X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=ofproto%2Fofproto.c;h=52a1964bfdfe3d1ffd95ed9ad8d68e07e0179da2;hb=db5ce51427906ed54aa3e8e1dffc0ba6d908aa8c;hp=3ab9cc9992c1162a836f1702caa67a51e5c91698;hpb=fe55ad159d8fd396a9e4914a03eea93d096d03b1;p=openvswitch diff --git a/ofproto/ofproto.c b/ofproto/ofproto.c index 3ab9cc99..52a1964b 100644 --- a/ofproto/ofproto.c +++ b/ofproto/ofproto.c @@ -1,5 +1,6 @@ /* - * 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"); * you may not use this file except in compliance with the License. @@ -18,207 +19,376 @@ #include "ofproto.h" #include #include +#include #include #include #include #include +#include "byte-order.h" #include "classifier.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" +#include "ofproto-sflow.h" #include "ofpbuf.h" #include "openflow/nicira-ext.h" #include "openflow/openflow.h" -#include "openflow/openflow-mgmt.h" #include "openvswitch/datapath-protocol.h" #include "packets.h" #include "pinsched.h" #include "pktbuf.h" #include "poll-loop.h" -#include "port-array.h" #include "rconn.h" #include "shash.h" #include "status.h" -#include "stp.h" #include "stream-ssl.h" #include "svec.h" #include "tag.h" #include "timeval.h" #include "unixctl.h" #include "vconn.h" -#include "xtoxll.h" - -#define THIS_MODULE VLM_ofproto #include "vlog.h" -enum { - DP_GROUP_FLOOD = 0, - DP_GROUP_ALL = 1 -}; - -enum { - TABLEID_HASH = 0, - TABLEID_CLASSIFIER = 1 -}; +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_ofconn_stuck); +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" + +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; }; static void ofport_free(struct ofport *); static void hton_ofp_phy_port(struct ofp_phy_port *); -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; + /* The ofproto. */ + struct ofproto *ofproto; - uint16_t idle_timeout; /* In seconds from time of last use. */ - uint16_t hard_timeout; /* In seconds from time of creation. */ - 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. */ + /* Flow to which the OpenFlow actions apply. xlate_actions() will modify + * this flow when actions change header fields. */ + struct flow flow; - /* 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; + /* The packet corresponding to 'flow', or a null pointer if we are + * revalidating without a packet to refer to. */ + const struct ofpbuf *packet; - /* OpenFlow actions. + /* If nonnull, called just before executing a resubmit action. * - * A subrule has no actions (it uses the super-rule's actions). */ - int n_actions; - union ofp_action *actions; + * 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 *, const struct rule *); - /* Datapath actions. - * - * 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 +/* 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 + * recently added ODPAT_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_create(struct ofproto *, struct rule *super, +static struct rule *rule_from_cls_rule(const struct cls_rule *); +static bool rule_is_hidden(const struct rule *); + +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); -static void rule_free(struct rule *); + uint16_t idle_timeout, uint16_t hard_timeout, + 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 *); -struct ofconn { - struct list node; - struct rconn *rconn; - struct pktbuf *pktbuf; - bool send_flow_exp; - int miss_send_len; +static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason); + +/* An exact-match instantiation of an OpenFlow flow. */ +struct facet { + long long int used; /* Time last used; time created if not used. */ + + /* 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 ODPPF_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. */ + + /* 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. */ +}; + +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 *); - struct rconn_packet_counter *packet_in_counter; +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 *); - /* Number of OpenFlow messages queued as replies to OpenFlow requests, and - * the maximum number before we stop reading OpenFlow requests. */ +static void facet_make_actions(struct ofproto *, struct facet *, + const struct ofpbuf *packet); +static void facet_update_stats(struct ofproto *, struct facet *, + const struct odp_flow_stats *); + +/* 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. */ + + /* 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. */ +}; + +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. */ + enum nx_flow_format flow_format; /* Currently selected flow format. */ + + /* 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? */ }; -static struct ofconn *ofconn_create(struct ofproto *, struct rconn *); +/* 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_run(struct ofconn *); 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); + struct ofproto { /* Settings. */ uint64_t datapath_id; /* Datapath ID. */ uint64_t fallback_dpid; /* Datapath ID if no better choice found. */ - uint64_t mgmt_id; /* Management channel identifier. */ - char *manufacturer; /* Manufacturer. */ - char *hardware; /* Hardware. */ - char *software; /* Software version. */ - char *serial; /* Serial number. */ + char *mfr_desc; /* Manufacturer. */ + char *hw_desc; /* Hardware. */ + char *sw_desc; /* Software version. */ + char *serial_desc; /* Serial number. */ + char *dp_desc; /* Datapath description. */ /* 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 status_category *ss_cat; - struct in_band *in_band; - struct discovery *discovery; struct fail_open *fail_open; - struct pinsched *miss_sched, *action_sched; 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; + int in_band_queue; /* Flow table. */ struct classifier cls; - bool need_revalidate; long long int next_expiration; + + /* Facets. */ + struct hmap facets; + bool need_revalidate; struct tag_set revalidate_set; /* OpenFlow connections. */ - struct list all_conns; - struct ofconn *controller; - struct pvconn **listeners; - size_t n_listeners; + 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; @@ -230,34 +400,32 @@ 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; static uint64_t pick_datapath_id(const struct ofproto *); static uint64_t pick_fallback_dpid(void); -static void send_packet_in_miss(struct ofpbuf *, void *ofproto); -static void send_packet_in_action(struct ofpbuf *, void *ofproto); -static void update_used(struct ofproto *); -static void update_stats(struct 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 handle_odp_msg(struct ofproto *, struct ofpbuf *); -static void handle_openflow(struct ofconn *, struct ofproto *, - struct ofpbuf *); +static void handle_openflow(struct ofconn *, struct ofpbuf *); -static void refresh_port_group(struct ofproto *, unsigned int group); +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 struct ofhooks *ofhooks, void *aux, +ofproto_create(const char *datapath, const char *datapath_type, + const struct ofhooks *ofhooks, void *aux, struct ofproto **ofprotop) { struct odp_stats stats; @@ -267,8 +435,10 @@ ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux, *ofprotop = NULL; + ofproto_unixctl_init(); + /* Connect to datapath and start listening for messages. */ - error = dpif_open(datapath, &dpif); + error = dpif_open(datapath, datapath_type, &dpif); if (error) { VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error)); return error; @@ -280,7 +450,7 @@ ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux, dpif_close(dpif); return error; } - error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION); + error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW); if (error) { VLOG_ERR("failed to listen on datapath %s: %s", datapath, strerror(error)); @@ -294,39 +464,42 @@ ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux, p = xzalloc(sizeof *p); p->fallback_dpid = pick_fallback_dpid(); p->datapath_id = p->fallback_dpid; - p->manufacturer = xstrdup("Nicira Networks, Inc."); - p->hardware = xstrdup("Reference Implementation"); - p->software = xstrdup(VERSION BUILDNR); - p->serial = xstrdup("None"); + p->mfr_desc = xstrdup(DEFAULT_MFR_DESC); + p->hw_desc = xstrdup(DEFAULT_HW_DESC); + p->sw_desc = xstrdup(DEFAULT_SW_DESC); + p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC); + p->dp_desc = xstrdup(DEFAULT_DP_DESC); /* 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; /* Initialize submodules. */ p->switch_status = switch_status_create(p); - p->in_band = NULL; - p->discovery = NULL; p->fail_open = NULL; - p->miss_sched = p->action_sched = NULL; p->netflow = NULL; + p->sflow = NULL; + + /* Initialize in-band control. */ + p->in_band = NULL; + p->in_band_queue = -1; /* Initialize flow table. */ classifier_init(&p->cls); - p->need_revalidate = false; p->next_expiration = time_msec() + 1000; + + /* Initialize facet table. */ + hmap_init(&p->facets); + p->need_revalidate = false; tag_set_init(&p->revalidate_set); /* Initialize OpenFlow connections. */ list_init(&p->all_conns); - p->controller = ofconn_create(p, rconn_create(5, 8)); - p->controller->pktbuf = pktbuf_create(); - p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN; - p->listeners = NULL; - p->n_listeners = 0; + hmap_init(&p->controllers); + hmap_init(&p->services); p->snoops = NULL; p->n_snoops = 0; @@ -341,20 +514,11 @@ ofproto_create(const char *datapath, const struct ofhooks *ofhooks, void *aux, p->ml = mac_learning_create(); } - /* Register switch status category. */ - p->ss_cat = switch_status_register(p->switch_status, "remote", - rconn_status_cb, p->controller->rconn); - - /* Almost done... */ - error = init_ports(p); - if (error) { - ofproto_destroy(p); - return error; - } - /* Pick final datapath ID. */ p->datapath_id = pick_datapath_id(p); - VLOG_INFO("using datapath ID %012"PRIx64, p->datapath_id); + VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id); + + shash_add_once(&all_ofprotos, dpif_name(p->dpif), p); *ofprotop = p; return 0; @@ -366,116 +530,392 @@ ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id) uint64_t old_dpid = p->datapath_id; p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p); if (p->datapath_id != old_dpid) { - VLOG_INFO("datapath ID changed to %012"PRIx64, p->datapath_id); - rconn_reconnect(p->controller->rconn); + VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id); + + /* Force all active connections to reconnect, since there is no way to + * notify a controller that the datapath ID has changed. */ + ofproto_reconnect_controllers(p); } } -void -ofproto_set_mgmt_id(struct ofproto *p, uint64_t mgmt_id) +static bool +is_discovery_controller(const struct ofproto_controller *c) { - p->mgmt_id = mgmt_id; + return !strcmp(c->target, "discover"); } -void -ofproto_set_probe_interval(struct ofproto *p, int probe_interval) +static bool +is_in_band_controller(const struct ofproto_controller *c) { - probe_interval = probe_interval ? MAX(probe_interval, 5) : 0; - rconn_set_probe_interval(p->controller->rconn, probe_interval); - if (p->fail_open) { - int trigger_duration = probe_interval ? probe_interval * 3 : 15; - fail_open_set_trigger_duration(p->fail_open, trigger_duration); + 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)); } -void -ofproto_set_max_backoff(struct ofproto *p, int max_backoff) +/* 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) { - rconn_set_max_backoff(p->controller->rconn, max_backoff); + return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn); } -void -ofproto_set_desc(struct ofproto *p, - const char *manufacturer, const char *hardware, - const char *software, const char *serial) +static struct ofconn * +find_controller_by_target(struct ofproto *ofproto, const char *target) { - if (manufacturer) { - free(p->manufacturer); - p->manufacturer = xstrdup(manufacturer); + struct ofconn *ofconn; + + HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node, + hash_string(target, 0), &ofproto->controllers) { + if (!strcmp(ofconn_get_target(ofconn), target)) { + return ofconn; + } } - if (hardware) { - free(p->hardware); - p->hardware = xstrdup(hardware); + 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, 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; + } } - if (software) { - free(p->software); - p->software = xstrdup(software); + for (i = 0; i < ofproto->n_extra_remotes; i++) { + addrs[n_addrs++] = ofproto->extra_in_band_remotes[i]; } - if (serial) { - free(p->serial); - p->serial = xstrdup(serial); + + /* 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); + } + in_band_set_queue(ofproto->in_band, ofproto->in_band_queue); + ofproto->next_in_band_update = time_msec() + 1000; + } else { + in_band_destroy(ofproto->in_band); + ofproto->in_band = NULL; } + + /* Clean up. */ + free(addrs); } -int -ofproto_set_in_band(struct ofproto *p, bool in_band) +static void +update_fail_open(struct ofproto *p) { - if (in_band != (p->in_band != NULL)) { - if (in_band) { - return in_band_create(p, p->dpif, p->switch_status, - p->controller->rconn, &p->in_band); - } else { - ofproto_set_discovery(p, false, NULL, true); - in_band_destroy(p->in_band); - p->in_band = NULL; + 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, hmap_node, &p->controllers) { + rconns[n++] = ofconn->rconn; } - rconn_reconnect(p->controller->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; } - return 0; } -int -ofproto_set_discovery(struct ofproto *p, bool discovery, - const char *re, bool update_resolv_conf) -{ - if (discovery != (p->discovery != NULL)) { - if (discovery) { - int error = ofproto_set_in_band(p, true); - if (error) { - return error; +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); } - error = discovery_create(re, update_resolv_conf, - p->dpif, p->switch_status, - &p->discovery); - if (error) { - return error; + } else if (!pvconn_verify_name(c->target)) { + if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) { + continue; } } else { - discovery_destroy(p->discovery); - p->discovery = NULL; + VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"", + dpif_name(p->dpif), c->target); + continue; } - rconn_disconnect(p->controller->rconn); - } else if (discovery) { - discovery_set_update_resolv_conf(p->discovery, update_resolv_conf); - return discovery_set_accept_controller_re(p->discovery, re); + + 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, 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, 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); } - return 0; } -int -ofproto_set_controller(struct ofproto *ofproto, const char *controller) +void +ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode) { - if (ofproto->discovery) { - return EINVAL; - } else if (controller) { - if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) { - return rconn_connect(ofproto->controller->rconn, controller); - } else { - return 0; + p->fail_mode = fail_mode; + update_fail_open(p); +} + +/* Drops the connections between 'ofproto' and all of its controllers, forcing + * them to reconnect. */ +void +ofproto_reconnect_controllers(struct ofproto *ofproto) +{ + struct ofconn *ofconn; + + LIST_FOR_EACH (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; } - } else { - rconn_disconnect(ofproto->controller->rconn); - return 0; + } + + return false; +} + +/* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s + * in-band control should guarantee access, in the same way that in-band + * control guarantees access to OpenFlow controllers. */ +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); + + 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) +{ + if (queue_id != ofproto->in_band_queue) { + ofproto->in_band_queue = queue_id; + update_in_band_remotes(ofproto); + } +} + +void +ofproto_set_desc(struct ofproto *p, + const char *mfr_desc, const char *hw_desc, + const char *sw_desc, const char *serial_desc, + const char *dp_desc) +{ + struct ofp_desc_stats *ods; + + if (mfr_desc) { + if (strlen(mfr_desc) >= sizeof ods->mfr_desc) { + VLOG_WARN("truncating mfr_desc, must be less than %zu characters", + sizeof ods->mfr_desc); + } + free(p->mfr_desc); + p->mfr_desc = xstrdup(mfr_desc); + } + if (hw_desc) { + if (strlen(hw_desc) >= sizeof ods->hw_desc) { + VLOG_WARN("truncating hw_desc, must be less than %zu characters", + sizeof ods->hw_desc); + } + free(p->hw_desc); + p->hw_desc = xstrdup(hw_desc); + } + if (sw_desc) { + if (strlen(sw_desc) >= sizeof ods->sw_desc) { + VLOG_WARN("truncating sw_desc, must be less than %zu characters", + sizeof ods->sw_desc); + } + free(p->sw_desc); + p->sw_desc = xstrdup(sw_desc); + } + if (serial_desc) { + if (strlen(serial_desc) >= sizeof ods->serial_num) { + VLOG_WARN("truncating serial_desc, must be less than %zu " + "characters", + sizeof ods->serial_num); + } + free(p->serial_desc); + p->serial_desc = xstrdup(serial_desc); + } + if (dp_desc) { + if (strlen(dp_desc) >= sizeof ods->dp_desc) { + VLOG_WARN("truncating dp_desc, must be less than %zu characters", + sizeof ods->dp_desc); + } + free(p->dp_desc); + p->dp_desc = xstrdup(dp_desc); } } @@ -517,12 +957,6 @@ set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp, return retval; } -int -ofproto_set_listeners(struct ofproto *ofproto, const struct svec *listeners) -{ - return set_pvconns(&ofproto->listeners, &ofproto->n_listeners, listeners); -} - int ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops) { @@ -546,57 +980,24 @@ ofproto_set_netflow(struct ofproto *ofproto, } void -ofproto_set_failure(struct ofproto *ofproto, bool fail_open) -{ - if (fail_open) { - struct rconn *rconn = ofproto->controller->rconn; - int trigger_duration = rconn_get_probe_interval(rconn) * 3; - if (!ofproto->fail_open) { - ofproto->fail_open = fail_open_create(ofproto, trigger_duration, - ofproto->switch_status, - rconn); - } else { - fail_open_set_trigger_duration(ofproto->fail_open, - trigger_duration); - } - } else { - fail_open_destroy(ofproto->fail_open); - ofproto->fail_open = NULL; - } -} - -void -ofproto_set_rate_limit(struct ofproto *ofproto, - int rate_limit, int burst_limit) -{ - if (rate_limit > 0) { - if (!ofproto->miss_sched) { - ofproto->miss_sched = pinsched_create(rate_limit, burst_limit, - ofproto->switch_status); - ofproto->action_sched = pinsched_create(rate_limit, burst_limit, - NULL); - } else { - pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit); - pinsched_set_limits(ofproto->action_sched, - rate_limit, burst_limit); +ofproto_set_sflow(struct ofproto *ofproto, + const struct ofproto_sflow_options *oso) +{ + struct ofproto_sflow *os = ofproto->sflow; + if (oso) { + if (!os) { + struct ofport *ofport; + + os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif); + HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) { + ofproto_sflow_add_port(os, ofport->odp_port, + netdev_get_name(ofport->netdev)); + } } + ofproto_sflow_set_options(os, oso); } else { - pinsched_destroy(ofproto->miss_sched); - ofproto->miss_sched = NULL; - pinsched_destroy(ofproto->action_sched); - ofproto->action_sched = NULL; - } -} - -int -ofproto_set_stp(struct ofproto *ofproto UNUSED, bool enable_stp) -{ - /* XXX */ - if (enable_stp) { - VLOG_WARN("STP is not yet implemented"); - return EINVAL; - } else { - return 0; + ofproto_sflow_destroy(os); + ofproto->sflow = NULL; } } @@ -606,50 +1007,16 @@ ofproto_get_datapath_id(const struct ofproto *ofproto) return ofproto->datapath_id; } -uint64_t -ofproto_get_mgmt_id(const struct ofproto *ofproto) +bool +ofproto_has_primary_controller(const struct ofproto *ofproto) { - return ofproto->mgmt_id; + return !hmap_is_empty(&ofproto->controllers); } -int -ofproto_get_probe_interval(const struct ofproto *ofproto) +enum ofproto_fail_mode +ofproto_get_fail_mode(const struct ofproto *p) { - return rconn_get_probe_interval(ofproto->controller->rconn); -} - -int -ofproto_get_max_backoff(const struct ofproto *ofproto) -{ - return rconn_get_max_backoff(ofproto->controller->rconn); -} - -bool -ofproto_get_in_band(const struct ofproto *ofproto) -{ - return ofproto->in_band != NULL; -} - -bool -ofproto_get_discovery(const struct ofproto *ofproto) -{ - return ofproto->discovery != NULL; -} - -const char * -ofproto_get_controller(const struct ofproto *ofproto) -{ - return rconn_get_name(ofproto->controller->rconn); -} - -void -ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners) -{ - size_t i; - - for (i = 0; i < ofproto->n_listeners; i++) { - svec_add(listeners, pvconn_get_name(ofproto->listeners[i])); - } + return p->fail_mode; } void @@ -665,44 +1032,50 @@ ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops) void ofproto_destroy(struct ofproto *p) { + struct ofservice *ofservice, *next_ofservice; struct ofconn *ofconn, *next_ofconn; - struct ofport *ofport; - unsigned int port_no; + struct ofport *ofport, *next_ofport; size_t i; if (!p) { return; } + shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif)); + + /* 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); + ofproto_flush_flows(p); classifier_destroy(&p->cls); + hmap_destroy(&p->facets); - LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node, - &p->all_conns) { + LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) { ofconn_destroy(ofconn); } + hmap_destroy(&p->controllers); 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); - in_band_destroy(p->in_band); - discovery_destroy(p->discovery); - fail_open_destroy(p->fail_open); - pinsched_destroy(p->miss_sched); - pinsched_destroy(p->action_sched); netflow_destroy(p->netflow); + ofproto_sflow_destroy(p->sflow); - switch_status_unregister(p->ss_cat); - - for (i = 0; i < p->n_listeners; i++) { - pvconn_close(p->listeners[i]); + HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) { + ofservice_destroy(p, ofservice); } - free(p->listeners); + hmap_destroy(&p->services); for (i = 0; i < p->n_snoops; i++) { pvconn_close(p->snoops[i]); @@ -711,6 +1084,14 @@ ofproto_destroy(struct ofproto *p) mac_learning_destroy(p->ml); + free(p->mfr_desc); + free(p->hw_desc); + free(p->sw_desc); + free(p->serial_desc); + free(p->dp_desc); + + hmap_destroy(&p->ports); + free(p); } @@ -735,17 +1116,64 @@ 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, 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; char *devname; int error; int i; + if (shash_is_empty(&p->port_by_name)) { + init_ports(p); + } + for (i = 0; i < 50; i++) { struct ofpbuf *buf; - int error; error = dpif_recv(p->dpif, &buf); if (error) { @@ -753,8 +1181,8 @@ ofproto_run1(struct ofproto *p) /* Someone destroyed the datapath behind our back. The caller * better destroy us and give up, because we're just going to * spin from here on out. */ - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); - VLOG_ERR_RL(&rl, "%s: datapath was destroyed externally", + static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally", dpif_name(p->dpif)); return ENODEV; } @@ -773,27 +1201,14 @@ ofproto_run1(struct ofproto *p) } if (p->in_band) { - in_band_run(p->in_band); - } - if (p->discovery) { - char *controller_name; - if (rconn_is_connectivity_questionable(p->controller->rconn)) { - discovery_question_connectivity(p->discovery); - } - if (discovery_run(p->discovery, &controller_name)) { - if (controller_name) { - rconn_connect(p->controller->rconn, controller_name); - } else { - rconn_disconnect(p->controller->rconn); - } + if (time_msec() >= p->next_in_band_update) { + update_in_band_remotes(p); } + in_band_run(p->in_band); } - pinsched_run(p->miss_sched, send_packet_in_miss, p); - pinsched_run(p->action_sched, send_packet_in_action, p); - LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node, - &p->all_conns) { - ofconn_run(ofconn, p); + LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) { + ofconn_run(ofconn); } /* Fail-open maintenance. Do this after processing the ofconns since @@ -802,13 +1217,23 @@ ofproto_run1(struct ofproto *p) fail_open_run(p->fail_open); } - for (i = 0; i < p->n_listeners; i++) { + HMAP_FOR_EACH (ofservice, node, &p->services) { struct vconn *vconn; int retval; - retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn); + retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn); if (!retval) { - ofconn_create(p, rconn_new_from_vconn("passive", vconn)); + 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)); } @@ -820,56 +1245,51 @@ ofproto_run1(struct ofproto *p) retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn); if (!retval) { - rconn_add_monitor(p->controller->rconn, vconn); + add_snooper(p, vconn); } else if (retval != EAGAIN) { VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); } } if (time_msec() >= p->next_expiration) { + int delay = ofproto_expire(p); + p->next_expiration = time_msec() + 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) { netflow_run(p->netflow); } + if (p->sflow) { + ofproto_sflow_run(p->sflow); + } 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; @@ -878,26 +1298,26 @@ ofproto_run2(struct ofproto *p, bool revalidate_all) void ofproto_wait(struct ofproto *p) { + struct ofservice *ofservice; struct ofconn *ofconn; size_t i; 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) { + LIST_FOR_EACH (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->discovery) { - discovery_wait(p->discovery); - } if (p->fail_open) { fail_open_wait(p->fail_open); } - pinsched_wait(p->miss_sched); - pinsched_wait(p->action_sched); + if (p->sflow) { + ofproto_sflow_wait(p->sflow); + } if (!tag_set_is_empty(&p->revalidate_set)) { poll_immediate_wake(); } @@ -906,10 +1326,10 @@ ofproto_wait(struct ofproto *p) VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); poll_immediate_wake(); } else if (p->next_expiration != LLONG_MAX) { - poll_timer_wait(p->next_expiration - time_msec()); + poll_timer_wait_until(p->next_expiration); } - for (i = 0; i < p->n_listeners; i++) { - pvconn_wait(p->listeners[i]); + HMAP_FOR_EACH (ofservice, node, &p->services) { + pvconn_wait(ofservice->pvconn); } for (i = 0; i < p->n_snoops; i++) { pvconn_wait(p->snoops[i]); @@ -931,77 +1351,124 @@ ofproto_get_revalidate_set(struct ofproto *ofproto) bool ofproto_is_alive(const struct ofproto *p) { - return p->discovery || rconn_is_alive(p->controller->rconn); + return !hmap_is_empty(&p->controllers); } +/* 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_send_packet(struct ofproto *p, const flow_t *flow, - const union ofp_action *actions, size_t n_actions, - const struct ofpbuf *packet) +ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port) { - struct odp_actions odp_actions; + struct ofport *ofport = get_port(ofproto, odp_port); + const char *name = ofport ? ofport->opp.name : ""; int error; - error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions, - NULL, NULL, NULL); + error = dpif_port_del(ofproto->dpif, odp_port); if (error) { - return 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); +} + +int +ofproto_send_packet(struct ofproto *p, const struct flow *flow, + const union ofp_action *actions, size_t n_actions, + const struct ofpbuf *packet) +{ + struct action_xlate_ctx ctx; + struct ofpbuf *odp_actions; + + action_xlate_ctx_init(&ctx, p, flow, packet); + odp_actions = xlate_actions(&ctx, actions, n_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); + dpif_execute(p->dpif, odp_actions->data, odp_actions->size, packet); + + ofpbuf_delete(odp_actions); + return 0; } +/* 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); - cls_rule_from_flow(&rule->cr, flow, wildcards, priority); - 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); + + 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_remove(ofproto, facet); + } + + 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); if (ofproto->in_band) { in_band_flushed(ofproto->in_band); @@ -1014,68 +1481,46 @@ ofproto_flush_flows(struct ofproto *ofproto) static void reinit_ports(struct ofproto *p) { - struct svec devnames; + struct shash_node *node; + struct shash devnames; struct ofport *ofport; - unsigned int port_no; struct odp_port *odp_ports; size_t n_odp_ports; size_t i; - svec_init(&devnames); - PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) { - svec_add (&devnames, (char *) ofport->opp.name); + COVERAGE_INC(ofproto_reinit_ports); + + shash_init(&devnames); + HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { + shash_add_once (&devnames, ofport->opp.name, NULL); } 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); + shash_add_once (&devnames, odp_ports[i].devname, NULL); } free(odp_ports); - svec_sort_unique(&devnames); - for (i = 0; i < devnames.n; i++) { - update_port(p, devnames.names[i]); - } - svec_destroy(&devnames); -} - -static void -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; - } + SHASH_FOR_EACH (node, &devnames) { + update_port(p, node->name); } - dpif_port_group_set(p->dpif, group, ports, n_ports); - free(ports); -} - -static void -refresh_port_groups(struct ofproto *p) -{ - refresh_port_group(p, DP_GROUP_FLOOD); - refresh_port_group(p, DP_GROUP_ALL); + shash_destroy(&devnames); } static struct ofport * make_ofport(const struct odp_port *odp_port) { + struct netdev_options netdev_options; enum netdev_flags flags; struct ofport *ofport; struct netdev *netdev; - bool carrier; int error; - error = netdev_open(odp_port->devname, NETDEV_ETH_TYPE_NONE, &netdev); + memset(&netdev_options, 0, sizeof netdev_options); + netdev_options.name = odp_port->devname; + netdev_options.type = odp_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)", @@ -1086,6 +1531,7 @@ make_ofport(const struct odp_port *odp_port) ofport = xmalloc(sizeof *ofport); ofport->netdev = netdev; + ofport->odp_port = odp_port->port; ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port); netdev_get_etheraddr(netdev, ofport->opp.hw_addr); memcpy(ofport->opp.name, odp_port->devname, @@ -1095,8 +1541,7 @@ make_ofport(const struct odp_port *odp_port) 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, @@ -1107,7 +1552,7 @@ make_ofport(const struct odp_port *odp_port) static bool ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port) { - if (port_array_get(&p->ports, odp_port->port)) { + if (get_port(p, odp_port->port)) { VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath", odp_port->port); return true; @@ -1129,7 +1574,7 @@ 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 @@ -1144,37 +1589,48 @@ send_port_status(struct ofproto *p, const struct ofport *ofport, { /* XXX Should limit the number of queued port status change messages. */ struct ofconn *ofconn; - LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { + LIST_FOR_EACH (ofconn, node, &p->all_conns) { struct ofp_port_status *ops; struct ofpbuf *b; + /* Primary controllers, even slaves, should always get port status + updates. Otherwise obey ofconn_receives_async_msgs(). */ + if (ofconn->type != OFCONN_PRIMARY + && !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) { + const char *netdev_name = ofport->opp.name; + netdev_monitor_add(p->netdev_monitor, ofport->netdev); - port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), - ofport); - shash_add(&p->port_by_name, (char *) ofport->opp.name, ofport); + 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, ofport->odp_port, netdev_name); + } } static void ofport_remove(struct ofproto *p, struct ofport *ofport) { netdev_monitor_remove(p->netdev_monitor, ofport->netdev); - port_array_set(&p->ports, ofp_port_to_odp_port(ofport->opp.port_no), NULL); + 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, ofport->odp_port); + } } static void @@ -1186,6 +1642,20 @@ ofport_free(struct ofport *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) { @@ -1213,7 +1683,7 @@ 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, odp_port.port); } } else if (error != ENOENT && error != ENODEV) { VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error " @@ -1229,9 +1699,9 @@ update_port(struct ofproto *p, const char *devname) return; } 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)) { @@ -1253,9 +1723,6 @@ update_port(struct ofproto *p, const char *devname) : !new_ofport ? OFPPR_DELETE : OFPPR_MODIFY)); ofport_free(old_ofport); - - /* Update port groups. */ - refresh_port_groups(p); } static int @@ -1281,20 +1748,22 @@ init_ports(struct ofproto *p) } } free(ports); - refresh_port_groups(p); return 0; } static struct ofconn * -ofconn_create(struct ofproto *p, struct rconn *rconn) +ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type) { - struct ofconn *ofconn = xmalloc(sizeof *ofconn); + struct ofconn *ofconn = xzalloc(sizeof *ofconn); + ofconn->ofproto = p; list_push_back(&p->all_conns, &ofconn->node); ofconn->rconn = rconn; + ofconn->type = type; + ofconn->flow_format = NXFF_OPENFLOW10; + ofconn->role = NX_ROLE_OTHER; + ofconn->packet_in_counter = rconn_packet_counter_create (); ofconn->pktbuf = NULL; - ofconn->send_flow_exp = false; ofconn->miss_send_len = 0; - ofconn->packet_in_counter = rconn_packet_counter_create (); ofconn->reply_counter = rconn_packet_counter_create (); return ofconn; } @@ -1302,7 +1771,13 @@ ofconn_create(struct ofproto *p, struct rconn *rconn) 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); @@ -1311,9 +1786,31 @@ ofconn_destroy(struct ofconn *ofconn) } static void -ofconn_run(struct ofconn *ofconn, struct ofproto *p) +ofconn_run(struct ofconn *ofconn) { + struct ofproto *p = ofconn->ofproto; 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); @@ -1328,12 +1825,12 @@ ofconn_run(struct ofconn *ofconn, struct ofproto *p) if (p->fail_open) { fail_open_maybe_recover(p->fail_open); } - handle_openflow(ofconn, p, of_msg); + handle_openflow(ofconn, of_msg); ofpbuf_delete(of_msg); } } - if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) { + if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) { ofconn_destroy(ofconn); } } @@ -1341,6 +1838,14 @@ ofconn_run(struct ofconn *ofconn, struct ofproto *p) 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); @@ -1348,28 +1853,146 @@ ofconn_wait(struct ofconn *ofconn) COVERAGE_INC(ofproto_ofconn_stuck); } } + +/* Returns true if 'ofconn' should receive asynchronous messages. */ +static bool +ofconn_receives_async_msgs(const struct ofconn *ofconn) +{ + 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; + } +} + +/* Returns a human-readable name for an OpenFlow connection between 'ofproto' + * and 'target', suitable for use in log messages for identifying the + * connection. + * + * 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, 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. */ +/* 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 +rule_is_hidden(const struct rule *rule) +{ + return rule->cr.priority > UINT16_MAX; +} + +/* Creates and returns a new rule initialized as specified. + * + * 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) + uint16_t idle_timeout, uint16_t hard_timeout, + 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->super = super; - if (super) { - list_push_back(&super->list, &rule->list); - } else { - list_init(&rule->list); + rule->send_flow_removed = send_flow_removed; + 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; } @@ -1384,34 +2007,28 @@ 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; @@ -1421,314 +2038,467 @@ rule_has_out_port(const struct rule *rule, uint16_t out_port) } for (oa = actions_first(&i, rule->actions, rule->n_actions); oa; oa = actions_next(&i)) { - if (oa->type == htons(OFPAT_OUTPUT) && oa->output.port == out_port) { + if (action_outputs_to_port(oa, out_port)) { return true; } } return false; } -/* Executes the actions indicated by 'rule' on 'packet', which is in flow - * 'flow' and is considered to have arrived on ODP port 'in_port'. +/* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on + * 'packet', which arrived on 'in_port'. * - * 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'. - * - * 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'. */ -static void -rule_execute(struct ofproto *ofproto, struct rule *rule, - struct ofpbuf *packet, const flow_t *flow) + * Takes ownership of 'packet'. */ +static bool +execute_odp_actions(struct ofproto *ofproto, uint16_t in_port, + const struct nlattr *odp_actions, size_t actions_len, + struct ofpbuf *packet) { - const union odp_action *actions; - size_t n_actions; - struct odp_actions a; + if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t)) + && odp_actions->nla_type == ODPAT_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; - /* 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)) { - return; - } - actions = a.actions; - n_actions = a.n_actions; + 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->arg = nl_attr_get_u64(odp_actions); + + send_packet_in(ofproto, packet); + + return true; } else { - actions = rule->odp_actions; - n_actions = rule->n_odp_actions; - } + int error; - /* Execute the ODP actions. */ - if (!dpif_execute(ofproto->dpif, flow->in_port, - actions, n_actions, packet)) { - struct odp_flow_stats stats; - flow_extract_stats(flow, packet, &stats); - update_stats(ofproto, rule, &stats); - rule->used = time_msec(); - netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used); + error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet); + ofpbuf_delete(packet); + return !error; } } +/* 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. + * + * 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'. + * + * 'facet' must have accurately composed ODP actions; that is, it must not be + * in need of revalidation. + * + * Takes ownership of 'packet'. */ static void -rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet, - uint16_t in_port) +facet_execute(struct ofproto *ofproto, struct facet *facet, + struct ofpbuf *packet) { - struct rule *displaced_rule; + struct odp_flow_stats stats; - /* 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)); - /* Send the packet and credit it to the rule. */ - if (packet) { - flow_t flow; - flow_extract(packet, in_port, &flow); - rule_execute(p, rule, packet, &flow); + flow_extract_stats(&facet->flow, packet, &stats); + if (execute_odp_actions(ofproto, facet->flow.in_port, + facet->actions, facet->actions_len, packet)) { + facet_update_stats(ofproto, facet, &stats); + facet->used = time_msec(); + netflow_flow_update_time(ofproto->netflow, + &facet->nf_flow, facet->used); } +} - /* 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); +/* 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. + * + * '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_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port, + struct ofpbuf *packet) +{ + struct action_xlate_ctx ctx; + struct ofpbuf *odp_actions; + struct facet *facet; + struct flow flow; + size_t size; + + 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; + } + + /* 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; + } + + /* 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, in_port, odp_actions->data, + odp_actions->size, packet)) { + rule->used = time_msec(); + rule->packet_count++; + rule->byte_count += size; } + 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); - COVERAGE_INC(ofproto_subrule_create); - cls_rule_from_flow(&subrule->cr, flow, 0, - (rule->cr.priority <= UINT16_MAX ? UINT16_MAX - : rule->cr.priority)); - 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); + + facet_make_actions(ofproto, facet, packet); - return subrule; + 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; +/* 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) +{ + 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 +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 int -do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags, +facet_put__(struct ofproto *ofproto, struct facet *facet, int flags, struct odp_flow_put *put) { 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; + odp_flow_key_from_flow(&put->flow.key, &facet->flow); + put->flow.actions = facet->actions; + put->flow.actions_len = facet->actions_len; + put->flow.flags = 0; put->flags = flags; return dpif_flow_put(ofproto->dpif, put); } +/* 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_install(struct ofproto *p, struct