X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=ofproto%2Fofproto.c;h=ffeb952484eb98e6840c4e4c73b54f18c003a77e;hb=9b45d7f5d;hp=897429f84cf11a0667d1e230c72ee3b850c1585c;hpb=b1da6250481f6879d0727710eea87d5f61d6a24c;p=sliver-openvswitch.git diff --git a/ofproto/ofproto.c b/ofproto/ofproto.c index 897429f84..ffeb95248 100644 --- a/ofproto/ofproto.c +++ b/ofproto/ofproto.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2009, 2010 Nicira Networks. + * Copyright (c) 2009, 2010, 2011 Nicira Networks. * Copyright (c) 2010 Jean Tourrilhes - HP-Labs. * * Licensed under the Apache License, Version 2.0 (the "License"); @@ -35,8 +35,10 @@ #include "hmap.h" #include "in-band.h" #include "mac-learning.h" +#include "multipath.h" #include "netdev.h" #include "netflow.h" +#include "netlink.h" #include "nx-match.h" #include "odp-util.h" #include "ofp-print.h" @@ -52,18 +54,47 @@ #include "poll-loop.h" #include "rconn.h" #include "shash.h" -#include "status.h" #include "stream-ssl.h" #include "svec.h" #include "tag.h" #include "timeval.h" +#include "unaligned.h" #include "unixctl.h" #include "vconn.h" #include "vlog.h" VLOG_DEFINE_THIS_MODULE(ofproto); -#include "sflow_api.h" +COVERAGE_DEFINE(facet_changed_rule); +COVERAGE_DEFINE(facet_revalidate); +COVERAGE_DEFINE(odp_overflow); +COVERAGE_DEFINE(ofproto_agg_request); +COVERAGE_DEFINE(ofproto_costly_flags); +COVERAGE_DEFINE(ofproto_ctlr_action); +COVERAGE_DEFINE(ofproto_del_rule); +COVERAGE_DEFINE(ofproto_error); +COVERAGE_DEFINE(ofproto_expiration); +COVERAGE_DEFINE(ofproto_expired); +COVERAGE_DEFINE(ofproto_flows_req); +COVERAGE_DEFINE(ofproto_flush); +COVERAGE_DEFINE(ofproto_invalidated); +COVERAGE_DEFINE(ofproto_no_packet_in); +COVERAGE_DEFINE(ofproto_ofconn_stuck); +COVERAGE_DEFINE(ofproto_ofp2odp); +COVERAGE_DEFINE(ofproto_packet_in); +COVERAGE_DEFINE(ofproto_packet_out); +COVERAGE_DEFINE(ofproto_queue_req); +COVERAGE_DEFINE(ofproto_recv_openflow); +COVERAGE_DEFINE(ofproto_reinit_ports); +COVERAGE_DEFINE(ofproto_unexpected_rule); +COVERAGE_DEFINE(ofproto_uninstallable); +COVERAGE_DEFINE(ofproto_update_port); + +/* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a + * flow translation. */ +#define MAX_RESUBMIT_RECURSION 16 + +struct rule; struct ofport { struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */ @@ -75,11 +106,53 @@ struct ofport { 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 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); +struct action_xlate_ctx { +/* action_xlate_ctx_init() initializes these members. */ + + /* The ofproto. */ + struct ofproto *ofproto; + + /* Flow to which the OpenFlow actions apply. xlate_actions() will modify + * this flow when actions change header fields. */ + struct flow flow; + + /* The packet corresponding to 'flow', or a null pointer if we are + * revalidating without a packet to refer to. */ + const struct ofpbuf *packet; + + /* If nonnull, called just before executing a resubmit action. + * + * This is normally null so the client has to set it manually after + * calling action_xlate_ctx_init(). */ + void (*resubmit_hook)(struct action_xlate_ctx *, struct rule *); + + /* If true, the speciality of 'flow' should be checked before executing + * its actions. If special_cb returns false on 'flow' rendered + * uninstallable and no actions will be executed. */ + bool check_special; + +/* xlate_actions() initializes and uses these members. The client might want + * to look at them after it returns. */ + + struct ofpbuf *odp_actions; /* Datapath actions. */ + tag_type tags; /* Tags associated with OFPP_NORMAL actions. */ + bool may_set_up_flow; /* True ordinarily; false if the actions must + * be reassessed for every packet. */ + uint16_t nf_output_iface; /* Output interface index for NetFlow. */ + +/* xlate_actions() initializes and uses these members, but the client has no + * reason to look at them. */ + + int recurse; /* Recursion level, via xlate_table_action. */ + int last_pop_priority; /* Offset in 'odp_actions' just past most + * recent ODP_ACTION_ATTR_SET_PRIORITY. */ +}; + +static void action_xlate_ctx_init(struct action_xlate_ctx *, + struct ofproto *, const struct flow *, + const struct ofpbuf *); +static struct ofpbuf *xlate_actions(struct action_xlate_ctx *, + const union ofp_action *in, size_t n_in); /* An OpenFlow flow. */ struct rule { @@ -128,6 +201,8 @@ static void rule_insert(struct ofproto *, struct rule *); static void rule_remove(struct ofproto *, struct rule *); static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason); +static void rule_get_stats(const struct rule *, uint64_t *packets, + uint64_t *bytes); /* An exact-match instantiation of an OpenFlow flow. */ struct facet { @@ -140,7 +215,8 @@ struct facet { * * - Do include packets and bytes that were obtained from the datapath * when a flow was deleted (e.g. dpif_flow_del()) or when its - * statistics were reset (e.g. dpif_flow_put() with ODPPF_ZERO_STATS). + * statistics were reset (e.g. dpif_flow_put() with + * DPIF_FP_ZERO_STATS). * * - Do not include any packets or bytes that can currently be obtained * from the datapath by, e.g., dpif_flow_get(). @@ -148,6 +224,13 @@ struct facet { uint64_t packet_count; /* Number of packets received. */ uint64_t byte_count; /* Number of bytes received. */ + uint64_t dp_packet_count; /* Last known packet count in the datapath. */ + uint64_t dp_byte_count; /* Last known byte count in the datapath. */ + + uint64_t rs_packet_count; /* Packets pushed to resubmit children. */ + uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */ + long long int rs_used; /* Used time pushed to resubmit children. */ + /* Number of bytes passed to account_cb. This may include bytes that can * currently obtained from the datapath (thus, it can be greater than * byte_count). */ @@ -160,8 +243,8 @@ struct facet { bool installed; /* Installed in datapath? */ bool may_install; /* True ordinarily; false if actions must * be reassessed for every packet. */ - int n_actions; /* Number of elements in actions[]. */ - union odp_action *actions; /* Datapath actions. */ + size_t actions_len; /* Number of bytes in actions[]. */ + struct nlattr *actions; /* Datapath actions. */ tag_type tags; /* Tags (set only by hooks). */ struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ }; @@ -182,7 +265,8 @@ static void facet_flush_stats(struct ofproto *, struct facet *); static void facet_make_actions(struct ofproto *, struct facet *, const struct ofpbuf *packet); static void facet_update_stats(struct ofproto *, struct facet *, - const struct odp_flow_stats *); + const struct dpif_flow_stats *); +static void facet_push_stats(struct ofproto *, struct facet *); /* ofproto supports two kinds of OpenFlow connections: * @@ -230,11 +314,12 @@ struct ofconn { struct list node; /* In struct ofproto's "all_conns" list. */ struct rconn *rconn; /* OpenFlow connection. */ enum ofconn_type type; /* Type. */ - int flow_format; /* One of NXFF_*. */ + enum nx_flow_format flow_format; /* Currently selected flow format. */ /* OFPT_PACKET_IN related data. */ struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */ - struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */ +#define N_SCHEDULERS 2 + struct pinsched *schedulers[N_SCHEDULERS]; struct pktbuf *pktbuf; /* OpenFlow packet buffers. */ int miss_send_len; /* Bytes to send of buffered packets. */ @@ -248,19 +333,9 @@ struct ofconn { 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? */ }; -/* 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); @@ -274,8 +349,9 @@ static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst); static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, struct rconn_packet_counter *counter); -static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg); -static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn); +static void send_packet_in(struct ofproto *, struct dpif_upcall *, + const struct flow *, bool clone); +static void do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn); struct ofproto { /* Settings. */ @@ -295,7 +371,6 @@ struct ofproto { uint32_t max_ports; /* Configuration. */ - struct switch_status *switch_status; struct fail_open *fail_open; struct netflow *netflow; struct ofproto_sflow *sflow; @@ -334,6 +409,9 @@ struct ofproto { struct mac_learning *ml; }; +/* Map from dpif name to struct ofproto, for use by unixctl commands. */ +static struct shash all_ofprotos = SHASH_INITIALIZER(&all_ofprotos); + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); static const struct ofhooks default_ofhooks; @@ -342,8 +420,11 @@ static uint64_t pick_datapath_id(const struct ofproto *); static uint64_t pick_fallback_dpid(void); static int ofproto_expire(struct ofproto *); +static void flow_push_stats(struct ofproto *, const struct rule *, + struct flow *, uint64_t packets, uint64_t bytes, + long long int used); -static void handle_odp_msg(struct ofproto *, struct ofpbuf *); +static void handle_upcall(struct ofproto *, struct dpif_upcall *); static void handle_openflow(struct ofconn *, struct ofpbuf *); @@ -352,32 +433,31 @@ static void update_port(struct ofproto *, const char *devname); static int init_ports(struct ofproto *); static void reinit_ports(struct ofproto *); +static void ofproto_unixctl_init(void); + int ofproto_create(const char *datapath, const char *datapath_type, const struct ofhooks *ofhooks, void *aux, struct ofproto **ofprotop) { - struct odp_stats stats; struct ofproto *p; struct dpif *dpif; int error; *ofprotop = NULL; + ofproto_unixctl_init(); + /* Connect to datapath and start listening for messages. */ error = dpif_open(datapath, datapath_type, &dpif); if (error) { VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error)); return error; } - error = dpif_get_dp_stats(dpif, &stats); - if (error) { - VLOG_ERR("failed to obtain stats for datapath %s: %s", - datapath, strerror(error)); - dpif_close(dpif); - return error; - } - error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW); + error = dpif_recv_set_mask(dpif, + ((1u << DPIF_UC_MISS) | + (1u << DPIF_UC_ACTION) | + (1u << DPIF_UC_SAMPLE))); if (error) { VLOG_ERR("failed to listen on datapath %s: %s", datapath, strerror(error)); @@ -402,10 +482,9 @@ ofproto_create(const char *datapath, const char *datapath_type, p->netdev_monitor = netdev_monitor_create(); hmap_init(&p->ports); shash_init(&p->port_by_name); - p->max_ports = stats.max_ports; + p->max_ports = dpif_get_max_ports(dpif); /* Initialize submodules. */ - p->switch_status = switch_status_create(p); p->fail_open = NULL; p->netflow = NULL; p->sflow = NULL; @@ -445,6 +524,8 @@ ofproto_create(const char *datapath, const char *datapath_type, p->datapath_id = pick_datapath_id(p); VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id); + shash_add_once(&all_ofprotos, dpif_name(p->dpif), p); + *ofprotop = p; return 0; } @@ -486,8 +567,7 @@ add_controller(struct ofproto *ofproto, const struct ofproto_controller *c) if (is_discovery_controller(c)) { int error = discovery_create(c->accept_re, c->update_resolv_conf, - ofproto->dpif, ofproto->switch_status, - &discovery); + ofproto->dpif, &discovery); if (error) { return; } @@ -598,8 +678,7 @@ update_in_band_remotes(struct ofproto *ofproto) * 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); + in_band_create(ofproto, ofproto->dpif, &ofproto->in_band); } if (ofproto->in_band) { in_band_set_remotes(ofproto->in_band, addrs, n_addrs); @@ -626,7 +705,7 @@ update_fail_open(struct ofproto *p) size_t n; if (!p->fail_open) { - p->fail_open = fail_open_create(p, p->switch_status); + p->fail_open = fail_open_create(p); } n = 0; @@ -651,7 +730,6 @@ ofproto_set_controllers(struct ofproto *p, 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. @@ -679,7 +757,6 @@ ofproto_set_controllers(struct ofproto *p, /* 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; @@ -688,9 +765,6 @@ ofproto_set_controllers(struct ofproto *p, ofconn_destroy(ofconn); } else { update_controller(ofconn, c); - if (ofconn->ss) { - ss_exists = true; - } } } @@ -712,13 +786,6 @@ ofproto_set_controllers(struct ofproto *p, 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 @@ -966,6 +1033,8 @@ ofproto_destroy(struct ofproto *p) return; } + shash_find_and_delete(&all_ofprotos, dpif_name(p->dpif)); + /* Destroy fail-open and in-band early, since they touch the classifier. */ fail_open_destroy(p->fail_open); p->fail_open = NULL; @@ -991,7 +1060,6 @@ ofproto_destroy(struct ofproto *p) } shash_destroy(&p->port_by_name); - switch_status_destroy(p->switch_status); netflow_destroy(p->netflow); ofproto_sflow_destroy(p->sflow); @@ -1096,9 +1164,9 @@ ofproto_run1(struct ofproto *p) } for (i = 0; i < 50; i++) { - struct ofpbuf *buf; + struct dpif_upcall packet; - error = dpif_recv(p->dpif, &buf); + error = dpif_recv(p->dpif, &packet); if (error) { if (error == ENODEV) { /* Someone destroyed the datapath behind our back. The caller @@ -1112,7 +1180,7 @@ ofproto_run1(struct ofproto *p) break; } - handle_odp_msg(p, buf); + handle_upcall(p, &packet); } while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) { @@ -1277,6 +1345,68 @@ ofproto_is_alive(const struct ofproto *p) return !hmap_is_empty(&p->controllers); } +void +ofproto_get_ofproto_controller_info(const struct ofproto *ofproto, + struct shash *info) +{ + const struct ofconn *ofconn; + + shash_init(info); + + HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) { + const struct rconn *rconn = ofconn->rconn; + time_t now = time_now(); + time_t last_connection = rconn_get_last_connection(rconn); + time_t last_disconnect = rconn_get_last_disconnect(rconn); + const int last_error = rconn_get_last_error(rconn); + struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo); + + shash_add(info, rconn_get_target(rconn), cinfo); + + cinfo->is_connected = rconn_is_connected(rconn); + cinfo->role = ofconn->role; + + cinfo->pairs.n = 0; + + if (last_error) { + cinfo->pairs.keys[cinfo->pairs.n] = "last_error"; + cinfo->pairs.values[cinfo->pairs.n++] = + xstrdup(ovs_retval_to_string(last_error)); + } + + cinfo->pairs.keys[cinfo->pairs.n] = "state"; + cinfo->pairs.values[cinfo->pairs.n++] = + xstrdup(rconn_get_state(rconn)); + + if (last_connection != TIME_MIN) { + cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_connect"; + cinfo->pairs.values[cinfo->pairs.n++] + = xasprintf("%ld", (long int) (now - last_connection)); + } + + if (last_disconnect != TIME_MIN) { + cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_disconnect"; + cinfo->pairs.values[cinfo->pairs.n++] + = xasprintf("%ld", (long int) (now - last_disconnect)); + } + } +} + +void +ofproto_free_ofproto_controller_info(struct shash *info) +{ + struct shash_node *node; + + SHASH_FOR_EACH (node, info) { + struct ofproto_controller_info *cinfo = node->data; + while (cinfo->pairs.n) { + free((char *) cinfo->pairs.values[--cinfo->pairs.n]); + } + free(cinfo); + } + shash_destroy(info); +} + /* Deletes port number 'odp_port' from the datapath for 