X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=ofproto%2Fofproto.c;h=2857757295295dc90eea6a5870134f4b980a801e;hb=3b6a2571f07e153e850a9bf2044699d8d4434ef0;hp=4b9ceebb738743d5b2fb78236f99b07b0c973b0e;hpb=750638bbefd179b91f8fb3c9c1bde855af05cf45;p=sliver-openvswitch.git diff --git a/ofproto/ofproto.c b/ofproto/ofproto.c index 4b9ceebb7..0b85e618c 100644 --- a/ofproto/ofproto.c +++ b/ofproto/ofproto.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2009, 2010 Nicira Networks. + * Copyright (c) 2009, 2010, 2011 Nicira Networks. * Copyright (c) 2010 Jean Tourrilhes - HP-Labs. * * Licensed under the Apache License, Version 2.0 (the "License"); @@ -24,16 +24,24 @@ #include #include #include +#include "autopath.h" +#include "byte-order.h" +#include "cfm.h" #include "classifier.h" +#include "connmgr.h" #include "coverage.h" -#include "discovery.h" #include "dpif.h" #include "dynamic-string.h" #include "fail-open.h" +#include "hash.h" +#include "hmap.h" #include "in-band.h" #include "mac-learning.h" +#include "multipath.h" #include "netdev.h" #include "netflow.h" +#include "netlink.h" +#include "nx-match.h" #include "odp-util.h" #include "ofp-print.h" #include "ofp-util.h" @@ -46,218 +54,226 @@ #include "pinsched.h" #include "pktbuf.h" #include "poll-loop.h" -#include "port-array.h" #include "rconn.h" #include "shash.h" -#include "status.h" +#include "sset.h" #include "stream-ssl.h" -#include "svec.h" #include "tag.h" +#include "timer.h" #include "timeval.h" +#include "unaligned.h" #include "unixctl.h" #include "vconn.h" #include "vlog.h" -#include "xtoxll.h" -VLOG_DEFINE_THIS_MODULE(ofproto) +VLOG_DEFINE_THIS_MODULE(ofproto); + +COVERAGE_DEFINE(facet_changed_rule); +COVERAGE_DEFINE(facet_revalidate); +COVERAGE_DEFINE(odp_overflow); +COVERAGE_DEFINE(ofproto_agg_request); +COVERAGE_DEFINE(ofproto_costly_flags); +COVERAGE_DEFINE(ofproto_ctlr_action); +COVERAGE_DEFINE(ofproto_del_rule); +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); -#include "sflow_api.h" +/* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a + * flow translation. */ +#define MAX_RESUBMIT_RECURSION 16 -enum { - TABLEID_HASH = 0, - TABLEID_CLASSIFIER = 1 -}; +struct rule; struct ofport { + struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */ struct netdev *netdev; struct ofp_phy_port opp; /* In host byte order. */ + uint16_t odp_port; + struct cfm *cfm; /* Connectivity Fault Management, if any. */ }; static void ofport_free(struct ofport *); -static void hton_ofp_phy_port(struct ofp_phy_port *); +static void ofport_run(struct ofproto *, struct ofport *); +static void ofport_wait(struct ofport *); -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_set_up_flow, uint16_t *nf_output_iface); +struct action_xlate_ctx { +/* action_xlate_ctx_init() initializes these members. */ -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. */ - tag_type tags; /* Tags (set only by hooks). */ - struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ + /* The ofproto. */ + struct ofproto *ofproto; - /* 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 - * wildcard rule 'super'. In this case, 'list' is an element of the - * super-rule's list. - * - * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of - * a list of subrules. A super-rule with no wildcards (where - * cr.wc.wildcards is 0) will never have any subrules. */ - struct rule *super; - struct list list; + /* Flow to which the OpenFlow actions apply. xlate_actions() will modify + * this flow when actions change header fields. */ + struct flow flow; - /* OpenFlow actions. - * - * 'n_actions' is the number of elements in the 'actions' array. A single - * action may take up more more than one element's worth of space. - * - * A subrule has no actions (it uses the super-rule's actions). */ - int n_actions; - union ofp_action *actions; + /* The packet corresponding to 'flow', or a null pointer if we are + * revalidating without a packet to refer to. */ + const struct ofpbuf *packet; - /* Datapath actions. + /* If nonnull, called just before executing a resubmit action. * - * A super-rule with wildcard fields never has ODP actions (since the - * datapath only supports exact-match flows). */ - bool installed; /* Installed in datapath? */ - bool may_install; /* True ordinarily; false if actions must + * This is normally null so the client has to set it manually after + * calling action_xlate_ctx_init(). */ + void (*resubmit_hook)(struct action_xlate_ctx *, struct rule *); + + /* If true, the speciality of 'flow' should be checked before executing + * its actions. If special_cb returns false on 'flow' rendered + * uninstallable and no actions will be executed. */ + bool check_special; + +/* xlate_actions() initializes and uses these members. The client might want + * to look at them after it returns. */ + + struct ofpbuf *odp_actions; /* Datapath actions. */ + tag_type tags; /* Tags associated with OFPP_NORMAL actions. */ + bool may_set_up_flow; /* True ordinarily; false if the actions must * be reassessed for every packet. */ - int n_odp_actions; - union odp_action *odp_actions; + uint16_t nf_output_iface; /* Output interface index for NetFlow. */ + +/* xlate_actions() initializes and uses these members, but the client has no + * reason to look at them. */ + + int recurse; /* Recursion level, via xlate_table_action. */ + int last_pop_priority; /* Offset in 'odp_actions' just past most + * recent ODP_ACTION_ATTR_SET_PRIORITY. */ }; -static inline bool -rule_is_hidden(const struct rule *rule) -{ - /* Subrules are merely an implementation detail, so hide them from the - * controller. */ - if (rule->super != NULL) { - return true; - } +static void action_xlate_ctx_init(struct action_xlate_ctx *, + struct ofproto *, const struct flow *, + const struct ofpbuf *); +static struct ofpbuf *xlate_actions(struct action_xlate_ctx *, + const union ofp_action *in, size_t n_in); - /* Rules with priority higher than UINT16_MAX are set up by ofproto itself - * (e.g. by in-band control) and are intentionally hidden from the - * controller. */ - if (rule->cr.priority > UINT16_MAX) { - return true; - } +/* An OpenFlow flow. */ +struct rule { + long long int used; /* Time last used; time created if not used. */ + long long int created; /* Creation time. */ - return false; -} + /* These statistics: + * + * - Do include packets and bytes from facets that have been deleted or + * whose own statistics have been folded into the rule. + * + * - Do include packets and bytes sent "by hand" that were accounted to + * the rule without any facet being involved (this is a rare corner + * case in rule_execute()). + * + * - Do not include packet or bytes that can be obtained from any facet's + * packet_count or byte_count member or that can be obtained from the + * datapath by, e.g., dpif_flow_get() for any facet. + */ + uint64_t packet_count; /* Number of packets received. */ + uint64_t byte_count; /* Number of bytes received. */ + + ovs_be64 flow_cookie; /* Controller-issued identifier. */ + + struct cls_rule cr; /* In owning ofproto's classifier. */ + 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? */ + int n_actions; /* Number of elements in actions[]. */ + union ofp_action *actions; /* OpenFlow actions. */ + struct list facets; /* List of "struct facet"s. */ +}; + +static struct rule *rule_from_cls_rule(const struct cls_rule *); +static bool rule_is_hidden(const struct rule *); -static struct rule *rule_create(struct ofproto *, struct rule *super, +static struct rule *rule_create(const struct cls_rule *, 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 *); + ovs_be64 flow_cookie, bool send_flow_removed); static void rule_destroy(struct ofproto *, struct rule *); -static struct rule *rule_from_cls_rule(const struct cls_rule *); -static void rule_insert(struct ofproto *, struct rule *, - struct ofpbuf *packet, uint16_t in_port); +static void rule_free(struct rule *); + +static struct rule *rule_lookup(struct ofproto *, const struct flow *); +static void rule_insert(struct ofproto *, struct rule *); static void rule_remove(struct ofproto *, struct rule *); -static bool rule_make_actions(struct ofproto *, struct rule *, - const struct ofpbuf *packet); -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); - -/* ofproto supports two kinds of OpenFlow connections: - * - * - "Primary" connections to ordinary OpenFlow controllers. ofproto - * maintains persistent connections to these controllers and by default - * sends them asynchronous messages such as packet-ins. - * - * - "Service" connections, e.g. from ovs-ofctl. When these connections - * drop, it is the other side's responsibility to reconnect them if - * necessary. ofproto does not send them asynchronous messages by default. - * - * Currently, active (tcp, ssl, unix) connections are always "primary" - * connections and passive (ptcp, pssl, punix) connections are always "service" - * connections. There is no inherent reason for this, but it reflects the - * common case. - */ -enum ofconn_type { - OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */ - OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */ -}; -/* A listener for incoming OpenFlow "service" connections. */ -struct ofservice { - struct hmap_node node; /* In struct ofproto's "services" hmap. */ - struct pvconn *pvconn; /* OpenFlow connection listener. */ +static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason); +static void rule_get_stats(const struct rule *, uint64_t *packets, + uint64_t *bytes); - /* These are not used by ofservice directly. They are settings for - * accepted "struct ofconn"s from the pvconn. */ - int probe_interval; /* Max idle time before probing, in seconds. */ - int rate_limit; /* Max packet-in rate in packets per second. */ - int burst_limit; /* Limit on accumulating packet credits. */ -}; +/* An exact-match instantiation of an OpenFlow flow. */ +struct facet { + long long int used; /* Time last used; time created if not used. */ -static struct ofservice *ofservice_lookup(struct ofproto *, - const char *target); -static int ofservice_create(struct ofproto *, - const struct ofproto_controller *); -static void ofservice_reconfigure(struct ofservice *, - const struct ofproto_controller *); -static void ofservice_destroy(struct ofproto *, struct ofservice *); - -/* An OpenFlow connection. */ -struct ofconn { - struct ofproto *ofproto; /* The ofproto that owns this connection. */ - struct list node; /* In struct ofproto's "all_conns" list. */ - struct rconn *rconn; /* OpenFlow connection. */ - enum ofconn_type type; /* Type. */ - - /* OFPT_PACKET_IN related data. */ - struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */ - struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */ - struct pktbuf *pktbuf; /* OpenFlow packet buffers. */ - int miss_send_len; /* Bytes to send of buffered packets. */ - - /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow - * requests, and the maximum number before we stop reading OpenFlow - * requests. */ -#define OFCONN_REPLY_MAX 100 - struct rconn_packet_counter *reply_counter; - - /* type == OFCONN_PRIMARY only. */ - enum nx_role role; /* Role. */ - struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */ - struct discovery *discovery; /* Controller discovery object, if enabled. */ - struct status_category *ss; /* Switch status category. */ - enum ofproto_band band; /* In-band or out-of-band? */ + /* These statistics: + * + * - Do include packets and bytes sent "by hand", e.g. with + * dpif_execute(). + * + * - Do include packets and bytes that were obtained from the datapath + * when a flow was deleted (e.g. dpif_flow_del()) or when its + * statistics were reset (e.g. dpif_flow_put() with + * DPIF_FP_ZERO_STATS). + * + * - Do not include any packets or bytes that can currently be obtained + * from the datapath by, e.g., dpif_flow_get(). + */ + uint64_t packet_count; /* Number of packets received. */ + uint64_t byte_count; /* Number of bytes received. */ + + uint64_t dp_packet_count; /* Last known packet count in the datapath. */ + uint64_t dp_byte_count; /* Last known byte count in the datapath. */ + + uint64_t rs_packet_count; /* Packets pushed to resubmit children. */ + uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */ + long long int rs_used; /* Used time pushed to resubmit children. */ + + /* Number of bytes passed to account_cb. This may include bytes that can + * currently obtained from the datapath (thus, it can be greater than + * byte_count). */ + uint64_t accounted_bytes; + + struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ + struct list list_node; /* In owning rule's 'facets' list. */ + struct rule *rule; /* Owning rule. */ + struct flow flow; /* Exact-match flow. */ + bool installed; /* Installed in datapath? */ + bool may_install; /* True ordinarily; false if actions must + * be reassessed for every packet. */ + size_t actions_len; /* Number of bytes in actions[]. */ + struct nlattr *actions; /* Datapath actions. */ + tag_type tags; /* Tags (set only by hooks). */ + struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ }; -/* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's - * "schedulers" array. Their values are 0 and 1, and their meanings and values - * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In - * case anything ever changes, check their values here. */ -#define N_SCHEDULERS 2 -BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0); -BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR); -BUILD_ASSERT_DECL(OFPR_ACTION == 1); -BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR); - -static struct ofconn *ofconn_create(struct ofproto *, struct rconn *, - enum ofconn_type); -static void ofconn_destroy(struct ofconn *); -static void ofconn_run(struct ofconn *, struct ofproto *); -static void ofconn_wait(struct ofconn *); -static bool ofconn_receives_async_msgs(const struct ofconn *); -static char *ofconn_make_name(const struct ofproto *, const char *target); -static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst); - -static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, - struct rconn_packet_counter *counter); - -static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg); -static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn); +static struct facet *facet_create(struct ofproto *, struct rule *, + const struct flow *, + const struct ofpbuf *packet); +static void facet_remove(struct ofproto *, struct facet *); +static void facet_free(struct facet *); + +static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *); +static bool facet_revalidate(struct ofproto *, struct facet *); + +static void facet_install(struct ofproto *, struct facet *, bool zero_stats); +static void facet_uninstall(struct ofproto *, struct facet *); +static void facet_flush_stats(struct ofproto *, struct facet *); + +static void facet_make_actions(struct ofproto *, struct facet *, + const struct ofpbuf *packet); +static void facet_update_stats(struct ofproto *, struct facet *, + const struct dpif_flow_stats *); +static void facet_push_stats(struct ofproto *, struct facet *); + +static void send_packet_in(struct ofproto *, struct dpif_upcall *, + const struct flow *, bool clone); struct ofproto { /* Settings. */ @@ -272,39 +288,25 @@ struct ofproto { /* Datapath. */ struct dpif *dpif; struct netdev_monitor *netdev_monitor; - struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is - * OFP port nr. */ + struct hmap ports; /* Contains "struct ofport"s. */ struct shash port_by_name; uint32_t max_ports; /* Configuration. */ - struct switch_status *switch_status; - struct fail_open *fail_open; struct netflow *netflow; struct ofproto_sflow *sflow; - /* In-band control. */ - struct in_band *in_band; - long long int next_in_band_update; - struct sockaddr_in *extra_in_band_remotes; - size_t n_extra_remotes; - /* Flow table. */ struct classifier cls; + struct timer next_expiration; + + /* Facets. */ + struct hmap facets; bool need_revalidate; - long long int next_expiration; struct tag_set revalidate_set; - bool tun_id_from_cookie; /* OpenFlow connections. */ - struct hmap controllers; /* Controller "struct ofconn"s. */ - struct list all_conns; /* Contains "struct ofconn"s. */ - enum ofproto_fail_mode fail_mode; - - /* OpenFlow listeners. */ - struct hmap services; /* Contains "struct ofservice"s. */ - struct pvconn **snoops; - size_t n_snoops; + struct connmgr *connmgr; /* Hooks for ovs-vswitchd. */ const struct ofhooks *ofhooks; @@ -314,6 +316,9 @@ struct ofproto { struct mac_learning *ml; }; +/* Map from dpif name to struct ofproto, for use by unixctl commands. */ +static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos); + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); static const struct ofhooks default_ofhooks; @@ -321,59 +326,62 @@ static const struct ofhooks default_ofhooks; static uint64_t pick_datapath_id(const struct ofproto *); static uint64_t pick_fallback_dpid(void); -static void update_used(struct ofproto *); -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_); +static int ofproto_expire(struct ofproto *); +static void flow_push_stats(struct ofproto *, const struct rule *, + struct flow *, uint64_t packets, uint64_t bytes, + long long int used); -static void handle_odp_msg(struct ofproto *, struct ofpbuf *); +static void handle_upcall(struct ofproto *, struct dpif_upcall *); -static void handle_openflow(struct ofconn *, struct ofproto *, - struct ofpbuf *); - -static void refresh_port_groups(struct ofproto *); +static void handle_openflow(struct ofconn *, struct ofpbuf *); +static struct ofport *get_port(const struct ofproto *, uint16_t odp_port); static void update_port(struct ofproto *, const char *devname); static int init_ports(struct ofproto *); static void reinit_ports(struct ofproto *); +static void ofproto_unixctl_init(void); + int ofproto_create(const char *datapath, const char *datapath_type, const struct ofhooks *ofhooks, void *aux, struct ofproto **ofprotop) { - struct odp_stats stats; + char local_name[IF_NAMESIZE]; struct ofproto *p; struct dpif *dpif; int error; *ofprotop = NULL; + ofproto_unixctl_init(); + /* Connect to datapath and start listening for messages. */ error = dpif_open(datapath, datapath_type, &dpif); if (error) { VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error)); return error; } - error = dpif_get_dp_stats(dpif, &stats); + error = dpif_recv_set_mask(dpif, + ((1u << DPIF_UC_MISS) | + (1u << DPIF_UC_ACTION) | + (1u << DPIF_UC_SAMPLE))); if (error) { - VLOG_ERR("failed to obtain stats for datapath %s: %s", + VLOG_ERR("failed to listen on datapath %s: %s", datapath, strerror(error)); dpif_close(dpif); return error; } - error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW); + dpif_flow_flush(dpif); + dpif_recv_purge(dpif); + + error = dpif_port_get_name(dpif, ODPP_LOCAL, + local_name, sizeof local_name); if (error) { - VLOG_ERR("failed to listen on datapath %s: %s", + VLOG_ERR("%s: cannot get name of datapath local port (%s)", datapath, strerror(error)); - dpif_close(dpif); return error; } - dpif_flow_flush(dpif); - dpif_recv_purge(dpif); /* Initialize settings. */ p = xzalloc(sizeof *p); @@ -388,30 +396,23 @@ ofproto_create(const char *datapath, const char *datapath_type, /* Initialize datapath. */ p->dpif = dpif; p->netdev_monitor = netdev_monitor_create(); - port_array_init(&p->ports); + hmap_init(&p->ports); shash_init(&p->port_by_name); - p->max_ports = stats.max_ports; + p->max_ports = dpif_get_max_ports(dpif); /* Initialize submodules. */ - p->switch_status = switch_status_create(p); - p->in_band = NULL; - p->fail_open = NULL; p->netflow = NULL; p->sflow = NULL; /* Initialize flow table. */ classifier_init(&p->cls); + timer_set_duration(&p->next_expiration, 1000); + + /* Initialize facet table. */ + hmap_init(&p->facets); p->need_revalidate = false; - p->next_expiration = time_msec() + 1000; tag_set_init(&p->revalidate_set); - /* Initialize OpenFlow connections. */ - list_init(&p->all_conns); - hmap_init(&p->controllers); - hmap_init(&p->services); - p->snoops = NULL; - p->n_snoops = 0; - /* Initialize hooks. */ if (ofhooks) { p->ofhooks = ofhooks; @@ -427,6 +428,11 @@ ofproto_create(const char *datapath, const char *datapath_type, p->datapath_id = pick_datapath_id(p); VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id); + shash_add_once(&all_ofprotos, dpif_name(p->dpif), p); + + /* Initialize OpenFlow connections. */ + p->connmgr = connmgr_create(p, datapath, local_name); + *ofprotop = p; return 0; } @@ -445,270 +451,18 @@ ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id) } } -static bool -is_discovery_controller(const struct ofproto_controller *c) -{ - return !strcmp(c->target, "discover"); -} - -static bool -is_in_band_controller(const struct ofproto_controller *c) -{ - return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND; -} - -/* Creates a new controller in 'ofproto'. Some of the settings are initially - * drawn from 'c', but update_controller() needs to be called later to finish - * the new ofconn's configuration. */ -static void -add_controller(struct ofproto *ofproto, const struct ofproto_controller *c) -{ - struct discovery *discovery; - struct ofconn *ofconn; - - if (is_discovery_controller(c)) { - int error = discovery_create(c->accept_re, c->update_resolv_conf, - ofproto->dpif, ofproto->switch_status, - &discovery); - if (error) { - return; - } - } else { - discovery = NULL; - } - - ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY); - ofconn->pktbuf = pktbuf_create(); - ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN; - if (discovery) { - ofconn->discovery = discovery; - } else { - char *name = ofconn_make_name(ofproto, c->target); - rconn_connect(ofconn->rconn, c->target, name); - free(name); - } - hmap_insert(&ofproto->controllers, &ofconn->hmap_node, - hash_string(c->target, 0)); -} - -/* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's - * target or turn discovery on or off (these are done by creating new ofconns - * and deleting old ones), but it can update the rest of an ofconn's - * settings. */ -static void -update_controller(struct ofconn *ofconn, const struct ofproto_controller *c) -{ - int probe_interval; - - ofconn->band = (is_in_band_controller(c) - ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND); - - rconn_set_max_backoff(ofconn->rconn, c->max_backoff); - - probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0; - rconn_set_probe_interval(ofconn->rconn, probe_interval); - - if (ofconn->discovery) { - discovery_set_update_resolv_conf(ofconn->discovery, - c->update_resolv_conf); - discovery_set_accept_controller_re(ofconn->discovery, c->accept_re); - } - - ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit); -} - -static const char * -ofconn_get_target(const struct ofconn *ofconn) -{ - return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn); -} - -static struct ofconn * -find_controller_by_target(struct ofproto *ofproto, const char *target) -{ - struct ofconn *ofconn; - - HMAP_FOR_EACH_WITH_HASH (ofconn, struct ofconn, hmap_node, - hash_string(target, 0), &ofproto->controllers) { - if (!strcmp(ofconn_get_target(ofconn), target)) { - return ofconn; - } - } - return NULL; -} - -static void -update_in_band_remotes(struct ofproto *ofproto) -{ - const struct ofconn *ofconn; - struct sockaddr_in *addrs; - size_t max_addrs, n_addrs; - bool discovery; - size_t i; - - /* Allocate enough memory for as many remotes as we could possibly have. */ - max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers); - addrs = xmalloc(max_addrs * sizeof *addrs); - n_addrs = 0; - - /* Add all the remotes. */ - discovery = false; - HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &ofproto->controllers) { - struct sockaddr_in *sin = &addrs[n_addrs]; - - if (ofconn->band == OFPROTO_OUT_OF_BAND) { - continue; - } - - sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn); - if (sin->sin_addr.s_addr) { - sin->sin_port = rconn_get_remote_port(ofconn->rconn); - n_addrs++; - } - if (ofconn->discovery) { - discovery = true; - } - } - for (i = 0; i < ofproto->n_extra_remotes; i++) { - addrs[n_addrs++] = ofproto->extra_in_band_remotes[i]; - } - - /* Create or update or destroy in-band. - * - * Ordinarily we only enable in-band if there's at least one remote - * address, but discovery needs the in-band rules for DHCP to be installed - * even before we know any remote addresses. */ - if (n_addrs || discovery) { - if (!ofproto->in_band) { - in_band_create(ofproto, ofproto->dpif, ofproto->switch_status, - &ofproto->in_band); - } - if (ofproto->in_band) { - in_band_set_remotes(ofproto->in_band, addrs, n_addrs); - } - ofproto->next_in_band_update = time_msec() + 1000; - } else { - in_band_destroy(ofproto->in_band); - ofproto->in_band = NULL; - } - - /* Clean up. */ - free(addrs); -} - -static void -update_fail_open(struct ofproto *p) -{ - struct ofconn *ofconn; - - if (!hmap_is_empty(&p->controllers) - && p->fail_mode == OFPROTO_FAIL_STANDALONE) { - struct rconn **rconns; - size_t n; - - if (!p->fail_open) { - p->fail_open = fail_open_create(p, p->switch_status); - } - - n = 0; - rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns); - HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) { - rconns[n++] = ofconn->rconn; - } - - fail_open_set_controllers(p->fail_open, rconns, n); - /* p->fail_open takes ownership of 'rconns'. */ - } else { - fail_open_destroy(p->fail_open); - p->fail_open = NULL; - } -} - void ofproto_set_controllers(struct ofproto *p, const struct ofproto_controller *controllers, size_t n_controllers) { - struct shash new_controllers; - struct ofconn *ofconn, *next_ofconn; - struct ofservice *ofservice, *next_ofservice; - bool ss_exists; - size_t i; - - /* Create newly configured controllers and services. - * Create a name to ofproto_controller mapping in 'new_controllers'. */ - shash_init(&new_controllers); - for (i = 0; i < n_controllers; i++) { - const struct ofproto_controller *c = &controllers[i]; - - if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) { - if (!find_controller_by_target(p, c->target)) { - add_controller(p, c); - } - } else if (!pvconn_verify_name(c->target)) { - if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) { - continue; - } - } else { - VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"", - dpif_name(p->dpif), c->target); - continue; - } - - shash_add_once(&new_controllers, c->target, &controllers[i]); - } - - /* Delete controllers that are no longer configured. - * Update configuration of all now-existing controllers. */ - ss_exists = false; - HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, hmap_node, - &p->controllers) { - struct ofproto_controller *c; - - c = shash_find_data(&new_controllers, ofconn_get_target(ofconn)); - if (!c) { - ofconn_destroy(ofconn); - } else { - update_controller(ofconn, c); - if (ofconn->ss) { - ss_exists = true; - } - } - } - - /* Delete services that are no longer configured. - * Update configuration of all now-existing services. */ - HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, struct ofservice, node, - &p->services) { - struct ofproto_controller *c; - - c = shash_find_data(&new_controllers, - pvconn_get_name(ofservice->pvconn)); - if (!c) { - ofservice_destroy(p, ofservice); - } else { - ofservice_reconfigure(ofservice, c); - } - } - - shash_destroy(&new_controllers); - - update_in_band_remotes(p); - update_fail_open(p); - - if (!hmap_is_empty(&p->controllers) && !ss_exists) { - ofconn = CONTAINER_OF(hmap_first(&p->controllers), - struct ofconn, hmap_node); - ofconn->ss = switch_status_register(p->switch_status, "remote", - rconn_status_cb, ofconn->rconn); - } + connmgr_set_controllers(p->connmgr, controllers, n_controllers); } void ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode) { - p->fail_mode = fail_mode; - update_fail_open(p); + connmgr_set_fail_mode(p->connmgr, fail_mode); } /* Drops the connections between 'ofproto' and all of its controllers, forcing @@ -716,34 +470,7 @@ ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode) void ofproto_reconnect_controllers(struct ofproto *ofproto) { - struct ofconn *ofconn; - - LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) { - rconn_reconnect(ofconn->rconn); - } -} - -static bool -any_extras_changed(const struct ofproto *ofproto, - const struct sockaddr_in *extras, size_t n) -{ - size_t i; - - if (n != ofproto->n_extra_remotes) { - return true; - } - - for (i = 0; i < n; i++) { - const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i]; - const struct sockaddr_in *new = &extras[i]; - - if (old->sin_addr.s_addr != new->sin_addr.s_addr || - old->sin_port != new->sin_port) { - return true; - } - } - - return false; + connmgr_reconnect(ofproto->connmgr); } /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s @@ -753,15 +480,16 @@ void ofproto_set_extra_in_band_remotes(struct ofproto *ofproto, const struct sockaddr_in *extras, size_t n) { - if (!any_extras_changed(ofproto, extras, n)) { - return; - } - - free(ofproto->extra_in_band_remotes); - ofproto->n_extra_remotes = n; - ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras); + connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n); +} - update_in_band_remotes(ofproto); +/* Sets the OpenFlow queue used by flows set up by in-band control on + * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control + * flows will use the default queue. */ +void +ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id) +{ + connmgr_set_in_band_queue(ofproto->connmgr, queue_id); } void @@ -815,55 +543,17 @@ ofproto_set_desc(struct ofproto *p, } } -static int -set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp, - const struct svec *svec) -{ - struct pvconn **pvconns = *pvconnsp; - size_t n_pvconns = *n_pvconnsp; - int retval = 0; - size_t i; - - for (i = 0; i < n_pvconns; i++) { - pvconn_close(pvconns[i]); - } - free(pvconns); - - pvconns = xmalloc(svec->n * sizeof *pvconns); - n_pvconns = 0; - for (i = 0; i < svec->n; i++) { - const char *name = svec->names[i]; - struct pvconn *pvconn; - int error; - - error = pvconn_open(name, &pvconn); - if (!error) { - pvconns[n_pvconns++] = pvconn; - } else { - VLOG_ERR("failed to listen on %s: %s", name, strerror(error)); - if (!retval) { - retval = error; - } - } - } - - *pvconnsp = pvconns; - *n_pvconnsp = n_pvconns; - - return retval; -} - int -ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops) +ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops) { - return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops); + return connmgr_set_snoops(ofproto->connmgr, snoops); } int ofproto_set_netflow(struct ofproto *ofproto, const struct netflow_options *nf_options) { - if (nf_options && nf_options->collectors.n) { + if (nf_options && !sset_is_empty(&nf_options->collectors)) { if (!ofproto->netflow) { ofproto->netflow = netflow_create(); } @@ -883,12 +573,10 @@ ofproto_set_sflow(struct ofproto *ofproto, 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, + HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) { + ofproto_sflow_add_port(os, ofport->odp_port, netdev_get_name(ofport->netdev)); } } @@ -898,7 +586,70 @@ ofproto_set_sflow(struct ofproto *ofproto, ofproto->sflow = NULL; } } + +/* Connectivity Fault Management configuration. */ + +/* Clears the CFM configuration from 'port_no' on 'ofproto'. */ +void +ofproto_iface_clear_cfm(struct ofproto *ofproto, uint32_t port_no) +{ + struct ofport *ofport = get_port(ofproto, port_no); + if (ofport && ofport->cfm){ + cfm_destroy(ofport->cfm); + ofport->cfm = NULL; + } +} + +/* Configures connectivity fault management on 'port_no' 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'. + * + * This function has no effect if 'ofproto' does not have a port 'port_no'. */ +void +ofproto_iface_set_cfm(struct ofproto *ofproto, uint32_t port_no, + const struct cfm *cfm, + const uint16_t *remote_mps, size_t n_remote_mps) +{ + struct ofport *ofport; + + ofport = get_port(ofproto, port_no); + if (!ofport) { + VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu32, + dpif_name(ofproto->dpif), port_no); + return; + } + + if (!ofport->cfm) { + ofport->cfm = cfm_create(); + } + + ofport->cfm->mpid = cfm->mpid; + ofport->cfm->interval = cfm->interval; + memcpy(ofport->cfm->maid, cfm->maid, CCM_MAID_LEN); + + cfm_update_remote_mps(ofport->cfm, remote_mps, n_remote_mps); + + if (!cfm_configure(ofport->cfm)) { + VLOG_WARN("%s: CFM configuration on port %"PRIu32" (%s) failed", + dpif_name(ofproto->dpif), port_no, + netdev_get_name(ofport->netdev)); + cfm_destroy(ofport->cfm); + ofport->cfm = NULL; + } +} +/* Returns the connectivity fault management object associated with 'port_no' + * within 'ofproto', or a null pointer if 'ofproto' does not have a port + * 'port_no' or if that port does not have CFM configured. The caller must not + * modify or destroy the returned object. */ +const struct cfm * +ofproto_iface_get_cfm(struct ofproto *ofproto, uint32_t port_no) +{ + struct ofport *ofport = get_port(ofproto, port_no); + return ofport ? ofport->cfm : NULL; +} + uint64_t ofproto_get_datapath_id(const struct ofproto *ofproto) { @@ -908,77 +659,54 @@ ofproto_get_datapath_id(const struct ofproto *ofproto) bool ofproto_has_primary_controller(const struct ofproto *ofproto) { - return !hmap_is_empty(&ofproto->controllers); + return connmgr_has_controllers(ofproto->connmgr); } enum ofproto_fail_mode ofproto_get_fail_mode(const struct ofproto *p) { - return p->fail_mode; + return connmgr_get_fail_mode(p->connmgr); } -void -ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops) +bool +ofproto_has_snoops(const struct ofproto *ofproto) { - size_t i; + return connmgr_has_snoops(ofproto->connmgr); +} - for (i = 0; i < ofproto->n_snoops; i++) { - svec_add(snoops, pvconn_get_name(ofproto->snoops[i])); - } +void +ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops) +{ + connmgr_get_snoops(ofproto->connmgr, snoops); } void ofproto_destroy(struct ofproto *p) { - struct ofservice *ofservice, *next_ofservice; - struct ofconn *ofconn, *next_ofconn; - struct ofport *ofport; - unsigned int port_no; - size_t i; + struct ofport *ofport, *next_ofport; if (!p) { return; } - /* Destroy fail-open and in-band early, since they touch the classifier. */ - fail_open_destroy(p->fail_open); - p->fail_open = NULL; - - in_band_destroy(p->in_band); - p->in_band = NULL; - free(p->extra_in_band_remotes); + shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif)); ofproto_flush_flows(p); + connmgr_destroy(p->connmgr); classifier_destroy(&p->cls); - - LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node, - &p->all_conns) { - ofconn_destroy(ofconn); - } - hmap_destroy(&p->controllers); + hmap_destroy(&p->facets); dpif_close(p->dpif); netdev_monitor_destroy(p->netdev_monitor); - PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) { + HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) { + hmap_remove(&p->ports, &ofport->hmap_node); ofport_free(ofport); } shash_destroy(&p->port_by_name); - switch_status_destroy(p->switch_status); netflow_destroy(p->netflow); ofproto_sflow_destroy(p->sflow); - HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, struct ofservice, node, - &p->services) { - ofservice_destroy(p, ofservice); - } - hmap_destroy(&p->services); - - for (i = 0; i < p->n_snoops; i++) { - pvconn_close(p->snoops[i]); - } - free(p->snoops); - mac_learning_destroy(p->ml); free(p->mfr_desc); @@ -987,7 +715,7 @@ ofproto_destroy(struct ofproto *p) free(p->serial_desc); free(p->dp_desc); - port_array_destroy(&p->ports); + hmap_destroy(&p->ports); free(p); } @@ -1013,54 +741,10 @@ process_port_change(struct ofproto *ofproto, int error, char *devname) } } -/* Returns a "preference level" for snooping 'ofconn'. A higher return value - * means that 'ofconn' is more interesting for monitoring than a lower return - * value. */ -static int -snoop_preference(const struct ofconn *ofconn) -{ - switch (ofconn->role) { - case NX_ROLE_MASTER: - return 3; - case NX_ROLE_OTHER: - return 2; - case NX_ROLE_SLAVE: - return 1; - default: - /* Shouldn't happen. */ - return 0; - } -} - -/* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'. - * Connects this vconn to a controller. */ -static void -add_snooper(struct ofproto *ofproto, struct vconn *vconn) -{ - struct ofconn *ofconn, *best; - - /* Pick a controller for monitoring. */ - best = NULL; - LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) { - if (ofconn->type == OFCONN_PRIMARY - && (!best || snoop_preference(ofconn) > snoop_preference(best))) { - best = ofconn; - } - } - - if (best) { - rconn_add_monitor(best->rconn, vconn); - } else { - VLOG_INFO_RL(&rl, "no controller connection to snoop"); - vconn_close(vconn); - } -} - int ofproto_run1(struct ofproto *p) { - struct ofconn *ofconn, *next_ofconn; - struct ofservice *ofservice; + struct ofport *ofport; char *devname; int error; int i; @@ -1070,10 +754,9 @@ ofproto_run1(struct ofproto *p) } for (i = 0; i < 50; i++) { - struct ofpbuf *buf; - int error; + struct dpif_upcall packet; - error = dpif_recv(p->dpif, &buf); + error = dpif_recv(p->dpif, &packet); if (error) { if (error == ENODEV) { /* Someone destroyed the datapath behind our back. The caller @@ -1087,7 +770,7 @@ ofproto_run1(struct ofproto *p) break; } - handle_odp_msg(p, buf); + handle_upcall(p, &packet); } while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) { @@ -1098,73 +781,16 @@ ofproto_run1(struct ofproto *p) process_port_change(p, error, devname); } - if (p->in_band) { - if (time_msec() >= p->next_in_band_update) { - update_in_band_remotes(p); - } - in_band_run(p->in_band); - } - - 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); - } - - HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) { - struct vconn *vconn; - int retval; - - retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn); - if (!retval) { - struct ofconn *ofconn; - struct rconn *rconn; - char *name; - - rconn = rconn_create(ofservice->probe_interval, 0); - name = ofconn_make_name(p, vconn_get_name(vconn)); - rconn_connect_unreliably(rconn, vconn, name); - free(name); - - ofconn = ofconn_create(p, rconn, OFCONN_SERVICE); - ofconn_set_rate_limit(ofconn, ofservice->rate_limit, - ofservice->burst_limit); - } else if (retval != EAGAIN) { - VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); - } + HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { + ofport_run(p, ofport); } - for (i = 0; i < p->n_snoops; i++) { - struct vconn *vconn; - int retval; - - retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn); - if (!retval) { - add_snooper(p, vconn); - } else if (retval != EAGAIN) { - VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); - } - } + connmgr_run(p->connmgr, handle_openflow); - if (time_msec() >= p->next_expiration) { + if (timer_expired(&p->next_expiration)) { + int delay = ofproto_expire(p); + timer_set_duration(&p->next_expiration, delay); COVERAGE_INC(ofproto_expiration); - p->next_expiration = time_msec() + 1000; - update_used(p); - - classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p); - - /* Let the hook know that we're at a stable point: all outstanding data - * in existing flows has been accounted to the account_cb. Thus, the - * hook can now reasonably do operations that depend on having accurate - * flow volume accounting (currently, that's just bond rebalancing). */ - if (p->ofhooks->account_checkpoint_cb) { - p->ofhooks->account_checkpoint_cb(p->aux); - } } if (p->netflow) { @@ -1177,27 +803,29 @@ ofproto_run1(struct ofproto *p) return 0; } -struct revalidate_cbdata { - struct ofproto *ofproto; - bool revalidate_all; /* Revalidate all exact-match rules? */ - bool revalidate_subrules; /* Revalidate all exact-match subrules? */ - struct tag_set revalidate_set; /* Set of tags to revalidate. */ -}; - int ofproto_run2(struct ofproto *p, bool revalidate_all) { - if (p->need_revalidate || revalidate_all - || !tag_set_is_empty(&p->revalidate_set)) { - struct revalidate_cbdata cbdata; - cbdata.ofproto = p; - cbdata.revalidate_all = revalidate_all; - cbdata.revalidate_subrules = p->need_revalidate; - cbdata.revalidate_set = p->revalidate_set; - tag_set_init(&p->revalidate_set); - COVERAGE_INC(ofproto_revalidate); - classifier_for_each(&p->cls, CLS_INC_EXACT, revalidate_cb, &cbdata); - p->need_revalidate = false; + /* Figure out what we need to revalidate now, if anything. */ + struct tag_set revalidate_set = p->revalidate_set; + if (p->need_revalidate) { + revalidate_all = true; + } + + /* Clear the revalidation flags. */ + tag_set_init(&p->revalidate_set); + p->need_revalidate = false; + + /* Now revalidate if there's anything to do. */ + if (revalidate_all || !tag_set_is_empty(&revalidate_set)) { + struct facet *facet, *next; + + HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) { + if (revalidate_all + || tag_set_intersects(&revalidate_set, facet->tags)) { + facet_revalidate(p, facet); + } + } } return 0; @@ -1206,23 +834,14 @@ ofproto_run2(struct ofproto *p, bool revalidate_all) void ofproto_wait(struct ofproto *p) { - struct ofservice *ofservice; - struct ofconn *ofconn; - size_t i; + struct ofport *ofport; + HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { + ofport_wait(ofport); + } dpif_recv_wait(p->dpif); dpif_port_poll_wait(p->dpif); netdev_monitor_poll_wait(p->netdev_monitor); - LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { - ofconn_wait(ofconn); - } - if (p->in_band) { - poll_timer_wait_until(p->next_in_band_update); - in_band_wait(p->in_band); - } - if (p->fail_open) { - fail_open_wait(p->fail_open); - } if (p->sflow) { ofproto_sflow_wait(p->sflow); } @@ -1233,15 +852,10 @@ ofproto_wait(struct ofproto *p) /* Shouldn't happen, but if it does just go around again. */ VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); poll_immediate_wake(); - } else if (p->next_expiration != LLONG_MAX) { - poll_timer_wait_until(p->next_expiration); - } - HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) { - pvconn_wait(ofservice->pvconn); - } - for (i = 0; i < p->n_snoops; i++) { - pvconn_wait(p->snoops[i]); + } else { + timer_wait(&p->next_expiration); } + connmgr_wait(p->connmgr); } void @@ -1259,183 +873,221 @@ ofproto_get_revalidate_set(struct ofproto *ofproto) bool ofproto_is_alive(const struct ofproto *p) { - return !hmap_is_empty(&p->controllers); + return connmgr_has_controllers(p->connmgr); +} + +void +ofproto_get_ofproto_controller_info(const struct ofproto *ofproto, + struct shash *info) +{ + connmgr_get_controller_info(ofproto->connmgr, info); +} + +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); +} + +/* Deletes port number 'odp_port' from the datapath for 'ofproto'. + * + * This is almost the same as calling dpif_port_del() directly on the + * datapath, but it also makes 'ofproto' close its open netdev for the port + * (if any). This makes it possible to create a new netdev of a different + * type under the same name, which otherwise the netdev library would refuse + * to do because of the conflict. (The netdev would eventually get closed on + * the next trip through ofproto_run(), but this interface is more direct.) + * + * Returns 0 if successful, otherwise a positive errno. */ +int +ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port) +{ + struct ofport *ofport = get_port(ofproto, odp_port); + const char *name = ofport ? ofport->opp.name : ""; + int error; + + error = dpif_port_del(ofproto->dpif, odp_port); + if (error) { + VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)", + dpif_name(ofproto->dpif), odp_port, name, strerror(error)); + } else if (ofport) { + /* 'name' is ofport->opp.name and update_port() is going to destroy + * 'ofport'. Just in case update_port() refers to 'name' after it + * destroys 'ofport', make a copy of it around the update_port() + * call. */ + char *devname = xstrdup(name); + update_port(ofproto, devname); + free(devname); + } + return error; +} + +/* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns + * true if 'odp_port' exists and should be included, false otherwise. */ +bool +ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port) +{ + struct ofport *ofport = get_port(ofproto, odp_port); + return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD); } +/* Sends 'packet' out of port 'port_no' within 'p'. If 'vlan_tci' is zero the + * packet will not have any 802.1Q hader; if it is nonzero, then the packet + * will be sent with the VLAN TCI specified by 'vlan_tci & ~VLAN_CFI'. + * + * Returns 0 if successful, otherwise a positive errno value. */ int -ofproto_send_packet(struct ofproto *p, const flow_t *flow, - const union ofp_action *actions, size_t n_actions, +ofproto_send_packet(struct ofproto *ofproto, + uint32_t port_no, uint16_t vlan_tci, const struct ofpbuf *packet) { - struct odp_actions odp_actions; + struct ofpbuf odp_actions; int error; - error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions, - NULL, NULL, NULL); - if (error) { - return error; + ofpbuf_init(&odp_actions, 32); + if (vlan_tci != 0) { + nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_SET_DL_TCI, + ntohs(vlan_tci & ~VLAN_CFI)); } + nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, port_no); + error = dpif_execute(ofproto->dpif, odp_actions.data, odp_actions.size, + packet); + ofpbuf_uninit(&odp_actions); - /* XXX Should we translate the dpif_execute() errno value into an OpenFlow - * error code? */ - dpif_execute(p->dpif, flow->in_port, odp_actions.actions, - odp_actions.n_actions, packet); - return 0; + if (error) { + VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)", + dpif_name(ofproto->dpif), port_no, strerror(error)); + } + return error; } +/* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and + * performs the 'n_actions' actions in 'actions'. The new flow will not + * timeout. + * + * If cls_rule->priority is in the range of priorities supported by OpenFlow + * (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'. */ void -ofproto_add_flow(struct ofproto *p, - const flow_t *flow, uint32_t wildcards, unsigned int priority, - const union ofp_action *actions, size_t n_actions, - int idle_timeout) +ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule, + const union ofp_action *actions, size_t n_actions) { struct rule *rule; - rule = rule_create(p, NULL, actions, n_actions, - idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, - 0, 0, false); - cls_rule_from_flow(flow, wildcards, priority, &rule->cr); - rule_insert(p, rule, NULL, 0); + rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false); + rule_insert(p, rule); } void -ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow, - uint32_t wildcards, unsigned int priority) +ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target) { struct rule *rule; rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls, - flow, wildcards, - priority)); + target)); if (rule) { rule_remove(ofproto, rule); } } -static void -destroy_rule(struct cls_rule *rule_, void *ofproto_) -{ - struct