X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=ofproto%2Fofproto.c;h=f84510fabe8c75bb7ad02f91946c377bd750cfa4;hb=97f63e57a8f31952107e23d164b9347eda2ddc98;hp=0625ccfa196e2b173e6aae6b939d20af250c733e;hpb=13b1b2ae700284e8a752ced7c87e16a7c8f9d76c;p=sliver-openvswitch.git diff --git a/ofproto/ofproto.c b/ofproto/ofproto.c index 0625ccfa1..f84510fab 100644 --- a/ofproto/ofproto.c +++ b/ofproto/ofproto.c @@ -21,6 +21,7 @@ #include #include #include +#include #include "bitmap.h" #include "byte-order.h" #include "classifier.h" @@ -75,7 +76,8 @@ enum ofproto_state { enum ofoperation_type { OFOPERATION_ADD, OFOPERATION_DELETE, - OFOPERATION_MODIFY + OFOPERATION_MODIFY, + OFOPERATION_REPLACE }; /* A single OpenFlow request can execute any number of operations. The @@ -120,10 +122,8 @@ struct ofoperation { struct rule *rule; /* Rule being operated upon. */ enum ofoperation_type type; /* Type of operation. */ - /* OFOPERATION_ADD. */ - struct rule *victim; /* Rule being replaced, if any.. */ - - /* OFOPERATION_MODIFY: The old actions, if the actions are changing. */ + /* OFOPERATION_MODIFY, OFOPERATION_REPLACE: The old actions, if the actions + * are changing. */ struct ofpact *ofpacts; size_t ofpacts_len; uint32_t meter_id; @@ -131,8 +131,11 @@ struct ofoperation { /* OFOPERATION_DELETE. */ enum ofp_flow_removed_reason reason; /* Reason flow was removed. */ - ovs_be64 flow_cookie; /* Rule's old flow cookie. */ - enum ofperr error; /* 0 if no error. */ + ovs_be64 flow_cookie; /* Rule's old flow cookie. */ + uint16_t idle_timeout; /* Rule's old idle timeout. */ + uint16_t hard_timeout; /* Rule's old hard timeout. */ + enum ofputil_flow_mod_flags flags; /* Rule's old flags. */ + enum ofperr error; /* 0 if no error. */ }; static struct ofoperation *ofoperation_create(struct ofopgroup *, @@ -152,9 +155,11 @@ static void oftable_enable_eviction(struct oftable *, const struct mf_subfield *fields, size_t n_fields); -static void oftable_remove_rule(struct rule *); -static struct rule *oftable_replace_rule(struct rule *); -static void oftable_substitute_rule(struct rule *old, struct rule *new); +static void oftable_remove_rule(struct rule *rule) OVS_RELEASES(rule->rwlock); +static void oftable_remove_rule__(struct ofproto *ofproto, + struct classifier *cls, struct rule *rule) + OVS_REQ_WRLOCK(cls->rwlock) OVS_RELEASES(rule->rwlock); +static void oftable_insert_rule(struct rule *); /* A set of rules within a single OpenFlow table (oftable) that have the same * values for the oftable's eviction_fields. A rule to be evicted, when one is @@ -178,9 +183,44 @@ struct eviction_group { struct heap rules; /* Contains "struct rule"s. */ }; -static struct rule *choose_rule_to_evict(struct oftable *); +static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep) + OVS_TRY_WRLOCK(true, (*rulep)->rwlock); static void ofproto_evict(struct ofproto *); static uint32_t rule_eviction_priority(struct rule *); +static void eviction_group_add_rule(struct rule *); +static void eviction_group_remove_rule(struct rule *); + +/* Criteria that flow_mod and other operations use for selecting rules on + * which to operate. */ +struct rule_criteria { + /* An OpenFlow table or 255 for all tables. */ + uint8_t table_id; + + /* OpenFlow matching criteria. Interpreted different in "loose" way by + * collect_rules_loose() and "strict" way by collect_rules_strict(), as + * defined in the OpenFlow spec. */ + struct cls_rule cr; + + /* 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'. + * The rule will not be selected if M is 1 and B != C. */ + ovs_be64 cookie; + ovs_be64 cookie_mask; + + /* Selection based on actions within a rule: + * + * If out_port != OFPP_ANY, selects only rules that output to out_port. + * If out_group != OFPG_ALL, select only rules that output to out_group. */ + ofp_port_t out_port; + uint32_t out_group; +}; + +static void rule_criteria_init(struct rule_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); +static void rule_criteria_destroy(struct rule_criteria *); /* ofport. */ static void ofport_destroy__(struct ofport *); @@ -191,6 +231,7 @@ static int init_ports(struct ofproto *); static void reinit_ports(struct ofproto *); /* rule. */ +static void ofproto_rule_destroy(struct rule *); static void ofproto_rule_destroy__(struct rule *); static void ofproto_rule_send_removed(struct rule *, uint8_t reason); static bool rule_is_modifiable(const struct rule *); @@ -199,8 +240,14 @@ static bool rule_is_modifiable(const struct rule *); static enum ofperr add_flow(struct ofproto *, struct ofconn *, struct ofputil_flow_mod *, const struct ofp_header *); -static void delete_flow__(struct rule *, struct ofopgroup *, - enum ofp_flow_removed_reason); +static enum ofperr modify_flows__(struct ofproto *, struct ofconn *, + struct ofputil_flow_mod *, + const struct ofp_header *, + const struct rule_collection *); +static void delete_flow__(struct rule *rule, struct ofopgroup *, + enum ofp_flow_removed_reason) + OVS_RELEASES(rule->rwlock); +static enum ofperr add_group(struct ofproto *, struct ofputil_group_mod *); static bool handle_openflow(struct ofconn *, const struct ofpbuf *); static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *, struct ofputil_flow_mod *, @@ -224,6 +271,7 @@ static size_t n_ofproto_classes; static size_t allocated_ofproto_classes; unsigned flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT; +unsigned n_handler_threads; enum ofproto_flow_miss_model flow_miss_model = OFPROTO_HANDLE_MISS_AUTO; /* Map from datapath name to struct ofproto, for use by unixctl commands. */ @@ -425,11 +473,13 @@ ofproto_create(const char *datapath_name, const char *datapath_type, hmap_init(&ofproto->ports); shash_init(&ofproto->port_by_name); simap_init(&ofproto->ofp_requests); - ofproto->max_ports = OFPP_MAX; + ofproto->max_ports = ofp_to_u16(OFPP_MAX); + ofproto->eviction_group_timer = LLONG_MIN; ofproto->tables = NULL; ofproto->n_tables = 0; hindex_init(&ofproto->cookies); list_init(&ofproto->expirable); + ovs_mutex_init_recursive(&ofproto->expirable_mutex); ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name); ofproto->state = S_OPENFLOW; list_init(&ofproto->pending); @@ -442,6 +492,8 @@ ofproto_create(const char *datapath_name, const char *datapath_type, ofproto->vlan_bitmap = NULL; ofproto->vlans_changed = false; ofproto->min_mtu = INT_MAX; + ovs_rwlock_init(&ofproto->groups_rwlock); + hmap_init(&ofproto->groups); error = ofproto->ofproto_class->construct(ofproto); if (error) { @@ -453,7 +505,7 @@ ofproto_create(const char *datapath_name, const char *datapath_type, /* The "max_ports" member should have been set by ->construct(ofproto). * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */ - ofproto->ofp_port_ids = bitmap_allocate(ofp_to_u16(ofproto->max_ports)); + ofproto->ofp_port_ids = bitmap_allocate(ofproto->max_ports); bitmap_set1(ofproto->ofp_port_ids, 0); /* Check that hidden tables, if any, are at the end. */ @@ -512,9 +564,9 @@ ofproto_init_tables(struct ofproto *ofproto, int n_tables) * Reserved ports numbered OFPP_MAX and higher are special and not subject to * the 'max_ports' restriction. */ void -ofproto_init_max_ports(struct ofproto *ofproto, ofp_port_t max_ports) +ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports) { - ovs_assert(ofp_to_u16(max_ports) <= ofp_to_u16(OFPP_MAX)); + ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX)); ofproto->max_ports = max_ports; } @@ -622,6 +674,19 @@ ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time, } } +/* Sets number of upcall handler threads. The default is + * (number of online cores - 2). */ +void +ofproto_set_n_handler_threads(unsigned limit) +{ + if (limit) { + n_handler_threads = limit; + } else { + int n_proc = sysconf(_SC_NPROCESSORS_ONLN); + n_handler_threads = n_proc > 2 ? n_proc - 2 : 1; + } +} + void ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc) { @@ -1017,6 +1082,7 @@ ofproto_configure_table(struct ofproto *ofproto, int table_id, } table->max_flows = s->max_flows; + ovs_rwlock_rdlock(&table->cls.rwlock); if (classifier_count(&table->cls) > table->max_flows && table->eviction_fields) { /* 'table' contains more flows than allowed. We might not be able to @@ -1032,6 +1098,7 @@ ofproto_configure_table(struct ofproto *ofproto, int table_id, break; } } + ovs_rwlock_unlock(&table->cls.rwlock); } bool @@ -1046,17 +1113,47 @@ ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops) connmgr_get_snoops(ofproto->connmgr, snoops); } +/* Deletes 'rule' from 'cls' within 'ofproto'. + * + * Within an ofproto implementation, this function allows an ofproto + * implementation to destroy any rules that remain when its ->destruct() + * function is called. This function is not suitable for use elsewhere in an + * ofproto implementation. + * + * This function is also used internally in ofproto.c. + * + * This function implements steps 4.4 and 4.5 in the section titled "Rule Life + * Cycle" in ofproto-provider.h. + + * The 'cls' argument is redundant (it is &ofproto->tables[rule->table_id].cls) + * but it allows Clang to do better checking. */ +void +ofproto_rule_delete(struct ofproto *ofproto, struct classifier *cls, + struct rule *rule) + OVS_REQ_WRLOCK(cls->rwlock) +{ + struct ofopgroup *group; + + ovs_assert(!rule->pending); + ovs_assert(cls == &ofproto->tables[rule->table_id].cls); + + group = ofopgroup_create_unattached(ofproto); + ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE); + ovs_rwlock_wrlock(&rule->rwlock); + oftable_remove_rule__(ofproto, cls, rule); + ofproto->ofproto_class->rule_delete(rule); + ofopgroup_submit(group); +} + static void ofproto_flush__(struct ofproto *ofproto) { - struct ofopgroup *group; struct oftable *table; if (ofproto->ofproto_class->flush) { ofproto->ofproto_class->flush(ofproto); } - group = ofopgroup_create_unattached(ofproto); OFPROTO_FOR_EACH_TABLE (table, ofproto) { struct rule *rule, *next_rule; struct cls_cursor cursor; @@ -1065,19 +1162,19 @@ ofproto_flush__(struct ofproto *ofproto) continue; } + ovs_rwlock_wrlock(&table->cls.rwlock); cls_cursor_init(&cursor, &table->cls, NULL); CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { if (!rule->pending) { - ofoperation_create(group, rule, OFOPERATION_DELETE, - OFPRR_DELETE); - oftable_remove_rule(rule); - ofproto->ofproto_class->rule_destruct(rule); + ofproto_rule_delete(ofproto, &table->cls, rule); } } + ovs_rwlock_unlock(&table->cls.rwlock); } - ofopgroup_submit(group); } +static void delete_group(struct ofproto *ofproto, uint32_t group_id); + static void ofproto_destroy__(struct ofproto *ofproto) { @@ -1086,10 +1183,9 @@ ofproto_destroy__(struct ofproto *ofproto) ovs_assert(list_is_empty(&ofproto->pending)); ovs_assert(!ofproto->n_pending); - if (ofproto->meters) { - meter_delete(ofproto, 1, ofproto->meter_features.max_meters); - free(ofproto->meters); - } + delete_group(ofproto, OFPG_ALL); + ovs_rwlock_destroy(&ofproto->groups_rwlock); + hmap_destroy(&ofproto->groups); connmgr_destroy(ofproto->connmgr); @@ -1115,6 +1211,7 @@ ofproto_destroy__(struct ofproto *ofproto) free(ofproto->vlan_bitmap); + ovs_mutex_destroy(&ofproto->expirable_mutex); ofproto->ofproto_class->dealloc(ofproto); } @@ -1127,6 +1224,13 @@ ofproto_destroy(struct ofproto *p) return; } + if (p->meters) { + meter_delete(p, 1, p->meter_features.max_meters); + p->meter_features.max_meters = 0; + free(p->meters); + p->meters = NULL; + } + ofproto_flush__(p); HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) { ofport_destroy(ofport); @@ -1222,6 +1326,38 @@ ofproto_run(struct ofproto *p) VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error)); } + /* Restore the eviction group heap invariant occasionally. */ + if (p->eviction_group_timer < time_msec()) { + size_t i; + + p->eviction_group_timer = time_msec() + 1000; + + for (i = 0; i < p->n_tables; i++) { + struct oftable *table = &p->tables[i]; + struct eviction_group *evg; + struct cls_cursor cursor; + struct rule *rule; + + if (!table->eviction_fields) { + continue; + } + + HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) { + heap_rebuild(&evg->rules); + } + + ovs_rwlock_rdlock(&table->cls.rwlock); + cls_cursor_init(&cursor, &table->cls, NULL); + CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { + if (!rule->eviction_group + && (rule->idle_timeout || rule->hard_timeout)) { + eviction_group_add_rule(rule); + } + } + ovs_rwlock_unlock(&table->cls.rwlock); + } + } + if (p->ofproto_class->port_poll) { char *devname; @@ -1385,7 +1521,9 @@ ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage) n_rules = 0; OFPROTO_FOR_EACH_TABLE (table, ofproto) { + ovs_rwlock_rdlock(&table->cls.rwlock); n_rules += classifier_count(&table->cls); + ovs_rwlock_unlock(&table->cls.rwlock); } simap_increase(usage, "rules", n_rules); @@ -1594,6 +1732,33 @@ ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port) return error; } +static int +simple_flow_mod(struct ofproto *ofproto, + const struct match *match, unsigned int priority, + const struct ofpact *ofpacts, size_t ofpacts_len, + enum ofp_flow_mod_command command) +{ + struct ofputil_flow_mod fm; + + 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; + return handle_flow_mod__(ofproto, NULL, &fm, NULL); +} + /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and * performs the 'n_actions' actions in 'actions'. The new flow will not * timeout. @@ -1612,20 +1777,21 @@ ofproto_add_flow(struct ofproto *ofproto, const struct match *match, { const struct rule *rule; + /* 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. */ + ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock); rule = rule_from_cls_rule(classifier_find_match_exactly( &ofproto->tables[0].cls, match, priority)); + ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock); + + /* If there's no such rule or the rule doesn't have the actions we want, + * fall back to a executing a full flow mod. We can't optimize this at + * all because we didn't take enough locks above to ensure that the flow + * table didn't already change beneath us. */ if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len, ofpacts, ofpacts_len)) { - struct ofputil_flow_mod fm; - - memset(&fm, 0, sizeof fm); - fm.match = *match; - fm.priority = priority; - fm.buffer_id = UINT32_MAX; - fm.ofpacts = xmemdup(ofpacts, ofpacts_len); - fm.ofpacts_len = ofpacts_len; - add_flow(ofproto, NULL, &fm, NULL); - free(fm.ofpacts); + simple_flow_mod(ofproto, match, priority, ofpacts, ofpacts_len, + OFPFC_MODIFY_STRICT); } } @@ -1648,27 +1814,24 @@ bool ofproto_delete_flow(struct ofproto *ofproto, const struct match *target, unsigned int priority) { + struct classifier *cls = &ofproto->tables[0].cls; struct rule *rule; - rule = rule_from_cls_rule(classifier_find_match_exactly( - &ofproto->tables[0].cls, target, priority)); + /* First do a cheap check whether the rule we're looking for has already + * been deleted. If so, then we're done. */ + ovs_rwlock_rdlock(&cls->rwlock); + rule = rule_from_cls_rule(classifier_find_match_exactly(cls, target, + priority)); + ovs_rwlock_unlock(&cls->rwlock); if (!rule) { - /* No such rule -> success. */ - return true; - } else if (rule->pending) { - /* An operation on the rule is already pending -> failure. - * Caller must retry later if it's important. */ - return false; - } else { - /* Initiate deletion -> success. */ - struct ofopgroup *group = ofopgroup_create_unattached(ofproto); - ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE); - oftable_remove_rule(rule); - ofproto->ofproto_class->rule_destruct(rule); - ofopgroup_submit(group); return true; } + /* Fall back to a executing a full flow mod. We can't optimize this at all + * because we didn't take enough locks above to ensure that the flow table + * didn't already change beneath us. */ + return simple_flow_mod(ofproto, target, priority, NULL, 0, + OFPFC_DELETE_STRICT) != OFPROTO_POSTPONE; } /* Starts the process of deleting all of the flows from all of ofproto's flow @@ -1709,32 +1872,28 @@ reinit_ports(struct ofproto *p) static ofp_port_t alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name) { - uint16_t max_ports = ofp_to_u16(ofproto->max_ports); uint16_t port_idx; port_idx = simap_get(&ofproto->ofp_requests, netdev_name); - if (!port_idx) { - port_idx = UINT16_MAX; - } + port_idx = port_idx ? port_idx : UINT16_MAX; - if (port_idx >= max_ports + if (port_idx >= ofproto->max_ports || bitmap_is_set(ofproto->ofp_port_ids, port_idx)) { - uint16_t end_port_no = ofp_to_u16(ofproto->alloc_port_no); - uint16_t alloc_port_no = end_port_no; + uint16_t end_port_no = ofproto->alloc_port_no; /* Search for a free OpenFlow port number. We try not to * immediately reuse them to prevent problems due to old * flows. */ for (;;) { - if (++alloc_port_no >= max_ports) { - alloc_port_no = 0; + if (++ofproto->alloc_port_no >= ofproto->max_ports) { + ofproto->alloc_port_no = 0; } - if (!bitmap_is_set(ofproto->ofp_port_ids, alloc_port_no)) { - port_idx = alloc_port_no; - ofproto->alloc_port_no = u16_to_ofp(alloc_port_no); + if (!bitmap_is_set(ofproto->ofp_port_ids, + ofproto->alloc_port_no)) { + port_idx = ofproto->alloc_port_no; break; } - if (alloc_port_no == end_port_no) { + if (ofproto->alloc_port_no == end_port_no) { return OFPP_NONE; } } @@ -1746,7 +1905,7 @@ alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name) static void dealloc_ofp_port(const struct ofproto *ofproto, ofp_port_t ofp_port) { - if (ofp_to_u16(ofp_port) < ofp_to_u16(ofproto->max_ports)) { + if (ofp_to_u16(ofp_port) < ofproto->max_ports) { bitmap_set0(ofproto->ofp_port_ids, ofp_to_u16(ofp_port)); } } @@ -1789,8 +1948,8 @@ ofport_open(struct ofproto *ofproto, pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN; netdev_get_features(netdev, &pp->curr, &pp->advertised, &pp->supported, &pp->peer); - pp->curr_speed = netdev_features_to_bps(pp->curr, 0); - pp->max_speed = netdev_features_to_bps(pp->supported, 0); + pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000; + pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000; return netdev; } @@ -2070,6 +2229,10 @@ init_ports(struct ofproto *p) netdev = ofport_open(p, &ofproto_port, &pp); if (netdev) { ofport_install(p, netdev, &pp); + if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) { + p->alloc_port_no = MAX(p->alloc_port_no, + ofp_to_u16(ofproto_port.ofp_port)); + } } } } @@ -2160,29 +2323,22 @@ update_mtu(struct ofproto *p, struct ofport *port) } static void -ofproto_rule_destroy__(struct rule *rule) +ofproto_rule_destroy(struct rule *rule) { if (rule) { - cls_rule_destroy(&rule->cr); - free(rule->ofpacts); - rule->ofproto->ofproto_class->rule_dealloc(rule); + rule->ofproto->ofproto_class->rule_destruct(rule); + ofproto_rule_destroy__(rule); } } -/* This function allows an ofproto implementation to destroy any rules that - * remain when its ->destruct() function is called. The caller must have - * already uninitialized any derived members of 'rule' (step 5 described in the - * large comment in ofproto/ofproto-provider.h titled "Life Cycle"). - * This function implements steps 6 and 7. - * - * This function should only be called from an ofproto implementation's - * ->destruct() function. It is not suitable elsewhere. */ -void -ofproto_rule_destroy(struct rule *rule) +static void +ofproto_rule_destroy__(struct rule *rule) { - ovs_assert(!rule->pending); - oftable_remove_rule(rule); - ofproto_rule_destroy__(rule); + cls_rule_destroy(&rule->cr); + free(rule->ofpacts); + ovs_mutex_destroy(&rule->timeout_mutex); + ovs_rwlock_destroy(&rule->rwlock); + rule->ofproto->ofproto_class->rule_dealloc(rule); } /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action @@ -2194,6 +2350,14 @@ ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port) || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port)); } +/* Returns true if 'rule' has group and equals group_id. */ +bool +ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id) +{ + return (group_id == OFPG11_ANY + || ofpacts_output_to_group(rule->ofpacts, rule->ofpacts_len, group_id)); +} + /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or * OFPAT_ENQUEUE action that outputs to 'out_port'. */ bool @@ -2205,12 +2369,11 @@ ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port) switch (op->type) { case OFOPERATION_ADD: - return op->victim && ofproto_rule_has_out_port(op->victim, out_port); - case OFOPERATION_DELETE: return false; case OFOPERATION_MODIFY: + case OFOPERATION_REPLACE: return ofpacts_output_to_port(op->ofpacts, op->ofpacts_len, out_port); } @@ -2218,8 +2381,7 @@ ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port) } /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s - * statistics appropriately. 