X-Git-Url: https://git.proxmox.com/?a=blobdiff_plain;f=ofproto%2Fconnmgr.c;h=1e9ef5b5d1002855bce82099df75ea6576e3a5c9;hb=7ed58d4a0daeffec52d603ad034a85d0d3011c8d;hp=0cd7d53fa791c3410340490341609a9df5726cb2;hpb=9886b6620ea710a4a8afd0a88271e563a64e7ea2;p=mirror_ovs.git diff --git a/ofproto/connmgr.c b/ofproto/connmgr.c index 0cd7d53fa..bf43fbd73 100644 --- a/ofproto/connmgr.c +++ b/ofproto/connmgr.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks. + * 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. @@ -15,37 +15,50 @@ */ #include - -#include "connmgr.h" - #include #include +#include "bundles.h" +#include "connmgr.h" #include "coverage.h" #include "fail-open.h" #include "in-band.h" #include "odp-util.h" -#include "ofp-util.h" -#include "ofpbuf.h" #include "ofproto-provider.h" +#include "openvswitch/dynamic-string.h" +#include "openvswitch/ofp-actions.h" +#include "openvswitch/ofp-msgs.h" +#include "openvswitch/ofp-monitor.h" +#include "openvswitch/ofpbuf.h" +#include "openvswitch/vconn.h" +#include "openvswitch/vlog.h" +#include "ovs-atomic.h" #include "pinsched.h" -#include "poll-loop.h" -#include "pktbuf.h" -#include "rconn.h" -#include "shash.h" +#include "openvswitch/poll-loop.h" +#include "openvswitch/rconn.h" +#include "openvswitch/shash.h" +#include "simap.h" #include "stream.h" #include "timeval.h" -#include "vconn.h" -#include "vlog.h" +#include "util.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. */ @@ -53,26 +66,21 @@ struct ofconn { enum ofconn_type type; /* Type. */ enum ofproto_band band; /* In-band or out-of-band? */ bool enable_async_msgs; /* Initially enable async messages? */ + bool want_packet_in_on_miss; /* State that should be cleared from one connection to the next. */ /* OpenFlow state. */ - enum nx_role role; /* Role. */ - enum nx_flow_format flow_format; /* Currently selected flow format. */ + enum ofp12_controller_role role; /* Role. */ + enum ofputil_protocol protocol; /* Current protocol variant. */ enum nx_packet_in_format packet_in_format; /* OFPT_PACKET_IN format. */ - bool flow_mod_table_id; /* NXT_FLOW_MOD_TABLE_ID enabled? */ - - /* 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 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 @@ -80,26 +88,76 @@ 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 */ + 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; + long long int next_bundle_expiry_check; +}; + +/* vswitchd/ovs-vswitchd.8.in documents the value of BUNDLE_IDLE_LIFETIME in + * seconds. That documentation must be kept in sync with the value below. */ +enum { + BUNDLE_EXPIRY_INTERVAL = 1000, /* Check bundle expiry every 1 sec. */ + BUNDLE_IDLE_TIMEOUT = 10000, /* Expire idle bundles after 10 seconds. */ }; 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); @@ -109,7 +167,7 @@ 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 { @@ -122,11 +180,15 @@ struct ofservice { 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 */ + uint32_t allowed_versions; /* OpenFlow protocol versions that may + * be negotiated for a session. */ }; 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 *mgr, const char *target, + uint32_t allowed_versions, uint8_t dscp); static void ofservice_destroy(struct connmgr *, struct ofservice *); static struct ofservice *ofservice_lookup(struct connmgr *, const char *target); @@ -138,8 +200,14 @@ 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. All modifications are + protected by ofproto_mutex, so that any + traversals from other threads can be made + safe by holding the ofproto_mutex. */ + uint64_t master_election_id; /* monotonically increasing sequence number + * for master election */ + bool master_election_id_defined; /* OpenFlow listeners. */ struct hmap services; /* Contains "struct ofservice"s. */ @@ -155,10 +223,14 @@ struct connmgr { struct sockaddr_in *extra_in_band_remotes; size_t n_extra_remotes; int in_band_queue; + + ATOMIC(int) want_packet_in_on_miss; /* Sum of ofconns' values. */ }; static void update_in_band_remotes(struct connmgr *); static void add_snooper(struct connmgr *, struct vconn *); +static void ofmonitor_run(struct connmgr *); +static void ofmonitor_wait(struct connmgr *); /* Creates and returns a new connection manager owned by 'ofproto'. 