X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=ofproto%2Fofproto-dpif.c;h=6af49747fa851f14958e18a32e02d3ee4b224fa9;hb=a6b7506dab305d91fc5f2ac6416a714e5fa09dd4;hp=8bbc5953f091de7ceaca6a0717f4652f1c7483d7;hpb=0b4f207828c9354761efc60bd9ab90898d9cc623;p=sliver-openvswitch.git diff --git a/ofproto/ofproto-dpif.c b/ofproto/ofproto-dpif.c index 8bbc5953f..6af49747f 100644 --- a/ofproto/ofproto-dpif.c +++ b/ofproto/ofproto-dpif.c @@ -25,12 +25,14 @@ #include "bond.h" #include "bundle.h" #include "byte-order.h" +#include "connectivity.h" #include "connmgr.h" #include "coverage.h" #include "cfm.h" #include "dpif.h" #include "dynamic-string.h" #include "fail-open.h" +#include "guarded-list.h" #include "hmapx.h" #include "lacp.h" #include "learn.h" @@ -51,9 +53,12 @@ #include "ofproto-dpif-governor.h" #include "ofproto-dpif-ipfix.h" #include "ofproto-dpif-mirror.h" +#include "ofproto-dpif-monitor.h" #include "ofproto-dpif-sflow.h" +#include "ofproto-dpif-upcall.h" #include "ofproto-dpif-xlate.h" #include "poll-loop.h" +#include "seq.h" #include "simap.h" #include "smap.h" #include "timer.h" @@ -66,13 +71,14 @@ VLOG_DEFINE_THIS_MODULE(ofproto_dpif); COVERAGE_DEFINE(ofproto_dpif_expired); -COVERAGE_DEFINE(facet_changed_rule); COVERAGE_DEFINE(facet_revalidate); COVERAGE_DEFINE(facet_unexpected); -COVERAGE_DEFINE(facet_suppress); +COVERAGE_DEFINE(facet_create); +COVERAGE_DEFINE(facet_remove); +COVERAGE_DEFINE(subfacet_create); +COVERAGE_DEFINE(subfacet_destroy); COVERAGE_DEFINE(subfacet_install_fail); COVERAGE_DEFINE(packet_in_overflow); -COVERAGE_DEFINE(flow_mod_overflow); /* Number of implemented OpenFlow tables. */ enum { N_TABLES = 255 }; @@ -82,11 +88,51 @@ BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255); struct flow_miss; struct facet; -static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *, - const struct flow *, - struct flow_wildcards *wc); +struct rule_dpif { + struct rule up; + + /* These statistics: + * + * - Do include packets and bytes from facets that have been deleted or + * whose own statistics have been folded into the rule. + * + * - Do include packets and bytes sent "by hand" that were accounted to + * the rule without any facet being involved (this is a rare corner + * case in rule_execute()). + * + * - Do not include packet or bytes that can be obtained from any facet's + * packet_count or byte_count member or that can be obtained from the + * datapath by, e.g., dpif_flow_get() for any subfacet. + */ + struct ovs_mutex stats_mutex; + uint64_t packet_count OVS_GUARDED; /* Number of packets received. */ + uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */ +}; static void rule_get_stats(struct rule *, uint64_t *packets, uint64_t *bytes); +static struct rule_dpif *rule_dpif_cast(const struct rule *); + +struct group_dpif { + struct ofgroup up; + + /* These statistics: + * + * - Do include packets and bytes from facets that have been deleted or + * whose own statistics have been folded into the rule. + * + * - Do include packets and bytes sent "by hand" that were accounted to + * the rule without any facet being involved (this is a rare corner + * case in rule_execute()). + * + * - Do not include packet or bytes that can be obtained from any facet's + * packet_count or byte_count member or that can be obtained from the + * datapath by, e.g., dpif_flow_get() for any subfacet. + */ + struct ovs_mutex stats_mutex; + uint64_t packet_count OVS_GUARDED; /* Number of packets received. */ + uint64_t byte_count OVS_GUARDED; /* Number of bytes received. */ + struct bucket_counter *bucket_stats OVS_GUARDED; /* Bucket statistics. */ +}; struct ofbundle { struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ @@ -155,7 +201,6 @@ struct subfacet { struct facet *facet; /* Owning facet. */ struct dpif_backer *backer; /* Owning backer. */ - enum odp_key_fitness key_fitness; struct nlattr *key; int key_len; @@ -170,8 +215,8 @@ struct subfacet { #define SUBFACET_DESTROY_MAX_BATCH 50 -static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss, - long long int now); +static struct subfacet *subfacet_create(struct facet *, struct flow_miss *, + uint32_t key_hash); static struct subfacet *subfacet_find(struct dpif_backer *, const struct nlattr *key, size_t key_len, uint32_t key_hash); @@ -209,8 +254,7 @@ static void subfacet_uninstall(struct subfacet *); * Flow expiration works in terms of subfacets, so a facet must have at * least one subfacet or it will never expire, leaking memory. */ struct facet { - /* Owners. */ - struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ + /* Owner. */ struct ofproto_dpif *ofproto; /* Owned data. */ @@ -242,12 +286,9 @@ struct facet { long long int prev_used; /* Used time from last stats push. */ /* Accounting. */ - uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ - struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ - uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */ + uint16_t tcp_flags; /* TCP flags seen for this 'rule'. */ struct xlate_out xout; - bool fail_open; /* Facet matched the fail open rule. */ /* Storage for a single subfacet, to reduce malloc() time and space * overhead. (A facet always has at least one subfacet and in the common @@ -259,9 +300,7 @@ struct facet { long long int learn_rl; /* Rate limiter for facet_learn(). */ }; -static struct facet *facet_create(const struct flow_miss *, struct rule_dpif *, - struct xlate_out *, - struct dpif_flow_stats *); +static struct facet *facet_create(const struct flow_miss *); static void facet_remove(struct facet *); static void facet_free(struct facet *); @@ -274,9 +313,10 @@ static bool facet_check_consistency(struct facet *); static void facet_flush_stats(struct facet *); static void facet_reset_counters(struct facet *); +static void flow_push_stats(struct ofproto_dpif *, struct flow *, + struct dpif_flow_stats *, bool may_learn); static void facet_push_stats(struct facet *, bool may_learn); static void facet_learn(struct facet *); -static void facet_account(struct facet *); static void push_all_stats(void); static bool facet_is_controller_flow(struct facet *); @@ -292,6 +332,7 @@ struct ofport_dpif { struct bfd *bfd; /* BFD, if any. */ bool may_enable; /* May be enabled in bonds. */ bool is_tunnel; /* This port is a tunnel. */ + bool is_layer3; /* This is a layer 3 port. */ long long int carrier_seq; /* Carrier status changes. */ struct ofport_dpif *peer; /* Peer if patch port. */ @@ -344,13 +385,9 @@ ofport_dpif_cast(const struct ofport *ofport) } static void port_run(struct ofport_dpif *); -static void port_run_fast(struct ofport_dpif *); -static void port_wait(struct ofport_dpif *); static int set_bfd(struct ofport *, const struct smap *); static int set_cfm(struct ofport *, const struct cfm_settings *); static void ofport_update_peer(struct ofport_dpif *); -static void run_fast_rl(void); -static int run_fast(struct ofproto *); struct dpif_completion { struct list list_node; @@ -382,29 +419,16 @@ COVERAGE_DEFINE(rev_flow_table); COVERAGE_DEFINE(rev_mac_learning); COVERAGE_DEFINE(rev_inconsistency); -/* Drop keys are odp flow keys which have drop flows installed in the kernel. - * These are datapath flows which have no associated ofproto, if they did we - * would use facets. */ -struct drop_key { - struct hmap_node hmap_node; - struct nlattr *key; - size_t key_len; -}; - -struct avg_subfacet_rates { - double add_rate; /* Moving average of new flows created per minute. */ - double del_rate; /* Moving average of flows deleted per minute. */ -}; - /* All datapaths of a given type share a single dpif backer instance. */ struct dpif_backer { char *type; int refcount; struct dpif *dpif; + struct udpif *udpif; struct timer next_expiration; struct ovs_rwlock odp_to_ofport_lock; - struct hmap odp_to_ofport_map OVS_GUARDED; /* ODP port to ofport map. */ + struct hmap odp_to_ofport_map OVS_GUARDED; /* Contains "struct ofport"s. */ struct simap tnl_backers; /* Set of dpif ports backing tunnels. */ @@ -424,31 +448,18 @@ struct dpif_backer { * exposed via "ovs-appctl dpif/show". The goal is to learn about * traffic patterns in ways that we can use later to improve Open vSwitch * performance in new situations. */ - long long int created; /* Time when it is created. */ unsigned max_n_subfacet; /* Maximum number of flows */ unsigned avg_n_subfacet; /* Average number of flows. */ long long int avg_subfacet_life; /* Average life span of subfacets. */ - /* The average number of subfacets... */ - struct avg_subfacet_rates hourly; /* ...over the last hour. */ - struct avg_subfacet_rates daily; /* ...over the last day. */ - struct avg_subfacet_rates lifetime; /* ...over the switch lifetime. */ - long long int last_minute; /* Last time 'hourly' was updated. */ - - /* Number of subfacets added or deleted since 'last_minute'. */ - unsigned subfacet_add_count; - unsigned subfacet_del_count; - - /* Number of subfacets added or deleted from 'created' to 'last_minute.' */ - unsigned long long int total_subfacet_add_count; - unsigned long long int total_subfacet_del_count; + /* Number of upcall handling threads. */ + unsigned int n_handler_threads; }; /* All existing ofproto_backer instances, indexed by ofproto->up.type. */ static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers); static void drop_key_clear(struct dpif_backer *); -static void update_moving_averages(struct dpif_backer *backer); struct ofproto_dpif { struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ @@ -467,17 +478,16 @@ struct ofproto_dpif { struct hmap bundles; /* Contains "struct ofbundle"s. */ struct mac_learning *ml; bool has_bonded_bundles; + bool lacp_enabled; struct mbridge *mbridge; /* Facets. */ struct classifier facets; /* Contains 'struct facet's. */ long long int consistency_rl; - /* Support for debugging async flow mods. */ - struct list completions; - - struct netdev_stats stats; /* To account packets generated and consumed in - * userspace. */ + struct ovs_mutex stats_mutex; + struct netdev_stats stats OVS_GUARDED; /* To account packets generated and + * consumed in userspace. */ /* Spanning tree. */ struct stp *stp; @@ -493,25 +503,16 @@ struct ofproto_dpif { struct sset ghost_ports; /* Ports with no datapath port. */ struct sset port_poll_set; /* Queued names for port_poll() reply. */ int port_poll_errno; /* Last errno for port_poll() reply. */ + uint64_t change_seq; /* Connectivity status changes. */ /* Per ofproto's dpif stats. */ uint64_t n_hit; uint64_t n_missed; /* Work queues. */ - struct ovs_mutex flow_mod_mutex; - struct list flow_mods OVS_GUARDED; - size_t n_flow_mods OVS_GUARDED; - - struct ovs_mutex pin_mutex; - struct list pins OVS_GUARDED; - size_t n_pins OVS_GUARDED; + struct guarded_list pins; /* Contains "struct ofputil_packet_in"s. */ }; -/* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only - * for debugging the asynchronous flow_mod implementation.) */ -static bool clogged; - /* By default, flows in the datapath are wildcarded (megaflows). They * may be disabled with the "ovs-appctl dpif/disable-megaflows" command. */ static bool enable_megaflows = true; @@ -531,62 +532,42 @@ ofproto_dpif_cast(const struct ofproto *ofproto) static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port); static void ofproto_trace(struct ofproto_dpif *, const struct flow *, - const struct ofpbuf *packet, struct ds *); + const struct ofpbuf *packet, + const struct ofpact[], size_t ofpacts_len, + struct ds *); /* Upcalls. */ -#define FLOW_MISS_MAX_BATCH 50 -static int handle_upcalls(struct dpif_backer *, unsigned int max_batch); +static void handle_upcalls(struct dpif_backer *); /* Flow expiration. */ static int expire(struct dpif_backer *); -/* NetFlow. */ -static void send_netflow_active_timeouts(struct ofproto_dpif *); - -/* Utilities. */ -static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet); - /* Global variables. */ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); /* Initial mappings of port to bridge mappings. */ static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports); -/* Executes and takes ownership of 'fm'. */ +/* Executes 'fm'. The caller retains ownership of 'fm' and everything in + * it. */ void ofproto_dpif_flow_mod(struct ofproto_dpif *ofproto, struct ofputil_flow_mod *fm) { - ovs_mutex_lock(&ofproto->flow_mod_mutex); - if (ofproto->n_flow_mods > 1024) { - ovs_mutex_unlock(&ofproto->flow_mod_mutex); - COVERAGE_INC(flow_mod_overflow); - free(fm->ofpacts); - free(fm); - return; - } - - list_push_back(&ofproto->flow_mods, &fm->list_node); - ofproto->n_flow_mods++; - ovs_mutex_unlock(&ofproto->flow_mod_mutex); + ofproto_flow_mod(&ofproto->up, fm); } +/* Appends 'pin' to the queue of "packet ins" to be sent to the controller. + * Takes ownership of 'pin' and pin->packet. */ void ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto, - struct ofputil_packet_in *pin) + struct ofproto_packet_in *pin) { - ovs_mutex_lock(&ofproto->pin_mutex); - if (ofproto->n_pins > 1024) { - ovs_mutex_unlock(&ofproto->pin_mutex); + if (!guarded_list_push_back(&ofproto->pins, &pin->list_node, 1024)) { COVERAGE_INC(packet_in_overflow); - free(CONST_CAST(void *, pin->packet)); + free(CONST_CAST(void *, pin->up.packet)); free(pin); - return; } - - list_push_back(&ofproto->pins, &pin->list_node); - ofproto->n_pins++; - ovs_mutex_unlock(&ofproto->pin_mutex); } /* Factory functions. */ @@ -688,6 +669,8 @@ type_run(const char *type) dpif_run(backer->dpif); + handle_upcalls(backer); + /* The most natural place to push facet statistics is when they're pulled * from the datapath. However, when there are many flows in the datapath, * this expensive operation can occur so frequently, that it reduces our @@ -708,13 +691,24 @@ type_run(const char *type) error = dpif_recv_set(backer->dpif, backer->recv_set_enable); if (error) { + udpif_recv_set(backer->udpif, 0, false); VLOG_ERR("Failed to enable receiving packets in dpif."); return error; } + udpif_recv_set(backer->udpif, n_handler_threads, + backer->recv_set_enable); dpif_flow_flush(backer->dpif); backer->need_revalidate = REV_RECONFIGURE; } + /* If the n_handler_threads is reconfigured, call udpif_recv_set() + * to reset the handler threads. */ + if (backer->n_handler_threads != n_handler_threads) { + udpif_recv_set(backer->udpif, n_handler_threads, + backer->recv_set_enable); + backer->n_handler_threads = n_handler_threads; + } + if (backer->need_revalidate) { struct ofproto_dpif *ofproto; struct simap_node *node; @@ -804,10 +798,9 @@ type_run(const char *type) ofproto->no_packet_in_rule, ofproto->ml, ofproto->stp, ofproto->mbridge, ofproto->sflow, ofproto->ipfix, - ofproto->up.frag_handling, + ofproto->netflow, ofproto->up.frag_handling, ofproto->up.forward_bpdu, - connmgr_has_in_band(ofproto->up.connmgr), - ofproto->netflow != NULL); + connmgr_has_in_band(ofproto->up.connmgr)); HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { xlate_bundle_set(ofproto, bundle, bundle->name, @@ -820,14 +813,14 @@ type_run(const char *type) HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { int stp_port = ofport->stp_port ? stp_port_no(ofport->stp_port) - : 0; + : -1; xlate_ofport_set(ofproto, ofport->bundle, ofport, ofport->up.ofp_port, ofport->odp_port, ofport->up.netdev, ofport->cfm, ofport->bfd, ofport->peer, stp_port, ofport->qdscp, ofport->n_qdscp, - ofport->up.pp.config, ofport->is_tunnel, - ofport->may_enable); + ofport->up.pp.config, ofport->up.pp.state, + ofport->is_tunnel, ofport->may_enable); } ovs_rwlock_unlock(&xlate_rwlock); @@ -838,9 +831,10 @@ type_run(const char *type) ovs_rwlock_unlock(&ofproto->facets.rwlock); CLS_CURSOR_FOR_EACH_SAFE (facet, next, cr, &cursor) { facet_revalidate(facet); - run_fast_rl(); } } + + udpif_revalidate(backer->udpif); } if (!backer->recv_set_enable) { @@ -1003,85 +997,6 @@ process_dpif_port_error(struct dpif_backer *backer, int error) } } -static int -dpif_backer_run_fast(struct dpif_backer *backer, int max_batch) -{ - unsigned int work; - - /* If recv_set_enable is false, we should not handle upcalls. */ - if (!backer->recv_set_enable) { - return 0; - } - - /* Handle one or more batches of upcalls, until there's nothing left to do - * or until we do a fixed total amount of work. - * - * We do work in batches because it can be much cheaper to set up a number - * of flows and fire off their patches all at once. We do multiple batches - * because in some cases handling a packet can cause another packet to be - * queued almost immediately as part of the return flow. Both - * optimizations can make major improvements on some benchmarks and - * presumably for real traffic as well. */ - work = 0; - while (work < max_batch) { - int retval = handle_upcalls(backer, max_batch - work); - if (retval <= 0) { - return -retval; - } - work += retval; - } - - return 0; -} - -static int -type_run_fast(const char *type) -{ - struct dpif_backer *backer; - - backer = shash_find_data(&all_dpif_backers, type); - if (!backer) { - /* This is not necessarily a problem, since backers are only - * created on demand. */ - return 0; - } - - return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH); -} - -static void -run_fast_rl(void) -{ - static long long int port_rl = LLONG_MIN; - static unsigned int backer_rl = 0; - - if (time_msec() >= port_rl) { - struct ofproto_dpif *ofproto; - - HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { - run_fast(&ofproto->up); - } - port_rl = time_msec() + 200; - } - - /* XXX: We have to be careful not to do too much work in this function. If - * we call dpif_backer_run_fast() too often, or with too large a batch, - * performance improves signifcantly, but at a cost. It's possible for the - * number of flows in the datapath to increase without bound, and for poll - * loops to take 10s of seconds. The correct solution to this problem, - * long term, is to separate flow miss handling into it's own thread so it - * isn't affected by revalidations, and expirations. Until then, this is - * the best we can do. */ - if (++backer_rl >= 10) { - struct shash_node *node; - - backer_rl = 0; - SHASH_FOR_EACH (node, &all_dpif_backers) { - dpif_backer_run_fast(node->data, 1); - } - } -} - static void type_wait(const char *type) { @@ -1099,6 +1014,8 @@ type_wait(const char *type) } timer_wait(&backer->next_expiration); + dpif_wait(backer->dpif); + udpif_wait(backer->udpif); } /* Basic life-cycle. */ @@ -1122,8 +1039,6 @@ dealloc(struct ofproto *ofproto_) static void close_dpif_backer(struct dpif_backer *backer) { - struct shash_node *node; - ovs_assert(backer->refcount > 0); if (--backer->refcount) { @@ -1133,12 +1048,13 @@ close_dpif_backer(struct dpif_backer *backer) drop_key_clear(backer); hmap_destroy(&backer->drop_keys); + udpif_destroy(backer->udpif); + simap_destroy(&backer->tnl_backers); ovs_rwlock_destroy(&backer->odp_to_ofport_lock); hmap_destroy(&backer->odp_to_ofport_map); - node = shash_find(&all_dpif_backers, backer->type); + shash_find_and_delete(&all_dpif_backers, backer->type); free(backer->type); - shash_delete(&all_dpif_backers, node); dpif_close(backer->dpif); ovs_assert(hmap_is_empty(&backer->subfacets)); @@ -1208,6 +1124,7 @@ open_dpif_backer(const char *type, struct dpif_backer **backerp) free(backer); return error; } + backer->udpif = udpif_create(backer, backer->dpif); backer->type = xstrdup(type); backer->governor = NULL; @@ -1255,17 +1172,11 @@ open_dpif_backer(const char *type, struct dpif_backer **backerp) close_dpif_backer(backer); return error; } + udpif_recv_set(backer->udpif, n_handler_threads, + backer->recv_set_enable); + backer->n_handler_threads = n_handler_threads; backer->max_n_subfacet = 0; - backer->created = time_msec(); - backer->last_minute = backer->created; - memset(&backer->hourly, 0, sizeof backer->hourly); - memset(&backer->daily, 0, sizeof backer->daily); - memset(&backer->lifetime, 0, sizeof backer->lifetime); - backer->subfacet_add_count = 0; - backer->subfacet_del_count = 0; - backer->total_subfacet_add_count = 0; - backer->total_subfacet_del_count = 0; backer->avg_n_subfacet = 0; backer->avg_subfacet_life = 0; @@ -1277,7 +1188,6 @@ construct(struct ofproto *ofproto_) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); struct shash_node *node, *next; - odp_port_t max_ports; int error; error = open_dpif_backer(ofproto->up.type, &ofproto->backer); @@ -1285,10 +1195,6 @@ construct(struct ofproto *ofproto_) return error; } - max_ports = dpif_get_max_ports(ofproto->backer->dpif); - ofproto_init_max_ports(ofproto_, u16_to_ofp(MIN(odp_to_u32(max_ports), - ofp_to_u16(OFPP_MAX)))); - ofproto->netflow = NULL; ofproto->sflow = NULL; ofproto->ipfix = NULL; @@ -1297,24 +1203,14 @@ construct(struct ofproto *ofproto_) ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME); ofproto->mbridge = mbridge_create(); ofproto->has_bonded_bundles = false; - ovs_mutex_init(&ofproto->vsp_mutex, PTHREAD_MUTEX_NORMAL); + ofproto->lacp_enabled = false; + ovs_mutex_init(&ofproto->stats_mutex); + ovs_mutex_init(&ofproto->vsp_mutex); - classifier_init(&ofproto->facets); + classifier_init(&ofproto->facets, NULL); ofproto->consistency_rl = LLONG_MIN; - list_init(&ofproto->completions); - - ovs_mutex_init(&ofproto->flow_mod_mutex, PTHREAD_MUTEX_NORMAL); - ovs_mutex_lock(&ofproto->flow_mod_mutex); - list_init(&ofproto->flow_mods); - ofproto->n_flow_mods = 0; - ovs_mutex_unlock(&ofproto->flow_mod_mutex); - - ovs_mutex_init(&ofproto->pin_mutex, PTHREAD_MUTEX_NORMAL); - ovs_mutex_lock(&ofproto->pin_mutex); - list_init(&ofproto->pins); - ofproto->n_pins = 0; - ovs_mutex_unlock(&ofproto->pin_mutex); + guarded_list_init(&ofproto->pins); ofproto_dpif_unixctl_init(); @@ -1325,6 +1221,7 @@ construct(struct ofproto *ofproto_) sset_init(&ofproto->ghost_ports); sset_init(&ofproto->port_poll_set); ofproto->port_poll_errno = 0; + ofproto->change_seq = 0; SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) { struct iface_hint *iface_hint = node->data; @@ -1387,9 +1284,12 @@ add_internal_flow(struct ofproto_dpif *ofproto, int id, return error; } - *rulep = rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, - TBL_INTERNAL); - ovs_assert(*rulep != NULL); + if (rule_dpif_lookup_in_table(ofproto, &fm.match.flow, NULL, TBL_INTERNAL, + rulep)) { + rule_dpif_unref(*rulep); + } else { + NOT_REACHED(); + } return 0; } @@ -1429,69 +1329,57 @@ add_internal_flows(struct ofproto_dpif *ofproto) return error; } -static void -complete_operations(struct ofproto_dpif *ofproto) -{ - struct dpif_completion *c, *next; - - LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) { - ofoperation_complete(c->op, 0); - list_remove(&c->list_node); - free(c); - } -} - static void destruct(struct ofproto *ofproto_) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); struct rule_dpif *rule, *next_rule; - struct ofputil_flow_mod *pin, *next_pin; - struct ofputil_flow_mod *fm, *next_fm; + struct ofproto_packet_in *pin, *next_pin; + struct facet *facet, *next_facet; + struct cls_cursor cursor; struct oftable *table; + struct list pins; + + ovs_rwlock_rdlock(&ofproto->facets.rwlock); + cls_cursor_init(&cursor, &ofproto->facets, NULL); + ovs_rwlock_unlock(&ofproto->facets.rwlock); + CLS_CURSOR_FOR_EACH_SAFE (facet, next_facet, cr, &cursor) { + facet_remove(facet); + } ofproto->backer->need_revalidate = REV_RECONFIGURE; ovs_rwlock_wrlock(&xlate_rwlock); xlate_remove_ofproto(ofproto); ovs_rwlock_unlock(&xlate_rwlock); + /* Discard any flow_miss_batches queued up for 'ofproto', avoiding a + * use-after-free error. */ + udpif_revalidate(ofproto->backer->udpif); + hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node); - complete_operations(ofproto); OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { struct cls_cursor cursor; - 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, up.cr, &cursor) { - ofproto_rule_destroy(&ofproto->up, &table->cls, &rule->up); + ofproto_rule_delete(&ofproto->up, &rule->up); } - ovs_rwlock_unlock(&table->cls.rwlock); - } - - ovs_mutex_lock(&ofproto->flow_mod_mutex); - LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &ofproto->flow_mods) { - list_remove(&fm->list_node); - ofproto->n_flow_mods--; - free(fm->ofpacts); - free(fm); } - ovs_mutex_unlock(&ofproto->flow_mod_mutex); - ovs_mutex_destroy(&ofproto->flow_mod_mutex); - ovs_mutex_lock(&ofproto->pin_mutex); - LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &ofproto->pins) { + guarded_list_pop_all(&ofproto->pins, &pins); + LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) { list_remove(&pin->list_node); - ofproto->n_pins--; - free(pin->ofpacts); + free(CONST_CAST(void *, pin->up.packet)); free(pin); } - ovs_mutex_unlock(&ofproto->pin_mutex); - ovs_mutex_destroy(&ofproto->pin_mutex); + guarded_list_destroy(&ofproto->pins); mbridge_unref(ofproto->mbridge); - netflow_destroy(ofproto->netflow); + netflow_unref(ofproto->netflow); dpif_sflow_unref(ofproto->sflow); hmap_destroy(&ofproto->bundles); mac_learning_unref(ofproto->ml); @@ -1505,85 +1393,17 @@ destruct(struct ofproto *ofproto_) sset_destroy(&ofproto->ghost_ports); sset_destroy(&ofproto->port_poll_set); + ovs_mutex_destroy(&ofproto->stats_mutex); ovs_mutex_destroy(&ofproto->vsp_mutex); close_dpif_backer(ofproto->backer); } -static int -run_fast(struct ofproto *ofproto_) -{ - struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - struct ofputil_packet_in *pin, *next_pin; - struct ofputil_flow_mod *fm, *next_fm; - struct list flow_mods, pins; - struct ofport_dpif *ofport; - - /* Do not perform any periodic activity required by 'ofproto' while - * waiting for flow restore to complete. */ - if (ofproto_get_flow_restore_wait()) { - return 0; - } - - ovs_mutex_lock(&ofproto->flow_mod_mutex); - if (ofproto->n_flow_mods) { - flow_mods = ofproto->flow_mods; - list_moved(&flow_mods); - list_init(&ofproto->flow_mods); - ofproto->n_flow_mods = 0; - } else { - list_init(&flow_mods); - } - ovs_mutex_unlock(&ofproto->flow_mod_mutex); - - LIST_FOR_EACH_SAFE (fm, next_fm, list_node, &flow_mods) { - int error = ofproto_flow_mod(&ofproto->up, fm); - if (error && !VLOG_DROP_WARN(&rl)) { - VLOG_WARN("learning action failed to modify flow table (%s)", - ofperr_get_name(error)); - } - - list_remove(&fm->list_node); - free(fm->ofpacts); - free(fm); - } - - ovs_mutex_lock(&ofproto->pin_mutex); - if (ofproto->n_pins) { - pins = ofproto->pins; - list_moved(&pins); - list_init(&ofproto->pins); - ofproto->n_pins = 0; - } else { - list_init(&pins); - } - ovs_mutex_unlock(&ofproto->pin_mutex); - - LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) { - connmgr_send_packet_in(ofproto->up.connmgr, pin); - list_remove(&pin->list_node); - free(CONST_CAST(void *, pin->packet)); - free(pin); - } - - HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { - port_run_fast(ofport); - } - - return 0; -} - static int run(struct ofproto *ofproto_) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - struct ofport_dpif *ofport; - struct ofbundle *bundle; - int error; - - if (!clogged) { - complete_operations(ofproto); - } + uint64_t new_seq; if (mbridge_need_revalidate(ofproto->mbridge)) { ofproto->backer->need_revalidate = REV_RECONFIGURE; @@ -1592,31 +1412,47 @@ run(struct ofproto *ofproto_) ovs_rwlock_unlock(&ofproto->ml->rwlock); } - /* Do not perform any periodic activity below required by 'ofproto' while + /* Do not perform any periodic activity required by 'ofproto' while * waiting for flow restore to complete. */ - if (ofproto_get_flow_restore_wait()) { - return 0; - } + if (!ofproto_get_flow_restore_wait()) { + struct ofproto_packet_in *pin, *next_pin; + struct list pins; - error = run_fast(ofproto_); - if (error) { - return error; + guarded_list_pop_all(&ofproto->pins, &pins); + LIST_FOR_EACH_SAFE (pin, next_pin, list_node, &pins) { + connmgr_send_packet_in(ofproto->up.connmgr, pin); + list_remove(&pin->list_node); + free(CONST_CAST(void *, pin->up.packet)); + free(pin); + } } if (ofproto->netflow) { - if (netflow_run(ofproto->netflow)) { - send_netflow_active_timeouts(ofproto); - } + netflow_run(ofproto->netflow); } if (ofproto->sflow) { dpif_sflow_run(ofproto->sflow); } + if (ofproto->ipfix) { + dpif_ipfix_run(ofproto->ipfix); + } + + new_seq = seq_read(connectivity_seq_get()); + if (ofproto->change_seq != new_seq) { + struct ofport_dpif *ofport; - HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { - port_run(ofport); + HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { + port_run(ofport); + } + + ofproto->change_seq = new_seq; } - HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { - bundle_run(bundle); + if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) { + struct ofbundle *bundle; + + HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { + bundle_run(bundle); + } } stp_run(ofproto); @@ -1631,14 +1467,14 @@ run(struct ofproto *ofproto_) if (time_msec() >= ofproto->consistency_rl && !classifier_is_empty(&ofproto->facets) && !ofproto->backer->need_revalidate) { - struct cls_table *table; + struct cls_subtable *table; struct cls_rule *cr; struct facet *facet; ofproto->consistency_rl = time_msec() + 250; - table = CONTAINER_OF(hmap_random_node(&ofproto->facets.tables), - struct cls_table, hmap_node); + table = CONTAINER_OF(hmap_random_node(&ofproto->facets.subtables), + struct cls_subtable, hmap_node); cr = CONTAINER_OF(hmap_random_node(&table->rules), struct cls_rule, hmap_node); facet = CONTAINER_OF(cr, struct