rule *rule, struct rule *displaced_rule) +facet_install(struct ofproto *p, struct facet *facet, bool zero_stats) { - assert(!rule->cr.wc.wildcards); - - if (rule->may_install) { + if (facet->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); - } + int flags; + + flags = ODPPF_CREATE | ODPPF_MODIFY; + if (zero_stats) { + flags |= ODPPF_ZERO_STATS; + } + if (!facet_put__(p, facet, flags, &put)) { + facet->installed = true; } - } else if (displaced_rule) { - rule_uninstall(p, displaced_rule); } } +/* 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_reinstall(struct ofproto *ofproto, struct rule *rule) +facet_account(struct ofproto *ofproto, + struct facet *facet, uint64_t extra_bytes) { - 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); + uint64_t total_bytes = facet->byte_count + extra_bytes; + + if (ofproto->ofhooks->account_flow_cb + && total_bytes > facet->accounted_bytes) + { + ofproto->ofhooks->account_flow_cb( + &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_update_actions(struct ofproto *ofproto, struct rule *rule) +facet_uninstall(struct ofproto *p, struct facet *facet) { - 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); + if (facet->installed) { + struct odp_flow odp_flow; + + odp_flow_key_from_flow(&odp_flow.key, &facet->flow); + odp_flow.actions = NULL; + odp_flow.actions_len = 0; + odp_flow.flags = 0; + if (!dpif_flow_del(p->dpif, &odp_flow)) { + facet_update_stats(p, facet, &odp_flow.stats); } - } else { - rule_uninstall(ofproto, rule); + facet->installed = false; } } +/* 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 +facet_is_controller_flow(struct facet *facet) +{ + 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. */ static void -rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes) +facet_flush_stats(struct ofproto *ofproto, struct facet *facet) { - uint64_t total_bytes = rule->byte_count + extra_bytes; + facet_account(ofproto, facet, 0); - if (ofproto->ofhooks->account_flow_cb - && total_bytes > rule->accounted_bytes) - { - ofproto->ofhooks->account_flow_cb( - &rule->cr.flow, rule->odp_actions, rule->n_odp_actions, - total_bytes - rule->accounted_bytes, ofproto->aux); - rule->accounted_bytes = total_bytes; + if (ofproto->netflow && !facet_is_controller_flow(facet)) { + struct ofexpired 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); } + + facet->rule->packet_count += facet->packet_count; + facet->rule->byte_count += facet->byte_count; + + /* Reset counters to prevent double counting if 'facet' ever gets + * reinstalled. */ + facet->packet_count = 0; + facet->byte_count = 0; + facet->accounted_bytes = 0; + + netflow_flow_clear(&facet->nf_flow); } -static void -rule_uninstall(struct ofproto *p, struct rule *rule) +/* 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) { - assert(!rule->cr.wc.wildcards); - if (rule->installed) { - struct odp_flow odp_flow; + struct facet *facet; - odp_flow.key = rule->cr.flow; - odp_flow.actions = NULL; - odp_flow.n_actions = 0; - if (!dpif_flow_del(p->dpif, &odp_flow)) { - update_stats(p, rule, &odp_flow.stats); + HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0), + &ofproto->facets) { + if (flow_equal(flow, &facet->flow)) { + return facet; } - rule->installed = false; - - rule_post_uninstall(p, rule); } + + return NULL; } -static bool -is_controller_rule(struct rule *rule) +/* 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) { - /* 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. */ + struct facet *facet = facet_find(ofproto, flow); - if (rule && rule->super) { - struct rule *super = rule->super; - - return super->n_actions == 1 && - super->actions[0].type == htons(OFPAT_OUTPUT) && - super->actions[0].output.port == htons(OFPP_CONTROLLER); + /* 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; } - return false; + return facet; } -static void -rule_post_uninstall(struct ofproto *ofproto, struct rule *rule) +/* 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) { - struct rule *super = rule->super; + struct action_xlate_ctx ctx; + struct ofpbuf *odp_actions; + struct rule *new_rule; + bool actions_changed; - rule_account(ofproto, rule, 0); + COVERAGE_INC(facet_revalidate); - if (ofproto->netflow && !is_controller_rule(rule)) { - struct ofexpired expired; - expired.flow = rule->cr.flow; - expired.packet_count = rule->packet_count; - expired.byte_count = rule->byte_count; - expired.used = rule->used; - netflow_expire(ofproto->netflow, &rule->nf_flow, &expired); + /* 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; } - 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; + /* 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 || facet->may_install != facet->installed) { + if (facet->may_install) { + struct odp_flow_put put; + + memset(&put.flow.stats, 0, sizeof put.flow.stats); + odp_flow_key_from_flow(&put.flow.key, &facet->flow); + put.flow.actions = odp_actions->data; + put.flow.actions_len = odp_actions->size; + put.flow.flags = 0; + put.flags = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS; + dpif_flow_put(ofproto->dpif, &put); + + facet_update_stats(ofproto, facet, &put.flow.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); + } - netflow_flow_clear(&rule->nf_flow); + /* 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; } + + ofpbuf_delete(odp_actions); + + return true; } static void @@ -1741,34 +2511,15 @@ queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, } } -static void -send_error(const struct ofconn *ofconn, const struct ofp_header *oh, - int error, const void *data, size_t len) -{ - struct ofpbuf *buf; - struct ofp_error_msg *oem; - - if (!(error >> 16)) { - VLOG_WARN_RL(&rl, "not sending bad error code %d to controller", - error); - return; - } - - 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); -} - static void send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh, int error) { - size_t oh_length = ntohs(oh->length); - send_error(ofconn, oh, error, oh, MIN(oh_length, 64)); + struct ofpbuf *buf = make_ofp_error_msg(error, oh); + if (buf) { + COVERAGE_INC(ofproto_error); + queue_tx(buf, ofconn, ofconn->reply_counter); + } } static void @@ -1784,28 +2535,25 @@ hton_ofp_phy_port(struct ofp_phy_port *opp) } 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); + queue_tx(make_echo_reply(oh), ofconn, ofconn->reply_counter); 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 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(ofconn->ofproto->datapath_id); osf->n_buffers = htonl(pktbuf_capacity()); osf->n_tables = 2; osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS | - OFPC_PORT_STATS | OFPC_MULTI_PHY_TX); + OFPC_PORT_STATS | OFPC_ARP_MATCH_IP); osf->actions = htonl((1u << OFPAT_OUTPUT) | (1u << OFPAT_SET_VLAN_VID) | (1u << OFPAT_SET_VLAN_PCP) | @@ -1814,10 +2562,12 @@ handle_features_request(struct ofproto *p, struct ofconn *ofconn, (1u << OFPAT_SET_DL_DST) | (1u << OFPAT_SET_NW_SRC) | (1u << OFPAT_SET_NW_DST) | + (1u << OFPAT_SET_NW_TOS) | (1u << OFPAT_SET_TP_SRC) | - (1u << OFPAT_SET_TP_DST)); + (1u << OFPAT_SET_TP_DST) | + (1u << OFPAT_ENQUEUE)); - PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { + HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) { hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp)); } @@ -1826,8 +2576,7 @@ handle_features_request(struct ofproto *p, struct ofconn *ofconn, } 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 ofpbuf *buf; struct ofp_switch_config *osc; @@ -1835,11 +2584,8 @@ handle_get_config_request(struct ofproto *p, struct ofconn *ofconn, bool drop_frags; /* Figure out flags. */ - dpif_get_drop_frags(p->dpif, &drop_frags); + dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags); flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL; - if (ofconn->send_flow_exp) { - flags |= OFPC_SEND_FLOW_EXP; - } /* Send reply. */ osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf); @@ -1851,27 +2597,17 @@ handle_get_config_request(struct ofproto *p, struct ofconn *ofconn, } 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); + uint16_t flags = ntohs(osc->flags); - ofconn->send_flow_exp = (flags & OFPC_SEND_FLOW_EXP) != 0; - - if (ofconn == p->controller) { + if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) { switch (flags & OFPC_FRAG_MASK) { case OFPC_FRAG_NORMAL: - dpif_set_drop_frags(p->dpif, false); + dpif_set_drop_frags(ofconn->ofproto->dpif, false); break; case OFPC_FRAG_DROP: - dpif_set_drop_frags(p->dpif, true); + dpif_set_drop_frags(ofconn->ofproto->dpif, true); break; default: VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")", @@ -1880,54 +2616,14 @@ handle_set_config(struct ofproto *p, struct ofconn *ofconn, } } - if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) { - if (ntohs(osc->miss_send_len) != 0) { - ofconn->pktbuf = pktbuf_create(); - } else { - pktbuf_destroy(ofconn->pktbuf); - } - } - ofconn->miss_send_len = 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, - const struct ofp_action_output *oao) -{ - union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER); - a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX; -} - -struct action_xlate_ctx { - /* Input. */ - const 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 16 static void do_xlate_actions(const union ofp_action *in, size_t n_in, struct action_xlate_ctx *ctx); @@ -1935,7 +2631,7 @@ static void do_xlate_actions(const union ofp_action *in, size_t n_in, 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) { @@ -1950,71 +2646,82 @@ 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, ODPAT_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 xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port) { - if (!ctx->recurse) { + if (ctx->recurse < MAX_RESUBMIT_RECURSION) { + uint16_t old_in_port; struct rule *rule; - flow_t flow; - flow = *ctx->flow; - flow.in_port = in_port; + /* Look up a flow with 'in_port' as the input port. Then restore the + * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will + * have surprising behavior). */ + old_in_port = ctx->flow.in_port; + ctx->flow.in_port = in_port; + rule = rule_lookup(ctx->ofproto, &ctx->flow); + ctx->flow.in_port = old_in_port; - rule = lookup_valid_rule(ctx->ofproto, &flow); - 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 { + 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 -xlate_output_action(struct action_xlate_ctx *ctx, - const struct ofp_action_output *oao) +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, ODPAT_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) { uint16_t odp_port; uint16_t prev_nf_output_iface = ctx->nf_output_iface; ctx->nf_output_iface = NF_OUT_DROP; - switch (ntohs(oao->port)) { + switch (port) { case OFPP_IN_PORT: - add_output_action(ctx, ctx->flow->in_port); + add_output_action(ctx, ctx->flow.in_port); break; case OFPP_TABLE: - xlate_table_action(ctx, ctx->flow->in_port); + xlate_table_action(ctx, ctx->flow.in_port); break; case OFPP_NORMAL: - if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet, - ctx->out, ctx->tags, + if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet, + ctx->odp_actions, &ctx->tags, &ctx->nf_output_iface, ctx->ofproto->aux)) { COVERAGE_INC(ofproto_uninstallable); @@ -2022,21 +2729,22 @@ 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, oao); + nl_msg_put_u64(ctx->odp_actions, ODPAT_CONTROLLER, max_len); break; case OFPP_LOCAL: add_output_action(ctx, ODPP_LOCAL); break; default: - odp_port = ofp_port_to_odp_port(ntohs(oao->port)); - if (odp_port != ctx->flow->in_port) { + odp_port = ofp_port_to_odp_port(port); + if (odp_port != ctx->flow.in_port) { add_output_action(ctx, odp_port); } break; @@ -2053,166 +2761,380 @@ xlate_output_action(struct action_xlate_ctx *ctx, } static void -xlate_nicira_action(struct action_xlate_ctx *ctx, - const struct nx_action_header *nah) +xlate_output_action(struct action_xlate_ctx *ctx, + const struct ofp_action_output *oao) { - const struct nx_action_resubmit *nar; - int subtype = ntohs(nah->subtype); + xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); +} - assert(nah->vendor == htonl(NX_VENDOR_ID)); - switch (subtype) { - case NXAST_RESUBMIT: - nar = (const struct nx_action_resubmit *) nah; - xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port))); - break; +/* If the final ODP action in 'ctx' is "pop priority", drop it, as an + * optimization, because we're going to add another action that sets the + * priority immediately after, or because there are no actions following the + * pop. */ +static void +remove_pop_action(struct action_xlate_ctx *ctx) +{ + if (ctx->odp_actions->size == ctx->last_pop_priority) { + ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN); + ctx->last_pop_priority = -1; + } +} - default: - VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype); - break; +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, ODPAT_POP_PRIORITY); + ctx->last_pop_priority = ctx->odp_actions->size; } } static void -do_xlate_actions(const union ofp_action *in, size_t n_in, - struct action_xlate_ctx *ctx) +xlate_enqueue_action(struct action_xlate_ctx *ctx, + const struct ofp_action_enqueue *oae) { - struct actions_iterator iter; - const union ofp_action *ia; - const struct ofport *port; + uint16_t ofp_port, odp_port; + uint32_t priority; + int error; - port = port_array_get(&ctx->ofproto->ports, 