X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=ofproto%2Fofproto.c;h=7b9974176061f16bfa421da6612f3f6a1ae781ab;hb=80d5aefd65b9dd953d873f9995b949bc8b8d19d5;hp=cead7f76fcb275be9f6454493743a884cd683878;hpb=5bf0e941d91cc1236b0fa9551a75a287ca31282f;p=sliver-openvswitch.git diff --git a/ofproto/ofproto.c b/ofproto/ofproto.c index cead7f76f..7b9974176 100644 --- a/ofproto/ofproto.c +++ b/ofproto/ofproto.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2009, 2010, 2011 Nicira Networks. + * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks. * Copyright (c) 2010 Jean Tourrilhes - HP-Labs. * * Licensed under the Apache License, Version 2.0 (the "License"); @@ -19,106 +19,186 @@ #include "ofproto.h" #include #include -#include -#include -#include #include #include -#include "autopath.h" #include "bitmap.h" -#include "bond.h" #include "byte-order.h" -#include "cfm.h" #include "classifier.h" #include "connmgr.h" #include "coverage.h" #include "dynamic-string.h" -#include "fail-open.h" #include "hash.h" #include "hmap.h" -#include "hmapx.h" -#include "in-band.h" -#include "lacp.h" -#include "mac-learning.h" -#include "multipath.h" +#include "meta-flow.h" #include "netdev.h" -#include "netflow.h" -#include "netlink.h" #include "nx-match.h" -#include "odp-util.h" +#include "ofp-errors.h" #include "ofp-print.h" #include "ofp-util.h" -#include "ofproto-sflow.h" #include "ofpbuf.h" +#include "ofproto-provider.h" #include "openflow/nicira-ext.h" #include "openflow/openflow.h" -#include "openvswitch/datapath-protocol.h" #include "packets.h" #include "pinsched.h" #include "pktbuf.h" #include "poll-loop.h" -#include "private.h" -#include "rconn.h" +#include "random.h" #include "shash.h" #include "sset.h" -#include "stream-ssl.h" -#include "tag.h" -#include "timer.h" #include "timeval.h" #include "unaligned.h" #include "unixctl.h" -#include "vconn.h" -#include "vlan-bitmap.h" #include "vlog.h" VLOG_DEFINE_THIS_MODULE(ofproto); -COVERAGE_DEFINE(odp_overflow); -COVERAGE_DEFINE(ofproto_agg_request); -COVERAGE_DEFINE(ofproto_costly_flags); -COVERAGE_DEFINE(ofproto_ctlr_action); COVERAGE_DEFINE(ofproto_error); -COVERAGE_DEFINE(ofproto_expiration); -COVERAGE_DEFINE(ofproto_expired); -COVERAGE_DEFINE(ofproto_flows_req); COVERAGE_DEFINE(ofproto_flush); -COVERAGE_DEFINE(ofproto_invalidated); COVERAGE_DEFINE(ofproto_no_packet_in); -COVERAGE_DEFINE(ofproto_ofp2odp); -COVERAGE_DEFINE(ofproto_packet_in); COVERAGE_DEFINE(ofproto_packet_out); COVERAGE_DEFINE(ofproto_queue_req); COVERAGE_DEFINE(ofproto_recv_openflow); COVERAGE_DEFINE(ofproto_reinit_ports); -COVERAGE_DEFINE(ofproto_unexpected_rule); COVERAGE_DEFINE(ofproto_uninstallable); COVERAGE_DEFINE(ofproto_update_port); -static void ofport_destroy__(struct ofport *); -static void ofport_destroy(struct ofport *); +enum ofproto_state { + S_OPENFLOW, /* Processing OpenFlow commands. */ + S_EVICT, /* Evicting flows from over-limit tables. */ + S_FLUSH, /* Deleting all flow table rules. */ +}; -static int rule_create(struct ofproto *, const struct cls_rule *, - const union ofp_action *, size_t n_actions, - uint16_t idle_timeout, uint16_t hard_timeout, - ovs_be64 flow_cookie, bool send_flow_removed, - struct rule **rulep); +enum ofoperation_type { + OFOPERATION_ADD, + OFOPERATION_DELETE, + OFOPERATION_MODIFY +}; -static uint64_t pick_datapath_id(const struct ofproto *); -static uint64_t pick_fallback_dpid(void); +/* A single OpenFlow request can execute any number of operations. The + * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the + * ofconn to which an error reply should be sent if necessary. + * + * ofproto initiates some operations internally. These operations are still + * assigned to groups but will not have an associated ofconn. */ +struct ofopgroup { + struct ofproto *ofproto; /* Owning ofproto. */ + struct list ofproto_node; /* In ofproto's "pending" list. */ + struct list ops; /* List of "struct ofoperation"s. */ + + /* Data needed to send OpenFlow reply on failure or to send a buffered + * packet on success. + * + * If list_is_empty(ofconn_node) then this ofopgroup never had an + * associated ofconn or its ofconn's connection dropped after it initiated + * the operation. In the latter case 'ofconn' is a wild pointer that + * refers to freed memory, so the 'ofconn' member must be used only if + * !list_is_empty(ofconn_node). + */ + struct list ofconn_node; /* In ofconn's list of pending opgroups. */ + struct ofconn *ofconn; /* ofconn for reply (but see note above). */ + struct ofp_header *request; /* Original request (truncated at 64 bytes). */ + uint32_t buffer_id; /* Buffer id from original request. */ + int error; /* 0 if no error yet, otherwise error code. */ +}; -static void ofproto_destroy__(struct ofproto *); -static void ofproto_flush_flows__(struct ofproto *); +static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *); +static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *, + const struct ofp_header *, + uint32_t buffer_id); +static void ofopgroup_submit(struct ofopgroup *); +static void ofopgroup_destroy(struct ofopgroup *); + +/* A single flow table operation. */ +struct ofoperation { + struct ofopgroup *group; /* Owning group. */ + struct list group_node; /* In ofopgroup's "ops" list. */ + struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */ + struct rule *rule; /* Rule being operated upon. */ + enum ofoperation_type type; /* Type of operation. */ + int status; /* -1 if pending, otherwise 0 or error code. */ + struct rule *victim; /* OFOPERATION_ADDING: Replaced rule. */ + union ofp_action *actions; /* OFOPERATION_MODIFYING: Replaced actions. */ + int n_actions; /* OFOPERATION_MODIFYING: # of old actions. */ + ovs_be64 flow_cookie; /* Rule's old flow cookie. */ +}; -static void ofproto_rule_destroy__(struct rule *); -static void ofproto_rule_send_removed(struct rule *, uint8_t reason); -static void ofproto_rule_remove(struct rule *); +static void ofoperation_create(struct ofopgroup *, struct rule *, + enum ofoperation_type); +static void ofoperation_destroy(struct ofoperation *); + +/* oftable. */ +static void oftable_init(struct oftable *); +static void oftable_destroy(struct oftable *); + +static void oftable_set_name(struct oftable *, const char *name); + +static void oftable_disable_eviction(struct oftable *); +static void oftable_enable_eviction(struct oftable *, + const struct mf_subfield *fields, + size_t n_fields); + +static void oftable_remove_rule(struct rule *); +static struct rule *oftable_replace_rule(struct rule *); +static void oftable_substitute_rule(struct rule *old, struct rule *new); -static void handle_openflow(struct ofconn *, struct ofpbuf *); +/* A set of rules within a single OpenFlow table (oftable) that have the same + * values for the oftable's eviction_fields. A rule to be evicted, when one is + * needed, is taken from the eviction group that contains the greatest number + * of rules. + * + * An oftable owns any number of eviction groups, each of which contains any + * number of rules. + * + * Membership in an eviction group is imprecise, based on the hash of the + * oftable's eviction_fields (in the eviction_group's id_node.hash member). + * That is, if two rules have different eviction_fields, but those + * eviction_fields hash to the same value, then they will belong to the same + * eviction_group anyway. + * + * (When eviction is not enabled on an oftable, we don't track any eviction + * groups, to save time and space.) */ +struct eviction_group { + struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */ + struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */ + struct heap rules; /* Contains "struct rule"s. */ +}; + +static struct rule *choose_rule_to_evict(struct oftable *); +static void ofproto_evict(struct ofproto *); +static uint32_t rule_eviction_priority(struct rule *); + +/* ofport. */ +static void ofport_destroy__(struct ofport *); +static void ofport_destroy(struct ofport *); static void update_port(struct ofproto *, const char *devname); static int init_ports(struct ofproto *); static void reinit_ports(struct ofproto *); +/* 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 *); +static bool rule_is_hidden(const struct rule *); + +/* OpenFlow. */ +static enum ofperr add_flow(struct ofproto *, struct ofconn *, + const struct ofputil_flow_mod *, + const struct ofp_header *); +static void delete_flow__(struct rule *, struct ofopgroup *); +static bool handle_openflow(struct ofconn *, struct ofpbuf *); +static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *, + const struct ofputil_flow_mod *, + const struct ofp_header *); + +/* ofproto. */ +static uint64_t pick_datapath_id(const struct ofproto *); +static uint64_t pick_fallback_dpid(void); +static void ofproto_destroy__(struct ofproto *); +static void set_internal_devs_mtu(struct ofproto *); + +/* unixctl. */ static void ofproto_unixctl_init(void); /* All registered ofproto classes, in probe order. */ @@ -126,7 +206,7 @@ static const struct ofproto_class **ofproto_classes; static size_t n_ofproto_classes; static size_t allocated_ofproto_classes; -/* Map from dpif name to struct ofproto, for use by unixctl commands. */ +/* Map from datapath name to struct ofproto, for use by unixctl commands. */ static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos); static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); @@ -286,17 +366,27 @@ ofproto_create(const char *datapath_name, const char *datapath_type, hmap_insert(&all_ofprotos, &ofproto->hmap_node, hash_string(ofproto->name, 0)); ofproto->datapath_id = 0; + ofproto_set_flow_eviction_threshold(ofproto, + OFPROTO_FLOW_EVICTON_THRESHOLD_DEFAULT); + ofproto->forward_bpdu = false; ofproto->fallback_dpid = pick_fallback_dpid(); ofproto->mfr_desc = xstrdup(DEFAULT_MFR_DESC); ofproto->hw_desc = xstrdup(DEFAULT_HW_DESC); ofproto->sw_desc = xstrdup(DEFAULT_SW_DESC); ofproto->serial_desc = xstrdup(DEFAULT_SERIAL_DESC); ofproto->dp_desc = xstrdup(DEFAULT_DP_DESC); - ofproto->netdev_monitor = netdev_monitor_create(); + ofproto->frag_handling = OFPC_FRAG_NORMAL; hmap_init(&ofproto->ports); shash_init(&ofproto->port_by_name); - classifier_init(&ofproto->cls); + ofproto->tables = NULL; + ofproto->n_tables = 0; ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name); + ofproto->state = S_OPENFLOW; + list_init(&ofproto->pending); + ofproto->n_pending = 0; + hmap_init(&ofproto->deletions); + ofproto->vlan_bitmap = NULL; + ofproto->vlans_changed = false; error = ofproto->ofproto_class->construct(ofproto); if (error) { @@ -306,6 +396,8 @@ ofproto_create(const char *datapath_name, const char *datapath_type, return error; } + assert(ofproto->n_tables); + ofproto->datapath_id = pick_datapath_id(ofproto); VLOG_INFO("using datapath ID %016"PRIx64, ofproto->datapath_id); init_ports(ofproto); @@ -314,6 +406,21 @@ ofproto_create(const char *datapath_name, const char *datapath_type, return 0; } +void +ofproto_init_tables(struct ofproto *ofproto, int n_tables) +{ + struct oftable *table; + + assert(!ofproto->n_tables); + assert(n_tables >= 1 && n_tables <= 255); + + ofproto->n_tables = n_tables; + ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables); + OFPROTO_FOR_EACH_TABLE (table, ofproto) { + oftable_init(table); + } +} + void ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id) { @@ -369,6 +476,43 @@ ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id) connmgr_set_in_band_queue(ofproto->connmgr, queue_id); } +/* Sets the number of flows at which eviction from the kernel flow table + * will occur. */ +void +ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold) +{ + if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) { + ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN; + } else { + ofproto->flow_eviction_threshold = threshold; + } +} + +/* If forward_bpdu is true, the NORMAL action will forward frames with + * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false, + * the NORMAL action will drop these frames. */ +void +ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu) +{ + bool old_val = ofproto->forward_bpdu; + ofproto->forward_bpdu = forward_bpdu; + if (old_val != ofproto->forward_bpdu) { + if (ofproto->ofproto_class->forward_bpdu_changed) { + ofproto->ofproto_class->forward_bpdu_changed(ofproto); + } + } +} + +/* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to + * 'idle_time', in seconds. */ +void +ofproto_set_mac_idle_time(struct ofproto *ofproto, unsigned idle_time) +{ + if (ofproto->ofproto_class->set_mac_idle_time) { + ofproto->ofproto_class->set_mac_idle_time(ofproto, idle_time); + } +} + void ofproto_set_desc(struct ofproto *p, const char *mfr_desc, const char *hw_desc, @@ -456,6 +600,106 @@ ofproto_set_sflow(struct ofproto *ofproto, } } +/* Spanning Tree Protocol (STP) configuration. */ + +/* Configures STP on 'ofproto' using the settings defined in 's'. If + * 's' is NULL, disables STP. + * + * Returns 0 if successful, otherwise a positive errno value. */ +int +ofproto_set_stp(struct ofproto *ofproto, + const struct ofproto_stp_settings *s) +{ + return (ofproto->ofproto_class->set_stp + ? ofproto->ofproto_class->set_stp(ofproto, s) + : EOPNOTSUPP); +} + +/* Retrieves STP status of 'ofproto' and stores it in 's'. If the + * 'enabled' member of 's' is false, then the other members are not + * meaningful. + * + * Returns 0 if successful, otherwise a positive errno value. */ +int +ofproto_get_stp_status(struct ofproto *ofproto, + struct ofproto_stp_status *s) +{ + return (ofproto->ofproto_class->get_stp_status + ? ofproto->ofproto_class->get_stp_status(ofproto, s) + : EOPNOTSUPP); +} + +/* Configures STP on 'ofp_port' of 'ofproto' using the settings defined + * in 's'. The caller is responsible for assigning STP port numbers + * (using the 'port_num' member in the range of 1 through 255, inclusive) + * and ensuring there are no duplicates. If the 's' is NULL, then STP + * is disabled on the port. + * + * Returns 0 if successful, otherwise a positive errno value.*/ +int +ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port, + const struct ofproto_port_stp_settings *s) +{ + struct ofport *ofport = ofproto_get_port(ofproto, ofp_port); + if (!ofport) { + VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16, + ofproto->name, ofp_port); + return ENODEV; + } + + return (ofproto->ofproto_class->set_stp_port + ? ofproto->ofproto_class->set_stp_port(ofport, s) + : EOPNOTSUPP); +} + +/* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in + * 's'. If the 'enabled' member in 's' is false, then the other members + * are not meaningful. + * + * Returns 0 if successful, otherwise a positive errno value.