X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=ofproto%2Fofproto-dpif.c;h=55d030f68a097c0ab98562404025c1ba7686c055;hb=454a77e5b4b3161632945fd8ed75dc931d61a39f;hp=4d8540ecfbcb9134e84ce3afad3fad128fcc90a7;hpb=6472ba1117d19919c9cd1726e4f8cef0669059d6;p=sliver-openvswitch.git diff --git a/ofproto/ofproto-dpif.c b/ofproto/ofproto-dpif.c index 4d8540ecf..55d030f68 100644 --- a/ofproto/ofproto-dpif.c +++ b/ofproto/ofproto-dpif.c @@ -20,7 +20,7 @@ #include -#include "autopath.h" +#include "bfd.h" #include "bond.h" #include "bundle.h" #include "byte-order.h" @@ -36,6 +36,7 @@ #include "mac-learning.h" #include "meta-flow.h" #include "multipath.h" +#include "netdev-vport.h" #include "netdev.h" #include "netlink.h" #include "nx-match.h" @@ -46,11 +47,13 @@ #include "ofp-parse.h" #include "ofp-print.h" #include "ofproto-dpif-governor.h" +#include "ofproto-dpif-ipfix.h" #include "ofproto-dpif-sflow.h" #include "poll-loop.h" #include "simap.h" #include "smap.h" #include "timer.h" +#include "tunnel.h" #include "unaligned.h" #include "unixctl.h" #include "vlan-bitmap.h" @@ -77,6 +80,7 @@ BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255); struct ofport_dpif; struct ofproto_dpif; struct flow_miss; +struct facet; struct rule_dpif { struct rule up; @@ -117,8 +121,7 @@ static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto, static void rule_credit_stats(struct rule_dpif *, const struct dpif_flow_stats *); -static void flow_push_stats(struct rule_dpif *, const struct flow *, - const struct dpif_flow_stats *); +static void flow_push_stats(struct facet *, const struct dpif_flow_stats *); static tag_type rule_calculate_tag(const struct flow *, const struct minimask *, uint32_t basis); static void rule_invalidate(const struct rule_dpif *); @@ -215,6 +218,22 @@ struct action_xlate_ctx { * this flow when actions change header fields. */ struct flow flow; + /* Flow at the last commit. */ + struct flow base_flow; + + /* Tunnel IP destination address as received. This is stored separately + * as the base_flow.tunnel is cleared on init to reflect the datapath + * behavior. Used to make sure not to send tunneled output to ourselves, + * which might lead to an infinite loop. This could happen easily + * if a tunnel is marked as 'ip_remote=flow', and the flow does not + * actually set the tun_dst field. */ + ovs_be32 orig_tunnel_ip_dst; + + /* stack for the push and pop actions. + * Each stack element is of the type "union mf_subvalue". */ + struct ofpbuf stack; + union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)]; + /* The packet corresponding to 'flow', or a null pointer if we are * revalidating without a packet to refer to. */ const struct ofpbuf *packet; @@ -275,19 +294,33 @@ struct action_xlate_ctx { int recurse; /* Recursion level, via xlate_table_action. */ bool max_resubmit_trigger; /* Recursed too deeply during translation. */ - struct flow base_flow; /* Flow at the last commit. */ uint32_t orig_skb_priority; /* Priority when packet arrived. */ uint8_t table_id; /* OpenFlow table ID where flow was found. */ uint32_t sflow_n_outputs; /* Number of output ports. */ uint32_t sflow_odp_port; /* Output port for composing sFlow action. */ uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */ bool exit; /* No further actions should be processed. */ - struct flow orig_flow; /* Copy of original flow. */ +}; + +/* Initial values of fields of the packet that may be changed during + * flow processing and needed later. */ +struct initial_vals { + /* This is the value of vlan_tci in the packet as actually received from + * dpif. This is the same as the facet's flow.vlan_tci unless the packet + * was received via a VLAN splinter. In that case, this value is 0 + * (because the packet as actually received from the dpif had no 802.1Q + * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter + * represents. + * + * This member should be removed when the VLAN splinters feature is no + * longer needed. */ + ovs_be16 vlan_tci; }; static void action_xlate_ctx_init(struct action_xlate_ctx *, struct ofproto_dpif *, const struct flow *, - ovs_be16 initial_tci, struct rule_dpif *, + const struct initial_vals *initial_vals, + struct rule_dpif *, uint8_t tcp_flags, const struct ofpbuf *); static void xlate_actions(struct action_xlate_ctx *, const struct ofpact *ofpacts, size_t ofpacts_len, @@ -295,11 +328,14 @@ static void xlate_actions(struct action_xlate_ctx *, static void xlate_actions_for_side_effects(struct action_xlate_ctx *, const struct ofpact *ofpacts, size_t ofpacts_len); +static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port, + uint8_t table_id, bool may_packet_in); static size_t put_userspace_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions, const struct flow *, - const union user_action_cookie *); + const union user_action_cookie *, + const size_t); static void compose_slow_path(const struct ofproto_dpif *, const struct flow *, enum slow_path_reason, @@ -339,16 +375,12 @@ struct subfacet { struct list list_node; /* In struct facet's 'facets' list. */ struct facet *facet; /* Owning facet. */ - /* Key. - * - * To save memory in the common case, 'key' is NULL if 'key_fitness' is - * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately - * regenerate the ODP flow key from ->facet->flow. */ enum odp_key_fitness key_fitness; struct nlattr *key; int key_len; long long int used; /* Time last used; time created if not used. */ + long long int created; /* Time created. */ uint64_t dp_packet_count; /* Last known packet count in the datapath. */ uint64_t dp_byte_count; /* Last known byte count in the datapath. */ @@ -363,10 +395,8 @@ struct subfacet { enum slow_path_reason slow; /* 0 if fast path may be used. */ enum subfacet_path path; /* Installed in datapath? */ - /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN - * splinters can cause it to differ. This value should be removed when - * the VLAN splinters feature is no longer needed. */ - ovs_be16 initial_tci; /* Initial VLAN TCI value. */ + /* Initial values of the packet that may be needed later. */ + struct initial_vals initial_vals; /* Datapath port the packet arrived on. This is needed to remove * flows for ports that are no longer part of the bridge. Since the @@ -382,22 +412,18 @@ static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss, long long int now); static struct subfacet *subfacet_find(struct ofproto_dpif *, const struct nlattr *key, size_t key_len, - uint32_t key_hash, - const struct flow *flow); + uint32_t key_hash); static void subfacet_destroy(struct subfacet *); static void subfacet_destroy__(struct subfacet *); static void subfacet_destroy_batch(struct ofproto_dpif *, struct subfacet **, int n); -static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *, - struct ofpbuf *key); static void subfacet_reset_dp_stats(struct subfacet *, struct dpif_flow_stats *); static void subfacet_update_time(struct subfacet *, long long int used); static void subfacet_update_stats(struct subfacet *, const struct dpif_flow_stats *); static void subfacet_make_actions(struct subfacet *, - const struct ofpbuf *packet, - struct ofpbuf *odp_actions); + const struct ofpbuf *packet); static int subfacet_install(struct subfacet *, const struct nlattr *actions, size_t actions_len, struct dpif_flow_stats *, enum slow_path_reason); @@ -471,8 +497,12 @@ struct facet { /* Storage for a single subfacet, to reduce malloc() time and space * overhead. (A facet always has at least one subfacet and in the common - * case has exactly one subfacet.) */ + * case has exactly one subfacet. However, 'one_subfacet' may not + * always be valid, since it could have been removed after newer + * subfacets were pushed onto the 'subfacets' list.) */ struct subfacet one_subfacet; + + long long int learn_rl; /* Rate limiter for facet_learn(). */ }; static struct facet *facet_create(struct rule_dpif *, @@ -484,7 +514,7 @@ static struct facet *facet_find(struct ofproto_dpif *, const struct flow *, uint32_t hash); static struct facet *facet_lookup_valid(struct ofproto_dpif *, const struct flow *, uint32_t hash); -static void facet_revalidate(struct facet *); +static bool facet_revalidate(struct facet *); static bool facet_check_consistency(struct facet *); static void facet_flush_stats(struct facet *); @@ -494,6 +524,9 @@ static void facet_reset_counters(struct facet *); static void facet_push_stats(struct facet *); static void facet_learn(struct facet *); static void facet_account(struct facet *); +static void push_all_stats(void); + +static struct subfacet *facet_get_subfacet(struct facet *); static bool facet_is_controller_flow(struct facet *); @@ -505,10 +538,11 @@ struct ofport_dpif { struct ofbundle *bundle; /* Bundle that contains this port, if any. */ struct list bundle_node; /* In struct ofbundle's "ports" list. */ struct cfm *cfm; /* Connectivity Fault Management, if any. */ + struct bfd *bfd; /* BFD, if any. */ tag_type tag; /* Tag associated with this port. */ - uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */ bool may_enable; /* May be enabled in bonds. */ long long int carrier_seq; /* Carrier status changes. */ + struct tnl_port *tnl_port; /* Tunnel handle, or null. */ /* Spanning tree. */ struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ @@ -565,15 +599,16 @@ static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *, static struct ofport_dpif * ofport_dpif_cast(const struct ofport *ofport) { - assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class); return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; } 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_clear_priorities(struct ofport_dpif *); +static void run_fast_rl(void); struct dpif_completion { struct list list_node; @@ -612,6 +647,15 @@ COVERAGE_DEFINE(rev_port_toggled); COVERAGE_DEFINE(rev_flow_table); 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; +}; + /* All datapaths of a given type share a single dpif backer instance. */ struct dpif_backer { char *type; @@ -619,14 +663,33 @@ struct dpif_backer { struct dpif *dpif; struct timer next_expiration; struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */ + + struct simap tnl_backers; /* Set of dpif ports backing tunnels. */ + + /* Facet revalidation flags applying to facets which use this backer. */ + enum revalidate_reason need_revalidate; /* Revalidate every facet. */ + struct tag_set revalidate_set; /* Revalidate only matching facets. */ + + struct hmap drop_keys; /* Set of dropped odp keys. */ }; /* 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 struct ofport_dpif * odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port); +static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, + uint64_t delta); +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. */ +}; +static void show_dp_rates(struct ds *ds, const char *heading, + const struct avg_subfacet_rates *rates); +static void exp_mavg(double *avg, int base, double new); + struct ofproto_dpif { struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */ struct ofproto up; @@ -642,6 +705,7 @@ struct ofproto_dpif { /* Bridging. */ struct netflow *netflow; struct dpif_sflow *sflow; + struct dpif_ipfix *ipfix; struct hmap bundles; /* Contains "struct ofbundle"s. */ struct mac_learning *ml; struct ofmirror *mirrors[MAX_MIRRORS]; @@ -652,11 +716,10 @@ struct ofproto_dpif { struct hmap facets; struct hmap subfacets; struct governor *governor; + long long int consistency_rl; /* Revalidation. */ struct table_dpif tables[N_TABLES]; - enum revalidate_reason need_revalidate; - struct tag_set revalidate_set; /* Support for debugging async flow mods. */ struct list completions; @@ -674,10 +737,56 @@ struct ofproto_dpif { struct hmap vlandev_map; /* vlandev -> (realdev,vid). */ /* Ports. */ - struct sset ports; /* Set of port names. */ + struct sset ports; /* Set of standard port names. */ + 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. */ + + /* Per ofproto's dpif stats. */ + uint64_t n_hit; + uint64_t n_missed; + + /* Subfacet statistics. + * + * These keep track of the total number of subfacets added and deleted and + * flow life span. They are useful for computing the flow rates stats + * 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 int max_n_subfacet; /* Maximum number of flows */ + + /* The average number of subfacets... */ + struct avg_subfacet_rates hourly; /* ...over the last hour. */ + struct avg_subfacet_rates daily; /* ...over the last day. */ + long long int last_minute; /* Last time 'hourly' was updated. */ + + /* Number of subfacets added or deleted since 'last_minute'. */ + unsigned int subfacet_add_count; + unsigned int 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; + + /* Sum of the number of milliseconds that each subfacet existed, + * over the subfacets that have been added and then later deleted. */ + unsigned long long int total_subfacet_life_span; + + /* Incremented by the number of currently existing subfacets, each + * time we pull statistics from the kernel. */ + unsigned long long int total_subfacet_count; + + /* Number of times we pull statistics from the kernel. */ + unsigned long long int n_update_stats; }; +static unsigned long long int avg_subfacet_life_span( + const struct ofproto_dpif *); +static double avg_subfacet_count(const struct ofproto_dpif *ofproto); +static void update_moving_averages(struct ofproto_dpif *ofproto); +static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, + uint64_t delta); +static void update_max_subfacet_count(struct ofproto_dpif *ofproto); /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only * for debugging the asynchronous flow_mod implementation.) */ @@ -691,7 +800,7 @@ static void ofproto_dpif_unixctl_init(void); static struct ofproto_dpif * ofproto_dpif_cast(const struct ofproto *ofproto) { - assert(ofproto->ofproto_class == &ofproto_dpif_class); + ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class); return CONTAINER_OF(ofproto, struct ofproto_dpif, up); } @@ -700,8 +809,8 @@ static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *, static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *, uint32_t odp_port); static void ofproto_trace(struct ofproto_dpif *, const struct flow *, - const struct ofpbuf *, ovs_be16 initial_tci, - struct ds *); + const struct ofpbuf *, + const struct initial_vals *, struct ds *); /* Packet processing. */ static void update_learning_table(struct ofproto_dpif *, @@ -722,6 +831,9 @@ static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet); static size_t compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions, const struct flow *, uint32_t odp_port); +static void compose_ipfix_action(const struct ofproto_dpif *, + struct ofpbuf *odp_actions, + const struct flow *); static void add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *flow); /* Global variables. */ @@ -811,6 +923,7 @@ lookup_ofproto_dpif_by_port_name(const char *name) static int type_run(const char *type) { + static long long int push_timer = LLONG_MIN; struct dpif_backer *backer; char *devname; int error; @@ -824,6 +937,109 @@ type_run(const char *type) dpif_run(backer->dpif); + /* 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 + * ability to quickly set up flows. To reduce the cost, we push statistics + * here instead. */ + if (time_msec() > push_timer) { + push_timer = time_msec() + 2000; + push_all_stats(); + } + + if (backer->need_revalidate + || !tag_set_is_empty(&backer->revalidate_set)) { + struct tag_set revalidate_set = backer->revalidate_set; + bool need_revalidate = backer->need_revalidate; + struct ofproto_dpif *ofproto; + struct simap_node *node; + struct simap tmp_backers; + + /* Handle tunnel garbage collection. */ + simap_init(&tmp_backers); + simap_swap(&backer->tnl_backers, &tmp_backers); + + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { + struct ofport_dpif *iter; + + if (backer != ofproto->backer) { + continue; + } + + HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) { + const char *dp_port; + + if (!iter->tnl_port) { + continue; + } + + dp_port = netdev_vport_get_dpif_port(iter->up.netdev); + node = simap_find(&tmp_backers, dp_port); + if (node) { + simap_put(&backer->tnl_backers, dp_port, node->data); + simap_delete(&tmp_backers, node); + node = simap_find(&backer->tnl_backers, dp_port); + } else { + node = simap_find(&backer->tnl_backers, dp_port); + if (!node) { + uint32_t odp_port = UINT32_MAX; + + if (!dpif_port_add(backer->dpif, iter->up.netdev, + &odp_port)) { + simap_put(&backer->tnl_backers, dp_port, odp_port); + node = simap_find(&backer->tnl_backers, dp_port); + } + } + } + + iter->odp_port = node ? node->data : OVSP_NONE; + if (tnl_port_reconfigure(&iter->up, iter->odp_port, + &iter->tnl_port)) { + backer->need_revalidate = REV_RECONFIGURE; + } + } + } + + SIMAP_FOR_EACH (node, &tmp_backers) { + dpif_port_del(backer->dpif, node->data); + } + simap_destroy(&tmp_backers); + + switch (backer->need_revalidate) { + case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break; + case REV_STP: COVERAGE_INC(rev_stp); break; + case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break; + case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break; + case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break; + } + + if (backer->need_revalidate) { + /* Clear the drop_keys in case we should now be accepting some + * formerly dropped flows. */ + drop_key_clear(backer); + } + + /* Clear the revalidation flags. */ + tag_set_init(&backer->revalidate_set); + backer->need_revalidate = 0; + + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { + struct facet *facet, *next; + + if (ofproto->backer != backer) { + continue; + } + + HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) { + if (need_revalidate + || tag_set_intersects(&revalidate_set, facet->tags)) { + facet_revalidate(facet); + run_fast_rl(); + } + } + } + } + if (timer_expired(&backer->next_expiration)) { int delay = expire(backer); timer_set_duration(&backer->next_expiration, delay); @@ -839,6 +1055,13 @@ type_run(const char *type) goto next; } + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, + &all_ofproto_dpifs) { + if (simap_contains(&ofproto->backer->tnl_backers, devname)) { + goto next; + } + } + ofproto = lookup_ofproto_dpif_by_port_name(devname); if (dpif_port_query_by_name(backer->dpif, devname, &port)) { /* The port was removed. If we know the datapath, @@ -879,18 +1102,10 @@ type_run(const char *type) } static int -type_run_fast(const char *type) +dpif_backer_run_fast(struct dpif_backer *backer, int max_batch) { - struct dpif_backer *backer; unsigned int work; - 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; - } - /* Handle one or more batches of upcalls, until there's nothing left to do * or until we do a fixed total amount of work. * @@ -901,8 +1116,8 @@ type_run_fast(const char *type) * optimizations can make major improvements on some benchmarks and * presumably for real traffic as well. */ work = 0; - while (work < FLOW_MISS_MAX_BATCH) { - int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work); + while (work < max_batch) { + int retval = handle_upcalls(backer, max_batch - work); if (retval <= 0) { return -retval; } @@ -912,6 +1127,58 @@ type_run_fast(const char *type) 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; + struct ofport_dpif *ofport; + + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { + + HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { + port_run_fast(ofport); + } + } + 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) { @@ -950,12 +1217,16 @@ close_dpif_backer(struct dpif_backer *backer) { struct shash_node *node; - assert(backer->refcount > 0); + ovs_assert(backer->refcount > 0); if (--backer->refcount) { return; } + drop_key_clear(backer); + hmap_destroy(&backer->drop_keys); + + simap_destroy(&backer->tnl_backers); hmap_destroy(&backer->odp_to_ofport_map); node = shash_find(&all_dpif_backers, backer->type); free(backer->type); @@ -1022,13 +1293,18 @@ open_dpif_backer(const char *type, struct dpif_backer **backerp) if (error) { VLOG_ERR("failed to open datapath of type %s: %s", type, strerror(error)); + free(backer); return error; } backer->type = xstrdup(type); backer->refcount = 1; hmap_init(&backer->odp_to_ofport_map); + hmap_init(&backer->drop_keys); timer_set_duration(&backer->next_expiration, 1000); + backer->need_revalidate = 0; + simap_init(&backer->tnl_backers); + tag_set_init(&backer->revalidate_set); *backerp = backer; dpif_flow_flush(backer->dpif); @@ -1087,6 +1363,7 @@ construct(struct ofproto *ofproto_) ofproto->netflow = NULL; ofproto->sflow = NULL; + ofproto->ipfix = NULL; ofproto->stp = NULL; hmap_init(&ofproto->bundles); ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME); @@ -1098,6 +1375,7 @@ construct(struct ofproto *ofproto_) hmap_init(&ofproto->facets); hmap_init(&ofproto->subfacets); ofproto->governor = NULL; + ofproto->consistency_rl = LLONG_MIN; for (i = 0; i < N_TABLES; i++) { struct table_dpif *table = &ofproto->tables[i]; @@ -1106,8 +1384,6 @@ construct(struct ofproto *ofproto_) table->other_table = NULL; table->basis = random_uint32(); } - ofproto->need_revalidate = 0; - tag_set_init(&ofproto->revalidate_set); list_init(&ofproto->completions); @@ -1120,6 +1396,7 @@ construct(struct ofproto *ofproto_) hmap_init(&ofproto->realdev_vid_map); sset_init(&ofproto->ports); + sset_init(&ofproto->ghost_ports); sset_init(&ofproto->port_poll_set); ofproto->port_poll_errno = 0; @@ -1147,6 +1424,22 @@ construct(struct ofproto *ofproto_) error = add_internal_flows(ofproto); ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY; + ofproto->n_hit = 0; + ofproto->n_missed = 0; + + ofproto->max_n_subfacet = 0; + ofproto->created = time_msec(); + ofproto->last_minute = ofproto->created; + memset(&ofproto->hourly, 0, sizeof ofproto->hourly); + memset(&ofproto->daily, 0, sizeof ofproto->daily); + ofproto->subfacet_add_count = 0; + ofproto->subfacet_del_count = 0; + ofproto->total_subfacet_add_count = 0; + ofproto->total_subfacet_del_count = 0; + ofproto->total_subfacet_life_span = 0; + ofproto->total_subfacet_count = 0; + ofproto->n_update_stats = 0; + return error; } @@ -1181,7 +1474,7 @@ add_internal_flow(struct ofproto_dpif *ofproto, int id, } *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL); - assert(*rulep != NULL); + ovs_assert(*rulep != NULL); return 0; } @@ -1264,6 +1557,7 @@ destruct(struct ofproto *ofproto_) hmap_destroy(&ofproto->realdev_vid_map); sset_destroy(&ofproto->ports); + sset_destroy(&ofproto->ghost_ports); sset_destroy(&ofproto->port_poll_set); close_dpif_backer(ofproto->backer); @@ -1316,44 +1610,22 @@ run(struct ofproto *ofproto_) } stp_run(ofproto); - mac_learning_run(ofproto->ml, &ofproto->revalidate_set); - - /* Now revalidate if there's anything to do. */ - if (ofproto->need_revalidate - || !tag_set_is_empty(&ofproto->revalidate_set)) { - struct tag_set revalidate_set = ofproto->revalidate_set; - bool revalidate_all = ofproto->need_revalidate; - struct facet *facet; - - switch (ofproto->need_revalidate) { - case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break; - case REV_STP: COVERAGE_INC(rev_stp); break; - case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break; - case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break; - case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break; - } - - /* Clear the revalidation flags. */ - tag_set_init(&ofproto->revalidate_set); - ofproto->need_revalidate = 0; - - HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { - if (revalidate_all - || tag_set_intersects(&revalidate_set, facet->tags)) { - facet_revalidate(facet); - } - } - } + mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set); /* Check the consistency of a random facet, to aid debugging. */ - if (!hmap_is_empty(&ofproto->facets) && !ofproto->need_revalidate) { + if (time_msec() >= ofproto->consistency_rl + && !hmap_is_empty(&ofproto->facets) + && !ofproto->backer->need_revalidate) { struct facet *facet; + ofproto->consistency_rl = time_msec() + 250; + facet = CONTAINER_OF(hmap_random_node(&ofproto->facets), struct facet, hmap_node); - if (!tag_set_intersects(&ofproto->revalidate_set, facet->tags)) { + if (!tag_set_intersects(&ofproto->backer->revalidate_set, + facet->tags)) { if (!facet_check_consistency(facet)) { - ofproto->need_revalidate = REV_INCONSISTENCY; + ofproto->backer->need_revalidate = REV_INCONSISTENCY; } } } @@ -1395,7 +1667,7 @@ wait(struct ofproto *ofproto_) if (ofproto->sflow) { dpif_sflow_wait(ofproto->sflow); } - if (!tag_set_is_empty(&ofproto->revalidate_set)) { + if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) { poll_immediate_wake(); } HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { @@ -1409,7 +1681,7 @@ wait(struct ofproto *ofproto_) } mac_learning_wait(ofproto->ml); stp_wait(ofproto); - if (ofproto->need_revalidate) { + if (ofproto->backer->need_revalidate) { /* Shouldn't happen, but if it does just go around again. */ VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); poll_immediate_wake(); @@ -1507,23 +1779,36 @@ port_construct(struct ofport *port_) { struct ofport_dpif *port = ofport_dpif_cast(port_); struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); + const struct netdev *netdev = port->up.netdev; struct dpif_port dpif_port; int error; - ofproto->need_revalidate = REV_RECONFIGURE; + ofproto->backer->need_revalidate = REV_RECONFIGURE; port->bundle = NULL; port->cfm = NULL; + port->bfd = NULL; port->tag = tag_create_random(); port->may_enable = true; port->stp_port = NULL; port->stp_state = STP_DISABLED; + port->tnl_port = NULL; hmap_init(&port->priorities); port->realdev_ofp_port = 0; port->vlandev_vid = 0; - port->carrier_seq = netdev_get_carrier_resets(port->up.netdev); + port->carrier_seq = netdev_get_carrier_resets(netdev); + + if (netdev_vport_is_patch(netdev)) { + /* By bailing out here, we don't submit the port to the sFlow module + * to be considered for counter polling export. This is correct + * because the patch port represents an interface that sFlow considers + * to be "internal" to the switch as a whole, and therefore not an + * candidate for counter polling. */ + port->odp_port = OVSP_NONE; + return 0; + } error = dpif_port_query_by_name(ofproto->backer->dpif, - netdev_get_name(port->up.netdev), + netdev_vport_get_dpif_port(netdev), &dpif_port); if (error) { return error; @@ -1531,16 +1816,22 @@ port_construct(struct ofport *port_) port->odp_port = dpif_port.port_no; - /* Sanity-check that a mapping doesn't already exist. This - * shouldn't happen. */ - if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) { - VLOG_ERR("port %s already has an OpenFlow port number\n", - dpif_port.name); - return EBUSY; - } + if (netdev_get_tunnel_config(netdev)) { + port->tnl_port = tnl_port_add(&port->up, port->odp_port); + } else { + /* Sanity-check that a mapping doesn't already exist. This + * shouldn't happen for non-tunnel ports. */ + if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) { + VLOG_ERR("port %s already has an OpenFlow port number", + dpif_port.name); + dpif_port_destroy(&dpif_port); + return EBUSY; + } - hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node, - hash_int(port->odp_port, 0)); + hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node, + hash_int(port->odp_port, 0)); + } + dpif_port_destroy(&dpif_port); if (ofproto->sflow) { dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port); @@ -1554,21 +1845,31 @@ port_destruct(struct ofport *port_) { struct ofport_dpif *port = ofport_dpif_cast(port_); struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); + const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev); const char *devname = netdev_get_name(port->up.netdev); - if (dpif_port_exists(ofproto->backer->dpif, devname)) { + if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { /* The underlying device is still there, so delete it. This * happens when the ofproto is being destroyed, since the caller * assumes that removal of attached ports will happen as part of * destruction. */ - dpif_port_del(ofproto->backer->dpif, port->odp_port); + if (!port->tnl_port) { + dpif_port_del(ofproto->backer->dpif, port->odp_port); + } + ofproto->backer->need_revalidate = REV_RECONFIGURE; } + if (port->odp_port != OVSP_NONE && !port->tnl_port) { + hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node); + } + + tnl_port_del(port->tnl_port); sset_find_and_delete(&ofproto->ports, devname); - hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node); - ofproto->need_revalidate = REV_RECONFIGURE; + sset_find_and_delete(&ofproto->ghost_ports, devname); + ofproto->backer->need_revalidate = REV_RECONFIGURE; bundle_remove(port_); set_cfm(port_, NULL); + set_bfd(port_, NULL); if (ofproto->sflow) { dpif_sflow_del_port(ofproto->sflow, port->odp_port); } @@ -1597,7 +1898,7 @@ port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config) if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP | OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD | OFPUTIL_PC_NO_PACKET_IN)) { - ofproto->need_revalidate = REV_RECONFIGURE; + ofproto->backer->need_revalidate = REV_RECONFIGURE; if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) { bundle_update(port->bundle); @@ -1620,19 +1921,45 @@ set_sflow(struct ofproto *ofproto_, HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port); } - ofproto->need_revalidate = REV_RECONFIGURE; + ofproto->backer->need_revalidate = REV_RECONFIGURE; } dpif_sflow_set_options(ds, sflow_options); } else { if (ds) { dpif_sflow_destroy(ds); - ofproto->need_revalidate = REV_RECONFIGURE; + ofproto->backer->need_revalidate = REV_RECONFIGURE; ofproto->sflow = NULL; } } return 0; } +static int +set_ipfix( + struct ofproto *ofproto_, + const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options, + const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options, + size_t n_flow_exporters_options) +{ + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); + struct dpif_ipfix *di = ofproto->ipfix; + + if (bridge_exporter_options || flow_exporters_options) { + if (!di) { + di = ofproto->ipfix = dpif_ipfix_create(); + } + dpif_ipfix_set_options( + di, bridge_exporter_options, flow_exporters_options, + n_flow_exporters_options); + } else { + if (di) { + dpif_ipfix_destroy(di); + ofproto->ipfix = NULL; + } + } + return 0; +} + static int set_cfm(struct ofport *ofport_, const struct cfm_settings *s) { @@ -1646,8 +1973,8 @@ set_cfm(struct ofport *ofport_, const struct cfm_settings *s) struct ofproto_dpif *ofproto; ofproto = ofproto_dpif_cast(ofport->up.ofproto); - ofproto->need_revalidate = REV_RECONFIGURE; - ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev)); + ofproto->backer->need_revalidate = REV_RECONFIGURE; + ofport->cfm = cfm_create(ofport->up.netdev); } if (cfm_configure(ofport->cfm, s)) { @@ -1661,43 +1988,51 @@ set_cfm(struct ofport *ofport_, const struct cfm_settings *s) return error; } -static int -get_cfm_fault(const struct ofport *ofport_) +static bool +get_cfm_status(const struct ofport *ofport_, + struct ofproto_cfm_status *status) { struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); - return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1; + if (ofport->cfm) { + status->faults = cfm_get_fault(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); + return true; + } else { + return false; + } } static int -get_cfm_opup(const struct ofport *ofport_) +set_bfd(struct ofport *ofport_, const struct smap *cfg) { + struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto); struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); + struct bfd *old; + + old = ofport->bfd; + ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg); + if (ofport->bfd != old) { + ofproto->backer->need_revalidate = REV_RECONFIGURE; + } - return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1; + return 0; } static int -get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps, - size_t *n_rmps) +get_bfd_status(struct ofport *ofport_, struct smap *smap) { struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); - if (ofport->cfm) { - cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps); + if (ofport->bfd) { + bfd_get_status(ofport->bfd, smap); return 0; } else { - return -1; + return ENOENT; } } - -static int -get_cfm_health(const struct ofport *ofport_) -{ - struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); - - return ofport->cfm ? cfm_get_health(ofport->cfm) : -1; -} /* Spanning Tree. */ @@ -1734,7 +2069,7 @@ set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) /* Only revalidate flows if the configuration changed. */ if (!s != !ofproto->stp) { - ofproto->need_revalidate = REV_RECONFIGURE; + ofproto->backer->need_revalidate = REV_RECONFIGURE; } if (s) { @@ -1802,12 +2137,13 @@ update_stp_port_state(struct ofport_dpif *ofport) if (stp_learn_in_state(ofport->stp_state) != stp_learn_in_state(state)) { /* xxx Learning action flows should also be flushed. */ - mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); + mac_learning_flush(ofproto->ml, + &ofproto->backer->revalidate_set); } fwd_change = stp_forward_in_state(ofport->stp_state) != stp_forward_in_state(state); - ofproto->need_revalidate = REV_STP; + ofproto->backer->need_revalidate = REV_STP; ofport->stp_state = state; ofport->stp_state_entered = time_msec(); @@ -1907,7 +2243,7 @@ stp_run(struct ofproto_dpif *ofproto) } if (stp_check_and_reset_fdb_flush(ofproto->stp)) { - mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); + mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); } } } @@ -2005,12 +2341,12 @@ set_queues(struct ofport *ofport_, pdscp = xmalloc(sizeof *pdscp); pdscp->priority = priority; pdscp->dscp = dscp; - ofproto->need_revalidate = REV_RECONFIGURE; + ofproto->backer->need_revalidate = REV_RECONFIGURE; } if (pdscp->dscp != dscp) { pdscp->dscp = dscp; - ofproto->need_revalidate = REV_RECONFIGURE; + ofproto->backer->need_revalidate = REV_RECONFIGURE; } hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0)); @@ -2018,7 +2354,7 @@ set_queues(struct ofport *ofport_, if (!hmap_is_empty(&ofport->priorities)) { ofport_clear_priorities(ofport); - ofproto->need_revalidate = REV_RECONFIGURE; + ofproto->backer->need_revalidate = REV_RECONFIGURE; } hmap_swap(&new, &ofport->priorities); @@ -2045,7 +2381,7 @@ bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) struct mac_learning *ml = ofproto->ml; struct mac_entry *mac, *next_mac; - ofproto->need_revalidate = REV_RECONFIGURE; + ofproto->backer->need_revalidate = REV_RECONFIGURE; LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { if (mac->port.p == bundle) { if (all_ofprotos) { @@ -2058,7 +2394,6 @@ bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos) e = mac_learning_lookup(o->ml, mac->mac, mac->vlan, NULL); if (e) { - tag_set_add(&o->revalidate_set, e->tag); mac_learning_expire(o->ml, e); } } @@ -2122,7 +2457,7 @@ bundle_del_port(struct ofport_dpif *port) { struct ofbundle *bundle = port->bundle; - bundle->ofproto->need_revalidate = REV_RECONFIGURE; + bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; list_remove(&port->bundle_node); port->bundle = NULL; @@ -2138,9 +2473,8 @@ bundle_del_port(struct ofport_dpif *port) } static bool -bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, - struct lacp_slave_settings *lacp, - uint32_t bond_stable_id) +bundle_add_port(struct ofbundle *bundle, uint16_t ofp_port, + struct lacp_slave_settings *lacp) { struct ofport_dpif *port; @@ -2150,7 +2484,7 @@ bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, } if (port->bundle != bundle) { - bundle->ofproto->need_revalidate = REV_RECONFIGURE; + bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; if (port->bundle) { bundle_del_port(port); } @@ -2163,12 +2497,10 @@ bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, } } if (lacp) { - port->bundle->ofproto->need_revalidate = REV_RECONFIGURE; + bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE; lacp_slave_register(bundle->lacp, port, lacp); } - port->bond_stable_id = bond_stable_id; - return true; } @@ -2191,7 +2523,7 @@ bundle_destroy(struct ofbundle *bundle) mirror_destroy(m); } else if (hmapx_find_and_delete(&m->srcs, bundle) || hmapx_find_and_delete(&m->dsts, bundle)) { - ofproto->need_revalidate = REV_RECONFIGURE; + ofproto->backer->need_revalidate = REV_RECONFIGURE; } } } @@ -2227,8 +2559,8 @@ bundle_set(struct ofproto *ofproto_, void *aux, return 0; } - assert(s->n_slaves == 1 || s->bond != NULL); - assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); + ovs_assert(s->n_slaves == 1 || s->bond != NULL); + ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); bundle = bundle_lookup(ofproto, aux); if (!bundle) { @@ -2263,7 +2595,7 @@ bundle_set(struct ofproto *ofproto_, void *aux, /* LACP. */ if (s->lacp) { if (!bundle->lacp) { - ofproto->need_revalidate = REV_RECONFIGURE; + ofproto->backer->need_revalidate = REV_RECONFIGURE; bundle->lacp = lacp_create(); } lacp_configure(bundle->lacp, s->lacp); @@ -2276,8 +2608,7 @@ bundle_set(struct ofproto *ofproto_, void *aux, ok = true; for (i = 0; i < s->n_slaves; i++) { if (!bundle_add_port(bundle, s->slaves[i], - s->lacp ? &s->lacp_slaves[i] : NULL, - s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) { + s->lacp ? &s->lacp_slaves[i] : NULL)) { ok = false; } } @@ -2295,7 +2626,7 @@ bundle_set(struct ofproto *ofproto_, void *aux, found: ; } } - assert(list_size(&bundle->ports) <= s->n_slaves); + ovs_assert(list_size(&bundle->ports) <= s->n_slaves); if (list_is_empty(&bundle->ports)) { bundle_destroy(bundle); @@ -2369,16 +2700,15 @@ bundle_set(struct ofproto *ofproto_, void *aux, bundle->ofproto->has_bonded_bundles = true; if (bundle->bond) { if (bond_reconfigure(bundle->bond, s->bond)) { - ofproto->need_revalidate = REV_RECONFIGURE; + ofproto->backer->need_revalidate = REV_RECONFIGURE; } } else { bundle->bond = bond_create(s->bond); - ofproto->need_revalidate = REV_RECONFIGURE; + ofproto->backer->need_revalidate = REV_RECONFIGURE; } LIST_FOR_EACH (port, bundle_node, &bundle->ports) { - bond_slave_register(bundle->bond, port, port->bond_stable_id, - port->up.netdev); + bond_slave_register(bundle->bond, port, port->up.netdev); } } else { bond_destroy(bundle->bond); @@ -2493,7 +2823,7 @@ bundle_run(struct ofbundle *bundle) bond_slave_set_may_enable(bundle->bond, port, port->may_enable); } - bond_run(bundle->bond, &bundle->ofproto->revalidate_set, + bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set, lacp_status(bundle->lacp)); if (bond_should_send_learning_packets(bundle->bond)) { bundle_send_learning_packets(bundle); @@ -2678,9 +3008,10 @@ mirror_set(struct ofproto *ofproto_, void *aux, } } - ofproto->need_revalidate = REV_RECONFIGURE; + ofproto->backer->need_revalidate = REV_RECONFIGURE; ofproto->has_mirrors = true; - mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); + mac_learning_flush(ofproto->ml, + &ofproto->backer->revalidate_set); mirror_update_dups(ofproto); return 0; @@ -2699,8 +3030,8 @@ mirror_destroy(struct ofmirror *mirror) } ofproto = mirror->ofproto; - ofproto->need_revalidate = REV_RECONFIGURE; - mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); + ofproto->backer->need_revalidate = REV_RECONFIGURE; + mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); mirror_bit = MIRROR_MASK_C(1) << mirror->idx; HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { @@ -2740,6 +3071,8 @@ mirror_get_stats(struct ofproto *ofproto_, void *aux, return 0; } + push_all_stats(); + *packets = mirror->packet_count; *bytes = mirror->byte_count; @@ -2751,7 +3084,7 @@ set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { - mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); + mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); } return 0; } @@ -2768,7 +3101,7 @@ static void forward_bpdu_changed(struct ofproto *ofproto_) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - ofproto->need_revalidate = REV_RECONFIGURE; + ofproto->backer->need_revalidate = REV_RECONFIGURE; } static void @@ -2806,6 +3139,28 @@ ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto, ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no); } +static struct ofport_dpif * +ofport_get_peer(const struct ofport_dpif *ofport_dpif) +{ + const struct ofproto_dpif *ofproto; + const char *peer; + + peer = netdev_vport_patch_peer(ofport_dpif->up.netdev); + if (!peer) { + return NULL; + } + + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { + struct ofport *ofport; + + ofport = shash_find_data(&ofproto->up.port_by_name, peer); + if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) { + return ofport_dpif_cast(ofport); + } + } + return NULL; +} + static void port_run_fast(struct ofport_dpif *ofport) { @@ -2817,6 +3172,15 @@ port_run_fast(struct ofport_dpif *ofport) 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); + } } static void @@ -2829,6 +3193,13 @@ port_run(struct ofport_dpif *ofport) ofport->carrier_seq = carrier_seq; port_run_fast(ofport); + + if (ofport->tnl_port + && tnl_port_reconfigure(&ofport->up, ofport->odp_port, + &ofport->tnl_port)) { + ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true; + } + if (ofport->cfm) { int cfm_opup = cfm_get_opup(ofport->cfm); @@ -2840,6 +3211,11 @@ port_run(struct ofport_dpif *ofport) } } + if (ofport->bfd) { + bfd_run(ofport->bfd); + enable = enable && bfd_forwarding(ofport->bfd); + } + if (ofport->bundle) { enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); if (carrier_changed) { @@ -2851,7 +3227,7 @@ port_run(struct ofport_dpif *ofport) struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); if (ofproto->has_bundle_action) { - ofproto->need_revalidate = REV_PORT_TOGGLED; + ofproto->backer->need_revalidate = REV_PORT_TOGGLED; } } @@ -2864,6 +3240,10 @@ port_wait(struct ofport_dpif *ofport) if (ofport->cfm) { cfm_wait(ofport->cfm); } + + if (ofport->bfd) { + bfd_wait(ofport->bfd); + } } static int @@ -2874,6 +3254,24 @@ port_query_by_name(const struct ofproto *ofproto_, const char *devname, struct dpif_port dpif_port; int error; + if (sset_contains(&ofproto->ghost_ports, devname)) { + const char *type = netdev_get_type_from_name(devname); + + /* We may be called before ofproto->up.port_by_name is populated with + * the appropriate ofport. For this reason, we must get the name and + * type from the netdev layer directly. */ + if (type) { + const struct ofport *ofport; + + ofport = shash_find_data(&ofproto->up.port_by_name, devname); + ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE; + ofproto_port->name = xstrdup(devname); + ofproto_port->type = xstrdup(type); + return 0; + } + return ENODEV; + } + if (!sset_contains(&ofproto->ports, devname)) { return ENODEV; } @@ -2889,29 +3287,52 @@ static int port_add(struct ofproto *ofproto_, struct netdev *netdev) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - uint32_t odp_port = UINT32_MAX; - int error; + const char *dp_port_name = netdev_vport_get_dpif_port(netdev); + const char *devname = netdev_get_name(netdev); - error = dpif_port_add(ofproto->backer->dpif, netdev, &odp_port); - if (!error) { - sset_add(&ofproto->ports, netdev_get_name(netdev)); + if (netdev_vport_is_patch(netdev)) { + sset_add(&ofproto->ghost_ports, netdev_get_name(netdev)); + return 0; } - return error; + + if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) { + uint32_t port_no = UINT32_MAX; + int error; + + error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no); + if (error) { + return error; + } + if (netdev_get_tunnel_config(netdev)) { + simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no); + } + } + + if (netdev_get_tunnel_config(netdev)) { + sset_add(&ofproto->ghost_ports, devname); + } else { + sset_add(&ofproto->ports, devname); + } + return 0; } static int port_del(struct ofproto *ofproto_, uint16_t ofp_port) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - uint32_t odp_port = ofp_port_to_odp_port(ofproto, ofp_port); + struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); int error = 0; - if (odp_port != OFPP_NONE) { - error = dpif_port_del(ofproto->backer->dpif, odp_port); + if (!ofport) { + return 0; } - if (!error) { - struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); - if (ofport) { + + sset_find_and_delete(&ofproto->ghost_ports, + netdev_get_name(ofport->up.netdev)); + ofproto->backer->need_revalidate = REV_RECONFIGURE; + if (!ofport->tnl_port) { + 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 * bonded port, then the bond is using that netdev, so remove it * from the bond. The client will need to reconfigure everything @@ -2928,9 +3349,11 @@ port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); int error; + push_all_stats(); + error = netdev_get_stats(ofport->up.netdev, stats); - if (!error && ofport->odp_port == OVSP_LOCAL) { + if (!error && ofport_->ofp_port == OFPP_LOCAL) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); /* ofproto->stats.tx_packets represents packets that we created @@ -2962,57 +3385,56 @@ port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats) return error; } -/* Account packets for LOCAL port. */ -static void -ofproto_update_local_port_stats(const struct ofproto *ofproto_, - size_t tx_size, size_t rx_size) -{ - struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - - if (rx_size) { - ofproto->stats.rx_packets++; - ofproto->stats.rx_bytes += rx_size; - } - if (tx_size) { - ofproto->stats.tx_packets++; - ofproto->stats.tx_bytes += tx_size; - } -} - struct port_dump_state { uint32_t bucket; uint32_t offset; + bool ghost; + + struct ofproto_port port; + bool has_port; }; static int port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep) { - struct port_dump_state *state; - - *statep = state = xmalloc(sizeof *state); - state->bucket = 0; - state->offset = 0; + *statep = xzalloc(sizeof(struct port_dump_state)); return 0; } static int -port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_, +port_dump_next(const struct ofproto *ofproto_, void *state_, struct ofproto_port *port) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); struct port_dump_state *state = state_; + const struct sset *sset; struct sset_node *node; - while ((node = sset_at_position(&ofproto->ports, &state->bucket, - &state->offset))) { + if (state->has_port) { + ofproto_port_destroy(&state->port); + state->has_port = false; + } + sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports; + while ((node = sset_at_position(sset, &state->bucket, &state->offset))) { int error; - error = port_query_by_name(ofproto_, node->name, port); - if (error != ENODEV) { + error = port_query_by_name(ofproto_, node->name, &state->port); + if (!error) { + *port = state->port; + state->has_port = true; + return 0; + } else if (error != ENODEV) { return error; } } + if (!state->ghost) { + state->ghost = true; + state->bucket = 0; + state->offset = 0; + return port_dump_next(ofproto_, state_, port); + } + return EOF; } @@ -3021,6 +3443,9 @@ port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) { struct port_dump_state *state = state_; + if (state->has_port) { + ofproto_port_destroy(&state->port); + } free(state); return 0; } @@ -3078,7 +3503,7 @@ struct flow_miss { enum odp_key_fitness key_fitness; const struct nlattr *key; size_t key_len; - ovs_be16 initial_tci; + struct initial_vals initial_vals; struct list packets; enum dpif_upcall_type upcall_type; uint32_t odp_in_port; @@ -3086,7 +3511,6 @@ struct flow_miss { struct flow_miss_op { struct dpif_op dpif_op; - struct subfacet *subfacet; /* Subfacet */ void *garbage; /* Pointer to pass to free(), NULL if none. */ uint64_t stub[1024 / 8]; /* Temporary buffer. */ }; @@ -3117,19 +3541,20 @@ send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet, static enum slow_path_reason process_special(struct ofproto_dpif *ofproto, const struct flow *flow, - const struct ofpbuf *packet) + const struct ofport_dpif *ofport, const struct ofpbuf *packet) { - struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port); - if (!ofport) { return 0; - } - - if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) { + } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) { if (packet) { cfm_process_heartbeat(ofport->cfm, packet); } return SLOW_CFM; + } else if (ofport->bfd && bfd_should_process_flow(flow)) { + if (packet) { + bfd_process_packet(ofport->bfd, flow, packet); + } + return SLOW_BFD; } else if (ofport->bundle && ofport->bundle->lacp && flow->dl_type == htons(ETH_TYPE_LACP)) { if (packet) { @@ -3141,17 +3566,19 @@ process_special(struct ofproto_dpif *ofproto, const struct flow *flow, stp_process_packet(ofport, packet); } return SLOW_STP; + } else { + return 0; } - return 0; } static struct flow_miss * -flow_miss_find(struct hmap *todo, const struct flow *flow, uint32_t hash) +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 (flow_equal(&miss->flow, flow)) { + if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) { return miss; } } @@ -3168,7 +3595,7 @@ static void init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet, struct flow_miss_op *op) { - if (miss->flow.vlan_tci != miss->initial_tci) { + if (miss->flow.vlan_tci != miss->initial_vals.