling add_flow_finish(), or add_flow_revert() if the operation needs to + * be reverted. * - * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, - * if any. */ + * The caller retains ownership of 'fm->ofpacts'. */ static enum ofperr -add_flow(struct ofproto *ofproto, struct ofputil_flow_mod *fm, - const struct flow_mod_requester *req) +add_flow_start(struct ofproto *ofproto, struct ofproto_flow_mod *ofm) OVS_REQUIRES(ofproto_mutex) { - const struct rule_actions *actions; + struct ofputil_flow_mod *fm = &ofm->fm; + struct rule **old_rule = &ofm->old_rules.stub[0]; + struct rule **new_rule = &ofm->new_rules.stub[0]; struct oftable *table; struct cls_rule cr; struct rule *rule; uint8_t table_id; - int error = 0; + struct cls_conjunction *conjs; + size_t n_conjs; + enum ofperr error; if (!check_table_id(ofproto, fm->table_id)) { error = OFPERR_OFPBRC_BAD_TABLE_ID; @@ -3985,59 +4571,138 @@ add_flow(struct ofproto *ofproto, struct ofputil_flow_mod *fm, return OFPERR_OFPBRC_EPERM; } - if (!(fm->flags & OFPUTIL_FF_HIDDEN_FIELDS)) { - if (!match_has_default_hidden_fields(&fm->match)) { - VLOG_WARN_RL(&rl, "%s: (add_flow) only internal flows can set " - "non-default values to hidden fields", ofproto->name); - return OFPERR_OFPBRC_EPERM; - } + if (!(fm->flags & OFPUTIL_FF_HIDDEN_FIELDS) + && !match_has_default_hidden_fields(&fm->match)) { + VLOG_WARN_RL(&rl, "%s: (add_flow) only internal flows can set " + "non-default values to hidden fields", ofproto->name); + return OFPERR_OFPBRC_EPERM; } cls_rule_init(&cr, &fm->match, fm->priority); - /* Transform "add" into "modify" if there's an existing identical flow. */ - rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr)); - if (rule) { - struct rule_collection rules; - - cls_rule_destroy(&cr); + /* Check for the existence of an identical rule. + * This will not return rules earlier marked for removal. */ + rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr, + ofm->version)); + *old_rule = rule; + if (!rule) { + /* Check for overlap, if requested. */ + if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP + && classifier_rule_overlaps(&table->cls, &cr, ofm->version)) { + cls_rule_destroy(&cr); + return OFPERR_OFPFMFC_OVERLAP; + } - rule_collection_init(&rules); - rule_collection_add(&rules, rule); + /* If necessary, evict an existing rule to clear out space. */ + if (table->n_flows >= table->max_flows) { + if (!choose_rule_to_evict(table, &rule)) { + error = OFPERR_OFPFMFC_TABLE_FULL; + cls_rule_destroy(&cr); + return error; + } + eviction_group_remove_rule(rule); + /* Marks '*old_rule' as an evicted rule rather than replaced rule. + */ + fm->delete_reason = OFPRR_EVICTION; + *old_rule = rule; + } + } else { fm->modify_cookie = true; - error = modify_flows__(ofproto, fm, &rules, req); - rule_collection_destroy(&rules); + } + /* Allocate new rule. */ + error = replace_rule_create(ofproto, fm, &cr, table - ofproto->tables, + rule, new_rule); + if (error) { return error; } - /* Check for overlap, if requested. */ - if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP) { - if (classifier_rule_overlaps(&table->cls, &cr)) { - cls_rule_destroy(&cr); - return OFPERR_OFPFMFC_OVERLAP; - } + get_conjunctions(fm, &conjs, &n_conjs); + replace_rule_start(ofproto, ofm->version, rule, *new_rule, conjs, n_conjs); + free(conjs); + + return 0; +} + +/* Revert the effects of add_flow_start(). */ +static void +add_flow_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm) + OVS_REQUIRES(ofproto_mutex) +{ + struct ofputil_flow_mod *fm = &ofm->fm; + struct rule *old_rule = ofm->old_rules.stub[0]; + struct rule *new_rule = ofm->new_rules.stub[0]; + + if (old_rule && fm->delete_reason == OFPRR_EVICTION) { + /* Revert the eviction. */ + eviction_group_add_rule(old_rule); } - /* If necessary, evict an existing rule to clear out space. */ - error = evict_rules_from_table(table, 1); - if (error) { - cls_rule_destroy(&cr); - return error; + replace_rule_revert(ofproto, old_rule, new_rule); +} + +/* To be called after version bump. */ +static void +add_flow_finish(struct ofproto *ofproto, struct ofproto_flow_mod *ofm, + const struct flow_mod_requester *req) + OVS_REQUIRES(ofproto_mutex) +{ + struct ofputil_flow_mod *fm = &ofm->fm; + struct rule *old_rule = ofm->old_rules.stub[0]; + struct rule *new_rule = ofm->new_rules.stub[0]; + struct ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies); + + replace_rule_finish(ofproto, fm, req, old_rule, new_rule, &dead_cookies); + learned_cookies_flush(ofproto, &dead_cookies); + + if (old_rule) { + ovsrcu_postpone(remove_rule_rcu, old_rule); + } else { + if (minimask_get_vid_mask(new_rule->cr.match.mask) == VLAN_VID_MASK) { + if (ofproto->vlan_bitmap) { + uint16_t vid = miniflow_get_vid(new_rule->cr.match.flow); + + if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) { + bitmap_set1(ofproto->vlan_bitmap, vid); + ofproto->vlans_changed = true; + } + } else { + ofproto->vlans_changed = true; + } + } + + ofmonitor_report(ofproto->connmgr, new_rule, NXFME_ADDED, 0, + req ? req->ofconn : NULL, + req ? req->request->xid : 0, NULL); } + send_buffered_packet(req, fm->buffer_id, new_rule); +} + +/* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */ + +/* Create a new rule based on attributes in 'fm', match in 'cr', 'table_id', + * and 'old_rule'. Note that the rule is NOT inserted into a any data + * structures yet. Takes ownership of 'cr'. */ +static enum ofperr +replace_rule_create(struct ofproto *ofproto, struct ofputil_flow_mod *fm, + struct cls_rule *cr, uint8_t table_id, + struct rule *old_rule, struct rule **new_rule) +{ + struct rule *rule; + enum ofperr error; + /* 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; + cls_rule_destroy(cr); + VLOG_WARN_RL(&rl, "%s: failed to allocate a rule.", ofproto->name); + return OFPERR_OFPFMFC_UNKNOWN; } /* Initialize base state. */ *CONST_CAST(struct ofproto **, &rule->ofproto) = ofproto; - cls_rule_move(CONST_CAST(struct cls_rule *, &rule->cr), &cr); + cls_rule_move(CONST_CAST(struct cls_rule *, &rule->cr), cr); ovs_refcount_init(&rule->ref_count); rule->flow_cookie = fm->new_cookie; rule->created = rule->modified = time_msec(); @@ -4046,12 +4711,13 @@ add_flow(struct ofproto *ofproto, struct ofputil_flow_mod *fm, ovs_mutex_lock(&rule->mutex); rule->idle_timeout = fm->idle_timeout; rule->hard_timeout = fm->hard_timeout; - ovs_mutex_unlock(&rule->mutex); + *CONST_CAST(uint16_t *, &rule->importance) = fm->importance; + rule->removed_reason = OVS_OFPRR_NONE; - *CONST_CAST(uint8_t *, &rule->table_id) = table - ofproto->tables; + *CONST_CAST(uint8_t *, &rule->table_id) = table_id; rule->flags = fm->flags & OFPUTIL_FF_STATE; - actions = rule_actions_create(fm->ofpacts, fm->ofpacts_len); - ovsrcu_set(&rule->actions, actions); + *CONST_CAST(const struct rule_actions **, &rule->actions) + = rule_actions_create(fm->ofpacts, fm->ofpacts_len); list_init(&rule->meter_list_node); rule->eviction_group = NULL; list_init(&rule->expirable); @@ -4059,6 +4725,27 @@ add_flow(struct ofproto *ofproto, struct ofputil_flow_mod *fm, rule->add_seqno = 0; rule->modify_seqno = 0; + /* Copy values from old rule for modify semantics. */ + if (old_rule && fm->delete_reason != OFPRR_EVICTION) { + bool change_cookie = (fm->modify_cookie + && fm->new_cookie != OVS_BE64_MAX + && fm->new_cookie != old_rule->flow_cookie); + + ovs_mutex_lock(&old_rule->mutex); + if (fm->command != OFPFC_ADD) { + rule->idle_timeout = old_rule->idle_timeout; + rule->hard_timeout = old_rule->hard_timeout; + *CONST_CAST(uint16_t *, &rule->importance) = old_rule->importance; + rule->flags = old_rule->flags; + rule->created = old_rule->created; + } + if (!change_cookie) { + rule->flow_cookie = old_rule->flow_cookie; + } + ovs_mutex_unlock(&old_rule->mutex); + } + ovs_mutex_unlock(&rule->mutex); + /* Construct rule, initializing derived state. */ error = ofproto->ofproto_class->rule_construct(rule); if (error) { @@ -4066,165 +4753,175 @@ add_flow(struct ofproto *ofproto, struct ofputil_flow_mod *fm, return error; } - if (fm->hard_timeout || fm->idle_timeout) { - list_insert(&ofproto->expirable, &rule->expirable); - } - cookies_insert(ofproto, rule); - eviction_group_add_rule(rule); - if (actions->has_meter) { - meter_insert_rule(rule); - } + rule->removed = true; /* Not yet in ofproto data structures. */ + + *new_rule = rule; + return 0; +} - classifier_insert(&table->cls, CONST_CAST(struct cls_rule *, &rule->cr)); +static void +replace_rule_start(struct ofproto *ofproto, cls_version_t version, + struct rule *old_rule, struct rule *new_rule, + struct cls_conjunction *conjs, size_t n_conjs) +{ + struct oftable *table = &ofproto->tables[new_rule->table_id]; - error = ofproto->ofproto_class->rule_insert(rule); - if (error) { - oftable_remove_rule(rule); - ofproto_rule_unref(rule); - return error; + /* 'old_rule' may be either an evicted rule or replaced rule. */ + if (old_rule) { + /* Mark the old rule for removal in the next version. */ + cls_rule_make_invisible_in_version(&old_rule->cr, version); + } else { + table->n_flows++; } - learned_cookies_inc(ofproto, actions); + /* Insert flow to the classifier, so that later flow_mods may relate + * to it. This is reversible, in case later errors require this to + * be reverted. */ + ofproto_rule_insert__(ofproto, new_rule); + /* Make the new rule visible for classifier lookups only from the next + * version. */ + classifier_insert(&table->cls, &new_rule->cr, version, conjs, n_conjs); +} - if (minimask_get_vid_mask(&rule->cr.match.mask) == VLAN_VID_MASK) { - if (ofproto->vlan_bitmap) { - uint16_t vid = miniflow_get_vid(&rule->cr.match.flow); - if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) { - bitmap_set1(ofproto->vlan_bitmap, vid); - ofproto->vlans_changed = true; - } - } else { - ofproto->vlans_changed = true; - } - } +static void replace_rule_revert(struct ofproto *ofproto, + struct rule *old_rule, struct rule *new_rule) +{ + struct oftable *table = &ofproto->tables[new_rule->table_id]; - ofmonitor_report(ofproto->connmgr, rule, NXFME_ADDED, 0, - req ? req->ofconn : NULL, req ? req->xid : 0, NULL); + if (old_rule) { + /* Restore the original visibility of the old rule. */ + cls_rule_restore_visibility(&old_rule->cr); + } else { + /* Restore table's rule count. */ + table->n_flows--; + } - return req ? send_buffered_packet(req->ofconn, fm->buffer_id, rule) : 0; + /* Remove the new rule immediately. It was never visible to lookups. */ + if (!classifier_remove(&table->cls, &new_rule->cr)) { + OVS_NOT_REACHED(); + } + ofproto_rule_remove__(ofproto, new_rule); + /* The rule was not inserted to the ofproto provider, so we can + * release it without deleting it from the ofproto provider. */ + ofproto_rule_unref(new_rule); } - -/* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */ -/* Modifies the rules listed in 'rules', changing their actions to match those - * in 'fm'. - * - * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id, - * if any. - * - * Returns 0 on success, otherwise an OpenFlow error code. */ -static enum ofperr -modify_flows__(struct ofproto *ofproto, struct ofputil_flow_mod *fm, - const struct rule_collection *rules, - const struct flow_mod_requester *req) +/* Adds the 'new_rule', replacing the 'old_rule'. */ +static void +replace_rule_finish(struct ofproto *ofproto, struct ofputil_flow_mod *fm, + const struct flow_mod_requester *req, + struct rule *old_rule, struct rule *new_rule, + struct ovs_list *dead_cookies) OVS_REQUIRES(ofproto_mutex) { - struct list dead_cookies = LIST_INITIALIZER(&dead_cookies); - enum nx_flow_update_event event; - size_t i; + bool forward_stats = !(fm->flags & OFPUTIL_FF_RESET_COUNTS); + struct rule *replaced_rule; + + replaced_rule = fm->delete_reason != OFPRR_EVICTION ? old_rule : NULL; + + /* Insert the new flow to the ofproto provider. A non-NULL 'replaced_rule' + * is a duplicate rule the 'new_rule' is replacing. The provider should + * link the stats from the old rule to the new one if 'forward_stats' is + * 'true'. The 'replaced_rule' will be deleted right after this call. */ + ofproto->ofproto_class->rule_insert(new_rule, replaced_rule, + forward_stats); + learned_cookies_inc(ofproto, rule_get_actions(new_rule)); + + if (old_rule) { + const struct rule_actions *old_actions = rule_get_actions(old_rule); + + /* Remove the old rule from data structures. Removal from the + * classifier and the deletion of the rule is RCU postponed by the + * caller. */ + ofproto_rule_remove__(ofproto, old_rule); + learned_cookies_dec(ofproto, old_actions, dead_cookies); + + if (replaced_rule) { + enum nx_flow_update_event event = fm->command == OFPFC_ADD + ? NXFME_ADDED : NXFME_MODIFIED; + + bool change_cookie = (fm->modify_cookie + && fm->new_cookie != OVS_BE64_MAX + && fm->new_cookie != old_rule->flow_cookie); + + bool change_actions = !ofpacts_equal(fm->ofpacts, + fm->ofpacts_len, + old_actions->ofpacts, + old_actions->ofpacts_len); + + if (event != NXFME_MODIFIED || change_actions || change_cookie) { + ofmonitor_report(ofproto->connmgr, new_rule, event, 0, + req ? req->ofconn : NULL, + req ? req->request->xid : 0, + change_actions ? old_actions : NULL); + } + } else { + /* XXX: This is slight duplication with delete_flows_finish__() */ - if (ofproto->ofproto_class->rule_premodify_actions) { - for (i = 0; i < rules->n; i++) { - struct rule *rule = rules->rules[i]; - enum ofperr error; + old_rule->removed_reason = OFPRR_EVICTION; - error = ofproto->ofproto_class->rule_premodify_actions( - rule, fm->ofpacts, fm->ofpacts_len); - if (error) { - return error; - } + ofmonitor_report(ofproto->connmgr, old_rule, NXFME_DELETED, + OFPRR_EVICTION, + req ? req->ofconn : NULL, + req ? req->request->xid : 0, NULL); } } +} - event = fm->command == OFPFC_ADD ? NXFME_ADDED : NXFME_MODIFIED; - for (i = 0; i < rules->n; i++) { - struct rule *rule = rules->rules[i]; - - /* 'fm' says that */ - bool change_cookie = (fm->modify_cookie - && fm->new_cookie != OVS_BE64_MAX - && fm->new_cookie != rule->flow_cookie); - - const struct rule_actions *actions = rule_get_actions(rule); - bool change_actions = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len, - actions->ofpacts, - actions->ofpacts_len); - - bool reset_counters = (fm->flags & OFPUTIL_FF_RESET_COUNTS) != 0; - - long long int now = time_msec(); +static enum ofperr +modify_flows_start__(struct ofproto *ofproto, struct ofproto_flow_mod *ofm) + OVS_REQUIRES(ofproto_mutex) +{ + struct ofputil_flow_mod *fm = &ofm->fm; + struct rule_collection *old_rules = &ofm->old_rules; + struct rule_collection *new_rules = &ofm->new_rules; + enum ofperr error; - /* FIXME: Implement OFPFUTIL_FF_RESET_COUNTS */ + rule_collection_init(new_rules); - if (change_cookie) { - cookies_remove(ofproto, rule); - } + if (old_rules->n > 0) { + struct cls_conjunction *conjs; + size_t n_conjs; + size_t i; - ovs_mutex_lock(&rule->mutex); - if (fm->command == OFPFC_ADD) { - rule->idle_timeout = fm->idle_timeout; - rule->hard_timeout = fm->hard_timeout; - rule->flags = fm->flags & OFPUTIL_FF_STATE; - rule->created = now; - } - if (change_cookie) { - rule->flow_cookie = fm->new_cookie; - } - rule->modified = now; - ovs_mutex_unlock(&rule->mutex); + /* Create a new 'modified' rule for each old rule. */ + for (i = 0; i < old_rules->n; i++) { + struct rule *old_rule = old_rules->rules[i]; + struct rule *new_rule; + struct cls_rule cr; - if (change_cookie) { - cookies_insert(ofproto, rule); - } - if (fm->command == OFPFC_ADD) { - if (fm->idle_timeout || fm->hard_timeout) { - if (!rule->eviction_group) { - eviction_group_add_rule(rule); - } + cls_rule_clone(&cr, &old_rule->cr); + error = replace_rule_create(ofproto, fm, &cr, old_rule->table_id, + old_rule, &new_rule); + if (!error) { + rule_collection_add(new_rules, new_rule); } else { - eviction_group_remove_rule(rule); + rule_collection_unref(new_rules); + rule_collection_destroy(new_rules); + return error; } } + ovs_assert(new_rules->n == old_rules->n); - if (change_actions) { - ovsrcu_set(&rule->actions, rule_actions_create(fm->ofpacts, - fm->ofpacts_len)); - } - - if (change_actions || reset_counters) { - ofproto->ofproto_class->rule_modify_actions(rule, reset_counters); - } - - if (event != NXFME_MODIFIED || change_actions || change_cookie) { - ofmonitor_report(ofproto->connmgr, rule, event, 0, - req ? req->ofconn : NULL, req ? req->xid : 0, - change_actions ? actions : NULL); + get_conjunctions(fm, &conjs, &n_conjs); + for (i = 0; i < old_rules->n; i++) { + replace_rule_start(ofproto, ofm->version, old_rules->rules[i], + new_rules->rules[i], conjs, n_conjs); } - - if (change_actions) { - learned_cookies_inc(ofproto, rule_get_actions(rule)); - learned_cookies_dec(ofproto, actions, &dead_cookies); - rule_actions_destroy(actions); + free(conjs); + } else if (!