X-Git-Url: http://git.cascardo.eti.br/?a=blobdiff_plain;f=ofproto%2Fconnmgr.c;h=295c03ca64500d50089f5e7bbad0a8092007c40b;hb=af7bc161f5ea9b6ae3e1405d6c619b734c2529a7;hp=29321d50f6bd462493f3852be3558315e7864e99;hpb=f4f1ea7eaca92e2ca44c9624b3bb7d6426b2ddea;p=cascardo%2Fovs.git diff --git a/ofproto/connmgr.c b/ofproto/connmgr.c index 29321d50f..295c03ca6 100644 --- a/ofproto/connmgr.c +++ b/ofproto/connmgr.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc. + * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 Nicira, Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -22,6 +22,7 @@ #include #include "coverage.h" +#include "dynamic-string.h" #include "fail-open.h" #include "in-band.h" #include "odp-util.h" @@ -38,17 +39,28 @@ #include "simap.h" #include "stream.h" #include "timeval.h" -#include "vconn.h" -#include "vlog.h" +#include "openvswitch/vconn.h" +#include "openvswitch/vlog.h" + +#include "bundles.h" VLOG_DEFINE_THIS_MODULE(connmgr); static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); -/* An OpenFlow connection. */ +/* An OpenFlow connection. + * + * + * Thread-safety + * ============= + * + * 'ofproto_mutex' must be held whenever an ofconn is created or destroyed or, + * more or less equivalently, whenever an ofconn is added to or removed from a + * connmgr. 'ofproto_mutex' doesn't protect the data inside the ofconn, except + * as specifically noted below. */ struct ofconn { /* Configuration that persists from one connection to the next. */ - struct list node; /* In struct connmgr's "all_conns" list. */ + struct ovs_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. */ @@ -64,11 +76,6 @@ struct ofconn { 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'. */ #define N_SCHEDULERS 2 @@ -83,33 +90,68 @@ struct ofconn { #define OFCONN_REPLY_MAX 100 struct rconn_packet_counter *reply_counter; - /* Asynchronous message configuration in each possible roles. + /* Asynchronous message configuration in each possible role. * * 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 */ - - /* Flow monitors. */ - struct hmap monitors; /* Contains "struct ofmonitor"s. */ - struct list updates; /* List of "struct ofpbuf"s. */ - bool sent_abbrev_update; /* Does 'updates' contain NXFME_ABBREV? */ - struct rconn_packet_counter *monitor_counter; - uint64_t monitor_paused; + struct ofputil_async_cfg *async_cfg; + + /* Flow table operation logging. */ + int n_add, n_delete, n_modify; /* Number of unreported ops of each kind. */ + long long int first_op, last_op; /* Range of times for unreported ops. */ + long long int next_op_report; /* Time to report ops, or LLONG_MAX. */ + long long int op_backoff; /* Earliest time to report ops again. */ + +/* Flow monitors (e.g. NXST_FLOW_MONITOR). */ + + /* Configuration. Contains "struct ofmonitor"s. */ + struct hmap monitors OVS_GUARDED_BY(ofproto_mutex); + + /* Flow control. + * + * When too many flow monitor notifications back up in the transmit buffer, + * we pause the transmission of further notifications. These members track + * the flow control state. + * + * When notifications are flowing, 'monitor_paused' is 0. When + * notifications are paused, 'monitor_paused' is the value of + * 'monitor_seqno' at the point we paused. + * + * 'monitor_counter' counts the OpenFlow messages and bytes currently in + * flight. This value growing too large triggers pausing. */ + uint64_t monitor_paused OVS_GUARDED_BY(ofproto_mutex); + struct rconn_packet_counter *monitor_counter OVS_GUARDED_BY(ofproto_mutex); + + /* State of monitors for a single ongoing flow_mod. + * + * 'updates' is a list of "struct ofpbuf"s that contain + * NXST_FLOW_MONITOR_REPLY messages representing the changes made by the + * current flow_mod. + * + * When 'updates' is nonempty, 'sent_abbrev_update' is true if 'updates' + * contains an update event of type NXFME_ABBREV and false otherwise.. */ + struct ovs_list updates OVS_GUARDED_BY(ofproto_mutex); + bool sent_abbrev_update OVS_GUARDED_BY(ofproto_mutex); + + /* Active bundles. Contains "struct ofp_bundle"s. */ + struct hmap bundles; }; static struct ofconn *ofconn_create(struct connmgr *, struct rconn *, - enum ofconn_type, bool enable_async_msgs); -static void ofconn_destroy(struct ofconn *); -static void ofconn_flush(struct ofconn *); + enum ofconn_type, bool enable_async_msgs) + OVS_REQUIRES(ofproto_mutex); +static void ofconn_destroy(struct ofconn *) OVS_REQUIRES(ofproto_mutex); +static void ofconn_flush(struct ofconn *) OVS_REQUIRES(ofproto_mutex); static void ofconn_reconfigure(struct ofconn *, const struct ofproto_controller *); static void ofconn_run(struct ofconn *, - bool (*handle_openflow)(struct ofconn *, - struct ofpbuf *ofp_msg)); -static void ofconn_wait(struct ofconn *, bool handling_openflow); + void (*handle_openflow)(struct ofconn *, + const struct ofpbuf *ofp_msg)); +static void ofconn_wait(struct ofconn *); + +static void ofconn_log_flow_mods(struct ofconn *); static const char *ofconn_get_target(const struct ofconn *); static char *ofconn_make_name(const struct connmgr *, const char *target); @@ -119,7 +161,7 @@ static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst); static void ofconn_send(const struct ofconn *, struct ofpbuf *, struct rconn_packet_counter *); -static void do_send_packet_in(struct ofpbuf *, void *ofconn_); +static void do_send_packet_ins(struct ofconn *, struct ovs_list *txq); /* A listener for incoming OpenFlow "service" connections. */ struct ofservice { @@ -152,8 +194,8 @@ struct connmgr { char *local_port_name; /* OpenFlow connections. */ - struct hmap controllers; /* Controller "struct ofconn"s. */ - struct list all_conns; /* Contains "struct ofconn"s. */ + struct hmap controllers; /* All OFCONN_PRIMARY controllers. */ + struct ovs_list all_conns; /* All controllers. */ uint64_t master_election_id; /* monotonically increasing sequence number * for master election */ bool master_election_id_defined; @@ -226,9 +268,12 @@ connmgr_destroy(struct connmgr *mgr) return; } + ovs_mutex_lock(&ofproto_mutex); LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &mgr->all_conns) { ofconn_destroy(ofconn); } + ovs_mutex_unlock(&ofproto_mutex); + hmap_destroy(&mgr->controllers); HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &mgr->services) { @@ -253,29 +298,21 @@ connmgr_destroy(struct connmgr *mgr) free(mgr); } -/* 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. */ +/* 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. + * 'handle_openflow' must not modify or free the message. */ void connmgr_run(struct connmgr *mgr, - bool (*handle_openflow)(struct ofconn *, struct ofpbuf *ofp_msg)) + void (*handle_openflow)(struct ofconn *, + const struct ofpbuf *ofp_msg)) + OVS_EXCLUDED(ofproto_mutex) { struct ofconn *ofconn, *next_ofconn; struct ofservice *ofservice; size_t i; - if (handle_openflow && mgr->in_band) { + if (mgr->in_band) { if (!in_band_run(mgr->in_band)) { in_band_destroy(mgr->in_band); mgr->in_band = NULL; @@ -289,7 +326,7 @@ 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 (handle_openflow && mgr->fail_open) { + if (mgr->fail_open) { fail_open_run(mgr->fail_open); } @@ -309,12 +346,15 @@ connmgr_run(struct connmgr *mgr, rconn_connect_unreliably(rconn, vconn, name); free(name); + ovs_mutex_lock(&ofproto_mutex); ofconn = ofconn_create(mgr, rconn, OFCONN_SERVICE, ofservice->enable_async_msgs); + ovs_mutex_unlock(&ofproto_mutex); + ofconn_set_rate_limit(ofconn, ofservice->rate_limit, ofservice->burst_limit); } else if (retval != EAGAIN) { - VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); + VLOG_WARN_RL(&rl, "accept failed (%s)", ovs_strerror(retval)); } } @@ -326,31 +366,27 @@ connmgr_run(struct connmgr *mgr, if (!retval) { add_snooper(mgr, vconn); } else if (retval != EAGAIN) { - VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); + VLOG_WARN_RL(&rl, "accept failed (%s)", ovs_strerror(retval)); } } } -/* 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. */ +/* Causes the poll loop to wake up when connmgr_run() needs to run. */ void -connmgr_wait(struct connmgr *mgr, bool handling_openflow) +connmgr_wait(struct connmgr *mgr) { struct ofservice *ofservice; struct ofconn *ofconn; size_t i; LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { - ofconn_wait(ofconn, handling_openflow); + ofconn_wait(ofconn); } ofmonitor_wait(mgr); - if (handling_openflow && mgr->in_band) { + if (mgr->in_band) { in_band_wait(mgr->in_band); } - if (handling_openflow && mgr->fail_open) { + if (mgr->fail_open) { fail_open_wait(mgr->fail_open); } HMAP_FOR_EACH (ofservice, node, &mgr->services) { @@ -377,7 +413,10 @@ connmgr_get_memory_usage(const struct connmgr *mgr, struct simap *usage) packets += rconn_count_txqlen(ofconn->rconn); for (i = 0; i < N_SCHEDULERS; i++) { - packets += pinsched_count_txqlen(ofconn->schedulers[i]); + struct pinsched_stats stats; + + pinsched_get_stats(ofconn->schedulers[i], &stats); + packets += stats.n_queued; } packets += pktbuf_count_packets(ofconn->pktbuf); } @@ -391,27 +430,15 @@ 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, uint8_t dscp, - uint32_t allowed_versions); + uint32_t allowed_versions) + OVS_REQUIRES(ofproto_mutex); static struct ofconn *find_controller_by_target(struct connmgr *, const char *target); -static void update_fail_open(struct connmgr *); +static void update_fail_open(struct connmgr *) OVS_EXCLUDED(ofproto_mutex); static int set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp, const struct sset *); @@ -446,34 +473,50 @@ connmgr_get_controller_info(struct connmgr *mgr, struct shash *info) 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); + int i; shash_add(info, target, cinfo); cinfo->is_connected = rconn_is_connected(rconn); cinfo->role = ofconn->role; - cinfo->pairs.n = 0; - + smap_init(&cinfo->pairs); if (last_error) { - cinfo->pairs.keys[cinfo->pairs.n] = "last_error"; - cinfo->pairs.values[cinfo->pairs.n++] - = xstrdup(ovs_retval_to_string(last_error)); + smap_add(&cinfo->pairs, "last_error", + ovs_retval_to_string(last_error)); } - cinfo->pairs.keys[cinfo->pairs.n] = "state"; - cinfo->pairs.values[cinfo->pairs.n++] - = xstrdup(rconn_get_state(rconn)); + smap_add(&cinfo->pairs, "state", 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)); + smap_add_format(&cinfo->pairs, "sec_since_connect", + "%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)); + smap_add_format(&cinfo->pairs, "sec_since_disconnect", + "%ld", (long int) (now - last_disconnect)); + } + + for (i = 0; i < N_SCHEDULERS; i++) { + if (ofconn->schedulers[i]) { + const char *name = i ? "miss" : "action"; + struct pinsched_stats stats; + + pinsched_get_stats(ofconn->schedulers[i], &stats); + smap_add_nocopy(&cinfo->pairs, + xasprintf("packet-in-%s-backlog", name), + xasprintf("%u", stats.n_queued)); + smap_add_nocopy(&cinfo->pairs, + xasprintf("packet-in-%s-bypassed", name), + xasprintf("%llu", stats.n_normal)); + smap_add_nocopy(&cinfo->pairs, + xasprintf("packet-in-%s-queued", name), + xasprintf("%llu", stats.n_limited)); + smap_add_nocopy(&cinfo->pairs, + xasprintf("packet-in-%s-dropped", name), + xasprintf("%llu", stats.n_queue_dropped)); + } } } } @@ -486,9 +529,7 @@ connmgr_free_controller_info(struct shash *info) SHASH_FOR_EACH (node, info) { struct ofproto_controller_info *cinfo = node->data; - while (cinfo->pairs.n) { - free(CONST_CAST(char *, cinfo->pairs.values[--cinfo->pairs.n])); - } + smap_destroy(&cinfo->pairs); free(cinfo); } shash_destroy(info); @@ -500,6 +541,7 @@ void connmgr_set_controllers(struct connmgr *mgr, const struct ofproto_controller *controllers, size_t n_controllers, uint32_t allowed_versions) + OVS_EXCLUDED(ofproto_mutex) { bool had_controllers = connmgr_has_controllers(mgr); struct shash new_controllers; @@ -507,6 +549,10 @@ connmgr_set_controllers(struct connmgr *mgr, struct ofservice *ofservice, *next_ofservice; size_t i; + /* Required to add and remove ofconns. This could probably be narrowed to + * cover a smaller amount of code, if that yielded some benefit. */ + ovs_mutex_lock(&ofproto_mutex); + /* Create newly configured controllers and services. * Create a name to ofproto_controller mapping in 'new_controllers'. */ shash_init(&new_controllers); @@ -589,6 +635,8 @@ connmgr_set_controllers(struct connmgr *mgr, shash_destroy(&new_controllers); + ovs_mutex_unlock(&ofproto_mutex); + update_in_band_remotes(mgr); update_fail_open(mgr); if (had_controllers != connmgr_has_controllers(mgr)) { @@ -641,6 +689,7 @@ connmgr_has_snoops(const struct connmgr *mgr) static void add_controller(struct connmgr *mgr, const char *target, uint8_t dscp, uint32_t allowed_versions) + OVS_REQUIRES(ofproto_mutex) { char *name = ofconn_make_name(mgr, target); struct ofconn *ofconn; @@ -683,18 +732,16 @@ 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; - } - - if (stream_parse_target_with_default_ports(target, - OFP_TCP_PORT, - OFP_SSL_PORT, - sin)) { - n_addrs++; + union { + struct sockaddr_storage ss; + struct sockaddr_in in; + } sa; + + if (ofconn->band == OFPROTO_IN_BAND + && stream_parse_target_with_default_port(target, OFP_PORT, &sa.ss) + && sa.ss.ss_family == AF_INET) { + addrs[n_addrs++] = sa.in; } } for (i = 0; i < mgr->n_extra_remotes; i++) { @@ -721,6 +768,7 @@ update_in_band_remotes(struct connmgr *mgr) static void update_fail_open(struct connmgr *mgr) + OVS_EXCLUDED(ofproto_mutex) { if (connmgr_has_controllers(mgr) && mgr->fail_mode == OFPROTO_FAIL_STANDALONE) { @@ -757,7 +805,7 @@ set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp, if (!error) { pvconns[n_pvconns++] = pvconn; } else { - VLOG_ERR("failed to listen on %s: %s", name, strerror(error)); + VLOG_ERR("failed to listen on %s: %s", name, ovs_strerror(error)); if (!retval) { retval = error; } @@ -823,6 +871,17 @@ ofconn_get_type(const struct ofconn *ofconn) return ofconn->type; } +/* If a master election id is defined, stores it into '*idp' and returns + * true. Otherwise, stores UINT64_MAX into '*idp' and returns false. */ +bool +ofconn_get_master_election_id(const struct ofconn *ofconn, uint64_t *idp) +{ + *idp = (ofconn->connmgr->master_election_id_defined + ? ofconn->connmgr->master_election_id + : UINT64_MAX); + return ofconn->connmgr->master_election_id_defined; +} + /* Sets the master election id. * * Returns true if successful, false if the id is stale @@ -852,17 +911,34 @@ ofconn_get_role(const struct ofconn *ofconn) return ofconn->role; } +void +ofconn_send_role_status(struct ofconn *ofconn, uint32_t role, uint8_t reason) +{ + struct ofputil_role_status status; + struct ofpbuf *buf; + + status.reason = reason; + status.role = role; + ofconn_get_master_election_id(ofconn, &status.generation_id); + + buf = ofputil_encode_role_status(&status, ofconn_get_protocol(ofconn)); + if (buf) { + ofconn_send(ofconn, buf, NULL); + } +} + /* Changes 'ofconn''s role to 'role'. If 'role' is OFPCR12_ROLE_MASTER then * any existing master is demoted to a slave. */ void ofconn_set_role(struct ofconn *ofconn, enum ofp12_controller_role role) { - if (role == OFPCR12_ROLE_MASTER) { + if (role != ofconn->role && role == OFPCR12_ROLE_MASTER) { struct ofconn *other; - HMAP_FOR_EACH (other, hmap_node, &ofconn->connmgr->controllers) { + LIST_FOR_EACH (other, node, &ofconn->connmgr->all_conns) { if (other->role == OFPCR12_ROLE_MASTER) { other->role = OFPCR12_ROLE_SLAVE; + ofconn_send_role_status(other, OFPCR12_ROLE_SLAVE, OFPCRR_MASTER_REQUEST); } } } @@ -872,19 +948,22 @@ ofconn_set_role(struct ofconn *ofconn, enum ofp12_controller_role role) void ofconn_set_invalid_ttl_to_controller(struct ofconn *ofconn, bool enable) { + struct ofputil_async_cfg ac = ofconn_get_async_config(ofconn); uint32_t bit = 1u << OFPR_INVALID_TTL; if (enable) { - ofconn->master_async_config[OAM_PACKET_IN] |= bit; + ac.master[OAM_PACKET_IN] |= bit; } else { - ofconn->master_async_config[OAM_PACKET_IN] &= ~bit; + ac.master[OAM_PACKET_IN] &= ~bit; } + ofconn_set_async_config(ofconn, &ac); } bool ofconn_get_invalid_ttl_to_controller(struct ofconn *ofconn) { + struct ofputil_async_cfg ac = ofconn_get_async_config(ofconn); uint32_t bit = 1u << OFPR_INVALID_TTL; - return (ofconn->master_async_config[OAM_PACKET_IN] & bit) != 0; + return (ac.master[OAM_PACKET_IN] & bit) != 0; } /* Returns the currently configured protocol for 'ofconn', one of OFPUTIL_P_*. @@ -965,12 +1044,25 @@ ofconn_set_miss_send_len(struct ofconn *ofconn, int 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]) + const struct