X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Flearning-switch.c;h=4ce307e4bddceadfc739571437a06875fdfff2e0;hb=748dca871d1f8a06d33a1002e1dfabb74b16c9b6;hp=90749c018c6320b699e2cf2e0a4f5d0c4d9d2dc8;hpb=ad67e568887f56d3b70a8225c3a476d94f70c20b;p=sliver-openvswitch.git diff --git a/lib/learning-switch.c b/lib/learning-switch.c index 90749c018..4ce307e4b 100644 --- a/lib/learning-switch.c +++ b/lib/learning-switch.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2008, 2009, 2010 Nicira Networks. + * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira Networks. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -23,22 +23,31 @@ #include #include +#include "byte-order.h" +#include "classifier.h" #include "flow.h" +#include "hmap.h" #include "mac-learning.h" #include "ofpbuf.h" +#include "ofp-errors.h" #include "ofp-parse.h" #include "ofp-print.h" #include "ofp-util.h" #include "openflow/openflow.h" #include "poll-loop.h" -#include "queue.h" #include "rconn.h" +#include "shash.h" #include "timeval.h" #include "vconn.h" #include "vlog.h" -#include "xtoxll.h" -VLOG_DEFINE_THIS_MODULE(learning_switch) +VLOG_DEFINE_THIS_MODULE(learning_switch); + +struct lswitch_port { + struct hmap_node hmap_node; /* Hash node for port number. */ + uint16_t port_no; /* OpenFlow port number, in host byte order. */ + uint32_t queue_id; /* OpenFlow queue number. */ +}; struct lswitch { /* If nonnegative, the switch sets up flows that expire after the given @@ -49,9 +58,13 @@ struct lswitch { unsigned long long int datapath_id; time_t last_features_request; struct mac_learning *ml; /* NULL to act as hub instead of switch. */ - uint32_t wildcards; /* Wildcards to apply to flows. */ + struct flow_wildcards wc; /* Wildcards to apply to flows. */ bool action_normal; /* Use OFPP_NORMAL? */ - uint32_t queue; /* OpenFlow queue to use, or UINT32_MAX. */ + + /* Queue distribution. */ + uint32_t default_queue; /* Default OpenFlow queue, or UINT32_MAX. */ + struct hmap queue_numbers; /* Map from port number to lswitch_port. */ + struct shash queue_names; /* Map from port name to lswitch_port. */ /* Number of outgoing queued packets on the rconn. */ struct rconn_packet_counter *queued; @@ -64,10 +77,12 @@ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); static void queue_tx(struct lswitch *, struct rconn *, struct ofpbuf *); static void send_features_request(struct lswitch *, struct rconn *); -typedef void packet_handler_func(struct lswitch *, struct rconn *, void *); -static packet_handler_func process_switch_features; -static packet_handler_func process_packet_in; -static packet_handler_func process_echo_request; +static void process_switch_features(struct lswitch *, + struct ofp_switch_features *); +static void process_packet_in(struct lswitch *, struct rconn *, + const struct ofp_packet_in *); +static void process_echo_request(struct lswitch *, struct rconn *, + const struct ofp_header *); /* Creates and returns a new learning switch whose configuration is given by * 'cfg'. @@ -76,31 +91,66 @@ static packet_handler_func process_echo_request; struct lswitch * lswitch_create(struct