X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=ofproto%2Fofproto.c;h=68b2493def2686c41dd661d8a8c9c430dd919808;hb=ca0f572cfe87f284018e14fa7f1de58fbaef4c87;hp=52e4fe3b622ad9f9ec435fceb1c1419f512f7444;hpb=abdfe47476cc1a192e329f428b3740a3fae8390d;p=openvswitch diff --git a/ofproto/ofproto.c b/ofproto/ofproto.c index 52e4fe3b..68b2493d 100644 --- a/ofproto/ofproto.c +++ b/ofproto/ofproto.c @@ -30,6 +30,8 @@ #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" @@ -46,11 +48,9 @@ #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" @@ -69,9 +69,12 @@ enum { TABLEID_CLASSIFIER = 1 }; + 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 *); @@ -86,7 +89,7 @@ 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. + uint64_t flow_cookie; /* Controller-issued identifier. (Kept in network-byte order.) */ uint16_t idle_timeout; /* In seconds from time of last use. */ uint16_t hard_timeout; /* In seconds from time of creation. */ @@ -170,19 +173,44 @@ static void send_flow_removed(struct ofproto *p, struct rule *rule, /* 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. */ @@ -202,7 +230,7 @@ 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. */ @@ -227,6 +255,7 @@ static void ofconn_run(struct ofconn *, struct ofproto *); static void ofconn_wait(struct ofconn *); static bool ofconn_receives_async_msgs(const struct ofconn *); static char *ofconn_make_name(const struct ofproto *, const char *target); +static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst); static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, struct rconn_packet_counter *counter); @@ -247,8 +276,7 @@ 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; @@ -275,8 +303,9 @@ struct ofproto { struct hmap controllers; /* Controller "struct ofconn"s. */ struct list all_conns; /* Contains "struct ofconn"s. */ enum ofproto_fail_mode fail_mode; - struct pvconn **listeners; - size_t n_listeners; + + /* OpenFlow listeners. */ + struct hmap services; /* Contains "struct ofservice"s. */ struct pvconn **snoops; size_t n_snoops; @@ -310,6 +339,7 @@ static void handle_openflow(struct ofconn *, struct ofproto *, static void refresh_port_groups(struct ofproto *); +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 *); @@ -362,7 +392,7 @@ 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; @@ -382,8 +412,7 @@ 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; @@ -452,7 +481,7 @@ 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) { @@ -473,9 +502,7 @@ 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); @@ -491,21 +518,7 @@ 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 * @@ -621,22 +634,38 @@ ofproto_set_controllers(struct ofproto *p, size_t n_controllers) { struct shash new_controllers; - 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]); } + /* 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, + HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, hmap_node, &p->controllers) { struct ofproto_controller *c; @@ -650,10 +679,25 @@ ofproto_set_controllers(struct ofproto *p, } } } + + /* Delete services that are no longer configured. + * Update configuration of all now-existing services. */ + HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, struct ofservice, node, + &p->services) { + struct ofproto_controller *c; + + c = shash_find_data(&new_controllers, + pvconn_get_name(ofservice->pvconn)); + if (!c) { + ofservice_destroy(p, ofservice); + } else { + ofservice_reconfigure(ofservice, c); + } + } + shash_destroy(&new_controllers); update_in_band_remotes(p); - update_fail_open(p); if (!hmap_is_empty(&p->controllers) && !ss_exists) { @@ -813,12 +857,6 @@ 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) { @@ -849,12 +887,11 @@ 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, struct ofport, hmap_node, &ofproto->ports) { + ofproto_sflow_add_port(os, ofport->odp_port, netdev_get_name(ofport->netdev)); } } @@ -865,18 +902,6 @@ 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) { @@ -884,7 +909,7 @@ 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); } @@ -895,16 +920,6 @@ ofproto_get_fail_mode(const struct ofproto *p) return p->fail_mode; } -void -ofproto_get_listeners(const struct ofproto *ofproto, struct svec *listeners) -{ - size_t i; - - for (i = 0; i < ofproto->n_listeners; i++) { - svec_add(listeners, pvconn_get_name(ofproto->listeners[i])); - } -} - void ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops) { @@ -918,9 +933,9 @@ 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) { @@ -946,7 +961,9 @@ ofproto_destroy(struct ofproto *p) 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, struct ofport, hmap_node, + &p->ports) { + hmap_remove(&p->ports, &ofport->hmap_node); ofport_free(ofport); } shash_destroy(&p->port_by_name); @@ -955,10 +972,11 @@ 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, struct 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]); @@ -973,7 +991,7 @@ ofproto_destroy(struct ofproto *p) free(p->serial_desc); free(p->dp_desc); - port_array_destroy(&p->ports); + hmap_destroy(&p->ports); free(p); } @@ -1028,7 +1046,7 @@ 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 + if (ofconn->type == OFCONN_PRIMARY && (!best || snoop_preference(ofconn) > snoop_preference(best))) { best = ofconn; } @@ -1046,6 +1064,7 @@ int ofproto_run1(struct ofproto *p) { struct ofconn *ofconn, *next_ofconn; + struct ofservice *ofservice; char *devname; int error; int i; @@ -1056,7 +1075,6 @@ ofproto_run1(struct ofproto *p) for (i = 0; i < 50; i++) { struct ofpbuf *buf; - int error; error = dpif_recv(p->dpif, &buf); if (error) { @@ -1101,21 +1119,23 @@ ofproto_run1(struct ofproto *p) fail_open_run(p->fail_open); } - for (i = 0; i < p->n_listeners; i++) { + HMAP_FOR_EACH (ofservice, struct 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) { struct rconn *rconn; char *name; - rconn = rconn_create(60, 0); + 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_create(p, rconn, OFCONN_TRANSIENT); + 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)); } @@ -1188,6 +1208,7 @@ ofproto_run2(struct ofproto *p, bool revalidate_all) void ofproto_wait(struct ofproto *p) { + struct ofservice *ofservice; struct ofconn *ofconn; size_t i; @@ -1217,8 +1238,8 @@ ofproto_wait(struct ofproto *p) } else if (p->next_expiration != LLONG_MAX) { poll_timer_wait_until(p->next_expiration); } - for (i = 0; i < p->n_listeners; i++) { - pvconn_wait(p->listeners[i]); + HMAP_FOR_EACH (ofservice, struct ofservice, node, &p->services) { + pvconn_wait(ofservice->pvconn); } for (i = 0; i < p->n_snoops; i++) { pvconn_wait(p->snoops[i]); @@ -1272,7 +1293,7 @@ ofproto_add_flow(struct ofproto *p, { 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_insert(p, rule, NULL, 0); @@ -1326,13 +1347,12 @@ 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; svec_init(&devnames); - PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) { + HMAP_FOR_EACH (ofport, struct ofport, hmap_node, &p->ports) { svec_add (&devnames, (char *) ofport->opp.name); } dpif_port_list(p->dpif, &odp_ports, &n_odp_ports); @@ -1354,15 +1374,14 @@ 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); + ports = xmalloc(hmap_count(&p->ports) * sizeof *ports); n_ports = 0; - PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { + HMAP_FOR_EACH (port, struct ofport, hmap_node, &p->ports) { if (group == DP_GROUP_ALL || !(port->opp.config & OFPPC_NO_FLOOD)) { - ports[n_ports++] = port_no; + ports[n_ports++] = port->odp_port; } } dpif_port_group_set(p->dpif, group, ports, n_ports); @@ -1406,6 +1425,7 @@ 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, @@ -1427,7 +1447,7 @@ 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; @@ -1486,28 +1506,25 @@ send_port_status(struct ofproto *p, const struct ofport *ofport, static void ofport_install(struct ofproto *p, struct ofport *ofport) { - uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no); const char *netdev_name = (const char *) ofport->opp.name; netdev_monitor_add(p->netdev_monitor, ofport->netdev); - port_array_set(&p->ports, odp_port, ofport); + hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0)); shash_add(&p->port_by_name, netdev_name, ofport); if (p->sflow) { - ofproto_sflow_add_port(p->sflow, odp_port, netdev_name); + ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name); } } static void ofport_remove(struct ofproto *p, struct ofport *ofport) { - uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no); - netdev_monitor_remove(p->netdev_monitor, ofport->netdev); - port_array_delete(&p->ports, odp_port); + hmap_remove(&p->ports, &ofport->hmap_node); shash_delete(&p->port_by_name, shash_find(&p->port_by_name, (char *) ofport->opp.name)); if (p->sflow) { - ofproto_sflow_del_port(p->sflow, odp_port); + ofproto_sflow_del_port(p->sflow, ofport->odp_port); } } @@ -1520,6 +1537,20 @@ 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, struct ofport, 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) { @@ -1547,7 +1578,7 @@ 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 " @@ -1638,7 +1669,7 @@ 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); @@ -1724,14 +1755,13 @@ 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; } } @@ -1747,6 +1777,85 @@ ofconn_make_name(const struct ofproto *ofproto, const char *target) { return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target); } + +static void +ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst) +{ + int i; + + for (i = 0; i < N_SCHEDULERS; i++) { + struct pinsched **s = &ofconn->schedulers[i]; + + if (rate > 0) { + if (!*s) { + *s = pinsched_create(rate, burst, + ofconn->ofproto->switch_status); + } else { + pinsched_set_limits(*s, rate, burst); + } + } else { + pinsched_destroy(*s); + *s = NULL; + } + } +} + +static void +ofservice_reconfigure(struct ofservice *ofservice, + const struct ofproto_controller *c) +{ + ofservice->probe_interval = c->probe_interval; + ofservice->rate_limit = c->rate_limit; + ofservice->burst_limit = c->burst_limit; +} + +/* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a + * positive errno value. */ +static int +ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c) +{ + struct ofservice *ofservice; + struct pvconn *pvconn; + int error; + + error = pvconn_open(c->target, &pvconn); + if (error) { + return error; + } + + ofservice = xzalloc(sizeof *ofservice); + hmap_insert(&ofproto->services, &ofservice->node, + hash_string(c->target, 0)); + ofservice->pvconn = pvconn; + + ofservice_reconfigure(ofservice, c); + + return 0; +} + +static void +ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice) +{ + hmap_remove(&ofproto->services, &ofservice->node); + pvconn_close(ofservice->pvconn); + free(ofservice); +} + +/* Finds and returns the ofservice within 'ofproto' that has the given + * 'target', or a null pointer if none exists. */ +static struct ofservice * +ofservice_lookup(struct ofproto *ofproto, const char *target) +{ + struct ofservice *ofservice; + + HMAP_FOR_EACH_WITH_HASH (ofservice, struct ofservice, node, + hash_string(target, 0), &ofproto->services) { + if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) { + return ofservice; + } + } + return NULL; +} /* Caller is responsible for initializing the 'cr' member of the returned * rule. */ @@ -1830,8 +1939,44 @@ rule_has_out_port(const struct rule *rule, uint16_t out_port) 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, in_port, + 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, @@ -1843,15 +1988,20 @@ 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) { 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 @@ -1862,6 +2012,7 @@ 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; @@ -1872,16 +2023,21 @@ 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) @@ -2062,7 +2218,7 @@ 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; } @@ -2197,7 +2353,6 @@ handle_features_request(struct ofproto *p, struct ofconn *ofconn, { struct ofp_switch_features *osf; struct ofpbuf *buf; - unsigned int port_no; struct ofport *port; osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf); @@ -2219,7 +2374,7 @@ handle_features_request(struct ofproto *p, struct ofconn *ofconn, (1u << OFPAT_SET_TP_DST) | (1u << OFPAT_ENQUEUE)); - PORT_ARRAY_FOR_EACH (port, &p->ports, port_no) { + HMAP_FOR_EACH (port, struct ofport, hmap_node, &p->ports) { hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp)); } @@ -2262,7 +2417,7 @@ 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); @@ -2317,13 +2472,17 @@ 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) { @@ -2364,7 +2523,7 @@ 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; @@ -2385,6 +2544,11 @@ 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); } } @@ -2525,6 +2689,12 @@ 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; + /* If you add a new action here that modifies flow