ofputil_async_cfg *ac) { - size_t size = sizeof ofconn->master_async_config; - memcpy(ofconn->master_async_config, master_masks, size); - memcpy(ofconn->slave_async_config, slave_masks, size); + if (!ofconn->async_cfg) { + ofconn->async_cfg = xmalloc(sizeof *ofconn->async_cfg); + } + *ofconn->async_cfg = *ac; +} + +struct ofputil_async_cfg +ofconn_get_async_config(const struct ofconn *ofconn) +{ + if (ofconn->async_cfg) { + return *ofconn->async_cfg; + } + + int version = rconn_get_version(ofconn->rconn); + return (version < 0 || !ofconn->enable_async_msgs + ? OFPUTIL_ASYNC_CFG_INIT + : ofputil_async_cfg_default(version)); } /* Sends 'msg' on 'ofconn', accounting it as a reply. (If there is a @@ -986,12 +1078,11 @@ ofconn_send_reply(const struct ofconn *ofconn, struct ofpbuf *msg) /* 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) +ofconn_send_replies(const struct ofconn *ofconn, struct ovs_list *replies) { - struct ofpbuf *reply, *next; + struct ofpbuf *reply; - LIST_FOR_EACH_SAFE (reply, next, list_node, replies) { - list_remove(&reply->list_node); + LIST_FOR_EACH_POP (reply, list_node, replies) { ofconn_send_reply(ofconn, reply); } } @@ -1027,31 +1118,92 @@ ofconn_send_error(const struct ofconn *ofconn, /* Same as pktbuf_retrieve(), using the pktbuf owned by 'ofconn'. */ enum ofperr ofconn_pktbuf_retrieve(struct ofconn *ofconn, uint32_t id, - struct ofpbuf **bufferp, uint16_t *in_port) + struct dp_packet **bufferp, ofp_port_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) +/* Reports that a flow_mod operation of the type specified by 'command' was + * successfully executed by 'ofconn', so that the connmgr can log it. */ +void +ofconn_report_flow_mod(struct ofconn *ofconn, + enum ofp_flow_mod_command command) +{ + long long int now; + + switch (command) { + case OFPFC_ADD: + ofconn->n_add++; + break; + + case OFPFC_MODIFY: + case OFPFC_MODIFY_STRICT: + ofconn->n_modify++; + break; + + case OFPFC_DELETE: + case OFPFC_DELETE_STRICT: + ofconn->n_delete++; + break; + } + + now = time_msec(); + if (ofconn->next_op_report == LLONG_MAX) { + ofconn->first_op = now; + ofconn->next_op_report = MAX(now + 10 * 1000, ofconn->op_backoff); + ofconn->op_backoff = ofconn->next_op_report + 60 * 1000; + } + ofconn->last_op = now; +} + +/* OpenFlow 1.4 bundles. */ + +static inline uint32_t +bundle_hash(uint32_t id) { - return !list_is_empty(&ofconn->opgroups); + return hash_int(id, 0); } -/* 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) +struct ofp_bundle * +ofconn_get_bundle(struct ofconn *ofconn, uint32_t id) +{ + struct ofp_bundle *bundle; + + HMAP_FOR_EACH_IN_BUCKET(bundle, node, bundle_hash(id), &ofconn->bundles) { + if (bundle->id == id) { + return bundle; + } + } + + return NULL; +} + +enum ofperr +ofconn_insert_bundle(struct ofconn *ofconn, struct ofp_bundle *bundle) +{ + /* XXX: Check the limit of open bundles */ + + hmap_insert(&ofconn->bundles, &bundle->node, bundle_hash(bundle->id)); + + return 0; +} + +enum ofperr +ofconn_remove_bundle(struct ofconn *ofconn, struct ofp_bundle *bundle) { - list_push_back(&ofconn->opgroups, ofconn_node); + hmap_remove(&ofconn->bundles, &bundle->node); + + return 0; +} + +static void +bundle_remove_all(struct ofconn *ofconn) +{ + struct ofp_bundle *b, *next; + + HMAP_FOR_EACH_SAFE (b, next, node, &ofconn->bundles) { + ofp_bundle_remove__(ofconn, b, false); + } } /* Private ofconn functions. */ @@ -1075,11 +1227,11 @@ ofconn_create(struct connmgr *mgr, struct rconn *rconn, enum ofconn_type type, ofconn->type = type; ofconn->enable_async_msgs = enable_async_msgs; - list_init(&ofconn->opgroups); - hmap_init(&ofconn->monitors); list_init(&ofconn->updates); + hmap_init(&ofconn->bundles); + ofconn_flush(ofconn); return ofconn; @@ -1089,26 +1241,17 @@ ofconn_create(struct connmgr *mgr, struct rconn *rconn, enum ofconn_type type, * connection to the next. */ static void ofconn_flush(struct ofconn *ofconn) + OVS_REQUIRES(ofproto_mutex) { struct ofmonitor *monitor, *next_monitor; int i; + ofconn_log_flow_mods(ofconn); + ofconn->role = OFPCR12_ROLE_EQUAL; ofconn_set_protocol(ofconn, OFPUTIL_P_NONE); 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++) { @@ -1132,33 +1275,13 @@ ofconn_flush(struct ofconn *ofconn) 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); - } + free(ofconn->async_cfg); + ofconn->async_cfg = NULL; + + ofconn->n_add = ofconn->n_delete = ofconn->n_modify = 0; + ofconn->first_op = ofconn->last_op = LLONG_MIN; + ofconn->next_op_report = LLONG_MAX; + ofconn->op_backoff = LLONG_MIN; HMAP_FOR_EACH_SAFE (monitor, next_monitor, ofconn_node, &ofconn->monitors) { @@ -1171,6 +1294,7 @@ ofconn_flush(struct ofconn *ofconn) static void ofconn_destroy(struct ofconn *ofconn) + OVS_REQUIRES(ofproto_mutex) { ofconn_flush(ofconn); @@ -1178,6 +1302,10 @@ ofconn_destroy(struct ofconn *ofconn) hmap_remove(&ofconn->connmgr->controllers, &ofconn->hmap_node); } + bundle_remove_all(ofconn); + hmap_destroy(&ofconn->bundles); + + hmap_destroy(&ofconn->monitors); list_remove(&ofconn->node); rconn_destroy(ofconn->rconn); rconn_packet_counter_destroy(ofconn->packet_in_counter); @@ -1216,57 +1344,57 @@ ofconn_reconfigure(struct ofconn *ofconn, const struct ofproto_controller *c) static bool ofconn_may_recv(const struct ofconn *ofconn) { - int count = ofconn->reply_counter->n_packets; - return (!ofconn->blocked || ofconn->retry) && count < OFCONN_REPLY_MAX; + int count = rconn_packet_counter_n_packets(ofconn->reply_counter); + return count < OFCONN_REPLY_MAX; } static void ofconn_run(struct ofconn *ofconn, - bool (*handle_openflow)(struct ofconn *, struct ofpbuf *ofp_msg)) + void (*handle_openflow)(struct ofconn *, + const struct ofpbuf *ofp_msg)) { struct connmgr *mgr = ofconn->connmgr; size_t i; for (i = 0; i < N_SCHEDULERS; i++) { - pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn); + struct ovs_list txq; + + pinsched_run(ofconn->schedulers[i], &txq); + do_send_packet_ins(ofconn, &txq); } rconn_run(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); - } + /* Limit the number of iterations to avoid starving other tasks. */ + for (i = 0; i < 50 && ofconn_may_recv(ofconn); i++) { + struct ofpbuf *of_msg = rconn_recv(ofconn->rconn); + if (!of_msg) { + break; + } - if (handle_openflow(ofconn, of_msg)) { - ofpbuf_delete(of_msg); - ofconn->blocked = NULL; - } else { - ofconn->blocked = of_msg; - ofconn->retry = false; - } + if (mgr->fail_open) { + fail_open_maybe_recover(mgr->fail_open); } + + handle_openflow(ofconn, of_msg); + ofpbuf_delete(of_msg); + } + + if (time_msec() >= ofconn->next_op_report) { + ofconn_log_flow_mods(ofconn); } + ovs_mutex_lock(&ofproto_mutex); if (!rconn_is_alive(ofconn->rconn)) { ofconn_destroy(ofconn); } else if (!rconn_is_connected(ofconn->rconn)) { ofconn_flush(ofconn); } + ovs_mutex_unlock(&ofproto_mutex); } static void -ofconn_wait(struct ofconn *ofconn, bool handling_openflow) +ofconn_wait(struct ofconn *ofconn) { int i; @@ -1274,9 +1402,53 @@ ofconn_wait(struct ofconn *ofconn, bool handling_openflow) pinsched_wait(ofconn->schedulers[i]); } rconn_run_wait(ofconn->rconn); - if (handling_openflow && ofconn_may_recv(ofconn)) { + if (ofconn_may_recv(ofconn)) { rconn_recv_wait(ofconn->rconn); } + if (ofconn->next_op_report != LLONG_MAX) { + poll_timer_wait_until(ofconn->next_op_report); + } +} + +static void +ofconn_log_flow_mods(struct ofconn *ofconn) +{ + int n_flow_mods = ofconn->n_add + ofconn->n_delete + ofconn->n_modify; + if (n_flow_mods) { + long long int ago = (time_msec() - ofconn->first_op) / 1000; + long long int interval = (ofconn->last_op - ofconn->first_op) / 1000; + struct ds s; + + ds_init(&s); + ds_put_format(&s, "%d flow_mods ", n_flow_mods); + if (interval == ago) { + ds_put_format(&s, "in the last %lld s", ago); + } else if (interval) { + ds_put_format(&s, "in the %lld s starting %lld s ago", + interval, ago); + } else { + ds_put_format(&s, "%lld s ago", ago); + } + + ds_put_cstr(&s, " ("); + if (ofconn->n_add) { + ds_put_format(&s, "%d adds, ", ofconn->n_add); + } + if (ofconn->n_delete) { + ds_put_format(&s, "%d deletes, ", ofconn->n_delete); + } + if (ofconn->n_modify) { + ds_put_format(&s, "%d modifications, ", ofconn->n_modify); + } + s.length -= 2; + ds_put_char(&s, ')'); + + VLOG_INFO("%s: %s", rconn_get_name(ofconn->rconn), ds_cstr(&s)); + ds_destroy(&s); + + ofconn->n_add = ofconn->n_delete = ofconn->n_modify = 0; + } + ofconn->next_op_report = LLONG_MAX; } /* Returns true if 'ofconn' should receive asynchronous messages of the given @@ -1286,19 +1458,12 @@ ofconn_wait(struct ofconn *ofconn, bool handling_openflow) * 'ofconn'. */ static bool ofconn_receives_async_msg(const struct ofconn *ofconn, - enum ofconn_async_msg_type type, + enum ofputil_async_msg_type type, unsigned int reason) { - const uint32_t *async_config; - ovs_assert(reason < 32); ovs_assert((unsigned int) type < OAM_N_TYPES); - if (ofconn_get_protocol(ofconn) == OFPUTIL_P_NONE - || !rconn_is_connected(ofconn->rconn)) { - return false; - } - /* Keep the following code in sync with the documentation in the * "Asynchronous Messages" section in DESIGN. */ @@ -1308,16 +1473,73 @@ ofconn_receives_async_msg(const struct ofconn *ofconn, return false; } - async_config = (ofconn->role == OFPCR12_ROLE_SLAVE - ? ofconn->slave_async_config - : ofconn->master_async_config); - if (!(async_config[type] & (1u << reason))) { - return false; - } + struct ofputil_async_cfg ac = ofconn_get_async_config(ofconn); + uint32_t *masks = (ofconn->role == OFPCR12_ROLE_SLAVE + ? ac.slave + : ac.master); + return (masks[type] & (1u << reason)) != 0; +} +/* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the + * packet rather than to send the packet to the controller. + * + * This function returns false to indicate the packet should be dropped if + * the controller action was the result of the default table-miss behaviour + * and the controller is using OpenFlow1.3+. + * + * Otherwise true is returned to indicate the packet should be forwarded to + * the controller */ +static bool +ofconn_wants_packet_in_on_miss(struct ofconn *ofconn, + const struct ofproto_packet_in *pin) +{ + if (pin->miss_type == OFPROTO_PACKET_IN_MISS_WITHOUT_FLOW) { + enum ofputil_protocol protocol = ofconn_get_protocol(ofconn); + + if (protocol != OFPUTIL_P_NONE + && ofputil_protocol_to_ofp_version(protocol) >= OFP13_VERSION + && (ofproto_table_get_miss_config(ofconn->connmgr->ofproto, + pin->up.table_id) + == OFPUTIL_TABLE_MISS_DEFAULT)) { + return false; + } + } return true; } +/* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the + * packet rather than to send the packet to the controller. + * + * This function returns true to indicate that a packet_in message + * for a "table-miss" should be sent to at least one controller. + * That is there is at least one controller with controller_id 0 + * which connected using an OpenFlow version earlier than OpenFlow1.3. + * + * False otherwise. + * + * This logic assumes that "table-miss" packet_in