X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=ofproto%2Fofproto.c;h=00c9c6964a222ebdae1d3851d862e3e9371743a7;hb=abaad8cf1b5089e17a4af0ab1ff644bfcf63cad9;hp=7650068e88c7340e0454204cfe158c8c1b615f0d;hpb=f1acd62b54376a425a975f9af501c4c8c5689b39;p=openvswitch diff --git a/ofproto/ofproto.c b/ofproto/ofproto.c index 7650068e..00c9c696 100644 --- a/ofproto/ofproto.c +++ b/ofproto/ofproto.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2009 Nicira Networks. + * Copyright (c) 2009, 2010 Nicira Networks. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -27,7 +27,6 @@ #include "discovery.h" #include "dpif.h" #include "dynamic-string.h" -#include "executer.h" #include "fail-open.h" #include "in-band.h" #include "mac-learning.h" @@ -35,10 +34,10 @@ #include "netflow.h" #include "odp-util.h" #include "ofp-print.h" +#include "ofproto-sflow.h" #include "ofpbuf.h" #include "openflow/nicira-ext.h" #include "openflow/openflow.h" -#include "openflow/openflow-mgmt.h" #include "openvswitch/datapath-protocol.h" #include "packets.h" #include "pinsched.h" @@ -49,21 +48,18 @@ #include "shash.h" #include "status.h" #include "stp.h" +#include "stream-ssl.h" #include "svec.h" #include "tag.h" #include "timeval.h" #include "unixctl.h" #include "vconn.h" -#include "vconn-ssl.h" #include "xtoxll.h" #define THIS_MODULE VLM_ofproto #include "vlog.h" -enum { - DP_GROUP_FLOOD = 0, - DP_GROUP_ALL = 1 -}; +#include "sflow_api.h" enum { TABLEID_HASH = 0, @@ -82,21 +78,23 @@ static int xlate_actions(const union ofp_action *in, size_t n_in, const flow_t *flow, struct ofproto *ofproto, const struct ofpbuf *packet, struct odp_actions *out, tag_type *tags, - bool *may_setup_flow); + bool *may_set_up_flow, uint16_t *nf_output_iface); struct rule { struct cls_rule cr; + uint64_t flow_cookie; /* Controller-issued identifier. + (Kept in network-byte order.) */ uint16_t idle_timeout; /* In seconds from time of last use. */ uint16_t hard_timeout; /* In seconds from time of creation. */ + bool send_flow_removed; /* Send a flow removed message? */ long long int used; /* Last-used time (0 if never used). */ long long int created; /* Creation time. */ uint64_t packet_count; /* Number of packets received. */ uint64_t byte_count; /* Number of bytes received. */ uint64_t accounted_bytes; /* Number of bytes passed to account_cb. */ - uint8_t tcp_flags; /* Bitwise-OR of all TCP flags seen. */ - uint8_t ip_tos; /* Last-seen IP type-of-service. */ tag_type tags; /* Tags (set only by hooks). */ + struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ /* If 'super' is non-NULL, this rule is a subrule, that is, it is an * exact-match rule (having cr.wc.wildcards of 0) generated from the @@ -145,9 +143,10 @@ rule_is_hidden(const struct rule *rule) return false; } -static struct rule *rule_create(struct rule *super, const union ofp_action *, - size_t n_actions, uint16_t idle_timeout, - uint16_t hard_timeout); +static struct rule *rule_create(struct ofproto *, struct rule *super, + const union ofp_action *, size_t n_actions, + uint16_t idle_timeout, uint16_t hard_timeout, + uint64_t flow_cookie, bool send_flow_removed); static void rule_free(struct rule *); static void rule_destroy(struct ofproto *, struct rule *); static struct rule *rule_from_cls_rule(const struct cls_rule *); @@ -160,12 +159,13 @@ static void rule_install(struct ofproto *, struct rule *, struct rule *displaced_rule); static void rule_uninstall(struct ofproto *, struct rule *); static void rule_post_uninstall(struct ofproto *, struct rule *); +static void send_flow_removed(struct ofproto *p, struct rule *rule, + long long int now, uint8_t reason); struct ofconn { struct list node; struct rconn *rconn; struct pktbuf *pktbuf; - bool send_flow_exp; int miss_send_len; struct rconn_packet_counter *packet_in_counter; @@ -177,7 +177,7 @@ struct ofconn { }; static struct ofconn *ofconn_create(struct ofproto *, struct rconn *); -static void ofconn_destroy(struct ofconn *, struct ofproto *); +static void ofconn_destroy(struct ofconn *); static void ofconn_run(struct ofconn *, struct ofproto *); static void ofconn_wait(struct ofconn *); static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, @@ -187,11 +187,11 @@ struct ofproto { /* Settings. */ uint64_t datapath_id; /* Datapath ID. */ uint64_t fallback_dpid; /* Datapath ID if no better choice found. */ - uint64_t mgmt_id; /* Management channel identifier. */ char *manufacturer; /* Manufacturer. */ char *hardware; /* Hardware. */ char *software; /* Software version. */ char *serial; /* Serial number. */ + char *dp_desc; /* Datapath description. */ /* Datapath. */ struct dpif *dpif; @@ -208,8 +208,8 @@ struct ofproto { struct discovery *discovery; struct fail_open *fail_open; struct pinsched *miss_sched, *action_sched; - struct executer *executer; struct netflow *netflow; + struct ofproto_sflow *sflow; /* Flow table. */ struct classifier cls; @@ -242,8 +242,10 @@ static uint64_t pick_fallback_dpid(void); static void send_packet_in_miss(struct ofpbuf *, void *ofproto); static void send_packet_in_action(struct ofpbuf *, void *ofproto); static void update_used(struct ofproto *); -static void update_stats(struct rule *, const struct odp_flow_stats *); +static void update_stats(struct ofproto *, struct rule *, + const struct odp_flow_stats *); static void expire_rule(struct cls_rule *, void *ofproto); +static void active_timeout(struct ofproto *ofproto, struct rule *rule); static bool revalidate_rule(struct ofproto *p, struct rule *rule); static void revalidate_cb(struct cls_rule *rule_, void *p_); @@ -252,13 +254,15 @@ static void handle_odp_msg(struct ofproto *, struct ofpbuf *); static void handle_openflow(struct ofconn *, struct ofproto *, struct ofpbuf *); -static void refresh_port_group(struct ofproto *, unsigned int group); +static void refresh_port_groups(struct ofproto *); + static void update_port(struct ofproto *, const char *devname); static int init_ports(struct ofproto *); static void reinit_ports(struct ofproto *); int -ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux, +ofproto_create(const char *datapath, const char *datapath_type, + const struct ofhooks *ofhooks, void *aux, struct ofproto **ofprotop) { struct odp_stats stats; @@ -269,7 +273,7 @@ ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux, *ofprotop = NULL; /* Connect to datapath and start listening for messages. */ - error = dpif_open(datapath, &dpif); + error = dpif_open(datapath, datapath_type, &dpif); if (error) { VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error)); return error; @@ -281,7 +285,7 @@ ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux, dpif_close(dpif); return error; } - error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION); + error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW); if (error) { VLOG_ERR("failed to listen on datapath %s: %s", datapath, strerror(error)); @@ -292,13 +296,14 @@ ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux, dpif_recv_purge(dpif); /* Initialize settings. */ - p = xcalloc(1, sizeof *p); + p = xzalloc(sizeof *p); p->fallback_dpid = pick_fallback_dpid(); p->datapath_id = p->fallback_dpid; p->manufacturer = xstrdup("Nicira Networks, Inc."); p->hardware = xstrdup("Reference Implementation"); p->software = xstrdup(VERSION BUILDNR); p->serial = xstrdup("None"); + p->dp_desc = xstrdup("None"); /* Initialize datapath. */ p->dpif = dpif; @@ -313,8 +318,8 @@ ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux, p->discovery = NULL; p->fail_open = NULL; p->miss_sched = p->action_sched = NULL; - p->executer = NULL; p->netflow = NULL; + p->sflow = NULL; /* Initialize flow table. */ classifier_init(&p->cls); @@ -347,16 +352,9 @@ ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux, p->ss_cat = switch_status_register(p->switch_status, "remote", rconn_status_cb, p->controller->rconn); - /* Almost done... */ - error = init_ports(p); - if (error) { - ofproto_destroy(p); - return error; - } - /* Pick final datapath ID. */ p->datapath_id = pick_datapath_id(p); - VLOG_INFO("using datapath ID %012"PRIx64, p->datapath_id); + VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id); *ofprotop = p; return 0; @@ -368,17 +366,11 @@ ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id) uint64_t old_dpid = p->datapath_id; p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p); if (p->datapath_id != old_dpid) { - VLOG_INFO("datapath ID changed to %012"PRIx64, p->datapath_id); + VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id); rconn_reconnect(p->controller->rconn); } } -void -ofproto_set_mgmt_id(struct ofproto *p, uint64_t mgmt_id) -{ - p->mgmt_id = mgmt_id; -} - void ofproto_set_probe_interval(struct ofproto *p, int probe_interval) { @@ -399,7 +391,8 @@ ofproto_set_max_backoff(struct ofproto *p, int max_backoff) void ofproto_set_desc(struct ofproto *p, const char *manufacturer, const char *hardware, - const char *software, const char *serial) + const char *software, const char *serial, + const char *dp_desc) { if (manufacturer) { free(p->manufacturer); @@ -417,6 +410,10 @@ ofproto_set_desc(struct ofproto *p, free(p->serial); p->serial = xstrdup(serial); } + if (dp_desc) { + free(p->dp_desc); + p->dp_desc = xstrdup(dp_desc); + } } int @@ -532,16 +529,14 @@ ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops) } int -ofproto_set_netflow(struct ofproto *ofproto, const struct svec *collectors, - uint8_t engine_type, uint8_t engine_id, bool add_id_to_iface) +ofproto_set_netflow(struct ofproto *ofproto, + const struct netflow_options *nf_options) { - if (collectors && collectors->n) { + if (nf_options && nf_options->collectors.n) { if (!ofproto->netflow) { ofproto->netflow = netflow_create(); } - netflow_set_engine(ofproto->netflow, engine_type, engine_id, - add_id_to_iface); - return netflow_set_collectors(ofproto->netflow, collectors); + return netflow_set_options(ofproto->netflow, nf_options); } else { netflow_destroy(ofproto->netflow); ofproto->netflow = NULL; @@ -549,6 +544,30 @@ ofproto_set_netflow(struct ofproto *ofproto, const struct svec *collectors, } } +void +ofproto_set_sflow(struct ofproto *ofproto, + const struct ofproto_sflow_options *oso) +{ + struct ofproto_sflow *os = ofproto->sflow; + if (oso) { + if (!os) { + struct ofport *ofport; + unsigned int odp_port; + + os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif); + refresh_port_groups(ofproto); + PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) { + ofproto_sflow_add_port(os, odp_port, + netdev_get_name(ofport->netdev)); + } + } + ofproto_sflow_set_options(os, oso); + } else { + ofproto_sflow_destroy(os); + ofproto->sflow = NULL; + } +} + void ofproto_set_failure(struct ofproto *ofproto, bool fail_open) { @@ -593,7 +612,7 @@ ofproto_set_rate_limit(struct ofproto *ofproto, } int -ofproto_set_stp(struct ofproto *ofproto UNUSED, bool enable_stp) +ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp) { /* XXX */ if (enable_stp) { @@ -604,36 +623,12 @@ ofproto_set_stp(struct ofproto *ofproto UNUSED, bool enable_stp) } } -int -ofproto_set_remote_execution(struct ofproto *ofproto, const char *command_acl, - const char *command_dir) -{ - if (command_acl) { - if (!ofproto->executer) { - return executer_create(command_acl, command_dir, - &ofproto->executer); - } else { - executer_set_acl(ofproto->executer, command_acl, command_dir); - } - } else { - executer_destroy(ofproto->executer); - ofproto->executer = NULL; - } - return 0; -} - uint64_t ofproto_get_datapath_id(const struct ofproto *ofproto) { return ofproto->datapath_id; } -uint64_t -ofproto_get_mgmt_id(const struct ofproto *ofproto) -{ - return ofproto->mgmt_id; -} - int ofproto_get_probe_interval(const struct ofproto *ofproto) { @@ -696,12 +691,15 @@ ofproto_destroy(struct ofproto *p) return; } + /* Destroy fail-open early, because it touches the classifier. */ + ofproto_set_failure(p, false); + ofproto_flush_flows(p); classifier_destroy(&p->cls); LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node, &p->all_conns) { - ofconn_destroy(ofconn, p); + ofconn_destroy(ofconn); } dpif_close(p->dpif); @@ -714,11 +712,10 @@ ofproto_destroy(struct ofproto *p) switch_status_destroy(p->switch_status); in_band_destroy(p->in_band); discovery_destroy(p->discovery); - fail_open_destroy(p->fail_open); pinsched_destroy(p->miss_sched); pinsched_destroy(p->action_sched); - executer_destroy(p->executer); netflow_destroy(p->netflow); + ofproto_sflow_destroy(p->sflow); switch_status_unregister(p->ss_cat); @@ -766,6 +763,10 @@ ofproto_run1(struct ofproto *p) int error; int i; + if (shash_is_empty(&p->port_by_name)) { + init_ports(p); + } + for (i = 0; i < 50; i++) { struct ofpbuf *buf; int error; @@ -811,20 +812,20 @@ ofproto_run1(struct ofproto *p) } } } - if (p->fail_open) { - fail_open_run(p->fail_open); - } pinsched_run(p->miss_sched, send_packet_in_miss, p); pinsched_run(p->action_sched, send_packet_in_action, p); - if (p->executer) { - executer_run(p->executer); - } LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node, &p->all_conns) { ofconn_run(ofconn, p); } + /* Fail-open maintenance. Do this after processing the ofconns since + * fail-open checks the status of the controller rconn. */ + if (p->fail_open) { + fail_open_run(p->fail_open); + } + for (i = 0; i < p->n_listeners; i++) { struct vconn *vconn; int retval; @@ -868,6 +869,9 @@ ofproto_run1(struct ofproto *p) if (p->netflow) { netflow_run(p->netflow); } + if (p->sflow) { + ofproto_sflow_run(p->sflow); + } return 0; } @@ -921,8 +925,8 @@ ofproto_wait(struct ofproto *p) } pinsched_wait(p->miss_sched); pinsched_wait(p->action_sched); - if (p->executer) { - executer_wait(p->executer); + if (p->sflow) { + ofproto_sflow_wait(p->sflow); } if (!tag_set_is_empty(&p->revalidate_set)) { poll_immediate_wake(); @@ -969,7 +973,7 @@ ofproto_send_packet(struct ofproto *p, const flow_t *flow, int error; error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions, - NULL, NULL); + NULL, NULL, NULL); if (error) { return error; } @@ -988,8 +992,9 @@ ofproto_add_flow(struct ofproto *p, int idle_timeout) { struct rule *rule; - rule = rule_create(NULL, actions, n_actions, - idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, 0); + rule = rule_create(p, NULL, actions, n_actions, + idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, + 0, 0, false); cls_rule_from_flow(&rule->cr, flow, wildcards, priority); rule_insert(p, rule, NULL, 0); } @@ -1064,7 +1069,7 @@ reinit_ports(struct ofproto *p) svec_destroy(&devnames); } -static void +static size_t refresh_port_group(struct ofproto *p, unsigned int group) { uint16_t *ports; @@ -1083,25 +1088,36 @@ refresh_port_group(struct ofproto *p, unsigned int group) } dpif_port_group_set(p->dpif, group, ports, n_ports); free(ports); + + return n_ports; } static void refresh_port_groups(struct ofproto *p) { - refresh_port_group(p, DP_GROUP_FLOOD); - refresh_port_group(p, DP_GROUP_ALL); + size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD); + size_t n_all = refresh_port_group(p, DP_GROUP_ALL); + if (p->sflow) { + ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all); + } } static struct ofport * make_ofport(const struct odp_port *odp_port) { + struct netdev_options netdev_options; enum netdev_flags flags; struct ofport *ofport; struct netdev *netdev; bool carrier; int error; - error = netdev_open(odp_port->devname, NETDEV_ETH_TYPE_NONE, &netdev); + memset(&netdev_options, 0, sizeof netdev_options); + netdev_options.name = odp_port->devname; + netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; + netdev_options.may_open = true; + + error = netdev_open(&netdev_options, &netdev); if (error) { VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s " "cannot be opened (%s)", @@ -1188,19 +1204,29 @@ send_port_status(struct ofproto *p, const struct ofport *ofport, static void ofport_install(struct ofproto *p, struct ofport *ofport) { + uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no); + const char *netdev_name = (const char *) ofport->opp.name; + netdev_monitor_add(p->netdev_monitor, ofport->netdev); - port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), - ofport); - shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport); + port_array_set(&p->ports, odp_port, ofport); + shash_add(&p->port_by_name, netdev_name, ofport); + if (p->sflow) { + ofproto_sflow_add_port(p->sflow, odp_port, netdev_name); + } } static void ofport_remove(struct ofproto *p, struct ofport *ofport) { + uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no); + netdev_monitor_remove(p->netdev_monitor, ofport->netdev); - port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), NULL); + port_array_set(&p->ports, odp_port, NULL); shash_delete(&p->port_by_name, shash_find(&p->port_by_name, (char *) ofport->opp.name)); + if (p->sflow) { + ofproto_sflow_del_port(p->sflow, odp_port); + } } static void @@ -1318,7 +1344,6 @@ ofconn_create(struct ofproto *p, struct rconn *rconn) list_push_back(&p->all_conns, &ofconn->node); ofconn->rconn = rconn; ofconn->pktbuf = NULL; - ofconn->send_flow_exp = false; ofconn->miss_send_len = 0; ofconn->packet_in_counter = rconn_packet_counter_create (); ofconn->reply_counter = rconn_packet_counter_create (); @@ -1326,12 +1351,8 @@ ofconn_create(struct ofproto *p, struct rconn *rconn) } static void -ofconn_destroy(struct ofconn *ofconn, struct ofproto *p) +ofconn_destroy(struct ofconn *ofconn) { - if (p->executer) { - executer_rconn_closing(p->executer, ofconn->rconn); - } - list_remove(&ofconn->node); rconn_destroy(ofconn->rconn); rconn_packet_counter_destroy(ofconn->packet_in_counter); @@ -1355,13 +1376,16 @@ ofconn_run(struct ofconn *ofconn, struct ofproto *p) if (!of_msg) { break; } + if (p->fail_open) { + fail_open_maybe_recover(p->fail_open); + } handle_openflow(ofconn, p, of_msg); ofpbuf_delete(of_msg); } } if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) { - ofconn_destroy(ofconn, p); + ofconn_destroy(ofconn); } } @@ -1379,14 +1403,17 @@ ofconn_wait(struct ofconn *ofconn) /* Caller is responsible for initializing the 'cr' member of the returned * rule. */ static struct rule * -rule_create(struct rule *super, +rule_create(struct ofproto *ofproto, struct rule *super, const union ofp_action *actions, size_t n_actions, - uint16_t idle_timeout, uint16_t hard_timeout) + uint16_t idle_timeout, uint16_t hard_timeout, + uint64_t flow_cookie, bool send_flow_removed) { - struct rule *rule = xcalloc(1, sizeof *rule); + struct rule *rule = xzalloc(sizeof *rule); rule->idle_timeout = idle_timeout; rule->hard_timeout = hard_timeout; + rule->flow_cookie = flow_cookie; rule->used = rule->created = time_msec(); + rule->send_flow_removed = send_flow_removed; rule->super = super; if (super) { list_push_back(&super->list, &rule->list); @@ -1395,6 +1422,9 @@ rule_create(struct rule *super, } rule->n_actions = n_actions; rule->actions = xmemdup(actions, n_actions * sizeof *actions); + netflow_flow_clear(&rule->nf_flow); + netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->created); + return rule; } @@ -1483,7 +1513,7 @@ rule_execute(struct ofproto *ofproto, struct rule *rule, if (rule->cr.wc.wildcards || !flow_equal(flow, &rule->cr.flow)) { struct rule *super = rule->super ? rule->super : rule; if (xlate_actions(super->actions, super->n_actions, flow, ofproto, - packet, &a, NULL, 0)) { + packet, &a, NULL, 0, NULL)) { return; } actions = a.actions; @@ -1498,8 +1528,9 @@ rule_execute(struct ofproto *ofproto, struct rule *rule, actions, n_actions, packet)) { struct odp_flow_stats stats; flow_extract_stats(flow, packet, &stats); - update_stats(rule, &stats); + update_stats(ofproto, rule, &stats); rule->used = time_msec(); + netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used); } } @@ -1541,8 +1572,9 @@ static struct rule * rule_create_subrule(struct ofproto *ofproto, struct rule *rule, const flow_t *flow) { - struct rule *subrule = rule_create(rule, NULL, 0, - rule->idle_timeout, rule->hard_timeout); + struct rule *subrule = rule_create(ofproto, rule, NULL, 0, + rule->idle_timeout, rule->hard_timeout, + 0, false); COVERAGE_INC(ofproto_subrule_create); cls_rule_from_flow(&subrule->cr, flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX @@ -1579,7 +1611,8 @@ rule_make_actions(struct ofproto *p, struct rule *rule, super = rule->super ? rule->super : rule; rule->tags = 