X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;ds=sidebyside;f=ofproto%2Fofproto.c;h=52a1964bfdfe3d1ffd95ed9ad8d68e07e0179da2;hb=db5ce51427906ed54aa3e8e1dffc0ba6d908aa8c;hp=d5f258ffe8c879a53c1287b3cf1a2ca495e1b5c5;hpb=bbc32a883f790a331b412eecc6de4b7cdf1e63cb;p=openvswitch diff --git a/ofproto/ofproto.c b/ofproto/ofproto.c index d5f258ff..52a1964b 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" @@ -90,6 +92,8 @@ 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 +104,48 @@ 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 *); + +/* 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 + * recently added ODPAT_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 { @@ -185,8 +226,8 @@ 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. */ }; @@ -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; @@ -377,6 +421,8 @@ 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, @@ -389,6 +435,8 @@ ofproto_create(const char *datapath, const char *datapath_type, *ofprotop = NULL; + ofproto_unixctl_init(); + /* Connect to datapath and start listening for messages. */ error = dpif_open(datapath, datapath_type, &dpif); if (error) { @@ -470,6 +518,8 @@ 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 +1041,8 @@ 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; @@ -1349,18 +1401,18 @@ 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); + 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; } @@ -1429,7 +1481,8 @@ ofproto_flush_flows(struct ofproto *ofproto) static void reinit_ports(struct ofproto *p) { - struct svec devnames; + struct shash_node *node; + struct shash devnames; struct ofport *ofport; struct odp_port *odp_ports; size_t n_odp_ports; @@ -1437,21 +1490,20 @@ reinit_ports(struct ofproto *p) 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); + shash_add_once (&devnames, odp_ports[i].devname, 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 * @@ -1999,10 +2051,11 @@ 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, + 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 == ODPAT_CONTROLLER) { /* As an optimization, avoid a round-trip from userspace to kernel to * userspace. This also avoids possibly filling up kernel packet * buffers along the way. */ @@ -2012,8 +2065,7 @@ execute_odp_actions(struct ofproto *ofproto, uint16_t in_port, 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; + msg->arg = nl_attr_get_u64(odp_actions); send_packet_in(ofproto, packet); @@ -2021,7 +2073,7 @@ execute_odp_actions(struct ofproto *ofproto, uint16_t in_port, } 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; } @@ -2049,7 +2101,7 @@ facet_execute(struct ofproto *ofproto, struct facet *facet, flow_extract_stats(&facet->flow, packet, &stats); if (execute_odp_actions(ofproto, facet->flow.in_port, - facet->actions, facet->n_actions, packet)) { + facet->actions, facet->actions_len, packet)) { facet_update_stats(ofproto, facet, &stats); facet->used = time_msec(); netflow_flow_update_time(ofproto->netflow, @@ -2069,8 +2121,9 @@ 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 +2149,16 @@ 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, in_port, 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. */ @@ -2195,20 +2246,23 @@ 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 @@ -2218,7 +2272,7 @@ facet_put__(struct ofproto *ofproto, struct facet *facet, int flags, 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.actions_len = facet->actions_len; put->flow.flags = 0; put->flags = flags; return dpif_flow_put(ofproto->dpif, put); @@ -2256,7 +2310,7 @@ 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; } @@ -2271,7 +2325,7 @@ facet_uninstall(struct ofproto *p, struct facet *facet) odp_flow_key_from_flow(&odp_flow.key, &facet->flow); odp_flow.actions = NULL; - odp_flow.n_actions = 0; + odp_flow.actions_len = 0; odp_flow.flags = 0; if (!dpif_flow_del(p->dpif, &odp_flow)) { facet_update_stats(p, facet, &odp_flow.stats); @@ -2375,10 +2429,9 @@ 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); @@ -2393,15 +2446,14 @@ 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. */ @@ -2411,8 +2463,8 @@ facet_revalidate(struct ofproto *ofproto, struct facet *facet) 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.actions = odp_actions->data; + put.flow.actions_len = odp_actions->size; put.flow.flags = 0; put.flags = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS; dpif_flow_put(ofproto->dpif, &put); @@ -2428,11 +2480,13 @@ 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 +2496,8 @@ facet_revalidate(struct ofproto *ofproto, struct facet *facet) facet->used = new_rule->created; } + ofpbuf_delete(odp_actions); + return true; } @@ -2565,33 +2621,9 @@ 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 +2646,7 @@ 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, ODPAT_OUTPUT, port); ctx->nf_output_iface = port; } @@ -2639,13 +2671,17 @@ 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 +2690,14 @@ 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, ODPAT_OUTPUT, odp_port); } } *nf_output_iface = NF_OUT_FLOOD; @@ -2685,7 +2721,7 @@ 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 +2730,14 @@ 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, ODPAT_CONTROLLER, max_len); break; case OFPP_LOCAL: add_output_action(ctx, ODPP_LOCAL); @@ -2738,9 +2774,18 @@ 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, ODPAT_POP_PRIORITY); + ctx->last_pop_priority = ctx->odp_actions->size; } } @@ -2770,10 +2815,9 @@ 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, ODPAT_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 +2843,7 @@ 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, ODPAT_SET_PRIORITY, priority); } static void @@ -2808,10 +2851,10 @@ 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, ODPAT_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, ODPAT_SET_DL_TCI, + tci & ~htons(VLAN_CFI)); } } @@ -2835,8 +2878,9 @@ 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); + ovs_be64 tun_id; assert(nah->vendor == htonl(NX_VENDOR_ID)); switch (subtype) { @@ -2847,13 +2891,14 @@ 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, ODPAT_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, ODPAT_DROP_SPOOFED_ARP); } break; @@ -2863,7 +2908,7 @@ 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: @@ -2878,11 +2923,23 @@ xlate_nicira_action(struct action_xlate_ctx *ctx, /* 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, ODPAT_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 +2961,8 @@ 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 +2988,47 @@ 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, ODPAT_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, ODPAT_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, ODPAT_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, ODPAT_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, ODPAT_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, ODPAT_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, ODPAT_SET_TP_DST, + ia->tp_port.tp_port); + ctx->flow.tp_dst = ia->tp_port.tp_port; break; case OFPAT_VENDOR: @@ -2980,55 +3040,46 @@ 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; +} +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; + do_xlate_actions(in, n_in, ctx); + remove_pop_action(ctx); /* Check with in-band control to see if we're allowed to set up this * flow. */ - 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 +3108,8 @@ 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 +3156,10 @@ 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 +3265,7 @@ 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; } @@ -3418,9 +3468,9 @@ 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, &ofs->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); @@ -3542,19 +3592,17 @@ 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 { @@ -4287,12 +4335,12 @@ handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) /* 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; + 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, ODPAT_OUTPUT, ODPP_LOCAL); + dpif_execute(p->dpif, odp_actions.data, odp_actions.size, &payload); + ofpbuf_uninit(&odp_actions); } facet = facet_lookup_valid(p, &flow); @@ -4637,9 +4685,9 @@ 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 +4705,10 @@ 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; @@ -4878,9 +4926,177 @@ 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_be32 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 = ntohl(strtoul(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 +5127,9 @@ 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, ODPAT_OUTPUT, out_port); *nf_output_iface = out_port; } else { /* Drop. */