X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;ds=sidebyside;f=ofproto%2Fofproto-dpif.c;h=3074881eaca801f0367795ab49358d53c4c36bd6;hb=64624b3cf405d0ca089fe978898896087337cb90;hp=cdc21bca65c12516403e52c589a97ec25fb6f2cd;hpb=084f529087b7d3a3b3e0df470ad4ad39800cbab5;p=openvswitch diff --git a/ofproto/ofproto-dpif.c b/ofproto/ofproto-dpif.c index cdc21bca..3074881e 100644 --- a/ofproto/ofproto-dpif.c +++ b/ofproto/ofproto-dpif.c @@ -60,7 +60,7 @@ COVERAGE_DEFINE(facet_invalidated); COVERAGE_DEFINE(facet_revalidate); COVERAGE_DEFINE(facet_unexpected); -/* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a +/* Maximum depth of flow table recursion (due to resubmit actions) in a * flow translation. */ #define MAX_RESUBMIT_RECURSION 16 @@ -96,8 +96,8 @@ static struct rule_dpif *rule_dpif_cast(const struct rule *rule) return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL; } -static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *ofproto, - const struct flow *flow); +static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *, + const struct flow *, uint8_t table); #define MAX_MIRRORS 32 typedef uint32_t mirror_mask_t; @@ -188,6 +188,7 @@ struct action_xlate_ctx { uint32_t priority; /* Current flow priority. 0 if none. */ struct flow base_flow; /* Flow at the last commit. */ uint32_t base_priority; /* Priority at the last commit. */ + uint8_t table_id; /* OpenFlow table ID where flow was found. */ }; static void action_xlate_ctx_init(struct action_xlate_ctx *, @@ -206,12 +207,8 @@ struct facet { * dpif_execute(). * * - 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 + * when its 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(). */ uint64_t packet_count; /* Number of packets received. */ uint64_t byte_count; /* Number of bytes received. */ @@ -223,10 +220,7 @@ struct facet { 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). */ - uint64_t accounted_bytes; + uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ struct list list_node; /* In owning rule's 'facets' list. */ @@ -268,10 +262,10 @@ static void facet_update_time(struct ofproto_dpif *, struct facet *, long long int used); static void facet_update_stats(struct ofproto_dpif *, struct facet *, const struct dpif_flow_stats *); +static void facet_reset_counters(struct facet *); static void facet_reset_dp_stats(struct facet *, struct dpif_flow_stats *); static void facet_push_stats(struct facet *); -static void facet_account(struct ofproto_dpif *, struct facet *, - uint64_t extra_bytes); +static void facet_account(struct ofproto_dpif *, struct facet *); static bool facet_is_controller_flow(struct facet *); @@ -419,7 +413,7 @@ dealloc(struct ofproto *ofproto_) } static int -construct(struct ofproto *ofproto_) +construct(struct ofproto *ofproto_, int *n_tablesp) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); const char *name = ofproto->up.name; @@ -464,14 +458,11 @@ construct(struct ofproto *ofproto_) list_init(&ofproto->completions); - ofproto->up.tables = xmalloc(sizeof *ofproto->up.tables); - classifier_init(&ofproto->up.tables[0]); - ofproto->up.n_tables = 1; - ofproto_dpif_unixctl_init(); ofproto->has_bundle_action = false; + *n_tablesp = 255; return 0; } @@ -492,14 +483,18 @@ destruct(struct ofproto *ofproto_) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); struct rule_dpif *rule, *next_rule; - struct cls_cursor cursor; + struct classifier *table; int i; complete_operations(ofproto); - cls_cursor_init(&cursor, &ofproto->up.tables[0], NULL); - CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { - ofproto_rule_destroy(&rule->up); + OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { + struct cls_cursor cursor; + + cls_cursor_init(&cursor, table, NULL); + CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { + ofproto_rule_destroy(&rule->up); + } } for (i = 0; i < MAX_MIRRORS; i++) { @@ -564,6 +559,8 @@ run(struct ofproto *ofproto_) bundle_run(bundle); } + mac_learning_run(ofproto->ml, &ofproto->revalidate_set); + /* Now revalidate if there's anything to do. */ if (ofproto->need_revalidate || !tag_set_is_empty(&ofproto->revalidate_set)) { @@ -611,6 +608,7 @@ wait(struct ofproto *ofproto_) HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { bundle_wait(bundle); } + mac_learning_wait(ofproto->ml); if (ofproto->need_revalidate) { /* Shouldn't happen, but if it does just go around again. */ VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); @@ -662,7 +660,7 @@ static void get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - struct odp_stats s; + struct ovs_dp_stats s; strcpy(ots->name, "classifier"); @@ -815,6 +813,20 @@ get_cfm_fault(const struct ofport *ofport_) return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1; } + +static int +get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps, + size_t *n_rmps) +{ + struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); + + if (ofport->cfm) { + cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps); + return 0; + } else { + return -1; + } +} /* Bundles. */ @@ -1113,7 +1125,7 @@ bundle_remove(struct ofport *port_) } static void -send_pdu_cb(void *port_, const struct lacp_pdu *pdu) +send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) { static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); struct ofport_dpif *port = port_; @@ -1122,13 +1134,14 @@ send_pdu_cb(void *port_, const struct lacp_pdu *pdu) error = netdev_get_etheraddr(port->up.netdev, ea); if (!error) { - struct lacp_pdu *packet_pdu; struct ofpbuf packet; + void *packet_pdu; ofpbuf_init(&packet, 0); packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, - sizeof *packet_pdu); - *packet_pdu = *pdu; + pdu_size); + memcpy(packet_pdu, pdu, pdu_size); + error = netdev_send(port->up.netdev, &packet); if (error) { VLOG_WARN_RL(&rl, "port %s: sending LACP PDU on iface %s failed " @@ -1393,6 +1406,14 @@ is_mirror_output_bundle(struct ofproto *ofproto_, void *aux) struct ofbundle *bundle = bundle_lookup(ofproto, aux); return bundle && bundle->mirror_out != 0; } + +static void +forward_bpdu_changed(struct ofproto *ofproto_) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + /* Revalidate cached flows whenever forward_bpdu option changes. */ + ofproto->need_revalidate = true; +} /* Ports. */ @@ -1607,19 +1628,21 @@ static bool process_special(struct ofproto_dpif *ofproto, const struct flow *flow, const struct ofpbuf *packet) { - if (cfm_should_process_flow(flow)) { - struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port); - if (packet && ofport && ofport->cfm) { + struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port); + + if (!ofport) { + return false; + } + + if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) { + if (packet) { cfm_process_heartbeat(ofport->cfm, packet); } return true; - } else if (flow->dl_type == htons(ETH_TYPE_LACP)) { - struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port); - if (packet && port && port->bundle && port->bundle->lacp) { - const struct lacp_pdu *pdu = parse_lacp_packet(packet); - if (pdu) { - lacp_process_pdu(port->bundle->lacp, port, pdu); - } + } else if (ofport->bundle && ofport->bundle->lacp + && flow->dl_type == htons(ETH_TYPE_LACP)) { + if (packet) { + lacp_process_packet(ofport->bundle->lacp, ofport, packet); } return true; } @@ -1648,12 +1671,12 @@ handle_miss_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall) /* Check with in-band control to see if this packet should be sent * to the local port regardless of the flow table. */ if (connmgr_msg_in_hook(ofproto->up.connmgr, &flow, upcall->packet)) { - send_packet(ofproto, ODPP_LOCAL, upcall->packet); + send_packet(ofproto, OVSP_LOCAL, upcall->packet); } facet = facet_lookup_valid(ofproto, &flow); if (!facet) { - struct rule_dpif *rule = rule_dpif_lookup(ofproto, &flow); + struct rule_dpif *rule = rule_dpif_lookup(ofproto, &flow, 0); if (!rule) { /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ struct ofport_dpif *port = get_ofp_port(ofproto, flow.in_port); @@ -1747,7 +1770,7 @@ static int expire(struct ofproto_dpif *ofproto) { struct rule_dpif *rule, *next_rule; - struct cls_cursor cursor; + struct classifier *table; int dp_max_idle; /* Update stats for each flow in the datapath. */ @@ -1758,9 +1781,13 @@ expire(struct ofproto_dpif *ofproto) expire_facets(ofproto, dp_max_idle); /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ - cls_cursor_init(&cursor, &ofproto->up.tables[0], NULL); - CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { - rule_expire(rule); + OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { + struct cls_cursor cursor; + + cls_cursor_init(&cursor, table, NULL); + CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { + rule_expire(rule); + } } /* All outstanding data in existing flows has been accounted, so it's a @@ -1807,7 +1834,7 @@ update_stats(struct ofproto_dpif *p) ds_init(&s); odp_flow_key_format(key, key_len, &s); - VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s", + VLOG_WARN_RL(&rl, "failed to convert datapath flow key to flow: %s", ds_cstr(&s)); ds_destroy(&s); @@ -1834,7 +1861,7 @@ update_stats(struct ofproto_dpif *p) facet->dp_byte_count = stats->n_bytes; facet_update_time(p, facet, stats->used); - facet_account(p, facet, stats->n_bytes); + facet_account(p, facet); facet_push_stats(facet); } else { /* There's a flow in the datapath that we know nothing about. @@ -2049,7 +2076,7 @@ execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, struct ofpbuf *packet) { if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t)) - && odp_actions->nla_type == ODP_ACTION_ATTR_USERSPACE) { + && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE) { /* 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. */ @@ -2091,8 +2118,8 @@ execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, * applying flow_extract() to 'packet' would yield the same flow as * 'facet->flow'. * - * 'facet' must have accurately composed ODP actions; that is, it must not be - * in need of revalidation. + * 'facet' must have accurately composed datapath actions; that is, it must + * not be in need of revalidation. * * Takes ownership of 'packet'. */ static void @@ -2128,7 +2155,7 @@ facet_remove(struct ofproto_dpif *ofproto, struct facet *facet) facet_free(facet); } -/* Composes the ODP actions for 'facet' based on its rule's actions. */ +/* Composes the datapath actions for 'facet' based on its rule's actions. */ static void facet_make_actions(struct ofproto_dpif *p, struct facet *facet, const struct ofpbuf *packet) @@ -2209,10 +2236,9 @@ vlan_tci_to_openflow_vlan(ovs_be16 vlan_tci) } static void -facet_account(struct ofproto_dpif *ofproto, - struct facet *facet, uint64_t extra_bytes) +facet_account(struct ofproto_dpif *ofproto, struct facet *facet) { - uint64_t total_bytes, n_bytes; + uint64_t n_bytes; struct ofbundle *in_bundle; const struct nlattr *a; tag_type dummy = 0; @@ -2220,12 +2246,11 @@ facet_account(struct ofproto_dpif *ofproto, ovs_be16 vlan_tci; int vlan; - total_bytes = facet->byte_count + extra_bytes; - if (total_bytes <= facet->accounted_bytes) { + if (facet->byte_count <= facet->accounted_bytes) { return; } - n_bytes = total_bytes - facet->accounted_bytes; - facet->accounted_bytes = total_bytes; + n_bytes = facet->byte_count - facet->accounted_bytes; + facet->accounted_bytes = facet->byte_count; /* Test