'ofproto'. * * This is almost the same as calling dpif_port_del() directly on the @@ -1291,7 +1421,7 @@ 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 : ""; + const char *name = ofport ? ofport->opp.name : ""; int error; error = dpif_port_del(ofproto->dpif, odp_port); @@ -1324,18 +1454,20 @@ ofproto_send_packet(struct ofproto *p, const struct flow *flow, const union ofp_action *actions, size_t n_actions, const struct ofpbuf *packet) { - struct odp_actions odp_actions; - int error; + struct action_xlate_ctx ctx; + struct ofpbuf *odp_actions; - error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions, - NULL, NULL, NULL); - if (error) { - return error; - } + action_xlate_ctx_init(&ctx, p, flow, packet); + /* Always xlate packets originated in this function. */ + ctx.check_special = false; + odp_actions = xlate_actions(&ctx, actions, n_actions); /* XXX Should we translate the dpif_execute() errno value into an OpenFlow * error code? */ - dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions, packet); + dpif_execute(p->dpif, odp_actions->data, odp_actions->size, packet); + + ofpbuf_delete(odp_actions); + return 0; } @@ -1369,19 +1501,12 @@ ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target) } } -static void -destroy_rule(struct cls_rule *rule_, void *ofproto_) -{ - struct rule *rule = rule_from_cls_rule(rule_); - struct ofproto *ofproto = ofproto_; - - rule_remove(ofproto, rule); -} - void ofproto_flush_flows(struct ofproto *ofproto) { struct facet *facet, *next_facet; + struct rule *rule, *next_rule; + struct cls_cursor cursor; COVERAGE_INC(ofproto_flush); @@ -1391,9 +1516,16 @@ ofproto_flush_flows(struct ofproto *ofproto) * individually since we are about to blow away all the facets with * dpif_flow_flush(). */ facet->installed = false; + facet->dp_packet_count = 0; + facet->dp_byte_count = 0; facet_remove(ofproto, facet); } - classifier_for_each(&ofproto->cls, destroy_rule, ofproto); + + cls_cursor_init(&cursor, &ofproto->cls, NULL); + CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { + rule_remove(ofproto, rule); + } + dpif_flow_flush(ofproto->dpif); if (ofproto->in_band) { in_band_flushed(ofproto->in_band); @@ -1406,33 +1538,30 @@ ofproto_flush_flows(struct ofproto *ofproto) static void reinit_ports(struct ofproto *p) { - struct svec devnames; + struct dpif_port_dump dump; + struct shash_node *node; + struct shash devnames; struct ofport *ofport; - struct odp_port *odp_ports; - size_t n_odp_ports; - size_t i; + struct dpif_port dpif_port; COVERAGE_INC(ofproto_reinit_ports); - svec_init(&devnames); + shash_init(&devnames); HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { - svec_add (&devnames, (char *) ofport->opp.name); + shash_add_once (&devnames, ofport->opp.name, NULL); } - dpif_port_list(p->dpif, &odp_ports, &n_odp_ports); - for (i = 0; i < n_odp_ports; i++) { - svec_add (&devnames, odp_ports[i].devname); + DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) { + shash_add_once (&devnames, dpif_port.name, NULL); } - free(odp_ports); - svec_sort_unique(&devnames); - for (i = 0; i < devnames.n; i++) { - update_port(p, devnames.names[i]); + SHASH_FOR_EACH (node, &devnames) { + update_port(p, node->name); } - svec_destroy(&devnames); + shash_destroy(&devnames); } static struct ofport * -make_ofport(const struct odp_port *odp_port) +make_ofport(const struct dpif_port *dpif_port) { struct netdev_options netdev_options; enum netdev_flags flags; @@ -1441,26 +1570,25 @@ make_ofport(const struct odp_port *odp_port) int error; memset(&netdev_options, 0, sizeof netdev_options); - netdev_options.name = odp_port->devname; + netdev_options.name = dpif_port->name; + netdev_options.type = dpif_port->type; netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; error = netdev_open(&netdev_options, &netdev); if (error) { VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s " "cannot be opened (%s)", - odp_port->devname, odp_port->port, - odp_port->devname, strerror(error)); + dpif_port->name, dpif_port->port_no, + dpif_port->name, strerror(error)); return NULL; } - ofport = xmalloc(sizeof *ofport); + ofport = xzalloc(sizeof *ofport); ofport->netdev = netdev; - ofport->odp_port = odp_port->port; - ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port); + ofport->odp_port = dpif_port->port_no; + ofport->opp.port_no = odp_port_to_ofp_port(dpif_port->port_no); netdev_get_etheraddr(netdev, ofport->opp.hw_addr); - memcpy(ofport->opp.name, odp_port->devname, - MIN(sizeof ofport->opp.name, sizeof odp_port->devname)); - ofport->opp.name[sizeof ofport->opp.name - 1] = '\0'; + ovs_strlcpy(ofport->opp.name, dpif_port->name, sizeof ofport->opp.name); netdev_get_flags(netdev, &flags); ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN; @@ -1474,15 +1602,15 @@ make_ofport(const struct odp_port *odp_port) } static bool -ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port) +ofport_conflicts(const struct ofproto *p, const struct dpif_port *dpif_port) { - if (get_port(p, odp_port->port)) { + if (get_port(p, dpif_port->port_no)) { VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath", - odp_port->port); + dpif_port->port_no); return true; - } else if (shash_find(&p->port_by_name, odp_port->devname)) { + } else if (shash_find(&p->port_by_name, dpif_port->name)) { VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath", - odp_port->devname); + dpif_port->name); return true; } else { return false; @@ -1498,7 +1626,7 @@ ofport_equal(const struct ofport *a_, const struct ofport *b_) BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */ return (a->port_no == b->port_no && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr) - && !strcmp((char *) a->name, (char *) b->name) + && !strcmp(a->name, b->name) && a->state == b->state && a->config == b->config && a->curr == b->curr @@ -1517,7 +1645,10 @@ send_port_status(struct ofproto *p, const struct ofport *ofport, struct ofp_port_status *ops; struct ofpbuf *b; - if (!ofconn_receives_async_msgs(ofconn)) { + /* Primary controllers, even slaves, should always get port status + updates. Otherwise obey ofconn_receives_async_msgs(). */ + if (ofconn->type != OFCONN_PRIMARY + && !ofconn_receives_async_msgs(ofconn)) { continue; } @@ -1532,7 +1663,7 @@ send_port_status(struct ofproto *p, const struct ofport *ofport, static void ofport_install(struct ofproto *p, struct ofport *ofport) { - const char *netdev_name = (const char *) ofport->opp.name; + const char *netdev_name = ofport->opp.name; netdev_monitor_add(p->netdev_monitor, ofport->netdev); hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0)); @@ -1548,7 +1679,7 @@ ofport_remove(struct ofproto *p, struct ofport *ofport) netdev_monitor_remove(p->netdev_monitor, ofport->netdev); hmap_remove(&p->ports, &ofport->hmap_node); shash_delete(&p->port_by_name, - shash_find(&p->port_by_name, (char *) ofport->opp.name)); + shash_find(&p->port_by_name, ofport->opp.name)); if (p->sflow) { ofproto_sflow_del_port(p->sflow, ofport->odp_port); } @@ -1580,7 +1711,7 @@ get_port(const struct ofproto *ofproto, uint16_t odp_port) static void update_port(struct ofproto *p, const char *devname) { - struct odp_port odp_port; + struct dpif_port dpif_port; struct ofport *old_ofport; struct ofport *new_ofport; int error; @@ -1588,7 +1719,7 @@ update_port(struct ofproto *p, const char *devname) COVERAGE_INC(ofproto_update_port); /* Query the datapath for port information. */ - error = dpif_port_query_by_name(p->dpif, devname, &odp_port); + error = dpif_port_query_by_name(p->dpif, devname, &dpif_port); /* Find the old ofport. */ old_ofport = shash_find_data(&p->port_by_name, devname); @@ -1604,31 +1735,31 @@ update_port(struct ofproto *p, const char *devname) * reliably but more portably by comparing the old port's MAC * against the new port's MAC. However, this code isn't that smart * and always sends an OFPPR_MODIFY (XXX). */ - old_ofport = get_port(p, odp_port.port); + old_ofport = get_port(p, dpif_port.port_no); } } else if (error != ENOENT && error != ENODEV) { VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error " "%s", strerror(error)); - return; + goto exit; } /* Create a new ofport. */ - new_ofport = !error ? make_ofport(&odp_port) : NULL; + new_ofport = !error ? make_ofport(&dpif_port) : NULL; /* Eliminate a few pathological cases. */ if (!old_ofport && !new_ofport) { - return; + goto exit; } else if (old_ofport && new_ofport) { /* Most of the 'config' bits are OpenFlow soft state, but - * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow - * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and - * leaves the other bits 0.) */ + * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the + * OpenFlow bits from old_ofport. (make_ofport() only sets + * OFPPC_PORT_DOWN and leaves the other bits 0.) */ new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN; if (ofport_equal(old_ofport, new_ofport)) { /* False alarm--no change. */ ofport_free(new_ofport); - return; + goto exit; } } @@ -1644,31 +1775,26 @@ update_port(struct ofproto *p, const char *devname) : !new_ofport ? OFPPR_DELETE : OFPPR_MODIFY)); ofport_free(old_ofport); + +exit: + dpif_port_destroy(&dpif_port); } static int init_ports(struct ofproto *p) { - struct odp_port *ports; - size_t n_ports; - size_t i; - int error; + struct dpif_port_dump dump; + struct dpif_port dpif_port; - error = dpif_port_list(p->dpif, &ports, &n_ports); - if (error) { - return error; - } - - for (i = 0; i < n_ports; i++) { - const struct odp_port *odp_port = &ports[i]; - if (!ofport_conflicts(p, odp_port)) { - struct ofport *ofport = make_ofport(odp_port); + DPIF_PORT_FOR_EACH (&dpif_port, &dump, p->dpif) { + if (!ofport_conflicts(p, &dpif_port)) { + struct ofport *ofport = make_ofport(&dpif_port); if (ofport) { ofport_install(p, ofport); } } } - free(ports); + return 0; } @@ -1698,7 +1824,6 @@ ofconn_destroy(struct ofconn *ofconn) 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); @@ -1723,6 +1848,7 @@ ofconn_run(struct ofconn *ofconn) char *ofconn_name = ofconn_make_name(p, controller_name); rconn_connect(ofconn->rconn, controller_name, ofconn_name); free(ofconn_name); + free(controller_name); } else { rconn_disconnect(ofconn->rconn); } @@ -1812,8 +1938,7 @@ ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst) if (rate > 0) { if (!*s) { - *s = pinsched_create(rate, burst, - ofconn->ofproto->switch_status); + *s = pinsched_create(rate, burst); } else { pinsched_set_limits(*s, rate, burst); } @@ -1971,30 +2096,33 @@ rule_has_out_port(const struct rule *rule, ovs_be16 out_port) * * Takes ownership of 'packet'. */ static bool -execute_odp_actions(struct ofproto *ofproto, uint16_t in_port, - const union odp_action *actions, size_t n_actions, +execute_odp_actions(struct ofproto *ofproto, const struct flow *flow, + const struct nlattr *odp_actions, size_t actions_len, struct ofpbuf *packet) { - if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) { + if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t)) + && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) { /* As an optimization, avoid a round-trip from userspace to kernel to * userspace. This also avoids possibly filling up kernel packet * buffers along the way. */ - struct odp_msg *msg; + struct dpif_upcall upcall; - msg = ofpbuf_push_uninit(packet, sizeof *msg); - msg->type = _ODPL_ACTION_NR; - msg->length = sizeof(struct odp_msg) + packet->size; - msg->port = in_port; - msg->reserved = 0; - msg->arg = actions[0].controller.arg; + upcall.type = DPIF_UC_ACTION; + upcall.packet = packet; + upcall.key = NULL; + upcall.key_len = 0; + upcall.userdata = nl_attr_get_u64(odp_actions); + upcall.sample_pool = 0; + upcall.actions = NULL; + upcall.actions_len = 0; - send_packet_in(ofproto, packet); + send_packet_in(ofproto, &upcall, flow, false); return true; } else { int error; - error = dpif_execute(ofproto->dpif, actions, n_actions, packet); + error = dpif_execute(ofproto->dpif, odp_actions, actions_len, packet); ofpbuf_delete(packet); return !error; } @@ -2016,17 +2144,15 @@ static void facet_execute(struct ofproto *ofproto, struct facet *facet, struct ofpbuf *packet) { - struct odp_flow_stats stats; + struct dpif_flow_stats stats; assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); flow_extract_stats(&facet->flow, packet, &stats); - if (execute_odp_actions(ofproto, facet->flow.in_port, - facet->actions, facet->n_actions, packet)) { + stats.used = time_msec(); + if (execute_odp_actions(ofproto, &facet->flow, + facet->actions, facet->actions_len, packet)) { facet_update_stats(ofproto, facet, &stats); - facet->used = time_msec(); - netflow_flow_update_time(ofproto->netflow, - &facet->nf_flow, facet->used); } } @@ -2042,8 +2168,9 @@ static void rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port, struct ofpbuf *packet) { + struct action_xlate_ctx ctx; + struct ofpbuf *odp_actions; struct facet *facet; - struct odp_actions a; struct flow flow; size_t size; @@ -2069,18 +2196,17 @@ rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port, /* We can't account anything to a facet. If we were to try, then that * facet would have a non-matching rule, busting our invariants. */ - if (xlate_actions(rule->actions, rule->n_actions, &flow, ofproto, - packet, &a, NULL, 0, NULL)) { - ofpbuf_delete(packet); - return; - } + action_xlate_ctx_init(&ctx, ofproto, &flow, packet); + odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions); size = packet->size; - if (execute_odp_actions(ofproto, in_port, - a.actions, a.n_actions, packet)) { + if (execute_odp_actions(ofproto, &flow, odp_actions->data, + odp_actions->size, packet)) { rule->used = time_msec(); rule->packet_count++; rule->byte_count += size; + flow_push_stats(ofproto, rule, &flow, 1, size, rule->used); } + ofpbuf_delete(odp_actions); } /* Inserts 'rule' into 'p''s flow table. */ @@ -2168,33 +2294,47 @@ facet_make_actions(struct ofproto *p, struct facet *facet, const struct ofpbuf *packet) { const struct rule *rule = facet->rule; - struct odp_actions a; - size_t actions_len; + struct ofpbuf *odp_actions; + struct action_xlate_ctx ctx; - xlate_actions(rule->actions, rule->n_actions, &facet->flow, p, - packet, &a, &facet->tags, &facet->may_install, - &facet->nf_flow.output_iface); + action_xlate_ctx_init(&ctx, p, &facet->flow, packet); + odp_actions = xlate_actions(&ctx, rule->actions, rule->n_actions); + facet->tags = ctx.tags; + facet->may_install = ctx.may_set_up_flow; + facet->nf_flow.output_iface = ctx.nf_output_iface; - actions_len = a.n_actions * sizeof *a.actions; - if (facet->n_actions != a.n_actions - || memcmp(facet->actions, a.actions, actions_len)) { + if (facet->actions_len != odp_actions->size + || memcmp(facet->actions, odp_actions->data, odp_actions->size)) { free(facet->actions); - facet->n_actions = a.n_actions; - facet->actions = xmemdup(a.actions, actions_len); + facet->actions_len = odp_actions->size; + facet->actions = xmemdup(odp_actions->data, odp_actions->size); } + + ofpbuf_delete(odp_actions); } static