X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=ofproto%2Fofproto.c;h=39d345736375efa04bf1a28a484507668ba0c79c;hb=a4a2643672e4d2f68f8246e406749009a3f3211b;hp=d5f258ffe8c879a53c1287b3cf1a2ca495e1b5c5;hpb=bbc32a883f790a331b412eecc6de4b7cdf1e63cb;p=openvswitch diff --git a/ofproto/ofproto.c b/ofproto/ofproto.c index d5f258ff..39d34573 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" @@ -57,6 +59,7 @@ #include "svec.h" #include "tag.h" #include "timeval.h" +#include "unaligned.h" #include "unixctl.h" #include "vconn.h" #include "vlog.h" @@ -90,6 +93,8 @@ COVERAGE_DEFINE(ofproto_update_port); #include "sflow_api.h" +struct rule; + struct ofport { struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */ struct netdev *netdev; @@ -100,11 +105,53 @@ 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 *, const 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 { @@ -165,7 +212,8 @@ 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(). @@ -185,8 +233,8 @@ 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. */ }; @@ -207,7 +255,7 @@ 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 *); /* ofproto supports two kinds of OpenFlow connections: * @@ -259,7 +307,8 @@ struct ofconn { /* 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. */ @@ -277,15 +326,6 @@ struct ofconn { 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); @@ -299,8 +339,9 @@ 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. */ @@ -359,6 +400,9 @@ 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; @@ -368,7 +412,7 @@ static uint64_t pick_fallback_dpid(void); static int ofproto_expire(struct ofproto *); -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 *); @@ -377,32 +421,31 @@ 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)); @@ -427,7 +470,7 @@ 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); @@ -470,6 +513,8 @@ 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; } @@ -991,6 +1036,8 @@ 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; @@ -1121,9 +1168,9 @@ 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 @@ -1137,7 +1184,7 @@ ofproto_run1(struct ofproto *p) break; } - handle_odp_msg(p, buf); + handle_upcall(p, &packet); } while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) { @@ -1302,6 +1349,57 @@ 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; + 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)); + + cinfo->pairs.keys[cinfo->pairs.n] = "time_in_state"; + cinfo->pairs.values[cinfo->pairs.n++] = + xasprintf("%u", rconn_get_state_elapsed(rconn)); + } +} + +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 @@ -1349,18 +1447,20 @@ 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; } @@ -1429,33 +1529,30 @@ 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, 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; @@ -1464,27 +1561,25 @@ 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.type = odp_port->type; + 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; @@ -1498,15 +1593,15 @@ 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; @@ -1607,7 +1702,7 @@ 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; @@ -1615,7 +1710,7 @@ 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); @@ -1631,20 +1726,20 @@ 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 by the kernel. So transfer the @@ -1655,7 +1750,7 @@ update_port(struct ofproto *p, const char *devname) if (ofport_equal(old_ofport, new_ofport)) { /* False alarm--no change. */ ofport_free(new_ofport); - return; + goto exit; } } @@ -1671,31 +1766,26 @@ 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; - - error = dpif_port_list(p->dpif, &ports, &n_ports); - if (error) { - return error; - } + struct dpif_port_dump dump; + struct dpif_port dpif_port; - 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; } @@ -1998,30 +2088,33 @@ 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; } @@ -2043,13 +2136,13 @@ 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)) { + 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, @@ -2069,8 +2162,9 @@ 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; @@ -2096,18 +2190,16 @@ 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; } + ofpbuf_delete(odp_actions); } /* Inserts 'rule' into 'p''s flow table. */ @@ -2195,33 +2287,45 @@ 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; + } + + 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 @@ -2230,17 +2334,12 @@ 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; } } @@ -2256,7 +2355,7 @@ 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; } @@ -2267,14 +2366,16 @@ 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; } @@ -2375,10 +2476,9 @@ 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); @@ -2393,31 +2493,24 @@ 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); } @@ -2428,11 +2521,13 @@ 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); @@ -2442,6 +2537,8 @@ facet_revalidate(struct ofproto *ofproto, struct facet *facet) facet->used = new_rule->created; } + ofpbuf_delete(odp_actions); + return true; } @@ -2459,7 +2556,7 @@ 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); @@ -2565,33 +2662,9 @@ handle_set_config(struct ofconn *ofconn, const 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 +#define MAX_RESUBMIT_RECURSION 16 static void do_xlate_actions(const union ofp_action *in, size_t n_in, struct action_xlate_ctx *ctx); @@ -2614,7 +2687,7 @@ 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; } @@ -2639,13 +2712,17 @@ 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); @@ -2654,14 +2731,14 @@ 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; @@ -2685,7 +2762,7 @@ 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); @@ -2694,14 +2771,14 @@ 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); @@ -2738,9 +2815,18 @@ 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; } } @@ -2770,10 +2856,9 @@ 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) { @@ -2799,8 +2884,7 @@ 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 @@ -2808,24 +2892,37 @@ xlate_set_dl_tci(struct action_xlate_ctx *ctx) { ovs_be16 tci = ctx->flow.vlan_tci; if (!(tci & htons(VLAN_CFI))) { - odp_actions_add(ctx->out, ODPAT_STRIP_VLAN); + 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 = tci & ~htons(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_tci = ctx->flow.vlan_tci; - - nxm_execute_reg_move(narm, &ctx->flow); + state->vlan_tci = ctx->flow.vlan_tci; + state->tun_id = ctx->flow.tun_id; +} - if (ctx->flow.vlan_tci != old_tci) { +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 @@ -2835,8 +2932,10 @@ 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) { @@ -2847,13 +2946,15 @@ 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; @@ -2863,26 +2964,44 @@ 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; } } @@ -2904,8 +3023,8 @@ 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: @@ -2931,44 +3050,47 @@ do_xlate_actions(const union ofp_action *in, size_t n_in, 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: @@ -2980,55 +3102,56 @@ 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 @@ -3057,7 +3180,8 @@ handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh) 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; @@ -3104,12 +3228,10 @@ handle_packet_out(struct ofconn *ofconn, const 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); @@ -3215,7 +3337,7 @@ 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; } @@ -3279,8 +3401,8 @@ 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; @@ -3301,18 +3423,18 @@ 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 @@ -3347,42 +3469,33 @@ static void query_stats(struct ofproto *p, struct rule *rule, uint64_t *packet_countp, uint64_t *byte_countp) { + uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S]; uint64_t packet_count, byte_count; struct facet *facet; - struct odp_flow *odp_flows; - size_t n_odp_flows; + struct ofpbuf key; /* 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. + /* 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; + ofpbuf_use_stack(&key, keybuf, sizeof keybuf); 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; - } + struct dpif_flow_stats stats; - /* 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; + ofpbuf_clear(&key); + odp_flow_key_from_flow(&key, &facet->flow); + dpif_flow_get(p->dpif, key.data, key.size, NULL, &stats); - 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; - } + packet_count += stats.n_packets + facet->packet_count; + byte_count += stats.n_bytes + facet->byte_count; } - free(odp_flows); /* Return the stats to the caller. */ *packet_countp = packet_count; @@ -3403,6 +3516,7 @@ put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule, { 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, out_port)) { @@ -3418,15 +3532,16 @@ put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule, ofs->length = htons(len); ofs->table_id = 0; ofs->pad = 0; - ofputil_cls_rule_to_match(&rule->cr, 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); } @@ -3480,8 +3595,8 @@ put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule, act_len = sizeof *rule->actions * rule->n_actions; - start_len = (*replyp)->size; 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); @@ -3542,19 +3657,17 @@ handle_nxst_flow(struct ofconn *ofconn, const struct ofp_header *oh) static void flow_stats_ds(struct ofproto *ofproto, struct rule *rule, struct ds *results) { - struct ofp_match match; uint64_t packet_count, byte_count; size_t act_len = sizeof *rule->actions * rule->n_actions; query_stats(ofproto, rule, &packet_count, &byte_count); - ofputil_cls_rule_to_match(&rule->cr, NXFF_OPENFLOW10, &match); ds_put_format(results, "duration=%llds, ", (time_msec() - rule->created) / 