rule *rule = rule_from_cls_rule(rule_); - struct ofproto *ofproto = ofproto_; - - /* Mark the flow as not installed, even though it might really be - * installed, so that rule_remove() doesn't bother trying to uninstall it. - * There is no point in uninstalling it individually since we are about to - * blow away all the flows with dpif_flow_flush(). */ - rule->installed = false; - - rule_remove(ofproto, rule); -} - void ofproto_flush_flows(struct ofproto *ofproto) { + struct facet *facet, *next_facet; + struct rule *rule, *next_rule; + struct cls_cursor cursor; + COVERAGE_INC(ofproto_flush); - classifier_for_each(&ofproto->cls, CLS_INC_ALL, destroy_rule, ofproto); - dpif_flow_flush(ofproto->dpif); - if (ofproto->in_band) { - in_band_flushed(ofproto->in_band); + + HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { + /* Mark the facet as not installed so that facet_remove() doesn't + * bother trying to uninstall it. There is no point in uninstalling it + * individually since we are about to blow away all the facets with + * dpif_flow_flush(). */ + facet->installed = false; + facet->dp_packet_count = 0; + facet->dp_byte_count = 0; + facet_remove(ofproto, facet); } - if (ofproto->fail_open) { - fail_open_flushed(ofproto->fail_open); + + cls_cursor_init(&cursor, &ofproto->cls, NULL); + CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { + rule_remove(ofproto, rule); } + + dpif_flow_flush(ofproto->dpif); + connmgr_flushed(ofproto->connmgr); } static void reinit_ports(struct ofproto *p) { - struct svec devnames; + struct dpif_port_dump dump; + struct sset devnames; struct ofport *ofport; - unsigned int port_no; - struct odp_port *odp_ports; - size_t n_odp_ports; - size_t i; + struct dpif_port dpif_port; + const char *devname; - svec_init(&devnames); - PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) { - svec_add (&devnames, (char *) ofport->opp.name); - } - dpif_port_list(p->dpif, &odp_ports, &n_odp_ports); - for (i = 0; i < n_odp_ports; i++) { - svec_add (&devnames, odp_ports[i].devname); - } - free(odp_ports); + COVERAGE_INC(ofproto_reinit_ports); - svec_sort_unique(&devnames); - for (i = 0; i < devnames.n; i++) { - update_port(p, devnames.names[i]); + sset_init(&devnames); + HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { + sset_add(&devnames, ofport->opp.name); } - svec_destroy(&devnames); -} - -static size_t -refresh_port_group(struct ofproto *p, unsigned int group) -{ - uint16_t *ports; - size_t n_ports; - struct ofport *port; - unsigned int port_no; - - assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD); - - ports = xmalloc(port_array_count(&p->ports) * sizeof *ports); - n_ports = 0; - PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { - if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) { - ports[n_ports++] = port_no; - } + DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) { + sset_add(&devnames, dpif_port.name); } - dpif_port_group_set(p->dpif, group, ports, n_ports); - free(ports); - - return n_ports; -} -static void -refresh_port_groups(struct ofproto *p) -{ - 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); + SSET_FOR_EACH (devname, &devnames) { + update_port(p, devname); } + sset_destroy(&devnames); } static struct ofport * -make_ofport(const struct odp_port *odp_port) +make_ofport(const struct dpif_port *dpif_port) { struct netdev_options netdev_options; enum netdev_flags flags; struct ofport *ofport; struct netdev *netdev; - bool carrier; int error; memset(&netdev_options, 0, sizeof netdev_options); - netdev_options.name = odp_port->devname; + netdev_options.name = dpif_port->name; + netdev_options.type = dpif_port->type; netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; error = netdev_open(&netdev_options, &netdev); if (error) { VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s " "cannot be opened (%s)", - odp_port->devname, odp_port->port, - odp_port->devname, strerror(error)); + dpif_port->name, dpif_port->port_no, + dpif_port->name, strerror(error)); return NULL; } - ofport = xmalloc(sizeof *ofport); + ofport = xzalloc(sizeof *ofport); ofport->netdev = netdev; - ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port); + ofport->odp_port = dpif_port->port_no; + ofport->opp.port_no = odp_port_to_ofp_port(dpif_port->port_no); netdev_get_etheraddr(netdev, ofport->opp.hw_addr); - memcpy(ofport->opp.name, odp_port->devname, - MIN(sizeof ofport->opp.name, sizeof odp_port->devname)); - ofport->opp.name[sizeof ofport->opp.name - 1] = '\0'; + ovs_strlcpy(ofport->opp.name, dpif_port->name, sizeof ofport->opp.name); netdev_get_flags(netdev, &flags); ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN; - netdev_get_carrier(netdev, &carrier); - ofport->opp.state = carrier ? 0 : OFPPS_LINK_DOWN; + ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN; netdev_get_features(netdev, &ofport->opp.curr, &ofport->opp.advertised, @@ -1444,15 +1096,15 @@ make_ofport(const struct odp_port *odp_port) } static bool -ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port) +ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port) { - if (port_array_get(&p->ports, odp_port->port)) { + if (get_port(p, dpif_port->port_no)) { VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath", - odp_port->port); + dpif_port->port_no); return true; - } else if (shash_find(&p->port_by_name, odp_port->devname)) { + } else if (shash_find(&p->port_by_name, dpif_port->name)) { VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath", - odp_port->devname); + dpif_port->name); return true; } else { return false; @@ -1468,7 +1120,7 @@ ofport_equal(const struct ofport *a_, const struct ofport *b_) BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */ return (a->port_no == b->port_no && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr) - && !strcmp((char *) a->name, (char *) b->name) + && !strcmp(a->name, b->name) && a->state == b->state && a->config == b->config && a->curr == b->curr @@ -1477,56 +1129,56 @@ ofport_equal(const struct ofport *a_, const struct ofport *b_) && a->peer == b->peer); } -static void -send_port_status(struct ofproto *p, const struct ofport *ofport, - uint8_t reason) -{ - /* XXX Should limit the number of queued port status change messages. */ - struct ofconn *ofconn; - LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { - struct ofp_port_status *ops; - struct ofpbuf *b; - - if (!ofconn_receives_async_msgs(ofconn)) { - continue; - } - - ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b); - ops->reason = reason; - ops->desc = ofport->opp; - hton_ofp_phy_port(&ops->desc); - queue_tx(b, ofconn, NULL); - } - if (p->ofhooks->port_changed_cb) { - p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux); - } -} - 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; + const char *netdev_name = ofport->opp.name; netdev_monitor_add(p->netdev_monitor, ofport->netdev); - port_array_set(&p->ports, odp_port, ofport); + hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0)); shash_add(&p->port_by_name, netdev_name, ofport); if (p->sflow) { - ofproto_sflow_add_port(p->sflow, odp_port, netdev_name); + ofproto_sflow_add_port(p->sflow, ofport->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_delete(&p->ports, odp_port); + hmap_remove(&p->ports, &ofport->hmap_node); shash_delete(&p->port_by_name, - shash_find(&p->port_by_name, (char *) ofport->opp.name)); + shash_find(&p->port_by_name, ofport->opp.name)); if (p->sflow) { - ofproto_sflow_del_port(p->sflow, odp_port); + ofproto_sflow_del_port(p->sflow, ofport->odp_port); + } +} + +static void +ofport_run(struct ofproto *ofproto, struct ofport *ofport) +{ + if (ofport->cfm) { + cfm_run(ofport->cfm); + + if (cfm_should_send_ccm(ofport->cfm)) { + struct ofpbuf packet; + struct ccm *ccm; + + ofpbuf_init(&packet, 0); + ccm = eth_compose(&packet, eth_addr_ccm, ofport->opp.hw_addr, + ETH_TYPE_CFM, sizeof *ccm); + cfm_compose_ccm(ofport->cfm, ccm); + ofproto_send_packet(ofproto, ofport->odp_port, 0, &packet); + ofpbuf_uninit(&packet); + } + } +} + +static void +ofport_wait(struct ofport *ofport) +{ + if (ofport->cfm) { + cfm_wait(ofport->cfm); } } @@ -1534,15 +1186,30 @@ static void ofport_free(struct ofport *ofport) { if (ofport) { + cfm_destroy(ofport->cfm); netdev_close(ofport->netdev); free(ofport); } } +static struct ofport * +get_port(const struct ofproto *ofproto, uint16_t odp_port) +{ + struct ofport *port; + + HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, + hash_int(odp_port, 0), &ofproto->ports) { + if (port->odp_port == odp_port) { + return port; + } + } + return NULL; +} + static void update_port(struct ofproto *p, const char *devname) { - struct odp_port odp_port; + struct dpif_port dpif_port; struct ofport *old_ofport; struct ofport *new_ofport; int error; @@ -1550,7 +1217,7 @@ update_port(struct ofproto *p, const char *devname) COVERAGE_INC(ofproto_update_port); /* Query the datapath for port information. */ - error = dpif_port_query_by_name(p->dpif, devname, &odp_port); + error = dpif_port_query_by_name(p->dpif, devname, &dpif_port); /* Find the old ofport. */ old_ofport = shash_find_data(&p->port_by_name, devname); @@ -1566,31 +1233,31 @@ update_port(struct ofproto *p, const char *devname) * reliably but more portably by comparing the old port's MAC * against the new port's MAC. However, this code isn't that smart * and always sends an OFPPR_MODIFY (XXX). */ - old_ofport = port_array_get(&p->ports, odp_port.port); + old_ofport = get_port(p, dpif_port.port_no); } } else if (error != ENOENT && error != ENODEV) { VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error " "%s", strerror(error)); - return; + goto exit; } /* Create a new ofport. */ - new_ofport = !error ? make_ofport(&odp_port) : NULL; + new_ofport = !error ? make_ofport(&dpif_port) : NULL; /* Eliminate a few pathological cases. */ if (!old_ofport && !new_ofport) { - return; + goto exit; } else if (old_ofport && new_ofport) { /* Most of the 'config' bits are OpenFlow soft state, but - * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow - * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and - * leaves the other bits 0.) */ + * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the + * OpenFlow bits from old_ofport. (make_ofport() only sets + * OFPPC_PORT_DOWN and leaves the other bits 0.) */ new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN; if (ofport_equal(old_ofport, new_ofport)) { /* False alarm--no change. */ ofport_free(new_ofport); - return; + goto exit; } } @@ -1601,274 +1268,68 @@ update_port(struct ofproto *p, const char *devname) if (new_ofport) { ofport_install(p, new_ofport); } - send_port_status(p, new_ofport ? new_ofport : old_ofport, - (!old_ofport ? OFPPR_ADD - : !new_ofport ? OFPPR_DELETE - : OFPPR_MODIFY)); + connmgr_send_port_status(p->connmgr, + new_ofport ? &new_ofport->opp : &old_ofport->opp, + (!old_ofport ? OFPPR_ADD + : !new_ofport ? OFPPR_DELETE + : OFPPR_MODIFY)); ofport_free(old_ofport); - /* Update port groups. */ - refresh_port_groups(p); +exit: + dpif_port_destroy(&dpif_port); } static int init_ports(struct ofproto *p) { - struct odp_port *ports; - size_t n_ports; - size_t i; - int error; - - error = dpif_port_list(p->dpif, &ports, &n_ports); - if (error) { - return error; - } + struct dpif_port_dump dump; + struct dpif_port dpif_port; - for (i = 0; i < n_ports; i++) { - const struct odp_port *odp_port = &ports[i]; - if (!ofport_conflicts(p, odp_port)) { - struct ofport *ofport = make_ofport(odp_port); + DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) { + if (!ofport_conflicts(p, &dpif_port)) { + struct ofport *ofport = make_ofport(&dpif_port); if (ofport) { ofport_install(p, ofport); } } } - free(ports); - refresh_port_groups(p); + return 0; } -static struct ofconn * -ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type) -{ - struct ofconn *ofconn = xzalloc(sizeof *ofconn); - ofconn->ofproto = p; - list_push_back(&p->all_conns, &ofconn->node); - ofconn->rconn = rconn; - ofconn->type = type; - ofconn->role = NX_ROLE_OTHER; - ofconn->packet_in_counter = rconn_packet_counter_create (); - ofconn->pktbuf = NULL; - ofconn->miss_send_len = 0; - ofconn->reply_counter = rconn_packet_counter_create (); - return ofconn; -} - -static void -ofconn_destroy(struct ofconn *ofconn) -{ - if (ofconn->type == OFCONN_PRIMARY) { - hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node); - } - discovery_destroy(ofconn->discovery); - - list_remove(&ofconn->node); - switch_status_unregister(ofconn->ss); - rconn_destroy(ofconn->rconn); - rconn_packet_counter_destroy(ofconn->packet_in_counter); - rconn_packet_counter_destroy(ofconn->reply_counter); - pktbuf_destroy(ofconn->pktbuf); - free(ofconn); -} - -static void -ofconn_run(struct ofconn *ofconn, struct ofproto *p) -{ - int iteration; - size_t i; - - if (ofconn->discovery) { - char *controller_name; - if (rconn_is_connectivity_questionable(ofconn->rconn)) { - discovery_question_connectivity(ofconn->discovery); - } - if (discovery_run(ofconn->discovery, &controller_name)) { - if (controller_name) { - char *ofconn_name = ofconn_make_name(p, controller_name); - rconn_connect(ofconn->rconn, controller_name, ofconn_name); - free(ofconn_name); - } else { - rconn_disconnect(ofconn->rconn); - } - } - } - - for (i = 0; i < N_SCHEDULERS; i++) { - pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn); - } - - rconn_run(ofconn->rconn); - - if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { - /* Limit the number of iterations to prevent other tasks from - * starving. */ - for (iteration = 0; iteration < 50; iteration++) { - struct ofpbuf *of_msg = rconn_recv(ofconn->rconn); - 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->discovery && !rconn_is_alive(ofconn->rconn)) { - ofconn_destroy(ofconn); - } -} - -static void -ofconn_wait(struct ofconn *ofconn) -{ - int i; - - if (ofconn->discovery) { - discovery_wait(ofconn->discovery); - } - for (i = 0; i < N_SCHEDULERS; i++) { - pinsched_wait(ofconn->schedulers[i]); - } - rconn_run_wait(ofconn->rconn); - if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { - rconn_recv_wait(ofconn->rconn); - } else { - COVERAGE_INC(ofproto_ofconn_stuck); - } -} - -/* Returns true if 'ofconn' should receive asynchronous messages. */ +/* Returns true if 'rule' should be hidden from the controller. + * + * Rules with priority higher than UINT16_MAX are set up by ofproto itself + * (e.g. by in-band control) and are intentionally hidden from the + * controller. */ static bool -ofconn_receives_async_msgs(const struct ofconn *ofconn) +rule_is_hidden(const struct rule *rule) { - if (ofconn->type == OFCONN_PRIMARY) { - /* Primary controllers always get asynchronous messages unless they - * have configured themselves as "slaves". */ - return ofconn->role != NX_ROLE_SLAVE; - } else { - /* Service connections don't get asynchronous messages unless they have - * explicitly asked for them by setting a nonzero miss send length. */ - return ofconn->miss_send_len > 0; - } + return rule->cr.priority > UINT16_MAX; } -/* Returns a human-readable name for an OpenFlow connection between 'ofproto' - * and 'target', suitable for use in log messages for identifying the - * connection. +/* Creates and returns a new rule initialized as specified. * - * The name is dynamically allocated. The caller should free it (with free()) - * when it is no longer needed. */ -static char * -ofconn_make_name(const struct ofproto *ofproto, const char *target) -{ - return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target); -} - -static void -ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst) -{ - int i; - - for (i = 0; i < N_SCHEDULERS; i++) { - struct pinsched **s = &ofconn->schedulers[i]; - - if (rate > 0) { - if (!*s) { - *s = pinsched_create(rate, burst, - ofconn->ofproto->switch_status); - } else { - pinsched_set_limits(*s, rate, burst); - } - } else { - pinsched_destroy(*s); - *s = NULL; - } - } -} - -static void -ofservice_reconfigure(struct ofservice *ofservice, - const struct ofproto_controller *c) -{ - ofservice->probe_interval = c->probe_interval; - ofservice->rate_limit = c->rate_limit; - ofservice->burst_limit = c->burst_limit; -} - -/* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a - * positive errno value. */ -static int -ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c) -{ - struct ofservice *ofservice; - struct pvconn *pvconn; - int error; - - error = pvconn_open(c->target, &pvconn); - if (error) { - return error; - } - - ofservice = xzalloc(sizeof *ofservice); - hmap_insert(&ofproto->services, &ofservice->node, - hash_string(c->target, 0)); - ofservice->pvconn = pvconn; - - ofservice_reconfigure(ofservice, c); - - return 0; -} - -static void -ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice) -{ - hmap_remove(&ofproto->services, &ofservice->node); - pvconn_close(ofservice->pvconn); - free(ofservice); -} - -/* Finds and returns the ofservice within 'ofproto' that has the given - * 'target', or a null pointer if none exists. */ -static struct ofservice * -ofservice_lookup(struct ofproto *ofproto, const char *target) -{ - struct ofservice *ofservice; - - HMAP_FOR_EACH_WITH_HASH (ofservice, struct ofservice, node, - hash_string(target, 0), &ofproto->services) { - if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) { - return ofservice; - } - } - return NULL; -} - -/* Caller is responsible for initializing the 'cr' member of the returned - * rule. */ + * The caller is responsible for inserting the rule into the classifier (with + * rule_insert()). */ static struct rule * -rule_create(struct ofproto *ofproto, struct rule *super, +rule_create(const struct cls_rule *cls_rule, const union ofp_action *actions, size_t n_actions, uint16_t idle_timeout, uint16_t hard_timeout, - uint64_t flow_cookie, bool send_flow_removed) + ovs_be64 flow_cookie, bool send_flow_removed) { struct rule *rule = xzalloc(sizeof *rule); + rule->cr = *cls_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); - } else { - list_init(&rule->list); + list_init(&rule->facets); + if (n_actions > 0) { + rule->n_actions = n_actions; + rule->actions = xmemdup(actions, n_actions * sizeof *actions); } - 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; } @@ -1883,34 +1344,28 @@ static void rule_free(struct rule *rule) { free(rule->actions); - free(rule->odp_actions); free(rule); } -/* Destroys 'rule'. If 'rule' is a subrule, also removes it from its - * super-rule's list of subrules. If 'rule' is a super-rule, also iterates - * through all of its subrules and revalidates them, destroying any that no - * longer has a super-rule (which is probably all of them). +/* 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). * - * Before calling this function, the caller must make have removed 'rule' from - * the classifier. If 'rule' is an exact-match rule, the caller is also - * responsible for ensuring that it has been uninstalled from the datapath. */ + * The caller must have already removed 'rule' from the classifier. */ static void rule_destroy(struct ofproto *ofproto, struct rule *rule) { - if (!rule->super) { - struct rule *subrule, *next; - LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) { - revalidate_rule(ofproto, subrule); - } - } else { - list_remove(&rule->list); + struct facet *facet, *next_facet; + LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { + facet_revalidate(ofproto, facet); } rule_free(rule); } +/* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action + * 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, uint16_t out_port) +rule_has_out_port(const struct rule *rule, ovs_be16 out_port) { const union ofp_action *oa; struct actions_iterator i; @@ -1932,420 +1387,501 @@ rule_has_out_port(const struct rule *rule, uint16_t out_port) * * Takes ownership of 'packet'. */ static bool -execute_odp_actions(struct ofproto *ofproto, uint16_t in_port, - const union odp_action *actions, size_t n_actions, +execute_odp_actions(struct ofproto *ofproto, const struct flow *flow, + const struct nlattr *odp_actions, size_t actions_len, struct ofpbuf *packet) { - if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) { + if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t)) + && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) { /* As an optimization, avoid a round-trip from userspace to kernel to * userspace. This also avoids possibly filling up kernel packet * buffers along the way. */ - struct odp_msg *msg; + struct dpif_upcall upcall; - msg = ofpbuf_push_uninit(packet, sizeof *msg); - msg->type = _ODPL_ACTION_NR; - msg->length = sizeof(struct odp_msg) + packet->size; - msg->port = in_port; - msg->reserved = 0; - msg->arg = actions[0].controller.arg; + upcall.type = DPIF_UC_ACTION; + upcall.packet = packet; + upcall.key = NULL; + upcall.key_len = 0; + upcall.userdata = nl_attr_get_u64(odp_actions); + upcall.sample_pool = 0; + upcall.actions = NULL; + upcall.actions_len = 0; - send_packet_in(ofproto, packet); + send_packet_in(ofproto, &upcall, flow, false); return true; } else { int error; - error = dpif_execute(ofproto->dpif, in_port, - actions, n_actions, packet); + error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet); ofpbuf_delete(packet); return !error; } } -/* Executes the actions indicated by 'rule' on 'packet', which is in flow - * 'flow' and is considered to have arrived on ODP port 'in_port'. 'packet' - * must have at least sizeof(struct ofp_packet_in) bytes of headroom. +/* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s + * statistics appropriately. 