facet, cr); @@ -1656,27 +1492,23 @@ static void wait(struct ofproto *ofproto_) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - struct ofport_dpif *ofport; - struct ofbundle *bundle; - - if (!clogged && !list_is_empty(&ofproto->completions)) { - poll_immediate_wake(); - } if (ofproto_get_flow_restore_wait()) { return; } - dpif_wait(ofproto->backer->dpif); - dpif_recv_wait(ofproto->backer->dpif); if (ofproto->sflow) { dpif_sflow_wait(ofproto->sflow); } - HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { - port_wait(ofport); + if (ofproto->ipfix) { + dpif_ipfix_wait(ofproto->ipfix); } - HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { - bundle_wait(bundle); + if (ofproto->lacp_enabled || ofproto->has_bonded_bundles) { + struct ofbundle *bundle; + + HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { + bundle_wait(bundle); + } } if (ofproto->netflow) { netflow_wait(ofproto->netflow); @@ -1821,6 +1653,7 @@ port_construct(struct ofport *port_) port->realdev_ofp_port = 0; port->vlandev_vid = 0; port->carrier_seq = netdev_get_carrier_resets(netdev); + port->is_layer3 = netdev_vport_is_layer3(netdev); if (netdev_vport_is_patch(netdev)) { /* By bailing out here, we don't submit the port to the sFlow module @@ -1933,6 +1766,13 @@ port_modified(struct ofport *port_) cfm_set_netdev(port->cfm, port->up.netdev); } + if (port->bfd) { + bfd_set_netdev(port->bfd, port->up.netdev); + } + + ofproto_dpif_monitor_port_update(port, port->bfd, port->cfm, + port->up.pp.hw_addr); + if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev, port->odp_port)) { ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate = @@ -1997,20 +1837,25 @@ set_ipfix( { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); struct dpif_ipfix *di = ofproto->ipfix; + bool has_options = bridge_exporter_options || flow_exporters_options; - if (bridge_exporter_options || flow_exporters_options) { - if (!di) { - di = ofproto->ipfix = dpif_ipfix_create(); - } + if (has_options && !di) { + di = ofproto->ipfix = dpif_ipfix_create(); + } + + if (di) { + /* Call set_options in any case to cleanly flush the flow + * caches in the last exporters that are to be destroyed. */ dpif_ipfix_set_options( di, bridge_exporter_options, flow_exporters_options, n_flow_exporters_options); - } else { - if (di) { + + if (!has_options) { dpif_ipfix_unref(di); ofproto->ipfix = NULL; } } + return 0; } @@ -2018,11 +1863,9 @@ static int set_cfm(struct ofport *ofport_, const struct cfm_settings *s) { struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); - int error; + int error = 0; - if (!s) { - error = 0; - } else { + if (s) { if (!ofport->cfm) { struct ofproto_dpif *ofproto; @@ -2032,13 +1875,17 @@ set_cfm(struct ofport *ofport_, const struct cfm_settings *s) } if (cfm_configure(ofport->cfm, s)) { - return 0; + error = 0; + goto out; } error = EINVAL; } cfm_unref(ofport->cfm); ofport->cfm = NULL; +out: + ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm, + ofport->up.pp.hw_addr); return error; } @@ -2050,6 +1897,7 @@ get_cfm_status(const struct ofport *ofport_, if (ofport->cfm) { status->faults = cfm_get_fault(ofport->cfm); + status->flap_count = cfm_get_flap_count(ofport->cfm); status->remote_opstate = cfm_get_opup(ofport->cfm); status->health = cfm_get_health(ofport->cfm); cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps); @@ -2067,11 +1915,13 @@ set_bfd(struct ofport *ofport_, const struct smap *cfg) struct bfd *old; old = ofport->bfd; - ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg); + ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), + cfg, ofport->up.netdev); if (ofport->bfd != old) { ofproto->backer->need_revalidate = REV_RECONFIGURE; } - + ofproto_dpif_monitor_port_update(ofport, ofport->bfd, ofport->cfm, + ofport->up.pp.hw_addr); return 0; } @@ -2109,7 +1959,7 @@ send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_) VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " "with unknown MAC", ofproto->up.name, port_num); } else { - send_packet(ofport, pkt); + ofproto_dpif_send_packet(ofport, pkt); } } ofpbuf_delete(pkt); @@ -2272,6 +2122,24 @@ get_stp_port_status(struct ofport *ofport_, s->state = stp_port_get_state(sp); s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000; s->role = stp_port_get_role(sp); + + return 0; +} + +static int +get_stp_port_stats(struct ofport *ofport_, + struct ofproto_port_stp_stats *s) +{ + struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); + struct stp_port *sp = ofport->stp_port; + + if (!ofproto->stp || !sp) { + s->enabled = false; + return 0; + } + + s->enabled = true; stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count); return 0; @@ -2401,6 +2269,7 @@ bundle_update(struct ofbundle *bundle) bundle->floodable = true; LIST_FOR_EACH (port, bundle_node, &bundle->ports) { if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD + || port->is_layer3 || !stp_forward_in_state(port->stp_state)) { bundle->floodable = false; break; @@ -2442,12 +2311,13 @@ bundle_add_port(struct ofbundle *bundle, ofp_port_t ofp_port, if (port->bundle != bundle) { bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; if (port->bundle) { - bundle_del_port(port); + bundle_remove(&port->up); } port->bundle = bundle; list_push_back(&bundle->ports, &port->bundle_node); if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD + || port->is_layer3 || !stp_forward_in_state(port->stp_state)) { bundle->floodable = false; } @@ -2540,6 +2410,7 @@ bundle_set(struct ofproto *ofproto_, void *aux, /* LACP. */ if (s->lacp) { + ofproto->lacp_enabled = true; if (!bundle->lacp) { ofproto->backer->need_revalidate = REV_RECONFIGURE; bundle->lacp = lacp_create(); @@ -2705,7 +2576,7 @@ send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) pdu_size); memcpy(packet_pdu, pdu, pdu_size); - send_packet(port, &packet); + ofproto_dpif_send_packet(port, &packet); ofpbuf_uninit(&packet); } else { VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " @@ -2718,35 +2589,39 @@ static void bundle_send_learning_packets(struct ofbundle *bundle) { struct ofproto_dpif *ofproto = bundle->ofproto; + struct ofpbuf *learning_packet; int error, n_packets, n_errors; struct mac_entry *e; + struct list packets; - error = n_packets = n_errors = 0; + list_init(&packets); ovs_rwlock_rdlock(&ofproto->ml->rwlock); LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { if (e->port.p != bundle) { - struct ofpbuf *learning_packet; - struct ofport_dpif *port; void *port_void; - int ret; - /* The assignment to "port" is unnecessary but makes "grep"ing for - * struct ofport_dpif more effective. */ learning_packet = bond_compose_learning_packet(bundle->bond, e->mac, e->vlan, &port_void); - port = port_void; - ret = send_packet(port, learning_packet); - ofpbuf_delete(learning_packet); - if (ret) { - error = ret; - n_errors++; - } - n_packets++; + learning_packet->private_p = port_void; + list_push_back(&packets, &learning_packet->list_node); } } ovs_rwlock_unlock(&ofproto->ml->rwlock); + error = n_packets = n_errors = 0; + LIST_FOR_EACH (learning_packet, list_node, &packets) { + int ret; + + ret = ofproto_dpif_send_packet(learning_packet->private_p, learning_packet); + if (ret) { + error = ret; + n_errors++; + } + n_packets++; + } + ofpbuf_list_delete(&packets); + if (n_errors) { static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " @@ -2883,13 +2758,6 @@ get_ofp_port(const struct ofproto_dpif *ofproto, ofp_port_t ofp_port) return ofport ? ofport_dpif_cast(ofport) : NULL; } -static struct ofport_dpif * -get_odp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port) -{ - struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port); - return port && &ofproto->up == port->up.ofproto ? port : NULL; -} - static void ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto, struct ofproto_port *ofproto_port, @@ -2905,7 +2773,7 @@ ofport_update_peer(struct ofport_dpif *ofport) { const struct ofproto_dpif *ofproto; struct dpif_backer *backer; - const char *peer_name; + char *peer_name; if (!netdev_vport_is_patch(ofport->up.netdev)) { return; @@ -2927,7 +2795,7 @@ ofport_update_peer(struct ofport_dpif *ofport) HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { struct ofport *peer_ofport; struct ofport_dpif *peer; - const char *peer_peer; + char *peer_peer; if (ofproto->backer != backer) { continue; @@ -2945,31 +2813,11 @@ ofport_update_peer(struct ofport_dpif *ofport) ofport->peer = peer; ofport->peer->peer = ofport; } + free(peer_peer); - return; - } -} - -static void -port_run_fast(struct ofport_dpif *ofport) -{ - if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) { - struct ofpbuf packet; - - ofpbuf_init(&packet, 0); - cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr); - send_packet(ofport, &packet); - ofpbuf_uninit(&packet); - } - - if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) { - struct ofpbuf packet; - - ofpbuf_init(&packet, 0); - bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr); - send_packet(ofport, &packet); - ofpbuf_uninit(&packet); + break; } + free(peer_name); } static void @@ -2978,25 +2826,27 @@ port_run(struct ofport_dpif *ofport) long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev); bool carrier_changed = carrier_seq != ofport->carrier_seq; bool enable = netdev_get_carrier(ofport->up.netdev); + bool cfm_enable = false; + bool bfd_enable = false; ofport->carrier_seq = carrier_seq; - port_run_fast(ofport); - if (ofport->cfm) { int cfm_opup = cfm_get_opup(ofport->cfm); - cfm_run(ofport->cfm); - enable = enable && !cfm_get_fault(ofport->cfm); + cfm_enable = !cfm_get_fault(ofport->cfm); if (cfm_opup >= 0) { - enable = enable && cfm_opup; + cfm_enable = cfm_enable && cfm_opup; } } if (ofport->bfd) { - bfd_run(ofport->bfd); - enable = enable && bfd_forwarding(ofport->bfd); + bfd_enable = bfd_forwarding(ofport->bfd); + } + + if (ofport->bfd || ofport->cfm) { + enable = enable && (cfm_enable || bfd_enable); } if (ofport->bundle) { @@ -3014,18 +2864,6 @@ port_run(struct ofport_dpif *ofport) ofport->may_enable = enable; } -static void -port_wait(struct ofport_dpif *ofport) -{ - if (ofport->cfm) { - cfm_wait(ofport->cfm); - } - - if (ofport->bfd) { - bfd_wait(ofport->bfd); - } -} - static int port_query_by_name(const struct ofproto *ofproto_, const char *devname, struct ofproto_port *ofproto_port) @@ -3113,7 +2951,7 @@ port_del(struct ofproto *ofproto_, ofp_port_t ofp_port) sset_find_and_delete(&ofproto->ghost_ports, netdev_get_name(ofport->up.netdev)); ofproto->backer->need_revalidate = REV_RECONFIGURE; - if (!ofport->is_tunnel) { + if (!ofport->is_tunnel && !netdev_vport_is_patch(ofport->up.netdev)) { error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port); if (!error) { /* The caller is going to close ofport->up.netdev. If this is a @@ -3139,10 +2977,11 @@ port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) if (!error && ofport_->ofp_port == OFPP_LOCAL) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); + ovs_mutex_lock(&ofproto->stats_mutex); /* ofproto->stats.tx_packets represents packets that we created * internally and sent to some port (e.g. packets sent with - * send_packet()). Account for them as if they had come from - * OFPP_LOCAL and got forwarded. */ + * ofproto_dpif_send_packet()). Account for them as if they had + * come from OFPP_LOCAL and got forwarded. */ if (stats->rx_packets != UINT64_MAX) { stats->rx_packets += ofproto->stats.tx_packets; @@ -3163,6 +3002,7 @@ port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) if (stats->tx_bytes != UINT64_MAX) { stats->tx_bytes += ofproto->stats.rx_bytes; } + ovs_mutex_unlock(&ofproto->stats_mutex); } return error; @@ -3270,26 +3110,6 @@ port_is_lacp_current(const struct ofport *ofport_) /* Upcall handling. */ -/* Flow miss batching. - * - * Some dpifs implement operations faster when you hand them off in a batch. - * To allow batching, "struct flow_miss" queues the dpif-related work needed - * for a given flow. Each "struct flow_miss" corresponds to sending one or - * more packets, plus possibly installing the flow in the dpif. - * - * So far we only batch the operations that affect flow setup time the most. - * It's possible to batch more than that, but the benefit might be minimal. */ -struct flow_miss { - struct hmap_node hmap_node; - struct ofproto_dpif *ofproto; - struct flow flow; - enum odp_key_fitness key_fitness; - const struct nlattr *key; - size_t key_len; - struct list packets; - enum dpif_upcall_type upcall_type; -}; - struct flow_miss_op { struct dpif_op dpif_op; @@ -3305,96 +3125,6 @@ struct flow_miss_op { struct subfacet *subfacet; }; -/* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each - * OpenFlow controller as necessary according to their individual - * configurations. */ -static void -send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet, - const struct flow *flow) -{ - struct ofputil_packet_in pin; - - pin.packet = packet->data; - pin.packet_len = packet->size; - pin.reason = OFPR_NO_MATCH; - pin.controller_id = 0; - - pin.table_id = 0; - pin.cookie = 0; - - pin.send_len = 0; /* not used for flow table misses */ - - flow_get_metadata(flow, &pin.fmd); - - connmgr_send_packet_in(ofproto->up.connmgr, &pin); -} - -static struct flow_miss * -flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto, - const struct flow *flow, uint32_t hash) -{ - struct flow_miss *miss; - - HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) { - if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) { - return miss; - } - } - - return NULL; -} - -/* Partially Initializes 'op' as an "execute" operation for 'miss' and - * 'packet'. The caller must initialize op->actions and op->actions_len. If - * 'miss' is associated with a subfacet the caller must also initialize the - * returned op->subfacet, and if anything needs to be freed after processing - * the op, the caller must initialize op->garbage also. */ -static void -init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet, - struct flow_miss_op *op) -{ - if (miss->flow.in_port.ofp_port - != vsp_realdev_to_vlandev(miss->ofproto, miss->flow.in_port.ofp_port, - miss->flow.vlan_tci)) { - /* This packet was received on a VLAN splinter port. We - * added a VLAN to the packet to make the packet resemble - * the flow, but the actions were composed assuming that - * the packet contained no VLAN. So, we must remove the - * VLAN header from the packet before trying to execute the - * actions. */ - eth_pop_vlan(packet); - } - - op->subfacet = NULL; - op->xout_garbage = false; - op->dpif_op.type = DPIF_OP_EXECUTE; - op->dpif_op.u.execute.key = miss->key; - op->dpif_op.u.execute.key_len = miss->key_len; - op->dpif_op.u.execute.packet = packet; - ofpbuf_use_stack(&op->mask, &op->maskbuf, sizeof op->maskbuf); -} - -/* Helper for handle_flow_miss_without_facet() and - * handle_flow_miss_with_facet(). */ -static void -handle_flow_miss_common(struct ofproto_dpif *ofproto, struct ofpbuf *packet, - const struct flow *flow, bool fail_open) -{ - if (fail_open) { - /* - * Extra-special case for fail-open mode. - * - * We are in fail-open mode and the packet matched the fail-open - * rule, but we are connected to a controller too. We should send - * the packet up to the controller in the hope that it will try to - * set up a flow and thereby allow us to exit fail-open. - * - * See the top-level comment in fail-open.c for more information. - */ - send_packet_in_miss(ofproto, packet, flow); - } -} - /* Figures out whether a flow that missed in 'ofproto', whose details are in * 'miss' masked by 'wc', is likely to be worth tracking in detail in userspace * and (usually) installing a datapath flow. The answer is usually "yes" (a @@ -3403,7 +3133,7 @@ handle_flow_miss_common(struct ofproto_dpif *ofproto, struct ofpbuf *packet, * flows we impose some heuristics to decide which flows are likely to be worth * tracking. */ static bool -flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc) +flow_miss_should_make_facet(struct flow_miss *miss) { struct dpif_backer *backer = miss->ofproto->backer; uint32_t hash; @@ -3428,95 +3158,51 @@ flow_miss_should_make_facet(struct flow_miss *miss, struct flow_wildcards *wc) backer->governor = governor_create(); } - hash = flow_hash_in_wildcards(&miss->flow, wc, 0); + hash = flow_hash_in_wildcards(&miss->flow, &miss->xout.wc, 0); return governor_should_install_flow(backer->governor, hash, - list_size(&miss->packets)); -} - -/* Handles 'miss' without creating a facet or subfacet or creating any datapath - * flow. 