rconn *rconn, const struct lswitch_config *cfg) { - const struct ofpbuf *b; struct lswitch *sw; sw = xzalloc(sizeof *sw); sw->max_idle = cfg->max_idle; sw->datapath_id = 0; sw->last_features_request = time_now() - 1; - sw->ml = cfg->mode == LSW_LEARN ? mac_learning_create() : NULL; + sw->ml = (cfg->mode == LSW_LEARN + ? mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME) + : NULL); sw->action_normal = cfg->mode == LSW_NORMAL; - if (cfg->exact_flows) { - /* Exact match. */ - sw->wildcards = 0; - } else { - /* We cannot wildcard all fields. - * We need in_port to detect moves. - * We need both SA and DA to do learning. */ - sw->wildcards = (OFPFW_DL_TYPE | OFPFW_NW_SRC_MASK | OFPFW_NW_DST_MASK - | OFPFW_NW_PROTO | OFPFW_TP_SRC | OFPFW_TP_DST); + + flow_wildcards_init_exact(&sw->wc); + if (cfg->wildcards) { + uint32_t ofpfw; + + if (cfg->wildcards == UINT32_MAX) { + /* Try to wildcard as many fields as possible, but we cannot + * wildcard all fields. We need in_port to detect moves. We need + * Ethernet source and dest and VLAN VID to do L2 learning. */ + ofpfw = (OFPFW_DL_TYPE | OFPFW_DL_VLAN_PCP + | OFPFW_NW_SRC_ALL | OFPFW_NW_DST_ALL + | OFPFW_NW_TOS | OFPFW_NW_PROTO + | OFPFW_TP_SRC | OFPFW_TP_DST); + } else { + ofpfw = cfg->wildcards; + } + + ofputil_wildcard_from_openflow(ofpfw, &sw->wc); } - sw->queue = cfg->queue_id; + + sw->default_queue = cfg->default_queue; + hmap_init(&sw->queue_numbers); + shash_init(&sw->queue_names); + if (cfg->port_queues) { + struct shash_node *node; + + SHASH_FOR_EACH (node, cfg->port_queues) { + struct lswitch_port *port = xmalloc(sizeof *port); + hmap_node_nullify(&port->hmap_node); + port->queue_id = (uintptr_t) node->data; + shash_add(&sw->queue_names, node->name, port); + } + } + sw->queued = rconn_packet_counter_create(); send_features_request(sw, rconn); - for (b = cfg->default_flows; b; b = b->next) { - queue_tx(sw, rconn, ofpbuf_clone(b)); + if (cfg->default_flows) { + const struct ofpbuf *b; + + LIST_FOR_EACH (b, list_node, cfg->default_flows) { + struct ofpbuf *copy = ofpbuf_clone(b); + int error = rconn_send(rconn, copy, NULL); + if (error) { + VLOG_INFO_RL(&rl, "%s: failed to queue default flows (%s)", + rconn_get_name(rconn), strerror(error)); + ofpbuf_delete(copy); + break; + } + } } return sw; @@ -111,6 +161,13 @@ void lswitch_destroy(struct lswitch *sw) { if (sw) { + struct lswitch_port *node, *next; + + HMAP_FOR_EACH_SAFE (node, next, hmap_node, &sw->queue_numbers) { + hmap_remove(&sw->queue_numbers, &node->hmap_node); + free(node); + } + shash_destroy(&sw->queue_names); mac_learning_destroy(sw->ml); rconn_packet_counter_destroy(sw->queued); free(sw); @@ -143,38 +200,9 @@ void lswitch_process_packet(struct lswitch *sw, struct rconn *rconn, const struct ofpbuf *msg) { - struct processor { - uint8_t type; - size_t min_size; - packet_handler_func *handler; - }; - static const struct processor processors[] = { - { - OFPT_ECHO_REQUEST, - sizeof(struct ofp_header), - process_echo_request - }, - { - OFPT_FEATURES_REPLY, - sizeof(struct ofp_switch_features), - process_switch_features - }, - { - OFPT_PACKET_IN, - offsetof(struct