'packet' must have at least sizeof(struct - * ofp10_packet_in) bytes of headroom. + * statistics appropriately. * * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited * with statistics for 'packet' either way. @@ -2231,8 +2393,6 @@ rule_execute(struct rule *rule, ofp_port_t in_port, struct ofpbuf *packet) struct flow flow; union flow_in_port in_port_; - ovs_assert(ofpbuf_headroom(packet) >= sizeof(struct ofp10_packet_in)); - in_port_.ofp_port = in_port; flow_extract(packet, 0, 0, NULL, &in_port_, &flow); return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet); @@ -2386,30 +2546,6 @@ reject_slave_controller(struct ofconn *ofconn) } } -/* Finds the OFPACT_METER action, if any, in the 'ofpacts_len' bytes of - * 'ofpacts'. If found, returns its meter ID; if not, returns 0. - * - * This function relies on the order of 'ofpacts' being correct (as checked by - * ofpacts_verify()). */ -static uint32_t -find_meter(const struct ofpact ofpacts[], size_t ofpacts_len) -{ - const struct ofpact *a; - - OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) { - enum ovs_instruction_type inst; - - inst = ovs_instruction_type_from_ofpact_type(a->type); - if (a->type == OFPACT_METER) { - return ofpact_get_METER(a)->meter_id; - } else if (inst > OVSINST_OFPIT13_METER) { - break; - } - } - - return 0; -} - /* Checks that the 'ofpacts_len' bytes of actions in 'ofpacts' are appropriate * for a packet with the prerequisites satisfied by 'flow' in table 'table_id'. * 'flow' may be temporarily modified, but is restored at return. @@ -2422,13 +2558,13 @@ ofproto_check_ofpacts(struct ofproto *ofproto, enum ofperr error; uint32_t mid; - error = ofpacts_check(ofpacts, ofpacts_len, flow, ofproto->max_ports, - table_id); + error = ofpacts_check(ofpacts, ofpacts_len, flow, + u16_to_ofp(ofproto->max_ports), table_id); if (error) { return error; } - mid = find_meter(ofpacts, ofpacts_len); + mid = ofpacts_get_meter(ofpacts, ofpacts_len); if (mid && ofproto_get_provider_meter_id(ofproto, mid) == UINT32_MAX) { return OFPERR_OFPMMFC_INVALID_METER; } @@ -2460,7 +2596,7 @@ handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh) if (error) { goto exit_free_ofpacts; } - if (ofp_to_u16(po.in_port) >= ofp_to_u16(p->max_ports) + if (ofp_to_u16(po.in_port) >= p->max_ports && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) { error = OFPERR_OFPBRC_BAD_PORT; goto exit_free_ofpacts; @@ -2474,8 +2610,8 @@ handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh) goto exit_free_ofpacts; } } else { - payload = xmalloc(sizeof *payload); - ofpbuf_use_const(payload, po.packet, po.packet_len); + /* Ensure that the L3 header is 32-bit aligned. */ + payload = ofpbuf_clone_data_with_headroom(po.packet, po.packet_len, 2); } /* Verify actions against packet, then send packet if successful. */ @@ -2613,7 +2749,9 @@ handle_table_stats_request(struct ofconn *ofconn, ots[i].instructions = htonl(OFPIT11_ALL); ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK); ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */ + ovs_rwlock_rdlock(&p->tables[i].cls.rwlock); ots[i].active_count = htonl(classifier_count(&p->tables[i].cls)); + ovs_rwlock_unlock(&p->tables[i].cls.rwlock); } p->ofproto_class->get_tables(p, ots); @@ -2739,7 +2877,9 @@ ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule, if (new_cookie != rule->flow_cookie) { cookies_remove(ofproto, rule); + ovs_rwlock_wrlock(&rule->rwlock); rule->flow_cookie = new_cookie; + ovs_rwlock_unlock(&rule->rwlock); cookies_insert(ofproto, rule); } @@ -2822,158 +2962,205 @@ next_matching_table(const struct ofproto *ofproto, (TABLE) != NULL; \ (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID)) -/* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if - * 'table_id' is 0xff) that match 'match' in the "loose" way required for - * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list - * 'rules'. +/* Initializes 'criteria' in a straightforward way based on the other + * parameters. * - * If 'out_port' is anything other than OFPP_ANY, then only rules that output - * to 'out_port' are included. + * For "loose" matching, the 'priority' parameter is unimportant and may be + * 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) +{ + criteria->table_id = table_id; + cls_rule_init(&criteria->cr, match, priority); + criteria->cookie = cookie; + criteria->cookie_mask = cookie_mask; + criteria->out_port = out_port; + criteria->out_group = out_group; +} + +static void +rule_criteria_destroy(struct rule_criteria *criteria) +{ + cls_rule_destroy(&criteria->cr); +} + +void +rule_collection_init(struct rule_collection *rules) +{ + rules->rules = rules->stub; + rules->n = 0; + rules->capacity = ARRAY_SIZE(rules->stub); +} + +void +rule_collection_add(struct rule_collection *rules, struct rule *rule) +{ + if (rules->n >= rules->capacity) { + size_t old_size, new_size; + + old_size = rules->capacity * sizeof *rules->rules; + rules->capacity *= 2; + new_size = rules->capacity * sizeof *rules->rules; + + if (rules->rules == rules->stub) { + rules->rules = xmalloc(new_size); + memcpy(rules->rules, rules->stub, old_size); + } else { + rules->rules = xrealloc(rules->rules, new_size); + } + } + + rules->rules[rules->n++] = rule; +} + +void +rule_collection_destroy(struct rule_collection *rules) +{ + if (rules->rules != rules->stub) { + free(rules->rules); + } +} + +static enum ofperr +collect_rule(struct rule *rule, const struct rule_criteria *c, + struct rule_collection *rules) +{ + if (ofproto_rule_is_hidden(rule)) { + return 0; + } else if (rule->pending) { + return OFPROTO_POSTPONE; + } else { + if ((c->table_id == rule->table_id || c->table_id == 0xff) + && 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_collection_add(rules, rule); + } + return 0; + } +} + +/* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches + * on classifiers rules are done in the "loose" way required for OpenFlow + * OFPFC_MODIFY and OFPFC_DELETE requests. Puts the selected rules on list + * 'rules'. * * Hidden rules are always omitted. * * Returns 0 on success, otherwise an OpenFlow error code. */ static enum ofperr -collect_rules_loose(struct ofproto *ofproto, uint8_t table_id, - const struct match *match, - ovs_be64 cookie, ovs_be64 cookie_mask, - ofp_port_t out_port, struct list *rules) +collect_rules_loose(struct ofproto *ofproto, + const struct rule_criteria *criteria, + struct rule_collection *rules) { struct oftable *table; - struct cls_rule cr; enum ofperr error; - error = check_table_id(ofproto, table_id); + rule_collection_init(rules); + + error = check_table_id(ofproto, criteria->table_id); if (error) { - return error; + goto exit; } - list_init(rules); - cls_rule_init(&cr, match, 0); - - if (cookie_mask == htonll(UINT64_MAX)) { + if (criteria->cookie_mask == htonll(UINT64_MAX)) { struct rule *rule; - HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, hash_cookie(cookie), + HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, + hash_cookie(criteria->cookie), &ofproto->cookies) { - if (table_id != rule->table_id && table_id != 0xff) { - continue; - } - if (ofproto_rule_is_hidden(rule)) { - continue; - } - if (cls_rule_is_loose_match(&rule->cr, &cr.match)) { - if (rule->pending) { - error = OFPROTO_POSTPONE; - goto exit; - } - if (rule->flow_cookie == cookie /* Hash collisions possible. */ - && ofproto_rule_has_out_port(rule, out_port)) { - list_push_back(rules, &rule->ofproto_node); + if (cls_rule_is_loose_match(&rule->cr, &criteria->cr.match)) { + error = collect_rule(rule, criteria, rules); + if (error) { + break; } } } - goto exit; - } - - FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) { - struct cls_cursor cursor; - struct rule *rule; + } else { + FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) { + struct cls_cursor cursor; + struct rule *rule; - cls_cursor_init(&cursor, &table->cls, &cr); - CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { - if (rule->pending) { - error = OFPROTO_POSTPONE; - goto exit; - } - if (!ofproto_rule_is_hidden(rule) - && ofproto_rule_has_out_port(rule, out_port) - && !((rule->flow_cookie ^ cookie) & cookie_mask)) { - list_push_back(rules, &rule->ofproto_node); + ovs_rwlock_rdlock(&table->cls.rwlock); + cls_cursor_init(&cursor, &table->cls, &criteria->cr); + CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { + error = collect_rule(rule, criteria, rules); + if (error) { + break; + } } + ovs_rwlock_unlock(&table->cls.rwlock); } } exit: - cls_rule_destroy(&cr); + if (error) { + rule_collection_destroy(rules); + } return error; } -/* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if - * 'table_id' is 0xff) that match 'match' in the "strict" way required for - * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them - * on list 'rules'. - * - * If 'out_port' is anything other than OFPP_ANY, then only rules that output - * to 'out_port' are included. +/* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches + * on classifiers rules are done in the "strict" way required for OpenFlow + * OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests. Puts the selected + * rules on list 'rules'. * * Hidden rules are always omitted. * * Returns 0 on success, otherwise an OpenFlow error code. */ static enum ofperr -collect_rules_strict(struct ofproto *ofproto, uint8_t table_id, - const struct match *match, unsigned int priority, - ovs_be64 cookie, ovs_be64 cookie_mask, - ofp_port_t out_port, struct list *rules) +collect_rules_strict(struct ofproto *ofproto, + const struct rule_criteria *criteria, + struct rule_collection *rules) { struct oftable *table; - struct cls_rule cr; int error; - error = check_table_id(ofproto, table_id); + rule_collection_init(rules); + + error = check_table_id(ofproto, criteria->table_id); if (error) { - return error; + goto exit; } - list_init(rules); - cls_rule_init(&cr, match, priority); - - if (cookie_mask == htonll(UINT64_MAX)) { + if (criteria->cookie_mask == htonll(UINT64_MAX)) { struct rule *rule; - HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, hash_cookie(cookie), + HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, + hash_cookie(criteria->cookie), &ofproto->cookies) { - if (table_id != rule->table_id && table_id != 0xff) { - continue; - } - if (ofproto_rule_is_hidden(rule)) { - continue; - } - if (cls_rule_equal(&rule->cr, &cr)) { - if (rule->pending) { - error = OFPROTO_POSTPONE; - goto exit; - } - if (rule->flow_cookie == cookie /* Hash collisions possible. */ - && ofproto_rule_has_out_port(rule, out_port)) { - list_push_back(rules, &rule->ofproto_node); + if (cls_rule_equal(&rule->cr, &criteria->cr)) { + error = collect_rule(rule, criteria, rules); + if (error) { + break; } } } - goto exit; - } - - FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) { - struct rule *rule; + } else { + FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) { + struct rule *rule; - rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, - &cr)); - if (rule) { - if (rule->pending) { - error = OFPROTO_POSTPONE; - goto exit; - } - if (!ofproto_rule_is_hidden(rule) - && ofproto_rule_has_out_port(rule, out_port) - && !