'name' is * a name for the ofproto suitable for using in log messages. @@ -176,7 +248,9 @@ connmgr_create(struct ofproto *ofproto, mgr->local_port_name = xstrdup(local_port_name); hmap_init(&mgr->controllers); - list_init(&mgr->all_conns); + ovs_list_init(&mgr->all_conns); + mgr->master_election_id = 0; + mgr->master_election_id_defined = false; hmap_init(&mgr->services); mgr->snoops = NULL; @@ -190,12 +264,43 @@ connmgr_create(struct ofproto *ofproto, mgr->n_extra_remotes = 0; mgr->in_band_queue = -1; + atomic_init(&mgr->want_packet_in_on_miss, 0); + return mgr; } +/* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the + * packet rather than to send the packet to the controller. + * + * This function maintains the count of pre-OpenFlow1.3 with controller_id 0, + * as we assume these are the controllers that should receive "table-miss" + * notifications. */ +static void +update_want_packet_in_on_miss(struct ofconn *ofconn) +{ + /* We want a packet-in on miss when controller_id is zero and OpenFlow is + * lower than version 1.3. */ + enum ofputil_protocol p = ofconn->protocol; + int new_want = (ofconn->controller_id == 0 && + (p == OFPUTIL_P_NONE || + ofputil_protocol_to_ofp_version(p) < OFP13_VERSION)); + + /* Update the setting and the count if necessary. */ + int old_want = ofconn->want_packet_in_on_miss; + if (old_want != new_want) { + atomic_int *dst = &ofconn->connmgr->want_packet_in_on_miss; + int count; + atomic_read_relaxed(dst, &count); + atomic_store_relaxed(dst, count - old_want + new_want); + + ofconn->want_packet_in_on_miss = new_want; + } +} + /* Frees 'mgr' and all of its resources. */ void connmgr_destroy(struct connmgr *mgr) + OVS_REQUIRES(ofproto_mutex) { struct ofservice *ofservice, *next_ofservice; struct ofconn *ofconn, *next_ofconn; @@ -208,6 +313,7 @@ connmgr_destroy(struct connmgr *mgr) LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &mgr->all_conns) { ofconn_destroy(ofconn); } + hmap_destroy(&mgr->controllers); HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &mgr->services) { @@ -232,29 +338,21 @@ 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; @@ -264,10 +362,11 @@ connmgr_run(struct connmgr *mgr, LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &mgr->all_conns) { ofconn_run(ofconn, handle_openflow); } + ofmonitor_run(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); } @@ -275,22 +374,27 @@ connmgr_run(struct connmgr *mgr, struct vconn *vconn; int retval; - retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn); + retval = pvconn_accept(ofservice->pvconn, &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, + vconn_get_allowed_versions(vconn)); name = ofconn_make_name(mgr, vconn_get_name(vconn)); 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)); } } @@ -298,34 +402,31 @@ 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], &vconn); 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); } - if (handling_openflow && mgr->in_band) { + ofmonitor_wait(mgr); + 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) { @@ -336,32 +437,47 @@ connmgr_wait(struct connmgr *mgr, bool handling_openflow) } } -/* 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. */ +/* Adds some memory usage statistics for 'mgr' into 'usage', for use with + * memory_report(). */ void -connmgr_retry(struct connmgr *mgr) +connmgr_get_memory_usage(const struct connmgr *mgr, struct simap *usage) { - struct ofconn *ofconn; + const struct ofconn *ofconn; + unsigned int packets = 0; + unsigned int ofconns = 0; LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { - ofconn->retry = true; + int i; + + ofconns++; + + packets += rconn_count_txqlen(ofconn->rconn); + for (i = 0; i < N_SCHEDULERS; i++) { + struct pinsched_stats stats; + + pinsched_get_stats(ofconn->schedulers[i], &stats); + packets += stats.n_queued; + } } + 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; } /* OpenFlow configuration. */ -static void add_controller(struct connmgr *, const char *target); +static void add_controller(struct connmgr *, const char *target, uint8_t dscp, + 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 *); @@ -396,34 +512,50 @@ 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)); + } } } } @@ -436,9 +568,7 @@ connmgr_free_controller_info(struct shash *info) SHASH_FOR_EACH (node, info) { struct ofproto_controller_info *cinfo = node->data; - while (cinfo->pairs.n) { - free((char *) cinfo->pairs.values[--cinfo->pairs.n]); - } + smap_destroy(&cinfo->pairs); free(cinfo); } shash_destroy(info); @@ -449,7 +579,8 @@ connmgr_free_controller_info(struct shash *info) void connmgr_set_controllers(struct connmgr *mgr, const struct ofproto_controller *controllers, - size_t n_controllers) + size_t n_controllers, uint32_t allowed_versions) + OVS_EXCLUDED(ofproto_mutex) { bool had_controllers = connmgr_has_controllers(mgr); struct shash new_controllers; @@ -457,6 +588,10 @@ 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); @@ -464,12 +599,37 @@ connmgr_set_controllers(struct connmgr *mgr, const struct ofproto_controller *c = &controllers[i]; if (!vconn_verify_name(c->target)) { - if (!find_controller_by_target(mgr, c->target)) { - add_controller(mgr, c->target); + bool