ofp_packet_in, data), - process_packet_in - }, - { - OFPT_FLOW_REMOVED, - sizeof(struct ofp_flow_removed), - NULL - }, - }; - const size_t n_processors = ARRAY_SIZE(processors); - const struct processor *p; - struct ofp_header *oh; - - oh = msg->data; + const struct ofp_header *oh = msg->data; + const struct ofputil_msg_type *type; + if (sw->datapath_id == 0 && oh->type != OFPT_ECHO_REQUEST && oh->type != OFPT_FEATURES_REPLY) { @@ -182,27 +210,75 @@ lswitch_process_packet(struct lswitch *sw, struct rconn *rconn, return; } - for (p = processors; p < &processors[n_processors]; p++) { - if (oh->type == p->type) { - if (msg->size < p->min_size) { - VLOG_WARN_RL(&rl, "%016llx: %s: too short (%zu bytes) for " - "type %"PRIu8" (min %zu)", sw->datapath_id, - rconn_get_name(rconn), msg->size, oh->type, - p->min_size); - return; - } - if (p->handler) { - (p->handler)(sw, rconn, msg->data); - } - return; + ofputil_decode_msg_type(oh, &type); + switch (ofputil_msg_type_code(type)) { + case OFPUTIL_OFPT_ECHO_REQUEST: + process_echo_request(sw, rconn, msg->data); + break; + + case OFPUTIL_OFPT_FEATURES_REPLY: + process_switch_features(sw, msg->data); + break; + + case OFPUTIL_OFPT_PACKET_IN: + process_packet_in(sw, rconn, msg->data); + break; + + case OFPUTIL_OFPT_FLOW_REMOVED: + /* Nothing to do. */ + break; + + case OFPUTIL_MSG_INVALID: + case OFPUTIL_OFPT_HELLO: + case OFPUTIL_OFPT_ERROR: + case OFPUTIL_OFPT_ECHO_REPLY: + case OFPUTIL_OFPT_FEATURES_REQUEST: + case OFPUTIL_OFPT_GET_CONFIG_REQUEST: + case OFPUTIL_OFPT_GET_CONFIG_REPLY: + case OFPUTIL_OFPT_SET_CONFIG: + case OFPUTIL_OFPT_PORT_STATUS: + case OFPUTIL_OFPT_PACKET_OUT: + case OFPUTIL_OFPT_FLOW_MOD: + case OFPUTIL_OFPT_PORT_MOD: + case OFPUTIL_OFPT_BARRIER_REQUEST: + case OFPUTIL_OFPT_BARRIER_REPLY: + case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST: + case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY: + case OFPUTIL_OFPST_DESC_REQUEST: + case OFPUTIL_OFPST_FLOW_REQUEST: + case OFPUTIL_OFPST_AGGREGATE_REQUEST: + case OFPUTIL_OFPST_TABLE_REQUEST: + case OFPUTIL_OFPST_PORT_REQUEST: + case OFPUTIL_OFPST_QUEUE_REQUEST: + case OFPUTIL_OFPST_DESC_REPLY: + case OFPUTIL_OFPST_FLOW_REPLY: + case OFPUTIL_OFPST_QUEUE_REPLY: + case OFPUTIL_OFPST_PORT_REPLY: + case OFPUTIL_OFPST_TABLE_REPLY: + case OFPUTIL_OFPST_AGGREGATE_REPLY: + case OFPUTIL_NXT_ROLE_REQUEST: + case OFPUTIL_NXT_ROLE_REPLY: + case OFPUTIL_NXT_FLOW_MOD_TABLE_ID: + case OFPUTIL_NXT_SET_FLOW_FORMAT: + case OFPUTIL_NXT_SET_PACKET_IN_FORMAT: + case OFPUTIL_NXT_PACKET_IN: + case OFPUTIL_NXT_FLOW_MOD: + case OFPUTIL_NXT_FLOW_REMOVED: + case OFPUTIL_NXT_FLOW_AGE: + case OFPUTIL_NXT_SET_ASYNC_CONFIG: + case OFPUTIL_NXT_SET_CONTROLLER_ID: + case OFPUTIL_NXST_FLOW_REQUEST: + case OFPUTIL_NXST_AGGREGATE_REQUEST: + case OFPUTIL_NXST_FLOW_REPLY: + case OFPUTIL_NXST_AGGREGATE_REPLY: + default: + if (VLOG_IS_DBG_ENABLED()) { + char *s = ofp_to_string(msg->data, msg->size, 2); + VLOG_DBG_RL(&rl, "%016llx: OpenFlow packet ignored: %s", + sw->datapath_id, s); + free(s); } } - if (VLOG_IS_DBG_ENABLED()) { - char *s = ofp_to_string(msg->data, msg->size, 