((rule->flow_cookie ^ cookie) & cookie_mask)) { - list_push_back(rules, &rule->ofproto_node); + ovs_rwlock_rdlock(&table->cls.rwlock); + rule = rule_from_cls_rule(classifier_find_rule_exactly( + &table->cls, &criteria->cr)); + ovs_rwlock_unlock(&table->cls.rwlock); + if (rule) { + error = collect_rule(rule, criteria, rules); + if (error) { + break; + } } } } exit: - cls_rule_destroy(&cr); - return 0; + if (error) { + rule_collection_destroy(rules); + } + return error; } /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and @@ -2992,25 +3179,28 @@ handle_flow_stats_request(struct ofconn *ofconn, { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); struct ofputil_flow_stats_request fsr; + struct rule_criteria criteria; + struct rule_collection rules; struct list replies; - struct list rules; - struct rule *rule; enum ofperr error; + size_t i; error = ofputil_decode_flow_stats_request(&fsr, request); if (error) { return error; } - error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match, - fsr.cookie, fsr.cookie_mask, - fsr.out_port, &rules); + rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, fsr.cookie, + fsr.cookie_mask, fsr.out_port, fsr.out_group); + error = collect_rules_loose(ofproto, &criteria, &rules); + rule_criteria_destroy(&criteria); if (error) { return error; } ofpmp_init(&replies, request); - LIST_FOR_EACH (rule, ofproto_node, &rules) { + for (i = 0; i < rules.n; i++) { + struct rule *rule = rules.rules[i]; long long int now = time_msec(); struct ofputil_flow_stats fs; @@ -3019,22 +3209,24 @@ handle_flow_stats_request(struct ofconn *ofconn, fs.cookie = rule->flow_cookie; fs.table_id = rule->table_id; calc_duration(rule->created, now, &fs.duration_sec, &fs.duration_nsec); - fs.idle_timeout = rule->idle_timeout; - fs.hard_timeout = rule->hard_timeout; fs.idle_age = age_secs(now - rule->used); fs.hard_age = age_secs(now - rule->modified); ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count, &fs.byte_count); fs.ofpacts = rule->ofpacts; fs.ofpacts_len = rule->ofpacts_len; - fs.flags = 0; - if (rule->send_flow_removed) { - fs.flags |= OFPFF_SEND_FLOW_REM; - /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS - and OFPFF13_NO_BYT_COUNTS */ - } + + ovs_mutex_lock(&rule->timeout_mutex); + fs.idle_timeout = rule->idle_timeout; + fs.hard_timeout = rule->hard_timeout; + ovs_mutex_unlock(&rule->timeout_mutex); + + fs.flags = rule->flags; + ofputil_append_flow_stats_reply(&fs, &replies); } + rule_collection_destroy(&rules); + ofconn_send_replies(ofconn, &replies); return 0; @@ -3073,10 +3265,12 @@ ofproto_get_all_flows(struct ofproto *p, struct ds *results) struct cls_cursor cursor; struct rule *rule; + ovs_rwlock_rdlock(&table->cls.rwlock); cls_cursor_init(&cursor, &table->cls, NULL); CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { flow_stats_ds(rule, results); } + ovs_rwlock_unlock(&table->cls.rwlock); } } @@ -3113,26 +3307,30 @@ handle_aggregate_stats_request(struct ofconn *ofconn, struct ofputil_flow_stats_request request; struct ofputil_aggregate_stats stats; bool unknown_packets, unknown_bytes; + struct rule_criteria criteria; + struct rule_collection rules; struct ofpbuf *reply; - struct list rules; - struct rule *rule; enum ofperr error; + size_t i; error = ofputil_decode_flow_stats_request(&request, oh); if (error) { return error; } - error = collect_rules_loose(ofproto, request.table_id, &request.match, - request.cookie, request.cookie_mask, - request.out_port, &rules); + rule_criteria_init(&criteria, request.table_id, &request.match, 0, + request.cookie, request.cookie_mask, + request.out_port, request.out_group); + error = collect_rules_loose(ofproto, &criteria, &rules); + rule_criteria_destroy(&criteria); if (error) { return error; } memset(&stats, 0, sizeof stats); unknown_packets = unknown_bytes = false; - LIST_FOR_EACH (rule, ofproto_node, &rules) { + for (i = 0; i < rules.n; i++) { + struct rule *rule = rules.rules[i]; uint64_t packet_count; uint64_t byte_count; @@ -3160,6 +3358,8 @@ handle_aggregate_stats_request(struct ofconn *ofconn, stats.byte_count = UINT64_MAX; } + rule_collection_destroy(&rules); + reply = ofputil_encode_aggregate_stats_reply(&stats, oh); ofconn_send_reply(ofconn, reply); @@ -3284,6 +3484,34 @@ is_flow_deletion_pending(const struct ofproto *ofproto, return false; } +static enum ofperr +evict_rule_from_table(struct ofproto *ofproto, struct oftable *table) +{ + struct rule *rule; + size_t n_rules; + + ovs_rwlock_rdlock(&table->cls.rwlock); + n_rules = classifier_count(&table->cls); + ovs_rwlock_unlock(&table->cls.rwlock); + + if (n_rules < table->max_flows) { + return 0; + } else if (!choose_rule_to_evict(table, &rule)) { + return OFPERR_OFPFMFC_TABLE_FULL; + } else if (rule->pending) { + ovs_rwlock_unlock(&rule->rwlock); + return OFPROTO_POSTPONE; + } else { + struct ofopgroup *group; + + group = ofopgroup_create_unattached(ofproto); + delete_flow__(rule, group, OFPRR_EVICTION); + ofopgroup_submit(group); + + return 0; + } +} + /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT * in which no matching flow already exists in the flow table. * @@ -3292,8 +3520,7 @@ is_flow_deletion_pending(const struct ofproto *ofproto, * error code on failure, or OFPROTO_POSTPONE if the operation cannot be * initiated now but may be retried later. * - * Upon successful return, takes ownership of 'fm->ofpacts'. On failure, - * ownership remains with the caller. + * The caller retains ownership of 'fm->ofpacts'. * * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, * if any. */ @@ -3303,7 +3530,7 @@ add_flow(struct ofproto *ofproto, struct ofconn *ofconn, { struct oftable *table; struct ofopgroup *group; - struct rule *victim; + struct cls_rule cr; struct rule *rule; uint8_t table_id; int error; @@ -3338,109 +3565,117 @@ add_flow(struct ofproto *ofproto, struct ofconn *ofconn, return OFPERR_OFPBRC_EPERM; } + cls_rule_init(&cr, &fm->match, fm->priority); + + /* Transform "add" into "modify" if there's an existing identical flow. */ + ovs_rwlock_rdlock(&table->cls.rwlock); + rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr)); + ovs_rwlock_unlock(&table->cls.rwlock); + if (rule) { + cls_rule_destroy(&cr); + if (!rule_is_modifiable(rule)) { + return OFPERR_OFPBRC_EPERM; + } else if (rule->pending) { + return OFPROTO_POSTPONE; + } else { + struct rule_collection rules; + + rule_collection_init(&rules); + rule_collection_add(&rules, rule); + fm->modify_cookie = true; + error = modify_flows__(ofproto, ofconn, fm, request, &rules); + rule_collection_destroy(&rules); + + return error; + } + } + /* Verify actions. */ error = ofproto_check_ofpacts(ofproto, fm->ofpacts, fm->ofpacts_len, &fm->match.flow, table_id); if (error) { + cls_rule_destroy(&cr); return error; } - /* Allocate new rule and initialize classifier rule. */ - rule = ofproto->ofproto_class->rule_alloc(); - if (!rule) { - VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)", - ofproto->name, ovs_strerror(error)); - return ENOMEM; - } - cls_rule_init(&rule->cr, &fm->match, fm->priority); - /* Serialize against pending deletion. */ - if (is_flow_deletion_pending(ofproto, &rule->cr, table_id)) { - cls_rule_destroy(&rule->cr); - ofproto->ofproto_class->rule_dealloc(rule); + if (is_flow_deletion_pending(ofproto, &cr, table_id)) { + cls_rule_destroy(&cr); return OFPROTO_POSTPONE; } /* Check for overlap, if requested. */ - if (fm->flags & OFPFF_CHECK_OVERLAP - && classifier_rule_overlaps(&table->cls, &rule->cr)) { - cls_rule_destroy(&rule->cr); - ofproto->ofproto_class->rule_dealloc(rule); - return OFPERR_OFPFMFC_OVERLAP; + if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP) { + bool overlaps; + + ovs_rwlock_rdlock(&table->cls.rwlock); + overlaps = classifier_rule_overlaps(&table->cls, &cr); + ovs_rwlock_unlock(&table->cls.rwlock); + + if (overlaps) { + cls_rule_destroy(&cr); + return OFPERR_OFPFMFC_OVERLAP; + } + } + + /* If necessary, evict an existing rule to clear out space. */ + error = evict_rule_from_table(ofproto, table); + if (error) { + cls_rule_destroy(&cr); + return error; } - /* FIXME: Implement OFPFF12_RESET_COUNTS */ + /* Allocate new rule. */ + rule = ofproto->ofproto_class->rule_alloc(); + if (!rule) { + cls_rule_destroy(&cr); + VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)", + ofproto->name, ovs_strerror(error)); + return ENOMEM; + } + /* Initialize base state. */ rule->ofproto = ofproto; + cls_rule_move(&rule->cr, &cr); rule->pending = NULL; rule->flow_cookie = fm->new_cookie; rule->created = rule->modified = rule->used = time_msec(); + + ovs_mutex_init(&rule->timeout_mutex); + ovs_mutex_lock(&rule->timeout_mutex); rule->idle_timeout = fm->idle_timeout; rule->hard_timeout = fm->hard_timeout; + ovs_mutex_unlock(&rule->timeout_mutex); + rule->table_id = table - ofproto->tables; - rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0; - /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS - and OFPFF13_NO_BYT_COUNTS */ + rule->flags = fm->flags & OFPUTIL_FF_STATE; + rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len); rule->ofpacts_len = fm->ofpacts_len; - rule->meter_id = find_meter(rule->ofpacts, rule->ofpacts_len); + rule->meter_id = ofpacts_get_meter(rule->ofpacts, rule->ofpacts_len); list_init(&rule->meter_list_node); - rule->evictable = true; rule->eviction_group = NULL; list_init(&rule->expirable); rule->monitor_flags = 0; rule->add_seqno = 0; rule->modify_seqno = 0; + ovs_rwlock_init(&rule->rwlock); - /* Insert new rule. */ - victim = oftable_replace_rule(rule); - if (victim && !rule_is_modifiable(victim)) { - error = OFPERR_OFPBRC_EPERM; - } else if (victim && victim->pending) { - error = OFPROTO_POSTPONE; - } else { - struct ofoperation *op; - struct rule *evict; - - if (classifier_count(&table->cls) > table->max_flows) { - bool was_evictable; - - was_evictable = rule->evictable; - rule->evictable = false; - evict = choose_rule_to_evict(table); - rule->evictable = was_evictable; - - if (!evict) { - error = OFPERR_OFPFMFC_TABLE_FULL; - goto exit; - } else if (evict->pending) { - error = OFPROTO_POSTPONE; - goto exit; - } - } else { - evict = NULL; - } - - group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id); - op = ofoperation_create(group, rule, OFOPERATION_ADD, 0); - op->victim = victim; - - error = ofproto->ofproto_class->rule_construct(rule); - if (error) { - op->group->n_running--; - ofoperation_destroy(rule->pending); - } else if (evict) { - delete_flow__(evict, group, OFPRR_EVICTION); - } - ofopgroup_submit(group); - } - -exit: - /* Back out if an error occurred. */ + /* Construct