X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=ofproto%2Fofproto.c;h=cd87cd0d55c2040324afa7f78f40f1ab5d6a8526;hb=f1defbf96df61a1fce8ec8ca63a56018bff4c940;hp=dcf8683ff37ea007a41d75e1fc375c2b1e0e71e6;hpb=e2bfacb6e490bb571f2fbd243c7a2d4658149fba;p=sliver-openvswitch.git diff --git a/ofproto/ofproto.c b/ofproto/ofproto.c index dcf8683ff..cd87cd0d5 100644 --- a/ofproto/ofproto.c +++ b/ofproto/ofproto.c @@ -19,22 +19,28 @@ #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 "netdev.h" #include "netflow.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" @@ -44,39 +50,33 @@ #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" -#include "sflow_api.h" +VLOG_DEFINE_THIS_MODULE(ofproto); -enum { - TABLEID_HASH = 0, - TABLEID_CLASSIFIER = 1 -}; +#include "sflow_api.h" 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 flow *, struct ofproto *, const struct ofpbuf *packet, struct odp_actions *out, tag_type *tags, bool *may_set_up_flow, uint16_t *nf_output_iface); @@ -84,8 +84,7 @@ static int xlate_actions(const union ofp_action *in, size_t n_in, struct rule { struct cls_rule cr; - uint64_t flow_cookie; /* Controller-issued identifier. - (Kept in network-byte order.) */ + ovs_be64 flow_cookie; /* Controller-issued identifier. */ 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? */ @@ -150,7 +149,7 @@ rule_is_hidden(const struct rule *rule) static struct rule *rule_create(struct ofproto *, struct rule *super, const union ofp_action *, size_t n_actions, uint16_t idle_timeout, uint16_t hard_timeout, - uint64_t flow_cookie, bool send_flow_removed); + ovs_be64 flow_cookie, bool send_flow_removed); static void rule_free(struct rule *); static void rule_destroy(struct ofproto *, struct rule *); static struct rule *rule_from_cls_rule(const struct cls_rule *); @@ -163,30 +162,55 @@ static void rule_install(struct ofproto *, struct rule *, struct rule *displaced_rule); static void rule_uninstall(struct ofproto *, struct rule *); static void rule_post_uninstall(struct ofproto *, struct rule *); -static void send_flow_removed(struct ofproto *p, struct rule *rule, - long long int now, uint8_t reason); +static void send_flow_removed(struct ofproto *, struct rule *, uint8_t reason); /* ofproto supports two kinds of OpenFlow connections: * - * - "Controller connections": Connections to ordinary OpenFlow controllers. - * ofproto maintains persistent connections to these controllers and by - * default sends them asynchronous messages such as packet-ins. + * - "Primary" connections to ordinary OpenFlow controllers. ofproto + * maintains persistent connections to these controllers and by default + * sends them asynchronous messages such as packet-ins. * - * - "Transient connections", e.g. from ovs-ofctl. When these connections + * - "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_CONTROLLER, /* An OpenFlow controller. */ - OFCONN_TRANSIENT /* A transient connection. */ + 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. */ + int flow_format; /* One of NXFF_*. */ /* OFPT_PACKET_IN related data. */ struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */ @@ -200,7 +224,7 @@ struct ofconn { #define OFCONN_REPLY_MAX 100 struct rconn_packet_counter *reply_counter; - /* type == OFCONN_CONTROLLER only. */ + /* 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. */ @@ -221,9 +245,11 @@ 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); @@ -244,8 +270,7 @@ struct ofproto { /* Datapath. */ struct dpif *dpif; struct netdev_monitor *netdev_monitor; - struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is - * OFP port nr. */ + struct hmap ports; /* Contains "struct ofport"s. */ struct shash port_by_name; uint32_t max_ports; @@ -266,13 +291,14 @@ struct ofproto { bool need_revalidate; long long int next_expiration; struct tag_set revalidate_set; - bool tun_id_from_cookie; /* OpenFlow connections. */ struct hmap controllers; /* Controller "struct ofconn"s. */ struct list all_conns; /* Contains "struct ofconn"s. */ - struct pvconn **listeners; - size_t n_listeners; + enum ofproto_fail_mode fail_mode; + + /* OpenFlow listeners. */ + struct hmap services; /* Contains "struct ofservice"s. */ struct pvconn **snoops; size_t n_snoops; @@ -291,21 +317,18 @@ static const struct ofhooks default_ofhooks; static uint64_t pick_datapath_id(const struct ofproto *); static uint64_t pick_fallback_dpid(void); -static void update_used(struct ofproto *); +static int ofproto_expire(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 void handle_odp_msg(struct ofproto *, struct ofpbuf *); -static void handle_openflow(struct ofconn *, struct ofproto *, - struct ofpbuf *); - -static void refresh_port_groups(struct ofproto *); +static void handle_openflow(struct ofconn *, struct ofpbuf *); +static struct ofport *get_port(const struct ofproto *, uint16_t odp_port); static void update_port(struct ofproto *, const char *devname); static int init_ports(struct ofproto *); static void reinit_ports(struct ofproto *); @@ -358,7 +381,7 @@ ofproto_create(const char *datapath, const char *datapath_type, /* Initialize datapath. */ p->dpif = dpif; p->netdev_monitor = netdev_monitor_create(); - port_array_init(&p->ports); + hmap_init(&p->ports); shash_init(&p->port_by_name); p->max_ports = stats.max_ports; @@ -378,8 +401,7 @@ ofproto_create(const char *datapath, const char *datapath_type, /* Initialize OpenFlow connections. */ list_init(&p->all_conns); hmap_init(&p->controllers); - p->listeners = NULL; - p->n_listeners = 0; + hmap_init(&p->services); p->snoops = NULL; p->n_snoops = 0; @@ -408,15 +430,11 @@ ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id) uint64_t old_dpid = p->datapath_id; p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p); if (p->datapath_id != old_dpid) { - struct ofconn *ofconn; - 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. */ - LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { - rconn_reconnect(ofconn->rconn); - } + ofproto_reconnect_controllers(p); } } @@ -452,13 +470,15 @@ add_controller(struct ofproto *ofproto, const struct ofproto_controller *c) discovery = NULL; } - ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_CONTROLLER); + 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 { - rconn_connect(ofconn->rconn, c->target); + 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)); @@ -471,9 +491,7 @@ add_controller(struct ofproto *ofproto, const struct ofproto_controller *c) static void update_controller(struct ofconn *ofconn, const struct ofproto_controller *c) { - struct ofproto *ofproto = ofconn->ofproto; int probe_interval; - int i; ofconn->band = (is_in_band_controller(c) ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND); @@ -489,27 +507,13 @@ update_controller(struct ofconn *ofconn, const struct ofproto_controller *c) discovery_set_accept_controller_re(ofconn->discovery, c->accept_re); } - for (i = 0; i < N_SCHEDULERS; i++) { - struct pinsched **s = &ofconn->schedulers[i]; - - if (c->rate_limit > 0) { - if (!*s) { - *s = pinsched_create(c->rate_limit, c->burst_limit, - ofproto->switch_status); - } else { - pinsched_set_limits(*s, c->rate_limit, c->burst_limit); - } - } else { - pinsched_destroy(*s); - *s = NULL; - } - } + ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit); } static const char * ofconn_get_target(const struct ofconn *ofconn) { - return ofconn->discovery ? "discover" : rconn_get_name(ofconn->rconn); + return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn); } static struct ofconn * @@ -517,7 +521,7 @@ find_controller_by_target(struct ofproto *ofproto, const char *target) { struct ofconn *ofconn; - HMAP_FOR_EACH_WITH_HASH (ofconn, struct ofconn, hmap_node, + HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node, hash_string(target, 0), &ofproto->controllers) { if (!strcmp(ofconn_get_target(ofconn), target)) { return ofconn; @@ -542,9 +546,13 @@ update_in_band_remotes(struct ofproto *ofproto) /* Add all the remotes. */ discovery = false; - HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &ofproto->controllers) { + 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); @@ -568,7 +576,9 @@ update_in_band_remotes(struct ofproto *ofproto) in_band_create(ofproto, ofproto->dpif, ofproto->switch_status, &ofproto->in_band); } - in_band_set_remotes(ofproto->in_band, addrs, n_addrs); + if (ofproto->in_band) { + in_band_set_remotes(ofproto->in_band, addrs, n_addrs); + } ofproto->next_in_band_update = time_msec() + 1000; } else { in_band_destroy(ofproto->in_band); @@ -579,31 +589,72 @@ update_in_band_remotes(struct ofproto *ofproto) free(addrs); } +static void +update_fail_open(struct ofproto *p) +{ + struct ofconn *ofconn; + + if (!hmap_is_empty(&p->controllers) + && p->fail_mode == OFPROTO_FAIL_STANDALONE) { + struct rconn **rconns; + size_t n; + + if (!p->fail_open) { + p->fail_open = fail_open_create(p, p->switch_status); + } + + n = 0; + rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns); + HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) { + rconns[n++] = ofconn->rconn; + } + + fail_open_set_controllers(p->fail_open, rconns, n); + /* p->fail_open takes ownership of 'rconns'. */ + } else { + fail_open_destroy(p->fail_open); + p->fail_open = NULL; + } +} + void ofproto_set_controllers(struct ofproto *p, const struct ofproto_controller *controllers, size_t n_controllers) { struct shash new_controllers; - enum ofproto_fail_mode fail_mode; - struct ofconn *ofconn, *next; + 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]; - shash_add_once(&new_controllers, c->target, &controllers[i]); - if (!find_controller_by_target(p, c->target)) { - add_controller(p, c); + if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) { + if (!find_controller_by_target(p, c->target)) { + add_controller(p, c); + } + } else if (!pvconn_verify_name(c->target)) { + if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) { + continue; + } + } else { + VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"", + dpif_name(p->dpif), c->target); + continue; } + + shash_add_once(&new_controllers, c->target, &controllers[i]); } - fail_mode = OFPROTO_FAIL_STANDALONE; + /* Delete controllers that are no longer configured. + * Update configuration of all now-existing controllers. */ ss_exists = false; - HMAP_FOR_EACH_SAFE (ofconn, next, struct ofconn, hmap_node, - &p->controllers) { + 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)); @@ -614,36 +665,27 @@ ofproto_set_controllers(struct ofproto *p, if (ofconn->ss) { ss_exists = true; } - if (c->fail == OFPROTO_FAIL_SECURE) { - fail_mode = OFPROTO_FAIL_SECURE; - } } } - shash_destroy(&new_controllers); - - update_in_band_remotes(p); - if (!hmap_is_empty(&p->controllers) - && fail_mode == OFPROTO_FAIL_STANDALONE) { - struct rconn **rconns; - size_t n; + /* 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; - if (!p->fail_open) { - p->fail_open = fail_open_create(p, p->switch_status); + c = shash_find_data(&new_controllers, + pvconn_get_name(ofservice->pvconn)); + if (!c) { + ofservice_destroy(p, ofservice); + } else { + ofservice_reconfigure(ofservice, c); } + } - n = 0; - rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns); - HMAP_FOR_EACH (ofconn, struct ofconn, hmap_node, &p->controllers) { - rconns[n++] = ofconn->rconn; - } + shash_destroy(&new_controllers); - 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; - } + update_in_band_remotes(p); + update_fail_open(p); if (!hmap_is_empty(&p->controllers) && !ss_exists) { ofconn = CONTAINER_OF(hmap_first(&p->controllers), @@ -653,6 +695,25 @@ ofproto_set_controllers(struct ofproto *p, } } +void +ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode) +{ + 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) @@ -783,12 +844,6 @@ 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) { @@ -819,12 +874,10 @@ ofproto_set_sflow(struct ofproto *ofproto, if (oso) { if (!os) { struct ofport *ofport; - unsigned int odp_port; os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif); - refresh_port_groups(ofproto); - PORT_ARRAY_FOR_EACH (ofport, &ofproto->ports, odp_port) { - ofproto_sflow_add_port(os, odp_port, + HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) { + ofproto_sflow_add_port(os, ofport->odp_port, netdev_get_name(ofport->netdev)); } } @@ -835,18 +888,6 @@ ofproto_set_sflow(struct ofproto *ofproto, } } -int -ofproto_set_stp(struct ofproto *ofproto OVS_UNUSED, bool enable_stp) -{ - /* XXX */ - if (enable_stp) { - VLOG_WARN("STP is not yet implemented"); - return EINVAL; - } else { - return 0; - } -} - uint64_t ofproto_get_datapath_id(const struct ofproto *ofproto) { @@ -854,19 +895,15 @@ ofproto_get_datapath_id(const struct ofproto *ofproto) } bool -ofproto_has_controller(const struct ofproto *ofproto) +ofproto_has_primary_controller(const struct ofproto *ofproto) { return !hmap_is_empty(&ofproto->controllers); } -void -ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners) +enum ofproto_fail_mode +ofproto_get_fail_mode(const struct ofproto *p) { - 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 @@ -882,9 +919,9 @@ 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) { @@ -902,15 +939,15 @@ ofproto_destroy(struct ofproto *p) ofproto_flush_flows(p); classifier_destroy(&p->cls); - 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); @@ -919,10 +956,10 @@ ofproto_destroy(struct ofproto *p) netflow_destroy(p->netflow); ofproto_sflow_destroy(p->sflow); - 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]); @@ -937,7 +974,7 @@ ofproto_destroy(struct ofproto *p) free(p->serial_desc); free(p->dp_desc); - port_array_destroy(&p->ports); + hmap_destroy(&p->ports); free(p); } @@ -991,8 +1028,8 @@ add_snooper(struct ofproto *ofproto, struct vconn *vconn) /* Pick a controller for monitoring. */ best = NULL; - LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) { - if (ofconn->type == OFCONN_CONTROLLER + LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { + if (ofconn->type == OFCONN_PRIMARY && (!best || snoop_preference(ofconn) > snoop_preference(best))) { best = ofconn; } @@ -1010,6 +1047,7 @@ int ofproto_run1(struct ofproto *p) { struct ofconn *ofconn, *next_ofconn; + struct ofservice *ofservice; char *devname; int error; int i; @@ -1020,7 +1058,6 @@ ofproto_run1(struct ofproto *p) for (i = 0; i < 50; i++) { struct ofpbuf *buf; - int error; error = dpif_recv(p->dpif, &buf); if (error) { @@ -1054,9 +1091,8 @@ ofproto_run1(struct ofproto *p) in_band_run(p->in_band); } - 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 @@ -1065,14 +1101,23 @@ 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), - OFCONN_TRANSIENT); + 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)); } @@ -1091,19 +1136,9 @@ ofproto_run1(struct ofproto *p) } 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) { @@ -1145,17 +1180,18 @@ 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(p->next_in_band_update - time_msec()); + poll_timer_wait_until(p->next_in_band_update); in_band_wait(p->in_band); } if (p->fail_open) { @@ -1172,10 +1208,10 @@ 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]); @@ -1200,8 +1236,50 @@ ofproto_is_alive(const struct ofproto *p) 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_port_del(struct ofproto *ofproto, uint16_t odp_port) +{ + struct ofport *ofport = get_port(ofproto, odp_port); + const char *name = ofport ? (char *) ofport->opp.name : ""; + int error; + + error = dpif_port_del(ofproto->dpif, odp_port); + if (error) { + VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)", + dpif_name(ofproto->dpif), odp_port, name, strerror(error)); + } else if (ofport) { + /* 'name' is ofport->opp.name and update_port() is going to destroy + * 'ofport'. Just in case update_port() refers to 'name' after it + * destroys 'ofport', make a copy of it around the update_port() + * call. */ + char *devname = xstrdup(name); + update_port(ofproto, devname); + free(devname); + } + return error; +} + +/* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns + * true if 'odp_port' exists and should be included, false otherwise. */ +bool +ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port) +{ + struct ofport *ofport = get_port(ofproto, odp_port); + return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD); +} + int -ofproto_send_packet(struct ofproto *p, const flow_t *flow, +ofproto_send_packet(struct ofproto *p, const struct flow *flow, const union ofp_action *actions, size_t n_actions, const struct ofpbuf *packet) { @@ -1216,34 +1294,30 @@ ofproto_send_packet(struct ofproto *p, const flow_t *flow, /* 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.actions, odp_actions.n_actions, packet); return 0; } void -ofproto_add_flow(struct ofproto *p, - const flow_t *flow, uint32_t wildcards, unsigned int priority, +ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule, const union ofp_action *actions, size_t n_actions, int idle_timeout) { struct rule *rule; rule = rule_create(p, NULL, actions, n_actions, - idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, + idle_timeout >= 0 ? idle_timeout : 5 /* XXX */, 0, 0, false); - cls_rule_from_flow(flow, wildcards, priority, &rule->cr); + rule->cr = *cls_rule; rule_insert(p, rule, NULL, 0); } 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); } @@ -1283,13 +1357,14 @@ reinit_ports(struct ofproto *p) { struct svec devnames; struct ofport *ofport; - unsigned int port_no; struct odp_port *odp_ports; size_t n_odp_ports; size_t i; + COVERAGE_INC(ofproto_reinit_ports); + svec_init(&devnames); - PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) { + HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { svec_add (&devnames, (char *) ofport->opp.name); } dpif_port_list(p->dpif, &odp_ports, &n_odp_ports); @@ -1305,39 +1380,6 @@ reinit_ports(struct ofproto *p) svec_destroy(&devnames); } -static size_t -refresh_port_group(struct ofproto *p, unsigned int group) -{ - uint16_t *ports; - size_t n_ports; - struct ofport *port; - unsigned int port_no; - - assert(group == DP_GROUP_ALL || group == DP_GROUP_FLOOD); - - ports = xmalloc(port_array_count(&p->ports) * sizeof *ports); - n_ports = 0; - PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { - if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) { - ports[n_ports++] = port_no; - } - } - dpif_port_group_set(p->dpif, group, ports, n_ports); - free(ports); - - return n_ports; -} - -static void -refresh_port_groups(struct ofproto *p) -{ - size_t n_flood = refresh_port_group(p, DP_GROUP_FLOOD); - size_t n_all = refresh_port_group(p, DP_GROUP_ALL); - if (p->sflow) { - ofproto_sflow_set_group_sizes(p->sflow, n_flood, n_all); - } -} - static struct ofport * make_ofport(const struct odp_port *odp_port) { @@ -1345,13 +1387,11 @@ make_ofport(const struct odp_port *odp_port) enum netdev_flags flags; struct ofport *ofport; struct netdev *netdev; - bool carrier; int error; memset(&netdev_options, 0, sizeof netdev_options); netdev_options.name = odp_port->devname; netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; - netdev_options.may_open = true; error = netdev_open(&netdev_options, &netdev); if (error) { @@ -1364,6 +1404,7 @@ 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, @@ -1373,8 +1414,7 @@ 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, @@ -1385,7 +1425,7 @@ 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; @@ -1422,7 +1462,7 @@ 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; @@ -1436,36 +1476,30 @@ send_port_status(struct ofproto *p, const struct ofport *ofport, hton_ofp_phy_port(&ops->desc); queue_tx(b, ofconn, NULL); } - if (p->ofhooks->port_changed_cb) { - p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux); - } } static void ofport_install(struct ofproto *p, struct ofport *ofport) { - uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no); const char *netdev_name = (const char *) ofport->opp.name; netdev_monitor_add(p->netdev_monitor, ofport->netdev); - port_array_set(&p->ports, odp_port, ofport); + hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0)); shash_add(&p->port_by_name, netdev_name, ofport); if (p->sflow) { - ofproto_sflow_add_port(p->sflow, odp_port, netdev_name); + ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name); } } static void ofport_remove(struct ofproto *p, struct ofport *ofport) { - uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no); - netdev_monitor_remove(p->netdev_monitor, ofport->netdev); - port_array_set(&p->ports, odp_port, NULL); + hmap_remove(&p->ports, &ofport->hmap_node); shash_delete(&p->port_by_name, shash_find(&p->port_by_name, (char *) ofport->opp.name)); if (p->sflow) { - ofproto_sflow_del_port(p->sflow, odp_port); + ofproto_sflow_del_port(p->sflow, ofport->odp_port); } } @@ -1478,6 +1512,20 @@ 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) { @@ -1505,7 +1553,7 @@ 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 " @@ -1545,9 +1593,6 @@ 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 @@ -1573,7 +1618,6 @@ init_ports(struct ofproto *p) } } free(ports); - refresh_port_groups(p); return 0; } @@ -1585,6 +1629,7 @@ ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type) 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; @@ -1596,7 +1641,7 @@ ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type) static void ofconn_destroy(struct ofconn *ofconn) { - if (ofconn->type == OFCONN_CONTROLLER) { + if (ofconn->type == OFCONN_PRIMARY) { hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node); } discovery_destroy(ofconn->discovery); @@ -1611,8 +1656,9 @@ 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; @@ -1623,7 +1669,9 @@ ofconn_run(struct ofconn *ofconn, struct ofproto *p) } if (discovery_run(ofconn->discovery, &controller_name)) { if (controller_name) { - rconn_connect(ofconn->rconn, 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); } @@ -1647,7 +1695,7 @@ 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); } } @@ -1680,17 +1728,107 @@ ofconn_wait(struct ofconn *ofconn) static bool ofconn_receives_async_msgs(const struct ofconn *ofconn) { - if (ofconn->type == OFCONN_CONTROLLER) { - /* Ordinary controllers always get asynchronous messages unless they + 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 { - /* Transient connections don't get asynchronous messages unless they - * have explicitly asked for them by setting a nonzero miss send - * length. */ + /* 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. */ @@ -1698,7 +1836,7 @@ static struct rule * rule_create(struct ofproto *ofproto, struct rule *super, const union ofp_action *actions, size_t n_actions, uint16_t idle_timeout, uint16_t hard_timeout, - uint64_t flow_cookie, bool send_flow_removed) + ovs_be64 flow_cookie, bool send_flow_removed) { struct rule *rule = xzalloc(sizeof *rule); rule->idle_timeout = idle_timeout; @@ -1712,8 +1850,10 @@ rule_create(struct ofproto *ofproto, struct rule *super, } else { list_init(&rule->list); } - rule->n_actions = n_actions; - rule->actions = xmemdup(actions, n_actions * sizeof *actions); + if (n_actions > 0) { + 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); @@ -1747,7 +1887,7 @@ 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) { + LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) { revalidate_rule(ofproto, subrule); } } else { @@ -1757,7 +1897,7 @@ rule_destroy(struct ofproto *ofproto, struct rule *rule) } 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; @@ -1767,15 +1907,50 @@ 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, within 'ofproto', the 'n_actions' actions in 'actions' on + * 'packet', which arrived on 'in_port'. + * + * Takes ownership of 'packet'. */ +static bool +execute_odp_actions(struct ofproto *ofproto, uint16_t in_port, + const union odp_action *actions, size_t n_actions, + struct ofpbuf *packet) +{ + if (n_actions == 1 && actions[0].