X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=ofproto%2Fconnmgr.c;h=18b80b8787c68ca706a095502cc0c456cc0c4cc6;hb=1a59dc2c8a23dcd6a240fa6ee099008602c9f73e;hp=4d9e7582513c16cfc94d8c461a4fccbcaa02b65c;hpb=9cfcdadf934213982a0fb96a29d100e8ba0187f7;p=openvswitch diff --git a/ofproto/connmgr.c b/ofproto/connmgr.c index 4d9e7582..18b80b87 100644 --- a/ofproto/connmgr.c +++ b/ofproto/connmgr.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2009, 2010, 2011 Nicira Networks. + * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -22,17 +22,20 @@ #include #include "coverage.h" -#include "dpif.h" #include "fail-open.h" #include "in-band.h" #include "odp-util.h" +#include "ofp-actions.h" #include "ofp-util.h" #include "ofpbuf.h" +#include "ofproto-provider.h" #include "pinsched.h" #include "poll-loop.h" #include "pktbuf.h" #include "rconn.h" #include "shash.h" +#include "simap.h" +#include "stream.h" #include "timeval.h" #include "vconn.h" #include "vlog.h" @@ -40,15 +43,30 @@ VLOG_DEFINE_THIS_MODULE(connmgr); static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); -COVERAGE_DEFINE(ofconn_stuck); - /* An OpenFlow connection. */ struct ofconn { - struct connmgr *connmgr; /* Connection's manager. */ +/* Configuration that persists from one connection to the next. */ + struct list node; /* In struct connmgr's "all_conns" list. */ + struct hmap_node hmap_node; /* In struct connmgr's "controllers" map. */ + + struct connmgr *connmgr; /* Connection's manager. */ struct rconn *rconn; /* OpenFlow connection. */ enum ofconn_type type; /* Type. */ - enum nx_flow_format flow_format; /* Currently selected flow format. */ + enum ofproto_band band; /* In-band or out-of-band? */ + bool enable_async_msgs; /* Initially enable async messages? */ + +/* State that should be cleared from one connection to the next. */ + + /* OpenFlow state. */ + enum nx_role role; /* Role. */ + enum ofputil_protocol protocol; /* Current protocol variant. */ + enum nx_packet_in_format packet_in_format; /* OFPT_PACKET_IN format. */ + + /* Asynchronous flow table operation support. */ + struct list opgroups; /* Contains pending "ofopgroups", if any. */ + struct ofpbuf *blocked; /* Postponed OpenFlow message, if any. */ + bool retry; /* True if 'blocked' is ready to try again. */ /* OFPT_PACKET_IN related data. */ struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */ @@ -56,6 +74,7 @@ struct ofconn { struct pinsched *schedulers[N_SCHEDULERS]; struct pktbuf *pktbuf; /* OpenFlow packet buffers. */ int miss_send_len; /* Bytes to send of buffered packets. */ + uint16_t controller_id; /* Connection controller ID. */ /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow * requests, and the maximum number before we stop reading OpenFlow @@ -63,31 +82,32 @@ struct ofconn { #define OFCONN_REPLY_MAX 100 struct rconn_packet_counter *reply_counter; - /* type == OFCONN_PRIMARY only. */ - enum nx_role role; /* Role. */ - struct hmap_node hmap_node; /* In struct connmgr's "controllers" map. */ - enum ofproto_band band; /* In-band or out-of-band? */ + /* Asynchronous message configuration in each possible roles. + * + * A 1-bit enables sending an asynchronous message for one possible reason + * that the message might be generated, a 0-bit disables it. */ + uint32_t master_async_config[OAM_N_TYPES]; /* master, other */ + uint32_t slave_async_config[OAM_N_TYPES]; /* slave */ }; static struct ofconn *ofconn_create(struct connmgr *, struct rconn *, - enum ofconn_type); + enum ofconn_type, bool enable_async_msgs); static void ofconn_destroy(struct ofconn *); +static void ofconn_flush(struct ofconn *); static void ofconn_reconfigure(struct ofconn *, const struct ofproto_controller *); static void ofconn_run(struct ofconn *, - void (*handle_openflow)(struct ofconn *, + bool (*handle_openflow)(struct ofconn *, struct ofpbuf *ofp_msg)); -static void ofconn_wait(struct ofconn *); +static void ofconn_wait(struct ofconn *, bool handling_openflow); static const char *ofconn_get_target(const struct ofconn *); static char *ofconn_make_name(const struct connmgr *, const char *target); static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst); -static bool ofconn_receives_async_msgs(const struct ofconn *); - static void ofconn_send(const struct ofconn *, struct ofpbuf *, struct rconn_packet_counter *); @@ -103,11 +123,13 @@ struct ofservice { int probe_interval; /* Max idle time before probing, in seconds. */ int rate_limit; /* Max packet-in rate in packets per second. */ int burst_limit; /* Limit on accumulating packet credits. */ + bool enable_async_msgs; /* Initially enable async messages? */ + uint8_t dscp; /* DSCP Value for controller connection */ }; static void ofservice_reconfigure(struct ofservice *, const struct ofproto_controller *); -static int ofservice_create(struct connmgr *, const char *target); +static int ofservice_create(struct connmgr *, const char *target, uint8_t dscp); static void ofservice_destroy(struct connmgr *, struct ofservice *); static struct ofservice *ofservice_lookup(struct connmgr *, const char *target); @@ -133,7 +155,6 @@ struct connmgr { /* In-band control. */ struct in_band *in_band; - long long int next_in_band_update; struct sockaddr_in *extra_in_band_remotes; size_t n_extra_remotes; int in_band_queue; @@ -168,7 +189,6 @@ connmgr_create(struct ofproto *ofproto, mgr->fail_mode = OFPROTO_FAIL_SECURE; mgr->in_band = NULL; - mgr->next_in_band_update = LLONG_MAX; mgr->extra_in_band_remotes = NULL; mgr->n_extra_remotes = 0; mgr->in_band_queue = -1; @@ -215,23 +235,33 @@ connmgr_destroy(struct connmgr *mgr) free(mgr); } -/* Does all of the periodic maintenance required by 'mgr'. Calls - * 'handle_openflow' for each message received on an OpenFlow connection, - * passing along the OpenFlow connection itself and the message that was sent. - * The 'handle_openflow' callback must not free the message. */ +/* Does all of the periodic maintenance required by 'mgr'. + * + * If 'handle_openflow' is nonnull, calls 'handle_openflow' for each message + * received on an OpenFlow connection, passing along the OpenFlow connection + * itself and the message that was sent. If 'handle_openflow' returns true, + * the message is considered to be fully processed. If 'handle_openflow' + * returns false, the message is considered not to have been processed at all; + * it will be stored and re-presented to 'handle_openflow' following the next + * call to connmgr_retry(). 'handle_openflow' must not modify or free the + * message. + * + * If 'handle_openflow' is NULL, no OpenFlow messages will be processed and + * other activities that could affect the flow table (in-band processing, + * fail-open processing) are suppressed too. */ void connmgr_run(struct connmgr *mgr, - void (*handle_openflow)(struct ofconn *, struct ofpbuf *ofp_msg)) + bool (*handle_openflow)(struct ofconn *, struct ofpbuf *ofp_msg)) { struct ofconn *ofconn, *next_ofconn; struct ofservice *ofservice; size_t i; - if (mgr->in_band) { - if (time_msec() >= mgr->next_in_band_update) { - update_in_band_remotes(mgr); + if (handle_openflow && mgr->in_band) { + if (!in_band_run(mgr->in_band)) { + in_band_destroy(mgr->in_band); + mgr->in_band = NULL; } - in_band_run(mgr->in_band); } LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &mgr->all_conns) { @@ -240,7 +270,7 @@ connmgr_run(struct connmgr *mgr, /* Fail-open maintenance. Do this after processing the ofconns since * fail-open checks the status of the controller rconn. */ - if (mgr->fail_open) { + if (handle_openflow && mgr->fail_open) { fail_open_run(mgr->fail_open); } @@ -248,17 +278,19 @@ connmgr_run(struct connmgr *mgr, struct vconn *vconn; int retval; - retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn); + retval = pvconn_accept(ofservice->pvconn, OFP10_VERSION, &vconn); if (!retval) { struct rconn *rconn; char *name; - rconn = rconn_create(ofservice->probe_interval, 0); + /* Passing default value for creation of the rconn */ + rconn = rconn_create(ofservice->probe_interval, 0, ofservice->dscp); name = ofconn_make_name(mgr, vconn_get_name(vconn)); rconn_connect_unreliably(rconn, vconn, name); free(name); - ofconn = ofconn_create(mgr, rconn, OFCONN_SERVICE); + ofconn = ofconn_create(mgr, rconn, OFCONN_SERVICE, + ofservice->enable_async_msgs); ofconn_set_rate_limit(ofconn, ofservice->rate_limit, ofservice->burst_limit); } else if (retval != EAGAIN) { @@ -270,7 +302,7 @@ connmgr_run(struct connmgr *mgr, struct vconn *vconn; int retval; - retval = pvconn_accept(mgr->snoops[i], OFP_VERSION, &vconn); + retval = pvconn_accept(mgr->snoops[i], OFP10_VERSION, &vconn); if (!retval) { add_snooper(mgr, vconn); } else if (retval != EAGAIN) { @@ -279,22 +311,25 @@ connmgr_run(struct connmgr *mgr, } } -/* Causes the poll loop to wake up when connmgr_run() needs to run. */ +/* Causes the poll loop to wake up when connmgr_run() needs to run. + * + * If 'handling_openflow' is true, arriving OpenFlow messages and other + * activities that affect the flow table will wake up the poll loop. If + * 'handling_openflow' is false, they will not. */ void -connmgr_wait(struct connmgr *mgr) +connmgr_wait(struct connmgr *mgr, bool handling_openflow) { struct ofservice *ofservice; struct ofconn *ofconn; size_t i; LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { - ofconn_wait(ofconn); + ofconn_wait(ofconn, handling_openflow); } - if (mgr->in_band) { - poll_timer_wait_until(mgr->next_in_band_update); + if (handling_openflow && mgr->in_band) { in_band_wait(mgr->in_band); } - if (mgr->fail_open) { + if (handling_openflow && mgr->fail_open) { fail_open_wait(mgr->fail_open); } HMAP_FOR_EACH (ofservice, node, &mgr->services) { @@ -305,16 +340,53 @@ connmgr_wait(struct connmgr *mgr) } } +/* Adds some memory usage statistics for 'mgr' into 'usage', for use with + * memory_report(). */ +void +connmgr_get_memory_usage(const struct connmgr *mgr, struct simap *usage) +{ + const struct ofconn *ofconn; + unsigned int packets = 0; + unsigned int ofconns = 0; + + LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { + int i; + + ofconns++; + + packets += rconn_count_txqlen(ofconn->rconn); + for (i = 0; i < N_SCHEDULERS; i++) { + packets += pinsched_count_txqlen(ofconn->schedulers[i]); + } + packets += pktbuf_count_packets(ofconn->pktbuf); + } + simap_increase(usage, "ofconns", ofconns); + simap_increase(usage, "packets", packets); +} + /* Returns the ofproto that owns 'ofconn''s connmgr. */ struct ofproto * ofconn_get_ofproto(const struct ofconn *ofconn) { return ofconn->connmgr->ofproto; } + +/* If processing of OpenFlow messages was blocked on any 'mgr' ofconns by + * returning false to the 'handle_openflow' callback to connmgr_run(), this + * re-enables them. */ +void +connmgr_retry(struct connmgr *mgr) +{ + struct ofconn *ofconn; + + LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { + ofconn->retry = true; + } +} /* OpenFlow configuration. */ -static void add_controller(struct connmgr *, const char *target); +static void add_controller(struct connmgr *, const char *target, uint8_t dscp); static struct ofconn *find_controller_by_target(struct connmgr *, const char *target); static void update_fail_open(struct connmgr *); @@ -342,45 +414,62 @@ connmgr_get_controller_info(struct connmgr *mgr, struct shash *info) { const struct ofconn *ofconn; - shash_init(info); - HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) { const struct rconn *rconn = ofconn->rconn; - time_t now = time_now(); - time_t last_connection = rconn_get_last_connection(rconn); - time_t last_disconnect = rconn_get_last_disconnect(rconn); - int last_error = rconn_get_last_error(rconn); - struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo); + const char *target = rconn_get_target(rconn); - shash_add(info, rconn_get_target(rconn), cinfo); + if (!shash_find(info, target)) { + struct ofproto_controller_info *cinfo = xmalloc(sizeof *cinfo); + time_t now = time_now(); + time_t last_connection = rconn_get_last_connection(rconn); + time_t last_disconnect = rconn_get_last_disconnect(rconn); + int last_error = rconn_get_last_error(rconn); - cinfo->is_connected = rconn_is_connected(rconn); - cinfo->role = ofconn->role; + shash_add(info, target, cinfo); - cinfo->pairs.n = 0; + cinfo->is_connected = rconn_is_connected(rconn); + cinfo->role = ofconn->role; - 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.n = 0; - cinfo->pairs.keys[cinfo->pairs.n] = "state"; - cinfo->pairs.values[cinfo->pairs.n++] = - xstrdup(rconn_get_state(rconn)); + if (last_error) { + cinfo->pairs.keys[cinfo->pairs.n] = "last_error"; + cinfo->pairs.values[cinfo->pairs.n++] + = xstrdup(ovs_retval_to_string(last_error)); + } - if (last_connection != TIME_MIN) { - cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_connect"; + cinfo->pairs.keys[cinfo->pairs.n] = "state"; cinfo->pairs.values[cinfo->pairs.n++] - = xasprintf("%ld", (long int) (now - last_connection)); + = xstrdup(rconn_get_state(rconn)); + + if (last_connection != TIME_MIN) { + cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_connect"; + cinfo->pairs.values[cinfo->pairs.n++] + = xasprintf("%ld", (long int) (now - last_connection)); + } + + if (last_disconnect != TIME_MIN) { + cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_disconnect"; + cinfo->pairs.values[cinfo->pairs.n++] + = xasprintf("%ld", (long int) (now - last_disconnect)); + } } + } +} - if (last_disconnect != TIME_MIN) { - cinfo->pairs.keys[cinfo->pairs.n] = "sec_since_disconnect"; - cinfo->pairs.values[cinfo->pairs.n++] - = xasprintf("%ld", (long int) (now - last_disconnect)); +void +connmgr_free_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); } /* Changes 'mgr''s set of controllers to the 'n_controllers' controllers in @@ -394,7 +483,6 @@ connmgr_set_controllers(struct connmgr *mgr, struct shash new_controllers; struct ofconn *ofconn, *next_ofconn; struct ofservice *ofservice, *next_ofservice; - bool ss_exists; size_t i; /* Create newly configured controllers and services. @@ -405,11 +493,15 @@ connmgr_set_controllers(struct connmgr *mgr, if (!vconn_verify_name(c->target)) { if (!find_controller_by_target(mgr, c->target)) { - add_controller(mgr, c->target); + VLOG_INFO("%s: added primary controller \"%s\"", + mgr->name, c->target); + add_controller(mgr, c->target, c->dscp); } } else if (!pvconn_verify_name(c->target)) { if (!ofservice_lookup(mgr, c->target)) { - ofservice_create(mgr, c->target); + VLOG_INFO("%s: added service controller \"%s\"", + mgr->name, c->target); + ofservice_create(mgr, c->target, c->dscp); } } else { VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"", @@ -422,12 +514,14 @@ connmgr_set_controllers(struct connmgr *mgr, /* Delete controllers that are no longer configured. * Update configuration of all now-existing controllers. */ - ss_exists = false; HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &mgr->controllers) { + const char *target = ofconn_get_target(ofconn); struct ofproto_controller *c; - c = shash_find_data(&new_controllers, ofconn_get_target(ofconn)); + c = shash_find_data(&new_controllers, target); if (!c) { + VLOG_INFO("%s: removed primary controller \"%s\"", + mgr->name, target); ofconn_destroy(ofconn); } else { ofconn_reconfigure(ofconn, c); @@ -437,11 +531,13 @@ connmgr_set_controllers(struct connmgr *mgr, /* Delete services that are no longer configured. * Update configuration of all now-existing services. */ HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &mgr->services) { + const char *target = pvconn_get_name(ofservice->pvconn); struct ofproto_controller *c; - c = shash_find_data(&new_controllers, - pvconn_get_name(ofservice->pvconn)); + c = shash_find_data(&new_controllers, target); if (!c) { + VLOG_INFO("%s: removed service controller \"%s\"", + mgr->name, target); ofservice_destroy(mgr, ofservice); } else { ofservice_reconfigure(ofservice, c); @@ -500,14 +596,13 @@ connmgr_has_snoops(const struct connmgr *mgr) /* Creates a new controller for 'target' in 'mgr'. update_controller() needs * to be called later to finish the new ofconn's configuration. */ static void -add_controller(struct connmgr *mgr, const char *target) +add_controller(struct connmgr *mgr, const char *target, uint8_t dscp) { char *name = ofconn_make_name(mgr, target); struct ofconn *ofconn; - ofconn = ofconn_create(mgr, rconn_create(5, 8), OFCONN_PRIMARY); + ofconn = ofconn_create(mgr, rconn_create(5, 8, dscp), OFCONN_PRIMARY, true); ofconn->pktbuf = pktbuf_create(); - ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN; rconn_connect(ofconn->rconn, target, name); hmap_insert(&mgr->controllers, &ofconn->hmap_node, hash_string(target, 0)); @@ -544,14 +639,16 @@ update_in_band_remotes(struct connmgr *mgr) /* Add all the remotes. */ HMAP_FOR_EACH (ofconn, hmap_node, &mgr->controllers) { struct sockaddr_in *sin = &addrs[n_addrs]; + const char *target = rconn_get_target(ofconn->rconn); if (ofconn->band == OFPROTO_OUT_OF_BAND) { continue; } - sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn); - if (sin->sin_addr.s_addr) { - sin->sin_port = rconn_get_remote_port(ofconn->rconn); + if (stream_parse_target_with_default_ports(target, + OFP_TCP_PORT, + OFP_SSL_PORT, + sin)) { n_addrs++; } } @@ -564,14 +661,13 @@ update_in_band_remotes(struct connmgr *mgr) if (!mgr->in_band) { in_band_create(mgr->ofproto, mgr->local_port_name, &mgr->in_band); } - if (mgr->in_band) { - in_band_set_remotes(mgr->in_band, addrs, n_addrs); - } in_band_set_queue(mgr->in_band, mgr->in_band_queue); - mgr->next_in_band_update = time_msec() + 1000; } else { - in_band_destroy(mgr->in_band); - mgr->in_band = NULL; + /* in_band_run() needs a chance to delete any existing in-band flows. + * We will destroy mgr->in_band after it's done with that. */ + } + if (mgr->in_band) { + in_band_set_remotes(mgr->in_band, addrs, n_addrs); } /* Clean up. */ @@ -613,7 +709,7 @@ set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp, struct pvconn *pvconn; int error; - error = pvconn_open(name, &pvconn); + error = pvconn_open(name, &pvconn, 0); if (!error) { pvconns[n_pvconns++] = pvconn; } else { @@ -708,20 +804,70 @@ ofconn_set_role(struct ofconn *ofconn, enum nx_role role) ofconn->role = role; } -/* Returns the currently configured flow format for 'ofconn', one of NXFF_*. +void +ofconn_set_invalid_ttl_to_controller(struct ofconn *ofconn, bool enable) +{ + uint32_t bit = 1u << OFPR_INVALID_TTL; + if (enable) { + ofconn->master_async_config[OAM_PACKET_IN] |= bit; + } else { + ofconn->master_async_config[OAM_PACKET_IN] &= ~bit; + } +} + +bool +ofconn_get_invalid_ttl_to_controller(struct ofconn *ofconn) +{ + uint32_t bit = 1u << OFPR_INVALID_TTL; + return (ofconn->master_async_config[OAM_PACKET_IN] & bit) != 0; +} + +/* Returns the currently configured protocol for 'ofconn', one of OFPUTIL_P_*. + * + * The default, if no other format has been set, is OFPUTIL_P_OPENFLOW10. */ +enum ofputil_protocol +ofconn_get_protocol(struct ofconn *ofconn) +{ + return ofconn->protocol; +} + +/* Sets the protocol for 'ofconn' to 'protocol' (one of OFPUTIL_P_*). + * + * (This doesn't actually send anything to accomplish this. Presumably the + * caller already did that.) */ +void +ofconn_set_protocol(struct ofconn *ofconn, enum ofputil_protocol protocol) +{ + ofconn->protocol = protocol; +} + +/* Returns the currently configured packet in format for 'ofconn', one of + * NXPIF_*. * - * The default, if no other format has been set, is NXFF_OPENFLOW10. */ -enum nx_flow_format -ofconn_get_flow_format(struct ofconn *ofconn) + * The default, if no other format has been set, is NXPIF_OPENFLOW10. */ +enum nx_packet_in_format +ofconn_get_packet_in_format(struct ofconn *ofconn) +{ + return ofconn->packet_in_format; +} + +/* Sets the packet in format for 'ofconn' to 'packet_in_format' (one of + * NXPIF_*). */ +void +ofconn_set_packet_in_format(struct ofconn *ofconn, + enum nx_packet_in_format packet_in_format) { - return ofconn->flow_format; + ofconn->packet_in_format = packet_in_format; } -/* Sets the flow format for 'ofconn' to 'flow_format' (one of NXFF_*). */ +/* Sets the controller connection ID for 'ofconn' to 'controller_id'. + * + * The connection controller ID is used for OFPP_CONTROLLER and + * NXAST_CONTROLLER actions. See "struct nx_action_controller" for details. */ void -ofconn_set_flow_format(struct ofconn *ofconn, enum nx_flow_format