data, don't forget to * update the flow key in ctx->flow at the same time. */ @@ -2542,9 +2712,9 @@ 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; @@ -2666,6 +2836,7 @@ 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); } @@ -2680,7 +2851,7 @@ xlate_actions(const union ofp_action *in, size_t n_in, static int reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh) { - 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; @@ -2790,8 +2961,7 @@ 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)) { @@ -2916,19 +3086,19 @@ 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->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); @@ -2953,7 +3123,6 @@ handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn, 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); @@ -2962,14 +3131,13 @@ handle_port_stats_request(struct ofproto *p, struct ofconn *ofconn, msg = start_stats_reply(osr, sizeof *ops * 16); if (psr->port_no != htons(OFPP_NONE)) { - port = port_array_get(&p->ports, - ofp_port_to_odp_port(ntohs(psr->port_no))); + 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, struct ofport, hmap_node, &p->ports) { + append_port_stat(port, ofconn, &msg); } } @@ -3147,7 +3315,7 @@ flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_) 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) @@ -3231,8 +3399,8 @@ handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn, struct queue_stats_cbdata { struct ofconn *ofconn; + struct ofport *ofport; struct ofpbuf *msg; - uint16_t port_no; }; static void @@ -3242,7 +3410,7 @@ put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id, struct ofp_queue_stats *reply; reply = append_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg); - reply->port_no = htons(cbdata->port_no); + reply->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); @@ -3261,11 +3429,10 @@ handle_queue_stats_dump_cb(uint32_t queue_id, } static void -handle_queue_stats_for_port(struct ofport *port, uint16_t port_no, - uint32_t queue_id, +handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id, struct queue_stats_cbdata *cbdata) { - cbdata->port_no = port_no; + cbdata->ofport = port; if (queue_id == OFPQ_ALL) { netdev_dump_queue_stats(port->netdev, handle_queue_stats_dump_cb, cbdata); @@ -3301,13 +3468,13 @@ handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn, port_no = ntohs(qsr->port_no); queue_id = ntohl(qsr->queue_id); if (port_no == OFPP_ALL) { - PORT_ARRAY_FOR_EACH (port, &ofproto->ports, port_no) { - handle_queue_stats_for_port(port, port_no, queue_id, &cbdata); + HMAP_FOR_EACH (port, struct ofport, hmap_node, &ofproto->ports) { + handle_queue_stats_for_port(port, queue_id, &cbdata); } } else if (port_no < ofproto->max_ports) { - port = port_array_get(&ofproto->ports, port_no); + port = get_port(ofproto, ofp_port_to_odp_port(port_no)); if (port) { - handle_queue_stats_for_port(port, port_no, queue_id, &cbdata); + handle_queue_stats_for_port(port, queue_id, &cbdata); } } else { ofpbuf_delete(cbdata.msg); @@ -3388,8 +3555,7 @@ update_stats(struct ofproto *ofproto, struct rule *rule, 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); + netflow_flow_update_flags(&rule->nf_flow, stats->tcp_flags); } } @@ -3440,7 +3606,6 @@ add_flow(struct ofproto *p, struct ofconn *ofconn, } rule_insert(p, rule, packet, in_port); - ofpbuf_delete(packet); return error; } @@ -3478,7 +3643,6 @@ send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn, flow_extract(packet, 0, in_port, &flow); rule_execute(ofproto, rule, packet, &flow); - ofpbuf_delete(packet); return 0; } @@ -3773,7 +3937,7 @@ handle_role_request(struct ofproto *ofproto, } 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); @@ -3955,7 +4119,7 @@ handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) 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); @@ -3986,9 +4150,6 @@ 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. @@ -4000,10 +4161,13 @@ 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 @@ -4211,7 +4375,7 @@ active_timeout(struct ofproto *ofproto, struct rule *rule) 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, + netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.tcp_flags); } } @@ -4412,7 +4576,7 @@ 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;