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; + + msg = ofpbuf_push_uninit(packet, sizeof *msg); + msg->type = _ODPL_ACTION_NR; + msg->length = sizeof(struct odp_msg) + packet->size; + msg->port = in_port; + msg->reserved = 0; + msg->arg = actions[0].controller.arg; + + send_packet_in(ofproto, packet); + + return true; + } else { + int error; + + error = dpif_execute(ofproto->dpif, actions, n_actions, packet); + ofpbuf_delete(packet); + return !error; + } +} + /* Executes the actions indicated by 'rule' on 'packet', which is in flow - * 'flow' and is considered to have arrived on ODP port 'in_port'. + * 'flow' and is considered to have arrived on ODP port 'in_port'. 'packet' + * must have at least sizeof(struct ofp_packet_in) bytes of headroom. * * 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, @@ -1787,15 +1962,20 @@ rule_has_out_port(const struct rule *rule, uint16_t out_port) * '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'. */ + * actions and apply them to 'packet'. + * + * Takes ownership of 'packet'. */ static void rule_execute(struct ofproto *ofproto, struct rule *rule, - struct ofpbuf *packet, const flow_t *flow) + struct ofpbuf *packet, const struct flow *flow) { const union odp_action *actions; + struct odp_flow_stats stats; size_t n_actions; struct odp_actions a; + assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); + /* Grab or compose the ODP actions. * * The special case for an exact-match 'rule' where 'flow' is not the @@ -1806,6 +1986,7 @@ rule_execute(struct ofproto *ofproto, struct rule *rule, struct rule *super = rule->super ? rule->super : rule; if (xlate_actions(super->actions, super->n_actions, flow, ofproto, packet, &a, NULL, 0, NULL)) { + ofpbuf_delete(packet); return; } actions = a.actions; @@ -1816,16 +1997,21 @@ rule_execute(struct ofproto *ofproto, struct rule *rule, } /* 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); + flow_extract_stats(flow, packet, &stats); + if (execute_odp_actions(ofproto, flow->in_port, + actions, n_actions, packet)) { update_stats(ofproto, rule, &stats); rule->used = time_msec(); netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, rule->used); } } +/* Inserts 'rule' into 'p''s flow table. + * + * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s + * actions on it and credits the statistics for sending the packet to 'rule'. + * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of + * headroom. */ static void rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet, uint16_t in_port) @@ -1840,7 +2026,7 @@ rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet, /* Send the packet and credit it to the rule. */ if (packet) { - flow_t flow; + struct flow flow; flow_extract(packet, 0, in_port, &flow); rule_execute(p, rule, packet, &flow); } @@ -1862,19 +2048,35 @@ rule_insert(struct ofproto *p, struct rule *rule, struct ofpbuf *packet, static struct rule * rule_create_subrule(struct ofproto *ofproto, struct rule *rule, - const flow_t *flow) + const struct flow *flow) { struct rule *subrule = rule_create(ofproto, rule, NULL, 0, rule->idle_timeout, rule->hard_timeout, 0, false); COVERAGE_INC(ofproto_subrule_create); - cls_rule_from_flow(flow, 0, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX - : rule->cr.priority), &subrule->cr); - classifier_insert_exact(&ofproto->cls, &subrule->cr); + cls_rule_init_exact(flow, (rule->cr.priority <= UINT16_MAX ? UINT16_MAX + : rule->cr.priority), + &subrule->cr); + + if (classifier_insert(&ofproto->cls, &subrule->cr)) { + /* Can't happen, */ + NOT_REACHED(); + } return subrule; } +/* Remove 'rule' from 'ofproto' and free up the associated memory: + * + * - If 'rule' was installed in the datapath, uninstalls it and updates + * 'rule''s statistics (or its super-rule's statistics, if it is a + * subrule), via rule_uninstall(). + * + * - Removes 'rule' from the classifier. + * + * - If 'rule' is a super-rule that has subrules, revalidates (and possibly + * uninstalls and destroys) its subrules, via rule_destroy(). + */ static void rule_remove(struct ofproto *ofproto, struct rule *rule) { @@ -1923,7 +2125,7 @@ do_put_flow(struct ofproto *ofproto, struct rule *rule, int flags, struct odp_flow_put *put) { memset(&put->flow.stats, 0, sizeof put->flow.stats); - put->flow.key = rule->cr.flow; + odp_flow_key_from_flow(&put->flow.key, &rule->cr.flow); put->flow.actions = rule->odp_actions; put->flow.n_actions = rule->n_odp_actions; put->flow.flags = 0; @@ -2006,12 +2208,20 @@ rule_account(struct ofproto *ofproto, struct rule *rule, uint64_t extra_bytes) && total_bytes > rule->accounted_bytes) { ofproto->ofhooks->account_flow_cb( - &rule->cr.flow, rule->odp_actions, rule->n_odp_actions, + &rule->cr.flow, rule->tags, rule->odp_actions, rule->n_odp_actions, total_bytes - rule->accounted_bytes, ofproto->aux); rule->accounted_bytes = total_bytes; } } +/* 'rule' must be an exact-match rule in 'p'. + * + * If 'rule' is installed in the datapath, uninstalls it and updates's + * statistics. If 'rule' is a subrule, the statistics that are updated are + * actually its super-rule's statistics; otherwise 'rule''s own statistics are + * updated. + * + * If 'rule' is not installed, this function has no effect. */ static void rule_uninstall(struct ofproto *p, struct rule *rule) { @@ -2019,7 +2229,7 @@ rule_uninstall(struct ofproto *p, struct rule *rule) if (rule->installed) { struct odp_flow odp_flow; - odp_flow.key = rule->cr.flow; + odp_flow_key_from_flow(&odp_flow.key, &rule->cr.flow); odp_flow.actions = NULL; odp_flow.n_actions = 0; odp_flow.flags = 0; @@ -2039,16 +2249,12 @@ is_controller_rule(struct rule *rule) * NetFlow expiration messages since it is just part of the control * logic for the network and not real traffic. */ - if (rule && rule->super) { - struct rule *super = rule->super; - - return super->n_actions == 1 && - super->actions[0].type == htons(OFPAT_OUTPUT) && - super->actions[0].output.port == htons(OFPP_CONTROLLER); - } - - return false; -} + return (rule + && rule->super + && rule->super->n_actions == 1 + && action_outputs_to_port(&rule->super->actions[0], + htons(OFPP_CONTROLLER))); +} static void rule_post_uninstall(struct ofproto *ofproto, struct rule *rule) @@ -2089,34 +2295,15 @@ 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 @@ -2140,16 +2327,14 @@ handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh) } static int -handle_features_request(struct ofproto *p, struct ofconn *ofconn, - struct ofp_header *oh) +handle_features_request(struct ofconn *ofconn, 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 | @@ -2164,9 +2349,10 @@ handle_features_request(struct ofproto *p, struct ofconn *ofconn, (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)); } @@ -2175,8 +2361,7 @@ 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, struct ofp_header *oh) { struct ofpbuf *buf; struct ofp_switch_config *osc; @@ -2184,7 +2369,7 @@ 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; /* Send reply. */ @@ -2197,8 +2382,7 @@ 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, struct ofp_switch_config *osc) { uint16_t flags; int error; @@ -2209,13 +2393,13 @@ handle_set_config(struct ofproto *p, struct ofconn *ofconn, } flags = ntohs(osc->flags); - if (ofconn->type == OFCONN_CONTROLLER && ofconn->role != NX_ROLE_SLAVE) { + 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")", @@ -2230,27 +2414,15 @@ handle_set_config(struct ofproto *p, struct ofconn *ofconn, } 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) +add_controller_action(struct odp_actions *actions, uint16_t max_len) { union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER); - a->controller.arg = ntohs(oao->max_len); + a->controller.arg = max_len; } struct action_xlate_ctx { /* Input. */ - flow_t flow; /* Flow to which these actions correspond. */ + struct flow 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 @@ -2265,13 +2437,17 @@ struct action_xlate_ctx { uint16_t nf_output_iface; /* Output interface index for NetFlow. */ }; +/* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a + * flow translation. */ +#define MAX_RESUBMIT_RECURSION 8 + static void do_xlate_actions(const union ofp_action *in, size_t n_in, struct action_xlate_ctx *ctx); static void add_output_action(struct action_xlate_ctx *ctx, uint16_t port) { - const struct ofport *ofport = port_array_get(&ctx->ofproto->ports, port); + const struct ofport *ofport = get_port(ctx->ofproto, port); if (ofport) { if (ofport->opp.config & OFPPC_NO_FWD) { @@ -2291,10 +2467,11 @@ add_output_action(struct action_xlate_ctx *ctx, uint16_t port) } static struct rule * -lookup_valid_rule(struct ofproto *ofproto, const flow_t *flow) +lookup_valid_rule(struct ofproto *ofproto, const struct flow *flow) { struct rule *rule; - rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow)); + rule = rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow, + CLS_INC_ALL)); /* 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. */ @@ -2312,7 +2489,7 @@ lookup_valid_rule(struct ofproto *ofproto, const flow_t *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; @@ -2333,19 +2510,39 @@ xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port) do_xlate_actions(rule->actions, rule->n_actions, ctx); ctx->recurse--; } + } else { + 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 odp_actions *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)) { + odp_actions_add(actions, ODPAT_OUTPUT)->output.port = 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); break; @@ -2362,20 +2559,21 @@ 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->out); 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->out); break; case OFPP_CONTROLLER: - add_controller_action(ctx->out, oao); + add_controller_action(ctx->out, max_len); break; case OFPP_LOCAL: add_output_action(ctx, ODPP_LOCAL); break; default: - odp_port = ofp_port_to_odp_port(ntohs(oao->port)); + odp_port = ofp_port_to_odp_port(port); if (odp_port != ctx->flow.in_port) { add_output_action(ctx, odp_port); } @@ -2392,12 +2590,92 @@ xlate_output_action(struct action_xlate_ctx *ctx, } } +static void +xlate_output_action(struct action_xlate_ctx *ctx, + const struct ofp_action_output *oao) +{ + xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); +} + +/* 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) +{ + size_t n = ctx->out->n_actions; + if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) { + ctx->out->n_actions--; + } +} + +static void +xlate_enqueue_action(struct action_xlate_ctx *ctx, + const struct ofp_action_enqueue *oae) +{ + uint16_t ofp_port, odp_port; + uint32_t priority; + int error; + + 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; + } + + /* 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; + } + + /* Add ODP actions. */ + remove_pop_action(ctx); + odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority + = priority; + add_output_action(ctx, odp_port); + odp_actions_add(ctx->out, ODPAT_POP_PRIORITY); + + /* 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; + } +} + +static void +xlate_set_queue_action(struct action_xlate_ctx *ctx, + const struct nx_action_set_queue *nasq) +{ + uint32_t priority; + int error; + + error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id), + &priority); + if (error) { + /* Couldn't translate queue to a priority, so ignore. A warning + * has already been logged. */ + return; + } + + remove_pop_action(ctx); + odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority + = priority; +} + 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; union odp_action *oa; int subtype = ntohs(nah->subtype); @@ -2414,8 +2692,23 @@ xlate_nicira_action(struct action_xlate_ctx *ctx, ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id; break; + case NXAST_DROP_SPOOFED_ARP: + if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) { + odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP); + } + break; + + case NXAST_SET_QUEUE: + nasq = (const struct nx_action_set_queue *) nah; + xlate_set_queue_action(ctx, nasq); + break; + + case NXAST_POP_QUEUE: + odp_actions_add(ctx->out, ODPAT_POP_PRIORITY); + break; + /* If you add a new action here that modifies flow data, don't forget to - * update the flow key in ctx->flow in the same