rule, initializing derived state. */ + error = ofproto->ofproto_class->rule_construct(rule); if (error) { - oftable_substitute_rule(rule, victim); ofproto_rule_destroy__(rule); + return error; } + + /* Insert rule. */ + oftable_insert_rule(rule); + + group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id); + ofoperation_create(group, rule, OFOPERATION_ADD, 0); + ofproto->ofproto_class->rule_insert(rule); + ofopgroup_submit(group); + return error; } @@ -3456,19 +3691,23 @@ exit: static enum ofperr modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn, struct ofputil_flow_mod *fm, const struct ofp_header *request, - struct list *rules) + const struct rule_collection *rules) { + enum ofoperation_type type; struct ofopgroup *group; - struct rule *rule; enum ofperr error; + size_t i; + type = fm->command == OFPFC_ADD ? OFOPERATION_REPLACE : OFOPERATION_MODIFY; group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id); error = OFPERR_OFPBRC_EPERM; - LIST_FOR_EACH (rule, ofproto_node, rules) { + for (i = 0; i < rules->n; i++) { + struct rule *rule = rules->rules[i]; struct ofoperation *op; bool actions_changed; + bool reset_counters; - /* FIXME: Implement OFPFF12_RESET_COUNTS */ + /* FIXME: Implement OFPFUTIL_FF_RESET_COUNTS */ if (rule_is_modifiable(rule)) { /* At least one rule is modifiable, don't report EPERM error. */ @@ -3479,7 +3718,7 @@ modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn, /* Verify actions. */ error = ofpacts_check(fm->ofpacts, fm->ofpacts_len, &fm->match.flow, - ofproto->max_ports, rule->table_id); + u16_to_ofp(ofproto->max_ports), rule->table_id); if (error) { return error; } @@ -3487,19 +3726,42 @@ modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn, actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len, rule->ofpacts, rule->ofpacts_len); - op = ofoperation_create(group, rule, OFOPERATION_MODIFY, 0); + op = ofoperation_create(group, rule, type, 0); if (fm->modify_cookie && fm->new_cookie != htonll(UINT64_MAX)) { ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie); } - if (actions_changed) { + if (type == OFOPERATION_REPLACE) { + ovs_mutex_lock(&rule->timeout_mutex); + rule->idle_timeout = fm->idle_timeout; + rule->hard_timeout = fm->hard_timeout; + ovs_mutex_unlock(&rule->timeout_mutex); + + rule->flags = fm->flags & OFPUTIL_FF_STATE; + if (fm->idle_timeout || fm->hard_timeout) { + if (!rule->eviction_group) { + eviction_group_add_rule(rule); + } + } else { + eviction_group_remove_rule(rule); + } + } + + reset_counters = (fm->flags & OFPUTIL_FF_RESET_COUNTS) != 0; + if (actions_changed || reset_counters) { op->ofpacts = rule->ofpacts; op->ofpacts_len = rule->ofpacts_len; op->meter_id = rule->meter_id; + + ovs_rwlock_wrlock(&rule->rwlock); rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len); rule->ofpacts_len = fm->ofpacts_len; - rule->meter_id = find_meter(rule->ofpacts, rule->ofpacts_len); - rule->ofproto->ofproto_class->rule_modify_actions(rule); + ovs_rwlock_unlock(&rule->rwlock); + + rule->meter_id = ofpacts_get_meter(rule->ofpacts, + rule->ofpacts_len); + rule->ofproto->ofproto_class->rule_modify_actions(rule, + reset_counters); } else { ofoperation_complete(op, 0); } @@ -3529,19 +3791,24 @@ modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn, struct ofputil_flow_mod *fm, const struct ofp_header *request) { - struct list rules; + struct rule_criteria criteria; + struct rule_collection rules; int error; - error = collect_rules_loose(ofproto, fm->table_id, &fm->match, - fm->cookie, fm->cookie_mask, - OFPP_ANY, &rules); - if (error) { - return error; - } else if (list_is_empty(&rules)) { - return modify_flows_add(ofproto, ofconn, fm, request); - } else { - return modify_flows__(ofproto, ofconn, fm, request, &rules); + rule_criteria_init(&criteria, fm->table_id, &fm->match, 0, + fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY); + error = collect_rules_loose(ofproto, &criteria, &rules); + rule_criteria_destroy(&criteria); + + if (!error) { + error = (rules.n > 0 + ? modify_flows__(ofproto, ofconn, fm, request, &rules) + : modify_flows_add(ofproto, ofconn, fm, request)); } + + rule_collection_destroy(&rules); + + return error; } /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error @@ -3554,22 +3821,26 @@ modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn, struct ofputil_flow_mod *fm, const struct ofp_header *request) { - struct list rules; + struct rule_criteria criteria; + struct rule_collection rules; int error; - error = collect_rules_strict(ofproto, fm->table_id, &fm->match, - fm->priority, fm->cookie, fm->cookie_mask, - OFPP_ANY, &rules); + rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority, + fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY); + error = collect_rules_strict(ofproto, &criteria, &rules); + rule_criteria_destroy(&criteria); - if (error) { - return error; - } else if (list_is_empty(&rules)) { - return modify_flows_add(ofproto, ofconn, fm, request); - } else { - return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn, - fm, request, &rules) - : 0; + if (!error) { + if (rules.n == 0) { + error = modify_flows_add(ofproto, ofconn, fm, request); + } else if (rules.n == 1) { + error = modify_flows__(ofproto, ofconn, fm, request, &rules); + } } + + rule_collection_destroy(&rules); + + return error; } /* OFPFC_DELETE implementation. */ @@ -3584,7 +3855,7 @@ delete_flow__(struct rule *rule, struct ofopgroup *group, ofoperation_create(group, rule, OFOPERATION_DELETE, reason); oftable_remove_rule(rule); - ofproto->ofproto_class->rule_destruct(rule); + ofproto->ofproto_class->rule_delete(rule); } /* Deletes the rules listed in 'rules'. @@ -3592,14 +3863,17 @@ delete_flow__(struct rule *rule, struct ofopgroup *group, * Returns 0 on success, otherwise an OpenFlow error code. */ static enum ofperr delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn, - const struct ofp_header *request, struct list *rules, + const struct ofp_header *request, + const struct rule_collection *rules, enum ofp_flow_removed_reason reason) { - struct rule *rule, *next; struct ofopgroup *group; + size_t i; group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX); - LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) { + for (i = 0; i < rules->n; i++) { + struct rule *rule = rules->rules[i]; + ovs_rwlock_wrlock(&rule->rwlock); delete_flow__(rule, group, reason); } ofopgroup_submit(group); @@ -3613,16 +3887,22 @@ delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn, const struct ofputil_flow_mod *fm, const struct ofp_header *request) { - struct list rules; + struct rule_criteria criteria; + struct rule_collection rules; enum ofperr error; - error = collect_rules_loose(ofproto, fm->table_id, &fm->match, - fm->cookie, fm->cookie_mask, - fm->out_port, &rules); - return (error ? error - : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request, - &rules, OFPRR_DELETE) - : 0); + rule_criteria_init(&criteria, fm->table_id, &fm->match, 0, + fm->cookie, fm->cookie_mask, + fm->out_port, fm->out_group); + error = collect_rules_loose(ofproto, &criteria, &rules); + rule_criteria_destroy(&criteria); + + if (!error && rules.n > 0) { + error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE); + } + rule_collection_destroy(&rules); + + return error; } /* Implements OFPFC_DELETE_STRICT. */ @@ -3631,17 +3911,22 @@ delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn, const struct ofputil_flow_mod *fm, const struct ofp_header *request) { - struct list rules; + struct rule_criteria criteria; + struct rule_collection rules; enum ofperr error; - error = collect_rules_strict(ofproto, fm->table_id, &fm->match, - fm->priority, fm->cookie, fm->cookie_mask, - fm->out_port, &rules); - return (error ? error - : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn, - request, &rules, - OFPRR_DELETE) - : 0); + rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority, + fm->cookie, fm->cookie_mask, + fm->out_port, fm->out_group); + error = collect_rules_strict(ofproto, &criteria, &rules); + rule_criteria_destroy(&criteria); + + if (!error && rules.n > 0) { + error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE); + } + rule_collection_destroy(&rules); + + return error; } static void @@ -3649,7 +3934,8 @@ ofproto_rule_send_removed(struct rule *rule, uint8_t reason) { struct ofputil_flow_removed fr; - if (ofproto_rule_is_hidden(rule) || !rule->send_flow_removed) { + if (ofproto_rule_is_hidden(rule) || + !(rule->flags & OFPUTIL_FF_SEND_FLOW_REM)) { return; } @@ -3660,28 +3946,16 @@ ofproto_rule_send_removed(struct rule *rule, uint8_t reason) fr.table_id = rule->table_id; calc_duration(rule->created, time_msec(), &fr.duration_sec, &fr.duration_nsec); + ovs_mutex_lock(&rule->timeout_mutex); fr.idle_timeout = rule->idle_timeout; fr.hard_timeout = rule->hard_timeout; + ovs_mutex_unlock(&rule->timeout_mutex); rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count, &fr.byte_count); connmgr_send_flow_removed(rule->ofproto->connmgr, &fr); } -void -ofproto_rule_update_used(struct rule *rule, long long int used) -{ - if (used > rule->used) { - struct eviction_group *evg = rule->eviction_group; - - rule->used = used; - if (evg) { - heap_change(&evg->rules, &rule->evg_node, - rule_eviction_priority(rule)); - } - } -} - /* Sends an OpenFlow "flow removed" message with the given 'reason' (either * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its * ofproto. @@ -3695,17 +3969,51 @@ void ofproto_rule_expire(struct rule *rule, uint8_t reason) { struct ofproto *ofproto = rule->ofproto; - struct ofopgroup *group; - - ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT); + struct classifier *cls = &ofproto->tables[rule->table_id].cls; + ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT + || reason == OFPRR_DELETE || reason == OFPRR_GROUP_DELETE); ofproto_rule_send_removed(rule, reason); - group = ofopgroup_create_unattached(ofproto); - ofoperation_create(group, rule, OFOPERATION_DELETE, reason); - oftable_remove_rule(rule); - ofproto->ofproto_class->rule_destruct(rule); - ofopgroup_submit(group); + ovs_rwlock_wrlock(&cls->rwlock); + ofproto_rule_delete(ofproto, cls, rule); + ovs_rwlock_unlock(&cls->rwlock); +} + +/* Reduces '*timeout' to no more than 'max'. A value of zero in either case + * means "infinite". */ +static void +reduce_timeout(uint16_t max, uint16_t *timeout) +{ + if (max && (!