X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=ofproto%2Fofproto.c;h=2ec7c8c4a035385b37a1d806029a929f384f49cb;hb=85da620e9788b473797b95212b916327974e6942;hp=fcc76d6cbfee19f9b75d6e1d30f9fccdd3e9ea7d;hpb=5f55c39b21e69025045437ffbd3bb98fe6ce2e89;p=sliver-openvswitch.git diff --git a/ofproto/ofproto.c b/ofproto/ofproto.c index fcc76d6cb..2ec7c8c4a 100644 --- a/ofproto/ofproto.c +++ b/ofproto/ofproto.c @@ -1,5 +1,6 @@ /* * Copyright (c) 2009, 2010 Nicira Networks. + * Copyright (c) 2010 Jean Tourrilhes - HP-Labs. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -18,6 +19,7 @@ #include "ofproto.h" #include #include +#include #include #include #include @@ -28,12 +30,15 @@ #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 "odp-util.h" #include "ofp-print.h" +#include "ofp-util.h" #include "ofproto-sflow.h" #include "ofpbuf.h" #include "openflow/nicira-ext.h" @@ -43,21 +48,19 @@ #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 "vlog.h" #include "xtoxll.h" -#define THIS_MODULE VLM_ofproto -#include "vlog.h" +VLOG_DEFINE_THIS_MODULE(ofproto) #include "sflow_api.h" @@ -66,16 +69,19 @@ 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 *); 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); @@ -83,7 +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. */ @@ -108,6 +114,9 @@ struct rule { struct list list; /* OpenFlow actions. + * + * 'n_actions' is the number of elements in the 'actions' array. A single + * action may take up more more than one element's worth of space. * * A subrule has no actions (it uses the super-rule's actions). */ int n_actions; @@ -162,27 +171,98 @@ static void rule_post_uninstall(struct ofproto *, struct rule *); static void send_flow_removed(struct ofproto *p, struct rule *rule, long long int now, uint8_t reason); -struct ofconn { - struct list node; - struct rconn *rconn; - struct pktbuf *pktbuf; - int miss_send_len; +/* ofproto supports two kinds of OpenFlow connections: + * + * - "Primary" connections to ordinary OpenFlow controllers. ofproto + * maintains persistent connections to these controllers and by default + * sends them asynchronous messages such as packet-ins. + * + * - "Service" connections, e.g. from ovs-ofctl. When these connections + * drop, it is the other side's responsibility to reconnect them if + * necessary. ofproto does not send them asynchronous messages by default. + * + * Currently, active (tcp, ssl, unix) connections are always "primary" + * connections and passive (ptcp, pssl, punix) connections are always "service" + * connections. There is no inherent reason for this, but it reflects the + * common case. + */ +enum ofconn_type { + OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */ + OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */ +}; + +/* A listener for incoming OpenFlow "service" connections. */ +struct ofservice { + struct hmap_node node; /* In struct ofproto's "services" hmap. */ + struct pvconn *pvconn; /* OpenFlow connection listener. */ - struct rconn_packet_counter *packet_in_counter; + /* 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 *); - /* Number of OpenFlow messages queued as replies to OpenFlow requests, and - * the maximum number before we stop reading OpenFlow requests. */ +/* An OpenFlow connection. */ +struct ofconn { + struct ofproto *ofproto; /* The ofproto that owns this connection. */ + struct list node; /* In struct ofproto's "all_conns" list. */ + struct rconn *rconn; /* OpenFlow connection. */ + enum ofconn_type type; /* Type. */ + + /* OFPT_PACKET_IN related data. */ + struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */ + struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */ + struct pktbuf *pktbuf; /* OpenFlow packet buffers. */ + int miss_send_len; /* Bytes to send of buffered packets. */ + + /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow + * requests, and the maximum number before we stop reading OpenFlow + * requests. */ #define OFCONN_REPLY_MAX 100 struct rconn_packet_counter *reply_counter; + + /* type == OFCONN_PRIMARY only. */ + enum nx_role role; /* Role. */ + struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */ + struct discovery *discovery; /* Controller discovery object, if enabled. */ + struct status_category *ss; /* Switch status category. */ + enum ofproto_band band; /* In-band or out-of-band? */ }; -static struct ofconn *ofconn_create(struct ofproto *, struct rconn *); +/* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's + * "schedulers" array. Their values are 0 and 1, and their meanings and values + * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In + * case anything ever changes, check their values here. */ +#define N_SCHEDULERS 2 +BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0); +BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR); +BUILD_ASSERT_DECL(OFPR_ACTION == 1); +BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR); + +static struct ofconn *ofconn_create(struct ofproto *, struct rconn *, + enum ofconn_type); static void ofconn_destroy(struct ofconn *); static void ofconn_run(struct ofconn *, struct ofproto *); static void ofconn_wait(struct ofconn *); +static bool ofconn_receives_async_msgs(const struct ofconn *); +static char *ofconn_make_name(const struct ofproto *, const char *target); +static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst); + static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, struct rconn_packet_counter *counter); +static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg); +static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn); + struct ofproto { /* Settings. */ uint64_t datapath_id; /* Datapath ID. */ @@ -196,32 +276,36 @@ struct ofproto { /* Datapath. */ struct dpif *dpif; struct netdev_monitor *netdev_monitor; - struct port_array ports; /* Index is ODP port nr; ofport->opp.port_no is - * OFP port nr. */ + struct hmap ports; /* Contains "struct ofport"s. */ struct shash port_by_name; uint32_t max_ports; /* Configuration. */ struct switch_status *switch_status; - struct status_category *ss_cat; - struct in_band *in_band; - struct discovery *discovery; struct fail_open *fail_open; - struct pinsched *miss_sched, *action_sched; struct netflow *netflow; struct ofproto_sflow *sflow; + /* In-band control. */ + struct in_band *in_band; + long long int next_in_band_update; + struct sockaddr_in *extra_in_band_remotes; + size_t n_extra_remotes; + /* Flow table. */ struct classifier cls; bool need_revalidate; long long int next_expiration; struct tag_set revalidate_set; + bool tun_id_from_cookie; /* OpenFlow connections. */ - struct list all_conns; - struct ofconn *controller; - struct pvconn **listeners; - size_t n_listeners; + struct hmap controllers; /* Controller "struct ofconn"s. */ + struct list all_conns; /* Contains "struct ofconn"s. */ + enum ofproto_fail_mode fail_mode; + + /* OpenFlow listeners. */ + struct hmap services; /* Contains "struct ofservice"s. */ struct pvconn **snoops; size_t n_snoops; @@ -239,13 +323,11 @@ static const struct ofhooks default_ofhooks; static uint64_t pick_datapath_id(const struct ofproto *); static uint64_t pick_fallback_dpid(void); -static void send_packet_in_miss(struct ofpbuf *, void *ofproto); -static void send_packet_in_action(struct ofpbuf *, void *ofproto); -static void update_used(struct ofproto *); + +static 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_); @@ -254,8 +336,7 @@ 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 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 *); @@ -308,16 +389,14 @@ 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; /* Initialize submodules. */ p->switch_status = switch_status_create(p); p->in_band = NULL; - p->discovery = NULL; p->fail_open = NULL; - p->miss_sched = p->action_sched = NULL; p->netflow = NULL; p->sflow = NULL; @@ -329,11 +408,8 @@ ofproto_create(const char *datapath, const char *datapath_type, /* Initialize OpenFlow connections. */ list_init(&p->all_conns); - p->controller = ofconn_create(p, rconn_create(5, 8)); - p->controller->pktbuf = pktbuf_create(); - p->controller->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN; - p->listeners = NULL; - p->n_listeners = 0; + hmap_init(&p->controllers); + hmap_init(&p->services); p->snoops = NULL; p->n_snoops = 0; @@ -348,10 +424,6 @@ ofproto_create(const char *datapath, const char *datapath_type, p->ml = mac_learning_create(); } - /* Register switch status category. */ - p->ss_cat = switch_status_register(p->switch_status, "remote", - rconn_status_cb, p->controller->rconn); - /* Pick final datapath ID. */ p->datapath_id = pick_datapath_id(p); VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id); @@ -367,25 +439,328 @@ ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id) p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p); if (p->datapath_id != old_dpid) { VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id); - rconn_reconnect(p->controller->rconn); + + /* Force all active connections to reconnect, since there is no way to + * notify a controller that the datapath ID has changed. */ + ofproto_reconnect_controllers(p); + } +} + +static bool +is_discovery_controller(const struct ofproto_controller *c) +{ + return !strcmp(c->target, "discover"); +} + +static bool +is_in_band_controller(const struct ofproto_controller *c) +{ + return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND; +} + +/* Creates a new controller in 'ofproto'. Some of the settings are initially + * drawn from 'c', but update_controller() needs to be called later to finish + * the new ofconn's configuration. */ +static void +add_controller(struct ofproto *ofproto, const struct ofproto_controller *c) +{ + struct discovery *discovery; + struct ofconn *ofconn; + + if (is_discovery_controller(c)) { + int error = discovery_create(c->accept_re, c->update_resolv_conf, + ofproto->dpif, ofproto->switch_status, + &discovery); + if (error) { + return; + } + } else { + discovery = NULL; + } + + ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY); + ofconn->pktbuf = pktbuf_create(); + ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN; + if (discovery) { + ofconn->discovery = discovery; + } else { + char *name = ofconn_make_name(ofproto, c->target); + rconn_connect(ofconn->rconn, c->target, name); + free(name); + } + hmap_insert(&ofproto->controllers, &ofconn->hmap_node, + hash_string(c->target, 0)); +} + +/* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's + * target or turn discovery on or off (these are done by creating new ofconns + * and deleting old ones), but it can update the rest of an ofconn's + * settings. */ +static void +update_controller(struct ofconn *ofconn, const struct ofproto_controller *c) +{ + int probe_interval; + + ofconn->band = (is_in_band_controller(c) + ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND); + + rconn_set_max_backoff(ofconn->rconn, c->max_backoff); + + probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0; + rconn_set_probe_interval(ofconn->rconn, probe_interval); + + if (ofconn->discovery) { + discovery_set_update_resolv_conf(ofconn->discovery, + c->update_resolv_conf); + discovery_set_accept_controller_re(ofconn->discovery, c->accept_re); + } + + ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit); +} + +static const char * +ofconn_get_target(const struct ofconn *ofconn) +{ + return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn); +} + +static struct ofconn * +find_controller_by_target(struct ofproto *ofproto, const char *target) +{ + struct ofconn *ofconn; + + HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node, + hash_string(target, 0), &ofproto->controllers) { + if (!strcmp(ofconn_get_target(ofconn), target)) { + return ofconn; + } + } + return NULL; +} + +static void +update_in_band_remotes(struct ofproto *ofproto) +{ + const struct ofconn *ofconn; + struct sockaddr_in *addrs; + size_t max_addrs, n_addrs; + bool discovery; + size_t i; + + /* Allocate enough memory for as many remotes as we could possibly have. */ + max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers); + addrs = xmalloc(max_addrs * sizeof *addrs); + n_addrs = 0; + + /* Add all the remotes. */ + discovery = false; + HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) { + struct sockaddr_in *sin = &addrs[n_addrs]; + + if (ofconn->band == OFPROTO_OUT_OF_BAND) { + continue; + } + + sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn); + if (sin->sin_addr.s_addr) { + sin->sin_port = rconn_get_remote_port(ofconn->rconn); + n_addrs++; + } + if (ofconn->discovery) { + discovery = true; + } + } + for (i = 0; i < ofproto->n_extra_remotes; i++) { + addrs[n_addrs++] = ofproto->extra_in_band_remotes[i]; + } + + /* Create or update or destroy in-band. + * + * Ordinarily we only enable in-band if there's at least one remote + * address, but discovery needs the in-band rules for DHCP to be installed + * even before we know any remote addresses. */ + if (n_addrs || discovery) { + if (!ofproto->in_band) { + in_band_create(ofproto, ofproto->dpif, ofproto->switch_status, + &ofproto->in_band); + } + if (ofproto->in_band) { + in_band_set_remotes(ofproto->in_band, addrs, n_addrs); + } + ofproto->next_in_band_update = time_msec() + 1000; + } else { + in_band_destroy(ofproto->in_band); + ofproto->in_band = NULL; + } + + /* Clean up. */ + free(addrs); +} + +static void +update_fail_open(struct ofproto *p) +{ + struct ofconn *ofconn; + + if (!hmap_is_empty(&p->controllers) + && p->fail_mode == OFPROTO_FAIL_STANDALONE) { + struct rconn **rconns; + size_t n; + + if (!p->fail_open) { + p->fail_open = fail_open_create(p, p->switch_status); + } + + n = 0; + rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns); + HMAP_FOR_EACH (ofconn, 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_probe_interval(struct ofproto *p, int probe_interval) +ofproto_set_controllers(struct ofproto *p, + const struct ofproto_controller *controllers, + size_t n_controllers) { - probe_interval = probe_interval ? MAX(probe_interval, 5) : 0; - rconn_set_probe_interval(p->controller->rconn, probe_interval); - if (p->fail_open) { - int trigger_duration = probe_interval ? probe_interval * 3 : 15; - fail_open_set_trigger_duration(p->fail_open, trigger_duration); + struct shash new_controllers; + struct ofconn *ofconn, *next_ofconn; + struct ofservice *ofservice, *next_ofservice; + bool ss_exists; + size_t i; + + /* Create newly configured controllers and services. + * Create a name to ofproto_controller mapping in 'new_controllers'. */ + shash_init(&new_controllers); + for (i = 0; i < n_controllers; i++) { + const struct ofproto_controller *c = &controllers[i]; + + if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) { + if (!find_controller_by_target(p, c->target)) { + add_controller(p, c); + } + } else if (!pvconn_verify_name(c->target)) { + if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) { + continue; + } + } else { + VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"", + dpif_name(p->dpif), c->target); + continue; + } + + shash_add_once(&new_controllers, c->target, &controllers[i]); + } + + /* Delete controllers that are no longer configured. + * Update configuration of all now-existing controllers. */ + ss_exists = false; + HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) { + struct ofproto_controller *c; + + c = shash_find_data(&new_controllers, ofconn_get_target(ofconn)); + if (!c) { + ofconn_destroy(ofconn); + } else { + update_controller(ofconn, c); + if (ofconn->ss) { + ss_exists = true; + } + } + } + + /* Delete services that are no longer configured. + * Update configuration of all now-existing services. */ + HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) { + struct ofproto_controller *c; + + c = shash_find_data(&new_controllers, + pvconn_get_name(ofservice->pvconn)); + if (!c) { + ofservice_destroy(p, ofservice); + } else { + ofservice_reconfigure(ofservice, c); + } + } + + shash_destroy(&new_controllers); + + update_in_band_remotes(p); + update_fail_open(p); + + if (!hmap_is_empty(&p->controllers) && !ss_exists) { + ofconn = CONTAINER_OF(hmap_first(&p->controllers), + struct ofconn, hmap_node); + ofconn->ss = switch_status_register(p->switch_status, "remote", + rconn_status_cb, ofconn->rconn); + } +} + +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) +{ + size_t i; + + if (n != ofproto->n_extra_remotes) { + return true; + } + + for (i = 0; i < n; i++) { + const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i]; + const struct sockaddr_in *new = &extras[i]; + + if (old->sin_addr.s_addr != new->sin_addr.s_addr || + old->sin_port != new->sin_port) { + return true; + } + } + + return false; +} + +/* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s + * in-band control should guarantee access, in the same way that in-band + * control guarantees access to OpenFlow controllers. */ void -ofproto_set_max_backoff(struct ofproto *p, int max_backoff) +ofproto_set_extra_in_band_remotes(struct ofproto *ofproto, + const struct sockaddr_in *extras, size_t n) { - rconn_set_max_backoff(p->controller->rconn, max_backoff); + if (!any_extras_changed(ofproto, extras, n)) { + return; + } + + free(ofproto->extra_in_band_remotes); + ofproto->n_extra_remotes = n; + ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras); + + update_in_band_remotes(ofproto); } void @@ -439,68 +814,6 @@ ofproto_set_desc(struct ofproto *p, } } -int -ofproto_set_in_band(struct ofproto *p, bool in_band) -{ - if (in_band != (p->in_band != NULL)) { - if (in_band) { - return in_band_create(p, p->dpif, p->switch_status, - p->controller->rconn, &p->in_band); - } else { - ofproto_set_discovery(p, false, NULL, true); - in_band_destroy(p->in_band); - p->in_band = NULL; - } - rconn_reconnect(p->controller->rconn); - } - return 0; -} - -int -ofproto_set_discovery(struct ofproto *p, bool discovery, - const char *re, bool update_resolv_conf) -{ - if (discovery != (p->discovery != NULL)) { - if (discovery) { - int error = ofproto_set_in_band(p, true); - if (error) { - return error; - } - error = discovery_create(re, update_resolv_conf, - p->dpif, p->switch_status, - &p->discovery); - if (error) { - return error; - } - } else { - discovery_destroy(p->discovery); - p->discovery = NULL; - } - rconn_disconnect(p->controller->rconn); - } else if (discovery) { - discovery_set_update_resolv_conf(p->discovery, update_resolv_conf); - return discovery_set_accept_controller_re(p->discovery, re); - } - return 0; -} - -int -ofproto_set_controller(struct ofproto *ofproto, const char *controller) -{ - if (ofproto->discovery) { - return EINVAL; - } else if (controller) { - if (strcmp(rconn_get_name(ofproto->controller->rconn), controller)) { - return rconn_connect(ofproto->controller->rconn, controller); - } else { - return 0; - } - } else { - rconn_disconnect(ofproto->controller->rconn); - return 0; - } -} - static int set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp, const struct svec *svec) @@ -539,12 +852,6 @@ 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) { @@ -575,12 +882,10 @@ 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)); } } @@ -591,105 +896,22 @@ ofproto_set_sflow(struct ofproto *ofproto, } } -void -ofproto_set_failure(struct ofproto *ofproto, bool fail_open) -{ - if (fail_open) { - struct rconn *rconn = ofproto->controller->rconn; - int trigger_duration = rconn_get_probe_interval(rconn) * 3; - if (!ofproto->fail_open) { - ofproto->fail_open = fail_open_create(ofproto, trigger_duration, - ofproto->switch_status, - rconn); - } else { - fail_open_set_trigger_duration(ofproto->fail_open, - trigger_duration); - } - } else { - fail_open_destroy(ofproto->fail_open); - ofproto->fail_open = NULL; - } -} - -void -ofproto_set_rate_limit(struct ofproto *ofproto, - int rate_limit, int burst_limit) -{ - if (rate_limit > 0) { - if (!ofproto->miss_sched) { - ofproto->miss_sched = pinsched_create(rate_limit, burst_limit, - ofproto->switch_status); - ofproto->action_sched = pinsched_create(rate_limit, burst_limit, - NULL); - } else { - pinsched_set_limits(ofproto->miss_sched, rate_limit, burst_limit); - pinsched_set_limits(ofproto->action_sched, - rate_limit, burst_limit); - } - } else { - pinsched_destroy(ofproto->miss_sched); - ofproto->miss_sched = NULL; - pinsched_destroy(ofproto->action_sched); - ofproto->action_sched = NULL; - } -} - -int -ofproto_set_stp(struct ofproto *ofproto 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) { return ofproto->datapath_id; } -int -ofproto_get_probe_interval(const struct ofproto *ofproto) -{ - return rconn_get_probe_interval(ofproto->controller->rconn); -} - -int -ofproto_get_max_backoff(const struct ofproto *ofproto) -{ - return rconn_get_max_backoff(ofproto->controller->rconn); -} - -bool -ofproto_get_in_band(const struct ofproto *ofproto) -{ - return ofproto->in_band != NULL; -} - bool -ofproto_get_discovery(const struct ofproto *ofproto) -{ - return ofproto->discovery != NULL; -} - -const char * -ofproto_get_controller(const struct ofproto *ofproto) +ofproto_has_primary_controller(const struct ofproto *ofproto) { - return rconn_get_name(ofproto->controller->rconn); + 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 @@ -705,47 +927,47 @@ ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops) void ofproto_destroy(struct ofproto *p) { + struct ofservice *ofservice, *next_ofservice; struct ofconn *ofconn, *next_ofconn; - struct ofport *ofport; - unsigned int port_no; + struct ofport *ofport, *next_ofport; size_t i; if (!p) { return; } - /* Destroy fail-open early, because it touches the classifier. */ - ofproto_set_failure(p, false); + /* Destroy fail-open and in-band early, since they touch the classifier. */ + fail_open_destroy(p->fail_open); + p->fail_open = NULL; + + in_band_destroy(p->in_band); + p->in_band = NULL; + free(p->extra_in_band_remotes); ofproto_flush_flows(p); classifier_destroy(&p->cls); - LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node, - &p->all_conns) { + LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) { ofconn_destroy(ofconn); } + hmap_destroy(&p->controllers); dpif_close(p->dpif); netdev_monitor_destroy(p->netdev_monitor); - PORT_ARRAY_FOR_EACH (ofport, &p->ports, port_no) { + HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) { + hmap_remove(&p->ports, &ofport->hmap_node); ofport_free(ofport); } shash_destroy(&p->port_by_name); switch_status_destroy(p->switch_status); - in_band_destroy(p->in_band); - discovery_destroy(p->discovery); - pinsched_destroy(p->miss_sched); - pinsched_destroy(p->action_sched); netflow_destroy(p->netflow); ofproto_sflow_destroy(p->sflow); - switch_status_unregister(p->ss_cat); - - for (i = 0; i < p->n_listeners; i++) { - pvconn_close(p->listeners[i]); + HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) { + ofservice_destroy(p, ofservice); } - free(p->listeners); + hmap_destroy(&p->services); for (i = 0; i < p->n_snoops; i++) { pvconn_close(p->snoops[i]); @@ -760,7 +982,7 @@ ofproto_destroy(struct ofproto *p) free(p->serial_desc); free(p->dp_desc); - port_array_destroy(&p->ports); + hmap_destroy(&p->ports); free(p); } @@ -786,21 +1008,64 @@ process_port_change(struct ofproto *ofproto, int error, char *devname) } } -int -ofproto_run1(struct ofproto *p) -{ - struct ofconn *ofconn, *next_ofconn; - char *devname; - int error; - int i; - - if (shash_is_empty(&p->port_by_name)) { - init_ports(p); +/* Returns a "preference level" for snooping 'ofconn'. A higher return value + * means that 'ofconn' is more interesting for monitoring than a lower return + * value. */ +static int +snoop_preference(const struct ofconn *ofconn) +{ + switch (ofconn->role) { + case NX_ROLE_MASTER: + return 3; + case NX_ROLE_OTHER: + return 2; + case NX_ROLE_SLAVE: + return 1; + default: + /* Shouldn't happen. */ + return 0; } +} - for (i = 0; i < 50; i++) { +/* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'. + * Connects this vconn to a controller. */ +static void +add_snooper(struct ofproto *ofproto, struct vconn *vconn) +{ + struct ofconn *ofconn, *best; + + /* Pick a controller for monitoring. */ + best = NULL; + LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { + if (ofconn->type == OFCONN_PRIMARY + && (!best || snoop_preference(ofconn) > snoop_preference(best))) { + best = ofconn; + } + } + + if (best) { + rconn_add_monitor(best->rconn, vconn); + } else { + VLOG_INFO_RL(&rl, "no controller connection to snoop"); + vconn_close(vconn); + } +} + +int +ofproto_run1(struct ofproto *p) +{ + struct ofconn *ofconn, *next_ofconn; + struct ofservice *ofservice; + char *devname; + int error; + int i; + + if (shash_is_empty(&p->port_by_name)) { + init_ports(p); + } + + for (i = 0; i < 50; i++) { struct ofpbuf *buf; - int error; error = dpif_recv(p->dpif, &buf); if (error) { @@ -828,26 +1093,13 @@ ofproto_run1(struct ofproto *p) } if (p->in_band) { - in_band_run(p->in_band); - } - if (p->discovery) { - char *controller_name; - if (rconn_is_connectivity_questionable(p->controller->rconn)) { - discovery_question_connectivity(p->discovery); - } - if (discovery_run(p->discovery, &controller_name)) { - if (controller_name) { - rconn_connect(p->controller->rconn, controller_name); - } else { - rconn_disconnect(p->controller->rconn); - } + if (time_msec() >= p->next_in_band_update) { + update_in_band_remotes(p); } + in_band_run(p->in_band); } - pinsched_run(p->miss_sched, send_packet_in_miss, p); - pinsched_run(p->action_sched, send_packet_in_action, p); - LIST_FOR_EACH_SAFE (ofconn, next_ofconn, struct ofconn, node, - &p->all_conns) { + LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) { ofconn_run(ofconn, p); } @@ -857,13 +1109,23 @@ ofproto_run1(struct ofproto *p) fail_open_run(p->fail_open); } - for (i = 0; i < p->n_listeners; i++) { + HMAP_FOR_EACH (ofservice, node, &p->services) { struct vconn *vconn; int retval; - retval = pvconn_accept(p->listeners[i], OFP_VERSION, &vconn); + retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn); if (!retval) { - ofconn_create(p, rconn_new_from_vconn("passive", vconn)); + struct rconn *rconn; + char *name; + + rconn = rconn_create(ofservice->probe_interval, 0); + name = ofconn_make_name(p, vconn_get_name(vconn)); + rconn_connect_unreliably(rconn, vconn, name); + free(name); + + ofconn = ofconn_create(p, rconn, OFCONN_SERVICE); + ofconn_set_rate_limit(ofconn, ofservice->rate_limit, + ofservice->burst_limit); } else if (retval != EAGAIN) { VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); } @@ -875,26 +1137,16 @@ ofproto_run1(struct ofproto *p) retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn); if (!retval) { - rconn_add_monitor(p->controller->rconn, vconn); + add_snooper(p, vconn); } else if (retval != EAGAIN) { VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); } } if (time_msec() >= p->next_expiration) { + int delay = ofproto_expire(p); + p->next_expiration = time_msec() + delay; COVERAGE_INC(ofproto_expiration); - p->next_expiration = time_msec() + 1000; - update_used(p); - - classifier_for_each(&p->cls, CLS_INC_ALL, expire_rule, p); - - /* Let the hook know that we're at a stable point: all outstanding data - * in existing flows has been accounted to the account_cb. Thus, the - * hook can now reasonably do operations that depend on having accurate - * flow volume accounting (currently, that's just bond rebalancing). */ - if (p->ofhooks->account_checkpoint_cb) { - p->ofhooks->account_checkpoint_cb(p->aux); - } } if (p->netflow) { @@ -936,26 +1188,23 @@ ofproto_run2(struct ofproto *p, bool revalidate_all) void ofproto_wait(struct ofproto *p) { + struct ofservice *ofservice; struct ofconn *ofconn; size_t i; dpif_recv_wait(p->dpif); dpif_port_poll_wait(p->dpif); netdev_monitor_poll_wait(p->netdev_monitor); - LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { + LIST_FOR_EACH (ofconn, node, &p->all_conns) { ofconn_wait(ofconn); } if (p->in_band) { + poll_timer_wait_until(p->next_in_band_update); in_band_wait(p->in_band); } - if (p->discovery) { - discovery_wait(p->discovery); - } if (p->fail_open) { fail_open_wait(p->fail_open); } - pinsched_wait(p->miss_sched); - pinsched_wait(p->action_sched); if (p->sflow) { ofproto_sflow_wait(p->sflow); } @@ -967,10 +1216,10 @@ 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]); @@ -992,11 +1241,53 @@ ofproto_get_revalidate_set(struct ofproto *ofproto) bool ofproto_is_alive(const struct ofproto *p) { - return p->discovery || rconn_is_alive(p->controller->rconn); + return !hmap_is_empty(&p->controllers); +} + +/* Deletes port number 'odp_port' from the datapath for 'ofproto'. + * + * This is almost the same as calling dpif_port_del() directly on the + * datapath, but it also makes 'ofproto' close its open netdev for the port + * (if any). This makes it possible to create a new netdev of a different + * type under the same name, which otherwise the netdev library would refuse + * to do because of the conflict. (The netdev would eventually get closed on + * the next trip through ofproto_run(), but this interface is more direct.) + * + * Returns 0 if successful, otherwise a positive errno. */ +int +ofproto_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) { @@ -1011,27 +1302,26 @@ 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 flow *flow, + uint32_t wildcards, unsigned int priority, 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(&rule->cr, flow, wildcards, priority); + cls_rule_from_flow(flow, wildcards, priority, &rule->cr); rule_insert(p, rule, NULL, 0); } void -ofproto_delete_flow(struct ofproto *ofproto, const flow_t *flow, +ofproto_delete_flow(struct ofproto *ofproto, const struct flow *flow, uint32_t wildcards, unsigned int priority) { struct rule *rule; @@ -1078,13 +1368,14 @@ 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); @@ -1100,39 +1391,6 @@ 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) { @@ -1140,13 +1398,11 @@ 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) { @@ -1159,6 +1415,7 @@ 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, @@ -1168,8 +1425,7 @@ 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, @@ -1180,7 +1436,7 @@ 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; @@ -1217,46 +1473,44 @@ 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; + if (!ofconn_receives_async_msgs(ofconn)) { + continue; + } + ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b); ops->reason = reason; ops->desc = ofport->opp; hton_ofp_phy_port(&ops->desc); queue_tx(b, ofconn, NULL); } - if (p->ofhooks->port_changed_cb) { - p->ofhooks->port_changed_cb(reason, &ofport->opp, p->aux); - } } static void ofport_install(struct ofproto *p, struct ofport *ofport) { - uint16_t odp_port = ofp_port_to_odp_port(ofport->opp.port_no); const char *netdev_name = (const char *) ofport->opp.name; 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); } } @@ -1269,6 +1523,20 @@ 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) { @@ -1296,7 +1564,7 @@ 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 " @@ -1336,9 +1604,6 @@ 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 @@ -1364,19 +1629,21 @@ init_ports(struct ofproto *p) } } free(ports); - refresh_port_groups(p); return 0; } static struct ofconn * -ofconn_create(struct ofproto *p, struct rconn *rconn) +ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type) { - struct ofconn *ofconn = xmalloc(sizeof *ofconn); + struct ofconn *ofconn = xzalloc(sizeof *ofconn); + ofconn->ofproto = p; list_push_back(&p->all_conns, &ofconn->node); ofconn->rconn = rconn; + ofconn->type = type; + ofconn->role = NX_ROLE_OTHER; + ofconn->packet_in_counter = rconn_packet_counter_create (); ofconn->pktbuf = NULL; ofconn->miss_send_len = 0; - ofconn->packet_in_counter = rconn_packet_counter_create (); ofconn->reply_counter = rconn_packet_counter_create (); return ofconn; } @@ -1384,7 +1651,13 @@ ofconn_create(struct ofproto *p, struct rconn *rconn) static void ofconn_destroy(struct ofconn *ofconn) { + if (ofconn->type == OFCONN_PRIMARY) { + hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node); + } + discovery_destroy(ofconn->discovery); + list_remove(&ofconn->node); + switch_status_unregister(ofconn->ss); rconn_destroy(ofconn->rconn); rconn_packet_counter_destroy(ofconn->packet_in_counter); rconn_packet_counter_destroy(ofconn->reply_counter); @@ -1396,6 +1669,27 @@ static void ofconn_run(struct ofconn *ofconn, struct ofproto *p) { int iteration; + size_t i; + + if (ofconn->discovery) { + char *controller_name; + if (rconn_is_connectivity_questionable(ofconn->rconn)) { + discovery_question_connectivity(ofconn->discovery); + } + if (discovery_run(ofconn->discovery, &controller_name)) { + if (controller_name) { + char *ofconn_name = ofconn_make_name(p, controller_name); + rconn_connect(ofconn->rconn, controller_name, ofconn_name); + free(ofconn_name); + } else { + rconn_disconnect(ofconn->rconn); + } + } + } + + for (i = 0; i < N_SCHEDULERS; i++) { + pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn); + } rconn_run(ofconn->rconn); @@ -1415,7 +1709,7 @@ ofconn_run(struct ofconn *ofconn, struct ofproto *p) } } - if (ofconn != p->controller && !rconn_is_alive(ofconn->rconn)) { + if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) { ofconn_destroy(ofconn); } } @@ -1423,6 +1717,14 @@ ofconn_run(struct ofconn *ofconn, struct ofproto *p) static void ofconn_wait(struct ofconn *ofconn) { + int i; + + if (ofconn->discovery) { + discovery_wait(ofconn->discovery); + } + for (i = 0; i < N_SCHEDULERS; i++) { + pinsched_wait(ofconn->schedulers[i]); + } rconn_run_wait(ofconn->rconn); if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { rconn_recv_wait(ofconn->rconn); @@ -1430,6 +1732,112 @@ ofconn_wait(struct ofconn *ofconn) COVERAGE_INC(ofproto_ofconn_stuck); } } + +/* Returns true if 'ofconn' should receive asynchronous messages. */ +static bool +ofconn_receives_async_msgs(const struct ofconn *ofconn) +{ + if (ofconn->type == OFCONN_PRIMARY) { + /* Primary controllers always get asynchronous messages unless they + * have configured themselves as "slaves". */ + return ofconn->role != NX_ROLE_SLAVE; + } else { + /* Service connections don't get asynchronous messages unless they have + * explicitly asked for them by setting a nonzero miss send length. */ + return ofconn->miss_send_len > 0; + } +} + +/* Returns a human-readable name for an OpenFlow connection between 'ofproto' + * and 'target', suitable for use in log messages for identifying the + * connection. + * + * The name is dynamically allocated. The caller should free it (with free()) + * when it is no longer needed. */ +static char * +ofconn_make_name(const struct ofproto *ofproto, const char *target) +{ + return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target); +} + +static void +ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst) +{ + int i; + + for (i = 0; i < N_SCHEDULERS; i++) { + struct pinsched **s = &ofconn->schedulers[i]; + + if (rate > 0) { + if (!*s) { + *s = pinsched_create(rate, burst, + ofconn->ofproto->switch_status); + } else { + pinsched_set_limits(*s, rate, burst); + } + } else { + pinsched_destroy(*s); + *s = NULL; + } + } +} + +static void +ofservice_reconfigure(struct ofservice *ofservice, + const struct ofproto_controller *c) +{ + ofservice->probe_interval = c->probe_interval; + ofservice->rate_limit = c->rate_limit; + ofservice->burst_limit = c->burst_limit; +} + +/* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a + * positive errno value. */ +static int +ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c) +{ + struct ofservice *ofservice; + struct pvconn *pvconn; + int error; + + error = pvconn_open(c->target, &pvconn); + if (error) { + return error; + } + + ofservice = xzalloc(sizeof *ofservice); + hmap_insert(&ofproto->services, &ofservice->node, + hash_string(c->target, 0)); + ofservice->pvconn = pvconn; + + ofservice_reconfigure(ofservice, c); + + return 0; +} + +static void +ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice) +{ + hmap_remove(&ofproto->services, &ofservice->node); + pvconn_close(ofservice->pvconn); + free(ofservice); +} + +/* Finds and returns the ofservice within 'ofproto' that has the given + * 'target', or a null pointer if none exists. */ +static struct ofservice * +ofservice_lookup(struct ofproto *ofproto, const char *target) +{ + struct ofservice *ofservice; + + HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0), + &ofproto->services) { + if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) { + return ofservice; + } + } + return NULL; +} /* Caller is responsible for initializing the 'cr' member of the returned * rule. */ @@ -1451,8 +1859,10 @@ 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); @@ -1486,7 +1896,7 @@ 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 { @@ -1506,15 +1916,50 @@ 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, @@ -1526,15 +1971,20 @@ 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 @@ -1545,6 +1995,7 @@ 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; @@ -1555,16 +2006,21 @@ 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) @@ -1579,8 +2035,8 @@ 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; - flow_extract(packet, in_port, &flow); + struct flow flow; + flow_extract(packet, 0, in_port, &flow); rule_execute(p, rule, packet, &flow); } @@ -1601,20 +2057,30 @@ 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(&subrule->cr, flow, 0, - (rule->cr.priority <= UINT16_MAX ? UINT16_MAX - : rule->cr.priority)); + 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); 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) { @@ -1663,9 +2129,10 @@ 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; put->flags = flags; return dpif_flow_put(ofproto->dpif, put); } @@ -1745,12 +2212,20 @@ 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) { @@ -1758,9 +2233,10 @@ 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; if (!dpif_flow_del(p->dpif, &odp_flow)) { update_stats(p, rule, &odp_flow.stats); } @@ -1777,15 +2253,11 @@ 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 @@ -1883,7 +2355,6 @@ 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); @@ -1902,9 +2373,10 @@ 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, &p->ports) { hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp)); } @@ -1947,7 +2419,7 @@ handle_set_config(struct ofproto *p, struct ofconn *ofconn, } flags = ntohs(osc->flags); - if (ofconn == p->controller) { + if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) { switch (flags & OFPC_FRAG_MASK) { case OFPC_FRAG_NORMAL: dpif_set_drop_frags(p->dpif, false); @@ -1962,41 +2434,21 @@ handle_set_config(struct ofproto *p, struct ofconn *ofconn, } } - if ((ntohs(osc->miss_send_len) != 0) != (ofconn->miss_send_len != 0)) { - if (ntohs(osc->miss_send_len) != 0) { - ofconn->pktbuf = pktbuf_create(); - } else { - pktbuf_destroy(ofconn->pktbuf); - } - } - ofconn->miss_send_len = ntohs(osc->miss_send_len); return 0; } static void -add_output_group_action(struct odp_actions *actions, uint16_t group, - uint16_t *nf_output_iface) -{ - odp_actions_add(actions, ODPAT_OUTPUT_GROUP)->output_group.group = group; - - if (group == DP_GROUP_ALL || group == DP_GROUP_FLOOD) { - *nf_output_iface = NF_OUT_FLOOD; - } -} - -static void -add_controller_action(struct odp_actions *actions, - const struct ofp_action_output *oao) +add_controller_action(struct odp_actions *actions, uint16_t max_len) { union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER); - a->controller.arg = oao->max_len ? ntohs(oao->max_len) : UINT32_MAX; + a->controller.arg = max_len; } struct action_xlate_ctx { /* Input. */ - const 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 @@ -2011,13 +2463,17 @@ 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) { @@ -2037,7 +2493,7 @@ 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)); @@ -2058,14 +2514,18 @@ 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; - flow_t flow; - flow = *ctx->flow; - flow.in_port = in_port; + /* Look up a flow with 'in_port' as the input port. Then restore the + * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will + * have surprising behavior). */ + old_in_port = ctx->flow.in_port; + ctx->flow.in_port = in_port; + rule = lookup_valid_rule(ctx->ofproto, &ctx->flow); + ctx->flow.in_port = old_in_port; - rule = lookup_valid_rule(ctx->ofproto, &flow); if (rule) { if (rule->super) { rule = rule->super; @@ -2075,27 +2535,47 @@ 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); + add_output_action(ctx, ctx->flow.in_port); break; case OFPP_TABLE: - xlate_table_action(ctx, ctx->flow->in_port); + xlate_table_action(ctx, ctx->flow.in_port); break; case OFPP_NORMAL: - if (!ctx->ofproto->ofhooks->normal_cb(ctx->flow, ctx->packet, + if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet, ctx->out, ctx->tags, &ctx->nf_output_iface, ctx->ofproto->aux)) { @@ -2104,21 +2584,22 @@ xlate_output_action(struct action_xlate_ctx *ctx, } break; case OFPP_FLOOD: - add_output_group_action(ctx->out, DP_GROUP_FLOOD, - &ctx->nf_output_iface); + flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD, + &ctx->nf_output_iface, ctx->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)); - if (odp_port != ctx->flow->in_port) { + odp_port = ofp_port_to_odp_port(port); + if (odp_port != ctx->flow.in_port) { add_output_action(ctx, odp_port); } break; @@ -2134,11 +2615,93 @@ 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); assert(nah->vendor == htonl(NX_VENDOR_ID)); @@ -2148,23 +2711,47 @@ xlate_nicira_action(struct action_xlate_ctx *ctx, xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port))); break; - default: - VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype); + case NXAST_SET_TUNNEL: + nast = (const struct nx_action_set_tunnel *) nah; + oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL); + ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id; break; - } -} -static void -do_xlate_actions(const union ofp_action *in, size_t n_in, - struct action_xlate_ctx *ctx) -{ + 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 at the same time. */ + + default: + VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype); + break; + } +} + +static void +do_xlate_actions(const union ofp_action *in, size_t n_in, + struct action_xlate_ctx *ctx) +{ struct actions_iterator iter; const union ofp_action *ia; const struct ofport *port; - port = 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; @@ -2180,60 +2767,74 @@ 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); - 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); - 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 = htons(OFP_VLAN_NONE); + ctx->flow.dl_vlan_pcp = 0; break; case OFPAT_SET_DL_SRC: oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC); memcpy(oa->dl_addr.dl_addr, ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); + memcpy(ctx->flow.dl_src, + ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); break; case OFPAT_SET_DL_DST: oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST); memcpy(oa->dl_addr.dl_addr, ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); + memcpy(ctx->flow.dl_dst, + ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); break; case OFPAT_SET_NW_SRC: oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC); - oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; + ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; break; case OFPAT_SET_NW_DST: oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST); - oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; + ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; break; case OFPAT_SET_NW_TOS: oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS); - oa->nw_tos.nw_tos = ia->nw_tos.nw_tos; + ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos; break; case OFPAT_SET_TP_SRC: oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC); - oa->tp_port.tp_port = ia->tp_port.tp_port; + ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port; break; case OFPAT_SET_TP_DST: oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST); - oa->tp_port.tp_port = ia->tp_port.tp_port; + ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port; break; case OFPAT_VENDOR: xlate_nicira_action(ctx, (const struct nx_action_header *) ia); break; + 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; @@ -2243,7 +2844,7 @@ 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) @@ -2252,7 +2853,7 @@ xlate_actions(const union ofp_action *in, size_t n_in, struct action_xlate_ctx ctx; COVERAGE_INC(ofproto_ofp2odp); odp_actions_init(out); - ctx.flow = flow; + ctx.flow = *flow; ctx.recurse = 0; ctx.ofproto = ofproto; ctx.packet = packet; @@ -2261,6 +2862,7 @@ 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. */ @@ -2275,12 +2877,36 @@ 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); } return 0; } +/* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow + * error message code (composed with ofp_mkerr()) for the caller to propagate + * upward. Otherwise, returns 0. + * + * 'oh' is used to make log messages more informative. */ +static int +reject_slave_controller(struct ofconn *ofconn, const struct ofp_header *oh) +{ + 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); + + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); + } else { + return 0; + } +} + static int handle_packet_out(struct ofproto *p, struct ofconn *ofconn, struct ofp_header *oh) @@ -2288,11 +2914,16 @@ handle_packet_out(struct ofproto *p, struct ofconn *ofconn, struct ofp_packet_out *opo; struct ofpbuf payload, *buffer; struct odp_actions actions; + struct flow flow; int n_actions; uint16_t in_port; - flow_t flow; int error; + error = reject_slave_controller(ofconn, oh); + if (error) { + return error; + } + error = check_ofp_packet_out(oh, &payload, &n_actions, p->max_ports); if (error) { return error; @@ -2311,15 +2942,14 @@ handle_packet_out(struct ofproto *p, struct ofconn *ofconn, buffer = NULL; } - flow_extract(&payload, ofp_port_to_odp_port(ntohs(opo->in_port)), &flow); + 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); if (error) { return error; } - dpif_execute(p->dpif, flow.in_port, actions.actions, actions.n_actions, - &payload); + dpif_execute(p->dpif, actions.actions, actions.n_actions, &payload); ofpbuf_delete(buffer); return 0; @@ -2337,37 +2967,38 @@ 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 ofp_header *oh) +handle_port_mod(struct ofproto *p, struct ofconn *ofconn, + struct ofp_header *oh) { const struct ofp_port_mod *opm; struct ofport *port; int error; + error = reject_slave_controller(ofconn, oh); + if (error) { + return error; + } error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm); if (error) { return error; } opm = (struct ofp_port_mod *) oh; - port = port_array_get(&p->ports, - ofp_port_to_odp_port(ntohs(opm->port_no))); + port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no))); if (!port) { return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT); } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) { @@ -2434,17 +3065,6 @@ 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, struct ofp_stats_request *request) @@ -2453,12 +3073,17 @@ handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn, struct ofpbuf *msg; struct odp_stats dpstats; int n_exact, n_subrules, n_wild; + struct rule *rule; msg = start_stats_reply(request, sizeof *ots * 2); /* Count rules of various kinds. */ n_subrules = 0; - classifier_for_each(&p->cls, CLS_INC_EXACT, count_subrules, &n_subrules); + CLASSIFIER_FOR_EACH_EXACT_RULE (rule, cr, &p->cls) { + if (rule->super) { + n_subrules++; + } + } n_exact = classifier_count_exact(&p->cls) - n_subrules; n_wild = classifier_count(&p->cls) - classifier_count_exact(&p->cls); @@ -2480,7 +3105,8 @@ handle_table_stats_request(struct ofproto *p, struct ofconn *ofconn, memset(ots, 0, sizeof *ots); ots->table_id = TABLEID_CLASSIFIER; strcpy(ots->name, "classifier"); - ots->wildcards = htonl(OFPFW_ALL); + ots->wildcards = p->tun_id_from_cookie ? htonl(OVSFW_ALL) + : htonl(OFPFW_ALL); ots->max_entries = htonl(65536); ots->active_count = htonl(n_wild); ots->lookup_count = htonll(0); /* XXX */ @@ -2491,19 +3117,19 @@ 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, - struct ofpbuf *msg) +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, &msg); - ops->port_no = htons(odp_port_to_ofp_port(port_no)); + ops = append_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); @@ -2528,7 +3154,6 @@ 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); @@ -2537,14 +3162,13 @@ 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, hmap_node, &p->ports) { + append_port_stat(port, ofconn, &msg); } } @@ -2588,13 +3212,13 @@ 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. */ @@ -2638,7 +3262,8 @@ flow_stats_cb(struct cls_rule *rule_, void *cbdata_) ofs->length = htons(len); ofs->table_id = rule->cr.wc.wildcards ? TABLEID_CLASSIFIER : TABLEID_HASH; ofs->pad = 0; - flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofs->match); + flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, + cbdata->ofproto->tun_id_from_cookie, &ofs->match); ofs->duration_sec = htonl(sec); ofs->duration_nsec = htonl(msec * 1000000); ofs->cookie = rule->flow_cookie; @@ -2648,7 +3273,9 @@ 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 @@ -2679,7 +3306,7 @@ handle_flow_stats_request(struct ofproto *p, struct ofconn *ofconn, cbdata.ofconn = ofconn; cbdata.out_port = fsr->out_port; cbdata.msg = start_stats_reply(osr, 1024); - cls_rule_from_match(&target, &fsr->match, 0); + cls_rule_from_match(&fsr->match, 0, false, 