key. */ + * update the flow key in ctx->flow at the same time. */ default: VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype); @@ -2431,9 +2724,9 @@ do_xlate_actions(const union ofp_action *in, size_t n_in, const union ofp_action *ia; const struct ofport *port; - port = port_array_get(&ctx->ofproto->ports, ctx->flow.in_port); + port = get_port(ctx->ofproto, ctx->flow.in_port); if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) && - port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, stp_eth_addr) + port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) { /* Drop this flow. */ return; @@ -2449,18 +2742,22 @@ do_xlate_actions(const union ofp_action *in, size_t n_in, break; case OFPAT_SET_VLAN_VID: - oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_VID); - ctx->flow.dl_vlan = oa->vlan_vid.vlan_vid = ia->vlan_vid.vlan_vid; + oa = odp_actions_add(ctx->out, ODPAT_SET_DL_TCI); + oa->dl_tci.tci = ia->vlan_vid.vlan_vid; + oa->dl_tci.tci |= htons(ctx->flow.dl_vlan_pcp << VLAN_PCP_SHIFT); + ctx->flow.dl_vlan = ia->vlan_vid.vlan_vid; break; case OFPAT_SET_VLAN_PCP: - oa = odp_actions_add(ctx->out, ODPAT_SET_VLAN_PCP); - ctx->flow.dl_vlan_pcp = oa->vlan_pcp.vlan_pcp = ia->vlan_pcp.vlan_pcp; + oa = odp_actions_add(ctx->out, ODPAT_SET_DL_TCI); + oa->dl_tci.tci = htons(ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT); + oa->dl_tci.tci |= ctx->flow.dl_vlan; + ctx->flow.dl_vlan_pcp = ia->vlan_pcp.vlan_pcp; break; case OFPAT_STRIP_VLAN: odp_actions_add(ctx->out, ODPAT_STRIP_VLAN); - ctx->flow.dl_vlan = OFP_VLAN_NONE; + ctx->flow.dl_vlan = htons(OFP_VLAN_NONE); ctx->flow.dl_vlan_pcp = 0; break; @@ -2509,6 +2806,10 @@ do_xlate_actions(const union ofp_action *in, size_t n_in, 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); break; @@ -2518,7 +2819,7 @@ do_xlate_actions(const union ofp_action *in, size_t n_in, static int xlate_actions(const union ofp_action *in, size_t n_in, - const flow_t *flow, struct ofproto *ofproto, + const struct flow *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) @@ -2536,6 +2837,7 @@ xlate_actions(const union ofp_action *in, size_t n_in, ctx.may_set_up_flow = true; ctx.nf_output_iface = NF_OUT_DROP; 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. */ @@ -2550,6 +2852,7 @@ xlate_actions(const union ofp_action *in, size_t n_in, *nf_output_iface = ctx.nf_output_iface; } if (odp_actions_overflow(out)) { + COVERAGE_INC(odp_overflow); odp_actions_init(out); return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY); } @@ -2560,18 +2863,14 @@ xlate_actions(const union ofp_action *in, size_t n_in, * error message code (composed with ofp_mkerr()) for the caller to propagate * upward. Otherwise, returns 0. * - * 'oh' is used to make log messages more informative. */ + * The log message mentions 'msg_type'. */ static int -reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh) +reject_slave_controller(struct ofconn *ofconn, const const char *msg_type) { - if (ofconn->type == OFCONN_CONTROLLER && ofconn->role == NX_ROLE_SLAVE) { + if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) { static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5); - char *type_name; - - type_name = ofp_message_type_to_string(oh->type); VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller", - type_name); - free(type_name); + msg_type); return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); } else { @@ -2580,29 +2879,42 @@ reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh) } static int -handle_packet_out(struct ofproto *p, struct ofconn *ofconn, - struct ofp_header *oh) +handle_packet_out(struct ofconn *ofconn, 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 odp_actions odp_actions; + struct ofpbuf request; + struct flow flow; + size_t n_ofp_actions; uint16_t in_port; - flow_t flow; int error; - error = reject_slave_controller(ofconn, oh); + COVERAGE_INC(ofproto_packet_out); + + error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT"); if (error) { return error; } - error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports); + /* Get ofp_packet_out. */ + request.data = oh; + request.size = ntohs(oh->length); + opo = ofpbuf_try_pull(&request, offsetof(struct ofp_packet_out, actions)); + if (!opo) { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); + } + + /* Get actions. */ + error = ofputil_pull_actions(&request, ntohs(opo->actions_len), + &ofp_actions, &n_ofp_actions); if (error) { return error; } - opo = (struct ofp_packet_out *) oh; - COVERAGE_INC(ofproto_packet_out); + /* Get payload. */ if (opo->buffer_id != htonl(UINT32_MAX)) { error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id), &buffer, &in_port); @@ -2611,20 +2923,28 @@ handle_packet_out(struct ofproto *p, struct ofconn *ofconn, } payload = *buffer; } else { + payload = request; buffer = NULL; } - flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow); - error = xlate_actions((const union ofp_action *) opo->actions, n_actions, - &flow, p, &payload, &actions, NULL, NULL, NULL); + /* Extract flow, check actions. */ + flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), + &flow); + error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports); if (error) { - return error; + goto exit; } - dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions, - &payload); - ofpbuf_delete(buffer); + /* Send. */ + error = xlate_actions(ofp_actions, n_ofp_actions, &flow, p, &payload, + &odp_actions, NULL, NULL, NULL); + if (!error) { + dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions, + &payload); + } +exit: + ofpbuf_delete(buffer); return 0; } @@ -2640,31 +2960,28 @@ update_port_config(struct ofproto *p, struct ofport *port, netdev_turn_flags_on(port->netdev, NETDEV_UP, true); } } -#define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD) +#define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \ + OFPPC_NO_FWD | OFPPC_NO_FLOOD) if (mask & REVALIDATE_BITS) { COVERAGE_INC(ofproto_costly_flags); port->opp.config ^= mask & REVALIDATE_BITS; p->need_revalidate = true; } #undef REVALIDATE_BITS - if (mask & OFPPC_NO_FLOOD) { - port->opp.config ^= OFPPC_NO_FLOOD; - refresh_port_groups(p); - } if (mask & OFPPC_NO_PACKET_IN) { port->opp.config ^= OFPPC_NO_PACKET_IN; } } static int -handle_port_mod(struct ofproto *p, struct ofconn *ofconn, - struct ofp_header *oh) +handle_port_mod(struct ofconn *ofconn, struct ofp_header *oh) { + struct ofproto *p = ofconn->ofproto; const struct ofp_port_mod *opm; struct ofport *port; int error; - error = reject_slave_controller(ofconn, oh); + error = reject_slave_controller(ofconn, "OFPT_PORT_MOD"); if (error) { return error; } @@ -2674,8 +2991,7 @@ handle_port_mod(struct ofproto *p, struct ofconn *ofconn, } 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)) { @@ -2690,7 +3006,7 @@ handle_port_mod(struct ofproto *p, struct ofconn *ofconn, } static struct ofpbuf * -make_stats_reply(uint32_t xid, uint16_t type, size_t body_len) +make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len) { struct ofp_stats_reply *osr; struct ofpbuf *msg; @@ -2703,34 +3019,72 @@ 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_stats_request *request, size_t body_len) { - return make_stats_reply(request->header.xid, request->type, body_len); + return make_ofp_stats_reply(request->header.xid, request->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 struct ofpbuf * +make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len) +{ + struct nicira_stats_msg *nsm; + struct ofpbuf *msg; + + msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX)); + nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg); + nsm->type = htons(OFPST_VENDOR); + nsm->flags = htons(0); + nsm->vendor = htonl(NX_VENDOR_ID); + nsm->subtype = htonl(subtype); + return msg; +} + +static struct ofpbuf * +start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len) +{ + return make_nxstats_reply(request->header.xid, request->subtype, body_len); +} + +static void +append_nxstats_reply(size_t nbytes, struct ofconn *ofconn, + struct ofpbuf **msgp) +{ + struct ofpbuf *msg = *msgp; + assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg)); + if (nbytes + msg->size > UINT16_MAX) { + struct nicira_stats_msg *reply = msg->data; + reply->flags = htons(OFPSF_REPLY_MORE); + *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes); + queue_tx(msg, ofconn, ofconn->reply_counter); + } + ofpbuf_prealloc_tailroom(*msgp, nbytes); +} + static int -handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn, - struct ofp_stats_request *request) +handle_desc_stats_request(struct ofconn *ofconn, + struct ofp_stats_request *request) { + struct ofproto *p = ofconn->ofproto; struct ofp_desc_stats *ods; struct ofpbuf *msg; - msg = start_stats_reply(request, sizeof *ods); - ods = append_stats_reply(sizeof *ods, ofconn, &msg); + msg = start_ofp_stats_reply(request, sizeof *ods); + ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg); memset(ods, 0, sizeof *ods); ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc); ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc); @@ -2742,56 +3096,34 @@ handle_desc_stats_request(struct ofproto *p, struct ofconn *ofconn, return 0; } -static void -count_subrules(struct cls_rule *cls_rule, void *n_subrules_) -{ - struct rule *rule = rule_from_cls_rule(cls_rule); - int *n_subrules = n_subrules_; - - if (rule->super) { - (*n_subrules)++; - } -} - static int -handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn, +handle_table_stats_request(struct ofconn *ofconn, struct ofp_stats_request *request) { + struct ofproto *p = ofconn->ofproto; struct ofp_table_stats *ots; struct ofpbuf *msg; - struct odp_stats dpstats; - int n_exact, n_subrules, n_wild; - - msg = start_stats_reply(request, sizeof *ots * 2); + struct rule *rule; + int n_rules; - /* 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); + msg = start_ofp_stats_reply(request, sizeof *ots * 2); - /* 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 */ + /* Count rules other than subrules. */ + n_rules = classifier_count(&p->cls); + CLASSIFIER_FOR_EACH_EXACT_RULE (rule, cr, &p->cls) { + if (rule->super) { + n_rules--; + } + } /* Classifier table. */ - ots = append_stats_reply(sizeof *ots, ofconn, &msg); + ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg); memset(ots, 0, sizeof *ots); - ots->table_id = TABLEID_CLASSIFIER; strcpy(ots->name, "classifier"); - ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL) - : htonl(OFPFW_ALL); - ots->max_entries = htonl(65536); - ots->active_count = htonl(n_wild); + ots->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(n_rules); ots->lookup_count = htonll(0); /* XXX */ ots->matched_count = htonll(0); /* XXX */ @@ -2800,19 +3132,19 @@ handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn, } static void -append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn, +append_port_stat(struct ofport *port, struct ofconn *ofconn, struct ofpbuf **msgp) { struct netdev_stats stats; struct ofp_port_stats *ops; - /* Intentionally ignore return value, since errors will set - * 'stats' to all-1s, which is correct for OpenFlow, and + /* Intentionally ignore return value, since errors will set + * 'stats' to all-1s, which is correct for OpenFlow, and * netdev_get_stats() will log errors. */ netdev_get_stats(port->netdev, &stats); - ops = append_stats_reply(sizeof *ops, ofconn, msgp); - ops->port_no = htons(odp_port_to_ofp_port(port_no)); + ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp); + ops->port_no = htons(port->opp.port_no); memset(ops->pad, 0, sizeof ops->pad); ops->rx_packets = htonll(stats.rx_packets); ops->tx_packets = htonll(stats.tx_packets); @@ -2829,31 +3161,29 @@ append_port_stat(struct ofport *port, uint16_t port_no, struct ofconn *ofconn, } static int -handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn, - struct ofp_stats_request *osr, +handle_port_stats_request(struct ofconn *ofconn, struct ofp_stats_request *osr, size_t arg_size) { + struct ofproto *p = ofconn->ofproto; struct