'packet' must have at least sizeof(struct + * ofp_packet_in) bytes of headroom. * - * The flow that 'packet' actually contains does not need to actually match - * 'rule'; the actions in 'rule' will be applied to it either way. Likewise, - * the packet and byte counters for 'rule' will be credited for the packet sent - * out whether or not the packet actually matches 'rule'. + * For correct results, 'packet' must actually be in 'facet''s flow; that is, + * applying flow_extract() to 'packet' would yield the same flow as + * 'facet->flow'. * - * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow, - * the caller must already have accurately composed ODP actions for it given - * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if - * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this - * function will compose a set of ODP actions based on 'rule''s OpenFlow - * actions and apply them to 'packet'. + * 'facet' must have accurately composed ODP actions; that is, it must not be + * in need of revalidation. * * Takes ownership of 'packet'. */ static void -rule_execute(struct ofproto *ofproto, struct rule *rule, - struct ofpbuf *packet, const flow_t *flow) +facet_execute(struct ofproto *ofproto, struct facet *facet, + struct ofpbuf *packet) { - const union odp_action *actions; - struct odp_flow_stats stats; - size_t n_actions; - struct odp_actions a; + struct dpif_flow_stats stats; assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); - /* Grab or compose the ODP actions. - * - * The special case for an exact-match 'rule' where 'flow' is not the - * rule's flow is important to avoid, e.g., sending a packet out its input - * port simply because the ODP actions were composed for the wrong - * scenario. */ - 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, NULL)) { - ofpbuf_delete(packet); - return; - } - actions = a.actions; - n_actions = a.n_actions; - } else { - actions = rule->odp_actions; - n_actions = rule->n_odp_actions; + flow_extract_stats(&facet->flow, packet, &stats); + stats.used = time_msec(); + if (execute_odp_actions(ofproto, &facet->flow, + facet->actions, facet->actions_len, packet)) { + facet_update_stats(ofproto, facet, &stats); } +} - /* Execute the ODP actions. */ - flow_extract_stats(flow, packet, &stats); - if (execute_odp_actions(ofproto, flow->in_port, - actions, n_actions, packet)) { - update_stats(ofproto, rule, &stats); - rule->used = time_msec(); - netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used); - } -} - -/* Inserts 'rule' into 'p''s flow table. +/* 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. * - * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s - * actions on it and credits the statistics for sending the packet to 'rule'. - * '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. + * + * Takes ownership of 'packet'. */ static void -rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet, - uint16_t in_port) +rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port, + struct ofpbuf *packet) { - struct rule *displaced_rule; + struct action_xlate_ctx ctx; + struct ofpbuf *odp_actions; + struct facet *facet; + struct flow flow; + size_t size; - /* Insert the rule in the classifier. */ - displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr)); - if (!rule->cr.wc.wildcards) { - rule_make_actions(p, rule, packet); + assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); + + flow_extract(packet, 0, in_port, &flow); + + /* First look for a related facet. If we find one, account it to that. */ + facet = facet_lookup_valid(ofproto, &flow); + if (facet && facet->rule == rule) { + facet_execute(ofproto, facet, packet); + return; } - /* Send the packet and credit it to the rule. */ - if (packet) { - flow_t flow; - flow_extract(packet, 0, in_port, &flow); - rule_execute(p, rule, packet, &flow); + /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then + * create a new facet for it and use that. */ + if (rule_lookup(ofproto, &flow) == rule) { + facet = facet_create(ofproto, rule, &flow, packet); + facet_execute(ofproto, facet, packet); + facet_install(ofproto, facet, true); + return; } - /* Install the rule in the datapath only after sending the packet, to - * avoid packet reordering. */ - if (rule->cr.wc.wildcards) { - COVERAGE_INC(ofproto_add_wc_flow); - p->need_revalidate = true; - } else { - rule_install(p, rule, displaced_rule); + /* We can't account anything to a facet. If we were to try, then that + * facet would have a non-matching rule, busting our invariants. */ + action_xlate_ctx_init(&ctx, ofproto, &flow, packet); + odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions); + size = packet->size; + if (execute_odp_actions(ofproto, &flow, odp_actions->data, + odp_actions->size, packet)) { + rule->used = time_msec(); + rule->packet_count++; + rule->byte_count += size; + flow_push_stats(ofproto, rule, &flow, 1, size, rule->used); } + ofpbuf_delete(odp_actions); +} + +/* Inserts 'rule' into 'p''s flow table. */ +static void +rule_insert(struct ofproto *p, struct rule *rule) +{ + struct rule *displaced_rule; - /* Free the rule that was displaced, if any. */ + displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr)); if (displaced_rule) { rule_destroy(p, displaced_rule); } + p->need_revalidate = true; } -static struct rule * -rule_create_subrule(struct ofproto *ofproto, struct rule *rule, - const flow_t *flow) +/* Creates and returns a new facet within 'ofproto' owned by 'rule', given a + * 'flow' and an example 'packet' within that flow. + * + * The caller must already have determined that no facet with an identical + * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in + * 'ofproto''s classifier table. */ +static struct facet * +facet_create(struct ofproto *ofproto, struct rule *rule, + const struct flow *flow, const struct ofpbuf *packet) { - 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(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX - : rule->cr.priority), &subrule->cr); - classifier_insert_exact(&ofproto->cls, &subrule->cr); + struct facet *facet; + + facet = xzalloc(sizeof *facet); + facet->used = time_msec(); + hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0)); + list_push_back(&rule->facets, &facet->list_node); + facet->rule = rule; + facet->flow = *flow; + netflow_flow_init(&facet->nf_flow); + netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); - return subrule; + facet_make_actions(ofproto, facet, packet); + + return facet; +} + +static void +facet_free(struct facet *facet) +{ + free(facet->actions); + free(facet); } +/* 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(). + */ static void rule_remove(struct ofproto *ofproto, struct rule *rule) { - if (rule->cr.wc.wildcards) { - COVERAGE_INC(ofproto_del_wc_flow); - ofproto->need_revalidate = true; - } else { - rule_uninstall(ofproto, rule); - } + COVERAGE_INC(ofproto_del_rule); + ofproto->need_revalidate = true; classifier_remove(&ofproto->cls, &rule->cr); rule_destroy(ofproto, rule); } -/* Returns true if the actions changed, false otherwise. */ -static bool -rule_make_actions(struct ofproto *p, struct rule *rule, - const struct ofpbuf *packet) -{ - const struct rule *super; - struct odp_actions a; - size_t actions_len; - - assert(!rule->cr.wc.wildcards); - - 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, - &rule->nf_flow.output_iface); - - actions_len = a.n_actions * sizeof *a.actions; - if (rule->n_odp_actions != a.n_actions - || memcmp(rule->odp_actions, a.actions, actions_len)) { - COVERAGE_INC(ofproto_odp_unchanged); - free(rule->odp_actions); - rule->n_odp_actions = a.n_actions; - rule->odp_actions = xmemdup(a.actions, actions_len); - return true; - } else { - return false; - } -} - -static int -do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags, - struct odp_flow_put *put) +/* Remove 'facet' from 'ofproto' and free up the associated memory: + * + * - If 'facet' was installed in the datapath, uninstalls it and updates its + * rule's statistics, via facet_uninstall(). + * + * - Removes 'facet' from its rule and from ofproto->facets. + */ +static void +facet_remove(struct ofproto *ofproto, struct facet *facet) { - memset(&put->flow.stats, 0, sizeof put->flow.stats); - put->flow.key = rule->cr.flow; - put->flow.actions = rule->odp_actions; - put->flow.n_actions = rule->n_odp_actions; - put->flow.flags = 0; - put->flags = flags; - return dpif_flow_put(ofproto->dpif, put); + facet_uninstall(ofproto, facet); + facet_flush_stats(ofproto, facet); + hmap_remove(&ofproto->facets, &facet->hmap_node); + list_remove(&facet->list_node); + facet_free(facet); } +/* Composes the ODP actions for 'facet' based on its rule's actions. */ static void -rule_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule) -{ - assert(!rule->cr.wc.wildcards); - - if (rule->may_install) { - struct odp_flow_put put; - if (!do_put_flow(p, rule, - ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS, - &put)) { - rule->installed = true; - if (displaced_rule) { - update_stats(p, displaced_rule, &put.flow.stats); - rule_post_uninstall(p, displaced_rule); - } - } - } else if (displaced_rule) { - rule_uninstall(p, displaced_rule); +facet_make_actions(struct ofproto *p, struct facet *facet, + const struct ofpbuf *packet) +{ + const struct rule *rule = facet->rule; + struct ofpbuf *odp_actions; + struct action_xlate_ctx ctx; + + action_xlate_ctx_init(&ctx, p, &facet->flow, packet); + odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions); + facet->tags = ctx.tags; + facet->may_install = ctx.may_set_up_flow; + facet->nf_flow.output_iface = ctx.nf_output_iface; + + if (facet->actions_len != odp_actions->size + || memcmp(facet->actions, odp_actions->data, odp_actions->size)) { + free(facet->actions); + facet->actions_len = odp_actions->size; + facet->actions = xmemdup(odp_actions->data, odp_actions->size); } + + ofpbuf_delete(odp_actions); } -static void -rule_reinstall(struct ofproto *ofproto, struct rule *rule) +static int +facet_put__(struct ofproto *ofproto, struct facet *facet, + const struct nlattr *actions, size_t actions_len, + struct dpif_flow_stats *stats) { - if (rule->installed) { - struct odp_flow_put put; - COVERAGE_INC(ofproto_dp_missed); - do_put_flow(ofproto, rule, ODPPF_CREATE | ODPPF_MODIFY, &put); - } else { - rule_install(ofproto, rule, NULL); + struct odputil_keybuf keybuf; + enum dpif_flow_put_flags flags; + struct ofpbuf key; + + flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; + if (stats) { + flags |= DPIF_FP_ZERO_STATS; + facet->dp_packet_count = 0; + facet->dp_byte_count = 0; } + + ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); + odp_flow_key_from_flow(&key, &facet->flow); + + return dpif_flow_put(ofproto->dpif, flags, key.data, key.size, + actions, actions_len, stats); } +/* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If + * 'zero_stats' is true, clears any existing statistics from the datapath for + * 'facet'. */ static void -rule_update_actions(struct ofproto *ofproto, struct rule *rule) +facet_install(struct ofproto *p, struct facet *facet, bool zero_stats) { - 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) { - struct odp_flow_put 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); - } - } else { - rule_uninstall(ofproto, rule); + struct dpif_flow_stats stats; + + if (facet->may_install + && !facet_put__(p, facet, facet->actions, facet->actions_len, + zero_stats ? &stats : NULL)) { + facet->installed = true; } } +/* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up + * to the accounting hook function in the ofhooks structure. */ static void -rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes) +facet_account(struct ofproto *ofproto, + struct facet *facet, uint64_t extra_bytes) { - uint64_t total_bytes = rule->byte_count + extra_bytes; + uint64_t total_bytes = facet->byte_count + extra_bytes; if (ofproto->ofhooks->account_flow_cb - && total_bytes > rule->accounted_bytes) + && total_bytes > facet->accounted_bytes) { ofproto->ofhooks->account_flow_cb( - &rule->cr.flow, rule->tags, rule->odp_actions, rule->n_odp_actions, - total_bytes - rule->accounted_bytes, ofproto->aux); - rule->accounted_bytes = total_bytes; + &facet->flow, facet->tags, facet->actions, facet->actions_len, + total_bytes - facet->accounted_bytes, ofproto->aux); + facet->accounted_bytes = total_bytes; } } +/* If 'rule' is installed in the datapath, uninstalls it. */ static void -rule_uninstall(struct ofproto *p, struct rule *rule) -{ - assert(!rule->cr.wc.wildcards); - if (rule->installed) { - struct odp_flow odp_flow; - - odp_flow.key = rule->cr.flow; - odp_flow.actions = NULL; - odp_flow.n_actions = 0; - odp_flow.flags = 0; - if (!dpif_flow_del(p->dpif, &odp_flow)) { - update_stats(p, rule, &odp_flow.stats); - } - rule->installed = false; +facet_uninstall(struct ofproto *p, struct facet *facet) +{ + if (facet->installed) { + struct odputil_keybuf keybuf; + struct dpif_flow_stats stats; + struct ofpbuf key; + + ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); + odp_flow_key_from_flow(&key, &facet->flow); - rule_post_uninstall(p, rule); + if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) { + facet_update_stats(p, facet, &stats); + } + facet->installed = false; + facet->dp_packet_count = 0; + facet->dp_byte_count = 0; + } else { + assert(facet->dp_packet_count == 0); + assert(facet->dp_byte_count == 0); } } +/* Returns true if the only action for 'facet' is to send to the controller. + * (We don't report NetFlow expiration messages for such facets because they + * are just part of the control logic for the network, not real traffic). */ static bool -is_controller_rule(struct rule *rule) +facet_is_controller_flow(struct facet *facet) { - /* 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. */ - - return (rule - && rule->super - && rule->super->n_actions == 1 - && action_outputs_to_port(&rule->super->actions[0], + return (facet + && facet->rule->n_actions == 1 + && action_outputs_to_port(&facet->rule->actions[0], htons(OFPP_CONTROLLER))); } +/* Folds all of 'facet''s statistics into its rule. Also updates the + * accounting ofhook and emits a NetFlow expiration if appropriate. All of + * 'facet''s statistics in the datapath should have been zeroed and folded into + * its packet and byte counts before this function is called. */ static void -rule_post_uninstall(struct ofproto *ofproto, struct rule *rule) +facet_flush_stats(struct ofproto *ofproto, struct facet *facet) { - struct rule *super = rule->super; + assert(!facet->dp_byte_count); + assert(!facet->dp_packet_count); - rule_account(ofproto, rule, 0); + facet_push_stats(ofproto, facet); + facet_account(ofproto, facet, 0); - if (ofproto->netflow && !is_controller_rule(rule)) { + if (ofproto->netflow && !facet_is_controller_flow(facet)) { struct ofexpired expired; - expired.flow = rule->cr.flow; - expired.packet_count = rule->packet_count; - expired.byte_count = rule->byte_count; - expired.used = rule->used; - netflow_expire(ofproto->netflow, &rule->nf_flow, &expired); + expired.flow = facet->flow; + expired.packet_count = facet->packet_count; + expired.byte_count = facet->byte_count; + expired.used = facet->used; + netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); } - if (super) { - super->packet_count += rule->packet_count; - super->byte_count += rule->byte_count; - /* Reset counters to prevent double counting if the rule ever gets - * reinstalled. */ - rule->packet_count = 0; - rule->byte_count = 0; - rule->accounted_bytes = 0; + facet->rule->packet_count += facet->packet_count; + facet->rule->byte_count += facet->byte_count; - netflow_flow_clear(&rule->nf_flow); - } + /* Reset counters to prevent double counting if 'facet' ever gets + * reinstalled. */ + facet->packet_count = 0; + facet->byte_count = 0; + facet->rs_packet_count = 0; + facet->rs_byte_count = 0; + facet->accounted_bytes = 0; + + netflow_flow_clear(&facet->nf_flow); } - -static void -queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, - struct rconn_packet_counter *counter) + +/* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. + * Returns it if found, otherwise a null pointer. + * + * The returned facet might need revalidation; use facet_lookup_valid() + * instead if that is important. */ +static struct facet * +facet_find(struct ofproto *ofproto, const struct flow *flow) { - update_openflow_length(msg); - if (rconn_send(ofconn->rconn, msg, counter)) { - ofpbuf_delete(msg); + struct facet *facet; + + HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0), + &ofproto->facets) { + if (flow_equal(flow, &facet->flow)) { + return facet; + } } + + return NULL; } -static void -send_error(const struct ofconn *ofconn, const struct ofp_header *oh, - int error, const void *data, size_t len) +/* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. + * Returns it if found, otherwise a null pointer. + * + * The returned facet is guaranteed to be valid. */ +static struct facet * +facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow) { - struct ofpbuf *buf; - struct ofp_error_msg *oem; + struct facet *facet = facet_find(ofproto, flow); - if (!(error >> 16)) { - VLOG_WARN_RL(&rl, "not sending bad error code %d to controller", - error); - return; + /* The facet we found might not be valid, since we could be in need of + * revalidation. If it is not valid, don't return it. */ + if (facet + && ofproto->need_revalidate + && !facet_revalidate(ofproto, facet)) { + COVERAGE_INC(ofproto_invalidated); + return NULL; } - COVERAGE_INC(ofproto_error); - oem = make_openflow_xid(len + sizeof *oem, OFPT_ERROR, - oh ? oh->xid : 0, &buf); - oem->type = htons((unsigned int) error >> 16); - oem->code = htons(error & 0xffff); - memcpy(oem->data, data, len); - queue_tx(buf, ofconn, ofconn->reply_counter); + return facet; } -static void -send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh, - int error) +/* Re-searches 'ofproto''s classifier for a rule matching 'facet': + * + * - If the rule found is different from 'facet''s current rule, moves + * 'facet' to the new rule and recompiles its actions. + * + * - If the rule found is the same as 'facet''s current rule, leaves 'facet' + * where it is and recompiles its actions anyway. + * + * - If there is none, destroys 'facet'. + * + * Returns true if 'facet' still exists, false if it has been destroyed. */ +static bool +facet_revalidate(struct ofproto *ofproto, struct facet *facet) { - size_t oh_length = ntohs(oh->length); - send_error(ofconn, oh, error, oh, MIN(oh_length, 64)); -} + struct action_xlate_ctx ctx; + struct ofpbuf *odp_actions; + struct rule *new_rule; + bool actions_changed; + + COVERAGE_INC(facet_revalidate); + + /* Determine the new rule. */ + new_rule = rule_lookup(ofproto, &facet->flow); + if (!new_rule) { + /* No new rule, so delete the facet. */ + facet_remove(ofproto, facet); + return false; + } + + /* Calculate new ODP actions. + * + * We do not modify any 'facet' state yet, because we might need to, e.g., + * emit a NetFlow expiration and, if so, we need to have the old state + * around to properly compose it. */ + action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL); + odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions); + actions_changed = (facet->actions_len != odp_actions->size + || memcmp(facet->actions, odp_actions->data, + facet->actions_len)); + + /* If the ODP actions changed or the installability changed, then we need + * to talk to the datapath. */ + if (actions_changed || ctx.may_set_up_flow != facet->installed) { + if (ctx.may_set_up_flow) { + struct dpif_flow_stats stats; + + facet_put__(ofproto, facet, + odp_actions->data, odp_actions->size, &stats); + facet_update_stats(ofproto, facet, &stats); + } else { + facet_uninstall(ofproto, facet); + } + /* The datapath flow is gone or has zeroed stats, so push stats out of + * 'facet' into 'rule'. */ + facet_flush_stats(ofproto, facet); + } + + /* Update 'facet' now that we've taken care of all the old state. */ + facet->tags = ctx.tags; + facet->nf_flow.output_iface = ctx.nf_output_iface; + facet->may_install = ctx.may_set_up_flow; + if (actions_changed) { + free(facet->actions); + facet->actions_len = odp_actions->size; + facet->actions = xmemdup(odp_actions->data, odp_actions->size); + } + if (facet->rule != new_rule) { + COVERAGE_INC(facet_changed_rule); + list_remove(&facet->list_node); + list_push_back(&new_rule->facets, &facet->list_node); + facet->rule = new_rule; + facet->used = new_rule->created; + facet->rs_used = facet->used; + } + + ofpbuf_delete(odp_actions); + + return true; +} + static void -hton_ofp_phy_port(struct ofp_phy_port *opp) +send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh, + int error) { - opp->port_no = htons(opp->port_no); - opp->config = htonl(opp->config); - opp->state = htonl(opp->state); - opp->curr = htonl(opp->curr); - opp->advertised = htonl(opp->advertised); - opp->supported = htonl(opp->supported); - opp->peer = htonl(opp->peer); + struct ofpbuf *buf = ofputil_encode_error_msg(error, oh); + if (buf) { + COVERAGE_INC(ofproto_error); + ofconn_send_reply(ofconn, buf); + } } static int -handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh) +handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh) { - struct ofp_header *rq = oh; - queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter); + ofconn_send_reply(ofconn, make_echo_reply(oh)); return 0; } static int -handle_features_request(struct ofproto *p, struct ofconn *ofconn, - struct ofp_header *oh) +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; - unsigned int port_no; struct ofport *port; osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf); - osf->datapath_id = htonll(p->datapath_id); + osf->datapath_id = htonll(ofproto->datapath_id); osf->n_buffers = htonl(pktbuf_capacity()); osf->n_tables = 2; osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS | @@ -2363,56 +1899,50 @@ handle_features_request(struct ofproto *p, struct ofconn *ofconn, (1u << OFPAT_SET_TP_DST) | (1u << OFPAT_ENQUEUE)); - PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { + HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) { hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp)); } - queue_tx(buf, ofconn, ofconn->reply_counter); + ofconn_send_reply(ofconn, buf); return 0; } static int -handle_get_config_request(struct ofproto *p, struct ofconn *ofconn, - struct ofp_header *oh) +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. */ - dpif_get_drop_frags(p->dpif, &drop_frags); + dpif_get_drop_frags(ofproto->dpif, &drop_frags); flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL; /* Send reply. */ osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf); osc->flags = htons(flags); - osc->miss_send_len = htons(ofconn->miss_send_len); - queue_tx(buf, ofconn, ofconn->reply_counter); + osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn)); + ofconn_send_reply(ofconn, buf); return 0; } static int -handle_set_config(struct ofproto *p, struct ofconn *ofconn, - struct ofp_switch_config *osc) +handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc) { - uint16_t flags; - int error; - - error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc); - if (error) { - return error; - } - flags = ntohs(osc->flags); + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); + uint16_t flags = ntohs(osc->flags); - if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) { + if (ofconn_get_type(ofconn) == OFCONN_PRIMARY + && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) { switch (flags & OFPC_FRAG_MASK) { case OFPC_FRAG_NORMAL: - dpif_set_drop_frags(p->dpif, false); + dpif_set_drop_frags(ofproto->dpif, false); break; case OFPC_FRAG_DROP: - dpif_set_drop_frags(p->dpif, true); + dpif_set_drop_frags(ofproto->dpif, true); break; default: VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")", @@ -2421,57 +1951,18 @@ handle_set_config(struct ofproto *p, struct ofconn *ofconn, } } - ofconn->miss_send_len = ntohs(osc->miss_send_len); + ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len)); return 0; } -static void -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 -add_controller_action(struct odp_actions *actions, uint16_t max_len) -{ - union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER); - a->controller.arg = max_len; -} - -struct action_xlate_ctx { - /* Input. */ - flow_t flow; /* Flow to which these actions correspond. */ - int recurse; /* Recursion level, via xlate_table_action. */ - struct ofproto *ofproto; - const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a - * null pointer if we are revalidating - * without a packet to refer to. */ - - /* Output. */ - struct odp_actions *out; /* Datapath actions. */ - tag_type *tags; /* Tags associated with OFPP_NORMAL actions. */ - 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. */ -}; - -/* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a - * flow translation. */ -#define MAX_RESUBMIT_RECURSION 8 - static void do_xlate_actions(const union ofp_action *in, size_t n_in, struct action_xlate_ctx *ctx); static void add_output_action(struct action_xlate_ctx *ctx, uint16_t port) { - const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port); + const struct ofport *ofport = get_port(ctx->ofproto, port); if (ofport) { if (ofport->opp.config & OFPPC_NO_FWD) { @@ -2486,27 +1977,14 @@ add_output_action(struct action_xlate_ctx *ctx, uint16_t port) */ } - odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port; + nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port); ctx->nf_output_iface = port; } static struct rule * -lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow) +rule_lookup(struct ofproto *ofproto, const struct flow *flow) { - struct rule *rule; - rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow)); - - /* The rule we found might not be valid, since we could be in need of - * revalidation. If it is not valid, don't return it. */ - if (rule - && rule->super - && ofproto->need_revalidate - && !revalidate_rule(ofproto, rule)) { - COVERAGE_INC(ofproto_invalidated); - return NULL; - } - - return rule; + return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow)); } static void @@ -2521,26 +1999,41 @@ xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port) * have surprising behavior). */ old_in_port = ctx->flow.in_port; ctx->flow.in_port = in_port; - rule = lookup_valid_rule(ctx->ofproto, &ctx->flow); + rule = rule_lookup(ctx->ofproto, &ctx->flow); ctx->flow.in_port = old_in_port; - if (rule) { - if (rule->super) { - rule = rule->super; - } + if (ctx->resubmit_hook) { + ctx->resubmit_hook(ctx, rule); + } + if (rule) { ctx->recurse++; do_xlate_actions(rule->actions, rule->n_actions, ctx); ctx->recurse--; } } else { - struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); + static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times", MAX_RESUBMIT_RECURSION); } } +static void +flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask, + uint16_t *nf_output_iface, struct ofpbuf *odp_actions) +{ + struct ofport *ofport; + + HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) { + uint16_t odp_port = ofport->odp_port; + if (odp_port != odp_in_port && !