'miss->flow' must have matched 'rule' and been xlated into 'xout'. - * May add an "execute" operation to 'ops' and increment '*n_ops'. */ -static void -handle_flow_miss_without_facet(struct rule_dpif *rule, struct xlate_out *xout, - struct flow_miss *miss, - struct flow_miss_op *ops, size_t *n_ops) -{ - struct ofpbuf *packet; - - LIST_FOR_EACH (packet, list_node, &miss->packets) { - - COVERAGE_INC(facet_suppress); - - handle_flow_miss_common(miss->ofproto, packet, &miss->flow, - rule->up.cr.priority == FAIL_OPEN_PRIORITY); - - if (xout->slow) { - struct xlate_in xin; - - xlate_in_init(&xin, miss->ofproto, &miss->flow, rule, 0, packet); - xlate_actions_for_side_effects(&xin); - } - - if (xout->odp_actions.size) { - struct flow_miss_op *op = &ops[*n_ops]; - struct dpif_execute *execute = &op->dpif_op.u.execute; - - init_flow_miss_execute_op(miss, packet, op); - xlate_out_copy(&op->xout, xout); - execute->actions = op->xout.odp_actions.data; - execute->actions_len = op->xout.odp_actions.size; - op->xout_garbage = true; - - (*n_ops)++; - } - } + miss->stats.n_packets); } /* Handles 'miss', which matches 'facet'. May add any required datapath * operations to 'ops', incrementing '*n_ops' for each new op. * - * All of the packets in 'miss' are considered to have arrived at time 'now'. - * This is really important only for new facets: if we just called time_msec() - * here, then the new subfacet or its packets could look (occasionally) as - * though it was used some time after the facet was used. That can make a - * one-packet flow look like it has a nonzero duration, which looks odd in - * e.g. NetFlow statistics. - * - * If non-null, 'stats' will be folded into 'facet'. */ + * All of the packets in 'miss' are considered to have arrived at time + * 'miss->stats.used'. This is really important only for new facets: if we + * just called time_msec() here, then the new subfacet or its packets could + * look (occasionally) as though it was used some time after the facet was + * used. That can make a one-packet flow look like it has a nonzero duration, + * which looks odd in e.g. NetFlow statistics. */ static void handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, - long long int now, struct dpif_flow_stats *stats, struct flow_miss_op *ops, size_t *n_ops) { enum subfacet_path want_path; struct subfacet *subfacet; - struct ofpbuf *packet; - - want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH; - - LIST_FOR_EACH (packet, list_node, &miss->packets) { - struct flow_miss_op *op = &ops[*n_ops]; - - handle_flow_miss_common(miss->ofproto, packet, &miss->flow, - facet->fail_open); - - if (want_path != SF_FAST_PATH) { - struct rule_dpif *rule; - struct xlate_in xin; - - rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL); - xlate_in_init(&xin, facet->ofproto, &miss->flow, rule, 0, packet); - xlate_actions_for_side_effects(&xin); - } - - if (facet->xout.odp_actions.size) { - struct dpif_execute *execute = &op->dpif_op.u.execute; + uint32_t key_hash; - init_flow_miss_execute_op(miss, packet, op); - execute->actions = facet->xout.odp_actions.data, - execute->actions_len = facet->xout.odp_actions.size; - (*n_ops)++; + /* Update facet stats. */ + facet->packet_count += miss->stats.n_packets; + facet->prev_packet_count += miss->stats.n_packets; + facet->byte_count += miss->stats.n_bytes; + facet->prev_byte_count += miss->stats.n_bytes; + + /* Look for an existing subfacet. If we find one, update its used time. */ + key_hash = odp_flow_key_hash(miss->key, miss->key_len); + if (!list_is_empty(&facet->subfacets)) { + subfacet = subfacet_find(miss->ofproto->backer, + miss->key, miss->key_len, key_hash); + if (subfacet) { + if (subfacet->facet == facet) { + subfacet->used = MAX(subfacet->used, miss->stats.used); + } else { + /* This shouldn't happen. */ + VLOG_ERR_RL(&rl, "subfacet with wrong facet"); + subfacet_destroy(subfacet); + subfacet = NULL; + } } + } else { + subfacet = NULL; } /* Don't install the flow if it's the result of the "userspace" @@ -3526,20 +3212,16 @@ handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, * be rejected as overlapping by the datapath. */ if (miss->upcall_type == DPIF_UC_ACTION && !list_is_empty(&facet->subfacets)) { - if (stats) { - facet->used = MAX(facet->used, stats->used); - facet->packet_count += stats->n_packets; - facet->byte_count += stats->n_bytes; - facet->tcp_flags |= stats->tcp_flags; - } return; } - subfacet = subfacet_create(facet, miss, now); - if (stats) { - subfacet_update_stats(subfacet, stats); + /* Create a subfacet, if we don't already have one. */ + if (!subfacet) { + subfacet = subfacet_create(facet, miss, key_hash); } + /* Install the subfacet, if it's not already installed. */ + want_path = facet->xout.slow ? SF_SLOW_PATH : SF_FAST_PATH; if (subfacet->path != want_path) { struct flow_miss_op *op = &ops[(*n_ops)++]; struct dpif_flow_put *put = &op->dpif_op.u.flow_put; @@ -3579,55 +3261,25 @@ static void handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops, size_t *n_ops) { - struct ofproto_dpif *ofproto = miss->ofproto; - struct dpif_flow_stats stats__; - struct dpif_flow_stats *stats = &stats__; - struct ofpbuf *packet; struct facet *facet; - long long int now; - now = time_msec(); - memset(stats, 0, sizeof *stats); - stats->used = now; - LIST_FOR_EACH (packet, list_node, &miss->packets) { - stats->tcp_flags |= packet_get_tcp_flags(packet, &miss->flow); - stats->n_bytes += packet->size; - stats->n_packets++; - } + miss->ofproto->n_missed += miss->stats.n_packets; - facet = facet_lookup_valid(ofproto, &miss->flow); + facet = facet_lookup_valid(miss->ofproto, &miss->flow); if (!facet) { - struct flow_wildcards wc; - struct rule_dpif *rule; - struct xlate_out xout; - struct xlate_in xin; - - flow_wildcards_init_catchall(&wc); - rule = rule_dpif_lookup(ofproto, &miss->flow, &wc); - rule_credit_stats(rule, stats); - - xlate_in_init(&xin, ofproto, &miss->flow, rule, stats->tcp_flags, - NULL); - xin.resubmit_stats = stats; - xin.may_learn = true; - xlate_actions(&xin, &xout); - flow_wildcards_or(&xout.wc, &xout.wc, &wc); - /* There does not exist a bijection between 'struct flow' and datapath * flow keys with fitness ODP_FIT_TO_LITTLE. This breaks a fundamental * assumption used throughout the facet and subfacet handling code. * Since we have to handle these misses in userspace anyway, we simply * skip facet creation, avoiding the problem altogether. */ if (miss->key_fitness == ODP_FIT_TOO_LITTLE - || !flow_miss_should_make_facet(miss, &xout.wc)) { - handle_flow_miss_without_facet(rule, &xout, miss, ops, n_ops); + || !flow_miss_should_make_facet(miss)) { return; } - facet = facet_create(miss, rule, &xout, stats); - stats = NULL; + facet = facet_create(miss); } - handle_flow_miss_with_facet(miss, facet, now, stats, ops, n_ops); + handle_flow_miss_with_facet(miss, facet, ops, n_ops); } static struct drop_key * @@ -3666,109 +3318,24 @@ drop_key_clear(struct dpif_backer *backer) } hmap_remove(&backer->drop_keys, &drop_key->hmap_node); - free(drop_key->key); - free(drop_key); + drop_key_destroy(drop_key); } + + udpif_drop_key_clear(backer->udpif); } static void -handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls, - size_t n_upcalls) +handle_flow_misses(struct dpif_backer *backer, struct flow_miss_batch *fmb) { - struct dpif_upcall *upcall; + struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH]; + struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH]; struct flow_miss *miss; - struct flow_miss misses[FLOW_MISS_MAX_BATCH]; - struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; - struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; - struct hmap todo; - int n_misses; - size_t n_ops; - size_t i; - - if (!n_upcalls) { - return; - } - - /* Construct the to-do list. - * - * This just amounts to extracting the flow from each packet and sticking - * the packets that have the same flow in the same "flow_miss" structure so - * that we can process them together. */ - hmap_init(&todo); - n_misses = 0; - for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { - struct flow_miss *miss = &misses[n_misses]; - struct flow_miss *existing_miss; - struct ofproto_dpif *ofproto; - odp_port_t odp_in_port; - struct flow flow; - uint32_t hash; - int error; - - error = xlate_receive(backer, upcall->packet, upcall->key, - upcall->key_len, &flow, &miss->key_fitness, - &ofproto, &odp_in_port); - if (error == ENODEV) { - struct drop_key *drop_key; - - /* Received packet on datapath port for which we couldn't - * associate an ofproto. This can happen if a port is removed - * while traffic is being received. Print a rate-limited message - * in case it happens frequently. Install a drop flow so - * that future packets of the flow are inexpensively dropped - * in the kernel. */ - VLOG_INFO_RL(&rl, "received packet on unassociated datapath port " - "%"PRIu32, odp_in_port); - - drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len); - if (!drop_key) { - int ret; - ret = dpif_flow_put(backer->dpif, - DPIF_FP_CREATE | DPIF_FP_MODIFY, - upcall->key, upcall->key_len, - NULL, 0, NULL, 0, NULL); - - if (!ret) { - drop_key = xmalloc(sizeof *drop_key); - drop_key->key = xmemdup(upcall->key, upcall->key_len); - drop_key->key_len = upcall->key_len; - - hmap_insert(&backer->drop_keys, &drop_key->hmap_node, - hash_bytes(drop_key->key, drop_key->key_len, 0)); - } - } - continue; - } - if (error) { - continue; - } - - ofproto->n_missed++; - flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark, - &flow.tunnel, &flow.in_port, &miss->flow); - - /* Add other packets to a to-do list. */ - hash = flow_hash(&miss->flow, 0); - existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash); - if (!existing_miss) { - hmap_insert(&todo, &miss->hmap_node, hash); - miss->ofproto = ofproto; - miss->key = upcall->key; - miss->key_len = upcall->key_len; - miss->upcall_type = upcall->type; - list_init(&miss->packets); - - n_misses++; - } else { - miss = existing_miss; - } - list_push_back(&miss->packets, &upcall->packet->list_node); - } + size_t n_ops, i; /* Process each element in the to-do list, constructing the set of * operations to batch. */ n_ops = 0; - HMAP_FOR_EACH (miss, hmap_node, &todo) { + HMAP_FOR_EACH (miss, hmap_node, &fmb->misses) { handle_flow_miss(miss, flow_miss_ops, &n_ops); } ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); @@ -3801,201 +3368,44 @@ handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls, subfacet->path = SF_NOT_INSTALLED; } - - /* Free memory. */ - if (flow_miss_ops[i].xout_garbage) { - xlate_out_uninit(&flow_miss_ops[i].xout); - } } - hmap_destroy(&todo); -} - -static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL, - IPFIX_UPCALL } -classify_upcall(const struct dpif_upcall *upcall) -{ - size_t userdata_len; - union user_action_cookie cookie; - - /* First look at the upcall type. */ - switch (upcall->type) { - case DPIF_UC_ACTION: - break; - - case DPIF_UC_MISS: - return MISS_UPCALL; - - case DPIF_N_UC_TYPES: - default: - VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); - return BAD_UPCALL; - } - - /* "action" upcalls need a closer look. */ - if (!upcall->userdata) { - VLOG_WARN_RL(&rl, "action upcall missing cookie"); - return BAD_UPCALL; - } - userdata_len = nl_attr_get_size(upcall->userdata); - if (userdata_len < sizeof cookie.type - || userdata_len > sizeof cookie) { - VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu", - userdata_len); - return BAD_UPCALL; - } - memset(&cookie, 0, sizeof cookie); - memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len); - if (userdata_len == sizeof cookie.sflow - && cookie.type == USER_ACTION_COOKIE_SFLOW) { - return SFLOW_UPCALL; - } else if (userdata_len == sizeof cookie.slow_path - && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) { - return MISS_UPCALL; - } else if (userdata_len == sizeof cookie.flow_sample - && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) { - return FLOW_SAMPLE_UPCALL; - } else if (userdata_len == sizeof cookie.ipfix - && cookie.type == USER_ACTION_COOKIE_IPFIX) { - return IPFIX_UPCALL; - } else { - VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16 - " and size %zu", cookie.type, userdata_len); - return BAD_UPCALL; - } -} - -static void -handle_sflow_upcall(struct dpif_backer *backer, - const struct dpif_upcall *upcall) -{ - struct ofproto_dpif *ofproto; - union user_action_cookie cookie; - struct flow flow; - odp_port_t odp_in_port; - - if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len, - &flow, NULL, &ofproto, &odp_in_port) - || !ofproto->sflow) { - return; - } - - memset(&cookie, 0, sizeof cookie); - memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow); - dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, - odp_in_port, &cookie); } static void -handle_flow_sample_upcall(struct dpif_backer *backer, - const struct dpif_upcall *upcall) +handle_upcalls(struct dpif_backer *backer) { - struct ofproto_dpif *ofproto; - union user_action_cookie cookie; - struct flow flow; - - if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len, - &flow, NULL, &ofproto, NULL) - || !ofproto->ipfix) { - return; - } - - memset(&cookie, 0, sizeof cookie); - memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample); - - /* The flow reflects exactly the contents of the packet. Sample - * the packet using it. */ - dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow, - cookie.flow_sample.collector_set_id, - cookie.flow_sample.probability, - cookie.flow_sample.obs_domain_id, - cookie.flow_sample.obs_point_id); -} - -static void -handle_ipfix_upcall(struct dpif_backer *backer, - const struct dpif_upcall *upcall) -{ - struct ofproto_dpif *ofproto; - struct flow flow; - - if (xlate_receive(backer, upcall->packet, upcall->key, upcall->key_len, - &flow, NULL, &ofproto, NULL) - || !ofproto->ipfix) { - return; - } - - /* The flow reflects exactly the contents of the packet. Sample - * the packet using it. */ - dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow); -} - -static int -handle_upcalls(struct dpif_backer *backer, unsigned int max_batch) -{ - struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; - struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH]; - uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8]; + struct flow_miss_batch *fmb; int n_processed; - int n_misses; - int i; - - ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH); - n_misses = 0; - for (n_processed = 0; n_processed < max_batch; n_processed++) { - struct dpif_upcall *upcall = &misses[n_misses]; - struct ofpbuf *buf = &miss_bufs[n_misses]; - int error; - - ofpbuf_use_stub(buf, miss_buf_stubs[n_misses], - sizeof miss_buf_stubs[n_misses]); - error = dpif_recv(backer->dpif, upcall, buf); - if (error) { - ofpbuf_uninit(buf); + for (n_processed = 0; n_processed < FLOW_MISS_MAX_BATCH; n_processed++) { + struct drop_key *drop_key = drop_key_next(backer->udpif); + if (!drop_key) { break; } - switch (classify_upcall(upcall)) { - case MISS_UPCALL: - /* Handle it later. */ - n_misses++; - break; - - case SFLOW_UPCALL: - handle_sflow_upcall(backer, upcall); - ofpbuf_uninit(buf); - break; - - case FLOW_SAMPLE_UPCALL: - handle_flow_sample_upcall(backer, upcall); - ofpbuf_uninit(buf); - break; - - case IPFIX_UPCALL: - handle_ipfix_upcall(backer, upcall); - ofpbuf_uninit(buf); - break; - - case BAD_UPCALL: - ofpbuf_uninit(buf); - break; + if (!drop_key_lookup(backer, drop_key->key, drop_key->key_len)) { + hmap_insert(&backer->drop_keys, &drop_key->hmap_node, + hash_bytes(drop_key->key, drop_key->key_len, 0)); + dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY, + drop_key->key, drop_key->key_len, + NULL, 0, NULL, 0, NULL); + } else { + drop_key_destroy(drop_key); } } - /* Handle deferred MISS_UPCALL processing. */ - handle_miss_upcalls(backer, misses, n_misses); - for (i = 0; i < n_misses; i++) { - ofpbuf_uninit(&miss_bufs[i]); + fmb = flow_miss_batch_next(backer->udpif); + if (fmb) { + handle_flow_misses(backer, fmb); + flow_miss_batch_destroy(fmb); } - - return n_processed; } /* Flow expiration. */ static int subfacet_max_idle(const struct dpif_backer *); static void update_stats(struct dpif_backer *); -static void rule_expire(struct rule_dpif *); +static void rule_expire(struct rule_dpif *) OVS_REQUIRES(ofproto_mutex); static void expire_subfacets(struct dpif_backer *, int dp_max_idle); /* This function is called periodically by run(). Its job is to collect @@ -4049,12 +3459,12 @@ expire(struct dpif_backer *backer) /* Expire OpenFlow flows whose idle_timeout or hard_timeout * has passed. */ - ovs_mutex_lock(&ofproto->up.expirable_mutex); + ovs_mutex_lock(&ofproto_mutex); LIST_FOR_EACH_SAFE (rule, next_rule, expirable, &ofproto->up.expirable) { rule_expire(rule_dpif_cast(rule)); } - ovs_mutex_unlock(&ofproto->up.expirable_mutex); + ovs_mutex_unlock(&ofproto_mutex); /* All outstanding data in existing flows has been accounted, so it's a * good time to do bond rebalancing. */ @@ -4103,10 +3513,8 @@ update_subfacet_stats(struct subfacet *subfacet, subfacet->dp_byte_count = stats->n_bytes; subfacet_update_stats(subfacet, &diff); - if (facet->accounted_bytes < facet->byte_count) { + if (diff.n_packets) { facet_learn(facet); - facet_account(facet); - facet->accounted_bytes = facet->byte_count; } } @@ -4175,11 +3583,8 @@ update_stats(struct dpif_backer *backer) delete_unexpected_flow(backer, key, key_len); break; } - run_fast_rl(); } dpif_flow_dump_done(&dump); - - update_moving_averages(backer); } /* Calculates and returns the number of milliseconds of idle time after which @@ -4220,7 +3625,7 @@ subfacet_max_idle(const struct dpif_backer *backer) * pass made by update_stats(), because the former function never looks at * uninstallable subfacets. */ - enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; + enum { BUCKET_WIDTH = 100 }; enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; int buckets[N_BUCKETS] = { 0 }; int total, subtotal, bucket; @@ -4315,35 +3720,31 @@ expire_subfacets(struct dpif_backer *backer, int dp_max_idle) * then delete it entirely. */ static void rule_expire(struct rule_dpif *rule) + OVS_REQUIRES(ofproto_mutex) { uint16_t idle_timeout, hard_timeout; - long long int now; - uint8_t reason; + long long int now = time_msec(); + int reason; - if (rule->up.pending) { - /* We'll have to expire it later. */ - return; - } + ovs_assert(!rule->up.pending); - ovs_mutex_lock(&rule->up.timeout_mutex); + /* Has 'rule' expired? */ + ovs_mutex_lock(&rule->up.mutex); hard_timeout = rule->up.hard_timeout; idle_timeout = rule->up.idle_timeout; - ovs_mutex_unlock(&rule->up.timeout_mutex); - - /* Has 'rule' expired? */ - now = time_msec(); if (hard_timeout && now > rule->up.modified + hard_timeout * 1000) { reason = OFPRR_HARD_TIMEOUT; } else if (idle_timeout && now > rule->up.used + idle_timeout * 1000) { reason = OFPRR_IDLE_TIMEOUT; } else { - return; + reason = -1; } + ovs_mutex_unlock(&rule->up.mutex); - COVERAGE_INC(ofproto_dpif_expired); - - /* Get rid of the rule. */ - ofproto_rule_expire(&rule->up, reason); + if (reason >= 0) { + COVERAGE_INC(ofproto_dpif_expired); + ofproto_rule_expire(&rule->up, reason); + } } /* Facets. */ @@ -4360,27 +3761,22 @@ rule_expire(struct rule_dpif *rule) * The facet will initially have no subfacets. The caller should create (at * least) one subfacet with subfacet_create(). */ static struct facet * -facet_create(const struct flow_miss *miss, struct rule_dpif *rule, - struct xlate_out *xout, struct dpif_flow_stats *stats) +facet_create(const struct flow_miss *miss) { struct ofproto_dpif *ofproto = miss->ofproto; struct facet *facet; struct match match; + COVERAGE_INC(facet_create); facet = xzalloc(sizeof *facet); facet->ofproto = miss->ofproto; - facet->packet_count = facet->prev_packet_count = stats->n_packets; - facet->byte_count = facet->prev_byte_count = stats->n_bytes; - facet->tcp_flags = stats->tcp_flags; - facet->used = stats->used; + facet->used = miss->stats.used; facet->flow = miss->flow; facet->learn_rl = time_msec() + 500; list_init(&facet->subfacets); - netflow_flow_init(&facet->nf_flow); - netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); - xlate_out_copy(&facet->xout, xout); + xlate_out_copy(&facet->xout, &miss->xout); match_init(&match, &facet->flow, &facet->xout.wc); cls_rule_init(&facet->cr, &match, OFP_DEFAULT_PRIORITY); @@ -4388,9 +3784,6 @@ facet_create(const struct flow_miss *miss, struct rule_dpif *rule, classifier_insert(&ofproto->facets, &facet->cr); ovs_rwlock_unlock(&ofproto->facets.rwlock); - facet->nf_flow.output_iface = facet->xout.nf_output_iface; - facet->fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY; - return facet; } @@ -4403,24 +3796,49 @@ facet_free(struct facet *facet) } } -/* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on - * 'packet', which arrived on 'in_port'. */ -static bool -execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, - const struct nlattr *odp_actions, size_t actions_len, - struct ofpbuf *packet) +/* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'. + * 'flow' must reflect the data in 'packet'. */ +int +ofproto_dpif_execute_actions(struct ofproto_dpif *ofproto, + const struct flow *flow, + struct rule_dpif *rule, + const struct ofpact *ofpacts, size_t ofpacts_len, + struct ofpbuf *packet) { struct odputil_keybuf keybuf; + struct dpif_flow_stats stats; + struct xlate_out xout; + struct xlate_in xin; + ofp_port_t in_port; struct ofpbuf key; int error; + ovs_assert((rule != NULL) != (ofpacts != NULL)); + + dpif_flow_stats_extract(flow, packet, time_msec(), &stats); + if (rule) { + rule_dpif_credit_stats(rule, &stats); + } + + xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet); + xin.ofpacts = ofpacts; + xin.ofpacts_len = ofpacts_len; + xin.resubmit_stats = &stats; + xlate_actions(&xin, &xout); + ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); - odp_flow_key_from_flow(&key, flow, - ofp_port_to_odp_port(ofproto, flow->in_port.ofp_port)); + in_port = flow->in_port.ofp_port; + if (in_port == OFPP_NONE) { + in_port = OFPP_LOCAL; + } + odp_flow_key_from_flow(&key, flow, ofp_port_to_odp_port(ofproto, in_port)); error = dpif_execute(ofproto->backer->dpif, key.data, key.size, - odp_actions, actions_len, packet); - return !error; + xout.odp_actions.data, xout.odp_actions.size, packet, + (xout.slow & SLOW_ACTION) != 0); + xlate_out_uninit(&xout); + + return error; } /* Remove 'facet' from its ofproto and free up the associated memory: @@ -4435,6 +3853,7 @@ facet_remove(struct facet *facet) { struct subfacet *subfacet, *next_subfacet; + COVERAGE_INC(facet_remove); ovs_assert(!list_is_empty(&facet->subfacets)); /* First uninstall all of the subfacets to get final statistics. */ @@ -4484,54 +3903,6 @@ facet_learn(struct facet *facet) facet_push_stats(facet, true); } -static void -facet_account(struct facet *facet) -{ - const struct nlattr *a; - unsigned int left; - ovs_be16 vlan_tci; - uint64_t n_bytes; - - if (!facet->xout.has_normal || !facet->ofproto->has_bonded_bundles) { - return; - } - n_bytes = facet->byte_count - facet->accounted_bytes; - - /* This loop feeds byte counters to bond_account() for rebalancing to use - * as a basis. We also need to track the actual VLAN on which the packet - * is going to be sent to ensure that it matches the one passed to - * bond_choose_output_slave(). (Otherwise, we will account to the wrong - * hash bucket.) - * - * We use the actions from an arbitrary subfacet because they should all - * be equally valid for our purpose. */ - vlan_tci = facet->flow.vlan_tci; - NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->xout.odp_actions.data, - facet->xout.odp_actions.size) { - const struct ovs_action_push_vlan *vlan; - struct ofport_dpif *port; - - switch (nl_attr_type(a)) { - case OVS_ACTION_ATTR_OUTPUT: - port = get_odp_port(facet->ofproto, nl_attr_get_odp_port(a)); - if (port && port->bundle && port->bundle->bond) { - bond_account(port->bundle->bond, &facet->flow, - vlan_tci_to_vid(vlan_tci), n_bytes); - } - break; - - case OVS_ACTION_ATTR_POP_VLAN: - vlan_tci = htons(0); - break; - - case OVS_ACTION_ATTR_PUSH_VLAN: - vlan = nl_attr_get(a); - vlan_tci = vlan->vlan_tci; - break; - } - } -} - /* Returns true if the only action for 'facet' is to send to the controller. * (We don't report NetFlow expiration messages for such facets because they * are just part of the control logic for the network, not real traffic). */ @@ -4540,16 +3911,24 @@ facet_is_controller_flow(struct facet *facet) { if (facet) { struct ofproto_dpif *ofproto = facet->ofproto; - const struct rule_dpif *rule = rule_dpif_lookup(ofproto, &facet->flow, - NULL); - const struct ofpact *ofpacts = rule->up.ofpacts; - size_t ofpacts_len = rule->up.ofpacts_len; - - if (ofpacts_len > 0 && - ofpacts->type == OFPACT_CONTROLLER && - ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) { - return true; - } + const struct ofpact *ofpacts; + struct rule_actions *actions; + struct rule_dpif *rule; + size_t ofpacts_len; + bool is_controller; + + rule_dpif_lookup(ofproto, &facet->flow, NULL, &rule); + actions = rule_dpif_get_actions(rule); + rule_dpif_unref(rule); + + ofpacts_len = actions->ofpacts_len; + ofpacts = actions->ofpacts; + is_controller = ofpacts_len > 0 + && ofpacts->type == OFPACT_CONTROLLER + && ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len); + rule_actions_unref(actions); + + return is_controller; } return false; } @@ -4570,25 +3949,15 @@ facet_flush_stats(struct facet *facet) } facet_push_stats(facet, false); - if (facet->accounted_bytes < facet->byte_count) { - facet_account(facet); - facet->accounted_bytes = facet->byte_count; - } if (ofproto->netflow && !facet_is_controller_flow(facet)) { - struct ofexpired expired; - expired.flow = facet->flow; - expired.packet_count = facet->packet_count; - expired.byte_count = facet->byte_count; - expired.used = facet->used; - netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); + netflow_expire(ofproto->netflow, &facet->flow); + netflow_flow_clear(ofproto->netflow, &facet->flow); } /* Reset counters to prevent double counting if 'facet' ever gets * reinstalled. */ facet_reset_counters(facet); - - netflow_flow_clear(&facet->nf_flow); facet->tcp_flags = 0; } @@ -4634,19 +4003,14 @@ facet_check_consistency(struct facet *facet) struct xlate_out xout; struct xlate_in xin; - - struct rule_dpif *rule; - bool ok, fail_open; + bool ok; /* Check the datapath actions for consistency. */ - rule = rule_dpif_lookup(facet->ofproto, &facet->flow, NULL); - xlate_in_init(&xin, facet->ofproto, &facet->flow, rule, 0, NULL); + xlate_in_init(&xin, facet->ofproto, &facet->flow, NULL, 0, NULL); xlate_actions(&xin, &xout); - fail_open = rule->up.cr.priority == FAIL_OPEN_PRIORITY; ok = ofpbuf_equal(&facet->xout.odp_actions, &xout.odp_actions) - && facet->xout.slow == xout.slow - && facet->fail_open == fail_open; + && facet->xout.slow == xout.slow; if (!ok && !VLOG_DROP_WARN(&rl)) { struct ds s = DS_EMPTY_INITIALIZER; @@ -4667,10 +4031,6 @@ facet_check_consistency(struct facet *facet) ds_put_format(&s, " slow path incorrect. should be %d", xout.slow); } - if (facet->fail_open != fail_open) { - ds_put_format(&s, " fail open incorrect. should be %s", - fail_open ? "true" : "false"); - } ds_destroy(&s); } xlate_out_uninit(&xout); @@ -4712,7 +4072,7 @@ facet_revalidate(struct facet *facet) error = xlate_receive(ofproto->backer, NULL, subfacet->key, subfacet->key_len, &recv_flow, NULL, - &recv_ofproto, NULL); + &recv_ofproto, NULL, NULL, NULL, NULL); if (error || recv_ofproto != ofproto || facet != facet_find(ofproto, &recv_flow)) { @@ -4722,7 +4082,7 @@ facet_revalidate(struct facet *facet) } flow_wildcards_init_catchall(&wc); - new_rule = rule_dpif_lookup(ofproto, &facet->flow, &wc); + rule_dpif_lookup(ofproto, &facet->flow, &wc, &new_rule); /* Calculate new datapath actions. * @@ -4732,6 +4092,10 @@ facet_revalidate(struct facet *facet) xlate_in_init(&xin, ofproto, &facet->flow, new_rule, 0, NULL); xlate_actions(&xin, &xout); flow_wildcards_or(&xout.wc, &xout.wc, &wc); + /* Make sure non -packet fields are not masked. If not cleared, + * the memcmp() below may fail, causing an otherwise valid facet + * to be removed. */ + flow_wildcards_clear_non_packet_fields(&xout.wc); /* A facet's slow path reason should only change under dramatic * circumstances. Rather than try to update everything, it's simpler to @@ -4745,6 +4109,7 @@ facet_revalidate(struct facet *facet) || memcmp(&facet->xout.wc, &xout.wc, sizeof xout.wc)) { facet_remove(facet); xlate_out_uninit(&xout); + rule_dpif_unref(new_rule); return false; } @@ -4772,11 +4137,13 @@ facet_revalidate(struct facet *facet) facet->xout.has_fin_timeout = xout.has_fin_timeout; facet->xout.nf_output_iface = xout.nf_output_iface; facet->xout.mirrors = xout.mirrors; - facet->nf_flow.output_iface = facet->xout.nf_output_iface; + + ovs_mutex_lock(&new_rule->up.mutex); facet->used = MAX(facet->used, new_rule->up.created); - facet->fail_open = new_rule->up.cr.priority == FAIL_OPEN_PRIORITY; + ovs_mutex_unlock(&new_rule->up.mutex); xlate_out_uninit(&xout); + rule_dpif_unref(new_rule); return true; } @@ -4787,7 +4154,18 @@ facet_reset_counters(struct facet *facet) facet->byte_count = 0; facet->prev_packet_count = 0; facet->prev_byte_count = 0; - facet->accounted_bytes = 0; +} + +static void +flow_push_stats(struct ofproto_dpif *ofproto, struct flow *flow, + struct dpif_flow_stats *stats, bool may_learn) +{ + struct xlate_in xin; + + xlate_in_init(&xin, ofproto, flow, NULL, stats->tcp_flags, NULL); + xin.resubmit_stats = stats; + xin.may_learn = may_learn; + xlate_actions_for_side_effects(&xin); } static void @@ -4805,38 +4183,15 @@ facet_push_stats(struct facet *facet, bool may_learn) stats.tcp_flags = facet->tcp_flags; if (may_learn || stats.n_packets || facet->used > facet->prev_used) { - struct ofproto_dpif *ofproto = facet->ofproto; - struct ofport_dpif *in_port; - struct rule_dpif *rule; - struct xlate_in xin; - facet->prev_packet_count = facet->packet_count; facet->prev_byte_count = facet->byte_count; facet->prev_used = facet->used; - - in_port = get_ofp_port(ofproto, facet->flow.in_port.ofp_port); - if (in_port && in_port->is_tunnel) { - netdev_vport_inc_rx(in_port->up.netdev, &stats); - } - - rule = rule_dpif_lookup(ofproto, &facet->flow, NULL); - rule_credit_stats(rule, &stats); - netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, - facet->used); - netflow_flow_update_flags(&facet->nf_flow, facet->tcp_flags); - mirror_update_stats(ofproto->mbridge, facet->xout.mirrors, - stats.n_packets, stats.n_bytes); - - xlate_in_init(&xin, ofproto, &facet->flow, rule, stats.tcp_flags, - NULL); - xin.resubmit_stats = &stats; - xin.may_learn = may_learn; - xlate_actions_for_side_effects(&xin); + flow_push_stats(facet->ofproto, &facet->flow, &stats, may_learn); } } static void -push_all_stats__(bool run_fast) +push_all_stats(void) { static long long int rl = LLONG_MIN; struct ofproto_dpif *ofproto; @@ -4853,9 +4208,6 @@ push_all_stats__(bool run_fast) cls_cursor_init(&cursor, &ofproto->facets, NULL); CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { facet_push_stats(facet, false); - if (run_fast) { - run_fast_rl(); - } } ovs_rwlock_unlock(&ofproto->facets.rwlock); } @@ -4863,21 +4215,51 @@ push_all_stats__(bool run_fast) rl = time_msec() + 100; } -static void -push_all_stats(void) -{ - push_all_stats__(true); -} - void -rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats) +rule_dpif_credit_stats(struct rule_dpif *rule, + const struct dpif_flow_stats *stats) { ovs_mutex_lock(&rule->stats_mutex); rule->packet_count += stats->n_packets; rule->byte_count += stats->n_bytes; - ofproto_rule_update_used(&rule->up, stats->used); + rule->up.used = MAX(rule->up.used, stats->used); ovs_mutex_unlock(&rule->stats_mutex); } + +bool +rule_dpif_is_fail_open(const struct rule_dpif *rule) +{ + return is_fail_open_rule(&rule->up); +} + +bool +rule_dpif_is_table_miss(const struct rule_dpif *rule) +{ + return rule_is_table_miss(&rule->up); +} + +ovs_be64 +rule_dpif_get_flow_cookie(const struct rule_dpif *rule) + OVS_REQUIRES(rule->up.mutex) +{ + return rule->up.flow_cookie; +} + +void +rule_dpif_reduce_timeouts(struct rule_dpif *rule, uint16_t idle_timeout, + uint16_t hard_timeout) +{ + ofproto_rule_reduce_timeouts(&rule->up, idle_timeout, hard_timeout); +} + +/* Returns 'rule''s actions. The caller owns a reference on the returned + * actions and must eventually release it (with rule_actions_unref()) to avoid + * a memory leak. */ +struct rule_actions * +rule_dpif_get_actions(const struct rule_dpif *rule) +{ + return rule_get_actions(&rule->up); +} /* Subfacets. */ @@ -4898,54 +4280,34 @@ subfacet_find(struct dpif_backer *backer, const struct nlattr *key, return NULL; } -/* Searches 'facet' (within 'ofproto') for a subfacet with the specified - * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the - * existing subfacet if there is one, otherwise creates and returns a - * new subfacet. */ +/* Creates and returns a new subfacet within 'facet' for the flow in 'miss'. + * 'key_hash' must be a hash over miss->key. The caller must have already + * ensured that no subfacet subfacet already exists. */ static struct subfacet * -subfacet_create(struct facet *facet, struct flow_miss *miss, - long long int now) +subfacet_create(struct facet *facet, struct flow_miss *miss, uint32_t key_hash) { struct dpif_backer *backer = miss->ofproto->backer; - enum odp_key_fitness key_fitness = miss->key_fitness; const struct nlattr *key = miss->key; size_t key_len = miss->key_len; - uint32_t key_hash; struct subfacet *subfacet; - key_hash = odp_flow_key_hash(key, key_len); - - if (list_is_empty(&facet->subfacets)) { - subfacet = &facet->one_subfacet; - } else { - subfacet = subfacet_find(backer, key, key_len, key_hash); - if (subfacet) { - if (subfacet->facet == facet) { - return subfacet; - } - - /* This shouldn't happen. */ - VLOG_ERR_RL(&rl, "subfacet with wrong facet"); - subfacet_destroy(subfacet); - } - - subfacet = xmalloc(sizeof *subfacet); - } + subfacet = (list_is_empty(&facet->subfacets) + ? &facet->one_subfacet + : xmalloc(sizeof *subfacet)); + COVERAGE_INC(subfacet_create); hmap_insert(&backer->subfacets, &subfacet->hmap_node, key_hash); list_push_back(&facet->subfacets, &subfacet->list_node); subfacet->facet = facet; - subfacet->key_fitness = key_fitness; subfacet->key = xmemdup(key, key_len); subfacet->key_len = key_len; - subfacet->used = now; - subfacet->created = now; + subfacet->used = miss->stats.used; + subfacet->created = subfacet->used; subfacet->dp_packet_count = 0; subfacet->dp_byte_count = 0; subfacet->path = SF_NOT_INSTALLED; subfacet->backer = backer; - backer->subfacet_add_count++; return subfacet; } @@ -4955,11 +4317,8 @@ static void subfacet_destroy__(struct subfacet *subfacet) { struct facet *facet = subfacet->facet; - struct ofproto_dpif *ofproto = facet->ofproto; - - /* Update ofproto stats before uninstall the subfacet. */ - ofproto->backer->subfacet_del_count++; + COVERAGE_INC(subfacet_destroy); subfacet_uninstall(subfacet); hmap_remove(&subfacet->backer->subfacets, &subfacet->hmap_node); list_remove(&subfacet->list_node); @@ -5006,7 +4365,6 @@ subfacet_destroy_batch(struct dpif_backer *backer, subfacet_reset_dp_stats(subfacets[i], &stats[i]); subfacets[i]->path = SF_NOT_INSTALLED; subfacet_destroy(subfacets[i]); - run_fast_rl(); } } @@ -5132,16 +4490,14 @@ subfacet_update_stats(struct subfacet *subfacet, /* Lookup 'flow' in 'ofproto''s classifier. If 'wc' is non-null, sets * the fields that were relevant as part of the lookup. */ -static struct rule_dpif * +void rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow, - struct flow_wildcards *wc) + struct flow_wildcards *wc, struct rule_dpif **rule) { struct ofport_dpif *port; - struct rule_dpif *rule; - rule = rule_dpif_lookup_in_table(ofproto, flow, wc, 0); - if (rule) { - return rule; + if (rule_dpif_lookup_in_table(ofproto, flow, wc, 0, rule)) { + return; } port = get_ofp_port(ofproto, flow->in_port.ofp_port); if (!port) { @@ -5149,21 +4505,22 @@ rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow, flow->in_port.ofp_port); } - return choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule, - ofproto->no_packet_in_rule); + choose_miss_rule(port ? port->up.pp.config : 0, ofproto->miss_rule, + ofproto->no_packet_in_rule, rule); } -struct rule_dpif * +bool rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, const struct flow *flow, struct flow_wildcards *wc, - uint8_t table_id) + uint8_t table_id, struct rule_dpif **rule) { - struct cls_rule *cls_rule; + const struct cls_rule *cls_rule; struct classifier *cls; bool frag; + *rule = NULL; if (table_id >= N_TABLES) { - return NULL; + return false; } if (wc) { @@ -5172,51 +4529,68 @@ rule_dpif_lookup_in_table(struct ofproto_dpif *ofproto, } cls = &ofproto->up.tables[table_id].cls; + ovs_rwlock_rdlock(&cls->rwlock); frag = (flow->nw_frag & FLOW_NW_FRAG_ANY) != 0; if (frag && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { /* We must pretend that transport ports are unavailable. */ struct flow ofpc_normal_flow = *flow; ofpc_normal_flow.tp_src = htons(0); ofpc_normal_flow.tp_dst = htons(0); - ovs_rwlock_rdlock(&cls->rwlock); cls_rule = classifier_lookup(cls, &ofpc_normal_flow, wc); - ovs_rwlock_unlock(&cls->rwlock); } else if (frag && ofproto->up.frag_handling == OFPC_FRAG_DROP) { cls_rule = &ofproto->drop_frags_rule->up.cr; - if (wc) { - flow_wildcards_init_exact(wc); - } + /* Frag mask in wc already set above. */ } else { - ovs_rwlock_rdlock(&cls->rwlock); cls_rule = classifier_lookup(cls, flow, wc); - ovs_rwlock_unlock(&cls->rwlock); } - return rule_dpif_cast(rule_from_cls_rule(cls_rule)); + + *rule = rule_dpif_cast(rule_from_cls_rule(cls_rule)); + rule_dpif_ref(*rule); + ovs_rwlock_unlock(&cls->rwlock); + + return *rule != NULL; } /* Given a port configuration (specified as zero if there's no port), chooses * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a * flow table miss. */ -struct rule_dpif * +void choose_miss_rule(enum ofputil_port_config config, struct rule_dpif *miss_rule, - struct rule_dpif *no_packet_in_rule) + struct rule_dpif *no_packet_in_rule, struct rule_dpif **rule) +{ + *rule = config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule; + rule_dpif_ref(*rule); +} + +void +rule_dpif_ref(struct rule_dpif *rule) +{ + if (rule) { + ofproto_rule_ref(&rule->up); + } +} + +void +rule_dpif_unref(struct rule_dpif *rule) { - return config & OFPUTIL_PC_NO_PACKET_IN ? no_packet_in_rule : miss_rule; + if (rule) { + ofproto_rule_unref(&rule->up); + } } static void complete_operation(struct rule_dpif *rule) + OVS_REQUIRES(ofproto_mutex) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); ofproto->backer->need_revalidate = REV_FLOW_TABLE; - if (clogged) { - struct dpif_completion *c = xmalloc(sizeof *c); - c->op = rule->up.pending; - list_push_back(&ofproto->completions, &c->list_node); - } else { - ofoperation_complete(rule->up.pending, 0); - } + ofoperation_complete(rule->up.pending, 0); +} + +static struct rule_dpif *rule_dpif_cast(const struct rule *rule) +{ + return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL; } static struct rule * @@ -5236,134 +4610,233 @@ rule_dealloc(struct rule *rule_) static enum ofperr rule_construct(struct rule *rule_) { - struct rule_dpif *rule = rule_dpif_cast(rule_); - ovs_mutex_init(&rule->stats_mutex, PTHREAD_MUTEX_NORMAL); - ovs_mutex_lock(&rule->stats_mutex); - rule->packet_count = 0; - rule->byte_count = 0; - ovs_mutex_unlock(&rule->stats_mutex); - complete_operation(rule); + struct rule_dpif *rule = rule_dpif_cast(rule_); + ovs_mutex_init(&rule->stats_mutex); + ovs_mutex_lock(&rule->stats_mutex); + rule->packet_count = 0; + rule->byte_count = 0; + ovs_mutex_unlock(&rule->stats_mutex); + return 0; +} + +static void +rule_insert(struct rule *rule_) + OVS_REQUIRES(ofproto_mutex) +{ + struct rule_dpif *rule = rule_dpif_cast(rule_); + complete_operation(rule); +} + +static void +rule_delete(struct rule *rule_) + OVS_REQUIRES(ofproto_mutex) +{ + struct rule_dpif *rule = rule_dpif_cast(rule_); + complete_operation(rule); +} + +static void +rule_destruct(struct rule *rule_) +{ + struct rule_dpif *rule = rule_dpif_cast(rule_); + ovs_mutex_destroy(&rule->stats_mutex); +} + +static void +rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) +{ + struct rule_dpif *rule = rule_dpif_cast(rule_); + + push_all_stats(); + + /* Start from historical data for 'rule' itself that are no longer tracked + * in facets. This counts, for example, facets that have expired. */ + ovs_mutex_lock(&rule->stats_mutex); + *packets = rule->packet_count; + *bytes = rule->byte_count; + ovs_mutex_unlock(&rule->stats_mutex); +} + +static void +rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow, + struct ofpbuf *packet) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); + + ofproto_dpif_execute_actions(ofproto, flow, rule, NULL, 0, packet); +} + +static enum ofperr +rule_execute(struct rule *rule, const struct flow *flow, + struct ofpbuf *packet) +{ + rule_dpif_execute(rule_dpif_cast(rule), flow, packet); + ofpbuf_delete(packet); + return 0; +} + +static void +rule_modify_actions(struct rule *rule_, bool reset_counters) + OVS_REQUIRES(ofproto_mutex) +{ + struct rule_dpif *rule = rule_dpif_cast(rule_); + + if (reset_counters) { + ovs_mutex_lock(&rule->stats_mutex); + rule->packet_count = 0; + rule->byte_count = 0; + ovs_mutex_unlock(&rule->stats_mutex); + } + + complete_operation(rule); +} + +static struct group_dpif *group_dpif_cast(const struct ofgroup *group) +{ + return group ? CONTAINER_OF(group, struct group_dpif, up) : NULL; +} + +static struct ofgroup * +group_alloc(void) +{ + struct group_dpif *group = xzalloc(sizeof *group); + return &group->up; +} + +static void +group_dealloc(struct ofgroup *group_) +{ + struct group_dpif *group = group_dpif_cast(group_); + free(group); +} + +static void +group_construct_stats(struct group_dpif *group) + OVS_REQUIRES(group->stats_mutex) +{ + group->packet_count = 0; + group->byte_count = 0; + if (!group->bucket_stats) { + group->bucket_stats = xcalloc(group->up.n_buckets, + sizeof *group->bucket_stats); + } else { + memset(group->bucket_stats, 0, group->up.n_buckets * + sizeof *group->bucket_stats); + } +} + +static enum ofperr +group_construct(struct ofgroup *group_) +{ + struct group_dpif *group = group_dpif_cast(group_); + ovs_mutex_init(&group->stats_mutex); + ovs_mutex_lock(&group->stats_mutex); + group_construct_stats(group); + ovs_mutex_unlock(&group->stats_mutex); return 0; } static void -rule_destruct(struct rule *rule_) +group_destruct__(struct group_dpif *group) + OVS_REQUIRES(group->stats_mutex) { - struct rule_dpif *rule = rule_dpif_cast(rule_); - complete_operation(rule); - ovs_mutex_destroy(&rule->stats_mutex); + free(group->bucket_stats); + group->bucket_stats = NULL; } static void -rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) +group_destruct(struct ofgroup *group_) { - struct rule_dpif *rule = rule_dpif_cast(rule_); + struct group_dpif *group = group_dpif_cast(group_); + ovs_mutex_lock(&group->stats_mutex); + group_destruct__(group); + ovs_mutex_unlock(&group->stats_mutex); + ovs_mutex_destroy(&group->stats_mutex); +} - /* push_all_stats() can handle flow misses which, when using the learn - * action, can cause rules to be added and deleted. This can corrupt our - * caller's datastructures which assume that rule_get_stats() doesn't have - * an impact on the flow table. To be safe, we disable miss handling. */ - push_all_stats__(false); +static enum ofperr +group_modify(struct ofgroup *group_, struct ofgroup *victim_) +{ + struct group_dpif *group = group_dpif_cast(group_); + struct group_dpif *victim = group_dpif_cast(victim_); - /* Start from historical data for 'rule' itself that are no longer tracked - * in facets. This counts, for example, facets that have expired. */ - ovs_mutex_lock(&rule->stats_mutex); - *packets = rule->packet_count; - *bytes = rule->byte_count; - ovs_mutex_unlock(&rule->stats_mutex); + ovs_mutex_lock(&group->stats_mutex); + if (victim->up.n_buckets < group->up.n_buckets) { + group_destruct__(group); + } + group_construct_stats(group); + ovs_mutex_unlock(&group->stats_mutex); + + return 0; } -static void -rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow, - struct ofpbuf *packet) +static enum ofperr +group_get_stats(const struct ofgroup *group_, struct ofputil_group_stats *ogs) { - struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); - struct dpif_flow_stats stats; - struct xlate_out xout; - struct xlate_in xin; + struct group_dpif *group = group_dpif_cast(group_); - dpif_flow_stats_extract(flow, packet, time_msec(), &stats); - rule_credit_stats(rule, &stats); + /* Start from historical data for 'group' itself that are no longer tracked + * in facets. This counts, for example, facets that have expired. */ + ovs_mutex_lock(&group->stats_mutex); + ogs->packet_count = group->packet_count; + ogs->byte_count = group->byte_count; + memcpy(ogs->bucket_stats, group->bucket_stats, + group->up.n_buckets * sizeof *group->bucket_stats); + ovs_mutex_unlock(&group->stats_mutex); - xlate_in_init(&xin, ofproto, flow, rule, stats.tcp_flags, packet); - xin.resubmit_stats = &stats; - xlate_actions(&xin, &xout); + return 0; +} + +bool +group_dpif_lookup(struct ofproto_dpif *ofproto, uint32_t group_id, + struct group_dpif **group) + OVS_TRY_RDLOCK(true, (*group)->up.rwlock) +{ + struct ofgroup *ofgroup; + bool found; - execute_odp_actions(ofproto, flow, xout.odp_actions.data, - xout.odp_actions.size, packet); + *group = NULL; + found = ofproto_group_lookup(&ofproto->up, group_id, &ofgroup); + *group = found ? group_dpif_cast(ofgroup) : NULL; - xlate_out_uninit(&xout); + return found; } -static enum ofperr -rule_execute(struct rule *rule, const struct flow *flow, - struct ofpbuf *packet) +void +group_dpif_release(struct group_dpif *group) + OVS_RELEASES(group->up.rwlock) { - rule_dpif_execute(rule_dpif_cast(rule), flow, packet); - ofpbuf_delete(packet); - return 0; + ofproto_group_release(&group->up); } -static void -rule_modify_actions(struct rule *rule_) +void +group_dpif_get_buckets(const struct group_dpif *group, + const struct list **buckets) { - struct rule_dpif *rule = rule_dpif_cast(rule_); + *buckets = &group->up.buckets; +} - complete_operation(rule); +enum ofp11_group_type +group_dpif_get_type(const struct group_dpif *group) +{ + return group->up.type; } /* Sends 'packet' out 'ofport'. * May modify 'packet'. * Returns 0 if successful, otherwise a positive errno value. */ -static int -send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) +int +ofproto_dpif_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); - uint64_t odp_actions_stub[1024 / 8]; - struct ofpbuf key, odp_actions; - struct dpif_flow_stats stats; - struct odputil_keybuf keybuf; - struct ofpact_output output; - struct xlate_out xout; - struct xlate_in xin; - struct flow flow; - union flow_in_port in_port_; int error; - ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); - ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); - - /* Use OFPP_NONE as the in_port to avoid special packet processing. */ - in_port_.ofp_port = OFPP_NONE; - flow_extract(packet, 0, 0, NULL, &in_port_, &flow); - odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(ofproto, - OFPP_LOCAL)); - dpif_flow_stats_extract(&flow, packet, time_msec(), &stats); - - ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output); - output.port = ofport->up.ofp_port; - output.max_len = 0; - - xlate_in_init(&xin, ofproto, &flow, NULL, 0, packet); - xin.ofpacts_len = sizeof output; - xin.ofpacts = &output.ofpact; - xin.resubmit_stats = &stats; - xlate_actions(&xin, &xout); - - error = dpif_execute(ofproto->backer->dpif, - key.data, key.size, - xout.odp_actions.data, xout.odp_actions.size, - packet); - xlate_out_uninit(&xout); - - if (error) { - VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)", - ofproto->up.name, netdev_get_name(ofport->up.netdev), - ovs_strerror(error)); - } + error = xlate_send_packet(ofport, packet); + ovs_mutex_lock(&ofproto->stats_mutex); ofproto->stats.tx_packets++; ofproto->stats.tx_bytes += packet->size; + ovs_mutex_unlock(&ofproto->stats_mutex); return error; } @@ -5426,30 +4899,9 @@ packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, const struct ofpact *ofpacts, size_t ofpacts_len) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - struct odputil_keybuf keybuf; - struct dpif_flow_stats stats; - struct xlate_out xout; - struct xlate_in xin; - struct ofpbuf key; - - - ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); - odp_flow_key_from_flow(&key, flow, - ofp_port_to_odp_port(ofproto, - flow->in_port.ofp_port)); - - dpif_flow_stats_extract(flow, packet, time_msec(), &stats); - - xlate_in_init(&xin, ofproto, flow, NULL, stats.tcp_flags, packet); - xin.resubmit_stats = &stats; - xin.ofpacts_len = ofpacts_len; - xin.ofpacts = ofpacts; - - xlate_actions(&xin, &xout); - dpif_execute(ofproto->backer->dpif, key.data, key.size, - xout.odp_actions.data, xout.odp_actions.size, packet); - xlate_out_uninit(&xout); + ofproto_dpif_execute_actions(ofproto, flow, NULL, ofpacts, + ofpacts_len, packet); return 0; } @@ -5469,7 +4921,7 @@ set_netflow(struct ofproto *ofproto_, return netflow_set_options(ofproto->netflow, netflow_options); } else if (ofproto->netflow) { ofproto->backer->need_revalidate = REV_RECONFIGURE; - netflow_destroy(ofproto->netflow); + netflow_unref(ofproto->netflow); ofproto->netflow = NULL; } @@ -5484,46 +4936,6 @@ get_netflow_ids(const struct ofproto *ofproto_, dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id); } - -static void -send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) -{ - if (!facet_is_controller_flow(facet) && - netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { - struct subfacet *subfacet; - struct ofexpired expired; - - LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { - if (subfacet->path == SF_FAST_PATH) { - struct dpif_flow_stats stats; - - subfacet_install(subfacet, &facet->xout.odp_actions, - &stats); - subfacet_update_stats(subfacet, &stats); - } - } - - expired.flow = facet->flow; - expired.packet_count = facet->packet_count; - expired.byte_count = facet->byte_count; - expired.used = facet->used; - netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); - } -} - -static void -send_netflow_active_timeouts(struct ofproto_dpif *ofproto) -{ - struct cls_cursor cursor; - struct facet *facet; - - ovs_rwlock_rdlock(&ofproto->facets.rwlock); - cls_cursor_init(&cursor, &ofproto->facets, NULL); - CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { - send_active_timeout(ofproto, facet); - } - ovs_rwlock_unlock(&ofproto->facets.rwlock); -} static struct ofproto_dpif * ofproto_dpif_lookup(const char *name) @@ -5613,21 +5025,32 @@ struct trace_ctx { static void trace_format_rule(struct ds *result, int level, const struct rule_dpif *rule) { + struct rule_actions *actions; + ovs_be64 cookie; + ds_put_char_multiple(result, '\t', level); if (!rule) { ds_put_cstr(result, "No match\n"); return; } + ovs_mutex_lock(&rule->up.mutex); + cookie = rule->up.flow_cookie; + ovs_mutex_unlock(&rule->up.mutex); + ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", - rule ? rule->up.table_id : 0, ntohll(rule->up.flow_cookie)); + rule ? rule->up.table_id : 0, ntohll(cookie)); cls_rule_format(&rule->up.cr, result); ds_put_char(result, '\n'); + actions = rule_dpif_get_actions(rule); + ds_put_char_multiple(result, '\t', level); - ds_put_cstr(result, "OpenFlow "); - ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result); + ds_put_cstr(result, "OpenFlow actions="); + ofpacts_format(actions->ofpacts, actions->ofpacts_len, result); ds_put_char(result, '\n'); + + rule_actions_unref(actions); } static void @@ -5654,7 +5077,7 @@ trace_format_regs(struct ds *result, int level, const char *title, ds_put_char_multiple(result, '\t', level); ds_put_format(result, "%s:", title); for (i = 0; i < FLOW_N_REGS; i++) { - ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); + ds_put_format(result, " reg%"PRIuSIZE"=0x%"PRIx32, i, trace->flow.regs[i]); } ds_put_char(result, '\n'); } @@ -5695,21 +5118,26 @@ trace_report(struct xlate_in *xin, const char *s, int recurse) ds_put_char(result, '\n'); } -static void -ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], - void *aux OVS_UNUSED) +/* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following + * forms are supported: + * + * - [dpname] odp_flow [-generate | packet] + * - bridge br_flow [-generate | packet] + * + * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure + * returns a nonnull error message. */ +static const char * +parse_flow_and_packet(int argc, const char *argv[], + struct ofproto_dpif **ofprotop, struct flow *flow, + struct ofpbuf **packetp) { - const struct dpif_backer *backer; - struct ofproto_dpif *ofproto; - struct ofpbuf odp_key, odp_mask; + const struct dpif_backer *backer = NULL; + const char *error = NULL; + struct simap port_names = SIMAP_INITIALIZER(&port_names); struct ofpbuf *packet; - struct ds result; - struct flow flow; - char *s; + struct ofpbuf odp_key; + struct ofpbuf odp_mask; - packet = NULL; - backer = NULL; - ds_init(&result); ofpbuf_init(&odp_key, 0); ofpbuf_init(&odp_mask, 0); @@ -5718,130 +5146,264 @@ ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], packet = ofpbuf_new(0); argc--; } else { - const char *error = eth_from_hex(argv[argc - 1], &packet); + error = eth_from_hex(argv[argc - 1], &packet); if (!error) { argc--; } else if (argc == 4) { /* The 3-argument form must end in "-generate' or a hex string. */ - unixctl_command_reply_error(conn, error); goto exit; } } + /* odp_flow can have its in_port specified as a name instead of port no. + * We do not yet know whether a given flow is a odp_flow or a br_flow. + * But, to know whether a flow is odp_flow through odp_flow_from_string(), + * we need to create a simap of name to port no. */ + if (argc == 3) { + const char *dp_type; + if (!strncmp(argv[1], "ovs-", 4)) { + dp_type = argv[1] + 4; + } else { + dp_type = argv[1]; + } + backer = shash_find_data(&all_dpif_backers, dp_type); + } else if (argc == 2) { + struct shash_node *node; + if (shash_count(&all_dpif_backers) == 1) { + node = shash_first(&all_dpif_backers); + backer = node->data; + } + } else { + error = "Syntax error"; + goto exit; + } + if (backer && backer->dpif) { + struct dpif_port dpif_port; + struct dpif_port_dump port_dump; + DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, backer->dpif) { + simap_put(&port_names, dpif_port.name, + odp_to_u32(dpif_port.port_no)); + } + } + /* Parse the flow and determine whether a datapath or * bridge is specified. If function odp_flow_key_from_string() * returns 0, the flow is a odp_flow. If function * parse_ofp_exact_flow() returns 0, the flow is a br_flow. */ - if (!odp_flow_from_string(argv[argc - 1], NULL, &odp_key, &odp_mask)) { - /* If the odp_flow is the second argument, - * the datapath name is the first argument. */ - if (argc == 3) { - const char *dp_type; - if (!strncmp(argv[1], "ovs-", 4)) { - dp_type = argv[1] + 4; - } else { - dp_type = argv[1]; - } - backer = shash_find_data(&all_dpif_backers, dp_type); - if (!backer) { - unixctl_command_reply_error(conn, "Cannot find datapath " - "of this name"); - goto exit; - } - } else { - /* No datapath name specified, so there should be only one - * datapath. */ - struct shash_node *node; - if (shash_count(&all_dpif_backers) != 1) { - unixctl_command_reply_error(conn, "Must specify datapath " - "name, there is more than one type of datapath"); - goto exit; - } - node = shash_first(&all_dpif_backers); - backer = node->data; + if (!odp_flow_from_string(argv[argc - 1], &port_names, + &odp_key, &odp_mask)) { + if (!backer) { + error = "Cannot find the datapath"; + goto exit; } - if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, &flow, - NULL, &ofproto, NULL)) { - unixctl_command_reply_error(conn, "Invalid datapath flow"); + if (xlate_receive(backer, NULL, odp_key.data, odp_key.size, flow, + NULL, ofprotop, NULL, NULL, NULL, NULL)) { + error = "Invalid datapath flow"; goto exit; } - ds_put_format(&result, "Bridge: %s\n", ofproto->up.name); - } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) { + } else if (!parse_ofp_exact_flow(flow, NULL, argv[argc - 1], NULL)) { if (argc != 3) { - unixctl_command_reply_error(conn, "Must specify bridge name"); + error = "Must specify bridge name"; goto exit; } - ofproto = ofproto_dpif_lookup(argv[1]); - if (!ofproto) { - unixctl_command_reply_error(conn, "Unknown bridge name"); + *ofprotop = ofproto_dpif_lookup(argv[1]); + if (!