ofp_port_stats_request *psr; struct ofp_port_stats *ops; struct ofpbuf *msg; struct ofport *port; - unsigned int port_no; if (arg_size != sizeof *psr) { return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); } psr = (struct ofp_port_stats_request *) osr->body; - msg = start_stats_reply(osr, sizeof *ops * 16); + msg = start_ofp_stats_reply(osr, sizeof *ops * 16); if (psr->port_no != htons(OFPP_NONE)) { - port = port_array_get(&p->ports, - ofp_port_to_odp_port(ntohs(psr->port_no))); + port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no))); if (port) { - append_port_stat(port, ntohs(psr->port_no), ofconn, &msg); + append_port_stat(port, ofconn, &msg); } } else { - PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { - append_port_stat(port, port_no, ofconn, &msg); + HMAP_FOR_EACH (port, hmap_node, &p->ports) { + append_port_stat(port, ofconn, &msg); } } @@ -2862,9 +3192,8 @@ handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn, } struct flow_stats_cbdata { - struct ofproto *ofproto; struct ofconn *ofconn; - uint16_t out_port; + ovs_be16 out_port; struct ofpbuf *msg; }; @@ -2897,13 +3226,13 @@ query_stats(struct ofproto *p, struct rule *rule, 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; + LIST_FOR_EACH (subrule, list, &rule->list) { + odp_flow_key_from_flow(&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; + odp_flow_key_from_flow(&odp_flows[0].key, &rule->cr.flow); } /* Fetch up-to-date statistics from the datapath and add them in. */ @@ -2922,6 +3251,14 @@ query_stats(struct ofproto *p, struct rule *rule, *byte_countp = byte_count; } +static void +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 flow_stats_cb(struct cls_rule *rule_, void *cbdata_) { @@ -2930,9 +3267,6 @@ flow_stats_cb(struct cls_rule *rule_, void *cbdata_) struct ofp_flow_stats *ofs; uint64_t packet_count, byte_count; size_t act_len, len; - long long int tdiff = time_msec() - rule->created; - uint32_t sec = tdiff / 1000; - uint32_t msec = tdiff - (sec * 1000); if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { return; @@ -2941,16 +3275,15 @@ flow_stats_cb(struct cls_rule *rule_, void *cbdata_) 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(cbdata->ofconn->ofproto, rule, &packet_count, &byte_count); - ofs = append_stats_reply(len, cbdata->ofconn, &cbdata->msg); + ofs = append_ofp_stats_reply(len, cbdata->ofconn, &cbdata->msg); ofs->length = htons(len); - ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH; + ofs->table_id = 0; ofs->pad = 0; flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, - cbdata->ofproto->tun_id_from_cookie, &ofs->match); - ofs->duration_sec = htonl(sec); - ofs->duration_nsec = htonl(msec * 1000000); + cbdata->ofconn->flow_format, &ofs->match); + calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec); ofs->cookie = rule->flow_cookie; ofs->priority = htons(rule->cr.priority); ofs->idle_timeout = htons(rule->idle_timeout); @@ -2958,22 +3291,20 @@ flow_stats_cb(struct cls_rule *rule_, void *cbdata_) 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 int table_id_to_include(uint8_t table_id) { - return (table_id == TABLEID_HASH ? CLS_INC_EXACT - : table_id == TABLEID_CLASSIFIER ? CLS_INC_WILD - : table_id == 0xff ? CLS_INC_ALL - : 0); + return table_id == 0 || table_id == 0xff ? CLS_INC_ALL : 0; } static int -handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn, - const struct ofp_stats_request *osr, - size_t arg_size) +handle_flow_stats_request(struct ofconn *ofconn, + const struct ofp_stats_request *osr, size_t arg_size) { struct ofp_flow_stats_request *fsr; struct flow_stats_cbdata cbdata; @@ -2985,18 +3316,84 @@ handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn, fsr = (struct ofp_flow_stats_request *) osr->body; COVERAGE_INC(ofproto_flows_req); - cbdata.ofproto = p; cbdata.ofconn = ofconn; cbdata.out_port = fsr->out_port; - cbdata.msg = start_stats_reply(osr, 1024); - cls_rule_from_match(&fsr->match, 0, false, 0, &target); - classifier_for_each_match(&p->cls, &target, + cbdata.msg = start_ofp_stats_reply(osr, 1024); + cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0, &target); + classifier_for_each_match(&ofconn->ofproto->cls, &target, table_id_to_include(fsr->table_id), flow_stats_cb, &cbdata); queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); return 0; } +static void +nx_flow_stats_cb(struct cls_rule *rule_, void *cbdata_) +{ + struct rule *rule = rule_from_cls_rule(rule_); + struct flow_stats_cbdata *cbdata = cbdata_; + struct nx_flow_stats *nfs; + uint64_t packet_count, byte_count; + size_t act_len, start_len; + + if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { + return; + } + + query_stats(cbdata->ofconn->ofproto, rule, &packet_count, &byte_count); + + act_len = sizeof *rule->actions * rule->n_actions; + + start_len = cbdata->msg->size; + append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, + cbdata->ofconn, &cbdata->msg); + nfs = ofpbuf_put_uninit(cbdata->msg, 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(cbdata->msg, &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(cbdata->msg, rule->actions, act_len); + } + nfs->length = htons(cbdata->msg->size - start_len); +} + +static int +handle_nxst_flow(struct ofconn *ofconn, struct ofpbuf *b) +{ + struct nx_flow_stats_request *nfsr; + struct flow_stats_cbdata cbdata; + struct cls_rule target; + int error; + + /* Dissect the message. */ + nfsr = ofpbuf_try_pull(b, sizeof *nfsr); + if (!nfsr) { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); + } + error = nx_pull_match(b, ntohs(nfsr->match_len), 0, &target); + if (error) { + return error; + } + + COVERAGE_INC(ofproto_flows_req); + cbdata.ofconn = ofconn; + cbdata.out_port = nfsr->out_port; + cbdata.msg = start_nxstats_reply(&nfsr->nsm, 1024); + classifier_for_each_match(&ofconn->ofproto->cls, &target, + table_id_to_include(nfsr->table_id), + nx_flow_stats_cb, &cbdata); + queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); + return 0; +} + struct flow_stats_ds_cbdata { struct ofproto *ofproto; struct ds *results; @@ -3019,7 +3416,7 @@ flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_) query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, - cbdata->ofproto->tun_id_from_cookie, &match); + NXFF_OPENFLOW10, &match); ds_put_format(results, "duration=%llds, ", (time_msec() - rule->created) / 1000); @@ -3027,11 +3424,15 @@ flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_) 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); + 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) @@ -3046,14 +3447,14 @@ ofproto_get_all_flows(struct ofproto *p, struct ds *results) cbdata.ofproto = p; cbdata.results = results; - cls_rule_from_match(&match, 0, false, 0, &target); + cls_rule_from_match(&match, 0, NXFF_OPENFLOW10, 0, &target); classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, flow_stats_ds_cb, &cbdata); } struct aggregate_stats_cbdata { struct ofproto *ofproto; - uint16_t out_port; + ovs_be16 out_port; uint64_t packet_count; uint64_t byte_count; uint32_t n_flows; @@ -3077,154 +3478,329 @@ aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_) cbdata->n_flows++; } +static void +query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target, + ovs_be16 out_port, uint8_t table_id, + struct ofp_aggregate_stats_reply *oasr) +{ + struct aggregate_stats_cbdata cbdata; + + COVERAGE_INC(ofproto_agg_request); + cbdata.ofproto = ofproto; + cbdata.out_port = out_port; + cbdata.packet_count = 0; + cbdata.byte_count = 0; + cbdata.n_flows = 0; + classifier_for_each_match(&ofproto->cls, target, + table_id_to_include(table_id), + aggregate_stats_cb, &cbdata); + + oasr->flow_count = htonl(cbdata.n_flows); + oasr->packet_count = htonll(cbdata.packet_count); + oasr->byte_count = htonll(cbdata.byte_count); + memset(oasr->pad, 0, sizeof oasr->pad); +} + static int -handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn, +handle_aggregate_stats_request(struct ofconn *ofconn, const struct ofp_stats_request *osr, size_t arg_size) { - struct ofp_aggregate_stats_request *asr; + struct ofp_aggregate_stats_request *request; struct ofp_aggregate_stats_reply *reply; - struct aggregate_stats_cbdata cbdata; struct cls_rule target; struct ofpbuf *msg; - if (arg_size != sizeof *asr) { + if (arg_size != sizeof *request) { return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); } - asr = (struct ofp_aggregate_stats_request *) osr->body; + request = (struct ofp_aggregate_stats_request *) osr->body; - COVERAGE_INC(ofproto_agg_request); - cbdata.ofproto = p; - cbdata.out_port = asr->out_port; - cbdata.packet_count = 0; - cbdata.byte_count = 0; - cbdata.n_flows = 0; - cls_rule_from_match(&asr->match, 0, false, 0, &target); - classifier_for_each_match(&p->cls, &target, - table_id_to_include(asr->table_id), - aggregate_stats_cb, &cbdata); + cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0, &target); - 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(osr, 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, struct ofpbuf *b) { - 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 *buf; int error; - error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr, - 1, &arg_size); + /* Dissect the message. */ + request = ofpbuf_try_pull(b, sizeof *request); + if (!request) { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); + } + error = nx_pull_match(b, ntohs(request->match_len), 0, &target); if (error) { return error; } - osr = (struct ofp_stats_request *) oh; - - switch (ntohs(osr->type)) { - case OFPST_DESC: - return handle_desc_stats_request(p, ofconn, osr); - case OFPST_FLOW: - return handle_flow_stats_request(p, ofconn, osr, arg_size); - - case OFPST_AGGREGATE: - return handle_aggregate_stats_request(p, ofconn, osr, arg_size); + /* 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_TABLE: - return handle_table_stats_request(p, ofconn, osr); + return 0; +} - case OFPST_PORT: - return handle_port_stats_request(p, ofconn, osr, arg_size); +struct queue_stats_cbdata { + struct ofconn *ofconn; + struct ofport *ofport; + struct ofpbuf *msg; +}; - case OFPST_VENDOR: - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); +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; - default: - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT); - } + 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); } -static long long int -msec_from_nsec(uint64_t sec, uint32_t nsec) +static void +handle_queue_stats_dump_cb(uint32_t queue_id, + struct netdev_queue_stats *stats, + void *cbdata_) { - return !sec ? 