int -facet_put__(struct ofproto *ofproto, struct facet *facet, int flags, - struct odp_flow_put *put) +facet_put__(struct ofproto *ofproto, struct facet *facet, + const struct nlattr *actions, size_t actions_len, + struct dpif_flow_stats *stats) { - memset(&put->flow.stats, 0, sizeof put->flow.stats); - odp_flow_key_from_flow(&put->flow.key, &facet->flow); - put->flow.actions = facet->actions; - put->flow.n_actions = facet->n_actions; - put->flow.flags = 0; - put->flags = flags; - return dpif_flow_put(ofproto->dpif, put); + uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S]; + enum dpif_flow_put_flags flags; + struct ofpbuf key; + + flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; + if (stats) { + flags |= DPIF_FP_ZERO_STATS; + facet->dp_packet_count = 0; + facet->dp_byte_count = 0; + } + + ofpbuf_use_stack(&key, keybuf, sizeof keybuf); + odp_flow_key_from_flow(&key, &facet->flow); + assert(key.base == keybuf); + + return dpif_flow_put(ofproto->dpif, flags, key.data, key.size, + actions, actions_len, stats); } /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If @@ -2203,17 +2343,12 @@ facet_put__(struct ofproto *ofproto, struct facet *facet, int flags, static void facet_install(struct ofproto *p, struct facet *facet, bool zero_stats) { - if (facet->may_install) { - struct odp_flow_put put; - int flags; + struct dpif_flow_stats stats; - flags = ODPPF_CREATE | ODPPF_MODIFY; - if (zero_stats) { - flags |= ODPPF_ZERO_STATS; - } - if (!facet_put__(p, facet, flags, &put)) { - facet->installed = true; - } + if (facet->may_install + && !facet_put__(p, facet, facet->actions, facet->actions_len, + zero_stats ? &stats : NULL)) { + facet->installed = true; } } @@ -2229,7 +2364,7 @@ facet_account(struct ofproto *ofproto, && total_bytes > facet->accounted_bytes) { ofproto->ofhooks->account_flow_cb( - &facet->flow, facet->tags, facet->actions, facet->n_actions, + &facet->flow, facet->tags, facet->actions, facet->actions_len, total_bytes - facet->accounted_bytes, ofproto->aux); facet->accounted_bytes = total_bytes; } @@ -2240,16 +2375,23 @@ static void facet_uninstall(struct ofproto *p, struct facet *facet) { if (facet->installed) { - struct odp_flow odp_flow; - - odp_flow_key_from_flow(&odp_flow.key, &facet->flow); - odp_flow.actions = NULL; - odp_flow.n_actions = 0; - odp_flow.flags = 0; - if (!dpif_flow_del(p->dpif, &odp_flow)) { - facet_update_stats(p, facet, &odp_flow.stats); + uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S]; + struct dpif_flow_stats stats; + struct ofpbuf key; + + ofpbuf_use_stack(&key, keybuf, sizeof keybuf); + odp_flow_key_from_flow(&key, &facet->flow); + assert(key.base == keybuf); + + if (!dpif_flow_del(p->dpif, key.data, key.size, &stats)) { + facet_update_stats(p, facet, &stats); } facet->installed = false; + facet->dp_packet_count = 0; + facet->dp_byte_count = 0; + } else { + assert(facet->dp_packet_count == 0); + assert(facet->dp_byte_count == 0); } } @@ -2266,10 +2408,16 @@ facet_is_controller_flow(struct facet *facet) } /* Folds all of 'facet''s statistics into its rule. Also updates the - * accounting ofhook and emits a NetFlow expiration if appropriate. */ + * accounting ofhook and emits a NetFlow expiration if appropriate. All of + * 'facet''s statistics in the datapath should have been zeroed and folded into + * its packet and byte counts before this function is called. */ static void facet_flush_stats(struct ofproto *ofproto, struct facet *facet) { + assert(!facet->dp_byte_count); + assert(!facet->dp_packet_count); + + facet_push_stats(ofproto, facet); facet_account(ofproto, facet, 0); if (ofproto->netflow && !facet_is_controller_flow(facet)) { @@ -2288,6 +2436,8 @@ facet_flush_stats(struct ofproto *ofproto, struct facet *facet) * reinstalled. */ facet->packet_count = 0; facet->byte_count = 0; + facet->rs_packet_count = 0; + facet->rs_byte_count = 0; facet->accounted_bytes = 0; netflow_flow_clear(&facet->nf_flow); @@ -2348,10 +2498,9 @@ facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow) static bool facet_revalidate(struct ofproto *ofproto, struct facet *facet) { + struct action_xlate_ctx ctx; + struct ofpbuf *odp_actions; struct rule *new_rule; - struct odp_actions a; - size_t actions_len; - uint16_t new_nf_output_iface; bool actions_changed; COVERAGE_INC(facet_revalidate); @@ -2366,31 +2515,24 @@ facet_revalidate(struct ofproto *ofproto, struct facet *facet) /* Calculate new ODP actions. * - * We are very cautious about actually modifying 'facet' state at this - * point, because we might need to, e.g., emit a NetFlow expiration and, if - * so, we need to have the old state around to properly compose it. */ - xlate_actions(new_rule->actions, new_rule->n_actions, &facet->flow, - ofproto, NULL, &a, &facet->tags, &facet->may_install, - &new_nf_output_iface); - actions_len = a.n_actions * sizeof *a.actions; - actions_changed = (facet->n_actions != a.n_actions - || memcmp(facet->actions, a.actions, actions_len)); + * We do not modify any 'facet' state yet, because we might need to, e.g., + * emit a NetFlow expiration and, if so, we need to have the old state + * around to properly compose it. */ + action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL); + odp_actions = xlate_actions(&ctx, new_rule->actions, new_rule->n_actions); + actions_changed = (facet->actions_len != odp_actions->size + || memcmp(facet->actions, odp_actions->data, + facet->actions_len)); /* If the ODP actions changed or the installability changed, then we need * to talk to the datapath. */ - if (actions_changed || facet->may_install != facet->installed) { - if (facet->may_install) { - struct odp_flow_put put; - - memset(&put.flow.stats, 0, sizeof put.flow.stats); - odp_flow_key_from_flow(&put.flow.key, &facet->flow); - put.flow.actions = a.actions; - put.flow.n_actions = a.n_actions; - put.flow.flags = 0; - put.flags = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS; - dpif_flow_put(ofproto->dpif, &put); - - facet_update_stats(ofproto, facet, &put.flow.stats); + if (actions_changed || ctx.may_set_up_flow != facet->installed) { + if (ctx.may_set_up_flow) { + struct dpif_flow_stats stats; + + facet_put__(ofproto, facet, + odp_actions->data, odp_actions->size, &stats); + facet_update_stats(ofproto, facet, &stats); } else { facet_uninstall(ofproto, facet); } @@ -2401,11 +2543,13 @@ facet_revalidate(struct ofproto *ofproto, struct facet *facet) } /* Update 'facet' now that we've taken care of all the old state. */ - facet->nf_flow.output_iface = new_nf_output_iface; + facet->tags = ctx.tags; + facet->nf_flow.output_iface = ctx.nf_output_iface; + facet->may_install = ctx.may_set_up_flow; if (actions_changed) { free(facet->actions); - facet->n_actions = a.n_actions; - facet->actions = xmemdup(a.actions, actions_len); + facet->actions_len = odp_actions->size; + facet->actions = xmemdup(odp_actions->data, odp_actions->size); } if (facet->rule != new_rule) { COVERAGE_INC(facet_changed_rule); @@ -2413,8 +2557,11 @@ facet_revalidate(struct ofproto *ofproto, struct facet *facet) list_push_back(&new_rule->facets, &facet->list_node); facet->rule = new_rule; facet->used = new_rule->created; + facet->rs_used = facet->used; } + ofpbuf_delete(odp_actions); + return true; } @@ -2432,7 +2579,7 @@ static void send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh, int error) { - struct ofpbuf *buf = make_ofp_error_msg(error, oh); + struct ofpbuf *buf = ofputil_encode_error_msg(error, oh); if (buf) { COVERAGE_INC(ofproto_error); queue_tx(buf, ofconn, ofconn->reply_counter); @@ -2452,15 +2599,14 @@ hton_ofp_phy_port(struct ofp_phy_port *opp) } static int -handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh) +handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh) { - struct ofp_header *rq = oh; - queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter); + queue_tx(make_echo_reply(oh), ofconn, ofconn->reply_counter); return 0; } static int -handle_features_request(struct ofconn *ofconn, struct ofp_header *oh) +handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh) { struct ofp_switch_features *osf; struct ofpbuf *buf; @@ -2494,7 +2640,7 @@ handle_features_request(struct ofconn *ofconn, struct ofp_header *oh) } static int -handle_get_config_request(struct ofconn *ofconn, struct ofp_header *oh) +handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh) { struct ofpbuf *buf; struct ofp_switch_config *osc; @@ -2515,16 +2661,9 @@ handle_get_config_request(struct ofconn *ofconn, struct ofp_header *oh) } static int -handle_set_config(struct ofconn *ofconn, struct ofp_switch_config *osc) +handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc) { - uint16_t flags; - int error; - - error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc); - if (error) { - return error; - } - flags = ntohs(osc->flags); + uint16_t flags = ntohs(osc->flags); if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) { switch (flags & OFPC_FRAG_MASK) { @@ -2546,34 +2685,6 @@ handle_set_config(struct ofconn *ofconn, struct ofp_switch_config *osc) return 0; } -static void -add_controller_action(struct odp_actions *actions, uint16_t max_len) -{ - union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER); - a->controller.arg = max_len; -} - -struct action_xlate_ctx { - /* Input. */ - 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 - * null pointer if we are revalidating - * without a packet to refer to. */ - - /* Output. */ - struct odp_actions *out; /* Datapath actions. */ - tag_type tags; /* Tags associated with OFPP_NORMAL actions. */ - bool may_set_up_flow; /* True ordinarily; false if the actions must - * be reassessed for every packet. */ - uint16_t nf_output_iface; /* Output interface index for NetFlow. */ -}; - -/* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a - * flow translation. */ -#define MAX_RESUBMIT_RECURSION 8 - static void do_xlate_actions(const union ofp_action *in, size_t n_in, struct action_xlate_ctx *ctx); @@ -2595,7 +2706,7 @@ add_output_action(struct action_xlate_ctx *ctx, uint16_t port) */ } - odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port; + nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, port); ctx->nf_output_iface = port; } @@ -2620,13 +2731,17 @@ xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port) rule = rule_lookup(ctx->ofproto, &ctx->flow); ctx->flow.in_port = old_in_port; + if (ctx->resubmit_hook) { + ctx->resubmit_hook(ctx, rule); + } + if (rule) { ctx->recurse++; do_xlate_actions(rule->actions, rule->n_actions, ctx); ctx->recurse--; } } else { - struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); + static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times", MAX_RESUBMIT_RECURSION); @@ -2635,14 +2750,14 @@ xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port) static void flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask, - uint16_t *nf_output_iface, struct odp_actions *actions) + uint16_t *nf_output_iface, struct ofpbuf *odp_actions) { struct ofport *ofport; HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) { uint16_t odp_port = ofport->odp_port; if (odp_port != odp_in_port && !(ofport->opp.config & mask)) { - odp_actions_add(actions, ODPAT_OUTPUT)->output.port = odp_port; + nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port); } } *nf_output_iface = NF_OUT_FLOOD; @@ -2666,7 +2781,7 @@ xlate_output_action__(struct action_xlate_ctx *ctx, break; case OFPP_NORMAL: if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet, - ctx->out, &ctx->tags, + ctx->odp_actions, &ctx->tags, &ctx->nf_output_iface, ctx->ofproto->aux)) { COVERAGE_INC(ofproto_uninstallable); @@ -2675,14 +2790,14 @@ xlate_output_action__(struct action_xlate_ctx *ctx, break; case OFPP_FLOOD: flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD, - &ctx->nf_output_iface, ctx->out); + &ctx->nf_output_iface, ctx->odp_actions); break; case OFPP_ALL: flood_packets(ctx->ofproto, ctx->flow.in_port, 0, - &ctx->nf_output_iface, ctx->out); + &ctx->nf_output_iface, ctx->odp_actions); break; case OFPP_CONTROLLER: - add_controller_action(ctx->out, max_len); + nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len); break; case OFPP_LOCAL: add_output_action(ctx, ODPP_LOCAL); @@ -2719,9 +2834,18 @@ xlate_output_action(struct action_xlate_ctx *ctx, static void remove_pop_action(struct action_xlate_ctx *ctx) { - size_t n = ctx->out->n_actions; - if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) { - ctx->out->n_actions--; + if (ctx->odp_actions->size == ctx->last_pop_priority) { + ctx->odp_actions->size -= NLA_ALIGN(NLA_HDRLEN); + ctx->last_pop_priority = -1; + } +} + +static void +add_pop_action(struct action_xlate_ctx *ctx) +{ + if (ctx->odp_actions->size != ctx->last_pop_priority) { + nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_POP_PRIORITY); + ctx->last_pop_priority = ctx->odp_actions->size; } } @@ -2751,10 +2875,9 @@ xlate_enqueue_action(struct action_xlate_ctx *ctx, /* Add ODP actions. */ remove_pop_action(ctx); - odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority - = priority; + nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority); add_output_action(ctx, odp_port); - odp_actions_add(ctx->out, ODPAT_POP_PRIORITY); + add_pop_action(ctx); /* Update NetFlow output port. */ if (ctx->nf_output_iface == NF_OUT_DROP) { @@ -2780,38 +2903,45 @@ xlate_set_queue_action(struct action_xlate_ctx *ctx, } remove_pop_action(ctx); - odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority - = priority; + nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, priority); } static void xlate_set_dl_tci(struct action_xlate_ctx *ctx) { - ovs_be16 dl_vlan = ctx->flow.dl_vlan; - uint8_t dl_vlan_pcp = ctx->flow.dl_vlan_pcp; - - if (dl_vlan == htons(OFP_VLAN_NONE)) { - odp_actions_add(ctx->out, ODPAT_STRIP_VLAN); + ovs_be16 tci = ctx->flow.vlan_tci; + if (!