*timeout || *timeout > max)) { + *timeout = max; + } +} + +/* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle + * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to + * 'idle_timeout' seconds. Similarly for 'hard_timeout'. + * + * Suitable for implementing OFPACT_FIN_TIMEOUT. */ +void +ofproto_rule_reduce_timeouts(struct rule *rule, + uint16_t idle_timeout, uint16_t hard_timeout) + OVS_EXCLUDED(rule->ofproto->expirable_mutex, rule->timeout_mutex) +{ + if (!idle_timeout && !hard_timeout) { + return; + } + + ovs_mutex_lock(&rule->ofproto->expirable_mutex); + if (list_is_empty(&rule->expirable)) { + list_insert(&rule->ofproto->expirable, &rule->expirable); + } + ovs_mutex_unlock(&rule->ofproto->expirable_mutex); + + ovs_mutex_lock(&rule->timeout_mutex); + reduce_timeout(idle_timeout, &rule->idle_timeout); + reduce_timeout(hard_timeout, &rule->hard_timeout); + ovs_mutex_unlock(&rule->timeout_mutex); } static enum ofperr @@ -3919,6 +4227,31 @@ handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh) return 0; } +static enum ofperr +handle_nxt_get_async_request(struct ofconn *ofconn, const struct ofp_header *oh) +{ + struct ofpbuf *buf; + uint32_t master[OAM_N_TYPES]; + uint32_t slave[OAM_N_TYPES]; + struct nx_async_config *msg; + + ofconn_get_async_config(ofconn, master, slave); + buf = ofpraw_alloc_reply(OFPRAW_OFPT13_GET_ASYNC_REPLY, oh, 0); + msg = ofpbuf_put_zeros(buf, sizeof *msg); + + msg->packet_in_mask[0] = htonl(master[OAM_PACKET_IN]); + msg->port_status_mask[0] = htonl(master[OAM_PORT_STATUS]); + msg->flow_removed_mask[0] = htonl(master[OAM_FLOW_REMOVED]); + + msg->packet_in_mask[1] = htonl(slave[OAM_PACKET_IN]); + msg->port_status_mask[1] = htonl(slave[OAM_PORT_STATUS]); + msg->flow_removed_mask[1] = htonl(slave[OAM_FLOW_REMOVED]); + + ofconn_send_reply(ofconn, buf); + + return 0; +} + static enum ofperr handle_nxt_set_controller_id(struct ofconn *ofconn, const struct ofp_header *oh) @@ -3958,7 +4291,7 @@ ofproto_compose_flow_refresh_update(const struct rule *rule, struct ofputil_flow_update fu; struct match match; - if (op && op->type == OFOPERATION_ADD && !op->victim) { + if (op && op->type == OFOPERATION_ADD) { /* We'll report the final flow when the operation completes. Reporting * it now would cause a duplicate report later. */ return; @@ -3967,8 +4300,10 @@ ofproto_compose_flow_refresh_update(const struct rule *rule, fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD) ? NXFME_ADDED : NXFME_MODIFIED); fu.reason = 0; + ovs_mutex_lock(&rule->timeout_mutex); fu.idle_timeout = rule->idle_timeout; fu.hard_timeout = rule->hard_timeout; + ovs_mutex_unlock(&rule->timeout_mutex); fu.table_id = rule->table_id; fu.cookie = rule->flow_cookie; minimatch_expand(&rule->cr.match, &match); @@ -3985,12 +4320,10 @@ ofproto_compose_flow_refresh_update(const struct rule *rule, * actions, so that when the operation commits we report the change. */ switch (op->type) { case OFOPERATION_ADD: - /* We already verified that there was a victim. */ - fu.ofpacts = op->victim->ofpacts; - fu.ofpacts_len = op->victim->ofpacts_len; - break; + NOT_REACHED(); case OFOPERATION_MODIFY: + case OFOPERATION_REPLACE: if (op->ofpacts) { fu.ofpacts = op->ofpacts; fu.ofpacts_len = op->ofpacts_len; @@ -4017,11 +4350,13 @@ ofproto_compose_flow_refresh_update(const struct rule *rule, } void -ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs) +ofmonitor_compose_refresh_updates(struct rule_collection *rules, + struct list *msgs) { - struct rule *rule; + size_t i; - LIST_FOR_EACH (rule, ofproto_node, rules) { + for (i = 0; i < rules->n; i++) { + struct rule *rule = rules->rules[i]; enum nx_flow_monitor_flags flags = rule->monitor_flags; rule->monitor_flags = 0; @@ -4032,7 +4367,7 @@ ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs) static void ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m, struct rule *rule, uint64_t seqno, - struct list *rules) + struct rule_collection *rules) { enum nx_flow_monitor_flags update; @@ -4063,7 +4398,7 @@ ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m, } if (!rule->monitor_flags) { - list_push_back(rules, &rule->ofproto_node); + rule_collection_add(rules, rule); } rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS); } @@ -4071,7 +4406,7 @@ ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m, static void ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m, uint64_t seqno, - struct list *rules) + struct rule_collection *rules) { const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn); const struct ofoperation *op; @@ -4083,11 +4418,13 @@ ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m, struct cls_cursor cursor; struct rule *rule; + ovs_rwlock_rdlock(&table->cls.rwlock); cls_cursor_init(&cursor, &table->cls, &target); CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { ovs_assert(!rule->pending); /* XXX */ ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules); } + ovs_rwlock_unlock(&table->cls.rwlock); } HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) { @@ -4105,7 +4442,7 @@ ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m, static void ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m, - struct list *rules) + struct rule_collection *rules) { if (m->flags & NXFMF_INITIAL) { ofproto_collect_ofmonitor_refresh_rules(m, 0, rules); @@ -4114,7 +4451,7 @@ ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m, void ofmonitor_collect_resume_rules(struct ofmonitor *m, - uint64_t seqno, struct list *rules) + uint64_t seqno, struct rule_collection *rules) { ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules); } @@ -4125,9 +4462,9 @@ handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh) struct ofproto *ofproto = ofconn_get_ofproto(ofconn); struct ofmonitor **monitors; size_t n_monitors, allocated_monitors; + struct rule_collection rules; struct list replies; enum ofperr error; - struct list rules; struct ofpbuf b; size_t i; @@ -4166,13 +4503,15 @@ handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh) monitors[n_monitors++] = m; } - list_init(&rules); + rule_collection_init(&rules); for (i = 0; i < n_monitors; i++) { ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules); } ofpmp_init(&replies, oh); ofmonitor_compose_refresh_updates(&rules, &replies); + rule_collection_destroy(&rules); + ofconn_send_replies(ofconn, &replies); free(monitors); @@ -4331,8 +4670,9 @@ handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh, { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); uint32_t meter_id = mm->meter.meter_id; + struct rule_collection rules; + enum ofperr error = 0; uint32_t first, last; - struct list rules; if (meter_id == OFPM13_ALL) { first = 1; @@ -4346,7 +4686,7 @@ handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh, /* First delete the rules that use this meter. If any of those rules are * currently being modified, postpone the whole operation until later. */ - list_init(&rules); + rule_collection_init(&rules); for (meter_id = first; meter_id <= last; ++meter_id) { struct meter *meter = ofproto->meters[meter_id]; if (meter && !list_is_empty(&meter->rules)) { @@ -4354,20 +4694,24 @@ handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh, LIST_FOR_EACH (rule, meter_list_node, &meter->rules) { if (rule->pending) { - return OFPROTO_POSTPONE; + error = OFPROTO_POSTPONE; + goto exit; } - list_push_back(&rules, &rule->ofproto_node); + rule_collection_add(&rules, rule); } } } - if (!list_is_empty(&rules)) { + if (rules.n > 0) { delete_flows__(ofproto, ofconn, oh, &rules, OFPRR_METER_DELETE); } /* Delete the meters. */ meter_delete(ofproto, first, last); - return 0; +exit: + rule_collection_destroy(&rules); + + return error; } static enum ofperr @@ -4396,9 +4740,12 @@ handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh) if (mm.command != OFPMC13_DELETE) { /* Fails also when meters are not implemented by the provider. */ - if (!meter_id || meter_id > ofproto->meter_features.max_meters) { + if (meter_id == 0 || meter_id > OFPM13_MAX) { error = OFPERR_OFPMMFC_INVALID_METER; goto exit_free_bands; + } else if (meter_id > ofproto->meter_features.max_meters) { + error = OFPERR_OFPMMFC_OUT_OF_METERS; + goto exit_free_bands; } if (mm.meter.n_bands > ofproto->meter_features.max_bands) { error = OFPERR_OFPMMFC_OUT_OF_BANDS; @@ -4515,6 +4862,426 @@ handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request, return 0; } +bool +ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id, + struct ofgroup **group) + OVS_TRY_RDLOCK(true, (*group)->rwlock) +{ + ovs_rwlock_rdlock(&ofproto->groups_rwlock); + HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node, + hash_int(group_id, 0), &ofproto->groups) { + if ((*group)->group_id == group_id) { + ovs_rwlock_rdlock(&(*group)->rwlock); + ovs_rwlock_unlock(&ofproto->groups_rwlock); + return true; + } + } + ovs_rwlock_unlock(&ofproto->groups_rwlock); + return false; +} + +void +ofproto_group_release(struct ofgroup *group) + OVS_RELEASES(group->rwlock) +{ + ovs_rwlock_unlock(&group->rwlock); +} + +static bool +ofproto_group_write_lookup(const struct ofproto *ofproto, uint32_t group_id, + struct ofgroup **group) + OVS_TRY_WRLOCK(true, ofproto->groups_rwlock) + OVS_TRY_WRLOCK(true, (*group)->rwlock) +{ + ovs_rwlock_wrlock(&ofproto->groups_rwlock); + HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node, + hash_int(group_id, 0), &ofproto->groups) { + if ((*group)->group_id == group_id) { + ovs_rwlock_wrlock(&(*group)->rwlock); + return true; + } + } + ovs_rwlock_unlock(&ofproto->groups_rwlock); + return false; +} + +static bool +ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id) + OVS_REQ_RDLOCK(ofproto->groups_rwlock) +{ + struct ofgroup *grp; + + HMAP_FOR_EACH_IN_BUCKET (grp, hmap_node, + hash_int(group_id, 0), &ofproto->groups) { + if (grp->group_id == group_id) { + return true; + } + } + return false; +} + +static void +append_group_stats(struct ofgroup *group, struct list *replies) + OVS_REQ_RDLOCK(group->rwlock) +{ + struct ofputil_group_stats ogs; + struct ofproto *ofproto = group->ofproto; + long long int now = time_msec(); + int error; + + ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats); + + error = (ofproto->ofproto_class->group_get_stats + ? ofproto->ofproto_class->group_get_stats(group, &ogs) + : EOPNOTSUPP); + if (error) { + ogs.ref_count = UINT32_MAX; + ogs.packet_count = UINT64_MAX; + ogs.byte_count = UINT64_MAX; + ogs.n_buckets = group->n_buckets; + memset(ogs.bucket_stats, 0xff, + ogs.n_buckets * sizeof *ogs.bucket_stats); + } + + ogs.group_id = group->group_id; + calc_duration(group->created, now, &ogs.duration_sec, &ogs.duration_nsec); + + ofputil_append_group_stats(replies, &ogs); + + free(ogs.bucket_stats); +} + +static enum ofperr +handle_group_stats_request(struct ofconn *ofconn, + const struct ofp_header *request) +{ + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); + struct list replies; + enum ofperr error; + struct ofgroup *group; + uint32_t group_id; + + error = ofputil_decode_group_stats_request(request, &group_id); + if (error) { + return error; + } + + ofpmp_init(&replies, request); + + if (group_id == OFPG_ALL) { + ovs_rwlock_rdlock(&ofproto->groups_rwlock); + HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) { + ovs_rwlock_rdlock(&group->rwlock); + append_group_stats(group, &replies); + ovs_rwlock_unlock(&group->rwlock); + } + ovs_rwlock_unlock(&ofproto->groups_rwlock); + } else { + if (ofproto_group_lookup(ofproto, group_id, &group)) { + append_group_stats(group, &replies); + ofproto_group_release(group); + } + } + + ofconn_send_replies(ofconn, &replies); + + return 0; +} + +static enum ofperr +handle_group_desc_stats_request(struct ofconn *ofconn, + const struct ofp_header *request) +{ + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); + struct list replies; + struct ofputil_group_desc gds; + struct ofgroup *group; + + ofpmp_init(&replies, request); + + ovs_rwlock_rdlock(&ofproto->groups_rwlock); + HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) { + gds.group_id = group->group_id; + gds.type = group->type; + ofputil_append_group_desc_reply(&gds, &group->buckets, &replies); + } + ovs_rwlock_unlock(&ofproto->groups_rwlock); + + ofconn_send_replies(ofconn, &replies); + + return 0; +} + +static enum ofperr +handle_group_features_stats_request(struct ofconn *ofconn, + const struct ofp_header *request) +{ + struct ofproto *p = ofconn_get_ofproto(ofconn); + struct ofpbuf *msg; + + msg = ofputil_encode_group_features_reply(&p->ogf, request); + if (msg) { + ofconn_send_reply(ofconn, msg); + } + + return 0; +} + +/* Implements OFPGC11_ADD + * in which no matching flow already exists in the flow table. + * + * Adds the flow specified by 'ofm', which is followed by 'n_actions' + * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow + * error code on failure, or OFPROTO_POSTPONE if the operation cannot be + * initiated now but may be retried later. + * + * Upon successful return, takes ownership of 'fm->ofpacts'. On failure, + * ownership remains with the caller. + * + * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, + * if any. */ +static enum ofperr +add_group(struct ofproto *ofproto, struct ofputil_group_mod *gm) +{ + struct ofgroup *ofgroup; + enum ofperr error; + + if (gm->group_id > OFPG_MAX) { + return OFPERR_OFPGMFC_INVALID_GROUP; + } + if (gm->type > OFPGT11_FF) { + return OFPERR_OFPGMFC_BAD_TYPE; + } + + /* Allocate new group and initialize it. */ + ofgroup = ofproto->ofproto_class->group_alloc(); + if (!ofgroup) { + VLOG_WARN_RL(&rl, "%s: failed to create group", ofproto->name); + return OFPERR_OFPGMFC_OUT_OF_GROUPS; + } + + ovs_rwlock_init(&ofgroup->rwlock); + ofgroup->ofproto = ofproto; + ofgroup->group_id = gm->group_id; + ofgroup->type = gm->type; + ofgroup->created = ofgroup->modified = time_msec(); + + list_move(&ofgroup->buckets, &gm->buckets); + ofgroup->n_buckets = list_size(&ofgroup->buckets); + + /* Construct called BEFORE any locks are held. */ + error = ofproto->ofproto_class->group_construct(ofgroup); + if (error) { + goto free_out; + } + + /* We wrlock as late as possible to minimize the time we jam any other + * threads: No visible state changes before acquiring the lock. */ + ovs_rwlock_wrlock(&ofproto->groups_rwlock); + + if (ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) { + error = OFPERR_OFPGMFC_OUT_OF_GROUPS; + goto unlock_out; + } + + if (ofproto_group_exists(ofproto, gm->group_id)) { + error = OFPERR_OFPGMFC_GROUP_EXISTS; + goto unlock_out; + } + + if (!error) { + /* Insert new group. */ + hmap_insert(&ofproto->groups, &ofgroup->hmap_node, + hash_int(ofgroup->group_id, 0)); + ofproto->n_groups[ofgroup->type]++; + + ovs_rwlock_unlock(&ofproto->groups_rwlock); + return error; + } + + unlock_out: + ovs_rwlock_unlock(&ofproto->groups_rwlock); + ofproto->ofproto_class->group_destruct(ofgroup); + free_out: + ofputil_bucket_list_destroy(&ofgroup->buckets); + ofproto->ofproto_class->group_dealloc(ofgroup); + + return error; +} + +/* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on + * failure. + * + * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id, + * if any. */ +static enum ofperr +modify_group(struct ofproto *ofproto, struct ofputil_group_mod *gm) +{ + struct ofgroup *ofgroup; + struct ofgroup *victim; + enum ofperr error; + + if (gm->group_id > OFPG_MAX) { + return OFPERR_OFPGMFC_INVALID_GROUP; + } + + if (gm->type > OFPGT11_FF) { + return OFPERR_OFPGMFC_BAD_TYPE; + } + + victim = ofproto->ofproto_class->group_alloc(); + if (!victim) { + VLOG_WARN_RL(&rl, "%s: failed to allocate group", ofproto->name); + return OFPERR_OFPGMFC_OUT_OF_GROUPS; + } + + if (!ofproto_group_write_lookup(ofproto, gm->group_id, &ofgroup)) { + error = OFPERR_OFPGMFC_UNKNOWN_GROUP; + goto free_out; + } + /* Both group's and its container's write locks held now. + * Also, n_groups[] is protected by ofproto->groups_rwlock. */ + if (ofgroup->type != gm->type + && ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) { + error = OFPERR_OFPGMFC_OUT_OF_GROUPS; + goto unlock_out; + } + + *victim = *ofgroup; + list_move(&victim->buckets, &ofgroup->buckets); + + ofgroup->type = gm->type; + list_move(&ofgroup->buckets, &gm->buckets); + ofgroup->n_buckets = list_size(&ofgroup->buckets); + + error = ofproto->ofproto_class->group_modify(ofgroup, victim); + if (!error) { + ofputil_bucket_list_destroy(&victim->buckets); + ofproto->n_groups[victim->type]--; + ofproto->n_groups[ofgroup->type]++; + ofgroup->modified = time_msec(); + } else { + ofputil_bucket_list_destroy(&ofgroup->buckets); + + *ofgroup = *victim; + list_move(&ofgroup->buckets, &victim->buckets); + } + + unlock_out: + ovs_rwlock_unlock(&ofgroup->rwlock); + ovs_rwlock_unlock(&ofproto->groups_rwlock); + free_out: + ofproto->ofproto_class->group_dealloc(victim); + return error; +} + +static void +delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup) + OVS_RELEASES(ofproto->groups_rwlock) +{ + /* Must wait until existing readers are done, + * while holding the container's write lock at the same time. */ + ovs_rwlock_wrlock(&ofgroup->rwlock); + hmap_remove(&ofproto->groups, &ofgroup->hmap_node); + /* No-one can find this group any more. */ + ofproto->n_groups[ofgroup->type]--; + ovs_rwlock_unlock(&ofproto->groups_rwlock); + + ofproto->ofproto_class->group_destruct(ofgroup); + ofputil_bucket_list_destroy(&ofgroup->buckets); + ovs_rwlock_unlock(&ofgroup->rwlock); + ovs_rwlock_destroy(&ofgroup->rwlock); + ofproto->ofproto_class->group_dealloc(ofgroup); +} + +/* Implements OFPGC_DELETE. */ +static void +delete_group(struct ofproto *ofproto, uint32_t group_id) +{ + struct ofgroup *ofgroup; + + ovs_rwlock_wrlock(&ofproto->groups_rwlock); + if (group_id == OFPG_ALL) { + for (;;) { + struct hmap_node *node = hmap_first(&ofproto->groups); + if (!node) { + break; + } + ofgroup = CONTAINER_OF(node, struct ofgroup, hmap_node); + delete_group__(ofproto, ofgroup); + /* Lock for each node separately, so that we will not jam the + * other threads for too long time. */ + ovs_rwlock_wrlock(&ofproto->groups_rwlock); + } + } else { + HMAP_FOR_EACH_IN_BUCKET (ofgroup, hmap_node, + hash_int(group_id, 0), &ofproto->groups) { + if (ofgroup->group_id == group_id) { + delete_group__(ofproto, ofgroup); + return; + } + } + } + ovs_rwlock_unlock(&ofproto->groups_rwlock); +} + +static enum ofperr +handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh) +{ + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); + struct ofputil_group_mod gm; + enum ofperr error; + + error = reject_slave_controller(ofconn); + if (error) { + return error; + } + + error = ofputil_decode_group_mod(oh, &gm); + if (error) { + return error; + } + + switch (gm.command) { + case OFPGC11_ADD: + return add_group(ofproto, &gm); + + case OFPGC11_MODIFY: + return modify_group(ofproto, &gm); + + case OFPGC11_DELETE: + delete_group(ofproto, gm.group_id); + return 0; + + default: + if (gm.command > OFPGC11_DELETE) { + VLOG_WARN_RL(&rl, "%s: Invalid group_mod command type %d", + ofproto->name, gm.command); + } + return OFPERR_OFPGMFC_BAD_COMMAND; + } +} + +static enum ofperr +handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh) +{ + struct ofputil_table_mod tm; + enum ofperr error; + + error = reject_slave_controller(ofconn); + if (error) { + return error; + } + + error = ofputil_decode_table_mod(oh, &tm); + if (error) { + return error; + } + + /* XXX Actual table mod support is not implemented yet. */ + return 0; +} + static enum ofperr handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) { @@ -4550,6 +5317,12 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) case OFPTYPE_FLOW_MOD: return handle_flow_mod(ofconn, oh); + case OFPTYPE_GROUP_MOD: + return handle_group_mod(ofconn, oh); + + case OFPTYPE_TABLE_MOD: + return handle_table_mod(ofconn, oh); + case OFPTYPE_METER_MOD: return handle_meter_mod(ofconn, oh); @@ -4586,6 +5359,9 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) case OFPTYPE_SET_ASYNC_CONFIG: return handle_nxt_set_async_config(ofconn, oh); + case OFPTYPE_GET_ASYNC_REQUEST: + return handle_nxt_get_async_request(ofconn, oh); + /* Statistics requests. */ case OFPTYPE_DESC_STATS_REQUEST: return handle_desc_stats_request(ofconn, oh); @@ -4618,12 +5394,17 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) case OFPTYPE_METER_FEATURES_STATS_REQUEST: return handle_meter_features_request(ofconn, oh); - /* FIXME: Change the following once they are implemented: */ - case OFPTYPE_QUEUE_GET_CONFIG_REQUEST: - case OFPTYPE_GET_ASYNC_REQUEST: case OFPTYPE_GROUP_STATS_REQUEST: + return handle_group_stats_request(ofconn, oh); + case OFPTYPE_GROUP_DESC_STATS_REQUEST: + return handle_group_desc_stats_request(ofconn, oh); + case OFPTYPE_GROUP_FEATURES_STATS_REQUEST: + return handle_group_features_stats_request(ofconn, oh); + + /* FIXME: Change the following once they are implemented: */ + case OFPTYPE_QUEUE_GET_CONFIG_REQUEST: case OFPTYPE_TABLE_FEATURES_STATS_REQUEST: return OFPERR_OFPBRC_BAD_TYPE; @@ -4799,15 +5580,27 @@ ofopgroup_complete(struct ofopgroup *group) && rule->flow_cookie == op->flow_cookie))) { /* Check that we can just cast from ofoperation_type to * nx_flow_update_event. */ - BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_ADD - == NXFME_ADDED); - BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_DELETE - == NXFME_DELETED); - BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_MODIFY - == NXFME_MODIFIED); - - ofmonitor_report(ofproto->connmgr, rule, - (enum nx_flow_update_event) op->type, + enum nx_flow_update_event event_type; + + switch (op->type) { + case OFOPERATION_ADD: + case OFOPERATION_REPLACE: + event_type = NXFME_ADDED; + break; + + case OFOPERATION_DELETE: + event_type = NXFME_DELETED; + break; + + case OFOPERATION_MODIFY: + event_type = NXFME_MODIFIED; + break; + + default: + NOT_REACHED(); + } + + ofmonitor_report(ofproto->connmgr, rule, event_type, op->reason, abbrev_ofconn, abbrev_xid); } @@ -4818,7 +5611,6 @@ ofopgroup_complete(struct ofopgroup *group) if (!op->error) { uint16_t vid_mask; - ofproto_rule_destroy__(op->victim); vid_mask = minimask_get_vid_mask(&rule->cr.match.mask); if (vid_mask == VLAN_VID_MASK) { if (ofproto->vlan_bitmap) { @@ -4832,29 +5624,45 @@ ofopgroup_complete(struct ofopgroup *group) } } } else { - oftable_substitute_rule(rule, op->victim); - ofproto_rule_destroy__(rule); + ovs_rwlock_wrlock(&rule->rwlock); + oftable_remove_rule(rule); + ofproto_rule_destroy(rule); } break; case OFOPERATION_DELETE: ovs_assert(!op->error); - ofproto_rule_destroy__(rule); + ofproto_rule_destroy(rule); op->rule = NULL; break; case OFOPERATION_MODIFY: + case OFOPERATION_REPLACE: if (!op->error) { - rule->modified = time_msec(); + long long int now = time_msec(); + + rule->modified = now; + if (op->type == OFOPERATION_REPLACE) { + rule->created = rule->used = now; + } } else { ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie); + ovs_mutex_lock(&rule->timeout_mutex); + rule->idle_timeout = op->idle_timeout; + rule->hard_timeout = op->hard_timeout; + ovs_mutex_unlock(&rule->timeout_mutex); if (op->ofpacts) { free(rule->ofpacts); + + ovs_rwlock_wrlock(&rule->rwlock); rule->ofpacts = op->ofpacts; rule->ofpacts_len = op->ofpacts_len; + ovs_rwlock_unlock(&rule->rwlock); + op->ofpacts = NULL; op->ofpacts_len = 0; } + rule->flags = op->flags; } break; @@ -4908,6 +5716,11 @@ ofoperation_create(struct ofopgroup *group, struct rule *rule, op->type = type; op->reason = reason; op->flow_cookie = rule->flow_cookie; + ovs_mutex_lock(&rule->timeout_mutex); + op->idle_timeout = rule->idle_timeout; + op->hard_timeout = rule->hard_timeout; + ovs_mutex_unlock(&rule->timeout_mutex); + op->flags = rule->flags; group->n_running++; @@ -4939,14 +5752,7 @@ ofoperation_destroy(struct ofoperation *op) * indicate success or an OpenFlow error code on failure. * * If 'error' is 0, indicating success, the operation will be committed - * permanently to the flow table. There is one interesting subcase: - * - * - If 'op' is an "add flow" operation that is replacing an existing rule in - * the flow table (the "victim" rule) by a new one, then the caller must - * have uninitialized any derived state in the victim rule, as in step 5 in - * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete() - * performs steps 6 and 7 for the victim rule, most notably by calling its - * ->rule_dealloc() function. + * permanently to the flow table. * * If 'error' is nonzero, then generally the operation will be rolled back: * @@ -4978,13 +5784,6 @@ ofoperation_complete(struct ofoperation *op, enum ofperr error) ofopgroup_complete(group); } } - -struct rule * -ofoperation_get_victim(struct ofoperation *op) -{ - ovs_assert(op->type == OFOPERATION_ADD); - return op->victim; -} static uint64_t pick_datapath_id(const struct ofproto *ofproto) @@ -5017,17 +5816,18 @@ pick_fallback_dpid(void) /* Table overflow policy. */ -/* Chooses and returns a rule to evict from 'table'. Returns NULL if the table - * is not configured to evict rules or if the table contains no evictable - * rules. (Rules with 'evictable' set to false or with no timeouts are not - * evictable.) */ -static struct rule * -choose_rule_to_evict(struct oftable *table) +/* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep' + * to NULL if the table is not configured to evict rules or if the table + * contains no evictable rules. (Rules with a readlock on their evict rwlock, + * or with no timeouts are not evictable.) */ +static bool +choose_rule_to_evict(struct oftable *table, struct rule **rulep) { struct eviction_group *evg; + *rulep = NULL; if (!table->eviction_fields) { - return NULL; + return false; } /* In the common case, the outer and inner loops here will each be entered @@ -5041,18 +5841,19 @@ choose_rule_to_evict(struct oftable *table) * group has no evictable rules. * * - The outer loop can exit only if table's 'max_flows' is all filled up - * by unevictable rules'. */ + * by unevictable rules. */ HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) { struct rule *rule; HEAP_FOR_EACH (rule, evg_node, &evg->rules) { - if (rule->evictable) { - return rule; + if (!ovs_rwlock_trywrlock(&rule->rwlock)) { + *rulep = rule; + return true; } } } - return NULL; + return false; } /* Searches 'ofproto' for tables that have more flows than their configured @@ -5069,19 +5870,31 @@ ofproto_evict(struct ofproto *ofproto) group = ofopgroup_create_unattached(ofproto); OFPROTO_FOR_EACH_TABLE (table, ofproto) { - while (classifier_count(&table->cls) > table->max_flows - && table->eviction_fields) { + while (table->eviction_fields) { struct rule *rule; + size_t n_rules; + + ovs_rwlock_rdlock(&table->cls.rwlock); + n_rules = classifier_count(&table->cls); + ovs_rwlock_unlock(&table->cls.rwlock); - rule = choose_rule_to_evict(table); - if (!rule || rule->pending) { + if (n_rules <= table->max_flows) { + break; + } + + if (!choose_rule_to_evict(table, &rule)) { + break; + } + + if (rule->pending) { + ovs_rwlock_unlock(&rule->rwlock); break; } ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_EVICTION); oftable_remove_rule(rule); - ofproto->ofproto_class->rule_destruct(rule); + ofproto->ofproto_class->rule_delete(rule); } } ofopgroup_submit(group); @@ -5218,6 +6031,7 @@ rule_eviction_priority(struct rule *rule) uint32_t expiration_offset; /* Calculate time of expiration. */ + ovs_mutex_lock(&rule->timeout_mutex); hard_expiration = (rule->hard_timeout ? rule->modified + rule->hard_timeout * 1000 : LLONG_MAX); @@ -5225,6 +6039,7 @@ rule_eviction_priority(struct rule *rule) ? rule->used + rule->idle_timeout * 1000 : LLONG_MAX); expiration = MIN(hard_expiration, idle_expiration); + ovs_mutex_unlock(&rule->timeout_mutex); if (expiration == LLONG_MAX) { return 0; } @@ -5251,9 +6066,13 @@ eviction_group_add_rule(struct rule *rule) { struct ofproto *ofproto = rule->ofproto; struct oftable *table = &ofproto->tables[rule->table_id]; + bool has_timeout; - if (table->eviction_fields - && (rule->hard_timeout || rule->idle_timeout)) { + ovs_mutex_lock(&rule->timeout_mutex); + has_timeout = rule->hard_timeout || rule->idle_timeout; + ovs_mutex_unlock(&rule->timeout_mutex); + + if (table->eviction_fields && has_timeout) { struct eviction_group *evg; evg = eviction_group_find(table, eviction_group_hash_rule(rule)); @@ -5282,7 +6101,9 @@ oftable_init(struct oftable *table) static void oftable_destroy(struct oftable *table) { + ovs_rwlock_rdlock(&table->cls.rwlock); ovs_assert(classifier_is_empty(&table->cls)); + ovs_rwlock_unlock(&table->cls.rwlock); oftable_disable_eviction(table); classifier_destroy(&table->cls); free(table->name); @@ -5362,77 +6183,73 @@ oftable_enable_eviction(struct oftable *table, hmap_init(&table->eviction_groups_by_id); heap_init(&table->eviction_groups_by_size); + ovs_rwlock_rdlock(&table->cls.rwlock); cls_cursor_init(&cursor, &table->cls, NULL); CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { eviction_group_add_rule(rule); } + ovs_rwlock_unlock(&table->cls.rwlock); } /* Removes 'rule' from the oftable that contains it. */ static void -oftable_remove_rule(struct rule *rule) +oftable_remove_rule__(struct ofproto *ofproto, struct classifier *cls, + struct rule *rule) + OVS_REQ_WRLOCK(cls->rwlock) OVS_RELEASES(rule->rwlock) { - struct ofproto *ofproto = rule->ofproto; - struct oftable *table = &ofproto->tables[rule->table_id]; - - classifier_remove(&table->cls, &rule->cr); - if (rule->meter_id) { - list_remove(&rule->meter_list_node); - } + classifier_remove(cls, &rule->cr); cookies_remove(ofproto, rule); eviction_group_remove_rule(rule); + ovs_mutex_lock(&ofproto->expirable_mutex); if (!list_is_empty(&rule->expirable)) { list_remove(&rule->expirable); } + ovs_mutex_unlock(&ofproto->expirable_mutex); if (!list_is_empty(&rule->meter_list_node)) { list_remove(&rule->meter_list_node); + list_init(&rule->meter_list_node); } + ovs_rwlock_unlock(&rule->rwlock); +} + +static void +oftable_remove_rule(struct rule *rule) +{ + struct ofproto *ofproto = rule->ofproto; + struct oftable *table = &ofproto->tables[rule->table_id]; + + ovs_rwlock_wrlock(&table->cls.rwlock); + oftable_remove_rule__(ofproto, &table->cls, rule); + ovs_rwlock_unlock(&table->cls.rwlock); } -/* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s - * oftable that has an identical cls_rule. Returns the rule that was removed, - * if any, and otherwise NULL. */ -static struct rule * -oftable_replace_rule(struct rule *rule) +/* Inserts 'rule' into its oftable, which must not already contain any rule for + * the same cls_rule. */ +static void +oftable_insert_rule(struct rule *rule) { struct ofproto *ofproto = rule->ofproto; struct oftable *table = &ofproto->tables[rule->table_id]; - struct rule *victim; - bool may_expire = rule->hard_timeout || rule->idle_timeout; + bool may_expire; + + ovs_mutex_lock(&rule->timeout_mutex); + may_expire = rule->hard_timeout || rule->idle_timeout; + ovs_mutex_unlock(&rule->timeout_mutex); if (may_expire) { + ovs_mutex_lock(&ofproto->expirable_mutex); list_insert(&ofproto->expirable, &rule->expirable); + ovs_mutex_unlock(&ofproto->expirable_mutex); } cookies_insert(ofproto, rule); if (rule->meter_id) { struct meter *meter = ofproto->meters[rule->meter_id]; list_insert(&meter->rules, &rule->meter_list_node); } - victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr)); - if (victim) { - if (victim->meter_id) { - list_remove(&victim->meter_list_node); - } - cookies_remove(ofproto, victim); - - if (!list_is_empty(&victim->expirable)) { - list_remove(&victim->expirable); - } - eviction_group_remove_rule(victim); - } + ovs_rwlock_wrlock(&table->cls.rwlock); + classifier_insert(&table->cls, &rule->cr); + ovs_rwlock_unlock(&table->cls.rwlock); eviction_group_add_rule(rule); - return victim; -} - -/* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */ -static void -oftable_substitute_rule(struct rule *old, struct rule *new) -{ - if (new) { - oftable_replace_rule(new); - } else { - oftable_remove_rule(old); - } } /* unixctl commands. */