0 : sec * 1000 + nsec / 1000000; + struct queue_stats_cbdata *cbdata = cbdata_; + + put_queue_stats(cbdata, queue_id, stats); } static void -update_time(struct ofproto *ofproto, struct rule *rule, - const struct odp_flow_stats *stats) +handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id, + struct queue_stats_cbdata *cbdata) { - 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; + 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); } - netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used); } } -static void -update_stats(struct ofproto *ofproto, struct rule *rule, - const struct odp_flow_stats *stats) +static int +handle_queue_stats_request(struct ofconn *ofconn, + const struct ofp_stats_request *osr, + size_t arg_size) { - 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); - } -} + struct ofproto *ofproto = ofconn->ofproto; + struct ofp_queue_stats_request *qsr; + struct queue_stats_cbdata cbdata; + struct ofport *port; + unsigned int port_no; + uint32_t queue_id; + + if (arg_size != sizeof *qsr) { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); + } + qsr = (struct ofp_queue_stats_request *) osr->body; + + COVERAGE_INC(ofproto_queue_req); + + cbdata.ofconn = ofconn; + cbdata.msg = start_ofp_stats_reply(osr, 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 int +handle_vendor_stats_request(struct ofconn *ofconn, + struct ofp_stats_request *osr, size_t arg_size) +{ + struct nicira_stats_msg *nsm; + struct ofpbuf b; + ovs_be32 vendor; + + if (arg_size < 4) { + VLOG_WARN_RL(&rl, "truncated vendor stats request body"); + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); + } + + memcpy(&vendor, osr->body, sizeof vendor); + if (vendor != htonl(NX_VENDOR_ID)) { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); + } + + if (ntohs(nsm->header.length) < sizeof(struct nicira_stats_msg)) { + VLOG_WARN_RL(&rl, "truncated Nicira stats request"); + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); + } + + nsm = (struct nicira_stats_msg *) osr; + b.data = nsm; + b.size = ntohs(nsm->header.length); + switch (ntohl(nsm->subtype)) { + case NXST_FLOW: + return handle_nxst_flow(ofconn, &b); + + case NXST_AGGREGATE: + return handle_nxst_aggregate(ofconn, &b); + + default: + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); + } +} + +static int +handle_stats_request(struct ofconn *ofconn, struct ofp_header *oh) +{ + struct ofp_stats_request *osr; + size_t arg_size; + int error; + + error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr, + 1, &arg_size); + if (error) { + return error; + } + osr = (struct ofp_stats_request *) oh; + + switch (ntohs(osr->type)) { + case OFPST_DESC: + return handle_desc_stats_request(ofconn, osr); + + case OFPST_FLOW: + return handle_flow_stats_request(ofconn, osr, arg_size); + + case OFPST_AGGREGATE: + return handle_aggregate_stats_request(ofconn, osr, arg_size); + + case OFPST_TABLE: + return handle_table_stats_request(ofconn, osr); + + case OFPST_PORT: + return handle_port_stats_request(ofconn, osr, arg_size); + + case OFPST_QUEUE: + return handle_queue_stats_request(ofconn, osr, arg_size); + + case OFPST_VENDOR: + return handle_vendor_stats_request(ofconn, osr, arg_size); + + default: + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT); + } +} + +static long long int +msec_from_nsec(uint64_t sec, uint32_t nsec) +{ + return !sec ? 0 : sec * 1000 + nsec / 1000000; +} + +static void +update_time(struct ofproto *ofproto, struct rule *rule, + const struct odp_flow_stats *stats) +{ + long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec); + if (used > rule->used) { + rule->used = used; + if (rule->super && used > rule->super->used) { + rule->super->used = used; + } + netflow_flow_update_time(ofproto->netflow, &rule->nf_flow, used); + } +} + +static void +update_stats(struct ofproto *ofproto, struct rule *rule, + 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->tcp_flags); + } +} + +struct flow_mod { + struct cls_rule cr; + ovs_be64 cookie; + uint16_t command; + uint16_t idle_timeout; + uint16_t hard_timeout; + uint32_t buffer_id; + uint16_t out_port; + uint16_t flags; + union ofp_action *actions; + size_t n_actions; +}; /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT * in which no matching flow already exists in the flow table. * * Adds the flow specified by 'ofm', which is followed by 'n_actions' - * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error - * code as encoded by ofp_mkerr() on failure. + * ofp_actions, to 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, - const 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; - if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) { - flow_t flow; - uint32_t wildcards; - - flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie, - &flow, &wildcards); - if (classifier_rule_overlaps(&p->cls, &flow, wildcards, - ntohs(ofm->priority))) { - return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP); - } + if (fm->flags & OFPFF_CHECK_OVERLAP + && classifier_rule_overlaps(&p->cls, &fm->cr)) { + return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP); } - rule = rule_create(p, NULL, (const union ofp_action *) ofm->actions, - n_actions, ntohs(ofm->idle_timeout), - ntohs(ofm->hard_timeout), ofm->cookie, - ofm->flags & htons(OFPFF_SEND_FLOW_REM)); - cls_rule_from_match(&ofm->match, ntohs(ofm->priority), - p->tun_id_from_cookie, ofm->cookie, &rule->cr); + rule = rule_create(p, NULL, fm->actions, fm->n_actions, + fm->idle_timeout, fm->hard_timeout, fm->cookie, + fm->flags & OFPFF_SEND_FLOW_REM); + rule->cr = fm->cr; 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; @@ -3232,45 +3808,35 @@ add_flow(struct ofproto *p, struct ofconn *ofconn, } rule_insert(p, rule, packet, in_port); - ofpbuf_delete(packet); return error; } static struct rule * -find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm) +find_flow_strict(struct ofproto *p, const struct flow_mod *fm) { - uint32_t wildcards; - flow_t flow; - - flow_from_match(&ofm->match, p->tun_id_from_cookie, ofm->cookie, - &flow, &wildcards); - return rule_from_cls_rule(classifier_find_rule_exactly( - &p->cls, &flow, wildcards, - ntohs(ofm->priority))); + return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr)); } static int -send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn, - struct rule *rule, const struct ofp_flow_mod *ofm) +send_buffered_packet(struct ofconn *ofconn, + struct rule *rule, uint32_t buffer_id) { struct ofpbuf *packet; uint16_t in_port; - flow_t flow; + struct flow flow; int error; - if (ofm->buffer_id == htonl(UINT32_MAX)) { + if (buffer_id == UINT32_MAX) { return 0; } - error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id), - &packet, &in_port); + error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port); if (error) { return error; } flow_extract(packet, 0, in_port, &flow); - rule_execute(ofproto, rule, packet, &flow); - ofpbuf_delete(packet); + rule_execute(ofconn->ofproto, rule, packet, &flow); return 0; } @@ -3279,13 +3845,12 @@ send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn, struct modify_flows_cbdata { struct ofproto *ofproto; - const struct ofp_flow_mod *ofm; - size_t n_actions; + const struct flow_mod *fm; struct rule *match; }; -static int modify_flow(struct ofproto *, const struct ofp_flow_mod *, - size_t n_actions, struct rule *); +static int modify_flow(struct ofproto *, const struct flow_mod *, + struct rule *); static void modify_flows_cb(struct cls_rule *, void *cbdata_); /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as @@ -3294,30 +3859,24 @@ static void modify_flows_cb(struct cls_rule *, void *cbdata_); * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, * if any. */ static int -modify_flows_loose(struct ofproto *p, struct ofconn *ofconn, - const struct ofp_flow_mod *ofm, size_t n_actions) +modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm) { struct modify_flows_cbdata cbdata; - struct cls_rule target; - cbdata.ofproto = p; - cbdata.ofm = ofm; - cbdata.n_actions = n_actions; + cbdata.ofproto = ofconn->ofproto; + cbdata.fm = fm; cbdata.match = NULL; - cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie, - &target); - - classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, + classifier_for_each_match(&ofconn->ofproto->cls, &fm->cr, CLS_INC_ALL, modify_flows_cb, &cbdata); if (cbdata.match) { /* This credits the packet to whichever flow happened to happened to * match last. That's weird. Maybe we should do a lookup for the * flow that actually matches the packet? Who knows. */ - send_buffered_packet(p, ofconn, cbdata.match, ofm); + send_buffered_packet(ofconn, cbdata.match, fm->buffer_id); return 0; } else { - return add_flow(p, ofconn, ofm, n_actions); + return add_flow(ofconn, fm); } } @@ -3327,15 +3886,15 @@ modify_flows_loose(struct ofproto *p, struct ofconn *ofconn, * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, * if any. */ static int -modify_flow_strict(struct ofproto *p, struct ofconn *ofconn, - struct ofp_flow_mod *ofm, size_t n_actions) +modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm) { - struct rule *rule = find_flow_strict(p, ofm); + struct ofproto *p = ofconn->ofproto; + struct rule *rule = find_flow_strict(p, fm); if (rule && !rule_is_hidden(rule)) { - modify_flow(p, ofm, n_actions, rule); - return send_buffered_packet(p, ofconn, rule, ofm); + modify_flow(p, fm, rule); + return send_buffered_packet(ofconn, rule, fm->buffer_id); } else { - return add_flow(p, ofconn, ofm, n_actions); + return add_flow(ofconn, fm); } } @@ -3348,7 +3907,7 @@ modify_flows_cb(struct cls_rule *rule_, void *cbdata_) if (!rule_is_hidden(rule)) { cbdata->match = rule; - modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule); + modify_flow(cbdata->ofproto, cbdata->fm, rule); } } @@ -3357,24 +3916,23 @@ modify_flows_cb(struct cls_rule *rule_, void *cbdata_) * the rule's actions to match those in 'ofm' (which is followed by 'n_actions' * ofp_action[] structures). */ static int -modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm, - size_t n_actions, struct rule *rule) +modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule) { - size_t actions_len = n_actions * sizeof *rule->actions; + size_t actions_len = fm->n_actions * sizeof *rule->actions; - rule->flow_cookie = ofm->cookie; + rule->flow_cookie = fm->cookie; /* If the actions are the same, do nothing. */ - if (n_actions == rule->n_actions - && !memcmp(ofm->actions, rule->actions, actions_len)) - { + if (fm->n_actions == rule->n_actions + && (!fm->n_actions + || !memcmp(fm->actions, rule->actions, actions_len))) { return 0; } /* Replace actions. */ free(rule->actions); - rule->actions = xmemdup(ofm->actions, actions_len); - rule->n_actions = n_actions; + rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL; + rule->n_actions = fm->n_actions; /* Make sure that the datapath gets updated properly. */ if (rule->cr.wc.wildcards) { @@ -3391,36 +3949,32 @@ modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm, struct delete_flows_cbdata { struct ofproto *ofproto; - uint16_t out_port; + ovs_be16 out_port; }; static void delete_flows_cb(struct cls_rule *, void *cbdata_); -static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port); +static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port); /* Implements OFPFC_DELETE. */ static void -delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm) +delete_flows_loose(struct ofproto *p, const struct flow_mod *fm) { struct delete_flows_cbdata cbdata; - struct cls_rule target; cbdata.ofproto = p; - cbdata.out_port = ofm->out_port; + cbdata.out_port = htons(fm->out_port); - cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie, - &target); - - classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, + classifier_for_each_match(&p->cls, &fm->cr, CLS_INC_ALL, delete_flows_cb, &cbdata); } /* Implements OFPFC_DELETE_STRICT. */ static void -delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm) +delete_flow_strict(struct ofproto *p, struct flow_mod *fm) { - struct rule *rule = find_flow_strict(p, ofm); + struct rule *rule = find_flow_strict(p, fm); if (rule) { - delete_flow(p, rule, ofm->out_port); + delete_flow(p, rule, htons(fm->out_port)); } } @@ -3443,7 +3997,7 @@ delete_flows_cb(struct cls_rule *rule_, void *cbdata_) * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the * specified 'out_port'. */ static void -delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port) +delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port) { if (rule_is_hidden(rule)) { return; @@ -3453,62 +4007,51 @@ delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port) return; } - send_flow_removed(p, rule, time_msec(), OFPRR_DELETE); + send_flow_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) +flow_mod_core(struct ofconn *ofconn, struct flow_mod *fm) { - size_t n_actions; + struct ofproto *p = ofconn->ofproto; int error; - error = reject_slave_controller(ofconn, &ofm->header); + error = reject_slave_controller(ofconn, "flow_mod"); if (error) { return error; } - error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm, - sizeof *ofm->actions, &n_actions); + + error = validate_actions(fm->actions, fm->n_actions, + &fm->cr.flow, p->max_ports); if (error) { return error; } /* We do not support the emergency flow cache. It will hopefully * get dropped from OpenFlow in the near future. */ - if (ofm->flags & htons(OFPFF_EMERG)) { + if (fm->flags & OFPFF_EMERG) { /* There isn't a good fit for an error code, so