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, stp_eth_addr) - ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) { - /* Drop this flow. */ + error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id), + &priority); + if (error) { + /* Fall back to ordinary output action. */ + xlate_output_action__(ctx, ntohs(oae->port), 0); return; } - for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) { - uint16_t type = ntohs(ia->type); - union odp_action *oa; + /* Figure out ODP output port. */ + ofp_port = ntohs(oae->port); + if (ofp_port != OFPP_IN_PORT) { + odp_port = ofp_port_to_odp_port(ofp_port); + } else { + odp_port = ctx->flow.in_port; + } - switch (type) { - case OFPAT_OUTPUT: - xlate_output_action(ctx, &ia->output); - break; + /* Add ODP actions. */ + remove_pop_action(ctx); + nl_msg_put_u32(ctx->odp_actions, ODPAT_SET_PRIORITY, priority); + add_output_action(ctx, odp_port); + add_pop_action(ctx); - case OFPAT_SET_VLAN_VID: - oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID); - oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid; - break; + /* Update NetFlow output port. */ + if (ctx->nf_output_iface == NF_OUT_DROP) { + ctx->nf_output_iface = odp_port; + } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { + ctx->nf_output_iface = NF_OUT_MULTI; + } +} - case OFPAT_SET_VLAN_PCP: - oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP); - oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp; - break; +static void +xlate_set_queue_action(struct action_xlate_ctx *ctx, + const struct nx_action_set_queue *nasq) +{ + uint32_t priority; + int error; - case OFPAT_STRIP_VLAN: - odp_actions_add(ctx->out, ODPAT_STRIP_VLAN); - break; + 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; + } - 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); - break; + remove_pop_action(ctx); + nl_msg_put_u32(ctx->odp_actions, ODPAT_SET_PRIORITY, priority); +} - 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); - break; +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, ODPAT_STRIP_VLAN); + } else { + nl_msg_put_be16(ctx->odp_actions, ODPAT_SET_DL_TCI, + tci & ~htons(VLAN_CFI)); + } +} - case OFPAT_SET_NW_SRC: - oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC); - oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; - break; +static void +xlate_reg_move_action(struct action_xlate_ctx *ctx, + const struct nx_action_reg_move *narm) +{ + ovs_be16 old_tci = ctx->flow.vlan_tci; - case OFPAT_SET_NW_DST: - oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST); - oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; - break; + nxm_execute_reg_move(narm, &ctx->flow); - case OFPAT_SET_TP_SRC: - oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC); - oa->tp_port.tp_port = ia->tp_port.tp_port; - break; + if (ctx->flow.vlan_tci != old_tci) { + xlate_set_dl_tci(ctx); + } +} + +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; + const struct nx_action_set_queue *nasq; + const struct nx_action_multipath *nam; + enum nx_action_subtype subtype = ntohs(nah->subtype); + ovs_be64 tun_id; + + assert(nah->vendor == htonl(NX_VENDOR_ID)); + switch (subtype) { + case NXAST_RESUBMIT: + nar = (const struct nx_action_resubmit *) nah; + xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port))); + break; + + case NXAST_SET_TUNNEL: + nast = (const struct nx_action_set_tunnel *) nah; + tun_id = htonll(ntohl(nast->tun_id)); + nl_msg_put_be64(ctx->odp_actions, ODPAT_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)) { + nl_msg_put_flag(ctx->odp_actions, ODPAT_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: + xlate_reg_move_action(ctx, (const struct nx_action_reg_move *) nah); + break; + + case NXAST_REG_LOAD: + nxm_execute_reg_load((const struct nx_action_reg_load *) nah, + &ctx->flow); + + 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, ODPAT_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; + + /* 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 %d", (int) subtype); + break; + } +} + +static void +do_xlate_actions(const union ofp_action *in, size_t n_in, + struct action_xlate_ctx *ctx) +{ + struct actions_iterator iter; + const union ofp_action *ia; + const struct ofport *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)) { + /* Drop this flow. */ + return; + } + + for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) { + enum ofp_action_type type = ntohs(ia->type); + const struct ofp_action_dl_addr *oada; + + switch (type) { + case OFPAT_OUTPUT: + xlate_output_action(ctx, &ia->output); + break; + + case OFPAT_SET_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: + 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: + ctx->flow.vlan_tci = htons(0); + xlate_set_dl_tci(ctx); + break; + + case OFPAT_SET_DL_SRC: + oada = ((struct ofp_action_dl_addr *) ia); + nl_msg_put_unspec(ctx->odp_actions, ODPAT_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: + oada = ((struct ofp_action_dl_addr *) ia); + nl_msg_put_unspec(ctx->odp_actions, ODPAT_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: + nl_msg_put_be32(ctx->odp_actions, ODPAT_SET_NW_SRC, + ia->nw_addr.nw_addr); + ctx->flow.nw_src = ia->nw_addr.nw_addr; + break; + + case OFPAT_SET_NW_DST: + nl_msg_put_be32(ctx->odp_actions, ODPAT_SET_NW_DST, + ia->nw_addr.nw_addr); + ctx->flow.nw_dst = ia->nw_addr.nw_addr; + break; + + case OFPAT_SET_NW_TOS: + nl_msg_put_u8(ctx->odp_actions, ODPAT_SET_NW_TOS, + ia->nw_tos.nw_tos); + ctx->flow.nw_tos = ia->nw_tos.nw_tos; + break; + + case OFPAT_SET_TP_SRC: + nl_msg_put_be16(ctx->odp_actions, ODPAT_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); - oa->tp_port.tp_port = ia->tp_port.tp_port; + nl_msg_put_be16(ctx->odp_actions, ODPAT_SET_TP_DST, + ia->tp_port.tp_port); + ctx->flow.tp_dst = ia->tp_port.tp_port; break; case OFPAT_VENDOR: xlate_nicira_action(ctx, (const struct nx_action_header *) ia); break; + case OFPAT_ENQUEUE: + xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia); + 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; +} + +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); + + 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; + 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 (!in_band_rule_check(ctx->ofproto->in_band, &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)) { - odp_actions_init(out); - return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY); + 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. + * + * The log message mentions 'msg_type'. */ +static int +reject_slave_controller(struct ofconn *ofconn, const const char *msg_type) +{ + if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) { + static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5); + VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller", + msg_type); + + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); + } else { + return 0; } - return 0; } 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->ofproto; 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 = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports); + COVERAGE_INC(ofproto_packet_out); + + error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT"); if (error) { return error; } - opo = (struct ofp_packet_out *) oh; - COVERAGE_INC(ofproto_packet_out); + /* 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; + } + + /* Get payload. */ if (opo->buffer_id != htonl(UINT32_MAX)) { error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id), &buffer, &in_port); @@ -2221,20 +3143,26 @@ handle_packet_out(struct ofproto *p, struct ofconn *ofconn, } payload = *buffer; } else { + payload = request; buffer = NULL; } - flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow); - error = xlate_actions((const union ofp_action *) opo->actions, n_actions, - &flow, p, &payload, &actions, NULL, NULL, 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; } @@ -2250,37 +3178,33 @@ 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_group(p, DP_GROUP_FLOOD); - } if (mask & OFPPC_NO_PACKET_IN) { port->opp.config ^= OFPPC_NO_PACKET_IN; } } static int -handle_port_mod(struct ofproto *p, 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->ofproto; + const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh; struct ofport *port; int 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)) { @@ -2295,7 +3219,7 @@ handle_port_mod(struct ofproto *p, struct ofp_header *oh) } 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; @@ -2308,92 +3232,103 @@ 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); + *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes); queue_tx(msg, ofconn, ofconn->reply_counter); } return ofpbuf_put_uninit(*msgp, nbytes); } -static int -handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn, - struct ofp_stats_request *request) +static struct ofpbuf * +make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len) { - struct ofp_desc_stats *ods; + struct nicira_stats_msg *nsm; struct ofpbuf *msg; - msg = start_stats_reply(request, sizeof *ods); - ods = append_stats_reply(sizeof *ods, ofconn, &msg); - strncpy(ods->mfr_desc, p->manufacturer, sizeof ods->mfr_desc); - strncpy(ods->hw_desc, p->hardware, sizeof ods->hw_desc); - strncpy(ods->sw_desc, p->software, sizeof ods->sw_desc); - strncpy(ods->serial_num, p->serial, sizeof ods->serial_num); - queue_tx(msg, ofconn, ofconn->reply_counter); + 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; +} - return 0; +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 -count_subrules(struct cls_rule *cls_rule, void *n_subrules_) +append_nxstats_reply(size_t nbytes, struct ofconn *ofconn, + struct ofpbuf **msgp) { - struct rule *rule = rule_from_cls_rule(cls_rule); - int *n_subrules = n_subrules_; - - if (rule->super) { - (*n_subrules)++; + 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); + queue_tx(msg, ofconn, ofconn->reply_counter); } + ofpbuf_prealloc_tailroom(*msgp, nbytes); } static int -handle_table_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 ofp_table_stats *ots; + struct ofproto *p = ofconn->ofproto; + struct ofp_desc_stats *ods; struct ofpbuf *msg; - struct odp_stats dpstats; - int n_exact, n_subrules, n_wild; - msg = start_stats_reply(request, sizeof *ots * 2); + 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); - /* 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); + return 0; +} - /* 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 */ +static int +handle_table_stats_request(struct ofconn *ofconn, + const struct ofp_header *request) +{ + struct ofproto *p = ofconn->ofproto; + struct ofp_table_stats *ots; + struct ofpbuf *msg; + + 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 = htonl(OFPFW_ALL); - ots->max_entries = htonl(65536); - ots->active_count = htonl(n_wild); + ots->wildcards = (ofconn->flow_format == 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)); ots->lookup_count = htonll(0); /* XXX */ ots->matched_count = htonll(0); /* XXX */ @@ -2401,81 +3336,96 @@ handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn, return 0; } +static void +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 + * netdev_get_stats() will log errors. */ + netdev_get_stats(port->netdev, &stats); + + 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); +} + static int -handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn, - struct ofp_stats_request *request) +handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) { + struct ofproto *p = ofconn->ofproto; + 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; - msg = start_stats_reply(request, sizeof *ops * 16); - PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { - struct netdev_stats stats; - - /* Intentionally ignore return value, since errors will set 'stats' to - * all-1s, which is correct for OpenFlow, and netdev_get_stats() will - * log errors. */ - netdev_get_stats(port->netdev, &stats); - - ops = append_stats_reply(sizeof *ops, ofconn, &msg); - ops->port_no = htons(odp_port_to_ofp_port(port_no)); - memset(ops->pad, 0, sizeof ops->pad); - ops->rx_packets = htonll(stats.rx_packets); - ops->tx_packets = htonll(stats.tx_packets); - ops->rx_bytes = htonll(stats.rx_bytes); - ops->tx_bytes = htonll(stats.tx_bytes); - ops->rx_dropped = htonll(stats.rx_dropped); - ops->tx_dropped = htonll(stats.tx_dropped); - ops->rx_errors = htonll(stats.rx_errors); - ops->tx_errors = htonll(stats.tx_errors); - ops->rx_frame_err = htonll(stats.rx_frame_errors); - ops->rx_over_err = htonll(stats.rx_over_errors); - ops->rx_crc_err = htonll(stats.rx_crc_errors); - ops->collisions = htonll(stats.collisions); + msg = start_ofp_stats_reply(oh, sizeof *ops * 16); + if (psr->port_no != htons(OFPP_NONE)) { + port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no))); + if (port) { + append_port_stat(port, ofconn, &msg); + } + } else { + HMAP_FOR_EACH (port, hmap_node, &p->ports) { + append_port_stat(port, ofconn, &msg); + } } queue_tx(msg, ofconn, ofconn->reply_counter); 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'. The returned statistics include + * statistics for all of 'rule''s facets. */ static void query_stats(struct ofproto *p, struct rule *rule, uint64_t *packet_countp, uint64_t *byte_countp) { uint64_t packet_count, byte_count; - struct rule *subrule; + struct facet *facet; 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, facets