*/ +int +ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port, + struct ofproto_port_stp_status *s) +{ + struct ofport *ofport = ofproto_get_port(ofproto, ofp_port); + if (!ofport) { + VLOG_WARN("%s: cannot get STP status on nonexistent port %"PRIu16, + ofproto->name, ofp_port); + return ENODEV; + } + + return (ofproto->ofproto_class->get_stp_port_status + ? ofproto->ofproto_class->get_stp_port_status(ofport, s) + : EOPNOTSUPP); +} + +/* Queue DSCP configuration. */ + +/* Registers meta-data associated with the 'n_qdscp' Qualities of Service + * 'queues' attached to 'ofport'. This data is not intended to be sufficient + * to implement QoS. Instead, it is used to implement features which require + * knowledge of what queues exist on a port, and some basic information about + * them. + * + * Returns 0 if successful, otherwise a positive errno value. */ +int +ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port, + const struct ofproto_port_queue *queues, + size_t n_queues) +{ + struct ofport *ofport = ofproto_get_port(ofproto, ofp_port); + + if (!ofport) { + VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16, + ofproto->name, ofp_port); + return ENODEV; + } + + return (ofproto->ofproto_class->set_queues + ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues) + : EOPNOTSUPP); +} + /* Connectivity Fault Management configuration. */ /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */ @@ -464,20 +708,19 @@ ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port) { struct ofport *ofport = ofproto_get_port(ofproto, ofp_port); if (ofport && ofproto->ofproto_class->set_cfm) { - ofproto->ofproto_class->set_cfm(ofport, NULL, NULL, 0); + ofproto->ofproto_class->set_cfm(ofport, NULL); } } /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes - * basic configuration from the configuration members in 'cfm', and the set of - * remote maintenance points from the 'n_remote_mps' elements in 'remote_mps'. - * Ignores the statistics members of 'cfm'. + * basic configuration from the configuration members in 'cfm', and the remote + * maintenance point ID from remote_mpid. Ignores the statistics members of + * 'cfm'. * * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */ void ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port, - const struct cfm *cfm, - const uint16_t *remote_mps, size_t n_remote_mps) + const struct cfm_settings *s) { struct ofport *ofport; int error; @@ -489,9 +732,11 @@ ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port, return; } + /* XXX: For configuration simplicity, we only support one remote_mpid + * outside of the CFM module. It's not clear if this is the correct long + * term solution or not. */ error = (ofproto->ofproto_class->set_cfm - ? ofproto->ofproto_class->set_cfm(ofport, cfm, - remote_mps, n_remote_mps) + ? ofproto->ofproto_class->set_cfm(ofport, s) : EOPNOTSUPP); if (error) { VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)", @@ -500,22 +745,6 @@ ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port, } } -/* Returns the connectivity fault management object associated with 'ofp_port' - * within 'ofproto', or a null pointer if 'ofproto' does not have a port - * 'ofp_port' or if that port does not have CFM configured. The caller must - * not modify or destroy the returned object. */ -const struct cfm * -ofproto_port_get_cfm(struct ofproto *ofproto, uint16_t ofp_port) -{ - struct ofport *ofport; - const struct cfm *cfm; - - ofport = ofproto_get_port(ofproto, ofp_port); - return (ofport - && ofproto->ofproto_class->get_cfm - && !ofproto->ofproto_class->get_cfm(ofport, &cfm)) ? cfm : NULL; -} - /* Checks the status of LACP negotiation for 'ofp_port' within ofproto. * Returns 1 if LACP partner information for 'ofp_port' is up-to-date, * 0 if LACP partner information is not current (generally indicating a @@ -562,10 +791,7 @@ ofproto_bundle_unregister(struct ofproto *ofproto, void *aux) /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'. * If 'aux' is already registered then this function updates its configuration - * to 's'. Otherwise, this function registers a new mirror. - * - * Mirrors affect only the treatment of packets output to the OFPP_NORMAL - * port. */ + * to 's'. Otherwise, this function registers a new mirror. */ int ofproto_mirror_register(struct ofproto *ofproto, void *aux, const struct ofproto_mirror_settings *s) @@ -583,6 +809,23 @@ ofproto_mirror_unregister(struct ofproto *ofproto, void *aux) return ofproto_mirror_register(ofproto, aux, NULL); } +/* Retrieves statistics from mirror associated with client data pointer + * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and + * 'bytes', respectively. If a particular counters is not supported, + * the appropriate argument is set to UINT64_MAX. */ +int +ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux, + uint64_t *packets, uint64_t *bytes) +{ + if (!ofproto->ofproto_class->mirror_get_stats) { + *packets = *bytes = UINT64_MAX; + return EOPNOTSUPP; + } + + return ofproto->ofproto_class->mirror_get_stats(ofproto, aux, + packets, bytes); +} + /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on * which all packets are flooded, instead of using MAC learning. If * 'flood_vlans' is NULL, then MAC learning applies to all VLANs. @@ -600,13 +843,66 @@ ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans) /* Returns true if 'aux' is a registered bundle that is currently in use as the * output for a mirror. */ bool -ofproto_is_mirror_output_bundle(struct ofproto *ofproto, void *aux) +ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux) { return (ofproto->ofproto_class->is_mirror_output_bundle ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux) : false); } +/* Configuration of OpenFlow tables. */ + +/* Returns the number of OpenFlow tables in 'ofproto'. */ +int +ofproto_get_n_tables(const struct ofproto *ofproto) +{ + return ofproto->n_tables; +} + +/* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the + * settings from 's'. 'table_id' must be in the range 0 through the number of + * OpenFlow tables in 'ofproto' minus 1, inclusive. + * + * For read-only tables, only the name may be configured. */ +void +ofproto_configure_table(struct ofproto *ofproto, int table_id, + const struct ofproto_table_settings *s) +{ + struct oftable *table; + + assert(table_id >= 0 && table_id < ofproto->n_tables); + table = &ofproto->tables[table_id]; + + oftable_set_name(table, s->name); + + if (table->flags & OFTABLE_READONLY) { + return; + } + + if (s->groups) { + oftable_enable_eviction(table, s->groups, s->n_groups); + } else { + oftable_disable_eviction(table); + } + + table->max_flows = s->max_flows; + if (classifier_count(&table->cls) > table->max_flows + && table->eviction_fields) { + /* 'table' contains more flows than allowed. We might not be able to + * evict them right away because of the asynchronous nature of flow + * table changes. Schedule eviction for later. */ + switch (ofproto->state) { + case S_OPENFLOW: + ofproto->state = S_EVICT; + break; + case S_EVICT: + case S_FLUSH: + /* We're already deleting flows, nothing more to do. */ + break; + } + } +} + bool ofproto_has_snoops(const struct ofproto *ofproto) { @@ -619,22 +915,66 @@ ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops) connmgr_get_snoops(ofproto->connmgr, snoops); } +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; + + if (table->flags & OFTABLE_HIDDEN) { + continue; + } + + 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); + oftable_remove_rule(rule); + ofproto->ofproto_class->rule_destruct(rule); + } + } + } + ofopgroup_submit(group); +} + static void ofproto_destroy__(struct ofproto *ofproto) { + struct oftable *table; + + assert(list_is_empty(&ofproto->pending)); + assert(!ofproto->n_pending); + connmgr_destroy(ofproto->connmgr); hmap_remove(&all_ofprotos, &ofproto->hmap_node); free(ofproto->name); + free(ofproto->type); free(ofproto->mfr_desc); free(ofproto->hw_desc); free(ofproto->sw_desc); free(ofproto->serial_desc); free(ofproto->dp_desc); - netdev_monitor_destroy(ofproto->netdev_monitor); hmap_destroy(&ofproto->ports); shash_destroy(&ofproto->port_by_name); - classifier_destroy(&ofproto->cls); + + OFPROTO_FOR_EACH_TABLE (table, ofproto) { + oftable_destroy(table); + } + free(ofproto->tables); + + hmap_destroy(&ofproto->deletions); + + free(ofproto->vlan_bitmap); ofproto->ofproto_class->dealloc(ofproto); } @@ -648,9 +988,8 @@ ofproto_destroy(struct ofproto *p) return; } - ofproto_flush_flows__(p); + ofproto_flush__(p); HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) { - hmap_remove(&p->ports, &ofport->hmap_node); ofport_destroy(ofport); } @@ -687,18 +1026,13 @@ process_port_change(struct ofproto *ofproto, int error, char *devname) int ofproto_run(struct ofproto *p) { + struct ofport *ofport; char *devname; int error; error = p->ofproto_class->run(p); - if (error == ENODEV) { - /* Someone destroyed the datapath behind our back. The caller - * better destroy us and give up, because we're just going to - * spin from here on out. */ - static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally", - p->name); - return ENODEV; + if (error && error != EAGAIN) { + VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error)); } if (p->ofproto_class->port_poll) { @@ -706,25 +1040,92 @@ ofproto_run(struct ofproto *p) process_port_change(p, error, devname); } } - while ((error = netdev_monitor_poll(p->netdev_monitor, - &devname)) != EAGAIN) { - process_port_change(p, error, devname); + + HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { + unsigned int change_seq = netdev_change_seq(ofport->netdev); + if (ofport->change_seq != change_seq) { + ofport->change_seq = change_seq; + update_port(p, netdev_get_name(ofport->netdev)); + } } - connmgr_run(p->connmgr, handle_openflow); + switch (p->state) { + case S_OPENFLOW: + connmgr_run(p->connmgr, handle_openflow); + break; - return 0; + case S_EVICT: + connmgr_run(p->connmgr, NULL); + ofproto_evict(p); + if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) { + p->state = S_OPENFLOW; + } + break; + + case S_FLUSH: + connmgr_run(p->connmgr, NULL); + ofproto_flush__(p); + if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) { + connmgr_flushed(p->connmgr); + p->state = S_OPENFLOW; + } + break; + + default: + NOT_REACHED(); + } + + return error; +} + +/* Performs periodic activity required by 'ofproto' that needs to be done + * with the least possible latency. + * + * It makes sense to call this function a couple of times per poll loop, to + * provide a significant performance boost on some benchmarks with the + * ofproto-dpif implementation. */ +int +ofproto_run_fast(struct ofproto *p) +{ + int error; + + error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0; + if (error && error != EAGAIN) { + VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)", + p->name, strerror(error)); + } + return error; } void ofproto_wait(struct ofproto *p) { + struct ofport *ofport; + p->ofproto_class->wait(p); if (p->ofproto_class->port_poll_wait) { p->ofproto_class->port_poll_wait(p); } - netdev_monitor_poll_wait(p->netdev_monitor); - connmgr_wait(p->connmgr); + + HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { + if (ofport->change_seq != netdev_change_seq(ofport->netdev)) { + poll_immediate_wake(); + } + } + + switch (p->state) { + case S_OPENFLOW: + connmgr_wait(p->connmgr, true); + break; + + case S_EVICT: + case S_FLUSH: + connmgr_wait(p->connmgr, false); + if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) { + poll_immediate_wake(); + } + break; + } } bool @@ -743,16 +1144,7 @@ ofproto_get_ofproto_controller_info(const struct ofproto *ofproto, void ofproto_free_ofproto_controller_info(struct shash *info) { - struct shash_node *node; - - SHASH_FOR_EACH (node, info) { - struct ofproto_controller_info *cinfo = node->data; - while (cinfo->pairs.n) { - free((char *) cinfo->pairs.values[--cinfo->pairs.n]); - } - free(cinfo); - } - shash_destroy(info); + connmgr_free_controller_info(info); } /* Makes a deep copy of 'old' into 'port'. */ @@ -897,7 +1289,7 @@ ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port) return error; } -/* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and +/* 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. * @@ -905,50 +1297,79 @@ ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port) * (0...65535, inclusive) then the flow will be visible to OpenFlow * controllers; otherwise, it will be hidden. * - * The caller retains ownership of 'cls_rule' and 'actions'. */ + * The caller retains ownership of 'cls_rule' and 'actions'. + * + * This is a helper function for in-band control and fail-open. */ void -ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule, +ofproto_add_flow(struct ofproto *ofproto, const struct cls_rule *cls_rule, const union ofp_action *actions, size_t n_actions) { - struct rule *rule; - rule_create(p, cls_rule, actions, n_actions, 0, 0, 0, false, &rule); -} + const struct rule *rule; -void -ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target) -{ - struct rule *rule; + rule = rule_from_cls_rule(classifier_find_rule_exactly( + &ofproto->tables[0].cls, cls_rule)); + if (!rule || !ofputil_actions_equal(rule->actions, rule->n_actions, + actions, n_actions)) { + struct ofputil_flow_mod fm; - rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls, - target)); - if (rule) { - ofproto_rule_remove(rule); + memset(&fm, 0, sizeof fm); + fm.cr = *cls_rule; + fm.buffer_id = UINT32_MAX; + fm.actions = (union ofp_action *) actions; + fm.n_actions = n_actions; + add_flow(ofproto, NULL, &fm, NULL); } } -static void -ofproto_flush_flows__(struct ofproto *ofproto) +/* Executes the flow modification specified in 'fm'. Returns 0 on success, an + * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the + * operation cannot be initiated now but may be retried later. + * + * This is a helper function for in-band control and fail-open. */ +int +ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm) { - struct rule *rule, *next_rule; - struct cls_cursor cursor; + return handle_flow_mod__(ofproto, NULL, fm, NULL); +} - COVERAGE_INC(ofproto_flush); +/* Searches for a rule with matching criteria exactly equal to 'target' in + * ofproto's table 0 and, if it finds one, deletes it. + * + * This is a helper function for in-band control and fail-open. */ +bool +ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target) +{ + struct rule *rule; - if (ofproto->ofproto_class->flush) { - ofproto->ofproto_class->flush(ofproto); + rule = rule_from_cls_rule(classifier_find_rule_exactly( + &ofproto->tables[0].cls, target)); + 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); + oftable_remove_rule(rule); + ofproto->ofproto_class->rule_destruct(rule); + ofopgroup_submit(group); + return true; } - cls_cursor_init(&cursor, &ofproto->cls, NULL); - CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { - ofproto_rule_remove(rule); - } } +/* Starts the process of deleting all of the flows from all of ofproto's flow + * tables and then reintroducing the flows required by in-band control and + * fail-open. The process will complete in a later call to ofproto_run(). */ void ofproto_flush_flows(struct ofproto *ofproto) { - ofproto_flush_flows__(ofproto); - connmgr_flushed(ofproto->connmgr); + COVERAGE_INC(ofproto_flush); + ofproto->state = S_FLUSH; } static void @@ -982,17 +1403,11 @@ static struct netdev * ofport_open(const struct ofproto_port *ofproto_port, struct ofp_phy_port *opp) { uint32_t curr, advertised, supported, peer; - struct netdev_options netdev_options; enum netdev_flags flags; struct netdev *netdev; int error; - memset(&netdev_options, 0, sizeof netdev_options); - netdev_options.name = ofproto_port->name; - netdev_options.type = ofproto_port->type; - netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; - - error = netdev_open(&netdev_options, &netdev); + error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev); if (error) { VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s " "cannot be opened (%s)", @@ -1042,6 +1457,7 @@ ofport_install(struct ofproto *p, { const char *netdev_name = netdev_get_name(netdev); struct ofport *ofport; + int dev_mtu; int error; /* Create ofport. */ @@ -1052,14 +1468,21 @@ ofport_install(struct ofproto *p, } ofport->ofproto = p; ofport->netdev = netdev; + ofport->change_seq = netdev_change_seq(netdev); ofport->opp = *opp; ofport->ofp_port = ntohs(opp->port_no); /* Add port to 'p'. */ - netdev_monitor_add(p->netdev_monitor, ofport->netdev); hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0)); shash_add(&p->port_by_name, netdev_name, ofport); + if (!netdev_get_mtu(netdev, &dev_mtu)) { + set_internal_devs_mtu(p); + ofport->mtu = dev_mtu; + } else { + ofport->mtu = 0; + } + /* Let the ofproto_class initialize its private data. */ error = p->ofproto_class->port_construct(ofport); if (error) { @@ -1098,7 +1521,7 @@ ofport_remove_with_name(struct ofproto *ofproto, const char *name) } } -/* Updates 'port' within 'ofproto' with the new 'netdev' and 'opp'. +/* Updates 'port' with new 'opp' description. * * Does not handle a name or port number change. The caller must implement * such a change as a delete followed by an add. */ @@ -1117,13 +1540,30 @@ ofport_modified(struct ofport *port, struct ofp_phy_port *opp) connmgr_send_port_status(port->ofproto->connmgr, &port->opp, OFPPR_MODIFY); } +/* Update OpenFlow 'state' in 'port' and notify controller. */ +void +ofproto_port_set_state(struct ofport *port, ovs_be32 state) +{ + if (port->opp.state != state) { + port->opp.state = state; + connmgr_send_port_status(port->ofproto->connmgr, &port->opp, + OFPPR_MODIFY); + } +} + void ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port) { struct ofport *port = ofproto_get_port(ofproto, ofp_port); if (port) { + if (port->ofproto->ofproto_class->set_realdev) { + port->ofproto->ofproto_class->set_realdev(port, 0, 0); + } + if (port->ofproto->ofproto_class->set_stp_port) { + port->ofproto->ofproto_class->set_stp_port(port, NULL); + } if (port->ofproto->ofproto_class->set_cfm) { - port->ofproto->ofproto_class->set_cfm(port, NULL, NULL, 0); + port->ofproto->ofproto_class->set_cfm(port, NULL); } if (port->ofproto->ofproto_class->bundle_remove) { port->ofproto->ofproto_class->bundle_remove(port); @@ -1137,7 +1577,6 @@ ofport_destroy__(struct ofport *port) struct ofproto *ofproto = port->ofproto; const char *name = netdev_get_name(port->netdev); - netdev_monitor_remove(ofproto->netdev_monitor, port->netdev); hmap_remove(&ofproto->ports, &port->hmap_node); shash_delete(&ofproto->port_by_name, shash_find(&ofproto->port_by_name, name)); @@ -1169,6 +1608,21 @@ ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port) return NULL; } +int +ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats) +{ + struct ofproto *ofproto = port->ofproto; + int error; + + if (ofproto->ofproto_class->port_get_stats) { + error = ofproto->ofproto_class->port_get_stats(port, stats); + } else { + error = EOPNOTSUPP; + } + + return error; +} + static void update_port(struct ofproto *ofproto, const char *name) { @@ -1186,21 +1640,34 @@ update_port(struct ofproto *ofproto, const char *name) if (netdev) { port = ofproto_get_port(ofproto, ofproto_port.ofp_port); if (port && !strcmp(netdev_get_name(port->netdev), name)) { + struct netdev *old_netdev = port->netdev; + int dev_mtu; + /* 'name' hasn't changed location. Any properties changed? */ if (!ofport_equal(&port->opp, &opp)) { ofport_modified(port, &opp); } - /* Install the newly opened netdev in case it has changed. */ - netdev_monitor_remove(ofproto->netdev_monitor, port->netdev); - netdev_monitor_add(ofproto->netdev_monitor, netdev); + /* If this is a non-internal port and the MTU changed, check + * if the datapath's MTU needs to be updated. */ + if (strcmp(netdev_get_type(netdev), "internal") + && !netdev_get_mtu(netdev, &dev_mtu) + && port->mtu != dev_mtu) { + set_internal_devs_mtu(ofproto); + port->mtu = dev_mtu; + } - netdev_close(port->netdev); + /* Install the newly opened netdev in case it has changed. + * Don't close the old netdev yet in case port_modified has to + * remove a retained reference to it.*/ port->netdev = netdev; + port->change_seq = netdev_change_seq(netdev); if (port->ofproto->ofproto_class->port_modified) { port->ofproto->ofproto_class->port_modified(port); } + + netdev_close(old_netdev); } else { /* If 'port' is nonnull then its name differs from 'name' and thus * we should delete it. If we think there's a port named 'name' @@ -1245,71 +1712,75 @@ init_ports(struct ofproto *p) return 0; } - -/* Creates a new rule initialized as specified, inserts it into 'ofproto''s - * flow table, and stores the new rule into '*rulep'. Returns 0 on success, - * otherwise a positive errno value or OpenFlow error code. */ + +/* Find the minimum MTU of all non-datapath devices attached to 'p'. + * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */ static int -rule_create(struct ofproto *ofproto, const struct cls_rule *cls_rule, - const union ofp_action *actions, size_t n_actions, - uint16_t idle_timeout, uint16_t hard_timeout, - ovs_be64 flow_cookie, bool send_flow_removed, - struct rule **rulep) +find_min_mtu(struct ofproto *p) { - struct rule *rule; - int error; + struct ofport *ofport; + int mtu = 0; - rule = ofproto->ofproto_class->rule_alloc(); - if (!rule) { - error = ENOMEM; - goto error; - } + HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { + struct netdev *netdev = ofport->netdev; + int dev_mtu; - rule->ofproto = ofproto; - rule->cr = *cls_rule; - rule->flow_cookie = flow_cookie; - rule->created = time_msec(); - rule->idle_timeout = idle_timeout; - rule->hard_timeout = hard_timeout; - rule->send_flow_removed = send_flow_removed; - if (n_actions > 0) { - rule->actions = xmemdup(actions, n_actions * sizeof *actions); - } else { - rule->actions = NULL; - } - rule->n_actions = n_actions; + /* Skip any internal ports, since that's what we're trying to + * set. */ + if (!strcmp(netdev_get_type(netdev), "internal")) { + continue; + } - error = ofproto->ofproto_class->rule_construct(rule); - if (error) { - ofproto_rule_destroy__(rule); - goto error; + if (netdev_get_mtu(netdev, &dev_mtu)) { + continue; + } + if (!mtu || dev_mtu < mtu) { + mtu = dev_mtu; + } } - *rulep = rule; - return 0; - -error: - VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)", - ofproto->name, strerror(error)); - *rulep = NULL; - return error; + return mtu ? mtu: ETH_PAYLOAD_MAX; } +/* Set the MTU of all datapath devices on 'p' to the minimum of the + * non-datapath ports. */ +static void +set_internal_devs_mtu(struct ofproto *p) +{ + struct ofport *ofport; + int mtu = find_min_mtu(p); + + HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { + struct netdev *netdev = ofport->netdev; + + if (!strcmp(netdev_get_type(netdev), "internal")) { + netdev_set_mtu(netdev, mtu); + } + } +} + static void ofproto_rule_destroy__(struct rule *rule) { - free(rule->actions); - rule->ofproto->ofproto_class->rule_dealloc(rule); + if (rule) { + free(rule->actions); + rule->ofproto->ofproto_class->rule_dealloc(rule); + } } -/* Destroys 'rule' and iterates through all of its facets and revalidates them, - * destroying any that no longer has a rule (which is probably all of them). +/* 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. * - * The caller must have already removed 'rule' from the classifier. */ + * 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) { - rule->ofproto->ofproto_class->rule_destruct(rule); + assert(!rule->pending); + oftable_remove_rule(rule); ofproto_rule_destroy__(rule); } @@ -1317,32 +1788,25 @@ ofproto_rule_destroy(struct rule *rule) * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't * count). */ static bool -rule_has_out_port(const struct rule *rule, ovs_be16 out_port) +rule_has_out_port(const struct rule *rule, uint16_t out_port) { const union ofp_action *oa; - struct actions_iterator i; + size_t left; - if (out_port == htons(OFPP_NONE)) { + if (out_port == OFPP_NONE) { return true; } - for (oa = actions_first(&i, rule->actions, rule->n_actions); oa; - oa = actions_next(&i)) { - if (action_outputs_to_port(oa, out_port)) { + OFPUTIL_ACTION_FOR_EACH_UNSAFE (oa, left, rule->actions, rule->n_actions) { + if (action_outputs_to_port(oa, htons(out_port))) { return true; } } return false; } -struct rule * -ofproto_rule_lookup(struct ofproto *ofproto, const struct flow *flow) -{ - return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow)); -} - /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s - * statistics (or the statistics for one of its facets) appropriately. - * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom. + * statistics appropriately. 'packet' must have at least sizeof(struct + * ofp_packet_in) bytes of headroom. * * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited * with statistics for 'packet' either way. @@ -1355,24 +1819,10 @@ rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet) assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); - flow_extract(packet, 0, in_port, &flow); + flow_extract(packet, 0, 0, in_port, &flow); return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet); } -/* Remove 'rule' from 'ofproto' and free up the associated memory: - * - * - Removes 'rule' from the classifier. - * - * - If 'rule' has facets, revalidates them (and possibly uninstalls and - * destroys them), via rule_destroy(). - */ -void -ofproto_rule_remove(struct rule *rule) -{ - rule->ofproto->ofproto_class->rule_remove(rule); - ofproto_rule_destroy(rule); -} - /* Returns true if 'rule' should be hidden from the controller. * * Rules with priority higher than UINT16_MAX are set up by ofproto itself @@ -1383,51 +1833,49 @@ rule_is_hidden(const struct rule *rule) { return rule->cr.priority > UINT16_MAX; } - -static void -send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh, - int error) + +static enum oftable_flags +rule_get_flags(const struct rule *rule) { - struct ofpbuf *buf = ofputil_encode_error_msg(error, oh); - if (buf) { - COVERAGE_INC(ofproto_error); - ofconn_send_reply(ofconn, buf); - } + return rule->ofproto->tables[rule->table_id].flags; } -static int +static bool +rule_is_modifiable(const struct rule *rule) +{ + return !(rule_get_flags(rule) & OFTABLE_READONLY); +} + +static enum ofperr handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh) { ofconn_send_reply(ofconn, make_echo_reply(oh)); return 0; } -static int +static enum ofperr handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); struct ofp_switch_features *osf; struct ofpbuf *buf; struct ofport *port; + bool arp_match_ip; + uint32_t actions; + + ofproto->ofproto_class->get_features(ofproto, &arp_match_ip, &actions); + assert(actions & (1 << OFPAT_OUTPUT)); /* sanity check */ osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf); osf->datapath_id = htonll(ofproto->datapath_id); osf->n_buffers = htonl(pktbuf_capacity()); - osf->n_tables = 2; + osf->n_tables = ofproto->n_tables; osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS | - OFPC_PORT_STATS | OFPC_ARP_MATCH_IP); - osf->actions = htonl((1u << OFPAT_OUTPUT) | - (1u << OFPAT_SET_VLAN_VID) | - (1u << OFPAT_SET_VLAN_PCP) | - (1u << OFPAT_STRIP_VLAN) | - (1u << OFPAT_SET_DL_SRC) | - (1u << OFPAT_SET_DL_DST) | - (1u << OFPAT_SET_NW_SRC) | - (1u << OFPAT_SET_NW_DST) | - (1u << OFPAT_SET_NW_TOS) | - (1u << OFPAT_SET_TP_SRC) | - (1u << OFPAT_SET_TP_DST) | - (1u << OFPAT_ENQUEUE)); + OFPC_PORT_STATS | OFPC_QUEUE_STATS); + if (arp_match_ip) { + osf->capabilities |= htonl(OFPC_ARP_MATCH_IP); + } + osf->actions = htonl(actions); HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) { ofpbuf_put(buf, &port->opp, sizeof port->opp); @@ -1437,21 +1885,20 @@ handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh) return 0; } -static int +static enum ofperr handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - struct ofpbuf *buf; struct ofp_switch_config *osc; - uint16_t flags; - bool drop_frags; - - /* Figure out flags. */ - drop_frags = ofproto->ofproto_class->get_drop_frags(ofproto); - flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL; + enum ofp_config_flags flags; + struct ofpbuf *buf; /* Send reply. */ osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf); + flags = ofproto->frag_handling; + if (ofconn_get_invalid_ttl_to_controller(ofconn)) { + flags |= OFPC_INVALID_TTL_TO_CONTROLLER; + } osc->flags = htons(flags); osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn)); ofconn_send_reply(ofconn, buf); @@ -1459,27 +1906,30 @@ handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh) return 0; } -static int +static enum ofperr handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); uint16_t flags = ntohs(osc->flags); - if (ofconn_get_type(ofconn) == OFCONN_PRIMARY - && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) { - switch (flags & OFPC_FRAG_MASK) { - case OFPC_FRAG_NORMAL: - ofproto->ofproto_class->set_drop_frags(ofproto, false); - break; - case OFPC_FRAG_DROP: - ofproto->ofproto_class->set_drop_frags(ofproto, true); - break; - default: - VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")", - osc->flags); - break; + if (ofconn_get_type(ofconn) != OFCONN_PRIMARY + || ofconn_get_role(ofconn) != NX_ROLE_SLAVE) { + enum ofp_config_flags cur = ofproto->frag_handling; + enum ofp_config_flags next = flags & OFPC_FRAG_MASK; + + assert((cur & OFPC_FRAG_MASK) == cur); + if (cur != next) { + if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) { + ofproto->frag_handling = next; + } else { + VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s", + ofproto->name, + ofputil_frag_handling_to_string(next)); + } } } + ofconn_set_invalid_ttl_to_controller(ofconn, + (flags & OFPC_INVALID_TTL_TO_CONTROLLER)); ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len)); @@ -1487,74 +1937,59 @@ handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc) } /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow - * error message code (composed with ofp_mkerr()) for the caller to propagate - * upward. Otherwise, returns 0. + * error message code for the caller to propagate upward. Otherwise, returns + * 0. * * The log message mentions 'msg_type'. */ -static int -reject_slave_controller(struct ofconn *ofconn, const const char *msg_type) +static enum ofperr +reject_slave_controller(struct ofconn *ofconn) { if (ofconn_get_type(ofconn) == OFCONN_PRIMARY && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) { - static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller", - msg_type); - - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); + return OFPERR_OFPBRC_EPERM; } else { return 0; } } -static int -handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh) +static enum ofperr +handle_packet_out(struct ofconn *ofconn, const struct ofp_packet_out *opo) { struct ofproto *p = ofconn_get_ofproto(ofconn); - struct ofp_packet_out *opo; - struct ofpbuf payload, *buffer; - union ofp_action *ofp_actions; - struct ofpbuf request; + struct ofputil_packet_out po; + struct ofpbuf *payload; struct flow flow; - size_t n_ofp_actions; - uint16_t in_port; - int error; + enum ofperr error; COVERAGE_INC(ofproto_packet_out); - error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT"); + error = reject_slave_controller(ofconn); if (error) { return error; } - /* Get ofp_packet_out. */ - ofpbuf_use_const(&request, oh, ntohs(oh->length)); - opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions)); - - /* Get actions. */ - error = ofputil_pull_actions(&request, ntohs(opo->actions_len), - &ofp_actions, &n_ofp_actions); + /* Decode message. */ + error = ofputil_decode_packet_out(&po, opo); if (error) { return error; } /* Get payload. */ - if (opo->buffer_id != htonl(UINT32_MAX)) { - error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id), - &buffer, &in_port); - if (error || !buffer) { + if (po.buffer_id != UINT32_MAX) { + error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL); + if (error || !payload) { return error; } - payload = *buffer; } else { - payload = request; - buffer = NULL; + payload = xmalloc(sizeof *payload); + ofpbuf_use_const(payload, po.packet, po.packet_len); } /* Send out packet. */ - flow_extract(&payload, 0, ntohs(opo->in_port), &flow); - error = p->ofproto_class->packet_out(p, &payload, &flow, - ofp_actions, n_ofp_actions); - ofpbuf_delete(buffer); + flow_extract(payload, 0, 0, po.in_port, &flow); + error = p->ofproto_class->packet_out(p, payload, &flow, + po.actions, po.n_actions); + ofpbuf_delete(payload); return error; } @@ -1581,7 +2016,7 @@ update_port_config(struct ofport *port, ovs_be32 config, ovs_be32 mask) } } -static int +static enum ofperr handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh) { struct ofproto *p = ofconn_get_ofproto(ofconn); @@ -1589,16 +2024,16 @@ handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh) struct ofport *port; int error; - error = reject_slave_controller(ofconn, "OFPT_PORT_MOD"); + error = reject_slave_controller(ofconn); if (error) { return error; } port = ofproto_get_port(p, ntohs(opm->port_no)); if (!port) { - return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT); + return OFPERR_OFPPMFC_BAD_PORT; } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) { - return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR); + return OFPERR_OFPPMFC_BAD_HW_ADDR; } else { update_port_config(port, opm->config, opm->mask); if (opm->advertise) { @@ -1608,89 +2043,15 @@ handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh) return 0; } -static struct ofpbuf * -make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len) -{ - struct ofp_stats_reply *osr; - struct ofpbuf *msg; - - msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX)); - osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg); - osr->type = type; - osr->flags = htons(0); - return msg; -} - -static struct ofpbuf * -start_ofp_stats_reply(const struct ofp_header *request, size_t body_len) -{ - const struct ofp_stats_request *osr - = (const struct ofp_stats_request *) request; - return make_ofp_stats_reply(osr->header.xid, osr->type, body_len); -} - -static void * -append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn, - struct ofpbuf **msgp) -{ - struct ofpbuf *msg = *msgp; - assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply)); - if (nbytes + msg->size > UINT16_MAX) { - struct ofp_stats_reply *reply = msg->data; - reply->flags = htons(OFPSF_REPLY_MORE); - *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes); - ofconn_send_reply(ofconn, msg); - } - return ofpbuf_put_uninit(*msgp, nbytes); -} - -static struct ofpbuf * -make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len) -{ - struct nicira_stats_msg *nsm; - struct ofpbuf *msg; - - msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX)); - nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg); - nsm->type = htons(OFPST_VENDOR); - nsm->flags = htons(0); - nsm->vendor = htonl(NX_VENDOR_ID); - nsm->subtype = subtype; - return msg; -} - -static struct ofpbuf * -start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len) -{ - return make_nxstats_reply(request->header.xid, request->subtype, body_len); -} - -static void -append_nxstats_reply(size_t nbytes, struct ofconn *ofconn, - struct ofpbuf **msgp) -{ - struct ofpbuf *msg = *msgp; - assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg)); - if (nbytes + msg->size > UINT16_MAX) { - struct nicira_stats_msg *reply = msg->data; - reply->flags = htons(OFPSF_REPLY_MORE); - *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes); - ofconn_send_reply(ofconn, msg); - } - ofpbuf_prealloc_tailroom(*msgp, nbytes); -} - -static int +static enum ofperr handle_desc_stats_request(struct ofconn *ofconn, - const struct ofp_header *request) + const struct ofp_stats_msg *request) { struct ofproto *p = ofconn_get_ofproto(ofconn); struct ofp_desc_stats *ods; struct ofpbuf *msg; - msg = start_ofp_stats_reply(request, sizeof *ods); - ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg); - memset(ods, 0, sizeof *ods); + ods = ofputil_make_stats_reply(sizeof *ods, request, &msg); ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc); ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc); ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc); @@ -1701,34 +2062,46 @@ handle_desc_stats_request(struct ofconn *ofconn, return 0; } -static int +static enum ofperr handle_table_stats_request(struct ofconn *ofconn, - const struct ofp_header *request) + const struct ofp_stats_msg *request) { struct ofproto *p = ofconn_get_ofproto(ofconn); struct ofp_table_stats *ots; struct ofpbuf *msg; + size_t i; + + ofputil_make_stats_reply(sizeof(struct ofp_stats_msg), request, &msg); + + ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables); + for (i = 0; i < p->n_tables; i++) { + ots[i].table_id = i; + sprintf(ots[i].name, "table%zu", i); + ots[i].wildcards = htonl(OFPFW_ALL); + ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */ + ots[i].active_count = htonl(classifier_count(&p->tables[i].cls)); + } + + p->ofproto_class->get_tables(p, ots); - msg = start_ofp_stats_reply(request, sizeof *ots * 2); + for (i = 0; i < p->n_tables; i++) { + const struct oftable *table = &p->tables[i]; - /* Classifier table. */ - ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg); - memset(ots, 0, sizeof *ots); - strcpy(ots->name, "classifier"); - ots->wildcards = (ofconn_get_flow_format(ofconn) == NXFF_OPENFLOW10 - ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL)); - ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */ - ots->active_count = htonl(classifier_count(&p->cls)); - put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */ - put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */ + if (table->name) { + ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name); + } + + if (table->max_flows < ntohl(ots[i].max_entries)) { + ots[i].max_entries = htonl(table->max_flows); + } + } ofconn_send_reply(ofconn, msg); return 0; } static void -append_port_stat(struct ofport *port, struct ofconn *ofconn, - struct ofpbuf **msgp) +append_port_stat(struct ofport *port, struct list *replies) { struct netdev_stats stats; struct ofp_port_stats *ops; @@ -1736,9 +2109,9 @@ append_port_stat(struct ofport *port, struct ofconn *ofconn, /* Intentionally ignore return value, since errors will set * 'stats' to all-1s, which is correct for OpenFlow, and * netdev_get_stats() will log errors. */ - netdev_get_stats(port->netdev, &stats); + ofproto_port_get_stats(port, &stats); - ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp); + ops = ofputil_append_stats_reply(sizeof *ops, replies); ops->port_no = port->opp.port_no; memset(ops->pad, 0, sizeof ops->pad); put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets)); @@ -1755,202 +2128,249 @@ append_port_stat(struct ofport *port, struct ofconn *ofconn, put_32aligned_be64(&ops->collisions, htonll(stats.collisions)); } -static int -handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) +static enum ofperr +handle_port_stats_request(struct ofconn *ofconn, + const struct ofp_port_stats_request *psr) { struct ofproto *p = ofconn_get_ofproto(ofconn); - const struct ofp_port_stats_request *psr = ofputil_stats_body(oh); - struct ofp_port_stats *ops; - struct ofpbuf *msg; struct ofport *port; + struct list replies; - msg = start_ofp_stats_reply(oh, sizeof *ops * 16); + ofputil_start_stats_reply(&psr->osm, &replies); if (psr->port_no != htons(OFPP_NONE)) { port = ofproto_get_port(p, ntohs(psr->port_no)); if (port) { - append_port_stat(port, ofconn, &msg); + append_port_stat(port, &replies); } } else { HMAP_FOR_EACH (port, hmap_node, &p->ports) { - append_port_stat(port, ofconn, &msg); + append_port_stat(port, &replies); } } - ofconn_send_reply(ofconn, msg); + ofconn_send_replies(ofconn, &replies); return 0; } static void -calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec) +calc_flow_duration__(long long int start, long long int now, + uint32_t *sec, uint32_t *nsec) { - long long int msecs = time_msec() - start; + long long int msecs = now - start; *sec = msecs / 1000; *nsec = (msecs % 1000) * (1000 * 1000); } -static void -calc_flow_duration(long long int start, ovs_be32 *sec_be, ovs_be32 *nsec_be) +/* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns + * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */ +static enum ofperr +check_table_id(const struct ofproto *ofproto, uint8_t table_id) { - uint32_t sec, nsec; + return (table_id == 0xff || table_id < ofproto->n_tables + ? 0 + : OFPERR_NXBRC_BAD_TABLE_ID); - calc_flow_duration__(start, &sec, &nsec); - *sec_be = htonl(sec); - *nsec_be = htonl(nsec); } -static void -put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule, - ovs_be16 out_port, struct ofpbuf **replyp) +static struct oftable * +next_visible_table(struct ofproto *ofproto, uint8_t table_id) { - struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - struct ofp_flow_stats *ofs; - uint64_t packet_count, byte_count; - ovs_be64 cookie; - size_t act_len, len; + struct oftable *table; - if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) { - return; + for (table = &ofproto->tables[table_id]; + table < &ofproto->tables[ofproto->n_tables]; + table++) { + if (!(table->flags & OFTABLE_HIDDEN)) { + return table; + } } - act_len = sizeof *rule->actions * rule->n_actions; - len = offsetof(struct ofp_flow_stats, actions) + act_len; - - ofproto->ofproto_class->rule_get_stats(rule, &packet_count, &byte_count); - - ofs = append_ofp_stats_reply(len, ofconn, replyp); - ofs->length = htons(len); - ofs->table_id = 0; - ofs->pad = 0; - ofputil_cls_rule_to_match(&rule->cr, ofconn_get_flow_format(ofconn), - &ofs->match, rule->flow_cookie, &cookie); - put_32aligned_be64(&ofs->cookie, cookie); - calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec); - ofs->priority = htons(rule->cr.priority); - ofs->idle_timeout = htons(rule->idle_timeout); - ofs->hard_timeout = htons(rule->hard_timeout); - memset(ofs->pad2, 0, sizeof ofs->pad2); - put_32aligned_be64(&ofs->packet_count, htonll(packet_count)); - put_32aligned_be64(&ofs->byte_count, htonll(byte_count)); - if (rule->n_actions > 0) { - memcpy(ofs->actions, rule->actions, act_len); - } + return NULL; } -static bool -is_valid_table(uint8_t table_id) +static struct oftable * +first_matching_table(struct ofproto *ofproto, uint8_t table_id) { - if (table_id == 0 || table_id == 0xff) { - return true; + if (table_id == 0xff) { + return next_visible_table(ofproto, 0); + } else if (table_id < ofproto->n_tables) { + return &ofproto->tables[table_id]; } else { - /* It would probably be better to reply with an error but there doesn't - * seem to be any appropriate value, so that might just be - * confusing. */ - VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8, - table_id); - return false; + return NULL; } } -static int -handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) +static struct oftable * +next_matching_table(struct ofproto *ofproto, + struct oftable *table, uint8_t table_id) { - const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh); - struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - struct ofpbuf *reply; + return (table_id == 0xff + ? next_visible_table(ofproto, (table - ofproto->tables) + 1) + : NULL); +} + +/* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO: + * + * - If TABLE_ID is 0xff, this iterates over every classifier table in + * OFPROTO, skipping tables marked OFTABLE_HIDDEN. + * + * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates + * only once, for that table. (This can be used to access tables marked + * OFTABLE_HIDDEN.) + * + * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be + * entered at all. (Perhaps you should have validated TABLE_ID with + * check_table_id().) + * + * All parameters are evaluated multiple times. + */ +#define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \ + for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \ + (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'. + * + * If 'out_port' is anything other than OFPP_NONE, then only rules that output + * to 'out_port' are included. + * + * 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 cls_rule *match, + ovs_be64 cookie, ovs_be64 cookie_mask, + uint16_t out_port, struct list *rules) +{ + struct oftable *table; + enum ofperr error; + + error = check_table_id(ofproto, table_id); + if (error) { + return error; + } - COVERAGE_INC(ofproto_flows_req); - reply = start_ofp_stats_reply(oh, 1024); - if (is_valid_table(fsr->table_id)) { + list_init(rules); + FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) { struct cls_cursor cursor; - struct cls_rule target; struct rule *rule; - ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0, - &target); - cls_cursor_init(&cursor, &ofproto->cls, &target); + cls_cursor_init(&cursor, &table->cls, match); CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { - put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply); + if (rule->pending) { + return OFPROTO_POSTPONE; + } + if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port) + && !