0; xlate_actions(super->actions, super->n_actions, &rule->cr.flow, p, - packet, &a, &rule->tags, &rule->may_install); + packet, &a, &rule->tags, &rule->may_install, + &rule->nf_flow.output_iface); actions_len = a.n_actions * sizeof *a.actions; if (rule->n_odp_actions != a.n_actions @@ -1618,7 +1651,7 @@ rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule) &put)) { rule->installed = true; if (displaced_rule) { - update_stats(rule, &put.flow.stats); + update_stats(p, displaced_rule, &put.flow.stats); rule_post_uninstall(p, displaced_rule); } } @@ -1642,14 +1675,27 @@ rule_reinstall(struct ofproto *ofproto, struct rule *rule) static void rule_update_actions(struct ofproto *ofproto, struct rule *rule) { - bool actions_changed = rule_make_actions(ofproto, rule, NULL); + bool actions_changed; + uint16_t new_out_iface, old_out_iface; + + old_out_iface = rule->nf_flow.output_iface; + actions_changed = rule_make_actions(ofproto, rule, NULL); + if (rule->may_install) { if (rule->installed) { if (actions_changed) { - /* XXX should really do rule_post_uninstall() for the *old* set - * of actions, and distinguish the old stats from the new. */ struct odp_flow_put put; - do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put); + do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY + | ODPPF_ZERO_STATS, &put); + update_stats(ofproto, rule, &put.flow.stats); + + /* Temporarily set the old output iface so that NetFlow + * messages have the correct output interface for the old + * stats. */ + new_out_iface = rule->nf_flow.output_iface; + rule->nf_flow.output_iface = old_out_iface; + rule_post_uninstall(ofproto, rule); + rule->nf_flow.output_iface = new_out_iface; } } else { rule_install(ofproto, rule, NULL); @@ -1685,7 +1731,7 @@ rule_uninstall(struct ofproto *p, struct rule *rule) odp_flow.actions = NULL; odp_flow.n_actions = 0; if (!dpif_flow_del(p->dpif, &odp_flow)) { - update_stats(rule, &odp_flow.stats); + update_stats(p, rule, &odp_flow.stats); } rule->installed = false; @@ -1693,39 +1739,51 @@ rule_uninstall(struct ofproto *p, struct rule *rule) } } +static bool +is_controller_rule(struct rule *rule) +{ + /* If the only action is send to the controller then don't report + * NetFlow expiration messages since it is just part of the control + * logic for the network and not real traffic. */ + + if (rule && rule->super) { + struct rule *super = rule->super; + + return super->n_actions == 1 && + super->actions[0].type == htons(OFPAT_OUTPUT) && + super->actions[0].output.port == htons(OFPP_CONTROLLER); + } + + return false; +} + static void rule_post_uninstall(struct ofproto *ofproto, struct rule *rule) { struct rule *super = rule->super; rule_account(ofproto, rule, 0); - if (ofproto->netflow && rule->byte_count) { + + if (ofproto->netflow && !is_controller_rule(rule)) { struct ofexpired expired; expired.flow = rule->cr.flow; expired.packet_count = rule->packet_count; expired.byte_count = rule->byte_count; expired.used = rule->used; - expired.created = rule->created; - expired.tcp_flags = rule->tcp_flags; - expired.ip_tos = rule->ip_tos; - netflow_expire(ofproto->netflow, &expired); + netflow_expire(ofproto->netflow, &rule->nf_flow, &expired); } if (super) { super->packet_count += rule->packet_count; super->byte_count += rule->byte_count; - super->tcp_flags |= rule->tcp_flags; - if (rule->packet_count) { - super->ip_tos = rule->ip_tos; - } - } - /* Reset counters to prevent double counting if the rule ever gets - * reinstalled. */ - rule->packet_count = 0; - rule->byte_count = 0; - rule->accounted_bytes = 0; - rule->tcp_flags = 0; - rule->ip_tos = 0; + /* Reset counters to prevent double counting if the rule ever gets + * reinstalled. */ + rule->packet_count = 0; + rule->byte_count = 0; + rule->accounted_bytes = 0; + + netflow_flow_clear(&rule->nf_flow); + } } static void @@ -1802,7 +1860,8 @@ handle_features_request(struct ofproto *p, struct ofconn *ofconn, osf->n_buffers = htonl(pktbuf_capacity()); osf->n_tables = 2; osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS | - OFPC_PORT_STATS | OFPC_MULTI_PHY_TX); + OFPC_PORT_STATS | OFPC_MULTI_PHY_TX | + OFPC_ARP_MATCH_IP); osf->actions = htonl((1u << OFPAT_OUTPUT) | (1u << OFPAT_SET_VLAN_VID) | (1u << OFPAT_SET_VLAN_PCP) | @@ -1811,6 +1870,7 @@ handle_features_request(struct ofproto *p, struct ofconn *ofconn, (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)); @@ -1834,9 +1894,6 @@ handle_get_config_request(struct ofproto *p, struct ofconn *ofconn, /* Figure out flags. */ dpif_get_drop_frags(p->dpif, &drop_frags); flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL; - if (ofconn->send_flow_exp) { - flags |= OFPC_SEND_FLOW_EXP; - } /* Send reply. */ osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf); @@ -1860,8 +1917,6 @@ handle_set_config(struct ofproto *p, struct ofconn *ofconn, } flags = ntohs(osc->flags); - ofconn->send_flow_exp = (flags & OFPC_SEND_FLOW_EXP) != 0; - if (ofconn == p->controller) { switch (flags & OFPC_FRAG_MASK) { case OFPC_FRAG_NORMAL: @@ -1891,9 +1946,14 @@ handle_set_config(struct ofproto *p, struct ofconn *ofconn, } static void -add_output_group_action(struct odp_actions *actions, uint16_t group) +add_output_group_action(struct odp_actions *actions, uint16_t group, + uint16_t *nf_output_iface) { odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group; + + if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) { + *nf_output_iface = NF_OUT_FLOOD; + } } static void @@ -1916,8 +1976,9 @@ struct action_xlate_ctx { /* Output. */ struct odp_actions *out; /* Datapath actions. */ tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */ - bool may_setup_flow; /* True ordinarily; false if the actions must + bool may_set_up_flow; /* True ordinarily; false if the actions must * be reassessed for every packet. */ + uint16_t nf_output_iface; /* Output interface index for NetFlow. */ }; static void do_xlate_actions(const union ofp_action *in, size_t n_in, @@ -1927,9 +1988,22 @@ static void add_output_action(struct action_xlate_ctx *ctx, uint16_t port) { const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port); - if (!ofport || !