add = false; + ofconn = find_controller_by_target(mgr, c->target); + if (!ofconn) { + VLOG_INFO("%s: added primary controller \"%s\"", + mgr->name, c->target); + add = true; + } else if (rconn_get_allowed_versions(ofconn->rconn) != + allowed_versions) { + VLOG_INFO("%s: re-added primary controller \"%s\"", + mgr->name, c->target); + add = true; + ofconn_destroy(ofconn); + } + if (add) { + add_controller(mgr, c->target, c->dscp, allowed_versions); } } else if (!pvconn_verify_name(c->target)) { - if (!ofservice_lookup(mgr, c->target)) { - ofservice_create(mgr, c->target); + bool add = false; + ofservice = ofservice_lookup(mgr, c->target); + if (!ofservice) { + VLOG_INFO("%s: added service controller \"%s\"", + mgr->name, c->target); + add = true; + } else if (ofservice->allowed_versions != allowed_versions) { + VLOG_INFO("%s: re-added service controller \"%s\"", + mgr->name, c->target); + ofservice_destroy(mgr, ofservice); + add = true; + } + if (add) { + ofservice_create(mgr, c->target, allowed_versions, c->dscp); } } else { VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"", @@ -483,10 +643,13 @@ connmgr_set_controllers(struct connmgr *mgr, /* Delete controllers that are no longer configured. * Update configuration of all now-existing controllers. */ 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); @@ -496,11 +659,13 @@ 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); @@ -509,6 +674,8 @@ 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)) { @@ -559,13 +726,15 @@ 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, + uint32_t allowed_versions) + OVS_REQUIRES(ofproto_mutex) { char *name = ofconn_make_name(mgr, target); struct ofconn *ofconn; - ofconn = ofconn_create(mgr, rconn_create(5, 8), OFCONN_PRIMARY, true); - ofconn->pktbuf = pktbuf_create(); + ofconn = ofconn_create(mgr, rconn_create(5, 8, dscp, allowed_versions), + OFCONN_PRIMARY, true); rconn_connect(ofconn->rconn, target, name); hmap_insert(&mgr->controllers, &ofconn->hmap_node, hash_string(target, 0)); @@ -601,18 +770,16 @@ 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++) { @@ -624,12 +791,12 @@ update_in_band_remotes(struct connmgr *mgr) if (!mgr->in_band) { in_band_create(mgr->ofproto, mgr->local_port_name, &mgr->in_band); } - in_band_set_queue(mgr->in_band, mgr->in_band_queue); } else { /* 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_queue(mgr->in_band, mgr->in_band_queue); in_band_set_remotes(mgr->in_band, addrs, n_addrs); } @@ -639,6 +806,7 @@ 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) { @@ -646,7 +814,9 @@ update_fail_open(struct connmgr *mgr) mgr->fail_open = fail_open_create(mgr->ofproto, mgr); } } else { + ovs_mutex_lock(&ofproto_mutex); fail_open_destroy(mgr->fail_open); + ovs_mutex_unlock(&ofproto_mutex); mgr->fail_open = NULL; } } @@ -671,12 +841,11 @@ set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp, SSET_FOR_EACH (name, sset) { struct pvconn *pvconn; int error; - - error = pvconn_open(name, &pvconn); + error = pvconn_open(name, 0, 0, &pvconn); 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; } @@ -696,12 +865,13 @@ static int snoop_preference(const struct ofconn *ofconn) { switch (ofconn->role) { - case NX_ROLE_MASTER: + case OFPCR12_ROLE_MASTER: return 3; - case NX_ROLE_OTHER: + case OFPCR12_ROLE_EQUAL: return 2; - case NX_ROLE_SLAVE: + case OFPCR12_ROLE_SLAVE: return 1; + case OFPCR12_ROLE_NOCHANGE: default: /* Shouldn't happen. */ return 0; @@ -741,26 +911,74 @@ 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 + */ +bool +ofconn_set_master_election_id(struct ofconn *ofconn, uint64_t id) +{ + if (ofconn->connmgr->master_election_id_defined + && + /* Unsigned difference interpreted as a two's complement signed + * value */ + (int64_t)(id - ofconn->connmgr->master_election_id) < 0) { + return false; + } + ofconn->connmgr->master_election_id = id; + ofconn->connmgr->master_election_id_defined = true; + + return true; +} + /* Returns the role configured for 'ofconn'. * - * The default role, if no other role has been set, is NX_ROLE_OTHER. */ -enum nx_role + * The default role, if no other role has been set, is OFPCR12_ROLE_EQUAL. */ +enum ofp12_controller_role ofconn_get_role(const struct ofconn *ofconn) { return ofconn->role; } -/* Changes 'ofconn''s role to 'role'. If 'role' is NX_ROLE_MASTER then any - * existing master is demoted to a slave. */ void -ofconn_set_role(struct ofconn *ofconn, enum nx_role role) +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 == NX_ROLE_MASTER) { + if (role != ofconn->role && role == OFPCR12_ROLE_MASTER) { struct ofconn *other; - HMAP_FOR_EACH (other, hmap_node, &ofconn->connmgr->controllers) { - if (other->role == NX_ROLE_MASTER) { - other->role = NX_ROLE_SLAVE; + 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); } } } @@ -770,41 +988,62 @@ ofconn_set_role(struct ofconn *ofconn, enum nx_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 flow format for 'ofconn', one of NXFF_*. +/* Returns the currently configured protocol for 'ofconn', one of OFPUTIL_P_*. * - * The default, if no other format has been set, is NXFF_OPENFLOW10. */ -enum nx_flow_format -ofconn_get_flow_format(struct ofconn *ofconn) -{ - return ofconn->flow_format; + * Returns OFPUTIL_P_NONE, which is not a valid protocol, if 'ofconn' hasn't + * completed version negotiation. This can't happen if at least one OpenFlow + * message, other than OFPT_HELLO, has been received on the connection (such as + * in ofproto.c's message handling code), since version negotiation is a + * prerequisite for starting to receive messages. This means that + * OFPUTIL_P_NONE is a special case that most callers need not worry about. */ +enum ofputil_protocol +ofconn_get_protocol(const struct ofconn *ofconn) +{ + if (ofconn->protocol == OFPUTIL_P_NONE && + rconn_is_connected(ofconn->rconn)) { + int version = rconn_get_version(ofconn->rconn); + if (version > 0) { + ofconn_set_protocol(CONST_CAST(struct ofconn *, ofconn), + ofputil_protocol_from_ofp_version(version)); + } + } + + return ofconn->protocol; } -/* Sets the flow format for 'ofconn' to 'flow_format' (one of NXFF_*). */ +/* 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_flow_format(struct ofconn *ofconn, enum nx_flow_format flow_format) +ofconn_set_protocol(struct ofconn *ofconn, enum ofputil_protocol protocol) { - ofconn->flow_format = flow_format; + ofconn->protocol = protocol; + update_want_packet_in_on_miss(ofconn); } /* Returns the currently configured packet in format for 'ofconn', one of * NXPIF_*. * - * The default, if no other format has been set, is NXPIF_OPENFLOW10. */ + * The default, if no other format has been set, is NXPIF_STANDARD. */ enum nx_packet_in_format ofconn_get_packet_in_format(struct ofconn *ofconn) { @@ -820,22 +1059,15 @@ ofconn_set_packet_in_format(struct ofconn *ofconn, ofconn->packet_in_format = packet_in_format; } -/* Returns true if the NXT_FLOW_MOD_TABLE_ID extension is enabled, false - * otherwise. +/* Sets the controller connection ID for 'ofconn' to 'controller_id'. * - * By default the extension is not enabled. */ -bool -ofconn_get_flow_mod_table_id(const struct ofconn *ofconn) -{ - return ofconn->flow_mod_table_id; -} - -/* Enables or disables (according to 'enable') the NXT_FLOW_MOD_TABLE_ID - * extension on 'ofconn'. */ + * The connection controller ID is used for OFPP_CONTROLLER and + * NXAST_CONTROLLER actions. See "struct nx_action_controller" for details. */ void -ofconn_set_flow_mod_table_id(struct ofconn *ofconn, bool enable) +ofconn_set_controller_id(struct ofconn *ofconn, uint16_t controller_id) { - ofconn->flow_mod_table_id = enable; + ofconn->controller_id = controller_id; + update_want_packet_in_on_miss(ofconn); } /* Returns the default miss send length for 'ofconn'. */ @@ -854,12 +1086,35 @@ 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) +{ + if (!ofconn->async_cfg) { + ofconn->async_cfg = xmalloc(sizeof *ofconn->async_cfg); + } + *ofconn->async_cfg = *ac; + + if (ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn)) + < OFP14_VERSION) { + if (ofconn->async_cfg->master[OAM_PACKET_IN] & (1u << OFPR_ACTION)) { + ofconn->async_cfg->master[OAM_PACKET_IN] |= OFPR14_ACTION_BITS; + } + if (ofconn->async_cfg->slave[OAM_PACKET_IN] & (1u << OFPR_ACTION)) { + ofconn->async_cfg->slave[OAM_PACKET_IN] |= OFPR14_ACTION_BITS; + } + } +} + +struct ofputil_async_cfg +ofconn_get_async_config(const struct ofconn *ofconn) { - 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) { + 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 @@ -875,76 +1130,131 @@ 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); } } -/* Sends 'error' on 'ofconn', as a reply to 'request'. Only at most the - * first 64 bytes of 'request' are used. */ +/* Sends 'error' on 'ofconn', as a reply to 'request'. */ void ofconn_send_error(const struct ofconn *ofconn, const struct ofp_header *request, enum ofperr error) { + static struct vlog_rate_limit err_rl = VLOG_RATE_LIMIT_INIT(10, 10); 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); + if (!VLOG_DROP_INFO(&err_rl)) { + const char *type_name; + size_t request_len; + enum ofpraw raw; + + request_len = ntohs(request->length); + type_name = (!ofpraw_decode_partial(&raw, request, + MIN(64, request_len)) + ? ofpraw_get_name(raw) + : "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'. */ -enum ofperr -ofconn_pktbuf_retrieve(struct ofconn *ofconn, uint32_t id, - struct ofpbuf **bufferp, uint16_t *in_port) +/* 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) { - return pktbuf_retrieve(ofconn->pktbuf, id, bufferp, in_port); + 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. */ -/* Returns true if 'ofconn' has any pending opgroups. */ -bool -ofconn_has_pending_opgroups(const struct ofconn *ofconn) +static inline uint32_t +bundle_hash(uint32_t id) { - return !list_is_empty(&ofconn->opgroups); + return hash_int(id, 0); +} + +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; +} + +void +ofconn_insert_bundle(struct ofconn *ofconn, struct ofp_bundle *bundle) +{ + hmap_insert(&ofconn->bundles, &bundle->node, bundle_hash(bundle->id)); } -/* 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) +ofconn_remove_bundle(struct ofconn *ofconn, struct ofp_bundle *bundle) +{ + hmap_remove(&ofconn->bundles, &bundle->node); +} + +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); + } +} + +static void +bundle_remove_expired(struct ofconn *ofconn, long long int now) { - list_push_back(&ofconn->opgroups, ofconn_node); + struct ofp_bundle *b, *next; + long long int limit = now - BUNDLE_IDLE_TIMEOUT; + + HMAP_FOR_EACH_SAFE (b, next, node, &ofconn->bundles) { + if (b->used <= limit) { + ofconn_send_error(ofconn, b->msg, OFPERR_OFPBFC_TIMEOUT); + ofp_bundle_remove__(ofconn, b); + } + } } /* Private ofconn functions. */ @@ -958,17 +1268,22 @@ ofconn_get_target(const struct ofconn *ofconn) static struct ofconn * ofconn_create(struct connmgr *mgr, struct rconn *rconn, enum ofconn_type type, bool enable_async_msgs) + OVS_REQUIRES(ofproto_mutex) { struct ofconn *ofconn; ofconn = xzalloc(sizeof *ofconn); ofconn->connmgr = mgr; - list_push_back(&mgr->all_conns, &ofconn->node); + ovs_list_push_back(&mgr->all_conns, &ofconn->node); ofconn->rconn = rconn; ofconn->type = type; ofconn->enable_async_msgs = enable_async_msgs; - list_init(&ofconn->opgroups); + hmap_init(&ofconn->monitors); + ovs_list_init(&ofconn->updates); + + hmap_init(&ofconn->bundles); + ofconn->next_bundle_expiry_check = time_msec() + BUNDLE_EXPIRY_INTERVAL; ofconn_flush(ofconn); @@ -979,25 +1294,16 @@ 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->role = NX_ROLE_OTHER; - ofconn->flow_format = NXFF_OPENFLOW10; - ofconn->packet_in_format = NXPIF_OPENFLOW10; - ofconn->flow_mod_table_id = false; + ofconn_log_flow_mods(ofconn); - /* 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; + ofconn->role = OFPCR12_ROLE_EQUAL; + ofconn_set_protocol(ofconn, OFPUTIL_P_NONE); + ofconn->packet_in_format = NXPIF_STANDARD; rconn_packet_counter_destroy(ofconn->packet_in_counter); ofconn->packet_in_counter = rconn_packet_counter_create(); @@ -1010,60 +1316,55 @@ ofconn_flush(struct ofconn *ofconn) 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); + 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) { + ofmonitor_destroy(monitor); } + rconn_packet_counter_destroy(ofconn->monitor_counter); + ofconn->monitor_counter = rconn_packet_counter_create(); + ofpbuf_list_delete(&ofconn->updates); /* ...but it should be empty. */ } static void ofconn_destroy(struct ofconn *ofconn) + OVS_REQUIRES(ofproto_mutex) { ofconn_flush(ofconn); + /* Force clearing of want_packet_in_on_miss to keep the global count + * accurate. */ + ofconn->controller_id = 1; + update_want_packet_in_on_miss(ofconn); + if (ofconn->type == OFCONN_PRIMARY) { hmap_remove(&ofconn->connmgr->controllers, &ofconn->hmap_node); } - list_remove(&ofconn->node); + bundle_remove_all(ofconn); + hmap_destroy(&ofconn->bundles); + + hmap_destroy(&ofconn->monitors); + ovs_list_remove(&ofconn->node); rconn_destroy(ofconn->rconn); rconn_packet_counter_destroy(ofconn->packet_in_counter); rconn_packet_counter_destroy(ofconn->reply_counter); - pktbuf_destroy(ofconn->pktbuf); + rconn_packet_counter_destroy(ofconn->monitor_counter); free(ofconn); } @@ -1075,6 +1376,7 @@ 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); @@ -1082,6 +1384,12 @@ 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 @@ -1089,57 +1397,64 @@ ofconn_reconfigure(struct ofconn *ofconn, const struct ofproto_controller *c) 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; + 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); } + long long int now = time_msec(); + + if (now >= ofconn->next_bundle_expiry_check) { + ofconn->next_bundle_expiry_check = now + BUNDLE_EXPIRY_INTERVAL; + bundle_remove_expired(ofconn, now); + } + + if (now >= 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; @@ -1147,9 +1462,53 @@ 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 @@ -1159,20 +1518,14 @@ 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; - - assert(reason < 32); - assert((unsigned