*ofprotop) { + error = "Unknown bridge name"; goto exit; } } else { - unixctl_command_reply_error(conn, "Bad flow syntax"); + error = "Bad flow syntax"; goto exit; } /* Generate a packet, if requested. */ if (packet) { if (!packet->size) { - flow_compose(packet, &flow); + flow_compose(packet, flow); } else { - union flow_in_port in_port_; - - in_port_ = flow.in_port; - ds_put_cstr(&result, "Packet: "); - s = ofp_packet_to_string(packet->data, packet->size); - ds_put_cstr(&result, s); - free(s); + union flow_in_port in_port = flow->in_port; /* Use the metadata from the flow and the packet argument * to reconstruct the flow. */ - flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL, - &in_port_, &flow); + flow_extract(packet, flow->skb_priority, flow->pkt_mark, NULL, + &in_port, flow); } } - ofproto_trace(ofproto, &flow, packet, &result); + error = NULL; + +exit: + if (error) { + ofpbuf_delete(packet); + packet = NULL; + } + *packetp = packet; + ofpbuf_uninit(&odp_key); + ofpbuf_uninit(&odp_mask); + simap_destroy(&port_names); + return error; +} + +static void +ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], + void *aux OVS_UNUSED) +{ + struct ofproto_dpif *ofproto; + struct ofpbuf *packet; + const char *error; + struct flow flow; + + error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet); + if (!error) { + struct ds result; + + ds_init(&result); + ofproto_trace(ofproto, &flow, packet, NULL, 0, &result); + unixctl_command_reply(conn, ds_cstr(&result)); + ds_destroy(&result); + ofpbuf_delete(packet); + } else { + unixctl_command_reply_error(conn, error); + } +} + +static void +ofproto_unixctl_trace_actions(struct unixctl_conn *conn, int argc, + const char *argv[], void *aux OVS_UNUSED) +{ + enum ofputil_protocol usable_protocols; + struct ofproto_dpif *ofproto; + bool enforce_consistency; + struct ofpbuf ofpacts; + struct ofpbuf *packet; + struct ds result; + struct flow flow; + uint16_t in_port; + + /* Three kinds of error return values! */ + enum ofperr retval; + const char *error; + char *rw_error; + + packet = NULL; + ds_init(&result); + ofpbuf_init(&ofpacts, 0); + + /* Parse actions. */ + rw_error = parse_ofpacts(argv[--argc], &ofpacts, &usable_protocols); + if (rw_error) { + unixctl_command_reply_error(conn, rw_error); + free(rw_error); + goto exit; + } + + /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of + * consistency between the flow and the actions. With -consistent, we + * enforce consistency even for a flow supported in OpenFlow 1.0. */ + if (!strcmp(argv[1], "-consistent")) { + enforce_consistency = true; + argv++; + argc--; + } else { + enforce_consistency = false; + } + + error = parse_flow_and_packet(argc, argv, &ofproto, &flow, &packet); + if (error) { + unixctl_command_reply_error(conn, error); + goto exit; + } + + /* Do the same checks as handle_packet_out() in ofproto.c. + * + * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't + * strictly correct because these actions aren't in any table, but it's OK + * because it 'table_id' is used only to check goto_table instructions, but + * packet-outs take a list of actions and therefore it can't include + * instructions. + * + * We skip the "meter" check here because meter is an instruction, not an + * action, and thus cannot appear in ofpacts. */ + in_port = ofp_to_u16(flow.in_port.ofp_port); + if (in_port >= ofproto->up.max_ports && in_port < ofp_to_u16(OFPP_MAX)) { + unixctl_command_reply_error(conn, "invalid in_port"); + goto exit; + } + if (enforce_consistency) { + retval = ofpacts_check_consistency(ofpacts.data, ofpacts.size, &flow, + u16_to_ofp(ofproto->up.max_ports), + 0, 0, usable_protocols); + } else { + retval = ofpacts_check(ofpacts.data, ofpacts.size, &flow, + u16_to_ofp(ofproto->up.max_ports), 0, 0, + &usable_protocols); + } + + if (retval) { + ds_clear(&result); + ds_put_format(&result, "Bad actions: %s", ofperr_to_string(retval)); + unixctl_command_reply_error(conn, ds_cstr(&result)); + goto exit; + } + + ofproto_trace(ofproto, &flow, packet, ofpacts.data, ofpacts.size, &result); unixctl_command_reply(conn, ds_cstr(&result)); exit: ds_destroy(&result); ofpbuf_delete(packet); - ofpbuf_uninit(&odp_key); - ofpbuf_uninit(&odp_mask); + ofpbuf_uninit(&ofpacts); } +/* Implements a "trace" through 'ofproto''s flow table, appending a textual + * description of the results to 'ds'. + * + * The trace follows a packet with the specified 'flow' through the flow + * table. 'packet' may be nonnull to trace an actual packet, with consequent + * side effects (if it is nonnull then its flow must be 'flow'). + * + * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to + * trace, otherwise the actions are determined by a flow table lookup. */ static void ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, - const struct ofpbuf *packet, struct ds *ds) + const struct ofpbuf *packet, + const struct ofpact ofpacts[], size_t ofpacts_len, + struct ds *ds) { struct rule_dpif *rule; + struct flow_wildcards wc; + ds_put_format(ds, "Bridge: %s\n", ofproto->up.name); ds_put_cstr(ds, "Flow: "); flow_format(ds, flow); ds_put_char(ds, '\n'); - rule = rule_dpif_lookup(ofproto, flow, NULL); + flow_wildcards_init_catchall(&wc); + if (ofpacts) { + rule = NULL; + } else { + rule_dpif_lookup(ofproto, flow, &wc, &rule); - trace_format_rule(ds, 0, rule); - if (rule == ofproto->miss_rule) { - ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n"); - } else if (rule == ofproto->no_packet_in_rule) { - ds_put_cstr(ds, "\nNo match, packets dropped because " - "OFPPC_NO_PACKET_IN is set on in_port.\n"); - } else if (rule == ofproto->drop_frags_rule) { - ds_put_cstr(ds, "\nPackets dropped because they are IP fragments " - "and the fragment handling mode is \"drop\".\n"); + trace_format_rule(ds, 0, rule); + if (rule == ofproto->miss_rule) { + ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n"); + } else if (rule == ofproto->no_packet_in_rule) { + ds_put_cstr(ds, "\nNo match, packets dropped because " + "OFPPC_NO_PACKET_IN is set on in_port.\n"); + } else if (rule == ofproto->drop_frags_rule) { + ds_put_cstr(ds, "\nPackets dropped because they are IP fragments " + "and the fragment handling mode is \"drop\".\n"); + } } - if (rule) { + if (rule || ofpacts) { uint64_t odp_actions_stub[1024 / 8]; struct ofpbuf odp_actions; struct trace_ctx trace; struct match match; - uint8_t tcp_flags; + uint16_t tcp_flags; tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0; trace.result = ds; @@ -5849,10 +5411,15 @@ ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); xlate_in_init(&trace.xin, ofproto, flow, rule, tcp_flags, packet); + if (ofpacts) { + trace.xin.ofpacts = ofpacts; + trace.xin.ofpacts_len = ofpacts_len; + } trace.xin.resubmit_hook = trace_resubmit; trace.xin.report_hook = trace_report; xlate_actions(&trace.xin, &trace.xout); + flow_wildcards_or(&trace.xout.wc, &trace.xout.wc, &wc); ds_put_char(ds, '\n'); trace_format_flow(ds, 0, "Final flow", &trace); @@ -5867,48 +5434,26 @@ ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, trace.xout.odp_actions.size); if (trace.xout.slow) { + enum slow_path_reason slow; + ds_put_cstr(ds, "\nThis flow is handled by the userspace " "slow path because it:"); - switch (trace.xout.slow) { - case SLOW_CFM: - ds_put_cstr(ds, "\n\t- Consists of CFM packets."); - break; - case SLOW_LACP: - ds_put_cstr(ds, "\n\t- Consists of LACP packets."); - break; - case SLOW_STP: - ds_put_cstr(ds, "\n\t- Consists of STP packets."); - break; - case SLOW_BFD: - ds_put_cstr(ds, "\n\t- Consists of BFD packets."); - break; - case SLOW_CONTROLLER: - ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages " - "to the OpenFlow controller."); - break; - case __SLOW_MAX: - NOT_REACHED(); + + slow = trace.xout.slow; + while (slow) { + enum slow_path_reason bit = rightmost_1bit(slow); + + ds_put_format(ds, "\n\t- %s.", + slow_path_reason_to_explanation(bit)); + + slow &= ~bit; } } xlate_out_uninit(&trace.xout); } -} - -static void -ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, - const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) -{ - clogged = true; - unixctl_command_reply(conn, NULL); -} -static void -ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED, - const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED) -{ - clogged = false; - unixctl_command_reply(conn, NULL); + rule_dpif_unref(rule); } /* Runs a self-check of flow translations in 'ofproto'. Appends a message to @@ -6006,14 +5551,6 @@ ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED, ds_destroy(&ds); } -static void -show_dp_rates(struct ds *ds, const char *heading, - const struct avg_subfacet_rates *rates) -{ - ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n", - heading, rates->add_rate, rates->del_rate); -} - static void dpif_show_backer(const struct dpif_backer *backer, struct ds *ds) { @@ -6021,7 +5558,6 @@ dpif_show_backer(const struct dpif_backer *backer, struct ds *ds) struct ofproto_dpif *ofproto; struct shash ofproto_shash; uint64_t n_hit, n_missed; - long long int minutes; size_t i; n_hit = n_missed = 0; @@ -6034,20 +5570,11 @@ dpif_show_backer(const struct dpif_backer *backer, struct ds *ds) ds_put_format(ds, "%s: hit:%"PRIu64" missed:%"PRIu64"\n", dpif_name(backer->dpif), n_hit, n_missed); - ds_put_format(ds, "\tflows: cur: %zu, avg: %u, max: %u," + ds_put_format(ds, "\tflows: cur: %"PRIuSIZE", avg: %u, max: %u," " life span: %lldms\n", hmap_count(&backer->subfacets), backer->avg_n_subfacet, backer->max_n_subfacet, backer->avg_subfacet_life); - minutes = (time_msec() - backer->created) / (1000 * 60); - if (minutes >= 60) { - show_dp_rates(ds, "\thourly avg:", &backer->hourly); - } - if (minutes >= 60 * 24) { - show_dp_rates(ds, "\tdaily avg:", &backer->daily); - } - show_dp_rates(ds, "\toverall avg:", &backer->lifetime); - shash_init(&ofproto_shash); ofprotos = get_ofprotos(&ofproto_shash); for (i = 0; i < shash_count(&ofproto_shash); i++) { @@ -6154,7 +5681,7 @@ ofproto_unixctl_dpif_dump_megaflows(struct unixctl_conn *conn, CLS_CURSOR_FOR_EACH (facet, cr, &cursor) { cls_rule_format(&facet->cr, &ds); ds_put_cstr(&ds, ", "); - ds_put_format(&ds, "n_subfacets:%zu, ", list_size(&facet->subfacets)); + ds_put_format(&ds, "n_subfacets:%"PRIuSIZE", ", list_size(&facet->subfacets)); ds_put_format(&ds, "used:%.3fs, ", (now - facet->used) / 1000.0); ds_put_cstr(&ds, "Datapath actions: "); if (facet->xout.slow) { @@ -6254,7 +5781,7 @@ ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn, } odp_flow_format(subfacet->key, subfacet->key_len, - mask.data, mask.size, &ds, false); + mask.data, mask.size, NULL, &ds, false); ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:", subfacet->dp_packet_count, subfacet->dp_byte_count); @@ -6321,16 +5848,16 @@ ofproto_dpif_unixctl_init(void) unixctl_command_register( "ofproto/trace", - "[dp_name]|bridge odp_flow|br_flow [-generate|packet]", + "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]", 1, 3, ofproto_unixctl_trace, NULL); + unixctl_command_register( + "ofproto/trace-packet-out", + "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions", + 2, 6, ofproto_unixctl_trace_actions, NULL); unixctl_command_register("fdb/flush", "[bridge]", 0, 1, ofproto_unixctl_fdb_flush, NULL); unixctl_command_register("fdb/show", "bridge", 1, 1, ofproto_unixctl_fdb_show, NULL); - unixctl_command_register("ofproto/clog", "", 0, 0, - ofproto_dpif_clog, NULL); - unixctl_command_register("ofproto/unclog", "", 0, 0, - ofproto_dpif_unclog, NULL); unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1, ofproto_dpif_self_check, NULL); unixctl_command_register("dpif/dump-dps", "", 0, 0, @@ -6475,7 +6002,7 @@ vlandev_find(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port) static ofp_port_t vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto, ofp_port_t vlandev_ofp_port, int *vid) - OVS_REQ_WRLOCK(ofproto->vsp_mutex) + OVS_REQUIRES(ofproto->vsp_mutex) { if (!hmap_is_empty(&ofproto->vlandev_map)) { const struct vlan_splinter *vsp; @@ -6604,51 +6131,6 @@ odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, odp_port_t odp_port) } } -/* Compute exponentially weighted moving average, adding 'new' as the newest, - * most heavily weighted element. 'base' designates the rate of decay: after - * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e - * (about .37). */ -static void -exp_mavg(double *avg, int base, double new) -{ - *avg = (*avg * (base - 1) + new) / base; -} - -static void -update_moving_averages(struct dpif_backer *backer) -{ - const int min_ms = 60 * 1000; /* milliseconds in one minute. */ - long long int minutes = (time_msec() - backer->created) / min_ms; - - if (minutes > 0) { - backer->lifetime.add_rate = (double) backer->total_subfacet_add_count - / minutes; - backer->lifetime.del_rate = (double) backer->total_subfacet_del_count - / minutes; - } else { - backer->lifetime.add_rate = 0.0; - backer->lifetime.del_rate = 0.0; - } - - /* Update hourly averages on the minute boundaries. */ - if (time_msec() - backer->last_minute >= min_ms) { - exp_mavg(&backer->hourly.add_rate, 60, backer->subfacet_add_count); - exp_mavg(&backer->hourly.del_rate, 60, backer->subfacet_del_count); - - /* Update daily averages on the hour boundaries. */ - if ((backer->last_minute - backer->created) / min_ms % 60 == 59) { - exp_mavg(&backer->daily.add_rate, 24, backer->hourly.add_rate); - exp_mavg(&backer->daily.del_rate, 24, backer->hourly.del_rate); - } - - backer->total_subfacet_add_count += backer->subfacet_add_count; - backer->total_subfacet_del_count += backer->subfacet_del_count; - backer->subfacet_add_count = 0; - backer->subfacet_del_count = 0; - backer->last_minute += min_ms; - } -} - const struct ofproto_class ofproto_dpif_class = { init, enumerate_types, @@ -6656,14 +6138,12 @@ const struct ofproto_class ofproto_dpif_class = { del, port_open_type, type_run, - type_run_fast, type_wait, alloc, construct, destruct, dealloc, run, - run_fast, wait, get_memory_usage, flush, @@ -6688,6 +6168,8 @@ const struct ofproto_class ofproto_dpif_class = { NULL, /* rule_choose_table */ rule_alloc, rule_construct, + rule_insert, + rule_delete, rule_destruct, rule_dealloc, rule_get_stats, @@ -6707,6 +6189,7 @@ const struct ofproto_class ofproto_dpif_class = { get_stp_status, set_stp_port, get_stp_port_status, + get_stp_port_stats, set_queues, bundle_set, bundle_remove, @@ -6721,4 +6204,10 @@ const struct ofproto_class ofproto_dpif_class = { NULL, /* meter_set */ NULL, /* meter_get */ NULL, /* meter_del */ + group_alloc, /* group_alloc */ + group_construct, /* group_construct */ + group_destruct, /* group_destruct */ + group_dealloc, /* group_dealloc */ + group_modify, /* group_modify */ + group_get_stats, /* group_get_stats */ };