messages + * are always sent to controller_id 0. */ +bool +connmgr_wants_packet_in_on_miss(struct connmgr *mgr) OVS_EXCLUDED(ofproto_mutex) +{ + struct ofconn *ofconn; + + ovs_mutex_lock(&ofproto_mutex); + LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { + enum ofputil_protocol protocol = ofconn_get_protocol(ofconn); + + if (ofconn->controller_id == 0 && + (protocol == OFPUTIL_P_NONE || + ofputil_protocol_to_ofp_version(protocol) < OFP13_VERSION)) { + ovs_mutex_unlock(&ofproto_mutex); + return true; + } + } + ovs_mutex_unlock(&ofproto_mutex); + + return false; +} + /* Returns a human-readable name for an OpenFlow connection between 'mgr' and * 'target', suitable for use in log messages for identifying the connection. * @@ -1360,12 +1582,15 @@ ofconn_send(const struct ofconn *ofconn, struct ofpbuf *msg, /* Sending asynchronous messages. */ -static void schedule_packet_in(struct ofconn *, struct ofputil_packet_in); +static void schedule_packet_in(struct ofconn *, struct ofproto_packet_in, + enum ofp_packet_in_reason wire_reason); /* Sends an OFPT_PORT_STATUS message with 'opp' and 'reason' to appropriate - * controllers managed by 'mgr'. */ + * controllers managed by 'mgr'. For messages caused by a controller + * OFPT_PORT_MOD, specify 'source' as the controller connection that sent the + * request; otherwise, specify 'source' as NULL. */ void -connmgr_send_port_status(struct connmgr *mgr, +connmgr_send_port_status(struct connmgr *mgr, struct ofconn *source, const struct ofputil_phy_port *pp, uint8_t reason) { /* XXX Should limit the number of queued port status change messages. */ @@ -1378,12 +1603,58 @@ connmgr_send_port_status(struct connmgr *mgr, if (ofconn_receives_async_msg(ofconn, OAM_PORT_STATUS, reason)) { struct ofpbuf *msg; + /* Before 1.5, OpenFlow specified that OFPT_PORT_MOD should not + * generate OFPT_PORT_STATUS messages. That requirement was a + * relic of how OpenFlow originally supported a single controller, + * so that one could expect the controller to already know the + * changes it had made. + * + * EXT-338 changes OpenFlow 1.5 OFPT_PORT_MOD to send + * OFPT_PORT_STATUS messages to every controller. This is + * obviously more useful in the multi-controller case. We could + * always implement it that way in OVS, but that would risk + * confusing controllers that are intended for single-controller + * use only. (Imagine a controller that generates an OFPT_PORT_MOD + * in response to any OFPT_PORT_STATUS!) + * + * So this compromises: for OpenFlow 1.4 and earlier, it generates + * OFPT_PORT_STATUS for OFPT_PORT_MOD, but not back to the + * originating controller. In a single-controller environment, in + * particular, this means that it will never generate + * OFPT_PORT_STATUS for OFPT_PORT_MOD at all. */ + if (ofconn == source + && rconn_get_version(ofconn->rconn) < OFP15_VERSION) { + continue; + } + msg = ofputil_encode_port_status(&ps, ofconn_get_protocol(ofconn)); ofconn_send(ofconn, msg, NULL); } } } +/* Sends an OFPT_REQUESTFORWARD message with 'request' and 'reason' to + * appropriate controllers managed by 'mgr'. For messages caused by a + * controller OFPT_GROUP_MOD and OFPT_METER_MOD, specify 'source' as the + * controller connection that sent the request; otherwise, specify 'source' + * as NULL. */ +void +connmgr_send_requestforward(struct connmgr *mgr, const struct ofconn *source, + const struct ofputil_requestforward *rf) +{ + struct ofconn *ofconn; + + LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { + if (ofconn_receives_async_msg(ofconn, OAM_REQUESTFORWARD, rf->reason) + && rconn_get_version(ofconn->rconn) >= OFP14_VERSION + && ofconn != source) { + enum ofputil_protocol protocol = ofconn_get_protocol(ofconn); + ofconn_send(ofconn, ofputil_encode_requestforward(rf, protocol), + NULL); + } + } +} + /* Sends an OFPT_FLOW_REMOVED or NXT_FLOW_REMOVED message based on 'fr' to * appropriate controllers managed by 'mgr'. */ void @@ -1407,74 +1678,128 @@ connmgr_send_flow_removed(struct connmgr *mgr, } } +/* Normally a send-to-controller action uses reason OFPR_ACTION. However, in + * OpenFlow 1.3 and later, packet_ins generated by a send-to-controller action + * in a "table-miss" flow (one with priority 0 and completely wildcarded) are + * sent as OFPR_NO_MATCH. This function returns the reason that should + * actually be sent on 'ofconn' for 'pin'. */ +static enum ofp_packet_in_reason +wire_reason(struct ofconn *ofconn, const struct ofproto_packet_in *pin) +{ + enum ofputil_protocol protocol = ofconn_get_protocol(ofconn); + + if (pin->miss_type == OFPROTO_PACKET_IN_MISS_FLOW + && pin->up.reason == OFPR_ACTION + && protocol != OFPUTIL_P_NONE + && ofputil_protocol_to_ofp_version(protocol) >= OFP13_VERSION) { + return OFPR_NO_MATCH; + } + + switch (pin->up.reason) { + case OFPR_ACTION_SET: + case OFPR_GROUP: + case OFPR_PACKET_OUT: + if (!