((rule->flow_cookie ^ cookie) & cookie_mask)) { + list_push_back(rules, &rule->ofproto_node); + } } } - ofconn_send_reply(ofconn, reply); - return 0; } -static void -put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule, - ovs_be16 out_port, struct ofpbuf **replyp) -{ - struct nx_flow_stats *nfs; - uint64_t packet_count, byte_count; - size_t act_len, start_len; - struct ofpbuf *reply; +/* 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_NONE, then only rules that output + * to 'out_port' are included. + * + * 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 cls_rule *match, + ovs_be64 cookie, ovs_be64 cookie_mask, + uint16_t out_port, struct list *rules) +{ + struct oftable *table; + int error; - if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) { - return; + error = check_table_id(ofproto, table_id); + if (error) { + return error; } - rule->ofproto->ofproto_class->rule_get_stats(rule, - &packet_count, &byte_count); + list_init(rules); + FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) { + struct rule *rule; - act_len = sizeof *rule->actions * rule->n_actions; - - append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp); - start_len = (*replyp)->size; - reply = *replyp; - - nfs = ofpbuf_put_uninit(reply, sizeof *nfs); - nfs->table_id = 0; - nfs->pad = 0; - calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec); - nfs->cookie = rule->flow_cookie; - nfs->priority = htons(rule->cr.priority); - nfs->idle_timeout = htons(rule->idle_timeout); - nfs->hard_timeout = htons(rule->hard_timeout); - nfs->match_len = htons(nx_put_match(reply, &rule->cr)); - memset(nfs->pad2, 0, sizeof nfs->pad2); - nfs->packet_count = htonll(packet_count); - nfs->byte_count = htonll(byte_count); - if (rule->n_actions > 0) { - ofpbuf_put(reply, rule->actions, act_len); + rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, + match)); + if (rule) { + if (rule->pending) { + return OFPROTO_POSTPONE; + } + if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port) + && !((rule->flow_cookie ^ cookie) & cookie_mask)) { + list_push_back(rules, &rule->ofproto_node); + } + } } - nfs->length = htons(reply->size - start_len); + return 0; } +/* Returns 'age_ms' (a duration in milliseconds), converted to seconds and + * forced into the range of a uint16_t. */ static int -handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh) +age_secs(long long int age_ms) { - struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - struct nx_flow_stats_request *nfsr; - struct cls_rule target; - struct ofpbuf *reply; - struct ofpbuf b; - int error; + return (age_ms < 0 ? 0 + : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX + : (unsigned int) age_ms / 1000); +} - ofpbuf_use_const(&b, oh, ntohs(oh->length)); +static enum ofperr +handle_flow_stats_request(struct ofconn *ofconn, + const struct ofp_stats_msg *osm) +{ + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); + struct ofputil_flow_stats_request fsr; + struct list replies; + struct list rules; + struct rule *rule; + enum ofperr error; - /* Dissect the message. */ - nfsr = ofpbuf_pull(&b, sizeof *nfsr); - error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target); + error = ofputil_decode_flow_stats_request(&fsr, &osm->header); if (error) { return error; } - if (b.size) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); - } - - COVERAGE_INC(ofproto_flows_req); - reply = start_nxstats_reply(&nfsr->nsm, 1024); - if (is_valid_table(nfsr->table_id)) { - struct cls_cursor cursor; - struct rule *rule; - cls_cursor_init(&cursor, &ofproto->cls, &target); - CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { - put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply); - } + error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match, + fsr.cookie, fsr.cookie_mask, + fsr.out_port, &rules); + if (error) { + return error; } - ofconn_send_reply(ofconn, reply); + + ofputil_start_stats_reply(osm, &replies); + LIST_FOR_EACH (rule, ofproto_node, &rules) { + long long int now = time_msec(); + struct ofputil_flow_stats fs; + + fs.rule = rule->cr; + fs.cookie = rule->flow_cookie; + fs.table_id = rule->table_id; + calc_flow_duration__(rule->created, now, &fs.duration_sec, + &fs.duration_nsec); + fs.idle_timeout = rule->idle_timeout; + fs.hard_timeout = rule->hard_timeout; + 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.actions = rule->actions; + fs.n_actions = rule->n_actions; + ofputil_append_flow_stats_reply(&fs, &replies); + } + ofconn_send_replies(ofconn, &replies); return 0; } @@ -1959,11 +2379,13 @@ static void flow_stats_ds(struct rule *rule, struct ds *results) { uint64_t packet_count, byte_count; - size_t act_len = sizeof *rule->actions * rule->n_actions; rule->ofproto->ofproto_class->rule_get_stats(rule, &packet_count, &byte_count); + if (rule->table_id != 0) { + ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id); + } ds_put_format(results, "duration=%llds, ", (time_msec() - rule->created) / 1000); ds_put_format(results, "priority=%u, ", rule->cr.priority); @@ -1971,8 +2393,8 @@ flow_stats_ds(struct rule *rule, struct ds *results) ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count); cls_rule_format(&rule->cr, results); ds_put_char(results, ','); - if (act_len > 0) { - ofp_print_actions(results, &rule->actions->header, act_len); + if (rule->n_actions > 0) { + ofp_print_actions(results, rule->actions, rule->n_actions); } else { ds_put_cstr(results, "drop"); } @@ -1984,12 +2406,16 @@ flow_stats_ds(struct rule *rule, struct ds *results) void ofproto_get_all_flows(struct ofproto *p, struct ds *results) { - struct cls_cursor cursor; - struct rule *rule; + struct oftable *table; - cls_cursor_init(&cursor, &p->cls, NULL); - CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { - flow_stats_ds(rule, results); + OFPROTO_FOR_EACH_TABLE (table, p) { + struct cls_cursor cursor; + struct rule *rule; + + cls_cursor_init(&cursor, &table->cls, NULL); + CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { + flow_stats_ds(rule, results); + } } } @@ -2002,102 +2428,102 @@ ofproto_get_netflow_ids(const struct ofproto *ofproto, ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id); } -static void -query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target, - ovs_be16 out_port, uint8_t table_id, - struct ofp_aggregate_stats_reply *oasr) +/* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns a + * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally + * indicating a connectivity problem). Returns zero if CFM is not faulted, + * and -1 if CFM is not enabled on 'port'. */ +int +ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port) { - uint64_t total_packets = 0; - uint64_t total_bytes = 0; - int n_flows = 0; - - COVERAGE_INC(ofproto_agg_request); - - if (is_valid_table(table_id)) { - struct cls_cursor cursor; - struct rule *rule; - - cls_cursor_init(&cursor, &ofproto->cls, target); - CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { - if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) { - uint64_t packet_count; - uint64_t byte_count; - - ofproto->ofproto_class->rule_get_stats(rule, &packet_count, - &byte_count); - - total_packets += packet_count; - total_bytes += byte_count; - n_flows++; - } - } - } - - oasr->flow_count = htonl(n_flows); - put_32aligned_be64(&oasr->packet_count, htonll(total_packets)); - put_32aligned_be64(&oasr->byte_count, htonll(total_bytes)); - memset(oasr->pad, 0, sizeof oasr->pad); + struct ofport *ofport = ofproto_get_port(ofproto, ofp_port); + return (ofport && ofproto->ofproto_class->get_cfm_fault + ? ofproto->ofproto_class->get_cfm_fault(ofport) + : -1); } -static int -handle_aggregate_stats_request(struct ofconn *ofconn, - const struct ofp_header *oh) +/* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port' + * within 'ofproto'. Populates 'rmps' with an array of MPIDs owned by + * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'. Returns a + * number less than 0 if CFM is not enabled on 'ofp_port'. */ +int +ofproto_port_get_cfm_remote_mpids(const struct ofproto *ofproto, + uint16_t ofp_port, const uint64_t **rmps, + size_t *n_rmps) { - const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh); - struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - struct ofp_aggregate_stats_reply *reply; - struct cls_rule target; - struct ofpbuf *msg; - - ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0, - &target); + struct ofport *ofport = ofproto_get_port(ofproto, ofp_port); - msg = start_ofp_stats_reply(oh, sizeof *reply); - reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg); - query_aggregate_stats(ofproto, &target, request->out_port, - request->table_id, reply); - ofconn_send_reply(ofconn, msg); - return 0; + *rmps = NULL; + *n_rmps = 0; + return (ofport && ofproto->ofproto_class->get_cfm_remote_mpids + ? ofproto->ofproto_class->get_cfm_remote_mpids(ofport, rmps, + n_rmps) + : -1); } -static int -handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh) +static enum ofperr +handle_aggregate_stats_request(struct ofconn *ofconn, + const struct ofp_stats_msg *osm) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - struct nx_aggregate_stats_request *request; - struct ofp_aggregate_stats_reply *reply; - struct cls_rule target; - struct ofpbuf b; - struct ofpbuf *buf; - int error; + struct ofputil_flow_stats_request request; + struct ofputil_aggregate_stats stats; + bool unknown_packets, unknown_bytes; + struct ofpbuf *reply; + struct list rules; + struct rule *rule; + enum ofperr error; - ofpbuf_use_const(&b, oh, ntohs(oh->length)); + error = ofputil_decode_flow_stats_request(&request, &osm->header); + if (error) { + return error; + } - /* Dissect the message. */ - request = ofpbuf_pull(&b, sizeof *request); - error = nx_pull_match(&b, ntohs(request->match_len), 0, &target); + error = collect_rules_loose(ofproto, request.table_id, &request.match, + request.cookie, request.cookie_mask, + request.out_port, &rules); if (error) { return error; } - if (b.size) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); + + memset(&stats, 0, sizeof stats); + unknown_packets = unknown_bytes = false; + LIST_FOR_EACH (rule, ofproto_node, &rules) { + uint64_t packet_count; + uint64_t byte_count; + + ofproto->ofproto_class->rule_get_stats(rule, &packet_count, + &byte_count); + + if (packet_count == UINT64_MAX) { + unknown_packets = true; + } else { + stats.packet_count += packet_count; + } + + if (byte_count == UINT64_MAX) { + unknown_bytes = true; + } else { + stats.byte_count += byte_count; + } + + stats.flow_count++; + } + if (unknown_packets) { + stats.packet_count = UINT64_MAX; + } + if (unknown_bytes) { + stats.byte_count = UINT64_MAX; } - /* Reply. */ - COVERAGE_INC(ofproto_flows_req); - buf = start_nxstats_reply(&request->nsm, sizeof *reply); - reply = ofpbuf_put_uninit(buf, sizeof *reply); - query_aggregate_stats(ofproto, &target, request->out_port, - request->table_id, reply); - ofconn_send_reply(ofconn, buf); + reply = ofputil_encode_aggregate_stats_reply(&stats, osm); + ofconn_send_reply(ofconn, reply); return 0; } struct queue_stats_cbdata { - struct ofconn *ofconn; struct ofport *ofport; - struct ofpbuf *msg; + struct list replies; }; static void @@ -2106,7 +2532,7 @@ put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id, { struct ofp_queue_stats *reply; - reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg); + reply = ofputil_append_stats_reply(sizeof *reply, &cbdata->replies); reply->port_no = cbdata->ofport->opp.port_no; memset(reply->pad, 0, sizeof reply->pad); reply->queue_id = htonl(queue_id); @@ -2142,25 +2568,19 @@ handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id, } } -static int -handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) +static enum ofperr +handle_queue_stats_request(struct ofconn *ofconn, + const struct ofp_queue_stats_request *qsr) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); - const struct ofp_queue_stats_request *qsr; struct queue_stats_cbdata cbdata; struct ofport *port; unsigned int port_no; uint32_t queue_id; - qsr = ofputil_stats_body(oh); - if (!qsr) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); - } - COVERAGE_INC(ofproto_queue_req); - cbdata.ofconn = ofconn; - cbdata.msg = start_ofp_stats_reply(oh, 128); + ofputil_start_stats_reply(&qsr->osm, &cbdata.replies); port_no = ntohs(qsr->port_no); queue_id = ntohl(qsr->queue_id); @@ -2174,232 +2594,322 @@ handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) handle_queue_stats_for_port(port, queue_id, &cbdata); } } else { - ofpbuf_delete(cbdata.msg); - return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT); + ofpbuf_list_delete(&cbdata.replies); + return OFPERR_OFPQOFC_BAD_PORT; } - ofconn_send_reply(ofconn, cbdata.msg); + ofconn_send_replies(ofconn, &cbdata.replies); return 0; } +static bool +is_flow_deletion_pending(const struct ofproto *ofproto, + const struct cls_rule *cls_rule, + uint8_t table_id) +{ + if (!hmap_is_empty(&ofproto->deletions)) { + struct ofoperation *op; + + HMAP_FOR_EACH_WITH_HASH (op, hmap_node, + cls_rule_hash(cls_rule, table_id), + &ofproto->deletions) { + if (cls_rule_equal(cls_rule, &op->rule->cr)) { + return true; + } + } + } + + return