(ofport->opp.config & OFPPC_NO_FWD)) { - odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port; + + if (ofport) { + if (ofport->opp.config & OFPPC_NO_FWD) { + /* Forwarding disabled on port. */ + return; + } + } else { + /* + * We don't have an ofport record for this port, but it doesn't hurt to + * allow forwarding to it anyhow. Maybe such a port will appear later + * and we're pre-populating the flow table. + */ } + + odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port; + ctx->nf_output_iface = port; } static struct rule * @@ -1979,6 +2053,9 @@ xlate_output_action(struct action_xlate_ctx *ctx, const struct ofp_action_output *oao) { uint16_t odp_port; + uint16_t prev_nf_output_iface = ctx->nf_output_iface; + + ctx->nf_output_iface = NF_OUT_DROP; switch (ntohs(oao->port)) { case OFPP_IN_PORT: @@ -1990,16 +2067,18 @@ xlate_output_action(struct action_xlate_ctx *ctx, case OFPP_NORMAL: if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet, ctx->out, ctx->tags, + &ctx->nf_output_iface, ctx->ofproto->aux)) { COVERAGE_INC(ofproto_uninstallable); - ctx->may_setup_flow = false; + ctx->may_set_up_flow = false; } break; case OFPP_FLOOD: - add_output_group_action(ctx->out, DP_GROUP_FLOOD); + add_output_group_action(ctx->out, DP_GROUP_FLOOD, + &ctx->nf_output_iface); break; case OFPP_ALL: - add_output_group_action(ctx->out, DP_GROUP_ALL); + add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface); break; case OFPP_CONTROLLER: add_controller_action(ctx->out, oao); @@ -2014,6 +2093,15 @@ xlate_output_action(struct action_xlate_ctx *ctx, } break; } + + if (prev_nf_output_iface == NF_OUT_FLOOD) { + ctx->nf_output_iface = NF_OUT_FLOOD; + } else if (ctx->nf_output_iface == NF_OUT_DROP) { + ctx->nf_output_iface = prev_nf_output_iface; + } else if (prev_nf_output_iface != NF_OUT_DROP && + ctx->nf_output_iface != NF_OUT_FLOOD) { + ctx->nf_output_iface = NF_OUT_MULTI; + } } static void @@ -2092,11 +2180,25 @@ do_xlate_actions(const union ofp_action *in, size_t n_in, oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; break; + case OFPAT_SET_NW_DST: + oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST); + oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; + + case OFPAT_SET_NW_TOS: + oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS); + oa->nw_tos.nw_tos = ia->nw_tos.nw_tos; + break; + case OFPAT_SET_TP_SRC: oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC); oa->tp_port.tp_port = ia->tp_port.tp_port; break; + case OFPAT_SET_TP_DST: + oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST); + oa->tp_port.tp_port = ia->tp_port.tp_port; + break; + case OFPAT_VENDOR: xlate_nicira_action(ctx, (const struct nx_action_header *) ia); break; @@ -2112,7 +2214,8 @@ static int xlate_actions(const union ofp_action *in, size_t n_in, const flow_t *flow, struct ofproto *ofproto, const struct ofpbuf *packet, - struct odp_actions *out, tag_type *tags, bool *may_setup_flow) + struct odp_actions *out, tag_type *tags, bool *may_set_up_flow, + uint16_t *nf_output_iface) { tag_type no_tags = 0; struct action_xlate_ctx ctx; @@ -2124,17 +2227,21 @@ xlate_actions(const union ofp_action *in, size_t n_in, ctx.packet = packet; ctx.out = out; ctx.tags = tags ? tags : &no_tags; - ctx.may_setup_flow = true; + ctx.may_set_up_flow = true; + ctx.nf_output_iface = NF_OUT_DROP; do_xlate_actions(in, n_in, &ctx); - /* Check with in-band control to see if we're allowed to setup this + /* Check with in-band control to see if we're allowed to set up this * flow. */ if (!in_band_rule_check(ofproto->in_band, flow, out)) { - ctx.may_setup_flow = false; + ctx.may_set_up_flow = false; } - if (may_setup_flow) { - *may_setup_flow = ctx.may_setup_flow; + if (may_set_up_flow) { + *may_set_up_flow = ctx.may_set_up_flow; + } + if (nf_output_iface) { + *nf_output_iface = ctx.nf_output_iface; } if (odp_actions_overflow(out)) { odp_actions_init(out); @@ -2165,7 +2272,7 @@ handle_packet_out(struct ofproto *p, struct ofconn *ofconn, if (opo->buffer_id != htonl(UINT32_MAX)) { error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id), &buffer, &in_port); - if (error) { + if (error || !buffer) { return error; } payload = *buffer; @@ -2175,7 +2282,7 @@ handle_packet_out(struct ofproto *p, struct ofconn *ofconn, flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow); error = xlate_actions((const union ofp_action *) opo->actions, n_actions, - &flow, p, &payload, &actions, NULL, NULL); + &flow, p, &payload, &actions, NULL, NULL, NULL); if (error) { return error; } @@ -2208,7 +2315,7 @@ update_port_config(struct ofproto *p, struct ofport *port, #undef REVALIDATE_BITS if (mask & OFPPC_NO_FLOOD) { port->opp.config ^= OFPPC_NO_FLOOD; - refresh_port_group(p, DP_GROUP_FLOOD); + refresh_port_groups(p); } if (mask & OFPPC_NO_PACKET_IN) { port->opp.config ^= OFPPC_NO_PACKET_IN; @@ -2289,6 +2396,7 @@ handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn, strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc); strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc); strncpy(ods->serial_num, p->serial, sizeof ods->serial_num); + strncpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc); queue_tx(msg, ofconn, ofconn->reply_counter); return 0; @@ -2350,39 +2458,62 @@ handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn, return 0; } +static void +append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn, + struct ofpbuf *msg) +{ + struct netdev_stats stats; + struct ofp_port_stats *ops; + + /* 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); + + ops = append_stats_reply(sizeof *ops, ofconn, &msg); + ops->port_no = htons(odp_port_to_ofp_port(port_no)); + memset(ops->pad, 0, sizeof ops->pad); + ops->rx_packets = htonll(stats.rx_packets); + ops->tx_packets = htonll(stats.tx_packets); + ops->rx_bytes = htonll(stats.rx_bytes); + ops->tx_bytes = htonll(stats.tx_bytes); + ops->rx_dropped = htonll(stats.rx_dropped); + ops->tx_dropped = htonll(stats.tx_dropped); + ops->rx_errors = htonll(stats.rx_errors); + ops->tx_errors = htonll(stats.tx_errors); + ops->rx_frame_err = htonll(stats.rx_frame_errors); + ops->rx_over_err = htonll(stats.rx_over_errors); + ops->rx_crc_err = htonll(stats.rx_crc_errors); + ops->collisions = htonll(stats.collisions); +} + static int handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn, - struct ofp_stats_request *request) + struct ofp_stats_request *osr, + size_t arg_size) { + struct ofp_port_stats_request *psr; struct ofp_port_stats *ops; struct ofpbuf *msg; struct ofport *port; unsigned int port_no; - msg = start_stats_reply(request, sizeof *ops * 16); - PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { - struct netdev_stats stats; - - /* 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); - - ops = append_stats_reply(sizeof *ops, ofconn, &msg); - ops->port_no = htons(odp_port_to_ofp_port(port_no)); - memset(ops->pad, 0, sizeof ops->pad); - ops->rx_packets = htonll(stats.rx_packets); - ops->tx_packets = htonll(stats.tx_packets); - ops->rx_bytes = htonll(stats.rx_bytes); - ops->tx_bytes = htonll(stats.tx_bytes); - ops->rx_dropped = htonll(stats.rx_dropped); - ops->tx_dropped = htonll(stats.tx_dropped); - ops->rx_errors = htonll(stats.rx_errors); - ops->tx_errors = htonll(stats.tx_errors); - ops->rx_frame_err = htonll(stats.rx_frame_errors); - ops->rx_over_err = htonll(stats.rx_over_errors); - ops->rx_crc_err = htonll(stats.rx_crc_errors); - ops->collisions = htonll(stats.collisions); + if (arg_size != sizeof *psr) { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); + } + psr = (struct ofp_port_stats_request *) osr->body; + + msg = start_stats_reply(osr, sizeof *ops * 16); + if (psr->port_no != htons(OFPP_NONE)) { + port = port_array_get(&p->ports, + ofp_port_to_odp_port(ntohs(psr->port_no))); + if (port) { + append_port_stat(port, ntohs(psr->port_no), ofconn, msg); + } + } else { + PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { + append_port_stat(port, port_no, ofconn, msg); + } } queue_tx(msg, ofconn, ofconn->reply_counter); @@ -2405,12 +2536,17 @@ query_stats(struct ofproto *p, struct rule *rule, struct odp_flow *odp_flows; size_t n_odp_flows; + packet_count = rule->packet_count; + byte_count = rule->byte_count; + n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1; - odp_flows = xcalloc(1, n_odp_flows * sizeof *odp_flows); + odp_flows = xzalloc(n_odp_flows * sizeof *odp_flows); if (rule->cr.wc.wildcards) { size_t i = 0; LIST_FOR_EACH (subrule, struct rule, list, &rule->list) { odp_flows[i++].key = subrule->cr.flow; + packet_count += subrule->packet_count; + byte_count += subrule->byte_count; } } else { odp_flows[0].key = rule->cr.flow; @@ -2440,6 +2576,9 @@ flow_stats_cb(struct cls_rule *rule_, void *cbdata_) struct ofp_flow_stats *ofs; uint64_t packet_count, byte_count; size_t act_len, len; + long long int tdiff = time_msec() - rule->created; + uint32_t sec = tdiff / 1000; + uint32_t msec = tdiff - (sec * 1000); if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { return; @@ -2455,7 +2594,9 @@ flow_stats_cb(struct cls_rule *rule_, void *cbdata_) ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH; ofs->pad = 0; flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match); - ofs->duration = htonl((time_msec() - rule->created) / 1000); + ofs->duration_sec = htonl(sec); + ofs->duration_nsec = htonl(msec * 1000000); + ofs->cookie = rule->flow_cookie; ofs->priority = htons(rule->cr.priority); ofs->idle_timeout = htons(rule->idle_timeout); ofs->hard_timeout = htons(rule->hard_timeout); @@ -2484,7 +2625,7 @@ handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn, struct cls_rule target; if (arg_size != sizeof *fsr) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH); + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); } fsr = (struct ofp_flow_stats_request *) osr->body; @@ -2522,7 +2663,7 @@ flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_) } query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); - flow_to_ovs_match(&rule->cr.flow, rule->cr.wc.wildcards, &match); + flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &match); ds_put_format(results, "duration=%llds, ", (time_msec() - rule->created) / 1000); @@ -2592,7 +2733,7 @@ handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn, struct ofpbuf *msg; if (arg_size != sizeof *asr) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH); + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); } asr = (struct ofp_aggregate_stats_request *) osr->body; @@ -2645,7 +2786,7 @@ handle_stats_request(struct ofproto *p, struct ofconn *ofconn, return handle_table_stats_request(p, ofconn, osr); case OFPST_PORT: - return handle_port_stats_request(p, ofconn, osr); + return handle_port_stats_request(p, ofconn, osr, arg_size); case OFPST_VENDOR: return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); @@ -2662,23 +2803,29 @@ msec_from_nsec(uint64_t sec, uint32_t nsec) } static void -update_time(struct rule *rule, const struct odp_flow_stats *stats) +update_time(struct ofproto *ofproto, struct rule *rule, + const struct odp_flow_stats *stats) { long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec); if (used > rule->used) { rule->used = used; + if (rule->super && used > rule->super->used) { + rule->super->used = used; + } + netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used); } } static void -update_stats(struct rule *rule, const struct odp_flow_stats *stats) +update_stats(struct ofproto *ofproto, struct rule *rule, + const struct odp_flow_stats *stats) { - update_time(rule, stats); - rule->packet_count += stats->n_packets; - rule->byte_count += stats->n_bytes; - rule->tcp_flags |= stats->tcp_flags; if (stats->n_packets) { - rule->ip_tos = stats->ip_tos; + update_time(ofproto, rule, stats); + rule->packet_count += stats->n_packets; + rule->byte_count += stats->n_bytes; + netflow_flow_update_flags(&rule->nf_flow, stats->ip_tos, + stats->tcp_flags); } } @@ -2691,16 +2838,30 @@ add_flow(struct ofproto *p, struct ofconn *ofconn, uint16_t in_port; int error; - rule = rule_create(NULL, (const union ofp_action *) ofm->actions, + if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) { + flow_t flow; + uint32_t wildcards; + + flow_from_match(&flow, &wildcards, &ofm->match); + if (classifier_rule_overlaps(&p->cls, &flow, wildcards, + ntohs(ofm->priority))) { + return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP); + } + } + + rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions, n_actions, ntohs(ofm->idle_timeout), - ntohs(ofm->hard_timeout)); + ntohs(ofm->hard_timeout), ofm->cookie, + ofm->flags & htons(OFPFF_SEND_FLOW_REM)); cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority)); - packet = NULL; error = 0; if (ofm->buffer_id != htonl(UINT32_MAX)) { error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id), &packet, &in_port); + } else { + packet = NULL; + in_port = UINT16_MAX; } rule_insert(p, rule, packet, in_port); @@ -2717,6 +2878,8 @@ modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm, } if (command == OFPFC_DELETE) { + long long int now = time_msec(); + send_flow_removed(p, rule, now, OFPRR_DELETE); rule_remove(p, rule); } else { size_t actions_len = n_actions * sizeof *rule->actions; @@ -2730,6 +2893,7 @@ modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm, free(rule->actions); rule->actions = xmemdup(ofm->actions, actions_len); rule->n_actions = n_actions; + rule->flow_cookie = ofm->cookie; if (rule->cr.wc.wildcards) { COVERAGE_INC(ofproto_mod_wc_flow); @@ -2824,6 +2988,14 @@ handle_flow_mod(struct ofproto *p, struct ofconn *ofconn, return error; } + /* We do not support the emergency flow cache. It will hopefully + * get dropped from OpenFlow in the near future. */ + if (ofm->flags & htons(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); + } + normalize_match(&ofm->match); if (!ofm->match.wildcards) { ofm->priority = htons(UINT16_MAX); @@ -2856,58 +3028,6 @@ handle_flow_mod(struct ofproto *p, struct ofconn *ofconn, } } -static void -send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid) -{ - struct ofmp_capability_reply *ocr; - struct ofpbuf *b; - char capabilities[] = "com.nicira.mgmt.manager=false\n"; - - ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b); - ocr->header.header.vendor = htonl(NX_VENDOR_ID); - ocr->header.header.subtype = htonl(NXT_MGMT); - ocr->header.type = htons(OFMPT_CAPABILITY_REPLY); - - ocr->format = htonl(OFMPCOF_SIMPLE); - ocr->mgmt_id = htonll(p->mgmt_id); - - ofpbuf_put(b, capabilities, strlen(capabilities)); - - queue_tx(b, ofconn, ofconn->reply_counter); -} - -static int -handle_ofmp(struct ofproto *p, struct ofconn *ofconn, - struct ofmp_header *ofmph) -{ - size_t msg_len = ntohs(ofmph->header.header.length); - if (msg_len < sizeof(*ofmph)) { - VLOG_WARN_RL(&rl, "dropping short managment message: %d\n", msg_len); - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH); - } - - if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) { - struct ofmp_capability_request *ofmpcr; - - if (msg_len < sizeof(struct ofmp_capability_request)) { - VLOG_WARN_RL(&rl, "dropping short capability request: %d\n", - msg_len); - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH); - } - - ofmpcr = (struct ofmp_capability_request *)ofmph; - if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) { - /* xxx Find a better type than bad subtype */ - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); - } - - send_capability_reply(p, ofconn, ofmph->header.header.xid); - return 0; - } else { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); - } -} - static int handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg) { @@ -2915,13 +3035,13 @@ handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg) struct nicira_header *nh; if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH); + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); } if (ovh->vendor != htonl(NX_VENDOR_ID)) { return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); } if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH); + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); } nh = msg; @@ -2929,26 +3049,24 @@ handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg) case NXT_STATUS_REQUEST: return switch_status_handle_request(p->switch_status, ofconn->rconn, msg); - - case NXT_ACT_SET_CONFIG: - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */ - - case NXT_ACT_GET_CONFIG: - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */ - - case NXT_COMMAND_REQUEST: - if (p->executer) { - return executer_handle_request(p->executer, ofconn->rconn, msg); - } - break; - - case NXT_MGMT: - return handle_ofmp(p, ofconn, msg); } return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); } +static int +handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh) +{ + struct ofp_header *ob; + struct ofpbuf *buf; + + /* Currently, everything executes synchronously, so we can just + * immediately send the barrier reply. */ + ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf); + queue_tx(buf, ofconn, ofconn->reply_counter); + return 0; +} + static void handle_openflow(struct ofconn *ofconn, struct ofproto *p, struct ofpbuf *ofp_msg) @@ -2998,6 +3116,10 @@ handle_openflow(struct ofconn *ofconn, struct ofproto *p, error = handle_vendor(p, ofconn, ofp_msg->data); break; + case OFPT_BARRIER_REQUEST: + error = handle_barrier_request(ofconn, oh); + break; + default: if (VLOG_IS_WARN_ENABLED()) { char *s = ofp_to_string(oh, ntohs(oh->length), 2); @@ -3014,7 +3136,7 @@ handle_openflow(struct ofconn *ofconn, struct ofproto *p, } static void -handle_odp_msg(struct ofproto *p, struct ofpbuf *packet) +handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) { struct odp_msg *msg = packet->data; uint16_t in_port = odp_port_to_ofp_port(msg->port); @@ -3022,14 +3144,6 @@ handle_odp_msg(struct ofproto *p, struct ofpbuf *packet) struct ofpbuf payload; flow_t flow; - /* Handle controller actions. */ - if (msg->type == _ODPL_ACTION_NR) { - COVERAGE_INC(ofproto_ctlr_action); - pinsched_send(p->action_sched, in_port, packet, - send_packet_in_action, p); - return; - } - payload.data = msg + 1; payload.size = msg->length - sizeof *msg; flow_extract(&payload, msg->port, &flow); @@ -3081,7 +3195,53 @@ handle_odp_msg(struct ofproto *p, struct ofpbuf *packet) rule_execute(p, rule, &payload, &flow); rule_reinstall(p, rule); - ofpbuf_delete(packet); + + if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY + && rconn_is_connected(p->controller->rconn)) { + /* + * Extra-special case for fail-open mode. + * + * We are in fail-open mode and the packet matched the fail-open rule, + * but we are connected to a controller too. We should send the packet + * up to the controller in the hope that it will try to set up a flow + * and thereby allow us to exit fail-open. + * + * See the top-level comment in fail-open.c for more information. + */ + pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p); + } else { + ofpbuf_delete(packet); + } +} + +static void +handle_odp_msg(struct ofproto *p, struct ofpbuf *packet) +{ + struct odp_msg *msg = packet->data; + + switch (msg->type) { + case _ODPL_ACTION_NR: + COVERAGE_INC(ofproto_ctlr_action); + pinsched_send(p->action_sched, odp_port_to_ofp_port(msg->port), packet, + send_packet_in_action, p); + break; + + case _ODPL_SFLOW_NR: + if (p->sflow) { + ofproto_sflow_received(p->sflow, msg); + } + ofpbuf_delete(packet); + break; + + case _ODPL_MISS_NR: + handle_odp_miss_msg(p, packet); + break; + + default: + VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32, + msg->type); + break; + } } static void @@ -3126,25 +3286,43 @@ revalidate_rule(struct ofproto *p, struct rule *rule) } static struct ofpbuf * -compose_flow_exp(const struct rule *rule, long long int now, uint8_t reason) +compose_flow_removed(const struct rule *rule, long long int now, uint8_t reason) { - struct ofp_flow_expired *ofe; + struct ofp_flow_removed *ofr; struct ofpbuf *buf; - - ofe = make_openflow(sizeof *ofe, OFPT_FLOW_EXPIRED, &buf); - flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofe->match); - ofe->priority = htons(rule->cr.priority); - ofe->reason = reason; - ofe->duration = (now - rule->created) / 1000; - ofe->packet_count = rule->packet_count; - ofe->byte_count = rule->byte_count; + long long int tdiff = time_msec() - rule->created; + uint32_t sec = tdiff / 1000; + uint32_t msec = tdiff - (sec * 1000); + + ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf); + flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofr->match); + ofr->cookie = rule->flow_cookie; + ofr->priority = htons(rule->cr.priority); + ofr->reason = reason; + ofr->duration_sec = htonl(sec); + ofr->duration_nsec = htonl(msec * 1000000); + ofr->idle_timeout = htons(rule->idle_timeout); + ofr->packet_count = htonll(rule->packet_count); + ofr->byte_count = htonll(rule->byte_count); return buf; } static void -send_flow_exp(struct ofproto *p, struct rule *rule, - long long int now, uint8_t reason) +uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule) +{ + assert(rule->installed); + assert(!rule->cr.wc.wildcards); + + if (rule->super) { + rule_remove(ofproto, rule); + } else { + rule_uninstall(ofproto, rule); + } +} +static void +send_flow_removed(struct ofproto *p, struct rule *rule, + long long int now, uint8_t reason) { struct ofconn *ofconn; struct ofconn *prev; @@ -3158,32 +3336,20 @@ send_flow_exp(struct ofproto *p, struct rule *rule, prev = NULL; LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { - if (ofconn->send_flow_exp && rconn_is_connected(ofconn->rconn)) { + if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn)) { if (prev) { - queue_tx(ofpbuf_clone(buf), prev, ofconn->reply_counter); + queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter); } else { - buf = compose_flow_exp(rule, now, reason); + buf = compose_flow_removed(rule, now, reason); } prev = ofconn; } } if (prev) { - queue_tx(buf, prev, ofconn->reply_counter); + queue_tx(buf, prev, prev->reply_counter); } } -static void -uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule) -{ - assert(rule->installed); - assert(!rule->cr.wc.wildcards); - - if (rule->super) { - rule_remove(ofproto, rule); - } else { - rule_uninstall(ofproto, rule); - } -} static void expire_rule(struct cls_rule *cls_rule, void *p_) @@ -3200,38 +3366,75 @@ expire_rule(struct cls_rule *cls_rule, void *p_) ? rule->used + rule->idle_timeout * 1000 : LLONG_MAX); expire = MIN(hard_expire, idle_expire); - if (expire == LLONG_MAX) { - if (rule->installed && time_msec() >= rule->used + 5000) { - uninstall_idle_flow(p, rule); - } - return; - } now = time_msec(); if (now < expire) { if (rule->installed && now >= rule->used + 5000) { uninstall_idle_flow(p, rule); + } else if (!rule->cr.wc.wildcards) { + active_timeout(p, rule); } + return; } COVERAGE_INC(ofproto_expired); + + /* Update stats. This code will be a no-op if the rule expired + * due to an idle timeout. */ if (rule->cr.wc.wildcards) { - /* Update stats. (This code will be a no-op if the rule expired - * due to an idle timeout, because in that case the rule has no - * subrules left.) */ struct rule *subrule, *next; LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) { rule_remove(p, subrule); } + } else { + rule_uninstall(p, rule); } - send_flow_exp(p, rule, now, - (now >= hard_expire - ? OFPER_HARD_TIMEOUT : OFPER_IDLE_TIMEOUT)); + if (!rule_is_hidden(rule)) { + send_flow_removed(p, rule, now, + (now >= hard_expire + ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT)); + } rule_remove(p, rule); } +static void +active_timeout(struct ofproto *ofproto, struct rule *rule) +{ + if (ofproto->netflow && !is_controller_rule(rule) && + netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) { + struct ofexpired expired; + struct odp_flow odp_flow; + + /* Get updated flow stats. */ + memset(&odp_flow, 0, sizeof odp_flow); + if (rule->installed) { + odp_flow.key = rule->cr.flow; + odp_flow.flags = ODPFF_ZERO_TCP_FLAGS; + dpif_flow_get(ofproto->dpif, &odp_flow); + + if (odp_flow.stats.n_packets) { + update_time(ofproto, rule, &odp_flow.stats); + netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos, + odp_flow.stats.tcp_flags); + } + } + + expired.flow = rule->cr.flow; + expired.packet_count = rule->packet_count + + odp_flow.stats.n_packets; + expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes; + expired.used = rule->used; + + netflow_expire(ofproto->netflow, &rule->nf_flow, &expired); + + /* Schedule us to send the accumulated records once we have + * collected all of them. */ + poll_immediate_wake(); + } +} + static void update_used(struct ofproto *p) { @@ -3257,7 +3460,7 @@ update_used(struct ofproto *p) continue; } - update_time(rule, &f->stats); + update_time(p, rule, &f->stats); rule_account(p, rule, f->stats.n_bytes); } free(flows); @@ -3267,25 +3470,22 @@ static void do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id, const struct ofpbuf *packet, int send_len) { - struct ofp_packet_in *opi; - struct ofpbuf payload, *buf; - struct odp_msg *msg; + struct odp_msg *msg = packet->data; + struct ofpbuf payload; + struct ofpbuf *opi; + uint8_t reason; - msg = packet->data; + /* Extract packet payload from 'msg'. */ payload.data = msg + 1; payload.size = msg->length - sizeof *msg; - send_len = MIN(send_len, payload.size); - buf = ofpbuf_new(sizeof *opi + send_len); - opi = put_openflow_xid(offsetof(struct ofp_packet_in, data), - OFPT_PACKET_IN, 0, buf); - opi->buffer_id = htonl(buffer_id); - opi->total_len = htons(payload.size); - opi->in_port = htons(odp_port_to_ofp_port(msg->port)); - opi->reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH; - ofpbuf_put(buf, payload.data, MIN(send_len, payload.size)); - update_openflow_length(buf); - rconn_send_with_limit(ofconn->rconn, buf, ofconn->packet_in_counter, 100); + /* Construct ofp_packet_in message. */ + reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH; + opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason, + &payload, send_len); + + /* Send. */ + rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100); } static void @@ -3308,6 +3508,7 @@ static void send_packet_in_miss(struct ofpbuf *packet, void *p_) { struct ofproto *p = p_; + bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open); struct ofconn *ofconn; struct ofpbuf payload; struct odp_msg *msg; @@ -3317,8 +3518,10 @@ send_packet_in_miss(struct ofpbuf *packet, void *p_) payload.size = msg->length - sizeof *msg; LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { if (ofconn->miss_send_len) { - uint32_t buffer_id = pktbuf_save(ofconn->pktbuf, &payload, - msg->port); + struct pktbuf *pb = ofconn->pktbuf; + uint32_t buffer_id = (in_fail_open + ? pktbuf_get_null() + : pktbuf_save(pb, &payload, msg->port)); int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len : UINT32_MAX); do_send_packet_in(ofconn, buffer_id, packet, send_len); @@ -3351,17 +3554,14 @@ static uint64_t pick_fallback_dpid(void) { uint8_t ea[ETH_ADDR_LEN]; - eth_addr_random(ea); - ea[0] = 0x00; /* Set Nicira OUI. */ - ea[1] = 0x23; - ea[2] = 0x20; + eth_addr_nicira_random(ea); return eth_addr_to_uint64(ea); } static bool default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, struct odp_actions *actions, tag_type *tags, - void *ofproto_) + uint16_t *nf_output_iface, void *ofproto_) { struct ofproto *ofproto = ofproto_; int out_port; @@ -3388,9 +3588,10 @@ default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, /* Determine output port. */ out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags); if (out_port < 0) { - add_output_group_action(actions, DP_GROUP_FLOOD); + add_output_group_action(actions, DP_GROUP_FLOOD, nf_output_iface); } else if (out_port != flow->in_port) { odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port; + *nf_output_iface = out_port; } else { /* Drop. */ }