(protocol & OFPUTIL_P_OF14_UP)) { + /* Only supported in OF1.4+ */ + return OFPR_ACTION; + } + /* Fall through. */ + case OFPR_NO_MATCH: + case OFPR_ACTION: + case OFPR_INVALID_TTL: + case OFPR_N_REASONS: + default: + return pin->up.reason; + } +} + /* Given 'pin', sends an OFPT_PACKET_IN message to each OpenFlow controller as * necessary according to their individual configurations. * * 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 ofputil_packet_in *pin) + const struct ofproto_packet_in *pin) { struct ofconn *ofconn; LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { - if (ofconn_receives_async_msg(ofconn, OAM_PACKET_IN, pin->reason) + enum ofp_packet_in_reason reason = wire_reason(ofconn, pin); + + if (ofconn_wants_packet_in_on_miss(ofconn, pin) + && ofconn_receives_async_msg(ofconn, OAM_PACKET_IN, reason) && ofconn->controller_id == pin->controller_id) { - schedule_packet_in(ofconn, *pin); + schedule_packet_in(ofconn, *pin, reason); } } } -/* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */ static void -do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn_) +do_send_packet_ins(struct ofconn *ofconn, struct ovs_list *txq) { - struct ofconn *ofconn = ofconn_; + struct ofpbuf *pin; - rconn_send_with_limit(ofconn->rconn, ofp_packet_in, - ofconn->packet_in_counter, 100); + LIST_FOR_EACH_POP (pin, list_node, txq) { + if (rconn_send_with_limit(ofconn->rconn, pin, + ofconn->packet_in_counter, 100) == EAGAIN) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); + + VLOG_INFO_RL(&rl, "%s: dropping packet-in due to queue overflow", + rconn_get_name(ofconn->rconn)); + } + } } /* 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, struct ofputil_packet_in pin) +schedule_packet_in(struct ofconn *ofconn, struct ofproto_packet_in pin, + enum ofp_packet_in_reason wire_reason) { struct connmgr *mgr = ofconn->connmgr; + uint16_t controller_max_len; + struct ovs_list txq; - pin.total_len = pin.packet_len; + pin.up.total_len = pin.up.packet_len; - /* Get OpenFlow buffer_id. */ - 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; + pin.up.reason = wire_reason; + if (pin.up.reason == OFPR_ACTION) { + controller_max_len = pin.send_len; /* max_len */ } else { - pin.buffer_id = pktbuf_save(ofconn->pktbuf, pin.packet, pin.packet_len, - pin.fmd.in_port); + controller_max_len = ofconn->miss_send_len; } - /* Figure out how much of the packet to send. */ - if (pin.reason == OFPR_NO_MATCH) { - pin.send_len = pin.packet_len; + /* Get OpenFlow buffer_id. + * For OpenFlow 1.2+, OFPCML_NO_BUFFER (== UINT16_MAX) specifies + * unbuffered. This behaviour doesn't violate prior versions, too. */ + if (controller_max_len == UINT16_MAX) { + pin.up.buffer_id = UINT32_MAX; + } else if (mgr->fail_open && fail_open_is_active(mgr->fail_open)) { + pin.up.buffer_id = pktbuf_get_null(); + } else if (!ofconn->pktbuf) { + pin.up.buffer_id = UINT32_MAX; } else { - /* Caller should have initialized 'send_len' to 'max_len' specified in - * output action. */ + pin.up.buffer_id = pktbuf_save(ofconn->pktbuf, + pin.up.packet, pin.up.packet_len, + pin.up.flow_metadata.flow.in_port.ofp_port); } - if (pin.buffer_id != UINT32_MAX) { - pin.send_len = MIN(pin.send_len, ofconn->miss_send_len); + + /* Figure out how much of the packet to send. + * If not buffered, send the entire packet. Otherwise, depending on + * the reason of packet-in, send what requested by the controller. */ + if (pin.up.buffer_id != UINT32_MAX + && controller_max_len < pin.up.packet_len) { + pin.up.packet_len = controller_max_len; } - /* 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[pin.reason == OFPR_NO_MATCH ? 0 : 1], - pin.fmd.in_port, - ofputil_encode_packet_in(&pin, ofconn_get_protocol(ofconn), + /* Make OFPT_PACKET_IN and hand over to packet scheduler. */ + pinsched_send(ofconn->schedulers[pin.up.reason == OFPR_NO_MATCH ? 0 : 1], + pin.up.flow_metadata.flow.in_port.ofp_port, + ofputil_encode_packet_in(&pin.up, + ofconn_get_protocol(ofconn), ofconn->packet_in_format), - do_send_packet_in, ofconn); + &txq); + do_send_packet_ins(ofconn, &txq); } /* Fail-open settings. */ @@ -1631,20 +1956,9 @@ any_extras_changed(const struct connmgr *mgr, /* In-band implementation. */ bool -connmgr_msg_in_hook(struct connmgr *mgr, const struct flow *flow, - const struct ofpbuf *packet) +connmgr_has_in_band(struct connmgr *mgr) { - return mgr->in_band && in_band_msg_in_hook(mgr->in_band, flow, packet); -} - -bool -connmgr_may_set_up_flow(struct connmgr *mgr, const struct flow *flow, - uint32_t local_odp_port, - const struct nlattr *odp_actions, - size_t actions_len) -{ - return !mgr->in_band || in_band_rule_check(flow, local_odp_port, - odp_actions, actions_len); + return mgr->in_band != NULL; } /* Fail-open and in-band implementation. */ @@ -1655,6 +1969,7 @@ connmgr_may_set_up_flow(struct connmgr *mgr, const struct flow *flow, * In-band control has more sophisticated code that manages flows itself. */ void connmgr_flushed(struct connmgr *mgr) + OVS_EXCLUDED(ofproto_mutex) { if (mgr->fail_open) { fail_open_flushed(mgr->fail_open); @@ -1671,14 +1986,31 @@ connmgr_flushed(struct connmgr *mgr) ofpbuf_init(&ofpacts, OFPACT_OUTPUT_SIZE); ofpact_put_OUTPUT(&ofpacts)->port = OFPP_NORMAL; - ofpact_pad(&ofpacts); match_init_catchall(&match); - ofproto_add_flow(mgr->ofproto, &match, 0, ofpacts.data, ofpacts.size); + ofproto_add_flow(mgr->ofproto, &match, 0, ofpacts.data, + ofpacts.size); ofpbuf_uninit(&ofpacts); } } + +/* Returns the number of hidden rules created by the in-band and fail-open + * implementations in table 0. (Subtracting this count from the number of + * rules in the table 0 classifier, as maintained in struct oftable, yields + * the number of flows that OVS should report via OpenFlow for table 0.) */ +int +connmgr_count_hidden_rules(const struct connmgr *mgr) +{ + int n_hidden = 0; + if (mgr->in_band) { + n_hidden += in_band_count_rules(mgr->in_band); + } + if (mgr->fail_open) { + n_hidden += fail_open_count_rules(mgr->fail_open); + } + return n_hidden; +} /* Creates a new ofservice for 'target' in 'mgr'. Returns 0 