just state that the * flow table is full. */ return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL); } - normalize_match(&ofm->match); - if (!ofm->match.wildcards) { - ofm->priority = htons(UINT16_MAX); - } - - error = validate_actions((const union ofp_action *) ofm->actions, - n_actions, p->max_ports); - if (error) { - return error; - } - - switch (ntohs(ofm->command)) { + switch (fm->command) { case OFPFC_ADD: - return add_flow(p, ofconn, ofm, n_actions); + return add_flow(ofconn, fm); case OFPFC_MODIFY: - return modify_flows_loose(p, ofconn, ofm, n_actions); + return modify_flows_loose(ofconn, fm); case OFPFC_MODIFY_STRICT: - return modify_flow_strict(p, ofconn, ofm, n_actions); + return modify_flow_strict(ofconn, fm); case OFPFC_DELETE: - delete_flows_loose(p, ofm); + delete_flows_loose(p, fm); return 0; case OFPFC_DELETE_STRICT: - delete_flow_strict(p, ofm); + delete_flow_strict(p, fm); return 0; default: @@ -3517,7 +4060,102 @@ handle_flow_mod(struct ofproto *p, struct ofconn *ofconn, } static int -handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg) +handle_ofpt_flow_mod(struct ofconn *ofconn, struct ofp_header *oh) +{ + struct ofp_match orig_match; + struct ofp_flow_mod *ofm; + struct flow_mod fm; + struct ofpbuf b; + int error; + + b.data = oh; + b.size = ntohs(oh->length); + + /* Dissect the message. */ + ofm = ofpbuf_try_pull(&b, sizeof *ofm); + if (!ofm) { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); + } + error = ofputil_pull_actions(&b, b.size, &fm.actions, &fm.n_actions); + if (error) { + return error; + } + + /* Normalize ofm->match. If normalization actually changes anything, then + * log the differences. */ + ofm->match.pad1[0] = ofm->match.pad2[0] = 0; + orig_match = ofm->match; + normalize_match(&ofm->match); + if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) { + static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1); + if (!VLOG_DROP_INFO(&normal_rl)) { + char *old = ofp_match_to_literal_string(&orig_match); + char *new = ofp_match_to_literal_string(&ofm->match); + VLOG_INFO("%s: normalization changed ofp_match, details:", + rconn_get_name(ofconn->rconn)); + VLOG_INFO(" pre: %s", old); + VLOG_INFO("post: %s", new); + free(old); + free(new); + } + } + + /* Translate the message. */ + cls_rule_from_match(&ofm->match, ntohs(ofm->priority), ofconn->flow_format, + ofm->cookie, &fm.cr); + fm.cookie = ofm->cookie; + fm.command = ntohs(ofm->command); + fm.idle_timeout = ntohs(ofm->idle_timeout); + fm.hard_timeout = ntohs(ofm->hard_timeout); + fm.buffer_id = ntohl(ofm->buffer_id); + fm.out_port = ntohs(ofm->out_port); + fm.flags = ntohs(ofm->flags); + + /* Execute the command. */ + return flow_mod_core(ofconn, &fm); +} + +static int +handle_nxt_flow_mod(struct ofconn *ofconn, struct ofp_header *oh) +{ + struct nx_flow_mod *nfm; + struct flow_mod fm; + struct ofpbuf b; + int error; + + b.data = oh; + b.size = ntohs(oh->length); + + /* Dissect the message. */ + nfm = ofpbuf_try_pull(&b, sizeof *nfm); + if (!nfm) { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); + } + error = nx_pull_match(&b, ntohs(nfm->match_len), ntohs(nfm->priority), + &fm.cr); + if (error) { + return error; + } + error = ofputil_pull_actions(&b, b.size, &fm.actions, &fm.n_actions); + if (error) { + return error; + } + + /* Translate the message. */ + fm.cookie = nfm->cookie; + fm.command = ntohs(nfm->command); + fm.idle_timeout = ntohs(nfm->idle_timeout); + fm.hard_timeout = ntohs(nfm->hard_timeout); + fm.buffer_id = ntohl(nfm->buffer_id); + fm.out_port = ntohs(nfm->out_port); + fm.flags = ntohs(nfm->flags); + + /* Execute the command. */ + return flow_mod_core(ofconn, &fm); +} + +static int +handle_tun_id_from_cookie(struct ofconn *ofconn, struct nxt_tun_id_cookie *msg) { int error; @@ -3526,13 +4164,12 @@ handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg) return error; } - p->tun_id_from_cookie = !!msg->set; + ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10; return 0; } static int -handle_role_request(struct ofproto *ofproto, - struct ofconn *ofconn, struct nicira_header *msg) +handle_role_request(struct ofconn *ofconn, struct nicira_header *msg) { struct nx_role_request *nrr; struct nx_role_request *reply; @@ -3540,13 +4177,13 @@ handle_role_request(struct ofproto *ofproto, uint32_t role; if (ntohs(msg->header.length) != sizeof *nrr) { - VLOG_WARN_RL(&rl, "received role request of length %zu (expected %zu)", + VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)", ntohs(msg->header.length), sizeof *nrr); return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); } nrr = (struct nx_role_request *) msg; - if (ofconn->type != OFCONN_CONTROLLER) { + if (ofconn->type != OFCONN_PRIMARY) { VLOG_WARN_RL(&rl, "ignoring role request on non-controller " "connection"); return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); @@ -3564,8 +4201,7 @@ handle_role_request(struct ofproto *ofproto, if (role == NX_ROLE_MASTER) { struct ofconn *other; - HMAP_FOR_EACH (other, struct ofconn, hmap_node, - &ofproto->controllers) { + HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) { if (other->role == NX_ROLE_MASTER) { other->role = NX_ROLE_SLAVE; } @@ -3573,10 +4209,8 @@ handle_role_request(struct ofproto *ofproto, } ofconn->role = role; - reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid, - &buf); - reply->nxh.vendor = htonl(NX_VENDOR_ID); - reply->nxh.subtype = htonl(NXT_ROLE_REPLY); + reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, msg->header.xid, + &buf); reply->role = htonl(role); queue_tx(buf, ofconn, ofconn->reply_counter); @@ -3584,13 +4218,37 @@ handle_role_request(struct ofproto *ofproto, } static int -handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg) +handle_nxt_set_flow_format(struct ofconn *ofconn, + struct nxt_set_flow_format *msg) +{ + uint32_t format; + int error; + + error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg); + if (error) { + return error; + } + + 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); + } +} + +static int +handle_vendor(struct ofconn *ofconn, void *msg) { + struct ofproto *p = ofconn->ofproto; struct ofp_vendor_header *ovh = msg; struct nicira_header *nh; if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) { - VLOG_WARN_RL(&rl, "received vendor message of length %zu " + VLOG_WARN_RL(&rl, "received vendor message of length %u " "(expected at least %zu)", ntohs(ovh->header.length), sizeof(struct ofp_vendor_header)); return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); @@ -3599,7 +4257,7 @@ handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg) return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); } if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) { - VLOG_WARN_RL(&rl, "received Nicira vendor message of length %zu " + VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u " "(expected at least %zu)", ntohs(ovh->header.length), sizeof(struct nicira_header)); return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); @@ -3612,10 +4270,16 @@ handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg) msg); case NXT_TUN_ID_FROM_COOKIE: - return handle_tun_id_from_cookie(p, msg); + return handle_tun_id_from_cookie(ofconn, msg); case NXT_ROLE_REQUEST: - return handle_role_request(p, ofconn, msg); + return handle_role_request(ofconn, msg); + + case NXT_SET_FLOW_FORMAT: + return handle_nxt_set_flow_format(ofconn, msg); + + case NXT_FLOW_MOD: + return handle_nxt_flow_mod(ofconn, &ovh->header); } return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); @@ -3635,8 +4299,7 @@ handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh) } static void -handle_openflow(struct ofconn *ofconn, struct ofproto *p, - struct ofpbuf *ofp_msg) +handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg) { struct ofp_header *oh = ofp_msg->data; int error; @@ -3652,35 +4315,35 @@ handle_openflow(struct ofconn *ofconn, struct ofproto *p, break; case OFPT_FEATURES_REQUEST: - error = handle_features_request(p, ofconn, oh); + error = handle_features_request(ofconn, oh); break; case OFPT_GET_CONFIG_REQUEST: - error = handle_get_config_request(p, ofconn, oh); + error = handle_get_config_request(ofconn, oh); break; case OFPT_SET_CONFIG: - error = handle_set_config(p, ofconn, ofp_msg->data); + error = handle_set_config(ofconn, ofp_msg->data); break; case OFPT_PACKET_OUT: - error = handle_packet_out(p, ofconn, ofp_msg->data); + error = handle_packet_out(ofconn, ofp_msg->data); break; case OFPT_PORT_MOD: - error = handle_port_mod(p, ofconn, oh); + error = handle_port_mod(ofconn, oh); break; case OFPT_FLOW_MOD: - error = handle_flow_mod(p, ofconn, ofp_msg->data); + error = handle_ofpt_flow_mod(ofconn, ofp_msg->data); break; case OFPT_STATS_REQUEST: - error = handle_stats_request(p, ofconn, oh); + error = handle_stats_request(ofconn, oh); break; case OFPT_VENDOR: - error = handle_vendor(p, ofconn, ofp_msg->data); + error = handle_vendor(ofconn, ofp_msg->data); break; case OFPT_BARRIER_REQUEST: @@ -3708,7 +4371,7 @@ handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) struct odp_msg *msg = packet->data; struct rule *rule; struct ofpbuf payload; - flow_t flow; + struct flow flow; payload.data = msg + 1; payload.size = msg->length - sizeof *msg; @@ -3722,13 +4385,13 @@ handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) 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); + dpif_execute(p->dpif, &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); + struct ofport *port = get_port(p, msg->port); if (port) { if (port->opp.config & OFPPC_NO_PACKET_IN) { COVERAGE_INC(ofproto_no_packet_in); @@ -3759,9 +4422,6 @@ handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) } } - rule_execute(p, rule, &payload, &flow); - rule_reinstall(p, rule); - if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) { /* * Extra-special case for fail-open mode. @@ -3773,10 +4433,13 @@ handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) * * See the top-level comment in fail-open.c for more information. */ - send_packet_in(p, packet); - } else { - ofpbuf_delete(packet); + send_packet_in(p, ofpbuf_clone_with_headroom(packet, + DPIF_RECV_MSG_PADDING)); } + + ofpbuf_pull(packet, sizeof *msg); + rule_execute(p, rule, packet, &flow); + rule_reinstall(p, rule); } static void @@ -3808,6 +4471,290 @@ handle_odp_msg(struct ofproto *p, struct ofpbuf *packet) } } +/* Flow expiration. */ + +struct expire_cbdata { + struct ofproto *ofproto; + int dp_max_idle; +}; + +static int ofproto_dp_max_idle(const struct ofproto *); +static void ofproto_update_used(struct ofproto *); +static void rule_expire(struct cls_rule *, void *cbdata); + +/* 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 expire_cbdata cbdata; + + /* Update 'used' for each flow in the datapath. */ + ofproto_update_used(ofproto); + + /* Expire idle flows. + * + * A wildcarded flow is idle only when all of its subrules have expired due + * to becoming idle, so iterate through the exact-match flows first. */ + cbdata.ofproto = ofproto; + cbdata.dp_max_idle = ofproto_dp_max_idle(ofproto); + classifier_for_each(&ofproto->cls, CLS_INC_EXACT, rule_expire, &cbdata); + classifier_for_each(&ofproto->cls, CLS_INC_WILD, rule_expire, &cbdata); + + /* 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(cbdata.dp_max_idle, 1000); +} + +/* Update 'used' member of each flow currently installed into the datapath. */ +static void +ofproto_update_used(struct ofproto *p) +{ + struct odp_flow *flows; + size_t n_flows; + size_t i; + int error; + + error = dpif_flow_list_all(p->dpif, &flows, &n_flows); + if (error) { + return; + } + + for (i = 0; i < n_flows; i++) { + struct odp_flow *f = &flows[i]; + struct cls_rule target; + struct rule *rule; + struct flow flow; + + odp_flow_key_to_flow(&f->key, &flow); + cls_rule_init_exact(&flow, UINT16_MAX, &target); + + rule = rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, + &target)); + + if (rule && rule->installed) { + update_time(p, rule, &f->stats); + rule_account(p, rule, 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); +} + +/* Calculates and returns the number of milliseconds of idle time after which + * flows 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) +{ + /* + * Idle time histogram. + * + * Most of the time a switch has a relatively small number of flows. 