that 'tags' is nonzero to ensure that only flows that include an * OFPP_NORMAL action are used for learning and bond slave rebalancing. @@ -2256,7 +2281,7 @@ facet_account(struct ofproto_dpif *ofproto, struct ofport_dpif *port; switch (nl_attr_type(a)) { - case ODP_ACTION_ATTR_OUTPUT: + case OVS_ACTION_ATTR_OUTPUT: port = get_odp_port(ofproto, nl_attr_get_u32(a)); if (port && port->bundle && port->bundle->bond) { bond_account(port->bundle->bond, &facet->flow, @@ -2264,11 +2289,11 @@ facet_account(struct ofproto_dpif *ofproto, } break; - case ODP_ACTION_ATTR_STRIP_VLAN: + case OVS_ACTION_ATTR_POP_VLAN: vlan_tci = htons(0); break; - case ODP_ACTION_ATTR_SET_DL_TCI: + case OVS_ACTION_ATTR_PUSH_VLAN: vlan_tci = nl_attr_get_be16(a); break; } @@ -2341,7 +2366,7 @@ facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet) assert(!facet->dp_packet_count); facet_push_stats(facet); - facet_account(ofproto, facet, 0); + facet_account(ofproto, facet); if (ofproto->netflow && !facet_is_controller_flow(facet)) { struct ofexpired expired; @@ -2357,11 +2382,7 @@ facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet) /* Reset counters to prevent double counting if 'facet' ever gets * reinstalled. */ - facet->packet_count = 0; - facet->byte_count = 0; - facet->rs_packet_count = 0; - facet->rs_byte_count = 0; - facet->accounted_bytes = 0; + facet_reset_counters(facet); netflow_flow_clear(&facet->nf_flow); } @@ -2429,14 +2450,14 @@ facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet) COVERAGE_INC(facet_revalidate); /* Determine the new rule. */ - new_rule = rule_dpif_lookup(ofproto, &facet->flow); + new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0); if (!new_rule) { /* No new rule, so delete the facet. */ facet_remove(ofproto, facet); return false; } - /* Calculate new ODP actions. + /* Calculate new datapath actions. * * 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 @@ -2448,8 +2469,8 @@ facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet) || 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 the datapath actions changed or the installability changed, + * then we need to talk to the datapath. */ if (actions_changed || ctx.may_set_up_flow != facet->installed) { if (ctx.may_set_up_flow) { struct dpif_flow_stats stats; @@ -2523,6 +2544,16 @@ facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet, } } +static void +facet_reset_counters(struct facet *facet) +{ + facet->packet_count = 0; + facet->byte_count = 0; + facet->rs_packet_count = 0; + facet->rs_byte_count = 0; + facet->accounted_bytes = 0; +} + static void facet_push_stats(struct facet *facet) { @@ -2587,10 +2618,11 @@ flow_push_stats(const struct rule_dpif *rule, /* Rules. */ static struct rule_dpif * -rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow) +rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow, + uint8_t table_id) { return rule_dpif_cast(rule_from_cls_rule( - classifier_lookup(&ofproto->up.tables[0], + classifier_lookup(&ofproto->up.tables[table_id], flow))); } @@ -2648,6 +2680,13 @@ rule_construct(struct rule *rule_) rule->facets = victim->facets; list_moved(&rule->facets); LIST_FOR_EACH (facet, list_node, &rule->facets) { + /* XXX: We're only clearing our local counters here. It's possible + * that quite a few packets are unaccounted for in the datapath + * statistics. These will be accounted to the new rule instead of + * cleared as required. This could be fixed by clearing out the + * datapath statistics for this facet, but currently it doesn't + * seem worth it. */ + facet_reset_counters(facet); facet->rule = rule; } } else { @@ -2712,7 +2751,7 @@ rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet) /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then * create a new facet for it