if successful, * otherwise a positive errno value. @@ -1769,6 +2101,7 @@ COVERAGE_DEFINE(ofmonitor_resume); enum ofperr ofmonitor_create(const struct ofputil_flow_monitor_request *request, struct ofconn *ofconn, struct ofmonitor **monitorp) + OVS_REQUIRES(ofproto_mutex) { struct ofmonitor *m; @@ -1776,7 +2109,7 @@ ofmonitor_create(const struct ofputil_flow_monitor_request *request, m = ofmonitor_lookup(ofconn, request->id); if (m) { - return OFPERR_NXBRC_FM_DUPLICATE_ID; + return OFPERR_OFPMOFC_MONITOR_EXISTS; } m = xmalloc(sizeof *m); @@ -1794,6 +2127,7 @@ ofmonitor_create(const struct ofputil_flow_monitor_request *request, struct ofmonitor * ofmonitor_lookup(struct ofconn *ofconn, uint32_t id) + OVS_REQUIRES(ofproto_mutex) { struct ofmonitor *m; @@ -1808,6 +2142,7 @@ ofmonitor_lookup(struct ofconn *ofconn, uint32_t id) void ofmonitor_destroy(struct ofmonitor *m) + OVS_REQUIRES(ofproto_mutex) { if (m) { minimatch_destroy(&m->match); @@ -1820,11 +2155,17 @@ void ofmonitor_report(struct connmgr *mgr, struct rule *rule, enum nx_flow_update_event event, enum ofp_flow_removed_reason reason, - const struct ofconn *abbrev_ofconn, ovs_be32 abbrev_xid) + const struct ofconn *abbrev_ofconn, ovs_be32 abbrev_xid, + const struct rule_actions *old_actions) + OVS_REQUIRES(ofproto_mutex) { enum nx_flow_monitor_flags update; struct ofconn *ofconn; + if (rule_is_hidden(rule)) { + return; + } + switch (event) { case NXFME_ADDED: update = NXFMF_ADD; @@ -1842,7 +2183,7 @@ ofmonitor_report(struct connmgr *mgr, struct rule *rule, default: case NXFME_ABBREV: - NOT_REACHED(); + OVS_NOT_REACHED(); } LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { @@ -1861,7 +2202,11 @@ ofmonitor_report(struct connmgr *mgr, struct rule *rule, HMAP_FOR_EACH (m, ofconn_node, &ofconn->monitors) { if (m->flags & update && (m->table_id == 0xff || m->table_id == rule->table_id) - && ofoperation_has_out_port(rule->pending, m->out_port) + && (ofproto_rule_has_out_port(rule, m->out_port) + || (old_actions + && ofpacts_output_to_port(old_actions->ofpacts, + old_actions->ofpacts_len, + m->out_port))) && cls_rule_is_loose_match(&rule->cr, &m->match)) { flags |= m->flags; } @@ -1873,22 +2218,28 @@ ofmonitor_report(struct connmgr *mgr, struct rule *rule, ofconn->sent_abbrev_update = false; } - if (ofconn != abbrev_ofconn || ofconn->monitor_paused) { + if (flags & NXFMF_OWN || ofconn != abbrev_ofconn + || ofconn->monitor_paused) { struct ofputil_flow_update fu; struct match match; fu.event = event; fu.reason = event == NXFME_DELETED ? reason : 0; - fu.idle_timeout = rule->idle_timeout; - fu.hard_timeout = rule->hard_timeout; fu.table_id = rule->table_id; fu.cookie = rule->flow_cookie; minimatch_expand(&rule->cr.match, &match); fu.match = &match; fu.priority = rule->cr.priority; + + ovs_mutex_lock(&rule->mutex); + fu.idle_timeout = rule->idle_timeout; + fu.hard_timeout = rule->hard_timeout; + ovs_mutex_unlock(&rule->mutex); + if (flags & NXFMF_ACTIONS) { - fu.ofpacts = rule->ofpacts; - fu.ofpacts_len = rule->ofpacts_len; + const struct rule_actions *actions = rule_get_actions(rule); + fu.ofpacts = actions->ofpacts; + fu.ofpacts_len = actions->ofpacts_len; } else { fu.ofpacts = NULL; fu.ofpacts_len = 0; @@ -1909,17 +2260,19 @@ ofmonitor_report(struct connmgr *mgr, struct rule *rule, void ofmonitor_flush(struct connmgr *mgr) + OVS_REQUIRES(ofproto_mutex) { struct ofconn *ofconn; LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { - struct ofpbuf *msg, *next; + struct ofpbuf *msg; + + LIST_FOR_EACH_POP (msg, list_node, &ofconn->updates) { + unsigned int n_bytes; - LIST_FOR_EACH_SAFE (msg, next, list_node, &ofconn->updates) { - list_remove(&msg->list_node); ofconn_send(ofconn, msg, ofconn->monitor_counter); - if (!ofconn->monitor_paused - && ofconn->monitor_counter->n_bytes > 128 * 1024) { + n_bytes = rconn_packet_counter_n_bytes(ofconn->monitor_counter); + if (!ofconn->monitor_paused && n_bytes > 128 * 1024) { struct ofpbuf *pause; COVERAGE_INC(ofmonitor_pause); @@ -1934,13 +2287,14 @@ ofmonitor_flush(struct connmgr *mgr) static void ofmonitor_resume(struct ofconn *ofconn) + OVS_REQUIRES(ofproto_mutex) { + struct rule_collection rules; struct ofpbuf *resumed; struct ofmonitor *m; - struct list rules; - struct list msgs; + struct ovs_list msgs; - list_init(&rules); + rule_collection_init(&rules); HMAP_FOR_EACH (m, ofconn_node, &ofconn->monitors) { ofmonitor_collect_resume_rules(m, ofconn->monitor_paused, &rules); } @@ -1956,17 +2310,27 @@ ofmonitor_resume(struct ofconn *ofconn) ofconn->monitor_paused = 0; } +static bool +ofmonitor_may_resume(const struct ofconn *ofconn) + OVS_REQUIRES(ofproto_mutex) +{ + return (ofconn->monitor_paused != 0 + && !rconn_packet_counter_n_packets(ofconn->monitor_counter)); +} + static void ofmonitor_run(struct connmgr *mgr) { struct ofconn *ofconn; + ovs_mutex_lock(&ofproto_mutex); LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { - if (ofconn->monitor_paused && !ofconn->monitor_counter->n_packets) { + if (ofmonitor_may_resume(ofconn)) { COVERAGE_INC(ofmonitor_resume); ofmonitor_resume(ofconn); } } + ovs_mutex_unlock(&ofproto_mutex); } static void @@ -1974,9 +2338,11 @@ ofmonitor_wait(struct connmgr *mgr) { struct ofconn *ofconn; + ovs_mutex_lock(&ofproto_mutex); LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { - if (ofconn->monitor_paused && !ofconn->monitor_counter->n_packets) { + if (ofmonitor_may_resume(ofconn)) { poll_immediate_wake(); } } + ovs_mutex_unlock(&ofproto_mutex); }