X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=ofproto%2Fofproto-dpif.c;h=40e897f31e875d0979d806f793333bd6d5993493;hb=29089a540cfa30a834e3ee19a8b4c52ff2e331b2;hp=6f63f831efa64fa150caca91eff4fcb747c2578b;hpb=b0408fcacca150694d116d4ead3930757e545bbf;p=sliver-openvswitch.git diff --git a/ofproto/ofproto-dpif.c b/ofproto/ofproto-dpif.c index 6f63f831e..40e897f31 100644 --- a/ofproto/ofproto-dpif.c +++ b/ofproto/ofproto-dpif.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc. + * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -20,7 +20,6 @@ #include -#include "autopath.h" #include "bond.h" #include "bundle.h" #include "byte-order.h" @@ -36,6 +35,7 @@ #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" @@ -46,10 +46,13 @@ #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" @@ -75,6 +78,8 @@ 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; @@ -115,8 +120,7 @@ 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 *); @@ -213,6 +217,11 @@ struct action_xlate_ctx { * this flow when actions change header fields. */ struct flow flow; + /* 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; @@ -280,12 +289,30 @@ struct action_xlate_ctx { 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; + + /* If received on a tunnel, the IP TOS value of the tunnel. */ + uint8_t tunnel_ip_tos; }; 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, @@ -293,11 +320,14 @@ 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, @@ -337,16 +367,12 @@ 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. */ @@ -361,22 +387,28 @@ 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 + * flow definition only has the OpenFlow port number and the port is + * no longer part of the bridge, we can't determine the datapath port + * number needed to delete the flow from the datapath. */ + uint32_t odp_in_port; }; -static struct subfacet *subfacet_create(struct facet *, enum odp_key_fitness, - const struct nlattr *key, - size_t key_len, ovs_be16 initial_tci, +#define SUBFACET_DESTROY_MAX_BATCH 50 + +static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss, long long int now); static struct subfacet *subfacet_find(struct ofproto_dpif *, - const struct nlattr *key, size_t key_len); + const struct nlattr *key, size_t key_len, + uint32_t key_hash); static void subfacet_destroy(struct subfacet *); static void subfacet_destroy__(struct subfacet *); -static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *, - struct ofpbuf *key); +static void subfacet_destroy_batch(struct ofproto_dpif *, + struct subfacet **, int n); static void subfacet_reset_dp_stats(struct subfacet *, struct dpif_flow_stats *); static void subfacet_update_time(struct subfacet *, long long int used); @@ -458,8 +490,12 @@ 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 *, @@ -481,10 +517,14 @@ 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 *); struct ofport_dpif { + struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */ struct ofport up; uint32_t odp_port; @@ -492,9 +532,9 @@ struct ofport_dpif { struct list bundle_node; /* In struct ofbundle's "ports" list. */ struct cfm *cfm; /* Connectivity Fault Management, 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. */ @@ -543,10 +583,15 @@ static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *); static void vsp_remove(struct ofport_dpif *); static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid); +static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *, + uint16_t ofp_port); +static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *, + uint32_t odp_port); + static struct ofport_dpif * ofport_dpif_cast(const struct ofport *ofport) { - assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class); + ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class); return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; } @@ -555,6 +600,7 @@ static void port_run_fast(struct ofport_dpif *); static void port_wait(struct ofport_dpif *); 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; @@ -593,10 +639,53 @@ 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; + int refcount; + 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; - struct dpif *dpif; + struct dpif_backer *backer; /* Special OpenFlow rules. */ struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */ @@ -608,24 +697,21 @@ 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]; bool has_mirrors; bool has_bonded_bundles; - /* Expiration. */ - struct timer next_expiration; - /* Facets. */ 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; @@ -641,7 +727,58 @@ struct ofproto_dpif { /* VLAN splinters. */ struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */ struct hmap vlandev_map; /* vlandev -> (realdev,vid). */ + + /* Ports. */ + 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.) */ @@ -655,7 +792,7 @@ 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); } @@ -664,8 +801,8 @@ 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 *, @@ -673,10 +810,10 @@ static void update_learning_table(struct ofproto_dpif *, struct ofbundle *); /* Upcalls. */ #define FLOW_MISS_MAX_BATCH 50 -static int handle_upcalls(struct ofproto_dpif *, unsigned int max_batch); +static int handle_upcalls(struct dpif_backer *, unsigned int max_batch); /* Flow expiration. */ -static int expire(struct ofproto_dpif *); +static int expire(struct dpif_backer *); /* NetFlow. */ static void send_netflow_active_timeouts(struct ofproto_dpif *); @@ -686,16 +823,35 @@ 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. */ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + +/* Initial mappings of port to bridge mappings. */ +static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports); /* Factory functions. */ static void -init(const struct shash *iface_hints OVS_UNUSED) +init(const struct shash *iface_hints) { + struct shash_node *node; + + /* Make a local copy, since we don't own 'iface_hints' elements. */ + SHASH_FOR_EACH(node, iface_hints) { + const struct iface_hint *orig_hint = node->data; + struct iface_hint *new_hint = xmalloc(sizeof *new_hint); + + new_hint->br_name = xstrdup(orig_hint->br_name); + new_hint->br_type = xstrdup(orig_hint->br_type); + new_hint->ofp_port = orig_hint->ofp_port; + + shash_add(&init_ofp_ports, node->name, new_hint); + } } static void @@ -707,7 +863,17 @@ enumerate_types(struct sset *types) static int enumerate_names(const char *type, struct sset *names) { - return dp_enumerate_names(type, names); + struct ofproto_dpif *ofproto; + + sset_clear(names); + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { + if (strcmp(type, ofproto->up.type)) { + continue; + } + sset_add(names, ofproto->up.name); + } + + return 0; } static int @@ -724,6 +890,302 @@ del(const char *type, const char *name) return error; } +static const char * +port_open_type(const char *datapath_type, const char *port_type) +{ + return dpif_port_open_type(datapath_type, port_type); +} + +/* Type functions. */ + +static struct ofproto_dpif * +lookup_ofproto_dpif_by_port_name(const char *name) +{ + struct ofproto_dpif *ofproto; + + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { + if (sset_contains(&ofproto->ports, name)) { + return ofproto; + } + } + + return NULL; +} + +static int +type_run(const char *type) +{ + static long long int push_timer = LLONG_MIN; + struct dpif_backer *backer; + char *devname; + int error; + + 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; + } + + 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); + } + + /* Check for port changes in the dpif. */ + while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) { + struct ofproto_dpif *ofproto; + struct dpif_port port; + + /* Don't report on the datapath's device. */ + if (!strcmp(devname, dpif_base_name(backer->dpif))) { + 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, + * report it through poll_set(). If we don't, it may be + * notifying us of a removal we initiated, so ignore it. + * If there's a pending ENOBUFS, let it stand, since + * everything will be reevaluated. */ + if (ofproto && ofproto->port_poll_errno != ENOBUFS) { + sset_add(&ofproto->port_poll_set, devname); + ofproto->port_poll_errno = 0; + } + } else if (!ofproto) { + /* The port was added, but we don't know with which + * ofproto we should associate it. Delete it. */ + dpif_port_del(backer->dpif, port.port_no); + } + dpif_port_destroy(&port); + + next: + free(devname); + } + + if (error != EAGAIN) { + struct ofproto_dpif *ofproto; + + /* There was some sort of error, so propagate it to all + * ofprotos that use this backer. */ + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, + &all_ofproto_dpifs) { + if (ofproto->backer == backer) { + sset_clear(&ofproto->port_poll_set); + ofproto->port_poll_errno = error; + } + } + } + + return 0; +} + +static int +dpif_backer_run_fast(struct dpif_backer *backer, int max_batch) +{ + unsigned int work; + + /* Handle one or more batches of upcalls, until there's nothing left to do + * or until we do a fixed total amount of work. + * + * We do work in batches because it can be much cheaper to set up a number + * of flows and fire off their patches all at once. We do multiple batches + * because in some cases handling a packet can cause another packet to be + * queued almost immediately as part of the return flow. Both + * optimizations can make major improvements on some benchmarks and + * presumably for real traffic as well. */ + work = 0; + while (work < max_batch) { + int retval = handle_upcalls(backer, max_batch - work); + if (retval <= 0) { + return -retval; + } + work += retval; + } + + return 0; +} + +static int +type_run_fast(const char *type) +{ + struct dpif_backer *backer; + + backer = shash_find_data(&all_dpif_backers, type); + if (!backer) { + /* This is not necessarily a problem, since backers are only + * created on demand. */ + return 0; + } + + return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH); +} + +static void +run_fast_rl(void) +{ + static long long int port_rl = LLONG_MIN; + static unsigned int backer_rl = 0; + + if (time_msec() >= port_rl) { + struct ofproto_dpif *ofproto; + 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) +{ + 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; + } + + timer_wait(&backer->next_expiration); +} + /* Basic life-cycle. */ static int add_internal_flows(struct ofproto_dpif *); @@ -742,38 +1204,158 @@ dealloc(struct ofproto *ofproto_) free(ofproto); } +static void +close_dpif_backer(struct dpif_backer *backer) +{ + struct shash_node *node; + + 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); + shash_delete(&all_dpif_backers, node); + dpif_close(backer->dpif); + + free(backer); +} + +/* Datapath port slated for removal from datapath. */ +struct odp_garbage { + struct list list_node; + uint32_t odp_port; +}; + +static int +open_dpif_backer(const char *type, struct dpif_backer **backerp) +{ + struct dpif_backer *backer; + struct dpif_port_dump port_dump; + struct dpif_port port; + struct shash_node *node; + struct list garbage_list; + struct odp_garbage *garbage, *next; + struct sset names; + char *backer_name; + const char *name; + int error; + + backer = shash_find_data(&all_dpif_backers, type); + if (backer) { + backer->refcount++; + *backerp = backer; + return 0; + } + + backer_name = xasprintf("ovs-%s", type); + + /* Remove any existing datapaths, since we assume we're the only + * userspace controlling the datapath. */ + sset_init(&names); + dp_enumerate_names(type, &names); + SSET_FOR_EACH(name, &names) { + struct dpif *old_dpif; + + /* Don't remove our backer if it exists. */ + if (!strcmp(name, backer_name)) { + continue; + } + + if (dpif_open(name, type, &old_dpif)) { + VLOG_WARN("couldn't open old datapath %s to remove it", name); + } else { + dpif_delete(old_dpif); + dpif_close(old_dpif); + } + } + sset_destroy(&names); + + backer = xmalloc(sizeof *backer); + + error = dpif_create_and_open(backer_name, type, &backer->dpif); + free(backer_name); + 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); + + /* Loop through the ports already on the datapath and remove any + * that we don't need anymore. */ + list_init(&garbage_list); + dpif_port_dump_start(&port_dump, backer->dpif); + while (dpif_port_dump_next(&port_dump, &port)) { + node = shash_find(&init_ofp_ports, port.name); + if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) { + garbage = xmalloc(sizeof *garbage); + garbage->odp_port = port.port_no; + list_push_front(&garbage_list, &garbage->list_node); + } + } + dpif_port_dump_done(&port_dump); + + LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) { + dpif_port_del(backer->dpif, garbage->odp_port); + list_remove(&garbage->list_node); + free(garbage); + } + + shash_add(&all_dpif_backers, type, backer); + + error = dpif_recv_set(backer->dpif, true); + if (error) { + VLOG_ERR("failed to listen on datapath of type %s: %s", + type, strerror(error)); + close_dpif_backer(backer); + return error; + } + + return error; +} + static int construct(struct ofproto *ofproto_) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - const char *name = ofproto->up.name; + struct shash_node *node, *next; int max_ports; int error; int i; - error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif); + error = open_dpif_backer(ofproto->up.type, &ofproto->backer); if (error) { - VLOG_ERR("failed to open datapath %s: %s", name, strerror(error)); return error; } - max_ports = dpif_get_max_ports(ofproto->dpif); + max_ports = dpif_get_max_ports(ofproto->backer->dpif); ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX)); ofproto->n_matches = 0; - dpif_flow_flush(ofproto->dpif); - dpif_recv_purge(ofproto->dpif); - - error = dpif_recv_set(ofproto->dpif, true); - if (error) { - VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error)); - dpif_close(ofproto->dpif); - return error; - } - 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); @@ -782,11 +1364,10 @@ construct(struct ofproto *ofproto_) } ofproto->has_bonded_bundles = false; - timer_set_duration(&ofproto->next_expiration, 1000); - 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]; @@ -795,8 +1376,6 @@ 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); @@ -808,6 +1387,27 @@ construct(struct ofproto *ofproto_) hmap_init(&ofproto->vlandev_map); 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; + + SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) { + struct iface_hint *iface_hint = node->data; + + if (!strcmp(iface_hint->br_name, ofproto->up.name)) { + /* Check if the datapath already has this port. */ + if (dpif_port_exists(ofproto->backer->dpif, node->name)) { + sset_add(&ofproto->ports, node->name); + } + + free(iface_hint->br_name); + free(iface_hint->br_type); + free(iface_hint); + shash_delete(&init_ofp_ports, node); + } + } + hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node, hash_string(ofproto->up.name, 0)); memset(&ofproto->stats, 0, sizeof ofproto->stats); @@ -816,6 +1416,22 @@ 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; } @@ -850,7 +1466,7 @@ 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; } @@ -932,7 +1548,11 @@ destruct(struct ofproto *ofproto_) hmap_destroy(&ofproto->vlandev_map); hmap_destroy(&ofproto->realdev_vid_map); - dpif_close(ofproto->dpif); + sset_destroy(&ofproto->ports); + sset_destroy(&ofproto->ghost_ports); + sset_destroy(&ofproto->port_poll_set); + + close_dpif_backer(ofproto->backer); } static int @@ -940,29 +1560,11 @@ run_fast(struct ofproto *ofproto_) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); struct ofport_dpif *ofport; - unsigned int work; HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { port_run_fast(ofport); } - /* Handle one or more batches of upcalls, until there's nothing left to do - * or until we do a fixed total amount of work. - * - * We do work in batches because it can be much cheaper to set up a number - * of flows and fire off their patches all at once. We do multiple batches - * because in some cases handling a packet can cause another packet to be - * queued almost immediately as part of the return flow. Both - * optimizations can make major improvements on some benchmarks and - * presumably for real traffic as well. */ - work = 0; - while (work < FLOW_MISS_MAX_BATCH) { - int retval = handle_upcalls(ofproto, FLOW_MISS_MAX_BATCH - work); - if (retval <= 0) { - return -retval; - } - work += retval; - } return 0; } @@ -977,18 +1579,12 @@ run(struct ofproto *ofproto_) if (!clogged) { complete_operations(ofproto); } - dpif_run(ofproto->dpif); error = run_fast(ofproto_); if (error) { return error; } - if (timer_expired(&ofproto->next_expiration)) { - int delay = expire(ofproto); - timer_set_duration(&ofproto->next_expiration, delay); - } - if (ofproto->netflow) { if (netflow_run(ofproto->netflow)) { send_netflow_active_timeouts(ofproto); @@ -1006,44 +1602,22 @@ 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; } } } @@ -1080,12 +1654,12 @@ wait(struct ofproto *ofproto_) poll_immediate_wake(); } - dpif_wait(ofproto->dpif); - dpif_recv_wait(ofproto->dpif); + dpif_wait(ofproto->backer->dpif); + dpif_recv_wait(ofproto->backer->dpif); 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) { @@ -1099,12 +1673,10 @@ 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(); - } else { - timer_wait(&ofproto->next_expiration); } if (ofproto->governor) { governor_wait(ofproto->governor); @@ -1124,23 +1696,27 @@ static void flush(struct ofproto *ofproto_) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - struct facet *facet, *next_facet; - - HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { - /* Mark the facet as not installed so that facet_remove() doesn't - * bother trying to uninstall it. There is no point in uninstalling it - * individually since we are about to blow away all the facets with - * dpif_flow_flush(). */ - struct subfacet *subfacet; + struct subfacet *subfacet, *next_subfacet; + struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; + int n_batch; - LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) { - subfacet->path = SF_NOT_INSTALLED; - subfacet->dp_packet_count = 0; - subfacet->dp_byte_count = 0; + n_batch = 0; + HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node, + &ofproto->subfacets) { + if (subfacet->path != SF_NOT_INSTALLED) { + batch[n_batch++] = subfacet; + if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { + subfacet_destroy_batch(ofproto, batch, n_batch); + n_batch = 0; + } + } else { + subfacet_destroy(subfacet); } - facet_remove(facet); } - dpif_flow_flush(ofproto->dpif); + + if (n_batch > 0) { + subfacet_destroy_batch(ofproto, batch, n_batch); + } } static void @@ -1170,7 +1746,8 @@ get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots) strcpy(ots->name, "classifier"); - dpif_get_dp_stats(ofproto->dpif, &s); + dpif_get_dp_stats(ofproto->backer->dpif, &s); + ots->lookup_count = htonll(s.n_hit + s.n_missed); ots->matched_count = htonll(s.n_hit + ofproto->n_matches); } @@ -1194,22 +1771,57 @@ 