(tci & htons(VLAN_CFI))) { + nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN); } else { - union odp_action *oa = odp_actions_add(ctx->out, ODPAT_SET_DL_TCI); - oa->dl_tci.tci = htons(ntohs(dl_vlan & htons(VLAN_VID_MASK)) - | (dl_vlan_pcp << VLAN_PCP_SHIFT) - | VLAN_CFI); + nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_TCI, + tci & ~htons(VLAN_CFI)); } } +struct xlate_reg_state { + ovs_be16 vlan_tci; + ovs_be64 tun_id; +}; + static void -xlate_reg_move_action(struct action_xlate_ctx *ctx, - const struct nx_action_reg_move *narm) +save_reg_state(const struct action_xlate_ctx *ctx, + struct xlate_reg_state *state) { - ovs_be16 old_vlan = ctx->flow.dl_vlan; - uint8_t old_pcp = ctx->flow.dl_vlan_pcp; - - nxm_execute_reg_move(narm, &ctx->flow); + state->vlan_tci = ctx->flow.vlan_tci; + state->tun_id = ctx->flow.tun_id; +} - if (ctx->flow.dl_vlan != old_vlan || ctx->flow.dl_vlan_pcp != old_pcp) { +static void +update_reg_state(struct action_xlate_ctx *ctx, + const struct xlate_reg_state *state) +{ + if (ctx->flow.vlan_tci != state->vlan_tci) { xlate_set_dl_tci(ctx); } + if (ctx->flow.tun_id != state->tun_id) { + nl_msg_put_be64(ctx->odp_actions, + ODP_ACTION_ATTR_SET_TUNNEL, ctx->flow.tun_id); + } } static void @@ -2821,8 +2951,10 @@ xlate_nicira_action(struct action_xlate_ctx *ctx, 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); + const struct nx_action_multipath *nam; + enum nx_action_subtype subtype = ntohs(nah->subtype); + struct xlate_reg_state state; + ovs_be64 tun_id; assert(nah->vendor == htonl(NX_VENDOR_ID)); switch (subtype) { @@ -2833,13 +2965,15 @@ xlate_nicira_action(struct action_xlate_ctx *ctx, case NXAST_SET_TUNNEL: nast = (const struct nx_action_set_tunnel *) nah; - oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL); - ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id; + tun_id = htonll(ntohl(nast->tun_id)); + nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id); + ctx->flow.tun_id = tun_id; break; case NXAST_DROP_SPOOFED_ARP: if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) { - odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP); + nl_msg_put_flag(ctx->odp_actions, + ODP_ACTION_ATTR_DROP_SPOOFED_ARP); } break; @@ -2849,23 +2983,44 @@ xlate_nicira_action(struct action_xlate_ctx *ctx, break; case NXAST_POP_QUEUE: - odp_actions_add(ctx->out, ODPAT_POP_PRIORITY); + add_pop_action(ctx); break; case NXAST_REG_MOVE: - xlate_reg_move_action(ctx, (const struct nx_action_reg_move *) nah); + save_reg_state(ctx, &state); + nxm_execute_reg_move((const struct nx_action_reg_move *) nah, + &ctx->flow); + update_reg_state(ctx, &state); break; case NXAST_REG_LOAD: + save_reg_state(ctx, &state); nxm_execute_reg_load((const struct nx_action_reg_load *) nah, &ctx->flow); + update_reg_state(ctx, &state); + break; + + case NXAST_NOTE: + /* Nothing to do. */ + break; + + case NXAST_SET_TUNNEL64: + tun_id = ((const struct nx_action_set_tunnel64 *) nah)->tun_id; + nl_msg_put_be64(ctx->odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, tun_id); + ctx->flow.tun_id = tun_id; + break; + + case NXAST_MULTIPATH: + nam = (const struct nx_action_multipath *) nah; + multipath_execute(nam, &ctx->flow); break; /* If you add a new action here that modifies flow data, don't forget to * update the flow key in ctx->flow at the same time. */ + case NXAST_SNAT__OBSOLETE: default: - VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype); + VLOG_DBG_RL(&rl, "unknown Nicira action type %d", (int) subtype); break; } } @@ -2887,8 +3042,8 @@ do_xlate_actions(const union ofp_action *in, size_t n_in, } for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) { - uint16_t type = ntohs(ia->type); - union odp_action *oa; + enum ofp_action_type type = ntohs(ia->type); + const struct ofp_action_dl_addr *oada; switch (type) { case OFPAT_OUTPUT: @@ -2896,60 +3051,65 @@ do_xlate_actions(const union ofp_action *in, size_t n_in, break; case OFPAT_SET_VLAN_VID: - ctx->flow.dl_vlan = ia->vlan_vid.vlan_vid; + ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); + ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI); xlate_set_dl_tci(ctx); break; case OFPAT_SET_VLAN_PCP: - ctx->flow.dl_vlan_pcp = ia->vlan_pcp.vlan_pcp; + ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); + ctx->flow.vlan_tci |= htons( + (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI); xlate_set_dl_tci(ctx); break; case OFPAT_STRIP_VLAN: - ctx->flow.dl_vlan = htons(OFP_VLAN_NONE); - ctx->flow.dl_vlan_pcp = 0; + ctx->flow.vlan_tci = htons(0); xlate_set_dl_tci(ctx); break; case OFPAT_SET_DL_SRC: - oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC); - memcpy(oa->dl_addr.dl_addr, - ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); - memcpy(ctx->flow.dl_src, - ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); + oada = ((struct ofp_action_dl_addr *) ia); + nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_SRC, + oada->dl_addr, ETH_ADDR_LEN); + memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN); break; case OFPAT_SET_DL_DST: - oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST); - memcpy(oa->dl_addr.dl_addr, - ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); - memcpy(ctx->flow.dl_dst, - ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); + oada = ((struct ofp_action_dl_addr *) ia); + nl_msg_put_unspec(ctx->odp_actions, ODP_ACTION_ATTR_SET_DL_DST, + oada->dl_addr, ETH_ADDR_LEN); + memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN); break; case OFPAT_SET_NW_SRC: - oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC); - ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; + nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_SRC, + ia->nw_addr.nw_addr); + ctx->flow.nw_src = ia->nw_addr.nw_addr; break; case OFPAT_SET_NW_DST: - oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST); - ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; + nl_msg_put_be32(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_DST, + ia->nw_addr.nw_addr); + ctx->flow.nw_dst = ia->nw_addr.nw_addr; break; case OFPAT_SET_NW_TOS: - oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS); - ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos; + nl_msg_put_u8(ctx->odp_actions, ODP_ACTION_ATTR_SET_NW_TOS, + ia->nw_tos.nw_tos); + ctx->flow.nw_tos = ia->nw_tos.nw_tos; break; case OFPAT_SET_TP_SRC: - oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC); - ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port; + nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_SRC, + ia->tp_port.tp_port); + ctx->flow.tp_src = ia->tp_port.tp_port; break; case OFPAT_SET_TP_DST: - oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST); - ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port; + nl_msg_put_be16(ctx->odp_actions, ODP_ACTION_ATTR_SET_TP_DST, + ia->tp_port.tp_port); + ctx->flow.tp_dst = ia->tp_port.tp_port; break; case OFPAT_VENDOR: @@ -2961,55 +3121,56 @@ do_xlate_actions(const union ofp_action *in, size_t n_in, break; default: - VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type); + VLOG_DBG_RL(&rl, "unknown action type %d", (int) type); break; } } } -static int -xlate_actions(const union ofp_action *in, size_t n_in, - const 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) +static void +action_xlate_ctx_init(struct action_xlate_ctx *ctx, + struct ofproto *ofproto, const struct flow *flow, + const struct ofpbuf *packet) { - struct action_xlate_ctx ctx; + ctx->ofproto = ofproto; + ctx->flow = *flow; + ctx->packet = packet; + ctx->resubmit_hook = NULL; + ctx->check_special = true; +} +static struct ofpbuf * +xlate_actions(struct action_xlate_ctx *ctx, + const union ofp_action *in, size_t n_in) +{ COVERAGE_INC(ofproto_ofp2odp); - odp_actions_init(out); - ctx.flow = *flow; - ctx.recurse = 0; - ctx.ofproto = ofproto; - ctx.packet = packet; - ctx.out = out; - ctx.tags = 0; - ctx.may_set_up_flow = true; - ctx.nf_output_iface = NF_OUT_DROP; - do_xlate_actions(in, n_in, &ctx); - remove_pop_action(&ctx); + + ctx->odp_actions = ofpbuf_new(512); + ctx->tags = 0; + ctx->may_set_up_flow = true; + ctx->nf_output_iface = NF_OUT_DROP; + ctx->recurse = 0; + ctx->last_pop_priority = -1; + + if (!ctx->check_special + || !ctx->ofproto->ofhooks->special_cb + || ctx->ofproto->ofhooks->special_cb(&ctx->flow, ctx->packet, + ctx->ofproto->aux)) { + do_xlate_actions(in, n_in, ctx); + } else { + ctx->may_set_up_flow = false; + } + + remove_pop_action(ctx); /* Check with in-band control to see if we're allowed to set up this * flow. */ - if (!in_band_rule_check(ofproto->in_band, flow, out)) { - ctx.may_set_up_flow = false; + if (!in_band_rule_check(ctx->ofproto->in_band, &ctx->flow, + ctx->odp_actions->data, ctx->odp_actions->size)) { + ctx->may_set_up_flow = false; } - if (tags) { - *tags = ctx.tags; - } - if (may_set_up_flow) { - *may_set_up_flow = ctx.may_set_up_flow; - } - if (nf_output_iface) { - *nf_output_iface = ctx.nf_output_iface; - } - if (odp_actions_overflow(out)) { - COVERAGE_INC(odp_overflow); - odp_actions_init(out); - return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY); - } - return 0; + return ctx->odp_actions; } /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow @@ -3032,13 +3193,14 @@ reject_slave_controller(struct ofconn *ofconn, const const char *msg_type) } static int -handle_packet_out(struct ofconn *ofconn, struct ofp_header *oh) +handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh) { struct ofproto *p = ofconn->ofproto; struct ofp_packet_out *opo; struct ofpbuf payload, *buffer; union ofp_action *ofp_actions; - struct odp_actions odp_actions; + struct action_xlate_ctx ctx; + struct ofpbuf *odp_actions; struct ofpbuf request; struct flow flow; size_t n_ofp_actions; @@ -3053,12 +3215,8 @@ handle_packet_out(struct ofconn *ofconn, struct ofp_header *oh) } /* Get ofp_packet_out. */ - request.data = oh; - request.size = ntohs(oh->length); - opo = ofpbuf_try_pull(&request, offsetof(struct ofp_packet_out, actions)); - if (!opo) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); - } + ofpbuf_use_const(&request, oh, ntohs(oh->length)); + opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions)); /* Get actions. */ error = ofputil_pull_actions(&request, ntohs(opo->actions_len), @@ -3089,12 +3247,10 @@ handle_packet_out(struct ofconn *ofconn, struct ofp_header *oh) } /* Send. */ - error = xlate_actions(ofp_actions, n_ofp_actions, &flow, p, &payload, - &odp_actions, NULL, NULL, NULL); - if (!error) { - dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions, - &payload); - } + action_xlate_ctx_init(&ctx, p, &flow, &payload); + odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions); + dpif_execute(p->dpif, odp_actions->data, odp_actions->size, &payload); + ofpbuf_delete(odp_actions); exit: ofpbuf_delete(buffer); @@ -3127,10 +3283,10 @@ update_port_config(struct ofproto *p, struct ofport *port, } static int -handle_port_mod(struct ofconn *ofconn, struct ofp_header *oh) +handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh) { struct ofproto *p = ofconn->ofproto; - const struct ofp_port_mod *opm; + const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh; struct ofport *port; int error; @@ -3138,11 +3294,6 @@ handle_port_mod(struct ofconn *ofconn, struct ofp_header *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 = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no))); if (!port) { @@ -3172,9 +3323,11 @@ make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len) } static struct ofpbuf * -start_ofp_stats_reply(const struct ofp_stats_request *request, size_t body_len) +start_ofp_stats_reply(const struct ofp_header *request, size_t body_len) { - return make_ofp_stats_reply(request->header.xid, request->type, body_len); + const struct ofp_stats_request *osr + = (const struct ofp_stats_request *) request; + return make_ofp_stats_reply(osr->header.xid, osr->type, body_len); } static void * @@ -3203,7 +3356,7 @@ make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len) nsm->type = htons(OFPST_VENDOR); nsm->flags = htons(0); nsm->vendor = htonl(NX_VENDOR_ID); - nsm->subtype = htonl(subtype); + nsm->subtype = subtype; return msg; } @@ -3230,7 +3383,7 @@ append_nxstats_reply(size_t nbytes, struct ofconn *ofconn, static int handle_desc_stats_request(struct ofconn *ofconn, - struct ofp_stats_request *request) + const struct ofp_header *request) { struct ofproto *p = ofconn->ofproto; struct ofp_desc_stats *ods; @@ -3251,7 +3404,7 @@ handle_desc_stats_request(struct ofconn *ofconn, static int handle_table_stats_request(struct ofconn *ofconn, - struct ofp_stats_request *request) + const struct ofp_header *request) { struct ofproto *p = ofconn->ofproto; struct ofp_table_stats *ots; @@ -3267,8 +3420,8 @@ handle_table_stats_request(struct ofconn *ofconn, ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL)); ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */ ots->active_count = htonl(classifier_count(&p->cls)); - ots->lookup_count = htonll(0); /* XXX */ - ots->matched_count = htonll(0); /* XXX */ + put_32aligned_be64(&ots->lookup_count, htonll(0)); /* XXX */ + put_32aligned_be64(&ots->matched_count, htonll(0)); /* XXX */ queue_tx(msg, ofconn, ofconn->reply_counter); return 0; @@ -3289,36 +3442,30 @@ append_port_stat(struct ofport *port, struct ofconn *ofconn, ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp); ops->port_no = htons(port->opp.port_no); memset(ops->pad, 0, sizeof ops->pad); - ops->rx_packets = htonll(stats.rx_packets); - ops->tx_packets = htonll(stats.tx_packets); - ops->rx_bytes = htonll(stats.rx_bytes); - ops->tx_bytes = htonll(stats.tx_bytes); - ops->rx_dropped = htonll(stats.rx_dropped); - ops->tx_dropped = htonll(stats.tx_dropped); - ops->rx_errors = htonll(stats.rx_errors); - ops->tx_errors = htonll(stats.tx_errors); - ops->rx_frame_err = htonll(stats.rx_frame_errors); - ops->rx_over_err = htonll(stats.rx_over_errors); - ops->rx_crc_err = htonll(stats.rx_crc_errors); - ops->collisions = htonll(stats.collisions); + put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets)); + put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets)); + put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes)); + put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes)); + put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped)); + put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped)); + put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors)); + put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors)); + put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors)); + put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors)); + put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors)); + put_32aligned_be64(&ops->collisions, htonll(stats.collisions)); } static int -handle_port_stats_request(struct ofconn *ofconn, struct ofp_stats_request *osr, - size_t arg_size) +handle_port_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) { struct ofproto *p = ofconn->ofproto; - struct ofp_port_stats_request *psr; + const struct ofp_port_stats_request *psr = ofputil_stats_body(oh); struct ofp_port_stats *ops; struct ofpbuf *msg; struct ofport *port; - if (arg_size != sizeof *psr) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); - } - psr = (struct ofp_port_stats_request *) osr->body; - - msg = start_ofp_stats_reply(osr, sizeof *ops * 16); + msg = start_ofp_stats_reply(oh, sizeof *ops * 16); if (psr->port_no != htons(OFPP_NONE)) { port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no))); if (port) { @@ -3334,61 +3481,6 @@ handle_port_stats_request(struct ofconn *ofconn, struct ofp_stats_request *osr, return 0; } -struct flow_stats_cbdata { - struct ofconn *ofconn; - ovs_be16 out_port; - struct ofpbuf *msg; -}; - -/* Obtains statistic counters for 'rule' within 'p' and stores them into - * '*packet_countp' and '*byte_countp'. The returned statistics include - * statistics for all of 'rule''s facets. */ -static void -query_stats(struct ofproto *p, struct rule *rule, - uint64_t *packet_countp, uint64_t *byte_countp) -{ - uint64_t