(fm->cookie_mask != htonll(0) + || fm->new_cookie == OVS_BE64_MAX)) { + /* No match, add a new flow. */ + error = add_flow_start(ofproto, ofm); + if (!error) { + ovs_assert(fm->delete_reason == OFPRR_EVICTION + || !old_rules->rules[0]); } + new_rules->n = 1; + } else { + error = 0; } - learned_cookies_flush(ofproto, &dead_cookies); - - if (fm->buffer_id != UINT32_MAX && req) { - return send_buffered_packet(req->ofconn, fm->buffer_id, - rules->rules[0]); - } - - return 0; -} -static enum ofperr -modify_flows_add(struct ofproto *ofproto, struct ofputil_flow_mod *fm, - const struct flow_mod_requester *req) - OVS_REQUIRES(ofproto_mutex) -{ - if (fm->cookie_mask != htonll(0) || fm->new_cookie == OVS_BE64_MAX) { - return 0; - } - return add_flow(ofproto, fm, req); + return error; } /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on @@ -4233,165 +4930,268 @@ modify_flows_add(struct ofproto *ofproto, struct ofputil_flow_mod *fm, * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id, * if any. */ static enum ofperr -modify_flows_loose(struct ofproto *ofproto, struct ofputil_flow_mod *fm, - const struct flow_mod_requester *req) +modify_flows_start_loose(struct ofproto *ofproto, struct ofproto_flow_mod *ofm) OVS_REQUIRES(ofproto_mutex) { + struct ofputil_flow_mod *fm = &ofm->fm; + struct rule_collection *old_rules = &ofm->old_rules; struct rule_criteria criteria; - struct rule_collection rules; - int error; + enum ofperr error; - rule_criteria_init(&criteria, fm->table_id, &fm->match, 0, - fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY); + rule_criteria_init(&criteria, fm->table_id, &fm->match, 0, CLS_MAX_VERSION, + fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG_ANY); rule_criteria_require_rw(&criteria, (fm->flags & OFPUTIL_FF_NO_READONLY) != 0); - error = collect_rules_loose(ofproto, &criteria, &rules); + error = collect_rules_loose(ofproto, &criteria, old_rules); rule_criteria_destroy(&criteria); if (!error) { - error = (rules.n > 0 - ? modify_flows__(ofproto, fm, &rules, req) - : modify_flows_add(ofproto, fm, req)); + error = modify_flows_start__(ofproto, ofm); } - rule_collection_destroy(&rules); - + if (error) { + rule_collection_destroy(old_rules); + } return error; } +static void +modify_flows_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm) + OVS_REQUIRES(ofproto_mutex) +{ + struct rule_collection *old_rules = &ofm->old_rules; + struct rule_collection *new_rules = &ofm->new_rules; + + /* Old rules were not changed yet, only need to revert new rules. */ + if (old_rules->n == 0 && new_rules->n == 1) { + add_flow_revert(ofproto, ofm); + } else if (old_rules->n > 0) { + for (size_t i = 0; i < old_rules->n; i++) { + replace_rule_revert(ofproto, old_rules->rules[i], + new_rules->rules[i]); + } + rule_collection_destroy(new_rules); + rule_collection_destroy(old_rules); + } +} + +static void +modify_flows_finish(struct ofproto *ofproto, struct ofproto_flow_mod *ofm, + const struct flow_mod_requester *req) + OVS_REQUIRES(ofproto_mutex) +{ + struct ofputil_flow_mod *fm = &ofm->fm; + struct rule_collection *old_rules = &ofm->old_rules; + struct rule_collection *new_rules = &ofm->new_rules; + + if (old_rules->n == 0 && new_rules->n == 1) { + add_flow_finish(ofproto, ofm, req); + } else if (old_rules->n > 0) { + struct ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies); + + ovs_assert(new_rules->n == old_rules->n); + + for (size_t i = 0; i < old_rules->n; i++) { + replace_rule_finish(ofproto, fm, req, old_rules->rules[i], + new_rules->rules[i], &dead_cookies); + } + learned_cookies_flush(ofproto, &dead_cookies); + rule_collection_remove_postponed(old_rules); + + send_buffered_packet(req, fm->buffer_id, new_rules->rules[0]); + rule_collection_destroy(new_rules); + } +} + /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error * code on failure. */ static enum ofperr -modify_flow_strict(struct ofproto *ofproto, struct ofputil_flow_mod *fm, - const struct flow_mod_requester *req) +modify_flow_start_strict(struct ofproto *ofproto, struct ofproto_flow_mod *ofm) OVS_REQUIRES(ofproto_mutex) { + struct ofputil_flow_mod *fm = &ofm->fm; + struct rule_collection *old_rules = &ofm->old_rules; struct rule_criteria criteria; - struct rule_collection rules; - int error; + enum ofperr error; rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority, - fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY); + CLS_MAX_VERSION, fm->cookie, fm->cookie_mask, OFPP_ANY, + OFPG_ANY); rule_criteria_require_rw(&criteria, (fm->flags & OFPUTIL_FF_NO_READONLY) != 0); - error = collect_rules_strict(ofproto, &criteria, &rules); + error = collect_rules_strict(ofproto, &criteria, old_rules); rule_criteria_destroy(&criteria); if (!error) { - if (rules.n == 0) { - error = modify_flows_add(ofproto, fm, req); - } else if (rules.n == 1) { - error = modify_flows__(ofproto, fm, &rules, req); - } + /* collect_rules_strict() can return max 1 rule. */ + error = modify_flows_start__(ofproto, ofm); } - rule_collection_destroy(&rules); - + if (error) { + rule_collection_destroy(old_rules); + } return error; } /* OFPFC_DELETE implementation. */ -/* Deletes the rules listed in 'rules'. */ static void -delete_flows__(const struct rule_collection *rules, - enum ofp_flow_removed_reason reason, - const struct flow_mod_requester *req) +delete_flows_start__(struct ofproto *ofproto, cls_version_t version, + const struct rule_collection *rules) + OVS_REQUIRES(ofproto_mutex) +{ + for (size_t i = 0; i < rules->n; i++) { + struct rule *rule = rules->rules[i]; + struct oftable *table = &ofproto->tables[rule->table_id]; + + table->n_flows--; + cls_rule_make_invisible_in_version(&rule->cr, version); + } +} + +static void +delete_flows_finish__(struct ofproto *ofproto, + struct rule_collection *rules, + enum ofp_flow_removed_reason reason, + const struct flow_mod_requester *req) OVS_REQUIRES(ofproto_mutex) { if (rules->n) { - struct list dead_cookies = LIST_INITIALIZER(&dead_cookies); - struct ofproto *ofproto = rules->rules[0]->ofproto; - size_t i; + struct ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies); - for (i = 0; i < rules->n; i++) { + for (size_t i = 0; i < rules->n; i++) { struct rule *rule = rules->rules[i]; - const struct rule_actions *actions = rule_get_actions(rule); - ofproto_rule_send_removed(rule, reason); + /* This value will be used to send the flow removed message right + * before the rule is actually destroyed. */ + rule->removed_reason = reason; ofmonitor_report(ofproto->connmgr, rule, NXFME_DELETED, reason, - req ? req->ofconn : NULL, req ? req->xid : 0, - NULL); - oftable_remove_rule(rule); - ofproto->ofproto_class->rule_delete(rule); - - learned_cookies_dec(ofproto, actions, &dead_cookies); + req ? req->ofconn : NULL, + req ? req->request->xid : 0, NULL); + ofproto_rule_remove__(ofproto, rule); + learned_cookies_dec(ofproto, rule_get_actions(rule), + &dead_cookies); } + rule_collection_remove_postponed(rules); + learned_cookies_flush(ofproto, &dead_cookies); + } +} + +/* Deletes the rules listed in 'rules'. + * The deleted rules will become invisible to the lookups in the next version. + * Destroys 'rules'. */ +static void +delete_flows__(struct rule_collection *rules, + enum ofp_flow_removed_reason reason, + const struct flow_mod_requester *req) + OVS_REQUIRES(ofproto_mutex) +{ + if (rules->n) { + struct ofproto *ofproto = rules->rules[0]->ofproto; + + delete_flows_start__(ofproto, ofproto->tables_version + 1, rules); + ofproto_bump_tables_version(ofproto); + delete_flows_finish__(ofproto, rules, reason, req); ofmonitor_flush(ofproto->connmgr); } } /* Implements OFPFC_DELETE. */ static enum ofperr -delete_flows_loose(struct ofproto *ofproto, - const struct ofputil_flow_mod *fm, - const struct flow_mod_requester *req) +delete_flows_start_loose(struct ofproto *ofproto, struct ofproto_flow_mod *ofm) OVS_REQUIRES(ofproto_mutex) { + const struct ofputil_flow_mod *fm = &ofm->fm; + struct rule_collection *rules = &ofm->old_rules; struct rule_criteria criteria; - struct rule_collection rules; enum ofperr error; - rule_criteria_init(&criteria, fm->table_id, &fm->match, 0, - fm->cookie, fm->cookie_mask, - fm->out_port, fm->out_group); + rule_criteria_init(&criteria, fm->table_id, &fm->match, 0, CLS_MAX_VERSION, + fm->cookie, fm->cookie_mask, fm->out_port, + fm->out_group); rule_criteria_require_rw(&criteria, (fm->flags & OFPUTIL_FF_NO_READONLY) != 0); - error = collect_rules_loose(ofproto, &criteria, &rules); + error = collect_rules_loose(ofproto, &criteria, rules); rule_criteria_destroy(&criteria); - if (!error && rules.n > 0) { - delete_flows__(&rules, fm->delete_reason, req); + if (!error) { + delete_flows_start__(ofproto, ofm->version, rules); } - rule_collection_destroy(&rules); return error; } +static void +delete_flows_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm) + OVS_REQUIRES(ofproto_mutex) +{ + struct rule_collection *rules = &ofm->old_rules; + + for (size_t i = 0; i < rules->n; i++) { + struct rule *rule = rules->rules[i]; + struct oftable *table = &ofproto->tables[rule->table_id]; + + /* Restore table's rule count. */ + table->n_flows++; + + /* Restore the original visibility of the rule. */ + cls_rule_restore_visibility(&rule->cr); + } + rule_collection_destroy(rules); +} + +static void +delete_flows_finish(struct ofproto *ofproto, + struct ofproto_flow_mod *ofm, + const struct flow_mod_requester *req) + OVS_REQUIRES(ofproto_mutex) +{ + delete_flows_finish__(ofproto, &ofm->old_rules, ofm->fm.delete_reason, + req); +} + /* Implements OFPFC_DELETE_STRICT. */ static enum ofperr -delete_flow_strict(struct ofproto *ofproto, const struct ofputil_flow_mod *fm, - const struct flow_mod_requester *req) +delete_flow_start_strict(struct ofproto *ofproto, + struct ofproto_flow_mod *ofm) OVS_REQUIRES(ofproto_mutex) { + const struct ofputil_flow_mod *fm = &ofm->fm; + struct rule_collection *rules = &ofm->old_rules; struct rule_criteria criteria; - struct rule_collection rules; enum ofperr error; rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority, - fm->cookie, fm->cookie_mask, + CLS_MAX_VERSION, fm->cookie, fm->cookie_mask, fm->out_port, fm->out_group); rule_criteria_require_rw(&criteria, (fm->flags & OFPUTIL_FF_NO_READONLY) != 0); - error = collect_rules_strict(ofproto, &criteria, &rules); + error = collect_rules_strict(ofproto, &criteria, rules); rule_criteria_destroy(&criteria); - if (!error && rules.n > 0) { - delete_flows__(&rules, fm->delete_reason, req); + if (!error) { + delete_flows_start__(ofproto, ofm->version, rules); } - rule_collection_destroy(&rules); return error; } +/* This may only be called by rule_destroy_cb()! */ static void -ofproto_rule_send_removed(struct rule *rule, uint8_t reason) - OVS_REQUIRES(ofproto_mutex) +ofproto_rule_send_removed(struct rule *rule) + OVS_EXCLUDED(ofproto_mutex) { struct ofputil_flow_removed fr; long long int used; - if (rule_is_hidden(rule) || - !(rule->flags & OFPUTIL_FF_SEND_FLOW_REM)) { - return; - } - minimatch_expand(&rule->cr.match, &fr.match); fr.priority = rule->cr.priority; + + ovs_mutex_lock(&ofproto_mutex); fr.cookie = rule->flow_cookie; - fr.reason = reason; + fr.reason = rule->removed_reason; fr.table_id = rule->table_id; calc_duration(rule->created, time_msec(), &fr.duration_sec, &fr.duration_nsec); @@ -4401,8 +5201,8 @@ ofproto_rule_send_removed(struct rule *rule, uint8_t reason) ovs_mutex_unlock(&rule->mutex); rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count, &fr.byte_count, &used); - connmgr_send_flow_removed(rule->ofproto->connmgr, &fr); + ovs_mutex_unlock(&ofproto_mutex); } /* Sends an OpenFlow "flow removed" message with the given 'reason' (either @@ -4415,7 +5215,12 @@ void ofproto_rule_expire(struct rule *rule, uint8_t reason) OVS_REQUIRES(ofproto_mutex) { - ofproto_rule_delete__(rule, reason); + struct rule_collection rules; + + rules.rules = rules.stub; + rules.n = 1; + rules.stub[0] = rule; + delete_flows__(&rules, reason, NULL); } /* Reduces '*timeout' to no more than 'max'. A value of zero in either case @@ -4459,7 +5264,7 @@ handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh) OVS_EXCLUDED(ofproto_mutex) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - struct ofputil_flow_mod fm; + struct ofproto_flow_mod ofm; uint64_t ofpacts_stub[1024 / 8]; struct ofpbuf ofpacts; enum ofperr error; @@ -4470,25 +5275,26 @@ handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh) } ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); - error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn), + error = ofputil_decode_flow_mod(&ofm.fm, oh, ofconn_get_protocol(ofconn), &ofpacts, u16_to_ofp(ofproto->max_ports), ofproto->n_tables); if (!error) { - error = ofproto_check_ofpacts(ofproto, fm.ofpacts, fm.ofpacts_len); + error = ofproto_check_ofpacts(ofproto, ofm.fm.ofpacts, + ofm.fm.ofpacts_len); } if (!error) { struct flow_mod_requester req; req.ofconn = ofconn; - req.xid = oh->xid; - error = handle_flow_mod__(ofproto, &fm, &req); + req.request = oh; + error = handle_flow_mod__(ofproto, &ofm, &req); } if (error) { goto exit_free_ofpacts; } - ofconn_report_flow_mod(ofconn, fm.command); + ofconn_report_flow_mod(ofconn, ofm.fm.command); exit_free_ofpacts: ofpbuf_uninit(&ofpacts); @@ -4497,42 +5303,18 @@ exit: } static enum ofperr -handle_flow_mod__(struct ofproto *ofproto, struct ofputil_flow_mod *fm, +handle_flow_mod__(struct ofproto *ofproto, struct ofproto_flow_mod *ofm, const struct flow_mod_requester *req) OVS_EXCLUDED(ofproto_mutex) { enum ofperr error; ovs_mutex_lock(&ofproto_mutex); - switch (fm->command) { - case OFPFC_ADD: - error = add_flow(ofproto, fm, req); - break; - - case OFPFC_MODIFY: - error = modify_flows_loose(ofproto, fm, req); - break; - - case OFPFC_MODIFY_STRICT: - error = modify_flow_strict(ofproto, fm, req); - break; - - case OFPFC_DELETE: - error = delete_flows_loose(ofproto, fm, req); - break; - - case OFPFC_DELETE_STRICT: - error = delete_flow_strict(ofproto, fm, req); - break; - - default: - if (fm->command > 0xff) { - VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but " - "flow_mod_table_id extension is not enabled", - ofproto->name); - } - error = OFPERR_OFPFMFC_BAD_COMMAND; - break; + ofm->version = ofproto->tables_version + 1; + error = ofproto_flow_mod_start(ofproto, ofm); + if (!error) { + ofproto_bump_tables_version(ofproto); + ofproto_flow_mod_finish(ofproto, ofm, req); } ofmonitor_flush(ofproto->connmgr); ovs_mutex_unlock(&ofproto_mutex); @@ -4555,7 +5337,8 @@ handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh) } if (request.role != OFPCR12_ROLE_NOCHANGE) { - if (request.have_generation_id + if (request.role != OFPCR12_ROLE_EQUAL + && request.have_generation_id && !ofconn_set_master_election_id(ofconn, request.generation_id)) { return OFPERR_OFPRRFC_STALE; } @@ -4624,17 +5407,14 @@ handle_nxt_set_packet_in_format(struct ofconn *ofconn, static enum ofperr handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh) { - const struct nx_async_config *msg = ofpmsg_body(oh); - uint32_t master[OAM_N_TYPES]; - uint32_t slave[OAM_N_TYPES]; - - master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]); - master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]); - master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]); + enum ofperr error; + uint32_t master[OAM_N_TYPES] = {0}; + uint32_t slave[OAM_N_TYPES] = {0}; - slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]); - slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]); - slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]); + error = ofputil_decode_set_async_config(oh, master, slave, false); + if (error) { + return error; + } ofconn_set_async_config(ofconn, master, slave); if (ofconn_get_type(ofconn) == OFCONN_SERVICE && @@ -4651,20 +5431,10 @@ 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]); + buf = ofputil_encode_get_async_config(oh, master, slave); ofconn_send_reply(ofconn, buf); return 0; @@ -4699,7 +5469,7 @@ handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh) static void ofproto_compose_flow_refresh_update(const struct rule *rule, enum nx_flow_monitor_flags flags, - struct list *msgs) + struct ovs_list *msgs) OVS_REQUIRES(ofproto_mutex) { const struct rule_actions *actions; @@ -4731,7 +5501,7 @@ ofproto_compose_flow_refresh_update(const struct rule *rule, void ofmonitor_compose_refresh_updates(struct rule_collection *rules, - struct list *msgs) + struct ovs_list *msgs) OVS_REQUIRES(ofproto_mutex) { size_t i; @@ -4797,7 +5567,7 @@ ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m, FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) { struct rule *rule; - CLS_FOR_EACH_TARGET (rule, cr, &table->cls, &target) { + CLS_FOR_EACH_TARGET (rule, cr, &table->cls, &target, CLS_MAX_VERSION) { ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules); } } @@ -4848,12 +5618,11 @@ handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh) struct ofmonitor **monitors; size_t n_monitors, allocated_monitors; struct rule_collection rules; - struct list replies; + struct ovs_list replies; enum ofperr error; struct ofpbuf b; size_t i; - error = 0; ofpbuf_use_const(&b, oh, ntohs(oh->length)); monitors = NULL; n_monitors = allocated_monitors = 0; @@ -4942,7 +5711,7 @@ handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh) */ struct meter { long long int created; /* Time created. */ - struct list rules; /* List of "struct rule_dpif"s. */ + struct ovs_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. */ @@ -5098,15 +5867,12 @@ handle_delete_meter(struct ofconn *ofconn, struct ofputil_meter_mod *mm) } } } - if (rules.n > 0) { - delete_flows__(&rules, OFPRR_METER_DELETE, NULL); - } + delete_flows__(&rules, OFPRR_METER_DELETE, NULL); /* Delete the meters. */ meter_delete(ofproto, first, last); ovs_mutex_unlock(&ofproto_mutex); - rule_collection_destroy(&rules); return error; } @@ -5168,6 +5934,14 @@ handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh) break; } + if (!error) { + struct ofputil_requestforward rf; + rf.xid = oh->xid; + rf.reason = OFPRFR_METER_MOD; + rf.meter_mod = &mm; + connmgr_send_requestforward(ofproto->connmgr, ofconn, &rf); + } + exit_free_bands: ofpbuf_uninit(&bands); return error; @@ -5197,7 +5971,7 @@ handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request, enum ofptype type) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - struct list replies; + struct ovs_list replies; uint64_t bands_stub[256 / 8]; struct ofpbuf bands; uint32_t meter_id, first, last; @@ -5333,8 +6107,8 @@ group_get_ref_count(struct ofgroup *group) uint32_t count; match_init_catchall(&match); - rule_criteria_init(&criteria, 0xff, &match, 0, htonll(0), htonll(0), - OFPP_ANY, group->group_id); + rule_criteria_init(&criteria, 0xff, &match, 0, CLS_MAX_VERSION, htonll(0), + htonll(0), OFPP_ANY, group->group_id); ovs_mutex_lock(&ofproto_mutex); error = collect_rules_loose(ofproto, &criteria, &rules); ovs_mutex_unlock(&ofproto_mutex); @@ -5347,7 +6121,7 @@ group_get_ref_count(struct ofgroup *group) } static void -append_group_stats(struct ofgroup *group, struct list *replies) +append_group_stats(struct ofgroup *group, struct ovs_list *replies) { struct ofputil_group_stats ogs; const struct ofproto *ofproto = group->ofproto; @@ -5381,11 +6155,11 @@ append_group_stats(struct ofgroup *group, struct list *replies) static void handle_group_request(struct ofconn *ofconn, const struct ofp_header *request, uint32_t group_id, - void (*cb)(struct ofgroup *, struct list *replies)) + void (*cb)(struct ofgroup *, struct ovs_list *replies)) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); struct ofgroup *group; - struct list replies; + struct ovs_list replies; ofpmp_init(&replies, request); if (group_id == OFPG_ALL) { @@ -5420,12 +6194,14 @@ handle_group_stats_request(struct ofconn *ofconn, } static void -append_group_desc(struct ofgroup *group, struct list *replies) +append_group_desc(struct ofgroup *group, struct ovs_list *replies) { struct ofputil_group_desc gds; gds.group_id = group->group_id; gds.type = group->type; + gds.props = group->props; + ofputil_append_group_desc_reply(&gds, &group->buckets, replies); } @@ -5528,6 +6304,9 @@ init_group(struct ofproto *ofproto, struct ofputil_group_mod *gm, *CONST_CAST(uint32_t *, &(*ofgroup)->n_buckets) = list_size(&(*ofgroup)->buckets); + memcpy(CONST_CAST(struct ofputil_group_props *, &(*ofgroup)->props), + &gm->props, sizeof (struct ofputil_group_props)); + /* Construct called BEFORE any locks are held. */ error = ofproto->ofproto_class->group_construct(*ofgroup); if (error) { @@ -5585,7 +6364,104 @@ add_group(struct ofproto *ofproto, struct ofputil_group_mod *gm) return error; } -/* Implements OFPGC11_MODIFY. Returns 0 on success or an OpenFlow error code +/* Adds all of the buckets from 'ofgroup' to 'new_ofgroup'. The buckets + * already in 'new_ofgroup' will be placed just after the (copy of the) bucket + * in 'ofgroup' with bucket ID 'command_bucket_id'. Special + * 'command_bucket_id' values OFPG15_BUCKET_FIRST and OFPG15_BUCKET_LAST are + * also honored. */ +static enum ofperr +copy_buckets_for_insert_bucket(const struct ofgroup *ofgroup, + struct ofgroup *new_ofgroup, + uint32_t command_bucket_id) +{ + struct ofputil_bucket *last = NULL; + + if (command_bucket_id <= OFPG15_BUCKET_MAX) { + /* Check here to ensure that a bucket corresponding to + * command_bucket_id exists in the old bucket list. + * + * The subsequent search of below of new_ofgroup covers + * both buckets in the old bucket list and buckets added + * by the insert buckets group mod message this function processes. */ + if (!ofputil_bucket_find(&ofgroup->buckets, command_bucket_id)) { + return OFPERR_OFPGMFC_UNKNOWN_BUCKET; + } + + if (!list_is_empty(&new_ofgroup->buckets)) { + last = ofputil_bucket_list_back(&new_ofgroup->buckets); + } + } + + ofputil_bucket_clone_list(&new_ofgroup->buckets, &ofgroup->buckets, NULL); + + if (ofputil_bucket_check_duplicate_id(&new_ofgroup->buckets)) { + VLOG_INFO_RL(&rl, "Duplicate bucket id"); + return OFPERR_OFPGMFC_BUCKET_EXISTS; + } + + /* Rearrange list according to command_bucket_id */ + if (command_bucket_id == OFPG15_BUCKET_LAST) { + if (!list_is_empty(&ofgroup->buckets)) { + struct ofputil_bucket *new_first; + const struct ofputil_bucket *first; + + first = ofputil_bucket_list_front(&ofgroup->buckets); + new_first = ofputil_bucket_find(&new_ofgroup->buckets, + first->bucket_id); + + list_splice(new_ofgroup->buckets.next, &new_first->list_node, + &new_ofgroup->buckets); + } + } else if (command_bucket_id <= OFPG15_BUCKET_MAX && last) { + struct ofputil_bucket *after; + + /* Presence of bucket is checked above so after should never be NULL */ + after = ofputil_bucket_find(&new_ofgroup->buckets, command_bucket_id); + + list_splice(after->list_node.next, new_ofgroup->buckets.next, + last->list_node.next); + } + + return 0; +} + +/* Appends all of the a copy of all the buckets from 'ofgroup' to 'new_ofgroup' + * with the exception of the bucket whose bucket id is 'command_bucket_id'. + * Special 'command_bucket_id' values OFPG15_BUCKET_FIRST, OFPG15_BUCKET_LAST + * and OFPG15_BUCKET_ALL are also honored. */ +static enum ofperr +copy_buckets_for_remove_bucket(const struct ofgroup *ofgroup, + struct ofgroup *new_ofgroup, + uint32_t command_bucket_id) +{ + const struct ofputil_bucket *skip = NULL; + + if (command_bucket_id == OFPG15_BUCKET_ALL) { + return 0; + } + + if (command_bucket_id == OFPG15_BUCKET_FIRST) { + if (!list_is_empty(&ofgroup->buckets)) { + skip = ofputil_bucket_list_front(&ofgroup->buckets); + } + } else if (command_bucket_id == OFPG15_BUCKET_LAST) { + if (!list_is_empty(&ofgroup->buckets)) { + skip = ofputil_bucket_list_back(&ofgroup->buckets); + } + } else { + skip = ofputil_bucket_find(&ofgroup->buckets, command_bucket_id); + if (!skip) { + return OFPERR_OFPGMFC_UNKNOWN_BUCKET; + } + } + + ofputil_bucket_clone_list(&new_ofgroup->buckets, &ofgroup->buckets, skip); + + return 0; +} + +/* Implements OFPGC11_MODIFY, OFPGC15_INSERT_BUCKET and + * OFPGC15_REMOVE_BUCKET. Returns 0 on success or an OpenFlow error code * on failure. * * Note that the group is re-created and then replaces the old group in @@ -5617,6 +6493,18 @@ modify_group(struct ofproto *ofproto, struct ofputil_group_mod *gm) goto out; } + /* Manipulate bucket list for bucket commands */ + if (gm->command == OFPGC15_INSERT_BUCKET) { + error = copy_buckets_for_insert_bucket(ofgroup, new_ofgroup, + gm->command_bucket_id); + } else if (gm->command == OFPGC15_REMOVE_BUCKET) { + error = copy_buckets_for_remove_bucket(ofgroup, new_ofgroup, + gm->command_bucket_id); + } + if (error) { + goto out; + } + /* The group creation time does not change during modification. */ *CONST_CAST(long long int *, &(new_ofgroup->created)) = ofgroup->created; *CONST_CAST(long long int *, &(new_ofgroup->modified)) = time_msec(); @@ -5647,14 +6535,15 @@ delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup) OVS_RELEASES(ofproto->groups_rwlock) { struct match match; - struct ofputil_flow_mod fm; + struct ofproto_flow_mod ofm; /* Delete all flow entries containing this group in a group action */ match_init_catchall(&match); - flow_mod_init(&fm, &match, 0, NULL, 0, OFPFC_DELETE); - fm.delete_reason = OFPRR_GROUP_DELETE; - fm.out_group = ofgroup->group_id; - handle_flow_mod__(ofproto, &fm, NULL); + flow_mod_init(&ofm.fm, &match, 0, NULL, 0, OFPFC_DELETE); + ofm.fm.delete_reason = OFPRR_GROUP_DELETE; + ofm.fm.out_group = ofgroup->group_id; + ofm.fm.table_id = OFPTT_ALL; + handle_flow_mod__(ofproto, &ofm, NULL); hmap_remove(&ofproto->groups, &ofgroup->hmap_node); /* No-one can find this group any more. */ @@ -5694,6 +6583,16 @@ delete_group(struct ofproto *ofproto, uint32_t group_id) ovs_rwlock_unlock(&ofproto->groups_rwlock); } +/* Delete all groups from 'ofproto'. + * + * This is intended for use within an ofproto provider's 'destruct' + * function. */ +void +ofproto_group_delete_all(struct ofproto *ofproto) +{ + delete_group(ofproto, OFPG_ALL); +} + static enum ofperr handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh) { @@ -5713,51 +6612,121 @@ handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh) switch (gm.command) { case OFPGC11_ADD: - return add_group(ofproto, &gm); + error = add_group(ofproto, &gm); + break; case OFPGC11_MODIFY: - return modify_group(ofproto, &gm); + error = modify_group(ofproto, &gm); + break; case OFPGC11_DELETE: delete_group(ofproto, gm.group_id); - return 0; + error = 0; + break; + + case OFPGC15_INSERT_BUCKET: + error = modify_group(ofproto, &gm); + break; + + case