2); - VLOG_DBG_RL(&rl, "%016llx: OpenFlow packet ignored: %s", - sw->datapath_id, s); - free(s); - } } static void @@ -243,26 +319,43 @@ queue_tx(struct lswitch *sw, struct rconn *rconn, struct ofpbuf *b) } static void -process_switch_features(struct lswitch *sw, struct rconn *rconn OVS_UNUSED, - void *osf_) +process_switch_features(struct lswitch *sw, struct ofp_switch_features *osf) { - struct ofp_switch_features *osf = osf_; + size_t n_ports; + size_t i; sw->datapath_id = ntohll(osf->datapath_id); + + n_ports = (ntohs(osf->header.length) - sizeof *osf) / sizeof *osf->ports; + for (i = 0; i < n_ports; i++) { + struct ofp_phy_port *opp = &osf->ports[i]; + struct lswitch_port *lp; + + opp->name[OFP_MAX_PORT_NAME_LEN - 1] = '\0'; + lp = shash_find_data(&sw->queue_names, opp->name); + if (lp && hmap_node_is_null(&lp->hmap_node)) { + lp->port_no = ntohs(opp->port_no); + hmap_insert(&sw->queue_numbers, &lp->hmap_node, + hash_int(lp->port_no, 0)); + } + } } static uint16_t -lswitch_choose_destination(struct lswitch *sw, const flow_t *flow) +lswitch_choose_destination(struct lswitch *sw, const struct flow *flow) { uint16_t out_port; /* Learn the source MAC. */ - if (sw->ml) { - if (mac_learning_learn(sw->ml, flow->dl_src, 0, flow->in_port, - GRAT_ARP_LOCK_NONE)) { + if (mac_learning_may_learn(sw->ml, flow->dl_src, 0)) { + struct mac_entry *mac = mac_learning_insert(sw->ml, flow->dl_src, 0); + if (mac_entry_is_new(mac) || mac->port.i != flow->in_port) { VLOG_DBG_RL(&rl, "%016llx: learned that "ETH_ADDR_FMT" is on " "port %"PRIu16, sw->datapath_id, ETH_ADDR_ARGS(flow->dl_src), flow->in_port); + + mac->port.i = flow->in_port; + mac_learning_changed(sw->ml, mac); } } @@ -273,9 +366,11 @@ lswitch_choose_destination(struct lswitch *sw, const flow_t *flow) out_port = OFPP_FLOOD; if (sw->ml) { - int learned_port = mac_learning_lookup(sw->ml, flow->dl_dst, 0, NULL); - if (learned_port >= 0) { - out_port = learned_port; + struct mac_entry *mac; + + mac = mac_learning_lookup(sw->ml, flow->dl_dst, 0, NULL); + if (mac) { + out_port = mac->port.i; if (out_port == flow->in_port) { /* Don't send a packet back out its input port. */ return OFPP_NONE; @@ -291,19 +386,37 @@ lswitch_choose_destination(struct lswitch *sw, const flow_t *flow) return out_port; } +static uint32_t +get_queue_id(const struct lswitch *sw, uint16_t in_port) +{ + const struct lswitch_port *port; + + HMAP_FOR_EACH_WITH_HASH (port, hmap_node, hash_int(in_port, 0), + &sw->queue_numbers) { + if (port->port_no == in_port) { + return port->queue_id; + } + } + + return sw->default_queue; +} + static void -process_packet_in(struct lswitch *sw, struct rconn *rconn, void *opi_) +process_packet_in(struct lswitch *sw, struct rconn *rconn, + const struct ofp_packet_in *opi) { - struct ofp_packet_in *opi = opi_; uint16_t in_port = ntohs(opi->in_port); + uint32_t queue_id; uint16_t out_port; struct ofp_action_header actions[2]; size_t actions_len; + struct ofputil_packet_out po; + size_t pkt_ofs, pkt_len; struct ofpbuf pkt; - flow_t flow; + struct flow flow; /* Ignore packets sent via output to OFPP_CONTROLLER. This