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; - port->odp_port = ofp_port_to_odp_port(port->up.ofp_port); + ofproto->backer->need_revalidate = REV_RECONFIGURE; port->bundle = NULL; port->cfm = 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)) { + /* XXX By bailing out here, we don't do required sFlow work. */ + port->odp_port = OVSP_NONE; + return 0; + } + + error = dpif_port_query_by_name(ofproto->backer->dpif, + netdev_vport_get_dpif_port(netdev), + &dpif_port); + if (error) { + return error; + } + + port->odp_port = dpif_port.port_no; + + 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)); + } + dpif_port_destroy(&dpif_port); if (ofproto->sflow) { - dpif_sflow_add_port(ofproto->sflow, port_); + dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port); } return 0; @@ -1220,8 +1832,28 @@ 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, 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. */ + if (!port->tnl_port) { + dpif_port_del(ofproto->backer->dpif, port->odp_port); + } + ofproto->backer->need_revalidate = REV_RECONFIGURE; + } - ofproto->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); + sset_find_and_delete(&ofproto->ghost_ports, devname); + ofproto->backer->need_revalidate = REV_RECONFIGURE; bundle_remove(port_); set_cfm(port_, NULL); if (ofproto->sflow) { @@ -1252,7 +1884,7 @@ 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); @@ -1271,23 +1903,49 @@ set_sflow(struct ofproto *ofproto_, if (!ds) { struct ofport_dpif *ofport; - ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif); + ds = ofproto->sflow = dpif_sflow_create(); HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { - dpif_sflow_add_port(ds, &ofport->up); + 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) { @@ -1301,7 +1959,7 @@ 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; + ofproto->backer->need_revalidate = REV_RECONFIGURE; ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev)); } @@ -1311,48 +1969,27 @@ set_cfm(struct ofport *ofport_, const struct cfm_settings *s) error = EINVAL; } - cfm_destroy(ofport->cfm); - ofport->cfm = NULL; - return error; -} - -static int -get_cfm_fault(const struct ofport *ofport_) -{ - struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); - - return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1; -} - -static int -get_cfm_opup(const struct ofport *ofport_) -{ - struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); - - return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1; + cfm_destroy(ofport->cfm); + ofport->cfm = NULL; + return error; } -static int -get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps, - size_t *n_rmps) +static bool +get_cfm_status(const struct ofport *ofport_, + struct ofproto_cfm_status *status) { struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); if (ofport->cfm) { - cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps); - return 0; + 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 -1; + return false; } } - -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. */ @@ -1389,7 +2026,7 @@ 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) { @@ -1457,12 +2094,13 @@ 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(); @@ -1562,7 +2200,7 @@ 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); } } } @@ -1648,7 +2286,7 @@ set_queues(struct ofport *ofport_, uint8_t dscp; dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK; - if (dpif_queue_to_priority(ofproto->dpif, qdscp_list[i].queue, + if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue, &priority)) { continue; } @@ -1660,12 +2298,12 @@ 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)); @@ -1673,7 +2311,7 @@ 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); @@ -1700,7 +2338,7 @@ 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) { @@ -1713,7 +2351,6 @@ 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); } } @@ -1777,7 +2414,7 @@ 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; @@ -1794,8 +2431,7 @@ 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) + struct lacp_slave_settings *lacp) { struct ofport_dpif *port; @@ -1805,7 +2441,7 @@ 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); } @@ -1818,12 +2454,10 @@ 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; } @@ -1846,7 +2480,7 @@ 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; } } } @@ -1882,8 +2516,8 @@ 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) { @@ -1918,7 +2552,7 @@ 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); @@ -1931,8 +2565,7 @@ 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; } } @@ -1950,7 +2583,7 @@ 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); @@ -2024,16 +2657,15 @@ 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); @@ -2148,7 +2780,7 @@ 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); @@ -2333,9 +2965,10 @@ 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; @@ -2354,8 +2987,8 @@ 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) { @@ -2395,6 +3028,8 @@ mirror_get_stats(struct ofproto *ofproto_, void *aux, return 0; } + push_all_stats(); + *packets = mirror->packet_count; *bytes = mirror->byte_count; @@ -2406,7 +3041,7 @@ 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; } @@ -2423,14 +3058,16 @@ 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 -set_mac_idle_time(struct ofproto *ofproto_, unsigned int idle_time) +set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time, + size_t max_entries) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); mac_learning_set_idle_time(ofproto->ml, idle_time); + mac_learning_set_max_entries(ofproto->ml, max_entries); } /* Ports. */ @@ -2445,16 +3082,40 @@ get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port) static struct ofport_dpif * get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) { - return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port)); + struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port); + return port && &ofproto->up == port->up.ofproto ? port : NULL; } static void -ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port, +ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto, + struct ofproto_port *ofproto_port, struct dpif_port *dpif_port) { ofproto_port->name = dpif_port->name; ofproto_port->type = dpif_port->type; - ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no); + 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 @@ -2480,6 +3141,13 @@ 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); @@ -2502,7 +3170,7 @@ 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; } } @@ -2525,37 +3193,85 @@ port_query_by_name(const struct ofproto *ofproto_, const char *devname, struct dpif_port dpif_port; int error; - error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port); + 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; + } + error = dpif_port_query_by_name(ofproto->backer->dpif, + devname, &dpif_port); if (!error) { - ofproto_port_from_dpif_port(ofproto_port, &dpif_port); + ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port); } return error; } static int -port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp) +port_add(struct ofproto *ofproto_, struct netdev *netdev) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - uint32_t odp_port = *ofp_portp != OFPP_NONE ? *ofp_portp : 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->dpif, netdev, &odp_port); - if (!error) { - *ofp_portp = odp_port_to_ofp_port(odp_port); + 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_); - int error; + struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); + int error = 0; - error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port)); - if (!error) { - struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); - if (ofport) { + if (!ofport) { + return 0; + } + + 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 @@ -2572,9 +3288,11 @@ 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 @@ -2624,37 +3342,56 @@ ofproto_update_local_port_stats(const struct ofproto *ofproto_, } struct port_dump_state { - struct dpif_port_dump dump; - bool done; + uint32_t bucket; + uint32_t offset; + bool ghost; + + struct ofproto_port port; + bool has_port; }; static int -port_dump_start(const struct ofproto *ofproto_, void **statep) +port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep) { - struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - struct port_dump_state *state; - - *statep = state = xmalloc(sizeof *state); - dpif_port_dump_start(&state->dump, ofproto->dpif); - state->done = false; + *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_; - struct dpif_port dpif_port; + const struct sset *sset; + struct sset_node *node; - if (dpif_port_dump_next(&state->dump, &dpif_port)) { - ofproto_port_from_dpif_port(port, &dpif_port); - return 0; - } else { - int error = dpif_port_dump_done(&state->dump); - state->done = true; - return error ? error : EOF; + 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, &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; } static int @@ -2662,8 +3399,8 @@ port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) { struct port_dump_state *state = state_; - if (!state->done) { - dpif_port_dump_done(&state->dump); + if (state->has_port) { + ofproto_port_destroy(&state->port); } free(state); return 0; @@ -2673,14 +3410,26 @@ static int port_poll(const struct ofproto *ofproto_, char **devnamep) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - return dpif_port_poll(ofproto->dpif, devnamep); + + if (ofproto->port_poll_errno) { + int error = ofproto->port_poll_errno; + ofproto->port_poll_errno = 0; + return error; + } + + if (sset_is_empty(&ofproto->port_poll_set)) { + return EAGAIN; + } + + *devnamep = sset_pop(&ofproto->port_poll_set); + return 0; } static void port_poll_wait(const struct ofproto *ofproto_) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - dpif_port_poll_wait(ofproto->dpif); + dpif_port_poll_wait(ofproto->backer->dpif); } static int @@ -2705,18 +3454,19 @@ port_is_lacp_current(const struct ofport *ofport_) * It's possible to batch more than that, but the benefit might be minimal. */ struct flow_miss { struct hmap_node hmap_node; + struct ofproto_dpif *ofproto; struct flow flow; enum odp_key_fitness key_fitness; const struct nlattr *key; size_t key_len; - ovs_be16 initial_tci; + struct initial_vals initial_vals; struct list packets; enum dpif_upcall_type upcall_type; + uint32_t odp_in_port; }; 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. */ }; @@ -2747,15 +3497,11 @@ 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); } @@ -2771,17 +3517,19 @@ 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; } } @@ -2798,7 +3546,7 @@ 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 @@ -2808,7 +3556,6 @@ 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; @@ -2891,8 +3638,8 @@ handle_flow_miss_without_facet(struct flow_miss *miss, 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, 0, packet); ctx.resubmit_stats = &stats; xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions); @@ -2931,9 +3678,7 @@ handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, struct subfacet *subfacet; struct ofpbuf *packet; - subfacet = subfacet_create(facet, - miss->key_fitness, miss->key, miss->key_len, - miss->initial_tci, now); + subfacet = subfacet_create(facet, miss, now); LIST_FOR_EACH (packet, list_node, &miss->packets) { struct flow_miss_op *op = &ops[*n_ops]; @@ -2954,7 +3699,6 @@ 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; @@ -2976,7 +3720,8 @@ 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; @@ -2994,12 +3739,13 @@ handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet, } } -/* Handles flow miss 'miss' on 'ofproto'. May add any required datapath - * operations to 'ops', incrementing '*n_ops' for each new op. */ +/* Handles flow miss 'miss'. May add any required datapath operations + * to 'ops', incrementing '*n_ops' for each new op. */ static void -handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss, - struct flow_miss_op *ops, size_t *n_ops) +handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops, + size_t *n_ops) { + struct ofproto_dpif *ofproto = miss->ofproto; struct facet *facet; long long int now; uint32_t hash; @@ -3025,67 +3771,166 @@ handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss, handle_flow_miss_with_facet(miss, facet, now, ops, n_ops); } -/* Like odp_flow_key_to_flow(), this function converts the 'key_len' bytes of - * OVS_KEY_ATTR_* attributes in 'key' to a flow structure in 'flow' and returns - * an ODP_FIT_* value that indicates how well 'key' fits our expectations for - * what a flow key should contain. +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 + * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto' + * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that + * 'packet' ingressed. * - * This function also includes some logic to help make VLAN splinters - * transparent to the rest of the upcall processing logic. In particular, if - * the extracted in_port is a VLAN splinter port, it replaces flow->in_port by - * the "real" port, sets flow->vlan_tci correctly for the VLAN of the VLAN - * splinter port, and pushes a VLAN header onto 'packet' (if it is nonnull). + * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets + * 'flow''s in_port to OFPP_NONE. * - * 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.) - */ -static enum odp_key_fitness -ofproto_dpif_extract_flow_key(const struct ofproto_dpif *ofproto, - const struct nlattr *key, size_t key_len, - struct flow *flow, ovs_be16 *initial_tci, - struct ofpbuf *packet) + * This function does post-processing on data returned from + * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest + * of the upcall processing logic. In particular, if the extracted in_port is + * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets + * 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 '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.) Also, if received on an IP tunnel, sets + * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS. + * + * 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. */ +static int +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, + struct initial_vals *initial_vals) { + const struct ofport_dpif *port; enum odp_key_fitness fitness; + int error = ENODEV; fitness = odp_flow_key_to_flow(key, key_len, flow); - flow->in_port = odp_port_to_ofp_port(flow->in_port); if (fitness == ODP_FIT_ERROR) { - return fitness; + error = EINVAL; + goto exit; } - *initial_tci = flow->vlan_tci; - if (vsp_adjust_flow(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'. - * - * 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. */ - eth_push_vlan(packet, flow->vlan_tci); + if (initial_vals) { + initial_vals->vlan_tci = flow->vlan_tci; + initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos; + } + + if (odp_in_port) { + *odp_in_port = flow->in_port; + } + + if (tnl_port_should_receive(flow)) { + const struct ofport *ofport = tnl_port_receive(flow); + if (!ofport) { + flow->in_port = OFPP_NONE; + goto exit; + } + port = ofport_dpif_cast(ofport); + + /* We can't reproduce 'key' from 'flow'. */ + fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness; + + /* XXX: Since the tunnel module is not scoped per backer, 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); + } else { + port = odp_port_to_ofport(backer, flow->in_port); + if (!port) { + flow->in_port = OFPP_NONE; + goto exit; } - /* Let the caller know that we can't reproduce 'key' from 'flow'. */ - if (fitness == ODP_FIT_PERFECT) { - fitness = ODP_FIT_TOO_MUCH; + 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'. + * + * 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. */ + eth_push_vlan(packet, flow->vlan_tci); + } + /* 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); + } - return fitness; +exit: + if (fitnessp) { + *fitnessp = fitness; + } + return error; } static void -handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls, +handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls, size_t n_upcalls) { struct dpif_upcall *upcall; @@ -3112,28 +3957,58 @@ handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls, for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { struct flow_miss *miss = &misses[n_misses]; struct flow_miss *existing_miss; + struct ofproto_dpif *ofproto; + uint32_t odp_in_port; struct flow flow; uint32_t hash; + int error; - /* Obtain metadata and check userspace/kernel agreement on flow match, - * then set 'flow''s header pointers. */ - miss->key_fitness = ofproto_dpif_extract_flow_key( - ofproto, upcall->key, upcall->key_len, - &flow, &miss->initial_tci, upcall->packet); - if (miss->key_fitness == ODP_FIT_ERROR) { + error = ofproto_receive(backer, upcall->packet, upcall->key, + upcall->key_len, &flow, &miss->key_fitness, + &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. 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; } - flow_extract(upcall->packet, flow.skb_priority, + + 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; miss->key = upcall->key; miss->key_len = upcall->key_len; miss->upcall_type = upcall->type; + miss->odp_in_port = odp_in_port; list_init(&miss->packets); n_misses++; @@ -3147,42 +4022,28 @@ handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls, * operations to batch. */ n_ops = 0; HMAP_FOR_EACH (miss, hmap_node, &todo) { - handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops); + 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++) { dpif_ops[i] = &flow_miss_ops[i].dpif_op; } - dpif_operate(ofproto->dpif, dpif_ops, n_ops); + 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. */ @@ -3200,43 +4061,105 @@ 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; } } static void -handle_sflow_upcall(struct ofproto_dpif *ofproto, +handle_sflow_upcall(struct dpif_backer *backer, const struct dpif_upcall *upcall) { + struct ofproto_dpif *ofproto; union user_action_cookie cookie; - enum odp_key_fitness fitness; - ovs_be16 initial_tci; struct flow flow; + uint32_t odp_in_port; - fitness = ofproto_dpif_extract_flow_key(ofproto, upcall->key, - upcall->key_len, &flow, - &initial_tci, upcall->packet); - if (fitness == ODP_FIT_ERROR) { + if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len, + &flow, NULL, &ofproto, &odp_in_port, NULL) + || !ofproto->sflow) { + return; + } + + memset(&cookie, 0, sizeof cookie); + memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow); + dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, + odp_in_port, &cookie); +} + +static void +handle_flow_sample_upcall(struct dpif_backer *backer, + const struct dpif_upcall *upcall) +{ + 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; } - memcpy(&cookie, &upcall->userdata, sizeof(cookie)); - dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie); + 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 ofproto_dpif *ofproto, unsigned int max_batch) +handle_upcalls(struct dpif_backer *backer, unsigned