flow_format) +ofconn_set_controller_id(struct ofconn *ofconn, uint16_t controller_id) { - ofconn->flow_format = flow_format; + ofconn->controller_id = controller_id; } /* Returns the default miss send length for 'ofconn'. */ @@ -738,6 +884,16 @@ ofconn_set_miss_send_len(struct ofconn *ofconn, int miss_send_len) ofconn->miss_send_len = miss_send_len; } +void +ofconn_set_async_config(struct ofconn *ofconn, + const uint32_t master_masks[OAM_N_TYPES], + const uint32_t slave_masks[OAM_N_TYPES]) +{ + size_t size = sizeof ofconn->master_async_config; + memcpy(ofconn->master_async_config, master_masks, size); + memcpy(ofconn->slave_async_config, slave_masks, size); +} + /* Sends 'msg' on 'ofconn', accounting it as a reply. (If there is a * sufficient number of OpenFlow replies in-flight on a single ofconn, then the * connmgr will stop accepting new OpenFlow requests on that ofconn until the @@ -748,13 +904,80 @@ ofconn_send_reply(const struct ofconn *ofconn, struct ofpbuf *msg) ofconn_send(ofconn, msg, ofconn->reply_counter); } +/* Sends each of the messages in list 'replies' on 'ofconn' in order, + * accounting them as replies. */ +void +ofconn_send_replies(const struct ofconn *ofconn, struct list *replies) +{ + struct ofpbuf *reply, *next; + + LIST_FOR_EACH_SAFE (reply, next, list_node, replies) { + list_remove(&reply->list_node); + ofconn_send_reply(ofconn, reply); + } +} + +/* Sends 'error' on 'ofconn', as a reply to 'request'. Only at most the + * first 64 bytes of 'request' are used. */ +void +ofconn_send_error(const struct ofconn *ofconn, + const struct ofp_header *request, enum ofperr error) +{ + struct ofpbuf *reply; + + reply = ofperr_encode_reply(error, request); + if (reply) { + static struct vlog_rate_limit err_rl = VLOG_RATE_LIMIT_INIT(10, 10); + + if (!VLOG_DROP_INFO(&err_rl)) { + const struct ofputil_msg_type *type; + const char *type_name; + size_t request_len; + + request_len = ntohs(request->length); + type_name = (!ofputil_decode_msg_type_partial(request, + MIN(64, request_len), + &type) + ? ofputil_msg_type_name(type) + : "invalid"); + + VLOG_INFO("%s: sending %s error reply to %s message", + rconn_get_name(ofconn->rconn), ofperr_to_string(error), + type_name); + } + ofconn_send_reply(ofconn, reply); + } +} + /* Same as pktbuf_retrieve(), using the pktbuf owned by 'ofconn'. */ -int +enum ofperr ofconn_pktbuf_retrieve(struct ofconn *ofconn, uint32_t id, struct ofpbuf **bufferp, uint16_t *in_port) { return pktbuf_retrieve(ofconn->pktbuf, id, bufferp, in_port); } + +/* Returns true if 'ofconn' has any pending opgroups. */ +bool +ofconn_has_pending_opgroups(const struct ofconn *ofconn) +{ + return !list_is_empty(&ofconn->opgroups); +} + +/* Adds 'ofconn_node' to 'ofconn''s list of pending opgroups. + * + * If 'ofconn' is destroyed or its connection drops, then 'ofconn' will remove + * 'ofconn_node' from the list and re-initialize it with list_init(). The + * client may, therefore, use list_is_empty(ofconn_node) to determine whether + * 'ofconn_node' is still associated with an active ofconn. + * + * The client may also remove ofconn_node from the list itself, with + * list_remove(). */ +void +ofconn_add_opgroup(struct ofconn *ofconn, struct list *ofconn_node) +{ + list_push_back(&ofconn->opgroups, ofconn_node); +} /* Private ofconn functions. */ @@ -765,25 +988,105 @@ ofconn_get_target(const struct ofconn *ofconn) } static struct ofconn * -ofconn_create(struct connmgr *mgr, struct rconn *rconn, enum ofconn_type type) +ofconn_create(struct connmgr *mgr, struct rconn *rconn, enum ofconn_type type, + bool enable_async_msgs) { - struct ofconn *ofconn = xzalloc(sizeof *ofconn); + struct ofconn *ofconn; + + ofconn = xzalloc(sizeof *ofconn); ofconn->connmgr = mgr; list_push_back(&mgr->all_conns, &ofconn->node); ofconn->rconn = rconn; ofconn->type = type; - ofconn->flow_format = NXFF_OPENFLOW10; - ofconn->role = NX_ROLE_OTHER; - ofconn->packet_in_counter = rconn_packet_counter_create (); - ofconn->pktbuf = NULL; - ofconn->miss_send_len = 0; - ofconn->reply_counter = rconn_packet_counter_create (); + ofconn->enable_async_msgs = enable_async_msgs; + + list_init(&ofconn->opgroups); + + ofconn_flush(ofconn); + return ofconn; } +/* Clears all of the state in 'ofconn' that should not persist from one + * connection to the next. */ +static void +ofconn_flush(struct ofconn *ofconn) +{ + int i; + + ofconn->role = NX_ROLE_OTHER; + ofconn->protocol = OFPUTIL_P_OF10; + ofconn->packet_in_format = NXPIF_OPENFLOW10; + + /* Disassociate 'ofconn' from all of the ofopgroups that it initiated that + * have not yet completed. (Those ofopgroups will still run to completion + * in the usual way, but any errors that they run into will not be reported + * on any OpenFlow channel.) + * + * Also discard any blocked operation on 'ofconn'. */ + while (!list_is_empty(&ofconn->opgroups)) { + list_init(list_pop_front(&ofconn->opgroups)); + } + ofpbuf_delete(ofconn->blocked); + ofconn->blocked = NULL; + + rconn_packet_counter_destroy(ofconn->packet_in_counter); + ofconn->packet_in_counter = rconn_packet_counter_create(); + for (i = 0; i < N_SCHEDULERS; i++) { + if (ofconn->schedulers[i]) { + int rate, burst; + + pinsched_get_limits(ofconn->schedulers[i], &rate, &burst); + pinsched_destroy(ofconn->schedulers[i]); + ofconn->schedulers[i] = pinsched_create(rate, burst); + } + } + if (ofconn->pktbuf) { + pktbuf_destroy(ofconn->pktbuf); + ofconn->pktbuf = pktbuf_create(); + } + ofconn->miss_send_len = (ofconn->type == OFCONN_PRIMARY + ? OFP_DEFAULT_MISS_SEND_LEN + : 0); + ofconn->controller_id = 0; + + rconn_packet_counter_destroy(ofconn->reply_counter); + ofconn->reply_counter = rconn_packet_counter_create(); + + if (ofconn->enable_async_msgs) { + uint32_t *master = ofconn->master_async_config; + uint32_t *slave = ofconn->slave_async_config; + + /* "master" and "other" roles get all asynchronous messages by default, + * except that the controller needs to enable nonstandard "packet-in" + * reasons itself. */ + master[OAM_PACKET_IN] = (1u << OFPR_NO_MATCH) | (1u << OFPR_ACTION); + master[OAM_PORT_STATUS] = ((1u << OFPPR_ADD) + | (1u << OFPPR_DELETE) + | (1u << OFPPR_MODIFY)); + master[OAM_FLOW_REMOVED] = ((1u << OFPRR_IDLE_TIMEOUT) + | (1u << OFPRR_HARD_TIMEOUT) + | (1u << OFPRR_DELETE)); + + /* "slave" role gets port status updates by default. */ + slave[OAM_PACKET_IN] = 0; + slave[OAM_PORT_STATUS] = ((1u << OFPPR_ADD) + | (1u << OFPPR_DELETE) + | (1u << OFPPR_MODIFY)); + slave[OAM_FLOW_REMOVED] = 0; + } else { + memset(ofconn->master_async_config, 0, + sizeof ofconn->master_async_config); + memset(ofconn->slave_async_config, 0, + sizeof ofconn->slave_async_config); + } +} + static void ofconn_destroy(struct ofconn *ofconn) { + ofconn_flush(ofconn); + if (ofconn->type == OFCONN_PRIMARY) { hmap_remove(&ofconn->connmgr->controllers, &ofconn->hmap_node); } @@ -804,6 +1107,7 @@ ofconn_reconfigure(struct ofconn *ofconn, const struct ofproto_controller *c) int probe_interval; ofconn->band = c->band; + ofconn->enable_async_msgs = c->enable_async_msgs; rconn_set_max_backoff(ofconn->rconn, c->max_backoff); @@ -811,14 +1115,28 @@ ofconn_reconfigure(struct ofconn *ofconn, const struct ofproto_controller *c) rconn_set_probe_interval(ofconn->rconn, probe_interval); ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit); + + /* If dscp value changed reconnect. */ + if (c->dscp != rconn_get_dscp(ofconn->rconn)) { + rconn_set_dscp(ofconn->rconn, c->dscp); + rconn_reconnect(ofconn->rconn); + } +} + +/* Returns true if it makes sense for 'ofconn' to receive and process OpenFlow + * messages. */ +static bool +ofconn_may_recv(const struct ofconn *ofconn) +{ + int count = rconn_packet_counter_read (ofconn->reply_counter); + return (!ofconn->blocked || ofconn->retry) && count < OFCONN_REPLY_MAX; } static void ofconn_run(struct ofconn *ofconn, - void (*handle_openflow)(struct ofconn *, struct ofpbuf *ofp_msg)) + bool (*handle_openflow)(struct ofconn *, struct ofpbuf *ofp_msg)) { struct connmgr *mgr = ofconn->connmgr; - int iteration; size_t i; for (i = 0; i < N_SCHEDULERS; i++) { @@ -827,29 +1145,40 @@ ofconn_run(struct ofconn *ofconn, rconn_run(ofconn->rconn); - if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { - /* Limit the number of iterations to prevent other tasks from - * starving. */ - for (iteration = 0; iteration < 50; iteration++) { - struct ofpbuf *of_msg = rconn_recv(ofconn->rconn); + if (handle_openflow) { + /* Limit the number of iterations to avoid starving other tasks. */ + for (i = 0; i < 50 && ofconn_may_recv(ofconn); i++) { + struct ofpbuf *of_msg; + + of_msg = (ofconn->blocked + ? ofconn->blocked + : rconn_recv(ofconn->rconn)); if (!of_msg) { break; } if (mgr->fail_open) { fail_open_maybe_recover(mgr->fail_open); } - handle_openflow(ofconn, of_msg); - ofpbuf_delete(of_msg); + + if (handle_openflow(ofconn, of_msg)) { + ofpbuf_delete(of_msg); + ofconn->blocked = NULL; + } else { + ofconn->blocked = of_msg; + ofconn->retry = false; + } } } if (!rconn_is_alive(ofconn->rconn)) { ofconn_destroy(ofconn); + } else if (!rconn_is_connected(ofconn->rconn)) { + ofconn_flush(ofconn); } } static void -ofconn_wait(struct ofconn *ofconn) +ofconn_wait(struct ofconn *ofconn, bool handling_openflow) { int i; @@ -857,28 +1186,47 @@ ofconn_wait(struct ofconn *ofconn) pinsched_wait(ofconn->schedulers[i]); } rconn_run_wait(ofconn->rconn); - if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { + if (handling_openflow && ofconn_may_recv(ofconn)) { rconn_recv_wait(ofconn->rconn); - } else { - COVERAGE_INC(ofconn_stuck); } } -/* Returns true if 'ofconn' should receive asynchronous messages. */ +/* Returns true if 'ofconn' should receive asynchronous messages of the given + * OAM_* 'type' and 'reason', which should be a OFPR_* value for OAM_PACKET_IN, + * a OFPPR_* value for OAM_PORT_STATUS, or an OFPRR_* value for + * OAM_FLOW_REMOVED. Returns false if the message should not be sent on + * 'ofconn'. */ static bool -ofconn_receives_async_msgs(const struct ofconn *ofconn) +ofconn_receives_async_msg(const struct ofconn *ofconn, + enum ofconn_async_msg_type type, + unsigned int reason) { + const uint32_t *async_config; + + assert(reason < 