X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=ofproto%2Fofproto.c;h=8fc9916d5be5a295637c62a4f005c99709c13747;hb=a6b7506dab305d91fc5f2ac6416a714e5fa09dd4;hp=2b3f09482afc5f5fc25b5db711a3c8eb3a4ced02;hpb=2c91602834c1955a0c0acfc5103ba1d6127cc3f1;p=sliver-openvswitch.git diff --git a/ofproto/ofproto.c b/ofproto/ofproto.c index 2b3f09482..8fc9916d5 100644 --- a/ofproto/ofproto.c +++ b/ofproto/ofproto.c @@ -25,6 +25,7 @@ #include "bitmap.h" #include "byte-order.h" #include "classifier.h" +#include "connectivity.h" #include "connmgr.h" #include "coverage.h" #include "dynamic-string.h" @@ -47,8 +48,10 @@ #include "pktbuf.h" #include "poll-loop.h" #include "random.h" +#include "seq.h" #include "shash.h" #include "simap.h" +#include "smap.h" #include "sset.h" #include "timeval.h" #include "unaligned.h" @@ -57,14 +60,11 @@ VLOG_DEFINE_THIS_MODULE(ofproto); -COVERAGE_DEFINE(ofproto_error); COVERAGE_DEFINE(ofproto_flush); -COVERAGE_DEFINE(ofproto_no_packet_in); COVERAGE_DEFINE(ofproto_packet_out); COVERAGE_DEFINE(ofproto_queue_req); COVERAGE_DEFINE(ofproto_recv_openflow); COVERAGE_DEFINE(ofproto_reinit_ports); -COVERAGE_DEFINE(ofproto_uninstallable); COVERAGE_DEFINE(ofproto_update_port); enum ofproto_state { @@ -153,10 +153,9 @@ static void oftable_enable_eviction(struct oftable *, const struct mf_subfield *fields, size_t n_fields); -static void oftable_remove_rule(struct rule *rule) OVS_RELEASES(rule->rwlock); -static void oftable_remove_rule__(struct ofproto *ofproto, - struct classifier *cls, struct rule *rule) - OVS_REQ_WRLOCK(cls->rwlock) OVS_RELEASES(rule->rwlock); +static void oftable_remove_rule(struct rule *rule) OVS_REQUIRES(ofproto_mutex); +static void oftable_remove_rule__(struct ofproto *, struct rule *) + OVS_REQUIRES(ofproto_mutex); static void oftable_insert_rule(struct rule *); /* A set of rules within a single OpenFlow table (oftable) that have the same @@ -181,9 +180,8 @@ struct eviction_group { struct heap rules; /* Contains "struct rule"s. */ }; -static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep) - OVS_TRY_WRLOCK(true, (*rulep)->rwlock); -static void ofproto_evict(struct ofproto *); +static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep); +static void ofproto_evict(struct ofproto *) OVS_EXCLUDED(ofproto_mutex); static uint32_t rule_eviction_priority(struct rule *); static void eviction_group_add_rule(struct rule *); static void eviction_group_remove_rule(struct rule *); @@ -220,7 +218,10 @@ static void rule_criteria_init(struct rule_criteria *, uint8_t table_id, ofp_port_t out_port, uint32_t out_group); static void rule_criteria_destroy(struct rule_criteria *); -/* A packet that needs to be passed to rule_execute(). */ +/* A packet that needs to be passed to rule_execute(). + * + * (We can't do this immediately from ofopgroup_complete() because that holds + * ofproto_mutex, which rule_execute() needs released.) */ struct rule_execute { struct list list_node; /* In struct ofproto's "rule_executes" list. */ struct rule *rule; /* Owns a reference to the rule. */ @@ -228,17 +229,34 @@ struct rule_execute { struct ofpbuf *packet; /* Owns the packet. */ }; -static void run_rule_executes(struct ofproto *); +static void run_rule_executes(struct ofproto *) OVS_EXCLUDED(ofproto_mutex); static void destroy_rule_executes(struct ofproto *); /* ofport. */ -static void ofport_destroy__(struct ofport *); +static void ofport_destroy__(struct ofport *) OVS_EXCLUDED(ofproto_mutex); static void ofport_destroy(struct ofport *); static void update_port(struct ofproto *, const char *devname); static int init_ports(struct ofproto *); static void reinit_ports(struct ofproto *); +static long long int ofport_get_usage(const struct ofproto *, + ofp_port_t ofp_port); +static void ofport_set_usage(struct ofproto *, ofp_port_t ofp_port, + long long int last_used); +static void ofport_remove_usage(struct ofproto *, ofp_port_t ofp_port); + +/* Ofport usage. + * + * Keeps track of the currently used and recently used ofport values and is + * used to prevent immediate recycling of ofport values. */ +struct ofport_usage { + struct hmap_node hmap_node; /* In struct ofproto's "ofport_usage" hmap. */ + ofp_port_t ofp_port; /* OpenFlow port number. */ + long long int last_used; /* Last time the 'ofp_port' was used. LLONG_MAX + represents in-use ofports. */ +}; + /* rule. */ static void ofproto_rule_destroy__(struct rule *); static void ofproto_rule_send_removed(struct rule *, uint8_t reason); @@ -254,12 +272,19 @@ static enum ofperr modify_flows__(struct ofproto *, struct ofconn *, const struct rule_collection *); static void delete_flow__(struct rule *rule, struct ofopgroup *, enum ofp_flow_removed_reason) - OVS_RELEASES(rule->rwlock); + OVS_REQUIRES(ofproto_mutex); +static bool ofproto_group_exists__(const struct ofproto *ofproto, + uint32_t group_id) + OVS_REQ_RDLOCK(ofproto->groups_rwlock); +static bool ofproto_group_exists(const struct ofproto *ofproto, + uint32_t group_id) + OVS_EXCLUDED(ofproto->groups_rwlock); static enum ofperr add_group(struct ofproto *, struct ofputil_group_mod *); static bool handle_openflow(struct ofconn *, const struct ofpbuf *); static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *, struct ofputil_flow_mod *, - const struct ofp_header *); + const struct ofp_header *) + OVS_EXCLUDED(ofproto_mutex); static void calc_duration(long long int start, long long int now, uint32_t *sec, uint32_t *nsec); @@ -278,7 +303,8 @@ static const struct ofproto_class **ofproto_classes; static size_t n_ofproto_classes; static size_t allocated_ofproto_classes; -struct ovs_mutex ofproto_mutex; +/* Global lock that protects all flow table operations. */ +struct ovs_mutex ofproto_mutex = OVS_MUTEX_INITIALIZER; unsigned flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT; unsigned n_handler_threads; @@ -310,8 +336,6 @@ ofproto_init(const struct shash *iface_hints) struct shash_node *node; size_t i; - ovs_mutex_init_recursive(&ofproto_mutex); - ofproto_class_register(&ofproto_dpif_class); /* Make a local copy, since we don't own 'iface_hints' elements. */ @@ -410,6 +434,7 @@ ofproto_enumerate_types(struct sset *types) { size_t i; + sset_clear(types); for (i = 0; i < n_ofproto_classes; i++) { ofproto_classes[i]->enumerate_types(types); } @@ -436,7 +461,7 @@ ofproto_enumerate_names(const char *type, struct sset *names) { const struct ofproto_class *class = ofproto_class_find__(type); return class ? class->enumerate_names(type, names) : EAFNOSUPPORT; - } +} int ofproto_create(const char *datapath_name, const char *datapath_type, @@ -484,6 +509,7 @@ ofproto_create(const char *datapath_name, const char *datapath_type, ofproto->dp_desc = NULL; ofproto->frag_handling = OFPC_FRAG_NORMAL; hmap_init(&ofproto->ports); + hmap_init(&ofproto->ofport_usage); shash_init(&ofproto->port_by_name); simap_init(&ofproto->ofp_requests); ofproto->max_ports = ofp_to_u16(OFPP_MAX); @@ -508,6 +534,30 @@ ofproto_create(const char *datapath_name, const char *datapath_type, ovs_rwlock_init(&ofproto->groups_rwlock); hmap_init(&ofproto->groups); ovs_mutex_unlock(&ofproto_mutex); + ofproto->ogf.capabilities = OFPGFC_CHAINING | OFPGFC_SELECT_LIVENESS | + OFPGFC_SELECT_WEIGHT; + ofproto->ogf.max_groups[OFPGT11_ALL] = OFPG_MAX; + ofproto->ogf.max_groups[OFPGT11_SELECT] = OFPG_MAX; + ofproto->ogf.max_groups[OFPGT11_INDIRECT] = OFPG_MAX; + ofproto->ogf.max_groups[OFPGT11_FF] = OFPG_MAX; + ofproto->ogf.actions[0] = + (1 << OFPAT11_OUTPUT) | + (1 << OFPAT11_COPY_TTL_OUT) | + (1 << OFPAT11_COPY_TTL_IN) | + (1 << OFPAT11_SET_MPLS_TTL) | + (1 << OFPAT11_DEC_MPLS_TTL) | + (1 << OFPAT11_PUSH_VLAN) | + (1 << OFPAT11_POP_VLAN) | + (1 << OFPAT11_PUSH_MPLS) | + (1 << OFPAT11_POP_MPLS) | + (1 << OFPAT11_SET_QUEUE) | + (1 << OFPAT11_GROUP) | + (1 << OFPAT11_SET_NW_TTL) | + (1 << OFPAT11_DEC_NW_TTL) | + (1 << OFPAT12_SET_FIELD); +/* not supported: + * (1 << OFPAT13_PUSH_PBB) | + * (1 << OFPAT13_POP_PBB) */ error = ofproto->ofproto_class->construct(ofproto); if (error) { @@ -517,11 +567,6 @@ ofproto_create(const char *datapath_name, const char *datapath_type, return error; } - /* The "max_ports" member should have been set by ->construct(ofproto). - * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */ - ofproto->ofp_port_ids = bitmap_allocate(ofproto->max_ports); - bitmap_set1(ofproto->ofp_port_ids, 0); - /* Check that hidden tables, if any, are at the end. */ ovs_assert(ofproto->n_tables); for (i = 0; i + 1 < ofproto->n_tables; i++) { @@ -696,7 +741,7 @@ ofproto_set_n_handler_threads(unsigned limit) if (limit) { n_handler_threads = limit; } else { - int n_proc = sysconf(_SC_NPROCESSORS_ONLN); + int n_proc = count_cpu_cores(); n_handler_threads = n_proc > 2 ? n_proc - 2 : 1; } } @@ -842,6 +887,27 @@ ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port, ? ofproto->ofproto_class->get_stp_port_status(ofport, s) : EOPNOTSUPP); } + +/* Retrieves STP port statistics of 'ofp_port' on 'ofproto' and stores it in + * 's'. If the 'enabled' member in 's' is false, then the other members + * are not meaningful. + * + * Returns 0 if successful, otherwise a positive errno value.