(ofport->opp.config & mask)) { + nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port); + } + } + *nf_output_iface = NF_OUT_FLOOD; +} + static void xlate_output_action__(struct action_xlate_ctx *ctx, uint16_t port, uint16_t max_len) @@ -2559,7 +2052,7 @@ xlate_output_action__(struct action_xlate_ctx *ctx, break; case OFPP_NORMAL: if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet, - ctx->out, ctx->tags, + ctx->odp_actions, &ctx->tags, &ctx->nf_output_iface, ctx->ofproto->aux)) { COVERAGE_INC(ofproto_uninstallable); @@ -2567,14 +2060,15 @@ xlate_output_action__(struct action_xlate_ctx *ctx, } break; case OFPP_FLOOD: - add_output_group_action(ctx->out, DP_GROUP_FLOOD, - &ctx->nf_output_iface); + flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD, + &ctx->nf_output_iface, ctx->odp_actions); break; case OFPP_ALL: - add_output_group_action(ctx->out, DP_GROUP_ALL, &ctx->nf_output_iface); + flood_packets(ctx->ofproto, ctx->flow.in_port, 0, + &ctx->nf_output_iface, ctx->odp_actions); break; case OFPP_CONTROLLER: - add_controller_action(ctx->out, max_len); + nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len); break; case OFPP_LOCAL: add_output_action(ctx, ODPP_LOCAL); @@ -2611,9 +2105,18 @@ xlate_output_action(struct action_xlate_ctx *ctx, static void remove_pop_action(struct action_xlate_ctx *ctx) { - size_t n = ctx->out->n_actions; - if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) { - ctx->out->n_actions--; + if (ctx->odp_actions->size == ctx->last_pop_priority) { + ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN); + ctx->last_pop_priority = -1; + } +} + +static void +add_pop_action(struct action_xlate_ctx *ctx) +{ + if (ctx->odp_actions->size != ctx->last_pop_priority) { + nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY); + ctx->last_pop_priority = ctx->odp_actions->size; } } @@ -2643,10 +2146,9 @@ xlate_enqueue_action(struct action_xlate_ctx *ctx, /* Add ODP actions. */ remove_pop_action(ctx); - odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority - = priority; + nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority); add_output_action(ctx, odp_port); - odp_actions_add(ctx->out, ODPAT_POP_PRIORITY); + add_pop_action(ctx); /* Update NetFlow output port. */ if (ctx->nf_output_iface == NF_OUT_DROP) { @@ -2656,14 +2158,77 @@ xlate_enqueue_action(struct action_xlate_ctx *ctx, } } +static void +xlate_set_queue_action(struct action_xlate_ctx *ctx, + const struct nx_action_set_queue *nasq) +{ + uint32_t priority; + int error; + + error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id), + &priority); + if (error) { + /* Couldn't translate queue to a priority, so ignore. A warning + * has already been logged. */ + return; + } + + remove_pop_action(ctx); + nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority); +} + +static void +xlate_set_dl_tci(struct action_xlate_ctx *ctx) +{ + ovs_be16 tci = ctx->flow.vlan_tci; + if (!(tci & htons(VLAN_CFI))) { + nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN); + } else { + nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI, + tci & ~htons(VLAN_CFI)); + } +} + +struct xlate_reg_state { + ovs_be16 vlan_tci; + ovs_be64 tun_id; +}; + +static void +save_reg_state(const struct action_xlate_ctx *ctx, + struct xlate_reg_state *state) +{ + state->vlan_tci = ctx->flow.vlan_tci; + state->tun_id = ctx->flow.tun_id; +} + +static void +update_reg_state(struct action_xlate_ctx *ctx, + const struct xlate_reg_state *state) +{ + if (ctx->flow.vlan_tci != state->vlan_tci) { + xlate_set_dl_tci(ctx); + } + if (ctx->flow.tun_id != state->tun_id) { + nl_msg_put_be64(ctx->odp_actions, + ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id); + } +} + static void xlate_nicira_action(struct action_xlate_ctx *ctx, const struct nx_action_header *nah) { const struct nx_action_resubmit *nar; const struct nx_action_set_tunnel *nast; - union odp_action *oa; - int subtype = ntohs(nah->subtype); + const struct nx_action_set_queue *nasq; + const struct nx_action_multipath *nam; + const struct nx_action_autopath *naa; + enum nx_action_subtype subtype = ntohs(nah->subtype); + const struct ofhooks *ofhooks = ctx->ofproto->ofhooks; + struct xlate_reg_state state; + uint16_t autopath_port; + ovs_be64 tun_id; assert(nah->vendor == htonl(NX_VENDOR_ID)); switch (subtype) { @@ -2674,21 +2239,71 @@ xlate_nicira_action(struct action_xlate_ctx *ctx, case NXAST_SET_TUNNEL: nast = (const struct nx_action_set_tunnel *) nah; - oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL); - ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id; + tun_id = htonll(ntohl(nast->tun_id)); + nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id); + ctx->flow.tun_id = tun_id; break; case NXAST_DROP_SPOOFED_ARP: if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) { - odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP); + nl_msg_put_flag(ctx->odp_actions, + ODP_ACTION_ATTR_DROP_SPOOFED_ARP); } break; + case NXAST_SET_QUEUE: + nasq = (const struct nx_action_set_queue *) nah; + xlate_set_queue_action(ctx, nasq); + break; + + case NXAST_POP_QUEUE: + add_pop_action(ctx); + break; + + case NXAST_REG_MOVE: + save_reg_state(ctx, &state); + nxm_execute_reg_move((const struct nx_action_reg_move *) nah, + &ctx->flow); + update_reg_state(ctx, &state); + break; + + case NXAST_REG_LOAD: + save_reg_state(ctx, &state); + nxm_execute_reg_load((const struct nx_action_reg_load *) nah, + &ctx->flow); + update_reg_state(ctx, &state); + break; + + case NXAST_NOTE: + /* Nothing to do. */ + break; + + case NXAST_SET_TUNNEL64: + tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id; + nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id); + ctx->flow.tun_id = tun_id; + break; + + case NXAST_MULTIPATH: + nam = (const struct nx_action_multipath *) nah; + multipath_execute(nam, &ctx->flow); + break; + + case NXAST_AUTOPATH: + naa = (const struct nx_action_autopath *) nah; + autopath_port = (ofhooks->autopath_cb + ? ofhooks->autopath_cb(&ctx->flow, ntohl(naa->id), + &ctx->tags, ctx->ofproto->aux) + : OFPP_NONE); + autopath_execute(naa, &ctx->flow, autopath_port); + break; + /* If you add a new action here that modifies flow data, don't forget to * update the flow key in ctx->flow at the same time. */ + case NXAST_SNAT__OBSOLETE: default: - VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype); + VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype); break; } } @@ -2701,7 +2316,7 @@ do_xlate_actions(const union ofp_action *in, size_t n_in, const union ofp_action *ia; const struct ofport *port; - port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port); + port = get_port(ctx->ofproto, ctx->flow.in_port); if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) && port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) { @@ -2710,8 +2325,8 @@ do_xlate_actions(const union ofp_action *in, size_t n_in, } for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) { - uint16_t type = ntohs(ia->type); - union odp_action *oa; + enum ofp_action_type type = ntohs(ia->type); + const struct ofp_action_dl_addr *oada; switch (type) { case OFPAT_OUTPUT: @@ -2719,60 +2334,65 @@ do_xlate_actions(const union ofp_action *in, size_t n_in, break; case OFPAT_SET_VLAN_VID: - oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID); - ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid; + ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); + ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI); + xlate_set_dl_tci(ctx); break; case OFPAT_SET_VLAN_PCP: - oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP); - ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp; + ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); + ctx->flow.vlan_tci |= htons( + (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI); + xlate_set_dl_tci(ctx); break; case OFPAT_STRIP_VLAN: - odp_actions_add(ctx->out, ODPAT_STRIP_VLAN); - ctx->flow.dl_vlan = htons(OFP_VLAN_NONE); - ctx->flow.dl_vlan_pcp = 0; + ctx->flow.vlan_tci = htons(0); + xlate_set_dl_tci(ctx); break; case OFPAT_SET_DL_SRC: - oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC); - memcpy(oa->dl_addr.dl_addr, - ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); - memcpy(ctx->flow.dl_src, - ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); + oada = ((struct ofp_action_dl_addr *) ia); + nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC, + oada->dl_addr, ETH_ADDR_LEN); + memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN); break; case OFPAT_SET_DL_DST: - oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST); - memcpy(oa->dl_addr.dl_addr, - ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); - memcpy(ctx->flow.dl_dst, - ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); + oada = ((struct ofp_action_dl_addr *) ia); + nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST, + oada->dl_addr, ETH_ADDR_LEN); + memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN); break; case OFPAT_SET_NW_SRC: - oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC); - ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; + nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC, + ia->nw_addr.nw_addr); + ctx->flow.nw_src = ia->nw_addr.nw_addr; break; case OFPAT_SET_NW_DST: - oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST); - ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; + nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST, + ia->nw_addr.nw_addr); + ctx->flow.nw_dst = ia->nw_addr.nw_addr; break; case OFPAT_SET_NW_TOS: - oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS); - ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos; + nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS, + ia->nw_tos.nw_tos); + ctx->flow.nw_tos = ia->nw_tos.nw_tos; break; case OFPAT_SET_TP_SRC: - oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC); - ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port; + nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC, + ia->tp_port.tp_port); + ctx->flow.tp_src = ia->tp_port.tp_port; break; case OFPAT_SET_TP_DST: - oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST); - ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port; + nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST, + ia->tp_port.tp_port); + ctx->flow.tp_dst = ia->tp_port.tp_port; break; case OFPAT_VENDOR: @@ -2784,70 +2404,89 @@ do_xlate_actions(const union ofp_action *in, size_t n_in, break; default: - VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type); + VLOG_DBG_RL(&rl, "unknown action type %d", (int) type); break; } } } -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_set_up_flow, - uint16_t *nf_output_iface) +static void +action_xlate_ctx_init(struct action_xlate_ctx *ctx, + struct ofproto *ofproto, const struct flow *flow, + const struct ofpbuf *packet) +{ + ctx->ofproto = ofproto; + ctx->flow = *flow; + ctx->packet = packet; + ctx->resubmit_hook = NULL; + ctx->check_special = true; +} + +static void +ofproto_process_cfm(struct ofproto *ofproto, const struct flow *flow, + const struct ofpbuf *packet) +{ + struct ofport *ofport; + + ofport = get_port(ofproto, flow->in_port); + if (ofport && ofport->cfm) { + cfm_process_heartbeat(ofport->cfm, packet); + } +} + +static struct ofpbuf * +xlate_actions(struct action_xlate_ctx *ctx, + const union ofp_action *in, size_t n_in) { - tag_type no_tags = 0; - struct action_xlate_ctx ctx; COVERAGE_INC(ofproto_ofp2odp); - odp_actions_init(out); - ctx.flow = *flow; - ctx.recurse = 0; - ctx.ofproto = ofproto; - ctx.packet = packet; - ctx.out = out; - ctx.tags = tags ? tags : &no_tags; - ctx.may_set_up_flow = true; - ctx.nf_output_iface = NF_OUT_DROP; - do_xlate_actions(in, n_in, &ctx); - remove_pop_action(&ctx); + + ctx->odp_actions = ofpbuf_new(512); + ctx->tags = 0; + ctx->may_set_up_flow = true; + ctx->nf_output_iface = NF_OUT_DROP; + ctx->recurse = 0; + ctx->last_pop_priority = -1; + + if (ctx->check_special && cfm_should_process_flow(&ctx->flow)) { + if (ctx->packet) { + ofproto_process_cfm(ctx->ofproto, &ctx->flow, ctx->packet); + } + ctx->may_set_up_flow = false; + } else if (ctx->check_special + && ctx->ofproto->ofhooks->special_cb + && !ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet, + ctx->ofproto->aux)) { + ctx->may_set_up_flow = false; + } else { + do_xlate_actions(in, n_in, ctx); + } + + remove_pop_action(ctx); /* 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_set_up_flow = false; + if (!connmgr_may_set_up_flow(ctx->ofproto->connmgr, &ctx->flow, + ctx->odp_actions->data, + ctx->odp_actions->size)) { + ctx->may_set_up_flow = false; } - 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)) { - COVERAGE_INC(odp_overflow); - odp_actions_init(out); - return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY); - } - return 0; + return ctx->odp_actions; } /* 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. * - * 'oh' is used to make log messages more informative. */ + * The log message mentions 'msg_type'. */ static int -reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh) +reject_slave_controller(struct ofconn *ofconn, const const char *msg_type) { - if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) { + 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); - char *type_name; - - type_name = ofp_message_type_to_string(oh->type); VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller", - type_name); - free(type_name); + msg_type); return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); } else { @@ -2856,51 +2495,67 @@ reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh) } static int -handle_packet_out(struct ofproto *p, struct ofconn *ofconn, - struct ofp_header *oh) +handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh) { + struct ofproto *p = ofconn_get_ofproto(ofconn); struct ofp_packet_out *opo; struct ofpbuf payload, *buffer; - struct odp_actions actions; - int n_actions; + union ofp_action *ofp_actions; + struct action_xlate_ctx ctx; + struct ofpbuf *odp_actions; + struct ofpbuf request; + struct flow flow; + size_t n_ofp_actions; uint16_t in_port; - flow_t flow; int error; - error = reject_slave_controller(ofconn, oh); - if (error) { + COVERAGE_INC(ofproto_packet_out); + + error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT"); + if (error) { return error; } - error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports); + /* 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); if (error) { return error; } - opo = (struct ofp_packet_out *) oh; - COVERAGE_INC(ofproto_packet_out); + /* Get payload. */ if (opo->buffer_id != htonl(UINT32_MAX)) { - error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id), - &buffer, &in_port); + error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id), + &buffer, &in_port); if (error || !buffer) { return error; } payload = *buffer; } else { + payload = request; buffer = NULL; } - flow_extract(&payload, 0, 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, NULL); + /* Extract flow, check actions. */ + flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), + &flow); + error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports); if (error) { - return error; + goto exit; } - dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions, - &payload); - ofpbuf_delete(buffer); + /* Send. */ + action_xlate_ctx_init(&ctx, p, &flow, &payload); + odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions); + dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload); + ofpbuf_delete(odp_actions); +exit: + ofpbuf_delete(buffer); return 0; } @@ -2916,42 +2571,33 @@ update_port_config(struct ofproto *p, struct ofport *port, netdev_turn_flags_on(port->netdev, NETDEV_UP, true); } } -#define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD) +#define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \ + OFPPC_NO_FWD | OFPPC_NO_FLOOD) if (mask & REVALIDATE_BITS) { COVERAGE_INC(ofproto_costly_flags); port->opp.config ^= mask & REVALIDATE_BITS; p->need_revalidate = true; } #undef REVALIDATE_BITS - if (mask & OFPPC_NO_FLOOD) { - port->opp.config ^= OFPPC_NO_FLOOD; - refresh_port_groups(p); - } if (mask & OFPPC_NO_PACKET_IN) { port->opp.config ^= OFPPC_NO_PACKET_IN; } } static int -handle_port_mod(struct ofproto *p, struct ofconn *ofconn, - struct ofp_header *oh) +handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh) { - const struct ofp_port_mod *opm; + struct ofproto *p = ofconn_get_ofproto(ofconn); + const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh; struct ofport *port; int error; - error = reject_slave_controller(ofconn, oh); - if (error) { - return error; - } - error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm); + error = reject_slave_controller(ofconn, "OFPT_PORT_MOD"); if (error) { return error; } - opm = (struct ofp_port_mod *) oh; - port = port_array_get(&p->ports, - ofp_port_to_odp_port(ntohs(opm->port_no))); + port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no))); if (!port) { return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT); } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) { @@ -2966,7 +2612,7 @@ handle_port_mod(struct ofproto *p, struct ofconn *ofconn, } static struct ofpbuf * -make_stats_reply(uint32_t xid, uint16_t type, size_t body_len) +make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len) { struct ofp_stats_reply *osr; struct ofpbuf *msg; @@ -2979,420 +2625,470 @@ make_stats_reply(uint32_t xid, uint16_t type, size_t body_len) } static struct ofpbuf * -start_stats_reply(const struct ofp_stats_request *request, size_t body_len) +start_ofp_stats_reply(const struct ofp_header *request, size_t body_len) { - return make_stats_reply(request->header.xid, request->type, 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_stats_reply(size_t nbytes, struct ofconn *ofconn, struct ofpbuf **msgp) +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_stats_reply(reply->header.xid, reply->type, nbytes); - queue_tx(msg, ofconn, ofconn->reply_counter); + *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 -handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn, - struct ofp_stats_request *request) +handle_desc_stats_request(struct ofconn *ofconn, + const struct ofp_header *request) { + struct ofproto *p = ofconn_get_ofproto(ofconn); struct ofp_desc_stats *ods; struct ofpbuf *msg; - msg = start_stats_reply(request, sizeof *ods); - ods = append_stats_reply(sizeof *ods, ofconn, &msg); + msg = start_ofp_stats_reply(request, sizeof *ods); + ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg); memset(ods, 0, sizeof *ods); 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); ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num); ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc); - queue_tx(msg, ofconn, ofconn->reply_counter); + ofconn_send_reply(ofconn, msg); return 0; } -static void -count_subrules(struct cls_rule *cls_rule, void *n_subrules_) -{ - struct rule *rule = rule_from_cls_rule(cls_rule); - int *n_subrules = n_subrules_; - - if (rule->super) { - (*n_subrules)++; - } -} - static int -handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn, - struct ofp_stats_request *request) +handle_table_stats_request(struct ofconn *ofconn, + const struct ofp_header *request) { + struct ofproto *p = ofconn_get_ofproto(ofconn); struct ofp_table_stats *ots; struct ofpbuf *msg; - struct odp_stats dpstats; - int n_exact, n_subrules, n_wild; - - msg = start_stats_reply(request, sizeof *ots * 2); - /* Count rules of various kinds. */ - n_subrules = 0; - classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules); - n_exact = classifier_count_exact(&p->cls) - n_subrules; - n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls); - - /* Hash table. */ - dpif_get_dp_stats(p->dpif, &dpstats); - ots = append_stats_reply(sizeof *ots, ofconn, &msg); - memset(ots, 0, sizeof *ots); - ots->table_id = TABLEID_HASH; - strcpy(ots->name, "hash"); - ots->wildcards = htonl(0); - ots->max_entries = htonl(dpstats.max_capacity); - ots->active_count = htonl(n_exact); - ots->lookup_count = htonll(dpstats.n_frags + dpstats.n_hit + - dpstats.n_missed); - ots->matched_count = htonll(dpstats.n_hit); /* XXX */ + msg = start_ofp_stats_reply(request, sizeof *ots * 2); /* Classifier table. */ - ots = append_stats_reply(sizeof *ots, ofconn, &msg); + ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg); memset(ots, 0, sizeof *ots); - ots->table_id = TABLEID_CLASSIFIER; strcpy(ots->name, "classifier"); - ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL) - : htonl(OFPFW_ALL); - ots->max_entries = htonl(65536); - ots->active_count = htonl(n_wild); - ots->lookup_count = htonll(0); /* XXX */ - ots->matched_count = htonll(0); /* XXX */ - - queue_tx(msg, ofconn, ofconn->reply_counter); + 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 */ + + ofconn_send_reply(ofconn, msg); return 0; } static void -append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn, +append_port_stat(struct ofport *port, struct ofconn *ofconn, struct ofpbuf **msgp) { 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 + /* 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, msgp); - ops->port_no = htons(odp_port_to_ofp_port(port_no)); + ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp); + ops->port_no = htons(port->opp.