packet_count, byte_count; - struct facet *facet; - struct odp_flow *odp_flows; - size_t n_odp_flows; - - /* Start from historical data for 'rule' itself that are no longer tracked - * by the datapath. This counts, for example, facets that have expired. */ - packet_count = rule->packet_count; - byte_count = rule->byte_count; - - /* Prepare to ask the datapath for statistics on all of the rule's facets. - * - * Also, add any statistics that are not tracked by the datapath for each - * facet. This includes, for example, statistics for packets that were - * executed "by hand" by ofproto via dpif_execute() but must be accounted - * to a rule. */ - odp_flows = xzalloc(list_size(&rule->facets) * sizeof *odp_flows); - n_odp_flows = 0; - LIST_FOR_EACH (facet, list_node, &rule->facets) { - struct odp_flow *odp_flow = &odp_flows[n_odp_flows++]; - odp_flow_key_from_flow(&odp_flow->key, &facet->flow); - packet_count += facet->packet_count; - byte_count += facet->byte_count; - } - - /* Fetch up-to-date statistics from the datapath and add them in. */ - if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) { - size_t i; - - for (i = 0; i < n_odp_flows; i++) { - struct odp_flow *odp_flow = &odp_flows[i]; - packet_count += odp_flow->stats.n_packets; - byte_count += odp_flow->stats.n_bytes; - } - } - free(odp_flows); - - /* Return the stats to the caller. */ - *packet_countp = packet_count; - *byte_countp = byte_count; -} - static void calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec) { @@ -3398,37 +3490,37 @@ calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec) } static void -flow_stats_cb(struct cls_rule *rule_, void *cbdata_) +put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule, + ovs_be16 out_port, struct ofpbuf **replyp) { - struct rule *rule = rule_from_cls_rule(rule_); - struct flow_stats_cbdata *cbdata = cbdata_; struct ofp_flow_stats *ofs; uint64_t packet_count, byte_count; + ovs_be64 cookie; size_t act_len, len; - if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { + if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) { return; } act_len = sizeof *rule->actions * rule->n_actions; len = offsetof(struct ofp_flow_stats, actions) + act_len; - query_stats(cbdata->ofconn->ofproto, rule, &packet_count, &byte_count); + rule_get_stats(rule, &packet_count, &byte_count); - ofs = append_ofp_stats_reply(len, cbdata->ofconn, &cbdata->msg); + ofs = append_ofp_stats_reply(len, ofconn, replyp); ofs->length = htons(len); ofs->table_id = 0; ofs->pad = 0; - flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, - cbdata->ofconn->flow_format, &ofs->match); + ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match, + rule->flow_cookie, &cookie); + put_32aligned_be64(&ofs->cookie, cookie); calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec); - ofs->cookie = rule->flow_cookie; ofs->priority = htons(rule->cr.priority); ofs->idle_timeout = htons(rule->idle_timeout); ofs->hard_timeout = htons(rule->hard_timeout); memset(ofs->pad2, 0, sizeof ofs->pad2); - ofs->packet_count = htonll(packet_count); - ofs->byte_count = htonll(byte_count); + put_32aligned_be64(&ofs->packet_count, htonll(packet_count)); + put_32aligned_be64(&ofs->byte_count, htonll(byte_count)); if (rule->n_actions > 0) { memcpy(ofs->actions, rule->actions, act_len); } @@ -3437,58 +3529,65 @@ flow_stats_cb(struct cls_rule *rule_, void *cbdata_) static bool is_valid_table(uint8_t table_id) { - return table_id == 0 || table_id == 0xff; + if (table_id == 0 || table_id == 0xff) { + return true; + } else { + /* It would probably be better to reply with an error but there doesn't + * seem to be any appropriate value, so that might just be + * confusing. */ + VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8, + table_id); + return false; + } } static int -handle_flow_stats_request(struct ofconn *ofconn, - const struct ofp_stats_request *osr, size_t arg_size) +handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) { - struct ofp_flow_stats_request *fsr; - struct flow_stats_cbdata cbdata; - - if (arg_size != sizeof *fsr) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); - } - fsr = (struct ofp_flow_stats_request *) osr->body; + const struct ofp_flow_stats_request *fsr = ofputil_stats_body(oh); + struct ofpbuf *reply; COVERAGE_INC(ofproto_flows_req); - cbdata.msg = start_ofp_stats_reply(osr, 1024); + reply = start_ofp_stats_reply(oh, 1024); if (is_valid_table(fsr->table_id)) { + struct cls_cursor cursor; struct cls_rule target; + struct rule *rule; - cbdata.ofconn = ofconn; - cbdata.out_port = fsr->out_port; - cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0, &target); - classifier_for_each_match(&ofconn->ofproto->cls, &target, - flow_stats_cb, &cbdata); + ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0, + &target); + cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target); + CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { + put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply); + } } + queue_tx(reply, ofconn, ofconn->reply_counter); - queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); return 0; } static void -nx_flow_stats_cb(struct cls_rule *rule_, void *cbdata_) +put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule, + ovs_be16 out_port, struct ofpbuf **replyp) { - struct rule *rule = rule_from_cls_rule(rule_); - struct flow_stats_cbdata *cbdata = cbdata_; struct nx_flow_stats *nfs; uint64_t packet_count, byte_count; size_t act_len, start_len; + struct ofpbuf *reply; - if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { + if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) { return; } - query_stats(cbdata->ofconn->ofproto, rule, &packet_count, &byte_count); + rule_get_stats(rule, &packet_count, &byte_count); act_len = sizeof *rule->actions * rule->n_actions; - start_len = cbdata->msg->size; - append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, - cbdata->ofconn, &cbdata->msg); - nfs = ofpbuf_put_uninit(cbdata->msg, sizeof *nfs); + append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp); + start_len = (*replyp)->size; + reply = *replyp; + + nfs = ofpbuf_put_uninit(reply, sizeof *nfs); nfs->table_id = 0; nfs->pad = 0; calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec); @@ -3496,71 +3595,69 @@ nx_flow_stats_cb(struct cls_rule *rule_, void *cbdata_) nfs->priority = htons(rule->cr.priority); nfs->idle_timeout = htons(rule->idle_timeout); nfs->hard_timeout = htons(rule->hard_timeout); - nfs->match_len = htons(nx_put_match(cbdata->msg, &rule->cr)); + nfs->match_len = htons(nx_put_match(reply, &rule->cr)); memset(nfs->pad2, 0, sizeof nfs->pad2); nfs->packet_count = htonll(packet_count); nfs->byte_count = htonll(byte_count); if (rule->n_actions > 0) { - ofpbuf_put(cbdata->msg, rule->actions, act_len); + ofpbuf_put(reply, rule->actions, act_len); } - nfs->length = htons(cbdata->msg->size - start_len); + nfs->length = htons(reply->size - start_len); } static int -handle_nxst_flow(struct ofconn *ofconn, struct ofpbuf *b) +handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh) { struct nx_flow_stats_request *nfsr; - struct flow_stats_cbdata cbdata; struct cls_rule target; + struct ofpbuf *reply; + struct ofpbuf b; int error; + ofpbuf_use_const(&b, oh, ntohs(oh->length)); + /* Dissect the message. */ - nfsr = ofpbuf_try_pull(b, sizeof *nfsr); - if (!nfsr) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); - } - error = nx_pull_match(b, ntohs(nfsr->match_len), 0, &target); + nfsr = ofpbuf_pull(&b, sizeof *nfsr); + error = nx_pull_match(&b, ntohs(nfsr->match_len), 0, &target); if (error) { return error; } + if (b.size) { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); + } COVERAGE_INC(ofproto_flows_req); - cbdata.msg = start_nxstats_reply(&nfsr->nsm, 1024); + reply = start_nxstats_reply(&nfsr->nsm, 1024); if (is_valid_table(nfsr->table_id)) { - cbdata.ofconn = ofconn; - cbdata.out_port = nfsr->out_port; - classifier_for_each_match(&ofconn->ofproto->cls, &target, - nx_flow_stats_cb, &cbdata); + struct cls_cursor cursor; + struct rule *rule; + + cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target); + CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { + put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply); + } } - queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); + queue_tx(reply, ofconn, ofconn->reply_counter); + return 0; } -struct flow_stats_ds_cbdata { - struct ofproto *ofproto; - struct ds *results; -}; - static void -flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_) +flow_stats_ds(struct rule *rule, struct ds *results) { - struct rule *rule = rule_from_cls_rule(rule_); - struct flow_stats_ds_cbdata *cbdata = cbdata_; - struct ds *results = cbdata->results; - struct ofp_match match; uint64_t packet_count, byte_count; size_t act_len = sizeof *rule->actions * rule->n_actions; - query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); - flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, - NXFF_OPENFLOW10, &match); + rule_get_stats(rule, &packet_count, &byte_count); ds_put_format(results, "duration=%llds, ", (time_msec() - rule->created) / 1000); + ds_put_format(results, "idle=%.3fs, ", (time_msec() - rule->used) / 1000.0); ds_put_format(results, "priority=%u, ", rule->cr.priority); ds_put_format(results, "n_packets=%"PRIu64", ", packet_count); ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count); - ofp_print_match(results, &match, true); + cls_rule_format(&rule->cr, results); + ds_put_char(results, ','); if (act_len > 0) { ofp_print_actions(results, &rule->actions->header, act_len); } else { @@ -3570,48 +3667,17 @@ flow_stats_ds_cb(struct cls_rule *rule_, void *cbdata_) } /* Adds a pretty-printed description of all flows to 'results', including - * those marked hidden by secchan (e.g., by in-band control). */ + * hidden flows (e.g., set up by in-band control). */ void ofproto_get_all_flows(struct ofproto *p, struct ds *results) { - struct ofp_match match; - struct cls_rule target; - struct flow_stats_ds_cbdata cbdata; - - memset(&match, 0, sizeof match); - match.wildcards = htonl(OVSFW_ALL); - - cbdata.ofproto = p; - cbdata.results = results; - - cls_rule_from_match(&match, 0, NXFF_OPENFLOW10, 0, &target); - classifier_for_each_match(&p->cls, &target, flow_stats_ds_cb, &cbdata); -} - -struct aggregate_stats_cbdata { - struct ofproto *ofproto; - ovs_be16 out_port; - uint64_t packet_count; - uint64_t byte_count; - uint32_t n_flows; -}; - -static void -aggregate_stats_cb(struct cls_rule *rule_, void *cbdata_) -{ - struct rule *rule = rule_from_cls_rule(rule_); - struct aggregate_stats_cbdata *cbdata = cbdata_; - uint64_t packet_count, byte_count; + struct cls_cursor cursor; + struct rule *rule; - if (rule_is_hidden(rule) || !rule_has_out_port(rule, cbdata->out_port)) { - return; + cls_cursor_init(&cursor, &p->cls, NULL); + CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { + flow_stats_ds(rule, results); } - - query_stats(cbdata->ofproto, rule, &packet_count, &byte_count); - - cbdata->packet_count += packet_count; - cbdata->byte_count += byte_count; - cbdata->n_flows++; } static void @@ -3619,44 +3685,50 @@ query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target, ovs_be16 out_port, uint8_t table_id, struct ofp_aggregate_stats_reply *oasr) { - struct aggregate_stats_cbdata cbdata; + uint64_t total_packets = 0; + uint64_t total_bytes = 0; + int n_flows = 0; COVERAGE_INC(ofproto_agg_request); - cbdata.packet_count = 0; - cbdata.byte_count = 0; - cbdata.n_flows = 0; + if (is_valid_table(table_id)) { - cbdata.ofproto = ofproto; - cbdata.out_port = out_port; + struct cls_cursor cursor; + struct rule *rule; + + cls_cursor_init(&cursor, &ofproto->cls, target); + CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { + if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) { + uint64_t packet_count; + uint64_t byte_count; - classifier_for_each_match(&ofproto->cls, target, - aggregate_stats_cb, &cbdata); + rule_get_stats(rule, &packet_count, &byte_count); + + total_packets += packet_count; + total_bytes += byte_count; + n_flows++; + } + } } - oasr->flow_count = htonl(cbdata.n_flows); - oasr->packet_count = htonll(cbdata.packet_count); - oasr->byte_count = htonll(cbdata.byte_count); + oasr->flow_count = htonl(n_flows); + put_32aligned_be64(&oasr->packet_count, htonll(total_packets)); + put_32aligned_be64(&oasr->byte_count, htonll(total_bytes)); memset(oasr->pad, 0, sizeof oasr->pad); } static int handle_aggregate_stats_request(struct ofconn *ofconn, - const struct ofp_stats_request *osr, - size_t arg_size) + const struct ofp_header *oh) { - struct ofp_aggregate_stats_request *request; + const struct ofp_aggregate_stats_request *request = ofputil_stats_body(oh); struct ofp_aggregate_stats_reply *reply; struct cls_rule target; struct ofpbuf *msg; - if (arg_size != sizeof *request) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); - } - request = (struct ofp_aggregate_stats_request *) osr->body; - - cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0, &target); + ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0, + &target); - msg = start_ofp_stats_reply(osr, sizeof *reply); + msg = start_ofp_stats_reply(oh, sizeof *reply); reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg); query_aggregate_stats(ofconn->ofproto, &target, request->out_port, request->table_id, reply); @@ -3665,23 +3737,26 @@ handle_aggregate_stats_request(struct ofconn *ofconn, } static int -handle_nxst_aggregate(struct ofconn *ofconn, struct ofpbuf *b) +handle_nxst_aggregate(struct ofconn *ofconn, const struct ofp_header *oh) { struct nx_aggregate_stats_request *request; struct ofp_aggregate_stats_reply *reply; struct cls_rule target; + struct ofpbuf b; struct ofpbuf *buf; int error; + ofpbuf_use_const(&b, oh, ntohs(oh->length)); + /* Dissect the message. */ - request = ofpbuf_try_pull(b, sizeof *request); - if (!request) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); - } - error = nx_pull_match(b, ntohs(request->match_len), 0, &target); + request = ofpbuf_pull(&b, sizeof *request); + error = nx_pull_match(&b, ntohs(request->match_len), 0, &target); if (error) { return error; } + if (b.size) { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); + } /* Reply. */ COVERAGE_INC(ofproto_flows_req); @@ -3710,9 +3785,9 @@ put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id, reply->port_no = htons(cbdata->ofport->opp.port_no); memset(reply->pad, 0, sizeof reply->pad); reply->queue_id = htonl(queue_id); - reply->tx_bytes = htonll(stats->tx_bytes); - reply->tx_packets = htonll(stats->tx_packets); - reply->tx_errors = htonll(stats->tx_errors); + put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes)); + put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets)); + put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors)); } static void @@ -3743,26 +3818,24 @@ handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id, } static int -handle_queue_stats_request(struct