1000); ds_put_format(results, "priority=%u, ", rule->cr.priority); ds_put_format(results, "n_packets=%"PRIu64", ", packet_count); ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count); - ofp_print_match(results, &match, true); + cls_rule_format(&rule->cr, results); if (act_len > 0) { ofp_print_actions(results, &rule->actions->header, act_len); } else { @@ -3608,8 +3721,8 @@ query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target, } oasr->flow_count = htonl(n_flows); - oasr->packet_count = htonll(total_packets); - oasr->byte_count = htonll(total_bytes); + 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); } @@ -3682,9 +3795,9 @@ 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 @@ -3754,17 +3867,11 @@ handle_queue_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) return 0; } -static long long int -msec_from_nsec(uint64_t sec, uint32_t nsec) -{ - return !sec ? 0 : sec * 1000 + nsec / 1000000; -} - static void facet_update_time(struct ofproto *ofproto, struct facet *facet, - const struct odp_flow_stats *stats) + const struct dpif_flow_stats *stats) { - long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec); + long long int used = stats->used; if (used > facet->used) { facet->used = used; if (used > facet->rule->used) { @@ -3782,7 +3889,7 @@ 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); @@ -4269,30 +4376,33 @@ handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg) } 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; - ofpbuf_use_const(&payload, msg + 1, 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); - packet->l2 = payload.l2; - packet->l3 = payload.l3; - packet->l4 = payload.l4; - packet->l7 = payload.l7; + /* 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); @@ -4300,29 +4410,29 @@ 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) { @@ -4336,40 +4446,40 @@ 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; } } @@ -4422,36 +4532,40 @@ ofproto_expire(struct ofproto *ofproto) static void ofproto_update_used(struct ofproto *p) { - struct odp_flow *flows; - size_t n_flows; - size_t i; - int error; + const struct dpif_flow_stats *stats; + struct dpif_flow_dump dump; + const struct nlattr *key; + size_t key_len; - error = dpif_flow_list_all(p->dpif, &flows, &n_flows); - if (error) { - return; - } - - for (i = 0; i < n_flows; i++) { - struct odp_flow *f = &flows[i]; + 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); + facet_update_time(p, facet, stats); + facet_account(p, facet, stats->n_bytes); } else { /* There's a flow in the datapath that we know nothing about. * Delete it. */ COVERAGE_INC(ofproto_unexpected_rule); - dpif_flow_del(p->dpif, f); + 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 @@ -4550,32 +4664,19 @@ 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); } } @@ -4637,9 +4738,9 @@ 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); - ofputil_cls_rule_to_match(&rule->cr, 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); @@ -4657,10 +4758,10 @@ 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; @@ -4703,150 +4804,105 @@ rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason) } } -/* pinsched callback for sending 'packet' on 'ofconn'. */ +/* 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. +/* 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. * - * 'max_len' specifies the maximum number of bytes of the packet to send on - * 'ofconn' (INT_MAX specifies no limit). - * - * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise, - * ownership is transferred to this function. */ + * 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; - - ofpbuf_use_const(&payload, opi->data, - 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. - * - * 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()). +/* 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. * * Takes ownership of 'packet'. */ 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); } } @@ -4878,9 +4934,177 @@ 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, const struct rule *rule) +{ + struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx); + struct ds *result = trace->result; + + ds_put_char(result, '\n'); + trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); + trace_format_rule(result, ctx->recurse + 1, rule); +} + +static void +ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_, + void *aux OVS_UNUSED) +{ + char *dpname, *in_port_s, *tun_id_s, *packet_s; + char *args = xstrdup(args_); + char *save_ptr = NULL; + struct ofproto *ofproto; + struct ofpbuf packet; + struct rule *rule; + struct ds result; + struct flow flow; + uint16_t in_port; + ovs_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_; @@ -4911,9 +5135,9 @@ 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. */ @@ -4925,5 +5149,6 @@ 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 };