OFPGC15_REMOVE_BUCKET: + error = modify_group(ofproto, &gm); + break; default: if (gm.command > OFPGC11_DELETE) { - VLOG_WARN_RL(&rl, "%s: Invalid group_mod command type %d", + VLOG_INFO_RL(&rl, "%s: Invalid group_mod command type %d", ofproto->name, gm.command); } return OFPERR_OFPGMFC_BAD_COMMAND; } + + if (!error) { + struct ofputil_requestforward rf; + rf.xid = oh->xid; + rf.reason = OFPRFR_GROUP_MOD; + rf.group_mod = &gm; + connmgr_send_requestforward(ofproto->connmgr, ofconn, &rf); + } + return error; +} + +enum ofputil_table_miss +ofproto_table_get_miss_config(const struct ofproto *ofproto, uint8_t table_id) +{ + enum ofputil_table_miss miss; + + atomic_read_relaxed(&ofproto->tables[table_id].miss_config, &miss); + return miss; } -enum ofproto_table_config -ofproto_table_get_config(const struct ofproto *ofproto, uint8_t table_id) +static void +table_mod__(struct oftable *oftable, + const struct ofputil_table_mod *tm) { - unsigned int value; - atomic_read(&ofproto->tables[table_id].config, &value); - return (enum ofproto_table_config)value; + if (tm->miss == OFPUTIL_TABLE_MISS_DEFAULT) { + /* This is how an OFPT_TABLE_MOD decodes if it doesn't specify any + * table-miss configuration (because the protocol used doesn't have + * such a concept), so there's nothing to do. */ + } else { + atomic_store_relaxed(&oftable->miss_config, tm->miss); + } + + unsigned int new_eviction = oftable->eviction; + if (tm->eviction == OFPUTIL_TABLE_EVICTION_ON) { + new_eviction |= EVICTION_OPENFLOW; + } else if (tm->eviction == OFPUTIL_TABLE_EVICTION_OFF) { + new_eviction &= ~EVICTION_OPENFLOW; + } + + if (new_eviction != oftable->eviction) { + ovs_mutex_lock(&ofproto_mutex); + oftable_configure_eviction(oftable, new_eviction, + oftable->eviction_fields, + oftable->n_eviction_fields); + ovs_mutex_unlock(&ofproto_mutex); + } + + if (tm->vacancy != OFPUTIL_TABLE_VACANCY_DEFAULT) { + ovs_mutex_lock(&ofproto_mutex); + oftable->vacancy_enabled = (tm->vacancy == OFPUTIL_TABLE_VACANCY_ON + ? OFPTC14_VACANCY_EVENTS + : 0); + oftable->vacancy_down = tm->table_vacancy.vacancy_down; + oftable->vacancy_up = tm->table_vacancy.vacancy_up; + ovs_mutex_unlock(&ofproto_mutex); + } } static enum ofperr table_mod(struct ofproto *ofproto, const struct ofputil_table_mod *tm) { - /* Only accept currently supported configurations */ - if (tm->config & ~OFPTC11_TABLE_MISS_MASK) { + if (!check_table_id(ofproto, tm->table_id)) { + return OFPERR_OFPTMFC_BAD_TABLE; + } + + /* Don't allow the eviction flags to be changed (except to the only fixed + * value that OVS supports). OF1.4 says this is normal: "The + * OFPTMPT_EVICTION property usually cannot be modified using a + * OFP_TABLE_MOD request, because the eviction mechanism is switch + * defined". */ + if (tm->eviction_flags != UINT32_MAX + && tm->eviction_flags != OFPROTO_EVICTION_FLAGS) { return OFPERR_OFPTMFC_BAD_CONFIG; } if (tm->table_id == OFPTT_ALL) { - int i; - for (i = 0; i < ofproto->n_tables; i++) { - atomic_store(&ofproto->tables[i].config, - (unsigned int)tm->config); + struct oftable *oftable; + OFPROTO_FOR_EACH_TABLE (oftable, ofproto) { + if (!(oftable->flags & (OFTABLE_HIDDEN | OFTABLE_READONLY))) { + table_mod__(oftable, tm); + } } - } else if (!check_table_id(ofproto, tm->table_id)) { - return OFPERR_OFPTMFC_BAD_TABLE; } else { - atomic_store(&ofproto->tables[tm->table_id].config, - (unsigned int)tm->config); + struct oftable *oftable = &ofproto->tables[tm->table_id]; + if (oftable->flags & OFTABLE_READONLY) { + return OFPERR_OFPTMFC_EPERM; + } + table_mod__(oftable, tm); } return 0; @@ -5784,12 +6753,212 @@ handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh) } static enum ofperr -handle_bundle_control(struct ofconn *ofconn, const struct ofp_header *oh) +ofproto_flow_mod_start(struct ofproto *ofproto, struct ofproto_flow_mod *ofm) + OVS_REQUIRES(ofproto_mutex) +{ + switch (ofm->fm.command) { + case OFPFC_ADD: + return add_flow_start(ofproto, ofm); + /* , &be->old_rules.stub[0], + &be->new_rules.stub[0]); */ + case OFPFC_MODIFY: + return modify_flows_start_loose(ofproto, ofm); + case OFPFC_MODIFY_STRICT: + return modify_flow_start_strict(ofproto, ofm); + case OFPFC_DELETE: + return delete_flows_start_loose(ofproto, ofm); + + case OFPFC_DELETE_STRICT: + return delete_flow_start_strict(ofproto, ofm); + } + + return OFPERR_OFPFMFC_BAD_COMMAND; +} + +static void +ofproto_flow_mod_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm) + OVS_REQUIRES(ofproto_mutex) +{ + switch (ofm->fm.command) { + case OFPFC_ADD: + add_flow_revert(ofproto, ofm); + break; + + case OFPFC_MODIFY: + case OFPFC_MODIFY_STRICT: + modify_flows_revert(ofproto, ofm); + break; + + case OFPFC_DELETE: + case OFPFC_DELETE_STRICT: + delete_flows_revert(ofproto, ofm); + break; + + default: + break; + } +} + +static void +ofproto_flow_mod_finish(struct ofproto *ofproto, + struct ofproto_flow_mod *ofm, + const struct flow_mod_requester *req) + OVS_REQUIRES(ofproto_mutex) +{ + switch (ofm->fm.command) { + case OFPFC_ADD: + add_flow_finish(ofproto, ofm, req); + break; + + case OFPFC_MODIFY: + case OFPFC_MODIFY_STRICT: + modify_flows_finish(ofproto, ofm, req); + break; + + case OFPFC_DELETE: + case OFPFC_DELETE_STRICT: + delete_flows_finish(ofproto, ofm, req); + break; + + default: + break; + } +} + +/* Commit phases (all while locking ofproto_mutex): + * + * 1. Begin: Gather resources and make changes visible in the next version. + * - Mark affected rules for removal in the next version. + * - Create new replacement rules, make visible in the next + * version. + * - Do not send any events or notifications. + * + * 2. Revert: Fail if any errors are found. After this point no errors are + * possible. No visible changes were made, so rollback is minimal (remove + * added invisible rules, restore visibility of rules marked for removal). + * + * 3. Finish: Make the changes visible for lookups. Insert replacement rules to + * the ofproto provider. Remove replaced and deleted rules from ofproto data + * structures, and Schedule postponed removal of deleted rules from the + * classifier. Send notifications, buffered packets, etc. + */ +static enum ofperr +do_bundle_commit(struct ofconn *ofconn, uint32_t id, uint16_t flags) { + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); + cls_version_t version = ofproto->tables_version + 1; + struct ofp_bundle *bundle; + struct ofp_bundle_entry *be; enum ofperr error; + + bundle = ofconn_get_bundle(ofconn, id); + + if (!bundle) { + return OFPERR_OFPBFC_BAD_ID; + } + if (bundle->flags != flags) { + error = OFPERR_OFPBFC_BAD_FLAGS; + } else { + bool prev_is_port_mod = false; + + error = 0; + ovs_mutex_lock(&ofproto_mutex); + + /* 1. Begin. */ + LIST_FOR_EACH (be, node, &bundle->msg_list) { + if (be->type == OFPTYPE_PORT_MOD) { + /* Our port mods are not atomic. */ + if (flags & OFPBF_ATOMIC) { + error = OFPERR_OFPBFC_MSG_FAILED; + } else { + prev_is_port_mod = true; + error = port_mod_start(ofconn, &be->opm.pm, &be->opm.port); + } + } else if (be->type == OFPTYPE_FLOW_MOD) { + /* Flow mods between port mods are applied as a single + * version, but the versions are published only after + * we know the commit is successful. */ + if (prev_is_port_mod) { + ++version; + } + prev_is_port_mod = false; + /* Store the version in which the changes should take + * effect. */ + be->ofm.version = version; + error = ofproto_flow_mod_start(ofproto, &be->ofm); + } else { + OVS_NOT_REACHED(); + } + if (error) { + break; + } + } + + if (error) { + /* Send error referring to the original message. */ + if (error) { + ofconn_send_error(ofconn, be->ofp_msg, error); + error = OFPERR_OFPBFC_MSG_FAILED; + } + + /* 2. Revert. Undo all the changes made above. */ + LIST_FOR_EACH_REVERSE_CONTINUE(be, node, &bundle->msg_list) { + if (be->type == OFPTYPE_FLOW_MOD) { + ofproto_flow_mod_revert(ofproto, &be->ofm); + } + /* Nothing needs to be reverted for a port mod. */ + } + } else { + /* 4. Finish. */ + LIST_FOR_EACH (be, node, &bundle->msg_list) { + if (be->type == OFPTYPE_FLOW_MOD) { + struct flow_mod_requester req = { ofconn, be->ofp_msg }; + + /* Bump the lookup version to the one of the current + * message. This makes all the changes in the bundle at + * this version visible to lookups at once. */ + if (ofproto->tables_version < be->ofm.version) { + ofproto->tables_version = be->ofm.version; + ofproto->ofproto_class->set_tables_version( + ofproto, ofproto->tables_version); + } + + ofproto_flow_mod_finish(ofproto, &be->ofm, &req); + } else if (be->type == OFPTYPE_PORT_MOD) { + /* Perform the actual port mod. This is not atomic, i.e., + * the effects will be immediately seen by upcall + * processing regardless of the lookup version. It should + * be noted that port configuration changes can originate + * also from OVSDB changes asynchronously to all upcall + * processing. */ + port_mod_finish(ofconn, &be->opm.pm, be->opm.port); + } + } + } + + ofmonitor_flush(ofproto->connmgr); + ovs_mutex_unlock(&ofproto_mutex); + + run_rule_executes(ofproto); + } + + /* The bundle is discarded regardless the outcome. */ + ofp_bundle_remove__(ofconn, bundle, !error); + return error; +} + +static enum ofperr +handle_bundle_control(struct ofconn *ofconn, const struct ofp_header *oh) +{ struct ofputil_bundle_ctrl_msg bctrl; - struct ofpbuf *buf; struct ofputil_bundle_ctrl_msg reply; + struct ofpbuf *buf; + enum ofperr error; + + error = reject_slave_controller(ofconn); + if (error) { + return error; + } error = ofputil_decode_bundle_ctrl(oh, &bctrl); if (error) { @@ -5805,10 +6974,10 @@ handle_bundle_control(struct ofconn *ofconn, const struct ofp_header *oh) break; case OFPBCT_CLOSE_REQUEST: error = ofp_bundle_close(ofconn, bctrl.bundle_id, bctrl.flags); - reply.type = OFPBCT_CLOSE_REPLY;; + reply.type = OFPBCT_CLOSE_REPLY; break; case OFPBCT_COMMIT_REQUEST: - error = ofp_bundle_commit(ofconn, bctrl.bundle_id, bctrl.flags); + error = do_bundle_commit(ofconn, bctrl.bundle_id, bctrl.flags); reply.type = OFPBCT_COMMIT_REPLY; break; case OFPBCT_DISCARD_REQUEST: @@ -5831,26 +7000,103 @@ handle_bundle_control(struct ofconn *ofconn, const struct ofp_header *oh) return error; } - static enum ofperr handle_bundle_add(struct ofconn *ofconn, const struct ofp_header *oh) { + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); enum ofperr error; struct ofputil_bundle_add_msg badd; + struct ofp_bundle_entry *bmsg; + enum ofptype type; + + error = reject_slave_controller(ofconn); + if (error) { + return error; + } - error = ofputil_decode_bundle_add(oh, &badd); + error = ofputil_decode_bundle_add(oh, &badd, &type); if (error) { return error; } - return ofp_bundle_add_message(ofconn, &badd); + bmsg = ofp_bundle_entry_alloc(type, badd.msg); + + if (type == OFPTYPE_PORT_MOD) { + error = ofputil_decode_port_mod(badd.msg, &bmsg->opm.pm, false); + } else if (type == OFPTYPE_FLOW_MOD) { + struct ofpbuf ofpacts; + uint64_t ofpacts_stub[1024 / 8]; + + ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); + error = ofputil_decode_flow_mod(&bmsg->ofm.fm, badd.msg, + ofconn_get_protocol(ofconn), + &ofpacts, + u16_to_ofp(ofproto->max_ports), + ofproto->n_tables); + /* Move actions to heap. */ + bmsg->ofm.fm.ofpacts = ofpbuf_steal_data(&ofpacts); + + if (!error && bmsg->ofm.fm.ofpacts_len) { + error = ofproto_check_ofpacts(ofproto, bmsg->ofm.fm.ofpacts, + bmsg->ofm.fm.ofpacts_len); + } + } else { + OVS_NOT_REACHED(); + } + + if (!error) { + error = ofp_bundle_add_message(ofconn, badd.bundle_id, badd.flags, + bmsg); + } + + if (error) { + ofp_bundle_entry_free(bmsg); + } + + return error; +} + +static enum ofperr +handle_tlv_table_mod(struct ofconn *ofconn, const struct ofp_header *oh) +{ + struct ofputil_tlv_table_mod ttm; + enum ofperr error; + + error = reject_slave_controller(ofconn); + if (error) { + return error; + } + + error = ofputil_decode_tlv_table_mod(oh, &ttm); + if (error) { + return error; + } + + error = tun_metadata_table_mod(&ttm); + + ofputil_uninit_tlv_table(&ttm.mappings); + return error; +} + +static enum ofperr +handle_tlv_table_request(struct ofconn *ofconn, const struct ofp_header *oh) +{ + struct ofputil_tlv_table_reply ttr; + struct ofpbuf *b; + + tun_metadata_table_request(&ttr); + b = ofputil_encode_tlv_table_reply(oh, &ttr); + ofputil_uninit_tlv_table(&ttr.mappings); + + ofconn_send_reply(ofconn, b); + return 0; } static enum ofperr handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) OVS_EXCLUDED(ofproto_mutex) { - const struct ofp_header *oh = ofpbuf_data(msg); + const struct ofp_header *oh = msg->data; enum ofptype type; enum ofperr error; @@ -5947,6 +7193,12 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) case OFPTYPE_TABLE_STATS_REQUEST: return handle_table_stats_request(ofconn, oh); + case OFPTYPE_TABLE_FEATURES_STATS_REQUEST: + return handle_table_features_request(ofconn, oh); + + case OFPTYPE_TABLE_DESC_REQUEST: + return handle_table_desc_request(ofconn, oh); + case OFPTYPE_PORT_STATS_REQUEST: return handle_port_stats_request(ofconn, oh); @@ -5984,6 +7236,12 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) case OFPTYPE_BUNDLE_ADD_MESSAGE: return handle_bundle_add(ofconn, oh); + case OFPTYPE_NXT_TLV_TABLE_MOD: + return handle_tlv_table_mod(ofconn, oh); + + case OFPTYPE_NXT_TLV_TABLE_REQUEST: + return handle_tlv_table_request(ofconn, oh); + case OFPTYPE_HELLO: case OFPTYPE_ERROR: case OFPTYPE_FEATURES_REPLY: @@ -6011,9 +7269,11 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) case OFPTYPE_METER_STATS_REPLY: case OFPTYPE_METER_CONFIG_STATS_REPLY: case OFPTYPE_METER_FEATURES_STATS_REPLY: - case OFPTYPE_TABLE_FEATURES_STATS_REQUEST: case OFPTYPE_TABLE_FEATURES_STATS_REPLY: + case OFPTYPE_TABLE_DESC_REPLY: case OFPTYPE_ROLE_STATUS: + case OFPTYPE_REQUESTFORWARD: + case OFPTYPE_NXT_TLV_TABLE_REPLY: default: if (ofpmsg_is_stat_request(oh)) { return OFPERR_OFPBRC_BAD_STAT; @@ -6027,27 +7287,29 @@ static void handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg) OVS_EXCLUDED(ofproto_mutex) { - int error = handle_openflow__(ofconn, ofp_msg); + enum ofperr error = handle_openflow__(ofconn, ofp_msg); + if (error) { - ofconn_send_error(ofconn, ofpbuf_data(ofp_msg), error); + ofconn_send_error(ofconn, ofp_msg->data, error); } COVERAGE_INC(ofproto_recv_openflow); } /* Asynchronous operations. */ -static enum ofperr -send_buffered_packet(struct ofconn *ofconn, uint32_t buffer_id, +static void +send_buffered_packet(const struct flow_mod_requester *req, uint32_t buffer_id, struct rule *rule) OVS_REQUIRES(ofproto_mutex) { - enum ofperr error = 0; - if (ofconn && buffer_id != UINT32_MAX) { - struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - struct ofpbuf *packet; + if (req && req->ofconn && buffer_id != UINT32_MAX) { + struct ofproto *ofproto = ofconn_get_ofproto(req->ofconn); + struct dp_packet *packet; ofp_port_t in_port; + enum ofperr error; - error = ofconn_pktbuf_retrieve(ofconn, buffer_id, &packet, &in_port); + error = ofconn_pktbuf_retrieve(req->ofconn, buffer_id, &packet, + &in_port); if (packet) { struct rule_execute *re; @@ -6061,12 +7323,13 @@ send_buffered_packet(struct ofconn *ofconn, uint32_t buffer_id, if (!guarded_list_push_back(&ofproto->rule_executes, &re->list_node, 1024)) { ofproto_rule_unref(rule); - ofpbuf_delete(re->packet); + dp_packet_delete(re->packet); free(re); } + } else { + ofconn_send_error(req->ofconn, req->request, error); } } - return error; } static uint64_t @@ -6076,10 +7339,10 @@ pick_datapath_id(const struct ofproto *ofproto) port = ofproto_get_port(ofproto, OFPP_LOCAL); if (port) { - uint8_t ea[ETH_ADDR_LEN]; + struct eth_addr ea; int error; - error = netdev_get_etheraddr(port->netdev, ea); + error = netdev_get_etheraddr(port->netdev, &ea); if (!error) { return eth_addr_to_uint64(ea); } @@ -6093,8 +7356,8 @@ pick_datapath_id(const struct ofproto *ofproto) static uint64_t pick_fallback_dpid(void) { - uint8_t ea[ETH_ADDR_LEN]; - eth_addr_nicira_random(ea); + struct eth_addr ea; + eth_addr_nicira_random(&ea); return eth_addr_to_uint64(ea); } @@ -6111,7 +7374,7 @@ choose_rule_to_evict(struct oftable *table, struct rule **rulep) struct eviction_group *evg; *rulep = NULL; - if (!table->eviction_fields) { + if (!table->eviction) { return false; } @@ -6216,7 +7479,7 @@ eviction_group_hash_rule(struct rule *rule) uint32_t hash; hash = table->eviction_group_id_basis; - miniflow_expand(&rule->cr.match.flow, &flow); + miniflow_expand(rule->cr.match.flow, &flow); for (sf = table->eviction_fields; sf < &table->eviction_fields[table->n_eviction_fields]; sf++) @@ -6264,23 +7527,21 @@ eviction_group_find(struct oftable *table, uint32_t id) } /* Returns an eviction priority for 'rule'. The return value should be - * interpreted so that higher priorities make a rule more attractive candidates - * for eviction. - * Called only if have a timeout. */ -static uint32_t + * interpreted so that higher priorities make a rule a more attractive + * candidate for eviction. */ +static uint64_t rule_eviction_priority(struct ofproto *ofproto, struct rule *rule) OVS_REQUIRES(ofproto_mutex) { + /* Calculate absolute time when this flow will expire. If it will never + * expire, then return 0 to make it unevictable. */ long long int expiration = LLONG_MAX; - long long int modified; - uint32_t expiration_offset; - - /* 'modified' needs protection even when we hold 