int max_batch) { struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH]; @@ -3245,7 +4168,7 @@ handle_upcalls(struct ofproto_dpif *ofproto, 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++) { @@ -3255,7 +4178,7 @@ handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch) ofpbuf_use_stub(buf, miss_buf_stubs[n_misses], sizeof miss_buf_stubs[n_misses]); - error = dpif_recv(ofproto->dpif, upcall, buf); + error = dpif_recv(backer->dpif, upcall, buf); if (error) { ofpbuf_uninit(buf); break; @@ -3268,9 +4191,17 @@ handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch) break; case SFLOW_UPCALL: - if (ofproto->sflow) { - handle_sflow_upcall(ofproto, upcall); - } + handle_sflow_upcall(backer, upcall); + ofpbuf_uninit(buf); + break; + + case FLOW_SAMPLE_UPCALL: + handle_flow_sample_upcall(backer, upcall); + ofpbuf_uninit(buf); + break; + + case IPFIX_UPCALL: + handle_ipfix_upcall(backer, upcall); ofpbuf_uninit(buf); break; @@ -3281,7 +4212,7 @@ handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch) } /* Handle deferred MISS_UPCALL processing. */ - handle_miss_upcalls(ofproto, misses, n_misses); + handle_miss_upcalls(backer, misses, n_misses); for (i = 0; i < n_misses; i++) { ofpbuf_uninit(&miss_bufs[i]); } @@ -3292,7 +4223,7 @@ handle_upcalls(struct ofproto_dpif *ofproto, unsigned int max_batch) /* Flow expiration. */ static int subfacet_max_idle(const struct ofproto_dpif *); -static void update_stats(struct ofproto_dpif *); +static void update_stats(struct dpif_backer *); static void rule_expire(struct rule_dpif *); static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle); @@ -3303,42 +4234,56 @@ static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle); * * Returns the number of milliseconds after which it should be called again. */ static int -expire(struct ofproto_dpif *ofproto) +expire(struct dpif_backer *backer) { - struct rule_dpif *rule, *next_rule; - struct oftable *table; - int dp_max_idle; + struct ofproto_dpif *ofproto; + int max_idle = INT32_MAX; - /* Update stats for each flow in the datapath. */ - update_stats(ofproto); + /* Periodically clear out the drop keys in an effort to keep them + * relatively few. */ + drop_key_clear(backer); - /* Expire subfacets that have been idle too long. */ - dp_max_idle = subfacet_max_idle(ofproto); - expire_subfacets(ofproto, dp_max_idle); + /* Update stats for each flow in the backer. */ + update_stats(backer); - /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ - OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { - struct cls_cursor cursor; + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { + struct rule *rule, *next_rule; + int dp_max_idle; - cls_cursor_init(&cursor, &table->cls, NULL); - CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { - rule_expire(rule); + if (ofproto->backer != backer) { + continue; } - } - /* All outstanding data in existing flows has been accounted, so it's a - * good time to do bond rebalancing. */ - if (ofproto->has_bonded_bundles) { - struct ofbundle *bundle; + /* 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); + + max_idle = MIN(max_idle, dp_max_idle); + + /* Expire OpenFlow flows whose idle_timeout or hard_timeout + * has passed. */ + 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 + * good time to do bond rebalancing. */ + if (ofproto->has_bonded_bundles) { + struct ofbundle *bundle; - HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { - if (bundle->bond) { - bond_rebalance(bundle->bond, &ofproto->revalidate_set); + HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { + if (bundle->bond) { + bond_rebalance(bundle->bond, &backer->revalidate_set); + } } } } - return MIN(dp_max_idle, 1000); + return MIN(max_idle, 1000); } /* Updates flow table statistics given that the datapath just reported 'stats' @@ -3373,13 +4318,12 @@ 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 * about, or a flow that shouldn't be installed but was anyway. Delete it. */ static void -delete_unexpected_flow(struct dpif *dpif, +delete_unexpected_flow(struct ofproto_dpif *ofproto, const struct nlattr *key, size_t key_len) { if (!VLOG_DROP_WARN(&rl)) { @@ -3387,12 +4331,12 @@ delete_unexpected_flow(struct dpif *dpif, ds_init(&s); odp_flow_key_format(key, key_len, &s); - VLOG_WARN("unexpected flow from datapath %s", ds_cstr(&s)); + VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s)); ds_destroy(&s); } COVERAGE_INC(facet_unexpected); - dpif_flow_del(dpif, key, key_len, NULL); + dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL); } /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. @@ -3405,22 +4349,51 @@ delete_unexpected_flow(struct dpif *dpif, * 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 ofproto_dpif *p) +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, p->dpif); + 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; + struct ofport_dpif *ofport; + uint32_t key_hash; + + if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto, + NULL, NULL)) { + 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); + } - subfacet = subfacet_find(p, key, key_len); + key_hash = odp_flow_key_hash(key, key_len); + 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; @@ -3430,11 +4403,17 @@ update_stats(struct ofproto_dpif *p) case SF_NOT_INSTALLED: default: - delete_unexpected_flow(p->dpif, key, key_len); + 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 @@ -3525,35 +4504,6 @@ subfacet_max_idle(const struct ofproto_dpif *ofproto) return bucket * BUCKET_WIDTH; } -enum { EXPIRE_MAX_BATCH = 50 }; - -static void -expire_batch(struct ofproto_dpif *ofproto, struct subfacet **subfacets, int n) -{ - struct odputil_keybuf keybufs[EXPIRE_MAX_BATCH]; - struct dpif_op ops[EXPIRE_MAX_BATCH]; - struct dpif_op *opsp[EXPIRE_MAX_BATCH]; - struct ofpbuf keys[EXPIRE_MAX_BATCH]; - struct dpif_flow_stats stats[EXPIRE_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.stats = &stats[i]; - opsp[i] = &ops[i]; - } - - dpif_operate(ofproto->dpif, opsp, n); - for (i = 0; i < n; i++) { - subfacet_reset_dp_stats(subfacets[i], &stats[i]); - subfacets[i]->path = SF_NOT_INSTALLED; - subfacet_destroy(subfacets[i]); - } -} - static void expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) { @@ -3565,7 +4515,7 @@ expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) long long int special_cutoff = time_msec() - 10000; struct subfacet *subfacet, *next_subfacet; - struct subfacet *batch[EXPIRE_MAX_BATCH]; + struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH]; int n_batch; n_batch = 0; @@ -3579,8 +4529,8 @@ expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) if (subfacet->used < cutoff) { if (subfacet->path != SF_NOT_INSTALLED) { batch[n_batch++] = subfacet; - if (n_batch >= EXPIRE_MAX_BATCH) { - expire_batch(ofproto, batch, n_batch); + if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) { + subfacet_destroy_batch(ofproto, batch, n_batch); n_batch = 0; } } else { @@ -3590,7 +4540,7 @@ expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle) } if (n_batch > 0) { - expire_batch(ofproto, batch, n_batch); + subfacet_destroy_batch(ofproto, batch, n_batch); } } @@ -3660,6 +4610,8 @@ 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; } @@ -3670,9 +4622,7 @@ 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, @@ -3683,12 +4633,11 @@ execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, int error; ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); - odp_flow_key_from_flow(&key, flow, ofp_port_to_odp_port(flow->in_port)); + odp_flow_key_from_flow(&key, flow, + ofp_port_to_odp_port(ofproto, flow->in_port)); - error = dpif_execute(ofproto->dpif, key.data, key.size, + error = dpif_execute(ofproto->backer->dpif, key.data, key.size, odp_actions, actions_len, packet); - - ofpbuf_delete(packet); return !error; } @@ -3705,7 +4654,7 @@ 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) { @@ -3735,8 +4684,17 @@ 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 @@ -3745,7 +4703,7 @@ 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, @@ -3756,7 +4714,7 @@ 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; @@ -3775,8 +4733,6 @@ 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) { @@ -3835,8 +4791,8 @@ 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); @@ -3901,14 +4857,26 @@ 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))) { + && (ofproto->backer->need_revalidate + || tag_set_intersects(&ofproto->backer->revalidate_set, + facet->tags))) { facet_revalidate(facet); + + /* facet_revalidate() may have destroyed '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) { @@ -3989,13 +4957,11 @@ 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); @@ -4027,8 +4993,7 @@ 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) { @@ -4066,7 +5031,10 @@ 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. */ + * 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. */ static void facet_revalidate(struct facet *facet) { @@ -4087,6 +5055,25 @@ 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; + } + } + new_rule = rule_dpif_lookup(ofproto, &facet->flow); /* Calculate new datapath actions. @@ -4105,7 +5092,7 @@ facet_revalidate(struct facet *facet) 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); @@ -4193,9 +5180,9 @@ 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; @@ -4207,13 +5194,43 @@ 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); + 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) { @@ -4222,19 +5239,20 @@ 