32); + assert((unsigned int) type < OAM_N_TYPES); + if (!rconn_is_connected(ofconn->rconn)) { return false; - } else if (ofconn->type == OFCONN_PRIMARY) { - /* Primary controllers always get asynchronous messages unless they - * have configured themselves as "slaves". */ - return ofconn->role != NX_ROLE_SLAVE; - } else { + } + + /* Keep the following code in sync with the documentation in the + * "Asynchronous Messages" section in DESIGN. */ + + if (ofconn->type == OFCONN_SERVICE && !ofconn->miss_send_len) { /* Service connections don't get asynchronous messages unless they have * explicitly asked for them by setting a nonzero miss send length. */ - return ofconn->miss_send_len > 0; + return false; } + + async_config = (ofconn->role == NX_ROLE_SLAVE + ? ofconn->slave_async_config + : ofconn->master_async_config); + if (!(async_config[type] & (1u << reason))) { + return false; + } + + return true; } /* Returns a human-readable name for an OpenFlow connection between 'mgr' and @@ -918,40 +1266,32 @@ ofconn_send(const struct ofconn *ofconn, struct ofpbuf *msg, struct rconn_packet_counter *counter) { update_openflow_length(msg); - if (rconn_send(ofconn->rconn, msg, counter)) { - ofpbuf_delete(msg); - } + rconn_send(ofconn->rconn, msg, counter); } /* Sending asynchronous messages. */ -static void schedule_packet_in(struct ofconn *, const struct dpif_upcall *, - const struct flow *, struct ofpbuf *rw_packet); +static void schedule_packet_in(struct ofconn *, struct ofputil_packet_in); /* Sends an OFPT_PORT_STATUS message with 'opp' and 'reason' to appropriate * controllers managed by 'mgr'. */ void -connmgr_send_port_status(struct connmgr *mgr, const struct ofp_phy_port *opp, - uint8_t reason) +connmgr_send_port_status(struct connmgr *mgr, + const struct ofputil_phy_port *pp, uint8_t reason) { /* XXX Should limit the number of queued port status change messages. */ + struct ofputil_port_status ps; struct ofconn *ofconn; + ps.reason = reason; + ps.desc = *pp; LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { - struct ofp_port_status *ops; - struct ofpbuf *b; + if (ofconn_receives_async_msg(ofconn, OAM_PORT_STATUS, reason)) { + struct ofpbuf *msg; - /* Primary controllers, even slaves, should always get port status - updates. Otherwise obey ofconn_receives_async_msgs(). */ - if (ofconn->type != OFCONN_PRIMARY - && !ofconn_receives_async_msgs(ofconn)) { - continue; + msg = ofputil_encode_port_status(&ps, ofconn->protocol); + ofconn_send(ofconn, msg, NULL); } - - ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b); - ops->reason = reason; - ops->desc = *opp; - ofconn_send(ofconn, b, NULL); } } @@ -964,49 +1304,36 @@ connmgr_send_flow_removed(struct connmgr *mgr, struct ofconn *ofconn; LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { - struct ofpbuf *msg; - - if (!ofconn_receives_async_msgs(ofconn)) { - continue; + if (ofconn_receives_async_msg(ofconn, OAM_FLOW_REMOVED, fr->reason)) { + struct ofpbuf *msg; + + /* Account flow expirations as replies to OpenFlow requests. That + * works because preventing OpenFlow requests from being processed + * also prevents new flows from being added (and expiring). (It + * also prevents processing OpenFlow requests that would not add + * new flows, so it is imperfect.) */ + msg = ofputil_encode_flow_removed(fr, ofconn->protocol); + ofconn_send_reply(ofconn, msg); } - - /* Account flow expirations as replies to OpenFlow requests. That - * works because preventing OpenFlow requests from being processed also - * prevents new flows from being added (and expiring). (It also - * prevents processing OpenFlow requests that would not add new flows, - * so it is imperfect.) */ - msg = ofputil_encode_flow_removed(fr, ofconn->flow_format); - ofconn_send_reply(ofconn, msg); } } -/* 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. +/* Given 'pin', sends an OFPT_PACKET_IN message to each OpenFlow controller as + * necessary according to their individual configurations. * - * 'rw_packet' may be NULL. Otherwise, 'rw_packet' must contain the same data - * as upcall->packet. (rw_packet == upcall->packet is also valid.) Ownership - * of 'rw_packet' is transferred to this function. */ + * The caller doesn't need to fill in pin->buffer_id or pin->total_len. */ void -connmgr_send_packet_in(struct connmgr *mgr, const struct dpif_upcall *upcall, - const struct flow *flow, struct ofpbuf *rw_packet) +connmgr_send_packet_in(struct connmgr *mgr, + const struct ofputil_packet_in *pin) { - struct ofconn *ofconn, *prev; + struct ofconn *ofconn; - prev = NULL; LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { - if (ofconn_receives_async_msgs(ofconn)) { - if (prev) { - schedule_packet_in(prev, upcall, flow, NULL); - } - prev = ofconn; + if (ofconn_receives_async_msg(ofconn, OAM_PACKET_IN, pin->reason) + && ofconn->controller_id == pin->controller_id) { + schedule_packet_in(ofconn, *pin); } } - if (prev) { - schedule_packet_in(prev, upcall, flow, rw_packet); - } else { - ofpbuf_delete(rw_packet); - } } /* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */ @@ -1019,51 +1346,44 @@ do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn_) ofconn->packet_in_counter, 100); } -/* 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. - * - * 'rw_packet' may be NULL. Otherwise, 'rw_packet' must contain the same data - * as upcall->packet. (rw_packet == upcall->packet is also valid.) Ownership - * of 'rw_packet' is transferred to this function. */ +/* Takes 'pin', composes an OpenFlow packet-in message from it, and passes it + * to 'ofconn''s packet scheduler for sending. */ static void -schedule_packet_in(struct ofconn *ofconn, const struct dpif_upcall *upcall, - const struct flow *flow, struct ofpbuf *rw_packet) +schedule_packet_in(struct ofconn *ofconn, struct ofputil_packet_in pin) { struct connmgr *mgr = ofconn->connmgr; - struct ofputil_packet_in pin; - /* Figure out the easy parts. */ - pin.packet = upcall->packet; - pin.in_port = odp_port_to_ofp_port(flow->in_port); - pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION; + pin.total_len = pin.packet_len; /* Get OpenFlow buffer_id. */ - if (upcall->type == DPIF_UC_ACTION) { + if (pin.reason == OFPR_ACTION) { pin.buffer_id = UINT32_MAX; } else if (mgr->fail_open && fail_open_is_active(mgr->fail_open)) { pin.buffer_id = pktbuf_get_null(); } else if (!ofconn->pktbuf) { pin.buffer_id = UINT32_MAX; } else { - pin.buffer_id = pktbuf_save(ofconn->pktbuf, upcall->packet, - flow->in_port); + pin.buffer_id = pktbuf_save(ofconn->pktbuf, pin.packet, pin.packet_len, + pin.fmd.in_port); } /* Figure out how much of the packet to send. */ - pin.send_len = upcall->packet->size; + if (pin.reason == OFPR_NO_MATCH) { + pin.send_len = pin.packet_len; + } else { + /* Caller should have initialized 'send_len' to 'max_len' specified in + * output action. */ + } if (pin.buffer_id != UINT32_MAX) { pin.send_len = MIN(pin.send_len, ofconn->miss_send_len); } - if (upcall->type == DPIF_UC_ACTION) { - pin.send_len = MIN(pin.send_len, upcall->userdata); - } /* Make OFPT_PACKET_IN and 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[upcall->type == DPIF_UC_MISS ? 0 : 1], - flow->in_port, ofputil_encode_packet_in(&pin, rw_packet), + pinsched_send(ofconn->schedulers[pin.reason == OFPR_NO_MATCH ? 0 : 1], + pin.fmd.in_port, + ofputil_encode_packet_in(&pin, ofconn->packet_in_format), do_send_packet_in, ofconn); } @@ -1159,29 +1479,6 @@ connmgr_is_any_controller_admitted(const struct connmgr *mgr) } return false; } - -/* Sends 'packet' to each controller connected to 'mgr'. Takes ownership of - * 'packet'. */ -void -connmgr_broadcast(struct connmgr *mgr, struct ofpbuf *packet) -{ - struct ofconn *ofconn, *prev; - - prev = NULL; - LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { - if (prev) { - ofconn_send_reply(ofconn, ofpbuf_clone(packet)); - } - if (rconn_is_connected(ofconn->rconn)) { - prev = ofconn; - } - } - if (prev) { - ofconn_send_reply(prev, packet); - } else { - ofpbuf_delete(packet); - } -} /* In-band configuration. */ @@ -1261,13 +1558,12 @@ connmgr_may_set_up_flow(struct connmgr *mgr, const struct flow *flow, /* Fail-open and in-band implementation. */ /* Called by 'ofproto' after all flows have been flushed, to allow fail-open - * and in-band control to re-create their flows. */ + * and standalone mode to re-create their flows. + * + * In-band control has more sophisticated code that manages flows itself. */ void connmgr_flushed(struct connmgr *mgr) { - if (mgr->in_band) { - in_band_flushed(mgr->in_band); - } if (mgr->fail_open) { fail_open_flushed(mgr->fail_open); } @@ -1278,15 +1574,17 @@ connmgr_flushed(struct connmgr *mgr) * traffic until a controller has been defined and it tells us to do so. */ if (!connmgr_has_controllers(mgr) && mgr->fail_mode == OFPROTO_FAIL_STANDALONE) { - union ofp_action action; + struct ofpbuf ofpacts; struct cls_rule rule; - memset(&action, 0, sizeof action); - action.type = htons(OFPAT_OUTPUT); - action.output.len = htons(sizeof action); - action.output.port = htons(OFPP_NORMAL); + ofpbuf_init(&ofpacts, OFPACT_OUTPUT_SIZE); + ofpact_put_OUTPUT(&ofpacts)->port = OFPP_NORMAL; + ofpact_pad(&ofpacts); + cls_rule_init_catchall(&rule, 0); - ofproto_add_flow(mgr->ofproto, &rule, &action, 1); + ofproto_add_flow(mgr->ofproto, &rule, ofpacts.data, ofpacts.size); + + ofpbuf_uninit(&ofpacts); } } @@ -1296,13 +1594,13 @@ connmgr_flushed(struct connmgr *mgr) * ofservice_reconfigure() must be called to fully configure the new * ofservice. */ static int -ofservice_create(struct connmgr *mgr, const char *target) +ofservice_create(struct connmgr *mgr, const char *target, uint8_t dscp) { struct ofservice *ofservice; struct pvconn *pvconn; int error; - error = pvconn_open(target, &pvconn); + error = pvconn_open(target, &pvconn, dscp); if (error) { return error; } @@ -1329,6 +1627,8 @@ ofservice_reconfigure(struct ofservice *ofservice, ofservice->probe_interval = c->probe_interval; ofservice->rate_limit = c->rate_limit; ofservice->burst_limit = c->burst_limit; + ofservice->enable_async_msgs = c->enable_async_msgs; + ofservice->dscp = c->dscp; } /* Finds and returns the ofservice within 'mgr' that has the given