ofconn *ofconn, - const struct ofp_stats_request *osr, - size_t arg_size) +handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) { struct ofproto *ofproto = ofconn->ofproto; - struct ofp_queue_stats_request *qsr; + const struct ofp_queue_stats_request *qsr; struct queue_stats_cbdata cbdata; struct ofport *port; unsigned int port_no; uint32_t queue_id; - if (arg_size != sizeof *qsr) { + qsr = ofputil_stats_body(oh); + if (!qsr) { return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); } - qsr = (struct ofp_queue_stats_request *) osr->body; COVERAGE_INC(ofproto_queue_req); cbdata.ofconn = ofconn; - cbdata.msg = start_ofp_stats_reply(osr, 128); + cbdata.msg = start_ofp_stats_reply(oh, 128); port_no = ntohs(qsr->port_no); queue_id = ntohl(qsr->queue_id); @@ -3784,96 +3857,12 @@ handle_queue_stats_request(struct ofconn *ofconn, return 0; } -static int -handle_vendor_stats_request(struct ofconn *ofconn, - struct ofp_stats_request *osr, size_t arg_size) -{ - struct nicira_stats_msg *nsm; - struct ofpbuf b; - ovs_be32 vendor; - - if (arg_size < 4) { - VLOG_WARN_RL(&rl, "truncated vendor stats request body"); - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); - } - - memcpy(&vendor, osr->body, sizeof vendor); - if (vendor != htonl(NX_VENDOR_ID)) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); - } - - if (ntohs(osr->header.length) < sizeof(struct nicira_stats_msg)) { - VLOG_WARN_RL(&rl, "truncated Nicira stats request"); - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); - } - - nsm = (struct nicira_stats_msg *) osr; - b.data = nsm; - b.size = ntohs(nsm->header.length); - switch (ntohl(nsm->subtype)) { - case NXST_FLOW: - return handle_nxst_flow(ofconn, &b); - - case NXST_AGGREGATE: - return handle_nxst_aggregate(ofconn, &b); - - default: - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); - } -} - -static int -handle_stats_request(struct ofconn *ofconn, struct ofp_header *oh) -{ - struct ofp_stats_request *osr; - size_t arg_size; - int error; - - error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr, - 1, &arg_size); - if (error) { - return error; - } - osr = (struct ofp_stats_request *) oh; - - switch (ntohs(osr->type)) { - case OFPST_DESC: - return handle_desc_stats_request(ofconn, osr); - - case OFPST_FLOW: - return handle_flow_stats_request(ofconn, osr, arg_size); - - case OFPST_AGGREGATE: - return handle_aggregate_stats_request(ofconn, osr, arg_size); - - case OFPST_TABLE: - return handle_table_stats_request(ofconn, osr); - - case OFPST_PORT: - return handle_port_stats_request(ofconn, osr, arg_size); - - case OFPST_QUEUE: - return handle_queue_stats_request(ofconn, osr, arg_size); - - case OFPST_VENDOR: - return handle_vendor_stats_request(ofconn, osr, arg_size); - - default: - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT); - } -} - -static long long int -msec_from_nsec(uint64_t sec, uint32_t nsec) -{ - return !sec ? 0 : sec * 1000 + nsec / 1000000; -} - +/* Updates 'facet''s used time. Caller is responsible for calling + * facet_push_stats() to update the flows which 'facet' resubmits into. */ static void facet_update_time(struct ofproto *ofproto, struct facet *facet, - const struct odp_flow_stats *stats) + long long int used) { - long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec); if (used > facet->used) { facet->used = used; if (used > facet->rule->used) { @@ -3891,29 +3880,76 @@ facet_update_time(struct ofproto *ofproto, struct facet *facet, * cleared out of the datapath. */ static void facet_update_stats(struct ofproto *ofproto, struct facet *facet, - const struct odp_flow_stats *stats) + const struct dpif_flow_stats *stats) { - if (stats->n_packets) { - facet_update_time(ofproto, facet, stats); + if (stats->n_packets || stats->used > facet->used) { + facet_update_time(ofproto, facet, stats->used); facet->packet_count += stats->n_packets; facet->byte_count += stats->n_bytes; + facet_push_stats(ofproto, facet); netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); } } -struct flow_mod { - struct cls_rule cr; - ovs_be64 cookie; - uint16_t command; - uint16_t idle_timeout; - uint16_t hard_timeout; - uint32_t buffer_id; - uint16_t out_port; - uint16_t flags; - union ofp_action *actions; - size_t n_actions; +static void +facet_push_stats(struct ofproto *ofproto, struct facet *facet) +{ + uint64_t rs_packets, rs_bytes; + + assert(facet->packet_count >= facet->rs_packet_count); + assert(facet->byte_count >= facet->rs_byte_count); + assert(facet->used >= facet->rs_used); + + rs_packets = facet->packet_count - facet->rs_packet_count; + rs_bytes = facet->byte_count - facet->rs_byte_count; + + if (rs_packets || rs_bytes || facet->used > facet->rs_used) { + facet->rs_packet_count = facet->packet_count; + facet->rs_byte_count = facet->byte_count; + facet->rs_used = facet->used; + + flow_push_stats(ofproto, facet->rule, &facet->flow, + rs_packets, rs_bytes, facet->used); + } +} + +struct ofproto_push { + struct action_xlate_ctx ctx; + uint64_t packets; + uint64_t bytes; + long long int used; }; +static void +push_resubmit(struct action_xlate_ctx *ctx, struct rule *rule) +{ + struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx); + + if (rule) { + rule->packet_count += push->packets; + rule->byte_count += push->bytes; + rule->used = MAX(push->used, rule->used); + } +} + +/* Pushes flow statistics to the rules which 'flow' resubmits into given + * 'rule''s actions. */ +static void +flow_push_stats(struct ofproto *ofproto, const struct rule *rule, + struct flow *flow, uint64_t packets, uint64_t bytes, + long long int used) +{ + struct ofproto_push push; + + push.packets = packets; + push.bytes = bytes; + push.used = used; + + action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL); + push.ctx.resubmit_hook = push_resubmit; + ofpbuf_delete(xlate_actions(&push.ctx, rule->actions, rule->n_actions)); +} + /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT * in which no matching flow already exists in the flow table. * @@ -3994,7 +4030,6 @@ struct modify_flows_cbdata { static int modify_flow(struct ofproto *, const struct flow_mod *, struct rule *); -static void modify_flows_cb(struct cls_rule *, void *cbdata_); /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as * encoded by ofp_mkerr() on failure. @@ -4004,19 +4039,24 @@ static void modify_flows_cb(struct cls_rule *, void *cbdata_); static int modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm) { - struct modify_flows_cbdata cbdata; + struct ofproto *p = ofconn->ofproto; + struct rule *match = NULL; + struct cls_cursor cursor; + struct rule *rule; - cbdata.ofproto = ofconn->ofproto; - cbdata.fm = fm; - cbdata.match = NULL; + cls_cursor_init(&cursor, &p->cls, &fm->cr); + CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { + if (!rule_is_hidden(rule)) { + match = rule; + modify_flow(p, fm, rule); + } + } - classifier_for_each_match(&ofconn->ofproto->cls, &fm->cr, - 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(ofconn, cbdata.match, fm->buffer_id); + if (match) { + /* This credits the packet to whichever flow happened to match last. + * That's weird. Maybe we should do a lookup for the flow that + * actually matches the packet? Who knows. */ + send_buffered_packet(ofconn, match, fm->buffer_id); return 0; } else { return add_flow(ofconn, fm); @@ -4041,19 +4081,6 @@ modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm) } } -/* Callback for modify_flows_loose(). */ -static void -modify_flows_cb(struct cls_rule *rule_, void *cbdata_) -{ - struct rule *rule = rule_from_cls_rule(rule_); - struct modify_flows_cbdata *cbdata = cbdata_; - - if (!rule_is_hidden(rule)) { - cbdata->match = rule; - modify_flow(cbdata->ofproto, cbdata->fm, 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' @@ -4084,24 +4111,19 @@ modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule) /* OFPFC_DELETE implementation. */ -struct delete_flows_cbdata { - struct ofproto *ofproto; - ovs_be16 out_port; -}; - -static void delete_flows_cb(struct cls_rule *, void *cbdata_); static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port); /* Implements OFPFC_DELETE. */ static void delete_flows_loose(struct ofproto *p, const struct flow_mod *fm) { - struct delete_flows_cbdata cbdata; + struct rule *rule, *next_rule; + struct cls_cursor cursor; - cbdata.ofproto = p; - cbdata.out_port = htons(fm->out_port); - - classifier_for_each_match(&p->cls, &fm->cr, delete_flows_cb, &cbdata); + cls_cursor_init(&cursor, &p->cls, &fm->cr); + CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { + delete_flow(p, rule, htons(fm->out_port)); + } } /* Implements OFPFC_DELETE_STRICT. */ @@ -4114,16 +4136,6 @@ delete_flow_strict(struct ofproto *p, struct flow_mod *fm) } } -/* 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 @@ -4148,9 +4160,10 @@ delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port) } static int -flow_mod_core(struct ofconn *ofconn, struct flow_mod *fm) +handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh) { struct ofproto *p = ofconn->ofproto; + struct flow_mod fm; int error; error = reject_slave_controller(ofconn, "flow_mod"); @@ -4158,36 +4171,41 @@ flow_mod_core(struct ofconn *ofconn, struct flow_mod *fm) return error; } - error = validate_actions(fm->actions, fm->n_actions, - &fm->cr.flow, p->max_ports); + error = ofputil_decode_flow_mod(&fm, oh, ofconn->flow_format); if (error) { return error; } - /* We do not support the emergency flow cache. It will hopefully - * get dropped from OpenFlow in the near future. */ - if (fm->flags & OFPFF_EMERG) { + /* We do not support the emergency flow cache. It will hopefully get + * dropped from OpenFlow in the near future. */ + if (fm.flags & OFPFF_EMERG) { /* There isn't a good fit for an error code, so just state that the * flow table is full. */ return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL); } - switch (fm->command) { + error = validate_actions(fm.actions, fm.n_actions, + &fm.cr.flow, p->max_ports); + if (error) { + return error; + } + + switch (fm.command) { case OFPFC_ADD: - return add_flow(ofconn, fm); + return add_flow(ofconn, &fm); case OFPFC_MODIFY: - return modify_flows_loose(ofconn, fm); + return modify_flows_loose(ofconn, &fm); case OFPFC_MODIFY_STRICT: - return modify_flow_strict(ofconn, fm); + return modify_flow_strict(ofconn, &fm); case OFPFC_DELETE: - delete_flows_loose(p, fm); + delete_flows_loose(p, &fm); return 0; case OFPFC_DELETE_STRICT: - delete_flow_strict(p, fm); + delete_flow_strict(p, &fm); return 0; default: @@ -4196,129 +4214,23 @@ flow_mod_core(struct ofconn *ofconn, struct flow_mod *fm) } static int -handle_ofpt_flow_mod(struct ofconn *ofconn, struct ofp_header *oh) +handle_tun_id_from_cookie(struct ofconn *ofconn, const struct ofp_header *oh) { - struct ofp_match orig_match; - struct ofp_flow_mod *ofm; - struct flow_mod fm; - struct ofpbuf b; - int error; - - b.data = oh; - b.size = ntohs(oh->length); - - /* Dissect the message. */ - ofm = ofpbuf_try_pull(&b, sizeof *ofm); - if (!ofm) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); - } - error = ofputil_pull_actions(&b, b.size, &fm.actions, &fm.n_actions); - if (error) { - return error; - } - - /* Normalize ofm->match. If normalization actually changes anything, then - * log the differences. */ - ofm->match.pad1[0] = ofm->match.pad2[0] = 0; - orig_match = ofm->match; - normalize_match(&ofm->match); - if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) { - static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1); - if (!VLOG_DROP_INFO(&normal_rl)) { - char *old = ofp_match_to_literal_string(&orig_match); - char *new = ofp_match_to_literal_string(&ofm->match); - VLOG_INFO("%s: normalization changed ofp_match, details:", - rconn_get_name(ofconn->rconn)); - VLOG_INFO(" pre: %s", old); - VLOG_INFO("post: %s", new); - free(old); - free(new); - } - } - - /* Translate the message. */ - cls_rule_from_match(&ofm->match, ntohs(ofm->priority), ofconn->flow_format, - ofm->cookie, &fm.cr); - fm.cookie = ofm->cookie; - fm.command = ntohs(ofm->command); - fm.idle_timeout = ntohs(ofm->idle_timeout); - fm.hard_timeout = ntohs(ofm->hard_timeout); - fm.buffer_id = ntohl(ofm->buffer_id); - fm.out_port = ntohs(ofm->out_port); - fm.flags = ntohs(ofm->flags); - - /* Execute the command. */ - return flow_mod_core(ofconn, &fm); -} - -static int -handle_nxt_flow_mod(struct ofconn *ofconn, struct ofp_header *oh) -{ - struct nx_flow_mod *nfm; - struct flow_mod fm; - struct ofpbuf b; - int error; - - b.data = oh; - b.size = ntohs(oh->length); - - /* Dissect the message. */ - nfm = ofpbuf_try_pull(&b, sizeof *nfm); - if (!nfm) { - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); - } - error = nx_pull_match(&b, ntohs(nfm->match_len), ntohs(nfm->priority), - &fm.cr); - if (error) { - return error; - } - error = ofputil_pull_actions(&b, b.size, &fm.actions, &fm.n_actions); - if (error) { - return error; - } - - /* Translate the message. */ - fm.cookie = nfm->cookie; - fm.command = ntohs(nfm->command); - fm.idle_timeout = ntohs(nfm->idle_timeout); - fm.hard_timeout = ntohs(nfm->hard_timeout); - fm.buffer_id = ntohl(nfm->buffer_id); - fm.out_port = ntohs(nfm->out_port); - fm.flags = ntohs(nfm->flags); - - /* Execute the command. */ - return flow_mod_core(ofconn, &fm); -} - -static int -handle_tun_id_from_cookie(struct ofconn *ofconn, struct nxt_tun_id_cookie *msg) -{ - int error; - - error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg); - if (error) { - return error; - } + const struct nxt_tun_id_cookie *msg + = (const struct nxt_tun_id_cookie *) oh; ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10; return 0; } static int -handle_role_request(struct ofconn *ofconn, struct nicira_header *msg) +handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh) { - struct nx_role_request *nrr; + struct nx_role_request *nrr = (struct nx_role_request *) oh; struct nx_role_request *reply; struct ofpbuf *buf; uint32_t role; - if (ntohs(msg->header.length) != sizeof *nrr) { - VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)", - ntohs(msg->header.length), sizeof *nrr); - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); - } - nrr = (struct nx_role_request *) msg; - if (ofconn->type != OFCONN_PRIMARY) { VLOG_WARN_RL(&rl, "ignoring role request on non-controller " "connection"); @@ -4345,8 +4257,7 @@ handle_role_request(struct ofconn *ofconn, struct nicira_header *msg) } ofconn->role = role; - reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, msg->header.xid, - &buf); + reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf); reply->role = htonl(role); queue_tx(buf, ofconn, ofconn->reply_counter); @@ -4354,16 +4265,11 @@ handle_role_request(struct ofconn *ofconn, struct nicira_header *msg) } static int -handle_nxt_set_flow_format(struct ofconn *ofconn, - struct nxt_set_flow_format *msg) +handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh) { + const struct nxt_set_flow_format *msg + = (const struct nxt_set_flow_format *) oh; uint32_t format; - int error; - - error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg); - if (error) { - return error; - } format = ntohl(msg->format); if (format == NXFF_OPENFLOW10 @@ -4377,52 +4283,7 @@ handle_nxt_set_flow_format(struct ofconn *ofconn, } static int -handle_vendor(struct ofconn *ofconn, void *msg) -{ - struct ofproto *p = ofconn->ofproto; - struct ofp_vendor_header *ovh = msg; - struct nicira_header *nh; - - if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) { - VLOG_WARN_RL(&rl, "received vendor message of length %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(ofconn, msg); - - case NXT_ROLE_REQUEST: - return handle_role_request(ofconn, msg); - - case NXT_SET_FLOW_FORMAT: - return handle_nxt_set_flow_format(ofconn, msg); - - case NXT_FLOW_MOD: - return handle_nxt_flow_mod(ofconn, &ovh->header); - } - - return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); -} - -static int -handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh) +handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh) { struct ofp_header *ob; struct ofpbuf *buf; @@ -4434,94 +4295,160 @@ handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh) return 0; } -static void -handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg) +static int +handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) { - struct ofp_header *oh = ofp_msg->data; + const struct ofp_header *oh = msg->data; + const struct ofputil_msg_type *type; int error; - COVERAGE_INC(ofproto_recv_openflow); - switch (oh->type) { - case OFPT_ECHO_REQUEST: - error = handle_echo_request(ofconn, oh); - break; + error = ofputil_decode_msg_type(oh, &type); + if (error) { + return error; + } - case OFPT_ECHO_REPLY: - error = 0; - break; + switch (ofputil_msg_type_code(type)) { + /* OpenFlow requests. */ + case OFPUTIL_OFPT_ECHO_REQUEST: + return handle_echo_request(ofconn, oh); - case OFPT_FEATURES_REQUEST: - error = handle_features_request(ofconn, oh); - break; + case OFPUTIL_OFPT_FEATURES_REQUEST: + return handle_features_request(ofconn, oh); - case OFPT_GET_CONFIG_REQUEST: - error = handle_get_config_request(ofconn, oh); - break; + case OFPUTIL_OFPT_GET_CONFIG_REQUEST: + return handle_get_config_request(ofconn, oh); - case OFPT_SET_CONFIG: - error = handle_set_config(ofconn, ofp_msg->data); - break; + case OFPUTIL_OFPT_SET_CONFIG: + return handle_set_config(ofconn, msg->data); - case OFPT_PACKET_OUT: - error = handle_packet_out(ofconn, ofp_msg->data); - break; + case OFPUTIL_OFPT_PACKET_OUT: + return handle_packet_out(ofconn, oh); - case OFPT_PORT_MOD: - error = handle_port_mod(ofconn, oh); - break; + case OFPUTIL_OFPT_PORT_MOD: + return handle_port_mod(ofconn, oh); - case OFPT_FLOW_MOD: - error = handle_ofpt_flow_mod(ofconn, ofp_msg->data); - break; + case OFPUTIL_OFPT_FLOW_MOD: + return handle_flow_mod(ofconn, oh); - case OFPT_STATS_REQUEST: - error = handle_stats_request(ofconn, oh); - break; + case OFPUTIL_OFPT_BARRIER_REQUEST: + return handle_barrier_request(ofconn, oh); - case OFPT_VENDOR: - error = handle_vendor(ofconn, ofp_msg->data); - break; - - case OFPT_BARRIER_REQUEST: - error = handle_barrier_request(ofconn, oh); - break; + /* OpenFlow replies. */ + case OFPUTIL_OFPT_ECHO_REPLY: + return 0; + /* Nicira extension requests. */ + case OFPUTIL_NXT_TUN_ID_FROM_COOKIE: + return handle_tun_id_from_cookie(ofconn, oh); + + case OFPUTIL_NXT_ROLE_REQUEST: + return handle_role_request(ofconn, oh); + + case OFPUTIL_NXT_SET_FLOW_FORMAT: + return handle_nxt_set_flow_format(ofconn, oh); + + case OFPUTIL_NXT_FLOW_MOD: + return handle_flow_mod(ofconn, oh); + + /* OpenFlow statistics requests. */ + case OFPUTIL_OFPST_DESC_REQUEST: + return handle_desc_stats_request(ofconn, oh); + + case OFPUTIL_OFPST_FLOW_REQUEST: + return handle_flow_stats_request(ofconn, oh); + + case OFPUTIL_OFPST_AGGREGATE_REQUEST: + return handle_aggregate_stats_request(ofconn, oh); + + case OFPUTIL_OFPST_TABLE_REQUEST: + return handle_table_stats_request(ofconn, oh); + + case OFPUTIL_OFPST_PORT_REQUEST: + return handle_port_stats_request(ofconn, oh); + + case OFPUTIL_OFPST_QUEUE_REQUEST: + return handle_queue_stats_request(ofconn, oh); + + /* Nicira extension statistics requests. */ + case OFPUTIL_NXST_FLOW_REQUEST: + return handle_nxst_flow(ofconn, oh); + + case OFPUTIL_NXST_AGGREGATE_REQUEST: + return handle_nxst_aggregate(ofconn, oh); + + case OFPUTIL_INVALID: + case OFPUTIL_OFPT_HELLO: + case OFPUTIL_OFPT_ERROR: + case OFPUTIL_OFPT_FEATURES_REPLY: + case OFPUTIL_OFPT_GET_CONFIG_REPLY: + case OFPUTIL_OFPT_PACKET_IN: + case OFPUTIL_OFPT_FLOW_REMOVED: + case OFPUTIL_OFPT_PORT_STATUS: + case OFPUTIL_OFPT_BARRIER_REPLY: + case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST: + case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY: + case OFPUTIL_OFPST_DESC_REPLY: + case OFPUTIL_OFPST_FLOW_REPLY: + case OFPUTIL_OFPST_QUEUE_REPLY: + case OFPUTIL_OFPST_PORT_REPLY: + case OFPUTIL_OFPST_TABLE_REPLY: + case OFPUTIL_OFPST_AGGREGATE_REPLY: + case OFPUTIL_NXT_ROLE_REPLY: + case OFPUTIL_NXT_FLOW_REMOVED: + case OFPUTIL_NXST_FLOW_REPLY: + case OFPUTIL_NXST_AGGREGATE_REPLY: default: if (VLOG_IS_WARN_ENABLED()) { char *s = ofp_to_string(oh, ntohs(oh->length), 2); VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s); free(s); } - error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE); - break; + if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT); + } else { + return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE); + } } +} +static void +handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg) +{ + int error = handle_openflow__(ofconn, ofp_msg); if (error) { send_error_oh(ofconn, ofp_msg->data, error); } + COVERAGE_INC(ofproto_recv_openflow); } static void -handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) +handle_miss_upcall(struct ofproto *p, struct dpif_upcall *upcall) { - struct odp_msg *msg = packet->data; - struct ofpbuf payload; struct facet *facet; struct flow flow; - payload.data = msg + 1; - payload.size = msg->length - sizeof *msg; - flow_extract(&payload, msg->arg, msg->port, &flow); + /* Obtain in_port and tun_id, at least. */ + odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); + + /* Set header pointers in 'flow'. */ + flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow); + + if (p->ofhooks->special_cb + && !p->ofhooks->special_cb(&flow, upcall->packet, p->aux)) { + ofpbuf_delete(upcall->packet); + return; + } /* Check with in-band control to see if this packet should be sent * to the local port regardless of the flow table. */ - if (in_band_msg_in_hook(p->in_band, &flow, &payload)) { - union odp_action action; + if (in_band_msg_in_hook(p->in_band, &flow, upcall->packet)) { + struct ofpbuf odp_actions; - memset(&action, 0, sizeof(action)); - action.output.type = ODPAT_OUTPUT; - action.output.port = ODPP_LOCAL; - dpif_execute(p->dpif, &action, 1, &payload); + ofpbuf_init(&odp_actions, 32); + nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, ODPP_LOCAL); + dpif_execute(p->dpif, odp_actions.data, odp_actions.size, + upcall->packet); + ofpbuf_uninit(&odp_actions); } facet = facet_lookup_valid(p, &flow); @@ -4529,29 +4456,29 @@ handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) struct rule *rule = rule_lookup(p, &flow); if (!rule) { /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ - struct ofport *port = get_port(p, msg->port); + struct ofport *port = get_port(p, flow.in_port); if (port) { if (port->opp.config & OFPPC_NO_PACKET_IN) { COVERAGE_INC(ofproto_no_packet_in); /* XXX install 'drop' flow entry */ - ofpbuf_delete(packet); + ofpbuf_delete(upcall->packet); return; } } else { VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, - msg->port); + flow.in_port); } COVERAGE_INC(ofproto_packet_in); - send_packet_in(p, packet); + send_packet_in(p, upcall, &flow, false); return; } - facet = facet_create(p, rule, &flow, packet); + facet = facet_create(p, rule, &flow, upcall->packet); } else if (!facet->may_install) { /* The facet is not installable, that is, we need to process every * packet, so process the current packet's actions into 'facet'. */ - facet_make_actions(p, facet, packet); + facet_make_actions(p, facet, upcall->packet); } if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) { @@ -4565,54 +4492,49 @@ handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) * * See the top-level comment in fail-open.c for more information. */ - send_packet_in(p, ofpbuf_clone_with_headroom(packet, - DPIF_RECV_MSG_PADDING)); + send_packet_in(p, upcall, &flow, true); } - ofpbuf_pull(packet, sizeof *msg); - facet_execute(p, facet, packet); + facet_execute(p, facet, upcall->packet); facet_install(p, facet, false); } static void -handle_odp_msg(struct ofproto *p, struct ofpbuf *packet) +handle_upcall(struct ofproto *p, struct dpif_upcall *upcall) { - struct odp_msg *msg = packet->data; + struct flow flow; - switch (msg->type) { - case _ODPL_ACTION_NR: + switch (upcall->type) { + case DPIF_UC_ACTION: COVERAGE_INC(ofproto_ctlr_action); - send_packet_in(p, packet); + odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); + send_packet_in(p, upcall, &flow, false); break; - case _ODPL_SFLOW_NR: + case DPIF_UC_SAMPLE: if (p->sflow) { - ofproto_sflow_received(p->sflow, msg); + odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); + ofproto_sflow_received(p->sflow, upcall, &flow); } - ofpbuf_delete(packet); + ofpbuf_delete(upcall->packet); break; - case _ODPL_MISS_NR: - handle_odp_miss_msg(p, packet); + case DPIF_UC_MISS: + handle_miss_upcall(p, upcall); break; + case DPIF_N_UC_TYPES: default: - VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32, - msg->type); + VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); break; } } /* 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); +static void ofproto_update_stats(struct ofproto *); +static void rule_expire(struct ofproto *, struct rule *); static void ofproto_expire_facets(struct ofproto *, int dp_max_idle); /* This function is called periodically by ofproto_run(). Its job is to @@ -4624,18 +4546,22 @@ static void ofproto_expire_facets(struct ofproto *, int dp_max_idle); static int ofproto_expire(struct ofproto *ofproto) { - struct expire_cbdata cbdata; + struct rule *rule, *next_rule; + struct cls_cursor cursor; + int dp_max_idle; - /* Update 'used' for each flow in the datapath. */ - ofproto_update_used(ofproto); + /* Update stats for each flow in the datapath. */ + ofproto_update_stats(ofproto); /* Expire facets that have been idle too long. */ - cbdata.dp_max_idle = ofproto_dp_max_idle(ofproto); - ofproto_expire_facets(ofproto, cbdata.dp_max_idle); + dp_max_idle = ofproto_dp_max_idle(ofproto); + ofproto_expire_facets(ofproto, dp_max_idle); /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ - cbdata.ofproto = ofproto; - classifier_for_each(&ofproto->cls, rule_expire, &cbdata); + cls_cursor_init(&cursor, &ofproto->cls, NULL); + CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { + rule_expire(ofproto, rule); + } /* 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 @@ -4645,43 +4571,74 @@ ofproto_expire(struct ofproto *ofproto) ofproto->ofhooks->account_checkpoint_cb(ofproto->aux); } - return MIN(cbdata.dp_max_idle, 1000); + return MIN(dp_max_idle, 1000); } -/* Update 'used' member of installed facets. */ +/* Update 'packet_count', 'byte_count', and 'used' members of installed facets. + * + * This function also pushes statistics updates to rules which each facet + * resubmits into. Generally these statistics will be accurate. However, if a + * facet changes the rule it resubmits into at some time in between + * ofproto_update_stats() runs, it is possible that statistics accrued to the + * old rule will be incorrectly attributed to the new rule. This could be + * avoided by calling ofproto_update_stats() whenever rules are created or + * deleted. However, the performance impact of making so many calls to the + * datapath do not justify the benefit of having perfectly accurate statistics. + */ static void -ofproto_update_used(struct ofproto *p) +ofproto_update_stats(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; - } + const struct dpif_flow_stats *stats; + struct dpif_flow_dump dump; + const struct nlattr *key; + size_t key_len; - for (i = 0; i < n_flows; i++) { - struct odp_flow *f = &flows[i]; + dpif_flow_dump_start(&dump, p->dpif); + while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { struct facet *facet; struct flow flow; - odp_flow_key_to_flow(&f->key, &flow); + if (odp_flow_key_to_flow(key, key_len, &flow)) { + struct ds s; + + ds_init(&s); + odp_flow_key_format(key, key_len, &s); + VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s", + ds_cstr(&s)); + ds_destroy(&s); + + continue; + } facet = facet_find(p, &flow); if (facet && facet->installed) { - facet_update_time(p, facet, &f->stats); - facet_account(p, facet, f->stats.n_bytes); + + if (stats->n_packets >= facet->dp_packet_count) { + facet->packet_count += stats->n_packets - facet->dp_packet_count; + } else { + VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); + } + + if (stats->n_bytes >= facet->dp_byte_count) { + facet->byte_count += stats->n_bytes - facet->dp_byte_count; + } else { + VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); + } + + facet->dp_packet_count = stats->n_packets; + facet->dp_byte_count = stats->n_bytes; + + facet_update_time(p, facet, stats->used); + facet_account(p, facet, stats->n_bytes); + facet_push_stats(p, facet); } else { /* There's a flow in the datapath that we know nothing about. * Delete it. */ COVERAGE_INC(ofproto_unexpected_rule); - dpif_flow_del(p->dpif, f); + dpif_flow_del(p->dpif, key, key_len, NULL); } - } - free(flows); + dpif_flow_dump_done(&dump); } /* Calculates and returns the number of milliseconds of idle time after which @@ -4716,7 +4673,7 @@ ofproto_dp_max_idle(const struct ofproto *ofproto) * they receive additional data). * * This requires a second pass through the facets, in addition to the pass - * made by ofproto_update_used(), because the former function never looks + * made by ofproto_update_stats(), because the former function never looks * at uninstallable facets. */ enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; @@ -4780,32 +4737,19 @@ facet_active_timeout(struct ofproto *ofproto, struct facet *facet) if (ofproto->netflow && !facet_is_controller_flow(facet) && netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { struct ofexpired expired; - struct odp_flow odp_flow; - /* Get updated flow stats. - * - * 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 (facet->installed) { - odp_flow_key_from_flow(&odp_flow.key, &facet->flow); - odp_flow.flags = ODPFF_ZERO_TCP_FLAGS; - dpif_flow_get(ofproto->dpif, &odp_flow); - - if (odp_flow.stats.n_packets) { - facet_update_time(ofproto, facet, &odp_flow.stats); - netflow_flow_update_flags(&facet->nf_flow, - odp_flow.stats.tcp_flags); - } + struct dpif_flow_stats