*/ +int +ofproto_port_get_stp_stats(struct ofproto *ofproto, ofp_port_t ofp_port, + struct ofproto_port_stp_stats *s) +{ + struct ofport *ofport = ofproto_get_port(ofproto, ofp_port); + if (!ofport) { + VLOG_WARN_RL(&rl, "%s: cannot get STP stats on nonexistent " + "port %"PRIu16, ofproto->name, ofp_port); + return ENODEV; + } + + return (ofproto->ofproto_class->get_stp_port_stats + ? ofproto->ofproto_class->get_stp_port_stats(ofport, s) + : EOPNOTSUPP); +} /* Queue DSCP configuration. */ @@ -1096,7 +1162,7 @@ ofproto_configure_table(struct ofproto *ofproto, int table_id, } table->max_flows = s->max_flows; - ovs_rwlock_rdlock(&table->cls.rwlock); + ovs_rwlock_wrlock(&table->cls.rwlock); if (classifier_count(&table->cls) > table->max_flows && table->eviction_fields) { /* 'table' contains more flows than allowed. We might not be able to @@ -1112,6 +1178,10 @@ ofproto_configure_table(struct ofproto *ofproto, int table_id, break; } } + + classifier_set_prefix_fields(&table->cls, + s->prefix_fields, s->n_prefix_fields); + ovs_rwlock_unlock(&table->cls.rwlock); } @@ -1127,6 +1197,20 @@ ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops) connmgr_get_snoops(ofproto->connmgr, snoops); } +static void +ofproto_rule_delete__(struct ofproto *ofproto, struct rule *rule, + uint8_t reason) + OVS_REQUIRES(ofproto_mutex) +{ + struct ofopgroup *group; + + ovs_assert(!rule->pending); + + group = ofopgroup_create_unattached(ofproto); + delete_flow__(rule, group, reason); + ofopgroup_submit(group); +} + /* Deletes 'rule' from 'cls' within 'ofproto'. * * Within an ofproto implementation, this function allows an ofproto @@ -1134,33 +1218,29 @@ ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops) * function is called. This function is not suitable for use elsewhere in an * ofproto implementation. * - * This function is also used internally in ofproto.c. - * * This function implements steps 4.4 and 4.5 in the section titled "Rule Life - * Cycle" in ofproto-provider.h. - - * The 'cls' argument is redundant (it is &ofproto->tables[rule->table_id].cls) - * but it allows Clang to do better checking. */ + * Cycle" in ofproto-provider.h. */ void -ofproto_rule_delete(struct ofproto *ofproto, struct classifier *cls, - struct rule *rule) - OVS_REQ_WRLOCK(cls->rwlock) +ofproto_rule_delete(struct ofproto *ofproto, struct rule *rule) + OVS_EXCLUDED(ofproto_mutex) { struct ofopgroup *group; + ovs_mutex_lock(&ofproto_mutex); ovs_assert(!rule->pending); - ovs_assert(cls == &ofproto->tables[rule->table_id].cls); group = ofopgroup_create_unattached(ofproto); ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE); - ovs_rwlock_wrlock(&rule->rwlock); - oftable_remove_rule__(ofproto, cls, rule); + oftable_remove_rule__(ofproto, rule); ofproto->ofproto_class->rule_delete(rule); ofopgroup_submit(group); + + ovs_mutex_unlock(&ofproto_mutex); } static void ofproto_flush__(struct ofproto *ofproto) + OVS_EXCLUDED(ofproto_mutex) { struct oftable *table; @@ -1168,6 +1248,7 @@ ofproto_flush__(struct ofproto *ofproto) ofproto->ofproto_class->flush(ofproto); } + ovs_mutex_lock(&ofproto_mutex); OFPROTO_FOR_EACH_TABLE (table, ofproto) { struct rule *rule, *next_rule; struct cls_cursor cursor; @@ -1176,26 +1257,27 @@ ofproto_flush__(struct ofproto *ofproto) continue; } - ovs_rwlock_wrlock(&table->cls.rwlock); + ovs_rwlock_rdlock(&table->cls.rwlock); cls_cursor_init(&cursor, &table->cls, NULL); + ovs_rwlock_unlock(&table->cls.rwlock); CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { if (!rule->pending) { - ofproto_rule_delete(ofproto, &table->cls, rule); + ofproto_rule_delete__(ofproto, rule, OFPRR_DELETE); } } - ovs_rwlock_unlock(&table->cls.rwlock); } + ovs_mutex_unlock(&ofproto_mutex); } static void delete_group(struct ofproto *ofproto, uint32_t group_id); static void ofproto_destroy__(struct ofproto *ofproto) + OVS_EXCLUDED(ofproto_mutex) { struct oftable *table; ovs_assert(list_is_empty(&ofproto->pending)); - ovs_assert(!ofproto->n_pending); destroy_rule_executes(ofproto); guarded_list_destroy(&ofproto->rule_executes); @@ -1215,8 +1297,8 @@ ofproto_destroy__(struct ofproto *ofproto) free(ofproto->serial_desc); free(ofproto->dp_desc); hmap_destroy(&ofproto->ports); + hmap_destroy(&ofproto->ofport_usage); shash_destroy(&ofproto->port_by_name); - bitmap_free(ofproto->ofp_port_ids); simap_destroy(&ofproto->ofp_requests); OFPROTO_FOR_EACH_TABLE (table, ofproto) { @@ -1233,8 +1315,10 @@ ofproto_destroy__(struct ofproto *ofproto) void ofproto_destroy(struct ofproto *p) + OVS_EXCLUDED(ofproto_mutex) { struct ofport *ofport, *next_ofport; + struct ofport_usage *usage, *next_usage; if (!p) { return; @@ -1252,6 +1336,11 @@ ofproto_destroy(struct ofproto *p) ofport_destroy(ofport); } + HMAP_FOR_EACH_SAFE (usage, next_usage, hmap_node, &p->ofport_usage) { + hmap_remove(&p->ofport_usage, &usage->hmap_node); + free(usage); + } + p->ofproto_class->destruct(p); ofproto_destroy__(p); } @@ -1299,43 +1388,37 @@ ofproto_type_run(const char *datapath_type) return error; } -int -ofproto_type_run_fast(const char *datapath_type) +void +ofproto_type_wait(const char *datapath_type) { const struct ofproto_class *class; - int error; datapath_type = ofproto_normalize_type(datapath_type); class = ofproto_class_find__(datapath_type); - error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0; - if (error && error != EAGAIN) { - VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)", - datapath_type, ovs_strerror(error)); + if (class->type_wait) { + class->type_wait(datapath_type); } - return error; } -void -ofproto_type_wait(const char *datapath_type) +static bool +any_pending_ops(const struct ofproto *p) + OVS_EXCLUDED(ofproto_mutex) { - const struct ofproto_class *class; + bool b; - datapath_type = ofproto_normalize_type(datapath_type); - class = ofproto_class_find__(datapath_type); + ovs_mutex_lock(&ofproto_mutex); + b = !list_is_empty(&p->pending); + ovs_mutex_unlock(&ofproto_mutex); - if (class->type_wait) { - class->type_wait(datapath_type); - } + return b; } int ofproto_run(struct ofproto *p) { - struct sset changed_netdevs; - const char *changed_netdev; - struct ofport *ofport; int error; + uint64_t new_seq; error = p->ofproto_class->run(p); if (error && error != EAGAIN) { @@ -1360,6 +1443,7 @@ ofproto_run(struct ofproto *p) continue; } + ovs_mutex_lock(&ofproto_mutex); HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) { heap_rebuild(&evg->rules); } @@ -1373,6 +1457,7 @@ ofproto_run(struct ofproto *p) } } ovs_rwlock_unlock(&table->cls.rwlock); + ovs_mutex_unlock(&ofproto_mutex); } } @@ -1384,24 +1469,29 @@ ofproto_run(struct ofproto *p) } } - /* Update OpenFlow port status for any port whose netdev has changed. - * - * Refreshing a given 'ofport' can cause an arbitrary ofport to be - * destroyed, so it's not safe to update ports directly from the - * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we - * need this two-phase approach. */ - sset_init(&changed_netdevs); - HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { - unsigned int change_seq = netdev_change_seq(ofport->netdev); - if (ofport->change_seq != change_seq) { - ofport->change_seq = change_seq; - sset_add(&changed_netdevs, netdev_get_name(ofport->netdev)); + new_seq = seq_read(connectivity_seq_get()); + if (new_seq != p->change_seq) { + struct sset devnames; + const char *devname; + struct ofport *ofport; + + /* Update OpenFlow port status for any port whose netdev has changed. + * + * Refreshing a given 'ofport' can cause an arbitrary ofport to be + * destroyed, so it's not safe to update ports directly from the + * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we + * need this two-phase approach. */ + sset_init(&devnames); + HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { + sset_add(&devnames, netdev_get_name(ofport->netdev)); } + SSET_FOR_EACH (devname, &devnames) { + update_port(p, devname); + } + sset_destroy(&devnames); + + p->change_seq = new_seq; } - SSET_FOR_EACH (changed_netdev, &changed_netdevs) { - update_port(p, changed_netdev); - } - sset_destroy(&changed_netdevs); switch (p->state) { case S_OPENFLOW: @@ -1411,7 +1501,7 @@ ofproto_run(struct ofproto *p) case S_EVICT: connmgr_run(p->connmgr, NULL); ofproto_evict(p); - if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) { + if (!any_pending_ops(p)) { p->state = S_OPENFLOW; } break; @@ -1419,7 +1509,7 @@ ofproto_run(struct ofproto *p) case S_FLUSH: connmgr_run(p->connmgr, NULL); ofproto_flush__(p); - if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) { + if (!any_pending_ops(p)) { connmgr_flushed(p->connmgr); p->state = S_OPENFLOW; } @@ -1469,40 +1559,14 @@ ofproto_run(struct ofproto *p) return error; } -/* Performs periodic activity required by 'ofproto' that needs to be done - * with the least possible latency. - * - * It makes sense to call this function a couple of times per poll loop, to - * provide a significant performance boost on some benchmarks with the - * ofproto-dpif implementation. */ -int -ofproto_run_fast(struct ofproto *p) -{ - int error; - - error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0; - if (error && error != EAGAIN) { - VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)", - p->name, ovs_strerror(error)); - } - return error; -} - void ofproto_wait(struct ofproto *p) { - struct ofport *ofport; - p->ofproto_class->wait(p); if (p->ofproto_class->port_poll_wait) { p->ofproto_class->port_poll_wait(p); } - - HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { - if (ofport->change_seq != netdev_change_seq(ofport->netdev)) { - poll_immediate_wake(); - } - } + seq_wait(connectivity_seq_get(), p->change_seq); switch (p->state) { case S_OPENFLOW: @@ -1512,7 +1576,7 @@ ofproto_wait(struct ofproto *p) case S_EVICT: case S_FLUSH: connmgr_wait(p->connmgr, false); - if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) { + if (!any_pending_ops(p)) { poll_immediate_wake(); } break; @@ -1534,8 +1598,11 @@ ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage) unsigned int n_rules; simap_increase(usage, "ports", hmap_count(&ofproto->ports)); + + ovs_mutex_lock(&ofproto_mutex); simap_increase(usage, "ops", ofproto->n_pending + hmap_count(&ofproto->deletions)); + ovs_mutex_unlock(&ofproto_mutex); n_rules = 0; OFPROTO_FOR_EACH_TABLE (table, ofproto) { @@ -1693,12 +1760,18 @@ ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev, update_port(ofproto, netdev_name); } if (ofp_portp) { - struct ofproto_port ofproto_port; - - ofproto_port_query_by_name(ofproto, netdev_get_name(netdev), - &ofproto_port); - *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port; - ofproto_port_destroy(&ofproto_port); + *ofp_portp = OFPP_NONE; + if (!error) { + struct ofproto_port ofproto_port; + + error = ofproto_port_query_by_name(ofproto, + netdev_get_name(netdev), + &ofproto_port); + if (!error) { + *ofp_portp = ofproto_port.ofp_port; + ofproto_port_destroy(&ofproto_port); + } + } } return error; } @@ -1750,6 +1823,30 @@ ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port) return error; } +static void +flow_mod_init(struct ofputil_flow_mod *fm, + const struct match *match, unsigned int priority, + const struct ofpact *ofpacts, size_t ofpacts_len, + enum ofp_flow_mod_command command) +{ + memset(fm, 0, sizeof *fm); + fm->match = *match; + fm->priority = priority; + fm->cookie = 0; + fm->new_cookie = 0; + fm->modify_cookie = false; + fm->table_id = 0; + fm->command = command; + fm->idle_timeout = 0; + fm->hard_timeout = 0; + fm->buffer_id = UINT32_MAX; + fm->out_port = OFPP_ANY; + fm->out_group = OFPG_ANY; + fm->flags = 0; + fm->ofpacts = CONST_CAST(struct ofpact *, ofpacts); + fm->ofpacts_len = ofpacts_len; +} + static int simple_flow_mod(struct ofproto *ofproto, const struct match *match, unsigned int priority, @@ -1758,22 +1855,8 @@ simple_flow_mod(struct ofproto *ofproto, { struct ofputil_flow_mod fm; - memset(&fm, 0, sizeof fm); - fm.match = *match; - fm.priority = priority; - fm.cookie = 0; - fm.new_cookie = 0; - fm.modify_cookie = false; - fm.table_id = 0; - fm.command = command; - fm.idle_timeout = 0; - fm.hard_timeout = 0; - fm.buffer_id = UINT32_MAX; - fm.out_port = OFPP_ANY; - fm.out_group = OFPG_ANY; - fm.flags = 0; - fm.ofpacts = CONST_CAST(struct ofpact *, ofpacts); - fm.ofpacts_len = ofpacts_len; + flow_mod_init(&fm, match, priority, ofpacts, ofpacts_len, command); + return handle_flow_mod__(ofproto, NULL, &fm, NULL); } @@ -1792,6 +1875,7 @@ void ofproto_add_flow(struct ofproto *ofproto, const struct match *match, unsigned int priority, const struct ofpact *ofpacts, size_t ofpacts_len) + OVS_EXCLUDED(ofproto_mutex) { const struct rule *rule; bool must_add; @@ -1802,11 +1886,11 @@ ofproto_add_flow(struct ofproto *ofproto, const struct match *match, rule = rule_from_cls_rule(classifier_find_match_exactly( &ofproto->tables[0].cls, match, priority)); if (rule) { - ovs_rwlock_rdlock(&rule->rwlock); + ovs_mutex_lock(&rule->mutex); must_add = !ofpacts_equal(rule->actions->ofpacts, rule->actions->ofpacts_len, ofpacts, ofpacts_len); - ovs_rwlock_unlock(&rule->rwlock); + ovs_mutex_unlock(&rule->mutex); } else { must_add = true; } @@ -1826,9 +1910,11 @@ ofproto_add_flow(struct ofproto *ofproto, const struct match *match, * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the * operation cannot be initiated now but may be retried later. * - * This is a helper function for in-band control and fail-open. */ + * This is a helper function for in-band control and fail-open and the "learn" + * action. */ int ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm) + OVS_EXCLUDED(ofproto_mutex) { return handle_flow_mod__(ofproto, NULL, fm, NULL); } @@ -1840,6 +1926,7 @@ ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm) bool ofproto_delete_flow(struct ofproto *ofproto, const struct match *target, unsigned int priority) + OVS_EXCLUDED(ofproto_mutex) { struct classifier *cls = &ofproto->tables[0].cls; struct rule *rule; @@ -1905,35 +1992,56 @@ alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name) port_idx = port_idx ? port_idx : UINT16_MAX; if (port_idx >= ofproto->max_ports - || bitmap_is_set(ofproto->ofp_port_ids, port_idx)) { - uint16_t end_port_no = ofproto->alloc_port_no; + || ofport_get_usage(ofproto, u16_to_ofp(port_idx)) == LLONG_MAX) { + uint16_t lru_ofport = 0, end_port_no = ofproto->alloc_port_no; + long long int last_used_at, lru = LLONG_MAX; /* Search for a free OpenFlow port number. We try not to * immediately reuse them to prevent problems due to old - * flows. */ + * flows. + * + * We limit the automatically assigned port numbers to the lower half + * of the port range, to reserve the upper half for assignment by + * controllers. */ for (;;) { - if (++ofproto->alloc_port_no >= ofproto->max_ports) { - ofproto->alloc_port_no = 0; + if (++ofproto->alloc_port_no >= MIN(ofproto->max_ports, 32768)) { + ofproto->alloc_port_no = 1; } - if (!bitmap_is_set(ofproto->ofp_port_ids, - ofproto->alloc_port_no)) { + last_used_at = ofport_get_usage(ofproto, + u16_to_ofp(ofproto->alloc_port_no)); + if (!last_used_at) { + port_idx = ofproto->alloc_port_no; + break; + } else if ( last_used_at < time_msec() - 60*60*1000) { + /* If the port with ofport 'ofproto->alloc_port_no' was deleted + * more than an hour ago, consider it usable. */ + ofport_remove_usage(ofproto, + u16_to_ofp(ofproto->alloc_port_no)); port_idx = ofproto->alloc_port_no; break; + } else if (last_used_at < lru) { + lru = last_used_at; + lru_ofport = ofproto->alloc_port_no; } + if (ofproto->alloc_port_no == end_port_no) { + if (lru_ofport) { + port_idx = lru_ofport; + break; + } return OFPP_NONE; } } } - bitmap_set1(ofproto->ofp_port_ids, port_idx); + ofport_set_usage(ofproto, u16_to_ofp(port_idx), LLONG_MAX); return u16_to_ofp(port_idx); } static void -dealloc_ofp_port(const struct ofproto *ofproto, ofp_port_t ofp_port) +dealloc_ofp_port(struct ofproto *ofproto, ofp_port_t ofp_port) { if (ofp_to_u16(ofp_port) < ofproto->max_ports) { - bitmap_set0(ofproto->ofp_port_ids, ofp_to_u16(ofp_port)); + ofport_set_usage(ofproto, ofp_port, time_msec()); } } @@ -2018,7 +2126,6 @@ ofport_install(struct ofproto *p, } ofport->ofproto = p; ofport->netdev = netdev; - ofport->change_seq = netdev_change_seq(netdev); ofport->pp = *pp; ofport->ofp_port = pp->port_no; ofport->created = time_msec(); @@ -2158,6 +2265,55 @@ ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port) return NULL; } +static long long int +ofport_get_usage(const struct ofproto *ofproto, ofp_port_t ofp_port) +{ + struct ofport_usage *usage; + + HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port), + &ofproto->ofport_usage) { + if (usage->ofp_port == ofp_port) { + return usage->last_used; + } + } + return 0; +} + +static void +ofport_set_usage(struct ofproto *ofproto, ofp_port_t ofp_port, + long long int last_used) +{ + struct ofport_usage *usage; + HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port), + &ofproto->ofport_usage) { + if (usage->ofp_port == ofp_port) { + usage->last_used = last_used; + return; + } + } + ovs_assert(last_used == LLONG_MAX); + + usage = xmalloc(sizeof *usage); + usage->ofp_port = ofp_port; + usage->last_used = last_used; + hmap_insert(&ofproto->ofport_usage, &usage->hmap_node, + hash_ofp_port(ofp_port)); +} + +static void +ofport_remove_usage(struct ofproto *ofproto, ofp_port_t ofp_port) +{ + struct ofport_usage *usage; + HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port), + &ofproto->ofport_usage) { + if (usage->ofp_port == ofp_port) { + hmap_remove(&ofproto->ofport_usage, &usage->hmap_node); + free(usage); + break; + } + } +} + int ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats) { @@ -2204,7 +2360,6 @@ update_port(struct ofproto *ofproto, const char *name) * Don't close the old netdev yet in case port_modified has to * remove a retained reference to it.*/ port->netdev = netdev; - port->change_seq = netdev_change_seq(netdev); if (port->ofproto->ofproto_class->port_modified) { port->ofproto->ofproto_class->port_modified(port); @@ -2376,20 +2531,45 @@ ofproto_rule_unref(struct rule *rule) } } +struct rule_actions * +rule_get_actions(const struct rule *rule) + OVS_EXCLUDED(rule->mutex) +{ + struct rule_actions *actions; + + ovs_mutex_lock(&rule->mutex); + actions = rule_get_actions__(rule); + ovs_mutex_unlock(&rule->mutex); + + return actions; +} + +struct rule_actions * +rule_get_actions__(const struct rule *rule) + OVS_REQUIRES(rule->mutex) +{ + rule_actions_ref(rule->actions); + return rule->actions; +} + static void ofproto_rule_destroy__(struct rule *rule) + OVS_NO_THREAD_SAFETY_ANALYSIS { - cls_rule_destroy(&rule->cr); + cls_rule_destroy(CONST_CAST(struct cls_rule *, &rule->cr)); rule_actions_unref(rule->actions); - ovs_mutex_destroy(&rule->timeout_mutex); - ovs_rwlock_destroy(&rule->rwlock); + ovs_mutex_destroy(&rule->mutex); rule->ofproto->ofproto_class->rule_dealloc(rule); } +static uint32_t get_provider_meter_id(const struct ofproto *, + uint32_t of_meter_id); + /* Creates and returns a new 'struct rule_actions', with a ref_count of 1, * whose actions are a copy of from the 'ofpacts_len' bytes of 'ofpacts'. */ struct rule_actions * -rule_actions_create(const struct ofpact *ofpacts, size_t ofpacts_len) +rule_actions_create(const struct ofproto *ofproto, + const struct ofpact *ofpacts, size_t ofpacts_len) { struct rule_actions *actions; @@ -2397,7 +2577,10 @@ rule_actions_create(const struct ofpact *ofpacts, size_t ofpacts_len) atomic_init(&actions->ref_count, 1); actions->ofpacts = xmemdup(ofpacts, ofpacts_len); actions->ofpacts_len = ofpacts_len; - actions->meter_id = ofpacts_get_meter(ofpacts, ofpacts_len); + actions->provider_meter_id + = get_provider_meter_id(ofproto, + ofpacts_get_meter(ofpacts, ofpacts_len)); + return actions; } @@ -2423,6 +2606,7 @@ rule_actions_unref(struct rule_actions *actions) atomic_sub(&actions->ref_count, 1, &orig); if (orig == 1) { + free(actions->ofpacts); free(actions); } else { ovs_assert(orig != 0); @@ -2432,8 +2616,9 @@ rule_actions_unref(struct rule_actions *actions) /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */ -bool +static bool ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port) + OVS_REQUIRES(ofproto_mutex) { return (port == OFPP_ANY || ofpacts_output_to_port(rule->actions->ofpacts, @@ -2441,8 +2626,9 @@ ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port) } /* Returns true if 'rule' has group and equals group_id. */ -bool +static bool ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id) + OVS_REQUIRES(ofproto_mutex) { return (group_id == OFPG11_ANY || ofpacts_output_to_group(rule->actions->ofpacts, @@ -2453,6 +2639,7 @@ ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id) * OFPAT_ENQUEUE action that outputs to 'out_port'. */ bool ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port) + OVS_REQUIRES(ofproto_mutex) { if (ofproto_rule_has_out_port(op->rule, out_port)) { return true; @@ -2484,6 +2671,7 @@ rule_execute_destroy(struct rule_execute *e) * by passing them to the ofproto provider. */ static void run_rule_executes(struct ofproto *ofproto) + OVS_EXCLUDED(ofproto_mutex) { struct rule_execute *e, *next; struct list executes; @@ -2521,7 +2709,7 @@ destroy_rule_executes(struct ofproto *ofproto) * Rules with priority higher than UINT16_MAX are set up by ofproto itself * (e.g. by in-band control) and are intentionally hidden from the * controller. */ -bool +static bool ofproto_rule_is_hidden(const struct rule *rule) { return rule->cr.priority > UINT16_MAX; @@ -2664,28 +2852,33 @@ reject_slave_controller(struct ofconn *ofconn) } } -/* Checks that the 'ofpacts_len' bytes of actions in 'ofpacts' are appropriate - * for a packet with the prerequisites satisfied by 'flow' in table 'table_id'. - * 'flow' may be temporarily modified, but is restored at return. - */ +/* Checks that the 'ofpacts_len' bytes of action in 'ofpacts' are appropriate + * for 'ofproto': + * + * - If they use a meter, then 'ofproto' has that meter configured. + * + * - If they use any groups, then 'ofproto' has that group configured. + * + * Returns 0 if successful, otherwise an OpenFlow error. */ static enum ofperr ofproto_check_ofpacts(struct ofproto *ofproto, - const struct ofpact ofpacts[], size_t ofpacts_len, - struct flow *flow, uint8_t table_id) + const struct ofpact ofpacts[], size_t ofpacts_len) { - enum ofperr error; + const struct ofpact *a; uint32_t mid; - error = ofpacts_check(ofpacts, ofpacts_len, flow, - u16_to_ofp(ofproto->max_ports), table_id); - if (error) { - return error; - } - mid = ofpacts_get_meter(ofpacts, ofpacts_len); - if (mid && ofproto_get_provider_meter_id(ofproto, mid) == UINT32_MAX) { + if (mid && get_provider_meter_id(ofproto, mid) == UINT32_MAX) { return OFPERR_OFPMMFC_INVALID_METER; } + + OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) { + if (a->type == OFPACT_GROUP + && !ofproto_group_exists(ofproto, ofpact_get_GROUP(a)->group_id)) { + return OFPERR_OFPBAC_BAD_OUT_GROUP; + } + } + return 0; } @@ -2720,7 +2913,6 @@ handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh) goto exit_free_ofpacts; } - /* Get payload. */ if (po.buffer_id != UINT32_MAX) { error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL); @@ -2735,7 +2927,7 @@ handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh) /* Verify actions against packet, then send packet if successful. */ in_port_.ofp_port = po.in_port; flow_extract(payload, 0, 0, NULL, &in_port_, &flow); - error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len, &flow, 0); + error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len); if (!error) { error = p->ofproto_class->packet_out(p, payload, &flow, po.ofpacts, po.ofpacts_len); @@ -2855,15 +3047,15 @@ handle_table_stats_request(struct ofconn *ofconn, ots = xcalloc(p->n_tables, sizeof *ots); for (i = 0; i < p->n_tables; i++) { ots[i].table_id = i; - sprintf(ots[i].name, "table%zu", i); - ots[i].match = htonll(OFPXMT12_MASK); - ots[i].wildcards = htonll(OFPXMT12_MASK); + sprintf(ots[i].name, "table%"PRIuSIZE, i); + ots[i].match = htonll(OFPXMT13_MASK); + ots[i].wildcards = htonll(OFPXMT13_MASK); ots[i].write_actions = htonl(OFPAT11_OUTPUT); ots[i].apply_actions = htonl(OFPAT11_OUTPUT); - ots[i].write_setfields = htonll(OFPXMT12_MASK); - ots[i].apply_setfields = htonll(OFPXMT12_MASK); - ots[i].metadata_match = htonll(UINT64_MAX); - ots[i].metadata_write = htonll(UINT64_MAX); + ots[i].write_setfields = htonll(OFPXMT13_MASK); + ots[i].apply_setfields = htonll(OFPXMT13_MASK); + ots[i].metadata_match = OVS_BE64_MAX; + ots[i].metadata_write = OVS_BE64_MAX; ots[i].instructions = htonl(OFPIT11_ALL); ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK); ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */ @@ -2993,17 +3185,16 @@ cookies_remove(struct ofproto *ofproto, struct rule *rule) static void ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule, ovs_be64 new_cookie) + OVS_REQUIRES(ofproto_mutex) { if (new_cookie != rule->flow_cookie) { - ovs_mutex_lock(&ofproto_mutex); cookies_remove(ofproto, rule); - ovs_rwlock_wrlock(&rule->rwlock); + ovs_mutex_lock(&rule->mutex); rule->flow_cookie = new_cookie; - ovs_rwlock_unlock(&rule->rwlock); + ovs_mutex_unlock(&rule->mutex); cookies_insert(ofproto, rule); - ovs_mutex_unlock(&ofproto_mutex); } } @@ -3138,6 +3329,27 @@ rule_collection_add(struct rule_collection *rules, struct rule *rule) rules->rules[rules->n++] = rule; } +void +rule_collection_ref(struct rule_collection *rules) + OVS_REQUIRES(ofproto_mutex) +{ + size_t i; + + for (i = 0; i < rules->n; i++) { + ofproto_rule_ref(rules->rules[i]); + } +} + +void +rule_collection_unref(struct rule_collection *rules) +{ + size_t i; + + for (i = 0; i < rules->n; i++) { + ofproto_rule_unref(rules->rules[i]); + } +} + void rule_collection_destroy(struct rule_collection *rules) { @@ -3149,8 +3361,15 @@ rule_collection_destroy(struct rule_collection *rules) static enum ofperr collect_rule(struct rule *rule, const struct rule_criteria *c, struct rule_collection *rules) + OVS_REQUIRES(ofproto_mutex) { - if (ofproto_rule_is_hidden(rule)) { + /* We ordinarily want to skip hidden rules, but there has to be a way for + * code internal to OVS to modify and delete them, so if the criteria + * specify a priority that can only be for a hidden flow, then allow hidden + * rules to be selected. (This doesn't allow OpenFlow clients to meddle + * with hidden flows because OpenFlow uses only a 16-bit field to specify + * priority.) */ + if (ofproto_rule_is_hidden(rule) && c->cr.priority <= UINT16_MAX) { return 0; } else if (rule->pending) { return OFPROTO_POSTPONE; @@ -3177,6 +3396,7 @@ static enum ofperr collect_rules_loose(struct ofproto *ofproto, const struct rule_criteria *criteria, struct rule_collection *rules) + OVS_REQUIRES(ofproto_mutex) { struct oftable *table; enum ofperr error; @@ -3188,10 +3408,9 @@ collect_rules_loose(struct ofproto *ofproto, goto exit; } - if (criteria->cookie_mask == htonll(UINT64_MAX)) { + if (criteria->cookie_mask == OVS_BE64_MAX) { struct rule *rule; - ovs_mutex_lock(&ofproto_mutex); HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, hash_cookie(criteria->cookie), &ofproto->cookies) { @@ -3202,7 +3421,6 @@ collect_rules_loose(struct ofproto *ofproto, } } } - ovs_mutex_unlock(&ofproto_mutex); } else { FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) { struct cls_cursor cursor; @@ -3239,6 +3457,7 @@ static enum ofperr collect_rules_strict(struct ofproto *ofproto, const struct rule_criteria *criteria, struct rule_collection *rules) + OVS_REQUIRES(ofproto_mutex) { struct oftable *table; int error; @@ -3250,10 +3469,9 @@ collect_rules_strict(struct ofproto *ofproto, goto exit; } - if (criteria->cookie_mask == htonll(UINT64_MAX)) { + if (criteria->cookie_mask == OVS_BE64_MAX) { struct rule *rule; - ovs_mutex_lock(&ofproto_mutex); HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, hash_cookie(criteria->cookie), &ofproto->cookies) { @@ -3264,7 +3482,6 @@ collect_rules_strict(struct ofproto *ofproto, } } } - ovs_mutex_unlock(&ofproto_mutex); } else { FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) { struct rule *rule; @@ -3302,6 +3519,7 @@ age_secs(long long int age_ms) static enum ofperr handle_flow_stats_request(struct ofconn *ofconn, const struct ofp_header *request) + OVS_EXCLUDED(ofproto_mutex) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); struct ofputil_flow_stats_request fsr; @@ -3318,8 +3536,15 @@ handle_flow_stats_request(struct ofconn *ofconn, rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, fsr.cookie, fsr.cookie_mask, fsr.out_port, fsr.out_group); + + ovs_mutex_lock(&ofproto_mutex); error = collect_rules_loose(ofproto, &criteria, &rules); rule_criteria_destroy(&criteria); + if (!error) { + rule_collection_ref(&rules); + } + ovs_mutex_unlock(&ofproto_mutex); + if (error) { return error; } @@ -3329,28 +3554,39 @@ handle_flow_stats_request(struct ofconn *ofconn, struct rule *rule = rules.rules[i]; long long int now = time_msec(); struct ofputil_flow_stats fs; + long long int created, used, modified; + struct rule_actions *actions; + enum ofputil_flow_mod_flags flags; - minimatch_expand(&rule->cr.match, &fs.match); - fs.priority = rule->cr.priority; + ovs_mutex_lock(&rule->mutex); fs.cookie = rule->flow_cookie; - fs.table_id = rule->table_id; - calc_duration(rule->created, now, &fs.duration_sec, &fs.duration_nsec); - fs.idle_age = age_secs(now - rule->used); - fs.hard_age = age_secs(now - rule->modified); - ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count, - &fs.byte_count); - fs.ofpacts = rule->actions->ofpacts; - fs.ofpacts_len = rule->actions->ofpacts_len; - - ovs_mutex_lock(&rule->timeout_mutex); fs.idle_timeout = rule->idle_timeout; fs.hard_timeout = rule->hard_timeout; - ovs_mutex_unlock(&rule->timeout_mutex); + created = rule->created; + used = rule->used; + modified = rule->modified; + actions = rule_get_actions__(rule); + flags = rule->flags; + ovs_mutex_unlock(&rule->mutex); - fs.flags = rule->flags; + minimatch_expand(&rule->cr.match, &fs.match); + fs.table_id = rule->table_id; + calc_duration(created, now, &fs.duration_sec, &fs.duration_nsec); + fs.priority = rule->cr.priority; + fs.idle_age = age_secs(now - used); + fs.hard_age = age_secs(now - modified); + ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count, + &fs.byte_count); + fs.ofpacts = actions->ofpacts; + fs.ofpacts_len = actions->ofpacts_len; + fs.flags = flags; ofputil_append_flow_stats_reply(&fs, &replies); + + rule_actions_unref(actions); } + + rule_collection_unref(&rules); rule_collection_destroy(&rules); ofconn_send_replies(ofconn, &replies); @@ -3362,23 +3598,32 @@ static void flow_stats_ds(struct rule *rule, struct ds *results) { uint64_t packet_count, byte_count; + struct rule_actions *actions; + long long int created; rule->ofproto->ofproto_class->rule_get_stats(rule, &packet_count, &byte_count); + ovs_mutex_lock(&rule->mutex); + actions = rule_get_actions__(rule); + created = rule->created; + ovs_mutex_unlock(&rule->mutex); + if (rule->table_id != 0) { ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id); } - ds_put_format(results, "duration=%llds, ", - (time_msec() - rule->created) / 1000); - ds_put_format(results, "priority=%u, ", rule->cr.priority); + ds_put_format(results, "duration=%llds, ", (time_msec() - created) / 1000); ds_put_format(results, "n_packets=%"PRIu64", ", packet_count); ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count); cls_rule_format(&rule->cr, results); ds_put_char(results, ','); - ofpacts_format(rule->actions->ofpacts, rule->actions->ofpacts_len, - results); + + ds_put_cstr(results, "actions="); + ofpacts_format(actions->ofpacts, actions->ofpacts_len, results); + ds_put_cstr(results, "\n"); + + rule_actions_unref(actions); } /* Adds a pretty-printed description of all flows to 'results', including @@ -3429,6 +3674,7 @@ ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port, static enum ofperr handle_aggregate_stats_request(struct ofconn *ofconn, const struct ofp_header *oh) + OVS_EXCLUDED(ofproto_mutex) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); struct ofputil_flow_stats_request request; @@ -3448,8 +3694,15 @@ handle_aggregate_stats_request(struct ofconn *ofconn, rule_criteria_init(&criteria, request.table_id, &request.match, 0, request.cookie, request.cookie_mask, request.out_port, request.out_group); + + ovs_mutex_lock(&ofproto_mutex); error = collect_rules_loose(ofproto, &criteria, &rules); rule_criteria_destroy(&criteria); + if (!error) { + rule_collection_ref(&rules); + } + ovs_mutex_unlock(&ofproto_mutex); + if (error) { return error; } @@ -3485,6 +3738,7 @@ handle_aggregate_stats_request(struct ofconn *ofconn, stats.byte_count = UINT64_MAX; } + rule_collection_unref(&rules); rule_collection_destroy(&rules); reply = ofputil_encode_aggregate_stats_reply(&stats, oh); @@ -3595,6 +3849,7 @@ static bool is_flow_deletion_pending(const struct ofproto *ofproto, const struct cls_rule *cls_rule, uint8_t table_id) + OVS_REQUIRES(ofproto_mutex) { if (!hmap_is_empty(&ofproto->deletions)) { struct ofoperation *op; @@ -3611,32 +3866,34 @@ is_flow_deletion_pending(const struct ofproto *ofproto, return false; } -static enum ofperr -evict_rule_from_table(struct ofproto *ofproto, struct oftable *table) +static bool +should_evict_a_rule(struct oftable *table, unsigned int extra_space) + OVS_REQUIRES(ofproto_mutex) + OVS_NO_THREAD_SAFETY_ANALYSIS { - struct rule *rule; - size_t n_rules; - - ovs_rwlock_rdlock(&table->cls.rwlock); - n_rules = classifier_count(&table->cls); - ovs_rwlock_unlock(&table->cls.rwlock); - - if (n_rules < table->max_flows) { - return 0; - } else if (!choose_rule_to_evict(table, &rule)) { - return OFPERR_OFPFMFC_TABLE_FULL; - } else if (rule->pending) { - ovs_rwlock_unlock(&rule->rwlock); - return OFPROTO_POSTPONE; - } else { - struct ofopgroup *group; + return classifier_count(&table->cls) + extra_space > table->max_flows; +} - group = ofopgroup_create_unattached(ofproto); - delete_flow__(rule, group, OFPRR_EVICTION); - ofopgroup_submit(group); +static enum ofperr +evict_rules_from_table(struct ofproto *ofproto, struct oftable *table, + unsigned int extra_space) + OVS_REQUIRES(ofproto_mutex) +{ + while (should_evict_a_rule(table, extra_space)) { + struct rule *rule; - return 0; + if (!choose_rule_to_evict(table, &rule)) { + return OFPERR_OFPFMFC_TABLE_FULL; + } else if (rule->pending) { + return OFPROTO_POSTPONE; + } else { + struct ofopgroup *group = ofopgroup_create_unattached(ofproto); + delete_flow__(rule, group, OFPRR_EVICTION); + ofopgroup_submit(group); + } } + + return 0; } /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT @@ -3654,6 +3911,7 @@ evict_rule_from_table(struct ofproto *ofproto, struct oftable *table) static enum ofperr add_flow(struct ofproto *ofproto, struct ofconn *ofconn, struct ofputil_flow_mod *fm, const struct ofp_header *request) + OVS_REQUIRES(ofproto_mutex) { struct oftable *table; struct ofopgroup *group; @@ -3717,14 +3975,6 @@ add_flow(struct ofproto *ofproto, struct ofconn *ofconn, } } - /* Verify actions. */ - error = ofproto_check_ofpacts(ofproto, fm->ofpacts, fm->ofpacts_len, - &fm->match.flow, table_id); - if (error) { - cls_rule_destroy(&cr); - return error; - } - /* Serialize against pending deletion. */ if (is_flow_deletion_pending(ofproto, &cr, table_id)) { cls_rule_destroy(&cr); @@ -3746,7 +3996,7 @@ add_flow(struct ofproto *ofproto, struct ofconn *ofconn, } /* If necessary, evict an existing rule to clear out space. */ - error = evict_rule_from_table(ofproto, table); + error = evict_rules_from_table(ofproto, table, 1); if (error) { cls_rule_destroy(&cr); return error; @@ -3762,29 +4012,28 @@ add_flow(struct ofproto *ofproto, struct ofconn *ofconn, } /* Initialize base state. */ - rule->ofproto = ofproto; - cls_rule_move(&rule->cr, &cr); + *CONST_CAST(struct ofproto **, &rule->ofproto) = ofproto; + cls_rule_move(CONST_CAST(struct cls_rule *, &rule->cr), &cr); atomic_init(&rule->ref_count, 1); rule->pending = NULL; rule->flow_cookie = fm->new_cookie; rule->created = rule->modified = rule->used = time_msec(); - ovs_mutex_init(&rule->timeout_mutex); - ovs_mutex_lock(&rule->timeout_mutex); + ovs_mutex_init(&rule->mutex); + ovs_mutex_lock(&rule->mutex); rule->idle_timeout = fm->idle_timeout; rule->hard_timeout = fm->hard_timeout; - ovs_mutex_unlock(&rule->timeout_mutex); + ovs_mutex_unlock(&rule->mutex); - rule->table_id = table - ofproto->tables; + *CONST_CAST(uint8_t *, &rule->table_id) = table - ofproto->tables; rule->flags = fm->flags & OFPUTIL_FF_STATE; - rule->actions = rule_actions_create(fm->ofpacts, fm->ofpacts_len); + rule->actions = rule_actions_create(ofproto, fm->ofpacts, fm->ofpacts_len); list_init(&rule->meter_list_node); rule->eviction_group = NULL; list_init(&rule->expirable); rule->monitor_flags = 0; rule->add_seqno = 0; rule->modify_seqno = 0; - ovs_rwlock_init(&rule->rwlock); /* Construct rule, initializing derived state. */ error = ofproto->ofproto_class->rule_construct(rule); @@ -3817,6 +4066,7 @@ static enum ofperr modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn, struct ofputil_flow_mod *fm, const struct ofp_header *request, const struct rule_collection *rules) + OVS_REQUIRES(ofproto_mutex) { enum ofoperation_type type; struct ofopgroup *group; @@ -3841,27 +4091,20 @@ modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn, continue; } - /* Verify actions. */ - error = ofpacts_check(fm->ofpacts, fm->ofpacts_len, &fm->match.flow, - u16_to_ofp(ofproto->max_ports), rule->table_id); - if (error) { - return error; - } - actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len, rule->actions->ofpacts, rule->actions->ofpacts_len); op = ofoperation_create(group, rule, type, 0); - if (fm->modify_cookie && fm->new_cookie != htonll(UINT64_MAX)) { + if (fm->modify_cookie && fm->new_cookie != OVS_BE64_MAX) { ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie); } if (type == OFOPERATION_REPLACE) { - ovs_mutex_lock(&rule->timeout_mutex); + ovs_mutex_lock(&rule->mutex); rule->idle_timeout = fm->idle_timeout; rule->hard_timeout = fm->hard_timeout; - ovs_mutex_unlock(&rule->timeout_mutex); + ovs_mutex_unlock(&rule->mutex); rule->flags = fm->flags & OFPUTIL_FF_STATE; if (fm->idle_timeout || fm->hard_timeout) { @@ -3878,11 +4121,12 @@ modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn, struct rule_actions *new_actions; op->actions = rule->actions; - new_actions = rule_actions_create(fm->ofpacts, fm->ofpacts_len); + new_actions = rule_actions_create(ofproto, + fm->ofpacts, fm->ofpacts_len); - ovs_rwlock_wrlock(&rule->rwlock); + ovs_mutex_lock(&rule->mutex); rule->actions = new_actions; - ovs_rwlock_unlock(&rule->rwlock); + ovs_mutex_unlock(&rule->mutex); rule->ofproto->ofproto_class->rule_modify_actions(rule, reset_counters); @@ -3898,8 +4142,9 @@ modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn, static enum ofperr modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn, struct ofputil_flow_mod *fm, const struct ofp_header *request) + OVS_REQUIRES(ofproto_mutex) { - if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) { + if (fm->cookie_mask != htonll(0) || fm->new_cookie == OVS_BE64_MAX) { return 0; } return add_flow(ofproto, ofconn, fm, request); @@ -3914,6 +4159,7 @@ static enum ofperr modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn, struct ofputil_flow_mod *fm, const struct ofp_header *request) + OVS_REQUIRES(ofproto_mutex) { struct rule_criteria criteria; struct rule_collection rules; @@ -3944,6 +4190,7 @@ static enum ofperr modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn, struct ofputil_flow_mod *fm, const struct ofp_header *request) + OVS_REQUIRES(ofproto_mutex) { struct rule_criteria criteria; struct rule_collection rules; @@ -3972,6 +4219,7 @@ modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn, static void delete_flow__(struct rule *rule, struct ofopgroup *group, enum ofp_flow_removed_reason reason) + OVS_REQUIRES(ofproto_mutex) { struct ofproto *ofproto = rule->ofproto; @@ -3990,15 +4238,14 @@ delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn, const struct ofp_header *request, const struct rule_collection *rules, enum ofp_flow_removed_reason reason) + OVS_REQUIRES(ofproto_mutex) { struct ofopgroup *group; size_t i; group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX); for (i = 0; i < rules->n; i++) { - struct rule *rule = rules->rules[i]; - ovs_rwlock_wrlock(&rule->rwlock); - delete_flow__(rule, group, reason); + delete_flow__(rules->rules[i], group, reason); } ofopgroup_submit(group); @@ -4010,6 +4257,7 @@ static enum ofperr delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn, const struct ofputil_flow_mod *fm, const struct ofp_header *request) + OVS_REQUIRES(ofproto_mutex) { struct rule_criteria criteria; struct rule_collection rules; @@ -4034,6 +4282,7 @@ static enum ofperr delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn, const struct ofputil_flow_mod *fm, const struct ofp_header *request) + OVS_REQUIRES(ofproto_mutex) { struct rule_criteria criteria; struct rule_collection rules; @@ -4055,6 +4304,7 @@ delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn, static void ofproto_rule_send_removed(struct rule *rule, uint8_t reason) + OVS_REQUIRES(ofproto_mutex) { struct ofputil_flow_removed fr; @@ -4070,10 +4320,10 @@ ofproto_rule_send_removed(struct rule *rule, uint8_t reason) fr.table_id = rule->table_id; calc_duration(rule->created, time_msec(), &fr.duration_sec, &fr.duration_nsec); - ovs_mutex_lock(&rule->timeout_mutex); + ovs_mutex_lock(&rule->mutex); fr.idle_timeout = rule->idle_timeout; fr.hard_timeout = rule->hard_timeout; - ovs_mutex_unlock(&rule->timeout_mutex); + ovs_mutex_unlock(&rule->mutex); rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count, &fr.byte_count); @@ -4091,17 +4341,14 @@ ofproto_rule_send_removed(struct rule *rule, uint8_t reason) * OpenFlow flows. */ void ofproto_rule_expire(struct rule *rule, uint8_t reason) + OVS_REQUIRES(ofproto_mutex) { struct ofproto *ofproto = rule->ofproto; - struct classifier *cls = &ofproto->tables[rule->table_id].cls; ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT || reason == OFPRR_DELETE || reason == OFPRR_GROUP_DELETE); - ofproto_rule_send_removed(rule, reason); - ovs_rwlock_wrlock(&cls->rwlock); - ofproto_rule_delete(ofproto, cls, rule); - ovs_rwlock_unlock(&cls->rwlock); + ofproto_rule_delete__(ofproto, rule, reason); } /* Reduces '*timeout' to no more than 'max'. A value of zero in either case @@ -4122,7 +4369,7 @@ reduce_timeout(uint16_t max, uint16_t *timeout) void ofproto_rule_reduce_timeouts(struct rule *rule, uint16_t idle_timeout, uint16_t hard_timeout) - OVS_EXCLUDED(ofproto_mutex, rule->timeout_mutex) + OVS_EXCLUDED(ofproto_mutex, rule->mutex) { if (!idle_timeout && !hard_timeout) { return; @@ -4134,14 +4381,15 @@ ofproto_rule_reduce_timeouts(struct rule *rule, } ovs_mutex_unlock(&ofproto_mutex); - ovs_mutex_lock(&rule->timeout_mutex); + ovs_mutex_lock(&rule->mutex); reduce_timeout(idle_timeout, &rule->idle_timeout); reduce_timeout(hard_timeout, &rule->hard_timeout); - ovs_mutex_unlock(&rule->timeout_mutex); + ovs_mutex_unlock(&rule->mutex); } static enum ofperr handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh) + OVS_EXCLUDED(ofproto_mutex) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); struct ofputil_flow_mod fm; @@ -4157,7 +4405,12 @@ handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh) ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub); error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn), - &ofpacts); + &ofpacts, + u16_to_ofp(ofproto->max_ports), + ofproto->n_tables); + if (!error) { + error = ofproto_check_ofpacts(ofproto, fm.ofpacts, fm.ofpacts_len); + } if (!error) { error = handle_flow_mod__(ofproto, ofconn, &fm, oh); } @@ -4200,9 +4453,11 @@ exit: static enum ofperr handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn, struct ofputil_flow_mod *fm, const struct ofp_header *oh) + OVS_EXCLUDED(ofproto_mutex) { enum ofperr error; + ovs_mutex_lock(&ofproto_mutex); if (ofproto->n_pending < 50) { switch (fm->command) { case OFPFC_ADD: @@ -4238,6 +4493,7 @@ handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn, ovs_assert(!list_is_empty(&ofproto->pending)); error = OFPROTO_POSTPONE; } + ovs_mutex_unlock(&ofproto_mutex); run_rule_executes(ofproto); return error; @@ -4421,6 +4677,7 @@ static void ofproto_compose_flow_refresh_update(const struct rule *rule, enum nx_flow_monitor_flags flags, struct list *msgs) + OVS_REQUIRES(ofproto_mutex) { struct ofoperation *op = rule->pending; const struct rule_actions *actions; @@ -4436,10 +4693,10 @@ ofproto_compose_flow_refresh_update(const struct rule *rule, fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD) ? NXFME_ADDED : NXFME_MODIFIED); fu.reason = 0; - ovs_mutex_lock(&rule->timeout_mutex); + ovs_mutex_lock(&rule->mutex); fu.idle_timeout = rule->idle_timeout; fu.hard_timeout = rule->hard_timeout; - ovs_mutex_unlock(&rule->timeout_mutex); + ovs_mutex_unlock(&rule->mutex); fu.table_id = rule->table_id; fu.cookie = rule->flow_cookie; minimatch_expand(&rule->cr.match, &match); @@ -4482,6 +4739,7 @@ ofproto_compose_flow_refresh_update(const struct rule *rule, void ofmonitor_compose_refresh_updates(struct rule_collection *rules, struct list *msgs) + OVS_REQUIRES(ofproto_mutex) { size_t i; @@ -4498,6 +4756,7 @@ static void ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m, struct rule *rule, uint64_t seqno, struct rule_collection *rules) + OVS_REQUIRES(ofproto_mutex) { enum nx_flow_monitor_flags update; @@ -4537,6 +4796,7 @@ static void ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m, uint64_t seqno, struct rule_collection *rules) + OVS_REQUIRES(ofproto_mutex) { const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn); const struct ofoperation *op; @@ -4573,6 +4833,7 @@ ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m, static void ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m, struct rule_collection *rules) + OVS_REQUIRES(ofproto_mutex) { if (m->flags & NXFMF_INITIAL) { ofproto_collect_ofmonitor_refresh_rules(m, 0, rules); @@ -4582,12 +4843,14 @@ ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m, void ofmonitor_collect_resume_rules(struct ofmonitor *m, uint64_t seqno, struct rule_collection *rules) + OVS_REQUIRES(ofproto_mutex) { ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules); } static enum ofperr handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh) + OVS_EXCLUDED(ofproto_mutex) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); struct ofmonitor **monitors; @@ -4602,6 +4865,8 @@ handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh) ofpbuf_use_const(&b, oh, ntohs(oh->length)); monitors = NULL; n_monitors = allocated_monitors = 0; + + ovs_mutex_lock(&ofproto_mutex); for (;;) { struct ofputil_flow_monitor_request request; struct ofmonitor *m; @@ -4640,10 +4905,11 @@ handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh) ofpmp_init(&replies, oh); ofmonitor_compose_refresh_updates(&rules, &replies); + ovs_mutex_unlock(&ofproto_mutex); + rule_collection_destroy(&rules); ofconn_send_replies(ofconn, &replies); - free(monitors); return 0; @@ -4653,23 +4919,32 @@ error: ofmonitor_destroy(monitors[i]); } free(monitors); + ovs_mutex_unlock(&ofproto_mutex); + return error; } static enum ofperr handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh) + OVS_EXCLUDED(ofproto_mutex) { struct ofmonitor *m; + enum ofperr error; uint32_t id; id = ofputil_decode_flow_monitor_cancel(oh); + + ovs_mutex_lock(&ofproto_mutex); m = ofmonitor_lookup(ofconn, id); - if (!m) { - return OFPERR_NXBRC_FM_BAD_ID; + if (m) { + ofmonitor_destroy(m); + error = 0; + } else { + error = OFPERR_NXBRC_FM_BAD_ID; } + ovs_mutex_unlock(&ofproto_mutex); - ofmonitor_destroy(m); - return 0; + return error; } /* Meters implementation. @@ -4694,13 +4969,10 @@ struct meter { /* * This is used in instruction validation at flow set-up time, * as flows may not use non-existing meters. - * This is also used by ofproto-providers to translate OpenFlow meter_ids - * in METER instructions to the corresponding provider meter IDs. * Return value of UINT32_MAX signifies an invalid meter. */ -uint32_t -ofproto_get_provider_meter_id(const struct ofproto * ofproto, - uint32_t of_meter_id) +static uint32_t +get_provider_meter_id(const struct ofproto *ofproto, uint32_t of_meter_id) { if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) { const struct meter *meter = ofproto->meters[of_meter_id]; @@ -4740,6 +5012,7 @@ meter_create(const struct ofputil_meter_config *config, static void meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last) + OVS_REQUIRES(ofproto_mutex) { uint32_t mid; for (mid = first; mid <= last; ++mid) { @@ -4771,7 +5044,7 @@ handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm) ovs_assert(provider_meter_id.uint32 != UINT32_MAX); *meterp = meter_create(&mm->meter, provider_meter_id); } - return 0; + return error; } static enum ofperr @@ -4779,15 +5052,17 @@ handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm) { struct meter *meter = ofproto->meters[mm->meter.meter_id]; enum ofperr error; + uint32_t provider_meter_id; if (!meter) { return OFPERR_OFPMMFC_UNKNOWN_METER; } + provider_meter_id = meter->provider_meter_id.uint32; error = ofproto->ofproto_class->meter_set(ofproto, &meter->provider_meter_id, &mm->meter); - ovs_assert(meter->provider_meter_id.uint32 != UINT32_MAX); + ovs_assert(meter->provider_meter_id.uint32 == provider_meter_id); if (!error) { meter_update(meter, &mm->meter); } @@ -4797,6 +5072,7 @@ handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm) static enum ofperr handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh, struct ofputil_meter_mod *mm) + OVS_EXCLUDED(ofproto_mutex) { struct ofproto *ofproto = ofconn_get_ofproto(ofconn); uint32_t meter_id = mm->meter.meter_id; @@ -4817,6 +5093,7 @@ handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh, /* First delete the rules that use this meter. If any of those rules are * currently being modified, postpone the whole operation until later. */ rule_collection_init(&rules); + ovs_mutex_lock(&ofproto_mutex); for (meter_id = first; meter_id <= last; ++meter_id) { struct meter *meter = ofproto->meters[meter_id]; if (meter && !list_is_empty(&meter->rules)) { @@ -4839,6 +5116,7 @@ handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh, meter_delete(ofproto, first, last); exit: + ovs_mutex_unlock(&ofproto_mutex); rule_collection_destroy(&rules); return error; @@ -5036,7 +5314,7 @@ ofproto_group_write_lookup(const struct ofproto *ofproto, uint32_t group_id, } static bool -ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id) +ofproto_group_exists__(const struct ofproto *ofproto, uint32_t group_id) OVS_REQ_RDLOCK(ofproto->groups_rwlock) { struct ofgroup *grp; @@ -5050,6 +5328,44 @@ ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id) return false; } +static bool +ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id) + OVS_EXCLUDED(ofproto->groups_rwlock) +{ + bool exists; + + ovs_rwlock_rdlock(&ofproto->groups_rwlock); + exists = ofproto_group_exists__(ofproto, group_id); + ovs_rwlock_unlock(&ofproto->groups_rwlock); + + return exists; +} + +static uint32_t +group_get_ref_count(struct ofgroup *group) + OVS_EXCLUDED(ofproto_mutex) +{ + struct ofproto *ofproto = group->ofproto; + struct rule_criteria criteria; + struct rule_collection rules; + struct match match; + enum ofperr error; + uint32_t count; + + match_init_catchall(&match); + rule_criteria_init(&criteria, 0xff, &match, 0, htonll(0), htonll(0), + OFPP_ANY, group->group_id); + ovs_mutex_lock(&ofproto_mutex); + error = collect_rules_loose(ofproto, &criteria, &rules); + ovs_mutex_unlock(&ofproto_mutex); + rule_criteria_destroy(&criteria); + + count = !error && rules.n < UINT32_MAX ? rules.n : UINT32_MAX; + + rule_collection_destroy(&rules); + return count; +} + static void append_group_stats(struct ofgroup *group, struct list *replies) OVS_REQ_RDLOCK(group->rwlock) @@ -5061,14 +5377,16 @@ append_group_stats(struct ofgroup *group, struct list *replies) ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats); + /* Provider sets the packet and byte counts, we do the rest. */ + ogs.ref_count = group_get_ref_count(group); + ogs.n_buckets = group->n_buckets; + error = (ofproto->ofproto_class->group_get_stats ? ofproto->ofproto_class->group_get_stats(group, &ogs) : EOPNOTSUPP); if (error) { - ogs.ref_count = UINT32_MAX; ogs.packet_count = UINT64_MAX; ogs.byte_count = UINT64_MAX; - ogs.n_buckets = group->n_buckets; memset(ogs.bucket_stats, 0xff, ogs.n_buckets * sizeof *ogs.bucket_stats); } @@ -5157,6 +5475,49 @@ handle_group_features_stats_request(struct ofconn *ofconn, return 0; } +static enum ofperr +handle_queue_get_config_request(struct ofconn *ofconn, + const struct ofp_header *oh) +{ + struct ofproto *p = ofconn_get_ofproto(ofconn); + struct netdev_queue_dump queue_dump; + struct ofport *ofport; + unsigned int queue_id; + struct ofpbuf *reply; + struct smap details; + ofp_port_t request; + enum ofperr error; + + error = ofputil_decode_queue_get_config_request(oh, &request); + if (error) { + return error; + } + + ofport = ofproto_get_port(p, request); + if (!ofport) { + return OFPERR_OFPQOFC_BAD_PORT; + } + + reply = ofputil_encode_queue_get_config_reply(oh); + + smap_init(&details); + NETDEV_QUEUE_FOR_EACH (&queue_id, &details, &queue_dump, ofport->netdev) { + struct ofputil_queue_config queue; + + /* None of the existing queues have compatible properties, so we + * hard-code omitting min_rate and max_rate. */ + queue.queue_id = queue_id; + queue.min_rate = UINT16_MAX; + queue.max_rate = UINT16_MAX; + ofputil_append_queue_get_config_reply(reply, &queue); + } + smap_destroy(&details); + + ofconn_send_reply(ofconn, reply); + + return 0; +} + /* Implements OFPGC11_ADD * in which no matching flow already exists in the flow table. * @@ -5214,7 +5575,7 @@ add_group(struct ofproto *ofproto, struct ofputil_group_mod *gm) goto unlock_out; } - if (ofproto_group_exists(ofproto, gm->group_id)) { + if (ofproto_group_exists__(ofproto, gm->group_id)) { error = OFPERR_OFPGMFC_GROUP_EXISTS; goto unlock_out; } @@ -5309,6 +5670,15 @@ static void delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup) OVS_RELEASES(ofproto->groups_rwlock) { + struct match match; + struct ofputil_flow_mod fm; + + /* Delete all flow entries containing this group in a group action */ + match_init_catchall(&match); + flow_mod_init(&fm, &match, 0, NULL, 0, OFPFC_DELETE); + fm.out_group = ofgroup->group_id; + handle_flow_mod__(ofproto, NULL, &fm, NULL); + /* Must wait until existing readers are done, * while holding the container's write lock at the same time. */ ovs_rwlock_wrlock(&ofgroup->rwlock); @@ -5414,6 +5784,7 @@ handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh) static enum ofperr handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) + OVS_EXCLUDED(ofproto_mutex) { const struct ofp_header *oh = msg->data; enum ofptype type; @@ -5423,6 +5794,13 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) if (error) { return error; } + if (oh->version >= OFP13_VERSION && ofpmsg_is_stat_request(oh) + && ofpmp_more(oh)) { + /* We have no buffer implementation for multipart requests. + * Report overflow for requests which consists of multiple + * messages. */ + return OFPERR_OFPBRC_MULTIPART_BUFFER_OVERFLOW; + } switch (type) { /* OpenFlow requests. */ @@ -5533,10 +5911,8 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) case OFPTYPE_GROUP_FEATURES_STATS_REQUEST: return handle_group_features_stats_request(ofconn, oh); - /* FIXME: Change the following once they are implemented: */ case OFPTYPE_QUEUE_GET_CONFIG_REQUEST: - case OFPTYPE_TABLE_FEATURES_STATS_REQUEST: - return OFPERR_OFPBRC_BAD_TYPE; + return handle_queue_get_config_request(ofconn, oh); case OFPTYPE_HELLO: case OFPTYPE_ERROR: @@ -5565,14 +5941,21 @@ handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg) case OFPTYPE_METER_STATS_REPLY: case OFPTYPE_METER_CONFIG_STATS_REPLY: case OFPTYPE_METER_FEATURES_STATS_REPLY: + case OFPTYPE_TABLE_FEATURES_STATS_REQUEST: case OFPTYPE_TABLE_FEATURES_STATS_REPLY: + case OFPTYPE_ROLE_STATUS: default: - return OFPERR_OFPBRC_BAD_TYPE; + if (ofpmsg_is_stat_request(oh)) { + return OFPERR_OFPBRC_BAD_STAT; + } else { + return OFPERR_OFPBRC_BAD_TYPE; + } } } static bool handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg) + OVS_EXCLUDED(ofproto_mutex) { int error = handle_openflow__(ofconn, ofp_msg); if (error && error != OFPROTO_POSTPONE) { @@ -5591,6 +5974,7 @@ handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg) * ofoperation_create() and then submit it with ofopgroup_submit(). */ static struct ofopgroup * ofopgroup_create_unattached(struct ofproto *ofproto) + OVS_REQUIRES(ofproto_mutex) { struct ofopgroup *group = xzalloc(sizeof *group); group->ofproto = ofproto; @@ -5615,6 +5999,7 @@ ofopgroup_create_unattached(struct ofproto *ofproto) static struct ofopgroup * ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn, const struct ofp_header *request, uint32_t buffer_id) + OVS_REQUIRES(ofproto_mutex) { struct ofopgroup *group = ofopgroup_create_unattached(ofproto); if (ofconn) { @@ -5638,6 +6023,7 @@ ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn, * groups. */ static void ofopgroup_submit(struct ofopgroup *group) + OVS_REQUIRES(ofproto_mutex) { if (!group->n_running) { ofopgroup_complete(group); @@ -5649,6 +6035,7 @@ ofopgroup_submit(struct ofopgroup *group) static void ofopgroup_complete(struct ofopgroup *group) + OVS_REQUIRES(ofproto_mutex) { struct ofproto *ofproto = group->ofproto; @@ -5769,7 +6156,6 @@ ofopgroup_complete(struct ofopgroup *group) } } } else { - ovs_rwlock_wrlock(&rule->rwlock); oftable_remove_rule(rule); ofproto_rule_unref(rule); } @@ -5792,17 +6178,17 @@ ofopgroup_complete(struct ofopgroup *group) } } else { ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie); - ovs_mutex_lock(&rule->timeout_mutex); + ovs_mutex_lock(&rule->mutex); rule->idle_timeout = op->idle_timeout; rule->hard_timeout = op->hard_timeout; - ovs_mutex_unlock(&rule->timeout_mutex); + ovs_mutex_unlock(&rule->mutex); if (op->actions) { struct rule_actions *old_actions; - ovs_rwlock_wrlock(&rule->rwlock); + ovs_mutex_lock(&rule->mutex); old_actions = rule->actions; rule->actions = op->actions; - ovs_rwlock_unlock(&rule->rwlock); + ovs_mutex_unlock(&rule->mutex); op->actions = NULL; rule_actions_unref(old_actions); @@ -5848,6 +6234,7 @@ static struct ofoperation * ofoperation_create(struct ofopgroup *group, struct rule *rule, enum ofoperation_type type, enum ofp_flow_removed_reason reason) + OVS_REQUIRES(ofproto_mutex) { struct ofproto *ofproto = group->ofproto; struct ofoperation *op; @@ -5861,10 +6248,10 @@ ofoperation_create(struct ofopgroup *group, struct rule *rule, op->type = type; op->reason = reason; op->flow_cookie = rule->flow_cookie; - ovs_mutex_lock(&rule->timeout_mutex); + ovs_mutex_lock(&rule->mutex); op->idle_timeout = rule->idle_timeout; op->hard_timeout = rule->hard_timeout; - ovs_mutex_unlock(&rule->timeout_mutex); + ovs_mutex_unlock(&rule->mutex); op->flags = rule->flags; group->n_running++; @@ -5879,6 +6266,7 @@ ofoperation_create(struct ofopgroup *group, struct rule *rule, static void ofoperation_destroy(struct ofoperation *op) + OVS_REQUIRES(ofproto_mutex) { struct ofopgroup *group = op->group; @@ -5920,13 +6308,19 @@ ofoperation_complete(struct ofoperation *op, enum ofperr error) { struct ofopgroup *group = op->group; - ovs_assert(op->rule->pending == op); ovs_assert(group->n_running > 0); ovs_assert(!error || op->type != OFOPERATION_DELETE); op->error = error; if (!