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); @@ -4243,18 +5261,15 @@ flow_push_stats(struct rule_dpif *rule, /* Subfacets. */ 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) +subfacet_find(struct ofproto_dpif *ofproto, + 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; } } @@ -4263,26 +5278,30 @@ subfacet_find__(struct ofproto_dpif *ofproto, } /* Searches 'facet' (within 'ofproto') for a subfacet with the specified - * 'key_fitness', 'key', and 'key_len'. Returns the existing subfacet if - * there is one, otherwise creates and returns a new subfacet. + * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the + * existing subfacet if there is one, otherwise creates and returns a + * new subfacet. * * If the returned subfacet is new, then subfacet->actions will be NULL, in * which case the caller must populate the actions with * subfacet_make_actions(). */ static struct subfacet * -subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness, - const struct nlattr *key, size_t key_len, - ovs_be16 initial_tci, long long int now) +subfacet_create(struct facet *facet, struct flow_miss *miss, + long long int now) { struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto); - uint32_t key_hash = odp_flow_key_hash(key, key_len); + enum odp_key_fitness key_fitness = miss->key_fitness; + const struct nlattr *key = miss->key; + size_t key_len = miss->key_len; + uint32_t key_hash; struct subfacet *subfacet; + key_hash = odp_flow_key_hash(key, key_len); + if (list_is_empty(&facet->subfacets)) { subfacet = &facet->one_subfacet; } else { - subfacet = subfacet_find__(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; @@ -4300,14 +5319,10 @@ subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness, 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; @@ -4316,30 +5331,13 @@ subfacet_create(struct facet *facet, enum odp_key_fitness key_fitness, ? SLOW_MATCH : 0); subfacet->path = SF_NOT_INSTALLED; - subfacet->initial_tci = initial_tci; + subfacet->initial_vals = miss->initial_vals; + subfacet->odp_in_port = miss->odp_in_port; + ofproto->subfacet_add_count++; return subfacet; } -/* Searches 'ofproto' for a subfacet with the given 'key', 'key_len', and - * 'flow'. Returns the subfacet if one exists, otherwise NULL. */ -static struct subfacet * -subfacet_find(struct ofproto_dpif *ofproto, - const struct nlattr *key, size_t key_len) -{ - uint32_t key_hash = odp_flow_key_hash(key, key_len); - enum odp_key_fitness fitness; - struct flow flow; - - fitness = odp_flow_key_to_flow(key, key_len, &flow); - flow.in_port = odp_port_to_ofp_port(flow.in_port); - if (fitness == ODP_FIT_ERROR) { - return NULL; - } - - return subfacet_find__(ofproto, key, key_len, key_hash, &flow); -} - /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from * its facet within 'ofproto', and frees it. */ static void @@ -4348,6 +5346,10 @@ 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); @@ -4373,20 +5375,29 @@ subfacet_destroy(struct subfacet *subfacet) } } -/* 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) +subfacet_destroy_batch(struct ofproto_dpif *ofproto, + struct subfacet **subfacets, int n) { - if (!subfacet->key) { - struct flow *flow = &subfacet->facet->flow; + struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH]; + struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH]; + struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH]; + int i; - ofpbuf_use_stack(key, keybuf, sizeof *keybuf); - odp_flow_key_from_flow(key, flow, ofp_port_to_odp_port(flow->in_port)); - } else { - ofpbuf_use_const(key, subfacet->key, subfacet->key_len); + for (i = 0; i < n; i++) { + ops[i].type = DPIF_OP_FLOW_DEL; + 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]; + } + + dpif_operate(ofproto->backer->dpif, opsp, n); + for (i = 0; i < n; i++) { + subfacet_reset_dp_stats(subfacets[i], &stats[i]); + subfacets[i]->path = SF_NOT_INSTALLED; + subfacet_destroy(subfacets[i]); + run_fast_rl(); } } @@ -4403,8 +5414,8 @@ subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet, struct action_xlate_ctx ctx; - action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci, - rule, 0, packet); + 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; @@ -4438,9 +5449,7 @@ 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; @@ -4454,9 +5463,8 @@ subfacet_install(struct subfacet *subfacet, &actions, &actions_len); } - subfacet_get_key(subfacet, &keybuf, &key); - ret = dpif_flow_put(ofproto->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); @@ -4482,21 +5490,19 @@ 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->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); } } @@ -4548,7 +5554,6 @@ 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); } } @@ -4708,6 +5713,12 @@ 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; @@ -4722,15 +5733,13 @@ rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) } } -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; @@ -4738,8 +5747,10 @@ 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; + initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos; 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); @@ -4748,7 +5759,14 @@ 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; } @@ -4767,28 +5785,69 @@ static int send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) { const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); + uint64_t odp_actions_stub[1024 / 8]; struct ofpbuf key, odp_actions; struct odputil_keybuf keybuf; uint32_t odp_port; struct flow flow; int error; - flow_extract(packet, 0, NULL, 0, &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); + flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow); + if (netdev_vport_is_patch(ofport->up.netdev)) { + struct ofproto_dpif *peer_ofproto; + struct dpif_flow_stats stats; + struct ofport_dpif *peer; + struct rule_dpif *rule; + + peer = ofport_get_peer(ofport); + if (!peer) { + return ENODEV; + } + + dpif_flow_stats_extract(&flow, packet, time_msec(), &stats); + netdev_vport_inc_tx(ofport->up.netdev, &stats); + netdev_vport_inc_rx(peer->up.netdev, &stats); + + flow.in_port = peer->up.ofp_port; + peer_ofproto = ofproto_dpif_cast(peer->up.ofproto); + rule = rule_dpif_lookup(peer_ofproto, &flow); + rule_dpif_execute(rule, &flow, packet); + + return 0; + } + + ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); + + if (ofport->tnl_port) { + struct dpif_flow_stats stats; + + odp_port = tnl_port_send(ofport->tnl_port, &flow); + if (odp_port == OVSP_NONE) { + return ENODEV; + } + + dpif_flow_stats_extract(&flow, packet, time_msec(), &stats); + netdev_vport_inc_tx(ofport->up.netdev, &stats); + odp_put_tunnel_action(&flow.tunnel, &odp_actions); + odp_put_skb_mark_action(flow.skb_mark, &odp_actions); + } else { + 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_stack(&key, &keybuf, sizeof keybuf); - odp_flow_key_from_flow(&key, &flow, ofp_port_to_odp_port(flow.in_port)); + 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); + compose_ipfix_action(ofproto, &odp_actions, &flow); nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port); - error = dpif_execute(ofproto->dpif, + error = dpif_execute(ofproto->backer->dpif, key.data, key.size, odp_actions.data, odp_actions.size, packet); @@ -4804,6 +5863,7 @@ send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet) /* 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 *); @@ -4833,10 +5893,11 @@ compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow, ofpbuf_use_stack(&buf, stub, stub_size); if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) { - uint32_t pid = dpif_port_get_pid(ofproto->dpif, UINT16_MAX); - odp_put_userspace_action(pid, &cookie, &buf); + 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; @@ -4846,14 +5907,43 @@ 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->dpif, - ofp_port_to_odp_port(flow->in_port)); + pid = dpif_port_get_pid(ofproto->backer->dpif, + ofp_port_to_odp_port(ofproto, flow->in_port)); + + 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); - return odp_put_userspace_action(pid, cookie, odp_actions); + nl_msg_end_nested(odp_actions, actions_offset); + nl_msg_end_nested(odp_actions, sample_offset); + return cookie_offset; } static void @@ -4888,7 +5978,7 @@ 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, @@ -4897,32 +5987,60 @@ 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; +} + +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 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. */ +/* 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) { @@ -4933,6 +6051,14 @@ 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. */ @@ -4948,7 +6074,7 @@ 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); + 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); @@ -4959,37 +6085,118 @@ 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(ofp_port); - ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci; - uint8_t flow_nw_tos = ctx->flow.nw_tos; - uint16_t out_port; + ovs_be16 flow_vlan_tci; + uint32_t flow_skb_mark; + uint8_t flow_nw_tos; + struct priority_to_dscp *pdscp; + uint32_t out_port, odp_port; - if (ofport) { - struct priority_to_dscp *pdscp; + /* 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"); + return; + } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) { + xlate_report(ctx, "OFPPC_NO_FWD set, skipping output"); + return; + } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) { + xlate_report(ctx, "STP not in forwarding state, skipping output"); + 