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 @@ -3178,7 +3605,6 @@ init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet, eth_pop_vlan(packet); } - op->subfacet = NULL; op->garbage = NULL; op->dpif_op.type = DPIF_OP_EXECUTE; op->dpif_op.u.execute.key = miss->key; @@ -3256,13 +3682,15 @@ handle_flow_miss_without_facet(struct flow_miss *miss, COVERAGE_INC(facet_suppress); + handle_flow_miss_common(rule, packet, &miss->flow); + ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); dpif_flow_stats_extract(&miss->flow, packet, now, &stats); rule_credit_stats(rule, &stats); - action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci, - rule, 0, packet); + action_xlate_ctx_init(&ctx, ofproto, &miss->flow, &miss->initial_vals, + rule, stats.tcp_flags, packet); ctx.resubmit_stats = &stats; xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions); @@ -3306,13 +3734,19 @@ handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, LIST_FOR_EACH (packet, list_node, &miss->packets) { struct flow_miss_op *op = &ops[*n_ops]; struct dpif_flow_stats stats; - struct ofpbuf odp_actions; handle_flow_miss_common(facet->rule, packet, &miss->flow); - ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub); - if (!subfacet->actions || subfacet->slow) { - subfacet_make_actions(subfacet, packet, &odp_actions); + if (!subfacet->actions) { + subfacet_make_actions(subfacet, packet); + } else if (subfacet->slow) { + struct action_xlate_ctx ctx; + + action_xlate_ctx_init(&ctx, ofproto, &facet->flow, + &subfacet->initial_vals, facet->rule, 0, + packet); + xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts, + facet->rule->up.ofpacts_len); } dpif_flow_stats_extract(&facet->flow, packet, now, &stats); @@ -3322,20 +3756,10 @@ handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, struct dpif_execute *execute = &op->dpif_op.u.execute; init_flow_miss_execute_op(miss, packet, op); - op->subfacet = subfacet; - if (!subfacet->slow) { - execute->actions = subfacet->actions; - execute->actions_len = subfacet->actions_len; - ofpbuf_uninit(&odp_actions); - } else { - execute->actions = odp_actions.data; - execute->actions_len = odp_actions.size; - op->garbage = ofpbuf_get_uninit_pointer(&odp_actions); - } + execute->actions = subfacet->actions; + execute->actions_len = subfacet->actions_len; (*n_ops)++; - } else { - ofpbuf_uninit(&odp_actions); } } @@ -3344,7 +3768,8 @@ handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, struct flow_miss_op *op = &ops[(*n_ops)++]; struct dpif_flow_put *put = &op->dpif_op.u.flow_put; - op->subfacet = subfacet; + subfacet->path = want_path; + op->garbage = NULL; op->dpif_op.type = DPIF_OP_FLOW_PUT; put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; @@ -3381,7 +3806,13 @@ handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops, if (!facet) { struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow); - if (!flow_miss_should_make_facet(ofproto, miss, hash)) { + /* 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 alltogether. */ + if (miss->key_fitness == ODP_FIT_TOO_LITTLE + || !flow_miss_should_make_facet(ofproto, miss, hash)) { handle_flow_miss_without_facet(miss, rule, ops, n_ops); return; } @@ -3394,6 +3825,47 @@ handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops, handle_flow_miss_with_facet(miss, facet, now, ops, n_ops); } +static struct drop_key * +drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key, + size_t key_len) +{ + struct drop_key *drop_key; + + HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0), + &backer->drop_keys) { + if (drop_key->key_len == key_len + && !memcmp(drop_key->key, key, key_len)) { + return drop_key; + } + } + return NULL; +} + +static void +drop_key_clear(struct dpif_backer *backer) +{ + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15); + struct drop_key *drop_key, *next; + + HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) { + int error; + + error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len, + NULL); + if (error && !VLOG_DROP_WARN(&rl)) { + struct ds ds = DS_EMPTY_INITIALIZER; + odp_flow_key_format(drop_key->key, drop_key->key_len, &ds); + VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error), + ds_cstr(&ds)); + ds_destroy(&ds); + } + + hmap_remove(&backer->drop_keys, &drop_key->hmap_node); + free(drop_key->key); + free(drop_key); + } +} + /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key' * respectively), populates 'flow' with the result of odp_flow_key_to_flow(). * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as @@ -3411,10 +3883,15 @@ handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops, * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes * a VLAN header onto 'packet' (if it is nonnull). * - * Optionally, if nonnull, sets '*initial_tci' to the VLAN TCI with which the - * packet was really received, that is, the actual VLAN TCI extracted by - * odp_flow_key_to_flow(). (This differs from the value returned in - * flow->vlan_tci only for packets received on VLAN splinters.) + * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci' + * to the VLAN TCI with which the packet was really received, that is, the + * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from + * the value returned in flow->vlan_tci only for packets received on + * VLAN splinters.) + * + * Similarly, this function also includes some logic to help with tunnels. It + * may modify 'flow' as necessary to make the tunneling implementation + * transparent to the upcall processing logic. * * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport, * or some other positive errno if there are other problems. */ @@ -3423,11 +3900,11 @@ ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet, const struct nlattr *key, size_t key_len, struct flow *flow, enum odp_key_fitness *fitnessp, struct ofproto_dpif **ofproto, uint32_t *odp_in_port, - ovs_be16 *initial_tci) + struct initial_vals *initial_vals) { const struct ofport_dpif *port; enum odp_key_fitness fitness; - int error; + int error = ENODEV; fitness = odp_flow_key_to_flow(key, key_len, flow); if (fitness == ODP_FIT_ERROR) { @@ -3435,52 +3912,56 @@ ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet, goto exit; } - if (initial_tci) { - *initial_tci = flow->vlan_tci; + if (initial_vals) { + initial_vals->vlan_tci = flow->vlan_tci; } if (odp_in_port) { *odp_in_port = flow->in_port; } - port = odp_port_to_ofport(backer, flow->in_port); + port = (tnl_port_should_receive(flow) + ? ofport_dpif_cast(tnl_port_receive(flow)) + : odp_port_to_ofport(backer, flow->in_port)); + flow->in_port = port ? port->up.ofp_port : OFPP_NONE; if (!port) { - flow->in_port = OFPP_NONE; - error = ofproto ? ENODEV : 0; goto exit; } - if (ofproto) { - *ofproto = ofproto_dpif_cast(port->up.ofproto); - } + /* XXX: Since the tunnel module is not scoped per backer, for a tunnel port + * it's theoretically possible that we'll receive an ofport belonging to an + * entirely different datapath. In practice, this can't happen because no + * platforms has two separate datapaths which each support tunneling. */ + ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer); - flow->in_port = port->up.ofp_port; if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) { if (packet) { - /* Make the packet resemble the flow, so that it gets sent to an - * OpenFlow controller properly, so that it looks correct for - * sFlow, and so that flow_extract() will get the correct vlan_tci - * if it is called on 'packet'. + /* Make the packet resemble the flow, so that it gets sent to + * an OpenFlow controller properly, so that it looks correct + * for sFlow, and so that flow_extract() will get the correct + * vlan_tci if it is called on 'packet'. * * The allocated space inside 'packet' probably also contains - * 'key', that is, both 'packet' and 'key' are probably part of a - * struct dpif_upcall (see the large comment on that structure - * definition), so pushing data on 'packet' is in general not a - * good idea since it could overwrite 'key' or free it as a side - * effect. However, it's OK in this special case because we know - * that 'packet' is inside a Netlink attribute: pushing 4 bytes - * will just overwrite the 4-byte "struct nlattr", which is fine - * since we don't need that header anymore. */ + * 'key', that is, both 'packet' and 'key' are probably part of + * a struct dpif_upcall (see the large comment on that + * structure definition), so pushing data on 'packet' is in + * general not a good idea since it could overwrite 'key' or + * free it as a side effect. However, it's OK in this special + * case because we know that 'packet' is inside a Netlink + * attribute: pushing 4 bytes will just overwrite the 4-byte + * "struct nlattr", which is fine since we don't need that + * header anymore. */ eth_push_vlan(packet, flow->vlan_tci); } - - /* Let the caller know that we can't reproduce 'key' from 'flow'. */ - if (fitness == ODP_FIT_PERFECT) { - fitness = ODP_FIT_TOO_MUCH; - } + /* We can't reproduce 'key' from 'flow'. */ + fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness; } error = 0; + if (ofproto) { + *ofproto = ofproto_dpif_cast(port->up.ofproto); + } + exit: if (fitnessp) { *fitnessp = fitness; @@ -3524,24 +4005,43 @@ handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls, error = ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, &flow, &miss->key_fitness, - &ofproto, &odp_in_port, &miss->initial_tci); + &ofproto, &odp_in_port, &miss->initial_vals); if (error == ENODEV) { + struct drop_key *drop_key; + /* Received packet on 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. */ + * 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 port %"PRIu32, flow.in_port); + + drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len); + if (!drop_key) { + 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)); + dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY, + drop_key->key, drop_key->key_len, NULL, 0, NULL); + } + 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, &miss->flow, hash); + existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash); if (!existing_miss) { hmap_insert(&todo, &miss->hmap_node, hash); miss->ofproto = ofproto; @@ -3564,7 +4064,7 @@ handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls, HMAP_FOR_EACH (miss, hmap_node, &todo) { handle_flow_miss(miss, flow_miss_ops, &n_ops); } - assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); + ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); /* Execute batch. */ for (i = 0; i < n_ops; i++) { @@ -3572,32 +4072,18 @@ handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls, } dpif_operate(backer->dpif, dpif_ops, n_ops); - /* Free memory and update facets. */ + /* Free memory. */ for (i = 0; i < n_ops; i++) { - struct flow_miss_op *op = &flow_miss_ops[i]; - - switch (op->dpif_op.type) { - case DPIF_OP_EXECUTE: - break; - - case DPIF_OP_FLOW_PUT: - if (!op->dpif_op.error) { - op->subfacet->path = subfacet_want_path(op->subfacet->slow); - } - break; - - case DPIF_OP_FLOW_DEL: - NOT_REACHED(); - } - - free(op->garbage); + free(flow_miss_ops[i].garbage); } hmap_destroy(&todo); } -static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL } +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. */ @@ -3615,17 +4101,34 @@ classify_upcall(const struct dpif_upcall *upcall) } /* "action" upcalls need a closer look. */ - memcpy(&cookie, &upcall->userdata, sizeof(cookie)); - switch (cookie.type) { - case USER_ACTION_COOKIE_SFLOW: + 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; - - case USER_ACTION_COOKIE_SLOW_PATH: + } else if (userdata_len == sizeof cookie.slow_path + && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) { return MISS_UPCALL; - - case USER_ACTION_COOKIE_UNSPEC: - default: - VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata); + } 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; } } @@ -3645,11 +4148,56 @@ handle_sflow_upcall(struct dpif_backer *backer, return; } - memcpy(&cookie, &upcall->userdata, sizeof(cookie)); + 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) +{ + struct ofproto_dpif *ofproto; + union user_action_cookie cookie; + struct flow flow; + + if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, + &flow, NULL, &ofproto, NULL, 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 (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, + &flow, NULL, &ofproto, NULL, 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) { @@ -3660,7 +4208,7 @@ handle_upcalls(struct dpif_backer *backer, unsigned int max_batch) int n_misses; int i; - assert(max_batch <= FLOW_MISS_MAX_BATCH); + ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH); n_misses = 0; for (n_processed = 0; n_processed < max_batch; n_processed++) { @@ -3687,6 +4235,16 @@ handle_upcalls(struct dpif_backer *backer, unsigned int max_batch) 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; @@ -3721,18 +4279,24 @@ expire(struct dpif_backer *backer) struct ofproto_dpif *ofproto; int max_idle = INT32_MAX; + /* Periodically clear out the drop keys in an effort to keep them + * relatively few. */ + drop_key_clear(backer); + /* Update stats for each flow in the backer. */ update_stats(backer); HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { - struct rule_dpif *rule, *next_rule; - struct oftable *table; + struct rule *rule, *next_rule; int dp_max_idle; if (ofproto->backer != backer) { continue; } + /* Keep track of the max number of flows per ofproto_dpif. */ + update_max_subfacet_count(ofproto); + /* Expire subfacets that have been idle too long. */ dp_max_idle = subfacet_max_idle(ofproto); expire_subfacets(ofproto, dp_max_idle); @@ -3741,13 +4305,9 @@ expire(struct dpif_backer *backer) /* Expire OpenFlow flows whose idle_timeout or hard_timeout * has passed. */ - OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { - struct cls_cursor cursor; - - cls_cursor_init(&cursor, &table->cls, NULL); - CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { - rule_expire(rule); - } + LIST_FOR_EACH_SAFE (rule, next_rule, expirable, + &ofproto->up.expirable) { + rule_expire(rule_dpif_cast(rule)); } /* All outstanding data in existing flows has been accounted, so it's a @@ -3757,7 +4317,7 @@ expire(struct dpif_backer *backer) HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { if (bundle->bond) { - bond_rebalance(bundle->bond, &ofproto->revalidate_set); + bond_rebalance(bundle->bond, &backer->revalidate_set); } } } @@ -3798,7 +4358,6 @@ update_subfacet_stats(struct subfacet *subfacet, facet_account(facet); facet->accounted_bytes = facet->byte_count; } - facet_push_stats(facet); } /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing @@ -3830,6 +4389,11 @@ delete_unexpected_flow(struct ofproto_dpif *ofproto, * avoided by calling update_stats() whenever rules are created or * deleted. However, the performance impact of making so many calls to the * datapath do not justify the benefit of having perfectly accurate statistics. + * + * In addition, this function maintains per ofproto flow hit counts. The patch + * port is not treated specially. e.g. A packet ingress from br0 patched into + * br1 will increase the hit count of br0 by 1, however, does not affect + * the hit or miss counts of br1. */ static void update_stats(struct dpif_backer *backer) @@ -3837,38 +4401,39 @@ update_stats(struct dpif_backer *backer) const struct dpif_flow_stats *stats; struct dpif_flow_dump dump; const struct nlattr *key; + struct ofproto_dpif *ofproto; size_t key_len; dpif_flow_dump_start(&dump, backer->dpif); while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { struct flow flow; struct subfacet *subfacet; - enum odp_key_fitness fitness; - struct ofproto_dpif *ofproto; - struct ofport_dpif *port; + struct ofport_dpif *ofport; uint32_t key_hash; - fitness = odp_flow_key_to_flow(key, key_len, &flow); - if (fitness == ODP_FIT_ERROR) { + if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto, + NULL, NULL)) { continue; } - port = odp_port_to_ofport(backer, flow.in_port); - if (!port) { - /* This flow is for a port for which we couldn't associate an - * ofproto. This can happen if a port is removed while - * traffic is being received. Ignore this flow, since it - * will get timed out. */ - continue; + ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets); + ofproto->n_update_stats++; + + ofport = get_ofp_port(ofproto, flow.in_port); + if (ofport && ofport->tnl_port) { + netdev_vport_inc_rx(ofport->up.netdev, stats); } - ofproto = ofproto_dpif_cast(port->up.ofproto); - flow.in_port = port->up.ofp_port; key_hash = odp_flow_key_hash(key, key_len); - - subfacet = subfacet_find(ofproto, key, key_len, key_hash, &flow); + subfacet = subfacet_find(ofproto, key, key_len, key_hash); switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) { case SF_FAST_PATH: + /* Update ofproto_dpif's hit count. */ + if (stats->n_packets > subfacet->dp_packet_count) { + uint64_t delta = stats->n_packets - subfacet->dp_packet_count; + dpif_stats_update_hit_count(ofproto, delta); + } + update_subfacet_stats(subfacet, stats); break; @@ -3881,8 +4446,14 @@ update_stats(struct dpif_backer *backer) delete_unexpected_flow(ofproto, key, key_len); break; } + run_fast_rl(); } dpif_flow_dump_done(&dump); + + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { + update_moving_averages(ofproto); + } + } /* Calculates