that have expired. */ packet_count = rule->packet_count; byte_count = rule->byte_count; - 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; - } - - packet_count = rule->packet_count; - byte_count = rule->byte_count; + /* Prepare to ask the datapath for statistics on all of the rule's facets. + * + * Also, add any statistics that are not tracked by the datapath for each + * facet. This includes, for example, statistics for packets that were + * executed "by hand" by ofproto via dpif_execute() but must be accounted + * to a rule. */ + odp_flows = xzalloc(list_size(&rule->facets) * sizeof *odp_flows); + n_odp_flows = 0; + LIST_FOR_EACH (facet, list_node, &rule->facets) { + struct odp_flow *odp_flow = &odp_flows[n_odp_flows++]; + odp_flow_key_from_flow(&odp_flow->key, &facet->flow); + packet_count += facet->packet_count; + byte_count += facet->byte_count; + } + + /* 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; @@ -2484,233 +3434,374 @@ query_stats(struct ofproto *p, struct rule *rule, } free(odp_flows); + /* Return the stats to the caller. */ *packet_countp = packet_count; *byte_countp = byte_count; } static void -flow_stats_cb(struct cls_rule *rule_, void *cbdata_) +calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec) +{ + long long int msecs = time_msec() - start; + *sec = htonl(msecs / 1000); + *nsec = htonl((msecs % 1000) * (1000 * 1000)); +} + +static void +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; size_t act_len, len; - 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); + query_stats(ofconn->ofproto, 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, &ofs->match); - ofs->duration = htonl((time_msec() - rule->created) / 1000); + ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match, + rule->flow_cookie, &ofs->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); + if (rule->n_actions > 0) { + memcpy(ofs->actions, rule->actions, act_len); + } +} + +static bool +is_valid_table(uint8_t table_id) +{ + return table_id == 0 || table_id == 0xff; } static int -table_id_to_include(uint8_t table_id) +handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) +{ + const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh); + 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, &ofconn->ofproto->cls, &target); + CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { + put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply); + } + } + queue_tx(reply, ofconn, ofconn->reply_counter); + + return 0; +} + +static void +put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule, + ovs_be16 out_port, struct ofpbuf **replyp) { - return (table_id == TABLEID_HASH ? CLS_INC_EXACT - : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD - : table_id == 0xff ? CLS_INC_ALL - : 0); + 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; + } + + query_stats(ofconn->ofproto, rule, &packet_count, &byte_count); + + act_len = sizeof *rule->actions * rule->n_actions; + + start_len = (*replyp)->size; + append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp); + 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 nx_flow_stats_request *nfsr; struct cls_rule target; + struct ofpbuf *reply; + struct ofpbuf b; + int error; + + ofpbuf_use_const(&b, oh, ntohs(oh->length)); - if (arg_size != sizeof *fsr) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_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(&target, &fsr->match, 0); - 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, &ofconn->ofproto->cls, &target); + CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { + put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply); + } + } + queue_tx(reply, ofconn, ofconn->reply_counter); + 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 ofproto *ofproto, 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_ovs_match(&rule->cr.flow, rule->cr.wc.wildcards, &match); + query_stats(ofproto, rule, &packet_count, &byte_count); ds_put_format(results, "duration=%llds, ", (time_msec() - rule->created) / 1000); 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); + 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 +/* Adds a pretty-printed description of all flows to 'results', including * those marked hidden by secchan (e.g., 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(OFPFW_ALL); - - cbdata.ofproto = p; - cbdata.results = results; + struct cls_cursor cursor; + struct rule *rule; - cls_rule_from_match(&target, &match, 0); - 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(p, 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; + + 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; + + query_stats(ofproto, rule, &packet_count, &byte_count); + + total_packets += packet_count; + total_bytes += byte_count; + n_flows++; + } + } + } - cbdata->packet_count += packet_count; - cbdata->byte_count += byte_count; - cbdata->n_flows++; + oasr->flow_count = htonl(n_flows); + oasr->packet_count = htonll(total_packets); + 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 ofp_aggregate_stats_reply *reply; - struct aggregate_stats_cbdata cbdata; struct cls_rule target; struct ofpbuf *msg; - if (arg_size != sizeof *asr) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH); - } - asr = (struct ofp_aggregate_stats_request *) osr->body; + ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0, + &target); - 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(&target, &asr->match, 0); - 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); + msg = start_ofp_stats_reply(oh, sizeof *reply); + reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg); + query_aggregate_stats(ofconn->ofproto, &target, request->out_port, + request->table_id, reply); queue_tx(msg, ofconn, ofconn->reply_counter); return 0; } static int -handle_stats_request(struct ofproto *p, struct ofconn *ofconn, - struct ofp_header *oh) +handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh) { - struct ofp_stats_request *osr; - size_t arg_size; + struct nx_aggregate_stats_request *request; + struct ofp_aggregate_stats_reply *reply; + struct cls_rule target; + struct ofpbuf b; + struct ofpbuf *buf; int error; - error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr, - 1, &arg_size); + 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; } - osr = (struct ofp_stats_request *) oh; + if (b.size) { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); + } - switch (ntohs(osr->type)) { - case OFPST_DESC: - return handle_desc_stats_request(p, ofconn, osr); + /* Reply. */ + COVERAGE_INC(ofproto_flows_req); + buf = start_nxstats_reply(&request->nsm, sizeof *reply); + reply = ofpbuf_put_uninit(buf, sizeof *reply); + query_aggregate_stats(ofconn->ofproto, &target, request->out_port, + request->table_id, reply); + queue_tx(buf, ofconn, ofconn->reply_counter); - case OFPST_FLOW: - return handle_flow_stats_request(p, ofconn, osr, arg_size); + return 0; +} - case OFPST_AGGREGATE: - return handle_aggregate_stats_request(p, ofconn, osr, arg_size); +struct queue_stats_cbdata { + struct ofconn *ofconn; + struct ofport *ofport; + struct ofpbuf *msg; +}; - case OFPST_TABLE: - return handle_table_stats_request(p, ofconn, osr); +static void +put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id, + const struct netdev_queue_stats *stats) +{ + struct ofp_queue_stats *reply; - case OFPST_PORT: - return handle_port_stats_request(p, ofconn, osr); + 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); +} - case OFPST_VENDOR: - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); +static void +handle_queue_stats_dump_cb(uint32_t queue_id, + struct netdev_queue_stats *stats, + void *cbdata_) +{ + struct queue_stats_cbdata *cbdata = cbdata_; - default: - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT); + put_queue_stats(cbdata, queue_id, stats); +} + +static void +handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id, + struct queue_stats_cbdata *cbdata) +{ + 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; + + if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) { + put_queue_stats(cbdata, queue_id, &stats); + } } } +static int +handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) +{ + struct ofproto *ofproto = ofconn->ofproto; + const struct ofp_queue_stats_request *qsr; + struct queue_stats_cbdata cbdata; + struct ofport *port; + unsigned int port_no; + uint32_t queue_id; + + qsr = ofputil_stats_body(oh); + if (!qsr) { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); + } + + COVERAGE_INC(ofproto_queue_req); + + cbdata.ofconn = ofconn; + cbdata.msg = start_ofp_stats_reply(oh, 128); + + port_no = ntohs(qsr->port_no); + queue_id = ntohl(qsr->queue_id); + if (port_no == OFPP_ALL) { + 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 = get_port(ofproto, ofp_port_to_odp_port(port_no)); + if (port) { + 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); + + return 0; +} + static long long int msec_from_nsec(uint64_t sec, uint32_t nsec) { @@ -2718,428 +3809,571 @@ msec_from_nsec(uint64_t sec, uint32_t nsec) } static void -update_time(struct ofproto *ofproto, struct rule *rule, - const struct odp_flow_stats *stats) +facet_update_time(struct ofproto *ofproto, struct facet *facet, + 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; + if (used > facet->used) { + facet->used = used; + if (used > facet->rule->used) { + facet->rule->used = used; } - netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used); + netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); } } +/* 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 -update_stats(struct ofproto *ofproto, struct rule *rule, - const struct odp_flow_stats *stats) +facet_update_stats(struct ofproto *ofproto, struct facet *facet, + const struct odp_flow_stats *stats) { 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->ip_tos, - stats->tcp_flags); + facet_update_time(ofproto, facet, stats); + facet->packet_count += stats->n_packets; + facet->byte_count += stats->n_bytes; + netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); } } +/* 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 ofconn->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, - struct ofp_flow_mod *ofm, size_t n_actions) +add_flow(struct ofconn *ofconn, struct flow_mod *fm) { + struct ofproto *p = ofconn->ofproto; struct ofpbuf *packet; struct rule *rule; uint16_t in_port; int error; - rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions, - n_actions, ntohs(ofm->idle_timeout), - ntohs(ofm->hard_timeout)); - cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority)); + if (fm->flags & OFPFF_CHECK_OVERLAP + && classifier_rule_overlaps(&p->cls, &fm->cr)) { + return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP); + } - packet = NULL; error = 0; - if (ofm->buffer_id != htonl(UINT32_MAX)) { - error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id), + if (fm->buffer_id != UINT32_MAX) { + error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id, &packet, &in_port); + } else { + packet = NULL; + in_port = UINT16_MAX; } - rule_insert(p, rule, packet, in_port); - ofpbuf_delete(packet); + 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 flow_mod *fm) +{ + return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr)); +} + static int -modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm, - size_t n_actions, uint16_t command, struct rule *rule) +send_buffered_packet(struct ofconn *ofconn, + struct rule *rule, uint32_t buffer_id) { - if (rule_is_hidden(rule)) { + struct ofpbuf *packet; + uint16_t in_port; + int error; + + if (buffer_id == UINT32_MAX) { return 0; } - if (command == OFPFC_DELETE) { - rule_remove(p, rule); - } else { - size_t actions_len = n_actions * sizeof *rule->actions; + error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port); + if (error) { + return error; + } - if (n_actions == rule->n_actions - && !memcmp(ofm->actions, rule->actions, actions_len)) - { - return 0; - } + rule_execute(ofconn->ofproto, rule, in_port, packet); - free(rule->actions); - rule->actions = xmemdup(ofm->actions, actions_len); - rule->n_actions = n_actions; + return 0; +} + +/* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */ - if (rule->cr.wc.wildcards) { - COVERAGE_INC(ofproto_mod_wc_flow); - p->need_revalidate = true; - } else { - rule_update_actions(p, rule); +struct modify_flows_cbdata { + struct ofproto *ofproto; + const struct flow_mod *fm; + struct rule *match; +}; + +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. + * + * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, + * if any. */ +static int +modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm) +{ + struct ofproto *p = ofconn->ofproto; + struct rule *match = NULL; + struct cls_cursor cursor; + struct rule *rule; + + 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); } } - return 0; + if (match) { + /* This credits the packet to whichever flow happened to match last. + * That's weird. Maybe we should do a lookup for the flow that + * actually matches the packet? Who knows. */ + send_buffered_packet(ofconn, match, fm->buffer_id); + return 0; + } else { + return add_flow(ofconn, fm); + } } +/* Implements OFPFC_MODIFY_STRICT. 