and use that. */ - if (rule_dpif_lookup(ofproto, flow) == rule) { + if (rule_dpif_lookup(ofproto, flow, 0) == rule) { facet = facet_create(rule, flow, packet); facet_execute(ofproto, facet, packet); facet_install(ofproto, facet, true); @@ -2769,7 +2808,7 @@ send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port, odp_flow_key_from_flow(&key, &flow); ofpbuf_init(&odp_actions, 32); - nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port); + nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port); error = dpif_execute(ofproto->dpif, key.data, key.size, odp_actions.data, odp_actions.size, @@ -2783,7 +2822,7 @@ send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port, return error; } -/* OpenFlow to ODP action translation. */ +/* OpenFlow to datapath action translation. */ static void do_xlate_actions(const union ofp_action *in, size_t n_in, struct action_xlate_ctx *ctx); @@ -2797,63 +2836,66 @@ commit_odp_actions(struct action_xlate_ctx *ctx) struct ofpbuf *odp_actions = ctx->odp_actions; if (base->tun_id != flow->tun_id) { - nl_msg_put_be64(odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, flow->tun_id); + nl_msg_put_be64(odp_actions, OVS_ACTION_ATTR_SET_TUNNEL, flow->tun_id); base->tun_id = flow->tun_id; } if (base->nw_src != flow->nw_src) { - nl_msg_put_be32(odp_actions, ODP_ACTION_ATTR_SET_NW_SRC, flow->nw_src); + nl_msg_put_be32(odp_actions, OVS_ACTION_ATTR_SET_NW_SRC, flow->nw_src); base->nw_src = flow->nw_src; } if (base->nw_dst != flow->nw_dst) { - nl_msg_put_be32(odp_actions, ODP_ACTION_ATTR_SET_NW_DST, flow->nw_dst); + nl_msg_put_be32(odp_actions, OVS_ACTION_ATTR_SET_NW_DST, flow->nw_dst); base->nw_dst = flow->nw_dst; } if (base->nw_tos != flow->nw_tos) { - nl_msg_put_u8(odp_actions, ODP_ACTION_ATTR_SET_NW_TOS, flow->nw_tos); + nl_msg_put_u8(odp_actions, OVS_ACTION_ATTR_SET_NW_TOS, flow->nw_tos); base->nw_tos = flow->nw_tos; } if (base->vlan_tci != flow->vlan_tci) { if (!(flow->vlan_tci & htons(VLAN_CFI))) { - nl_msg_put_flag(odp_actions, ODP_ACTION_ATTR_STRIP_VLAN); + nl_msg_put_flag(odp_actions, OVS_ACTION_ATTR_POP_VLAN); } else { - nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_DL_TCI, + if (base->vlan_tci != OFP_VLAN_NONE) { + nl_msg_put_flag(odp_actions, OVS_ACTION_ATTR_POP_VLAN); + } + nl_msg_put_be16(odp_actions, OVS_ACTION_ATTR_PUSH_VLAN, flow->vlan_tci & ~htons(VLAN_CFI)); } base->vlan_tci = flow->vlan_tci; } if (base->tp_src != flow->tp_src) { - nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_TP_SRC, flow->tp_src); + nl_msg_put_be16(odp_actions, OVS_ACTION_ATTR_SET_TP_SRC, flow->tp_src); base->tp_src = flow->tp_src; } if (base->tp_dst != flow->tp_dst) { - nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_TP_DST, flow->tp_dst); + nl_msg_put_be16(odp_actions, OVS_ACTION_ATTR_SET_TP_DST, flow->tp_dst); base->tp_dst = flow->tp_dst; } if (!eth_addr_equals(base->dl_src, flow->dl_src)) { - nl_msg_put_unspec(odp_actions, ODP_ACTION_ATTR_SET_DL_SRC, + nl_msg_put_unspec(odp_actions, OVS_ACTION_ATTR_SET_DL_SRC, flow->dl_src, ETH_ADDR_LEN); memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN); } if (!eth_addr_equals(base->dl_dst, flow->dl_dst)) { - nl_msg_put_unspec(odp_actions, ODP_ACTION_ATTR_SET_DL_DST, + nl_msg_put_unspec(odp_actions, OVS_ACTION_ATTR_SET_DL_DST, flow->dl_dst, ETH_ADDR_LEN); memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN); } if (ctx->base_priority != ctx->priority) { if (ctx->priority) { - nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, + nl_msg_put_u32(odp_actions, OVS_ACTION_ATTR_SET_PRIORITY, ctx->priority); } else { - nl_msg_put_flag(odp_actions, ODP_ACTION_ATTR_POP_PRIORITY); + nl_msg_put_flag(odp_actions, OVS_ACTION_ATTR_POP_PRIORITY); } ctx->base_priority = ctx->priority; } @@ -2879,23 +2921,28 @@ add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) } commit_odp_actions(ctx); - nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port); + nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port); ctx->nf_output_iface = ofp_port; } static void -xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port) +xlate_table_action(struct action_xlate_ctx *ctx, + uint16_t in_port, uint8_t table_id) { if (ctx->recurse < MAX_RESUBMIT_RECURSION) { struct rule_dpif *rule; uint16_t old_in_port; + uint8_t old_table_id; + + old_table_id = ctx->table_id; + ctx->table_id = table_id; /* Look up a flow with 'in_port' as the input port. Then restore the * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will * have surprising behavior). */ old_in_port = ctx->flow.in_port; ctx->flow.in_port = in_port; - rule = rule_dpif_lookup(ctx->ofproto, &ctx->flow); + rule = rule_dpif_lookup(ctx->ofproto, &ctx->flow, table_id); ctx->flow.in_port = old_in_port; if (ctx->resubmit_hook) { @@ -2907,14 +2954,31 @@ xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port) do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx); ctx->recurse--; } + + ctx->table_id = old_table_id; } else { static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); - VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times", + VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times", MAX_RESUBMIT_RECURSION); } } +static void +xlate_resubmit_table(struct action_xlate_ctx *ctx, + const struct nx_action_resubmit *nar) +{ + uint16_t in_port; + uint8_t table_id; + + in_port = (nar->in_port == htons(OFPP_IN_PORT) + ? ctx->flow.in_port + : ntohs(nar->in_port)); + table_id = nar->table == 255 ? ctx->table_id : nar->table; + + xlate_table_action(ctx, in_port, table_id); +} + static void flood_packets(struct action_xlate_ctx *ctx, ovs_be32 mask) { @@ -2924,7 +2988,7 @@ flood_packets(struct action_xlate_ctx *ctx, ovs_be32 mask) HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) { uint16_t ofp_port = ofport->up.ofp_port; if (ofp_port != ctx->flow.in_port && !(ofport->up.opp.config & mask)) { - nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, + nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, ofport->odp_port); } } @@ -2945,7 +3009,7 @@ xlate_output_action__(struct action_xlate_ctx *ctx, add_output_action(ctx, ctx->flow.in_port); break; case OFPP_TABLE: - xlate_table_action(ctx, ctx->flow.in_port); + xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id); break; case OFPP_NORMAL: xlate_normal(ctx); @@ -2958,7 +3022,7 @@ xlate_output_action__(struct action_xlate_ctx *ctx, break; case OFPP_CONTROLLER: commit_odp_actions(ctx); - nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_USERSPACE, max_len); + nl_msg_put_u64(ctx->odp_actions, OVS_ACTION_ATTR_USERSPACE, max_len); break; case OFPP_LOCAL: add_output_action(ctx, OFPP_LOCAL); @@ -2982,6 +3046,19 @@ xlate_output_action__(struct action_xlate_ctx *ctx, } } +static void +xlate_output_reg_action(struct action_xlate_ctx *ctx, + const struct nx_action_output_reg *naor) +{ + uint64_t ofp_port; + + ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow); + + if (ofp_port <= UINT16_MAX) { + xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len)); + } +} + static void xlate_output_action(struct action_xlate_ctx *ctx, const struct ofp_action_output *oao) @@ -3005,14 +3082,14 @@ xlate_enqueue_action(struct action_xlate_ctx *ctx, return; } - /* Figure out ODP output port. */ + /* Figure out datapath output port. */ ofp_port = ntohs(oae->port); if (ofp_port == OFPP_IN_PORT) { ofp_port = ctx->flow.in_port; } odp_port = ofp_port_to_odp_port(ofp_port); - /* Add ODP actions. */ + /* Add datapath actions. */ ctx_priority = ctx->priority; ctx->priority = priority; add_output_action(ctx, odp_port); @@ -3117,6 +3194,7 @@ do_xlate_actions(const union ofp_action *in, size_t n_in, const struct nx_action_multipath *nam; const struct nx_action_autopath *naa; const struct nx_action_bundle *nab; + const struct nx_action_output_reg *naor; enum ofputil_action_code code; ovs_be64 tun_id; @@ -3177,7 +3255,11 @@ do_xlate_actions(const union ofp_action *in, size_t n_in, case OFPUTIL_NXAST_RESUBMIT: nar = (const struct nx_action_resubmit *) ia; - xlate_table_action(ctx, ntohs(nar->in_port)); + xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id); + break; + + case OFPUTIL_NXAST_RESUBMIT_TABLE: + xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia); break; case OFPUTIL_NXAST_SET_TUNNEL: @@ -3238,6 +3320,11 @@ do_xlate_actions(const union ofp_action *in, size_t n_in, bundle_execute_load(nab, &ctx->flow, slave_enabled_cb, ctx->ofproto); break; + + case OFPUTIL_NXAST_OUTPUT_REG: + naor = (const struct nx_action_output_reg *) ia; + xlate_output_reg_action(ctx, naor); + break; } } } @@ -3267,6 +3354,7 @@ xlate_actions(struct action_xlate_ctx *ctx, ctx->priority = 0; ctx->base_priority = 0; ctx->base_flow = ctx->flow; + ctx->table_id = 0; if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) { ctx->may_set_up_flow = false; @@ -3560,7 +3648,7 @@ compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan, continue; } nl_msg_put_u32(ctx->odp_actions, - ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port); + OVS_ACTION_ATTR_OUTPUT, dst->port->odp_port); } /* Then output the rest. */ @@ -3571,18 +3659,22 @@ compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan, } if (dst->vlan != cur_vlan) { if (dst->vlan == OFP_VLAN_NONE) { - nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN); + nl_msg_put_flag(ctx->odp_actions, OVS_ACTION_ATTR_POP_VLAN); } else { ovs_be16 tci; + + if (cur_vlan != OFP_VLAN_NONE) { + nl_msg_put_flag(ctx->odp_actions, OVS_ACTION_ATTR_POP_VLAN); + } tci = htons(dst->vlan & VLAN_VID_MASK); tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK); nl_msg_put_be16(ctx->odp_actions, - ODP_ACTION_ATTR_SET_DL_TCI, tci); + OVS_ACTION_ATTR_PUSH_VLAN, tci); } cur_vlan = dst->vlan; } nl_msg_put_u32(ctx->odp_actions, - ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port); + OVS_ACTION_ATTR_OUTPUT, dst->port->odp_port); } dst_set_free(&set); @@ -3735,8 +3827,10 @@ is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow, return false; } - /* Drop frames for reserved multicast addresses. */ - if (eth_addr_is_reserved(flow->dl_dst)) { + /* Drop frames for reserved multicast addresses + * only if forward_bpdu option is absent. */ + if (eth_addr_is_reserved(flow->dl_dst) && + !ofproto->up.forward_bpdu) { return false; } @@ -3920,7 +4014,8 @@ struct ofproto_trace { }; static void -trace_format_rule(struct ds *result, int level, const struct rule *rule) +trace_format_rule(struct ds *result, uint8_t table_id, int level, + const struct rule_dpif *rule) { ds_put_char_multiple(result, '\t', level); if (!rule) { @@ -3928,14 +4023,14 @@ trace_format_rule(struct ds *result, int level, const struct rule *rule) return; } - ds_put_format(result, "Rule: cookie=%#"PRIx64" ", - ntohll(rule->flow_cookie)); - cls_rule_format(&rule->cr, result); + ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", + table_id, ntohll(rule->up.flow_cookie)); + cls_rule_format(&rule->up.cr, result); ds_put_char(result, '\n'); ds_put_char_multiple(result, '\t', level); ds_put_cstr(result, "OpenFlow "); - ofp_print_actions(result, rule->actions, rule->n_actions); + ofp_print_actions(result, rule->up.actions, rule->up.n_actions); ds_put_char(result, '\n'); } @@ -3954,6 +4049,20 @@ trace_format_flow(struct ds *result, int level, const char *title, ds_put_char(result, '\n'); } +static void +trace_format_regs(struct ds *result, int