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 flows increases, the memory required to maintain + * statistics about them in userspace and in the kernel becomes + * significant. However, with a large number of flows 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 flows we can discard. + * + * The technique used to compute the idle time is to build a histogram with + * N_BUCKETS bucket whose width is BUCKET_WIDTH msecs each. Each flow 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 flows (but at least 1000 flows) are kept + * cached. At least the most-recently-used bucket of flows is kept, so + * actually an arbitrary number of flows 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 exact-match flows, in addition + * to the pass made by ofproto_update_used(), because the former function + * never looks at uninstallable flows. + */ + enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; + enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; + int buckets[N_BUCKETS] = { 0 }; + int total, bucket; + struct rule *rule; + long long int now; + int i; + + total = classifier_count_exact(&ofproto->cls); + if (total <= 1000) { + return N_BUCKETS * BUCKET_WIDTH; + } + + /* Build histogram. */ + now = time_msec(); + CLASSIFIER_FOR_EACH_EXACT_RULE (rule, cr, &ofproto->cls) { + long long int idle = now - rule->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; + } + } + + if (VLOG_IS_DBG_ENABLED()) { + struct ds s; + + 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]); + } + } + VLOG_INFO("%s: %s (msec:count)", + dpif_name(ofproto->dpif), ds_cstr(&s)); + ds_destroy(&s); + } + + return bucket * BUCKET_WIDTH; +} + +static void +rule_active_timeout(struct ofproto *ofproto, struct rule *rule) +{ + if (ofproto->netflow && !is_controller_rule(rule) && + netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) { + struct ofexpired expired; + struct odp_flow odp_flow; + + /* Get updated flow stats. + * + * 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_from_flow(&odp_flow.key, &rule->cr.flow); + odp_flow.flags = ODPFF_ZERO_TCP_FLAGS; + dpif_flow_get(ofproto->dpif, &odp_flow); + + if (odp_flow.stats.n_packets) { + update_time(ofproto, rule, &odp_flow.stats); + netflow_flow_update_flags(&rule->nf_flow, + odp_flow.stats.tcp_flags); + } + } + + expired.flow = rule->cr.flow; + expired.packet_count = rule->packet_count + + odp_flow.stats.n_packets; + expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes; + expired.used = rule->used; + + netflow_expire(ofproto->netflow, &rule->nf_flow, &expired); + } +} + +/* If 'cls_rule' is an OpenFlow rule, that has expired according to OpenFlow + * rules, then delete it entirely. + * + * If 'cls_rule' is a subrule, that has not been used recently, remove it from + * the datapath and fold its statistics back into its super-rule. + * + * (This is a callback function for classifier_for_each().) */ +static void +rule_expire(struct cls_rule *cls_rule, void *cbdata_) +{ + struct expire_cbdata *cbdata = cbdata_; + struct ofproto *ofproto = cbdata->ofproto; + struct rule *rule = rule_from_cls_rule(cls_rule); + long long int hard_expire, idle_expire, expire, now; + + /* Calculate OpenFlow expiration times for 'rule'. */ + 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); + + now = time_msec(); + if (now < expire) { + /* 'rule' has not expired according to OpenFlow rules. */ + if (!rule->cr.wc.wildcards) { + if (now >= rule->used + cbdata->dp_max_idle) { + /* This rule is idle, so drop it to free up resources. */ + if (rule->super) { + /* It's not part of the OpenFlow flow table, so we can + * delete it entirely and fold its statistics into its + * super-rule. */ + rule_remove(ofproto, rule); + } else { + /* It is part of the OpenFlow flow table, so we have to + * keep the rule but we can at least uninstall it from the + * datapath. */ + rule_uninstall(ofproto, rule); + } + } else { + /* Send NetFlow active timeout if appropriate. */ + rule_active_timeout(cbdata->ofproto, rule); + } + } + } else { + /* 'rule' has expired according to OpenFlow rules. */ + COVERAGE_INC(ofproto_expired); + + /* Update stats. (This is a no-op if the rule expired due to an idle + * timeout, because that only happens when the rule has no subrules + * left.) */ + if (rule->cr.wc.wildcards) { + struct rule *subrule, *next; + LIST_FOR_EACH_SAFE (subrule, next, list, &rule->list) { + rule_remove(cbdata->ofproto, subrule); + } + } else { + rule_uninstall(cbdata->ofproto, rule); + } + + /* Get rid of the rule. */ + if (!rule_is_hidden(rule)) { + send_flow_removed(cbdata->ofproto, rule, + (now >= hard_expire + ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT)); + } + rule_remove(cbdata->ofproto, rule); + } +} + static void revalidate_cb(struct cls_rule *sub_, void *cbdata_) { @@ -3824,12 +4771,13 @@ revalidate_cb(struct cls_rule *sub_, void *cbdata_) static bool revalidate_rule(struct ofproto *p, struct rule *rule) { - const flow_t *flow = &rule->cr.flow; + const struct flow *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)); + super = rule_from_cls_rule(classifier_lookup(&p->cls, flow, + CLS_INC_WILD)); if (!super) { rule_remove(p, rule); return false; @@ -3850,23 +4798,19 @@ revalidate_rule(struct ofproto *p, struct rule *rule) } static struct ofpbuf * -compose_flow_removed(struct ofproto *p, const struct rule *rule, - long long int now, uint8_t reason) +compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule, + uint8_t reason) { struct ofp_flow_removed *ofr; struct ofpbuf *buf; - long long int tdiff = now - rule->created; - uint32_t sec = tdiff / 1000; - uint32_t msec = tdiff - (sec * 1000); ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf); - flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie, + flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, ofconn->flow_format, &ofr->match); ofr->cookie = rule->flow_cookie; ofr->priority = htons(rule->cr.priority); ofr->reason = reason; - ofr->duration_sec = htonl(sec); - ofr->duration_nsec = htonl(msec * 1000000); + 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); @@ -3874,164 +4818,57 @@ compose_flow_removed(struct ofproto *p, const struct rule *rule, return buf; } -static void -uninstall_idle_flow(struct ofproto *ofproto, struct rule *rule) -{ - assert(rule->installed); - assert(!rule->cr.wc.wildcards); - - if (rule->super) { - rule_remove(ofproto, rule); - } else { - rule_uninstall(ofproto, rule); - } -} - -static void -send_flow_removed(struct ofproto *p, struct rule *rule, - long long int now, uint8_t reason) -{ - struct ofconn *ofconn; - struct ofconn *prev; - struct ofpbuf *buf = NULL; - - /* We limit the maximum number of queued flow expirations it by accounting - * them under the counter for replies. That works because preventing - * OpenFlow requests from being processed also prevents new flows from - * being added (and expiring). (It also prevents processing OpenFlow - * requests that would not add new flows, so it is imperfect.) */ - - prev = NULL; - LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { - if (rule->send_flow_removed && rconn_is_connected(ofconn->rconn) - && ofconn_receives_async_msgs(ofconn)) { - if (prev) { - queue_tx(ofpbuf_clone(buf), prev, prev->reply_counter); - } else { - buf = compose_flow_removed(p, rule, now, reason); - } - prev = ofconn; - } - } - if (prev) { - queue_tx(buf, prev, prev->reply_counter); - } -} - - -static void -expire_rule(struct cls_rule *cls_rule, void *p_) -{ - struct ofproto *p = p_; - struct rule *rule = rule_from_cls_rule(cls_rule); - long long int hard_expire, idle_expire, expire, now; - - 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); - - now = time_msec(); - if (now < expire) { - if (rule->installed && now >= rule->used + 5000) { - uninstall_idle_flow(p, rule); - } else if (!rule->cr.wc.wildcards) { - active_timeout(p, rule); - } - - return; - } - - COVERAGE_INC(ofproto_expired); - - /* Update stats. This code will be a no-op if the rule expired - * due to an idle timeout. */ - if (rule->cr.wc.wildcards) { - struct rule *subrule, *next; - LIST_FOR_EACH_SAFE (subrule, next, struct rule, list, &rule->list) { - rule_remove(p, subrule); - } - } else { - rule_uninstall(p, rule); - } - - if (!rule_is_hidden(rule)) { - send_flow_removed(p, rule, now, - (now >= hard_expire - ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT)); - } - rule_remove(p, rule); -} - -static void -active_timeout(struct ofproto *ofproto, struct rule *rule) +static struct ofpbuf * +compose_nx_flow_removed(const struct rule *rule, uint8_t reason) { - if (ofproto->netflow && !is_controller_rule(rule) && - netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) { - struct ofexpired expired; - struct odp_flow odp_flow; - - /* Get updated flow stats. */ - memset(&odp_flow, 0, sizeof odp_flow); - if (rule->installed) { - odp_flow.key = rule->cr.flow; - odp_flow.flags = ODPFF_ZERO_TCP_FLAGS; - dpif_flow_get(ofproto->dpif, &odp_flow); + struct nx_flow_removed *nfr; + struct ofpbuf *buf; + int match_len; - if (odp_flow.stats.n_packets) { - update_time(ofproto, rule, &odp_flow.stats); - netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos, - odp_flow.stats.tcp_flags); - } - } + nfr = make_nxmsg(sizeof *nfr, NXT_FLOW_REMOVED, &buf); - expired.flow = rule->cr.flow; - expired.packet_count = rule->packet_count + - odp_flow.stats.n_packets; - expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes; - expired.used = rule->used; + match_len = nx_put_match(buf, &rule->cr); - netflow_expire(ofproto->netflow, &rule->nf_flow, &expired); + 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); - /* Schedule us to send the accumulated records once we have - * collected all of them. */ - poll_immediate_wake(); - } + return buf; } static void -update_used(struct ofproto *p) +send_flow_removed(struct ofproto *p, struct rule *rule, uint8_t reason) { - struct odp_flow *flows; - size_t n_flows; - size_t i; - int error; + struct ofconn *ofconn; - error = dpif_flow_list_all(p->dpif, &flows, &n_flows); - if (error) { + if (!rule->send_flow_removed) { return; } - for (i = 0; i < n_flows; i++) { - struct odp_flow *f = &flows[i]; - struct rule *rule; + LIST_FOR_EACH (ofconn, node, &p->all_conns) { + struct ofpbuf *msg; - 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); + if (!rconn_is_connected(ofconn->rconn) + || !ofconn_receives_async_msgs(ofconn)) { continue; } - update_time(p, rule, &f->stats); - rule_account(p, rule, f->stats.n_bytes); + 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); } - free(flows); } /* pinsched callback for sending 'packet' on 'ofconn'. */ @@ -4165,7 +5002,7 @@ send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet) max_len = do_convert_to_packet_in(packet); prev = NULL; - LIST_FOR_EACH (ofconn, struct ofconn, node, &ofproto->all_conns) { + LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { if (ofconn_receives_async_msgs(ofconn)) { if (prev) { schedule_packet_in(prev, packet, max_len, true); @@ -4185,7 +5022,7 @@ 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; @@ -4209,7 +5046,7 @@ pick_fallback_dpid(void) } static bool -default_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, +default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, struct odp_actions *actions, tag_type *tags, uint16_t *nf_output_iface, void *ofproto_) { @@ -4224,7 +5061,8 @@ 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. */ @@ -4236,9 +5074,11 @@ 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, actions); } else if (out_port != flow->in_port) { odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port; *nf_output_iface = out_port; @@ -4250,7 +5090,6 @@ 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