0, &target); classifier_for_each_match(&p->cls, &target, table_id_to_include(fsr->table_id), flow_stats_cb, &cbdata); @@ -2708,7 +3335,8 @@ 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, &match); + flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, + cbdata->ofproto->tun_id_from_cookie, &match); ds_put_format(results, "duration=%llds, ", (time_msec() - rule->created) / 1000); @@ -2716,11 +3344,15 @@ 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) @@ -2730,12 +3362,12 @@ ofproto_get_all_flows(struct ofproto *p, struct ds *results) struct flow_stats_ds_cbdata cbdata; memset(&match, 0, sizeof match); - match.wildcards = htonl(OFPFW_ALL); + match.wildcards = htonl(OVSFW_ALL); cbdata.ofproto = p; cbdata.results = results; - cls_rule_from_match(&target, &match, 0); + cls_rule_from_match(&match, 0, false, 0, &target); classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, flow_stats_ds_cb, &cbdata); } @@ -2788,7 +3420,7 @@ handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn, cbdata.packet_count = 0; cbdata.byte_count = 0; cbdata.n_flows = 0; - cls_rule_from_match(&target, &asr->match, 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); @@ -2802,6 +3434,95 @@ handle_aggregate_stats_request(struct ofproto *p, struct ofconn *ofconn, return 0; } +struct queue_stats_cbdata { + struct ofconn *ofconn; + struct ofport *ofport; + struct ofpbuf *msg; +}; + +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; + + reply = append_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 void +handle_queue_stats_dump_cb(uint32_t queue_id, + struct netdev_queue_stats *stats, + void *cbdata_) +{ + struct queue_stats_cbdata *cbdata = cbdata_; + + put_queue_stats(cbdata, queue_id, stats); +} + +static void +handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id, + struct queue_stats_cbdata *cbdata) +{ + cbdata->ofport = port; + if (queue_id == OFPQ_ALL) { + netdev_dump_queue_stats(port->netdev, + handle_queue_stats_dump_cb, cbdata); + } else { + struct netdev_queue_stats stats; + + if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) { + put_queue_stats(cbdata, queue_id, &stats); + } + } +} + +static int +handle_queue_stats_request(struct ofproto *ofproto, struct ofconn *ofconn, + const struct ofp_stats_request *osr, + size_t arg_size) +{ + 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_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_stats_request(struct ofproto *p, struct ofconn *ofconn, struct ofp_header *oh) @@ -2833,6 +3554,9 @@ handle_stats_request(struct ofproto *p, struct ofconn *ofconn, case OFPST_PORT: return handle_port_stats_request(p, ofconn, osr, arg_size); + case OFPST_QUEUE: + return handle_queue_stats_request(p, ofconn, osr, arg_size); + case OFPST_VENDOR: return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); @@ -2869,14 +3593,22 @@ 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); } } +/* 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. + * + * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, + * if any. */ static int add_flow(struct ofproto *p, struct ofconn *ofconn, - struct ofp_flow_mod *ofm, size_t n_actions) + const struct ofp_flow_mod *ofm, size_t n_actions) { struct ofpbuf *packet; struct rule *rule; @@ -2884,10 +3616,11 @@ add_flow(struct ofproto *p, struct ofconn *ofconn, int error; if (ofm->flags & htons(OFPFF_CHECK_OVERLAP)) { - flow_t flow; + struct flow flow; uint32_t wildcards; - flow_from_match(&flow, &wildcards, &ofm->match); + 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); @@ -2898,7 +3631,8 @@ add_flow(struct ofproto *p, struct ofconn *ofconn, n_actions, ntohs(ofm->idle_timeout), ntohs(ofm->hard_timeout), ofm->cookie, ofm->flags & htons(OFPFF_SEND_FLOW_REM)); - cls_rule_from_match(&rule->cr, &ofm->match, ntohs(ofm->priority)); + cls_rule_from_match(&ofm->match, ntohs(ofm->priority), + p->tun_id_from_cookie, ofm->cookie, &rule->cr); error = 0; if (ofm->buffer_id != htonl(UINT32_MAX)) { @@ -2910,123 +3644,241 @@ add_flow(struct ofproto *p, struct ofconn *ofconn, } rule_insert(p, rule, packet, in_port); - ofpbuf_delete(packet); return error; } -static int -modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm, - size_t n_actions, uint16_t command, struct rule *rule) +static struct rule * +find_flow_strict(struct ofproto *p, const struct ofp_flow_mod *ofm) { - if (rule_is_hidden(rule)) { - return 0; - } + uint32_t wildcards; + struct flow flow; - if (command == OFPFC_DELETE) { - long long int now = time_msec(); - send_flow_removed(p, rule, now, OFPRR_DELETE); - rule_remove(p, rule); - } else { - size_t actions_len = n_actions * sizeof *rule->actions; + 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))); +} - if (n_actions == rule->n_actions - && !memcmp(ofm->actions, rule->actions, actions_len)) - { - return 0; - } +static int +send_buffered_packet(struct ofproto *ofproto, struct ofconn *ofconn, + struct rule *rule, const struct ofp_flow_mod *ofm) +{ + struct ofpbuf *packet; + uint16_t in_port; + struct flow flow; + int error; - free(rule->actions); - rule->actions = xmemdup(ofm->actions, actions_len); - rule->n_actions = n_actions; - rule->flow_cookie = ofm->cookie; + if (ofm->buffer_id == htonl(UINT32_MAX)) { + return 0; + } - if (rule->cr.wc.wildcards) { - COVERAGE_INC(ofproto_mod_wc_flow); - p->need_revalidate = true; - } else { - rule_update_actions(p, rule); - } + error = pktbuf_retrieve(ofconn->pktbuf, ntohl(ofm->buffer_id), + &packet, &in_port); + if (error) { + return error; } + flow_extract(packet, 0, in_port, &flow); + rule_execute(ofproto, rule, packet, &flow); + return 0; } + +/* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */ + +struct modify_flows_cbdata { + struct ofproto *ofproto; + const struct ofp_flow_mod *ofm; + size_t n_actions; + struct rule *match; +}; + +static int modify_flow(struct ofproto *, const struct ofp_flow_mod *, + size_t n_actions, struct rule *); +static void modify_flows_cb(struct cls_rule *, void *cbdata_); +/* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as + * encoded by ofp_mkerr() on failure. + * + * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, + * if any. */ static int -modify_flows_strict(struct ofproto *p, const struct ofp_flow_mod *ofm, - size_t n_actions, uint16_t command) +modify_flows_loose(struct ofproto *p, struct ofconn *ofconn, + const struct ofp_flow_mod *ofm, size_t n_actions) { - struct rule *rule; - uint32_t wildcards; - flow_t flow; + struct modify_flows_cbdata cbdata; + struct cls_rule target; - flow_from_match(&flow, &wildcards, &ofm->match); - rule = rule_from_cls_rule(classifier_find_rule_exactly( - &p->cls, &flow, wildcards, - ntohs(ofm->priority))); + cbdata.ofproto = p; + cbdata.ofm = ofm; + cbdata.n_actions = n_actions; + cbdata.match = NULL; - if (rule) { - if (command == OFPFC_DELETE - && ofm->out_port != htons(OFPP_NONE) - && !rule_has_out_port(rule, ofm->out_port)) { - return 0; - } + cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie, + &target); - modify_flow(p, ofm, n_actions, command, rule); + classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, + modify_flows_cb, &cbdata); + if (cbdata.match) { + /* This credits the packet to whichever flow happened to happened to + * match last. That's weird. Maybe we should do a lookup for the + * flow that actually matches the packet? Who knows. */ + send_buffered_packet(p, ofconn, cbdata.match, ofm); + return 0; + } else { + return add_flow(p, ofconn, ofm, n_actions); } - return 0; } -struct modify_flows_cbdata { - struct ofproto *ofproto; - const struct ofp_flow_mod *ofm; - uint16_t out_port; - size_t n_actions; - uint16_t command; -}; +/* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error + * code as encoded by ofp_mkerr() on failure. + * + * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, + * if any. */ +static int +modify_flow_strict(struct ofproto *p, struct ofconn *ofconn, + struct ofp_flow_mod *ofm, size_t n_actions) +{ + struct rule *rule = find_flow_strict(p, ofm); + if (rule && !rule_is_hidden(rule)) { + modify_flow(p, ofm, n_actions, rule); + return send_buffered_packet(p, ofconn, rule, ofm); + } else { + return add_flow(p, ofconn, ofm, n_actions); + } +} +/* Callback for modify_flows_loose(). */ static void modify_flows_cb(struct cls_rule *rule_, void *cbdata_) { struct rule *rule = rule_from_cls_rule(rule_); struct modify_flows_cbdata *cbdata = cbdata_; - if (cbdata->out_port != htons(OFPP_NONE) - && !rule_has_out_port(rule, cbdata->out_port)) { - return; + if (!rule_is_hidden(rule)) { + cbdata->match = rule; + modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, rule); } - - modify_flow(cbdata->ofproto, cbdata->ofm, cbdata->n_actions, - cbdata->command, rule); } +/* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has + * been identified as a flow in 'p''s flow table to be modified, by changing + * the rule's actions to match those in 'ofm' (which is followed by 'n_actions' + * ofp_action[] structures). */ static int -modify_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm, - size_t n_actions, uint16_t command) +modify_flow(struct ofproto *p, const struct ofp_flow_mod *ofm, + size_t n_actions, struct rule *rule) { - struct modify_flows_cbdata cbdata; + size_t actions_len = n_actions * sizeof *rule->actions; + + rule->flow_cookie = ofm->cookie; + + /* If the actions are the same, do nothing. */ + if (n_actions == rule->n_actions + && (!n_actions || !memcmp(ofm->actions, rule->actions, actions_len))) + { + return 0; + } + + /* Replace actions. */ + free(rule->actions); + rule->actions = n_actions ? xmemdup(ofm->actions, actions_len) : NULL; + rule->n_actions = n_actions; + + /* Make sure that the datapath gets updated properly. */ + if (rule->cr.wc.wildcards) { + COVERAGE_INC(ofproto_mod_wc_flow); + p->need_revalidate = true; + } else { + rule_update_actions(p, rule); + } + + return 0; +} + +/* OFPFC_DELETE implementation. */ + +struct delete_flows_cbdata { + struct