false; +} + /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT * in which no matching flow already exists in the flow table. * * Adds the flow specified by 'ofm', which is followed by 'n_actions' - * ofp_actions, to the ofproto's flow table. Returns 0 on success or an - * OpenFlow error code as encoded by ofp_mkerr() on failure. + * 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. * * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, * if any. */ -static int -add_flow(struct ofconn *ofconn, struct flow_mod *fm) +static enum ofperr +add_flow(struct ofproto *ofproto, struct ofconn *ofconn, + const struct ofputil_flow_mod *fm, const struct ofp_header *request) { - struct ofproto *p = ofconn_get_ofproto(ofconn); - struct ofpbuf *packet; + struct oftable *table; + struct ofopgroup *group; + struct rule *victim; struct rule *rule; - uint16_t in_port; - int buf_err; int error; - if (fm->flags & OFPFF_CHECK_OVERLAP - && classifier_rule_overlaps(&p->cls, &fm->cr)) { - return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP); - } - - buf_err = ofconn_pktbuf_retrieve(ofconn, fm->buffer_id, &packet, &in_port); - error = rule_create(p, &fm->cr, fm->actions, fm->n_actions, - fm->idle_timeout, fm->hard_timeout, fm->cookie, - fm->flags & OFPFF_SEND_FLOW_REM, &rule); + error = check_table_id(ofproto, fm->table_id); if (error) { - ofpbuf_delete(packet); return error; } - if (packet) { - assert(!buf_err); - return rule_execute(rule, in_port, packet); + /* Pick table. */ + if (fm->table_id == 0xff) { + uint8_t table_id; + if (ofproto->ofproto_class->rule_choose_table) { + error = ofproto->ofproto_class->rule_choose_table(ofproto, &fm->cr, + &table_id); + if (error) { + return error; + } + assert(table_id < ofproto->n_tables); + table = &ofproto->tables[table_id]; + } else { + table = &ofproto->tables[0]; + } + } else if (fm->table_id < ofproto->n_tables) { + table = &ofproto->tables[fm->table_id]; + } else { + return OFPERR_NXFMFC_BAD_TABLE_ID; } - return buf_err; -} -static struct rule * -find_flow_strict(struct ofproto *p, const struct flow_mod *fm) -{ - return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr)); -} + if (table->flags & OFTABLE_READONLY) { + return OFPERR_OFPBRC_EPERM; + } -static int -send_buffered_packet(struct ofconn *ofconn, - struct rule *rule, uint32_t buffer_id) -{ - struct ofpbuf *packet; - uint16_t in_port; - int error; + /* Check for overlap, if requested. */ + if (fm->flags & OFPFF_CHECK_OVERLAP + && classifier_rule_overlaps(&table->cls, &fm->cr)) { + return OFPERR_OFPFMFC_OVERLAP; + } - if (buffer_id == UINT32_MAX) { - return 0; + /* Serialize against pending deletion. */ + if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) { + return OFPROTO_POSTPONE; } - error = ofconn_pktbuf_retrieve(ofconn, buffer_id, &packet, &in_port); - if (error) { - return error; + /* Allocate new rule. */ + rule = ofproto->ofproto_class->rule_alloc(); + if (!rule) { + VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)", + ofproto->name, strerror(error)); + return ENOMEM; } + rule->ofproto = ofproto; + rule->cr = fm->cr; + rule->pending = NULL; + rule->flow_cookie = fm->cookie; + rule->created = rule->modified = rule->used = time_msec(); + rule->idle_timeout = fm->idle_timeout; + rule->hard_timeout = fm->hard_timeout; + rule->table_id = table - ofproto->tables; + rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0; + rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions); + rule->n_actions = fm->n_actions; + rule->evictable = true; + rule->eviction_group = NULL; + + /* 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 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_ALL_TABLES_FULL; + goto exit; + } else if (evict->pending) { + error = OFPROTO_POSTPONE; + goto exit; + } + } else { + evict = NULL; + } + + group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id); + ofoperation_create(group, rule, OFOPERATION_ADD); + rule->pending->victim = victim; - return rule_execute(rule, in_port, packet); + error = ofproto->ofproto_class->rule_construct(rule); + if (error) { + ofoperation_destroy(rule->pending); + } else if (evict) { + delete_flow__(evict, group); + } + ofopgroup_submit(group); + } + +exit: + /* Back out if an error occurred. */ + if (error) { + oftable_substitute_rule(rule, victim); + ofproto_rule_destroy__(rule); + } + return error; } /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */ -struct modify_flows_cbdata { - struct ofproto *ofproto; - const struct flow_mod *fm; - struct rule *match; -}; - -static int modify_flow(const struct flow_mod *, struct rule *); - -/* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as - * encoded by ofp_mkerr() on failure. +/* Modifies the rules listed in 'rules', changing their actions to match those + * in 'fm'. * - * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, - * if any. */ -static int -modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm) + * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id, + * if any. + * + * Returns 0 on success, otherwise an OpenFlow error code. */ +static enum ofperr +modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn, + const struct ofputil_flow_mod *fm, + const struct ofp_header *request, struct list *rules) { - struct ofproto *p = ofconn_get_ofproto(ofconn); - struct rule *match = NULL; - struct cls_cursor cursor; + struct ofopgroup *group; struct rule *rule; - int error; + enum ofperr error; + + group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id); + error = OFPERR_OFPBRC_EPERM; + LIST_FOR_EACH (rule, ofproto_node, rules) { + if (rule_is_modifiable(rule)) { + /* At least one rule is modifiable, don't report EPERM error. */ + error = 0; + } else { + continue; + } - error = 0; - cls_cursor_init(&cursor, &p->cls, &fm->cr); - CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { - if (!rule_is_hidden(rule)) { - int retval = modify_flow(fm, rule); - if (!retval) { - match = rule; - } else { - error = retval; - } + if (!ofputil_actions_equal(fm->actions, fm->n_actions, + rule->actions, rule->n_actions)) { + ofoperation_create(group, rule, OFOPERATION_MODIFY); + rule->pending->actions = rule->actions; + rule->pending->n_actions = rule->n_actions; + rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions); + rule->n_actions = fm->n_actions; + ofproto->ofproto_class->rule_modify_actions(rule); + } else { + rule->modified = time_msec(); } + rule->flow_cookie = fm->cookie; } + ofopgroup_submit(group); - if (error) { - return error; - } else if (match) { - /* This credits the packet to whichever flow happened to match last. - * That's weird. Maybe we should do a lookup for the flow that - * actually matches the packet? Who knows. */ - send_buffered_packet(ofconn, match, fm->buffer_id); - return 0; - } else { - return add_flow(ofconn, fm); - } + return error; } -/* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error - * code as encoded by ofp_mkerr() on failure. +/* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on + * failure. * - * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, + * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id, * if any. */ -static int -modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm) +static enum ofperr +modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn, + const struct ofputil_flow_mod *fm, + const struct ofp_header *request) { - struct ofproto *p = ofconn_get_ofproto(ofconn); - struct rule *rule = find_flow_strict(p, fm); - if (rule && !rule_is_hidden(rule)) { - int error = modify_flow(fm, rule); - if (!error) { - error = send_buffered_packet(ofconn, rule, fm->buffer_id); - } - return error; - } else { - return add_flow(ofconn, fm); - } + struct list rules; + int error; + + error = collect_rules_loose(ofproto, fm->table_id, &fm->cr, + fm->cookie, fm->cookie_mask, + OFPP_NONE, &rules); + return (error ? error + : list_is_empty(&rules) ? add_flow(ofproto, ofconn, fm, request) + : modify_flows__(ofproto, ofconn, fm, request, &rules)); } -/* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has - * been identified as a flow to be modified, by changing the rule's actions to - * match those in 'ofm' (which is followed by 'n_actions' ofp_action[] - * structures). */ -static int -modify_flow(const struct flow_mod *fm, struct rule *rule) +/* Implements OFPFC_MODIFY_STRICT. 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_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn, + const struct ofputil_flow_mod *fm, + const struct ofp_header *request) { - size_t actions_len = fm->n_actions * sizeof *rule->actions; + struct list rules; int error; - if (fm->n_actions == rule->n_actions - && (!fm->n_actions - || !memcmp(fm->actions, rule->actions, actions_len))) { - error = 0; - } else { - error = rule->ofproto->ofproto_class->rule_modify_actions( - rule, fm->actions, fm->n_actions); - if (!error) { - free(rule->actions); - rule->actions = (fm->n_actions - ? xmemdup(fm->actions, actions_len) - : NULL); - rule->n_actions = fm->n_actions; - } - } - - if (!error) { - rule->flow_cookie = fm->cookie; - } - - return error; + error = collect_rules_strict(ofproto, fm->table_id, &fm->cr, + fm->cookie, fm->cookie_mask, + OFPP_NONE, &rules); + return (error ? error + : list_is_empty(&rules) ? add_flow(ofproto, ofconn, fm, request) + : list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn, + fm, request, &rules) + : 0); } /* OFPFC_DELETE implementation. */ -static void delete_flow(struct rule *, ovs_be16 out_port); - -/* Implements OFPFC_DELETE. */ static void -delete_flows_loose(struct ofproto *p, const struct flow_mod *fm) +delete_flow__(struct rule *rule, struct ofopgroup *group) { - struct rule *rule, *next_rule; - struct cls_cursor cursor; + struct ofproto *ofproto = rule->ofproto; - cls_cursor_init(&cursor, &p->cls, &fm->cr); - CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { - delete_flow(rule, htons(fm->out_port)); - } + ofproto_rule_send_removed(rule, OFPRR_DELETE); + + ofoperation_create(group, rule, OFOPERATION_DELETE); + oftable_remove_rule(rule); + ofproto->ofproto_class->rule_destruct(rule); } -/* Implements OFPFC_DELETE_STRICT. */ -static void -delete_flow_strict(struct ofproto *p, struct flow_mod *fm) +/* Deletes the rules listed in 'rules'. + * + * 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) { - struct rule *rule = find_flow_strict(p, fm); - if (rule) { - delete_flow(rule, htons(fm->out_port)); + struct rule *rule, *next; + struct ofopgroup *group; + + group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX); + LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) { + delete_flow__(rule, group); } + ofopgroup_submit(group); + + return 0; } -/* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has - * been identified as a flow to delete from 'p''s flow table, by deleting the - * flow and sending out a OFPT_FLOW_REMOVED message to any interested - * controller. - * - * Will not delete 'rule' if it is hidden. Will delete 'rule' only if - * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the - * specified 'out_port'. */ -static void -delete_flow(struct rule *rule, ovs_be16 out_port) +/* Implements OFPFC_DELETE. */ +static enum ofperr +delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn, + const struct ofputil_flow_mod *fm, + const struct ofp_header *request) { - if (rule_is_hidden(rule)) { - return; - } + struct list rules; + enum ofperr error; - if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) { - return; - } + error = collect_rules_loose(ofproto, fm->table_id, &fm->cr, + fm->cookie, fm->cookie_mask, + fm->out_port, &rules); + return (error ? error + : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request, + &rules) + : 0); +} - ofproto_rule_send_removed(rule, OFPRR_DELETE); - ofproto_rule_remove(rule); +/* Implements OFPFC_DELETE_STRICT. */ +static enum ofperr +delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn, + const struct ofputil_flow_mod *fm, + const struct ofp_header *request) +{ + struct list rules; + enum ofperr error; + + error = collect_rules_strict(ofproto, fm->table_id, &fm->cr, + fm->cookie, fm->cookie_mask, + fm->out_port, &rules); + return (error ? error + : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn, + request, &rules) + : 0); } static void @@ -2414,7 +2924,8 @@ ofproto_rule_send_removed(struct rule *rule, uint8_t reason) fr.rule = rule->cr; fr.cookie = rule->flow_cookie; fr.reason = reason; - calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec); + calc_flow_duration__(rule->created, time_msec(), + &fr.duration_sec, &fr.duration_nsec); fr.idle_timeout = rule->idle_timeout; rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count, &fr.byte_count); @@ -2422,6 +2933,20 @@ ofproto_rule_send_removed(struct rule *rule, uint8_t reason) 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. @@ -2431,24 +2956,33 @@ ofproto_rule_send_removed(struct rule *rule, uint8_t reason) void ofproto_rule_expire(struct rule *rule, uint8_t reason) { + struct ofproto *ofproto = rule->ofproto; + struct ofopgroup *group; + assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT); + ofproto_rule_send_removed(rule, reason); - ofproto_rule_remove(rule); + + group = ofopgroup_create_unattached(ofproto); + ofoperation_create(group, rule, OFOPERATION_DELETE); + oftable_remove_rule(rule); + ofproto->ofproto_class->rule_destruct(rule); + ofopgroup_submit(group); } -static int +static enum ofperr handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh) { - struct ofproto *p = ofconn_get_ofproto(ofconn); - struct flow_mod fm; - int error; + struct ofputil_flow_mod fm; + enum ofperr error; - error = reject_slave_controller(ofconn, "flow_mod"); + error = reject_slave_controller(ofconn); if (error) { return error; } - error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_flow_format(ofconn)); + error = ofputil_decode_flow_mod(&fm, oh, + ofconn_get_flow_mod_table_id(ofconn)); if (error) { return error; } @@ -2458,46 +2992,48 @@ handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh) if (fm.flags & OFPFF_EMERG) { /* There isn't a good fit for an error code, so just state that the * flow table is full. */ - return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL); + return OFPERR_OFPFMFC_ALL_TABLES_FULL; } - switch (fm.command) { + return handle_flow_mod__(ofconn_get_ofproto(ofconn), ofconn, &fm, oh); +} + +static enum ofperr +handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn, + const struct ofputil_flow_mod *fm, + const struct ofp_header *oh) +{ + if (ofproto->n_pending >= 50) { + assert(!list_is_empty(&ofproto->pending)); + return