'ofproto_mutex'. */ - ovs_mutex_lock(&rule->mutex); - modified = rule->modified; - ovs_mutex_unlock(&rule->mutex); - if (rule->hard_timeout) { + /* 'modified' needs protection even when we hold 'ofproto_mutex'. */ + ovs_mutex_lock(&rule->mutex); + long long int modified = rule->modified; + ovs_mutex_unlock(&rule->mutex); + expiration = modified + rule->hard_timeout * 1000; } if (rule->idle_timeout) { @@ -6292,7 +7553,6 @@ rule_eviction_priority(struct ofproto *ofproto, struct rule *rule) idle_expiration = used + rule->idle_timeout * 1000; expiration = MIN(expiration, idle_expiration); } - if (expiration == LLONG_MAX) { return 0; } @@ -6302,10 +7562,19 @@ rule_eviction_priority(struct ofproto *ofproto, struct rule *rule) * * This should work OK for program runs that last UINT32_MAX seconds or * less. Therefore, please restart OVS at least once every 136 years. */ - expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10); + uint32_t expiration_ofs = (expiration >> 10) - (time_boot_msec() >> 10); - /* Invert the expiration offset because we're using a max-heap. */ - return UINT32_MAX - expiration_offset; + /* Combine expiration time with OpenFlow "importance" to form a single + * priority value. We want flows with relatively low "importance" to be + * evicted before even considering expiration time, so put "importance" in + * the most significant bits and expiration time in the least significant + * bits. + * + * Small 'priority' should be evicted before those with large 'priority'. + * The caller expects the opposite convention (a large return value being + * more attractive for eviction) so we invert it before returning. */ + uint64_t priority = ((uint64_t) rule->importance << 32) + expiration_ofs; + return UINT64_MAX - priority; } /* Adds 'rule' to an appropriate eviction group for its oftable's @@ -6326,7 +7595,7 @@ eviction_group_add_rule(struct rule *rule) * so no additional protection is needed. */ has_timeout = rule->hard_timeout || rule->idle_timeout; - if (table->eviction_fields && has_timeout) { + if (table->eviction && has_timeout) { struct eviction_group *evg; evg = eviction_group_find(table, eviction_group_hash_rule(rule)); @@ -6345,9 +7614,12 @@ static void oftable_init(struct oftable *table) { memset(table, 0, sizeof *table); - classifier_init(&table->cls, flow_segment_u32s); + classifier_init(&table->cls, flow_segment_u64s); table->max_flows = UINT_MAX; - atomic_init(&table->config, (unsigned int)OFPROTO_TABLE_MISS_DEFAULT); + table->n_flows = 0; + hmap_init(&table->eviction_groups_by_id); + heap_init(&table->eviction_groups_by_size); + atomic_init(&table->miss_config, OFPUTIL_TABLE_MISS_DEFAULT); classifier_set_prefix_fields(&table->cls, default_prefix_fields, ARRAY_SIZE(default_prefix_fields)); @@ -6363,7 +7635,13 @@ static void oftable_destroy(struct oftable *table) { ovs_assert(classifier_is_empty(&table->cls)); - oftable_disable_eviction(table); + + ovs_mutex_lock(&ofproto_mutex); + oftable_configure_eviction(table, 0, NULL, 0); + ovs_mutex_unlock(&ofproto_mutex); + + hmap_destroy(&table->eviction_groups_by_id); + heap_destroy(&table->eviction_groups_by_size); classifier_destroy(&table->cls); free(table->name); } @@ -6388,30 +7666,6 @@ oftable_set_name(struct oftable *table, const char *name) } } -/* oftables support a choice of two policies when adding a rule would cause the - * number of flows in the table to exceed the configured maximum number: either - * they can refuse to add the new flow or they can evict some existing flow. - * This function configures the former policy on 'table'. */ -static void -oftable_disable_eviction(struct oftable *table) - OVS_REQUIRES(ofproto_mutex) -{ - if (table->eviction_fields) { - struct eviction_group *evg, *next; - - HMAP_FOR_EACH_SAFE (evg, next, id_node, - &table->eviction_groups_by_id) { - eviction_group_destroy(table, evg); - } - hmap_destroy(&table->eviction_groups_by_id); - heap_destroy(&table->eviction_groups_by_size); - - free(table->eviction_fields); - table->eviction_fields = NULL; - table->n_eviction_fields = 0; - } -} - /* oftables support a choice of two policies when adding a rule would cause the * number of flows in the table to exceed the configured maximum number: either * they can refuse to add the new flow or they can evict some existing flow. @@ -6419,43 +7673,83 @@ oftable_disable_eviction(struct oftable *table) * on the values of the 'n_fields' fields specified in 'fields'. (Specifying * 'n_fields' as 0 disables fairness.) */ static void -oftable_enable_eviction(struct oftable *table, - const struct mf_subfield *fields, size_t n_fields) +oftable_configure_eviction(struct oftable *table, unsigned int eviction, + const struct mf_subfield *fields, size_t n_fields) OVS_REQUIRES(ofproto_mutex) { struct rule *rule; - if (table->eviction_fields + if ((table->eviction != 0) == (eviction != 0) && n_fields == table->n_eviction_fields && (!n_fields || !memcmp(fields, table->eviction_fields, n_fields * sizeof *fields))) { - /* No change. */ + /* The set of eviction fields did not change. If 'eviction' changed, + * it remains nonzero, so that we can just update table->eviction + * without fussing with the eviction groups. */ + table->eviction = eviction; return; } - oftable_disable_eviction(table); - - table->n_eviction_fields = n_fields; - table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields); - - table->eviction_group_id_basis = random_uint32(); + /* Destroy existing eviction groups, then destroy and recreate data + * structures to recover memory. */ + struct eviction_group *evg, *next; + HMAP_FOR_EACH_SAFE (evg, next, id_node, &table->eviction_groups_by_id) { + eviction_group_destroy(table, evg); + } + hmap_destroy(&table->eviction_groups_by_id); hmap_init(&table->eviction_groups_by_id); + heap_destroy(&table->eviction_groups_by_size); heap_init(&table->eviction_groups_by_size); - CLS_FOR_EACH (rule, cr, &table->cls) { - eviction_group_add_rule(rule); + /* Replace eviction groups by the new ones, if there is a change. Free the + * old fields only after allocating the new ones, because 'fields == + * table->eviction_fields' is possible. */ + struct mf_subfield *old_fields = table->eviction_fields; + table->n_eviction_fields = n_fields; + table->eviction_fields = (fields + ? xmemdup(fields, n_fields * sizeof *fields) + : NULL); + free(old_fields); + + /* Add the new eviction groups, if enabled. */ + table->eviction = eviction; + if (table->eviction) { + table->eviction_group_id_basis = random_uint32(); + CLS_FOR_EACH (rule, cr, &table->cls) { + eviction_group_add_rule(rule); + } } } -/* Removes 'rule' from the oftable that contains it. */ +/* Inserts 'rule' from the ofproto data structures BEFORE caller has inserted + * it to the classifier. */ static void -oftable_remove_rule__(struct ofproto *ofproto, struct rule *rule) +ofproto_rule_insert__(struct ofproto *ofproto, struct rule *rule) OVS_REQUIRES(ofproto_mutex) { - struct classifier *cls = &ofproto->tables[rule->table_id].cls; + const struct rule_actions *actions = rule_get_actions(rule); - classifier_remove(cls, CONST_CAST(struct cls_rule *, &rule->cr)); + ovs_assert(rule->removed); + + if (rule->hard_timeout || rule->idle_timeout) { + list_insert(&ofproto->expirable, &rule->expirable); + } + cookies_insert(ofproto, rule); + eviction_group_add_rule(rule); + if (actions->has_meter) { + meter_insert_rule(rule); + } + rule->removed = false; +} + +/* Removes 'rule' from the ofproto data structures. Caller may have deferred + * the removal from the classifier. */ +static void +ofproto_rule_remove__(struct ofproto *ofproto, struct rule *rule) + OVS_REQUIRES(ofproto_mutex) +{ + ovs_assert(!rule->removed); cookies_remove(ofproto, rule); @@ -6467,13 +7761,8 @@ oftable_remove_rule__(struct ofproto *ofproto, struct rule *rule) list_remove(&rule->meter_list_node); list_init(&rule->meter_list_node); } -} -static void -oftable_remove_rule(struct rule *rule) - OVS_REQUIRES(ofproto_mutex) -{ - oftable_remove_rule__(rule->ofproto, rule); + rule->removed = true; } /* unixctl commands. */ @@ -6547,9 +7836,10 @@ ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap) OFPROTO_FOR_EACH_TABLE (oftable, ofproto) { struct rule *rule; - CLS_FOR_EACH_TARGET (rule, cr, &oftable->cls, &target) { - if (minimask_get_vid_mask(&rule->cr.match.mask) == VLAN_VID_MASK) { - uint16_t vid = miniflow_get_vid(&rule->cr.match.flow); + CLS_FOR_EACH_TARGET (rule, cr, &oftable->cls, &target, + CLS_MAX_VERSION) { + if (minimask_get_vid_mask(rule->cr.match.mask) == VLAN_VID_MASK) { + uint16_t vid = miniflow_get_vid(rule->cr.match.flow); bitmap_set1(vlan_bitmap, vid); bitmap_set1(ofproto->vlan_bitmap, vid);