X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=ofproto%2Fofproto-dpif.c;h=58c50643696de7f3b33cfd5229f0da72ef951b2e;hb=60cda7d69b0bfd242045d346f2cd169836a3d78e;hp=52e47c298b732e4010bb714224e82df58ba95629;hpb=62c70c9d47edf2dbf5c9d0c8cd6c15c8c43dc7a8;p=sliver-openvswitch.git diff --git a/ofproto/ofproto-dpif.c b/ofproto/ofproto-dpif.c index 52e47c298..58c506436 100644 --- a/ofproto/ofproto-dpif.c +++ b/ofproto/ofproto-dpif.c @@ -78,6 +78,9 @@ enum { N_TABLES = 255 }; enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */ BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255); +/* No bfd/cfm status change. */ +#define NO_STATUS_CHANGE -1 + struct flow_miss; struct rule_dpif { @@ -314,6 +317,8 @@ struct ofproto_dpif { /* Work queues. */ struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */ + struct seq *pins_seq; /* For notifying 'pins' reception. */ + uint64_t pins_seqno; }; /* All existing ofproto_dpif instances, indexed by ->up.name. */ @@ -385,6 +390,9 @@ ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto, free(CONST_CAST(void *, pin->up.packet)); free(pin); } + + /* Wakes up main thread for packet-in I/O. */ + seq_change(ofproto->pins_seq); } /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the @@ -963,7 +971,7 @@ check_recirc(struct dpif_backer *backer) flow.dp_hash = 1; ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); - odp_flow_key_from_flow(&key, &flow, 0); + odp_flow_key_from_flow(&key, &flow, NULL, 0); error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY, ofpbuf_data(&key), ofpbuf_size(&key), NULL, 0, NULL, @@ -1096,7 +1104,7 @@ check_max_mpls_depth(struct dpif_backer *backer) flow_set_mpls_bos(&flow, n, 1); ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); - odp_flow_key_from_flow(&key, &flow, 0); + odp_flow_key_from_flow(&key, &flow, NULL, 0); error = dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY, ofpbuf_data(&key), ofpbuf_size(&key), NULL, 0, NULL, 0, NULL); @@ -1157,6 +1165,9 @@ construct(struct ofproto *ofproto_) sset_init(&ofproto->port_poll_set); ofproto->port_poll_errno = 0; ofproto->change_seq = 0; + ofproto->pins_seq = seq_create(); + ofproto->pins_seqno = seq_read(ofproto->pins_seq); + SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) { struct iface_hint *iface_hint = node->data; @@ -1330,6 +1341,8 @@ destruct(struct ofproto *ofproto_) ovs_mutex_destroy(&ofproto->stats_mutex); ovs_mutex_destroy(&ofproto->vsp_mutex); + seq_destroy(ofproto->pins_seq); + close_dpif_backer(ofproto->backer); } @@ -1338,7 +1351,6 @@ run(struct ofproto *ofproto_) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); uint64_t new_seq, new_dump_seq; - const bool enable_recirc = ofproto_dpif_get_enable_recirc(ofproto); if (mbridge_need_revalidate(ofproto->mbridge)) { ofproto->backer->need_revalidate = REV_RECONFIGURE; @@ -1362,6 +1374,12 @@ run(struct ofproto *ofproto_) } } + /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during + * flow restore. Even though nothing is processed during flow restore, + * all queued 'pins' will be handled immediately when flow restore + * completes. */ + ofproto->pins_seqno = seq_read(ofproto->pins_seq); + if (ofproto->netflow) { netflow_run(ofproto->netflow); } @@ -1416,17 +1434,12 @@ run(struct ofproto *ofproto_) /* All outstanding data in existing flows has been accounted, so it's a * good time to do bond rebalancing. */ - if (enable_recirc && ofproto->has_bonded_bundles) { + if (ofproto->has_bonded_bundles) { struct ofbundle *bundle; HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { - struct bond *bond = bundle->bond; - - if (bond && bond_may_recirc(bond, NULL, NULL)) { - bond_recirculation_account(bond); - if (bond_rebalance(bundle->bond)) { - bond_update_post_recirc_rules(bond, true); - } + if (bundle->bond) { + bond_rebalance(bundle->bond); } } } @@ -1471,6 +1484,7 @@ wait(struct ofproto *ofproto_) } seq_wait(udpif_dump_seq(ofproto->backer->udpif), ofproto->dump_seq); + seq_wait(ofproto->pins_seq, ofproto->pins_seqno); } static