OFPROTO_POSTPONE; + } + + switch (fm->command) { case OFPFC_ADD: - return add_flow(ofconn, &fm); + return add_flow(ofproto, ofconn, fm, oh); case OFPFC_MODIFY: - return modify_flows_loose(ofconn, &fm); + return modify_flows_loose(ofproto, ofconn, fm, oh); case OFPFC_MODIFY_STRICT: - return modify_flow_strict(ofconn, &fm); + return modify_flow_strict(ofproto, ofconn, fm, oh); case OFPFC_DELETE: - delete_flows_loose(p, &fm); - return 0; + return delete_flows_loose(ofproto, ofconn, fm, oh); case OFPFC_DELETE_STRICT: - delete_flow_strict(p, &fm); - return 0; + return delete_flow_strict(ofproto, ofconn, fm, oh); default: - return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND); + if (fm->command > 0xff) { + VLOG_WARN_RL(&rl, "flow_mod has explicit table_id but " + "flow_mod_table_id extension is not enabled"); + } + return OFPERR_OFPFMFC_BAD_COMMAND; } } -static int -handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh) -{ - const struct nxt_tun_id_cookie *msg - = (const struct nxt_tun_id_cookie *) oh; - enum nx_flow_format flow_format; - - flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10; - ofconn_set_flow_format(ofconn, flow_format); - - return 0; -} - -static int +static enum ofperr handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh) { struct nx_role_request *nrr = (struct nx_role_request *) oh; @@ -2505,18 +3041,15 @@ handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh) struct ofpbuf *buf; uint32_t role; - if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) { - VLOG_WARN_RL(&rl, "ignoring role request on service connection"); - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); - } - role = ntohl(nrr->role); if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER && role != NX_ROLE_SLAVE) { - VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role); + return OFPERR_NXBRC_BAD_ROLE; + } - /* There's no good error code for this. */ - return ofp_mkerr(OFPET_BAD_REQUEST, -1); + if (ofconn_get_role(ofconn) != role + && ofconn_has_pending_opgroups(ofconn)) { + return OFPROTO_POSTPONE; } ofconn_set_role(ofconn, role); @@ -2528,43 +3061,103 @@ handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh) return 0; } -static int +static enum ofperr +handle_nxt_flow_mod_table_id(struct ofconn *ofconn, + const struct ofp_header *oh) +{ + const struct nx_flow_mod_table_id *msg + = (const struct nx_flow_mod_table_id *) oh; + + ofconn_set_flow_mod_table_id(ofconn, msg->set != 0); + return 0; +} + +static enum ofperr handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh) { - const struct nxt_set_flow_format *msg - = (const struct nxt_set_flow_format *) oh; + const struct nx_set_flow_format *msg + = (const struct nx_set_flow_format *) oh; uint32_t format; format = ntohl(msg->format); - if (format == NXFF_OPENFLOW10 - || format == NXFF_TUN_ID_FROM_COOKIE - || format == NXFF_NXM) { - ofconn_set_flow_format(ofconn, format); - return 0; - } else { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); + if (format != NXFF_OPENFLOW10 && format != NXFF_NXM) { + return OFPERR_OFPBRC_EPERM; + } + + if (format != ofconn_get_flow_format(ofconn) + && ofconn_has_pending_opgroups(ofconn)) { + /* Avoid sending async messages in surprising flow format. */ + return OFPROTO_POSTPONE; } + + ofconn_set_flow_format(ofconn, format); + return 0; } -static int +static enum ofperr +handle_nxt_set_packet_in_format(struct ofconn *ofconn, + const struct ofp_header *oh) +{ + const struct nx_set_packet_in_format *msg; + uint32_t format; + + msg = (const struct nx_set_packet_in_format *) oh; + format = ntohl(msg->format); + if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) { + return OFPERR_OFPBRC_EPERM; + } + + if (format != ofconn_get_packet_in_format(ofconn) + && ofconn_has_pending_opgroups(ofconn)) { + /* Avoid sending async message in surprsing packet in format. */ + return OFPROTO_POSTPONE; + } + + ofconn_set_packet_in_format(ofconn, format); + return 0; +} + +static enum ofperr +handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh) +{ + const struct nx_async_config *msg = (const struct nx_async_config *) oh; + uint32_t master[OAM_N_TYPES]; + uint32_t slave[OAM_N_TYPES]; + + master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]); + master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]); + master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]); + + slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]); + slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]); + slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]); + + ofconn_set_async_config(ofconn, master, slave); + + return 0; +} + +static enum ofperr handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh) { struct ofp_header *ob; struct ofpbuf *buf; - /* Currently, everything executes synchronously, so we can just - * immediately send the barrier reply. */ + if (ofconn_has_pending_opgroups(ofconn)) { + return OFPROTO_POSTPONE; + } + ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf); ofconn_send_reply(ofconn, buf); return 0; } -static int +static enum ofperr handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) { const struct ofp_header *oh = msg->data; const struct ofputil_msg_type *type; - int error; + enum ofperr error; error = ofputil_decode_msg_type(oh, &type); if (error) { @@ -2586,7 +3179,7 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) return handle_set_config(ofconn, msg->data); case OFPUTIL_OFPT_PACKET_OUT: - return handle_packet_out(ofconn, oh); + return handle_packet_out(ofconn, msg->data); case OFPUTIL_OFPT_PORT_MOD: return handle_port_mod(ofconn, oh); @@ -2602,45 +3195,50 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) return 0; /* Nicira extension requests. */ - case OFPUTIL_NXT_TUN_ID_FROM_COOKIE: - return handle_tun_id_from_cookie(ofconn, oh); - case OFPUTIL_NXT_ROLE_REQUEST: return handle_role_request(ofconn, oh); + case OFPUTIL_NXT_FLOW_MOD_TABLE_ID: + return handle_nxt_flow_mod_table_id(ofconn, oh); + case OFPUTIL_NXT_SET_FLOW_FORMAT: return handle_nxt_set_flow_format(ofconn, oh); + case OFPUTIL_NXT_SET_PACKET_IN_FORMAT: + return handle_nxt_set_packet_in_format(ofconn, oh); + case OFPUTIL_NXT_FLOW_MOD: return handle_flow_mod(ofconn, oh); - /* OpenFlow statistics requests. */ + case OFPUTIL_NXT_FLOW_AGE: + /* Nothing to do. */ + return 0; + + case OFPUTIL_NXT_SET_ASYNC_CONFIG: + return handle_nxt_set_async_config(ofconn, oh); + + /* Statistics requests. */ case OFPUTIL_OFPST_DESC_REQUEST: - return handle_desc_stats_request(ofconn, oh); + return handle_desc_stats_request(ofconn, msg->data); case OFPUTIL_OFPST_FLOW_REQUEST: - return handle_flow_stats_request(ofconn, oh); + case OFPUTIL_NXST_FLOW_REQUEST: + return handle_flow_stats_request(ofconn, msg->data); case OFPUTIL_OFPST_AGGREGATE_REQUEST: - return handle_aggregate_stats_request(ofconn, oh); + case OFPUTIL_NXST_AGGREGATE_REQUEST: + return handle_aggregate_stats_request(ofconn, msg->data); case OFPUTIL_OFPST_TABLE_REQUEST: - return handle_table_stats_request(ofconn, oh); + return handle_table_stats_request(ofconn, msg->data); case OFPUTIL_OFPST_PORT_REQUEST: - return handle_port_stats_request(ofconn, oh); + return handle_port_stats_request(ofconn, msg->data); case OFPUTIL_OFPST_QUEUE_REQUEST: - return handle_queue_stats_request(ofconn, oh); - - /* Nicira extension statistics requests. */ - case OFPUTIL_NXST_FLOW_REQUEST: - return handle_nxst_flow(ofconn, oh); - - case OFPUTIL_NXST_AGGREGATE_REQUEST: - return handle_nxst_aggregate(ofconn, oh); + return handle_queue_stats_request(ofconn, msg->data); - case OFPUTIL_INVALID: + case OFPUTIL_MSG_INVALID: case OFPUTIL_OFPT_HELLO: case OFPUTIL_OFPT_ERROR: case OFPUTIL_OFPT_FEATURES_REPLY: @@ -2659,30 +3257,268 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) case OFPUTIL_OFPST_AGGREGATE_REPLY: case OFPUTIL_NXT_ROLE_REPLY: case OFPUTIL_NXT_FLOW_REMOVED: + case OFPUTIL_NXT_PACKET_IN: case OFPUTIL_NXST_FLOW_REPLY: case OFPUTIL_NXST_AGGREGATE_REPLY: default: - if (VLOG_IS_WARN_ENABLED()) { - char *s = ofp_to_string(oh, ntohs(oh->length), 2); - VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s); - free(s); - } if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT); + return OFPERR_OFPBRC_BAD_STAT; } else { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE); + return OFPERR_OFPBRC_BAD_TYPE; } } } -static void +static bool handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg) { int error = handle_openflow__(ofconn, ofp_msg); - if (error) { - send_error_oh(ofconn, ofp_msg->data, error); + if (error && error != OFPROTO_POSTPONE) { + ofconn_send_error(ofconn, ofp_msg->data, error); } COVERAGE_INC(ofproto_recv_openflow); + return error != OFPROTO_POSTPONE; +} + +/* Asynchronous operations. */ + +/* Creates and returns a new ofopgroup that is not associated with any + * OpenFlow connection. + * + * The caller should add operations to the returned group with + * ofoperation_create() and then submit it with ofopgroup_submit(). */ +static struct ofopgroup * +ofopgroup_create_unattached(struct ofproto *ofproto) +{ + struct ofopgroup *group = xzalloc(sizeof *group); + group->ofproto = ofproto; + list_init(&group->ofproto_node); + list_init(&group->ops); + list_init(&group->ofconn_node); + return group; +} + +/* Creates and returns a new ofopgroup for 'ofproto'. + * + * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow + * connection. The 'request' and 'buffer_id' arguments are ignored. + * + * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'. + * If the ofopgroup eventually fails, then the error reply will include + * 'request'. If the ofopgroup eventually succeeds, then the packet with + * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto. + * + * The caller should add operations to the returned group with + * ofoperation_create() and then submit it with ofopgroup_submit(). */ +static struct ofopgroup * +ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn, + const struct ofp_header *request, uint32_t buffer_id) +{ + struct ofopgroup *group = ofopgroup_create_unattached(ofproto); + if (ofconn) { + size_t request_len = ntohs(request->length); + + assert(ofconn_get_ofproto(ofconn) == ofproto); + + ofconn_add_opgroup(ofconn, &group->ofconn_node); + group->ofconn = ofconn; + group->request = xmemdup(request, MIN(request_len, 64)); + group->buffer_id = buffer_id; + } + return group; +} + +/* Submits 'group' for processing. + * + * If 'group' contains no operations (e.g. none were ever added, or all of the + * ones that were added completed synchronously), then it is destroyed + * immediately. Otherwise it is added to the ofproto's list of pending + * groups. */ +static void +ofopgroup_submit(struct ofopgroup *group) +{ + if (list_is_empty(&group->ops)) { + ofopgroup_destroy(group); + } else { + list_push_back(&group->ofproto->pending, &group->ofproto_node); + group->ofproto->n_pending++; + } +} + +static void +ofopgroup_destroy(struct ofopgroup *group) +{ + assert(list_is_empty(&group->ops)); + if (!list_is_empty(&group->ofproto_node)) { + assert(group->ofproto->n_pending > 0); + group->ofproto->n_pending--; + list_remove(&group->ofproto_node); + } + if (!list_is_empty(&group->ofconn_node)) { + list_remove(&group->ofconn_node); + if (group->error) { + ofconn_send_error(group->ofconn, group->request, group->error); + } + connmgr_retry(group->ofproto->connmgr); + } + free(group->request); + free(group); +} + +/* Initiates a new operation on 'rule', of the specified 'type', within + * 'group'. Prior to calling, 'rule' must not have any pending operation. */ +static void +ofoperation_create(struct ofopgroup *group, struct rule *rule, + enum ofoperation_type type) +{ + struct ofoperation *op; + + assert(!rule->pending); + + op = rule->pending = xzalloc(sizeof *op); + op->group = group; + list_push_back(&group->ops, &op->group_node); + op->rule = rule; + op->type = type; + op->status = -1; + op->flow_cookie = rule->flow_cookie; + + if (type == OFOPERATION_DELETE) { + hmap_insert(&op->group->ofproto->deletions, &op->hmap_node, + cls_rule_hash(&rule->cr, rule->table_id)); + } +} + +static void +ofoperation_destroy(struct ofoperation *op) +{ + struct ofopgroup *group = op->group; + + if (op->rule) { + op->rule->pending = NULL; + } + if (op->type == OFOPERATION_DELETE) { + hmap_remove(&group->ofproto->deletions, &op->hmap_node); + } + list_remove(&op->group_node); + free(op->actions); + free(op); + + if (list_is_empty(&group->ops) && !list_is_empty(&group->ofproto_node)) { + ofopgroup_destroy(group); + } +} + +/* Indicates that 'op' completed with status 'error', which is either 0 to + * 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. + * + * If 'error' is nonzero, then generally the operation will be rolled back: + * + * - If 'op' is an "add flow" operation, ofproto removes the new rule or + * restores the original rule. The caller must have uninitialized any + * derived state in the new rule, as in step 5 of in the "Life Cycle" in + * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and + * and 7 for the new rule, calling its ->rule_dealloc() function. + * + * - If 'op' is a "modify flow" operation, ofproto restores the original + * actions. + * + * - 'op' must not be a "delete flow" operation. Removing a rule is not + * allowed to fail. It must always succeed. + * + * Please see the large comment in ofproto/ofproto-provider.h titled + * "Asynchronous Operation Support" for more information. */ +void +ofoperation_complete(struct ofoperation *op, enum ofperr error) +{ + struct ofopgroup *group = op->group; + struct rule *rule = op->rule; + struct ofproto *ofproto = rule->ofproto; + + assert(rule->pending == op); + assert(op->status < 0); + + if (!error + && !group->error + && op->type != OFOPERATION_DELETE + && group->ofconn + && group->buffer_id != UINT32_MAX + && list_is_singleton(&op->group_node)) { + struct ofpbuf *packet; + uint16_t in_port; + + error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id, + &packet, &in_port); + if (packet) { + assert(!error); + error = rule_execute(rule, in_port, packet); + } + } + if (!group->error) { + group->error = error; + } + + switch (op->type) { + case OFOPERATION_ADD: + if (!error) { + ofproto_rule_destroy__(op->victim); + if ((rule->cr.wc.vlan_tci_mask & htons(VLAN_VID_MASK)) + == htons(VLAN_VID_MASK)) { + if (ofproto->vlan_bitmap) { + uint16_t vid = vlan_tci_to_vid(rule->cr.flow.vlan_tci); + + if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) { + bitmap_set1(ofproto->vlan_bitmap, vid); + ofproto->vlans_changed = true; + } + } else { + ofproto->vlans_changed = true; + } + } + } else { + oftable_substitute_rule(rule, op->victim); + ofproto_rule_destroy__(rule); + } + break; + + case OFOPERATION_DELETE: + assert(!error); + ofproto_rule_destroy__(rule); + op->rule = NULL; + break; + + case OFOPERATION_MODIFY: + if (!error) { + rule->modified = time_msec(); + } else { + free(rule->actions); + rule->actions = op->actions; + rule->n_actions = op->n_actions; + op->actions = NULL; + } + break; + + default: + NOT_REACHED(); + } + ofoperation_destroy(op); +} + +struct rule * +ofoperation_get_victim(struct ofoperation *op) +{ + assert(op->type == OFOPERATION_ADD); + return op->victim; } static uint64_t @@ -2713,6 +3549,395 @@ pick_fallback_dpid(void) return eth_addr_to_uint64(ea); } +/* 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) +{ + struct eviction_group *evg; + + if (!table->eviction_fields) { + return NULL; + } + + /* In the common case, the outer and inner loops here will each be entered + * exactly once: + * + * - The inner loop normally "return"s in its first iteration. If the + * eviction group has any evictable rules, then it always returns in + * some iteration. + * + * - The outer loop only iterates more than once if the largest eviction + * group has no evictable rules. + * + * - The outer loop can exit only if table's 'max_flows' is all filled up + * 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; + } + } + } + + return NULL; +} + +/* Searches 'ofproto' for tables that have more flows than their configured + * maximum and that have flow eviction enabled, and evicts as many flows as + * necessary and currently feasible from them. + * + * This triggers only when an OpenFlow table has N flows in it and then the + * client configures a maximum number of flows less than N. */ +static void +ofproto_evict(struct ofproto *ofproto) +{ + struct ofopgroup *group; + struct oftable *table; + + group = ofopgroup_create_unattached(ofproto); + OFPROTO_FOR_EACH_TABLE (table, ofproto) { + while (classifier_count(&table->cls) > table->max_flows + && table->eviction_fields) { + struct rule *rule; + + rule = choose_rule_to_evict(table); + if (!rule || rule->pending) { + break; + } + + ofoperation_create(group, rule, OFOPERATION_DELETE); + oftable_remove_rule(rule); + ofproto->ofproto_class->rule_destruct(rule); + } + } + ofopgroup_submit(group); +} + +/* Eviction groups. */ + +/* Returns the priority to use for an eviction_group that contains 'n_rules' + * rules. The priority contains low-order random bits to ensure that eviction + * groups with the same number of rules are prioritized randomly. */ +static uint32_t +eviction_group_priority(size_t n_rules) +{ + uint16_t size = MIN(UINT16_MAX, n_rules); + return (size << 16) | random_uint16(); +} + +/* Updates 'evg', an eviction_group within 'table', following a change that + * adds or removes rules in 'evg'. */ +static void +eviction_group_resized(struct oftable *table, struct eviction_group *evg) +{ + heap_change(&table->eviction_groups_by_size, &evg->size_node, + eviction_group_priority(heap_count(&evg->rules))); +} + +/* Destroys 'evg', an eviction_group within 'table': + * + * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the + * rules themselves, just removes them from the eviction group.) + * + * - Removes 'evg' from 'table'. + * + * - Frees 'evg'. */ +static void +eviction_group_destroy(struct oftable *table, struct eviction_group *evg) +{ + while (!heap_is_empty(&evg->rules)) { + struct rule *rule; + + rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node); + rule->eviction_group = NULL; + } + hmap_remove(&table->eviction_groups_by_id, &evg->id_node); + heap_remove(&table->eviction_groups_by_size, &evg->size_node); + heap_destroy(&evg->rules); + free(evg); +} + +/* Removes 'rule' from its eviction group, if any. */ +static void +eviction_group_remove_rule(struct rule *rule) +{ + if (rule->eviction_group) { + struct oftable *table = &rule->ofproto->tables[rule->table_id]; + struct eviction_group *evg = rule->eviction_group; + + rule->eviction_group = NULL; + heap_remove(&evg->rules, &rule->evg_node); + if (heap_is_empty(&evg->rules)) { + eviction_group_destroy(table, evg); + } else { + eviction_group_resized(table, evg); + } + } +} + +/* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and + * returns the hash value. */ +static uint32_t +eviction_group_hash_rule(struct rule *rule) +{ + struct oftable *table = &rule->ofproto->tables[rule->table_id]; + const struct mf_subfield *sf; + uint32_t hash; + + hash = table->eviction_group_id_basis; + for (sf = table->eviction_fields; + sf < &table->eviction_fields[table->n_eviction_fields]; + sf++) + { + if (mf_are_prereqs_ok(sf->field, &rule->cr.flow)) { + union mf_value value; + + mf_get_value(sf->field, &rule->cr.flow, &value); + if (sf->ofs) { + bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs); + } + if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) { + unsigned int start = sf->ofs + sf->n_bits; + bitwise_zero(&value, sf->field->n_bytes, start, + sf->field->n_bytes * 8 - start); + } + hash = hash_bytes(&value, sf->field->n_bytes, hash); + } else { + hash = hash_int(hash, 0); + } + } + + return hash; +} + +/* Returns an eviction group within 'table' with the given 'id', creating one + * if necessary. */ +static struct eviction_group * +eviction_group_find(struct oftable *table, uint32_t id) +{ + struct eviction_group *evg; + + HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) { + return evg; + } + + evg = xmalloc(sizeof *evg); + hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id); + heap_insert(&table->eviction_groups_by_size, &evg->size_node, + eviction_group_priority(0)); + heap_init(&evg->rules); + + return evg; +} + +/* Returns an eviction priority for 'rule'. The return value should be + * interpreted so that higher priorities make a rule more attractive candidates + * for eviction. */ +static uint32_t +rule_eviction_priority(struct rule *rule) +{ + long long int hard_expiration; + long long int idle_expiration; + long long int expiration; + uint32_t expiration_offset; + + /* Calculate time of expiration. */ + hard_expiration = (rule->hard_timeout + ? rule->modified + rule->hard_timeout * 1000 + : LLONG_MAX); + idle_expiration = (rule->idle_timeout + ? rule->used + rule->idle_timeout * 1000 + : LLONG_MAX); + expiration = MIN(hard_expiration, idle_expiration); + if (expiration == LLONG_MAX) { + return 0; + } + + /* Calculate the time of expiration as a number of (approximate) seconds + * after program startup. + * + * This should work OK for program runs that last UINT32_MAX seconds or + * less. Therefore, please restart OVS at least once every 136 years. */ + expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10); + + /* Invert the expiration offset because we're using a max-heap. */ + return UINT32_MAX - expiration_offset; +} + +/* Adds 'rule' to an appropriate eviction group for its oftable's + * configuration. Does nothing if 'rule''s oftable doesn't have eviction + * enabled, or if 'rule' is a permanent rule (one that will never expire on its + * own). + * + * The caller must ensure that 'rule' is not already in an eviction group. */ +static void +eviction_group_add_rule(struct rule *rule) +{ + struct ofproto *ofproto = rule->ofproto; + struct oftable *table = &ofproto->tables[rule->table_id]; + + if (table->eviction_fields + && (rule->hard_timeout || rule->idle_timeout)) { + struct eviction_group *evg; + + evg = eviction_group_find(table, eviction_group_hash_rule(rule)); + + rule->eviction_group = evg; + heap_insert(&evg->rules, &rule->evg_node, + rule_eviction_priority(rule)); + eviction_group_resized(table, evg); + } +} + +/* oftables. */ + +/* Initializes 'table'. */ +static void +oftable_init(struct oftable *table) +{ + memset(table, 0, sizeof *table); + classifier_init(&table->cls); +} + +/* Destroys 'table', including its classifier and eviction groups. + * + * The caller is responsible for freeing 'table' itself. */ +static void +oftable_destroy(struct oftable *table) +{ + assert(classifier_is_empty(&table->cls)); + oftable_disable_eviction(table); + classifier_destroy(&table->cls); + free(table->name); +} + +/* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty + * string, then 'table' will use its default name. + * + * This only affects the name exposed for a table exposed through the OpenFlow + * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */ +static void +oftable_set_name(struct oftable *table, const char *name) +{ + if (name && name[0]) { + int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN); + if (!table->name || strncmp(name, table->name, len)) { + free(table->name); + table->name = xmemdup0(name, len); + } + } else { + free(table->name); + table->name = NULL; + } +} + +/* oftables support a choice of two policies when adding a rule would cause the + * number of flows in the table to exceed the configured maximum number: either + * they can refuse to add the new flow or they can evict some existing flow. + * This function configures the former policy on 'table'. */ +static void +oftable_disable_eviction(struct oftable *table) +{ + if (table->eviction_fields) { + struct eviction_group *evg, *next; + + HMAP_FOR_EACH_SAFE (evg, next, id_node, + &table->eviction_groups_by_id) { + eviction_group_destroy(table, evg); + } + hmap_destroy(&table->eviction_groups_by_id); + heap_destroy(&table->eviction_groups_by_size); + + free(table->eviction_fields); + table->eviction_fields = NULL; + table->n_eviction_fields = 0; + } +} + +/* oftables support a choice of two policies when adding a rule would cause the + * number of flows in the table to exceed the configured maximum number: either + * they can refuse to add the new flow or they can evict some existing flow. + * This function configures the latter policy on 'table', with fairness based + * on the values of the 'n_fields' fields specified in 'fields'. (Specifying + * 'n_fields' as 0 disables fairness.) */ +static void +oftable_enable_eviction(struct oftable *table, + const struct mf_subfield *fields, size_t n_fields) +{ + struct cls_cursor cursor; + struct rule *rule; + + if (table->eviction_fields + && n_fields == table->n_eviction_fields + && (!n_fields + || !memcmp(fields, table->eviction_fields, + n_fields * sizeof *fields))) { + /* No change. */ + return; + } + + oftable_disable_eviction(table); + + table->n_eviction_fields = n_fields; + table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields); + + table->eviction_group_id_basis = random_uint32(); + hmap_init(&table->eviction_groups_by_id); + heap_init(&table->eviction_groups_by_size); + + cls_cursor_init(&cursor, &table->cls, NULL); + CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { + eviction_group_add_rule(rule); + } +} + +/* Removes 'rule' from the oftable that contains it. */ +static void +oftable_remove_rule(struct rule *rule) +{ + struct ofproto *ofproto = rule->ofproto; + struct oftable *table = &ofproto->tables[rule->table_id]; + + classifier_remove(&table->cls, &rule->cr); + eviction_group_remove_rule(rule); +} + +/* 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) +{ + struct ofproto *ofproto = rule->ofproto; + struct oftable *table = &ofproto->tables[rule->table_id]; + struct rule *victim; + + victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr)); + if (victim) { + eviction_group_remove_rule(victim); + } + eviction_group_add_rule(rule); + return victim; +} + +/* 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. */ struct ofproto * @@ -2730,8 +3955,8 @@ ofproto_lookup(const char *name) } static void -ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED, - void *aux OVS_UNUSED) +ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED, + const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) { struct ofproto *ofproto; struct ds results; @@ -2753,5 +3978,90 @@ ofproto_unixctl_init(void) } registered = true; - unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL); + unixctl_command_register("ofproto/list", "", 0, 0, + ofproto_unixctl_list, NULL); +} + +/* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) + * + * This is deprecated. It is only for compatibility with broken device drivers + * in old versions of Linux that do not properly support VLANs when VLAN + * devices are not used. When broken device drivers are no longer in + * widespread use, we will delete these interfaces. */ + +/* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched + * (exactly) by an OpenFlow rule in 'ofproto'. */ +void +ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap) +{ + const struct oftable *oftable; + + free(ofproto->vlan_bitmap); + ofproto->vlan_bitmap = bitmap_allocate(4096); + ofproto->vlans_changed = false; + + OFPROTO_FOR_EACH_TABLE (oftable, ofproto) { + const struct cls_table *table; + + HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) { + if ((table->wc.vlan_tci_mask & htons(VLAN_VID_MASK)) + == htons(VLAN_VID_MASK)) { + const struct cls_rule *rule; + + HMAP_FOR_EACH (rule, hmap_node, &table->rules) { + uint16_t vid = vlan_tci_to_vid(rule->flow.vlan_tci); + bitmap_set1(vlan_bitmap, vid); + bitmap_set1(ofproto->vlan_bitmap, vid); + } + } + } + } +} + +/* Returns true if new VLANs have come into use by the flow table since the + * last call to ofproto_get_vlan_usage(). + * + * We don't track when old VLANs stop being used. */ +bool +ofproto_has_vlan_usage_changed(const struct ofproto *ofproto) +{ + return ofproto->vlans_changed; +} + +/* Configures a VLAN splinter binding between the ports identified by OpenFlow + * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If + * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real + * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero, + * then the VLAN device is un-enslaved. */ +int +ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port, + uint16_t realdev_ofp_port, int vid) +{ + struct ofport *ofport; + int error; + + assert(vlandev_ofp_port != realdev_ofp_port); + + ofport = ofproto_get_port(ofproto, vlandev_ofp_port); + if (!ofport) { + VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16, + ofproto->name, vlandev_ofp_port); + return EINVAL; + } + + if (!ofproto->ofproto_class->set_realdev) { + if (!vlandev_ofp_port) { + return 0; + } + VLOG_WARN("%s: vlan splinters not supported", ofproto->name); + return EOPNOTSUPP; + } + + error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid); + if (error) { + VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)", + ofproto->name, vlandev_ofp_port, + netdev_get_name(ofport->netdev), strerror(error)); + } + return error; }