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); + put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets)); + put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets)); + put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes)); + put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes)); + put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped)); + put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped)); + put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors)); + put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors)); + put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors)); + put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors)); + put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors)); + put_32aligned_be64(&ops->collisions, htonll(stats.collisions)); } static int -handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn, - struct ofp_stats_request *osr, - size_t arg_size) +handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) { - 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; - unsigned int port_no; - 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); + msg = start_ofp_stats_reply(oh, 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))); + port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no))); if (port) { - append_port_stat(port, ntohs(psr->port_no), ofconn, &msg); + append_port_stat(port, ofconn, &msg); } } else { - PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { - append_port_stat(port, port_no, ofconn, &msg); + HMAP_FOR_EACH (port, hmap_node, &p->ports) { + append_port_stat(port, ofconn, &msg); } } - queue_tx(msg, ofconn, ofconn->reply_counter); + ofconn_send_reply(ofconn, msg); return 0; } -struct flow_stats_cbdata { - struct ofproto *ofproto; - struct ofconn *ofconn; - uint16_t out_port; - struct ofpbuf *msg; -}; - -/* Obtains statistic counters for 'rule' within 'p' and stores them into - * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the - * returned statistic include statistics for all of 'rule''s subrules. */ static void -query_stats(struct ofproto *p, struct rule *rule, - uint64_t *packet_countp, uint64_t *byte_countp) +calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec) { - uint64_t packet_count, byte_count; - struct rule *subrule; - struct odp_flow *odp_flows; - size_t n_odp_flows; - - /* Start from historical data for 'rule' itself that are no longer tracked - * by the datapath. This counts, for example, subrules that have - * expired. */ - packet_count = rule->packet_count; - byte_count = rule->byte_count; - - /* Prepare to ask the datapath for statistics on 'rule', or if it is - * wildcarded then on all of its subrules. - * - * Also, add any statistics that are not tracked by the datapath for each - * subrule. This includes, for example, statistics for packets that were - * executed "by hand" by ofproto via dpif_execute() but must be accounted - * to a flow. */ - n_odp_flows = rule->cr.wc.wildcards ? list_size(&rule->list) : 1; - 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; - } + long long int msecs = time_msec() - start; + *sec = msecs / 1000; + *nsec = (msecs % 1000) * (1000 * 1000); +} - /* Fetch up-to-date statistics from the datapath and add them in. */ - if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) { - size_t i; - for (i = 0; i < n_odp_flows; i++) { - struct odp_flow *odp_flow = &odp_flows[i]; - packet_count += odp_flow->stats.n_packets; - byte_count += odp_flow->stats.n_bytes; - } - } - free(odp_flows); +static void +calc_flow_duration(long long int start, ovs_be32 *sec_be, ovs_be32 *nsec_be) +{ + uint32_t sec, nsec; - /* Return the stats to the caller. */ - *packet_countp = packet_count; - *byte_countp = byte_count; + calc_flow_duration__(start, &sec, &nsec); + *sec_be = htonl(sec); + *nsec_be = htonl(nsec); } static void -flow_stats_cb(struct cls_rule *rule_, void *cbdata_) +put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule, + ovs_be16 out_port, struct ofpbuf **replyp) { - struct rule *rule = rule_from_cls_rule(rule_); - struct flow_stats_cbdata *cbdata = cbdata_; struct ofp_flow_stats *ofs; uint64_t packet_count, byte_count; + ovs_be64 cookie; 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)) { + if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) { return; } act_len = sizeof *rule->actions * rule->n_actions; len = offsetof(struct ofp_flow_stats, actions) + act_len; - query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); + rule_get_stats(rule, &packet_count, &byte_count); - ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg); + ofs = append_ofp_stats_reply(len, ofconn, replyp); ofs->length = htons(len); - ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH; + ofs->table_id = 0; ofs->pad = 0; - flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, - cbdata->ofproto->tun_id_from_cookie, &ofs->match); - ofs->duration_sec = htonl(sec); - ofs->duration_nsec = htonl(msec * 1000000); - ofs->cookie = rule->flow_cookie; + 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); - ofs->packet_count = htonll(packet_count); - ofs->byte_count = htonll(byte_count); - memcpy(ofs->actions, rule->actions, act_len); + 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); + } +} + +static bool +is_valid_table(uint8_t table_id) +{ + if (table_id == 0 || table_id == 0xff) { + return true; + } 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; + } } static int -table_id_to_include(uint8_t table_id) +handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) { - return (table_id == TABLEID_HASH ? CLS_INC_EXACT - : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD - : table_id == 0xff ? CLS_INC_ALL - : 0); + const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh); + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); + struct ofpbuf *reply; + + COVERAGE_INC(ofproto_flows_req); + reply = start_ofp_stats_reply(oh, 1024); + if (is_valid_table(fsr->table_id)) { + 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_FOR_EACH (rule, cr, &cursor) { + put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply); + } + } + 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; + + if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) { + return; + } + + rule_get_stats(rule, &packet_count, &byte_count); + + 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); + } + nfs->length = htons(reply->size - start_len); } static int -handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn, - const struct ofp_stats_request *osr, - size_t arg_size) +handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh) { - struct ofp_flow_stats_request *fsr; - struct flow_stats_cbdata cbdata; + 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; - if (arg_size != sizeof *fsr) { + ofpbuf_use_const(&b, oh, ntohs(oh->length)); + + /* Dissect the message. */ + nfsr = ofpbuf_pull(&b, sizeof *nfsr); + error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target); + if (error) { + return error; + } + if (b.size) { return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); } - fsr = (struct ofp_flow_stats_request *) osr->body; COVERAGE_INC(ofproto_flows_req); - cbdata.ofproto = p; - cbdata.ofconn = ofconn; - cbdata.out_port = fsr->out_port; - cbdata.msg = start_stats_reply(osr, 1024); - cls_rule_from_match(&fsr->match, 0, false, 0, &target); - classifier_for_each_match(&p->cls, &target, - table_id_to_include(fsr->table_id), - flow_stats_cb, &cbdata); - queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); + 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); + } + } + ofconn_send_reply(ofconn, reply); + return 0; } -struct flow_stats_ds_cbdata { - struct ofproto *ofproto; - struct ds *results; -}; - static void -flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_) +flow_stats_ds(struct rule *rule, struct ds *results) { - struct rule *rule = rule_from_cls_rule(rule_); - struct flow_stats_ds_cbdata *cbdata = cbdata_; - struct ds *results = cbdata->results; - struct ofp_match match; uint64_t packet_count, byte_count; size_t act_len = sizeof *rule->actions * rule->n_actions; - /* Don't report on subrules. */ - if (rule->super != NULL) { - return; - } - - query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); - flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, - cbdata->ofproto->tun_id_from_cookie, &match); + rule_get_stats(rule, &packet_count, &byte_count); ds_put_format(results, "duration=%llds, ", (time_msec() - rule->created) / 1000); + ds_put_format(results, "idle=%.3fs, ", (time_msec() - rule->used) / 1000.0); ds_put_format(results, "priority=%u, ", rule->cr.priority); ds_put_format(results, "n_packets=%"PRIu64", ", packet_count); ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count); - ofp_print_match(results, &match, true); - ofp_print_actions(results, &rule->actions->header, act_len); + cls_rule_format(&rule->cr, results); + ds_put_char(results, ','); + if (act_len > 0) { + ofp_print_actions(results, &rule->actions->header, act_len); + } else { + ds_put_cstr(results, "drop"); + } ds_put_cstr(results, "\n"); } -/* Adds a pretty-printed description of all flows to 'results', including - * those marked hidden by secchan (e.g., by in-band control). */ +/* Adds a pretty-printed description of all flows to 'results', including + * hidden flows (e.g., set up by in-band control). */ void ofproto_get_all_flows(struct ofproto *p, struct ds *results) { - struct ofp_match match; - struct cls_rule target; - struct flow_stats_ds_cbdata cbdata; - - memset(&match, 0, sizeof match); - match.wildcards = htonl(OVSFW_ALL); - - cbdata.ofproto = p; - cbdata.results = results; + struct cls_cursor cursor; + struct rule *rule; - cls_rule_from_match(&match, 0, false, 0, &target); - classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, - flow_stats_ds_cb, &cbdata); + cls_cursor_init(&cursor, &p->cls, NULL); + CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { + flow_stats_ds(rule, results); + } } -struct aggregate_stats_cbdata { - struct ofproto *ofproto; - uint16_t out_port; - uint64_t packet_count; - uint64_t byte_count; - uint32_t n_flows; -}; - static void -aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_) +query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target, + ovs_be16 out_port, uint8_t table_id, + struct ofp_aggregate_stats_reply *oasr) { - struct rule *rule = rule_from_cls_rule(rule_); - struct aggregate_stats_cbdata *cbdata = cbdata_; - uint64_t packet_count, byte_count; + uint64_t total_packets = 0; + uint64_t total_bytes = 0; + int n_flows = 0; - if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { - return; - } + COVERAGE_INC(ofproto_agg_request); - query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); + if (is_valid_table(table_id)) { + struct cls_cursor cursor; + struct rule *rule; - cbdata->packet_count += packet_count; - cbdata->byte_count += byte_count; - cbdata->n_flows++; + 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; + + 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); } static int -handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn, - const struct ofp_stats_request *osr, - size_t arg_size) +handle_aggregate_stats_request(struct ofconn *ofconn, + const struct ofp_header *oh) { - struct ofp_aggregate_stats_request *asr; + 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 aggregate_stats_cbdata cbdata; struct cls_rule target; struct ofpbuf *msg; - if (arg_size != sizeof *asr) { + ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0, + &target); + + 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; +} + +static int +handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh) +{ + 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; + + ofpbuf_use_const(&b, oh, ntohs(oh->length)); + + /* Dissect the message. */ + request = ofpbuf_pull(&b, sizeof *request); + error = nx_pull_match(&b, ntohs(request->match_len), 0, &target); + if (error) { + return error; + } + if (b.size) { return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); } - asr = (struct ofp_aggregate_stats_request *) osr->body; - COVERAGE_INC(ofproto_agg_request); - cbdata.ofproto = p; - cbdata.out_port = asr->out_port; - cbdata.packet_count = 0; - cbdata.byte_count = 0; - cbdata.n_flows = 0; - cls_rule_from_match(&asr->match, 0, false, 0, &target); - classifier_for_each_match(&p->cls, &target, - table_id_to_include(asr->table_id), - aggregate_stats_cb, &cbdata); - - msg = start_stats_reply(osr, sizeof *reply); - reply = append_stats_reply(sizeof *reply, ofconn, &msg); - reply->flow_count = htonl(cbdata.n_flows); - reply->packet_count = htonll(cbdata.packet_count); - reply->byte_count = htonll(cbdata.byte_count); - queue_tx(msg, ofconn, ofconn->reply_counter); + /* 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); + return 0; } struct queue_stats_cbdata { struct ofconn *ofconn; + struct ofport *ofport; struct ofpbuf *msg; - uint16_t port_no; }; static void @@ -3401,13 +3097,13 @@ put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id, { struct ofp_queue_stats *reply; - reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg); - reply->port_no = htons(cbdata->port_no); + reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg); + reply->port_no = htons(cbdata->ofport->opp.port_no); memset(reply->pad, 0, sizeof reply->pad); reply->queue_id = htonl(queue_id); - reply->tx_bytes = htonll(stats->tx_bytes); - reply->tx_packets = htonll(stats->tx_packets); - reply->tx_errors = htonll(stats->tx_errors); + put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes)); + put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets)); + put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors)); } static void @@ -3421,221 +3117,222 @@ handle_queue_stats_dump_cb(uint32_t queue_id, } static void -handle_queue_stats_for_port(struct ofport *port, uint16_t port_no, - uint32_t queue_id, +handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id, struct queue_stats_cbdata *cbdata) { - cbdata->port_no = port_no; + cbdata->ofport = port; if (queue_id == OFPQ_ALL) { netdev_dump_queue_stats(port->netdev, handle_queue_stats_dump_cb, cbdata); } else { struct netdev_queue_stats stats; - netdev_get_queue_stats(port->netdev, queue_id, &stats); - put_queue_stats(cbdata, queue_id, &stats); + if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) { + put_queue_stats(cbdata, queue_id, &stats); + } } } static int -handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn, - const struct ofp_stats_request *osr, - size_t arg_size) +handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) { - 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; - if (arg_size != sizeof *qsr) { + qsr = ofputil_stats_body(oh); + if (!qsr) { return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); } - qsr = (struct ofp_queue_stats_request *) osr->body; COVERAGE_INC(ofproto_queue_req); cbdata.ofconn = ofconn; - cbdata.msg = start_stats_reply(osr, 128); + cbdata.msg = start_ofp_stats_reply(oh, 128); port_no = ntohs(qsr->port_no); queue_id = ntohl(qsr->queue_id); if (port_no == OFPP_ALL) { - PORT_ARRAY_FOR_EACH (port, &ofproto->ports, port_no) { - handle_queue_stats_for_port(port, port_no, queue_id, &cbdata); + HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) { + handle_queue_stats_for_port(port, queue_id, &cbdata); } } else if (port_no < ofproto->max_ports) { - port = port_array_get(&ofproto->ports, port_no); + port = get_port(ofproto, ofp_port_to_odp_port(port_no)); if (port) { - handle_queue_stats_for_port(port, port_no, queue_id, &cbdata); + handle_queue_stats_for_port(port, queue_id, &cbdata); } } else { ofpbuf_delete(cbdata.msg); return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT); } - queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); + ofconn_send_reply(ofconn, cbdata.msg); return 0; } -static int -handle_stats_request(struct ofproto *p, struct ofconn *ofconn, - struct ofp_header *oh) +/* Updates 'facet''s used time. Caller is responsible for calling + * facet_push_stats() to update the flows which 'facet' resubmits into. */ +static void +facet_update_time(struct ofproto *ofproto, struct facet *facet, + long long int used) { - struct ofp_stats_request *osr; - size_t arg_size; - int error; - - error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr, - 1, &arg_size); - if (error) { - return error; + if (used > facet->used) { + facet->used = used; + if (used > facet->rule->used) { + facet->rule->used = used; + } + netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); } - osr = (struct ofp_stats_request *) oh; - - switch (ntohs(osr->type)) { - case OFPST_DESC: - return handle_desc_stats_request(p, ofconn, osr); - - case OFPST_FLOW: - return handle_flow_stats_request(p, ofconn, osr, arg_size); +} - case OFPST_AGGREGATE: - return handle_aggregate_stats_request(p, ofconn, osr, arg_size); +/* Folds the statistics from 'stats' into the counters in 'facet'. + * + * Because of the meaning of a facet's counters, it only makes sense to do this + * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a + * packet that was sent by hand or if it represents statistics that have been + * cleared out of the datapath. */ +static void +facet_update_stats(struct ofproto *ofproto, struct facet *facet, + const struct dpif_flow_stats *stats) +{ + if (stats->n_packets || stats->used > facet->used) { + facet_update_time(ofproto, facet, stats->used); + facet->packet_count += stats->n_packets; + facet->byte_count += stats->n_bytes; + facet_push_stats(ofproto, facet); + netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); + } +} - case OFPST_TABLE: - return handle_table_stats_request(p, ofconn, osr); +static void +facet_push_stats(struct ofproto *ofproto, struct facet *facet) +{ + uint64_t rs_packets, rs_bytes; - case OFPST_PORT: - return handle_port_stats_request(p, ofconn, osr, arg_size); + assert(facet->packet_count >= facet->rs_packet_count); + assert(facet->byte_count >= facet->rs_byte_count); + assert(facet->used >= facet->rs_used); - case OFPST_QUEUE: - return handle_queue_stats_request(p, ofconn, osr, arg_size); + rs_packets = facet->packet_count - facet->rs_packet_count; + rs_bytes = facet->byte_count - facet->rs_byte_count; - case OFPST_VENDOR: - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); + if (rs_packets || rs_bytes || facet->used > facet->rs_used) { + facet->rs_packet_count = facet->packet_count; + facet->rs_byte_count = facet->byte_count; + facet->rs_used = facet->used; - default: - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT); + flow_push_stats(ofproto, facet->rule, &facet->flow, + rs_packets, rs_bytes, facet->used); } } -static long long int -msec_from_nsec(uint64_t sec, uint32_t nsec) -{ - return !sec ? 0 : sec * 1000 + nsec / 1000000; -} +struct ofproto_push { + struct action_xlate_ctx ctx; + uint64_t packets; + uint64_t bytes; + long long int used; +}; static void -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); +push_resubmit(struct action_xlate_ctx *ctx, struct rule *rule) +{ + struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx); + + if (rule) { + rule->packet_count += push->packets; + rule->byte_count += push->bytes; + rule->used = MAX(push->used, rule->used); } } +/* Pushes flow statistics to the rules which 'flow' resubmits into given + * 'rule''s actions. */ static void -update_stats(struct ofproto *ofproto, struct rule *rule, - const struct odp_flow_stats *stats) +flow_push_stats(struct ofproto *ofproto, const struct rule *rule, + struct flow *flow, uint64_t packets, uint64_t bytes, + long long int used) { - if (stats->n_packets) { - 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->tcp_flags); - } + struct ofproto_push push; + + push.packets = packets; + push.bytes = bytes; + push.used = used; + + action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL); + push.ctx.resubmit_hook = push_resubmit; + ofpbuf_delete(xlate_actions(&push.ctx, rule->actions, rule->n_actions)); } /* 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 'p''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 or an + * OpenFlow error code as encoded by ofp_mkerr() on failure. * * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, * if any. */ static int -add_flow(struct ofproto *p, struct ofconn *ofconn, - const struct ofp_flow_mod *ofm, size_t n_actions) +add_flow(struct ofconn *ofconn, struct flow_mod *fm) { + struct ofproto *p = ofconn_get_ofproto(ofconn); struct ofpbuf *packet; struct rule *rule; uint16_t in_port; int error; - if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) { - flow_t flow; - uint32_t wildcards; - - flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie, - &flow, &wildcards); - if (classifier_rule_overlaps(&p->cls, &flow, wildcards, - ntohs(ofm->priority))) { - return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP); - } + if (fm->flags & OFPFF_CHECK_OVERLAP + && classifier_rule_overlaps(&p->cls, &fm->cr)) { + 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), ofm->cookie, - ofm->flags & htons(OFPFF_SEND_FLOW_REM)); - cls_rule_from_match(&ofm->match, ntohs(ofm->priority), - p->tun_id_from_cookie, ofm->cookie, &rule->cr); - error = 0; - if (ofm->buffer_id != htonl(UINT32_MAX)) { - error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id), - &packet, &in_port); + if (fm->buffer_id != UINT32_MAX) { + error = ofconn_pktbuf_retrieve(ofconn, fm->buffer_id, + &packet, &in_port); } else { packet = NULL; in_port = UINT16_MAX; } - rule_insert(p, rule, packet, in_port); + rule = rule_create(&fm->cr, fm->actions, fm->n_actions, + fm->idle_timeout, fm->hard_timeout, fm->cookie, + fm->flags & OFPFF_SEND_FLOW_REM); + rule_insert(p, rule); + if (packet) { + rule_execute(p, rule, in_port, packet); + } return error; } static struct rule * -find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm) +find_flow_strict(struct ofproto *p, const struct flow_mod *fm) { - uint32_t wildcards; - flow_t flow; - - flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie, - &flow, &wildcards); - return rule_from_cls_rule(classifier_find_rule_exactly( - &p->cls, &flow, wildcards, - ntohs(ofm->priority))); + return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr)); } static int -send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn, - struct rule *rule, const struct ofp_flow_mod *ofm) +send_buffered_packet(struct ofconn *ofconn, + struct rule *rule, uint32_t buffer_id) { + struct ofproto *ofproto = ofconn_get_ofproto(ofconn); struct ofpbuf *packet; uint16_t in_port; - flow_t flow; int error; - if (ofm->buffer_id == htonl(UINT32_MAX)) { + if (buffer_id == UINT32_MAX) { return 0; } - error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id), - &packet, &in_port); + error = ofconn_pktbuf_retrieve(ofconn, buffer_id, &packet, &in_port); if (error) { return error; } - flow_extract(packet, 0, in_port, &flow); - rule_execute(ofproto, rule, packet, &flow); + rule_execute(ofproto, rule, in_port, packet); return 0; } @@ -3644,14 +3341,12 @@ send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn, struct modify_flows_cbdata { struct ofproto *ofproto; - const struct ofp_flow_mod *ofm; - size_t n_actions; + const struct flow_mod *fm; struct rule *match; }; -static int modify_flow(struct ofproto *, const struct ofp_flow_mod *, - size_t n_actions, struct rule *); -static void modify_flows_cb(struct cls_rule *, void *cbdata_); +static int modify_flow(struct ofproto *, 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. @@ -3659,30 +3354,29 @@ static void modify_flows_cb(struct cls_rule *, void *cbdata_); * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, * if any. */ static int -modify_flows_loose(struct ofproto *p, struct ofconn *ofconn, - const struct ofp_flow_mod *ofm, size_t n_actions) +modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm) { - struct modify_flows_cbdata cbdata; - struct cls_rule target; + struct ofproto *p = ofconn_get_ofproto(ofconn); + struct rule *match = NULL; + struct cls_cursor cursor; + struct rule *rule; - cbdata.ofproto = p; - cbdata.ofm = ofm; - cbdata.n_actions = n_actions; - cbdata.match = NULL; - - cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie, - &target); - - classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, - modify_flows_cb, &cbdata); - if (cbdata.match) { - /* This credits the packet to whichever flow happened to 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(p, ofconn, cbdata.match, ofm); + cls_cursor_init(&cursor, &p->cls, &fm->cr); + CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { + if (!rule_is_hidden(rule)) { + match = rule; + modify_flow(p, fm, rule); + } + } + + 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(p, ofconn, ofm, n_actions); + return add_flow(ofconn, fm); } } @@ -3692,28 +3386,15 @@ modify_flows_loose(struct ofproto *p, struct ofconn *ofconn, * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, * if any. */ static int -modify_flow_strict(struct ofproto *p, struct ofconn *ofconn, - struct ofp_flow_mod *ofm, size_t n_actions) +modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm) { - struct rule *rule = find_flow_strict(p, ofm); + struct ofproto *p = ofconn_get_ofproto(ofconn); + struct rule *rule = find_flow_strict(p, fm); if (rule && !rule_is_hidden(rule)) { - modify_flow(p, ofm, n_actions, rule); - return send_buffered_packet(p, ofconn, rule, ofm); + modify_flow(p, fm, rule); + return send_buffered_packet(ofconn, rule, fm->buffer_id); } else { - return add_flow(p, ofconn, ofm, n_actions); - } -} - -/* Callback for modify_flows_loose(). */ -static void -modify_flows_cb(struct cls_rule *rule_, void *cbdata_) -{ - struct rule *rule = rule_from_cls_rule(rule_); - struct modify_flows_cbdata *cbdata = cbdata_; - - if (!rule_is_hidden(rule)) { - cbdata->match = rule; - modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule); + return add_flow(ofconn, fm); } } @@ -3722,83 +3403,56 @@ modify_flows_cb(struct cls_rule *rule_, void *cbdata_) * the rule's actions to match those in 'ofm' (which is followed by 'n_actions' * ofp_action[] structures). */ static int -modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm, - size_t n_actions, struct rule *rule) +modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule) { - size_t actions_len = n_actions * sizeof *rule->actions; + size_t actions_len = fm->n_actions * sizeof *rule->actions; - rule->flow_cookie = ofm->cookie; + rule->flow_cookie = fm->cookie; /* If the actions are the same, do nothing. */ - if (n_actions == rule->n_actions - && !memcmp(ofm->actions, rule->actions, actions_len)) - { + if (fm->n_actions == rule->n_actions + && (!fm->n_actions + || !memcmp(fm->actions, rule->actions, actions_len))) { return 0; } /* Replace actions. */ free(rule->actions); - rule->actions = xmemdup(ofm->actions, actions_len); - rule->n_actions = n_actions; + rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL; + rule->n_actions = fm->n_actions; - /* Make sure that the datapath gets updated properly. */ - if (rule->cr.wc.wildcards) { - COVERAGE_INC(ofproto_mod_wc_flow); - p->need_revalidate = true; - } else { - rule_update_actions(p, rule); - } + p->need_revalidate = true; return 0; } /* OFPFC_DELETE implementation. */ -struct delete_flows_cbdata { - struct ofproto *ofproto; - uint16_t out_port; -}; - -static void delete_flows_cb(struct cls_rule *, void *cbdata_); -static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port); +static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port); /* Implements OFPFC_DELETE. */ static void -delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm) +delete_flows_loose(struct ofproto *p, const struct flow_mod *fm) { - struct delete_flows_cbdata cbdata; - struct cls_rule target; - - cbdata.ofproto = p; - cbdata.out_port = ofm->out_port; + struct rule *rule, *next_rule; + struct cls_cursor cursor; - cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie, - &target); - - classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, - delete_flows_cb, &cbdata); + cls_cursor_init(&cursor, &p->cls, &fm->cr); + CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { + delete_flow(p, rule, htons(fm->out_port)); + } } /* Implements OFPFC_DELETE_STRICT. */ static void -delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm) +delete_flow_strict(struct ofproto *p, struct flow_mod *fm) { - struct rule *rule = find_flow_strict(p, ofm); + struct rule *rule = find_flow_strict(p, fm); if (rule) { - delete_flow(p, rule, ofm->out_port); + delete_flow(p, rule, htons(fm->out_port)); } } -/* Callback for delete_flows_loose(). */ -static void -delete_flows_cb(struct cls_rule *rule_, void *cbdata_) -{ - struct rule *rule = rule_from_cls_rule(rule_); - struct delete_flows_cbdata *cbdata = cbdata_; - - delete_flow(cbdata->ofproto, rule, cbdata->out_port); -} - /* 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 @@ -3808,7 +3462,7 @@ delete_flows_cb(struct cls_rule *rule_, void *cbdata_) * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the * specified 'out_port'. */ static void -delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port) +delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port) { if (rule_is_hidden(rule)) { return; @@ -3818,81 +3472,57 @@ delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port) return; } - send_flow_removed(p, rule, time_msec(), OFPRR_DELETE); + rule_send_removed(p, rule, OFPRR_DELETE); rule_remove(p, rule); } static int -handle_flow_mod(struct ofproto *p, struct ofconn *ofconn, - struct ofp_flow_mod *ofm) +handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh) { - struct ofp_match orig_match; - size_t n_actions; + struct ofproto *p = ofconn_get_ofproto(ofconn); + struct flow_mod fm; int error; - error = reject_slave_controller(ofconn, &ofm->header); + error = reject_slave_controller(ofconn, "flow_mod"); if (error) { return error; } - error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm, - sizeof *ofm->actions, &n_actions); + + error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_flow_format(ofconn)); if (error) { 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)) { + /* We do not support the emergency flow cache. It will hopefully get + * dropped from OpenFlow in the near future. */ + 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); } - /* Normalize ofp->match. If normalization actually changes anything, then - * log the differences. */ - ofm->match.pad1[0] = ofm->match.pad2[0] = 0; - orig_match = ofm->match; - normalize_match(&ofm->match); - if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) { - static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1); - if (!VLOG_DROP_INFO(&normal_rl)) { - char *old = ofp_match_to_literal_string(&orig_match); - char *new = ofp_match_to_literal_string(&ofm->match); - VLOG_INFO("%s: normalization changed ofp_match, details:", - rconn_get_name(ofconn->rconn)); - VLOG_INFO(" pre: %s", old); - VLOG_INFO("post: %s", new); - free(old); - free(new); - } - } - - if (!ofm->match.wildcards) { - ofm->priority = htons(UINT16_MAX); - } - - error = validate_actions((const union ofp_action *) ofm->actions, - n_actions, p->max_ports); + error = validate_actions(fm.actions, fm.n_actions, + &fm.cr.flow, p->max_ports); if (error) { return error; } - switch (ntohs(ofm->command)) { + switch (fm.command) { case OFPFC_ADD: - return add_flow(p, ofconn, ofm, n_actions); + return add_flow(ofconn, &fm); case OFPFC_MODIFY: - return modify_flows_loose(p, ofconn, ofm, n_actions); + return modify_flows_loose(ofconn, &fm); case OFPFC_MODIFY_STRICT: - return modify_flow_strict(p, ofconn, ofm, n_actions); + return modify_flow_strict(ofconn, &fm); case OFPFC_DELETE: - delete_flows_loose(p, ofm); + delete_flows_loose(p, &fm); return 0; case OFPFC_DELETE_STRICT: - delete_flow_strict(p, ofm); + delete_flow_strict(p, &fm); return 0; default: @@ -3901,38 +3531,28 @@ handle_flow_mod(struct ofproto *p, struct ofconn *ofconn, } static int -handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg) +handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh) { - int error; + const struct nxt_tun_id_cookie *msg + = (const struct nxt_tun_id_cookie *) oh; + enum nx_flow_format flow_format; - error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg); - if (error) { - return error; - } + flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10; + ofconn_set_flow_format(ofconn, flow_format); - p->tun_id_from_cookie = !!msg->set; return 0; } static int -handle_role_request(struct ofproto *ofproto, - struct ofconn *ofconn, struct nicira_header *msg) +handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh) { - struct nx_role_request *nrr; + struct nx_role_request *nrr = (struct nx_role_request *) oh; struct nx_role_request *reply; struct ofpbuf *buf; uint32_t role; - if (ntohs(msg->header.length) != sizeof *nrr) { - VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)", - ntohs(msg->header.length), sizeof *nrr); - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); - } - nrr = (struct nx_role_request *) msg; - - if (ofconn->type != OFCONN_PRIMARY) { - VLOG_WARN_RL(&rl, "ignoring role request on non-controller " - "connection"); + 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); } @@ -3945,68 +3565,35 @@ handle_role_request(struct ofproto *ofproto, return ofp_mkerr(OFPET_BAD_REQUEST, -1); } - if (role == NX_ROLE_MASTER) { - struct ofconn *other; + ofconn_set_role(ofconn, role); - HMAP_FOR_EACH (other, struct ofconn, hmap_node, - &ofproto->controllers) { - if (other->role == NX_ROLE_MASTER) { - other->role = NX_ROLE_SLAVE; - } - } - } - ofconn->role = role; - - reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid, - &buf); - reply->nxh.vendor = htonl(NX_VENDOR_ID); - reply->nxh.subtype = htonl(NXT_ROLE_REPLY); + reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf); reply->role = htonl(role); - queue_tx(buf, ofconn, ofconn->reply_counter); + ofconn_send_reply(ofconn, buf); return 0; } static int -handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg) +handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh) { - struct ofp_vendor_header *ovh = msg; - struct nicira_header *nh; - - if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) { - VLOG_WARN_RL(&rl, "received vendor message of length %u " - "(expected at least %zu)", - ntohs(ovh->header.length), sizeof(struct ofp_vendor_header)); - 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)) { - VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u " - "(expected at least %zu)", - ntohs(ovh->header.length), sizeof(struct nicira_header)); - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); - } - - nh = msg; - switch (ntohl(nh->subtype)) { - case NXT_STATUS_REQUEST: - return switch_status_handle_request(p->switch_status, ofconn->rconn, - msg); + const struct nxt_set_flow_format *msg + = (const struct nxt_set_flow_format *) oh; + uint32_t format; - case NXT_TUN_ID_FROM_COOKIE: - return handle_tun_id_from_cookie(p, msg); - - case NXT_ROLE_REQUEST: - return handle_role_request(p, ofconn, msg); + 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); } - - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); } static int -handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh) +handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh) { struct ofp_header *ob; struct ofpbuf *buf; @@ -4014,136 +3601,195 @@ handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh) /* 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); + ofconn_send_reply(ofconn, buf); return 0; } -static void -handle_openflow(struct ofconn *ofconn, struct ofproto *p, - struct ofpbuf *ofp_msg) +static int +handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) { - struct ofp_header *oh = ofp_msg->data; + const struct ofp_header *oh = msg->data; + const struct ofputil_msg_type *type; int error; - COVERAGE_INC(ofproto_recv_openflow); - switch (oh->type) { - case OFPT_ECHO_REQUEST: - error = handle_echo_request(ofconn, oh); - break; - - case OFPT_ECHO_REPLY: - error = 0; - break; + error = ofputil_decode_msg_type(oh, &type); + if (error) { + return error; + } - case OFPT_FEATURES_REQUEST: - error = handle_features_request(p, ofconn, oh); - break; + switch (ofputil_msg_type_code(type)) { + /* OpenFlow requests. */ + case OFPUTIL_OFPT_ECHO_REQUEST: + return handle_echo_request(ofconn, oh); - case OFPT_GET_CONFIG_REQUEST: - error = handle_get_config_request(p, ofconn, oh); - break; + case OFPUTIL_OFPT_FEATURES_REQUEST: + return handle_features_request(ofconn, oh); - case OFPT_SET_CONFIG: - error = handle_set_config(p, ofconn, ofp_msg->data); - break; + case OFPUTIL_OFPT_GET_CONFIG_REQUEST: + return handle_get_config_request(ofconn, oh); - case OFPT_PACKET_OUT: - error = handle_packet_out(p, ofconn, ofp_msg->data); - break; + case OFPUTIL_OFPT_SET_CONFIG: + return handle_set_config(ofconn, msg->data); - case OFPT_PORT_MOD: - error = handle_port_mod(p, ofconn, oh); - break; + case OFPUTIL_OFPT_PACKET_OUT: + return handle_packet_out(ofconn, oh); - case OFPT_FLOW_MOD: - error = handle_flow_mod(p, ofconn, ofp_msg->data); - break; + case OFPUTIL_OFPT_PORT_MOD: + return handle_port_mod(ofconn, oh); - case OFPT_STATS_REQUEST: - error = handle_stats_request(p, ofconn, oh); - break; + case OFPUTIL_OFPT_FLOW_MOD: + return handle_flow_mod(ofconn, oh); - case OFPT_VENDOR: - error = handle_vendor(p, ofconn, ofp_msg->data); - break; + case OFPUTIL_OFPT_BARRIER_REQUEST: + return handle_barrier_request(ofconn, oh); - case OFPT_BARRIER_REQUEST: - error = handle_barrier_request(ofconn, oh); - break; + /* OpenFlow replies. */ + case OFPUTIL_OFPT_ECHO_REPLY: + 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_SET_FLOW_FORMAT: + return handle_nxt_set_flow_format(ofconn, oh); + + case OFPUTIL_NXT_FLOW_MOD: + return handle_flow_mod(ofconn, oh); + + /* OpenFlow statistics requests. */ + case OFPUTIL_OFPST_DESC_REQUEST: + return handle_desc_stats_request(ofconn, oh); + + case OFPUTIL_OFPST_FLOW_REQUEST: + return handle_flow_stats_request(ofconn, oh); + + case OFPUTIL_OFPST_AGGREGATE_REQUEST: + return handle_aggregate_stats_request(ofconn, oh); + + case OFPUTIL_OFPST_TABLE_REQUEST: + return handle_table_stats_request(ofconn, oh); + + case OFPUTIL_OFPST_PORT_REQUEST: + return handle_port_stats_request(ofconn, oh); + + 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); + + case OFPUTIL_INVALID: + case OFPUTIL_OFPT_HELLO: + case OFPUTIL_OFPT_ERROR: + case OFPUTIL_OFPT_FEATURES_REPLY: + case OFPUTIL_OFPT_GET_CONFIG_REPLY: + case OFPUTIL_OFPT_PACKET_IN: + case OFPUTIL_OFPT_FLOW_REMOVED: + case OFPUTIL_OFPT_PORT_STATUS: + case OFPUTIL_OFPT_BARRIER_REPLY: + case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST: + case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY: + case OFPUTIL_OFPST_DESC_REPLY: + case OFPUTIL_OFPST_FLOW_REPLY: + case OFPUTIL_OFPST_QUEUE_REPLY: + case OFPUTIL_OFPST_PORT_REPLY: + case OFPUTIL_OFPST_TABLE_REPLY: + case OFPUTIL_OFPST_AGGREGATE_REPLY: + case OFPUTIL_NXT_ROLE_REPLY: + case OFPUTIL_NXT_FLOW_REMOVED: + 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); } - error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE); - break; + if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT); + } else { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE); + } } +} +static void +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); } + COVERAGE_INC(ofproto_recv_openflow); } static void -handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) +handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall) { - struct odp_msg *msg = packet->data; - struct rule *rule; - struct ofpbuf payload; - flow_t flow; + struct facet *facet; + struct flow flow; - payload.data = msg + 1; - payload.size = msg->length - sizeof *msg; - flow_extract(&payload, msg->arg, msg->port, &flow); + /* Obtain in_port and tun_id, at least. */ + odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); - /* Check with in-band control to see if this packet should be sent - * to the local port regardless of the flow table. */ - if (in_band_msg_in_hook(p->in_band, &flow, &payload)) { - union odp_action action; + /* Set header pointers in 'flow'. */ + flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow); - memset(&action, 0, sizeof(action)); - action.output.type = ODPAT_OUTPUT; - action.output.port = ODPP_LOCAL; - dpif_execute(p->dpif, flow.in_port, &action, 1, &payload); + if (cfm_should_process_flow(&flow)) { + ofproto_process_cfm(p, &flow, upcall->packet); + ofpbuf_delete(upcall->packet); + return; + } else if (p->ofhooks->special_cb + && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) { + ofpbuf_delete(upcall->packet); + return; } - rule = lookup_valid_rule(p, &flow); - if (!rule) { - /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ - struct ofport *port = port_array_get(&p->ports, msg->port); - if (port) { - if (port->opp.config & OFPPC_NO_PACKET_IN) { - COVERAGE_INC(ofproto_no_packet_in); - /* XXX install 'drop' flow entry */ - ofpbuf_delete(packet); - return; + /* Check with in-band control to see if this packet should be sent + * to the local port regardless of the flow table. */ + if (connmgr_msg_in_hook(p->connmgr, &flow, upcall->packet)) { + ofproto_send_packet(p, ODPP_LOCAL, 0, upcall->packet); + } + + facet = facet_lookup_valid(p, &flow); + if (!facet) { + struct rule *rule = rule_lookup(p, &flow); + if (!rule) { + /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ + struct ofport *port = get_port(p, flow.in_port); + if (port) { + if (port->opp.config & OFPPC_NO_PACKET_IN) { + COVERAGE_INC(ofproto_no_packet_in); + /* XXX install 'drop' flow entry */ + ofpbuf_delete(upcall->packet); + return; + } + } else { + VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, + flow.in_port); } - } else { - VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port); - } - - COVERAGE_INC(ofproto_packet_in); - send_packet_in(p, packet); - return; - } - if (rule->cr.wc.wildcards) { - rule = rule_create_subrule(p, rule, &flow); - rule_make_actions(p, rule, packet); - } else { - if (!rule->may_install) { - /* The rule is not installable, that is, we need to process every - * packet, so process the current packet and set its actions into - * 'subrule'. */ - rule_make_actions(p, rule, packet); - } else { - /* XXX revalidate rule if it needs it */ + COVERAGE_INC(ofproto_packet_in); + send_packet_in(p, upcall, &flow, false); + return; } + + facet = facet_create(p, rule, &flow, upcall->packet); + } else if (!facet->may_install) { + /* The facet is not installable, that is, we need to process every + * packet, so process the current packet's actions into 'facet'. */ + facet_make_actions(p, facet, upcall->packet); } - if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) { + if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) { /* * Extra-special case for fail-open mode. * @@ -4154,413 +3800,382 @@ handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) * * See the top-level comment in fail-open.c for more information. */ - send_packet_in(p, ofpbuf_clone(packet)); + send_packet_in(p, upcall, &flow, true); } - ofpbuf_pull(packet, sizeof *msg); - rule_execute(p, rule, packet, &flow); - rule_reinstall(p, rule); + facet_execute(p, facet, upcall->packet); + facet_install(p, facet, false); } static void -handle_odp_msg(struct ofproto *p, struct ofpbuf *packet) +handle_upcall(struct ofproto *p, struct dpif_upcall *upcall) { - struct odp_msg *msg = packet->data; + struct flow flow; - switch (msg->type) { - case _ODPL_ACTION_NR: + switch (upcall->type) { + case DPIF_UC_ACTION: COVERAGE_INC(ofproto_ctlr_action); - send_packet_in(p, packet); + odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); + send_packet_in(p, upcall, &flow, false); break; - case _ODPL_SFLOW_NR: + case DPIF_UC_SAMPLE: if (p->sflow) { - ofproto_sflow_received(p->sflow, msg); + odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); + ofproto_sflow_received(p->sflow, upcall, &flow); } - ofpbuf_delete(packet); + ofpbuf_delete(upcall->packet); break; - case _ODPL_MISS_NR: - handle_odp_miss_msg(p, packet); + case DPIF_UC_MISS: + handle_miss_upcall(p, upcall); break; + case DPIF_N_UC_TYPES: default: - VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32, - msg->type); + VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); break; } } -static void -revalidate_cb(struct cls_rule *sub_, void *cbdata_) +/* Flow expiration. */ + +static int ofproto_dp_max_idle(const struct ofproto *); +static void ofproto_update_stats(struct ofproto *); +static void rule_expire(struct ofproto *, struct rule *); +static void ofproto_expire_facets(struct ofproto *, int dp_max_idle); + +/* This function is called periodically by ofproto_run(). Its job is to + * collect updates for the flows that have been installed into the datapath, + * most importantly when they last were used, and then use that information to + * expire flows that have not been used recently. + * + * Returns the number of milliseconds after which it should be called again. */ +static int +ofproto_expire(struct ofproto *ofproto) { - struct rule *sub = rule_from_cls_rule(sub_); - struct revalidate_cbdata *cbdata = cbdata_; + struct rule *rule, *next_rule; + struct cls_cursor cursor; + int dp_max_idle; + + /* Update stats for each flow in the datapath. */ + ofproto_update_stats(ofproto); - if (cbdata->revalidate_all - || (cbdata->revalidate_subrules && sub->super) - || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) { - revalidate_rule(cbdata->ofproto, sub); + /* Expire facets that have been idle too long. */ + dp_max_idle = ofproto_dp_max_idle(ofproto); + ofproto_expire_facets(ofproto, dp_max_idle); + + /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ + cls_cursor_init(&cursor, &ofproto->cls, NULL); + CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { + rule_expire(ofproto, rule); } -} -static bool -revalidate_rule(struct ofproto *p, struct rule *rule) -{ - const flow_t *flow = &rule->cr.flow; - - COVERAGE_INC(ofproto_revalidate_rule); - if (rule->super) { - struct rule *super; - super = rule_from_cls_rule(classifier_lookup_wild(&p->cls, flow)); - if (!super) { - rule_remove(p, rule); - return false; - } else if (super != rule->super) { - COVERAGE_INC(ofproto_revalidate_moved); - list_remove(&rule->list); - list_push_back(&super->list, &rule->list); - rule->super = super; - rule->hard_timeout = super->hard_timeout; - rule->idle_timeout = super->idle_timeout; - rule->created = super->created; - rule->used = 0; - } + /* Let the hook know that we're at a stable point: all outstanding data + * in existing flows has been accounted to the account_cb. Thus, the + * hook can now reasonably do operations that depend on having accurate + * flow volume accounting (currently, that's just bond rebalancing). */ + if (ofproto->ofhooks->account_checkpoint_cb) { + ofproto->ofhooks->account_checkpoint_cb(ofproto->aux); } - rule_update_actions(p, rule); - return true; + return MIN(dp_max_idle, 1000); } -static struct ofpbuf * -compose_flow_removed(struct ofproto *p, const struct rule *rule, - long long int now, uint8_t reason) +/* Update 'packet_count', 'byte_count', and 'used' members of installed facets. + * + * This function also pushes statistics updates to rules which each facet + * resubmits into. Generally these statistics will be accurate. However, if a + * facet changes the rule it resubmits into at some time in between + * ofproto_update_stats() runs, it is possible that statistics accrued to the + * old rule will be incorrectly attributed to the new rule. This could be + * avoided