library never * uses such an action. You never know what experiments might be going on, @@ -315,17 +428,17 @@ process_packet_in(struct lswitch *sw, struct rconn *rconn, void *opi_) /* Extract flow data from 'opi' into 'flow'. */ pkt_ofs = offsetof(struct ofp_packet_in, data); pkt_len = ntohs(opi->header.length) - pkt_ofs; - pkt.data = opi->data; - pkt.size = pkt_len; - flow_extract(&pkt, 0, in_port, &flow); + ofpbuf_use_const(&pkt, opi->data, pkt_len); + flow_extract(&pkt, 0, 0, in_port, &flow); /* Choose output port. */ out_port = lswitch_choose_destination(sw, &flow); /* Make actions. */ + queue_id = get_queue_id(sw, in_port); if (out_port == OFPP_NONE) { actions_len = 0; - } else if (sw->queue == UINT32_MAX || out_port >= OFPP_MAX) { + } else if (queue_id == UINT32_MAX || out_port >= OFPP_MAX) { struct ofp_action_output oao; memset(&oao, 0, sizeof oao); @@ -342,47 +455,55 @@ process_packet_in(struct lswitch *sw, struct rconn *rconn, void *opi_) oae.type = htons(OFPAT_ENQUEUE); oae.len = htons(sizeof oae); oae.port = htons(out_port); - oae.queue_id = htonl(sw->queue); + oae.queue_id = htonl(queue_id); memcpy(actions, &oae, sizeof oae); actions_len = sizeof oae; } assert(actions_len <= sizeof actions); + /* Prepare packet_out in case we need one. */ + po.buffer_id = ntohl(opi->buffer_id); + if (po.buffer_id == UINT32_MAX) { + po.packet = pkt.data; + po.packet_len = pkt.size; + } else { + po.packet = NULL; + po.packet_len = 0; + } + po.in_port = in_port; + po.actions = (union ofp_action *) actions; + po.n_actions = actions_len / sizeof *actions; + /* Send the packet, and possibly the whole flow, to the output port. */ if (sw->max_idle >= 0 && (!sw->ml || out_port != OFPP_FLOOD)) { struct ofpbuf *buffer; - struct ofp_flow_mod *ofm; + struct cls_rule rule; /* The output port is known, or we always flood everything, so add a * new flow. */ - buffer = make_add_flow(&flow, ntohl(opi->buffer_id), + cls_rule_init(&flow, &sw->wc, 0, &rule); + buffer = make_add_flow(&rule, ntohl(opi->buffer_id), sw->max_idle, actions_len); ofpbuf_put(buffer, actions, actions_len); - ofm = buffer->data; - ofm->match.wildcards = htonl(sw->wildcards); queue_tx(sw, rconn, buffer); /* If the switch didn't buffer the packet, we need to send a copy. */ if (ntohl(opi->buffer_id) == UINT32_MAX && actions_len > 0) { - queue_tx(sw, rconn, - make_packet_out(&pkt, UINT32_MAX, in_port, - actions, actions_len / sizeof *actions)); + queue_tx(sw, rconn, ofputil_encode_packet_out(&po)); } } else { /* We don't know that MAC, or we don't set up flows. Send along the * packet without setting up a flow. */ if (ntohl(opi->buffer_id) != UINT32_MAX || actions_len > 0) { - queue_tx(sw, rconn, - make_packet_out(&pkt, ntohl(opi->buffer_id), in_port, - actions, actions_len / sizeof *actions)); + queue_tx(sw, rconn, ofputil_encode_packet_out(&po)); } } } static void -process_echo_request(struct lswitch *sw, struct rconn *rconn, void *rq_) +process_echo_request(struct lswitch *sw, struct rconn *rconn, + const struct ofp_header *rq) { - struct ofp_header *rq = rq_; queue_tx(sw, rconn, make_echo_reply(rq)); }