void @@ -1813,22 +1827,28 @@ out: return error; } -static bool +static int get_cfm_status(const struct ofport *ofport_, struct ofproto_cfm_status *status) { struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); + int ret = 0; if (ofport->cfm) { - status->faults = cfm_get_fault(ofport->cfm); - status->flap_count = cfm_get_flap_count(ofport->cfm); - status->remote_opstate = cfm_get_opup(ofport->cfm); - status->health = cfm_get_health(ofport->cfm); - cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps); - return true; + if (cfm_check_status_change(ofport->cfm)) { + status->faults = cfm_get_fault(ofport->cfm); + status->flap_count = cfm_get_flap_count(ofport->cfm); + status->remote_opstate = cfm_get_opup(ofport->cfm); + status->health = cfm_get_health(ofport->cfm); + cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps); + } else { + ret = NO_STATUS_CHANGE; + } } else { - return false; + ret = ENOENT; } + + return ret; } static int @@ -1853,13 +1873,19 @@ static int get_bfd_status(struct ofport *ofport_, struct smap *smap) { struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); + int ret = 0; if (ofport->bfd) { - bfd_get_status(ofport->bfd, smap); - return 0; + if (bfd_check_status_change(ofport->bfd)) { + bfd_get_status(ofport->bfd, smap); + } else { + ret = NO_STATUS_CHANGE; + } } else { - return ENOENT; + ret = ENOENT; } + + return ret; } /* Spanning Tree. */ @@ -3183,16 +3209,49 @@ rule_dpif_get_actions(const struct rule_dpif *rule) return rule_get_actions(&rule->up); } -static uint8_t -rule_dpif_lookup__ (struct ofproto_dpif *ofproto, const struct flow *flow, - struct flow_wildcards *wc, struct rule_dpif **rule) +/* Lookup 'flow' in table 0 of 'ofproto''s classifier. + * If 'wc' is non-null, sets the fields that were relevant as part of + * the lookup. Returns the table_id where a match or miss occurred. + * + * The return value will be zero unless there was a miss and + * OFPTC11_TABLE_MISS_CONTINUE is in effect for the sequence of tables + * where misses occur. + * + * The rule is returned in '*rule', which is valid at least until the next + * RCU quiescent period. If the '*rule' needs to stay around longer, + * a non-zero 'take_ref' must be passed in to cause a reference to be taken + * on it before this returns. */ +uint8_t +rule_dpif_lookup(struct ofproto_dpif *ofproto, struct flow *flow, + struct flow_wildcards *wc, struct rule_dpif **rule, + bool take_ref) { enum rule_dpif_lookup_verdict verdict; enum ofputil_port_config config = 0; - uint8_t table_id = TBL_INTERNAL; + uint8_t table_id; + + if (ofproto_dpif_get_enable_recirc(ofproto)) { + /* Always exactly match recirc_id since datapath supports + * recirculation. */ + if (wc) { + wc->masks.recirc_id = UINT32_MAX; + } + + /* Start looking up from internal table for post recirculation flows + * or packets. We can also simply send all, including normal flows + * or packets to the internal table. They will not match any post + * recirculation rules except the 'catch all' rule that resubmit + * them to table 0. + * + * As an optimization, we send normal flows and packets to table 0 + * directly, saving one table lookup. */ + table_id = flow->recirc_id ? TBL_INTERNAL : 0; + } else { + table_id = 0; + } verdict = rule_dpif_lookup_from_table(ofproto, flow, wc, true, - &table_id, rule); + &table_id, rule, take_ref); switch (verdict) { case RULE_DPIF_LOOKUP_VERDICT_MATCH: @@ -3221,30 +3280,17 @@ rule_dpif_lookup__ (struct ofproto_dpif *ofproto, const struct flow *flow, } choose_miss_rule(config, ofproto->miss_rule, - ofproto->no_packet_in_rule, rule); + ofproto->no_packet_in_rule, rule, take_ref); return table_id; } -/* Lookup 'flow' in table 0 of 'ofproto''s classifier. - * If 'wc' is non-null, sets the fields that were relevant as part of - * the lookup. Returns the table_id where a match or miss occurred. - * - * The return value will be zero unless there was a miss and - * OFPTC11_TABLE_MISS_CONTINUE is in effect for the sequence of tables - * where misses occur. */ -uint8_t -rule_dpif_lookup(struct ofproto_dpif *ofproto, struct flow *flow, - struct flow_wildcards *wc, struct rule_dpif **rule) -{ - /* Set metadata to the value of recirc_id to speed up internal - * rule lookup. */ - flow->metadata = htonll(flow->recirc_id); - return rule_dpif_lookup__(ofproto, flow, wc, rule); -} - +/* The returned rule is valid at least until the next RCU quiescent period. + * If the '*rule' needs to stay around longer, a non-zero 'take_ref' must be + * passed in to cause a reference to be taken on it before this returns. */ static struct rule_dpif * rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, uint8_t table_id, - const struct flow *flow, struct flow_wildcards *wc) + const struct flow *flow, struct flow_wildcards *wc, + bool take_ref) { struct classifier *cls = &ofproto->up.tables[table_id].cls; const struct cls_rule *cls_rule; @@ -3278,7 +3324,9 @@ rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, uint8_t table_id, } rule = rule_dpif_cast(rule_from_cls_rule(cls_rule)); - rule_dpif_ref(rule); + if (take_ref) { + rule_dpif_ref(rule); + } fat_rwlock_unlock(&cls->rwlock); return rule; @@ -3313,13 +3361,19 @@ rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, uint8_t table_id, * * - RULE_DPIF_LOOKUP_VERDICT_DEFAULT if no rule was found, * 'honor_table_miss' is true and a table miss configuration has - * not been specified in this case. */ + * not been specified in this case. + * + * The rule is returned in '*rule', which is valid at least until the next + * RCU quiescent period. If the '*rule' needs to stay around longer, + * a non-zero 'take_ref' must be passed in to cause a reference to be taken + * on it before this returns. */ enum rule_dpif_lookup_verdict rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto, const struct flow *flow, struct flow_wildcards *wc, bool honor_table_miss, - uint8_t *table_id, struct rule_dpif **rule) + uint8_t *table_id, struct rule_dpif **rule, + bool take_ref) { uint8_t next_id; @@ -3328,7 +3382,8 @@ rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto, next_id++, next_id += (next_id == TBL_INTERNAL)) { *table_id = next_id; - *rule = rule_dpif_lookup_in_table(ofproto, *table_id, flow, wc); + *rule = rule_dpif_lookup_in_table(ofproto, *table_id, flow, wc, + take_ref); if (*rule) { return RULE_DPIF_LOOKUP_VERDICT_MATCH; } else if (!honor_table_miss) { @@ -3355,13 +3410,21 @@ rule_dpif_lookup_from_table(struct ofproto_dpif *ofproto, /* Given a port configuration (specified as zero if there's no port), chooses * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a - * flow table miss. */ + * flow table miss. + * + * The rule is returned in '*rule', which is valid at least until the next + * RCU quiescent period. If the '*rule' needs to stay around longer, + * a reference must be taken on it (rule_dpif_ref()). + */ void choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule, - struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule) + struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule, + bool take_ref) { *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule; - rule_dpif_ref(*rule); + if (take_ref) { + rule_dpif_ref(*rule); + } } void @@ -4187,7 +4250,7 @@ ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow, if (ofpacts) { rule = NULL; } else { - rule_dpif_lookup(ofproto, flow, &trace.wc, &rule); + rule_dpif_lookup(ofproto, flow, &trace.wc, &rule, false); trace_format_rule(ds, 0, rule); if (rule == ofproto->miss_rule) { @@ -4243,8 +4306,6 @@ ofproto_trace(struct ofproto_dpif *ofproto, struct flow *flow, xlate_out_uninit(&trace.xout); } - - rule_dpif_unref(rule); } /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list @@ -4809,9 +4870,8 @@ ofproto_dpif_add_internal_flow(struct ofproto_dpif *ofproto, } rule = rule_dpif_lookup_in_table(ofproto, TBL_INTERNAL, &match->flow, - &match->wc); + &match->wc, false); if (rule) { - rule_dpif_unref(rule); *rulep = &rule->up; } else { OVS_NOT_REACHED();