int) type < OAM_N_TYPES); - - if (!rconn_is_connected(ofconn->rconn)) { - return false; - } + ovs_assert(reason < 32); + ovs_assert((unsigned int) type < OAM_N_TYPES); /* Keep the following code in sync with the documentation in the - * "Asynchronous Messages" section in DESIGN. */ + * "Asynchronous Messages" section in 'topics/design' */ if (ofconn->type == OFCONN_SERVICE && !ofconn->miss_send_len) { /* Service connections don't get asynchronous messages unless they have @@ -1180,14 +1533,24 @@ ofconn_receives_async_msg(const struct ofconn *ofconn, 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; - } + 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; +} - return true; +/* This function returns true to indicate that a packet_in message + * for a "table-miss" should be sent to at least one controller. + * + * False otherwise. */ +bool +connmgr_wants_packet_in_on_miss(struct connmgr *mgr) +{ + int count; + + atomic_read_relaxed(&mgr->want_packet_in_on_miss, &count); + return count > 0; } /* Returns a human-readable name for an OpenFlow connection between 'mgr' and @@ -1226,44 +1589,90 @@ static void 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); - } + ofpmsg_update_length(msg); + rconn_send(ofconn->rconn, msg, counter); } /* Sending asynchronous messages. */ -static void schedule_packet_in(struct ofconn *, struct ofputil_packet_in, - const struct flow *); - /* 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, const struct ofp_phy_port *opp, - uint8_t reason) +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. */ + struct ofputil_port_status ps; struct ofconn *ofconn; + ps.reason = reason; + ps.desc = *pp; LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { if (ofconn_receives_async_msg(ofconn, OAM_PORT_STATUS, reason)) { - struct ofp_port_status *ops; - struct ofpbuf *b; + struct ofpbuf *msg; - ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b); - ops->reason = reason; - ops->desc = *opp; - ofconn_send(ofconn, b, NULL); + /* 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'. */ + * appropriate controllers managed by 'mgr'. + * + * This may be called from the RCU thread. */ void connmgr_send_flow_removed(struct connmgr *mgr, const struct ofputil_flow_removed *fr) + OVS_REQUIRES(ofproto_mutex) { struct ofconn *ofconn; @@ -1276,77 +1685,88 @@ connmgr_send_flow_removed(struct connmgr *mgr, * 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); + msg = ofputil_encode_flow_removed(fr, ofconn_get_protocol(ofconn)); ofconn_send_reply(ofconn, msg); } } } -/* Given 'pin', sends an OFPT_PACKET_IN message to each OpenFlow controller as - * necessary according to their individual configurations. */ +/* Sends an OFPT_TABLE_STATUS message with 'reason' to appropriate controllers + * managed by 'mgr'. When the table state changes, the controller needs to be + * informed with the OFPT_TABLE_STATUS message. The reason values + * OFPTR_VACANCY_DOWN and OFPTR_VACANCY_UP identify a vacancy message. The + * vacancy events are generated when the remaining space in the flow table + * changes and crosses one of the vacancy thereshold specified by + * OFPT_TABLE_MOD. */ void -connmgr_send_packet_in(struct connmgr *mgr, - const struct ofputil_packet_in *pin, - const struct flow *flow) +connmgr_send_table_status(struct connmgr *mgr, + const struct ofputil_table_desc *td, + uint8_t reason) { + struct ofputil_table_status ts; struct ofconn *ofconn; + ts.reason = reason; + ts.desc = *td; + LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { - if (ofconn_receives_async_msg(ofconn, OAM_PACKET_IN, pin->reason)) { - schedule_packet_in(ofconn, *pin, flow); + if (ofconn_receives_async_msg(ofconn, OAM_TABLE_STATUS, reason)) { + struct ofpbuf *msg; + + msg = ofputil_encode_table_status(&ts, + ofconn_get_protocol(ofconn)); + if (msg) { + ofconn_send(ofconn, msg, NULL); + } } } } -/* pinsched callback for sending 'ofp_packet_in' on 'ofconn'. */ -static void -do_send_packet_in(struct ofpbuf *ofp_packet_in, void *ofconn_) +/* Given 'pin', sends an OFPT_PACKET_IN message to each OpenFlow controller as + * necessary according to their individual configurations. */ +void +connmgr_send_async_msg(struct connmgr *mgr, + const struct ofproto_async_msg *am) { - struct ofconn *ofconn = ofconn_; + struct ofconn *ofconn; + + LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { + enum ofputil_protocol protocol = ofconn_get_protocol(ofconn); + if (protocol == OFPUTIL_P_NONE || !rconn_is_connected(ofconn->rconn) + || ofconn->controller_id != am->controller_id + || !ofconn_receives_async_msg(ofconn, am->oam, + am->pin.up.base.reason)) { + continue; + } + + struct ofpbuf *msg = ofputil_encode_packet_in_private( + &am->pin.up, protocol, ofconn->packet_in_format); - rconn_send_with_limit(ofconn->rconn, ofp_packet_in, - ofconn->packet_in_counter, 100); + struct ovs_list txq; + bool is_miss = (am->pin.up.base.reason == OFPR_NO_MATCH || + am->pin.up.base.reason == OFPR_EXPLICIT_MISS || + am->pin.up.base.reason == OFPR_IMPLICIT_MISS); + pinsched_send(ofconn->schedulers[is_miss], + am->pin.up.base.flow_metadata.flow.in_port.ofp_port, + msg, &txq); + do_send_packet_ins(ofconn, &txq); + } } -/* Takes 'pin', 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. */ static void -schedule_packet_in(struct ofconn *ofconn, struct ofputil_packet_in pin, - const struct flow *flow) +do_send_packet_ins(struct ofconn *ofconn, struct ovs_list *txq) { - struct connmgr *mgr = ofconn->connmgr; + struct ofpbuf *pin; - /* 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; - } else { - pin.buffer_id = pktbuf_save(ofconn->pktbuf, pin.packet, pin.packet_len, - flow->in_port); - } + 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 rll = VLOG_RATE_LIMIT_INIT(5, 5); - /* Figure out how much of the packet to send. */ - if (pin.reason == OFPR_NO_MATCH) { - pin.send_len = pin.packet_len; - } else { - /* Caller should have initialized 'send_len' to 'max_len' specified in - * struct ofp_action_output. */ - } - if (pin.buffer_id != UINT32_MAX) { - pin.send_len = MIN(pin.send_len, ofconn->miss_send_len); + VLOG_INFO_RL(&rll, "%s: dropping packet-in due to queue overflow", + rconn_get_name(ofconn->rconn)); + } } - - /* 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], - flow->in_port, - ofputil_encode_packet_in(&pin, ofconn->packet_in_format), - do_send_packet_in, ofconn); } /* Fail-open settings. */ @@ -1441,29 +1861,6 @@ 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. */ @@ -1526,18 +1923,9 @@ 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, - const struct nlattr *odp_actions, - size_t actions_len) -{ - return !mgr->in_band || in_band_rule_check(flow, odp_actions, actions_len); + return mgr->in_band != NULL; } /* Fail-open and in-band implementation. */ @@ -1548,6 +1936,7 @@ 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); @@ -1559,16 +1948,35 @@ 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 cls_rule rule; + struct ofpbuf ofpacts; + struct match match; + + ofpbuf_init(&ofpacts, OFPACT_OUTPUT_SIZE); + ofpact_put_OUTPUT(&ofpacts)->port = OFPP_NORMAL; + + match_init_catchall(&match); + ofproto_add_flow(mgr->ofproto, &match, 0, ofpacts.data, + ofpacts.size); + + ofpbuf_uninit(&ofpacts); + } +} - memset(&action, 0, sizeof action); - action.type = htons(OFPAT_OUTPUT); - action.output.len = htons(sizeof action); - action.output.port = htons(OFPP_NORMAL); - cls_rule_init_catchall(&rule, 0); - ofproto_add_flow(mgr->ofproto, &rule, &action, 1); +/* 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, @@ -1577,13 +1985,14 @@ 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, + uint32_t allowed_versions, uint8_t dscp) { struct ofservice *ofservice; struct pvconn *pvconn; int error; - error = pvconn_open(target, &pvconn); + error = pvconn_open(target, allowed_versions, dscp, &pvconn); if (error) { return error; } @@ -1591,6 +2000,7 @@ ofservice_create(struct connmgr *mgr, const char *target) ofservice = xzalloc(sizeof *ofservice); hmap_insert(&mgr->services, &ofservice->node, hash_string(target, 0)); ofservice->pvconn = pvconn; + ofservice->allowed_versions = allowed_versions; return 0; } @@ -1611,6 +2021,7 @@ ofservice_reconfigure(struct ofservice *ofservice, 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 @@ -1628,3 +2039,288 @@ ofservice_lookup(struct connmgr *mgr, const char *target) } return NULL; } + +/* Flow monitors (NXST_FLOW_MONITOR). */ + +/* A counter incremented when something significant happens to an OpenFlow + * rule. + * + * - When a rule is added, its 'add_seqno' and 'modify_seqno' are set to + * the current value (which is then incremented). + * + * - When a rule is modified, its 'modify_seqno' is set to the current + * value (which is then incremented). + * + * Thus, by comparing an old value of monitor_seqno against a rule's + * 'add_seqno', one can tell whether the rule was added before or after the old + * value was read, and similarly for 'modify_seqno'. + * + * 32 bits should normally be sufficient (and would be nice, to save space in + * each rule) but then we'd have to have some special cases for wraparound. + * + * We initialize monitor_seqno to 1 to allow 0 to be used as an invalid + * value. */ +static uint64_t monitor_seqno = 1; + +COVERAGE_DEFINE(ofmonitor_pause); +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; + + *monitorp = NULL; + + m = ofmonitor_lookup(ofconn, request->id); + if (m) { + return