level, const char *title, + struct ofproto_trace *trace) +{ + size_t i; + + ds_put_char_multiple(result, '\t', level); + ds_put_format(result, "%s:", title); + for (i = 0; i < FLOW_N_REGS; i++) { + ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); + } + ds_put_char(result, '\n'); +} + static void trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) { @@ -3962,33 +4071,79 @@ trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) ds_put_char(result, '\n'); trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); - trace_format_rule(result, ctx->recurse + 1, &rule->up); + trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace); + trace_format_rule(result, ctx->table_id, 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 *dpname, *arg1, *arg2, *arg3; char *args = xstrdup(args_); char *save_ptr = NULL; struct ofproto_dpif *ofproto; - struct ofpbuf packet; + struct ofpbuf odp_key; + struct ofpbuf *packet; struct rule_dpif *rule; struct ds result; struct flow flow; - uint16_t in_port; - ovs_be64 tun_id; char *s; - ofpbuf_init(&packet, strlen(args) / 2); + packet = NULL; + ofpbuf_init(&odp_key, 0); 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) { + arg1 = strtok_r(NULL, " ", &save_ptr); + arg2 = strtok_r(NULL, " ", &save_ptr); + arg3 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */ + if (dpname && arg1 && !arg2 && !arg3) { + /* ofproto/trace dpname flow */ + int error; + + /* Convert string to datapath key. */ + ofpbuf_init(&odp_key, 0); + error = odp_flow_key_from_string(arg1, &odp_key); + if (error) { + unixctl_command_reply(conn, 501, "Bad flow syntax"); + goto exit; + } + + /* Convert odp_key to flow. */ + error = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow); + if (error) { + unixctl_command_reply(conn, 501, "Invalid flow"); + goto exit; + } + } else if (dpname && arg1 && arg2 && arg3) { + /* ofproto/trace dpname tun_id in_port packet */ + uint16_t in_port; + ovs_be64 tun_id; + + tun_id = htonll(strtoull(arg1, NULL, 0)); + in_port = ofp_port_to_odp_port(atoi(arg2)); + + packet = ofpbuf_new(strlen(args) / 2); + arg3 = ofpbuf_put_hex(packet, arg3, NULL); + arg3 += strspn(arg3, " "); + if (*arg3 != '\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); + } else { unixctl_command_reply(conn, 501, "Bad command syntax"); goto exit; } @@ -4000,39 +4155,19 @@ ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_, goto exit; } - tun_id = htonll(strtoull(tun_id_s, NULL, 0)); - 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_dpif_lookup(ofproto, &flow); - trace_format_rule(&result, 0, &rule->up); + rule = rule_dpif_lookup(ofproto, &flow, 0); + trace_format_rule(&result, 0, 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); + action_xlate_ctx_init(&trace.ctx, ofproto, &flow, packet); trace.ctx.resubmit_hook = trace_resubmit; odp_actions = xlate_actions(&trace.ctx, rule->up.actions, rule->up.n_actions); @@ -4042,13 +4177,23 @@ ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_, ds_put_cstr(&result, "Datapath actions: "); format_odp_actions(&result, odp_actions->data, odp_actions->size); ofpbuf_delete(odp_actions); + + if (!trace.ctx.may_set_up_flow) { + if (packet) { + ds_put_cstr(&result, "\nThis flow is not cachable."); + } else { + ds_put_cstr(&result, "\nThe datapath actions are incomplete--" + "for complete actions, please supply a packet."); + } + } } unixctl_command_reply(conn, 200, ds_cstr(&result)); exit: ds_destroy(&result); - ofpbuf_uninit(&packet); + ofpbuf_delete(packet); + ofpbuf_uninit(&odp_key); free(args); } @@ -4128,9 +4273,11 @@ const struct ofproto_class ofproto_dpif_class = { set_sflow, set_cfm, get_cfm_fault, + get_cfm_remote_mpids, bundle_set, bundle_remove, mirror_set, set_flood_vlans, is_mirror_output_bundle, + forward_bpdu_changed, };