stats; + + facet_put__(ofproto, facet, facet->actions, facet->actions_len, + &stats); + facet_update_stats(ofproto, facet, &stats); } expired.flow = facet->flow; - expired.packet_count = facet->packet_count + - odp_flow.stats.n_packets; - expired.byte_count = facet->byte_count + odp_flow.stats.n_bytes; + expired.packet_count = facet->packet_count; + expired.byte_count = facet->byte_count; expired.used = facet->used; - netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); } } @@ -4824,15 +4768,11 @@ ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle) } } -/* If 'cls_rule' is an OpenFlow rule, that has expired according to OpenFlow - * rules, then delete it entirely. - * - * (This is a callback function for classifier_for_each().) */ +/* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, + * then delete it entirely. */ static void -rule_expire(struct cls_rule *cls_rule, void *cbdata_) +rule_expire(struct ofproto *ofproto, struct rule *rule) { - struct expire_cbdata *cbdata = cbdata_; - struct rule *rule = rule_from_cls_rule(cls_rule); struct facet *facet, *next_facet; long long int now; uint8_t reason; @@ -4854,14 +4794,14 @@ rule_expire(struct cls_rule *cls_rule, void *cbdata_) /* Update stats. (This is a no-op if the rule expired due to an idle * timeout, because that only happens when the rule has no facets left.) */ LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { - facet_remove(cbdata->ofproto, facet); + facet_remove(ofproto, facet); } /* Get rid of the rule. */ if (!rule_is_hidden(rule)) { - rule_send_removed(cbdata->ofproto, rule, reason); + rule_send_removed(ofproto, rule, reason); } - rule_remove(cbdata->ofproto, rule); + rule_remove(ofproto, rule); } static struct ofpbuf * @@ -4871,10 +4811,9 @@ compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule, struct ofp_flow_removed *ofr; struct ofpbuf *buf; - ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf); - flow_to_match(&rule->cr.flow, rule->cr.wc.wildcards, ofconn->flow_format, - &ofr->match); - ofr->cookie = rule->flow_cookie; + ofr = make_openflow_xid(sizeof *ofr, OFPT_FLOW_REMOVED, htonl(0), &buf); + ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofr->match, + rule->flow_cookie, &ofr->cookie); ofr->priority = htons(rule->cr.priority); ofr->reason = reason; calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec); @@ -4892,10 +4831,10 @@ compose_nx_flow_removed(const struct rule *rule, uint8_t reason) struct ofpbuf *buf; int match_len; - nfr = make_nxmsg(sizeof *nfr, NXT_FLOW_REMOVED, &buf); - + make_nxmsg_xid(sizeof *nfr, NXT_FLOW_REMOVED, htonl(0), &buf); match_len = nx_put_match(buf, &rule->cr); + nfr = buf->data; nfr->cookie = rule->flow_cookie; nfr->priority = htons(rule->cr.priority); nfr->reason = reason; @@ -4938,149 +4877,131 @@ rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason) } } -/* pinsched callback for sending 'packet' on 'ofconn'. */ +/* Obtains statistics for 'rule' and stores them in '*packets' and '*bytes'. + * The returned statistics include statistics for all of 'rule''s facets. */ +static void +rule_get_stats(const struct rule *rule, uint64_t *packets, uint64_t *bytes) +{ + uint64_t p, b; + struct facet *facet; + + /* Start from historical data for 'rule' itself that are no longer tracked + * in facets. This counts, for example, facets that have expired. */ + p = rule->packet_count; + b = rule->byte_count; + + /* Add any statistics that are tracked by facets. This includes + * statistical data recently updated by ofproto_update_stats() as well as + * stats for packets that were executed "by hand" via dpif_execute(). */ + LIST_FOR_EACH (facet, list_node, &rule->facets) { + p += facet->packet_count; + b += facet->byte_count; + } + + *packets = p; + *bytes = b; +} + +/* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */ static void -do_send_packet_in(struct ofpbuf *packet, void *ofconn_) +do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn_) { struct ofconn *ofconn = ofconn_; - rconn_send_with_limit(ofconn->rconn, packet, + rconn_send_with_limit(ofconn->rconn, ofp_packet_in, 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). +/* Takes 'upcall', whose packet has the flow specified by 'flow', composes an + * OpenFlow packet-in message from it, and passes it to 'ofconn''s packet + * scheduler for sending. * - * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise, - * ownership is transferred to this function. */ + * If 'clone' is true, the caller retains ownership of 'upcall->packet'. + * Otherwise, ownership is transferred to this function. */ static void -schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len, - bool clone) +schedule_packet_in(struct ofconn *ofconn, struct dpif_upcall *upcall, + const struct flow *flow, bool clone) { + enum { OPI_SIZE = offsetof(struct ofp_packet_in, data) }; 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; + struct ofp_packet_in *opi; + int total_len, send_len; + struct ofpbuf *packet; uint32_t buffer_id; + int idx; - /* Get buffer. */ - if (opi->reason == OFPR_ACTION) { + /* Get OpenFlow buffer_id. */ + if (upcall->type == DPIF_UC_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); + buffer_id = pktbuf_save(ofconn->pktbuf, upcall->packet, flow->in_port); } /* Figure out how much of the packet to send. */ - send_len = ntohs(opi->total_len); + total_len = send_len = upcall->packet->size; if (buffer_id != UINT32_MAX) { send_len = MIN(send_len, ofconn->miss_send_len); } - send_len = MIN(send_len, max_len); + if (upcall->type == DPIF_UC_ACTION) { + send_len = MIN(send_len, upcall->userdata); + } - /* Adjust packet length and clone if necessary. */ - trim_size = offsetof(struct ofp_packet_in, data) + send_len; + /* Copy or steal buffer for OFPT_PACKET_IN. */ if (clone) { - packet = ofpbuf_clone_data(packet->data, trim_size); - opi = packet->data; + packet = ofpbuf_clone_data_with_headroom(upcall->packet->data, + send_len, OPI_SIZE); } else { - packet->size = trim_size; + packet = upcall->packet; + packet->size = send_len; } - /* Update packet headers. */ + /* Add OFPT_PACKET_IN. */ + opi = ofpbuf_push_zeros(packet, OPI_SIZE); + opi->header.version = OFP_VERSION; + opi->header.type = OFPT_PACKET_IN; + opi->total_len = htons(total_len); + opi->in_port = htons(odp_port_to_ofp_port(flow->in_port)); + opi->reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION; opi->buffer_id = htonl(buffer_id); update_openflow_length(packet); /* Hand over to packet scheduler. It might immediately call into * do_send_packet_in() or it might buffer it for a while (until a later * call to pinsched_run()). */ - pinsched_send(ofconn->schedulers[opi->reason], in_port, + idx = upcall->type == DPIF_UC_MISS ? 0 : 1; + pinsched_send(ofconn->schedulers[idx], flow->in_port, packet, do_send_packet_in, ofconn); } -/* Replace struct odp_msg header in 'packet' by equivalent struct - * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as - * returned by dpif_recv()). - * - * The conversion is not complete: the caller still needs to trim any unneeded - * payload off the end of the buffer, set the length in the OpenFlow header, - * and set buffer_id. Those require us to know the controller settings and so - * must be done on a per-controller basis. +/* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an + * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to + * their individual configurations. * - * Returns the maximum number of bytes of the packet that should be sent to - * the controller (INT_MAX if no limit). */ -static int -do_convert_to_packet_in(struct ofpbuf *packet) -{ - struct odp_msg *msg = packet->data; - struct ofp_packet_in *opi; - uint8_t reason; - uint16_t total_len; - uint16_t in_port; - int max_len; - - /* Extract relevant header fields */ - if (msg->type == _ODPL_ACTION_NR) { - reason = OFPR_ACTION; - max_len = msg->arg; - } else { - reason = OFPR_NO_MATCH; - max_len = INT_MAX; - } - total_len = msg->length - sizeof *msg; - in_port = odp_port_to_ofp_port(msg->port); - - /* Repurpose packet buffer by overwriting header. */ - ofpbuf_pull(packet, sizeof(struct odp_msg)); - opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data)); - opi->header.version = OFP_VERSION; - opi->header.type = OFPT_PACKET_IN; - opi->total_len = htons(total_len); - opi->in_port = htons(in_port); - opi->reason = reason; - - return max_len; -} - -/* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or - * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller - * as necessary according to their individual configurations. - * - * 'packet' must have sufficient headroom to convert it into a struct - * ofp_packet_in (e.g. as returned by dpif_recv()). - * - * Takes ownership of 'packet'. */ + * If 'clone' is true, the caller retains ownership of 'upcall->packet'. + * Otherwise, ownership is transferred to this function. */ static void -send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet) +send_packet_in(struct ofproto *ofproto, struct dpif_upcall *upcall, + const struct flow *flow, bool clone) { struct ofconn *ofconn, *prev; - int max_len; - - max_len = do_convert_to_packet_in(packet); prev = NULL; LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { if (ofconn_receives_async_msgs(ofconn)) { if (prev) { - schedule_packet_in(prev, packet, max_len, true); + schedule_packet_in(prev, upcall, flow, true); } prev = ofconn; } } if (prev) { - schedule_packet_in(prev, packet, max_len, false); - } else { - ofpbuf_delete(packet); + schedule_packet_in(prev, upcall, flow, clone); + } else if (!clone) { + ofpbuf_delete(upcall->packet); } } @@ -5112,9 +5033,177 @@ pick_fallback_dpid(void) return eth_addr_to_uint64(ea); } +static void +ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED, + void *aux OVS_UNUSED) +{ + const struct shash_node *node; + struct ds results; + + ds_init(&results); + SHASH_FOR_EACH (node, &all_ofprotos) { + ds_put_format(&results, "%s\n", node->name); + } + unixctl_command_reply(conn, 200, ds_cstr(&results)); + ds_destroy(&results); +} + +struct ofproto_trace { + struct action_xlate_ctx ctx; + struct flow flow; + struct ds *result; +}; + +static void +trace_format_rule(struct ds *result, int level, const struct rule *rule) +{ + ds_put_char_multiple(result, '\t', level); + if (!rule) { + ds_put_cstr(result, "No match\n"); + return; + } + + ds_put_format(result, "Rule: cookie=%#"PRIx64" ", + ntohll(rule->flow_cookie)); + cls_rule_format(&rule->cr, result); + ds_put_char(result, '\n'); + + ds_put_char_multiple(result, '\t', level); + ds_put_cstr(result, "OpenFlow "); + ofp_print_actions(result, (const struct ofp_action_header *) rule->actions, + rule->n_actions * sizeof *rule->actions); + ds_put_char(result, '\n'); +} + +static void +trace_format_flow(struct ds *result, int level, const char *title, + struct ofproto_trace *trace) +{ + ds_put_char_multiple(result, '\t', level); + ds_put_format(result, "%s: ", title); + if (flow_equal(&trace->ctx.flow, &trace->flow)) { + ds_put_cstr(result, "unchanged"); + } else { + flow_format(result, &trace->ctx.flow); + trace->flow = trace->ctx.flow; + } + ds_put_char(result, '\n'); +} + +static void +trace_resubmit(struct action_xlate_ctx *ctx, struct rule *rule) +{ + struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx); + struct ds *result = trace->result; + + ds_put_char(result, '\n'); + trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); + trace_format_rule(result, ctx->recurse + 1, rule); +} + +static void +ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_, + void *aux OVS_UNUSED) +{ + char *dpname, *in_port_s, *tun_id_s, *packet_s; + char *args = xstrdup(args_); + char *save_ptr = NULL; + struct ofproto *ofproto; + struct ofpbuf packet; + struct rule *rule; + struct ds result; + struct flow flow; + uint16_t in_port; + ovs_be64 tun_id; + char *s; + + ofpbuf_init(&packet, strlen(args) / 2); + ds_init(&result); + + dpname = strtok_r(args, " ", &save_ptr); + tun_id_s = strtok_r(NULL, " ", &save_ptr); + in_port_s = strtok_r(NULL, " ", &save_ptr); + packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */ + if (!dpname || !in_port_s || !packet_s) { + unixctl_command_reply(conn, 501, "Bad command syntax"); + goto exit; + } + + ofproto = shash_find_data(&all_ofprotos, dpname); + if (!ofproto) { + unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list " + "for help)"); + goto exit; + } + + tun_id = htonll(strtoull(tun_id_s, NULL, 10)); + in_port = ofp_port_to_odp_port(atoi(in_port_s)); + + packet_s = ofpbuf_put_hex(&packet, packet_s, NULL); + packet_s += strspn(packet_s, " "); + if (*packet_s != '\0') { + unixctl_command_reply(conn, 501, "Trailing garbage in command"); + goto exit; + } + if (packet.size < ETH_HEADER_LEN) { + unixctl_command_reply(conn, 501, "Packet data too short for Ethernet"); + goto exit; + } + + ds_put_cstr(&result, "Packet: "); + s = ofp_packet_to_string(packet.data, packet.size, packet.size); + ds_put_cstr(&result, s); + free(s); + + flow_extract(&packet, tun_id, in_port, &flow); + ds_put_cstr(&result, "Flow: "); + flow_format(&result, &flow); + ds_put_char(&result, '\n'); + + rule = rule_lookup(ofproto, &flow); + trace_format_rule(&result, 0, rule); + if (rule) { + struct ofproto_trace trace; + struct ofpbuf *odp_actions; + + trace.result = &result; + trace.flow = flow; + action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet); + trace.ctx.resubmit_hook = trace_resubmit; + odp_actions = xlate_actions(&trace.ctx, + rule->actions, rule->n_actions); + + ds_put_char(&result, '\n'); + trace_format_flow(&result, 0, "Final flow", &trace); + ds_put_cstr(&result, "Datapath actions: "); + format_odp_actions(&result, odp_actions->data, odp_actions->size); + ofpbuf_delete(odp_actions); + } + + unixctl_command_reply(conn, 200, ds_cstr(&result)); + +exit: + ds_destroy(&result); + ofpbuf_uninit(&packet); + free(args); +} + +static void +ofproto_unixctl_init(void) +{ + static bool registered; + if (registered) { + return; + } + registered = true; + + unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL); + unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL); +} + static bool default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, - struct odp_actions *actions, tag_type *tags, + struct ofpbuf *odp_actions, tag_type *tags, uint16_t *nf_output_iface, void *ofproto_) { struct ofproto *ofproto = ofproto_; @@ -5145,9 +5234,9 @@ default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, NULL); if (out_port < 0) { flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD, - nf_output_iface, actions); + nf_output_iface, odp_actions); } else if (out_port != flow->in_port) { - odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port; + nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_OUTPUT, out_port); *nf_output_iface = out_port; } else { /* Drop. */ @@ -5159,5 +5248,6 @@ default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, static const struct ofhooks default_ofhooks = { default_normal_ofhook_cb, NULL, + NULL, NULL };