OFPERR_OFPMOFC_MONITOR_EXISTS; + } + + m = xmalloc(sizeof *m); + m->ofconn = ofconn; + hmap_insert(&ofconn->monitors, &m->ofconn_node, hash_int(request->id, 0)); + m->id = request->id; + m->flags = request->flags; + m->out_port = request->out_port; + m->table_id = request->table_id; + minimatch_init(&m->match, &request->match); + + *monitorp = m; + return 0; +} + +struct ofmonitor * +ofmonitor_lookup(struct ofconn *ofconn, uint32_t id) + OVS_REQUIRES(ofproto_mutex) +{ + struct ofmonitor *m; + + HMAP_FOR_EACH_IN_BUCKET (m, ofconn_node, hash_int(id, 0), + &ofconn->monitors) { + if (m->id == id) { + return m; + } + } + return NULL; +} + +void +ofmonitor_destroy(struct ofmonitor *m) + OVS_REQUIRES(ofproto_mutex) +{ + if (m) { + minimatch_destroy(&m->match); + hmap_remove(&m->ofconn->monitors, &m->ofconn_node); + free(m); + } +} + +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 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; + rule->add_seqno = rule->modify_seqno = monitor_seqno++; + break; + + case NXFME_DELETED: + update = NXFMF_DELETE; + break; + + case NXFME_MODIFIED: + update = NXFMF_MODIFY; + rule->modify_seqno = monitor_seqno++; + break; + + default: + case NXFME_ABBREV: + OVS_NOT_REACHED(); + } + + LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { + enum nx_flow_monitor_flags flags = 0; + struct ofmonitor *m; + + if (ofconn->monitor_paused) { + /* Only send NXFME_DELETED notifications for flows that were added + * before we paused. */ + if (event != NXFME_DELETED + || rule->add_seqno > ofconn->monitor_paused) { + continue; + } + } + + HMAP_FOR_EACH (m, ofconn_node, &ofconn->monitors) { + if (m->flags & update + && (m->table_id == 0xff || m->table_id == rule->table_id) + && (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; + } + } + + if (flags) { + if (ovs_list_is_empty(&ofconn->updates)) { + ofputil_start_flow_update(&ofconn->updates); + ofconn->sent_abbrev_update = false; + } + + if (flags & NXFMF_OWN || ofconn != abbrev_ofconn + || ofconn->monitor_paused) { + struct ofputil_flow_update fu; + + fu.event = event; + fu.reason = event == NXFME_DELETED ? reason : 0; + fu.table_id = rule->table_id; + fu.cookie = rule->flow_cookie; + minimatch_expand(&rule->cr.match, &fu.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) { + 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; + } + ofputil_append_flow_update(&fu, &ofconn->updates, + ofproto_get_tun_tab(rule->ofproto)); + } else if (!ofconn->sent_abbrev_update) { + struct ofputil_flow_update fu; + + fu.event = NXFME_ABBREV; + fu.xid = abbrev_xid; + ofputil_append_flow_update(&fu, &ofconn->updates, + ofproto_get_tun_tab(rule->ofproto)); + + ofconn->sent_abbrev_update = true; + } + } + } +} + +void +ofmonitor_flush(struct connmgr *mgr) + OVS_REQUIRES(ofproto_mutex) +{ + struct ofconn *ofconn; + + LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { + struct rconn_packet_counter *counter = ofconn->monitor_counter; + + struct ofpbuf *msg; + LIST_FOR_EACH_POP (msg, list_node, &ofconn->updates) { + ofconn_send(ofconn, msg, counter); + } + + if (!ofconn->monitor_paused + && rconn_packet_counter_n_bytes(counter) > 128 * 1024) { + struct ofpbuf *pause; + + COVERAGE_INC(ofmonitor_pause); + ofconn->monitor_paused = monitor_seqno++; + pause = ofpraw_alloc_xid(OFPRAW_NXT_FLOW_MONITOR_PAUSED, + OFP10_VERSION, htonl(0), 0); + ofconn_send(ofconn, pause, counter); + } + } +} + +static void +ofmonitor_resume(struct ofconn *ofconn) + OVS_REQUIRES(ofproto_mutex) +{ + struct rule_collection rules; + struct ofpbuf *resumed; + struct ofmonitor *m; + struct ovs_list msgs; + + rule_collection_init(&rules); + HMAP_FOR_EACH (m, ofconn_node, &ofconn->monitors) { + ofmonitor_collect_resume_rules(m, ofconn->monitor_paused, &rules); + } + + ovs_list_init(&msgs); + ofmonitor_compose_refresh_updates(&rules, &msgs); + + resumed = ofpraw_alloc_xid(OFPRAW_NXT_FLOW_MONITOR_RESUMED, OFP10_VERSION, + htonl(0), 0); + ovs_list_push_back(&msgs, &resumed->list_node); + ofconn_send_replies(ofconn, &msgs); + + 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 (ofmonitor_may_resume(ofconn)) { + COVERAGE_INC(ofmonitor_resume); + ofmonitor_resume(ofconn); + } + } + ovs_mutex_unlock(&ofproto_mutex); +} + +static void +ofmonitor_wait(struct connmgr *mgr) +{ + struct ofconn *ofconn; + + ovs_mutex_lock(&ofproto_mutex); + LIST_FOR_EACH (ofconn, node, &mgr->all_conns) { + if (ofmonitor_may_resume(ofconn)) { + poll_immediate_wake(); + } + } + ovs_mutex_unlock(&ofproto_mutex); +} + +void +ofproto_async_msg_free(struct ofproto_async_msg *am) +{ + free(am->pin.up.base.packet); + free(am->pin.up.base.userdata); + free(am->pin.up.stack); + free(am->pin.up.actions); + free(am->pin.up.action_set); + free(am); +}