and returns the number of milliseconds of idle time after which @@ -3992,7 +4563,7 @@ expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) &ofproto->subfacets) { long long int cutoff; - cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP) + cutoff = (subfacet->slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP) ? special_cutoff : normal_cutoff); if (subfacet->used < cutoff) { @@ -4079,6 +4650,8 @@ facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash) netflow_flow_init(&facet->nf_flow); netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); + facet->learn_rl = time_msec() + 500; + return facet; } @@ -4089,9 +4662,7 @@ facet_free(struct facet *facet) } /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on - * 'packet', which arrived on 'in_port'. - * - * Takes ownership of 'packet'. */ + * '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, @@ -4107,8 +4678,6 @@ execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, error = dpif_execute(ofproto->backer->dpif, key.data, key.size, odp_actions, actions_len, packet); - - ofpbuf_delete(packet); return !error; } @@ -4125,7 +4694,7 @@ facet_remove(struct facet *facet) struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); struct subfacet *subfacet, *next_subfacet; - assert(!list_is_empty(&facet->subfacets)); + ovs_assert(!list_is_empty(&facet->subfacets)); /* First uninstall all of the subfacets to get final statistics. */ LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { @@ -4155,8 +4724,17 @@ static void facet_learn(struct facet *facet) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); + struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets), + struct subfacet, list_node); + long long int now = time_msec(); struct action_xlate_ctx ctx; + if (!facet->has_fin_timeout && now < facet->learn_rl) { + return; + } + + facet->learn_rl = now + 500; + if (!facet->has_learn && !facet->has_normal && (!facet->has_fin_timeout @@ -4165,7 +4743,7 @@ facet_learn(struct facet *facet) } action_xlate_ctx_init(&ctx, ofproto, &facet->flow, - facet->flow.vlan_tci, + &subfacet->initial_vals, facet->rule, facet->tcp_flags, NULL); ctx.may_learn = true; xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts, @@ -4176,7 +4754,7 @@ static void facet_account(struct facet *facet) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); - struct subfacet *subfacet; + struct subfacet *subfacet = facet_get_subfacet(facet); const struct nlattr *a; unsigned int left; ovs_be16 vlan_tci; @@ -4195,8 +4773,6 @@ facet_account(struct facet *facet) * * We use the actions from an arbitrary subfacet because they should all * be equally valid for our purpose. */ - subfacet = CONTAINER_OF(list_front(&facet->subfacets), - struct subfacet, list_node); vlan_tci = facet->flow.vlan_tci; NL_ATTR_FOR_EACH_UNSAFE (a, left, subfacet->actions, subfacet->actions_len) { @@ -4255,8 +4831,8 @@ facet_flush_stats(struct facet *facet) struct subfacet *subfacet; LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { - assert(!subfacet->dp_byte_count); - assert(!subfacet->dp_packet_count); + ovs_assert(!subfacet->dp_byte_count); + ovs_assert(!subfacet->dp_packet_count); } facet_push_stats(facet); @@ -4274,9 +4850,6 @@ facet_flush_stats(struct facet *facet) netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); } - facet->rule->packet_count += facet->packet_count; - facet->rule->byte_count += facet->byte_count; - /* Reset counters to prevent double counting if 'facet' ever gets * reinstalled. */ facet_reset_counters(facet); @@ -4321,14 +4894,24 @@ facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow, facet = facet_find(ofproto, flow, hash); if (facet - && (ofproto->need_revalidate - || tag_set_intersects(&ofproto->revalidate_set, facet->tags))) { - facet_revalidate(facet); + && (ofproto->backer->need_revalidate + || tag_set_intersects(&ofproto->backer->revalidate_set, + facet->tags)) + && !facet_revalidate(facet)) { + facet = facet_find(ofproto, flow, hash); } return facet; } +/* Return a subfacet from 'facet'. A facet consists of one or more + * subfacets, and this function returns one of them. */ +static struct subfacet *facet_get_subfacet(struct facet *facet) +{ + return CONTAINER_OF(list_front(&facet->subfacets), struct subfacet, + list_node); +} + static const char * subfacet_path_to_string(enum subfacet_path path) { @@ -4409,13 +4992,11 @@ facet_check_consistency(struct facet *facet) ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { enum subfacet_path want_path; - struct odputil_keybuf keybuf; struct action_xlate_ctx ctx; - struct ofpbuf key; struct ds s; action_xlate_ctx_init(&ctx, ofproto, &facet->flow, - subfacet->initial_tci, rule, 0, NULL); + &subfacet->initial_vals, rule, 0, NULL); xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions); @@ -4426,11 +5007,6 @@ facet_check_consistency(struct facet *facet) } want_path = subfacet_want_path(subfacet->slow); - if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) { - /* The actions for slow-path flows may legitimately vary from one - * packet to the next. We're done. */ - continue; - } if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) { continue; @@ -4447,8 +5023,7 @@ facet_check_consistency(struct facet *facet) } ds_init(&s); - subfacet_get_key(subfacet, &keybuf, &key); - odp_flow_key_format(key.data, key.size, &s); + odp_flow_key_format(subfacet->key, subfacet->key_len, &s); ds_put_cstr(&s, ": inconsistency in subfacet"); if (want_path != subfacet->path) { @@ -4486,8 +5061,13 @@ facet_check_consistency(struct facet *facet) * 'facet' to the new rule and recompiles its actions. * * - If the rule found is the same as 'facet''s current rule, leaves 'facet' - * where it is and recompiles its actions anyway. */ -static void + * where it is and recompiles its actions anyway. + * + * - If any of 'facet''s subfacets correspond to a new flow according to + * ofproto_receive(), 'facet' is removed. + * + * Returns true if 'facet' is still valid. False if 'facet' was removed. */ +static bool facet_revalidate(struct facet *facet) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); @@ -4507,6 +5087,25 @@ facet_revalidate(struct facet *facet) COVERAGE_INC(facet_revalidate); + /* Check that child subfacets still correspond to this facet. Tunnel + * configuration changes could cause a subfacet's OpenFlow in_port to + * change. */ + LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { + struct ofproto_dpif *recv_ofproto; + struct flow recv_flow; + int error; + + error = ofproto_receive(ofproto->backer, NULL, subfacet->key, + subfacet->key_len, &recv_flow, NULL, + &recv_ofproto, NULL, NULL); + if (error + || recv_ofproto != ofproto + || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) { + facet_remove(facet); + return false; + } + } + new_rule = rule_dpif_lookup(ofproto, &facet->flow); /* Calculate new datapath actions. @@ -4522,19 +5121,16 @@ facet_revalidate(struct facet *facet) memset(&ctx, 0, sizeof ctx); ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { - enum slow_path_reason slow; - action_xlate_ctx_init(&ctx, ofproto, &facet->flow, - subfacet->initial_tci, new_rule, 0, NULL); + &subfacet->initial_vals, new_rule, 0, NULL); xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len, &odp_actions); - slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; - if (subfacet_should_install(subfacet, slow, &odp_actions)) { + if (subfacet_should_install(subfacet, ctx.slow, &odp_actions)) { struct dpif_flow_stats stats; - subfacet_install(subfacet, - odp_actions.data, odp_actions.size, &stats, slow); + subfacet_install(subfacet, odp_actions.data, odp_actions.size, + &stats, ctx.slow); subfacet_update_stats(subfacet, &stats); if (!new_actions) { @@ -4564,7 +5160,7 @@ facet_revalidate(struct facet *facet) i = 0; LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { - subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; + subfacet->slow = ctx.slow; if (new_actions && new_actions[i].odp_actions) { free(subfacet->actions); @@ -4583,6 +5179,8 @@ facet_revalidate(struct facet *facet) facet->used = new_rule->up.created; facet->prev_used = facet->used; } + + return true; } /* Updates 'facet''s used time. Caller is responsible for calling @@ -4613,9 +5211,9 @@ facet_push_stats(struct facet *facet) { struct dpif_flow_stats stats; - assert(facet->packet_count >= facet->prev_packet_count); - assert(facet->byte_count >= facet->prev_byte_count); - assert(facet->used >= facet->prev_used); + ovs_assert(facet->packet_count >= facet->prev_packet_count); + ovs_assert(facet->byte_count >= facet->prev_byte_count); + ovs_assert(facet->used >= facet->prev_used); stats.n_packets = facet->packet_count - facet->prev_packet_count; stats.n_bytes = facet->byte_count - facet->prev_byte_count; @@ -4627,13 +5225,44 @@ facet_push_stats(struct facet *facet) facet->prev_byte_count = facet->byte_count; facet->prev_used = facet->used; - flow_push_stats(facet->rule, &facet->flow, &stats); + rule_credit_stats(facet->rule, &stats); + flow_push_stats(facet, &stats); update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto), facet->mirrors, stats.n_packets, stats.n_bytes); } } +static void +push_all_stats__(bool run_fast) +{ + static long long int rl = LLONG_MIN; + struct ofproto_dpif *ofproto; + + if (time_msec() < rl) { + return; + } + + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { + struct facet *facet; + + HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { + facet_push_stats(facet); + if (run_fast) { + run_fast_rl(); + } + } + } + + rl = time_msec() + 100; +} + +static void +push_all_stats(void) +{ + push_all_stats__(true); +} + static void rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats) { @@ -4642,19 +5271,20 @@ rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats) ofproto_rule_update_used(&rule->up, stats->used); } -/* Pushes flow statistics to the rules which 'flow' resubmits into given - * 'rule''s actions and mirrors. */ +/* Pushes flow statistics to the rules which 'facet->flow' resubmits + * into given 'facet->rule''s actions and mirrors. */ static void -flow_push_stats(struct rule_dpif *rule, - const struct flow *flow, const struct dpif_flow_stats *stats) +flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats) { + struct rule_dpif *rule = facet->rule; struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); + struct subfacet *subfacet = facet_get_subfacet(facet); struct action_xlate_ctx ctx; ofproto_rule_update_used(&rule->up, stats->used); - action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule, - 0, NULL); + action_xlate_ctx_init(&ctx, ofproto, &facet->flow, + &subfacet->initial_vals, rule, 0, NULL); ctx.resubmit_stats = stats; xlate_actions_for_side_effects(&ctx, rule->up.ofpacts, rule->up.ofpacts_len); @@ -4664,17 +5294,14 @@ flow_push_stats(struct rule_dpif *rule, static struct subfacet * subfacet_find(struct ofproto_dpif *ofproto, - const struct nlattr *key, size_t key_len, uint32_t key_hash, - const struct flow *flow) + const struct nlattr *key, size_t key_len, uint32_t key_hash) { struct subfacet *subfacet; HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash, &ofproto->subfacets) { - if (subfacet->key - ? (subfacet->key_len == key_len - && !memcmp(key, subfacet->key, key_len)) - : flow_equal(flow, &subfacet->facet->flow)) { + if (subfacet->key_len == key_len + && !memcmp(key, subfacet->key, key_len)) { return subfacet; } } @@ -4706,8 +5333,7 @@ subfacet_create(struct facet *facet, struct flow_miss *miss, if (list_is_empty(&facet->subfacets)) { subfacet = &facet->one_subfacet; } else { - subfacet = subfacet_find(ofproto, key, key_len, key_hash, - &facet->flow); + subfacet = subfacet_find(ofproto, key, key_len, key_hash); if (subfacet) { if (subfacet->facet == facet) { return subfacet; @@ -4725,25 +5351,20 @@ subfacet_create(struct facet *facet, struct flow_miss *miss, list_push_back(&facet->subfacets, &subfacet->list_node); subfacet->facet = facet; subfacet->key_fitness = key_fitness; - if (key_fitness != ODP_FIT_PERFECT) { - subfacet->key = xmemdup(key, key_len); - subfacet->key_len = key_len; - } else { - subfacet->key = NULL; - subfacet->key_len = 0; - } + subfacet->key = xmemdup(key, key_len); + subfacet->key_len = key_len; subfacet->used = now; + subfacet->created = now; subfacet->dp_packet_count = 0; subfacet->dp_byte_count = 0; subfacet->actions_len = 0; subfacet->actions = NULL; - subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE - ? SLOW_MATCH - : 0); + subfacet->slow = 0; subfacet->path = SF_NOT_INSTALLED; - subfacet->initial_tci = miss->initial_tci; + subfacet->initial_vals = miss->initial_vals; subfacet->odp_in_port = miss->odp_in_port; + ofproto->subfacet_add_count++; return subfacet; } @@ -4755,6 +5376,10 @@ subfacet_destroy__(struct subfacet *subfacet) struct facet *facet = subfacet->facet; struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); + /* Update ofproto stats before uninstall the subfacet. */ + ofproto->subfacet_del_count++; + ofproto->total_subfacet_life_span += (time_msec() - subfacet->created); + subfacet_uninstall(subfacet); hmap_remove(&ofproto->subfacets, &subfacet->hmap_node); list_remove(&subfacet->list_node); @@ -4784,18 +5409,15 @@ static void subfacet_destroy_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n) { - struct odputil_keybuf keybufs[SUBFACET_DESTROY_MAX_BATCH]; struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH]; struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH]; - struct ofpbuf keys[SUBFACET_DESTROY_MAX_BATCH]; struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH]; int i; for (i = 0; i < n; i++) { ops[i].type = DPIF_OP_FLOW_DEL; - subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]); - ops[i].u.flow_del.key = keys[i].data; - ops[i].u.flow_del.key_len = keys[i].size; + ops[i].u.flow_del.key = subfacets[i]->key; + ops[i].u.flow_del.key_len = subfacets[i]->key_len; ops[i].u.flow_del.stats = &stats[i]; opsp[i] = &ops[i]; } @@ -4805,43 +5427,26 @@ subfacet_destroy_batch(struct ofproto_dpif *ofproto, subfacet_reset_dp_stats(subfacets[i], &stats[i]); subfacets[i]->path = SF_NOT_INSTALLED; subfacet_destroy(subfacets[i]); + run_fast_rl(); } } -/* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes - * that can be used to refer to 'subfacet'. The caller must provide 'keybuf' - * for use as temporary storage. */ -static void -subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf, - struct ofpbuf *key) -{ - - if (!subfacet->key) { - struct flow *flow = &subfacet->facet->flow; - - ofpbuf_use_stack(key, keybuf, sizeof *keybuf); - odp_flow_key_from_flow(key, flow, subfacet->odp_in_port); - } else { - ofpbuf_use_const(key, subfacet->key, subfacet->key_len); - } -} - -/* Composes the datapath actions for 'subfacet' based on its rule's actions. - * Translates the actions into 'odp_actions', which the caller must have - * initialized and is responsible for uninitializing. */ +/* Composes the datapath actions for 'subfacet' based on its rule's actions. */ static void -subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet, - struct ofpbuf *odp_actions) +subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet) { struct facet *facet = subfacet->facet; struct rule_dpif *rule = facet->rule; struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); struct action_xlate_ctx ctx; + struct ofpbuf odp_actions; + uint64_t stub[1024 / 8]; - action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci, - rule, 0, packet); - xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions); + ofpbuf_use_stub(&odp_actions, stub, sizeof stub); + action_xlate_ctx_init(&ctx, ofproto, &facet->flow, + &subfacet->initial_vals, rule, 0, packet); + xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions); facet->tags = ctx.tags; facet->has_learn = ctx.has_learn; facet->has_normal = ctx.has_normal; @@ -4849,13 +5454,11 @@ subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet, facet->nf_flow.output_iface = ctx.nf_output_iface; facet->mirrors = ctx.mirrors; - subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow; - if (subfacet->actions_len != odp_actions->size - || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) { - free(subfacet->actions); - subfacet->actions_len = odp_actions->size; - subfacet->actions = xmemdup(odp_actions->data, odp_actions->size); - } + subfacet->slow = ctx.slow; + + ovs_assert(!subfacet->actions); + subfacet->actions_len = odp_actions.size; + subfacet->actions = ofpbuf_steal_data(&odp_actions); } /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len' @@ -4874,9 +5477,7 @@ subfacet_install(struct subfacet *subfacet, struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); enum subfacet_path path = subfacet_want_path(slow); uint64_t slow_path_stub[128 / 8]; - struct odputil_keybuf keybuf; enum dpif_flow_put_flags flags; - struct ofpbuf key; int ret; flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; @@ -4890,9 +5491,8 @@ subfacet_install(struct subfacet *subfacet, &actions, &actions_len); } - subfacet_get_key(subfacet, &keybuf, &key); - ret = dpif_flow_put(ofproto->backer->dpif, flags, key.data, key.size, - actions, actions_len, stats); + ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key, + subfacet->key_len, actions, actions_len, stats); if (stats) { subfacet_reset_dp_stats(subfacet, stats); @@ -4918,22 +5518,19 @@ subfacet_uninstall(struct subfacet *subfacet) if (subfacet->path != SF_NOT_INSTALLED) { struct rule_dpif *rule = subfacet->facet->rule; struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); - struct odputil_keybuf keybuf; struct dpif_flow_stats stats; - struct ofpbuf key; int error; - subfacet_get_key(subfacet, &keybuf, &key); - error = dpif_flow_del(ofproto->backer->dpif, - key.data, key.size, &stats); + error = dpif_flow_del(ofproto->backer->dpif, subfacet->key, + subfacet->key_len, &stats); subfacet_reset_dp_stats(subfacet, &stats); if (!error) { subfacet_update_stats(subfacet, &stats); } subfacet->path = SF_NOT_INSTALLED; } else { - assert(subfacet->dp_packet_count == 0); - assert(subfacet->dp_byte_count == 0); + ovs_assert(subfacet->dp_packet_count == 0); + ovs_assert(subfacet->dp_byte_count == 0); } } @@ -4985,7 +5582,6 @@ subfacet_update_stats(struct subfacet *subfacet, facet->packet_count += stats->n_packets; facet->byte_count += stats->n_bytes; facet->tcp_flags |= stats->tcp_flags; - facet_push_stats(facet); netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); } } @@ -5143,31 +5739,26 @@ static void rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) { struct rule_dpif *rule = rule_dpif_cast(rule_); - struct facet *facet; + + /* 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); /* Start from historical data for 'rule' itself that are no longer tracked * in facets. This counts, for example, facets that have expired. */ *packets = rule->packet_count; *bytes = rule->byte_count; - - /* Add any statistics that are tracked by facets. This includes - * statistical data recently updated by ofproto_update_stats() as well as - * stats for packets that were executed "by hand" via dpif_execute(). */ - LIST_FOR_EACH (facet, list_node, &rule->facets) { - *packets += facet->packet_count; - *bytes += facet->byte_count; - } } -static enum ofperr -rule_execute(struct rule *rule_, const struct flow *flow, - struct ofpbuf *packet) +static void +rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow, + struct ofpbuf *packet) { - struct rule_dpif *rule = rule_dpif_cast(rule_); struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); - + struct initial_vals initial_vals; struct dpif_flow_stats stats; - struct action_xlate_ctx ctx; uint64_t odp_actions_stub[1024 / 8]; struct ofpbuf odp_actions; @@ -5175,8 +5766,9 @@ rule_execute(struct rule *rule_, const struct flow *flow, dpif_flow_stats_extract(flow, packet, time_msec(), &stats); rule_credit_stats(rule, &stats); + initial_vals.vlan_tci = flow->vlan_tci; ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); - action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, + action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, rule, stats.tcp_flags, packet); ctx.resubmit_stats = &stats; xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions); @@ -5185,7 +5777,14 @@ rule_execute(struct rule *rule_, const struct flow *flow, odp_actions.size, packet); ofpbuf_uninit(&odp_actions); +} +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; } @@ -5203,29 +5802,33 @@ rule_modify_actions(struct rule *rule_) static int send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) { - const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); + 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; - uint32_t odp_port; + struct ofpact_output output; + struct action_xlate_ctx ctx; struct flow flow; int error; - flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow); - odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port, - flow.vlan_tci); - if (odp_port != ofport->odp_port) { - eth_pop_vlan(packet); - flow.vlan_tci = htons(0); - } - + ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); - odp_flow_key_from_flow(&key, &flow, - ofp_port_to_odp_port(ofproto, flow.in_port)); - ofpbuf_init(&odp_actions, 32); - compose_sflow_action(ofproto, &odp_actions, &flow, odp_port); + /* Use OFPP_NONE as the in_port to avoid special packet processing. */ + flow_extract(packet, 0, 0, NULL, OFPP_NONE, &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; + + action_xlate_ctx_init(&ctx, ofproto, &flow, NULL, NULL, 0, packet); + ctx.resubmit_stats = &stats; + xlate_actions(&ctx, &output.ofpact, sizeof output, &odp_actions); - nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port); error = dpif_execute(ofproto->backer->dpif, key.data, key.size, odp_actions.data, odp_actions.size, @@ -5233,15 +5836,19 @@ send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) ofpbuf_uninit(&odp_actions); if (error) { - VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)", - ofproto->up.name, odp_port, strerror(error)); + VLOG_WARN_RL(&rl, "%s: failed to send packet on port %s (%s)", + ofproto->up.name, netdev_get_name(ofport->up.netdev), + strerror(error)); } - ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0); + + ofproto->stats.tx_packets++; + ofproto->stats.tx_bytes += packet->size; return error; } /* OpenFlow to datapath action translation. */ +static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *); static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len, struct action_xlate_ctx *); static void xlate_normal(struct action_xlate_ctx *); @@ -5270,11 +5877,12 @@ compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow, cookie.slow_path.reason = slow; ofpbuf_use_stack(&buf, stub, stub_size); - if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) { - uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT16_MAX); - odp_put_userspace_action(pid, &cookie, &buf); + if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) { + uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX); + odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf); } else { - put_userspace_action(ofproto, &buf, flow, &cookie); + put_userspace_action(ofproto, &buf, flow, &cookie, + sizeof cookie.slow_path); } *actionsp = buf.data; *actions_lenp = buf.size; @@ -5284,14 +5892,43 @@ static size_t put_userspace_action(const struct ofproto_dpif *ofproto, struct ofpbuf *odp_actions, const struct flow *flow, - const union user_action_cookie *cookie) + const union user_action_cookie *cookie, + const size_t cookie_size) { uint32_t pid; pid = dpif_port_get_pid(ofproto->backer->dpif, ofp_port_to_odp_port(ofproto, flow->in_port)); - return odp_put_userspace_action(pid, cookie, odp_actions); + return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions); +} + +/* Compose SAMPLE action for sFlow or IPFIX. The given probability is + * the number of packets out of UINT32_MAX to sample. The given + * cookie is passed back in the callback for each sampled packet. + */ +static size_t +compose_sample_action(const struct ofproto_dpif *ofproto, + struct ofpbuf *odp_actions, + const struct flow *flow, + const uint32_t probability, + const union user_action_cookie *cookie, + const size_t cookie_size) +{ + size_t sample_offset, actions_offset; + int cookie_offset; + + sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE); + + nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability); + + actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS); + cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie, + cookie_size); + + nl_msg_end_nested(odp_actions, actions_offset); + nl_msg_end_nested(odp_actions, sample_offset); + return cookie_offset; } static void @@ -5326,7 +5963,7 @@ compose_sflow_cookie(const struct ofproto_dpif *ofproto, } } -/* Compose SAMPLE action for sFlow. */ +/* Compose SAMPLE action for sFlow bridge sampling. */ static size_t compose_sflow_action(const struct ofproto_dpif *ofproto, struct ofpbuf *odp_actions, @@ -5335,32 +5972,60 @@ compose_sflow_action(const struct ofproto_dpif *ofproto, { uint32_t probability; union user_action_cookie cookie; - size_t sample_offset, actions_offset; - int cookie_offset; if (!ofproto->sflow || flow->in_port == OFPP_NONE) { return 0; } - sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE); - - /* Number of packets out of UINT_MAX to sample. */ probability = dpif_sflow_get_probability(ofproto->sflow); - nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability); - - actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS); compose_sflow_cookie(ofproto, htons(0), odp_port, odp_port == OVSP_NONE ? 0 : 1, &cookie); - cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie); - nl_msg_end_nested(odp_actions, actions_offset); - nl_msg_end_nested(odp_actions, sample_offset); - return cookie_offset; + return compose_sample_action(ofproto, odp_actions, flow, probability, + &cookie, sizeof cookie.sflow); +} + +static void +compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id, + uint32_t obs_domain_id, uint32_t obs_point_id, + union user_action_cookie *cookie) +{ + cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE; + cookie->flow_sample.probability = probability; + cookie->flow_sample.collector_set_id = collector_set_id; + cookie->flow_sample.obs_domain_id = obs_domain_id; + cookie->flow_sample.obs_point_id = obs_point_id; } -/* SAMPLE action must be first action in any given list of actions. - * At this point we do not have all information required to build it. So try to - * build sample action as complete as possible. */ +static void +compose_ipfix_cookie(union user_action_cookie *cookie) +{ + cookie->type = USER_ACTION_COOKIE_IPFIX; +} + +/* Compose SAMPLE action for IPFIX bridge sampling. */ +static void +compose_ipfix_action(const struct ofproto_dpif *ofproto, + struct ofpbuf *odp_actions, + const struct flow *flow) +{ + uint32_t probability; + union user_action_cookie cookie; + + if (!ofproto->ipfix || flow->in_port == OFPP_NONE) { + return; + } + + probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix); + compose_ipfix_cookie(&cookie); + + compose_sample_action(ofproto, odp_actions, flow, probability, + &cookie, sizeof cookie.ipfix); +} + +/* SAMPLE action for sFlow must be first action in any given list of + * actions. At this point we do not have all information required to + * build it. So try to build sample action as complete as possible. */ static void add_sflow_action(struct action_xlate_ctx *ctx) { @@ -5371,6 +6036,14 @@ add_sflow_action(struct action_xlate_ctx *ctx) ctx->sflow_n_outputs = 0; } +/* SAMPLE action for IPFIX must be 1st or 2nd action in any given list + * of actions, eventually after the SAMPLE action for sFlow. */ +static void +add_ipfix_action(struct action_xlate_ctx *ctx) +{ + compose_ipfix_action(ctx->ofproto, ctx->odp_actions, &ctx->flow); +} + /* Fix SAMPLE action according to data collected while composing ODP actions. * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested * USERSPACE action's user-cookie which is required for sflow. */ @@ -5385,8 +6058,8 @@ fix_sflow_action(struct action_xlate_ctx *ctx) } cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset, - sizeof(*cookie)); - assert(cookie->type == USER_ACTION_COOKIE_SFLOW); + sizeof cookie->sflow); + ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW); compose_sflow_cookie(ctx->ofproto, base->vlan_tci, ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie); @@ -5397,11 +6070,15 @@ compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port, bool check_stp) { const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); - uint32_t odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port); - ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci; - uint8_t flow_nw_tos = ctx->flow.nw_tos; + ovs_be16 flow_vlan_tci; + uint32_t flow_skb_mark; + uint8_t flow_nw_tos; struct priority_to_dscp *pdscp; - uint32_t out_port; + uint32_t out_port, odp_port; + + /* If 'struct flow' gets additional metadata, we'll need to zero it out + * before traversing a patch port. */ + BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20); if (!ofport) { xlate_report(ctx, "Nonexistent output port"); @@ -5414,16 +6091,100 @@ compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port, return; } + if (netdev_vport_is_patch(ofport->up.netdev)) { + struct ofport_dpif *peer = ofport_get_peer(ofport); + struct flow old_flow = ctx->flow; + const struct ofproto_dpif *peer_ofproto; + enum slow_path_reason special; + struct ofport_dpif *in_port; + + if (!peer) { + xlate_report(ctx, "Nonexistent patch port peer"); + return; + } + + peer_ofproto = ofproto_dpif_cast(peer->up.ofproto); + if (peer_ofproto->backer != ctx->ofproto->backer) { + xlate_report(ctx, "Patch port peer on a different datapath"); + return; + } + + ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto); + ctx->flow.in_port = peer->up.ofp_port; + ctx->flow.metadata = htonll(0); + memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel); + memset(ctx->flow.regs, 0, sizeof ctx->flow.regs); + + in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); + special = process_special(ctx->ofproto, &ctx->flow, in_port, + ctx->packet); + if (special) { + ctx->slow |= special; + } else if (!in_port || may_receive(in_port, ctx)) { + if (!in_port || stp_forward_in_state(in_port->stp_state)) { + xlate_table_action(ctx, ctx->flow.in_port, 0, true); + } else { + /* Forwarding is disabled by STP. Let OFPP_NORMAL and the + * learning action look at the packet, then drop it. */ + struct flow old_base_flow = ctx->base_flow; + size_t old_size = ctx->odp_actions->size; + xlate_table_action(ctx, ctx->flow.in_port, 0, true); + ctx->base_flow = old_base_flow; + ctx->odp_actions->size = old_size; + } + } + + ctx->flow = old_flow; + ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto); + + if (ctx->resubmit_stats) { + netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats); + netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats); + } + + return; + } + + flow_vlan_tci = ctx->flow.vlan_tci; + flow_skb_mark = ctx->flow.skb_mark; + flow_nw_tos = ctx->flow.nw_tos; + pdscp = get_priority(ofport, ctx->flow.skb_priority); if (pdscp) { ctx->flow.nw_tos &= ~IP_DSCP_MASK; ctx->flow.nw_tos |= pdscp->dscp; } - out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port, - ctx->flow.vlan_tci); - if (out_port != odp_port) { - ctx->flow.vlan_tci = htons(0); + if (ofport->tnl_port) { + /* Save tunnel metadata so that changes made due to + * the Logical (tunnel) Port are not visible for any further + * matches, while explicit set actions on tunnel metadata are. + */ + struct flow_tnl flow_tnl = ctx->flow.tunnel; + odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow); + if (odp_port == OVSP_NONE) { + xlate_report(ctx, "Tunneling decided against output"); + goto out; /* restore flow_nw_tos */ + } + if (ctx->flow.tunnel.ip_dst == ctx->orig_tunnel_ip_dst) { + xlate_report(ctx, "Not tunneling to our own address"); + goto out; /* restore flow_nw_tos */ + } + if (ctx->resubmit_stats) { + netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats); + } + out_port = odp_port; + commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow, + ctx->odp_actions); + ctx->flow.tunnel = flow_tnl; /* Restore tunnel metadata */ + } else { + odp_port = ofport->odp_port; + out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port, + ctx->flow.vlan_tci); + if (out_port != odp_port) { + ctx->flow.vlan_tci = htons(0); + } + ctx->flow.skb_mark &= ~IPSEC_MARK; } commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions); nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port); @@ -5431,7 +6192,11 @@ compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port, ctx->sflow_odp_port = odp_port; ctx->sflow_n_outputs++; ctx->nf_output_iface = ofp_port; + + /* Restore flow */ ctx->flow.vlan_tci = flow_vlan_tci; + ctx->flow.skb_mark = flow_skb_mark; + out: ctx->flow.nw_tos = flow_nw_tos; } @@ -5441,62 +6206,74 @@ compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) compose_output_action__(ctx, ofp_port, true); } +static void +tag_the_flow(struct action_xlate_ctx *ctx, struct rule_dpif *rule) +{ + struct ofproto_dpif *ofproto = ctx->ofproto; + uint8_t table_id = ctx->table_id; + + if (table_id > 0 && table_id < N_TABLES) { + struct table_dpif *table = &ofproto->tables[table_id]; + if (table->other_table) { + ctx->tags |= (rule && rule->tag + ? rule->tag + : rule_calculate_tag(&ctx->flow, + &table->other_table->mask, + table->basis)); + } + } +} + +/* Common rule processing in one place to avoid duplicating code. */ +static struct rule_dpif * +ctx_rule_hooks(struct action_xlate_ctx *ctx, struct rule_dpif *rule, + bool may_packet_in) +{ + if (ctx->resubmit_hook) { + ctx->resubmit_hook(ctx, rule); + } + if (rule == NULL && may_packet_in) { + /* XXX + * check if table configuration flags + * OFPTC_TABLE_MISS_CONTROLLER, default. + * OFPTC_TABLE_MISS_CONTINUE, + * OFPTC_TABLE_MISS_DROP + * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do? + */ + rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->flow); + } + if (rule && ctx->resubmit_stats) { + rule_credit_stats(rule, ctx->resubmit_stats); + } + return rule; +} + static void xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port, uint8_t table_id, bool may_packet_in) { if (ctx->recurse < MAX_RESUBMIT_RECURSION) { - struct ofproto_dpif *ofproto = ctx->ofproto; struct rule_dpif *rule; - uint16_t old_in_port; - uint8_t old_table_id; + uint16_t old_in_port = ctx->flow.in_port; + uint8_t old_table_id = ctx->table_id; - old_table_id = ctx->table_id; ctx->table_id = table_id; /* Look up a flow with 'in_port' as the input port. */ - old_in_port = ctx->flow.in_port; ctx->flow.in_port = in_port; - rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id); - - /* Tag the flow. */ - if (table_id > 0 && table_id < N_TABLES) { - struct table_dpif *table = &ofproto->tables[table_id]; - if (table->other_table) { - ctx->tags |= (rule && rule->tag - ? rule->tag - : rule_calculate_tag(&ctx->flow, - &table->other_table->mask, - table->basis)); - } - } + rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, table_id); + + tag_the_flow(ctx, rule); /* Restore the original input port. Otherwise OFPP_NORMAL and * OFPP_IN_PORT will have surprising behavior. */ ctx->flow.in_port = old_in_port; - if (ctx->resubmit_hook) { - ctx->resubmit_hook(ctx, rule); - } - - if (rule == NULL && may_packet_in) { - /* XXX - * check if table configuration flags - * OFPTC_TABLE_MISS_CONTROLLER, default. - * OFPTC_TABLE_MISS_CONTINUE, - * OFPTC_TABLE_MISS_DROP - * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do? - */ - rule = rule_dpif_miss_rule(ofproto, &ctx->flow); - } + rule = ctx_rule_hooks(ctx, rule, may_packet_in); if (rule) { struct rule_dpif *old_rule = ctx->rule; - if (ctx->resubmit_stats) { - rule_credit_stats(rule, ctx->resubmit_stats); - } - ctx->recurse++; ctx->rule = rule; do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx); @@ -5573,16 +6350,11 @@ execute_controller_action(struct action_xlate_ctx *ctx, int len, if (packet->l2 && packet->l3) { struct eth_header *eh; + uint16_t mpls_depth; eth_pop_vlan(packet); eh = packet->l2; - /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2 - * LLC frame. Calculating the Ethernet type of these frames is more - * trouble than seems appropriate for a simple assertion. */ - assert(ntohs(eh->eth_type) < ETH_TYPE_MIN - || eh->eth_type == ctx->flow.dl_type); - memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src); memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst); @@ -5590,6 +6362,16 @@ execute_controller_action(struct action_xlate_ctx *ctx, int len, eth_push_vlan(packet, ctx->flow.vlan_tci); } + mpls_depth = eth_mpls_depth(packet); + + if (mpls_depth < ctx->flow.mpls_depth) { + push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse); + } else if (mpls_depth > ctx->flow.mpls_depth) { + pop_mpls(packet, ctx->flow.dl_type); + } else if (mpls_depth) { + set_mpls_lse(packet, ctx->flow.mpls_lse); + } + if (packet->l4) { if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst, @@ -5622,6 +6404,46 @@ execute_controller_action(struct action_xlate_ctx *ctx, int len, ofpbuf_delete(packet); } +static void +execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type) +{ + ovs_assert(eth_type_mpls(eth_type)); + + if (ctx->base_flow.mpls_depth) { + ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK); + ctx->flow.mpls_depth++; + } else { + ovs_be32 label; + uint8_t tc, ttl; + + if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) { + label = htonl(0x2); /* IPV6 Explicit Null. */ + } else { + label = htonl(0x0); /* IPV4 Explicit Null. */ + } + tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2; + ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40; + ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label); + ctx->flow.mpls_depth = 1; + } + ctx->flow.dl_type = eth_type; +} + +static void +execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type) +{ + ovs_assert(eth_type_mpls(ctx->flow.dl_type)); + ovs_assert(!eth_type_mpls(eth_type)); + + if (ctx->flow.mpls_depth) { + ctx->flow.mpls_depth--; + ctx->flow.mpls_lse = htonl(0); + if (!ctx->flow.mpls_depth) { + ctx->flow.dl_type = eth_type; + } + } +} + static bool compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids) { @@ -5646,6 +6468,38 @@ compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids) } } +static bool +execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl) +{ + if (!eth_type_mpls(ctx->flow.dl_type)) { + return true; + } + + set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl); + return false; +} + +static bool +execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx) +{ + uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse); + + if (!eth_type_mpls(ctx->flow.dl_type)) { + return false; + } + + if (ttl > 1) { + ttl--; + set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl); + return false; + } else { + execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0); + + /* Stop processing for current table. */ + return true; + } +} + static void xlate_output_action(struct action_xlate_ctx *ctx, uint16_t port, uint16_t max_len, bool may_packet_in) @@ -5758,31 +6612,6 @@ xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id) } } -struct xlate_reg_state { - ovs_be16 vlan_tci; - ovs_be64 tun_id; -}; - -static void -xlate_autopath(struct action_xlate_ctx *ctx, - const struct ofpact_autopath *ap) -{ - uint16_t ofp_port = ap->port; - struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port); - - if (!port || !port->bundle) { - ofp_port = OFPP_NONE; - } else if (port->bundle->bond) { - /* Autopath does not support VLAN hashing. */ - struct ofport_dpif *slave = bond_choose_output_slave( - port->bundle->bond, &ctx->flow, 0, &ctx->tags); - if (slave) { - ofp_port = slave->up.ofp_port; - } - } - nxm_reg_load(&ap->dst, ofp_port, &ctx->flow); -} - static bool slave_enabled_cb(uint16_t ofp_port, void *ofproto_) { @@ -5863,6 +6692,23 @@ xlate_fin_timeout(struct action_xlate_ctx *ctx, } } +static void +xlate_sample_action(struct action_xlate_ctx *ctx, + const struct ofpact_sample *os) +{ + union user_action_cookie cookie; + /* Scale the probability from 16-bit to 32-bit while representing + * the same percentage. */ + uint32_t probability = (os->probability << 16) | os->probability; + + commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions); + + compose_flow_sample_cookie(os->probability, os->collector_set_id, + os->obs_domain_id, os->obs_point_id, &cookie); + compose_sample_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, + probability, &cookie, sizeof cookie.flow_sample); +} + static bool may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx) { @@ -5884,25 +6730,38 @@ may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx) return true; } +static bool +tunnel_ecn_ok(struct action_xlate_ctx *ctx) +{ + if (is_ip_any(&ctx->base_flow) + && (ctx->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) { + if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) { + VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE" + " but is not ECN capable"); + return false; + } else { + /* Set the ECN CE value in the tunneled packet. */ + ctx->flow.nw_tos |= IP_ECN_CE; + } + } + + return true; +} + static void do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, struct action_xlate_ctx *ctx) { - const struct ofport_dpif *port; bool was_evictable = true; const struct ofpact *a; - port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); - if (port && !may_receive(port, ctx)) { - /* Drop this flow. */ - return; - } - if (ctx->rule) { /* Don't let the rule we're working on get evicted underneath us. */ was_evictable = ctx->rule->up.evictable; ctx->rule->up.evictable = false; } + + do_xlate_actions_again: OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) { struct ofpact_controller *controller; const struct ofpact_metadata *metadata; @@ -5961,11 +6820,15 @@ do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, break; case OFPACT_SET_IPV4_SRC: - ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4; + if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { + ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4; + } break; case OFPACT_SET_IPV4_DST: - ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4; + if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) { + ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4; + } break; case OFPACT_SET_IPV4_DSCP: @@ -5977,11 +6840,15 @@ do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, break; case OFPACT_SET_L4_SRC_PORT: - ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port); + if (is_ip_any(&ctx->flow)) { + ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port); + } break; case OFPACT_SET_L4_DST_PORT: - ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port); + if (is_ip_any(&ctx->flow)) { + ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port); + } break; case OFPACT_RESUBMIT: @@ -6008,6 +6875,36 @@ do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow); break; + case OFPACT_STACK_PUSH: + nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow, + &ctx->stack); + break; + + case OFPACT_STACK_POP: + nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow, + &ctx->stack); + break; + + case OFPACT_PUSH_MPLS: + execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype); + break; + + case OFPACT_POP_MPLS: + execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype); + break; + + case OFPACT_SET_MPLS_TTL: + if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) { + goto out; + } + break; + + case OFPACT_DEC_MPLS_TTL: + if (execute_dec_mpls_ttl_action(ctx)) { + goto out; + } + break; + case OFPACT_DEC_TTL: if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) { goto out; @@ -6022,10 +6919,6 @@ do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow); break; - case OFPACT_AUTOPATH: - xlate_autopath(ctx, ofpact_get_AUTOPATH(a)); - break; - case OFPACT_BUNDLE: ctx->ofproto->has_bundle_action = true; xlate_bundle_action(ctx, ofpact_get_BUNDLE(a)); @@ -6066,23 +6959,44 @@ do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, break; case OFPACT_GOTO_TABLE: { - /* XXX remove recursion */ - /* It is assumed that goto-table is last action */ + /* It is assumed that goto-table is the last action. */ struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a); - assert(ctx->table_id < ogt->table_id); - xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true); + struct rule_dpif *rule; + + ovs_assert(ctx->table_id < ogt->table_id); + + ctx->table_id = ogt->table_id; + + /* Look up a flow from the new table. */ + rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, ctx->table_id); + + tag_the_flow(ctx, rule); + + rule = ctx_rule_hooks(ctx, rule, true); + + if (rule) { + if (ctx->rule) { + ctx->rule->up.evictable = was_evictable; + } + ctx->rule = rule; + was_evictable = rule->up.evictable; + rule->up.evictable = false; + + /* Tail recursion removal. */ + ofpacts = rule->up.ofpacts; + ofpacts_len = rule->up.ofpacts_len; + goto do_xlate_actions_again; + } break; } + + case OFPACT_SAMPLE: + xlate_sample_action(ctx, ofpact_get_SAMPLE(a)); + break; } } out: - /* We've let OFPP_NORMAL and the learning action look at the packet, - * so drop it now if forwarding is disabled. */ - if (port && !stp_forward_in_state(port->stp_state)) { - ofpbuf_clear(ctx->odp_actions); - add_sflow_action(ctx); - } if (ctx->rule) { ctx->rule->up.evictable = was_evictable; } @@ -6091,14 +7005,36 @@ out: static void action_xlate_ctx_init(struct action_xlate_ctx *ctx, struct ofproto_dpif *ofproto, const struct flow *flow, - ovs_be16 initial_tci, struct rule_dpif *rule, + const struct initial_vals *initial_vals, + struct rule_dpif *rule, uint8_t tcp_flags, const struct ofpbuf *packet) { + /* Flow initialization rules: + * - 'base_flow' must match the kernel's view of the packet at the + * time that action processing starts. 'flow' represents any + * transformations we wish to make through actions. + * - By default 'base_flow' and 'flow' are the same since the input + * packet matches the output before any actions are applied. + * - When using VLAN splinters, 'base_flow''s VLAN is set to the value + * of the received packet as seen by the kernel. If we later output + * to another device without any modifications this will cause us to + * insert a new tag since the original one was stripped off by the + * VLAN device. + * - Tunnel metadata as received is retained in 'flow'. This allows + * tunnel metadata matching also in later tables. + * Since a kernel action for setting the tunnel metadata will only be + * generated with actual tunnel output, changing the tunnel metadata + * values in 'flow' (such as tun_id) will only have effect with a later + * tunnel output action. + * - Tunnel 'base_flow' is completely cleared since that is what the + * kernel does. If we wish to maintain the original values an action + * needs to be generated. */ + ctx->ofproto = ofproto; ctx->flow = *flow; - memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel); ctx->base_flow = ctx->flow; - ctx->base_flow.vlan_tci = initial_tci; + memset(&ctx->base_flow.tunnel, 0, sizeof ctx->base_flow.tunnel); + ctx->orig_tunnel_ip_dst = flow->tunnel.ip_dst; ctx->rule = rule; ctx->packet = packet; ctx->may_learn = packet != NULL; @@ -6106,6 +7042,10 @@ action_xlate_ctx_init(struct action_xlate_ctx *ctx, ctx->resubmit_hook = NULL; ctx->report_hook = NULL; ctx->resubmit_stats = NULL; + + if (initial_vals) { + ctx->base_flow.vlan_tci = initial_vals->vlan_tci; + } } /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts' @@ -6121,6 +7061,8 @@ xlate_actions(struct action_xlate_ctx *ctx, static bool hit_resubmit_limit; enum slow_path_reason special; + struct ofport_dpif *in_port; + struct flow orig_flow; COVERAGE_INC(ofproto_dpif_xlate); @@ -6141,14 +7083,12 @@ xlate_actions(struct action_xlate_ctx *ctx, ctx->table_id = 0; ctx->exit = false; + ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack); + if (ctx->ofproto->has_mirrors || hit_resubmit_limit) { /* Do this conditionally because the copy is expensive enough that it - * shows up in profiles. - * - * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4 - * believe that I wasn't using it without initializing it if I kept it - * in a local variable. */ - ctx->orig_flow = ctx->flow; + * shows up in profiles. */ + orig_flow = ctx->flow; } if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) { @@ -6174,15 +7114,31 @@ xlate_actions(struct action_xlate_ctx *ctx, } } - special = process_special(ctx->ofproto, &ctx->flow, ctx->packet); + in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); + special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet); if (special) { ctx->slow |= special; } else { static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1); - ovs_be16 initial_tci = ctx->base_flow.vlan_tci; + struct initial_vals initial_vals; + size_t sample_actions_len; + uint32_t local_odp_port; + + initial_vals.vlan_tci = ctx->base_flow.vlan_tci; add_sflow_action(ctx); - do_xlate_actions(ofpacts, ofpacts_len, ctx); + add_ipfix_action(ctx); + sample_actions_len = ctx->odp_actions->size; + + if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) { + do_xlate_actions(ofpacts, ofpacts_len, ctx); + + /* We've let OFPP_NORMAL and the learning action look at the + * packet, so drop it now if forwarding is disabled. */ + if (in_port && !stp_forward_in_state(in_port->stp_state)) { + ctx->odp_actions->size = sample_actions_len; + } + } if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) { if (!hit_resubmit_limit) { @@ -6192,29 +7148,28 @@ xlate_actions(struct action_xlate_ctx *ctx, } else if (!VLOG_DROP_ERR(&trace_rl)) { struct ds ds = DS_EMPTY_INITIALIZER; - ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet, - initial_tci, &ds); + ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet, + &initial_vals, &ds); VLOG_ERR("Trace triggered by excessive resubmit " "recursion:\n%s", ds_cstr(&ds)); ds_destroy(&ds); } } - if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow, - ctx->odp_actions->data, - ctx->odp_actions->size)) { - ctx->slow |= SLOW_IN_BAND; - if (ctx->packet - && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow, - ctx->packet)) { - compose_output_action(ctx, OFPP_LOCAL); - } + local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL); + if (!connmgr_must_output_local(ctx->ofproto->up.connmgr, &ctx->flow, + local_odp_port, + ctx->odp_actions->data, + ctx->odp_actions->size)) { + compose_output_action(ctx, OFPP_LOCAL); } if (ctx->ofproto->has_mirrors) { - add_mirror_actions(ctx, &ctx->orig_flow); + add_mirror_actions(ctx, &orig_flow); } fix_sflow_action(ctx); } + + ofpbuf_uninit(&ctx->stack); } /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts' @@ -6587,7 +7542,7 @@ update_learning_table(struct ofproto_dpif *ofproto, in_bundle->name, vlan); mac->port.p = in_bundle; - tag_set_add(&ofproto->revalidate_set, + tag_set_add(&ofproto->backer->revalidate_set, mac_learning_changed(ofproto->ml, mac)); } } @@ -6863,7 +7818,7 @@ table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id) if (table->catchall_table != catchall || table->other_table != other) { table->catchall_table = catchall; table->other_table = other; - ofproto->need_revalidate = REV_FLOW_TABLE; + ofproto->backer->need_revalidate = REV_FLOW_TABLE; } } @@ -6881,13 +7836,13 @@ rule_invalidate(const struct rule_dpif *rule) table_update_taggable(ofproto, rule->up.table_id); - if (!ofproto->need_revalidate) { + if (!ofproto->backer->need_revalidate) { struct table_dpif *table = &ofproto->tables[rule->up.table_id]; if (table->other_table && rule->tag) { - tag_set_add(&ofproto->revalidate_set, rule->tag); + tag_set_add(&ofproto->backer->revalidate_set, rule->tag); } else { - ofproto->need_revalidate = REV_FLOW_TABLE; + ofproto->backer->need_revalidate = REV_FLOW_TABLE; } } } @@ -6897,9 +7852,8 @@ set_frag_handling(struct ofproto *ofproto_, enum ofp_config_flags frag_handling) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - if (frag_handling != OFPC_FRAG_REASM) { - ofproto->need_revalidate = REV_RECONFIGURE; + ofproto->backer->need_revalidate = REV_RECONFIGURE; return true; } else { return false; @@ -6912,6 +7866,7 @@ 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 initial_vals initial_vals; struct odputil_keybuf keybuf; struct dpif_flow_stats stats; @@ -6927,7 +7882,8 @@ packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, dpif_flow_stats_extract(flow, packet, time_msec(), &stats); - action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL, + initial_vals.vlan_tci = flow->vlan_tci; + action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL, packet_get_tcp_flags(packet, flow), packet); ctx.resubmit_stats = &stats; @@ -7031,10 +7987,10 @@ ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc, unixctl_command_reply_error(conn, "no such bridge"); return; } - mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); + mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); } else { HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { - mac_learning_flush(ofproto->ml, &ofproto->revalidate_set); + mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set); } } @@ -7163,112 +8119,113 @@ static void ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], void *aux OVS_UNUSED) { - const char *dpname = argv[1]; + const struct dpif_backer *backer; struct ofproto_dpif *ofproto; struct ofpbuf odp_key; struct ofpbuf *packet; - ovs_be16 initial_tci; + struct initial_vals initial_vals; struct ds result; struct flow flow; char *s; packet = NULL; - ofpbuf_init(&odp_key, 0); + backer = NULL; ds_init(&result); + ofpbuf_init(&odp_key, 0); - ofproto = ofproto_dpif_lookup(dpname); - if (!ofproto) { - unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list " - "for help)"); - goto exit; + /* Handle "-generate" or a hex string as the last argument. */ + if (!strcmp(argv[argc - 1], "-generate")) { + packet = ofpbuf_new(0); + argc--; + } else { + const char *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; + } } - if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) { - /* ofproto/trace dpname flow [-generate] */ - const char *flow_s = argv[2]; - const char *generate_s = argv[3]; - - /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like - * flow. We guess which type it is based on whether 'flow_s' contains - * an '(', since a datapath flow always contains '(') but an - * OpenFlow-like flow should not (in fact it's allowed but I believe - * that's not documented anywhere). - * - * An alternative would be to try to parse 'flow_s' both ways, but then - * it would be tricky giving a sensible error message. After all, do - * you just say "syntax error" or do you present both error messages? - * Both choices seem lousy. */ - if (strchr(flow_s, '(')) { - int error; - /* Convert string to datapath key. */ - ofpbuf_init(&odp_key, 0); - error = odp_flow_key_from_string(flow_s, NULL, &odp_key); - if (error) { - unixctl_command_reply_error(conn, "Bad flow syntax"); - goto exit; + /* 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_key_from_string(argv[argc - 1], NULL, &odp_key)) { + /* 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]; } - - /* XXX: Since we allow the user to specify an ofproto, it's - * possible they will specify a different ofproto than the one the - * port actually belongs too. Ideally we should simply remove the - * ability to specify the ofproto. */ - if (ofproto_receive(ofproto->backer, NULL, odp_key.data, - odp_key.size, &flow, NULL, NULL, NULL, - &initial_tci)) { - unixctl_command_reply_error(conn, "Invalid flow"); + 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 { - char *error_s; - - error_s = parse_ofp_exact_flow(&flow, argv[2]); - if (error_s) { - unixctl_command_reply_error(conn, error_s); - free(error_s); + /* 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; } - - initial_tci = flow.vlan_tci; + node = shash_first(&all_dpif_backers); + backer = node->data; } - /* Generate a packet, if requested. */ - if (generate_s) { - packet = ofpbuf_new(0); - flow_compose(packet, &flow); + /* Extract the ofproto_dpif object from the ofproto_receive() + * function. */ + if (ofproto_receive(backer, NULL, odp_key.data, + odp_key.size, &flow, NULL, &ofproto, NULL, + &initial_vals)) { + unixctl_command_reply_error(conn, "Invalid datapath flow"); + goto exit; } - } else if (argc == 7) { - /* ofproto/trace dpname priority tun_id in_port mark packet */ - const char *priority_s = argv[2]; - const char *tun_id_s = argv[3]; - const char *in_port_s = argv[4]; - const char *mark_s = argv[5]; - const char *packet_s = argv[6]; - uint32_t in_port = atoi(in_port_s); - ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0)); - uint32_t priority = atoi(priority_s); - uint32_t mark = atoi(mark_s); - const char *msg; - - msg = eth_from_hex(packet_s, &packet); - if (msg) { - unixctl_command_reply_error(conn, msg); + ds_put_format(&result, "Bridge: %s\n", ofproto->up.name); + } else if (!parse_ofp_exact_flow(&flow, argv[argc - 1])) { + if (argc != 3) { + unixctl_command_reply_error(conn, "Must specify bridge name"); goto exit; } - ds_put_cstr(&result, "Packet: "); - s = ofp_packet_to_string(packet->data, packet->size); - ds_put_cstr(&result, s); - free(s); - - flow_extract(packet, priority, mark, NULL, in_port, &flow); - flow.tunnel.tun_id = tun_id; - initial_tci = flow.vlan_tci; + ofproto = ofproto_dpif_lookup(argv[1]); + if (!ofproto) { + unixctl_command_reply_error(conn, "Unknown bridge name"); + goto exit; + } + initial_vals.vlan_tci = flow.vlan_tci; } else { - unixctl_command_reply_error(conn, "Bad command syntax"); + unixctl_command_reply_error(conn, "Bad flow syntax"); goto exit; } - ofproto_trace(ofproto, &flow, packet, initial_tci, &result); + /* Generate a packet, if requested. */ + if (packet) { + if (!packet->size) { + flow_compose(packet, &flow); + } else { + ds_put_cstr(&result, "Packet: "); + s = ofp_packet_to_string(packet->data, packet->size); + ds_put_cstr(&result, s); + free(s); + + /* Use the metadata from the flow and the packet argument + * to reconstruct the flow. */ + flow_extract(packet, flow.skb_priority, flow.skb_mark, NULL, + flow.in_port, &flow); + initial_vals.vlan_tci = flow.vlan_tci; + } + } + + ofproto_trace(ofproto, &flow, packet, &initial_vals, &result); unixctl_command_reply(conn, ds_cstr(&result)); exit: @@ -7279,8 +8236,8 @@ exit: static void ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, - const struct ofpbuf *packet, ovs_be16 initial_tci, - struct ds *ds) + const struct ofpbuf *packet, + const struct initial_vals *initial_vals, struct ds *ds) { struct rule_dpif *rule; @@ -7310,7 +8267,7 @@ ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, trace.flow = *flow; ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); - action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci, + action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals, rule, tcp_flags, packet); trace.ctx.resubmit_hook = trace_resubmit; trace.ctx.report_hook = trace_report; @@ -7341,32 +8298,17 @@ ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow, case SLOW_STP: ds_put_cstr(ds, "\n\t- Consists of STP packets."); break; - case SLOW_IN_BAND: - ds_put_cstr(ds, "\n\t- Needs in-band special case " - "processing."); - if (!packet) { - ds_put_cstr(ds, "\n\t (The datapath actions are " - "incomplete--for complete actions, " - "please supply a packet.)"); - } + 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_MATCH: - ds_put_cstr(ds, "\n\t- Needs more specific matching " - "than the datapath supports."); - break; } slow &= ~bit; } - - if (slow & ~SLOW_MATCH) { - ds_put_cstr(ds, "\nThe datapath actions above do not reflect " - "the special slow-path processing."); - } } } } @@ -7402,7 +8344,7 @@ ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply) } } if (errors) { - ofproto->need_revalidate = REV_INCONSISTENCY; + ofproto->backer->need_revalidate = REV_INCONSISTENCY; } if (errors) { @@ -7481,21 +8423,42 @@ ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED, static void show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds) { - struct dpif_dp_stats s; const struct shash_node **ports; int i; + struct avg_subfacet_rates lifetime; + unsigned long long int minutes; + const int min_ms = 60 * 1000; /* milliseconds in one minute. */ - dpif_get_dp_stats(ofproto->backer->dpif, &s); + minutes = (time_msec() - ofproto->created) / min_ms; + + if (minutes > 0) { + lifetime.add_rate = (double)ofproto->total_subfacet_add_count + / minutes; + lifetime.del_rate = (double)ofproto->total_subfacet_del_count + / minutes; + }else { + lifetime.add_rate = 0.0; + lifetime.del_rate = 0.0; + } ds_put_format(ds, "%s (%s):\n", ofproto->up.name, dpif_name(ofproto->backer->dpif)); - /* xxx It would be better to show bridge-specific stats instead - * xxx of dp ones. */ ds_put_format(ds, - "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n", - s.n_hit, s.n_missed, s.n_lost); - ds_put_format(ds, "\tflows: %zu\n", - hmap_count(&ofproto->subfacets)); + "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n", + ofproto->n_hit, ofproto->n_missed); + ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d," + " life span: %llu(ms)\n", + hmap_count(&ofproto->subfacets), + avg_subfacet_count(ofproto), + ofproto->max_n_subfacet, + avg_subfacet_life_span(ofproto)); + if (minutes >= 60) { + show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly); + } + if (minutes >= 60 * 24) { + show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily); + } + show_dp_rates(ds, "\t\toverall avg:", &lifetime); ports = shash_sort(&ofproto->up.port_by_name); for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) { @@ -7503,9 +8466,17 @@ show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds) struct ofport *ofport = node->data; const char *name = netdev_get_name(ofport->netdev); const char *type = netdev_get_type(ofport->netdev); + uint32_t odp_port; + + ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port); + + odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port); + if (odp_port != OVSP_NONE) { + ds_put_format(ds, "%"PRIu32":", odp_port); + } else { + ds_put_cstr(ds, "none:"); + } - ds_put_format(ds, "\t%s %u/%u:", name, ofport->ofp_port, - ofp_port_to_odp_port(ofproto, ofport->ofp_port)); if (strcmp(type, "system")) { struct netdev *netdev; int error; @@ -7595,12 +8566,10 @@ ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn, return; } - HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { - struct odputil_keybuf keybuf; - struct ofpbuf key; + update_stats(ofproto->backer); - subfacet_get_key(subfacet, &keybuf, &key); - odp_flow_key_format(key.data, key.size, &ds); + HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) { + odp_flow_key_format(subfacet->key, subfacet->key_len, &ds); ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:", subfacet->dp_packet_count, subfacet->dp_byte_count); @@ -7616,7 +8585,18 @@ ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn, } ds_put_cstr(&ds, ", actions:"); - format_odp_actions(&ds, subfacet->actions, subfacet->actions_len); + if (subfacet->slow) { + uint64_t slow_path_stub[128 / 8]; + const struct nlattr *actions; + size_t actions_len; + + compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow, + slow_path_stub, sizeof slow_path_stub, + &actions, &actions_len); + format_odp_actions(&ds, actions, actions_len); + } else { + format_odp_actions(&ds, subfacet->actions, subfacet->actions_len); + } ds_put_char(&ds, '\n'); } @@ -7655,8 +8635,8 @@ ofproto_dpif_unixctl_init(void) unixctl_command_register( "ofproto/trace", - "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}", - 2, 6, ofproto_unixctl_trace, NULL); + "[dp_name]|bridge odp_flow|br_flow [-generate|packet]", + 1, 3, ofproto_unixctl_trace, NULL); unixctl_command_register("fdb/flush", "[bridge]", 0, 1, ofproto_unixctl_fdb_flush, NULL); unixctl_command_register("fdb/show", "bridge", 1, 1, @@ -7695,7 +8675,7 @@ set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid) return 0; } - ofproto->need_revalidate = REV_RECONFIGURE; + ofproto->backer->need_revalidate = REV_RECONFIGURE; if (ofport->realdev_ofp_port) { vsp_remove(ofport); @@ -7895,6 +8875,85 @@ odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) return OFPP_NONE; } } +static unsigned long long int +avg_subfacet_life_span(const struct ofproto_dpif *ofproto) +{ + unsigned long long int dc; + unsigned long long int avg; + + dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count; + avg = dc ? ofproto->total_subfacet_life_span / dc : 0; + + return avg; +} + +static double +avg_subfacet_count(const struct ofproto_dpif *ofproto) +{ + double avg_c = 0.0; + + if (ofproto->n_update_stats) { + avg_c = (double)ofproto->total_subfacet_count + / ofproto->n_update_stats; + } + + return avg_c; +} + +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 +update_max_subfacet_count(struct ofproto_dpif *ofproto) +{ + ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet, + hmap_count(&ofproto->subfacets)); +} + +/* 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 ofproto_dpif *ofproto) +{ + const int min_ms = 60 * 1000; /* milliseconds in one minute. */ + + /* Update hourly averages on the minute boundaries. */ + if (time_msec() - ofproto->last_minute >= min_ms) { + exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count); + exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count); + + /* Update daily averages on the hour boundaries. */ + if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) { + exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate); + exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate); + } + + ofproto->total_subfacet_add_count += ofproto->subfacet_add_count; + ofproto->total_subfacet_del_count += ofproto->subfacet_del_count; + ofproto->subfacet_add_count = 0; + ofproto->subfacet_del_count = 0; + ofproto->last_minute += min_ms; + } +} + +static void +dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta) +{ + ofproto->n_hit += delta; +} const struct ofproto_class ofproto_dpif_class = { init, @@ -7945,11 +9004,11 @@ const struct ofproto_class ofproto_dpif_class = { set_netflow, get_netflow_ids, set_sflow, + set_ipfix, set_cfm, - get_cfm_fault, - get_cfm_opup, - get_cfm_remote_mpids, - get_cfm_health, + get_cfm_status, + set_bfd, + get_bfd_status, set_stp, get_stp_status, set_stp_port,