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 -modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm, - size_t n_actions, uint16_t command) +modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm) { - struct rule *rule; - uint32_t wildcards; - flow_t flow; + struct ofproto *p = ofconn->ofproto; + struct rule *rule = find_flow_strict(p, fm); + if (rule && !rule_is_hidden(rule)) { + modify_flow(p, fm, rule); + return send_buffered_packet(ofconn, rule, fm->buffer_id); + } else { + return add_flow(ofconn, fm); + } +} - flow_from_match(&flow, &wildcards, &ofm->match); - rule = rule_from_cls_rule(classifier_find_rule_exactly( - &p->cls, &flow, wildcards, - ntohs(ofm->priority))); +/* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has + * been identified as a flow in 'p''s flow table to be modified, by changing + * the rule's actions to match those in 'ofm' (which is followed by 'n_actions' + * ofp_action[] structures). */ +static int +modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule) +{ + size_t actions_len = fm->n_actions * sizeof *rule->actions; - if (rule) { - if (command == OFPFC_DELETE - && ofm->out_port != htons(OFPP_NONE) - && !rule_has_out_port(rule, ofm->out_port)) { - return 0; - } + rule->flow_cookie = fm->cookie; - modify_flow(p, ofm, n_actions, command, rule); + /* If the actions are the same, do nothing. */ + 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 = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL; + rule->n_actions = fm->n_actions; + + p->need_revalidate = true; + return 0; } + +/* OFPFC_DELETE implementation. */ -struct modify_flows_cbdata { - struct ofproto *ofproto; - const struct ofp_flow_mod *ofm; - uint16_t out_port; - size_t n_actions; - uint16_t command; -}; +static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port); +/* Implements OFPFC_DELETE. */ static void -modify_flows_cb(struct cls_rule *rule_, void *cbdata_) +delete_flows_loose(struct ofproto *p, const struct flow_mod *fm) { - struct rule *rule = rule_from_cls_rule(rule_); - struct modify_flows_cbdata *cbdata = cbdata_; + struct rule *rule, *next_rule; + struct cls_cursor cursor; - if (cbdata->out_port != htons(OFPP_NONE) - && !rule_has_out_port(rule, cbdata->out_port)) { - return; + 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)); } - - modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, - cbdata->command, rule); } -static int -modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm, - size_t n_actions, uint16_t command) +/* Implements OFPFC_DELETE_STRICT. */ +static void +delete_flow_strict(struct ofproto *p, struct flow_mod *fm) { - struct modify_flows_cbdata cbdata; - struct cls_rule target; + struct rule *rule = find_flow_strict(p, fm); + if (rule) { + delete_flow(p, rule, htons(fm->out_port)); + } +} - cbdata.ofproto = p; - cbdata.ofm = ofm; - cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port - : htons(OFPP_NONE)); - cbdata.n_actions = n_actions; - cbdata.command = command; +/* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has + * been identified as a flow to delete from 'p''s flow table, by deleting the + * flow and sending out a OFPT_FLOW_REMOVED message to any interested + * controller. + * + * Will not delete 'rule' if it is hidden. Will delete 'rule' only if + * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the + * specified 'out_port'. */ +static void +delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port) +{ + if (rule_is_hidden(rule)) { + return; + } - cls_rule_from_match(&target, &ofm->match, 0); + if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) { + return; + } - classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, - modify_flows_cb, &cbdata); - return 0; + 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) { - size_t n_actions; + struct ofproto *p = ofconn->ofproto; + struct flow_mod fm; int error; - error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm, - sizeof *ofm->actions, &n_actions); + error = reject_slave_controller(ofconn, "flow_mod"); if (error) { return error; } - normalize_match(&ofm->match); - if (!ofm->match.wildcards) { - ofm->priority = htons(UINT16_MAX); + error = ofputil_decode_flow_mod(&fm, oh, ofconn->flow_format); + if (error) { + return error; + } + + /* 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); } - 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, ofm, n_actions, OFPFC_MODIFY); + return modify_flows_loose(ofconn, &fm); case OFPFC_MODIFY_STRICT: - return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY); + return modify_flow_strict(ofconn, &fm); case OFPFC_DELETE: - return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE); + delete_flows_loose(p, &fm); + return 0; case OFPFC_DELETE_STRICT: - return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE); + delete_flow_strict(p, &fm); + return 0; default: return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND); } } -static void -send_capability_reply(struct ofproto *p, struct ofconn *ofconn, uint32_t xid) +static int +handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh) { - struct ofmp_capability_reply *ocr; - struct ofpbuf *b; - char capabilities[] = "com.nicira.mgmt.manager=false\n"; + const struct nxt_tun_id_cookie *msg + = (const struct nxt_tun_id_cookie *) oh; - ocr = make_openflow_xid(sizeof(*ocr), OFPT_VENDOR, xid, &b); - ocr->header.header.vendor = htonl(NX_VENDOR_ID); - ocr->header.header.subtype = htonl(NXT_MGMT); - ocr->header.type = htons(OFMPT_CAPABILITY_REPLY); - - ocr->format = htonl(OFMPCOF_SIMPLE); - ocr->mgmt_id = htonll(p->mgmt_id); - - ofpbuf_put(b, capabilities, strlen(capabilities)); - - queue_tx(b, ofconn, ofconn->reply_counter); + ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10; + return 0; } static int -handle_ofmp(struct ofproto *p, struct ofconn *ofconn, - struct ofmp_header *ofmph) +handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh) { - size_t msg_len = ntohs(ofmph->header.header.length); - if (msg_len < sizeof(*ofmph)) { - VLOG_WARN_RL(&rl, "dropping short managment message: %zu\n", msg_len); - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH); + struct nx_role_request *nrr = (struct nx_role_request *) oh; + struct nx_role_request *reply; + struct ofpbuf *buf; + uint32_t role; + + if (ofconn->type != OFCONN_PRIMARY) { + VLOG_WARN_RL(&rl, "ignoring role request on non-controller " + "connection"); + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); } - if (ofmph->type == htons(OFMPT_CAPABILITY_REQUEST)) { - struct ofmp_capability_request *ofmpcr; + role = ntohl(nrr->role); + if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER + && role != NX_ROLE_SLAVE) { + VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role); - if (msg_len < sizeof(struct ofmp_capability_request)) { - VLOG_WARN_RL(&rl, "dropping short capability request: %zu\n", - msg_len); - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH); - } + /* There's no good error code for this. */ + return ofp_mkerr(OFPET_BAD_REQUEST, -1); + } - ofmpcr = (struct ofmp_capability_request *)ofmph; - if (ofmpcr->format != htonl(OFMPCAF_SIMPLE)) { - /* xxx Find a better type than bad subtype */ - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); - } + if (role == NX_ROLE_MASTER) { + struct ofconn *other; - send_capability_reply(p, ofconn, ofmph->header.header.xid); - return 0; - } else { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); + HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) { + if (other->role == NX_ROLE_MASTER) { + other->role = NX_ROLE_SLAVE; + } + } } + ofconn->role = role; + + reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf); + reply->role = htonl(role); + queue_tx(buf, ofconn, ofconn->reply_counter); + + 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; + const struct nxt_set_flow_format *msg + = (const struct nxt_set_flow_format *) oh; + uint32_t format; - if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH); - } - if (ovh->vendor != htonl(NX_VENDOR_ID)) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); - } - if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LENGTH); + format = ntohl(msg->format); + if (format == NXFF_OPENFLOW10 + || format == NXFF_TUN_ID_FROM_COOKIE + || format == NXFF_NXM) { + ofconn->flow_format = format; + return 0; + } else { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); } +} - nh = msg; - switch (ntohl(nh->subtype)) { - case NXT_STATUS_REQUEST: - return switch_status_handle_request(p->switch_status, ofconn->rconn, - msg); - - case NXT_ACT_SET_CONFIG: - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */ - - case NXT_ACT_GET_CONFIG: - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); /* XXX */ - - case NXT_MGMT: - return handle_ofmp(p, ofconn, msg); - } +static int +handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh) +{ + struct ofp_header *ob; + struct ofpbuf *buf; - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); + /* Currently, everything executes synchronously, so we can just + * immediately send the barrier reply. */ + ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf); + queue_tx(buf, ofconn, ofconn->reply_counter); + return 0; } -static void -handle_openflow(struct ofconn *ofconn, struct ofproto *p, - struct ofpbuf *ofp_msg) +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; + error = ofputil_decode_msg_type(oh, &type); + if (error) { + return error; + } - case OFPT_ECHO_REPLY: - error = 0; - break; + switch (ofputil_msg_type_code(type)) { + /* OpenFlow requests. */ + case OFPUTIL_OFPT_ECHO_REQUEST: + return handle_echo_request(ofconn, oh); - case OFPT_FEATURES_REQUEST: - error = handle_features_request(p, ofconn, oh); - break; + case OFPUTIL_OFPT_FEATURES_REQUEST: + return handle_features_request(ofconn, oh); - case OFPT_GET_CONFIG_REQUEST: - error = handle_get_config_request(p, ofconn, oh); - break; + case OFPUTIL_OFPT_GET_CONFIG_REQUEST: + return handle_get_config_request(ofconn, oh); - case OFPT_SET_CONFIG: - error = handle_set_config(p, ofconn, ofp_msg->data); - break; + case OFPUTIL_OFPT_SET_CONFIG: + return handle_set_config(ofconn, msg->data); - case OFPT_PACKET_OUT: - error = handle_packet_out(p, ofconn, ofp_msg->data); - break; + case OFPUTIL_OFPT_PACKET_OUT: + return handle_packet_out(ofconn, oh); - case OFPT_PORT_MOD: - error = handle_port_mod(p, oh); - break; + case OFPUTIL_OFPT_PORT_MOD: + return handle_port_mod(ofconn, oh); - case OFPT_FLOW_MOD: - error = handle_flow_mod(p, ofconn, ofp_msg->data); - break; + case OFPUTIL_OFPT_FLOW_MOD: + return handle_flow_mod(ofconn, oh); - case OFPT_STATS_REQUEST: - error = handle_stats_request(p, ofconn, oh); - break; + case OFPUTIL_OFPT_BARRIER_REQUEST: + return handle_barrier_request(ofconn, oh); - case OFPT_VENDOR: - error = handle_vendor(p, ofconn, ofp_msg->data); - break; + /* OpenFlow replies. */ + case OFPUTIL_OFPT_ECHO_REPLY: + return 0; + /* Nicira extension requests. */ + case OFPUTIL_NXT_STATUS_REQUEST: + return switch_status_handle_request( + ofconn->ofproto->switch_status, ofconn->rconn, oh); + + 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_STATUS_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_msg(struct ofproto *p, struct ofpbuf *packet) +handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) { struct odp_msg *msg = packet->data; - uint16_t in_port = odp_port_to_ofp_port(msg->port); - struct rule *rule; struct ofpbuf payload; - flow_t flow; + struct facet *facet; + struct flow flow; - /* Handle controller actions. */ - if (msg->type == _ODPL_ACTION_NR) { - COVERAGE_INC(ofproto_ctlr_action); - pinsched_send(p->action_sched, in_port, packet, - send_packet_in_action, p); - return; - } + ofpbuf_use_const(&payload, msg + 1, msg->length - sizeof *msg); + flow_extract(&payload, msg->arg, msg->port, &flow); - payload.data = msg + 1; - payload.size = msg->length - sizeof *msg; - flow_extract(&payload, msg->port, &flow); + packet->l2 = payload.l2; + packet->l3 = payload.l3; + packet->l4 = payload.l4; + packet->l7 = payload.l7; /* 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; - - 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); - } - - 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; + struct ofpbuf odp_actions; + + ofpbuf_init(&odp_actions, 32); + nl_msg_put_u32(&odp_actions, ODPAT_OUTPUT, ODPP_LOCAL); + dpif_execute(p->dpif, odp_actions.data, odp_actions.size, &payload); + ofpbuf_uninit(&odp_actions); + } + + 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, 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; + } + } else { + VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, + msg->port); } - } else { - VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, msg->port); - } - COVERAGE_INC(ofproto_packet_in); - pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p); - 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, packet); + return; } - } - rule_execute(p, rule, &payload, &flow); - rule_reinstall(p, rule); + facet = facet_create(p, rule, &flow, 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, packet); + } - if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY - && rconn_is_connected(p->controller->rconn)) { + if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) { /* * Extra-special case for fail-open mode. * @@ -3150,290 +4384,518 @@ handle_odp_msg(struct ofproto *p, struct ofpbuf *packet) * * See the top-level comment in fail-open.c for more information. */ - pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p); - } else { - ofpbuf_delete(packet); + send_packet_in(p, ofpbuf_clone_with_headroom(packet, + DPIF_RECV_MSG_PADDING)); } + + ofpbuf_pull(packet, sizeof *msg); + facet_execute(p, facet, packet); + facet_install(p, facet, false); } - + static void -revalidate_cb(struct cls_rule *sub_, void *cbdata_) +handle_odp_msg(struct ofproto *p, struct ofpbuf *packet) { - struct rule *sub = rule_from_cls_rule(sub_); - struct revalidate_cbdata *cbdata = cbdata_; - - if (cbdata->revalidate_all - || (cbdata->revalidate_subrules && sub->super) - || (tag_set_intersects(&cbdata->revalidate_set, sub->tags))) { - revalidate_rule(cbdata->ofproto, sub); - } -} + struct odp_msg *msg = packet->data; -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; - } - } - - rule_update_actions(p, rule); - return true; -} + switch (msg->type) { + case _ODPL_ACTION_NR: + COVERAGE_INC(ofproto_ctlr_action); + send_packet_in(p, packet); + break; -static struct ofpbuf * -compose_flow_exp(const struct rule *rule, long long int now, uint8_t reason) -{ - struct ofp_flow_expired *ofe; - struct ofpbuf *buf; + case _ODPL_SFLOW_NR: + if (p->sflow) { + ofproto_sflow_received(p->sflow, msg); + } + ofpbuf_delete(packet); + break; - ofe = make_openflow(sizeof *ofe, OFPT_FLOW_EXPIRED, &buf); - flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofe->match); - ofe->priority = htons(rule->cr.priority); - ofe->reason = reason; - ofe->duration = htonl((now - rule->created) / 1000); - ofe->packet_count = htonll(rule->packet_count); - ofe->byte_count = htonll(rule->byte_count); + case _ODPL_MISS_NR: + handle_odp_miss_msg(p, packet); + break; - return buf; + default: + VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32, + msg->type); + break; + } } + +/* Flow expiration. */ -static void -send_flow_exp(struct ofproto *p, struct rule *rule, - long long int now, uint8_t reason) +static int ofproto_dp_max_idle(const struct ofproto *); +static void ofproto_update_used(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 ofconn *ofconn; - struct ofconn *prev; - struct ofpbuf *buf = NULL; + struct rule *rule, *next_rule; + struct cls_cursor cursor; + int dp_max_idle; - /* 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.) */ + /* Update 'used' for each flow in the datapath. */ + ofproto_update_used(ofproto); - prev = NULL; - LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { - if (ofconn->send_flow_exp && rconn_is_connected(ofconn->rconn)) { - if (prev) { - queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter); - } else { - buf = compose_flow_exp(rule, now, reason); - } - prev = ofconn; - } + /* 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); } - if (prev) { - queue_tx(buf, prev, prev->reply_counter); + + /* 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); } + + return MIN(dp_max_idle, 1000); } +/* Update 'used' member of installed facets. */ static void -uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule) +ofproto_update_used(struct ofproto *p) { - assert(rule->installed); - assert(!rule->cr.wc.wildcards); + struct odp_flow *flows; + size_t n_flows; + size_t i; + int error; - if (rule->super) { - rule_remove(ofproto, rule); - } else { - rule_uninstall(ofproto, rule); + 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 facet *facet; + struct flow flow; + + odp_flow_key_to_flow(&f->key, &flow); + facet = facet_find(p, &flow); + + if (facet && facet->installed) { + facet_update_time(p, facet, &f->stats); + facet_account(p, facet, f->stats.n_bytes); + } 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, f); + } + + } + free(flows); } -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_used(), 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_exp(p, rule, now, - (now >= hard_expire - ? OFPER_HARD_TIMEOUT : OFPER_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. */ + /* Get updated flow stats. + * + * XXX We could avoid this call entirely if (1) ofproto_update_used() + * updated TCP flags and (2) the dpif_flow_list_all() in + * ofproto_update_used() zeroed TCP flags. */ memset(&odp_flow, 0, sizeof odp_flow); - if (rule->installed) { - odp_flow.key = rule->cr.flow; + if (facet->installed) { + odp_flow_key_from_flow(&odp_flow.key, &facet->flow); odp_flow.flags = ODPFF_ZERO_TCP_FLAGS; dpif_flow_get(ofproto->dpif, &odp_flow); if (odp_flow.stats.n_packets) { - update_time(ofproto, rule, &odp_flow.stats); - netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos, + facet_update_time(ofproto, facet, &odp_flow.stats); + netflow_flow_update_flags(&facet->nf_flow, odp_flow.stats.tcp_flags); } } - expired.flow = rule->cr.flow; - expired.packet_count = rule->packet_count + + expired.flow = facet->flow; + expired.packet_count = facet->packet_count + odp_flow.stats.n_packets; - expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes; - expired.used = rule->used; + expired.byte_count = facet->byte_count + odp_flow.stats.n_bytes; + expired.used = facet->used; - netflow_expire(ofproto->netflow, &rule->nf_flow, &expired); + netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); + } +} - /* Schedule us to send the accumulated records once we have - * collected all of them. */ - poll_immediate_wake(); +static void +ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle) +{ + long long int cutoff = time_msec() - dp_max_idle; + struct facet *facet, *next_facet; + + HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { + facet_active_timeout(ofproto, facet); + if (facet->used < cutoff) { + facet_remove(ofproto, facet); + } } } +/* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, + * then delete it entirely. */ static void -update_used(struct ofproto *p) +rule_expire(struct ofproto *ofproto, struct rule *rule) { - struct odp_flow *flows; - size_t n_flows; - size_t i; - int error; + struct facet *facet, *next_facet; + long long int now; + uint8_t reason; - error = dpif_flow_list_all(p->dpif, &flows, &n_flows); - if (error) { + /* 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 { return; } - for (i = 0; i < n_flows; i++) { - struct odp_flow *f = &flows[i]; - struct rule *rule; + COVERAGE_INC(ofproto_expired); - 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; - } + /* 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); + } - update_time(p, rule, &f->stats); - rule_account(p, rule, f->stats.n_bytes); + /* Get rid of the rule. */ + if (!rule_is_hidden(rule)) { + rule_send_removed(ofproto, rule, reason); } - free(flows); + rule_remove(ofproto, rule); } + +static struct ofpbuf * +compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule, + uint8_t reason) +{ + struct ofp_flow_removed *ofr; + struct ofpbuf *buf; -static void -do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id, - const struct ofpbuf *packet, int send_len) + ofr = make_openflow_xid(sizeof *ofr, OFPT_FLOW_REMOVED, htonl(0), &buf); + ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofr->match, + rule->flow_cookie, &ofr->cookie); + ofr->priority = htons(rule->cr.priority); + ofr->reason = reason; + calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec); + ofr->idle_timeout = htons(rule->idle_timeout); + ofr->packet_count = htonll(rule->packet_count); + ofr->byte_count = htonll(rule->byte_count); + + return buf; +} + +static struct ofpbuf * +compose_nx_flow_removed(const struct rule *rule, uint8_t reason) { - struct odp_msg *msg = packet->data; - struct ofpbuf payload; - struct ofpbuf *opi; - uint8_t reason; + struct nx_flow_removed *nfr; + struct ofpbuf *buf; + int match_len; - /* Extract packet payload from 'msg'. */ - payload.data = msg + 1; - payload.size = msg->length - sizeof *msg; + make_nxmsg_xid(sizeof *nfr, NXT_FLOW_REMOVED, htonl(0), &buf); + match_len = nx_put_match(buf, &rule->cr); - /* Construct ofp_packet_in message. */ - reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH; - opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason, - &payload, send_len); + nfr = buf->data; + nfr->cookie = rule->flow_cookie; + nfr->priority = htons(rule->cr.priority); + nfr->reason = reason; + calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec); + nfr->idle_timeout = htons(rule->idle_timeout); + nfr->match_len = htons(match_len); + nfr->packet_count = htonll(rule->packet_count); + nfr->byte_count = htonll(rule->byte_count); - /* Send. */ - rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100); + return buf; } static void -send_packet_in_action(struct ofpbuf *packet, void *p_) +rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason) { - struct ofproto *p = p_; struct ofconn *ofconn; - struct odp_msg *msg; - msg = packet->data; - LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { - if (ofconn == p->controller || ofconn->miss_send_len) { - do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg); + if (!rule->send_flow_removed) { + return; + } + + LIST_FOR_EACH (ofconn, node, &p->all_conns) { + struct ofpbuf *msg; + + if (!rconn_is_connected(ofconn->rconn) + || !ofconn_receives_async_msgs(ofconn)) { + continue; } + + msg = (ofconn->flow_format == NXFF_NXM + ? compose_nx_flow_removed(rule, reason) + : compose_ofp_flow_removed(ofconn, rule, reason)); + + /* Account flow expirations under ofconn->reply_counter, the counter + * for replies to OpenFlow requests. 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.) */ + queue_tx(msg, ofconn, ofconn->reply_counter); } - ofpbuf_delete(packet); } +/* pinsched callback for sending 'packet' on 'ofconn'. */ static void -send_packet_in_miss(struct ofpbuf *packet, void *p_) +do_send_packet_in(struct ofpbuf *packet, void *ofconn_) { - struct ofproto *p = p_; - bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open); - struct ofconn *ofconn; - struct ofpbuf payload; - struct odp_msg *msg; + struct ofconn *ofconn = ofconn_; + + rconn_send_with_limit(ofconn->rconn, packet, + ofconn->packet_in_counter, 100); +} + +/* 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; + } else { + struct ofpbuf payload; + + ofpbuf_use_const(&payload, opi->data, + packet->size - offsetof(struct ofp_packet_in, data)); + buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port); + } + + /* 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); + } + 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; + } + + /* Update packet headers. */ + opi->buffer_id = htonl(buffer_id); + update_openflow_length(packet); + + /* 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); +} + +/* 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) +{ + 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; - msg = packet->data; - payload.data = msg + 1; - payload.size = msg->length - sizeof *msg; - LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { - if (ofconn->miss_send_len) { - struct pktbuf *pb = ofconn->pktbuf; - uint32_t buffer_id = (in_fail_open - ? pktbuf_get_null() - : pktbuf_save(pb, &payload, msg->port)); - int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len - : UINT32_MAX); - do_send_packet_in(ofconn, buffer_id, packet, send_len); + /* 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); + + /* 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; + + return max_len; +} + +/* 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. + * + * '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'. */ +static void +send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet) +{ + struct ofconn *ofconn, *prev; + int max_len; + + max_len = do_convert_to_packet_in(packet); + + prev = NULL; + LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { + if (ofconn_receives_async_msgs(ofconn)) { + if (prev) { + schedule_packet_in(prev, packet, max_len, true); + } + prev = ofconn; } } - ofpbuf_delete(packet); + if (prev) { + schedule_packet_in(prev, packet, max_len, false); + } else { + ofpbuf_delete(packet); + } } static uint64_t @@ -3441,7 +4903,7 @@ 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; @@ -3460,16 +4922,181 @@ static uint64_t pick_fallback_dpid(void) { uint8_t ea[ETH_ADDR_LEN]; - eth_addr_random(ea); - ea[0] = 0x00; /* Set Nicira OUI. */ - ea[1] = 0x23; - ea[2] = 0x20; + eth_addr_nicira_random(ea); return eth_addr_to_uint64(ea); } +static 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, const 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_be32 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 = ntohl(strtoul(tun_id_s, NULL, 10)); + 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_; @@ -3483,7 +5110,8 @@ 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); + 0, flow->in_port, + GRAT_ARP_LOCK_NONE); if (rev_tag) { /* The log messages here could actually be useful in debugging, * so keep the rate limit relatively high. */ @@ -3495,11 +5123,13 @@ default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, } /* Determine output port. */ - out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags); + 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); + flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD, + nf_output_iface, odp_actions); } else if (out_port != flow->in_port) { - odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port; + nl_msg_put_u32(odp_actions, ODPAT_OUTPUT, out_port); *nf_output_iface = out_port; } else { /* Drop. */ @@ -3509,7 +5139,6 @@ default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, } static const struct ofhooks default_ofhooks = { - NULL, default_normal_ofhook_cb, NULL, NULL