--group->n_running && !list_is_empty(&group->ofproto_node)) { + /* This function can be called from ->rule_construct(), in which case + * ofproto_mutex is held, or it can be called from ->run(), in which + * case ofproto_mutex is not held. But only in the latter case can we + * arrive here, so we can safely take ofproto_mutex now. */ + ovs_mutex_lock(&ofproto_mutex); + ovs_assert(op->rule->pending == op); ofopgroup_complete(group); + ovs_mutex_unlock(&ofproto_mutex); } } @@ -5967,6 +6361,7 @@ pick_fallback_dpid(void) * or with no timeouts are not evictable.) */ static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep) + OVS_REQUIRES(ofproto_mutex) { struct eviction_group *evg; @@ -5991,10 +6386,8 @@ choose_rule_to_evict(struct oftable *table, struct rule **rulep) struct rule *rule; HEAP_FOR_EACH (rule, evg_node, &evg->rules) { - if (!ovs_rwlock_trywrlock(&rule->rwlock)) { - *rulep = rule; - return true; - } + *rulep = rule; + return true; } } @@ -6010,39 +6403,13 @@ choose_rule_to_evict(struct oftable *table, struct rule **rulep) static void ofproto_evict(struct ofproto *ofproto) { - struct ofopgroup *group; struct oftable *table; - group = ofopgroup_create_unattached(ofproto); + ovs_mutex_lock(&ofproto_mutex); OFPROTO_FOR_EACH_TABLE (table, ofproto) { - while (table->eviction_fields) { - struct rule *rule; - size_t n_rules; - - ovs_rwlock_rdlock(&table->cls.rwlock); - n_rules = classifier_count(&table->cls); - ovs_rwlock_unlock(&table->cls.rwlock); - - if (n_rules <= table->max_flows) { - break; - } - - if (!choose_rule_to_evict(table, &rule)) { - break; - } - - if (rule->pending) { - ovs_rwlock_unlock(&rule->rwlock); - break; - } - - ofoperation_create(group, rule, - OFOPERATION_DELETE, OFPRR_EVICTION); - oftable_remove_rule(rule); - ofproto->ofproto_class->rule_delete(rule); - } + evict_rules_from_table(ofproto, table, 0); } - ofopgroup_submit(group); + ovs_mutex_unlock(&ofproto_mutex); } /* Eviction groups. */ @@ -6061,6 +6428,7 @@ eviction_group_priority(size_t n_rules) * adds or removes rules in 'evg'. */ static void eviction_group_resized(struct oftable *table, struct eviction_group *evg) + OVS_REQUIRES(ofproto_mutex) { heap_change(&table->eviction_groups_by_size, &evg->size_node, eviction_group_priority(heap_count(&evg->rules))); @@ -6076,6 +6444,7 @@ eviction_group_resized(struct oftable *table, struct eviction_group *evg) * - Frees 'evg'. */ static void eviction_group_destroy(struct oftable *table, struct eviction_group *evg) + OVS_REQUIRES(ofproto_mutex) { while (!heap_is_empty(&evg->rules)) { struct rule *rule; @@ -6092,6 +6461,7 @@ eviction_group_destroy(struct oftable *table, struct eviction_group *evg) /* Removes 'rule' from its eviction group, if any. */ static void eviction_group_remove_rule(struct rule *rule) + OVS_REQUIRES(ofproto_mutex) { if (rule->eviction_group) { struct oftable *table = &rule->ofproto->tables[rule->table_id]; @@ -6111,6 +6481,7 @@ eviction_group_remove_rule(struct rule *rule) * returns the hash value. */ static uint32_t eviction_group_hash_rule(struct rule *rule) + OVS_REQUIRES(ofproto_mutex) { struct oftable *table = &rule->ofproto->tables[rule->table_id]; const struct mf_subfield *sf; @@ -6148,6 +6519,7 @@ eviction_group_hash_rule(struct rule *rule) * if necessary. */ static struct eviction_group * eviction_group_find(struct oftable *table, uint32_t id) + OVS_REQUIRES(ofproto_mutex) { struct eviction_group *evg; @@ -6169,6 +6541,7 @@ eviction_group_find(struct oftable *table, uint32_t id) * for eviction. */ static uint32_t rule_eviction_priority(struct rule *rule) + OVS_REQUIRES(ofproto_mutex) { long long int hard_expiration; long long int idle_expiration; @@ -6176,7 +6549,7 @@ rule_eviction_priority(struct rule *rule) uint32_t expiration_offset; /* Calculate time of expiration. */ - ovs_mutex_lock(&rule->timeout_mutex); + ovs_mutex_lock(&rule->mutex); hard_expiration = (rule->hard_timeout ? rule->modified + rule->hard_timeout * 1000 : LLONG_MAX); @@ -6184,7 +6557,7 @@ rule_eviction_priority(struct rule *rule) ? rule->used + rule->idle_timeout * 1000 : LLONG_MAX); expiration = MIN(hard_expiration, idle_expiration); - ovs_mutex_unlock(&rule->timeout_mutex); + ovs_mutex_unlock(&rule->mutex); if (expiration == LLONG_MAX) { return 0; } @@ -6208,14 +6581,15 @@ rule_eviction_priority(struct rule *rule) * The caller must ensure that 'rule' is not already in an eviction group. */ static void eviction_group_add_rule(struct rule *rule) + OVS_REQUIRES(ofproto_mutex) { struct ofproto *ofproto = rule->ofproto; struct oftable *table = &ofproto->tables[rule->table_id]; bool has_timeout; - ovs_mutex_lock(&rule->timeout_mutex); + ovs_mutex_lock(&rule->mutex); has_timeout = rule->hard_timeout || rule->idle_timeout; - ovs_mutex_unlock(&rule->timeout_mutex); + ovs_mutex_unlock(&rule->mutex); if (table->eviction_fields && has_timeout) { struct eviction_group *evg; @@ -6236,7 +6610,7 @@ static void oftable_init(struct oftable *table) { memset(table, 0, sizeof *table); - classifier_init(&table->cls); + classifier_init(&table->cls, flow_segment_u32s); table->max_flows = UINT_MAX; } @@ -6280,6 +6654,7 @@ oftable_set_name(struct oftable *table, const char *name) * This function configures the former policy on 'table'. */ static void oftable_disable_eviction(struct oftable *table) + OVS_REQUIRES(ofproto_mutex) { if (table->eviction_fields) { struct eviction_group *evg, *next; @@ -6306,6 +6681,7 @@ oftable_disable_eviction(struct oftable *table) static void oftable_enable_eviction(struct oftable *table, const struct mf_subfield *fields, size_t n_fields) + OVS_REQUIRES(ofproto_mutex) { struct cls_cursor cursor; struct rule *rule; @@ -6338,69 +6714,62 @@ oftable_enable_eviction(struct oftable *table, /* Removes 'rule' from the oftable that contains it. */ static void -oftable_remove_rule__(struct ofproto *ofproto, struct classifier *cls, - struct rule *rule) - OVS_REQ_WRLOCK(cls->rwlock) OVS_RELEASES(rule->rwlock) +oftable_remove_rule__(struct ofproto *ofproto, struct rule *rule) + OVS_REQUIRES(ofproto_mutex) { - classifier_remove(cls, &rule->cr); + struct classifier *cls = &ofproto->tables[rule->table_id].cls; + + ovs_rwlock_wrlock(&cls->rwlock); + classifier_remove(cls, CONST_CAST(struct cls_rule *, &rule->cr)); + ovs_rwlock_unlock(&cls->rwlock); - ovs_mutex_lock(&ofproto_mutex); cookies_remove(ofproto, rule); - ovs_mutex_unlock(&ofproto_mutex); eviction_group_remove_rule(rule); - ovs_mutex_lock(&ofproto_mutex); if (!list_is_empty(&rule->expirable)) { list_remove(&rule->expirable); } - ovs_mutex_unlock(&ofproto_mutex); if (!list_is_empty(&rule->meter_list_node)) { list_remove(&rule->meter_list_node); list_init(&rule->meter_list_node); } - ovs_rwlock_unlock(&rule->rwlock); } static void oftable_remove_rule(struct rule *rule) + OVS_REQUIRES(ofproto_mutex) { - struct ofproto *ofproto = rule->ofproto; - struct oftable *table = &ofproto->tables[rule->table_id]; - - ovs_rwlock_wrlock(&table->cls.rwlock); - oftable_remove_rule__(ofproto, &table->cls, rule); - ovs_rwlock_unlock(&table->cls.rwlock); + oftable_remove_rule__(rule->ofproto, rule); } /* Inserts 'rule' into its oftable, which must not already contain any rule for * the same cls_rule. */ static void oftable_insert_rule(struct rule *rule) + OVS_REQUIRES(ofproto_mutex) { struct ofproto *ofproto = rule->ofproto; struct oftable *table = &ofproto->tables[rule->table_id]; bool may_expire; - ovs_mutex_lock(&rule->timeout_mutex); + ovs_mutex_lock(&rule->mutex); may_expire = rule->hard_timeout || rule->idle_timeout; - ovs_mutex_unlock(&rule->timeout_mutex); + ovs_mutex_unlock(&rule->mutex); if (may_expire) { - ovs_mutex_lock(&ofproto_mutex); list_insert(&ofproto->expirable, &rule->expirable); - ovs_mutex_unlock(&ofproto_mutex); } - ovs_mutex_lock(&ofproto_mutex); cookies_insert(ofproto, rule); - ovs_mutex_unlock(&ofproto_mutex); - if (rule->actions->meter_id) { - struct meter *meter = ofproto->meters[rule->actions->meter_id]; + if (rule->actions->provider_meter_id != UINT32_MAX) { + uint32_t meter_id = ofpacts_get_meter(rule->actions->ofpacts, + rule->actions->ofpacts_len); + struct meter *meter = ofproto->meters[meter_id]; list_insert(&meter->rules, &rule->meter_list_node); } ovs_rwlock_wrlock(&table->cls.rwlock); - classifier_insert(&table->cls, &rule->cr); + classifier_insert(&table->cls, CONST_CAST(struct cls_rule *, &rule->cr)); ovs_rwlock_unlock(&table->cls.rwlock); eviction_group_add_rule(rule); } @@ -6468,9 +6837,10 @@ ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap) ofproto->vlans_changed = false; OFPROTO_FOR_EACH_TABLE (oftable, ofproto) { - const struct cls_table *table; + const struct cls_subtable *table; - HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) { + ovs_rwlock_rdlock(&oftable->cls.rwlock); + HMAP_FOR_EACH (table, hmap_node, &oftable->cls.subtables) { if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) { const struct cls_rule *rule; @@ -6481,6 +6851,7 @@ ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap) } } } + ovs_rwlock_unlock(&oftable->cls.rwlock); } }