ofproto *ofproto; + uint16_t out_port; +}; + +static void delete_flows_cb(struct cls_rule *, void *cbdata_); +static void delete_flow(struct ofproto *, struct rule *, uint16_t out_port); + +/* Implements OFPFC_DELETE. */ +static void +delete_flows_loose(struct ofproto *p, const struct ofp_flow_mod *ofm) +{ + struct delete_flows_cbdata cbdata; struct cls_rule target; cbdata.ofproto = p; - cbdata.ofm = ofm; - cbdata.out_port = (command == OFPFC_DELETE ? ofm->out_port - : htons(OFPP_NONE)); - cbdata.n_actions = n_actions; - cbdata.command = command; + cbdata.out_port = ofm->out_port; - cls_rule_from_match(&target, &ofm->match, 0); + cls_rule_from_match(&ofm->match, 0, p->tun_id_from_cookie, ofm->cookie, + &target); classifier_for_each_match(&p->cls, &target, CLS_INC_ALL, - modify_flows_cb, &cbdata); - return 0; + delete_flows_cb, &cbdata); +} + +/* Implements OFPFC_DELETE_STRICT. */ +static void +delete_flow_strict(struct ofproto *p, struct ofp_flow_mod *ofm) +{ + struct rule *rule = find_flow_strict(p, ofm); + if (rule) { + delete_flow(p, rule, ofm->out_port); + } +} + +/* Callback for delete_flows_loose(). */ +static void +delete_flows_cb(struct cls_rule *rule_, void *cbdata_) +{ + struct rule *rule = rule_from_cls_rule(rule_); + struct delete_flows_cbdata *cbdata = cbdata_; + + delete_flow(cbdata->ofproto, rule, cbdata->out_port); } +/* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has + * been identified as a flow to delete from 'p''s flow table, by deleting the + * flow and sending out a OFPT_FLOW_REMOVED message to any interested + * controller. + * + * Will not delete 'rule' if it is hidden. Will delete 'rule' only if + * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the + * specified 'out_port'. */ +static void +delete_flow(struct ofproto *p, struct rule *rule, uint16_t out_port) +{ + if (rule_is_hidden(rule)) { + return; + } + + if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) { + return; + } + + send_flow_removed(p, rule, time_msec(), OFPRR_DELETE); + rule_remove(p, rule); +} + static int handle_flow_mod(struct ofproto *p, struct ofconn *ofconn, struct ofp_flow_mod *ofm) { + struct ofp_match orig_match; size_t n_actions; int error; + error = reject_slave_controller(ofconn, &ofm->header); + if (error) { + return error; + } error = check_ofp_message_array(&ofm->header, OFPT_FLOW_MOD, sizeof *ofm, sizeof *ofm->actions, &n_actions); if (error) { @@ -3041,7 +3893,25 @@ handle_flow_mod(struct ofproto *p, struct ofconn *ofconn, return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL); } + /* Normalize ofp->match. If normalization actually changes anything, then + * log the differences. */ + ofm->match.pad1[0] = ofm->match.pad2[0] = 0; + orig_match = ofm->match; normalize_match(&ofm->match); + if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) { + static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1); + if (!VLOG_DROP_INFO(&normal_rl)) { + char *old = ofp_match_to_literal_string(&orig_match); + char *new = ofp_match_to_literal_string(&ofm->match); + VLOG_INFO("%s: normalization changed ofp_match, details:", + rconn_get_name(ofconn->rconn)); + VLOG_INFO(" pre: %s", old); + VLOG_INFO("post: %s", new); + free(old); + free(new); + } + } + if (!ofm->match.wildcards) { ofm->priority = htons(UINT16_MAX); } @@ -3057,16 +3927,18 @@ handle_flow_mod(struct ofproto *p, struct ofconn *ofconn, return add_flow(p, ofconn, ofm, n_actions); case OFPFC_MODIFY: - return modify_flows_loose(p, ofm, n_actions, OFPFC_MODIFY); + return modify_flows_loose(p, ofconn, ofm, n_actions); case OFPFC_MODIFY_STRICT: - return modify_flows_strict(p, ofm, n_actions, OFPFC_MODIFY); + return modify_flow_strict(p, ofconn, ofm, n_actions); case OFPFC_DELETE: - return modify_flows_loose(p, ofm, n_actions, OFPFC_DELETE); + delete_flows_loose(p, ofm); + return 0; case OFPFC_DELETE_STRICT: - return modify_flows_strict(p, ofm, n_actions, OFPFC_DELETE); + delete_flow_strict(p, ofm); + return 0; default: return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND); @@ -3074,26 +3946,104 @@ handle_flow_mod(struct ofproto *p, struct ofconn *ofconn, } static int -handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg) +handle_tun_id_from_cookie(struct ofproto *p, struct nxt_tun_id_cookie *msg) { - struct ofp_vendor_header *ovh = msg; - struct nicira_header *nh; + int error; - if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); - } - if (ovh->vendor != htonl(NX_VENDOR_ID)) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); + error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg); + if (error) { + return error; } - if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) { + + p->tun_id_from_cookie = !!msg->set; + return 0; +} + +static int +handle_role_request(struct ofproto *ofproto, + struct ofconn *ofconn, struct nicira_header *msg) +{ + struct nx_role_request *nrr; + struct nx_role_request *reply; + struct ofpbuf *buf; + uint32_t role; + + if (ntohs(msg->header.length) != sizeof *nrr) { + VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)", + ntohs(msg->header.length), sizeof *nrr); return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); } + nrr = (struct nx_role_request *) msg; - nh = msg; - switch (ntohl(nh->subtype)) { + if (ofconn->type != OFCONN_PRIMARY) { + VLOG_WARN_RL(&rl, "ignoring role request on non-controller " + "connection"); + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); + } + + role = ntohl(nrr->role); + if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER + && role != NX_ROLE_SLAVE) { + VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role); + + /* There's no good error code for this. */ + return ofp_mkerr(OFPET_BAD_REQUEST, -1); + } + + if (role == NX_ROLE_MASTER) { + struct ofconn *other; + + HMAP_FOR_EACH (other, hmap_node, &ofproto->controllers) { + if (other->role == NX_ROLE_MASTER) { + other->role = NX_ROLE_SLAVE; + } + } + } + ofconn->role = role; + + reply = make_openflow_xid(sizeof *reply, OFPT_VENDOR, msg->header.xid, + &buf); + reply->nxh.vendor = htonl(NX_VENDOR_ID); + reply->nxh.subtype = htonl(NXT_ROLE_REPLY); + reply->role = htonl(role); + queue_tx(buf, ofconn, ofconn->reply_counter); + + return 0; +} + +static int +handle_vendor(struct ofproto *p, struct ofconn *ofconn, void *msg) +{ + struct ofp_vendor_header *ovh = msg; + struct nicira_header *nh; + + if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) { + VLOG_WARN_RL(&rl, "received vendor message of length %u " + "(expected at least %zu)", + ntohs(ovh->header.length), sizeof(struct ofp_vendor_header)); + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); + } + if (ovh->vendor != htonl(NX_VENDOR_ID)) { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); + } + if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) { + VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u " + "(expected at least %zu)", + ntohs(ovh->header.length), sizeof(struct nicira_header)); + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); + } + + nh = msg; + switch (ntohl(nh->subtype)) { case NXT_STATUS_REQUEST: return switch_status_handle_request(p->switch_status, ofconn->rconn, msg); + + case NXT_TUN_ID_FROM_COOKIE: + return handle_tun_id_from_cookie(p, msg); + + case NXT_ROLE_REQUEST: + return handle_role_request(p, ofconn, msg); } return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); @@ -3146,7 +4096,7 @@ handle_openflow(struct ofconn *ofconn, struct ofproto *p, break; case OFPT_PORT_MOD: - error = handle_port_mod(p, oh); + error = handle_port_mod(p, ofconn, oh); break; case OFPT_FLOW_MOD: @@ -3184,14 +4134,13 @@ static void handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) { struct odp_msg *msg = packet->data; - uint16_t in_port = odp_port_to_ofp_port(msg->port); struct rule *rule; struct ofpbuf payload; - flow_t flow; + struct flow flow; payload.data = msg + 1; payload.size = msg->length - sizeof *msg; - flow_extract(&payload, msg->port, &flow); + flow_extract(&payload, msg->arg, msg->port, &flow); /* Check with in-band control to see if this packet should be sent * to the local port regardless of the flow table. */ @@ -3201,13 +4150,13 @@ 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); @@ -3220,7 +4169,7 @@ handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) } COVERAGE_INC(ofproto_packet_in); - pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p); + send_packet_in(p, packet); return; } @@ -3238,11 +4187,7 @@ 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 - && rconn_is_connected(p->controller->rconn)) { + if (rule->super && rule->super->cr.priority == FAIL_OPEN_PRIORITY) { /* * Extra-special case for fail-open mode. * @@ -3253,10 +4198,13 @@ handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) * * See the top-level comment in fail-open.c for more information. */ - pinsched_send(p->miss_sched, in_port, packet, send_packet_in_miss, p); - } else { - ofpbuf_delete(packet); + send_packet_in(p, ofpbuf_clone_with_headroom(packet, + DPIF_RECV_MSG_PADDING)); } + + ofpbuf_pull(packet, sizeof *msg); + rule_execute(p, rule, packet, &flow); + rule_reinstall(p, rule); } static void @@ -3267,8 +4215,7 @@ handle_odp_msg(struct ofproto *p, struct ofpbuf *packet) switch (msg->type) { case _ODPL_ACTION_NR: COVERAGE_INC(ofproto_ctlr_action); - pinsched_send(p->action_sched, odp_port_to_ofp_port(msg->port), packet, - send_packet_in_action, p); + send_packet_in(p, packet); break; case _ODPL_SFLOW_NR: @@ -3289,6 +4236,288 @@ 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 rule *rule; + struct flow flow; + + odp_flow_key_to_flow(&f->key, &flow); + + rule = rule_from_cls_rule( + classifier_find_rule_exactly(&p->cls, &flow, 0, UINT16_MAX)); + + 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, + (now >= hard_expire + ? OFPRR_HARD_TIMEOUT : OFPRR_IDLE_TIMEOUT)); + } + rule_remove(cbdata->ofproto, rule); + } +} + static void revalidate_cb(struct cls_rule *sub_, void *cbdata_) { @@ -3305,7 +4534,7 @@ 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) { @@ -3331,7 +4560,8 @@ revalidate_rule(struct ofproto *p, struct rule *rule) } static struct ofpbuf * -compose_flow_removed(const struct rule *rule, long long int now, uint8_t reason) +compose_flow_removed(struct ofproto *p, const struct rule *rule, + long long int now, uint8_t reason) { struct ofp_flow_removed *ofr; struct ofpbuf *buf; @@ -3340,7 +4570,8 @@ compose_flow_removed(const struct rule *rule, long long int now, uint8_t reason) uint32_t msec = tdiff - (sec * 1000); ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf); - flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, &ofr->match); + flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, p->tun_id_from_cookie, + &ofr->match); ofr->cookie = rule->flow_cookie; ofr->priority = htons(rule->cr.priority); ofr->reason = reason; @@ -3353,18 +4584,6 @@ compose_flow_removed(const struct rule *rule, long long int now, uint8_t reason) 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) @@ -3373,6 +4592,10 @@ send_flow_removed(struct ofproto *p, struct rule *rule, struct ofconn *prev; struct ofpbuf *buf = NULL; + if (!rule->send_flow_removed) { + return; + } + /* 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 @@ -3380,12 +4603,13 @@ send_flow_removed(struct ofproto *p, struct rule *rule, * 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)) { + LIST_FOR_EACH (ofconn, node, &p->all_conns) { + if (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(rule, now, reason); + buf = compose_flow_removed(p, rule, now, reason); } prev = ofconn; } @@ -3395,184 +4619,150 @@ send_flow_removed(struct ofproto *p, struct rule *rule, } } - +/* pinsched callback for sending 'packet' on 'ofconn'. */ static void -expire_rule(struct cls_rule *cls_rule, void *p_) +do_send_packet_in(struct ofpbuf *packet, void *ofconn_) { - 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); + struct ofconn *ofconn = ofconn_; - 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); + rconn_send_with_limit(ofconn->rconn, packet, + ofconn->packet_in_counter, 100); } +/* Takes 'packet', which has been converted with do_convert_to_packet_in(), and + * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s + * packet scheduler for sending. + * + * 'max_len' specifies the maximum number of bytes of the packet to send on + * 'ofconn' (INT_MAX specifies no limit). + * + * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise, + * ownership is transferred to this function. */ static void -active_timeout(struct ofproto *ofproto, struct rule *rule) -{ - if (ofproto->netflow && !is_controller_rule(rule) && - netflow_active_timeout_expired(ofproto->netflow, &rule->nf_flow)) { - struct ofexpired expired; - struct odp_flow odp_flow; - - /* Get updated flow stats. */ - memset(&odp_flow, 0, sizeof odp_flow); - if (rule->installed) { - odp_flow.key = rule->cr.flow; - odp_flow.flags = ODPFF_ZERO_TCP_FLAGS; - dpif_flow_get(ofproto->dpif, &odp_flow); - - if (odp_flow.stats.n_packets) { - update_time(ofproto, rule, &odp_flow.stats); - netflow_flow_update_flags(&rule->nf_flow, odp_flow.stats.ip_tos, - odp_flow.stats.tcp_flags); - } - } - - expired.flow = rule->cr.flow; - expired.packet_count = rule->packet_count + - odp_flow.stats.n_packets; - expired.byte_count = rule->byte_count + odp_flow.stats.n_bytes; - expired.used = rule->used; - - netflow_expire(ofproto->netflow, &rule->nf_flow, &expired); - - /* Schedule us to send the accumulated records once we have - * collected all of them. */ - poll_immediate_wake(); +schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len, + bool clone) +{ + struct ofproto *ofproto = ofconn->ofproto; + struct ofp_packet_in *opi = packet->data; + uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port)); + int send_len, trim_size; + uint32_t buffer_id; + + /* Get buffer. */ + if (opi->reason == OFPR_ACTION) { + buffer_id = UINT32_MAX; + } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) { + buffer_id = pktbuf_get_null(); + } else if (!ofconn->pktbuf) { + buffer_id = UINT32_MAX; + } else { + struct ofpbuf payload; + payload.data = opi->data; + payload.size = packet->size - offsetof(struct ofp_packet_in, data); + buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port); } -} -static void -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; + /* Figure out how much of the packet to send. */ + send_len = ntohs(opi->total_len); + if (buffer_id != UINT32_MAX) { + send_len = MIN(send_len, ofconn->miss_send_len); } + send_len = MIN(send_len, max_len); - for (i = 0; i < n_flows; i++) { - struct odp_flow *f = &flows[i]; - struct rule *rule; + /* Adjust packet length and clone if necessary. */ + trim_size = offsetof(struct ofp_packet_in, data) + send_len; + if (clone) { + packet = ofpbuf_clone_data(packet->data, trim_size); + opi = packet->data; + } else { + packet->size = trim_size; + } - rule = rule_from_cls_rule( - classifier_find_rule_exactly(&p->cls, &f->key, 0, UINT16_MAX)); - if (!rule || !rule->installed) { - COVERAGE_INC(ofproto_unexpected_rule); - dpif_flow_del(p->dpif, f); - continue; - } + /* Update packet headers. */ + opi->buffer_id = htonl(buffer_id); + update_openflow_length(packet); - update_time(p, rule, &f->stats); - rule_account(p, rule, f->stats.n_bytes); - } - free(flows); + /* Hand over to packet scheduler. It might immediately call into + * do_send_packet_in() or it might buffer it for a while (until a later + * call to pinsched_run()). */ + pinsched_send(ofconn->schedulers[opi->reason], in_port, + packet, do_send_packet_in, ofconn); } -static void -do_send_packet_in(struct ofconn *ofconn, uint32_t buffer_id, - const struct ofpbuf *packet, int send_len) +/* Replace struct odp_msg header in 'packet' by equivalent struct + * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as + * returned by dpif_recv()). + * + * The conversion is not complete: the caller still needs to trim any unneeded + * payload off the end of the buffer, set the length in the OpenFlow header, + * and set buffer_id. Those require us to know the controller settings and so + * must be done on a per-controller basis. + * + * Returns the maximum number of bytes of the packet that should be sent to + * the controller (INT_MAX if no limit). */ +static int +do_convert_to_packet_in(struct ofpbuf *packet) { struct odp_msg *msg = packet->data; - struct ofpbuf payload; - struct ofpbuf *opi; + struct ofp_packet_in *opi; uint8_t reason; + uint16_t total_len; + uint16_t in_port; + int max_len; - /* Extract packet payload from 'msg'. */ - payload.data = msg + 1; - payload.size = msg->length - sizeof *msg; + /* Extract relevant header fields */ + if (msg->type == _ODPL_ACTION_NR) { + reason = OFPR_ACTION; + max_len = msg->arg; + } else { + reason = OFPR_NO_MATCH; + max_len = INT_MAX; + } + total_len = msg->length - sizeof *msg; + in_port = odp_port_to_ofp_port(msg->port); - /* Construct ofp_packet_in message. */ - reason = msg->type == _ODPL_ACTION_NR ? OFPR_ACTION : OFPR_NO_MATCH; - opi = make_packet_in(buffer_id, odp_port_to_ofp_port(msg->port), reason, - &payload, send_len); + /* Repurpose packet buffer by overwriting header. */ + ofpbuf_pull(packet, sizeof(struct odp_msg)); + opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data)); + opi->header.version = OFP_VERSION; + opi->header.type = OFPT_PACKET_IN; + opi->total_len = htons(total_len); + opi->in_port = htons(in_port); + opi->reason = reason; - /* Send. */ - rconn_send_with_limit(ofconn->rconn, opi, ofconn->packet_in_counter, 100); + return max_len; } +/* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or + * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller + * as necessary according to their individual configurations. + * + * 'packet' must have sufficient headroom to convert it into a struct + * ofp_packet_in (e.g. as returned by dpif_recv()). + * + * Takes ownership of 'packet'. */ static void -send_packet_in_action(struct ofpbuf *packet, void *p_) +send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet) { - struct ofproto *p = p_; - struct ofconn *ofconn; - struct odp_msg *msg; + struct ofconn *ofconn, *prev; + int max_len; - msg = packet->data; - LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { - if (ofconn == p->controller || ofconn->miss_send_len) { - do_send_packet_in(ofconn, UINT32_MAX, packet, msg->arg); - } - } - ofpbuf_delete(packet); -} + max_len = do_convert_to_packet_in(packet); -static void -send_packet_in_miss(struct ofpbuf *packet, void *p_) -{ - struct ofproto *p = p_; - bool in_fail_open = p->fail_open && fail_open_is_active(p->fail_open); - struct ofconn *ofconn; - struct ofpbuf payload; - struct odp_msg *msg; - - msg = packet->data; - payload.data = msg + 1; - payload.size = msg->length - sizeof *msg; - LIST_FOR_EACH (ofconn, struct ofconn, node, &p->all_conns) { - if (ofconn->miss_send_len) { - struct pktbuf *pb = ofconn->pktbuf; - uint32_t buffer_id = (in_fail_open - ? pktbuf_get_null() - : pktbuf_save(pb, &payload, msg->port)); - int send_len = (buffer_id != UINT32_MAX ? ofconn->miss_send_len - : UINT32_MAX); - do_send_packet_in(ofconn, buffer_id, packet, send_len); + prev = NULL; + LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { + if (ofconn_receives_async_msgs(ofconn)) { + if (prev) { + schedule_packet_in(prev, packet, max_len, true); + } + prev = ofconn; } } - ofpbuf_delete(packet); + if (prev) { + schedule_packet_in(prev, packet, max_len, false); + } else { + ofpbuf_delete(packet); + } } static uint64_t @@ -3580,7 +4770,7 @@ 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; @@ -3604,7 +4794,7 @@ 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_) { @@ -3619,7 +4809,8 @@ 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. */ @@ -3631,9 +4822,11 @@ 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; @@ -3645,7 +4838,6 @@ 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