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 (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) { - xlate_report(ctx, "OFPPC_NO_FWD set, skipping output"); + if (!peer) { + xlate_report(ctx, "Nonexistent patch port peer"); return; - } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) { - xlate_report(ctx, "STP not in forwarding state, skipping output"); + } + + 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; } - pdscp = get_priority(ofport, ctx->flow.skb_priority); - if (pdscp) { - ctx->flow.nw_tos &= ~IP_DSCP_MASK; - ctx->flow.nw_tos |= pdscp->dscp; + 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; + } } - } else { - /* We may not have an ofport record for this port, but it doesn't hurt - * to allow forwarding to it anyhow. Maybe such a port will appear - * later and we're pre-populating the flow table. */ + + 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->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); @@ -4997,7 +6204,11 @@ 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; } @@ -5007,62 +6218,74 @@ 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) { - /* TODO: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); @@ -5139,16 +6362,11 @@ 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); @@ -5156,6 +6374,16 @@ 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, @@ -5184,28 +6412,100 @@ execute_controller_action(struct action_xlate_ctx *ctx, int len, pin.send_len = len; flow_get_metadata(&ctx->flow, &pin.fmd); - connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin); - ofpbuf_delete(packet); + connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin); + 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) +{ + if (ctx->flow.dl_type != htons(ETH_TYPE_IP) && + ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) { + return false; + } + + if (ctx->flow.nw_ttl > 1) { + ctx->flow.nw_ttl--; + return false; + } else { + size_t i; + + for (i = 0; i < ids->n_controllers; i++) { + execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, + ids->cnt_ids[i]); + } + + /* Stop processing for current table. */ + return true; + } +} + +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 -compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids) +execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx) { - if (ctx->flow.dl_type != htons(ETH_TYPE_IP) && - ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) { + uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse); + + if (!eth_type_mpls(ctx->flow.dl_type)) { return false; } - if (ctx->flow.nw_ttl > 1) { - ctx->flow.nw_ttl--; + if (ttl > 1) { + ttl--; + set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl); return false; } else { - size_t i; - - for (i = 0; i < ids->n_controllers; i++) { - execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, - ids->cnt_ids[i]); - } + execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0); /* Stop processing for current table. */ return true; @@ -5281,7 +6581,8 @@ xlate_enqueue_action(struct action_xlate_ctx *ctx, int error; /* Translate queue to priority. */ - error = dpif_queue_to_priority(ctx->ofproto->dpif, queue_id, &priority); + error = dpif_queue_to_priority(ctx->ofproto->backer->dpif, + queue_id, &priority); if (error) { /* Fall back to ordinary output action. */ xlate_output_action(ctx, enqueue->port, 0, false); @@ -5314,7 +6615,8 @@ xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id) { uint32_t skb_priority; - if (!dpif_queue_to_priority(ctx->ofproto->dpif, queue_id, &skb_priority)) { + if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif, + queue_id, &skb_priority)) { ctx->flow.skb_priority = skb_priority; } else { /* Couldn't translate queue to a priority. Nothing to do. A warning @@ -5322,31 +6624,6 @@ 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_) { @@ -5427,6 +6704,23 @@ 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) { @@ -5448,25 +6742,38 @@ 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->base_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; @@ -5510,7 +6817,7 @@ do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, break; case OFPACT_PUSH_VLAN: - /* TODO:XXX 802.1AD(QinQ) */ + /* XXX 802.1AD(QinQ) */ ctx->flow.vlan_tci = htons(VLAN_CFI); break; @@ -5525,11 +6832,15 @@ 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: @@ -5541,11 +6852,15 @@ 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: @@ -5572,6 +6887,36 @@ 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; @@ -5586,10 +6931,6 @@ 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)); @@ -5616,7 +6957,7 @@ do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, break; case OFPACT_CLEAR_ACTIONS: - /* TODO:XXX + /* XXX * Nothing to do because writa-actions is not supported for now. * When writa-actions is supported, clear-actions also must * be supported at the same time. @@ -5630,23 +6971,44 @@ do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len, break; case OFPACT_GOTO_TABLE: { - /* TODO: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; } @@ -5655,14 +7017,40 @@ 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) { + ovs_be64 initial_tun_id = flow->tunnel.tun_id; + + /* 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 'flow' is largely cleared when transitioning between + * the input and output stages since it does not make sense to output + * a packet with the exact headers that it was received with (i.e. + * the destination IP is us). The one exception is the tun_id, which + * is preserved to allow use in later resubmit lookups and loads into + * registers. + * - 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; - memset(&ctx->base_flow.tunnel, 0, sizeof ctx->base_flow.tunnel); - ctx->base_flow.vlan_tci = initial_tci; + ctx->base_flow.vlan_tci = initial_vals->vlan_tci; + ctx->base_flow.tunnel.ip_tos = initial_vals->tunnel_ip_tos; + ctx->flow.tunnel.tun_id = initial_tun_id; ctx->rule = rule; ctx->packet = packet; ctx->may_learn = packet != NULL; @@ -5685,6 +7073,8 @@ 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); @@ -5705,14 +7095,12 @@ 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) { @@ -5738,15 +7126,32 @@ 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; + uint32_t local_odp_port; + + initial_vals.vlan_tci = ctx->base_flow.vlan_tci; + initial_vals.tunnel_ip_tos = ctx->base_flow.tunnel.ip_tos; add_sflow_action(ctx); - do_xlate_actions(ofpacts, ofpacts_len, ctx); + add_ipfix_action(ctx); + + 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)) { + ofpbuf_clear(ctx->odp_actions); + add_sflow_action(ctx); + add_ipfix_action(ctx); + } + } if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) { if (!hit_resubmit_limit) { @@ -5756,15 +7161,17 @@ 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); } } + local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL); if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow, + local_odp_port, ctx->odp_actions->data, ctx->odp_actions->size)) { ctx->slow |= SLOW_IN_BAND; @@ -5775,10 +7182,12 @@ xlate_actions(struct action_xlate_ctx *ctx, } } 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' @@ -6151,7 +7560,7 @@ 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)); } } @@ -6427,7 +7836,7 @@ 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; } } @@ -6445,13 +7854,13 @@ 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; } } } @@ -6461,9 +7870,8 @@ 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; @@ -6476,6 +7884,7 @@ 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; @@ -6486,18 +7895,21 @@ packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, struct ofpbuf odp_actions; ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); - odp_flow_key_from_flow(&key, flow, ofp_port_to_odp_port(flow->in_port)); + odp_flow_key_from_flow(&key, flow, + ofp_port_to_odp_port(ofproto, flow->in_port)); 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; + initial_vals.tunnel_ip_tos = 0; + action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL, packet_get_tcp_flags(packet, flow), packet); ctx.resubmit_stats = &stats; ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub); xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions); - dpif_execute(ofproto->dpif, key.data, key.size, + dpif_execute(ofproto->backer->dpif, key.data, key.size, odp_actions.data, odp_actions.size, packet); ofpbuf_uninit(&odp_actions); @@ -6530,7 +7942,7 @@ get_netflow_ids(const struct ofproto *ofproto_, { struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); - dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id); + dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id); } static void @@ -6594,10 +8006,10 @@ 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); } } @@ -6730,7 +8142,7 @@ ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], 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; @@ -6771,14 +8183,16 @@ ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], goto exit; } - /* Convert odp_key to flow. */ - error = ofproto_dpif_extract_flow_key(ofproto, odp_key.data, - odp_key.size, &flow, - &initial_tci, NULL); - if (error == ODP_FIT_ERROR) { + /* The user might have specified the wrong ofproto but within the + * same backer. That's OK, ofproto_receive() can find the right + * one for us. */ + if (ofproto_receive(ofproto->backer, NULL, odp_key.data, + odp_key.size, &flow, NULL, &ofproto, NULL, + &initial_vals)) { unixctl_command_reply_error(conn, "Invalid flow"); goto exit; } + ds_put_format(&result, "Bridge: %s\n", ofproto->up.name); } else { char *error_s; @@ -6789,8 +8203,8 @@ ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], goto exit; } - initial_tci = flow.vlan_tci; - vsp_adjust_flow(ofproto, &flow); + initial_vals.vlan_tci = flow.vlan_tci; + initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos; } /* Generate a packet, if requested. */ @@ -6798,15 +8212,17 @@ ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], packet = ofpbuf_new(0); flow_compose(packet, &flow); } - } else if (argc == 6) { - /* ofproto/trace dpname priority tun_id in_port packet */ + } 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 *packet_s = argv[5]; + 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); @@ -6820,15 +8236,16 @@ ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[], ds_put_cstr(&result, s); free(s); - flow_extract(packet, priority, NULL, in_port, &flow); + flow_extract(packet, priority, mark, NULL, in_port, &flow); flow.tunnel.tun_id = tun_id; - initial_tci = flow.vlan_tci; + initial_vals.vlan_tci = flow.vlan_tci; + initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos; } else { unixctl_command_reply_error(conn, "Bad command syntax"); goto exit; } - ofproto_trace(ofproto, &flow, packet, initial_tci, &result); + ofproto_trace(ofproto, &flow, packet, &initial_vals, &result); unixctl_command_reply(conn, ds_cstr(&result)); exit: @@ -6839,8 +8256,8 @@ 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; @@ -6870,7 +8287,7 @@ 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; @@ -6962,7 +8379,7 @@ 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) { @@ -6998,6 +8415,250 @@ ofproto_dpif_self_check(struct unixctl_conn *conn, ds_destroy(&reply); } +/* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list + * of the 'ofproto_shash' nodes. It is the responsibility of the caller + * to destroy 'ofproto_shash' and free the returned value. */ +static const struct shash_node ** +get_ofprotos(struct shash *ofproto_shash) +{ + const struct ofproto_dpif *ofproto; + + HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) { + char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name); + shash_add_nocopy(ofproto_shash, name, ofproto); + } + + return shash_sort(ofproto_shash); +} + +static void +ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED, + const char *argv[] OVS_UNUSED, + void *aux OVS_UNUSED) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + struct shash ofproto_shash; + const struct shash_node **sorted_ofprotos; + int i; + + shash_init(&ofproto_shash); + sorted_ofprotos = get_ofprotos(&ofproto_shash); + for (i = 0; i < shash_count(&ofproto_shash); i++) { + const struct shash_node *node = sorted_ofprotos[i]; + ds_put_format(&ds, "%s\n", node->name); + } + + shash_destroy(&ofproto_shash); + free(sorted_ofprotos); + + unixctl_command_reply(conn, ds_cstr(&ds)); + ds_destroy(&ds); +} + +static void +show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds) +{ + 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. */ + + 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)); + ds_put_format(ds, + "\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++) { + const struct shash_node *node = ports[i]; + 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:"); + } + + if (strcmp(type, "system")) { + struct netdev *netdev; + int error; + + ds_put_format(ds, " (%s", type); + + error = netdev_open(name, type, &netdev); + if (!error) { + struct smap config; + + smap_init(&config); + error = netdev_get_config(netdev, &config); + if (!error) { + const struct smap_node **nodes; + size_t i; + + nodes = smap_sort(&config); + for (i = 0; i < smap_count(&config); i++) { + const struct smap_node *node = nodes[i]; + ds_put_format(ds, "%c %s=%s", i ? ',' : ':', + node->key, node->value); + } + free(nodes); + } + smap_destroy(&config); + + netdev_close(netdev); + } + ds_put_char(ds, ')'); + } + ds_put_char(ds, '\n'); + } + free(ports); +} + +static void +ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc, + const char *argv[], void *aux OVS_UNUSED) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + const struct ofproto_dpif *ofproto; + + if (argc > 1) { + int i; + for (i = 1; i < argc; i++) { + ofproto = ofproto_dpif_lookup(argv[i]); + if (!ofproto) { + ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps " + "for help)", argv[i]); + unixctl_command_reply_error(conn, ds_cstr(&ds)); + return; + } + show_dp_format(ofproto, &ds); + } + } else { + struct shash ofproto_shash; + const struct shash_node **sorted_ofprotos; + int i; + + shash_init(&ofproto_shash); + sorted_ofprotos = get_ofprotos(&ofproto_shash); + for (i = 0; i < shash_count(&ofproto_shash); i++) { + const struct shash_node *node = sorted_ofprotos[i]; + show_dp_format(node->data, &ds); + } + + shash_destroy(&ofproto_shash); + free(sorted_ofprotos); + } + + unixctl_command_reply(conn, ds_cstr(&ds)); + ds_destroy(&ds); +} + +static void +ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn, + int argc OVS_UNUSED, const char *argv[], + void *aux OVS_UNUSED) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + const struct ofproto_dpif *ofproto; + struct subfacet *subfacet; + + ofproto = ofproto_dpif_lookup(argv[1]); + if (!ofproto) { + unixctl_command_reply_error(conn, "no such bridge"); + return; + } + + update_stats(ofproto->backer); + + 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); + if (subfacet->used) { + ds_put_format(&ds, "%.3fs", + (time_msec() - subfacet->used) / 1000.0); + } else { + ds_put_format(&ds, "never"); + } + if (subfacet->facet->tcp_flags) { + ds_put_cstr(&ds, ", flags:"); + packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags); + } + + ds_put_cstr(&ds, ", actions:"); + 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'); + } + + unixctl_command_reply(conn, ds_cstr(&ds)); + ds_destroy(&ds); +} + +static void +ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn, + int argc OVS_UNUSED, const char *argv[], + void *aux OVS_UNUSED) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + struct ofproto_dpif *ofproto; + + ofproto = ofproto_dpif_lookup(argv[1]); + if (!ofproto) { + unixctl_command_reply_error(conn, "no such bridge"); + return; + } + + flush(&ofproto->up); + + unixctl_command_reply(conn, ds_cstr(&ds)); + ds_destroy(&ds); +} + static void ofproto_dpif_unixctl_init(void) { @@ -7009,8 +8670,8 @@ ofproto_dpif_unixctl_init(void) unixctl_command_register( "ofproto/trace", - "bridge {tun_id in_port packet | odp_flow [-generate]}", - 2, 5, ofproto_unixctl_trace, NULL); + "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}", + 2, 6, 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, @@ -7021,6 +8682,14 @@ ofproto_dpif_unixctl_init(void) ofproto_dpif_unclog, NULL); unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1, ofproto_dpif_self_check, NULL); + unixctl_command_register("dpif/dump-dps", "", 0, 0, + ofproto_unixctl_dpif_dump_dps, NULL); + unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX, + ofproto_unixctl_dpif_show, NULL); + unixctl_command_register("dpif/dump-flows", "bridge", 1, 1, + ofproto_unixctl_dpif_dump_flows, NULL); + unixctl_command_register("dpif/del-flows", "bridge", 1, 1, + ofproto_unixctl_dpif_del_flows, NULL); } /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.) @@ -7041,7 +8710,7 @@ 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); @@ -7080,16 +8749,17 @@ vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto, uint32_t realdev_odp_port, ovs_be16 vlan_tci) { if (!hmap_is_empty(&ofproto->realdev_vid_map)) { - uint16_t realdev_ofp_port = odp_port_to_ofp_port(realdev_odp_port); + uint16_t realdev_ofp_port; int vid = vlan_tci_to_vid(vlan_tci); const struct vlan_splinter *vsp; + realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port); HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node, hash_realdev_vid(realdev_ofp_port, vid), &ofproto->realdev_vid_map) { if (vsp->realdev_ofp_port == realdev_ofp_port && vsp->vid == vid) { - return ofp_port_to_odp_port(vsp->vlandev_ofp_port); + return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port); } } } @@ -7204,12 +8874,131 @@ vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid) VLOG_ERR("duplicate vlan device record"); } } - + +static uint32_t +ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port) +{ + const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); + return ofport ? ofport->odp_port : OVSP_NONE; +} + +static struct ofport_dpif * +odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port) +{ + struct ofport_dpif *port; + + HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node, + hash_int(odp_port, 0), + &backer->odp_to_ofport_map) { + if (port->odp_port == odp_port) { + return port; + } + } + + return NULL; +} + +static uint16_t +odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port) +{ + struct ofport_dpif *port; + + port = odp_port_to_ofport(ofproto->backer, odp_port); + if (port && &ofproto->up == port->up.ofproto) { + return port->up.ofp_port; + } else { + 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, enumerate_types, enumerate_names, del, + port_open_type, + type_run, + type_run_fast, + type_wait, alloc, construct, destruct, @@ -7250,11 +9039,9 @@ 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_stp, get_stp_status, set_stp_port, @@ -7267,6 +9054,6 @@ const struct ofproto_class ofproto_dpif_class = { set_flood_vlans, is_mirror_output_bundle, forward_bpdu_changed, - set_mac_idle_time, + set_mac_table_config, set_realdev, };