by calling ofproto_update_stats() whenever rules are created or + * deleted. However, the performance impact of making so many calls to the + * datapath do not justify the benefit of having perfectly accurate statistics. + */ +static void +ofproto_update_stats(struct ofproto *p) { - struct ofp_flow_removed *ofr; - struct ofpbuf *buf; - long long int tdiff = now - rule->created; - uint32_t sec = tdiff / 1000; - uint32_t msec = tdiff - (sec * 1000); + const struct dpif_flow_stats *stats; + struct dpif_flow_dump dump; + const struct nlattr *key; + size_t key_len; - ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf); - flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie, - &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); + dpif_flow_dump_start(&dump, p->dpif); + while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { + struct facet *facet; + struct flow flow; - return buf; -} + if (odp_flow_key_to_flow(key, key_len, &flow)) { + struct ds s; -static void -uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule) -{ - assert(rule->installed); - assert(!rule->cr.wc.wildcards); + ds_init(&s); + odp_flow_key_format(key, key_len, &s); + VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s", + ds_cstr(&s)); + ds_destroy(&s); - if (rule->super) { - rule_remove(ofproto, rule); - } else { - rule_uninstall(ofproto, rule); - } -} + continue; + } + facet = facet_find(p, &flow); -static void -send_flow_removed(struct ofproto *p, struct rule *rule, - long long int now, uint8_t reason) -{ - struct ofconn *ofconn; - struct ofconn *prev; - struct ofpbuf *buf = NULL; - - /* We limit the maximum number of queued flow expirations it by accounting - * them under the counter for replies. That works because preventing - * OpenFlow requests from being processed also prevents new flows from - * being added (and expiring). (It also prevents processing OpenFlow - * requests that would not add new flows, so it is imperfect.) */ - - prev = NULL; - LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { - if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn) - && ofconn_receives_async_msgs(ofconn)) { - if (prev) { - queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter); + if (facet && facet->installed) { + + if (stats->n_packets >= facet->dp_packet_count) { + facet->packet_count += stats->n_packets - facet->dp_packet_count; + } else { + VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); + } + + if (stats->n_bytes >= facet->dp_byte_count) { + facet->byte_count += stats->n_bytes - facet->dp_byte_count; } else { - buf = compose_flow_removed(p, rule, now, reason); + VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); } - prev = ofconn; + + facet->dp_packet_count = stats->n_packets; + facet->dp_byte_count = stats->n_bytes; + + facet_update_time(p, facet, stats->used); + facet_account(p, facet, stats->n_bytes); + facet_push_stats(p, facet); + } else { + /* There's a flow in the datapath that we know nothing about. + * Delete it. */ + COVERAGE_INC(ofproto_unexpected_rule); + dpif_flow_del(p->dpif, key, key_len, NULL); } } - if (prev) { - queue_tx(buf, prev, prev->reply_counter); - } + dpif_flow_dump_done(&dump); } - -static void -expire_rule(struct cls_rule *cls_rule, void *p_) +/* Calculates and returns the number of milliseconds of idle time after which + * facets should expire from the datapath and we should fold their statistics + * into their parent rules in userspace. */ +static int +ofproto_dp_max_idle(const struct ofproto *ofproto) { - struct ofproto *p = p_; - struct rule *rule = rule_from_cls_rule(cls_rule); - long long int hard_expire, idle_expire, expire, now; + /* + * Idle time histogram. + * + * Most of the time a switch has a relatively small number of facets. When + * this is the case we might as well keep statistics for all of them in + * userspace and to cache them in the kernel datapath for performance as + * well. + * + * As the number of facets increases, the memory required to maintain + * statistics about them in userspace and in the kernel becomes + * significant. However, with a large number of facets it is likely that + * only a few of them are "heavy hitters" that consume a large amount of + * bandwidth. At this point, only heavy hitters are worth caching in the + * kernel and maintaining in userspaces; other facets we can discard. + * + * The technique used to compute the idle time is to build a histogram with + * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet + * that is installed in the kernel gets dropped in the appropriate bucket. + * After the histogram has been built, we compute the cutoff so that only + * the most-recently-used 1% of facets (but at least 1000 flows) are kept + * cached. At least the most-recently-used bucket of facets is kept, so + * actually an arbitrary number of facets can be kept in any given + * expiration run (though the next run will delete most of those unless + * they receive additional data). + * + * This requires a second pass through the facets, in addition to the pass + * made by ofproto_update_stats(), because the former function never looks + * at uninstallable facets. + */ + enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; + enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; + int buckets[N_BUCKETS] = { 0 }; + struct facet *facet; + int total, bucket; + long long int now; + int i; - hard_expire = (rule->hard_timeout - ? rule->created + rule->hard_timeout * 1000 - : LLONG_MAX); - idle_expire = (rule->idle_timeout - && (rule->super || list_is_empty(&rule->list)) - ? rule->used + rule->idle_timeout * 1000 - : LLONG_MAX); - expire = MIN(hard_expire, idle_expire); + total = hmap_count(&ofproto->facets); + if (total <= 1000) { + return N_BUCKETS * BUCKET_WIDTH; + } + /* Build histogram. */ 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); + HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { + long long int idle = now - facet->used; + int bucket = (idle <= 0 ? 0 + : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 + : (unsigned int) idle / BUCKET_WIDTH); + buckets[bucket]++; + } + + /* Find the first bucket whose flows should be expired. */ + for (bucket = 0; bucket < N_BUCKETS; bucket++) { + if (buckets[bucket]) { + int subtotal = 0; + do { + subtotal += buckets[bucket++]; + } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100)); + break; } - - return; } - COVERAGE_INC(ofproto_expired); + if (VLOG_IS_DBG_ENABLED()) { + struct ds s; - /* Update stats. This code will be a no-op if the rule expired - * due to an idle timeout. */ - if (rule->cr.wc.wildcards) { - struct rule *subrule, *next; - LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) { - rule_remove(p, subrule); + ds_init(&s); + ds_put_cstr(&s, "keep"); + for (i = 0; i < N_BUCKETS; i++) { + if (i == bucket) { + ds_put_cstr(&s, ", drop"); + } + if (buckets[i]) { + ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); + } } - } else { - rule_uninstall(p, rule); + VLOG_INFO("%s: %s (msec:count)", + dpif_name(ofproto->dpif), ds_cstr(&s)); + ds_destroy(&s); } - if (!rule_is_hidden(rule)) { - send_flow_removed(p, rule, now, - (now >= hard_expire - ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT)); - } - rule_remove(p, rule); + return bucket * BUCKET_WIDTH; } static void -active_timeout(struct ofproto *ofproto, struct rule *rule) +facet_active_timeout(struct ofproto *ofproto, struct facet *facet) { - if (ofproto->netflow && !is_controller_rule(rule) && - netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) { + if (ofproto->netflow && !facet_is_controller_flow(facet) && + netflow_active_timeout_expired(ofproto->netflow, &facet->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.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; + if (facet->installed) { + struct dpif_flow_stats stats; - netflow_expire(ofproto->netflow, &rule->nf_flow, &expired); + facet_put__(ofproto, facet, facet->actions, facet->actions_len, + &stats); + facet_update_stats(ofproto, facet, &stats); + } - /* Schedule us to send the accumulated records once we have - * collected all of them. */ - poll_immediate_wake(); + expired.flow = facet->flow; + expired.packet_count = facet->packet_count; + expired.byte_count = facet->byte_count; + expired.used = facet->used; + netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); } } static void -update_used(struct ofproto *p) +ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle) { - struct odp_flow *flows; - size_t n_flows; - size_t i; - int error; - - error = dpif_flow_list_all(p->dpif, &flows, &n_flows); - if (error) { - return; - } - - for (i = 0; i < n_flows; i++) { - struct odp_flow *f = &flows[i]; - struct rule *rule; + long long int cutoff = time_msec() - dp_max_idle; + struct facet *facet, *next_facet; - rule = rule_from_cls_rule( - classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX)); - if (!rule || !rule->installed) { - COVERAGE_INC(ofproto_unexpected_rule); - dpif_flow_del(p->dpif, f); - continue; + HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { + facet_active_timeout(ofproto, facet); + if (facet->used < cutoff) { + facet_remove(ofproto, facet); } - - update_time(p, rule, &f->stats); - rule_account(p, rule, f->stats.n_bytes); } - free(flows); } -/* pinsched callback for sending 'packet' on 'ofconn'. */ +/* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, + * then delete it entirely. */ static void -do_send_packet_in(struct ofpbuf *packet, void *ofconn_) +rule_expire(struct ofproto *ofproto, struct rule *rule) { - struct ofconn *ofconn = ofconn_; - - rconn_send_with_limit(ofconn->rconn, packet, - ofconn->packet_in_counter, 100); -} + struct facet *facet, *next_facet; + long long int now; + uint8_t reason; -/* Takes 'packet', which has been converted with do_convert_to_packet_in(), and - * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s - * packet scheduler for sending. - * - * 'max_len' specifies the maximum number of bytes of the packet to send on - * 'ofconn' (INT_MAX specifies no limit). - * - * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise, - * ownership is transferred to this function. */ -static void -schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len, - bool clone) -{ - struct ofproto *ofproto = ofconn->ofproto; - struct ofp_packet_in *opi = packet->data; - uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port)); - int send_len, trim_size; - uint32_t buffer_id; - - /* Get buffer. */ - if (opi->reason == OFPR_ACTION) { - buffer_id = UINT32_MAX; - } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) { - buffer_id = pktbuf_get_null(); - } else if (!ofconn->pktbuf) { - buffer_id = UINT32_MAX; + /* Has 'rule' expired? */ + now = time_msec(); + if (rule->hard_timeout + && now > rule->created + rule->hard_timeout * 1000) { + reason = OFPRR_HARD_TIMEOUT; + } else if (rule->idle_timeout && list_is_empty(&rule->facets) + && now >rule->used + rule->idle_timeout * 1000) { + reason = OFPRR_IDLE_TIMEOUT; } else { - struct ofpbuf payload; - payload.data = opi->data; - payload.size = packet->size - offsetof(struct ofp_packet_in, data); - buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port); + return; } - /* Figure out how much of the packet to send. */ - send_len = ntohs(opi->total_len); - if (buffer_id != UINT32_MAX) { - send_len = MIN(send_len, ofconn->miss_send_len); + COVERAGE_INC(ofproto_expired); + + /* Update stats. (This is a no-op if the rule expired due to an idle + * timeout, because that only happens when the rule has no facets left.) */ + LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { + facet_remove(ofproto, facet); } - send_len = MIN(send_len, max_len); - /* Adjust packet length and clone if necessary. */ - trim_size = offsetof(struct ofp_packet_in, data) + send_len; - if (clone) { - packet = ofpbuf_clone_data(packet->data, trim_size); - opi = packet->data; - } else { - packet->size = trim_size; + /* Get rid of the rule. */ + if (!rule_is_hidden(rule)) { + rule_send_removed(ofproto, rule, reason); + } + rule_remove(ofproto, rule); +} + +static void +rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason) +{ + struct ofputil_flow_removed fr; + + if (!rule->send_flow_removed) { + return; } - /* Update packet headers. */ - opi->buffer_id = htonl(buffer_id); - update_openflow_length(packet); + fr.rule = rule->cr; + fr.cookie = rule->flow_cookie; + fr.reason = reason; + calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec); + fr.idle_timeout = rule->idle_timeout; + fr.packet_count = rule->packet_count; + fr.byte_count = rule->byte_count; - /* Hand over to packet scheduler. It might immediately call into - * do_send_packet_in() or it might buffer it for a while (until a later - * call to pinsched_run()). */ - pinsched_send(ofconn->schedulers[opi->reason], in_port, - packet, do_send_packet_in, ofconn); + connmgr_send_flow_removed(p->connmgr, &fr); } -/* Replace struct odp_msg header in 'packet' by equivalent struct - * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as - * returned by dpif_recv()). - * - * The conversion is not complete: the caller still needs to trim any unneeded - * payload off the end of the buffer, set the length in the OpenFlow header, - * and set buffer_id. Those require us to know the controller settings and so - * must be done on a per-controller basis. - * - * Returns the maximum number of bytes of the packet that should be sent to - * the controller (INT_MAX if no limit). */ -static int -do_convert_to_packet_in(struct ofpbuf *packet) +/* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'. + * The returned statistics include statistics for all of 'rule''s facets. */ +static void +rule_get_stats(const struct rule *rule, uint64_t *packets, uint64_t *bytes) { - struct odp_msg *msg = packet->data; - struct ofp_packet_in *opi; - uint8_t reason; - uint16_t total_len; - uint16_t in_port; - int max_len; + uint64_t p, b; + struct facet *facet; - /* Extract relevant header fields */ - if (msg->type == _ODPL_ACTION_NR) { - reason = OFPR_ACTION; - max_len = msg->arg; - } else { - reason = OFPR_NO_MATCH; - max_len = INT_MAX; - } - total_len = msg->length - sizeof *msg; - in_port = odp_port_to_ofp_port(msg->port); + /* Start from historical data for 'rule' itself that are no longer tracked + * in facets. This counts, for example, facets that have expired. */ + p = rule->packet_count; + b = rule->byte_count; - /* Repurpose packet buffer by overwriting header. */ - ofpbuf_pull(packet, sizeof(struct odp_msg)); - opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data)); - opi->header.version = OFP_VERSION; - opi->header.type = OFPT_PACKET_IN; - opi->total_len = htons(total_len); - opi->in_port = htons(in_port); - opi->reason = reason; + /* Add any statistics that are tracked by facets. This includes + * statistical data recently updated by ofproto_update_stats() as well as + * stats for packets that were executed "by hand" via dpif_execute(). */ + LIST_FOR_EACH (facet, list_node, &rule->facets) { + p += facet->packet_count; + b += facet->byte_count; + } - return max_len; + *packets = p; + *bytes = b; } -/* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or - * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller - * as necessary according to their individual configurations. +/* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an + * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to + * their individual configurations. * - * 'packet' must have sufficient headroom to convert it into a struct - * ofp_packet_in (e.g. as returned by dpif_recv()). - * - * Takes ownership of 'packet'. */ + * If 'clone' is true, the caller retains ownership of 'upcall->packet'. + * Otherwise, ownership is transferred to this function. */ static void -send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet) +send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall, + const struct flow *flow, bool clone) { - struct ofconn *ofconn, *prev; - int max_len; + struct ofputil_packet_in pin; - max_len = do_convert_to_packet_in(packet); - - prev = NULL; - LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) { - if (ofconn_receives_async_msgs(ofconn)) { - if (prev) { - schedule_packet_in(prev, packet, max_len, true); - } - prev = ofconn; - } - } - if (prev) { - schedule_packet_in(prev, packet, max_len, false); - } else { - ofpbuf_delete(packet); - } + pin.packet = upcall->packet; + pin.in_port = odp_port_to_ofp_port(flow->in_port); + pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION; + pin.buffer_id = 0; /* not yet known */ + pin.send_len = upcall->userdata; + connmgr_send_packet_in(ofproto->connmgr, upcall, flow, + clone ? NULL : upcall->packet); } static uint64_t @@ -4568,7 +4183,7 @@ pick_datapath_id(const struct ofproto *ofproto) { const struct ofport *port; - port = port_array_get(&ofproto->ports, ODPP_LOCAL); + port = get_port(ofproto, ODPP_LOCAL); if (port) { uint8_t ea[ETH_ADDR_LEN]; int error; @@ -4591,13 +4206,181 @@ pick_fallback_dpid(void) return eth_addr_to_uint64(ea); } +static void +ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED, + void *aux OVS_UNUSED) +{ + const struct shash_node *node; + struct ds results; + + ds_init(&results); + SHASH_FOR_EACH (node, &all_ofprotos) { + ds_put_format(&results, "%s\n", node->name); + } + unixctl_command_reply(conn, 200, ds_cstr(&results)); + ds_destroy(&results); +} + +struct ofproto_trace { + struct action_xlate_ctx ctx; + struct flow flow; + struct ds *result; +}; + +static void +trace_format_rule(struct ds *result, int level, const struct rule *rule) +{ + ds_put_char_multiple(result, '\t', level); + if (!rule) { + ds_put_cstr(result, "No match\n"); + return; + } + + ds_put_format(result, "Rule: cookie=%#"PRIx64" ", + ntohll(rule->flow_cookie)); + cls_rule_format(&rule->cr, result); + ds_put_char(result, '\n'); + + ds_put_char_multiple(result, '\t', level); + ds_put_cstr(result, "OpenFlow "); + ofp_print_actions(result, (const struct ofp_action_header *) rule->actions, + rule->n_actions * sizeof *rule->actions); + ds_put_char(result, '\n'); +} + +static void +trace_format_flow(struct ds *result, int level, const char *title, + struct ofproto_trace *trace) +{ + ds_put_char_multiple(result, '\t', level); + ds_put_format(result, "%s: ", title); + if (flow_equal(&trace->ctx.flow, &trace->flow)) { + ds_put_cstr(result, "unchanged"); + } else { + flow_format(result, &trace->ctx.flow); + trace->flow = trace->ctx.flow; + } + ds_put_char(result, '\n'); +} + +static void +trace_resubmit(struct action_xlate_ctx *ctx, struct rule *rule) +{ + struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx); + struct ds *result = trace->result; + + ds_put_char(result, '\n'); + trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); + trace_format_rule(result, ctx->recurse + 1, rule); +} + +static void +ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_, + void *aux OVS_UNUSED) +{ + char *dpname, *in_port_s, *tun_id_s, *packet_s; + char *args = xstrdup(args_); + char *save_ptr = NULL; + struct ofproto *ofproto; + struct ofpbuf packet; + struct rule *rule; + struct ds result; + struct flow flow; + uint16_t in_port; + ovs_be64 tun_id; + char *s; + + ofpbuf_init(&packet, strlen(args) / 2); + ds_init(&result); + + dpname = strtok_r(args, " ", &save_ptr); + tun_id_s = strtok_r(NULL, " ", &save_ptr); + in_port_s = strtok_r(NULL, " ", &save_ptr); + packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */ + if (!dpname || !in_port_s || !packet_s) { + unixctl_command_reply(conn, 501, "Bad command syntax"); + goto exit; + } + + ofproto = shash_find_data(&all_ofprotos, dpname); + if (!ofproto) { + unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list " + "for help)"); + goto exit; + } + + tun_id = htonll(strtoull(tun_id_s, NULL, 0)); + in_port = ofp_port_to_odp_port(atoi(in_port_s)); + + packet_s = ofpbuf_put_hex(&packet, packet_s, NULL); + packet_s += strspn(packet_s, " "); + if (*packet_s != '\0') { + unixctl_command_reply(conn, 501, "Trailing garbage in command"); + goto exit; + } + if (packet.size < ETH_HEADER_LEN) { + unixctl_command_reply(conn, 501, "Packet data too short for Ethernet"); + goto exit; + } + + ds_put_cstr(&result, "Packet: "); + s = ofp_packet_to_string(packet.data, packet.size, packet.size); + ds_put_cstr(&result, s); + free(s); + + flow_extract(&packet, tun_id, in_port, &flow); + ds_put_cstr(&result, "Flow: "); + flow_format(&result, &flow); + ds_put_char(&result, '\n'); + + rule = rule_lookup(ofproto, &flow); + trace_format_rule(&result, 0, rule); + if (rule) { + struct ofproto_trace trace; + struct ofpbuf *odp_actions; + + trace.result = &result; + trace.flow = flow; + action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet); + trace.ctx.resubmit_hook = trace_resubmit; + odp_actions = xlate_actions(&trace.ctx, + rule->actions, rule->n_actions); + + ds_put_char(&result, '\n'); + trace_format_flow(&result, 0, "Final flow", &trace); + ds_put_cstr(&result, "Datapath actions: "); + format_odp_actions(&result, odp_actions->data, odp_actions->size); + ofpbuf_delete(odp_actions); + } + + unixctl_command_reply(conn, 200, ds_cstr(&result)); + +exit: + ds_destroy(&result); + ofpbuf_uninit(&packet); + free(args); +} + +static void +ofproto_unixctl_init(void) +{ + static bool registered; + if (registered) { + return; + } + registered = true; + + unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL); + unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL); +} + static bool -default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, - struct odp_actions *actions, tag_type *tags, +default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, + struct ofpbuf *odp_actions, tag_type *tags, uint16_t *nf_output_iface, void *ofproto_) { struct ofproto *ofproto = ofproto_; - int out_port; + struct mac_entry *dst_mac; /* Drop frames for reserved multicast addresses. */ if (eth_addr_is_reserved(flow->dl_dst)) { @@ -4605,38 +4388,46 @@ default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, } /* Learn source MAC (but don't try to learn from revalidation). */ - if (packet != NULL) { - tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src, - 0, flow->in_port, - GRAT_ARP_LOCK_NONE); - if (rev_tag) { + if (packet != NULL + && mac_learning_may_learn(ofproto->ml, flow->dl_src, 0)) { + struct mac_entry *src_mac; + + src_mac = mac_learning_insert(ofproto->ml, flow->dl_src, 0); + if (mac_entry_is_new(src_mac) || src_mac->port.i != flow->in_port) { /* The log messages here could actually be useful in debugging, * so keep the rate limit relatively high. */ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16, ETH_ADDR_ARGS(flow->dl_src), flow->in_port); - ofproto_revalidate(ofproto, rev_tag); + + ofproto_revalidate(ofproto, + mac_learning_changed(ofproto->ml, src_mac)); + src_mac->port.i = flow->in_port; } } /* Determine output port. */ - out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags, - NULL); - if (out_port < 0) { - 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; + dst_mac = mac_learning_lookup(ofproto->ml, flow->dl_dst, 0, tags); + if (!dst_mac) { + flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD, + nf_output_iface, odp_actions); } else { - /* Drop. */ + int out_port = dst_mac->port.i; + if (out_port != flow->in_port) { + nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port); + *nf_output_iface = out_port; + } else { + /* Drop. */ + } } return true; } static const struct ofhooks default_ofhooks = { - NULL, default_normal_ofhook_cb, NULL, + NULL, + NULL, NULL };